Update HISTORY and version

Fix a race condition in FIFO size-based compaction where concurrent threads could select the same non-L0 file (#13946 )
Summary: **Context/Summary:** Fix a race condition (illustrated below) in FIFO size-based compaction where concurrent threads could select the same non-L0 file, causing assertion failures in debug builds or "Cannot delete table file from LSM tree" errors in release builds. ``` Thread 1 Thread 2 -------- -------- FIFO size-based compaction ↓ Pick L2 file ↓ Mark: file.being_compacted = true (file.being_compacted was false) ↓ WriteManifestStart (unlock mutex) ─→ FIFO size-based compaction starts ↓ ↓ Continue manifest write... Pick SAME L2 file ↓ Mark: file.being_compacted = true (file.being_compacted was true) ❌ ↓ ↓ ↓ Unlock mutex, wait for manifest ↓ ↓ Lock mutex ←─────────────────────────────────┘ ↓ Delete L2 file ✅ ↓ Complete ─────────────────────────────→ Try delete same file ❌ ↓ ERROR: "file not in LSM tree" 🐛 BUG: Both threads pick the same file! Thread 2 doesn't properly check file.being_compacted flag ``` **Test** New test that fails before the fix and passes after Pull Request resolved: https://github.com/facebook/rocksdb/pull/13946 Reviewed By: xingbowang Differential Revision: D82279731 Pulled By: hx235 fbshipit-source-id: b426517f2d1b23dd7d4951157822a2d322fe1435
2026-11-06 11:02:36 -05:00 · 2025-09-15 14:24:48 -07:00 · 2025-09-15 14:07:59 -07:00 · 2025-09-06 18:09:37 -07:00 · 2025-09-06 18:08:38 -07:00 · 2025-09-06 18:08:25 -07:00
225 changed files with 11811 additions and 4316 deletions
@@ -4,8 +4,8 @@ runs:
  steps:
  - name: Install Maven
    run: |
-      wget --no-check-certificate https://dlcdn.apache.org/maven/maven-3/3.9.10/binaries/apache-maven-3.9.10-bin.tar.gz
-      tar zxf apache-maven-3.9.10-bin.tar.gz
-      echo "export M2_HOME=$(pwd)/apache-maven-3.9.10" >> $GITHUB_ENV
-      echo "$(pwd)/apache-maven-3.9.10/bin" >> $GITHUB_PATH
+      wget --no-check-certificate https://archive.apache.org/dist/maven/maven-3/3.9.11/binaries/apache-maven-3.9.11-bin.tar.gz
+      tar zxf apache-maven-3.9.11-bin.tar.gz
+      echo "export M2_HOME=$(pwd)/apache-maven-3.9.11" >> $GITHUB_ENV
+      echo "$(pwd)/apache-maven-3.9.11/bin" >> $GITHUB_PATH
    shell: bash
@@ -27,18 +27,6 @@ jobs:
        git config --global --add safe.directory /__w/rocksdb/rocksdb
        tools/check_format_compatible.sh
    - uses: "./.github/actions/post-steps"
-  build-linux-run-microbench:
-    if: ${{ github.repository_owner == 'facebook' }}
-    runs-on:
-      labels: 16-core-ubuntu
-    container:
-      image: zjay437/rocksdb:0.6
-      options: --shm-size=16gb
-    steps:
-    - uses: actions/checkout@v4.1.0
-    - uses: "./.github/actions/pre-steps"
-    - run: DEBUG_LEVEL=0 make -j32 run_microbench
-    - uses: "./.github/actions/post-steps"
  build-linux-non-shm:
    if: ${{ github.repository_owner == 'facebook' }}
    runs-on:
@@ -91,15 +79,6 @@ jobs:
    steps:
    - uses: actions/checkout@v4.1.0
    - uses: "./.github/actions/windows-build-steps"
-  build-windows-vs2022:
-    if: ${{ github.repository_owner == 'facebook' }}
-    runs-on: windows-2022
-    env:
-      CMAKE_GENERATOR: Visual Studio 17 2022
-      CMAKE_PORTABLE: 1
-    steps:
-    - uses: actions/checkout@v4.1.0
-    - uses: "./.github/actions/windows-build-steps"
  build-linux-arm-test-full:
    if: ${{ github.repository_owner == 'facebook' }}
    runs-on:
@@ -110,3 +89,59 @@ jobs:
      - run: sudo apt-get update && sudo apt-get install -y build-essential libgflags-dev
      - run: make V=1 J=4 -j4 check
      - uses: "./.github/actions/post-steps"
+  build-examples:
+    if: ${{ github.repository_owner == 'facebook' }}
+    runs-on:
+      labels: 4-core-ubuntu
+    container:
+      image: zjay437/rocksdb:0.6
+      options: --shm-size=16gb
+    steps:
+    - uses: actions/checkout@v4.1.0
+    - uses: "./.github/actions/pre-steps"
+    - name: Build examples
+      run: make V=1 -j4 static_lib && cd examples && make V=1 -j4
+    - uses: "./.github/actions/post-steps"
+  build-fuzzers:
+    if: ${{ github.repository_owner == 'facebook' }}
+    runs-on:
+      labels: 4-core-ubuntu
+    container:
+      image: zjay437/rocksdb:0.6
+      options: --shm-size=16gb
+    steps:
+    - uses: actions/checkout@v4.1.0
+    - uses: "./.github/actions/pre-steps"
+    - name: Build rocksdb lib
+      run: CC=clang-13 CXX=clang++-13 USE_CLANG=1 make -j4 static_lib
+    - name: Build fuzzers
+      run: cd fuzz && make sst_file_writer_fuzzer db_fuzzer db_map_fuzzer
+    - uses: "./.github/actions/post-steps"
+  build-linux-gcc-11-no_test_run:
+    if: ${{ github.repository_owner == 'facebook' }}
+    runs-on:
+      labels: 16-core-ubuntu
+    container:
+      image: zjay437/rocksdb:0.6
+      options: --shm-size=16gb
+    steps:
+    - uses: actions/checkout@v4.1.0
+    - uses: "./.github/actions/pre-steps"
+    - run: LIB_MODE=static CC=gcc-11 CXX=g++-11 V=1 make -j32 all microbench
+    - uses: "./.github/actions/post-steps"
+  build-linux-cmake-with-folly-lite-no-test:
+    if: ${{ github.repository_owner == 'facebook' }}
+    runs-on:
+      labels: 16-core-ubuntu
+    container:
+      image: zjay437/rocksdb:0.6
+      options: --shm-size=16gb
+    env:
+      CC: gcc-10
+      CXX: g++-10
+    steps:
+    - uses: actions/checkout@v4.1.0
+    - uses: "./.github/actions/pre-steps"
+    - uses: "./.github/actions/setup-folly"
+    - run: "(mkdir build && cd build && cmake -DUSE_FOLLY_LITE=1 -DWITH_GFLAGS=1 .. && make V=1 -j20)"
+    - uses: "./.github/actions/post-steps"
@@ -109,22 +109,6 @@ jobs:
    - uses: "./.github/actions/build-folly"
    - run: "(mkdir build && cd build && cmake -DUSE_FOLLY=1 -DWITH_GFLAGS=1 -DROCKSDB_BUILD_SHARED=0 .. && make V=1 -j20 && ctest -j20)"
    - uses: "./.github/actions/post-steps"
-  build-linux-cmake-with-folly-lite-no-test:
-    if: ${{ github.repository_owner == 'facebook' }}
-    runs-on:
-      labels: 16-core-ubuntu
-    container:
-      image: zjay437/rocksdb:0.6
-      options: --shm-size=16gb
-    env:
-      CC: gcc-10
-      CXX: g++-10
-    steps:
-    - uses: actions/checkout@v4.1.0
-    - uses: "./.github/actions/pre-steps"
-    - uses: "./.github/actions/setup-folly"
-    - run: "(mkdir build && cd build && cmake -DUSE_FOLLY_LITE=1 -DWITH_GFLAGS=1 .. && make V=1 -j20)"
-    - uses: "./.github/actions/post-steps"
  build-linux-make-with-folly:
    if: ${{ github.repository_owner == 'facebook' }}
    runs-on:
@@ -198,7 +182,7 @@ jobs:
    - uses: actions/checkout@v4.1.0
    - uses: "./.github/actions/pre-steps"
    - run: ENCRYPTED_ENV=1 ROCKSDB_DISABLE_SNAPPY=1 ROCKSDB_DISABLE_ZLIB=1 ROCKSDB_DISABLE_BZIP=1 ROCKSDB_DISABLE_LZ4=1 ROCKSDB_DISABLE_ZSTD=1 make V=1 J=32 -j32 check
-    - run: "./sst_dump --help | grep -E -q 'Supported compression types: kNoCompression$' # Verify no compiled in compression\n"
+    - run: "./sst_dump --help | grep -E -q 'Supported built-in compression types: kNoCompression$' # Verify no compiled in compression\n"
    - uses: "./.github/actions/post-steps"
  # ======================== Linux No Test Runs ======================= #
  build-linux-release:
@@ -212,63 +196,20 @@ jobs:
    - uses: actions/checkout@v4.1.0
    - run: make V=1 -j32 LIB_MODE=shared release
    - run: ls librocksdb.so
-    - run: "./db_stress --version"
+    - run: "./trace_analyzer --version" # A tool dependent on gflags that can run in release build
    - run: make clean
-    - run: make V=1 -j32 release
+    - run: USE_RTTI=1 make V=1 -j32 release
    - run: ls librocksdb.a
-    - run: "./db_stress --version"
+    - run: "./trace_analyzer --version"
    - run: make clean
    - run: apt-get remove -y libgflags-dev
    - run: make V=1 -j32 LIB_MODE=shared release
    - run: ls librocksdb.so
-    - run: if ./db_stress --version; then false; else true; fi
+    - run: if ./trace_analyzer --version; then false; else true; fi
    - run: make clean
-    - run: make V=1 -j32 release
+    - run: USE_RTTI=1 make V=1 -j32 release
    - run: ls librocksdb.a
-    - run: if ./db_stress --version; then false; else true; fi
-    - uses: "./.github/actions/post-steps"
-  build-linux-release-rtti:
-    if: ${{ github.repository_owner == 'facebook' }}
-    runs-on:
-      labels: 8-core-ubuntu
-    container:
-      image: zjay437/rocksdb:0.6
-      options: --shm-size=16gb
-    steps:
-    - uses: actions/checkout@v4.1.0
-    - run: USE_RTTI=1 DEBUG_LEVEL=0 make V=1 -j16 static_lib tools db_bench
-    - run: "./db_stress --version"
-    - run: make clean
-    - run: apt-get remove -y libgflags-dev
-    - run: USE_RTTI=1 DEBUG_LEVEL=0 make V=1 -j16 static_lib tools db_bench
-    - run: if ./db_stress --version; then false; else true; fi
-  build-examples:
-    if: ${{ github.repository_owner == 'facebook' }}
-    runs-on:
-      labels: 4-core-ubuntu
-    container:
-      image: zjay437/rocksdb:0.6
-      options: --shm-size=16gb
-    steps:
-    - uses: actions/checkout@v4.1.0
-    - uses: "./.github/actions/pre-steps"
-    - name: Build examples
-      run: make V=1 -j4 static_lib && cd examples && make V=1 -j4
-    - uses: "./.github/actions/post-steps"
-  build-fuzzers:
-    if: ${{ github.repository_owner == 'facebook' }}
-    runs-on:
-      labels: 4-core-ubuntu
-    container:
-      image: zjay437/rocksdb:0.6
-      options: --shm-size=16gb
-    steps:
-    - uses: actions/checkout@v4.1.0
-    - uses: "./.github/actions/pre-steps"
-    - name: Build rocksdb lib
-      run: CC=clang-13 CXX=clang++-13 USE_CLANG=1 make -j4 static_lib
-    - name: Build fuzzers
-      run: cd fuzz && make sst_file_writer_fuzzer db_fuzzer db_map_fuzzer
+    - run: if ./trace_analyzer --version; then false; else true; fi
    - uses: "./.github/actions/post-steps"
  build-linux-clang-no_test_run:
    if: ${{ github.repository_owner == 'facebook' }}
@@ -292,6 +233,8 @@ jobs:
    - uses: actions/checkout@v4.1.0
    - uses: "./.github/actions/pre-steps"
    - run: CC=clang-13 CXX=clang++-13 USE_CLANG=1 make -j32 all microbench
+    - run: make clean
+    - run: CC=clang-13 CXX=clang++-13 USE_CLANG=1 DEBUG_LEVEL=0 make -j32 release
    - uses: "./.github/actions/post-steps"
  build-linux-gcc-8-no_test_run:
    if: ${{ github.repository_owner == 'facebook' }}
@@ -317,18 +260,7 @@ jobs:
    - uses: "./.github/actions/pre-steps"
    - run: CC=gcc-10 CXX=g++-10 V=1 ROCKSDB_CXX_STANDARD=c++20 make -j32 all
    - uses: "./.github/actions/post-steps"
-  build-linux-gcc-11-no_test_run:
-    if: ${{ github.repository_owner == 'facebook' }}
-    runs-on:
-      labels: 16-core-ubuntu
-    container:
-      image: zjay437/rocksdb:0.6
-      options: --shm-size=16gb
-    steps:
-    - uses: actions/checkout@v4.1.0
-    - uses: "./.github/actions/pre-steps"
-    - run: LIB_MODE=static CC=gcc-11 CXX=g++-11 V=1 make -j32 all microbench
-    - uses: "./.github/actions/post-steps"
+
  # ======================== Linux Other Checks ======================= #
  build-linux-clang10-clang-analyze:
    if: ${{ github.repository_owner == 'facebook' }}
@@ -376,7 +308,7 @@ jobs:
    - run: ulimit -S -n `ulimit -H -n` && make V=1 -j8 CRASH_TEST_EXT_ARGS='--duration=960 --max_key=2500000' blackbox_crash_test_with_atomic_flush
    - uses: "./.github/actions/post-steps"
  # ======================= Linux with Sanitizers ===================== #
-  build-linux-clang10-asan:
+  build-linux-clang10-asan-ubsan:
    if: ${{ github.repository_owner == 'facebook' }}
    runs-on:
      labels: 32-core-ubuntu
@@ -386,19 +318,7 @@ jobs:
    steps:
    - uses: actions/checkout@v4.1.0
    - uses: "./.github/actions/pre-steps"
-    - run: COMPILE_WITH_ASAN=1 CC=clang-10 CXX=clang++-10 ROCKSDB_DISABLE_ALIGNED_NEW=1 USE_CLANG=1 make V=1 -j32 check
-    - uses: "./.github/actions/post-steps"
-  build-linux-clang10-ubsan:
-    if: ${{ github.repository_owner == 'facebook' }}
-    runs-on:
-      labels: 16-core-ubuntu
-    container:
-      image: zjay437/rocksdb:0.6
-      options: --shm-size=16gb
-    steps:
-    - uses: actions/checkout@v4.1.0
-    - uses: "./.github/actions/pre-steps"
-    - run: COMPILE_WITH_UBSAN=1 CC=clang-10 CXX=clang++-10 ROCKSDB_DISABLE_ALIGNED_NEW=1 USE_CLANG=1 make V=1 -j32 ubsan_check
+    - run: COMPILE_WITH_ASAN=1 COMPILE_WITH_UBSAN=1 CC=clang-10 CXX=clang++-10 ROCKSDB_DISABLE_ALIGNED_NEW=1 USE_CLANG=1 make V=1 -j40 check
    - uses: "./.github/actions/post-steps"
  build-linux-clang13-mini-tsan:
    if: ${{ github.repository_owner == 'facebook' }}
@@ -469,11 +389,11 @@ jobs:
    - uses: "./.github/actions/post-steps"
  # ======================== Windows with Tests ======================= #
  # NOTE: some windows jobs are in "nightly" to save resources
-  build-windows-vs2019:
+  build-windows-vs2022:
    if: ${{ github.repository_owner == 'facebook' }}
-    runs-on: windows-2019
+    runs-on: windows-2022
    env:
-      CMAKE_GENERATOR: Visual Studio 16 2019
+      CMAKE_GENERATOR: Visual Studio 17 2022
      CMAKE_PORTABLE: 1
    steps:
    - uses: actions/checkout@v4.1.0
@@ -484,7 +404,7 @@ jobs:
    runs-on:
      labels: 4-core-ubuntu
    container:
-      image: evolvedbinary/rocksjava:centos6_x64-be
+      image: evolvedbinary/rocksjava:centos7_x64-be
      options: --shm-size=16gb
    steps:
    # The docker image is intentionally based on an OS that has an older GLIBC version.
@@ -512,7 +432,7 @@ jobs:
    runs-on:
      labels: 4-core-ubuntu
    container:
-      image: evolvedbinary/rocksjava:centos6_x64-be
+      image: evolvedbinary/rocksjava:centos7_x64-be
      options: --shm-size=16gb
    steps:
    # The docker image is intentionally based on an OS that has an older GLIBC version.
@@ -607,7 +527,7 @@ jobs:
    runs-on:
      labels: 4-core-ubuntu
    container:
-      image: evolvedbinary/rocksjava:rockylinux8_x64-be
+      image: evolvedbinary/rocksjava:alpine3_x64-be
      options: --shm-size=16gb
    steps:
    - uses: actions/checkout@v4.1.0
@@ -88,6 +88,7 @@ cpp_library_wrapper(name="rocksdb_lib", srcs=[
        "db/memtable_list.cc",
        "db/merge_helper.cc",
        "db/merge_operator.cc",
+        "db/multi_scan.cc",
        "db/output_validator.cc",
        "db/periodic_task_scheduler.cc",
        "db/range_del_aggregator.cc",
@@ -113,6 +114,7 @@ cpp_library_wrapper(name="rocksdb_lib", srcs=[
        "db/write_controller.cc",
        "db/write_stall_stats.cc",
        "db/write_thread.cc",
+        "db_stress_tool/db_stress_compression_manager.cc",
        "env/composite_env.cc",
        "env/env.cc",
        "env/env_chroot.cc",
@@ -249,7 +251,7 @@ cpp_library_wrapper(name="rocksdb_lib", srcs=[
        "trace_replay/trace_record_result.cc",
        "trace_replay/trace_replay.cc",
        "util/async_file_reader.cc",
-        "util/auto_skip_compressor.cc",
+        "util/auto_tune_compressor.cc",
        "util/build_version.cc",
        "util/cleanable.cc",
        "util/coding.cc",
@@ -421,6 +423,7 @@ rocks_cpp_library_wrapper(name="rocksdb_stress_lib", srcs=[
        "db_stress_tool/batched_ops_stress.cc",
        "db_stress_tool/cf_consistency_stress.cc",
        "db_stress_tool/db_stress_common.cc",
+        "db_stress_tool/db_stress_compression_manager.cc",
        "db_stress_tool/db_stress_driver.cc",
        "db_stress_tool/db_stress_filters.cc",
        "db_stress_tool/db_stress_gflags.cc",
@@ -5193,6 +5196,12 @@ cpp_unittest_wrapper(name="inlineskiplist_test",
            extra_compiler_flags=[])


+cpp_unittest_wrapper(name="interval_test",
+            srcs=["util/interval_test.cc"],
+            deps=[":rocksdb_test_lib"],
+            extra_compiler_flags=[])
+
+
 cpp_unittest_wrapper(name="io_posix_test",
            srcs=["env/io_posix_test.cc"],
            deps=[":rocksdb_test_lib"],
@@ -721,6 +721,7 @@ set(SOURCES
        db/memtable_list.cc
        db/merge_helper.cc
        db/merge_operator.cc
+        db/multi_scan.cc
        db/output_validator.cc
        db/periodic_task_scheduler.cc
        db/range_del_aggregator.cc
@@ -746,6 +747,7 @@ set(SOURCES
        db/write_controller.cc
        db/write_stall_stats.cc
        db/write_thread.cc
+        db_stress_tool/db_stress_compression_manager.cc
        env/composite_env.cc
        env/env.cc
        env/env_chroot.cc
@@ -874,7 +876,7 @@ set(SOURCES
        trace_replay/trace_record.cc
        trace_replay/trace_replay.cc
        util/async_file_reader.cc
-        util/auto_skip_compressor.cc
+        util/auto_tune_compressor.cc
        util/cleanable.cc
        util/coding.cc
        util/compaction_job_stats_impl.cc
@@ -1,6 +1,79 @@
 # Rocksdb Change Log
 > NOTE: Entries for next release do not go here. Follow instructions in `unreleased_history/README.txt`

+## 10.6.2 (09/15/2025)
+### Bug Fixes
+* Fix a race condition in FIFO size-based compaction where concurrent threads could select the same non-L0 file, causing assertion failures in debug builds or "Cannot delete table file from LSM tree" errors in release builds.
+
+## 10.6.1 (09/05/2025)
+### New Features
+* Add the fail_if_no_udi_on_open flag in BlockBasedTableOption to control whether a missing user defined index block in a SST is a hard error or not.
+* Add new option `MultiScanArgs::max_prefetch_size` that limits the memory usage of per file pinning of prefetched blocks.
+
+## 10.6.0 (08/22/2025)
+### New Features
+* Introduce column family option `cf_allow_ingest_behind`. This option aims to replace `DBOptions::allow_ingest_behind` to enable ingest behind at the per-CF level. `DBOptions::allow_ingest_behind` is deprecated.
+* Introduce `MultiScanArgs::io_coalesce_threshold` to allow a configurable IO coalescing threshold.
+
+### Public API Changes
+* `IngestExternalFileOptions::allow_db_generated_files` now allows files ingestion of any DB generated SST file, instead of only the ones with all keys having sequence number 0.
+* `decouple_partitioned_filters = true` is now the default in BlockBasedTableOptions.
+* GetTtl() API is now available in TTL DB
+* Minimum supported version of LZ4 library is now 1.7.0 (r129 from 2015)
+* Some changes to experimental Compressor and CompressionManager APIs
+* A new Filesystem::SyncFile function is added for syncing a file that was already written, such as on file ingestion. The default implementation matches previous RocksDB behavior: re-open the file for read-write, sync it, and close it. We recommend overriding for FileSystems that do not require syncing for crash recovery or do not handle (well) re-opening for writes.
+
+### Behavior Changes
+* When `allow_ingest_behind` is enabled, compaction will no longer drop tombstones based on the absence of underlying data. Tombstones will be preserved to apply to ingested files.
+
+### Bug Fixes
+* Files in dropped column family won't be returned to the caller upon successful, offline MANIFEST iteration in `GetFileChecksumsFromCurrentManifest`.
+* Fix a bug in MultiScan that causes it to fall back to a normal scan when dictionary compression is enabled.
+* Fix a crash in iterator Prepare() when fill_cache=false
+* Fix a bug in MultiScan where incorrect results can be returned when a Scan's range is across multiple files.
+* Fixed a bug in remote compaction that may mistakenly delete live SST file(s) during the cleanup phase when no keys survive the compaction (all expired)
+* Allow a user defined index to be configured from a string.
+* Make the User Defined Index interface consistently use the user key format, fixing the previous mixed usage of internal and user key.
+
+### Performance Improvements
+* Small improvement to CPU efficiency of compression using built-in algorithms, and a dramatic efficiency improvement for LZ4HC, based on reusing data structures between invocations.
+
+## 10.5.0 (07/18/2025)
+### Public API Changes
+* DB option skip_checking_sst_file_sizes_on_db_open is deprecated, in favor of validating file size in parallel in a thread pool, when db is opened. When DB is opened, with paranoid check enabled, a file with the wrong size would fail the DB open. With paranoid check disabled, the DB open would succeed, the column family with the corrupted file would not be read or write, while the other healthy column families could be read and write normally. When max_open_files option is not set to -1, only a subset of the files will be opened and checked. The rest of the files will be opened and checked when they are accessed.
+
+### Behavior Changes
+* PessimisticTransaction::GetWaitingTxns now returns waiting transaction information even if the current transaction has timed out. This allows the information to be surfaced to users for debugging purposes once it is known that the timeout has occured.
+* A new API GetFileSize is added to FSRandomAccessFile interface class. It uses fstat vs stat on the posix implementation which is more efficient. Caller could use it to get file size faster. This function might be required in the future for FileSystem implementation outside of the RocksDB code base.
+* RocksDB now triggers eligible compactions every 12 hours when periodic compaction is configured. This solves a limitation of the compaction trigger mechanism, which would only trigger compaction after specific events like flush, compaction, or SetOptions.
+
+### Bug Fixes
+* Fix a bug in BackupEngine that can crash backup due to a null FSWritableFile passed to WritableFileWriter.
+* Fix DB::NewMultiScan iterator to respect the scan upper bound specified in ScanOptions
+
+### Performance Improvements
+* Optimized MultiScan using BlockBasedTable to coalesce I/Os and prefetch all data blocks.
+
+## 10.4.0 (06/20/2025)
+### New Features
+* Add a new CF option `memtable_avg_op_scan_flush_trigger` that supports triggering memtable flush when an iterator scans through an expensive range of keys, with the average number of skipped keys from the active memtable exceeding the threshold.
+* Vector based memtable now supports concurrent writers (DBOptions::allow_concurrent_memtable_write) #13675.
+* Add new experimental `TransactionOptions::large_txn_commit_optimize_byte_threshold` to enable optimizations for large transaction commit by transaction batch data size.
+* Add a new option `CompactionOptionsUniversal::reduce_file_locking` and if it's true, auto universal compaction picking will adjust to minimize locking of input files when bottom priority compactions are waiting to run. This can increase the likelihood of existing L0s being selected for compaction, thereby improving write stall and reducing read regression.
+* Add new `format_version=7` to aid experimental support of custom compression algorithms with CompressionManager and block-based table. This format version includes changing the format of `TableProperties::compression_name`.
+
+### Public API Changes
+* Change NewExternalTableFactory to return a unique_ptr instead of shared_ptr.
+* Add an optional min file size requirement for deletion triggered compaction. It can be specified when creating `CompactOnDeletionCollectorFactory`.
+
+### Behavior Changes
+* `TransactionOptions::large_txn_commit_optimize_threshold` now has default value 0 for disabled. `TransactionDBOptions::txn_commit_bypass_memtable_threshold` now has no effect on transactions.
+
+### Bug Fixes
+* Fix a bug where CreateColumnFamilyWithImport() could miss the SST file for the memtable flush it triggered. The exported CF then may not contain the updates in the memtable when CreateColumnFamilyWithImport() is called.
+* Fix iterator operations returning NotImplemented status if disallow_memtable_writes and paranoid_memory_checks CF options are both set.
+* Fixed handling of file checksums in IngestExternalFile() to allow providing checksums using recognized but not necessarily the DB's preferred checksum function, to ease migration between checksum functions.
+
 ## 10.3.0 (05/17/2025)
 ### New Features
 * Add new experimental `CompactionOptionsFIFO::allow_trivial_copy_when_change_temperature` along with `CompactionOptionsFIFO::trivial_copy_buffer_size` to allow optimizing FIFO compactions with tiering when kChangeTemperature to move files from source tier FileSystem to another tier FileSystem via trivial and direct copying raw sst file instead of reading thru the content of the SST file then rebuilding the table files.
@@ -1357,6 +1357,9 @@ filter_bench: $(OBJ_DIR)/util/filter_bench.o $(LIBRARY)
 db_stress: $(OBJ_DIR)/db_stress_tool/db_stress.o $(STRESS_LIBRARY) $(TOOLS_LIBRARY) $(LIBRARY)
 	$(AM_LINK)

+db_stress_compression_manager: $(OBJ_DIR)/db_stress_tool/db_stress_compression_manager.o $(LIBRARY)
+	$(AM_LINK)
+
 write_stress: $(OBJ_DIR)/tools/write_stress.o $(LIBRARY)
 	$(AM_LINK)

@@ -2037,6 +2040,9 @@ wide_column_serialization_test: $(OBJ_DIR)/db/wide/wide_column_serialization_tes
 wide_columns_helper_test: $(OBJ_DIR)/db/wide/wide_columns_helper_test.o $(TEST_LIBRARY) $(LIBRARY)
 	$(AM_LINK)

+interval_test: $(OBJ_DIR)/util/interval_test.o $(TEST_LIBRARY) $(LIBRARY)
+	$(AM_LINK)
+
 #-------------------------------------------------
 # make install related stuff
 PREFIX ?= /usr/local
@@ -2147,14 +2153,14 @@ ZLIB_DOWNLOAD_BASE ?= http://zlib.net
 BZIP2_VER ?= 1.0.8
 BZIP2_SHA256 ?= ab5a03176ee106d3f0fa90e381da478ddae405918153cca248e682cd0c4a2269
 BZIP2_DOWNLOAD_BASE ?= http://sourceware.org/pub/bzip2
-SNAPPY_VER ?= 1.2.1
-SNAPPY_SHA256 ?= 736aeb64d86566d2236ddffa2865ee5d7a82d26c9016b36218fcc27ea4f09f86
+SNAPPY_VER ?= 1.2.2
+SNAPPY_SHA256 ?= 90f74bc1fbf78a6c56b3c4a082a05103b3a56bb17bca1a27e052ea11723292dc
 SNAPPY_DOWNLOAD_BASE ?= https://github.com/google/snappy/archive
-LZ4_VER ?= 1.9.4
-LZ4_SHA256 ?= 0b0e3aa07c8c063ddf40b082bdf7e37a1562bda40a0ff5272957f3e987e0e54b
+LZ4_VER ?= 1.10.0
+LZ4_SHA256 ?= 537512904744b35e232912055ccf8ec66d768639ff3abe5788d90d792ec5f48b
 LZ4_DOWNLOAD_BASE ?= https://github.com/lz4/lz4/archive
-ZSTD_VER ?= 1.5.5
-ZSTD_SHA256 ?= 98e9c3d949d1b924e28e01eccb7deed865eefebf25c2f21c702e5cd5b63b85e1
+ZSTD_VER ?= 1.5.7
+ZSTD_SHA256 ?= 37d7284556b20954e56e1ca85b80226768902e2edabd3b649e9e72c0c9012ee3
 ZSTD_DOWNLOAD_BASE ?= https://github.com/facebook/zstd/archive
 CURL_SSL_OPTS ?= --tlsv1

@@ -2375,27 +2381,27 @@ rocksdbjavastaticreleasedocker: rocksdbjavastaticosx rocksdbjavastaticdockerx86

 rocksdbjavastaticdockerx86:
 	mkdir -p java/target
-	docker run --rm --name rocksdb_linux_x86-be --platform linux/386 --attach stdin --attach stdout --attach stderr --volume $(HOME)/.m2:/root/.m2:ro --volume `pwd`:/rocksdb-host:ro --volume /rocksdb-local-build --volume `pwd`/java/target:/rocksdb-java-target --env DEBUG_LEVEL=$(DEBUG_LEVEL) --env J=$(J) evolvedbinary/rocksjava:centos6_x86-be /rocksdb-host/java/crossbuild/docker-build-linux.sh
+	docker run --rm --name rocksdb_linux_x86-be --platform linux/386 --attach stdin --attach stdout --attach stderr --volume $(HOME)/.m2:/root/.m2:ro --volume `pwd`:/rocksdb-host:ro --volume /rocksdb-local-build --volume `pwd`/java/target:/rocksdb-java-target --env DEBUG_LEVEL=$(DEBUG_LEVEL) --env J=$(J) evolvedbinary/rocksjava:centos7_x86-be /rocksdb-host/java/crossbuild/docker-build-linux.sh

 rocksdbjavastaticdockerx86_64:
 	mkdir -p java/target
-	docker run --rm --name rocksdb_linux_x64-be --attach stdin --attach stdout --attach stderr --volume $(HOME)/.m2:/root/.m2:ro --volume `pwd`:/rocksdb-host:ro --volume /rocksdb-local-build --volume `pwd`/java/target:/rocksdb-java-target --env DEBUG_LEVEL=$(DEBUG_LEVEL) --env J=$(J) evolvedbinary/rocksjava:centos6_x64-be /rocksdb-host/java/crossbuild/docker-build-linux.sh
+	docker run --rm --name rocksdb_linux_x64-be --platform linux/amd64 --attach stdin --attach stdout --attach stderr --volume $(HOME)/.m2:/root/.m2:ro --volume `pwd`:/rocksdb-host:ro --volume /rocksdb-local-build --volume `pwd`/java/target:/rocksdb-java-target --env DEBUG_LEVEL=$(DEBUG_LEVEL) --env J=$(J) evolvedbinary/rocksjava:centos7_x64-be /rocksdb-host/java/crossbuild/docker-build-linux.sh

 rocksdbjavastaticdockerppc64le:
 	mkdir -p java/target
-	docker run --rm --name rocksdb_linux_ppc64le-be --attach stdin --attach stdout --attach stderr --volume $(HOME)/.m2:/root/.m2:ro --volume `pwd`:/rocksdb-host:ro --volume /rocksdb-local-build --volume `pwd`/java/target:/rocksdb-java-target --env DEBUG_LEVEL=$(DEBUG_LEVEL) --env J=$(J) evolvedbinary/rocksjava:centos7_ppc64le-be /rocksdb-host/java/crossbuild/docker-build-linux.sh
+	docker run --rm --name rocksdb_linux_ppc64le-be --platform linux/ppc64le --attach stdin --attach stdout --attach stderr --volume $(HOME)/.m2:/root/.m2:ro --volume `pwd`:/rocksdb-host:ro --volume /rocksdb-local-build --volume `pwd`/java/target:/rocksdb-java-target --env DEBUG_LEVEL=$(DEBUG_LEVEL) --env J=$(J) evolvedbinary/rocksjava:centos7_ppc64le-be /rocksdb-host/java/crossbuild/docker-build-linux.sh

 rocksdbjavastaticdockerarm64v8:
 	mkdir -p java/target
-	docker run --rm --name rocksdb_linux_arm64v8-be --attach stdin --attach stdout --attach stderr --volume $(HOME)/.m2:/root/.m2:ro --volume `pwd`:/rocksdb-host:ro --volume /rocksdb-local-build --volume `pwd`/java/target:/rocksdb-java-target --env DEBUG_LEVEL=$(DEBUG_LEVEL) --env J=$(J) evolvedbinary/rocksjava:centos7_arm64v8-be /rocksdb-host/java/crossbuild/docker-build-linux.sh
+	docker run --rm --name rocksdb_linux_arm64v8-be --platform linux/aarch64 --attach stdin --attach stdout --attach stderr --volume $(HOME)/.m2:/root/.m2:ro --volume `pwd`:/rocksdb-host:ro --volume /rocksdb-local-build --volume `pwd`/java/target:/rocksdb-java-target --env DEBUG_LEVEL=$(DEBUG_LEVEL) --env J=$(J) evolvedbinary/rocksjava:centos7_arm64v8-be /rocksdb-host/java/crossbuild/docker-build-linux.sh

 rocksdbjavastaticdockers390x:
 	mkdir -p java/target
-	docker run --rm --name rocksdb_linux_s390x-be --attach stdin --attach stdout --attach stderr --volume $(HOME)/.m2:/root/.m2:ro --volume `pwd`:/rocksdb-host:ro --volume /rocksdb-local-build --volume `pwd`/java/target:/rocksdb-java-target --env DEBUG_LEVEL=$(DEBUG_LEVEL) --env J=$(J) evolvedbinary/rocksjava:ubuntu18_s390x-be /rocksdb-host/java/crossbuild/docker-build-linux.sh
+	docker run --rm --name rocksdb_linux_s390x-be --platform linux/s390x --attach stdin --attach stdout --attach stderr --volume $(HOME)/.m2:/root/.m2:ro --volume `pwd`:/rocksdb-host:ro --volume /rocksdb-local-build --volume `pwd`/java/target:/rocksdb-java-target --env DEBUG_LEVEL=$(DEBUG_LEVEL) --env J=$(J) evolvedbinary/rocksjava:ubuntu18_s390x-be /rocksdb-host/java/crossbuild/docker-build-linux.sh

 rocksdbjavastaticdockerriscv64:
 	mkdir -p java/target
-	docker run --rm --name rocksdb_linux_riscv64-be --attach stdin --attach stdout --attach stderr --volume $(HOME)/.m2:/root/.m2:ro --volume `pwd`:/rocksdb-host:ro --volume /rocksdb-local-build --volume `pwd`/java/target:/rocksdb-java-target --env DEBUG_LEVEL=$(DEBUG_LEVEL) --env J=$(J) evolvedbinary/rocksjava:ubuntu20_riscv64-be /rocksdb-host/java/crossbuild/docker-build-linux.sh
+	docker run --rm --name rocksdb_linux_riscv64-be --platform linux/riscv64 --attach stdin --attach stdout --attach stderr --volume $(HOME)/.m2:/root/.m2:ro --volume `pwd`:/rocksdb-host:ro --volume /rocksdb-local-build --volume `pwd`/java/target:/rocksdb-java-target --env DEBUG_LEVEL=$(DEBUG_LEVEL) --env J=$(J) evolvedbinary/rocksjava:ubuntu20_riscv64-be /rocksdb-host/java/crossbuild/docker-build-linux.sh

 rocksdbjavastaticdockerx86musl:
 	mkdir -p java/target
@@ -2403,19 +2409,19 @@ rocksdbjavastaticdockerx86musl:

 rocksdbjavastaticdockerx86_64musl:
 	mkdir -p java/target
-	docker run --rm --name rocksdb_linux_x64-musl-be --attach stdin --attach stdout --attach stderr --volume $(HOME)/.m2:/root/.m2:ro --volume `pwd`:/rocksdb-host:ro --volume /rocksdb-local-build --volume `pwd`/java/target:/rocksdb-java-target --env DEBUG_LEVEL=$(DEBUG_LEVEL) --env J=$(J) evolvedbinary/rocksjava:alpine3_x64-be /rocksdb-host/java/crossbuild/docker-build-linux.sh
+	docker run --rm --name rocksdb_linux_x64-musl-be --platform linux/amd64 --attach stdin --attach stdout --attach stderr --volume $(HOME)/.m2:/root/.m2:ro --volume `pwd`:/rocksdb-host:ro --volume /rocksdb-local-build --volume `pwd`/java/target:/rocksdb-java-target --env DEBUG_LEVEL=$(DEBUG_LEVEL) --env J=$(J) evolvedbinary/rocksjava:alpine3_x64-be /rocksdb-host/java/crossbuild/docker-build-linux.sh

 rocksdbjavastaticdockerppc64lemusl:
 	mkdir -p java/target
-	docker run --rm --name rocksdb_linux_ppc64le-musl-be --attach stdin --attach stdout --attach stderr --volume $(HOME)/.m2:/root/.m2:ro --volume `pwd`:/rocksdb-host:ro --volume /rocksdb-local-build --volume `pwd`/java/target:/rocksdb-java-target --env DEBUG_LEVEL=$(DEBUG_LEVEL) --env J=$(J) evolvedbinary/rocksjava:alpine3_ppc64le-be /rocksdb-host/java/crossbuild/docker-build-linux.sh
+	docker run --rm --name rocksdb_linux_ppc64le-musl-be --platform linux/ppc64le --attach stdin --attach stdout --attach stderr --volume $(HOME)/.m2:/root/.m2:ro --volume `pwd`:/rocksdb-host:ro --volume /rocksdb-local-build --volume `pwd`/java/target:/rocksdb-java-target --env DEBUG_LEVEL=$(DEBUG_LEVEL) --env J=$(J) evolvedbinary/rocksjava:alpine3_ppc64le-be /rocksdb-host/java/crossbuild/docker-build-linux.sh

 rocksdbjavastaticdockerarm64v8musl:
 	mkdir -p java/target
-	docker run --rm --name rocksdb_linux_arm64v8-musl-be --attach stdin --attach stdout --attach stderr --volume $(HOME)/.m2:/root/.m2:ro --volume `pwd`:/rocksdb-host:ro --volume /rocksdb-local-build --volume `pwd`/java/target:/rocksdb-java-target --env DEBUG_LEVEL=$(DEBUG_LEVEL) --env J=$(J) evolvedbinary/rocksjava:alpine3_arm64v8-be /rocksdb-host/java/crossbuild/docker-build-linux.sh
+	docker run --rm --name rocksdb_linux_arm64v8-musl-be --platform linux/aarch64 --attach stdin --attach stdout --attach stderr --volume $(HOME)/.m2:/root/.m2:ro --volume `pwd`:/rocksdb-host:ro --volume /rocksdb-local-build --volume `pwd`/java/target:/rocksdb-java-target --env DEBUG_LEVEL=$(DEBUG_LEVEL) --env J=$(J) evolvedbinary/rocksjava:alpine3_arm64v8-be /rocksdb-host/java/crossbuild/docker-build-linux.sh

 rocksdbjavastaticdockers390xmusl:
 	mkdir -p java/target
-	docker run --rm --name rocksdb_linux_s390x-musl-be --attach stdin --attach stdout --attach stderr --volume $(HOME)/.m2:/root/.m2:ro --volume `pwd`:/rocksdb-host:ro --volume /rocksdb-local-build --volume `pwd`/java/target:/rocksdb-java-target --env DEBUG_LEVEL=$(DEBUG_LEVEL) --env J=$(J) evolvedbinary/rocksjava:alpine3_s390x-be /rocksdb-host/java/crossbuild/docker-build-linux.sh
+	docker run --rm --name rocksdb_linux_s390x-musl-be --platform linux/s390x --attach stdin --attach stdout --attach stderr --volume $(HOME)/.m2:/root/.m2:ro --volume `pwd`:/rocksdb-host:ro --volume /rocksdb-local-build --volume `pwd`/java/target:/rocksdb-java-target --env DEBUG_LEVEL=$(DEBUG_LEVEL) --env J=$(J) evolvedbinary/rocksjava:alpine3_s390x-be /rocksdb-host/java/crossbuild/docker-build-linux.sh

 rocksdbjavastaticpublish: rocksdbjavastaticrelease rocksdbjavastaticpublishcentral

@@ -2470,8 +2476,8 @@ jtest_run:
 jtest: rocksdbjava
 	cd java;$(MAKE) sample test

-jpmd: rocksdbjava rocksdbjavageneratepom
-	cd java;$(MAKE) pmd
+jpmd: rocksdbjavageneratepom
+	cd java;$(MAKE) java java_test pmd

 jdb_bench:
 	cd java;$(MAKE) db_bench;
@@ -2492,7 +2498,7 @@ checkout_folly:
 	fi
 	@# Pin to a particular version for public CI, so that PR authors don't
 	@# need to worry about folly breaking our integration. Update periodically
-	cd third-party/folly && git reset --hard d17bf897cb5bbf8f07b122a614e8cffdc38edcde
+	cd third-party/folly && git reset --hard e95383b7c8b5b1e46cf47acf2f317d54f93c8268
 	@# Apparently missing include
 	perl -pi -e 's/(#include <atomic>)/$$1\n#include <cstring>/' third-party/folly/folly/lang/Exception.h
 	@# Warning-as-error on memcpy
@@ -55,8 +55,11 @@ fi
 # we currently depend on POSIX platform
 COMMON_FLAGS="-DROCKSDB_PLATFORM_POSIX -DROCKSDB_LIB_IO_POSIX"

-# Default to fbcode gcc on internal fb machines
-if [ -z "$ROCKSDB_NO_FBCODE" -a -d /mnt/gvfs/third-party ]; then
+# Default to fbcode gcc on Meta internal machines
+IS_META_HOST="$(hostname | grep -E '(facebook|meta).com|fbinfra.net')"
+if [ -z "$ROCKSDB_NO_FBCODE" -a "$IS_META_HOST" ]; then
+  if [ -d /mnt/gvfs/third-party ]; then
+    echo "NOTE: Using fbcode build" >&2
    FBCODE_BUILD="true"
    # If we're compiling with TSAN or shared lib, we need pic build
    PIC_BUILD=$COMPILE_WITH_TSAN
@@ -64,6 +67,11 @@ if [ -z "$ROCKSDB_NO_FBCODE" -a -d /mnt/gvfs/third-party ]; then
      PIC_BUILD=1
    fi
    source "$PWD/build_tools/fbcode_config_platform010.sh"
+  else
+    echo "************************************************************************" >&2
+    echo "WARNING: -d /mnt/gvfs/third-party failed; no fbcode build" >&2
+    echo "************************************************************************" >&2
+  fi
 fi

 # Delete existing output, if it exists
@@ -71,7 +79,9 @@ rm -f "$OUTPUT"
 touch "$OUTPUT"

 if test -z "$CC"; then
-    if [ -x "$(command -v cc)" ]; then
+    if [ "$USE_CLANG" -a -x "$(command -v clang)" ]; then
+        CC=clang
+    elif [ -x "$(command -v cc)" ]; then
        CC=cc
    elif [ -x "$(command -v clang)" ]; then
        CC=clang
@@ -81,7 +91,9 @@ if test -z "$CC"; then
 fi

 if test -z "$CXX"; then
-    if [ -x "$(command -v g++)" ]; then
+    if [ "$USE_CLANG" -a -x "$(command -v clang++)" ]; then
+        CXX=clang++
+    elif [ -x "$(command -v g++)" ]; then
        CXX=g++
    elif [ -x "$(command -v clang++)" ]; then
        CXX=clang++
@@ -91,7 +103,9 @@ if test -z "$CXX"; then
 fi

 if test -z "$AR"; then
-    if [ -x "$(command -v gcc-ar)" ]; then
+    if [ "$USE_CLANG" -a -x "$(command -v llvm-ar)" ]; then
+        AR=llvm-ar
+    elif [ -x "$(command -v gcc-ar)" ]; then
        AR=gcc-ar
    elif [ -x "$(command -v llvm-ar)" ]; then
        AR=llvm-ar
@@ -60,6 +60,8 @@ DEFINE_uint32(value_bytes, 8 * KiB, "Size of each value added.");
 DEFINE_uint32(value_bytes_estimate, 0,
              "If > 0, overrides estimated_entry_charge or "
              "min_avg_entry_charge depending on cache_type.");
+DEFINE_double(compressible_to_ratio, 0.5,
+              "Approximate size ratio that values can be compressed to.");

 DEFINE_int32(
    degenerate_hash_bits, 0,
@@ -291,10 +293,19 @@ struct KeyGen {

 Cache::ObjectPtr createValue(Random64& rnd, MemoryAllocator* alloc) {
  char* rv = AllocateBlock(FLAGS_value_bytes, alloc).release();
-  // Fill with some filler data, and take some CPU time
-  for (uint32_t i = 0; i < FLAGS_value_bytes; i += 8) {
+  // Fill with some filler data, and take some CPU time, but add redundancy
+  // as requested for compressibility.
+  uint32_t random_fill_size = std::max(
+      uint32_t{1}, std::min(FLAGS_value_bytes,
+                            static_cast<uint32_t>(FLAGS_compressible_to_ratio *
+                                                  FLAGS_value_bytes)));
+  uint32_t i = 0;
+  for (; i < random_fill_size; i += 8) {
    EncodeFixed64(rv + i, rnd.Next());
  }
+  for (; i < FLAGS_value_bytes; i++) {
+    rv[i] = rv[i % random_fill_size];
+  }
  return rv;
 }

@@ -309,16 +320,16 @@ Status SaveToFn(Cache::ObjectPtr from_obj, size_t /*from_offset*/,

 Status CreateFn(const Slice& data, CompressionType /*type*/,
                CacheTier /*source*/, Cache::CreateContext* /*context*/,
-                MemoryAllocator* /*allocator*/, Cache::ObjectPtr* out_obj,
+                MemoryAllocator* alloc, Cache::ObjectPtr* out_obj,
                size_t* out_charge) {
-  *out_obj = new char[data.size()];
+  *out_obj = AllocateBlock(data.size(), alloc).release();
  memcpy(*out_obj, data.data(), data.size());
  *out_charge = data.size();
  return Status::OK();
 };

 void DeleteFn(Cache::ObjectPtr value, MemoryAllocator* alloc) {
-  CustomDeleter{alloc}(static_cast<char*>(value));
+  CacheAllocationDeleter{alloc}(static_cast<char*>(value));
 }

 Cache::CacheItemHelper helper1_wos(CacheEntryRole::kDataBlock, DeleteFn);
@@ -16,6 +16,31 @@
 #include "util/string_util.h"

 namespace ROCKSDB_NAMESPACE {
+namespace {
+// Format of values in CompressedSecondaryCache:
+// If enable_custom_split_merge:
+//  * A chain of CacheValueChunk representing the sequence of bytes for a tagged
+//    value. The overall length of the tagged value is determined by the chain
+//    of CacheValueChunks.
+// If !enable_custom_split_merge:
+//  * A LengthPrefixedSlice (starts with varint64 size) of a tagged value.
+//
+// A tagged value has a 2-byte header before the "saved" or compressed block
+// data:
+//  * 1 byte for "source" CacheTier indicating which tier is responsible for
+//    compression/decompression.
+//  * 1 byte for compression type which is generated/used by
+//    CompressedSecondaryCache iff source == CacheTier::kVolatileCompressedTier
+//    (original entry passed in was uncompressed). Otherwise, the compression
+//    type is preserved from the entry passed in.
+constexpr uint32_t kTagSize = 2;
+
+// Size of tag + varint size prefix when applicable
+uint32_t GetHeaderSize(size_t data_size, bool enable_split_merge) {
+  return (enable_split_merge ? 0 : VarintLength(kTagSize + data_size)) +
+         kTagSize;
+}
+}  // namespace

 CompressedSecondaryCache::CompressedSecondaryCache(
    const CompressedSecondaryCacheOptions& opts)
@@ -40,13 +65,9 @@ std::unique_ptr<SecondaryCacheResultHandle> CompressedSecondaryCache::Lookup(
    Cache::CreateContext* create_context, bool /*wait*/, bool advise_erase,
    Statistics* stats, bool& kept_in_sec_cache) {
  assert(helper);
-  // This is a minor optimization. Its ok to skip it in TSAN in order to
-  // avoid a false positive.
-#ifndef __SANITIZE_THREAD__
-  if (disable_cache_) {
+  if (disable_cache_.LoadRelaxed()) {
    return nullptr;
  }
-#endif

  std::unique_ptr<SecondaryCacheResultHandle> handle;
  kept_in_sec_cache = false;
@@ -62,75 +83,58 @@ std::unique_ptr<SecondaryCacheResultHandle> CompressedSecondaryCache::Lookup(
    return nullptr;
  }

-  CacheAllocationPtr* ptr{nullptr};
-  CacheAllocationPtr merged_value;
-  size_t handle_value_charge{0};
-  const char* data_ptr = nullptr;
-  CacheTier source = CacheTier::kVolatileCompressedTier;
-  CompressionType type = cache_options_.compression_type;
+  std::string merged_value;
+  Slice tagged_data;
  if (cache_options_.enable_custom_split_merge) {
    CacheValueChunk* value_chunk_ptr =
-        reinterpret_cast<CacheValueChunk*>(handle_value);
-    merged_value = MergeChunksIntoValue(value_chunk_ptr, handle_value_charge);
-    ptr = &merged_value;
-    data_ptr = ptr->get();
+        static_cast<CacheValueChunk*>(handle_value);
+    merged_value = MergeChunksIntoValue(value_chunk_ptr);
+    tagged_data = Slice(merged_value);
  } else {
-    uint32_t type_32 = static_cast<uint32_t>(type);
-    uint32_t source_32 = static_cast<uint32_t>(source);
-    ptr = reinterpret_cast<CacheAllocationPtr*>(handle_value);
-    handle_value_charge = cache_->GetCharge(lru_handle);
-    data_ptr = ptr->get();
-    const char* limit = ptr->get() + handle_value_charge;
-    data_ptr =
-        GetVarint32Ptr(data_ptr, limit, static_cast<uint32_t*>(&type_32));
-    type = static_cast<CompressionType>(type_32);
-    data_ptr =
-        GetVarint32Ptr(data_ptr, limit, static_cast<uint32_t*>(&source_32));
-    source = static_cast<CacheTier>(source_32);
-    uint64_t data_size = 0;
-    data_ptr =
-        GetVarint64Ptr(data_ptr, limit, static_cast<uint64_t*>(&data_size));
-    assert(handle_value_charge > data_size);
-    handle_value_charge = data_size;
+    tagged_data = GetLengthPrefixedSlice(static_cast<char*>(handle_value));
  }
-  MemoryAllocator* allocator = cache_options_.memory_allocator.get();

-  Status s;
-  Cache::ObjectPtr value{nullptr};
-  size_t charge{0};
+  auto source = lossless_cast<CacheTier>(tagged_data[0]);
+  auto type = lossless_cast<CompressionType>(tagged_data[1]);
+
+  std::unique_ptr<char[]> uncompressed;
+  Slice saved(tagged_data.data() + kTagSize, tagged_data.size() - kTagSize);
  if (source == CacheTier::kVolatileCompressedTier) {
-    if (cache_options_.compression_type == kNoCompression ||
-        cache_options_.do_not_compress_roles.Contains(helper->role)) {
-      s = helper->create_cb(Slice(data_ptr, handle_value_charge),
-                            kNoCompression, CacheTier::kVolatileTier,
-                            create_context, allocator, &value, &charge);
-    } else {
-      // TODO: can we work some magic with create_cb, which might be based on
-      // custom compression, to decompress without an extra copy in create_cb?
+    if (type != kNoCompression) {
+      // TODO: can we do something to avoid yet another allocation?
      Decompressor::Args args;
-      args.compressed_data = Slice(data_ptr, handle_value_charge);
-      args.compression_type = cache_options_.compression_type;
-      s = decompressor_->ExtractUncompressedSize(args);
-      assert(s.ok());
+      args.compressed_data = saved;
+      args.compression_type = type;
+      Status s = decompressor_->ExtractUncompressedSize(args);
+      assert(s.ok());  // in-memory data
      if (s.ok()) {
-        auto uncompressed = std::make_unique<char[]>(args.uncompressed_size);
+        uncompressed = std::make_unique<char[]>(args.uncompressed_size);
        s = decompressor_->DecompressBlock(args, uncompressed.get());
-        assert(s.ok());
-        if (s.ok()) {
-          s = helper->create_cb(
-              Slice(uncompressed.get(), args.uncompressed_size), kNoCompression,
-              CacheTier::kVolatileTier, create_context, allocator, &value,
-              &charge);
-        }
+        assert(s.ok());  // in-memory data
      }
+      if (!s.ok()) {
+        cache_->Release(lru_handle, /*erase_if_last_ref=*/true);
+        return nullptr;
+      }
+      saved = Slice(uncompressed.get(), args.uncompressed_size);
+      type = kNoCompression;
+      // Free temporary compressed data as early as we can. This could matter
+      // for unusually large blocks because we also have
+      // * Another compressed copy above (from lru_cache).
+      // * The uncompressed copy in `uncompressed`.
+      // * Another uncompressed copy in `result_value` below.
+      // Let's try to max out at 3 copies instead of 4.
+      merged_value = std::string();
    }
-  } else {
-    // The item was not compressed by us. Let the helper create_cb
-    // uncompress it
-    s = helper->create_cb(Slice(data_ptr, handle_value_charge), type, source,
-                          create_context, allocator, &value, &charge);
+    // Reduced as if it came from primary cache
+    source = CacheTier::kVolatileTier;
  }

+  Cache::ObjectPtr result_value = nullptr;
+  size_t result_charge = 0;
+  Status s = helper->create_cb(saved, type, source, create_context,
+                               cache_options_.memory_allocator.get(),
+                               &result_value, &result_charge);
  if (!s.ok()) {
    cache_->Release(lru_handle, /*erase_if_last_ref=*/true);
    return nullptr;
@@ -148,7 +152,8 @@ std::unique_ptr<SecondaryCacheResultHandle> CompressedSecondaryCache::Lookup(
    kept_in_sec_cache = true;
    cache_->Release(lru_handle, /*erase_if_last_ref=*/false);
  }
-  handle.reset(new CompressedSecondaryCacheResultHandle(value, charge));
+  handle.reset(
+      new CompressedSecondaryCacheResultHandle(result_value, result_charge));
  RecordTick(stats, COMPRESSED_SECONDARY_CACHE_HITS);
  return handle;
 }
@@ -171,85 +176,111 @@ bool CompressedSecondaryCache::MaybeInsertDummy(const Slice& key) {

 Status CompressedSecondaryCache::InsertInternal(
    const Slice& key, Cache::ObjectPtr value,
-    const Cache::CacheItemHelper* helper, CompressionType type,
+    const Cache::CacheItemHelper* helper, CompressionType from_type,
    CacheTier source) {
-  if (source != CacheTier::kVolatileCompressedTier &&
-      cache_options_.enable_custom_split_merge) {
-    // We don't support custom split/merge for the tiered case
-    return Status::OK();
-  }
+  bool enable_split_merge = cache_options_.enable_custom_split_merge;
+  const Cache::CacheItemHelper* internal_helper = GetHelper(enable_split_merge);

-  auto internal_helper = GetHelper(cache_options_.enable_custom_split_merge);
-  char header[20];
-  char* payload = header;
-  payload = EncodeVarint32(payload, static_cast<uint32_t>(type));
-  payload = EncodeVarint32(payload, static_cast<uint32_t>(source));
-  size_t data_size = (*helper->size_cb)(value);
-  char* data_size_ptr = payload;
-  payload = EncodeVarint64(payload, data_size);
+  // TODO: variant of size_cb that also returns a pointer to the data if
+  // already available. Saves an allocation if we keep the compressed version.
+  const size_t data_size_original = (*helper->size_cb)(value);

-  size_t header_size = payload - header;
-  size_t total_size = data_size + header_size;
-  CacheAllocationPtr ptr =
-      AllocateBlock(total_size, cache_options_.memory_allocator.get());
-  char* data_ptr = ptr.get() + header_size;
+  // Allocate enough memory for header/tag + original data because (a) we might
+  // not be attempting compression at all, and (b) we might keep the original if
+  // compression is insufficient. But we don't need the length prefix with
+  // enable_split_merge. TODO: be smarter with CacheValueChunk to save an
+  // allocation in the enable_split_merge case.
+  size_t header_size = GetHeaderSize(data_size_original, enable_split_merge);
+  CacheAllocationPtr allocation = AllocateBlock(
+      header_size + data_size_original, cache_options_.memory_allocator.get());
+  char* data_ptr = allocation.get() + header_size;
+  Slice tagged_data(data_ptr - kTagSize, data_size_original + kTagSize);
+  assert(tagged_data.data() >= allocation.get());

-  Status s = (*helper->saveto_cb)(value, 0, data_size, data_ptr);
+  Status s = (*helper->saveto_cb)(value, 0, data_size_original, data_ptr);
  if (!s.ok()) {
    return s;
  }
-  Slice val(data_ptr, data_size);

-  std::string compressed_val;
-  if (cache_options_.compression_type != kNoCompression &&
-      type == kNoCompression &&
+  std::unique_ptr<char[]> tagged_compressed_data;
+  CompressionType to_type = kNoCompression;
+  if (compressor_ && from_type == kNoCompression &&
      !cache_options_.do_not_compress_roles.Contains(helper->role)) {
-    PERF_COUNTER_ADD(compressed_sec_cache_uncompressed_bytes, data_size);
+    assert(source == CacheTier::kVolatileCompressedTier);

-    CompressionType to_type = kNoCompression;
-    s = compressor_->CompressBlock(val, &compressed_val, &to_type,
+    // TODO: consider malloc sizes for max acceptable compressed size
+    // Or maybe max_compressed_bytes_per_kb
+    size_t data_size_compressed = data_size_original - 1;
+    tagged_compressed_data =
+        std::make_unique<char[]>(data_size_compressed + kTagSize);
+    s = compressor_->CompressBlock(Slice(data_ptr, data_size_original),
+                                   tagged_compressed_data.get() + kTagSize,
+                                   &data_size_compressed, &to_type,
                                   nullptr /*working_area*/);
    if (!s.ok()) {
      return s;
    }
-    // TODO: allow values not compressed when there's no size savings?
-    assert(to_type == cache_options_.compression_type);
-    if (to_type != cache_options_.compression_type) {
-      return Status::Corruption("Failed to compress value.");
-    }
-
-    val = Slice(compressed_val);
-    data_size = compressed_val.size();
-    payload = EncodeVarint64(data_size_ptr, data_size);
-    header_size = payload - header;
-    total_size = header_size + data_size;
-    PERF_COUNTER_ADD(compressed_sec_cache_compressed_bytes, data_size);
-
-    if (!cache_options_.enable_custom_split_merge) {
-      ptr = AllocateBlock(total_size, cache_options_.memory_allocator.get());
-      data_ptr = ptr.get() + header_size;
-      memcpy(data_ptr, compressed_val.data(), data_size);
+    PERF_COUNTER_ADD(compressed_sec_cache_uncompressed_bytes,
+                     data_size_original);
+    if (to_type == kNoCompression) {
+      // Compression rejected or otherwise aborted/failed
+      to_type = kNoCompression;
+      tagged_compressed_data.reset();
+      // TODO: consider separate counters for rejected compressions
+      PERF_COUNTER_ADD(compressed_sec_cache_compressed_bytes,
+                       data_size_original);
+    } else {
+      PERF_COUNTER_ADD(compressed_sec_cache_compressed_bytes,
+                       data_size_compressed);
+      if (enable_split_merge) {
+        // Only need tagged_data for copying into CacheValueChunks.
+        tagged_data = Slice(tagged_compressed_data.get(),
+                            data_size_compressed + kTagSize);
+        allocation.reset();
+      } else {
+        // Replace allocation with compressed version, copied from string
+        header_size = GetHeaderSize(data_size_compressed, enable_split_merge);
+        allocation = AllocateBlock(header_size + data_size_compressed,
+                                   cache_options_.memory_allocator.get());
+        data_ptr = allocation.get() + header_size;
+        // Ignore unpopulated tag on tagged_compressed_data; will only be
+        // populated on the new allocation.
+        std::memcpy(data_ptr, tagged_compressed_data.get() + kTagSize,
+                    data_size_compressed);
+        tagged_data =
+            Slice(data_ptr - kTagSize, data_size_compressed + kTagSize);
+        assert(tagged_data.data() >= allocation.get());
+      }
    }
  }

  PERF_COUNTER_ADD(compressed_sec_cache_insert_real_count, 1);
-  if (cache_options_.enable_custom_split_merge) {
+
+  // Save the tag fields
+  const_cast<char*>(tagged_data.data())[0] = lossless_cast<char>(source);
+  const_cast<char*>(tagged_data.data())[1] = lossless_cast<char>(
+      source == CacheTier::kVolatileCompressedTier ? to_type : from_type);
+
+  if (enable_split_merge) {
    size_t split_charge{0};
-    CacheValueChunk* value_chunks_head = SplitValueIntoChunks(
-        val, cache_options_.compression_type, split_charge);
-    return cache_->Insert(key, value_chunks_head, internal_helper,
-                          split_charge);
+    CacheValueChunk* value_chunks_head =
+        SplitValueIntoChunks(tagged_data, split_charge);
+    s = cache_->Insert(key, value_chunks_head, internal_helper, split_charge);
+    assert(s.ok());  // LRUCache::Insert() with handle==nullptr always OK
  } else {
+    // Save the size prefix
+    char* ptr = allocation.get();
+    ptr = EncodeVarint64(ptr, tagged_data.size());
+    assert(ptr == tagged_data.data());
 #ifdef ROCKSDB_MALLOC_USABLE_SIZE
-    size_t charge = malloc_usable_size(ptr.get());
+    size_t charge = malloc_usable_size(allocation.get());
 #else
-    size_t charge = total_size;
+    size_t charge = tagged_data.size();
 #endif
-    std::memcpy(ptr.get(), header, header_size);
-    CacheAllocationPtr* buf = new CacheAllocationPtr(std::move(ptr));
-    charge += sizeof(CacheAllocationPtr);
-    return cache_->Insert(key, buf, internal_helper, charge);
+    s = cache_->Insert(key, allocation.release(), internal_helper, charge);
+    assert(s.ok());  // LRUCache::Insert() with handle==nullptr always OK
  }
+  return Status::OK();
 }

 Status CompressedSecondaryCache::Insert(const Slice& key,
@@ -271,7 +302,17 @@ Status CompressedSecondaryCache::Insert(const Slice& key,
 Status CompressedSecondaryCache::InsertSaved(
    const Slice& key, const Slice& saved, CompressionType type = kNoCompression,
    CacheTier source = CacheTier::kVolatileTier) {
+  if (source == CacheTier::kVolatileCompressedTier) {
+    // Unexpected, would violate InsertInternal preconditions
+    assert(source != CacheTier::kVolatileCompressedTier);
+    return Status::OK();
+  }
  if (type == kNoCompression) {
+    // Not currently supported (why?)
+    return Status::OK();
+  }
+  if (cache_options_.enable_custom_split_merge) {
+    // We don't support custom split/merge for the tiered case (why?)
    return Status::OK();
  }

@@ -291,7 +332,7 @@ Status CompressedSecondaryCache::SetCapacity(size_t capacity) {
  MutexLock l(&capacity_mutex_);
  cache_options_.capacity = capacity;
  cache_->SetCapacity(capacity);
-  disable_cache_ = capacity == 0;
+  disable_cache_.StoreRelaxed(capacity == 0);
  return Status::OK();
 }

@@ -321,9 +362,14 @@ std::string CompressedSecondaryCache::GetPrintableOptions() const {
  return ret;
 }

+// FIXME: this could use a lot of attention, including:
+// * Use allocator
+// * We shouldn't be worse than non-split; be more pro-actively aware of
+// internal fragmentation
+// * Consider a unified object/chunk structure that may or may not split
+// * Optimize size overhead of chunks
 CompressedSecondaryCache::CacheValueChunk*
 CompressedSecondaryCache::SplitValueIntoChunks(const Slice& value,
-                                               CompressionType compression_type,
                                               size_t& charge) {
  assert(!value.empty());
  const char* src_ptr = value.data();
@@ -344,15 +390,14 @@ CompressedSecondaryCache::SplitValueIntoChunks(const Slice& value,
    // size, or there is no compression.
    if (upper == malloc_bin_sizes_.begin() ||
        upper == malloc_bin_sizes_.end() ||
-        *upper - predicted_chunk_size < malloc_bin_sizes_.front() ||
-        compression_type == kNoCompression) {
+        *upper - predicted_chunk_size < malloc_bin_sizes_.front()) {
      tmp_size = predicted_chunk_size;
    } else {
      tmp_size = *(--upper);
    }

    CacheValueChunk* new_chunk =
-        reinterpret_cast<CacheValueChunk*>(new char[tmp_size]);
+        static_cast<CacheValueChunk*>(static_cast<void*>(new char[tmp_size]));
    current_chunk->next = new_chunk;
    current_chunk = current_chunk->next;
    actual_chunk_size = tmp_size - sizeof(CacheValueChunk) + 1;
@@ -367,28 +412,24 @@ CompressedSecondaryCache::SplitValueIntoChunks(const Slice& value,
  return dummy_head.next;
 }

-CacheAllocationPtr CompressedSecondaryCache::MergeChunksIntoValue(
-    const void* chunks_head, size_t& charge) {
-  const CacheValueChunk* head =
-      reinterpret_cast<const CacheValueChunk*>(chunks_head);
+std::string CompressedSecondaryCache::MergeChunksIntoValue(
+    const CacheValueChunk* head) {
  const CacheValueChunk* current_chunk = head;
-  charge = 0;
+  size_t total_size = 0;
  while (current_chunk != nullptr) {
-    charge += current_chunk->size;
+    total_size += current_chunk->size;
    current_chunk = current_chunk->next;
  }

-  CacheAllocationPtr ptr =
-      AllocateBlock(charge, cache_options_.memory_allocator.get());
+  std::string result;
+  result.reserve(total_size);
  current_chunk = head;
-  size_t pos{0};
  while (current_chunk != nullptr) {
-    memcpy(ptr.get() + pos, current_chunk->data, current_chunk->size);
-    pos += current_chunk->size;
+    result.append(current_chunk->data, current_chunk->size);
    current_chunk = current_chunk->next;
  }
-
-  return ptr;
+  assert(result.size() == total_size);
+  return result;
 }

 const Cache::CacheItemHelper* CompressedSecondaryCache::GetHelper(
@@ -402,16 +443,16 @@ const Cache::CacheItemHelper* CompressedSecondaryCache::GetHelper(
            CacheValueChunk* tmp_chunk = chunks_head;
            chunks_head = chunks_head->next;
            tmp_chunk->Free();
-            obj = nullptr;
          }
        }};
    return &kHelper;
  } else {
    static const Cache::CacheItemHelper kHelper{
        CacheEntryRole::kMisc,
-        [](Cache::ObjectPtr obj, MemoryAllocator* /*alloc*/) {
-          delete static_cast<CacheAllocationPtr*>(obj);
-          obj = nullptr;
+        [](Cache::ObjectPtr obj, MemoryAllocator* alloc) {
+          if (obj != nullptr) {
+            CacheAllocationDeleter{alloc}(static_cast<char*>(obj));
+          }
        }};
    return &kHelper;
  }
@@ -422,12 +463,7 @@ size_t CompressedSecondaryCache::TEST_GetCharge(const Slice& key) {
  if (lru_handle == nullptr) {
    return 0;
  }
-
  size_t charge = cache_->GetCharge(lru_handle);
-  if (cache_->Value(lru_handle) != nullptr &&
-      !cache_options_.enable_custom_split_merge) {
-    charge -= 10;
-  }
  cache_->Release(lru_handle, /*erase_if_last_ref=*/false);
  return charge;
 }
@@ -10,13 +10,12 @@
 #include <memory>

 #include "cache/cache_reservation_manager.h"
-#include "cache/lru_cache.h"
 #include "memory/memory_allocator_impl.h"
+#include "rocksdb/advanced_compression.h"
 #include "rocksdb/secondary_cache.h"
 #include "rocksdb/slice.h"
 #include "rocksdb/status.h"
-#include "util/compression.h"
-#include "util/mutexlock.h"
+#include "util/atomic.h"

 namespace ROCKSDB_NAMESPACE {

@@ -124,14 +123,9 @@ class CompressedSecondaryCache : public SecondaryCache {
  // Split value into chunks to better fit into jemalloc bins. The chunks
  // are stored in CacheValueChunk and extra charge is needed for each chunk,
  // so the cache charge is recalculated here.
-  CacheValueChunk* SplitValueIntoChunks(const Slice& value,
-                                        CompressionType compression_type,
-                                        size_t& charge);
+  CacheValueChunk* SplitValueIntoChunks(const Slice& value, size_t& charge);

-  // After merging chunks, the extra charge for each chunk is removed, so
-  // the charge is recalculated.
-  CacheAllocationPtr MergeChunksIntoValue(const void* chunks_head,
-                                          size_t& charge);
+  std::string MergeChunksIntoValue(const CacheValueChunk* head);

  bool MaybeInsertDummy(const Slice& key);

@@ -149,7 +143,7 @@ class CompressedSecondaryCache : public SecondaryCache {
  std::shared_ptr<Decompressor> decompressor_;
  mutable port::Mutex capacity_mutex_;
  std::shared_ptr<ConcurrentCacheReservationManager> cache_res_mgr_;
-  bool disable_cache_;
+  RelaxedAtomic<bool> disable_cache_;
 };

 }  // namespace ROCKSDB_NAMESPACE
@@ -24,6 +24,14 @@ namespace ROCKSDB_NAMESPACE {
 using secondary_cache_test_util::GetTestingCacheTypes;
 using secondary_cache_test_util::WithCacheType;

+// Read and reset a statistic
+template <typename T>
+T Pop(T& var) {
+  T ret = var;
+  var = T();
+  return ret;
+}
+
 // 16 bytes for HCC compatibility
 const std::string key0 = "____    ____key0";
 const std::string key1 = "____    ____key1";
@@ -51,7 +59,7 @@ class CompressedSecondaryCacheTestBase : public testing::Test,

    Random rnd(301);
    // Insert and Lookup the item k1 for the first time.
-    std::string str1(rnd.RandomString(1000));
+    std::string str1 = test::CompressibleString(&rnd, 0.5, 1000);
    TestItem item1(str1.data(), str1.length());
    // A dummy handle is inserted if the item is inserted for the first time.
    ASSERT_OK(sec_cache->Insert(key1, &item1, GetHelper(), false));
@@ -68,7 +76,14 @@ class CompressedSecondaryCacheTestBase : public testing::Test,
    ASSERT_OK(sec_cache->Insert(key1, &item1, GetHelper(), false));
    ASSERT_EQ(get_perf_context()->compressed_sec_cache_insert_real_count, 1);

-    ASSERT_GT(comp_sec_cache->TEST_GetCharge(key1), 1000);
+    if (sec_cache_is_compressed) {
+      ASSERT_GT(comp_sec_cache->TEST_GetCharge(key1), str1.length() / 4);
+      ASSERT_LT(comp_sec_cache->TEST_GetCharge(key1), str1.length() * 3 / 4);
+    } else {
+      ASSERT_GE(comp_sec_cache->TEST_GetCharge(key1), str1.length());
+      // NOTE: split-merge is worse (1048 vs. 1024)
+      ASSERT_LE(comp_sec_cache->TEST_GetCharge(key1), 1048U);
+    }

    std::unique_ptr<SecondaryCacheResultHandle> handle1_2 =
        sec_cache->Lookup(key1, GetHelper(), this, true, /*advise_erase=*/true,
@@ -76,10 +91,13 @@ class CompressedSecondaryCacheTestBase : public testing::Test,
    ASSERT_NE(handle1_2, nullptr);
    ASSERT_FALSE(kept_in_sec_cache);
    if (sec_cache_is_compressed) {
-      ASSERT_EQ(get_perf_context()->compressed_sec_cache_uncompressed_bytes,
-                1000);
-      ASSERT_EQ(get_perf_context()->compressed_sec_cache_compressed_bytes,
-                1007);
+      ASSERT_EQ(
+          Pop(get_perf_context()->compressed_sec_cache_uncompressed_bytes),
+          str1.length());
+      ASSERT_LT(get_perf_context()->compressed_sec_cache_compressed_bytes,
+                str1.length() * 3 / 4);
+      ASSERT_GT(Pop(get_perf_context()->compressed_sec_cache_compressed_bytes),
+                str1.length() / 4);
    } else {
      ASSERT_EQ(get_perf_context()->compressed_sec_cache_uncompressed_bytes, 0);
      ASSERT_EQ(get_perf_context()->compressed_sec_cache_compressed_bytes, 0);
@@ -97,7 +115,7 @@ class CompressedSecondaryCacheTestBase : public testing::Test,
    ASSERT_EQ(handle1_3, nullptr);

    // Insert and Lookup the item k2.
-    std::string str2(rnd.RandomString(1000));
+    std::string str2 = test::CompressibleString(&rnd, 0.5, 1017);
    TestItem item2(str2.data(), str2.length());
    ASSERT_OK(sec_cache->Insert(key2, &item2, GetHelper(), false));
    ASSERT_EQ(get_perf_context()->compressed_sec_cache_insert_dummy_count, 2);
@@ -109,10 +127,13 @@ class CompressedSecondaryCacheTestBase : public testing::Test,
    ASSERT_OK(sec_cache->Insert(key2, &item2, GetHelper(), false));
    ASSERT_EQ(get_perf_context()->compressed_sec_cache_insert_real_count, 2);
    if (sec_cache_is_compressed) {
-      ASSERT_EQ(get_perf_context()->compressed_sec_cache_uncompressed_bytes,
-                2000);
-      ASSERT_EQ(get_perf_context()->compressed_sec_cache_compressed_bytes,
-                2014);
+      ASSERT_EQ(
+          Pop(get_perf_context()->compressed_sec_cache_uncompressed_bytes),
+          str2.length());
+      ASSERT_LT(get_perf_context()->compressed_sec_cache_compressed_bytes,
+                str2.length() * 3 / 4);
+      ASSERT_GT(Pop(get_perf_context()->compressed_sec_cache_compressed_bytes),
+                str2.length() / 4);
    } else {
      ASSERT_EQ(get_perf_context()->compressed_sec_cache_uncompressed_bytes, 0);
      ASSERT_EQ(get_perf_context()->compressed_sec_cache_compressed_bytes, 0);
@@ -126,9 +147,48 @@ class CompressedSecondaryCacheTestBase : public testing::Test,
    ASSERT_NE(val2, nullptr);
    ASSERT_EQ(memcmp(val2->Buf(), item2.Buf(), item2.Size()), 0);

+    // Release handles
    std::vector<SecondaryCacheResultHandle*> handles = {handle1_2.get(),
                                                        handle2_2.get()};
    sec_cache->WaitAll(handles);
+    handle1_2.reset();
+    handle2_2.reset();
+
+    // Insert and Lookup a non-compressible item k3.
+    std::string str3 = rnd.RandomBinaryString(480);
+    TestItem item3(str3.data(), str3.length());
+    ASSERT_OK(sec_cache->Insert(key3, &item3, GetHelper(), false));
+    ASSERT_EQ(get_perf_context()->compressed_sec_cache_insert_dummy_count, 3);
+    std::unique_ptr<SecondaryCacheResultHandle> handle3_1 =
+        sec_cache->Lookup(key3, GetHelper(), this, true, /*advise_erase=*/false,
+                          /*stats=*/nullptr, kept_in_sec_cache);
+    ASSERT_EQ(handle3_1, nullptr);
+
+    ASSERT_OK(sec_cache->Insert(key3, &item3, GetHelper(), false));
+    ASSERT_EQ(get_perf_context()->compressed_sec_cache_insert_real_count, 3);
+    if (sec_cache_is_compressed) {
+      // TODO: consider a compression rejected stat?
+      ASSERT_EQ(
+          Pop(get_perf_context()->compressed_sec_cache_uncompressed_bytes),
+          str3.length());
+      ASSERT_EQ(Pop(get_perf_context()->compressed_sec_cache_compressed_bytes),
+                str3.length());
+    } else {
+      ASSERT_EQ(get_perf_context()->compressed_sec_cache_uncompressed_bytes, 0);
+      ASSERT_EQ(get_perf_context()->compressed_sec_cache_compressed_bytes, 0);
+    }
+
+    std::unique_ptr<SecondaryCacheResultHandle> handle3_2 =
+        sec_cache->Lookup(key3, GetHelper(), this, true, /*advise_erase=*/false,
+                          /*stats=*/nullptr, kept_in_sec_cache);
+    ASSERT_NE(handle3_2, nullptr);
+    std::unique_ptr<TestItem> val3 =
+        std::unique_ptr<TestItem>(static_cast<TestItem*>(handle3_2->Value()));
+    ASSERT_NE(val3, nullptr);
+    ASSERT_EQ(memcmp(val3->Buf(), item3.Buf(), item3.Size()), 0);
+
+    EXPECT_GE(comp_sec_cache->TEST_GetCharge(key3), str3.length());
+    EXPECT_LE(comp_sec_cache->TEST_GetCharge(key3), 512);

    sec_cache.reset();
  }
@@ -178,8 +238,9 @@ class CompressedSecondaryCacheTestBase : public testing::Test,
      secondary_cache_opts.compression_type = CompressionType::kNoCompression;
    }

-    secondary_cache_opts.capacity = 1100;
+    secondary_cache_opts.capacity = 1400;
    secondary_cache_opts.num_shard_bits = 0;
+    secondary_cache_opts.strict_capacity_limit = true;
    std::shared_ptr<SecondaryCache> sec_cache =
        NewCompressedSecondaryCache(secondary_cache_opts);

@@ -193,7 +254,7 @@ class CompressedSecondaryCacheTestBase : public testing::Test,
    ASSERT_OK(sec_cache->Insert(key1, &item1, GetHelper(), false));

    // Insert and Lookup the second item.
-    std::string str2(rnd.RandomString(200));
+    std::string str2(rnd.RandomString(500));
    TestItem item2(str2.data(), str2.length());
    // Insert a dummy handle, k1 is not evicted.
    ASSERT_OK(sec_cache->Insert(key2, &item2, GetHelper(), false));
@@ -201,16 +262,23 @@ class CompressedSecondaryCacheTestBase : public testing::Test,
    std::unique_ptr<SecondaryCacheResultHandle> handle1 =
        sec_cache->Lookup(key1, GetHelper(), this, true, /*advise_erase=*/false,
                          /*stats=*/nullptr, kept_in_sec_cache);
-    ASSERT_EQ(handle1, nullptr);
+    ASSERT_NE(handle1, nullptr);
+    std::unique_ptr<TestItem> val1{static_cast<TestItem*>(handle1->Value())};
+    ASSERT_NE(val1, nullptr);
+    ASSERT_EQ(val1->ToString(), str1);
+    handle1.reset();

    // Insert k2 and k1 is evicted.
    ASSERT_OK(sec_cache->Insert(key2, &item2, GetHelper(), false));
+    handle1 =
+        sec_cache->Lookup(key1, GetHelper(), this, true, /*advise_erase=*/false,
+                          /*stats=*/nullptr, kept_in_sec_cache);
+    ASSERT_EQ(handle1, nullptr);
    std::unique_ptr<SecondaryCacheResultHandle> handle2 =
        sec_cache->Lookup(key2, GetHelper(), this, true, /*advise_erase=*/false,
                          /*stats=*/nullptr, kept_in_sec_cache);
    ASSERT_NE(handle2, nullptr);
-    std::unique_ptr<TestItem> val2 =
-        std::unique_ptr<TestItem>(static_cast<TestItem*>(handle2->Value()));
+    std::unique_ptr<TestItem> val2{static_cast<TestItem*>(handle2->Value())};
    ASSERT_NE(val2, nullptr);
    ASSERT_EQ(memcmp(val2->Buf(), item2.Buf(), item2.Size()), 0);

@@ -232,7 +300,7 @@ class CompressedSecondaryCacheTestBase : public testing::Test,
    // Save Fails.
    std::string str3 = rnd.RandomString(10);
    TestItem item3(str3.data(), str3.length());
-    // The Status is OK because a dummy handle is inserted.
+    // The first Status is OK because a dummy handle is inserted.
    ASSERT_OK(sec_cache->Insert(key3, &item3, GetHelperFail(), false));
    ASSERT_NOK(sec_cache->Insert(key3, &item3, GetHelperFail(), false));

@@ -265,11 +333,11 @@ class CompressedSecondaryCacheTestBase : public testing::Test,

    get_perf_context()->Reset();
    Random rnd(301);
-    std::string str1 = rnd.RandomString(1001);
+    std::string str1 = test::CompressibleString(&rnd, 0.5, 1001);
    auto item1_1 = new TestItem(str1.data(), str1.length());
    ASSERT_OK(cache->Insert(key1, item1_1, GetHelper(), str1.length()));

-    std::string str2 = rnd.RandomString(1012);
+    std::string str2 = test::CompressibleString(&rnd, 0.5, 1012);
    auto item2_1 = new TestItem(str2.data(), str2.length());
    // After this Insert, primary cache contains k2 and secondary cache contains
    // k1's dummy item.
@@ -278,7 +346,7 @@ class CompressedSecondaryCacheTestBase : public testing::Test,
    ASSERT_EQ(get_perf_context()->compressed_sec_cache_uncompressed_bytes, 0);
    ASSERT_EQ(get_perf_context()->compressed_sec_cache_compressed_bytes, 0);

-    std::string str3 = rnd.RandomString(1024);
+    std::string str3 = test::CompressibleString(&rnd, 0.5, 1024);
    auto item3_1 = new TestItem(str3.data(), str3.length());
    // After this Insert, primary cache contains k3 and secondary cache contains
    // k1's dummy item and k2's dummy item.
@@ -297,10 +365,13 @@ class CompressedSecondaryCacheTestBase : public testing::Test,
    ASSERT_OK(cache->Insert(key2, item2_2, GetHelper(), str2.length()));
    ASSERT_EQ(get_perf_context()->compressed_sec_cache_insert_real_count, 1);
    if (sec_cache_is_compressed) {
-      ASSERT_EQ(get_perf_context()->compressed_sec_cache_uncompressed_bytes,
+      ASSERT_EQ(
+          Pop(get_perf_context()->compressed_sec_cache_uncompressed_bytes),
+          str1.length());
+      ASSERT_LT(get_perf_context()->compressed_sec_cache_compressed_bytes,
                str1.length());
-      ASSERT_EQ(get_perf_context()->compressed_sec_cache_compressed_bytes,
-                1008);
+      ASSERT_GT(Pop(get_perf_context()->compressed_sec_cache_compressed_bytes),
+                str1.length() / 10);
    } else {
      ASSERT_EQ(get_perf_context()->compressed_sec_cache_uncompressed_bytes, 0);
      ASSERT_EQ(get_perf_context()->compressed_sec_cache_compressed_bytes, 0);
@@ -312,10 +383,13 @@ class CompressedSecondaryCacheTestBase : public testing::Test,
    ASSERT_OK(cache->Insert(key3, item3_2, GetHelper(), str3.length()));
    ASSERT_EQ(get_perf_context()->compressed_sec_cache_insert_real_count, 2);
    if (sec_cache_is_compressed) {
-      ASSERT_EQ(get_perf_context()->compressed_sec_cache_uncompressed_bytes,
-                str1.length() + str2.length());
-      ASSERT_EQ(get_perf_context()->compressed_sec_cache_compressed_bytes,
-                2027);
+      ASSERT_EQ(
+          Pop(get_perf_context()->compressed_sec_cache_uncompressed_bytes),
+          str2.length());
+      ASSERT_LT(get_perf_context()->compressed_sec_cache_compressed_bytes,
+                str2.length());
+      ASSERT_GT(Pop(get_perf_context()->compressed_sec_cache_compressed_bytes),
+                str2.length() / 10);
    } else {
      ASSERT_EQ(get_perf_context()->compressed_sec_cache_uncompressed_bytes, 0);
      ASSERT_EQ(get_perf_context()->compressed_sec_cache_compressed_bytes, 0);
@@ -641,8 +715,7 @@ class CompressedSecondaryCacheTestBase : public testing::Test,
    size_t str_size{8500};
    std::string str = rnd.RandomString(static_cast<int>(str_size));
    size_t charge{0};
-    CacheValueChunk* chunks_head =
-        sec_cache->SplitValueIntoChunks(str, kLZ4Compression, charge);
+    CacheValueChunk* chunks_head = sec_cache->SplitValueIntoChunks(str, charge);
    ASSERT_EQ(charge, str_size + 3 * (sizeof(CacheValueChunk) - 1));

    CacheValueChunk* current_chunk = chunks_head;
@@ -688,12 +761,9 @@ class CompressedSecondaryCacheTestBase : public testing::Test,
    std::unique_ptr<CompressedSecondaryCache> sec_cache =
        std::make_unique<CompressedSecondaryCache>(
            CompressedSecondaryCacheOptions(1000, 0, true, 0.5, 0.0));
-    size_t charge{0};
-    CacheAllocationPtr value =
-        sec_cache->MergeChunksIntoValue(chunks_head, charge);
-    ASSERT_EQ(charge, size1 + size2 + size3);
-    std::string value_str{value.get(), charge};
-    ASSERT_EQ(strcmp(value_str.data(), str.data()), 0);
+    std::string value_str = sec_cache->MergeChunksIntoValue(chunks_head);
+    ASSERT_EQ(value_str.size(), size1 + size2 + size3);
+    ASSERT_EQ(value_str, str);

    while (chunks_head != nullptr) {
      CacheValueChunk* tmp_chunk = chunks_head;
@@ -725,15 +795,12 @@ class CompressedSecondaryCacheTestBase : public testing::Test,
    size_t str_size{8500};
    std::string str = rnd.RandomString(static_cast<int>(str_size));
    size_t charge{0};
-    CacheValueChunk* chunks_head =
-        sec_cache->SplitValueIntoChunks(str, kLZ4Compression, charge);
+    CacheValueChunk* chunks_head = sec_cache->SplitValueIntoChunks(str, charge);
    ASSERT_EQ(charge, str_size + 3 * (sizeof(CacheValueChunk) - 1));

-    CacheAllocationPtr value =
-        sec_cache->MergeChunksIntoValue(chunks_head, charge);
-    ASSERT_EQ(charge, str_size);
-    std::string value_str{value.get(), charge};
-    ASSERT_EQ(strcmp(value_str.data(), str.data()), 0);
+    std::string value_str = sec_cache->MergeChunksIntoValue(chunks_head);
+    ASSERT_EQ(value_str.size(), str_size);
+    ASSERT_EQ(value_str, str);

    sec_cache->GetHelper(true)->del_cb(chunks_head, /*alloc*/ nullptr);
  }
@@ -896,8 +963,8 @@ TEST_P(CompressedSecondaryCacheTestWithCompressionParam, EntryRoles) {

  std::shared_ptr<SecondaryCache> sec_cache = NewCompressedSecondaryCache(opts);

-  // Fixed seed to ensure consistent compressibility (doesn't compress)
-  std::string junk(Random(301).RandomString(1000));
+  Random rnd(301);
+  std::string junk = test::CompressibleString(&rnd, 0.5, 1000);

  for (uint32_t i = 0; i < kNumCacheEntryRoles; ++i) {
    CacheEntryRole role = static_cast<CacheEntryRole>(i);
@@ -930,9 +997,11 @@ TEST_P(CompressedSecondaryCacheTestWithCompressionParam, EntryRoles) {
        sec_cache_is_compressed_ && !do_not_compress.Contains(role);
    if (compressed) {
      ASSERT_EQ(get_perf_context()->compressed_sec_cache_uncompressed_bytes,
-                1000);
-      ASSERT_EQ(get_perf_context()->compressed_sec_cache_compressed_bytes,
-                1007);
+                junk.length());
+      ASSERT_LT(get_perf_context()->compressed_sec_cache_compressed_bytes,
+                junk.length() * 3 / 4);
+      ASSERT_GT(get_perf_context()->compressed_sec_cache_compressed_bytes,
+                junk.length() / 4);
    } else {
      ASSERT_EQ(get_perf_context()->compressed_sec_cache_uncompressed_bytes, 0);
      ASSERT_EQ(get_perf_context()->compressed_sec_cache_compressed_bytes, 0);
@@ -121,16 +121,13 @@ CacheWithSecondaryAdapter::~CacheWithSecondaryAdapter() {
    assert(s.ok());
    assert(placeholder_usage_ == 0);
    assert(reserved_usage_ == 0);
-    bool pri_cache_res_mismatch =
-        pri_cache_res_->GetTotalMemoryUsed() != sec_capacity;
-    if (pri_cache_res_mismatch) {
-      fprintf(stderr,
+    if (pri_cache_res_->GetTotalMemoryUsed() != sec_capacity) {
+      fprintf(stdout,
              "~CacheWithSecondaryAdapter: Primary cache reservation: "
              "%zu, Secondary cache capacity: %zu, "
              "Secondary cache reserved: %zu\n",
              pri_cache_res_->GetTotalMemoryUsed(), sec_capacity,
              sec_reserved_);
-      assert(pri_cache_res_mismatch);
    }
  }
 #endif  // NDEBUG
@@ -489,12 +486,10 @@ const char* CacheWithSecondaryAdapter::Name() const {
 // as well. At the moment, we don't have a good way of handling the case
 // where the new capacity < total cache reservations.
 void CacheWithSecondaryAdapter::SetCapacity(size_t capacity) {
-  size_t sec_capacity = static_cast<size_t>(
-      capacity * (distribute_cache_res_ ? sec_cache_res_ratio_ : 0.0));
-  size_t old_sec_capacity = 0;
-
  if (distribute_cache_res_) {
    MutexLock m(&cache_res_mutex_);
+    size_t sec_capacity = static_cast<size_t>(capacity * sec_cache_res_ratio_);
+    size_t old_sec_capacity = 0;

    Status s = secondary_cache_->GetCapacity(old_sec_capacity);
    if (!s.ok()) {
@@ -613,6 +608,7 @@ Status CacheWithSecondaryAdapter::UpdateCacheReservationRatio(
    //    cache utilization (increase in capacity - increase in share of cache
    //    reservation)
    // 3. Increase secondary cache capacity
+    assert(new_sec_reserved >= sec_reserved_);
    s = secondary_cache_->Deflate(new_sec_reserved - sec_reserved_);
    assert(s.ok());
    s = pri_cache_res_->UpdateCacheReservation(
@@ -98,7 +98,7 @@ class ArenaWrappedDBIter : public Iterator {

  bool PrepareValue() override { return db_iter_->PrepareValue(); }

-  void Prepare(const std::vector<ScanOptions>& scan_opts) override {
+  void Prepare(const MultiScanArgs& scan_opts) override {
    db_iter_->Prepare(scan_opts);
  }

@@ -201,8 +201,7 @@ Status BuildTable(
    CompactionIterator c_iter(
        iter, ucmp, &merge, kMaxSequenceNumber, &snapshots, earliest_snapshot,
        earliest_write_conflict_snapshot, job_snapshot, snapshot_checker, env,
-        ShouldReportDetailedTime(env, ioptions.stats),
-        true /* internal key corruption is not ok */, range_del_agg.get(),
+        ShouldReportDetailedTime(env, ioptions.stats), range_del_agg.get(),
        blob_file_builder.get(), ioptions.allow_data_in_errors,
        ioptions.enforce_single_del_contracts,
        /*manual_compaction_canceled=*/kManualCompactionCanceledFalse,
@@ -925,6 +925,10 @@ void rocksdb_backup_engine_options_destroy(
  delete options;
 }

+void rocksdb_status_ptr_get_error(rocksdb_status_ptr_t* status, char** errptr) {
+  SaveError(errptr, *(status->rep));
+}
+
 rocksdb_checkpoint_t* rocksdb_checkpoint_object_create(rocksdb_t* db,
                                                       char** errptr) {
  Checkpoint* checkpoint;
@@ -3061,6 +3065,10 @@ uint64_t rocksdb_flushjobinfo_smallest_seqno(
  return info->rep.smallest_seqno;
 }

+uint32_t rocksdb_flushjobinfo_flush_reason(const rocksdb_flushjobinfo_t* info) {
+  return static_cast<uint32_t>(info->rep.flush_reason);
+}
+
 void rocksdb_reset_status(rocksdb_status_ptr_t* status_ptr) {
  auto ptr = status_ptr->rep;
  *ptr = Status::OK();
@@ -3192,6 +3200,11 @@ int rocksdb_subcompactionjobinfo_output_level(
  return info->rep.output_level;
 }

+uint32_t rocksdb_subcompactionjobinfo_compaction_reason(
+    const rocksdb_subcompactionjobinfo_t* info) {
+  return static_cast<uint32_t>(info->rep.compaction_reason);
+}
+
 /* ExternalFileIngestionInfo */

 const char* rocksdb_externalfileingestioninfo_cf_name(
@@ -3626,6 +3639,11 @@ void rocksdb_options_set_write_buffer_manager(
  opt->rep.write_buffer_manager = wbm->rep;
 }

+void rocksdb_options_set_sst_file_manager(rocksdb_options_t* opt,
+                                          rocksdb_sst_file_manager_t* sfm) {
+  opt->rep.sst_file_manager = sfm->rep;
+}
+
 size_t rocksdb_options_get_write_buffer_size(rocksdb_options_t* opt) {
  return opt->rep.write_buffer_size;
 }
@@ -4733,6 +4751,15 @@ void rocksdb_options_add_compact_on_deletion_collector_factory_del_ratio(
  opt->rep.table_properties_collector_factories.emplace_back(compact_on_del);
 }

+void rocksdb_options_add_compact_on_deletion_collector_factory_min_file_size(
+    rocksdb_options_t* opt, size_t window_size, size_t num_dels_trigger,
+    double deletion_ratio, uint64_t min_file_size) {
+  std::shared_ptr<ROCKSDB_NAMESPACE::TablePropertiesCollectorFactory>
+      compact_on_del = NewCompactOnDeletionCollectorFactory(
+          window_size, num_dels_trigger, deletion_ratio, min_file_size);
+  opt->rep.table_properties_collector_factories.emplace_back(compact_on_del);
+}
+
 void rocksdb_set_perf_level(int v) {
  PerfLevel level = static_cast<PerfLevel>(v);
  SetPerfLevel(level);
@@ -2230,10 +2230,6 @@ int main(int argc, char** argv) {
    rocksdb_options_set_skip_stats_update_on_db_open(o, 1);
    CheckCondition(1 == rocksdb_options_get_skip_stats_update_on_db_open(o));

-    rocksdb_options_set_skip_checking_sst_file_sizes_on_db_open(o, 1);
-    CheckCondition(
-        1 == rocksdb_options_get_skip_checking_sst_file_sizes_on_db_open(o));
-
    rocksdb_options_set_max_write_buffer_number(o, 97);
    CheckCondition(97 == rocksdb_options_get_max_write_buffer_number(o));

@@ -2493,8 +2489,6 @@ int main(int argc, char** argv) {
    CheckCondition(2.0 ==
                   rocksdb_options_get_max_bytes_for_level_multiplier(copy));
    CheckCondition(1 == rocksdb_options_get_skip_stats_update_on_db_open(copy));
-    CheckCondition(
-        1 == rocksdb_options_get_skip_checking_sst_file_sizes_on_db_open(copy));
    CheckCondition(97 == rocksdb_options_get_max_write_buffer_number(copy));
    CheckCondition(23 ==
                   rocksdb_options_get_min_write_buffer_number_to_merge(copy));
@@ -2681,12 +2675,6 @@ int main(int argc, char** argv) {
    CheckCondition(0 == rocksdb_options_get_skip_stats_update_on_db_open(copy));
    CheckCondition(1 == rocksdb_options_get_skip_stats_update_on_db_open(o));

-    rocksdb_options_set_skip_checking_sst_file_sizes_on_db_open(copy, 0);
-    CheckCondition(
-        0 == rocksdb_options_get_skip_checking_sst_file_sizes_on_db_open(copy));
-    CheckCondition(
-        1 == rocksdb_options_get_skip_checking_sst_file_sizes_on_db_open(o));
-
    rocksdb_options_set_max_write_buffer_number(copy, 2000);
    CheckCondition(2000 == rocksdb_options_get_max_write_buffer_number(copy));
    CheckCondition(97 == rocksdb_options_get_max_write_buffer_number(o));
@@ -280,7 +280,8 @@ ColumnFamilyOptions SanitizeCfOptions(const ImmutableDBOptions& db_options,
  }

  if (result.compaction_style == kCompactionStyleUniversal &&
-      db_options.allow_ingest_behind && result.num_levels < 3) {
+      (db_options.allow_ingest_behind || result.cf_allow_ingest_behind) &&
+      result.num_levels < 3) {
    result.num_levels = 3;
  }

@@ -1331,7 +1332,7 @@ Compaction* ColumnFamilyData::CompactRange(
    const InternalKey* begin, const InternalKey* end,
    InternalKey** compaction_end, bool* conflict,
    uint64_t max_file_num_to_ignore, const std::string& trim_ts) {
-  auto* result = compaction_picker_->CompactRange(
+  auto* result = compaction_picker_->PickCompactionForCompactRange(
      GetName(), mutable_cf_options, mutable_db_options,
      current_->storage_info(), input_level, output_level,
      compact_range_options, begin, end, compaction_end, conflict,
@@ -1633,6 +1634,8 @@ Status ColumnFamilyData::SetOptions(
  Status s = GetColumnFamilyOptionsFromMap(config_opts, cf_opts, options_map,
                                           &cf_opts);
  if (s.ok()) {
+    // FIXME: we should call SanitizeOptions() too or consolidate it with
+    // ValidateOptions().
    s = ValidateOptions(db_opts, cf_opts);
  }
  if (s.ok()) {
@@ -600,6 +600,11 @@ class ColumnFamilyData {
    return (mem_->IsEmpty() ? 0 : 1) + imm_.NumNotFlushed();
  }

+  // thread-safe, DB mutex not needed.
+  bool AllowIngestBehind() const {
+    return ioptions_.cf_allow_ingest_behind || ioptions_.allow_ingest_behind;
+  }
+
 private:
  friend class ColumnFamilySet;
  ColumnFamilyData(
@@ -441,6 +441,10 @@ TEST_F(CompactFilesTest, SentinelCompressionType) {
 }

 TEST_F(CompactFilesTest, CompressionWithBlockAlign) {
+  if (!Snappy_Supported()) {
+    ROCKSDB_GTEST_SKIP("Test requires Snappy support");
+    return;
+  }
  Options options;
  options.compression = CompressionType::kNoCompression;
  options.create_if_missing = true;
@@ -284,9 +284,9 @@ Compaction::Compaction(
    CompressionOptions _compression_opts, Temperature _output_temperature,
    uint32_t _max_subcompactions, std::vector<FileMetaData*> _grandparents,
    std::optional<SequenceNumber> _earliest_snapshot,
-    const SnapshotChecker* _snapshot_checker, bool _manual_compaction,
-    const std::string& _trim_ts, double _score, bool _deletion_compaction,
-    bool l0_files_might_overlap, CompactionReason _compaction_reason,
+    const SnapshotChecker* _snapshot_checker,
+    CompactionReason _compaction_reason, const std::string& _trim_ts,
+    double _score, bool l0_files_might_overlap,
    BlobGarbageCollectionPolicy _blob_garbage_collection_policy,
    double _blob_garbage_collection_age_cutoff)
    : input_vstorage_(vstorage),
@@ -304,7 +304,9 @@ Compaction::Compaction(
      output_compression_(_compression),
      output_compression_opts_(_compression_opts),
      output_temperature_(_output_temperature),
-      deletion_compaction_(_deletion_compaction),
+      deletion_compaction_(_compaction_reason == CompactionReason::kFIFOTtl ||
+                           _compaction_reason ==
+                               CompactionReason::kFIFOMaxSize),
      l0_files_might_overlap_(l0_files_might_overlap),
      inputs_(PopulateWithAtomicBoundaries(vstorage, std::move(_inputs))),
      grandparents_(std::move(_grandparents)),
@@ -321,7 +323,8 @@ Compaction::Compaction(
              ? false
              : IsBottommostLevel(output_level_, vstorage, inputs_)),
      is_full_compaction_(IsFullCompaction(vstorage, inputs_)),
-      is_manual_compaction_(_manual_compaction),
+      is_manual_compaction_(_compaction_reason ==
+                            CompactionReason::kManualCompaction),
      trim_ts_(_trim_ts),
      is_trivial_move_(false),
      compaction_reason_(_compaction_reason),
@@ -349,9 +352,6 @@ Compaction::Compaction(
                                      immutable_options_, start_level_,
                                      output_level_)) {
  MarkFilesBeingCompacted(true);
-  if (is_manual_compaction_) {
-    compaction_reason_ = CompactionReason::kManualCompaction;
-  }
  if (max_subcompactions_ == 0) {
    max_subcompactions_ = _mutable_db_options.max_subcompactions;
  }
@@ -647,6 +647,8 @@ bool Compaction::KeyNotExistsBeyondOutputLevel(
    return true;
  } else if (output_level_ != 0 &&
             cfd_->ioptions().compaction_style == kCompactionStyleLevel) {
+    // TODO: apply the optimization here to other compaction styles and
+    // compaction/flush to L0.
    // Maybe use binary search to find right entry instead of linear search?
    const Comparator* user_cmp = cfd_->user_comparator();
    for (int lvl = output_level_ + 1; lvl < number_levels_; lvl++) {
@@ -94,10 +94,9 @@ class Compaction {
             std::vector<FileMetaData*> grandparents,
             std::optional<SequenceNumber> earliest_snapshot,
             const SnapshotChecker* snapshot_checker,
-             bool manual_compaction = false, const std::string& trim_ts = "",
-             double score = -1, bool deletion_compaction = false,
+             CompactionReason compaction_reason,
+             const std::string& trim_ts = "", double score = -1,
             bool l0_files_might_overlap = true,
-             CompactionReason compaction_reason = CompactionReason::kUnknown,
             BlobGarbageCollectionPolicy blob_garbage_collection_policy =
                 BlobGarbageCollectionPolicy::kUseDefault,
             double blob_garbage_collection_age_cutoff = -1);
@@ -28,7 +28,7 @@ CompactionIterator::CompactionIterator(
    SequenceNumber earliest_snapshot,
    SequenceNumber earliest_write_conflict_snapshot,
    SequenceNumber job_snapshot, const SnapshotChecker* snapshot_checker,
-    Env* env, bool report_detailed_time, bool expect_valid_internal_key,
+    Env* env, bool report_detailed_time,
    CompactionRangeDelAggregator* range_del_agg,
    BlobFileBuilder* blob_file_builder, bool allow_data_in_errors,
    bool enforce_single_del_contracts,
@@ -42,8 +42,8 @@ CompactionIterator::CompactionIterator(
    : CompactionIterator(
          input, cmp, merge_helper, last_sequence, snapshots, earliest_snapshot,
          earliest_write_conflict_snapshot, job_snapshot, snapshot_checker, env,
-          report_detailed_time, expect_valid_internal_key, range_del_agg,
-          blob_file_builder, allow_data_in_errors, enforce_single_del_contracts,
+          report_detailed_time, range_del_agg, blob_file_builder,
+          allow_data_in_errors, enforce_single_del_contracts,
          manual_compaction_canceled,
          compaction ? std::make_unique<RealCompaction>(compaction) : nullptr,
          must_count_input_entries, compaction_filter, shutting_down, info_log,
@@ -55,7 +55,7 @@ CompactionIterator::CompactionIterator(
    SequenceNumber earliest_snapshot,
    SequenceNumber earliest_write_conflict_snapshot,
    SequenceNumber job_snapshot, const SnapshotChecker* snapshot_checker,
-    Env* env, bool report_detailed_time, bool expect_valid_internal_key,
+    Env* env, bool report_detailed_time,
    CompactionRangeDelAggregator* range_del_agg,
    BlobFileBuilder* blob_file_builder, bool allow_data_in_errors,
    bool enforce_single_del_contracts,
@@ -76,16 +76,14 @@ CompactionIterator::CompactionIterator(
      env_(env),
      clock_(env_->GetSystemClock().get()),
      report_detailed_time_(report_detailed_time),
-      expect_valid_internal_key_(expect_valid_internal_key),
      range_del_agg_(range_del_agg),
      blob_file_builder_(blob_file_builder),
      compaction_(std::move(compaction)),
      compaction_filter_(compaction_filter),
      shutting_down_(shutting_down),
      manual_compaction_canceled_(manual_compaction_canceled),
-      bottommost_level_(!compaction_ ? false
-                                     : compaction_->bottommost_level() &&
-                                           !compaction_->allow_ingest_behind()),
+      bottommost_level_(compaction_ && compaction_->bottommost_level() &&
+                        !compaction_->allow_ingest_behind()),
      // snapshots_ cannot be nullptr, but we will assert later in the body of
      // the constructor.
      visible_at_tip_(snapshots_ ? snapshots_->empty() : false),
@@ -161,6 +159,7 @@ void CompactionIterator::Next() {
      // MergeUntil stops when it encounters a corrupt key and does not
      // include them in the result, so we expect the keys here to be valid.
      if (!s.ok()) {
+        // FIXME: should fail compaction after this fatal logging.
        ROCKS_LOG_FATAL(
            info_log_, "Invalid ikey %s in compaction. %s",
            allow_data_in_errors_ ? key_.ToString(true).c_str() : "hidden",
@@ -464,18 +463,9 @@ void CompactionIterator::NextFromInput() {
    if (!pik_status.ok()) {
      iter_stats_.num_input_corrupt_records++;

-      // If `expect_valid_internal_key_` is false, return the corrupted key
-      // and let the caller decide what to do with it.
-      if (expect_valid_internal_key_) {
-        status_ = pik_status;
-        return;
-      }
-      key_ = current_key_.SetInternalKey(key_);
-      has_current_user_key_ = false;
-      current_user_key_sequence_ = kMaxSequenceNumber;
-      current_user_key_snapshot_ = 0;
-      validity_info_.SetValid(ValidContext::kParseKeyError);
-      break;
+      // Always fail compaction when encountering corrupted internal keys
+      status_ = pik_status;
+      return;
    }
    TEST_SYNC_POINT_CALLBACK("CompactionIterator:ProcessKV", &ikey_);
    if (is_range_del_) {
@@ -642,7 +632,8 @@ void CompactionIterator::NextFromInput() {
    } else if (ikey_.type == kTypeSingleDeletion) {
      // We can compact out a SingleDelete if:
      // 1) We encounter the corresponding PUT -OR- we know that this key
-      //    doesn't appear past this output level
+      //    doesn't appear past this output level and  we are not in
+      //    ingest_behind mode.
      // =AND=
      // 2) We've already returned a record in this snapshot -OR-
      //    there are no earlier earliest_write_conflict_snapshot.
@@ -731,6 +722,8 @@ void CompactionIterator::NextFromInput() {
            "CompactionIterator::NextFromInput:SingleDelete:1",
            const_cast<Compaction*>(c));
        if (last_key_seq_zeroed_) {
+          // Drop SD and the next key since they are both in the last
+          // snapshot (since last key has seqno zeroed).
          ++iter_stats_.num_record_drop_hidden;
          ++iter_stats_.num_record_drop_obsolete;
          assert(bottommost_level_);
@@ -841,7 +834,7 @@ void CompactionIterator::NextFromInput() {
        // iteration. If the next key is corrupt, we return before the
        // comparison, so the value of has_current_user_key does not matter.
        has_current_user_key_ = false;
-        if (compaction_ != nullptr &&
+        if (compaction_ != nullptr && !compaction_->allow_ingest_behind() &&
            DefinitelyInSnapshot(ikey_.sequence, earliest_snapshot_) &&
            compaction_->KeyNotExistsBeyondOutputLevel(ikey_.user_key,
                                                       &level_ptrs_) &&
@@ -854,6 +847,9 @@ void CompactionIterator::NextFromInput() {
            ++iter_stats_.num_optimized_del_drop_obsolete;
          }
        } else if (last_key_seq_zeroed_) {
+          // Sequence number zeroing requires bottommost_level_, which is
+          // false with ingest_behind.
+          assert(!compaction_->allow_ingest_behind());
          // Skip.
          ++iter_stats_.num_record_drop_hidden;
          ++iter_stats_.num_record_drop_obsolete;
@@ -870,6 +866,7 @@ void CompactionIterator::NextFromInput() {
    } else if (last_sequence != kMaxSequenceNumber &&
               (last_snapshot == current_user_key_snapshot_ ||
                last_snapshot < current_user_key_snapshot_)) {
+      // rule (A):
      // If the earliest snapshot is which this key is visible in
      // is the same as the visibility of a previous instance of the
      // same key, then this kv is not visible in any snapshot.
@@ -878,6 +875,15 @@ void CompactionIterator::NextFromInput() {
      // Note: Dropping this key will not affect TransactionDB write-conflict
      // checking since there has already been a record returned for this key
      // in this snapshot.
+      // When ingest_behind is enabled, it's ok that we drop an overwritten
+      // Delete here. The overwritting key still covers whatever that will be
+      // ingested. Note that we will not drop SingleDelete here as SingleDelte
+      // is handled entirely in its own if clause. This is important, see
+      // example: from new to old: SingleDelete_1, PUT_1, SingleDelete_2, PUT_2,
+      // where all operations are on the same key and PUT_2 is ingested with
+      // ingest_behind=true. If SingleDelete_2 is dropped due to being compacted
+      // together with PUT_1, and then PUT_1 is compacted away together with
+      // SingleDelete_1, PUT_2 can incorrectly becomes visible.
      if (last_sequence < current_user_key_sequence_) {
        ROCKS_LOG_FATAL(info_log_,
                        "key %s, last_sequence (%" PRIu64
@@ -887,12 +893,13 @@ void CompactionIterator::NextFromInput() {
        assert(false);
      }

-      ++iter_stats_.num_record_drop_hidden;  // rule (A)
+      ++iter_stats_.num_record_drop_hidden;
      AdvanceInputIter();
    } else if (compaction_ != nullptr &&
               (ikey_.type == kTypeDeletion ||
                (ikey_.type == kTypeDeletionWithTimestamp &&
                 cmp_with_history_ts_low_ < 0)) &&
+               !compaction_->allow_ingest_behind() &&
               DefinitelyInSnapshot(ikey_.sequence, earliest_snapshot_) &&
               compaction_->KeyNotExistsBeyondOutputLevel(ikey_.user_key,
                                                          &level_ptrs_)) {
@@ -928,11 +935,13 @@ void CompactionIterator::NextFromInput() {
                (ikey_.type == kTypeDeletionWithTimestamp &&
                 cmp_with_history_ts_low_ < 0)) &&
               bottommost_level_) {
+      assert(compaction_);
+      assert(!compaction_->allow_ingest_behind());  // bottommost_level_ is true
      // Handle the case where we have a delete key at the bottom most level
      // We can skip outputting the key iff there are no subsequent puts for
      // this key
-      assert(!compaction_ || compaction_->KeyNotExistsBeyondOutputLevel(
-                                 ikey_.user_key, &level_ptrs_));
+      assert(compaction_->KeyNotExistsBeyondOutputLevel(ikey_.user_key,
+                                                        &level_ptrs_));
      ParsedInternalKey next_ikey;
      AdvanceInputIter();
 #ifndef NDEBUG
@@ -974,6 +983,12 @@ void CompactionIterator::NextFromInput() {
                (compaction_ != nullptr &&
                 compaction_->KeyNotExistsBeyondOutputLevel(ikey_.user_key,
                                                            &level_ptrs_)))) {
+      // FIXME: it's possible that we are setting sequence number to 0 as
+      // preferred sequence number here. If cf_ingest_behind is enabled, this
+      // may fail ingestions since they expect all keys above the last level
+      // to have non-zero sequence number. We should probably not allow seqno
+      // zeroing here.
+      //
      // This section that attempts to swap preferred sequence number will not
      // be invoked if this is a CompactionIterator created for flush, since
      // `compaction_` will be nullptr and it's not bottommost either.
@@ -1274,11 +1289,11 @@ void CompactionIterator::PrepareOutput() {
    //
    // Can we do the same for levels above bottom level as long as
    // KeyNotExistsBeyondOutputLevel() return true?
-    if (Valid() && compaction_ != nullptr &&
-        !compaction_->allow_ingest_behind() && bottommost_level_ &&
+    if (Valid() && bottommost_level_ &&
        DefinitelyInSnapshot(ikey_.sequence, earliest_snapshot_) &&
        ikey_.type != kTypeMerge && current_key_committed_ &&
        ikey_.sequence <= preserve_seqno_after_ && !is_range_del_) {
+      assert(compaction_ != nullptr && !compaction_->allow_ingest_behind());
      if (ikey_.type == kTypeDeletion ||
          (ikey_.type == kTypeSingleDeletion && timestamp_size_ == 0)) {
        ROCKS_LOG_FATAL(
@@ -145,7 +145,8 @@ class CompactionIterator {
    }

    bool allow_ingest_behind() const override {
-      return compaction_->immutable_options().allow_ingest_behind;
+      return compaction_->immutable_options().cf_allow_ingest_behind ||
+             compaction_->immutable_options().allow_ingest_behind;
    }

    bool allow_mmap_reads() const override {
@@ -192,7 +193,7 @@ class CompactionIterator {
      SequenceNumber earliest_snapshot,
      SequenceNumber earliest_write_conflict_snapshot,
      SequenceNumber job_snapshot, const SnapshotChecker* snapshot_checker,
-      Env* env, bool report_detailed_time, bool expect_valid_internal_key,
+      Env* env, bool report_detailed_time,
      CompactionRangeDelAggregator* range_del_agg,
      BlobFileBuilder* blob_file_builder, bool allow_data_in_errors,
      bool enforce_single_del_contracts,
@@ -212,7 +213,7 @@ class CompactionIterator {
                     SequenceNumber earliest_write_conflict_snapshot,
                     SequenceNumber job_snapshot,
                     const SnapshotChecker* snapshot_checker, Env* env,
-                     bool report_detailed_time, bool expect_valid_internal_key,
+                     bool report_detailed_time,
                     CompactionRangeDelAggregator* range_del_agg,
                     BlobFileBuilder* blob_file_builder,
                     bool allow_data_in_errors,
@@ -347,7 +348,6 @@ class CompactionIterator {
  Env* env_;
  SystemClock* clock_;
  const bool report_detailed_time_;
-  const bool expect_valid_internal_key_;
  CompactionRangeDelAggregator* range_del_agg_;
  BlobFileBuilder* blob_file_builder_;
  std::unique_ptr<CompactionProxy> compaction_;
@@ -417,13 +417,15 @@ class CompactionIterator {
  // NextFromInput()).
  ParsedInternalKey ikey_;

-  // Stores whether ikey_.user_key is valid. If set to false, the user key is
-  // not compared against the current key in the underlying iterator.
+  // Stores whether current_user_key_ is valid. If so, current_user_key_
+  // stores the user key of the last key seen by the iterator.
+  // If false, treat the next key to read as a new user key.
  bool has_current_user_key_ = false;
  // If false, the iterator holds a copy of the current compaction iterator
  // output (or current key in the underlying iterator during NextFromInput()).
  bool at_next_ = false;

+  // A copy of the current internal key.
  IterKey current_key_;
  Slice current_user_key_;
  std::string curr_ts_;
@@ -433,8 +435,9 @@ class CompactionIterator {
  // True if the iterator has already returned a record for the current key.
  bool has_outputted_key_ = false;

-  // truncated the value of the next key and output it without applying any
-  // compaction rules.  This is used for outputting a put after a single delete.
+  // Truncate the value of the next key and output it without applying any
+  // compaction rules. This is an optimization for outputting a put after
+  // a single delete. See more in `NextFromInput()` under Optimization 3.
  bool clear_and_output_next_key_ = false;

  MergeOutputIterator merge_out_iter_;
@@ -294,7 +294,7 @@ class CompactionIteratorTest : public testing::TestWithParam<bool> {
        snapshots_.empty() ? kMaxSequenceNumber : snapshots_.at(0),
        earliest_write_conflict_snapshot, kMaxSequenceNumber,
        snapshot_checker_.get(), Env::Default(),
-        false /* report_detailed_time */, false, range_del_agg_.get(),
+        false /* report_detailed_time */, range_del_agg_.get(),
        nullptr /* blob_file_builder */, true /*allow_data_in_errors*/,
        true /*enforce_single_del_contracts*/,
        /*manual_compaction_canceled=*/kManualCompactionCanceledFalse_,
@@ -374,8 +374,7 @@ TEST_P(CompactionIteratorTest, EmptyResult) {
  ASSERT_FALSE(c_iter_->Valid());
 }

-// If there is a corruption after a single deletion, the corrupted key should
-// be preserved.
+// If there is a corruption after a single deletion, the compaction should fail.
 TEST_P(CompactionIteratorTest, CorruptionAfterSingleDeletion) {
  InitIterators({test::KeyStr("a", 5, kTypeSingleDeletion),
                 test::KeyStr("a", 3, kTypeValue, true),
@@ -386,14 +385,10 @@ TEST_P(CompactionIteratorTest, CorruptionAfterSingleDeletion) {
  ASSERT_EQ(test::KeyStr("a", 5, kTypeSingleDeletion),
            c_iter_->key().ToString());
  c_iter_->Next();
-  ASSERT_TRUE(c_iter_->Valid());
-  ASSERT_EQ(test::KeyStr("a", 3, kTypeValue, true), c_iter_->key().ToString());
-  c_iter_->Next();
-  ASSERT_TRUE(c_iter_->Valid());
-  ASSERT_EQ(test::KeyStr("b", 10, kTypeValue), c_iter_->key().ToString());
-  c_iter_->Next();
-  ASSERT_OK(c_iter_->status());
+  // The iterator should now fail when encountering the corrupted key
  ASSERT_FALSE(c_iter_->Valid());
+  ASSERT_FALSE(c_iter_->status().ok());
+  ASSERT_TRUE(c_iter_->status().IsCorruption());
 }

 // Tests compatibility of TimedPut and SingleDelete. TimedPut should act as if
@@ -668,16 +668,17 @@ void CompactionJob::GenSubcompactionBoundaries() {
               extra_num_subcompaction_threads_reserved_));
 }

-Status CompactionJob::Run() {
+void CompactionJob::InitializeCompactionRun() {
  AutoThreadOperationStageUpdater stage_updater(
      ThreadStatus::STAGE_COMPACTION_RUN);
  TEST_SYNC_POINT("CompactionJob::Run():Start");
  log_buffer_->FlushBufferToLog();
  LogCompaction();
+}

+void CompactionJob::RunSubcompactions() {
  const size_t num_threads = compact_->sub_compact_states.size();
  assert(num_threads > 0);
-  const uint64_t start_micros = db_options_.clock->NowMicros();
  compact_->compaction->GetOrInitInputTableProperties();

  // Launch a thread for each of subcompactions 1...num_threads-1
@@ -696,25 +697,43 @@ Status CompactionJob::Run() {
  for (auto& thread : thread_pool) {
    thread.join();
  }
+  RemoveEmptyOutputs();

+  ReleaseSubcompactionResources();
+  TEST_SYNC_POINT("CompactionJob::ReleaseSubcompactionResources");
+}
+
+void CompactionJob::UpdateTimingStats(uint64_t start_micros) {
  internal_stats_.SetMicros(db_options_.clock->NowMicros() - start_micros);

  for (auto& state : compact_->sub_compact_states) {
    internal_stats_.AddCpuMicros(state.compaction_job_stats.cpu_micros);
-    state.RemoveLastEmptyOutput();
  }

  RecordTimeToHistogram(stats_, COMPACTION_TIME,
                        internal_stats_.output_level_stats.micros);
  RecordTimeToHistogram(stats_, COMPACTION_CPU_TIME,
                        internal_stats_.output_level_stats.cpu_micros);
+}

-  TEST_SYNC_POINT("CompactionJob::Run:BeforeVerify");
+void CompactionJob::RemoveEmptyOutputs() {
+  for (auto& state : compact_->sub_compact_states) {
+    state.RemoveLastEmptyOutput();
+  }
+}

-  // Check if any thread encountered an error during execution
+bool CompactionJob::HasNewBlobFiles() const {
+  for (const auto& state : compact_->sub_compact_states) {
+    if (state.Current().HasBlobFileAdditions()) {
+      return true;
+    }
+  }
+  return false;
+}
+
+Status CompactionJob::CollectSubcompactionErrors() {
  Status status;
  IOStatus io_s;
-  bool wrote_new_blob_files = false;

  for (const auto& state : compact_->sub_compact_states) {
    if (!state.status.ok()) {
@@ -722,125 +741,131 @@ Status CompactionJob::Run() {
      io_s = state.io_status;
      break;
    }
-
-    if (state.Current().HasBlobFileAdditions()) {
-      wrote_new_blob_files = true;
-    }
  }

  if (io_status_.ok()) {
    io_status_ = io_s;
  }
-  if (status.ok()) {
-    constexpr IODebugContext* dbg = nullptr;

-    if (output_directory_) {
-      io_s = output_directory_->FsyncWithDirOptions(
-          IOOptions(), dbg,
-          DirFsyncOptions(DirFsyncOptions::FsyncReason::kNewFileSynced));
-    }
+  return status;
+}

-    if (io_s.ok() && wrote_new_blob_files && blob_output_directory_ &&
-        blob_output_directory_ != output_directory_) {
-      io_s = blob_output_directory_->FsyncWithDirOptions(
-          IOOptions(), dbg,
-          DirFsyncOptions(DirFsyncOptions::FsyncReason::kNewFileSynced));
-    }
+Status CompactionJob::SyncOutputDirectories() {
+  Status status;
+  IOStatus io_s;
+  constexpr IODebugContext* dbg = nullptr;
+  const bool wrote_new_blob_files = HasNewBlobFiles();
+  if (output_directory_) {
+    io_s = output_directory_->FsyncWithDirOptions(
+        IOOptions(), dbg,
+        DirFsyncOptions(DirFsyncOptions::FsyncReason::kNewFileSynced));
  }
+
+  if (io_s.ok() && wrote_new_blob_files && blob_output_directory_ &&
+      blob_output_directory_ != output_directory_) {
+    io_s = blob_output_directory_->FsyncWithDirOptions(
+        IOOptions(), dbg,
+        DirFsyncOptions(DirFsyncOptions::FsyncReason::kNewFileSynced));
+  }
+
  if (io_status_.ok()) {
    io_status_ = io_s;
  }
  if (status.ok()) {
    status = io_s;
  }
-  if (status.ok()) {
-    thread_pool.clear();
-    std::vector<const CompactionOutputs::Output*> files_output;
-    for (const auto& state : compact_->sub_compact_states) {
-      for (const auto& output : state.GetOutputs()) {
-        files_output.emplace_back(&output);
-      }
+
+  return status;
+}
+
+Status CompactionJob::VerifyOutputFiles() {
+  Status status;
+  std::vector<port::Thread> thread_pool;
+  std::vector<const CompactionOutputs::Output*> files_output;
+  for (const auto& state : compact_->sub_compact_states) {
+    for (const auto& output : state.GetOutputs()) {
+      files_output.emplace_back(&output);
    }
-    ColumnFamilyData* cfd = compact_->compaction->column_family_data();
-    std::atomic<size_t> next_file_idx(0);
-    auto verify_table = [&](Status& output_status) {
-      while (true) {
-        size_t file_idx = next_file_idx.fetch_add(1);
-        if (file_idx >= files_output.size()) {
-          break;
-        }
-        // Verify that the table is usable
-        // We set for_compaction to false and don't
-        // OptimizeForCompactionTableRead here because this is a special case
-        // after we finish the table building No matter whether
-        // use_direct_io_for_flush_and_compaction is true, we will regard this
-        // verification as user reads since the goal is to cache it here for
-        // further user reads
-        ReadOptions verify_table_read_options(Env::IOActivity::kCompaction);
-        verify_table_read_options.rate_limiter_priority =
-            GetRateLimiterPriority();
-        InternalIterator* iter = cfd->table_cache()->NewIterator(
-            verify_table_read_options, file_options_,
-            cfd->internal_comparator(), files_output[file_idx]->meta,
-            /*range_del_agg=*/nullptr,
-            compact_->compaction->mutable_cf_options(),
-            /*table_reader_ptr=*/nullptr,
-            cfd->internal_stats()->GetFileReadHist(
-                compact_->compaction->output_level()),
-            TableReaderCaller::kCompactionRefill, /*arena=*/nullptr,
-            /*skip_filters=*/false, compact_->compaction->output_level(),
-            MaxFileSizeForL0MetaPin(compact_->compaction->mutable_cf_options()),
-            /*smallest_compaction_key=*/nullptr,
-            /*largest_compaction_key=*/nullptr,
-            /*allow_unprepared_value=*/false);
-        auto s = iter->status();
+  }
+  ColumnFamilyData* cfd = compact_->compaction->column_family_data();
+  std::atomic<size_t> next_file_idx(0);
+  auto verify_table = [&](Status& output_status) {
+    while (true) {
+      size_t file_idx = next_file_idx.fetch_add(1);
+      if (file_idx >= files_output.size()) {
+        break;
+      }
+      // Verify that the table is usable
+      // We set for_compaction to false and don't
+      // OptimizeForCompactionTableRead here because this is a special case
+      // after we finish the table building No matter whether
+      // use_direct_io_for_flush_and_compaction is true, we will regard this
+      // verification as user reads since the goal is to cache it here for
+      // further user reads
+      ReadOptions verify_table_read_options(Env::IOActivity::kCompaction);
+      verify_table_read_options.rate_limiter_priority =
+          GetRateLimiterPriority();
+      InternalIterator* iter = cfd->table_cache()->NewIterator(
+          verify_table_read_options, file_options_, cfd->internal_comparator(),
+          files_output[file_idx]->meta,
+          /*range_del_agg=*/nullptr, compact_->compaction->mutable_cf_options(),
+          /*table_reader_ptr=*/nullptr,
+          cfd->internal_stats()->GetFileReadHist(
+              compact_->compaction->output_level()),
+          TableReaderCaller::kCompactionRefill, /*arena=*/nullptr,
+          /*skip_filters=*/false, compact_->compaction->output_level(),
+          MaxFileSizeForL0MetaPin(compact_->compaction->mutable_cf_options()),
+          /*smallest_compaction_key=*/nullptr,
+          /*largest_compaction_key=*/nullptr,
+          /*allow_unprepared_value=*/false);
+      auto s = iter->status();

-        if (s.ok() && paranoid_file_checks_) {
-          OutputValidator validator(cfd->internal_comparator(),
-                                    /*_enable_hash=*/true);
-          for (iter->SeekToFirst(); iter->Valid(); iter->Next()) {
-            s = validator.Add(iter->key(), iter->value());
-            if (!s.ok()) {
-              break;
-            }
-          }
-          if (s.ok()) {
-            s = iter->status();
-          }
-          if (s.ok() &&
-              !validator.CompareValidator(files_output[file_idx]->validator)) {
-            s = Status::Corruption("Paranoid checksums do not match");
+      if (s.ok() && paranoid_file_checks_) {
+        OutputValidator validator(cfd->internal_comparator(),
+                                  /*_enable_hash=*/true);
+        for (iter->SeekToFirst(); iter->Valid(); iter->Next()) {
+          s = validator.Add(iter->key(), iter->value());
+          if (!s.ok()) {
+            break;
          }
        }
-
-        delete iter;
-
-        if (!s.ok()) {
-          output_status = s;
-          break;
+        if (s.ok()) {
+          s = iter->status();
+        }
+        if (s.ok() &&
+            !validator.CompareValidator(files_output[file_idx]->validator)) {
+          s = Status::Corruption("Paranoid checksums do not match");
        }
      }
-    };
-    for (size_t i = 1; i < compact_->sub_compact_states.size(); i++) {
-      thread_pool.emplace_back(
-          verify_table, std::ref(compact_->sub_compact_states[i].status));
-    }
-    verify_table(compact_->sub_compact_states[0].status);
-    for (auto& thread : thread_pool) {
-      thread.join();
-    }

-    for (const auto& state : compact_->sub_compact_states) {
-      if (!state.status.ok()) {
-        status = state.status;
+      delete iter;
+
+      if (!s.ok()) {
+        output_status = s;
        break;
      }
    }
+  };
+  for (size_t i = 1; i < compact_->sub_compact_states.size(); i++) {
+    thread_pool.emplace_back(verify_table,
+                             std::ref(compact_->sub_compact_states[i].status));
+  }
+  verify_table(compact_->sub_compact_states[0].status);
+  for (auto& thread : thread_pool) {
+    thread.join();
  }

-  ReleaseSubcompactionResources();
-  TEST_SYNC_POINT("CompactionJob::ReleaseSubcompactionResources");
+  for (const auto& state : compact_->sub_compact_states) {
+    if (!state.status.ok()) {
+      status = state.status;
+      break;
+    }
+  }

+  return status;
+}
+
+void CompactionJob::SetOutputTableProperties() {
  for (const auto& state : compact_->sub_compact_states) {
    for (const auto& output : state.GetOutputs()) {
      auto fn =
@@ -850,7 +875,9 @@ Status CompactionJob::Run() {
                                                     output.table_properties);
    }
  }
+}

+void CompactionJob::AggregateSubcompactionStats() {
  // Before the compaction starts, is_remote_compaction was set to true if
  // compaction_service is set. We now know whether each sub_compaction was
  // done remotely or not. Reset is_remote_compaction back to false and allow
@@ -859,68 +886,86 @@ Status CompactionJob::Run() {

  // Finish up all bookkeeping to unify the subcompaction results.
  compact_->AggregateCompactionStats(internal_stats_, *job_stats_);
+}

-  uint64_t num_input_range_del = 0;
-  bool ok = BuildStatsFromInputTableProperties(&num_input_range_del);
-  // (Sub)compactions returned ok, do sanity check on the number of input
-  // keys.
-  if (status.ok() && ok) {
-    if (job_stats_->has_num_input_records) {
-      status = VerifyInputRecordCount(num_input_range_del);
-      if (!status.ok()) {
-        ROCKS_LOG_WARN(
-            db_options_.info_log, "[%s] [JOB %d] Compaction with status: %s",
-            compact_->compaction->column_family_data()->GetName().c_str(),
-            job_context_->job_id, status.ToString().c_str());
-      }
+Status CompactionJob::VerifyCompactionRecordCounts(
+    bool stats_built_from_input_table_prop, uint64_t num_input_range_del) {
+  Status status;
+  if (stats_built_from_input_table_prop && job_stats_->has_num_input_records) {
+    status = VerifyInputRecordCount(num_input_range_del);
+    if (!status.ok()) {
+      return status;
    }
+  }
+
+  const auto& mutable_cf_options = compact_->compaction->mutable_cf_options();
+  if ((mutable_cf_options.table_factory->IsInstanceOf(
+           TableFactory::kBlockBasedTableName()) ||
+       mutable_cf_options.table_factory->IsInstanceOf(
+           TableFactory::kPlainTableName()))) {
+    status = VerifyOutputRecordCount();
+    if (!status.ok()) {
+      return status;
+    }
+  }
+  return status;
+}
+
+void CompactionJob::FinalizeCompactionRun(
+    const Status& input_status, bool stats_built_from_input_table_prop,
+    uint64_t num_input_range_del) {
+  if (stats_built_from_input_table_prop) {
    UpdateCompactionJobInputStats(internal_stats_, num_input_range_del);
  }
  UpdateCompactionJobOutputStats(internal_stats_);
-
-  // Verify number of output records
-  // Only verify on table with format collects table properties
-  const auto& mutable_cf_options = compact_->compaction->mutable_cf_options();
-  if (status.ok() &&
-      (mutable_cf_options.table_factory->IsInstanceOf(
-           TableFactory::kBlockBasedTableName()) ||
-       mutable_cf_options.table_factory->IsInstanceOf(
-           TableFactory::kPlainTableName())) &&
-      db_options_.compaction_verify_record_count) {
-    uint64_t total_output_num = 0;
-    for (const auto& state : compact_->sub_compact_states) {
-      for (const auto& output : state.GetOutputs()) {
-        total_output_num += output.table_properties->num_entries -
-                            output.table_properties->num_range_deletions;
-      }
-    }
-
-    uint64_t expected = internal_stats_.output_level_stats.num_output_records;
-    if (internal_stats_.has_proximal_level_output) {
-      expected += internal_stats_.proximal_level_stats.num_output_records;
-    }
-    if (expected != total_output_num) {
-      char scratch[2345];
-      compact_->compaction->Summary(scratch, sizeof(scratch));
-      std::string msg =
-          "Number of keys in compaction output SST files does not match "
-          "number of keys added. Expected " +
-          std::to_string(expected) + " but there are " +
-          std::to_string(total_output_num) +
-          " in output SST files. Compaction summary: " + scratch;
-      ROCKS_LOG_WARN(
-          db_options_.info_log, "[%s] [JOB %d] Compaction with status: %s",
-          compact_->compaction->column_family_data()->GetName().c_str(),
-          job_context_->job_id, msg.c_str());
-      status = Status::Corruption(msg);
-    }
-  }
-
  RecordCompactionIOStats();
+
  LogFlush(db_options_.info_log);
  TEST_SYNC_POINT("CompactionJob::Run():End");
-  compact_->status = status;
-  TEST_SYNC_POINT_CALLBACK("CompactionJob::Run():EndStatusSet", &status);
+  compact_->status = input_status;
+  TEST_SYNC_POINT_CALLBACK("CompactionJob::Run():EndStatusSet",
+                           const_cast<Status*>(&input_status));
+}
+
+Status CompactionJob::Run() {
+  InitializeCompactionRun();
+
+  const uint64_t start_micros = db_options_.clock->NowMicros();
+
+  RunSubcompactions();
+
+  UpdateTimingStats(start_micros);
+
+  TEST_SYNC_POINT("CompactionJob::Run:BeforeVerify");
+
+  Status status = CollectSubcompactionErrors();
+
+  if (status.ok()) {
+    status = SyncOutputDirectories();
+  }
+
+  if (status.ok()) {
+    status = VerifyOutputFiles();
+  }
+
+  if (status.ok()) {
+    SetOutputTableProperties();
+  }
+
+  AggregateSubcompactionStats();
+
+  uint64_t num_input_range_del = 0;
+  bool stats_built_from_input_table_prop =
+      BuildStatsFromInputFiles(&num_input_range_del);
+
+  if (status.ok()) {
+    status = VerifyCompactionRecordCounts(stats_built_from_input_table_prop,
+                                          num_input_range_del);
+  }
+
+  FinalizeCompactionRun(status, stats_built_from_input_table_prop,
+                        num_input_range_del);
+
  return status;
 }

@@ -1365,8 +1410,7 @@ void CompactionJob::ProcessKeyValueCompaction(SubcompactionState* sub_compact) {
      &(job_context_->snapshot_seqs), earliest_snapshot_,
      job_context_->earliest_write_conflict_snapshot, job_snapshot_seq,
      job_context_->snapshot_checker, env_,
-      ShouldReportDetailedTime(env_, stats_),
-      /*expect_valid_internal_key=*/true, sub_compact->RangeDelAgg(),
+      ShouldReportDetailedTime(env_, stats_), sub_compact->RangeDelAgg(),
      blob_file_builder.get(), db_options_.allow_data_in_errors,
      db_options_.enforce_single_del_contracts, manual_compaction_canceled_,
      sub_compact->compaction
@@ -1736,14 +1780,11 @@ Status CompactionJob::FinishCompactionOutputFile(
  if (s.ok()) {
    tp = outputs.GetTableProperties();
  }
-
  if (s.ok() && current_entries == 0 && tp.num_range_deletions == 0) {
    // If there is nothing to output, no necessary to generate a sst file.
    // This happens when the output level is bottom level, at the same time
    // the sub_compact output nothing.
-    std::string fname =
-        TableFileName(sub_compact->compaction->immutable_options().cf_paths,
-                      meta->fd.GetNumber(), meta->fd.GetPathId());
+    std::string fname = GetTableFileName(meta->fd.GetNumber());

    // TODO(AR) it is not clear if there are any larger implications if
    // DeleteFile fails here
@@ -1942,6 +1983,10 @@ Status CompactionJob::OpenCompactionOutputFile(SubcompactionState* sub_compact,

  // no need to lock because VersionSet::next_file_number_ is atomic
  uint64_t file_number = versions_->NewFileNumber();
+#ifndef NDEBUG
+  TEST_SYNC_POINT_CALLBACK(
+      "CompactionJob::OpenCompactionOutputFile::NewFileNumber", &file_number);
+#endif
  std::string fname = GetTableFileName(file_number);
  // Fire events.
  ColumnFamilyData* cfd = sub_compact->compaction->column_family_data();
@@ -2105,8 +2150,7 @@ void CopyPrefix(const Slice& src, size_t prefix_length, std::string* dst) {
 }
 }  // namespace

-bool CompactionJob::BuildStatsFromInputTableProperties(
-    uint64_t* num_input_range_del) {
+bool CompactionJob::BuildStatsFromInputFiles(uint64_t* num_input_range_del) {
  assert(compact_);

  Compaction* compaction = compact_->compaction;
@@ -2377,6 +2421,11 @@ Status CompactionJob::VerifyInputRecordCount(
          "number of keys processed. Expected " +
          std::to_string(expected) + " but processed " +
          std::to_string(actual) + ". Compaction summary: " + scratch;
+      ROCKS_LOG_WARN(
+          db_options_.info_log,
+          "[%s] [JOB %d] VerifyInputRecordCount() Status: %s",
+          compact_->compaction->column_family_data()->GetName().c_str(),
+          job_context_->job_id, msg.c_str());
      if (db_options_.compaction_verify_record_count) {
        return Status::Corruption(msg);
      }
@@ -2385,4 +2434,38 @@ Status CompactionJob::VerifyInputRecordCount(
  return Status::OK();
 }

+Status CompactionJob::VerifyOutputRecordCount() const {
+  uint64_t total_output_num = 0;
+  for (const auto& state : compact_->sub_compact_states) {
+    for (const auto& output : state.GetOutputs()) {
+      total_output_num += output.table_properties->num_entries -
+                          output.table_properties->num_range_deletions;
+    }
+  }
+
+  uint64_t expected = internal_stats_.output_level_stats.num_output_records;
+  if (internal_stats_.has_proximal_level_output) {
+    expected += internal_stats_.proximal_level_stats.num_output_records;
+  }
+  if (expected != total_output_num) {
+    char scratch[2345];
+    compact_->compaction->Summary(scratch, sizeof(scratch));
+    std::string msg =
+        "Number of keys in compaction output SST files does not match "
+        "number of keys added. Expected " +
+        std::to_string(expected) + " but there are " +
+        std::to_string(total_output_num) +
+        " in output SST files. Compaction summary: " + scratch;
+    ROCKS_LOG_WARN(
+        db_options_.info_log,
+        "[%s] [JOB %d] VerifyOutputRecordCount() status: %s",
+        compact_->compaction->column_family_data()->GetName().c_str(),
+        job_context_->job_id, msg.c_str());
+    if (db_options_.compaction_verify_record_count) {
+      return Status::Corruption(msg);
+    }
+  }
+  return Status::OK();
+}
+
 }  // namespace ROCKSDB_NAMESPACE
@@ -225,7 +225,7 @@ class CompactionJob {
 private:
  friend class CompactionJobTestBase;

-  // Collect the following stats from Input Table Properties
+  // Collect the following stats from input files and table properties
  // - num_input_files_in_non_output_levels
  // - num_input_files_in_output_level
  // - bytes_read_non_output_levels
@@ -242,14 +242,14 @@ class CompactionJob {
  // num_input_range_del are calculated successfully.
  //
  // This should be called only once for compactions (not per subcompaction)
-  bool BuildStatsFromInputTableProperties(
-      uint64_t* num_input_range_del = nullptr);
+  bool BuildStatsFromInputFiles(uint64_t* num_input_range_del = nullptr);

  void UpdateCompactionJobInputStats(
      const InternalStats::CompactionStatsFull& internal_stats,
      uint64_t num_input_range_del) const;

  Status VerifyInputRecordCount(uint64_t num_input_range_del) const;
+  Status VerifyOutputRecordCount() const;

  // Generates a histogram representing potential divisions of key ranges from
  // the input. It adds the starting and/or ending keys of certain input files
@@ -278,6 +278,22 @@ class CompactionJob {
  // Release all reserved threads and update the compaction limits.
  void ReleaseSubcompactionResources();

+  void InitializeCompactionRun();
+  void RunSubcompactions();
+  void UpdateTimingStats(uint64_t start_micros);
+  void RemoveEmptyOutputs();
+  bool HasNewBlobFiles() const;
+  Status CollectSubcompactionErrors();
+  Status SyncOutputDirectories();
+  Status VerifyOutputFiles();
+  void SetOutputTableProperties();
+  void AggregateSubcompactionStats();
+  Status VerifyCompactionRecordCounts(bool stats_built_from_input_table_prop,
+                                      uint64_t num_input_range_del);
+  void FinalizeCompactionRun(const Status& status,
+                             bool stats_built_from_input_table_prop,
+                             uint64_t num_input_range_del);
+
  CompactionServiceJobStatus ProcessKeyValueCompactionWithCompactionService(
      SubcompactionState* sub_compact);

@@ -651,7 +651,8 @@ class CompactionJobTestBase : public testing::Test {
        mutable_cf_options_.max_compaction_bytes, 0, kNoCompression,
        cfd->GetLatestMutableCFOptions().compression_opts,
        Temperature::kUnknown, max_subcompactions, grandparents,
-        /*earliest_snapshot*/ std::nullopt, /*snapshot_checker*/ nullptr, true);
+        /*earliest_snapshot*/ std::nullopt, /*snapshot_checker*/ nullptr,
+        CompactionReason::kManualCompaction);
    compaction.FinalizeInputInfo(cfd->current());

    assert(db_options_.info_log);
@@ -333,11 +333,13 @@ bool CompactionPicker::AreFilesInCompaction(
  return false;
 }

-Compaction* CompactionPicker::CompactFiles(
+Compaction* CompactionPicker::PickCompactionForCompactFiles(
    const CompactionOptions& compact_options,
    const std::vector<CompactionInputFiles>& input_files, int output_level,
    VersionStorageInfo* vstorage, const MutableCFOptions& mutable_cf_options,
-    const MutableDBOptions& mutable_db_options, uint32_t output_path_id) {
+    const MutableDBOptions& mutable_db_options, uint32_t output_path_id,
+    std::optional<SequenceNumber> earliest_snapshot,
+    const SnapshotChecker* snapshot_checker) {
 #ifndef NDEBUG
  assert(input_files.size());
  // This compaction output should not overlap with a running compaction as
@@ -380,8 +382,8 @@ Compaction* CompactionPicker::CompactFiles(
      GetCompressionOptions(mutable_cf_options, vstorage, output_level),
      mutable_cf_options.default_write_temperature,
      compact_options.max_subcompactions,
-      /* grandparents */ {}, /* earliest_snapshot */ std::nullopt,
-      /* snapshot_checker */ nullptr, true);
+      /* grandparents */ {}, earliest_snapshot, snapshot_checker,
+      CompactionReason::kManualCompaction);
  RegisterCompaction(c);
  return c;
 }
@@ -601,7 +603,7 @@ void CompactionPicker::GetGrandparents(
  }
 }

-Compaction* CompactionPicker::CompactRange(
+Compaction* CompactionPicker::PickCompactionForCompactRange(
    const std::string& cf_name, const MutableCFOptions& mutable_cf_options,
    const MutableDBOptions& mutable_db_options, VersionStorageInfo* vstorage,
    int input_level, int output_level,
@@ -617,8 +619,8 @@ Compaction* CompactionPicker::CompactRange(
    // Universal compaction with more than one level always compacts all the
    // files together to the last level.
    assert(vstorage->num_levels() > 1);
-    int max_output_level =
-        vstorage->MaxOutputLevel(ioptions_.allow_ingest_behind);
+    int max_output_level = vstorage->MaxOutputLevel(
+        ioptions_.cf_allow_ingest_behind || ioptions_.allow_ingest_behind);
    // DBImpl::CompactRange() set output level to be the last level
    assert(output_level == max_output_level);
    // DBImpl::RunManualCompaction will make full range for universal compaction
@@ -680,10 +682,9 @@ Compaction* CompactionPicker::CompactRange(
        mutable_cf_options.default_write_temperature,
        compact_range_options.max_subcompactions,
        /* grandparents */ {}, /* earliest_snapshot */ std::nullopt,
-        /* snapshot_checker */ nullptr,
-        /* is manual */ true, trim_ts, /* score */ -1,
-        /* deletion_compaction */ false, /* l0_files_might_overlap */ true,
-        CompactionReason::kUnknown,
+        /* snapshot_checker */ nullptr, CompactionReason::kManualCompaction,
+        trim_ts, /* score */ -1,
+        /* l0_files_might_overlap */ true,
        compact_range_options.blob_garbage_collection_policy,
        compact_range_options.blob_garbage_collection_age_cutoff);

@@ -873,9 +874,8 @@ Compaction* CompactionPicker::CompactRange(
      mutable_cf_options.default_write_temperature,
      compact_range_options.max_subcompactions, std::move(grandparents),
      /* earliest_snapshot */ std::nullopt, /* snapshot_checker */ nullptr,
-      /* is manual */ true, trim_ts, /* score */ -1,
-      /* deletion_compaction */ false, /* l0_files_might_overlap */ true,
-      CompactionReason::kUnknown,
+      CompactionReason::kManualCompaction, trim_ts, /* score */ -1,
+      /* l0_files_might_overlap */ true,
      compact_range_options.blob_garbage_collection_policy,
      compact_range_options.blob_garbage_collection_age_cutoff);

@@ -75,7 +75,7 @@ class CompactionPicker {
  // *compaction_end should point to valid InternalKey!
  // REQUIRES: If not compacting all levels (input_level == kCompactAllLevels),
  // then levels between input_level and output_level should be empty.
-  virtual Compaction* CompactRange(
+  virtual Compaction* PickCompactionForCompactRange(
      const std::string& cf_name, const MutableCFOptions& mutable_cf_options,
      const MutableDBOptions& mutable_db_options, VersionStorageInfo* vstorage,
      int input_level, int output_level,
@@ -117,12 +117,17 @@ class CompactionPicker {
  // Caller must provide a set of input files that has been passed through
  // `SanitizeAndConvertCompactionInputFiles` earlier. The lock should not be
  // released between that call and this one.
-  Compaction* CompactFiles(const CompactionOptions& compact_options,
-                           const std::vector<CompactionInputFiles>& input_files,
-                           int output_level, VersionStorageInfo* vstorage,
-                           const MutableCFOptions& mutable_cf_options,
-                           const MutableDBOptions& mutable_db_options,
-                           uint32_t output_path_id);
+  //
+  //  TODO - Remove default values for earliest_snapshot and snapshot_checker
+  //  and require all callers to pass them in so that DB::CompactFiles() can
+  //  also benefit from Standalone Range Tombstone Optimization
+  Compaction* PickCompactionForCompactFiles(
+      const CompactionOptions& compact_options,
+      const std::vector<CompactionInputFiles>& input_files, int output_level,
+      VersionStorageInfo* vstorage, const MutableCFOptions& mutable_cf_options,
+      const MutableDBOptions& mutable_db_options, uint32_t output_path_id,
+      std::optional<SequenceNumber> earliest_snapshot = std::nullopt,
+      const SnapshotChecker* snapshot_checker = nullptr);

  // Converts a set of compaction input file numbers into
  // a list of CompactionInputFiles.
@@ -278,18 +283,17 @@ class NullCompactionPicker : public CompactionPicker {
  }

  // Always return "nullptr"
-  Compaction* CompactRange(const std::string& /*cf_name*/,
-                           const MutableCFOptions& /*mutable_cf_options*/,
-                           const MutableDBOptions& /*mutable_db_options*/,
-                           VersionStorageInfo* /*vstorage*/,
-                           int /*input_level*/, int /*output_level*/,
-                           const CompactRangeOptions& /*compact_range_options*/,
-                           const InternalKey* /*begin*/,
-                           const InternalKey* /*end*/,
-                           InternalKey** /*compaction_end*/,
-                           bool* /*manual_conflict*/,
-                           uint64_t /*max_file_num_to_ignore*/,
-                           const std::string& /*trim_ts*/) override {
+  Compaction* PickCompactionForCompactRange(
+      const std::string& /*cf_name*/,
+      const MutableCFOptions& /*mutable_cf_options*/,
+      const MutableDBOptions& /*mutable_db_options*/,
+      VersionStorageInfo* /*vstorage*/, int /*input_level*/,
+      int /*output_level*/,
+      const CompactRangeOptions& /*compact_range_options*/,
+      const InternalKey* /*begin*/, const InternalKey* /*end*/,
+      InternalKey** /*compaction_end*/, bool* /*manual_conflict*/,
+      uint64_t /*max_file_num_to_ignore*/,
+      const std::string& /*trim_ts*/) override {
    return nullptr;
  }

@@ -127,11 +127,9 @@ Compaction* FIFOCompactionPicker::PickTTLCompaction(
      mutable_cf_options.compression_opts,
      mutable_cf_options.default_write_temperature,
      /* max_subcompactions */ 0, {}, /* earliest_snapshot */ std::nullopt,
-      /* snapshot_checker */ nullptr,
-      /* is manual */ false,
+      /* snapshot_checker */ nullptr, CompactionReason::kFIFOTtl,
      /* trim_ts */ "", vstorage->CompactionScore(0),
-      /* is deletion compaction */ true, /* l0_files_might_overlap */ true,
-      CompactionReason::kFIFOTtl);
+      /* l0_files_might_overlap */ true);
  return c;
 }

@@ -200,11 +198,10 @@ Compaction* FIFOCompactionPicker::PickSizeCompaction(
            mutable_cf_options.default_write_temperature,
            0 /* max_subcompactions */, {},
            /* earliest_snapshot */ std::nullopt,
-            /* snapshot_checker */ nullptr, /* is manual */ false,
+            /* snapshot_checker */ nullptr,
+            CompactionReason::kFIFOReduceNumFiles,
            /* trim_ts */ "", vstorage->CompactionScore(0),
-            /* is deletion compaction */ false,
-            /* l0_files_might_overlap */ true,
-            CompactionReason::kFIFOReduceNumFiles);
+            /* l0_files_might_overlap */ true);
        return c;
      }
    }
@@ -261,6 +258,9 @@ Compaction* FIFOCompactionPicker::PickSizeCompaction(
    // better serves a major type of FIFO use cases where smaller keys are
    // associated with older data.
    for (const auto& f : last_level_files) {
+      if (f->being_compacted) {
+        continue;
+      }
      total_size -= f->fd.file_size;
      inputs[0].files.push_back(f);
      char tmp_fsize[16];
@@ -297,11 +297,9 @@ Compaction* FIFOCompactionPicker::PickSizeCompaction(
      mutable_cf_options.compression_opts,
      mutable_cf_options.default_write_temperature,
      /* max_subcompactions */ 0, {}, /* earliest_snapshot */ std::nullopt,
-      /* snapshot_checker */ nullptr,
-      /* is manual */ false,
+      /* snapshot_checker */ nullptr, CompactionReason::kFIFOMaxSize,
      /* trim_ts */ "", vstorage->CompactionScore(0),
-      /* is deletion compaction */ true,
-      /* l0_files_might_overlap */ true, CompactionReason::kFIFOMaxSize);
+      /* l0_files_might_overlap */ true);
  return c;
 }

@@ -416,10 +414,9 @@ Compaction* FIFOCompactionPicker::PickTemperatureChangeCompaction(
      mutable_cf_options.compression, mutable_cf_options.compression_opts,
      compaction_target_temp,
      /* max_subcompactions */ 0, {}, /* earliest_snapshot */ std::nullopt,
-      /* snapshot_checker */ nullptr,
-      /* is manual */ false, /* trim_ts */ "", vstorage->CompactionScore(0),
-      /* is deletion compaction */ false, /* l0_files_might_overlap */ true,
-      CompactionReason::kChangeTemperature);
+      /* snapshot_checker */ nullptr, CompactionReason::kChangeTemperature,
+      /* trim_ts */ "", vstorage->CompactionScore(0),
+      /* l0_files_might_overlap */ true);
  return c;
 }

@@ -446,7 +443,7 @@ Compaction* FIFOCompactionPicker::PickCompaction(
  return c;
 }

-Compaction* FIFOCompactionPicker::CompactRange(
+Compaction* FIFOCompactionPicker::PickCompactionForCompactRange(
    const std::string& cf_name, const MutableCFOptions& mutable_cf_options,
    const MutableDBOptions& mutable_db_options, VersionStorageInfo* vstorage,
    int input_level, int output_level,
@@ -26,16 +26,14 @@ class FIFOCompactionPicker : public CompactionPicker {
      VersionStorageInfo* version, LogBuffer* log_buffer,
      bool /* require_max_output_level*/ = false) override;

-  Compaction* CompactRange(const std::string& cf_name,
-                           const MutableCFOptions& mutable_cf_options,
-                           const MutableDBOptions& mutable_db_options,
-                           VersionStorageInfo* vstorage, int input_level,
-                           int output_level,
-                           const CompactRangeOptions& compact_range_options,
-                           const InternalKey* begin, const InternalKey* end,
-                           InternalKey** compaction_end, bool* manual_conflict,
-                           uint64_t max_file_num_to_ignore,
-                           const std::string& trim_ts) override;
+  Compaction* PickCompactionForCompactRange(
+      const std::string& cf_name, const MutableCFOptions& mutable_cf_options,
+      const MutableDBOptions& mutable_db_options, VersionStorageInfo* vstorage,
+      int input_level, int output_level,
+      const CompactRangeOptions& compact_range_options,
+      const InternalKey* begin, const InternalKey* end,
+      InternalKey** compaction_end, bool* manual_conflict,
+      uint64_t max_file_num_to_ignore, const std::string& trim_ts) override;

  // The maximum allowed output level.  Always returns 0.
  int MaxOutputLevel() const override { return 0; }
@@ -145,7 +145,6 @@ class LevelCompactionBuilder {
  int parent_index_ = -1;
  int base_index_ = -1;
  double start_level_score_ = 0;
-  bool is_manual_ = false;
  bool is_l0_trivial_move_ = false;
  CompactionInputFiles start_level_inputs_;
  std::vector<CompactionInputFiles> compaction_inputs_;
@@ -561,9 +560,8 @@ Compaction* LevelCompactionBuilder::GetCompaction() {
      mutable_cf_options_.default_write_temperature,
      /* max_subcompactions */ 0, std::move(grandparents_),
      /* earliest_snapshot */ std::nullopt, /* snapshot_checker */ nullptr,
-      is_manual_,
-      /* trim_ts */ "", start_level_score_, false /* deletion_compaction */,
-      l0_files_might_overlap, compaction_reason_);
+      compaction_reason_,
+      /* trim_ts */ "", start_level_score_, l0_files_might_overlap);

  // If it's level 0 compaction, make sure we don't execute any other level 0
  // compactions in parallel
@@ -544,41 +544,48 @@ TEST_F(CompactionPickerTest, NeedsCompactionUniversal) {
 }

 TEST_F(CompactionPickerTest, CompactionUniversalIngestBehindReservedLevel) {
-  const uint64_t kFileSize = 100000;
-  NewVersionStorage(3 /* num_levels */, kCompactionStyleUniversal);
-  ioptions_.allow_ingest_behind = true;
-  ioptions_.num_levels = 3;
-  UniversalCompactionPicker universal_compaction_picker(ioptions_, &icmp_);
-  UpdateVersionStorageInfo();
-  // must return false when there's no files.
-  ASSERT_EQ(universal_compaction_picker.NeedsCompaction(vstorage_.get()),
-            false);
+  for (bool cf_option : {false, true}) {
+    SCOPED_TRACE("cf_option = " + std::to_string(cf_option));
+    const uint64_t kFileSize = 100000;
+    NewVersionStorage(3 /* num_levels */, kCompactionStyleUniversal);
+    if (cf_option) {
+      ioptions_.cf_allow_ingest_behind = true;
+    } else {
+      ioptions_.allow_ingest_behind = true;
+    }
+    ioptions_.num_levels = 3;
+    UniversalCompactionPicker universal_compaction_picker(ioptions_, &icmp_);
+    UpdateVersionStorageInfo();
+    // must return false when there's no files.
+    ASSERT_EQ(universal_compaction_picker.NeedsCompaction(vstorage_.get()),
+              false);

-  NewVersionStorage(3, kCompactionStyleUniversal);
+    NewVersionStorage(3, kCompactionStyleUniversal);

-  Add(0, 1U, "150", "200", kFileSize, 0, 500, 550);
-  Add(0, 2U, "201", "250", kFileSize, 0, 401, 450);
-  Add(0, 4U, "260", "300", kFileSize, 0, 260, 300);
-  Add(1, 5U, "100", "151", kFileSize, 0, 200, 251);
-  Add(1, 3U, "301", "350", kFileSize, 0, 101, 150);
-  Add(2, 6U, "120", "200", kFileSize, 0, 20, 100);
+    Add(0, 1U, "150", "200", kFileSize, 0, 500, 550);
+    Add(0, 2U, "201", "250", kFileSize, 0, 401, 450);
+    Add(0, 4U, "260", "300", kFileSize, 0, 260, 300);
+    Add(1, 5U, "100", "151", kFileSize, 0, 200, 251);
+    Add(1, 3U, "301", "350", kFileSize, 0, 101, 150);
+    Add(2, 6U, "120", "200", kFileSize, 0, 20, 100);

-  UpdateVersionStorageInfo();
+    UpdateVersionStorageInfo();

-  std::unique_ptr<Compaction> compaction(
-      universal_compaction_picker.PickCompaction(
-          cf_name_, mutable_cf_options_, mutable_db_options_,
-          /*existing_snapshots=*/{}, /* snapshot_checker */ nullptr,
-          vstorage_.get(), &log_buffer_));
+    std::unique_ptr<Compaction> compaction(
+        universal_compaction_picker.PickCompaction(
+            cf_name_, mutable_cf_options_, mutable_db_options_,
+            /*existing_snapshots=*/{}, /* snapshot_checker */ nullptr,
+            vstorage_.get(), &log_buffer_));

-  // output level should be the one above the bottom-most
-  ASSERT_EQ(1, compaction->output_level());
+    // output level should be the one above the bottom-most
+    ASSERT_EQ(1, compaction->output_level());

-  // input should not include the reserved level
-  const std::vector<CompactionInputFiles>* inputs = compaction->inputs();
-  for (const auto& compaction_input : *inputs) {
-    if (!compaction_input.empty()) {
-      ASSERT_LT(compaction_input.level, 2);
+    // input should not include the reserved level
+    const std::vector<CompactionInputFiles>* inputs = compaction->inputs();
+    for (const auto& compaction_input : *inputs) {
+      if (!compaction_input.empty()) {
+        ASSERT_LT(compaction_input.level, 2);
+      }
    }
  }
 }
@@ -2689,13 +2696,14 @@ TEST_F(CompactionPickerTest, CompactRangeMaxCompactionBytes) {
  bool manual_conflict = false;
  InternalKey manual_end;
  InternalKey* manual_end_ptr = &manual_end;
-  std::unique_ptr<Compaction> compaction(level_compaction_picker.CompactRange(
-      cf_name_, mutable_cf_options_, mutable_db_options_, vstorage_.get(),
-      /*input_level=*/1, /*output_level=*/2,
-      /*compact_range_options*/ {}, /*begin=*/nullptr, /*end=*/nullptr,
-      &manual_end_ptr, &manual_conflict,
-      /*max_file_num_to_ignore=*/std::numeric_limits<uint64_t>::max(),
-      /*trim_ts=*/""));
+  std::unique_ptr<Compaction> compaction(
+      level_compaction_picker.PickCompactionForCompactRange(
+          cf_name_, mutable_cf_options_, mutable_db_options_, vstorage_.get(),
+          /*input_level=*/1, /*output_level=*/2,
+          /*compact_range_options*/ {}, /*begin=*/nullptr, /*end=*/nullptr,
+          &manual_end_ptr, &manual_conflict,
+          /*max_file_num_to_ignore=*/std::numeric_limits<uint64_t>::max(),
+          /*trim_ts=*/""));
  ASSERT_TRUE(compaction.get() != nullptr);
  ASSERT_EQ(1U, compaction->num_input_levels());
  ASSERT_EQ(2, compaction->output_level());
@@ -3627,7 +3635,7 @@ TEST_F(CompactionPickerTest, UniversalMarkedManualCompaction) {
  bool manual_conflict = false;
  InternalKey* manual_end = nullptr;
  std::unique_ptr<Compaction> compaction(
-      universal_compaction_picker.CompactRange(
+      universal_compaction_picker.PickCompactionForCompactRange(
          cf_name_, mutable_cf_options_, mutable_db_options_, vstorage_.get(),
          ColumnFamilyData::kCompactAllLevels, 6, CompactRangeOptions(),
          nullptr, nullptr, &manual_end, &manual_conflict,
@@ -3831,9 +3839,10 @@ TEST_F(CompactionPickerU64TsTest, Overlap) {
  std::vector<CompactionInputFiles> input_files;
  ASSERT_OK(level_compaction_picker.GetCompactionInputsFromFileNumbers(
      &input_files, &input, vstorage_.get(), CompactionOptions()));
-  std::unique_ptr<Compaction> comp1(level_compaction_picker.CompactFiles(
-      CompactionOptions(), input_files, level, vstorage_.get(),
-      mutable_cf_options_, mutable_db_options_, /*output_path_id=*/0));
+  std::unique_ptr<Compaction> comp1(
+      level_compaction_picker.PickCompactionForCompactFiles(
+          CompactionOptions(), input_files, level, vstorage_.get(),
+          mutable_cf_options_, mutable_db_options_, /*output_path_id=*/0));

  {
    // [600, ts=50000] to [600, ts=50000] is the range to check.
@@ -3942,9 +3951,10 @@ TEST_P(PerKeyPlacementCompactionPickerTest, OverlapWithNormalCompaction) {
  ASSERT_OK(level_compaction_picker.GetCompactionInputsFromFileNumbers(
      &input_files, &input_set, vstorage_.get(), comp_options));

-  std::unique_ptr<Compaction> comp1(level_compaction_picker.CompactFiles(
-      comp_options, input_files, 5, vstorage_.get(), mutable_cf_options_,
-      mutable_db_options_, 0));
+  std::unique_ptr<Compaction> comp1(
+      level_compaction_picker.PickCompactionForCompactFiles(
+          comp_options, input_files, 5, vstorage_.get(), mutable_cf_options_,
+          mutable_db_options_, 0));

  input_set.clear();
  input_files.clear();
@@ -3988,9 +3998,10 @@ TEST_P(PerKeyPlacementCompactionPickerTest, NormalCompactionOverlap) {
  ASSERT_OK(level_compaction_picker.GetCompactionInputsFromFileNumbers(
      &input_files, &input_set, vstorage_.get(), comp_options));

-  std::unique_ptr<Compaction> comp1(level_compaction_picker.CompactFiles(
-      comp_options, input_files, 6, vstorage_.get(), mutable_cf_options_,
-      mutable_db_options_, 0));
+  std::unique_ptr<Compaction> comp1(
+      level_compaction_picker.PickCompactionForCompactFiles(
+          comp_options, input_files, 6, vstorage_.get(), mutable_cf_options_,
+          mutable_db_options_, 0));

  input_set.clear();
  input_files.clear();
@@ -4030,9 +4041,10 @@ TEST_P(PerKeyPlacementCompactionPickerTest,
  ASSERT_OK(universal_compaction_picker.GetCompactionInputsFromFileNumbers(
      &input_files, &input_set, vstorage_.get(), comp_options));

-  std::unique_ptr<Compaction> comp1(universal_compaction_picker.CompactFiles(
-      comp_options, input_files, 5, vstorage_.get(), mutable_cf_options_,
-      mutable_db_options_, 0));
+  std::unique_ptr<Compaction> comp1(
+      universal_compaction_picker.PickCompactionForCompactFiles(
+          comp_options, input_files, 5, vstorage_.get(), mutable_cf_options_,
+          mutable_db_options_, 0));

  input_set.clear();
  input_files.clear();
@@ -4077,9 +4089,10 @@ TEST_P(PerKeyPlacementCompactionPickerTest, NormalCompactionOverlapUniversal) {
  ASSERT_OK(universal_compaction_picker.GetCompactionInputsFromFileNumbers(
      &input_files, &input_set, vstorage_.get(), comp_options));

-  std::unique_ptr<Compaction> comp1(universal_compaction_picker.CompactFiles(
-      comp_options, input_files, 6, vstorage_.get(), mutable_cf_options_,
-      mutable_db_options_, 0));
+  std::unique_ptr<Compaction> comp1(
+      universal_compaction_picker.PickCompactionForCompactFiles(
+          comp_options, input_files, 6, vstorage_.get(), mutable_cf_options_,
+          mutable_db_options_, 0));

  input_set.clear();
  input_files.clear();
@@ -4125,9 +4138,10 @@ TEST_P(PerKeyPlacementCompactionPickerTest, ProximalOverlapUniversal) {
  ASSERT_OK(universal_compaction_picker.GetCompactionInputsFromFileNumbers(
      &input_files, &input_set, vstorage_.get(), comp_options));

-  std::unique_ptr<Compaction> comp1(universal_compaction_picker.CompactFiles(
-      comp_options, input_files, 6, vstorage_.get(), mutable_cf_options_,
-      mutable_db_options_, 0));
+  std::unique_ptr<Compaction> comp1(
+      universal_compaction_picker.PickCompactionForCompactFiles(
+          comp_options, input_files, 6, vstorage_.get(), mutable_cf_options_,
+          mutable_db_options_, 0));

  input_set.clear();
  input_files.clear();
@@ -4176,9 +4190,10 @@ TEST_P(PerKeyPlacementCompactionPickerTest, LastLevelOnlyOverlapUniversal) {
  ASSERT_OK(universal_compaction_picker.GetCompactionInputsFromFileNumbers(
      &input_files, &input_set, vstorage_.get(), comp_options));

-  std::unique_ptr<Compaction> comp1(universal_compaction_picker.CompactFiles(
-      comp_options, input_files, 6, vstorage_.get(), mutable_cf_options_,
-      mutable_db_options_, 0));
+  std::unique_ptr<Compaction> comp1(
+      universal_compaction_picker.PickCompactionForCompactFiles(
+          comp_options, input_files, 6, vstorage_.get(), mutable_cf_options_,
+          mutable_db_options_, 0));

  // cannot compact file 41 if the preclude_last_level feature is on, otherwise
  // compact file 41 is okay.
@@ -4234,9 +4249,10 @@ TEST_P(PerKeyPlacementCompactionPickerTest,
  ASSERT_OK(universal_compaction_picker.GetCompactionInputsFromFileNumbers(
      &input_files, &input_set, vstorage_.get(), comp_options));

-  std::unique_ptr<Compaction> comp1(universal_compaction_picker.CompactFiles(
-      comp_options, input_files, 6, vstorage_.get(), mutable_cf_options_,
-      mutable_db_options_, 0));
+  std::unique_ptr<Compaction> comp1(
+      universal_compaction_picker.PickCompactionForCompactFiles(
+          comp_options, input_files, 6, vstorage_.get(), mutable_cf_options_,
+          mutable_db_options_, 0));

  ASSERT_TRUE(comp1);
  ASSERT_EQ(comp1->GetProximalLevel(), Compaction::kInvalidLevel);
@@ -4252,9 +4268,10 @@ TEST_P(PerKeyPlacementCompactionPickerTest,
  ASSERT_FALSE(universal_compaction_picker.FilesRangeOverlapWithCompaction(
      input_files, 5, Compaction::kInvalidLevel));

-  std::unique_ptr<Compaction> comp2(universal_compaction_picker.CompactFiles(
-      comp_options, input_files, 5, vstorage_.get(), mutable_cf_options_,
-      mutable_db_options_, 0));
+  std::unique_ptr<Compaction> comp2(
+      universal_compaction_picker.PickCompactionForCompactFiles(
+          comp_options, input_files, 5, vstorage_.get(), mutable_cf_options_,
+          mutable_db_options_, 0));
  ASSERT_TRUE(comp2);
  ASSERT_EQ(Compaction::kInvalidLevel, comp2->GetProximalLevel());
 }
@@ -4290,9 +4307,10 @@ TEST_P(PerKeyPlacementCompactionPickerTest,
  ASSERT_OK(universal_compaction_picker.GetCompactionInputsFromFileNumbers(
      &input_files, &input_set, vstorage_.get(), comp_options));

-  std::unique_ptr<Compaction> comp1(universal_compaction_picker.CompactFiles(
-      comp_options, input_files, 5, vstorage_.get(), mutable_cf_options_,
-      mutable_db_options_, 0));
+  std::unique_ptr<Compaction> comp1(
+      universal_compaction_picker.PickCompactionForCompactFiles(
+          comp_options, input_files, 5, vstorage_.get(), mutable_cf_options_,
+          mutable_db_options_, 0));

  ASSERT_TRUE(comp1);
  ASSERT_EQ(comp1->GetProximalLevel(), Compaction::kInvalidLevel);
@@ -4310,9 +4328,10 @@ TEST_P(PerKeyPlacementCompactionPickerTest,
                    vstorage_.get(), mutable_cf_options_, ioptions_, 6, 6)));

  if (!enable_per_key_placement_) {
-    std::unique_ptr<Compaction> comp2(universal_compaction_picker.CompactFiles(
-        comp_options, input_files, 6, vstorage_.get(), mutable_cf_options_,
-        mutable_db_options_, 0));
+    std::unique_ptr<Compaction> comp2(
+        universal_compaction_picker.PickCompactionForCompactFiles(
+            comp_options, input_files, 6, vstorage_.get(), mutable_cf_options_,
+            mutable_db_options_, 0));
    ASSERT_TRUE(comp2);
    ASSERT_EQ(Compaction::kInvalidLevel, comp2->GetProximalLevel());
  }
@@ -4350,9 +4369,10 @@ TEST_P(PerKeyPlacementCompactionPickerTest,
  ASSERT_OK(universal_compaction_picker.GetCompactionInputsFromFileNumbers(
      &input_files, &input_set, vstorage_.get(), comp_options));

-  std::unique_ptr<Compaction> comp1(universal_compaction_picker.CompactFiles(
-      comp_options, input_files, 5, vstorage_.get(), mutable_cf_options_,
-      mutable_db_options_, 0));
+  std::unique_ptr<Compaction> comp1(
+      universal_compaction_picker.PickCompactionForCompactFiles(
+          comp_options, input_files, 5, vstorage_.get(), mutable_cf_options_,
+          mutable_db_options_, 0));

  ASSERT_TRUE(comp1);
  ASSERT_EQ(comp1->GetProximalLevel(), Compaction::kInvalidLevel);
@@ -4370,9 +4390,10 @@ TEST_P(PerKeyPlacementCompactionPickerTest,
                                        ioptions_, 6, 6)));

  // 2 compactions can be run in parallel
-  std::unique_ptr<Compaction> comp2(universal_compaction_picker.CompactFiles(
-      comp_options, input_files, 6, vstorage_.get(), mutable_cf_options_,
-      mutable_db_options_, 0));
+  std::unique_ptr<Compaction> comp2(
+      universal_compaction_picker.PickCompactionForCompactFiles(
+          comp_options, input_files, 6, vstorage_.get(), mutable_cf_options_,
+          mutable_db_options_, 0));
  ASSERT_TRUE(comp2);
  if (enable_per_key_placement_) {
    ASSERT_NE(Compaction::kInvalidLevel, comp2->GetProximalLevel());
@@ -48,7 +48,9 @@ class UniversalCompactionBuilder {
        vstorage_(vstorage),
        picker_(picker),
        log_buffer_(log_buffer),
-        require_max_output_level_(require_max_output_level) {
+        require_max_output_level_(require_max_output_level),
+        allow_ingest_behind_(ioptions.cf_allow_ingest_behind ||
+                             ioptions.allow_ingest_behind) {
    assert(icmp_);
    const auto* ucmp = icmp_->user_comparator();
    assert(ucmp);
@@ -422,8 +424,7 @@ class UniversalCompactionBuilder {
  bool MeetsOutputLevelRequirements(int output_level) const {
    return !require_max_output_level_ ||
           Compaction::OutputToNonZeroMaxOutputLevel(
-               output_level,
-               vstorage_->MaxOutputLevel(ioptions_.allow_ingest_behind));
+               output_level, vstorage_->MaxOutputLevel(allow_ingest_behind_));
  }

  const ImmutableOptions& ioptions_;
@@ -437,7 +438,6 @@ class UniversalCompactionBuilder {
  VersionStorageInfo* vstorage_;
  UniversalCompactionPicker* picker_;
  LogBuffer* log_buffer_;
-  bool require_max_output_level_;
  // Optional earliest snapshot at time of compaction picking. This is only
  // provided if the column family doesn't enable user-defined timestamps.
  // And this information is only passed to `Compaction` picked by deletion
@@ -448,6 +448,8 @@ class UniversalCompactionBuilder {
  // marked for compaction. This is only populated when snapshot info is
  // populated.
  std::map<uint64_t, size_t> file_marked_for_compaction_to_sorted_run_index_;
+  bool require_max_output_level_;
+  bool allow_ingest_behind_;

  std::vector<UniversalCompactionBuilder::SortedRun> CalculateSortedRuns(
      const VersionStorageInfo& vstorage, int last_level,
@@ -733,8 +735,7 @@ bool UniversalCompactionBuilder::ShouldSkipMarkedFile(
 Compaction* UniversalCompactionBuilder::PickCompaction() {
  const int kLevel0 = 0;
  score_ = vstorage_->CompactionScore(kLevel0);
-  const int max_output_level =
-      vstorage_->MaxOutputLevel(ioptions_.allow_ingest_behind);
+  const int max_output_level = vstorage_->MaxOutputLevel(allow_ingest_behind_);
  const int file_num_compaction_trigger =
      mutable_cf_options_.level0_file_num_compaction_trigger;
  const unsigned int ratio =
@@ -781,8 +782,7 @@ Compaction* UniversalCompactionBuilder::PickCompaction() {
        "UniversalCompactionBuilder::PickCompaction:Return", nullptr);
    return nullptr;
  }
-  assert(c->output_level() <=
-         vstorage_->MaxOutputLevel(ioptions_.allow_ingest_behind));
+  assert(c->output_level() <= vstorage_->MaxOutputLevel(allow_ingest_behind_));
  assert(MeetsOutputLevelRequirements(c->output_level()));

  if (mutable_cf_options_.compaction_options_universal.allow_trivial_move ==
@@ -1024,8 +1024,7 @@ Compaction* UniversalCompactionBuilder::PickCompactionToReduceSortedRuns(
  int start_level = sorted_runs_[start_index].level;
  int output_level;
  // last level is reserved for the files ingested behind
-  int max_output_level =
-      vstorage_->MaxOutputLevel(ioptions_.allow_ingest_behind);
+  int max_output_level = vstorage_->MaxOutputLevel(allow_ingest_behind_);
  if (first_index_after == sorted_runs_.size()) {
    output_level = max_output_level;
  } else if (sorted_runs_[first_index_after].level == 0) {
@@ -1097,10 +1096,9 @@ Compaction* UniversalCompactionBuilder::PickCompactionToReduceSortedRuns(
                        mutable_cf_options_.default_write_temperature,
                        /* max_subcompactions */ 0, grandparents,
                        /* earliest_snapshot */ std::nullopt,
-                        /* snapshot_checker */ nullptr,
-                        /* is manual */ false, /* trim_ts */ "", score_,
-                        false /* deletion_compaction */,
-                        /* l0_files_might_overlap */ true, compaction_reason);
+                        /* snapshot_checker */ nullptr, compaction_reason,
+                        /* trim_ts */ "", score_,
+                        /* l0_files_might_overlap */ true);
 }

 // Look at overall size amplification. If size amplification
@@ -1447,10 +1445,9 @@ Compaction* UniversalCompactionBuilder::PickIncrementalForReduceSizeAmp(
      /* max_subcompactions */ 0, /* grandparents */ {},
      /* earliest_snapshot */ std::nullopt,
      /* snapshot_checker */ nullptr,
-      /* is manual */ false,
-      /* trim_ts */ "", score_, false /* deletion_compaction */,
-      /* l0_files_might_overlap */ true,
-      CompactionReason::kUniversalSizeAmplification);
+      CompactionReason::kUniversalSizeAmplification,
+      /* trim_ts */ "", score_,
+      /* l0_files_might_overlap */ true);
 }

 // Pick files marked for compaction. Typically, files are marked by
@@ -1519,8 +1516,7 @@ Compaction* UniversalCompactionBuilder::PickDeleteTriggeredCompaction() {
      return nullptr;
    }

-    int max_output_level =
-        vstorage_->MaxOutputLevel(ioptions_.allow_ingest_behind);
+    int max_output_level = vstorage_->MaxOutputLevel(allow_ingest_behind_);
    // Pick the first non-empty level after the start_level
    for (output_level = start_level + 1; output_level <= max_output_level;
         output_level++) {
@@ -1600,11 +1596,9 @@ Compaction* UniversalCompactionBuilder::PickDeleteTriggeredCompaction() {
      GetCompressionOptions(mutable_cf_options_, vstorage_, output_level),
      mutable_cf_options_.default_write_temperature,
      /* max_subcompactions */ 0, grandparents, earliest_snapshot_,
-      snapshot_checker_,
-      /* is manual */ false,
-      /* trim_ts */ "", score_, false /* deletion_compaction */,
-      /* l0_files_might_overlap */ true,
-      CompactionReason::kFilesMarkedForCompaction);
+      snapshot_checker_, CompactionReason::kFilesMarkedForCompaction,
+      /* trim_ts */ "", score_,
+      /* l0_files_might_overlap */ true);
 }

 Compaction* UniversalCompactionBuilder::PickCompactionToOldest(
@@ -1625,8 +1619,7 @@ Compaction* UniversalCompactionBuilder::PickCompactionWithSortedRunRange(
  uint32_t path_id =
      GetPathId(ioptions_, mutable_cf_options_, estimated_total_size);
  int start_level = sorted_runs_[start_index].level;
-  int max_output_level =
-      vstorage_->MaxOutputLevel(ioptions_.allow_ingest_behind);
+  int max_output_level = vstorage_->MaxOutputLevel(allow_ingest_behind_);
  std::vector<CompactionInputFiles> inputs(max_output_level + 1);
  for (size_t i = 0; i < inputs.size(); ++i) {
    inputs[i].level = start_level + static_cast<int>(i);
@@ -1700,10 +1693,9 @@ Compaction* UniversalCompactionBuilder::PickCompactionWithSortedRunRange(
      mutable_cf_options_.default_write_temperature,
      /* max_subcompactions */ 0, /* grandparents */ {},
      /* earliest_snapshot */ std::nullopt,
-      /* snapshot_checker */ nullptr,
-      /* is manual */ false,
-      /* trim_ts */ "", score_, false /* deletion_compaction */,
-      /* l0_files_might_overlap */ true, compaction_reason);
+      /* snapshot_checker */ nullptr, compaction_reason,
+      /* trim_ts */ "", score_,
+      /* l0_files_might_overlap */ true);
 }

 Compaction* UniversalCompactionBuilder::PickPeriodicCompaction() {
@@ -461,6 +461,97 @@ TEST_F(CompactionServiceTest, ManualCompaction) {
  ASSERT_EQ(handles_[1]->GetName(), info.cf_name);
 }

+TEST_F(CompactionServiceTest, StandaloneDeleteRangeTombstoneOptimization) {
+  Options options = CurrentOptions();
+  options.compaction_style = CompactionStyle::kCompactionStyleUniversal;
+  ReopenWithCompactionService(&options);
+
+  size_t num_files_after_filtered = 0;
+  SyncPoint::GetInstance()->SetCallBack(
+      "VersionSet::MakeInputIterator:NewCompactionMergingIterator",
+      [&](void* arg) {
+        num_files_after_filtered = *static_cast<size_t*>(arg);
+      });
+
+  SyncPoint::GetInstance()->EnableProcessing();
+
+  std::vector<std::string> files;
+  {
+    // Writes first version of data in range partitioned files.
+    SstFileWriter sst_file_writer(EnvOptions(), options);
+    std::string file1 = dbname_ + "file1.sst";
+    ASSERT_OK(sst_file_writer.Open(file1));
+    ASSERT_OK(sst_file_writer.Put("a", "a1"));
+    ASSERT_OK(sst_file_writer.Put("b", "b1"));
+    ExternalSstFileInfo file1_info;
+    ASSERT_OK(sst_file_writer.Finish(&file1_info));
+    files.push_back(std::move(file1));
+
+    std::string file2 = dbname_ + "file2.sst";
+    ASSERT_OK(sst_file_writer.Open(file2));
+    ASSERT_OK(sst_file_writer.Put("x", "x1"));
+    ASSERT_OK(sst_file_writer.Put("y", "y1"));
+    ExternalSstFileInfo file2_info;
+    ASSERT_OK(sst_file_writer.Finish(&file2_info));
+    files.push_back(std::move(file2));
+  }
+
+  IngestExternalFileOptions ifo;
+  ASSERT_OK(db_->IngestExternalFile(files, ifo));
+  ASSERT_EQ(Get("a"), "a1");
+  ASSERT_EQ(Get("b"), "b1");
+  ASSERT_EQ(Get("x"), "x1");
+  ASSERT_EQ(Get("y"), "y1");
+  ASSERT_EQ(2, NumTableFilesAtLevel(6));
+
+  auto my_cs = GetCompactionService();
+  uint64_t comp_num = my_cs->GetCompactionNum();
+
+  {
+    // Atomically delete old version of data with one range delete file.
+    // And a new batch of range partitioned files with new version of data.
+    files.clear();
+    SstFileWriter sst_file_writer(EnvOptions(), options);
+    std::string file2 = dbname_ + "file2.sst";
+    ASSERT_OK(sst_file_writer.Open(file2));
+    ASSERT_OK(sst_file_writer.DeleteRange("a", "z"));
+    ExternalSstFileInfo file2_info;
+    ASSERT_OK(sst_file_writer.Finish(&file2_info));
+    files.push_back(std::move(file2));
+
+    std::string file3 = dbname_ + "file3.sst";
+    ASSERT_OK(sst_file_writer.Open(file3));
+    ASSERT_OK(sst_file_writer.Put("a", "a2"));
+    ASSERT_OK(sst_file_writer.Put("b", "b2"));
+    ExternalSstFileInfo file3_info;
+    ASSERT_OK(sst_file_writer.Finish(&file3_info));
+    files.push_back(std::move(file3));
+
+    std::string file4 = dbname_ + "file4.sst";
+    ASSERT_OK(sst_file_writer.Open(file4));
+    ASSERT_OK(sst_file_writer.Put("x", "x2"));
+    ASSERT_OK(sst_file_writer.Put("y", "y2"));
+    ExternalSstFileInfo file4_info;
+    ASSERT_OK(sst_file_writer.Finish(&file4_info));
+    files.push_back(std::move(file4));
+  }
+
+  ASSERT_OK(db_->IngestExternalFile(files, ifo));
+  ASSERT_OK(db_->WaitForCompact(WaitForCompactOptions()));
+  ASSERT_GE(my_cs->GetCompactionNum(), comp_num + 1);
+
+  CompactionServiceResult result;
+  my_cs->GetResult(&result);
+  ASSERT_OK(result.status);
+  ASSERT_TRUE(result.stats.is_manual_compaction);
+  ASSERT_TRUE(result.stats.is_remote_compaction);
+
+  ASSERT_EQ(num_files_after_filtered, 1);
+
+  Close();
+  SyncPoint::GetInstance()->DisableProcessing();
+}
+
 TEST_F(CompactionServiceTest, CompactionOutputFileIOError) {
  Options options = CurrentOptions();
  options.disable_auto_compactions = true;
@@ -793,6 +884,79 @@ TEST_F(CompactionServiceTest, VerifyInputRecordCount) {
  SyncPoint::GetInstance()->ClearAllCallBacks();
 }

+TEST_F(CompactionServiceTest, EmptyResult) {
+  Options options = CurrentOptions();
+  options.disable_auto_compactions = true;
+  ReopenWithCompactionService(&options);
+  GenerateTestData();
+
+  auto my_cs = GetCompactionService();
+
+  uint64_t comp_num = my_cs->GetCompactionNum();
+  ASSERT_OK(db_->CompactRange(CompactRangeOptions(), nullptr, nullptr));
+  ASSERT_GE(my_cs->GetCompactionNum(), comp_num + 1);
+
+  // Delete range to cover entire range
+  ASSERT_OK(db_->DeleteRange(WriteOptions(), "key", "keyz"));
+  ASSERT_OK(Flush());
+
+  // In this unit test, both remote compaction and primary db instance are
+  // running in the same process, so NewFileNumber will never have a collision.
+  // In the real-world remote compactions, when the compaction is indeed running
+  // in another process, this is not going to be the case.
+  // To simulate the SST file with the same name created in the tmp directory,
+  // override the file number in remote compaction to re-use old SST file
+  // number.
+  bool need_to_override_file_number = false;
+  SyncPoint::GetInstance()->SetCallBack(
+      "DBImplSecondary::OpenAndCompact::BeforeLoadingOptions:0",
+      [&](void*) { need_to_override_file_number = true; });
+
+  SyncPoint::GetInstance()->SetCallBack(
+      "CompactionJob::OpenCompactionOutputFile::NewFileNumber",
+      [&](void* file_number) {
+        if (need_to_override_file_number) {
+          auto n = static_cast<uint64_t*>(file_number);
+          ColumnFamilyMetaData cf_meta;
+          db_->GetColumnFamilyMetaData(&cf_meta);
+          for (const auto& level : cf_meta.levels) {
+            for (const auto& file : level.files) {
+              // Use one of the existing file name
+              *n = test::GetFileNumber(file.name);
+              need_to_override_file_number = false;
+              return;
+            }
+          }
+        }
+      });
+
+  // Inject failure, so that the remote compaction fails after
+  // ProcessKeyValueCompaction()
+  SyncPoint::GetInstance()->SetCallBack(
+      "DBImplSecondary::CompactWithoutInstallation::End", [&](void* status) {
+        // override job status
+        auto s = static_cast<Status*>(status);
+        *s = Status::Aborted("MyTestCompactionService failed to compact!");
+      });
+  SyncPoint::GetInstance()->EnableProcessing();
+
+  // Compaction should fail and SST files in the primary db should exist
+  {
+    ASSERT_NOK(db_->CompactRange(CompactRangeOptions(), nullptr, nullptr));
+    ColumnFamilyMetaData meta;
+    db_->GetColumnFamilyMetaData(&meta);
+    for (const auto& level : meta.levels) {
+      for (const auto& file : level.files) {
+        std::string fname = file.db_path + "/" + file.name;
+        ASSERT_OK(db_->GetEnv()->FileExists(fname));
+      }
+    }
+  }
+  Close();
+  SyncPoint::GetInstance()->DisableProcessing();
+  SyncPoint::GetInstance()->ClearAllCallBacks();
+}
+
 TEST_F(CompactionServiceTest, CorruptedOutput) {
  Options options = CurrentOptions();
  options.disable_auto_compactions = true;
@@ -556,6 +556,74 @@ TEST_F(CorruptionTest, TableFileFooterNotMagic) {
  ASSERT_TRUE(s.ToString().find(".sst") != std::string::npos);
 }

+TEST_F(CorruptionTest, DBOpenWithWrongFileSize) {
+  // Validate that when paranoid flag is true, DB::Open() fails if one of the
+  // file corrupted. Validate that when paranoid flag is false, DB::Open()
+  // succeed if one of the file corrupted, and the healthy file is readable.
+  CloseDb();
+
+  const std::string test_cf_name = "test_cf";
+  std::vector<ColumnFamilyDescriptor> cf_descs;
+  cf_descs.emplace_back(kDefaultColumnFamilyName, ColumnFamilyOptions());
+  cf_descs.emplace_back(test_cf_name, ColumnFamilyOptions());
+
+  {
+    options_.create_missing_column_families = true;
+    std::vector<ColumnFamilyHandle*> cfhs;
+    ASSERT_OK(DB::Open(options_, dbname_, cf_descs, &cfhs, &db_));
+    assert(db_ != nullptr);  // suppress false clang-analyze report
+
+    ASSERT_OK(db_->Put(WriteOptions(), cfhs[0], "k", "v"));
+    ASSERT_OK(db_->Put(WriteOptions(), cfhs[1], "k1", "v1"));
+    ASSERT_OK(db_->Put(WriteOptions(), cfhs[0], "k2", "v2"));
+    for (auto* cfh : cfhs) {
+      delete cfh;
+    }
+    DBImpl* dbi = static_cast_with_check<DBImpl>(db_);
+    ASSERT_OK(dbi->TEST_FlushMemTable());
+
+    // ********************************************
+    // Corrupt the file by making the file bigger
+    std::vector<LiveFileMetaData> metadata;
+    db_->GetLiveFilesMetaData(&metadata);
+    std::string filename = dbname_ + metadata[0].name;
+    const auto& fs = options_.env->GetFileSystem();
+    {
+      std::unique_ptr<FSWritableFile> f;
+      ASSERT_OK(fs->ReopenWritableFile(filename, FileOptions(), &f, nullptr));
+      ASSERT_OK(f->Append("blahblah", IOOptions(), nullptr));
+      ASSERT_OK(f->Close(IOOptions(), nullptr));
+    }
+    CloseDb();
+  }
+
+  // DB failed to open due to one of the file is corrupted, as paranoid flag is
+  // true
+  options_.paranoid_checks = true;
+  std::vector<ColumnFamilyHandle*> cfhs;
+  auto s = DB::Open(options_, dbname_, cf_descs, &cfhs, &db_);
+  ASSERT_TRUE(s.IsCorruption());
+  ASSERT_TRUE(s.ToString().find("file size mismatch") != std::string::npos);
+
+  // DB opened successfully, as paranoid flag is false, validate the one that is
+  // healthy is still accessible
+  options_.paranoid_checks = false;
+  ASSERT_OK(DB::Open(options_, dbname_, cf_descs, &cfhs, &db_));
+  assert(db_ != nullptr);  // suppress false clang-analyze report
+
+  std::string v;
+  ASSERT_OK(db_->Get(ReadOptions(), cfhs[1], "k1", &v));
+  ASSERT_EQ(v, "v1");
+
+  // Validate the default column family is corrupted
+  Check(0, 0);
+  s = db_->Get(ReadOptions(), cfhs[0], "k1", &v);
+  ASSERT_TRUE(s.IsCorruption());
+
+  delete cfhs[1];
+  delete cfhs[0];
+}
+
 TEST_F(CorruptionTest, TableFileWrongSize) {
  Build(100);
  DBImpl* dbi = static_cast_with_check<DBImpl>(db_);
@@ -579,13 +647,16 @@ TEST_F(CorruptionTest, TableFileWrongSize) {
  // DB actually accepts this without paranoid checks, relying on size
  // recorded in manifest to locate the SST footer.
  options_.paranoid_checks = false;
-  options_.skip_checking_sst_file_sizes_on_db_open = false;
  Reopen();
-  Check(100, 100);
+  // As footer could not be extraced, file is completely unreadable
+  Check(0, 0);
+  std::string v;
+  auto s = db_->Get(ReadOptions(), "k1", &v);
+  ASSERT_TRUE(s.IsCorruption());

  // But reports the issue with paranoid checks
  options_.paranoid_checks = true;
-  Status s = TryReopen();
+  s = TryReopen();
  ASSERT_TRUE(s.IsCorruption());
  ASSERT_TRUE(s.ToString().find("file size mismatch") != std::string::npos);

@@ -74,6 +74,43 @@ class CompactionStatsCollector : public EventListener {
  std::vector<std::atomic<int>> compaction_completed_;
 };

+class DeletionTriggeredCompactionWithMinFileSizeTestListener
+    : public EventListener {
+ public:
+  explicit DeletionTriggeredCompactionWithMinFileSizeTestListener(
+      uint64_t min_file_size)
+      : min_file_size_(min_file_size) {}
+
+  void OnCompactionBegin(DB* db, const CompactionJobInfo& ci) override {
+    ASSERT_EQ(ci.compaction_reason,
+              CompactionReason::kFilesMarkedForCompaction);
+
+    auto env = db->GetEnv();
+    const std::vector<DbPath>& db_paths = db->GetOptions().db_paths;
+    for (const auto& file : ci.input_file_infos) {
+      uint64_t file_size = GetSstFileSize(env, db_paths, file.file_number);
+
+      // Assert that the file size respects the minimum threshold
+      ASSERT_GE(file_size, min_file_size_);
+    }
+  }
+
+ private:
+  static uint64_t GetSstFileSize(Env* env, const std::vector<DbPath>& db_paths,
+                                 uint64_t file_number) {
+    uint32_t path_id = 0;  // since only one path
+    std::string sst_file_name = TableFileName(db_paths, file_number, path_id);
+    uint64_t file_size = 0;
+    Status s = env->GetFileSize(sst_file_name, &file_size);
+    if (!s.ok()) {
+      return 0;
+    }
+    return file_size;
+  }
+
+  uint64_t min_file_size_;
+};
+
 class DBCompactionTest : public DBTestBase {
 public:
  DBCompactionTest()
@@ -1371,6 +1408,89 @@ TEST_F(DBCompactionTest, RecoverDuringMemtableCompaction) {
  } while (ChangeOptions());
 }

+TEST_F(DBCompactionTest, CompactionWithDeletionsAndMinFileSize) {
+  const uint64_t kMinFileSize = 32 * 1024;  // 32KB
+  const int kDeletionTriggerCount = 50;
+  const int kInitialKeyCount = 100;
+  const int kAdditionalKeyCount = 50;
+  const int kValueSize = 1024;
+  const int kSmallValueSize = 512;
+  const int kSeed = 301;
+
+  Options options = CurrentOptions();
+  options.compaction_style = kCompactionStyleLevel;
+  options.write_buffer_size = 1024 * 1024;  // 1MB
+  options.level0_file_num_compaction_trigger = 100;
+
+  options.table_properties_collector_factories = {
+      NewCompactOnDeletionCollectorFactory(
+          kInitialKeyCount /* sliding window size */, kDeletionTriggerCount,
+          0.5 /* deletion ratio */, kMinFileSize)};
+  auto listener =
+      new DeletionTriggeredCompactionWithMinFileSizeTestListener(kMinFileSize);
+  options.listeners.emplace_back(listener);
+
+  DestroyAndReopen(options);
+  Random rnd(kSeed);
+
+  // Create a large file that will be subject to DTC later
+  for (int i = 0; i < kInitialKeyCount; i++) {
+    ASSERT_OK(Put(Key(i), rnd.RandomString(kValueSize)));
+  }
+  ASSERT_OK(Flush());
+
+  std::vector<LiveFileMetaData> initial_metadata;
+  db_->GetLiveFilesMetaData(&initial_metadata);
+  ASSERT_EQ(initial_metadata.size(), 1);
+
+  // Create small files that should not trigger compaction
+  ASSERT_OK(Put("small_file_key1", rnd.RandomString(kSmallValueSize)));
+  ASSERT_OK(Put("small_file_key2", rnd.RandomString(kSmallValueSize)));
+  ASSERT_OK(Flush());
+  ASSERT_OK(Delete("small_file_key1"));
+  ASSERT_OK(Flush());
+
+  // Create a file with enough deletions and size to trigger DTC
+  // Delete keys from the large file to reach deletion threshold
+  for (int i = 0; i < kDeletionTriggerCount; i++) {
+    ASSERT_OK(Delete(Key(i)));
+  }
+
+  // Add new keys to ensure the deletion file meets the min_file_size threshold
+  for (int i = kInitialKeyCount; i < kInitialKeyCount + kAdditionalKeyCount;
+       i++) {
+    ASSERT_OK(Put(Key(i), rnd.RandomString(kValueSize)));
+  }
+  ASSERT_OK(Flush());
+
+  ASSERT_OK(dbfull()->TEST_WaitForCompact());
+
+  // Verify file count after compaction
+  ASSERT_EQ(NumTableFilesAtLevel(0), 2);  // Small file and deletion file
+  ASSERT_EQ(NumTableFilesAtLevel(1), 1);  // Compacted large file
+
+  // Verify deleted keys are gone
+  for (int i = 0; i < kDeletionTriggerCount; i++) {
+    std::string value;
+    ASSERT_TRUE(db_->Get(ReadOptions(), Key(i), &value).IsNotFound());
+  }
+
+  // Verify non-deleted keys from large file are still accessible
+  for (int i = kDeletionTriggerCount; i < kInitialKeyCount; i++) {
+    std::string value;
+    ASSERT_OK(db_->Get(ReadOptions(), Key(i), &value));
+    ASSERT_EQ(value.size(), kValueSize);
+  }
+
+  // Verify new keys are accessible
+  for (int i = kInitialKeyCount; i < kInitialKeyCount + kAdditionalKeyCount;
+       i++) {
+    std::string value;
+    ASSERT_OK(db_->Get(ReadOptions(), Key(i), &value));
+    ASSERT_EQ(value.size(), kValueSize);
+  }
+}
+
 TEST_P(DBCompactionTestWithParam, TrivialMoveOneFile) {
  int32_t trivial_move = 0;
  ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->SetCallBack(
@@ -6748,7 +6868,11 @@ INSTANTIATE_TEST_CASE_P(RoundRobinSubcompactionsAgainstPressureToken,
                        RoundRobinSubcompactionsAgainstPressureToken,
                        testing::Bool());

-TEST_P(RoundRobinSubcompactionsAgainstResources, SubcompactionsUsingResources) {
+// FIXME: the test is flaky and failing the assertion
+// ASSERT_EQ(actual_reserved_threads, expected_reserved_threads);
+// It's likely a test set up issue, fix if we are to use RoubdRobin compaction.
+TEST_P(RoundRobinSubcompactionsAgainstResources,
+       DISABLED_SubcompactionsUsingResources) {
  const int kKeysPerBuffer = 200;
  Options options = CurrentOptions();
  options.num_levels = 4;
@@ -7004,6 +7128,70 @@ TEST_F(DBCompactionTest, PartialManualCompaction) {
  ASSERT_OK(dbfull()->CompactRange(cro, nullptr, nullptr));
 }

+TEST_F(DBCompactionTest, ConcurrentFIFOPickingSameFileBug) {
+  Options opts = CurrentOptions();
+  opts.compaction_style = CompactionStyle::kCompactionStyleLevel;
+  opts.num_levels = 3;
+  opts.disable_auto_compactions = true;
+  opts.max_background_jobs = 3;
+
+  DestroyAndReopen(opts);
+
+  ASSERT_OK(Put("k1", "v1"));
+  ASSERT_OK(Flush());
+
+  // Create a non-L0 SST file for multi-level FIFO size-based compaction later
+  MoveFilesToLevel(2);
+
+  Options opts_new(opts);
+  opts_new.compaction_style = CompactionStyle::kCompactionStyleFIFO;
+  opts_new.max_open_files = -1;
+  // Set a low threshold to trigger multi-level size-based compaction
+  opts_new.compaction_options_fifo.max_table_files_size = 1;
+
+  Reopen(opts_new);
+
+  const CompactRangeOptions cro;
+  const Slice begin_key("k1");
+  const Slice end_key("k2");
+
+  std::unique_ptr<port::Thread> concurrent_compaction;
+
+  bool within_first_compaction = true;
+  SyncPoint::GetInstance()->SetCallBack(
+      "VersionSet::LogAndApply:WriteManifestStart", [&](void* /*arg*/) {
+        if (!within_first_compaction) {
+          return;
+        }
+        within_first_compaction = false;
+
+        // To allow the second/concurrent compaction to still see the non-L0
+        // SST file and coerce the bug of picking that file
+        SyncPoint::GetInstance()->LoadDependency({
+            {"DBImpl::BackgroundCompaction:BeforeCompaction",
+             "VersionSet::LogAndApply:WriteManifest"},
+        });
+
+        concurrent_compaction.reset(new port::Thread([&]() {
+          // Before the fix, the second CompactRange() will either fail the
+          // assertion of double file picking `being_compacted !=
+          // inputs_[i][j]->being_compacted` in debug mode or cause LSM shape
+          // corruption "Cannot delete table file XXX from level 2 since it is
+          // not in the LSM tree" in release mode
+          Status s = db_->CompactRange(cro, &begin_key, &end_key);
+          ASSERT_OK(s);
+        }));
+      });
+
+  SyncPoint::GetInstance()->EnableProcessing();
+  Status s = db_->CompactRange(cro, &begin_key, &end_key);
+  SyncPoint::GetInstance()->DisableProcessing();
+
+  ASSERT_OK(s);
+
+  concurrent_compaction->join();
+}
+
 TEST_F(DBCompactionTest, ManualCompactionFailsInReadOnlyMode) {
  // Regression test for bug where manual compaction hangs forever when the DB
  // is in read-only mode. Verify it now at least returns, despite failing.
@@ -11093,6 +11281,59 @@ TEST_F(DBCompactionTest, RecordNewestKeyTimeForTtlCompaction) {
  ASSERT_OK(dbfull()->TEST_WaitForCompact());
  ASSERT_EQ(NumTableFilesAtLevel(0), 0);
 }
+
+class PeriodicCompactionListener : public EventListener {
+ public:
+  explicit PeriodicCompactionListener() {}
+  void OnCompactionBegin(DB* /*db*/, const CompactionJobInfo& ci) override {
+    if (ci.compaction_reason == CompactionReason::kPeriodicCompaction) {
+      ++num_periodic_compactions;
+    }
+  }
+
+  std::atomic<int> num_periodic_compactions = 0;
+};
+
+TEST_F(DBCompactionTest, PeriodicTask) {
+  // Tests that when no trigger event is fired (flush/compaction/setoptions),
+  // periodic compaction is still triggered by a scheduled periodic function.
+  auto mock_clock = std::make_shared<MockSystemClock>(env_->GetSystemClock());
+  mock_clock->SetCurrentTime(100);
+  mock_clock->InstallTimedWaitFixCallback();
+  auto mock_env = std::make_unique<CompositeEnvWrapper>(env_, mock_clock);
+  SyncPoint::GetInstance()->SetCallBack(
+      "DBImpl::StartPeriodicTaskScheduler:Init", [&](void* arg) {
+        auto periodic_task_scheduler_ptr =
+            static_cast<PeriodicTaskScheduler*>(arg);
+        periodic_task_scheduler_ptr->TEST_OverrideTimer(mock_clock.get());
+      });
+
+  Options options;
+  options.env = mock_env.get();
+  options.compaction_style = kCompactionStyleUniversal;
+  options.statistics = CreateDBStatistics();
+  int kPeriodicCompactionSeconds = 7 * 24 * 60 * 60;  // 1 week
+  options.periodic_compaction_seconds = kPeriodicCompactionSeconds;
+  options.num_levels = 50;
+  auto listener = std::make_shared<PeriodicCompactionListener>();
+  options.listeners.push_back(listener);
+  ASSERT_OK(TryReopen(options));
+
+  Random* rnd = Random::GetTLSInstance();
+  for (int k = 0; k < 10; ++k) {
+    ASSERT_OK(Put(Key(k), rnd->RandomString(100)));
+  }
+  ASSERT_OK(Flush());
+  ASSERT_OK(db_->CompactRange({}, nullptr, nullptr));
+  ASSERT_EQ(1, NumTableFilesAtLevel(49));
+
+  dbfull()->TEST_WaitForPeriodicTaskRun(
+      [&] { mock_clock->MockSleepForSeconds(kPeriodicCompactionSeconds + 1); });
+  ASSERT_OK(db_->WaitForCompact({}));
+
+  ASSERT_EQ(listener->num_periodic_compactions, 1);
+  Close();
+}
 }  // namespace ROCKSDB_NAMESPACE

 int main(int argc, char** argv) {
@@ -17,9 +17,10 @@ class DBEncryptionTest : public DBTestBase {
 public:
  DBEncryptionTest()
      : DBTestBase("db_encryption_test", /*env_do_fsync=*/true) {}
-  Env* GetTargetEnv() {
+  Env* GetNonEncryptedEnv() {
    if (encrypted_env_ != nullptr) {
-      return (static_cast<EnvWrapper*>(encrypted_env_))->target();
+      return (static_cast_with_check<CompositeEnvWrapper>(encrypted_env_))
+          ->env_target();
    } else {
      return env_;
    }
@@ -38,7 +39,7 @@ TEST_F(DBEncryptionTest, CheckEncrypted) {
  auto status = env_->GetChildren(dbname_, &fileNames);
  ASSERT_OK(status);

-  Env* target = GetTargetEnv();
+  Env* target = GetNonEncryptedEnv();
  int hits = 0;
  for (auto it = fileNames.begin(); it != fileNames.end(); ++it) {
    if (*it == "LOCK") {
@@ -89,7 +90,7 @@ TEST_F(DBEncryptionTest, CheckEncrypted) {
 }

 TEST_F(DBEncryptionTest, ReadEmptyFile) {
-  auto defaultEnv = GetTargetEnv();
+  auto defaultEnv = GetNonEncryptedEnv();

  // create empty file for reading it back in later
  auto envOptions = EnvOptions(CurrentOptions());
@@ -116,6 +117,40 @@ TEST_F(DBEncryptionTest, ReadEmptyFile) {
  ASSERT_TRUE(data.empty());
 }

+TEST_F(DBEncryptionTest, NotSupportedGetFileSize) {
+  // Validate envrypted env does not support GetFileSize.
+  // The goal of the test is to validate the encrypted env/fs does not support
+  // GetFileSize API on FSRandomAccessFile interface.
+  // This test combined with the rest of the integration tests validate that
+  // the new API GetFileSize on FSRandomAccessFile interface is not required to
+  // be supported for database to work properly.
+  // The GetFileSize API is used in ReadFooterFromFile() API to get the file
+  // size. When GetFileSize API is not supported, the ReadFooterFromFile() API
+  // will use FileSystem GetFileSize API as fallback. Refer to the
+  // EncryptedRandomAccessFile class definition for more details.
+  if (!encrypted_env_) {
+    return;
+  }
+
+  auto fs = encrypted_env_->GetFileSystem();
+
+  // create empty file for reading it back in later
+  auto filePath = dbname_ + "/empty.empty";
+
+  // Create empty file
+  CreateFile(fs.get(), filePath, "", false);
+
+  // Open it for reading footer
+  std::unique_ptr<FSRandomAccessFile> randomAccessFile;
+  auto status = fs->NewRandomAccessFile(filePath, FileOptions(),
+                                        &randomAccessFile, nullptr);
+  ASSERT_OK(status);
+
+  uint64_t fileSize;
+  status = randomAccessFile->GetFileSize(&fileSize);
+  ASSERT_TRUE(status.IsNotSupported());
+}
+
 }  // namespace ROCKSDB_NAMESPACE

 int main(int argc, char** argv) {
@@ -253,6 +253,9 @@ DBImpl::DBImpl(const DBOptions& options, const std::string& dbname,
  periodic_task_functions_.emplace(
      PeriodicTaskType::kRecordSeqnoTime,
      [this]() { this->RecordSeqnoToTimeMapping(); });
+  periodic_task_functions_.emplace(
+      PeriodicTaskType::kTriggerCompaction,
+      [this]() { this->TriggerPeriodicCompaction(); });

  versions_.reset(new VersionSet(
      dbname_, &immutable_db_options_, file_options_, table_cache_.get(),
@@ -787,7 +790,8 @@ Status DBImpl::StartPeriodicTaskScheduler() {
    Status s = periodic_task_scheduler_.Register(
        PeriodicTaskType::kDumpStats,
        periodic_task_functions_.at(PeriodicTaskType::kDumpStats),
-        mutable_db_options_.stats_dump_period_sec);
+        mutable_db_options_.stats_dump_period_sec,
+        /*run_immediately=*/true);
    if (!s.ok()) {
      return s;
    }
@@ -796,7 +800,8 @@ Status DBImpl::StartPeriodicTaskScheduler() {
    Status s = periodic_task_scheduler_.Register(
        PeriodicTaskType::kPersistStats,
        periodic_task_functions_.at(PeriodicTaskType::kPersistStats),
-        mutable_db_options_.stats_persist_period_sec);
+        mutable_db_options_.stats_persist_period_sec,
+        /*run_immediately=*/true);
    if (!s.ok()) {
      return s;
    }
@@ -804,7 +809,15 @@ Status DBImpl::StartPeriodicTaskScheduler() {

  Status s = periodic_task_scheduler_.Register(
      PeriodicTaskType::kFlushInfoLog,
-      periodic_task_functions_.at(PeriodicTaskType::kFlushInfoLog));
+      periodic_task_functions_.at(PeriodicTaskType::kFlushInfoLog),
+      /*run_immediately=*/true);
+
+  if (s.ok()) {
+    s = periodic_task_scheduler_.Register(
+        PeriodicTaskType::kTriggerCompaction,
+        periodic_task_functions_.at(PeriodicTaskType::kTriggerCompaction),
+        /*run_immediately=*/false);
+  }

  return s;
 }
@@ -855,7 +868,7 @@ Status DBImpl::RegisterRecordSeqnoTimeWorker() {
    s = periodic_task_scheduler_.Register(
        PeriodicTaskType::kRecordSeqnoTime,
        periodic_task_functions_.at(PeriodicTaskType::kRecordSeqnoTime),
-        seqno_time_cadence);
+        seqno_time_cadence, /*run_immediately=*/true);
  }

  return s;
@@ -1365,7 +1378,7 @@ Status DBImpl::SetDBOptions(
        s = periodic_task_scheduler_.Register(
            PeriodicTaskType::kDumpStats,
            periodic_task_functions_.at(PeriodicTaskType::kDumpStats),
-            new_options.stats_dump_period_sec);
+            new_options.stats_dump_period_sec, /*run_immediately=*/true);
      }
      if (new_options.max_total_wal_size !=
          mutable_db_options_.max_total_wal_size) {
@@ -1380,7 +1393,7 @@ Status DBImpl::SetDBOptions(
          s = periodic_task_scheduler_.Register(
              PeriodicTaskType::kPersistStats,
              periodic_task_functions_.at(PeriodicTaskType::kPersistStats),
-              new_options.stats_persist_period_sec);
+              new_options.stats_persist_period_sec, /*run_immediately=*/true);
        }
      }
      mutex_.Lock();
@@ -3822,11 +3835,9 @@ bool DBImpl::KeyMayExist(const ReadOptions& read_options,

 std::unique_ptr<MultiScan> DBImpl::NewMultiScan(
    const ReadOptions& _read_options, ColumnFamilyHandle* column_family,
-    const std::vector<ScanOptions>& scan_opts) {
-  std::unique_ptr<Iterator> iter(NewIterator(_read_options, column_family));
-  iter->Prepare(scan_opts);
-  std::unique_ptr<MultiScan> ms_iter =
-      std::make_unique<MultiScan>(scan_opts, std::move(iter));
+    const MultiScanArgs& scan_opts) {
+  std::unique_ptr<MultiScan> ms_iter = std::make_unique<MultiScan>(
+      _read_options, scan_opts, this, column_family);
  return ms_iter;
 }

@@ -4334,7 +4345,7 @@ void DBImpl::ReleaseSnapshot(const Snapshot* s) {
    CfdList cf_scheduled;
    if (oldest_snapshot > bottommost_files_mark_threshold_) {
      for (auto* cfd : *versions_->GetColumnFamilySet()) {
-        if (!cfd->ioptions().allow_ingest_behind) {
+        if (!cfd->AllowIngestBehind()) {
          cfd->current()->storage_info()->UpdateOldestSnapshot(
              oldest_snapshot, /*allow_ingest_behind=*/false);
          if (!cfd->current()
@@ -4354,8 +4365,7 @@ void DBImpl::ReleaseSnapshot(const Snapshot* s) {
      // inaccurate.
      SequenceNumber new_bottommost_files_mark_threshold = kMaxSequenceNumber;
      for (auto* cfd : *versions_->GetColumnFamilySet()) {
-        if (CfdListContains(cf_scheduled, cfd) ||
-            cfd->ioptions().allow_ingest_behind) {
+        if (CfdListContains(cf_scheduled, cfd) || cfd->AllowIngestBehind()) {
          continue;
        }
        new_bottommost_files_mark_threshold = std::min(
@@ -5044,85 +5054,6 @@ void DBImpl::GetAllColumnFamilyMetaData(
  }
 }

-Status DBImpl::CheckConsistency() {
-  mutex_.AssertHeld();
-  std::vector<LiveFileMetaData> metadata;
-  versions_->GetLiveFilesMetaData(&metadata);
-  TEST_SYNC_POINT("DBImpl::CheckConsistency:AfterGetLiveFilesMetaData");
-
-  std::string corruption_messages;
-
-  if (immutable_db_options_.skip_checking_sst_file_sizes_on_db_open) {
-    // Instead of calling GetFileSize() for each expected file, call
-    // GetChildren() for the DB directory and check that all expected files
-    // are listed, without checking their sizes.
-    // Since sst files might be in different directories, do it for each
-    // directory separately.
-    std::map<std::string, std::vector<std::string>> files_by_directory;
-    for (const auto& md : metadata) {
-      // md.name has a leading "/". Remove it.
-      std::string fname = md.name;
-      if (!fname.empty() && fname[0] == '/') {
-        fname = fname.substr(1);
-      }
-      files_by_directory[md.db_path].push_back(fname);
-    }
-
-    IOOptions io_opts;
-    io_opts.do_not_recurse = true;
-    for (const auto& dir_files : files_by_directory) {
-      std::string directory = dir_files.first;
-      std::vector<std::string> existing_files;
-      Status s = fs_->GetChildren(directory, io_opts, &existing_files,
-                                  /*IODebugContext*=*/nullptr);
-      if (!s.ok()) {
-        corruption_messages +=
-            "Can't list files in " + directory + ": " + s.ToString() + "\n";
-        continue;
-      }
-      std::sort(existing_files.begin(), existing_files.end());
-
-      for (const std::string& fname : dir_files.second) {
-        if (!std::binary_search(existing_files.begin(), existing_files.end(),
-                                fname) &&
-            !std::binary_search(existing_files.begin(), existing_files.end(),
-                                Rocks2LevelTableFileName(fname))) {
-          corruption_messages +=
-              "Missing sst file " + fname + " in " + directory + "\n";
-        }
-      }
-    }
-  } else {
-    for (const auto& md : metadata) {
-      // md.name has a leading "/".
-      std::string file_path = md.db_path + md.name;
-
-      uint64_t fsize = 0;
-      TEST_SYNC_POINT("DBImpl::CheckConsistency:BeforeGetFileSize");
-      Status s = env_->GetFileSize(file_path, &fsize);
-      if (!s.ok() &&
-          env_->GetFileSize(Rocks2LevelTableFileName(file_path), &fsize).ok()) {
-        s = Status::OK();
-      }
-      if (!s.ok()) {
-        corruption_messages +=
-            "Can't access " + md.name + ": " + s.ToString() + "\n";
-      } else if (fsize != md.size) {
-        corruption_messages += "Sst file size mismatch: " + file_path +
-                               ". Size recorded in manifest " +
-                               std::to_string(md.size) + ", actual size " +
-                               std::to_string(fsize) + "\n";
-      }
-    }
-  }
-
-  if (corruption_messages.size() == 0) {
-    return Status::OK();
-  } else {
-    return Status::Corruption(corruption_messages);
-  }
-}
-
 Status DBImpl::GetDbIdentity(std::string& identity) const {
  identity.assign(db_id_);
  return Status::OK();
@@ -5829,10 +5760,6 @@ Status DBImpl::IngestExternalFiles(
  for (const auto& arg : args) {
    const IngestExternalFileOptions& ingest_opts = arg.options;
    if (ingest_opts.ingest_behind) {
-      if (!immutable_db_options_.allow_ingest_behind) {
-        return Status::InvalidArgument(
-            "can't ingest_behind file in DB with allow_ingest_behind=false");
-      }
      auto ucmp = arg.column_family->GetComparator();
      assert(ucmp);
      if (ucmp->timestamp_size() > 0) {
@@ -5840,6 +5767,14 @@ Status DBImpl::IngestExternalFiles(
            "Column family with user-defined "
            "timestamps enabled doesn't support ingest behind.");
      }
+
+      if (!static_cast<ColumnFamilyHandleImpl*>(arg.column_family)
+               ->cfd()
+               ->AllowIngestBehind()) {
+        return Status::InvalidArgument(
+            "Can't ingest_behind file in ColumnFamily %s with "
+            "cf_allow_ingest_behind=false");
+      }
    }
    if (arg.atomic_replace_range.has_value()) {
      if (ingest_opts.ingest_behind) {
@@ -6074,18 +6009,19 @@ Status DBImpl::IngestExternalFiles(
      // mutex when persisting MANIFEST file, and the snapshots taken during
      // that period will not be stable if VersionSet last seqno is updated
      // before LogAndApply.
-      int consumed_seqno_count =
-          ingestion_jobs[0].ConsumedSequenceNumbersCount();
+      SequenceNumber max_assigned_seqno =
+          ingestion_jobs[0].MaxAssignedSequenceNumber();
      for (size_t i = 1; i != num_cfs; ++i) {
-        consumed_seqno_count =
-            std::max(consumed_seqno_count,
-                     ingestion_jobs[i].ConsumedSequenceNumbersCount());
+        max_assigned_seqno = std::max(
+            max_assigned_seqno, ingestion_jobs[i].MaxAssignedSequenceNumber());
      }
-      if (consumed_seqno_count > 0) {
+      if (max_assigned_seqno > 0) {
        const SequenceNumber last_seqno = versions_->LastSequence();
-        versions_->SetLastAllocatedSequence(last_seqno + consumed_seqno_count);
-        versions_->SetLastPublishedSequence(last_seqno + consumed_seqno_count);
-        versions_->SetLastSequence(last_seqno + consumed_seqno_count);
+        if (max_assigned_seqno > last_seqno) {
+          versions_->SetLastAllocatedSequence(max_assigned_seqno);
+          versions_->SetLastPublishedSequence(max_assigned_seqno);
+          versions_->SetLastSequence(max_assigned_seqno);
+        }
      }
    }

@@ -6884,6 +6820,35 @@ void DBImpl::RecordSeqnoToTimeMapping() {
  sv_context.Clean();
 }

+void DBImpl::TriggerPeriodicCompaction() {
+  TEST_SYNC_POINT("DBImpl::TriggerPeriodicCompaction:StartRunning");
+  {
+    InstrumentedMutexLock l(&mutex_);
+    ROCKS_LOG_INFO(immutable_db_options_.info_log,
+                   "Running the periodic task to trigger compactions.");
+
+    for (ColumnFamilyData* cfd : *versions_->GetColumnFamilySet()) {
+      if (cfd->IsDropped()) {
+        continue;
+      }
+      if (cfd->GetLatestCFOptions().periodic_compaction_seconds &&
+          !cfd->queued_for_compaction()) {
+        cfd->current()->storage_info()->ComputeCompactionScore(
+            cfd->ioptions(), cfd->GetLatestMutableCFOptions());
+        EnqueuePendingCompaction(cfd);
+        if (cfd->queued_for_compaction()) {
+          ROCKS_LOG_INFO(immutable_db_options_.info_log,
+                         "Periodic task to trigger compaction queued Column "
+                         "family [%s] for compaction.",
+                         cfd->GetName().c_str());
+        }
+      }
+    }
+    MaybeScheduleFlushOrCompaction();
+    bg_cv_.SignalAll();
+  }
+}
+
 void DBImpl::TrackOrUntrackFiles(
    const std::vector<std::string>& existing_data_files, bool track) {
  auto sfm = static_cast_with_check<SstFileManagerImpl>(
@@ -386,7 +386,7 @@ class DBImpl : public DB {
  using DB::NewMultiScan;
  std::unique_ptr<MultiScan> NewMultiScan(
      const ReadOptions& _read_options, ColumnFamilyHandle* column_family,
-      const std::vector<ScanOptions>& scan_opts) override;
+      const MultiScanArgs& scan_opts) override;

  const Snapshot* GetSnapshot() override;
  void ReleaseSnapshot(const Snapshot* snapshot) override;
@@ -804,10 +804,6 @@ class DBImpl : public DB {
  // being detected.
  const Snapshot* GetSnapshotForWriteConflictBoundary();

-  // checks if all live files exist on file system and that their file sizes
-  // match to our in-memory records
-  virtual Status CheckConsistency();
-
  // max_file_num_to_ignore allows bottom level compaction to filter out newly
  // compacted SST files. Setting max_file_num_to_ignore to kMaxUint64 will
  // disable the filtering
@@ -1286,6 +1282,12 @@ class DBImpl : public DB {
  // For the background timer job
  void RecordSeqnoToTimeMapping();

+  // Compactions rely on an event triggers like flush/compaction/SetOptions.
+  // We need to trigger periodic compactions even when there is no such trigger.
+  // This function checks and schedules available compactions and will run
+  // periodically.
+  void TriggerPeriodicCompaction();
+
  // REQUIRES: DB mutex held
  std::pair<SequenceNumber, uint64_t> GetSeqnoToTimeSample() const;

@@ -1386,6 +1388,9 @@ class DBImpl : public DB {
  // logs_, cur_wal_number_. Refer to the definition of each variable below for
  // more description.
  //
+  // Protects access to most ColumnFamilyData methods, see more in comment for
+  // each method.
+  //
  // `mutex_` can be a hot lock in some workloads, so it deserves dedicated
  // cachelines.
  mutable CacheAlignedInstrumentedMutex mutex_;
@@ -1111,8 +1111,7 @@ Status DBImpl::CompactRangeInternal(const CompactRangeOptions& options,
      cfd->NumberLevels() > 1) {
    // Always compact all files together.
    final_output_level = cfd->NumberLevels() - 1;
-    // if bottom most level is reserved
-    if (immutable_db_options_.allow_ingest_behind) {
+    if (cfd->AllowIngestBehind()) {
      final_output_level--;
    }
    s = RunManualCompaction(cfd, ColumnFamilyData::kCompactAllLevels,
@@ -1460,7 +1459,7 @@ Status DBImpl::CompactFilesImpl(
    }
  }

-  if (cfd->ioptions().allow_ingest_behind &&
+  if (cfd->AllowIngestBehind() &&
      output_level >= cfd->ioptions().num_levels - 1) {
    return Status::InvalidArgument(
        "Exceed the maximum output level defined by "
@@ -1500,7 +1499,7 @@ Status DBImpl::CompactFilesImpl(

  std::unique_ptr<Compaction> c;
  assert(cfd->compaction_picker());
-  c.reset(cfd->compaction_picker()->CompactFiles(
+  c.reset(cfd->compaction_picker()->PickCompactionForCompactFiles(
      compact_options, input_files, output_level, version->storage_info(),
      cfd->GetLatestMutableCFOptions(), mutable_db_options_, output_path_id));
  // we already sanitized the set of input files and checked for conflicts
@@ -1849,10 +1848,9 @@ Status DBImpl::ReFitLevel(ColumnFamilyData* cfd, int level, int target_level) {
        0 /* max_subcompactions, not applicable */,
        {} /* grandparents, not applicable */,
        std::nullopt /* earliest_snapshot */, nullptr /* snapshot_checker */,
-        false /* is manual */, "" /* trim_ts */, -1 /* score, not applicable */,
-        false /* is deletion compaction, not applicable */,
-        false /* l0_files_might_overlap, not applicable */,
-        CompactionReason::kRefitLevel));
+        CompactionReason::kRefitLevel, "" /* trim_ts */,
+        -1 /* score, not applicable */,
+        false /* l0_files_might_overlap, not applicable */));
    cfd->compaction_picker()->RegisterCompaction(c.get());
    TEST_SYNC_POINT("DBImpl::ReFitLevel:PostRegisterCompaction");
    VersionEdit edit;
@@ -4156,6 +4154,7 @@ Status DBImpl::BackgroundCompaction(bool* made_progress,
                     ->current()
                     ->storage_info()
                     ->MaxOutputLevel(
+                         c->immutable_options().cf_allow_ingest_behind ||
                         immutable_db_options_.allow_ingest_behind)) &&
             env_->GetBackgroundThreads(Env::Priority::BOTTOM) > 0) {
    assert(thread_pri == Env::Priority::LOW);
@@ -4424,13 +4423,14 @@ Compaction* DBImpl::CreateIntendedCompactionForwardedToBottomPriorityPool(

  c->ReleaseCompactionFiles(Status::OK());

-  Compaction* intended_compaction = new Compaction(
-      vstorage, io, mo, mutable_db_options_, std::move(inputs),
-      c->output_level(), c->target_output_file_size(),
-      c->max_compaction_bytes(), c->output_path_id(), c->output_compression(),
-      c->output_compression_opts(), c->output_temperature(),
-      c->max_subcompactions(), c->grandparents(),
-      std::nullopt /* earliest_snapshot */, nullptr /* snapshot_checker */);
+  Compaction* intended_compaction =
+      new Compaction(vstorage, io, mo, mutable_db_options_, std::move(inputs),
+                     c->output_level(), c->target_output_file_size(),
+                     c->max_compaction_bytes(), c->output_path_id(),
+                     c->output_compression(), c->output_compression_opts(),
+                     c->output_temperature(), c->max_subcompactions(),
+                     c->grandparents(), std::nullopt /* earliest_snapshot */,
+                     nullptr /* snapshot_checker */, c->compaction_reason());

  cfd->compaction_picker()->RegisterCompaction(intended_compaction);
  vstorage->ComputeCompactionScore(io, mo);
@@ -4660,7 +4660,7 @@ void DBImpl::InstallSuperVersionAndScheduleWork(
  bottommost_files_mark_threshold_ = kMaxSequenceNumber;
  standalone_range_deletion_files_mark_threshold_ = kMaxSequenceNumber;
  for (auto* my_cfd : *versions_->GetColumnFamilySet()) {
-    if (!my_cfd->ioptions().allow_ingest_behind) {
+    if (!my_cfd->AllowIngestBehind()) {
      bottommost_files_mark_threshold_ = std::min(
          bottommost_files_mark_threshold_,
          my_cfd->current()->storage_info()->bottommost_files_mark_threshold());
@@ -70,9 +70,6 @@ Status DBImplFollower::Recover(
    }
    return s;
  }
-  if (immutable_db_options_.paranoid_checks && s.ok()) {
-    s = CheckConsistency();
-  }
  if (s.ok()) {
    default_cf_handle_ = new ColumnFamilyHandleImpl(
        versions_->GetColumnFamilySet()->GetDefault(), this, &mutex_);
@@ -191,12 +191,6 @@ DBOptions SanitizeOptions(const std::string& dbname, const DBOptions& src,
                   "wal_compression is disabled since only zstd is supported");
  }

-  if (!result.paranoid_checks) {
-    result.skip_checking_sst_file_sizes_on_db_open = true;
-    ROCKS_LOG_INFO(result.info_log,
-                   "file size check will be skipped during open.");
-  }
-
  return result;
 }

@@ -605,7 +599,7 @@ Status DBImpl::Recover(
        // allow_ingest_behind does not support Level Compaction,
        // and per_key_placement can have infinite compaction loop for Level
        // Compaction. Adjust to_level here just to be safe.
-        if (cfd->ioptions().allow_ingest_behind ||
+        if (cfd->AllowIngestBehind() ||
            moptions.preclude_last_level_data_seconds > 0) {
          to_level -= 1;
        }
@@ -694,9 +688,6 @@ Status DBImpl::Recover(
    s = MaybeUpdateNextFileNumber(recovery_ctx);
  }

-  if (immutable_db_options_.paranoid_checks && s.ok()) {
-    s = CheckConsistency();
-  }
  if (s.ok() && !read_only) {
    // TODO: share file descriptors (FSDirectory) with SetDirectories above
    std::map<std::string, std::shared_ptr<FSDirectory>> created_dirs;
@@ -1764,8 +1755,12 @@ Status DBImpl::MaybeHandleStopReplayForCorruptionForInconsistency(
        ROCKS_LOG_ERROR(immutable_db_options_.info_log,
                        "Column family inconsistency: SST file contains data"
                        " beyond the point of corruption.");
-        status = Status::Corruption("SST file is ahead of WALs in CF " +
-                                    cfd->GetName());
+        status = Status::Corruption(
+            "Column family inconsistency: SST file contains data"
+            " beyond the point of corruption in CF " +
+            cfd->GetName() +
+            ". WAL recovery stopped at corruption point, but SST files"
+            " contain newer data.");
        return status;
      }
    }
@@ -49,9 +49,6 @@ Status DBImplSecondary::Recover(
    }
    return s;
  }
-  if (immutable_db_options_.paranoid_checks && s.ok()) {
-    s = CheckConsistency();
-  }
  // Initial max_total_in_memory_state_ before recovery logs.
  max_total_in_memory_state_ = 0;
  for (auto cfd : *versions_->GetColumnFamilySet()) {
@@ -653,49 +650,6 @@ Status DBImplSecondary::NewIterators(
  return Status::OK();
 }

-Status DBImplSecondary::CheckConsistency() {
-  mutex_.AssertHeld();
-  Status s = DBImpl::CheckConsistency();
-  // If DBImpl::CheckConsistency() which is stricter returns success, then we
-  // do not need to give a second chance.
-  if (s.ok()) {
-    return s;
-  }
-  // It's possible that DBImpl::CheckConssitency() can fail because the primary
-  // may have removed certain files, causing the GetFileSize(name) call to
-  // fail and returning a PathNotFound. In this case, we take a best-effort
-  // approach and just proceed.
-  TEST_SYNC_POINT_CALLBACK(
-      "DBImplSecondary::CheckConsistency:AfterFirstAttempt", &s);
-
-  if (immutable_db_options_.skip_checking_sst_file_sizes_on_db_open) {
-    return Status::OK();
-  }
-
-  std::vector<LiveFileMetaData> metadata;
-  versions_->GetLiveFilesMetaData(&metadata);
-
-  std::string corruption_messages;
-  for (const auto& md : metadata) {
-    // md.name has a leading "/".
-    std::string file_path = md.db_path + md.name;
-
-    uint64_t fsize = 0;
-    s = env_->GetFileSize(file_path, &fsize);
-    if (!s.ok() &&
-        (env_->GetFileSize(Rocks2LevelTableFileName(file_path), &fsize).ok() ||
-         s.IsPathNotFound())) {
-      s = Status::OK();
-    }
-    if (!s.ok()) {
-      corruption_messages +=
-          "Can't access " + md.name + ": " + s.ToString() + "\n";
-    }
-  }
-  return corruption_messages.empty() ? Status::OK()
-                                     : Status::Corruption(corruption_messages);
-}
-
 Status DBImplSecondary::TryCatchUpWithPrimary() {
  assert(versions_.get() != nullptr);
  Status s;
@@ -894,7 +848,6 @@ Status DBImplSecondary::CompactWithoutInstallation(

  VersionStorageInfo* vstorage = version->storage_info();

-  // Use comp_options to reuse some CompactFiles functions
  CompactionOptions comp_options;
  comp_options.compression = kDisableCompressionOption;
  comp_options.output_file_size_limit = MaxFileSizeForLevel(
@@ -913,13 +866,27 @@ Status DBImplSecondary::CompactWithoutInstallation(
    return s;
  }

+  const int job_id = next_job_id_.fetch_add(1);
+  JobContext job_context(job_id, true /*create_superversion*/);
+  std::vector<SequenceNumber> snapshots = input.snapshots;
+
+  // TODO - snapshot_checker support in Remote Compaction
+  job_context.InitSnapshotContext(/*checker=*/nullptr,
+                                  /*managed_snapshot=*/nullptr,
+                                  kMaxSequenceNumber, std::move(snapshots));
+
+  // TODO - consider serializing the entire Compaction object and using it as
+  // input instead of recreating it in the remote worker
  std::unique_ptr<Compaction> c;
  assert(cfd->compaction_picker());
-  c.reset(cfd->compaction_picker()->CompactFiles(
+  c.reset(cfd->compaction_picker()->PickCompactionForCompactFiles(
      comp_options, input_files, input.output_level, vstorage,
-      cfd->GetLatestMutableCFOptions(), mutable_db_options_, 0));
+      cfd->GetLatestMutableCFOptions(), mutable_db_options_, 0,
+      /*earliest_snapshot=*/job_context.snapshot_seqs.empty()
+          ? kMaxSequenceNumber
+          : job_context.snapshot_seqs.front(),
+      job_context.snapshot_checker));
  assert(c != nullptr);
-
  c->FinalizeInputInfo(version);

  // Create output directory if it's not existed yet
@@ -932,11 +899,6 @@ Status DBImplSecondary::CompactWithoutInstallation(
  LogBuffer log_buffer(InfoLogLevel::INFO_LEVEL,
                       immutable_db_options_.info_log.get());

-  const int job_id = next_job_id_.fetch_add(1);
-  JobContext job_context(0, true /*create_superversion*/);
-  std::vector<SequenceNumber> snapshots = input.snapshots;
-  job_context.InitSnapshotContext(nullptr, nullptr, kMaxSequenceNumber,
-                                  std::move(snapshots));
  // use primary host's db_id for running the compaction, but db_session_id is
  // using the local one, which is to make sure the unique id is unique from
  // the remote compactors. Because the id is generated from db_id,
@@ -248,12 +248,6 @@ class DBImplSecondary : public DBImpl {
  Status MaybeInitLogReader(uint64_t log_number,
                            log::FragmentBufferedReader** log_reader);

-  // Check if all live files exist on file system and that their file sizes
-  // matche to the in-memory records. It is possible that some live files may
-  // have been deleted by the primary. In this case, CheckConsistency() does
-  // not flag the missing file as inconsistency.
-  Status CheckConsistency() override;
-
 #ifndef NDEBUG
  Status TEST_CompactWithoutInstallation(const OpenAndCompactOptions& options,
                                         ColumnFamilyHandle* cfh,
@@ -240,8 +240,8 @@ class DBIter final : public Iterator {

  bool PrepareValue() override;

-  void Prepare(const std::vector<ScanOptions>& scan_opts) override {
-    std::optional<std::vector<ScanOptions>> new_scan_opts;
+  void Prepare(const MultiScanArgs& scan_opts) override {
+    std::optional<MultiScanArgs> new_scan_opts;
    new_scan_opts.emplace(scan_opts);
    scan_opts_.swap(new_scan_opts);
    if (!scan_opts.empty()) {
@@ -505,7 +505,7 @@ class DBIter final : public Iterator {
  const Slice* const timestamp_lb_;
  const size_t timestamp_size_;
  std::string saved_timestamp_;
-  std::optional<std::vector<ScanOptions>> scan_opts_;
+  std::optional<MultiScanArgs> scan_opts_;
  ReadOnlyMemTable* const active_mem_;
  SequenceNumber memtable_seqno_lb_;
  uint32_t memtable_op_scan_flush_trigger_;
@@ -8,6 +8,8 @@
 // found in the LICENSE file. See the AUTHORS file for names of contributors.

 #include <functional>
+#include <iomanip>
+#include <iostream>

 #include "db/arena_wrapped_db_iter.h"
 #include "db/db_iter.h"
@@ -4139,6 +4141,240 @@ TEST_P(DBIteratorTest, AverageMemtableOpsScanFlushTriggerByOverwrites) {
  ASSERT_OK(db_->WaitForCompact({}));
  ASSERT_EQ(1, NumTableFilesAtLevel(0));
 }
+
+class DBMultiScanIteratorTest : public DBTestBase,
+                                public ::testing::WithParamInterface<bool> {
+ public:
+  DBMultiScanIteratorTest()
+      : DBTestBase("db_multi_scan_iterator_test", /*env_do_fsync=*/true) {}
+};
+
+INSTANTIATE_TEST_CASE_P(DBMultiScanIteratorTest, DBMultiScanIteratorTest,
+                        ::testing::Bool());
+
+TEST_P(DBMultiScanIteratorTest, BasicTest) {
+  // Create a file
+  for (int i = 0; i < 100; ++i) {
+    std::stringstream ss;
+    ss << std::setw(2) << std::setfill('0') << i;
+    ASSERT_OK(Put("k" + ss.str(), "val" + ss.str()));
+  }
+  ASSERT_OK(Flush());
+
+  std::vector<std::string> key_ranges({"k03", "k10", "k25", "k50"});
+  ReadOptions ro;
+  ro.fill_cache = GetParam();
+  MultiScanArgs scan_options(BytewiseComparator());
+  scan_options.insert(key_ranges[0], key_ranges[1]);
+  scan_options.insert(key_ranges[2], key_ranges[3]);
+  ColumnFamilyHandle* cfh = dbfull()->DefaultColumnFamily();
+  std::unique_ptr<MultiScan> iter =
+      dbfull()->NewMultiScan(ro, cfh, scan_options);
+  try {
+    int idx = 0;
+    int count = 0;
+    for (auto range : *iter) {
+      for (auto it : range) {
+        ASSERT_GE(it.first.ToString().compare(key_ranges[idx]), 0);
+        ASSERT_LT(it.first.ToString().compare(key_ranges[idx + 1]), 0);
+        count++;
+      }
+      idx += 2;
+    }
+    ASSERT_EQ(count, 32);
+  } catch (MultiScanException& ex) {
+    // Make sure exception contains the status
+    ASSERT_NOK(ex.status());
+    std::cerr << "Iterator returned status " << ex.what();
+    abort();
+  } catch (std::logic_error& ex) {
+    std::cerr << "Iterator returned logic error " << ex.what();
+    abort();
+  }
+  iter.reset();
+
+  // Test the overlapping scan case
+  key_ranges[1] = "k30";
+  scan_options = MultiScanArgs(BytewiseComparator());
+  scan_options.insert(key_ranges[0], key_ranges[1]);
+  scan_options.insert(key_ranges[2], key_ranges[3]);
+
+  iter = dbfull()->NewMultiScan(ro, cfh, scan_options);
+  try {
+    int idx = 0;
+    int count = 0;
+    for (auto range : *iter) {
+      for (auto it : range) {
+        ASSERT_GE(it.first.ToString().compare(key_ranges[idx]), 0);
+        ASSERT_LT(it.first.ToString().compare(key_ranges[idx + 1]), 0);
+        count++;
+      }
+      idx += 2;
+    }
+    ASSERT_EQ(count, 52);
+  } catch (MultiScanException& ex) {
+    // Make sure exception contains the status
+    ASSERT_NOK(ex.status());
+    std::cerr << "Iterator returned status " << ex.what();
+    abort();
+  } catch (std::logic_error& ex) {
+    std::cerr << "Iterator returned logic error " << ex.what();
+    abort();
+  }
+  iter.reset();
+
+  // Test the no limit scan case
+  scan_options = MultiScanArgs(BytewiseComparator());
+  scan_options.insert(key_ranges[0]);
+  scan_options.insert(key_ranges[2]);
+  iter = dbfull()->NewMultiScan(ro, cfh, scan_options);
+  try {
+    int idx = 0;
+    int count = 0;
+    for (auto range : *iter) {
+      for (auto it : range) {
+        ASSERT_GE(it.first.ToString().compare(key_ranges[idx]), 0);
+        if (it.first.ToString().compare(key_ranges[idx + 1]) == 0) {
+          break;
+        }
+        count++;
+      }
+      idx += 2;
+    }
+    ASSERT_EQ(count, 52);
+  } catch (MultiScanException& ex) {
+    // Make sure exception contains the status
+    ASSERT_NOK(ex.status());
+    std::cerr << "Iterator returned status " << ex.what();
+    abort();
+  } catch (std::logic_error& ex) {
+    std::cerr << "Iterator returned logic error " << ex.what();
+    abort();
+  }
+  iter.reset();
+}
+
+TEST_P(DBMultiScanIteratorTest, MixedBoundsTest) {
+  // Create a file
+  for (int i = 0; i < 100; ++i) {
+    std::stringstream ss;
+    ss << std::setw(2) << std::setfill('0') << i;
+    ASSERT_OK(Put("k" + ss.str(), "val" + ss.str()));
+  }
+  ASSERT_OK(Flush());
+
+  std::vector<std::string> key_ranges(
+      {"k03", "k10", "k25", "k50", "k75", "k90"});
+  ReadOptions ro;
+  ro.fill_cache = GetParam();
+  MultiScanArgs scan_options(BytewiseComparator());
+  scan_options.insert(key_ranges[0], key_ranges[1]);
+  scan_options.insert(key_ranges[2]);
+  scan_options.insert(key_ranges[4], key_ranges[5]);
+  ColumnFamilyHandle* cfh = dbfull()->DefaultColumnFamily();
+  std::unique_ptr<MultiScan> iter =
+      dbfull()->NewMultiScan(ro, cfh, scan_options);
+  try {
+    int idx = 0;
+    int count = 0;
+    for (auto range : *iter) {
+      for (auto it : range) {
+        ASSERT_GE(
+            it.first.ToString().compare(
+                scan_options.GetScanRanges()[idx].range.start->ToString()),
+            0);
+        if (scan_options.GetScanRanges()[idx].range.limit) {
+          ASSERT_LT(
+              it.first.ToString().compare(
+                  scan_options.GetScanRanges()[idx].range.limit->ToString()),
+              0);
+        }
+        count++;
+      }
+      idx++;
+    }
+    ASSERT_EQ(count, 97);
+  } catch (MultiScanException& ex) {
+    // Make sure exception contains the status
+    ASSERT_NOK(ex.status());
+    std::cerr << "Iterator returned status " << ex.what();
+    abort();
+  } catch (std::logic_error& ex) {
+    std::cerr << "Iterator returned logic error " << ex.what();
+    abort();
+  }
+  iter.reset();
+  scan_options = MultiScanArgs(BytewiseComparator());
+  scan_options.insert(key_ranges[0]);
+  scan_options.insert(key_ranges[2], key_ranges[3]);
+  scan_options.insert(key_ranges[4]);
+  iter = dbfull()->NewMultiScan(ro, cfh, scan_options);
+  try {
+    int idx = 0;
+    int count = 0;
+    for (auto range : *iter) {
+      for (auto it : range) {
+        ASSERT_GE(
+            it.first.ToString().compare(
+                scan_options.GetScanRanges()[idx].range.start->ToString()),
+            0);
+        if (scan_options.GetScanRanges()[idx].range.limit) {
+          ASSERT_LT(
+              it.first.ToString().compare(
+                  scan_options.GetScanRanges()[idx].range.limit->ToString()),
+              0);
+        }
+        count++;
+      }
+      idx++;
+    }
+    ASSERT_EQ(count, 147);
+  } catch (MultiScanException& ex) {
+    // Make sure exception contains the status
+    ASSERT_NOK(ex.status());
+    std::cerr << "Iterator returned status " << ex.what();
+    abort();
+  } catch (std::logic_error& ex) {
+    std::cerr << "Iterator returned logic error " << ex.what();
+    abort();
+  }
+  iter.reset();
+}
+
+TEST_P(DBMultiScanIteratorTest, RangeAcrossFiles) {
+  auto options = CurrentOptions();
+  options.target_file_size_base = 100 << 10;  // 20KB
+  options.compaction_style = kCompactionStyleUniversal;
+  options.num_levels = 50;
+  options.compression = kNoCompression;
+  DestroyAndReopen(options);
+
+  auto rnd = Random::GetTLSInstance();
+  // Write ~200KB data
+  for (int i = 0; i < 100; ++i) {
+    ASSERT_OK(Put(Key(i), rnd->RandomString(2 << 10)));
+  }
+  ASSERT_OK(Flush());
+
+  ASSERT_OK(db_->CompactRange({}, nullptr, nullptr));
+  ASSERT_EQ(2, NumTableFilesAtLevel(49));
+  std::vector<std::string> key_ranges({Key(10), Key(90)});
+  ReadOptions ro;
+  ro.fill_cache = GetParam();
+  MultiScanArgs scan_options(BytewiseComparator());
+  scan_options.insert(key_ranges[0], key_ranges[1]);
+  ColumnFamilyHandle* cfh = dbfull()->DefaultColumnFamily();
+  std::unique_ptr<MultiScan> iter =
+      dbfull()->NewMultiScan(ro, cfh, scan_options);
+  int i = 10;
+  for (auto range : *iter) {
+    for (auto it : range) {
+      ASSERT_EQ(it.first.ToString(), Key(i));
+      ++i;
+    }
+  }
+  ASSERT_EQ(i, 90);
+}
 }  // namespace ROCKSDB_NAMESPACE

 int main(int argc, char** argv) {
@@ -605,6 +605,9 @@ class TraceFileEnv : public EnvWrapper {
                  char* scratch) const override {
        return target_->Read(offset, n, result, scratch);
      }
+      Status GetFileSize(uint64_t* file_size) override {
+        return target_->GetFileSize(file_size);
+      }

     private:
      std::unique_ptr<RandomAccessFile> target_;
@@ -1291,46 +1294,6 @@ TEST_F(DBSecondaryTest, CatchUpAfterFlush) {
  ASSERT_OK(iter3->status());
 }

-TEST_F(DBSecondaryTest, CheckConsistencyWhenOpen) {
-  bool called = false;
-  Options options;
-  options.env = env_;
-  options.disable_auto_compactions = true;
-  Reopen(options);
-  SyncPoint::GetInstance()->DisableProcessing();
-  SyncPoint::GetInstance()->ClearAllCallBacks();
-  SyncPoint::GetInstance()->SetCallBack(
-      "DBImplSecondary::CheckConsistency:AfterFirstAttempt", [&](void* arg) {
-        ASSERT_NE(nullptr, arg);
-        called = true;
-        auto* s = static_cast<Status*>(arg);
-        ASSERT_NOK(*s);
-      });
-  SyncPoint::GetInstance()->LoadDependency(
-      {{"DBImpl::CheckConsistency:AfterGetLiveFilesMetaData",
-        "BackgroundCallCompaction:0"},
-       {"DBImpl::BackgroundCallCompaction:PurgedObsoleteFiles",
-        "DBImpl::CheckConsistency:BeforeGetFileSize"}});
-  SyncPoint::GetInstance()->EnableProcessing();
-
-  ASSERT_OK(Put("a", "value0"));
-  ASSERT_OK(Put("c", "value0"));
-  ASSERT_OK(Flush());
-  ASSERT_OK(Put("b", "value1"));
-  ASSERT_OK(Put("d", "value1"));
-  ASSERT_OK(Flush());
-  port::Thread thread([this]() {
-    Options opts;
-    opts.env = env_;
-    opts.max_open_files = -1;
-    OpenSecondary(opts);
-  });
-  ASSERT_OK(dbfull()->CompactRange(CompactRangeOptions(), nullptr, nullptr));
-  ASSERT_OK(dbfull()->TEST_WaitForCompact());
-  thread.join();
-  ASSERT_TRUE(called);
-}
-
 TEST_F(DBSecondaryTest, StartFromInconsistent) {
  Options options = CurrentOptions();
  DestroyAndReopen(options);
@@ -135,21 +135,6 @@ TEST_F(DBSSTTest, SSTsWithLdbSuffixHandling) {
  Destroy(options);
 }

-// Check that we don't crash when opening DB with
-// DBOptions::skip_checking_sst_file_sizes_on_db_open = true.
-TEST_F(DBSSTTest, SkipCheckingSSTFileSizesOnDBOpen) {
-  ASSERT_OK(Put("pika", "choo"));
-  ASSERT_OK(Flush());
-
-  // Just open the DB with the option set to true and check that we don't crash.
-  Options options;
-  options.env = env_;
-  options.skip_checking_sst_file_sizes_on_db_open = true;
-  Reopen(options);
-
-  ASSERT_EQ("choo", Get("pika"));
-}
-
 TEST_F(DBSSTTest, DontDeleteMovedFile) {
  // This test triggers move compaction and verifies that the file is not
  // deleted when it's part of move compaction
@@ -1748,45 +1733,6 @@ TEST_F(DBSSTTest, GetTotalSstFilesSize) {
  ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->DisableProcessing();
 }

-TEST_F(DBSSTTest, OpenDBWithoutGetFileSizeInvocations) {
-  Options options = CurrentOptions();
-  std::unique_ptr<MockEnv> env{MockEnv::Create(Env::Default())};
-  options.env = env.get();
-  options.disable_auto_compactions = true;
-  options.compression = kNoCompression;
-  options.enable_blob_files = true;
-  options.blob_file_size = 32;  // create one blob per file
-  options.skip_checking_sst_file_sizes_on_db_open = true;
-
-  DestroyAndReopen(options);
-  // Generate 5 files in L0
-  for (int i = 0; i < 5; i++) {
-    for (int j = 0; j < 10; j++) {
-      std::string val = "val_file_" + std::to_string(i);
-      ASSERT_OK(Put(Key(j), val));
-    }
-    ASSERT_OK(Flush());
-  }
-  Close();
-
-  bool is_get_file_size_called = false;
-  SyncPoint::GetInstance()->SetCallBack(
-      "MockFileSystem::GetFileSize:CheckFileType", [&](void* arg) {
-        std::string* filename = static_cast<std::string*>(arg);
-        if (filename->find(".blob") != std::string::npos) {
-          is_get_file_size_called = true;
-        }
-      });
-
-  SyncPoint::GetInstance()->EnableProcessing();
-  Reopen(options);
-  ASSERT_FALSE(is_get_file_size_called);
-  SyncPoint::GetInstance()->DisableProcessing();
-  SyncPoint::GetInstance()->ClearAllCallBacks();
-
-  Destroy(options);
-}
-
 TEST_F(DBSSTTest, GetTotalSstFilesSizeVersionsFilesShared) {
  Options options = CurrentOptions();
  options.disable_auto_compactions = true;
@@ -1991,6 +1937,70 @@ TEST_F(DBSSTTest, DBWithSFMForBlobFilesAtomicFlush) {
  ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->ClearAllCallBacks();
 }

+TEST_F(DBSSTTest, SstGetFileSizeFails) {
+  // Build an SST file
+  ASSERT_OK(Put("x", "zaphod"));
+  ASSERT_OK(Flush());
+  std::vector<LiveFileMetaData> metadata;
+  db_->GetLiveFilesMetaData(&metadata);
+  ASSERT_EQ(1U, metadata.size());
+  std::string filename = dbname_ + metadata[0].name;
+
+  // Prepare for fault injection
+  std::shared_ptr<FaultInjectionTestFS> fault_fs =
+      std::make_shared<FaultInjectionTestFS>(
+          CurrentOptions().env->GetFileSystem());
+  std::unique_ptr<Env> fault_fs_env(NewCompositeEnv(fault_fs));
+  Options options = CurrentOptions();
+  options.env = fault_fs_env.get();
+  options.paranoid_checks = false;  // don't check file sizes on open
+
+  for (int i = 0; i < 4; i++) {
+    SCOPED_TRACE("Iteration = " + std::to_string(i));
+    fault_fs->SetFailRandomAccessGetFileSizeSst(false);
+    fault_fs->SetFailFilesystemGetFileSizeSst(false);
+    Close();
+
+    if (i == 1) {
+      // Just FSRandomAccessFile::GetFileSize fails, which should be worked
+      // around
+      fault_fs->SetFailRandomAccessGetFileSizeSst(true);
+    } else if (i == 2) {
+      // FileSystem::GetFileSize fails, which should be worked around if
+      // FSRandomAccessFile::GetFileSize is supported
+      fault_fs->SetFailFilesystemGetFileSizeSst(true);
+    } else if (i == 3) {
+      // Both GetFileSize APIs fail with an IOError
+      fault_fs->SetFailRandomAccessGetFileSizeSst(true);
+      fault_fs->SetFailFilesystemGetFileSizeSst(true);
+    }
+
+    ASSERT_OK(TryReopen(options));
+    std::string value;
+    Status get_status = db_->Get({}, "x", &value);
+    if (i < 2) {
+      ASSERT_OK(get_status);
+    } else if (i == 2) {
+      if (encrypted_env_) {
+        // Can't recover because RandomAccessFile::GetFileSize is not supported
+        // on EncryptedEnv
+        // Fail with propagated IOError. (Not Corruption nor NotSupported!)
+        ASSERT_EQ(get_status.code(), Status::Code::kIOError);
+        ASSERT_STREQ(get_status.getState(), "FileSystem::GetFileSize failed");
+      } else {
+        // Never sees the FileSystem::GetFileSize failure
+        ASSERT_OK(get_status);
+      }
+    } else {
+      ASSERT_EQ(i, 3);
+      // Fail with propagated IOError. (Not Corruption nor NotSupported!)
+      ASSERT_EQ(get_status.code(), Status::Code::kIOError);
+      ASSERT_STREQ(get_status.getState(), "FileSystem::GetFileSize failed");
+    }
+  }
+  Close();
+}
+
 }  // namespace ROCKSDB_NAMESPACE

 int main(int argc, char** argv) {
@@ -1278,12 +1278,6 @@ class DelayFilterFactory : public CompactionFilterFactory {
 };
 }  // anonymous namespace

-static std::string CompressibleString(Random* rnd, int len) {
-  std::string r;
-  test::CompressibleString(rnd, 0.8, len, &r);
-  return r;
-}
-
 TEST_F(DBTest, FailMoreDbPaths) {
  Options options = CurrentOptions();
  options.db_paths.emplace_back(dbname_, 10000000);
@@ -5407,271 +5401,6 @@ TEST_F(DBTest, FlushOnDestroy) {
  CancelAllBackgroundWork(db_);
 }

-TEST_F(DBTest, DynamicLevelCompressionPerLevel) {
-  if (!Snappy_Supported()) {
-    return;
-  }
-  const int kNKeys = 120;
-  int keys[kNKeys];
-  for (int i = 0; i < kNKeys; i++) {
-    keys[i] = i;
-  }
-
-  Random rnd(301);
-  Options options;
-  options.env = env_;
-  options.create_if_missing = true;
-  options.db_write_buffer_size = 20480;
-  options.write_buffer_size = 20480;
-  options.max_write_buffer_number = 2;
-  options.level0_file_num_compaction_trigger = 2;
-  options.level0_slowdown_writes_trigger = 2;
-  options.level0_stop_writes_trigger = 2;
-  options.target_file_size_base = 20480;
-  options.level_compaction_dynamic_level_bytes = true;
-  options.max_bytes_for_level_base = 102400;
-  options.max_bytes_for_level_multiplier = 4;
-  options.max_background_compactions = 1;
-  options.num_levels = 5;
-  options.statistics = CreateDBStatistics();
-
-  options.compression_per_level.resize(3);
-  // No compression for L0
-  options.compression_per_level[0] = kNoCompression;
-  // No compression for the Ln whre L0 is compacted to
-  options.compression_per_level[1] = kNoCompression;
-  // Snappy compression for Ln+1
-  options.compression_per_level[2] = kSnappyCompression;
-
-  OnFileDeletionListener* listener = new OnFileDeletionListener();
-  options.listeners.emplace_back(listener);
-
-  DestroyAndReopen(options);
-
-  // Insert more than 80K. L4 should be base level. Neither L0 nor L4 should
-  // be compressed, so there shouldn't be any compression.
-  for (int i = 0; i < 20; i++) {
-    ASSERT_OK(Put(Key(keys[i]), CompressibleString(&rnd, 4000)));
-    ASSERT_OK(dbfull()->TEST_WaitForBackgroundWork());
-  }
-  ASSERT_OK(Flush());
-  ASSERT_OK(dbfull()->TEST_WaitForCompact());
-
-  ASSERT_EQ(NumTableFilesAtLevel(1), 0);
-  ASSERT_EQ(NumTableFilesAtLevel(2), 0);
-  ASSERT_EQ(NumTableFilesAtLevel(3), 0);
-  ASSERT_TRUE(NumTableFilesAtLevel(0) > 0 || NumTableFilesAtLevel(4) > 0);
-
-  // Verify there was no compression
-  auto num_block_compressed =
-      options.statistics->getTickerCount(NUMBER_BLOCK_COMPRESSED);
-  ASSERT_EQ(num_block_compressed, 0);
-
-  // Insert 400KB and there will be some files end up in L3. According to the
-  // above compression settings for each level, there will be some compression.
-  ASSERT_OK(options.statistics->Reset());
-  ASSERT_EQ(num_block_compressed, 0);
-  for (int i = 20; i < 120; i++) {
-    ASSERT_OK(Put(Key(keys[i]), CompressibleString(&rnd, 4000)));
-    ASSERT_OK(dbfull()->TEST_WaitForBackgroundWork());
-  }
-  ASSERT_OK(Flush());
-  ASSERT_OK(dbfull()->TEST_WaitForCompact());
-  ASSERT_EQ(NumTableFilesAtLevel(1), 0);
-  ASSERT_EQ(NumTableFilesAtLevel(2), 0);
-  ASSERT_GE(NumTableFilesAtLevel(3), 1);
-  ASSERT_GE(NumTableFilesAtLevel(4), 1);
-
-  // Verify there was compression
-  num_block_compressed =
-      options.statistics->getTickerCount(NUMBER_BLOCK_COMPRESSED);
-  ASSERT_GT(num_block_compressed, 0);
-
-  // Make sure data in files in L3 is not compacted by removing all files
-  // in L4 and calculate number of rows
-  ASSERT_OK(dbfull()->SetOptions({
-      {"disable_auto_compactions", "true"},
-  }));
-  ColumnFamilyMetaData cf_meta;
-  db_->GetColumnFamilyMetaData(&cf_meta);
-
-  // Ensure that L1+ files are non-overlapping and together with L0 encompass
-  // full key range between smallestkey and largestkey from CF file metadata.
-  int largestkey_in_prev_level = -1;
-  int keys_found = 0;
-  for (int level = (int)cf_meta.levels.size() - 1; level >= 0; level--) {
-    int files_in_level = (int)cf_meta.levels[level].files.size();
-    int largestkey_in_prev_file = -1;
-    for (int j = 0; j < files_in_level; j++) {
-      int smallestkey = IdFromKey(cf_meta.levels[level].files[j].smallestkey);
-      int largestkey = IdFromKey(cf_meta.levels[level].files[j].largestkey);
-      int num_entries = (int)cf_meta.levels[level].files[j].num_entries;
-      ASSERT_EQ(num_entries, largestkey - smallestkey + 1);
-      keys_found += num_entries;
-      if (level > 0) {
-        if (j == 0) {
-          ASSERT_GT(smallestkey, largestkey_in_prev_level);
-        }
-        if (j > 0) {
-          ASSERT_GT(smallestkey, largestkey_in_prev_file);
-        }
-        if (j == files_in_level - 1) {
-          largestkey_in_prev_level = largestkey;
-        }
-      }
-      largestkey_in_prev_file = largestkey;
-    }
-  }
-  ASSERT_EQ(keys_found, kNKeys);
-
-  for (const auto& file : cf_meta.levels[4].files) {
-    listener->SetExpectedFileName(dbname_ + file.name);
-    const RangeOpt ranges(file.smallestkey, file.largestkey);
-    // Given verification from above, we're guaranteed that by deleting all the
-    // files in [<smallestkey>, <largestkey>] range, we're effectively deleting
-    // that very single file and nothing more.
-    EXPECT_OK(dbfull()->DeleteFilesInRanges(dbfull()->DefaultColumnFamily(),
-                                            &ranges, true /* include_end */));
-  }
-  listener->VerifyMatchedCount(cf_meta.levels[4].files.size());
-
-  int num_keys = 0;
-  std::unique_ptr<Iterator> iter(db_->NewIterator(ReadOptions()));
-  for (iter->SeekToFirst(); iter->Valid(); iter->Next()) {
-    num_keys++;
-  }
-  ASSERT_OK(iter->status());
-
-  ASSERT_EQ(NumTableFilesAtLevel(1), 0);
-  ASSERT_EQ(NumTableFilesAtLevel(2), 0);
-  ASSERT_GE(NumTableFilesAtLevel(3), 1);
-  ASSERT_EQ(NumTableFilesAtLevel(4), 0);
-
-  ASSERT_GT(SizeAtLevel(0) + SizeAtLevel(3), num_keys * 4000U + num_keys * 10U);
-}
-
-TEST_F(DBTest, DynamicLevelCompressionPerLevel2) {
-  if (!Snappy_Supported() || !LZ4_Supported() || !Zlib_Supported()) {
-    return;
-  }
-  const int kNKeys = 500;
-  int keys[kNKeys];
-  for (int i = 0; i < kNKeys; i++) {
-    keys[i] = i;
-  }
-  RandomShuffle(std::begin(keys), std::end(keys));
-
-  Random rnd(301);
-  Options options;
-  options.create_if_missing = true;
-  options.db_write_buffer_size = 6000000;
-  options.write_buffer_size = 600000;
-  options.max_write_buffer_number = 2;
-  options.level0_file_num_compaction_trigger = 2;
-  options.level0_slowdown_writes_trigger = 2;
-  options.level0_stop_writes_trigger = 2;
-  options.soft_pending_compaction_bytes_limit = 1024 * 1024;
-  options.target_file_size_base = 20;
-  options.env = env_;
-  options.level_compaction_dynamic_level_bytes = true;
-  options.max_bytes_for_level_base = 200;
-  options.max_bytes_for_level_multiplier = 8;
-  options.max_background_compactions = 1;
-  options.num_levels = 5;
-  std::shared_ptr<mock::MockTableFactory> mtf(new mock::MockTableFactory);
-  options.table_factory = mtf;
-
-  options.compression_per_level.resize(3);
-  options.compression_per_level[0] = kNoCompression;
-  options.compression_per_level[1] = kLZ4Compression;
-  options.compression_per_level[2] = kZlibCompression;
-
-  DestroyAndReopen(options);
-  // When base level is L4, L4 is LZ4.
-  std::atomic<int> num_zlib(0);
-  std::atomic<int> num_lz4(0);
-  std::atomic<int> num_no(0);
-  ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->SetCallBack(
-      "LevelCompactionPicker::PickCompaction:Return", [&](void* arg) {
-        Compaction* compaction = static_cast<Compaction*>(arg);
-        if (compaction->output_level() == 4) {
-          ASSERT_TRUE(compaction->output_compression() == kLZ4Compression);
-          num_lz4.fetch_add(1);
-        }
-      });
-  ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->SetCallBack(
-      "FlushJob::WriteLevel0Table:output_compression", [&](void* arg) {
-        auto* compression = static_cast<CompressionType*>(arg);
-        ASSERT_TRUE(*compression == kNoCompression);
-        num_no.fetch_add(1);
-      });
-  ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->EnableProcessing();
-
-  for (int i = 0; i < 100; i++) {
-    std::string value = rnd.RandomString(200);
-    ASSERT_OK(Put(Key(keys[i]), value));
-    if (i % 25 == 24) {
-      ASSERT_OK(Flush());
-      ASSERT_OK(dbfull()->TEST_WaitForCompact());
-    }
-  }
-
-  ASSERT_OK(Flush());
-  ASSERT_OK(dbfull()->TEST_WaitForFlushMemTable());
-  ASSERT_OK(dbfull()->TEST_WaitForCompact());
-  ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->DisableProcessing();
-  ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->ClearAllCallBacks();
-
-  ASSERT_EQ(NumTableFilesAtLevel(1), 0);
-  ASSERT_EQ(NumTableFilesAtLevel(2), 0);
-  ASSERT_EQ(NumTableFilesAtLevel(3), 0);
-  ASSERT_GT(NumTableFilesAtLevel(4), 0);
-  ASSERT_GT(num_no.load(), 2);
-  ASSERT_GT(num_lz4.load(), 0);
-  int prev_num_files_l4 = NumTableFilesAtLevel(4);
-
-  // After base level turn L4->L3, L3 becomes LZ4 and L4 becomes Zlib
-  num_lz4.store(0);
-  num_no.store(0);
-  ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->SetCallBack(
-      "LevelCompactionPicker::PickCompaction:Return", [&](void* arg) {
-        Compaction* compaction = static_cast<Compaction*>(arg);
-        if (compaction->output_level() == 4 && compaction->start_level() == 3) {
-          ASSERT_TRUE(compaction->output_compression() == kZlibCompression);
-          num_zlib.fetch_add(1);
-        } else {
-          ASSERT_TRUE(compaction->output_compression() == kLZ4Compression);
-          num_lz4.fetch_add(1);
-        }
-      });
-  ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->SetCallBack(
-      "FlushJob::WriteLevel0Table:output_compression", [&](void* arg) {
-        auto* compression = static_cast<CompressionType*>(arg);
-        ASSERT_TRUE(*compression == kNoCompression);
-        num_no.fetch_add(1);
-      });
-  ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->EnableProcessing();
-
-  for (int i = 101; i < 500; i++) {
-    std::string value = rnd.RandomString(200);
-    ASSERT_OK(Put(Key(keys[i]), value));
-    if (i % 100 == 99) {
-      ASSERT_OK(Flush());
-      ASSERT_OK(dbfull()->TEST_WaitForCompact());
-    }
-  }
-
-  ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->ClearAllCallBacks();
-  ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->DisableProcessing();
-  ASSERT_EQ(NumTableFilesAtLevel(1), 0);
-  ASSERT_EQ(NumTableFilesAtLevel(2), 0);
-  ASSERT_GT(NumTableFilesAtLevel(3), 0);
-  ASSERT_GT(NumTableFilesAtLevel(4), prev_num_files_l4);
-  ASSERT_GT(num_no.load(), 2);
-  ASSERT_GT(num_lz4.load(), 0);
-  ASSERT_GT(num_zlib.load(), 0);
-}
-
 TEST_F(DBTest, DynamicCompactionOptions) {
  // minimum write buffer size is enforced at 64KB
  const uint64_t k32KB = 1 << 15;
@@ -7349,27 +7078,6 @@ TEST_F(DBTest, LastWriteBufferDelay) {
 }
 #endif  // !defined(ROCKSDB_DISABLE_STALL_NOTIFICATION)

-TEST_F(DBTest, FailWhenCompressionNotSupportedTest) {
-  CompressionType compressions[] = {kZlibCompression, kBZip2Compression,
-                                    kLZ4Compression, kLZ4HCCompression,
-                                    kXpressCompression};
-  for (auto comp : compressions) {
-    if (!CompressionTypeSupported(comp)) {
-      // not supported, we should fail the Open()
-      Options options = CurrentOptions();
-      options.compression = comp;
-      ASSERT_TRUE(!TryReopen(options).ok());
-      // Try if CreateColumnFamily also fails
-      options.compression = kNoCompression;
-      ASSERT_OK(TryReopen(options));
-      ColumnFamilyOptions cf_options(options);
-      cf_options.compression = comp;
-      ColumnFamilyHandle* handle;
-      ASSERT_TRUE(!db_->CreateColumnFamily(cf_options, "name", &handle).ok());
-    }
-  }
-}
-
 TEST_F(DBTest, CreateColumnFamilyShouldFailOnIncompatibleOptions) {
  Options options = CurrentOptions();
  options.max_open_files = 100;
@@ -366,11 +366,6 @@ Options DBTestBase::GetOptions(
    table_options.block_cache = NewLRUCache(/* too small */ 1);
  }

-  // Test anticipated new default as much as reasonably possible (and remove
-  // this code when obsolete)
-  assert(!table_options.decouple_partitioned_filters);
-  table_options.decouple_partitioned_filters = true;
-
  bool can_allow_mmap = IsMemoryMappedAccessSupported();
  switch (option_config) {
    case kHashSkipList:
@@ -1159,16 +1154,18 @@ size_t DBTestBase::CountLiveFiles() {
 }

 int DBTestBase::NumTableFilesAtLevel(int level, int cf) {
-  std::string property;
-  if (cf == 0) {
-    // default cfd
-    EXPECT_TRUE(db_->GetProperty(
-        "rocksdb.num-files-at-level" + std::to_string(level), &property));
-  } else {
-    EXPECT_TRUE(db_->GetProperty(
-        handles_[cf], "rocksdb.num-files-at-level" + std::to_string(level),
-        &property));
+  return NumTableFilesAtLevel(level,
+                              cf ? handles_[cf] : db_->DefaultColumnFamily());
+}
+
+int DBTestBase::NumTableFilesAtLevel(int level, ColumnFamilyHandle* cfh,
+                                     DB* db) {
+  if (!db) {
+    db = db_;
  }
+  std::string property;
+  EXPECT_TRUE(db->GetProperty(
+      cfh, "rocksdb.num-files-at-level" + std::to_string(level), &property));
  return atoi(property.c_str());
 }

@@ -1201,12 +1198,22 @@ int DBTestBase::TotalTableFiles(int cf, int levels) {

 // Return spread of files per level
 std::string DBTestBase::FilesPerLevel(int cf) {
-  int num_levels =
-      (cf == 0) ? db_->NumberLevels() : db_->NumberLevels(handles_[cf]);
+  if (cf == 0) {
+    return FilesPerLevel(db_->DefaultColumnFamily());
+  } else {
+    return FilesPerLevel(handles_[cf]);
+  }
+}
+
+std::string DBTestBase::FilesPerLevel(ColumnFamilyHandle* cfh, DB* db) {
+  if (!db) {
+    db = db_;
+  }
+  int num_levels = db->NumberLevels(cfh);
  std::string result;
  size_t last_non_zero_offset = 0;
  for (int level = 0; level < num_levels; level++) {
-    int f = NumTableFilesAtLevel(level, cf);
+    int f = NumTableFilesAtLevel(level, cfh, db);
    char buf[100];
    snprintf(buf, sizeof(buf), "%s%d", (level ? "," : ""), f);
    result += buf;
@@ -1339,12 +1346,14 @@ void DBTestBase::FillLevels(const std::string& smallest,
 }

 void DBTestBase::MoveFilesToLevel(int level, int cf) {
+  MoveFilesToLevel(level, cf ? handles_[cf] : db_->DefaultColumnFamily());
+}
+
+void DBTestBase::MoveFilesToLevel(int level, ColumnFamilyHandle* column_family,
+                                  DB* db) {
+  DBImpl* db_impl = db ? static_cast<DBImpl*>(db) : dbfull();
  for (int l = 0; l < level; ++l) {
-    if (cf > 0) {
-      EXPECT_OK(dbfull()->TEST_CompactRange(l, nullptr, nullptr, handles_[cf]));
-    } else {
-      EXPECT_OK(dbfull()->TEST_CompactRange(l, nullptr, nullptr));
-    }
+    EXPECT_OK(db_impl->TEST_CompactRange(l, nullptr, nullptr, column_family));
  }
 }

@@ -452,6 +452,10 @@ class SpecialEnv : public EnvWrapper {
        return s;
      }

+      Status GetFileSize(uint64_t* s) override {
+        return target_->GetFileSize(s);
+      }
+
     private:
      std::unique_ptr<RandomAccessFile> target_;
      anon::AtomicCounter* counter_;
@@ -478,6 +482,10 @@ class SpecialEnv : public EnvWrapper {
        return target_->Prefetch(offset, n);
      }

+      Status GetFileSize(uint64_t* s) override {
+        return target_->GetFileSize(s);
+      }
+
     private:
      std::unique_ptr<RandomAccessFile> target_;
      std::atomic<uint64_t>* fail_cnt_;
@@ -1272,6 +1280,9 @@ class DBTestBase : public testing::Test {

  int NumTableFilesAtLevel(int level, int cf = 0);

+  int NumTableFilesAtLevel(int level, ColumnFamilyHandle* column_family,
+                           DB* db = nullptr);
+
  double CompressionRatioAtLevel(int level, int cf = 0);

  int TotalTableFiles(int cf = 0, int levels = -1);
@@ -1281,6 +1292,8 @@ class DBTestBase : public testing::Test {
  // Return spread of files per level
  std::string FilesPerLevel(int cf = 0);

+  std::string FilesPerLevel(ColumnFamilyHandle* cfh, DB* db = nullptr);
+
  size_t CountFiles();

  Status CountFiles(size_t* count);
@@ -1312,6 +1325,9 @@ class DBTestBase : public testing::Test {

  void MoveFilesToLevel(int level, int cf = 0);

+  void MoveFilesToLevel(int level, ColumnFamilyHandle* column_family,
+                        DB* db = nullptr);
+
  void DumpFileCounts(const char* label);

  std::string DumpSSTableList();
@@ -2106,46 +2106,79 @@ TEST_F(DBTestUniversalCompaction2, OverlappingL0) {
 }

 TEST_F(DBTestUniversalCompaction2, IngestBehind) {
-  const int kNumKeys = 3000;
-  const int kWindowSize = 100;
-  const int kNumDelsTrigger = 90;
+  for (bool cf_option : {false, true}) {
+    SCOPED_TRACE("cf_option = " + std::to_string(cf_option));
+    const int kNumKeys = 3000;
+    const int kWindowSize = 100;
+    const int kNumDelsTrigger = 90;

-  Options opts = CurrentOptions();
-  opts.table_properties_collector_factories.emplace_back(
-      NewCompactOnDeletionCollectorFactory(kWindowSize, kNumDelsTrigger));
-  opts.compaction_style = kCompactionStyleUniversal;
-  opts.level0_file_num_compaction_trigger = 2;
-  opts.compression = kNoCompression;
-  opts.allow_ingest_behind = true;
-  opts.compaction_options_universal.size_ratio = 10;
-  opts.compaction_options_universal.min_merge_width = 2;
-  opts.compaction_options_universal.max_size_amplification_percent = 200;
-  Reopen(opts);
-
-  // add an L1 file to prevent tombstones from dropping due to obsolescence
-  // during flush
-  int i;
-  for (i = 0; i < 2000; ++i) {
-    ASSERT_OK(Put(Key(i), "val"));
-  }
-  ASSERT_OK(Flush());
-  //  MoveFilesToLevel(6);
-  ASSERT_OK(dbfull()->CompactRange(CompactRangeOptions(), nullptr, nullptr));
-
-  for (i = 1999; i < kNumKeys; ++i) {
-    if (i >= kNumKeys - kWindowSize &&
-        i < kNumKeys - kWindowSize + kNumDelsTrigger) {
-      ASSERT_OK(Delete(Key(i)));
+    Options opts = CurrentOptions();
+    opts.table_properties_collector_factories.emplace_back(
+        NewCompactOnDeletionCollectorFactory(kWindowSize, kNumDelsTrigger));
+    opts.compaction_style = kCompactionStyleUniversal;
+    opts.level0_file_num_compaction_trigger = 2;
+    opts.compression = kNoCompression;
+    if (cf_option) {
+      opts.cf_allow_ingest_behind = true;
    } else {
+      opts.allow_ingest_behind = true;
+    }
+    opts.compaction_options_universal.size_ratio = 10;
+    opts.compaction_options_universal.min_merge_width = 2;
+    opts.compaction_options_universal.max_size_amplification_percent = 200;
+    Reopen(opts);
+
+    // add an L1 file to prevent tombstones from dropping due to obsolescence
+    // during flush
+    int i;
+    for (i = 0; i < 2000; ++i) {
      ASSERT_OK(Put(Key(i), "val"));
    }
-  }
-  ASSERT_OK(Flush());
+    ASSERT_OK(Flush());
+    //  MoveFilesToLevel(6);
+    ASSERT_OK(dbfull()->CompactRange(CompactRangeOptions(), nullptr, nullptr));

-  ASSERT_OK(dbfull()->TEST_WaitForCompact());
-  ASSERT_EQ(0, NumTableFilesAtLevel(0));
-  ASSERT_EQ(0, NumTableFilesAtLevel(6));
-  ASSERT_GT(NumTableFilesAtLevel(5), 0);
+    for (i = 1999; i < kNumKeys; ++i) {
+      if (i >= kNumKeys - kWindowSize &&
+          i < kNumKeys - kWindowSize + kNumDelsTrigger) {
+        ASSERT_OK(Delete(Key(i)));
+      } else {
+        ASSERT_OK(Put(Key(i), "val"));
+      }
+    }
+    ASSERT_OK(Flush());
+
+    ASSERT_OK(dbfull()->TEST_WaitForCompact());
+    ASSERT_EQ(0, NumTableFilesAtLevel(0));
+    ASSERT_EQ(0, NumTableFilesAtLevel(6));
+    ASSERT_GT(NumTableFilesAtLevel(5), 0);
+
+    if (cf_option) {
+      // Test that another CF does not allow ingest behind
+      ColumnFamilyHandle* new_cfh;
+      Options new_cf_option;
+      new_cf_option.compaction_style = kCompactionStyleUniversal;
+      new_cf_option.num_levels = 7;
+      // CreateColumnFamilies({"new_cf"}, new_cf_option);
+      ASSERT_OK(db_->CreateColumnFamily(new_cf_option, "new_cf", &new_cfh));
+      // handles_.push_back(new_cfh);
+      for (i = 0; i < 10; ++i) {
+        // ASSERT_OK(Put(1, Key(i), "val"));
+        ASSERT_OK(db_->Put(WriteOptions(), new_cfh, Key(i), "val"));
+      }
+      ASSERT_OK(
+          db_->CompactRange(CompactRangeOptions(), new_cfh, nullptr, nullptr));
+      // This CF can use the last leve.
+      std::string property;
+      EXPECT_TRUE(db_->GetProperty(
+          new_cfh, "rocksdb.num-files-at-level" + std::to_string(6),
+          &property));
+      ASSERT_EQ(1, atoi(property.c_str()));
+
+      ASSERT_OK(db_->DropColumnFamily(new_cfh));
+      ASSERT_OK(db_->DestroyColumnFamilyHandle(new_cfh));
+    }
+  }
 }

 TEST_F(DBTestUniversalCompaction2, PeriodicCompactionDefault) {
@@ -2400,7 +2400,6 @@ class DataVisibilityTest : public DBBasicTestWithTimestampBase {
    }
  }
 };
-constexpr int DataVisibilityTest::kTestDataSize;

 // Application specifies timestamp but not snapshot.
 //           reader              writer
@@ -88,11 +88,12 @@ Status GetFileChecksumsFromCurrentManifest(FileSystem* fs,

  // Read all records from the manifest file...
  uint64_t manifest_file_size = std::numeric_limits<uint64_t>::max();
-  FileChecksumRetriever retriever(read_options, manifest_file_size,
-                                  *checksum_list);
+  FileChecksumRetriever retriever(read_options, manifest_file_size);
  retriever.Iterate(reader, &s);
-
-  return retriever.status();
+  if (!retriever.status().ok()) {
+    return retriever.status();
+  }
+  return retriever.FetchFileChecksumList(*checksum_list);
 }

 Status UpdateManifestForFilesState(
@@ -2567,7 +2567,14 @@ TEST_F(ExternalSSTFileBasicTest, IngestWithTemperature) {
    options.default_write_temperature = Temperature::kHot;
    SstFileWriter sst_file_writer(EnvOptions(), options);
    options.level0_file_num_compaction_trigger = 2;
-    options.allow_ingest_behind = (mode == "ingest_behind");
+    bool cf_option = Random::GetTLSInstance()->OneIn(2);
+    SCOPED_TRACE(std::string("Use ") + (cf_option ? "CF" : "DB") +
+                 " option for ingest behind");
+    if (cf_option) {
+      options.cf_allow_ingest_behind = (mode == "ingest_behind");
+    } else {
+      options.allow_ingest_behind = (mode == "ingest_behind");
+    }
    Reopen(options);
    Defer destroyer([&]() { Destroy(options); });

@@ -122,24 +122,28 @@ Status ExternalSstFileIngestionJob::Prepare(
    }
  }

-  if (ingestion_options_.ingest_behind && files_overlap_) {
-    return Status::NotSupported(
-        "Files with overlapping ranges cannot be ingested with ingestion "
-        "behind mode.");
-  }
+  if (files_overlap_) {
+    if (ingestion_options_.ingest_behind) {
+      return Status::NotSupported(
+          "Files with overlapping ranges cannot be ingested with ingestion "
+          "behind mode.");
+    }

-  // Overlapping files need at least two different sequence numbers. If settings
-  // disables global seqno, ingestion will fail anyway, so fail fast in prepare.
-  if (!ingestion_options_.allow_global_seqno && files_overlap_) {
-    return Status::InvalidArgument(
-        "Global seqno is required, but disabled (because external files key "
-        "range overlaps).");
-  }
+    // Overlapping files need at least two different sequence numbers. If
+    // settings disables global seqno, ingestion will fail anyway, so fail
+    // fast in prepare.
+    if (!ingestion_options_.allow_global_seqno &&
+        !ingestion_options_.allow_db_generated_files) {
+      return Status::InvalidArgument(
+          "Global seqno is required, but disabled (because external files key "
+          "range overlaps).");
+    }

-  if (ucmp_->timestamp_size() > 0 && files_overlap_) {
-    return Status::NotSupported(
-        "Files with overlapping ranges cannot be ingested to column "
-        "family with user-defined timestamp enabled.");
+    if (ucmp_->timestamp_size() > 0) {
+      return Status::NotSupported(
+          "Files with overlapping ranges cannot be ingested to column "
+          "family with user-defined timestamp enabled.");
+    }
  }

  // Copy/Move external files into DB
@@ -156,6 +160,14 @@ Status ExternalSstFileIngestionJob::Prepare(
        // It is unsafe to assume application had sync the file and file
        // directory before ingest the file. For integrity of RocksDB we need
        // to sync the file.
+
+        // TODO(xingbo), We should in general be moving away from production
+        // uses of ReuseWritableFile (except explicitly for WAL recycling),
+        // ReopenWritableFile, and NewRandomRWFile. We should create a
+        // FileSystem::SyncFile/FsyncFile API that by default does the
+        // re-open+sync+close combo but can (a) be reused easily, and (b) be
+        // overridden to do that more cleanly, e.g. in EncryptedEnv.
+        // https://github.com/facebook/rocksdb/issues/13741
        std::unique_ptr<FSWritableFile> file_to_sync;
        Status s = fs_->ReopenWritableFile(path_inside_db, env_options_,
                                           &file_to_sync, nullptr);
@@ -532,6 +544,8 @@ Status ExternalSstFileIngestionJob::Run() {

  // Find levels to ingest into
  std::optional<int> prev_batch_uppermost_level;
+  // batches at the front of file_batches_to_ingest_ contains older updates and
+  // are placed in smaller levels.
  for (auto& batch : file_batches_to_ingest_) {
    int batch_uppermost_level = 0;
    status = AssignLevelsForOneBatch(batch, super_version, force_global_seqno,
@@ -582,6 +596,15 @@ Status ExternalSstFileIngestionJob::AssignLevelsForOneBatch(
    if (!status.ok()) {
      return status;
    }
+
+    // If any ingested file overlaps with the DB, it will fail here.
+    if (ingestion_options_.allow_db_generated_files && assigned_seqno != 0) {
+      return Status::InvalidArgument(
+          "An ingested file overlaps with existing data in the DB and has been "
+          "assigned a non-zero sequence number, which is not allowed when "
+          "'allow_db_generated_files' is enabled.");
+    }
+
    if (smallest_parsed.sequence == 0 && assigned_seqno != 0) {
      UpdateInternalKey(file->smallest_internal_key.rep(), assigned_seqno,
                        smallest_parsed.type);
@@ -600,8 +623,8 @@ Status ExternalSstFileIngestionJob::AssignLevelsForOneBatch(
    assert(assigned_seqno == 0 || assigned_seqno == *last_seqno + 1);
    if (assigned_seqno > *last_seqno) {
      *last_seqno = assigned_seqno;
-      ++consumed_seqno_count_;
    }
+    max_assigned_seqno_ = std::max(max_assigned_seqno_, assigned_seqno);

    status = GenerateChecksumForIngestedFile(file);
    if (!status.ok()) {
@@ -624,15 +647,24 @@ Status ExternalSstFileIngestionJob::AssignLevelsForOneBatch(
        file->table_properties.num_range_deletions == 1 &&
        (file->table_properties.num_entries ==
         file->table_properties.num_range_deletions);
+    SequenceNumber smallest_seqno = file->assigned_seqno;
+    SequenceNumber largest_seqno = file->assigned_seqno;
+    if (ingestion_options_.allow_db_generated_files) {
+      assert(file->assigned_seqno == 0);
+      assert(file->smallest_seqno != kMaxSequenceNumber);
+      assert(file->largest_seqno != kMaxSequenceNumber);
+      smallest_seqno = file->smallest_seqno;
+      largest_seqno = file->largest_seqno;
+      max_assigned_seqno_ = std::max(max_assigned_seqno_, file->largest_seqno);
+    }
    FileMetaData f_metadata(
        file->fd.GetNumber(), file->fd.GetPathId(), file->fd.GetFileSize(),
-        file->smallest_internal_key, file->largest_internal_key,
-        file->assigned_seqno, file->assigned_seqno, false,
-        file->file_temperature, kInvalidBlobFileNumber, oldest_ancester_time,
-        current_time,
+        file->smallest_internal_key, file->largest_internal_key, smallest_seqno,
+        largest_seqno, false, file->file_temperature, kInvalidBlobFileNumber,
+        oldest_ancester_time, current_time,
        ingestion_options_.ingest_behind
            ? kReservedEpochNumberForFileIngestedBehind
-            : cfd_->NewEpochNumber(),
+            : cfd_->NewEpochNumber(),  // orders files ingested to L0
        file->file_checksum, file->file_checksum_func_name, file->unique_id, 0,
        tail_size, file->user_defined_timestamps_persisted);
    f_metadata.temperature = file->file_temperature;
@@ -689,10 +721,9 @@ void ExternalSstFileIngestionJob::CreateEquivalentFileIngestingCompactions() {
        0 /* max_subcompaction, not applicable */,
        {} /* grandparents, not applicable */,
        std::nullopt /* earliest_snapshot */, nullptr /* snapshot_checker */,
-        false /* is manual */, "" /* trim_ts */, -1 /* score, not applicable */,
-        false /* is deletion compaction, not applicable */,
-        files_overlap_ /* l0_files_might_overlap, not applicable */,
-        CompactionReason::kExternalSstIngestion));
+        CompactionReason::kExternalSstIngestion, "" /* trim_ts */,
+        -1 /* score, not applicable */,
+        files_overlap_ /* l0_files_might_overlap, not applicable */));
  }
 }

@@ -780,7 +811,6 @@ void ExternalSstFileIngestionJob::Cleanup(const Status& status) {
    // We failed to add the files to the database
    // remove all the files we copied
    DeleteInternalFiles();
-    consumed_seqno_count_ = 0;
    files_overlap_ = false;
  } else if (status.ok() && ingestion_options_.move_files) {
    // The files were moved and added successfully, remove original file links
@@ -994,6 +1024,32 @@ Status ExternalSstFileIngestionJob::GetIngestedFileInfo(
    return status;
  }

+  const bool allow_data_in_errors = db_options_.allow_data_in_errors;
+  ParsedInternalKey key;
+  if (ingestion_options_.allow_db_generated_files) {
+    // We are ingesting a DB generated SST file for which we don't reassign
+    // sequence numbers. We need its smallest sequence number and largest
+    // sequence number for FileMetaData.
+    Status seqno_status = GetSeqnoBoundaryForFile(
+        table_reader.get(), sv, file_to_ingest, allow_data_in_errors);
+
+    if (!seqno_status.ok()) {
+      return seqno_status;
+    }
+    assert(file_to_ingest->smallest_seqno <= file_to_ingest->largest_seqno);
+    assert(file_to_ingest->largest_seqno < kMaxSequenceNumber);
+  } else {
+    SequenceNumber largest_seqno =
+        table_reader.get()->GetTableProperties()->key_largest_seqno;
+    // UINT64_MAX means unknown and the file is generated before table property
+    // `key_largest_seqno` is introduced.
+    if (largest_seqno != UINT64_MAX && largest_seqno > 0) {
+      return Status::Corruption(
+          "External file has non zero largest sequence number " +
+          std::to_string(largest_seqno));
+    }
+  }
+
  if (ingestion_options_.verify_checksums_before_ingest) {
    // If customized readahead size is needed, we can pass a user option
    // all the way to here. Right now we just rely on the default readahead
@@ -1009,7 +1065,6 @@ Status ExternalSstFileIngestionJob::GetIngestedFileInfo(
    }
  }

-  ParsedInternalKey key;
  // TODO: plumb Env::IOActivity, Env::IOPriority
  ReadOptions ro;
  ro.fill_cache = ingestion_options_.fill_cache;
@@ -1018,7 +1073,6 @@ Status ExternalSstFileIngestionJob::GetIngestedFileInfo(
      /*skip_filters=*/false, TableReaderCaller::kExternalSSTIngestion));

  // Get first (smallest) and last (largest) key from file.
-  bool allow_data_in_errors = db_options_.allow_data_in_errors;
  iter->SeekToFirst();
  if (iter->Valid()) {
    Status pik_status =
@@ -1027,7 +1081,7 @@ Status ExternalSstFileIngestionJob::GetIngestedFileInfo(
      return Status::Corruption("Corrupted key in external file. ",
                                pik_status.getState());
    }
-    if (key.sequence != 0) {
+    if (key.sequence != 0 && !ingestion_options_.allow_db_generated_files) {
      return Status::Corruption("External file has non zero sequence number");
    }
    file_to_ingest->smallest_internal_key.SetFrom(key);
@@ -1064,41 +1118,13 @@ Status ExternalSstFileIngestionJob::GetIngestedFileInfo(
      return Status::Corruption("Corrupted key in external file. ",
                                pik_status.getState());
    }
-    if (key.sequence != 0) {
+    if (key.sequence != 0 && !ingestion_options_.allow_db_generated_files) {
      return Status::Corruption("External file has non zero sequence number");
    }
    file_to_ingest->largest_internal_key.SetFrom(key);
  } else if (!iter->status().ok()) {
    return iter->status();
  }
-  SequenceNumber largest_seqno =
-      table_reader.get()->GetTableProperties()->key_largest_seqno;
-  // UINT64_MAX means unknown and the file is generated before table property
-  // `key_largest_seqno` is introduced.
-  if (largest_seqno != UINT64_MAX && largest_seqno > 0) {
-    return Status::Corruption(
-        "External file has non zero largest sequence number " +
-        std::to_string(largest_seqno));
-  }
-  if (ingestion_options_.allow_db_generated_files &&
-      largest_seqno == UINT64_MAX) {
-    // Need to verify that all keys have seqno zero.
-    for (iter->SeekToFirst(); iter->Valid(); iter->Next()) {
-      Status pik_status =
-          ParseInternalKey(iter->key(), &key, allow_data_in_errors);
-      if (!pik_status.ok()) {
-        return Status::Corruption("Corrupted key in external file. ",
-                                  pik_status.getState());
-      }
-      if (key.sequence != 0) {
-        return Status::NotSupported(
-            "External file has a key with non zero sequence number.");
-      }
-    }
-    if (!iter->status().ok()) {
-      return iter->status();
-    }
-  }

  std::unique_ptr<InternalIterator> range_del_iter(
      table_reader->NewRangeTombstoneIterator(ro));
@@ -1113,7 +1139,7 @@ Status ExternalSstFileIngestionJob::GetIngestedFileInfo(
        return Status::Corruption("Corrupted key in external file. ",
                                  pik_status.getState());
      }
-      if (key.sequence != 0) {
+      if (key.sequence != 0 && !ingestion_options_.allow_db_generated_files) {
        return Status::Corruption(
            "External file has a range deletion with non zero sequence "
            "number.");
@@ -1161,12 +1187,14 @@ Status ExternalSstFileIngestionJob::AssignLevelAndSeqnoForIngestedFile(
  const size_t ts_sz = ucmp_->timestamp_size();
  assert(!prev_batch_uppermost_level.has_value() ||
         prev_batch_uppermost_level.value() < cfd_->NumberLevels());
-  bool must_assign_to_l0 = prev_batch_uppermost_level.has_value() &&
-                           prev_batch_uppermost_level.value() == 0;
-  if (force_global_seqno || files_overlap_ ||
-      compaction_style == kCompactionStyleFIFO || must_assign_to_l0) {
+  bool must_assign_to_l0 = (prev_batch_uppermost_level.has_value() &&
+                            prev_batch_uppermost_level.value() == 0) ||
+                           compaction_style == kCompactionStyleFIFO;
+
+  if (force_global_seqno || (!ingestion_options_.allow_db_generated_files &&
+                             (files_overlap_ || must_assign_to_l0))) {
    *assigned_seqno = last_seqno + 1;
-    if (compaction_style == kCompactionStyleFIFO || must_assign_to_l0) {
+    if (must_assign_to_l0) {
      assert(ts_sz == 0);
      file_to_ingest->picked_level = 0;
      if (ingestion_options_.fail_if_not_bottommost_level &&
@@ -1187,16 +1215,26 @@ Status ExternalSstFileIngestionJob::AssignLevelAndSeqnoForIngestedFile(
  ro.total_order_seek = true;
  int target_level = 0;
  auto* vstorage = cfd_->current()->storage_info();
-  assert(!must_assign_to_l0);
-  int exclusive_end_level = prev_batch_uppermost_level.has_value()
-                                ? prev_batch_uppermost_level.value()
-                                : cfd_->NumberLevels();
+  assert(!must_assign_to_l0 || ingestion_options_.allow_db_generated_files);
+  int assigned_level_exclusive_end = cfd_->NumberLevels();
+  if (must_assign_to_l0) {
+    assigned_level_exclusive_end = 0;
+  } else if (prev_batch_uppermost_level.has_value()) {
+    assigned_level_exclusive_end = prev_batch_uppermost_level.value();
+  }

-  for (int lvl = 0; lvl < exclusive_end_level; lvl++) {
+  // When ingesting db generated files, we require that ingested files do not
+  // overlap with any file in the DB. So we need to check all levels.
+  int overlap_checking_exclusive_end =
+      ingestion_options_.allow_db_generated_files
+          ? cfd_->NumberLevels()
+          : assigned_level_exclusive_end;
+  for (int lvl = 0; lvl < overlap_checking_exclusive_end; lvl++) {
    if (lvl > 0 && lvl < vstorage->base_level()) {
      continue;
    }
-    if (atomic_replace_range_.has_value()) {
+    if (lvl < assigned_level_exclusive_end &&
+        atomic_replace_range_.has_value()) {
      target_level = lvl;
      continue;
    }
@@ -1227,7 +1265,8 @@ Status ExternalSstFileIngestionJob::AssignLevelAndSeqnoForIngestedFile(

    // We don't overlap with any keys in this level, but we still need to check
    // if our file can fit in it
-    if (IngestedFileFitInLevel(file_to_ingest, lvl)) {
+    if (lvl < assigned_level_exclusive_end &&
+        IngestedFileFitInLevel(file_to_ingest, lvl)) {
      target_level = lvl;
    }
  }
@@ -1236,8 +1275,9 @@ Status ExternalSstFileIngestionJob::AssignLevelAndSeqnoForIngestedFile(
      target_level < cfd_->NumberLevels() - 1) {
    status = Status::TryAgain(
        "Files cannot be ingested to Lmax. Please make sure key range of Lmax "
-        "and ongoing compaction's output to Lmax"
-        "does not overlap with files to ingest.");
+        "and ongoing compaction's output to Lmax does not overlap with files "
+        "to ingest. Input files overlapping with each other can cause some "
+        "file to be assigned to non Lmax level.");
    return status;
  }

@@ -1258,11 +1298,6 @@ Status ExternalSstFileIngestionJob::AssignLevelAndSeqnoForIngestedFile(
    }
  }

-  if (ingestion_options_.allow_db_generated_files && *assigned_seqno != 0) {
-    return Status::InvalidArgument(
-        "An ingested file is assigned to a non-zero sequence number, which is "
-        "incompatible with ingestion option allow_db_generated_files.");
-  }
  return status;
 }

@@ -1279,13 +1314,13 @@ Status ExternalSstFileIngestionJob::CheckLevelForIngestedBehindFile(
        "at the last level!");
  }

-  // Second, check if despite allow_ingest_behind=true we still have 0 seqnums
-  // at some upper level
+  // Second, check if despite cf_allow_ingest_behind=true we still have 0
+  // seqnums at some upper level
  for (int lvl = 0; lvl < cfd_->NumberLevels() - 1; lvl++) {
    for (auto file : vstorage->LevelFiles(lvl)) {
      if (file->fd.smallest_seqno == 0) {
        return Status::InvalidArgument(
-            "Can't ingest_behind file as despite allow_ingest_behind=true "
+            "Can't ingest_behind file as despite cf_allow_ingest_behind=true "
            "there are files with 0 seqno in database at upper levels!");
      }
    }
@@ -1297,8 +1332,12 @@ Status ExternalSstFileIngestionJob::CheckLevelForIngestedBehindFile(

 Status ExternalSstFileIngestionJob::AssignGlobalSeqnoForIngestedFile(
    IngestedFileInfo* file_to_ingest, SequenceNumber seqno) {
+  if (ingestion_options_.allow_db_generated_files) {
+    assert(seqno == 0);
+    assert(file_to_ingest->original_seqno == 0);
+  }
  if (file_to_ingest->original_seqno == seqno) {
-    // This file already have the correct global seqno
+    // This file already has the correct global seqno.
    return Status::OK();
  } else if (!ingestion_options_.allow_global_seqno) {
    return Status::InvalidArgument("Global seqno is required, but disabled");
@@ -1412,4 +1451,86 @@ Status ExternalSstFileIngestionJob::SyncIngestedFile(TWritableFile* file) {
  }
 }

+Status ExternalSstFileIngestionJob::GetSeqnoBoundaryForFile(
+    TableReader* table_reader, SuperVersion* sv,
+    IngestedFileInfo* file_to_ingest, bool allow_data_in_errors) {
+  const bool has_largest_seqno =
+      table_reader->GetTableProperties()->HasKeyLargestSeqno();
+  SequenceNumber largest_seqno =
+      table_reader->GetTableProperties()->key_largest_seqno;
+  if (has_largest_seqno && largest_seqno == 0) {
+    file_to_ingest->largest_seqno = 0;
+    file_to_ingest->smallest_seqno = 0;
+    return Status::OK();
+  }
+  // The following file scan is only executed when ingesting files with
+  // non-zero seqno.
+  // TODO: record smallest_seqno in table properties to avoid the
+  // file scan here.
+  SequenceNumber smallest_seqno = kMaxSequenceNumber;
+
+  SequenceNumber largest_seqno_from_iter = 0;
+  ReadOptions ro;
+  ro.fill_cache = ingestion_options_.fill_cache;
+  std::unique_ptr<InternalIterator> iter(table_reader->NewIterator(
+      ro, sv->mutable_cf_options.prefix_extractor.get(), /*arena=*/nullptr,
+      /*skip_filters=*/false, TableReaderCaller::kExternalSSTIngestion));
+  ParsedInternalKey key;
+  iter->SeekToFirst();
+  while (iter->Valid()) {
+    Status pik_status =
+        ParseInternalKey(iter->key(), &key, allow_data_in_errors);
+    if (!pik_status.ok()) {
+      return Status::Corruption("Corrupted key in external file. ",
+                                pik_status.getState());
+    }
+    smallest_seqno = std::min(smallest_seqno, key.sequence);
+    largest_seqno_from_iter = std::max(largest_seqno_from_iter, key.sequence);
+    iter->Next();
+  }
+  if (!iter->status().ok()) {
+    return iter->status();
+  }
+
+  if (table_reader->GetTableProperties()->num_range_deletions > 0) {
+    std::unique_ptr<InternalIterator> range_del_iter(
+        table_reader->NewRangeTombstoneIterator(ro));
+    if (range_del_iter != nullptr) {
+      for (range_del_iter->SeekToFirst(); range_del_iter->Valid();
+           range_del_iter->Next()) {
+        Status pik_status =
+            ParseInternalKey(range_del_iter->key(), &key, allow_data_in_errors);
+        if (!pik_status.ok()) {
+          return Status::Corruption("Corrupted key in external file. ",
+                                    pik_status.getState());
+        }
+        smallest_seqno = std::min(smallest_seqno, key.sequence);
+        largest_seqno_from_iter =
+            std::max(largest_seqno_from_iter, key.sequence);
+      }
+      if (!range_del_iter->status().ok()) {
+        return range_del_iter->status();
+      }
+    }
+  }
+
+  file_to_ingest->smallest_seqno = smallest_seqno;
+  if (!has_largest_seqno) {
+    file_to_ingest->largest_seqno = largest_seqno_from_iter;
+  } else {
+    assert(largest_seqno == largest_seqno_from_iter);
+    file_to_ingest->largest_seqno = largest_seqno;
+  }
+
+  if (file_to_ingest->largest_seqno == kMaxSequenceNumber) {
+    return Status::InvalidArgument(
+        "Unknown smallest seqno for db generated file.");
+  }
+  if (file_to_ingest->smallest_seqno == kMaxSequenceNumber) {
+    return Status::InvalidArgument(
+        "Unknown largest seqno for db generated file.");
+  }
+  return Status::OK();
+}
+
 }  // namespace ROCKSDB_NAMESPACE
@@ -180,6 +180,9 @@ struct IngestedFileInfo : public KeyRangeInfo {
  // the user key's format in the external file matches the column family's
  // setting.
  bool user_defined_timestamps_persisted = true;
+
+  SequenceNumber largest_seqno = kMaxSequenceNumber;
+  SequenceNumber smallest_seqno = kMaxSequenceNumber;
 };

 // A batch of files.
@@ -230,7 +233,7 @@ class ExternalSstFileIngestionJob {
        directories_(directories),
        event_logger_(event_logger),
        job_start_time_(clock_->NowMicros()),
-        consumed_seqno_count_(0),
+        max_assigned_seqno_(0),
        io_tracer_(io_tracer) {
    assert(directories != nullptr);
    assert(cfd_);
@@ -287,8 +290,16 @@ class ExternalSstFileIngestionJob {
    return files_to_ingest_;
  }

-  // How many sequence numbers did we consume as part of the ingestion job?
-  int ConsumedSequenceNumbersCount() const { return consumed_seqno_count_; }
+  // Return the maximum assigned sequence number for all files in this job.
+  // When allow_db_generated_files = false, we may assign global sequence
+  // numbers to ingested files. The global sequence numbers are sequence numbers
+  // following versions_->LastSequence().
+  // When allow_db_generated_files = true, we ingest files that already have
+  // sequence numbers assigned. max_assigned_seqno_ will be the max sequence
+  // number among ingested files.
+  SequenceNumber MaxAssignedSequenceNumber() const {
+    return max_assigned_seqno_;
+  }

 private:
  Status ResetTableReader(const std::string& external_file,
@@ -349,7 +360,7 @@ class ExternalSstFileIngestionJob {
      std::optional<int> prev_batch_uppermost_level);

  // File that we want to ingest behind always goes to the lowest level;
-  // we just check that it fits in the level, that DB allows ingest_behind,
+  // we just check that it fits in the level, that the CF allows ingest_behind,
  // and that we don't have 0 seqnums at the upper levels.
  // REQUIRES: Mutex held
  Status CheckLevelForIngestedBehindFile(IngestedFileInfo* file_to_ingest);
@@ -369,6 +380,13 @@ class ExternalSstFileIngestionJob {
  template <typename TWritableFile>
  Status SyncIngestedFile(TWritableFile* file);

+  // Helper function to obtain the smallest and largest sequence number from a
+  // file. When OK is returned, file_to_ingest->smallest_seqno and
+  // file_to_ingest->largest_seqno will be updated.
+  Status GetSeqnoBoundaryForFile(TableReader* table_reader, SuperVersion* sv,
+                                 IngestedFileInfo* file_to_ingest,
+                                 bool allow_data_in_errors);
+
  // Create equivalent `Compaction` objects to this file ingestion job
  // , which will be used to check range conflict with other ongoing
  // compactions.
@@ -395,7 +413,7 @@ class ExternalSstFileIngestionJob {
  EventLogger* event_logger_;
  VersionEdit edit_;
  uint64_t job_start_time_;
-  int consumed_seqno_count_;
+  SequenceNumber max_assigned_seqno_;
  // Set in ExternalSstFileIngestionJob::Prepare(), if true all files are
  // ingested in L0
  bool files_overlap_{false};
@@ -7,6 +7,7 @@

 #include <functional>
 #include <memory>
+#include <sstream>

 #include "db/db_test_util.h"
 #include "db/dbformat.h"
@@ -2417,102 +2418,130 @@ TEST_F(ExternalSSTFileTest, SnapshotInconsistencyBug) {
 }

 TEST_P(ExternalSSTFileTest, IngestBehind) {
-  Options options = CurrentOptions();
-  options.compaction_style = kCompactionStyleUniversal;
-  options.num_levels = 3;
-  options.disable_auto_compactions = false;
-  DestroyAndReopen(options);
-  std::vector<std::pair<std::string, std::string>> file_data;
-  std::map<std::string, std::string> true_data;
+  for (bool cf_option : {false, true}) {
+    SCOPED_TRACE("cf_option = " + std::to_string(cf_option));
+    Options options = CurrentOptions();
+    options.compaction_style = kCompactionStyleUniversal;
+    options.num_levels = 3;
+    options.disable_auto_compactions = false;
+    DestroyAndReopen(options);
+    std::vector<std::pair<std::string, std::string>> file_data;
+    std::map<std::string, std::string> true_data;

-  // Insert 100 -> 200 into the memtable
-  for (int i = 100; i <= 200; i++) {
-    ASSERT_OK(Put(Key(i), "memtable"));
-  }
-
-  // Insert 100 -> 200 using IngestExternalFile
-  file_data.clear();
-  for (int i = 0; i <= 20; i++) {
-    file_data.emplace_back(Key(i), "ingest_behind");
-    true_data[Key(i)] = "ingest_behind";
-  }
-
-  bool allow_global_seqno = true;
-  bool ingest_behind = true;
-  bool write_global_seqno = std::get<0>(GetParam());
-  bool verify_checksums_before_ingest = std::get<1>(GetParam());
-
-  // Can't ingest behind since allow_ingest_behind isn't set to true
-  ASSERT_NOK(GenerateAndAddExternalFile(
-      options, file_data, -1, allow_global_seqno, write_global_seqno,
-      verify_checksums_before_ingest, ingest_behind, false /*sort_data*/,
-      &true_data));
-
-  options.allow_ingest_behind = true;
-  // check that we still can open the DB, as num_levels should be
-  // sanitized to 3
-  options.num_levels = 2;
-  DestroyAndReopen(options);
-
-  options.num_levels = 3;
-  DestroyAndReopen(options);
-  true_data.clear();
-  // Insert 100 -> 200 into the memtable
-  for (int i = 100; i <= 200; i++) {
-    ASSERT_OK(Put(Key(i), "memtable"));
-    true_data[Key(i)] = "memtable";
-  }
-  ASSERT_OK(db_->CompactRange(CompactRangeOptions(), nullptr, nullptr));
-  // Universal picker should go at second from the bottom level
-  ASSERT_EQ("0,1", FilesPerLevel());
-  ASSERT_OK(GenerateAndAddExternalFile(
-      options, file_data, -1, allow_global_seqno, write_global_seqno,
-      verify_checksums_before_ingest, true /*ingest_behind*/,
-      false /*sort_data*/, &true_data));
-  ASSERT_EQ("0,1,1", FilesPerLevel());
-  // this time ingest should fail as the file doesn't fit to the bottom level
-  ASSERT_NOK(GenerateAndAddExternalFile(
-      options, file_data, -1, allow_global_seqno, write_global_seqno,
-      verify_checksums_before_ingest, true /*ingest_behind*/,
-      false /*sort_data*/, &true_data));
-  ASSERT_EQ("0,1,1", FilesPerLevel());
-  std::vector<std::vector<FileMetaData>> level_to_files;
-  dbfull()->TEST_GetFilesMetaData(db_->DefaultColumnFamily(), &level_to_files);
-  uint64_t ingested_file_number = level_to_files[2][0].fd.GetNumber();
-  ASSERT_OK(db_->CompactRange(CompactRangeOptions(), nullptr, nullptr));
-  // Last level should not be compacted
-  ASSERT_EQ("0,1,1", FilesPerLevel());
-  dbfull()->TEST_GetFilesMetaData(db_->DefaultColumnFamily(), &level_to_files);
-  ASSERT_EQ(ingested_file_number, level_to_files[2][0].fd.GetNumber());
-  size_t kcnt = 0;
-  VerifyDBFromMap(true_data, &kcnt, false);
-
-  // Auto-compaction should not include the last level.
-  // Trigger compaction if size amplification exceeds 110%.
-  options.compaction_options_universal.max_size_amplification_percent = 110;
-  options.level0_file_num_compaction_trigger = 4;
-  ASSERT_OK(TryReopen(options));
-  Random rnd(301);
-  for (int i = 0; i < 4; ++i) {
-    for (int j = 0; j < 10; j++) {
-      true_data[Key(j)] = rnd.RandomString(1000);
-      ASSERT_OK(Put(Key(j), true_data[Key(j)]));
+    // Insert 100 -> 200 into the memtable
+    for (int i = 100; i <= 200; i++) {
+      ASSERT_OK(Put(Key(i), "memtable"));
    }
-    ASSERT_OK(Flush());
-  }
-  ASSERT_OK(dbfull()->TEST_WaitForCompact());
-  dbfull()->TEST_GetFilesMetaData(db_->DefaultColumnFamily(), &level_to_files);
-  ASSERT_EQ(1, level_to_files[2].size());
-  ASSERT_EQ(ingested_file_number, level_to_files[2][0].fd.GetNumber());

-  // Turning off the option allows DB to compact ingested files.
-  options.allow_ingest_behind = false;
-  ASSERT_OK(TryReopen(options));
-  ASSERT_OK(db_->CompactRange(CompactRangeOptions(), nullptr, nullptr));
-  dbfull()->TEST_GetFilesMetaData(db_->DefaultColumnFamily(), &level_to_files);
-  ASSERT_EQ(1, level_to_files[2].size());
-  ASSERT_NE(ingested_file_number, level_to_files[2][0].fd.GetNumber());
-  VerifyDBFromMap(true_data, &kcnt, false);
+    // Insert 100 -> 200 using IngestExternalFile
+    file_data.clear();
+    for (int i = 0; i <= 20; i++) {
+      file_data.emplace_back(Key(i), "ingest_behind");
+      true_data[Key(i)] = "ingest_behind";
+    }
+
+    bool allow_global_seqno = true;
+    bool ingest_behind = true;
+    bool write_global_seqno = std::get<0>(GetParam());
+    bool verify_checksums_before_ingest = std::get<1>(GetParam());
+
+    // Can't ingest behind since allow_ingest_behind isn't set to true
+    ASSERT_NOK(GenerateAndAddExternalFile(
+        options, file_data, -1, allow_global_seqno, write_global_seqno,
+        verify_checksums_before_ingest, ingest_behind, false /*sort_data*/,
+        &true_data));
+
+    if (cf_option) {
+      options.cf_allow_ingest_behind = true;
+    } else {
+      options.allow_ingest_behind = true;
+    }
+    // check that we still can open the DB, as num_levels should be
+    // sanitized to 3
+    options.num_levels = 2;
+    DestroyAndReopen(options);
+
+    options.num_levels = 3;
+    DestroyAndReopen(options);
+    true_data.clear();
+    // Insert 100 -> 200 into the memtable
+    for (int i = 100; i <= 200; i++) {
+      ASSERT_OK(Put(Key(i), "memtable"));
+      true_data[Key(i)] = "memtable";
+    }
+    ASSERT_OK(db_->CompactRange(CompactRangeOptions(), nullptr, nullptr));
+    // Universal picker should go at second from the bottom level
+    ASSERT_EQ("0,1", FilesPerLevel());
+    ASSERT_OK(GenerateAndAddExternalFile(
+        options, file_data, -1, allow_global_seqno, write_global_seqno,
+        verify_checksums_before_ingest, true /*ingest_behind*/,
+        false /*sort_data*/, &true_data));
+    ASSERT_EQ("0,1,1", FilesPerLevel());
+    // this time ingest should fail as the file doesn't fit to the bottom level
+    ASSERT_NOK(GenerateAndAddExternalFile(
+        options, file_data, -1, allow_global_seqno, write_global_seqno,
+        verify_checksums_before_ingest, true /*ingest_behind*/,
+        false /*sort_data*/, &true_data));
+    ASSERT_EQ("0,1,1", FilesPerLevel());
+    std::vector<std::vector<FileMetaData>> level_to_files;
+    dbfull()->TEST_GetFilesMetaData(db_->DefaultColumnFamily(),
+                                    &level_to_files);
+    uint64_t ingested_file_number = level_to_files[2][0].fd.GetNumber();
+    ASSERT_OK(db_->CompactRange(CompactRangeOptions(), nullptr, nullptr));
+    // Last level should not be compacted
+    ASSERT_EQ("0,1,1", FilesPerLevel());
+    dbfull()->TEST_GetFilesMetaData(db_->DefaultColumnFamily(),
+                                    &level_to_files);
+    ASSERT_EQ(ingested_file_number, level_to_files[2][0].fd.GetNumber());
+    size_t kcnt = 0;
+    VerifyDBFromMap(true_data, &kcnt, false);
+
+    // Auto-compaction should not include the last level.
+    // Trigger compaction if size amplification exceeds 110%.
+    options.compaction_options_universal.max_size_amplification_percent = 110;
+    options.level0_file_num_compaction_trigger = 4;
+    ASSERT_OK(TryReopen(options));
+    Random rnd(301);
+    for (int i = 0; i < 4; ++i) {
+      for (int j = 0; j < 10; j++) {
+        true_data[Key(j)] = rnd.RandomString(1000);
+        ASSERT_OK(Put(Key(j), true_data[Key(j)]));
+      }
+      ASSERT_OK(Flush());
+    }
+    ASSERT_OK(dbfull()->TEST_WaitForCompact());
+    dbfull()->TEST_GetFilesMetaData(db_->DefaultColumnFamily(),
+                                    &level_to_files);
+    ASSERT_EQ(1, level_to_files[2].size());
+    ASSERT_EQ(ingested_file_number, level_to_files[2][0].fd.GetNumber());
+
+    // Turning off the option allows DB to compact ingested files.
+    if (cf_option) {
+      // Test that another CF does not allow ingest behind
+      ColumnFamilyHandle* new_cfh;
+      Options new_cf_option;
+      ASSERT_OK(db_->CreateColumnFamily(new_cf_option, "new_cf", &new_cfh));
+      ASSERT_TRUE(GenerateAndAddExternalFile(
+                      new_cf_option, file_data, -1, allow_global_seqno,
+                      write_global_seqno, verify_checksums_before_ingest,
+                      true /*ingest_behind*/, false /*sort_data*/, nullptr,
+                      /*cfh=*/new_cfh)
+                      .IsInvalidArgument());
+      ASSERT_OK(db_->DropColumnFamily(new_cfh));
+      ASSERT_OK(db_->DestroyColumnFamilyHandle(new_cfh));
+
+      options.cf_allow_ingest_behind = false;
+    } else {
+      options.allow_ingest_behind = false;
+    }
+    ASSERT_OK(TryReopen(options));
+    ASSERT_OK(db_->CompactRange(CompactRangeOptions(), nullptr, nullptr));
+    dbfull()->TEST_GetFilesMetaData(db_->DefaultColumnFamily(),
+                                    &level_to_files);
+    ASSERT_EQ(1, level_to_files[2].size());
+    ASSERT_NE(ingested_file_number, level_to_files[2][0].fd.GetNumber());
+    VerifyDBFromMap(true_data, &kcnt, false);
+  }
 }

 TEST_F(ExternalSSTFileTest, SkipBloomFilter) {
@@ -3514,19 +3543,26 @@ TEST_F(ExternalSSTFileWithTimestampTest, SanityCheck) {
  // overlapping key ranges.
  ASSERT_TRUE(IngestExternalUDTFile({file1, file2}).IsNotSupported());

-  options.allow_ingest_behind = true;
-  DestroyAndReopen(options);
-  IngestExternalFileOptions opts;
+  for (bool cf_option : {false, true}) {
+    SCOPED_TRACE("cf_option = " + std::to_string(cf_option));
+    if (cf_option) {
+      options.cf_allow_ingest_behind = true;
+    } else {
+      options.allow_ingest_behind = true;
+    }
+    DestroyAndReopen(options);
+    IngestExternalFileOptions opts;

-  // TODO(yuzhangyu): support ingestion behind for user-defined timestamps?
-  // Ingesting external files with user-defined timestamps requires searching
-  // through the whole lsm tree to make sure there is no key range overlap with
-  // the db. Ingestion behind currently is doing a simply placing it at the
-  // bottom level step without a search, so we don't allow it either.
-  opts.ingest_behind = true;
-  ASSERT_TRUE(db_->IngestExternalFile({file1}, opts).IsNotSupported());
+    // TODO(yuzhangyu): support ingestion behind for user-defined timestamps?
+    // Ingesting external files with user-defined timestamps requires searching
+    // through the whole lsm tree to make sure there is no key range overlap
+    // with the db. Ingestion behind currently is doing a simply placing it at
+    // the bottom level step without a search, so we don't allow it either.
+    opts.ingest_behind = true;
+    ASSERT_TRUE(db_->IngestExternalFile({file1}, opts).IsNotSupported());

-  DestroyAndRecreateExternalSSTFilesDir();
+    DestroyAndRecreateExternalSSTFilesDir();
+  }
 }

 TEST_F(ExternalSSTFileWithTimestampTest, UDTSettingsCompatibilityCheck) {
@@ -3818,99 +3854,32 @@ TEST_P(IngestDBGeneratedFileTest, FailureCase) {
      ASSERT_OK(Put(1, Key(k), "cf1_" + Key(k)));
    }
    ASSERT_OK(Flush(/*cf=*/1));
-    {
-      // Verify that largest key of the file has non-zero seqno.
-      std::vector<std::vector<FileMetaData>> metadata;
-      dbfull()->TEST_GetFilesMetaData(handles_[1], &metadata, nullptr);
-      const FileMetaData& file = metadata[0][0];
-      ValueType vtype;
-      SequenceNumber seq;
-      UnPackSequenceAndType(ExtractInternalKeyFooter(file.largest.Encode()),
-                            &seq, &vtype);
-      ASSERT_GE(seq, 0);
-    }
-    std::vector<LiveFileMetaData> live_meta;
-    db_->GetLiveFilesMetaData(&live_meta);
-    ASSERT_EQ(live_meta.size(), 1);
-    std::vector<std::string> to_ingest_files;
-    to_ingest_files.emplace_back(live_meta[0].directory + "/" +
-                                 live_meta[0].relative_filename);
-    // Ingesting a file whose boundary key has non-zero seqno.
-    Status s = db_->IngestExternalFile(to_ingest_files, ingest_opts);
-    // This error msg is from checking seqno of boundary keys.
-    ASSERT_TRUE(
-        s.ToString().find("External file has non zero sequence number") !=
-        std::string::npos);
-    ASSERT_NOK(s);
-
-    {
-      // Only non-boundary key with non-zero seqno.
-      const Snapshot* snapshot = db_->GetSnapshot();
-      ASSERT_OK(Put(1, Key(70), "cf1_" + Key(70)));
-      ASSERT_OK(Flush(1));
-      CompactRangeOptions cro;
-      cro.bottommost_level_compaction =
-          BottommostLevelCompaction::kForceOptimized;
-      ASSERT_OK(db_->CompactRange(cro, handles_[1], nullptr, nullptr));
-
-      // Verify that only the non-boundary key of the file has non-zero seqno.
-      std::vector<std::vector<FileMetaData>> metadata;
-      // File may be at different level for different options.
-      dbfull()->TEST_GetFilesMetaData(handles_[1], &metadata, nullptr);
-      bool found_file = false;
-      for (const auto& level : metadata) {
-        if (level.empty()) {
-          continue;
-        }
-        ASSERT_FALSE(found_file);
-        found_file = true;
-        ASSERT_EQ(1, level.size());
-        const FileMetaData& file = level[0];
-        ValueType vtype;
-        SequenceNumber seq;
-        UnPackSequenceAndType(ExtractInternalKeyFooter(file.largest.Encode()),
-                              &seq, &vtype);
-        ASSERT_EQ(seq, 0);
-        UnPackSequenceAndType(ExtractInternalKeyFooter(file.smallest.Encode()),
-                              &seq, &vtype);
-        ASSERT_EQ(seq, 0);
-        ASSERT_GT(file.fd.largest_seqno, 0);
-      }
-      ASSERT_TRUE(found_file);
-      live_meta.clear();
-      db_->GetLiveFilesMetaData(&live_meta);
-      ASSERT_EQ(live_meta.size(), 1);
-      to_ingest_files[0] =
-          live_meta[0].directory + "/" + live_meta[0].relative_filename;
-      s = db_->IngestExternalFile(to_ingest_files, ingest_opts);
-      ASSERT_NOK(s);
-      // This error msg is from checking largest seqno in table property.
-      ASSERT_TRUE(s.ToString().find("non zero largest sequence number") !=
-                  std::string::npos);
-      db_->ReleaseSnapshot(snapshot);
-    }

+    Status s;
    CompactRangeOptions cro;
    cro.bottommost_level_compaction =
        BottommostLevelCompaction::kForceOptimized;
    ASSERT_OK(db_->CompactRange(cro, handles_[1], nullptr, nullptr));
-    live_meta.clear();
+
+    std::vector<LiveFileMetaData> live_meta;
+    std::vector<std::string> to_ingest_files;
    db_->GetLiveFilesMetaData(&live_meta);
    ASSERT_EQ(live_meta.size(), 1);
+    ASSERT_EQ(live_meta[0].column_family_name, "toto");
    ASSERT_EQ(0, live_meta[0].largest_seqno);
-    to_ingest_files[0] =
-        live_meta[0].directory + "/" + live_meta[0].relative_filename;
+    to_ingest_files.emplace_back(live_meta[0].directory + "/" +
+                                 live_meta[0].relative_filename);

+    // Ingesting a DB generated file with allow_db_generated_files = false
    ingest_opts.allow_db_generated_files = false;
-    // Ingesting a DB genrate file with allow_db_generated_files = false;
    s = db_->IngestExternalFile(to_ingest_files, ingest_opts);
    ASSERT_TRUE(s.ToString().find("External file version not found") !=
                std::string::npos);
    ASSERT_NOK(s);

    const std::string err =
-        "An ingested file is assigned to a non-zero sequence number, which is "
-        "incompatible with ingestion option allow_db_generated_files";
+        "An ingested file overlaps with existing data in the DB and has been "
+        "assigned a non-zero sequence number";
    ingest_opts.allow_db_generated_files = true;
    s = db_->IngestExternalFile(to_ingest_files, ingest_opts);
    ASSERT_TRUE(s.ToString().find(err) != std::string::npos);
@@ -4111,6 +4080,440 @@ TEST_P(IngestDBGeneratedFileTest2, NotOverlapWithDB) {
    }
  } while (ChangeOptions(kSkipPlainTable | kSkipFIFOCompaction));
 }
+
+TEST_P(IngestDBGeneratedFileTest2, NonZeroSeqno) {
+  // Test ingestion of DB-generated SST files that contain non-zero sequence
+  // numbers.
+  IngestExternalFileOptions ingest_opts;
+  ingest_opts.allow_db_generated_files = true;
+  // This only works since we are ingesting without snapshot
+  // Failure case will be tested below.
+  ingest_opts.snapshot_consistency = std::get<0>(GetParam());
+  ingest_opts.allow_global_seqno = std::get<1>(GetParam());
+  ingest_opts.allow_blocking_flush = std::get<2>(GetParam());
+  ingest_opts.fail_if_not_bottommost_level = std::get<3>(GetParam());
+  ingest_opts.link_files = std::get<4>(GetParam());
+  Random* rnd = Random::GetTLSInstance();
+
+  do {
+    SCOPED_TRACE("option_config_ = " + std::to_string(option_config_));
+
+    Options options = CurrentOptions();
+    options.allow_concurrent_memtable_write =
+        false;  // Required for VectorRepFactory
+    CreateAndReopenWithCF({"non_overlap", "overlap"}, options);
+
+    ColumnFamilyHandle* non_overlap_cf = handles_[1];
+    ColumnFamilyHandle* overlap_cf = handles_[2];
+
+    std::vector<std::string> expected_values;
+    expected_values.resize(100);
+    WriteOptions wo;
+    // Setup target CF with non-overlapping base data Key1 and Key99
+    // Will ingest keys [1, 98] below.
+    expected_values[0] = rnd->RandomString(100);
+    ASSERT_OK(db_->Put(wo, non_overlap_cf, Key(0), expected_values[0]));
+    ASSERT_OK(db_->Flush({}, non_overlap_cf));
+    expected_values[99] = rnd->RandomString(100);
+    ASSERT_OK(db_->Put(wo, non_overlap_cf, Key(99), expected_values[99]));
+
+    // Set up overlapping cf
+    ASSERT_OK(db_->Put(wo, overlap_cf, Key(50), rnd->RandomString(100)));
+
+    // Create temp CF/DB
+    Options temp_cf_opts;
+    ColumnFamilyHandle* temp_cfh = nullptr;
+    DB* from_db = nullptr;
+    std::string temp_db_name;
+    // Using a separate DB also validates that latest sequence number
+    // of target db is updated after ingestion (to the max sequence number
+    // in ingested files).
+    const bool use_temp_db = rnd->OneIn(2);
+    SCOPED_TRACE("use_temp_db: " + std::to_string(use_temp_db));
+
+    std::vector<std::string> sst_file_paths;
+    // optional L5: files in key range [70, 98]
+    // L6: files in key range [1, 79]
+    temp_cf_opts.target_file_size_base =
+        4 << 10;  // Small files to create multiple SSTs
+    temp_cf_opts.num_levels = 7;
+    temp_cf_opts.disable_auto_compactions = true;  // Manually set up LSM
+    temp_cf_opts.env = options.env;
+
+    if (use_temp_db) {
+      temp_cf_opts.create_if_missing = true;
+      temp_db_name = dbname_ + "/temp_db_" + std::to_string(rnd->Next());
+      ASSERT_OK(DB::Open(temp_cf_opts, temp_db_name, &from_db));
+      temp_cfh = from_db->DefaultColumnFamily();
+    } else {
+      from_db = db_;
+      ASSERT_OK(
+          from_db->CreateColumnFamily(temp_cf_opts, "temp_cf", &temp_cfh));
+    }
+
+    // Use snapshot to ensure non-zero sequence numbers after compaction
+    const Snapshot* snapshot = from_db->GetSnapshot();
+
+    for (int k = 1; k < 99; ++k) {
+      expected_values[k] = rnd->RandomString(500);
+      ASSERT_OK(from_db->Put(wo, temp_cfh, Key(k), expected_values[k]));
+    }
+    ASSERT_OK(from_db->Flush({}, temp_cfh));
+    CompactRangeOptions cro;
+    cro.bottommost_level_compaction =
+        BottommostLevelCompaction::kForceOptimized;
+    ASSERT_OK(from_db->CompactRange(cro, temp_cfh, nullptr, nullptr));
+
+    ASSERT_GT(NumTableFilesAtLevel(6, temp_cfh, from_db), 1);
+
+    const bool multi_level_ingestion = rnd->OneIn(2);
+    SCOPED_TRACE("Multi-level ingestion: " +
+                 std::to_string(multi_level_ingestion));
+    if (multi_level_ingestion) {
+      for (int k = 80; k < 99; ++k) {
+        expected_values[k] = rnd->RandomString(500);
+        ASSERT_OK(from_db->Put(wo, temp_cfh, Key(k), expected_values[k]));
+      }
+      ASSERT_OK(from_db->Flush({}, temp_cfh));
+
+      // Do some overwrites, and overlap with previous L0 to avoid trivial move
+      for (int k = 70; k < 82; ++k) {
+        expected_values[k] = rnd->RandomString(500);
+        ASSERT_OK(from_db->Put(wo, temp_cfh, Key(k), expected_values[k]));
+      }
+      ASSERT_OK(from_db->Flush({}, temp_cfh));
+
+      if (rnd->OneIn(2)) {
+        MoveFilesToLevel(5, temp_cfh, from_db);
+        ASSERT_GT(NumTableFilesAtLevel(5, temp_cfh, from_db), 0);
+      }
+      ASSERT_GT(NumTableFilesAtLevel(6, temp_cfh, from_db), 0);
+    }
+    SCOPED_TRACE("LSM of from_db " + FilesPerLevel(temp_cfh, from_db));
+
+    ColumnFamilyMetaData cf_meta;
+    from_db->GetColumnFamilyMetaData(temp_cfh, &cf_meta);
+
+    // Iterate in reverse since IngestExternalFiles expect files to be ordered
+    // from old to new
+    for (auto level_meta = cf_meta.levels.rbegin();
+         level_meta != cf_meta.levels.rend(); ++level_meta) {
+      // L0 files need to be added in reverse order.
+      for (auto file_meta = level_meta->files.rbegin();
+           file_meta != level_meta->files.rend(); ++file_meta) {
+        // Validate that files contain non-zero sequence numbers
+        ASSERT_GT(file_meta->smallest_seqno, 0);
+        ASSERT_GE(file_meta->largest_seqno, file_meta->smallest_seqno);
+        sst_file_paths.emplace_back(file_meta->directory + "/" +
+                                    file_meta->relative_filename);
+      }
+    }
+    from_db->ReleaseSnapshot(snapshot);
+
+    Status s;
+    // Perform ingestion and validate results
+    if (multi_level_ingestion && options.num_levels > 1) {
+      // fail_if_bottommost requres ingesting all files into the last level,
+      // so it fails if we are assiging files to multiple levels.
+      ingest_opts.fail_if_not_bottommost_level = true;
+      s = db_->IngestExternalFile(non_overlap_cf, sst_file_paths, ingest_opts);
+      ASSERT_NOK(s);
+      ASSERT_TRUE(s.ToString().find("Files cannot be ingested to Lmax") !=
+                  std::string::npos);
+      ingest_opts.fail_if_not_bottommost_level = false;
+    }
+    if (ingest_opts.snapshot_consistency) {
+      // snapshot_consisteny requires global sequence number assignment to
+      // ingested files if there is any live snapshot.
+      snapshot = db_->GetSnapshot();
+      s = db_->IngestExternalFile(non_overlap_cf, sst_file_paths, ingest_opts);
+      ASSERT_NOK(s);
+      ASSERT_TRUE(s.ToString().find(
+          "An ingested file overlaps with existing data in the DB and has been "
+          "assigned a non-zero sequence number"));
+      db_->ReleaseSnapshot(snapshot);
+    }
+    ASSERT_OK(
+        db_->IngestExternalFile(non_overlap_cf, sst_file_paths, ingest_opts));
+
+    // Validate ingested data.
+    ReadOptions ro;
+    std::string val;
+    for (int k = 0; k < 100; ++k) {
+      s = db_->Get(ro, handles_[1], Key(k), &val);
+      ASSERT_OK(s) << "Should find ingested key " << Key(k);
+      ASSERT_EQ(val, expected_values[k]) << "key: " << Key(k);
+    }
+
+    // Overlap with data in the CF
+    if (ingest_opts.allow_blocking_flush) {
+      s = db_->IngestExternalFile(overlap_cf, sst_file_paths, ingest_opts);
+
+      ASSERT_NOK(s);
+      ASSERT_TRUE(s.ToString().find("An ingested file overlaps with existing "
+                                    "data in the DB and has been "
+                                    "assigned a non-zero sequence number") !=
+                  std::string::npos)
+          << s.ToString();
+    }
+
+    // Cleanup
+    // FIXME: Without this, the test triggers some data race between dropping
+    // CF and background compaction.
+    ASSERT_OK(db_->WaitForCompact({}));
+    if (use_temp_db) {
+      ASSERT_OK(from_db->Close());
+      delete from_db;
+      ASSERT_OK(DestroyDB(temp_db_name, temp_cf_opts));
+    } else {
+      ASSERT_OK(db_->DropColumnFamily(temp_cfh));
+      ASSERT_OK(db_->DestroyColumnFamilyHandle(temp_cfh));
+    }
+  } while (ChangeOptions(kSkipPlainTable | kSkipFIFOCompaction));
+}
+
+std::string GenSecondaryKey(const std::string& pk, const std::string& val) {
+  return "index_" + val + "_" + pk;
+};
+
+TEST_P(IngestDBGeneratedFileTest2, ZeroAndNonZeroSeqno) {
+  // Test ingestion of SST files with zero and with non-zero sequence numbers.
+  // Generate data using a temp CF and a temp DB:
+  // 1. Temp CF with cf_allow_ingest_behind enabled to preserve non-zero seqno.
+  // 2. Temp DB with everything compacted to have zero seqno.
+  // Then ingest both types of files together into a target CF.
+  // This mimics a user case where temp DB contains data read from a
+  // snapshot while temp CF contains live writes after a snapshot is taken.
+  IngestExternalFileOptions ingest_opts;
+  ingest_opts.allow_db_generated_files = true;
+  ingest_opts.snapshot_consistency = std::get<0>(GetParam());
+  ingest_opts.allow_global_seqno = std::get<1>(GetParam());
+  ingest_opts.allow_blocking_flush = std::get<2>(GetParam());
+  ingest_opts.fail_if_not_bottommost_level = std::get<3>(GetParam());
+  ingest_opts.link_files = std::get<4>(GetParam());
+
+  Random* rnd = Random::GetTLSInstance();
+
+  do {
+    SCOPED_TRACE("option_config_ = " + std::to_string(option_config_));
+    Options options = CurrentOptions();
+    options.allow_concurrent_memtable_write = false;
+    // Force more flushes/compactions and more files to be generated
+    options.target_file_size_base = 1 << 10;     // 1KB
+    options.max_bytes_for_level_base = 2 << 10;  // 2KB
+    options.max_bytes_for_level_multiplier = 2;
+    options.level0_file_num_compaction_trigger = 2;
+    options.level_compaction_dynamic_level_bytes = true;
+    DestroyAndReopen(options);
+    CreateAndReopenWithCF({"target_cf"}, options);
+    auto* target_cfh = handles_[1];
+
+    Options live_write_cf_opts = options;
+    live_write_cf_opts.memtable_factory.reset(new VectorRepFactory());
+    live_write_cf_opts.compaction_style = kCompactionStyleUniversal;
+    live_write_cf_opts.cf_allow_ingest_behind = true;
+    live_write_cf_opts.num_levels = 50;
+    ColumnFamilyHandle* live_write_cfh;
+    ASSERT_OK(db_->CreateColumnFamily(live_write_cf_opts, "live_write_cf",
+                                      &live_write_cfh));
+
+    // Expected value and key
+    std::map<std::string, std::string> expected;
+    std::unordered_set<std::string> deleted;
+    std::stringstream debug_info;
+
+    // Setup base data in target CF, will ingest keys with different prefixes
+    // so they don't overlap with the base data.
+    WriteOptions wo;
+    for (int k = 0; k < 100; ++k) {
+      int random_val = rnd->Uniform(20);
+      expected[Key(k)] = std::to_string(random_val);
+      ASSERT_OK(db_->Put(wo, target_cfh, Key(k), expected[Key(k)]));
+
+      // Force flush every 20 keys to create multiple SST files
+      if (rnd->OneIn(20)) {
+        ASSERT_OK(db_->Flush({}, target_cfh));
+        debug_info << "Flush after " << k
+                   << ", LSM state: " << FilesPerLevel(target_cfh) << "\n";
+      }
+    }
+
+    // Temp DB for snapshot data
+    Options temp_db_opts;
+    temp_db_opts.create_if_missing = true;
+    temp_db_opts.target_file_size_base = 1 << 10;
+    temp_db_opts.write_buffer_size = 1 << 10;
+    temp_db_opts.memtable_factory.reset(new VectorRepFactory());
+    temp_db_opts.allow_concurrent_memtable_write = false;
+    temp_db_opts.compaction_style = kCompactionStyleUniversal;
+    temp_db_opts.env = env_;
+    temp_db_opts.num_levels = 7;
+
+    std::string temp_db_name =
+        dbname_ + "/temp_db_" + std::to_string(rnd->Next());
+    DB* temp_db = nullptr;
+    ASSERT_OK(DB::Open(temp_db_opts, temp_db_name, &temp_db));
+
+    const Snapshot* snapshot = db_->GetSnapshot();
+    ReadOptions ro;
+    ro.snapshot = snapshot;
+    ro.total_order_seek = true;
+    std::unique_ptr<Iterator> iter{db_->NewIterator(ro, target_cfh)};
+    // transform data read from snapshot and write to temp DB
+    // Varying the number of files in temp DB.
+    const int kValSize = rnd->Uniform(200);
+    for (iter->SeekToFirst(); iter->Valid(); iter->Next()) {
+      std::string key = iter->key().ToString();
+      std::string value = iter->value().ToString();
+      std::string sk = GenSecondaryKey(key, value);
+      // Usually value is empty, here we use a larger value to generate
+      // multiple SST files in temp_db.
+      std::string sk_val = rnd->RandomString(kValSize);
+      ASSERT_OK(temp_db->Put(wo, sk, sk_val));
+      expected[sk] = sk_val;
+      debug_info << "Snapshot data: " << sk << " -> \n";
+    }
+    ASSERT_OK(iter->status());
+
+    // Do some live writes into target CF and live write CF.
+    for (int i = 0; i < 10; ++i) {
+      WriteBatch wb;
+      for (int j = 0; j < 5; ++j) {
+        std::string key = Key(rnd->Uniform(100));
+        std::string old_val = expected[key];
+        // Value range is 0-19, allow some PK to have the same value.
+        int random_val = rnd->Uniform(20);
+        std::string new_val = std::to_string(random_val);
+        std::string old_index_key = GenSecondaryKey(key, old_val);
+        std::string new_index_key = GenSecondaryKey(key, new_val);
+        ASSERT_OK(wb.SingleDelete(live_write_cfh, old_index_key));
+        std::string sk_val = rnd->RandomString(kValSize);
+        ASSERT_OK(wb.Put(live_write_cfh, new_index_key, sk_val));
+        ASSERT_OK(wb.Put(target_cfh, key, new_val));
+        expected[key] = new_val;
+        expected.erase(old_index_key);
+        expected[new_index_key] = sk_val;
+        deleted.insert(old_index_key);
+        deleted.erase(new_index_key);
+
+        debug_info << "Live write: SD " << old_index_key << "\n";
+        debug_info << "Live write: " << key << " -> " << new_val << "\n";
+        debug_info << "Live write: " << new_index_key << " -> \n";
+      }
+      ASSERT_OK(db_->Write(wo, &wb));
+      if (rnd->OneIn(3)) {
+        debug_info << "Flush after " << i << " live writes\n";
+        ASSERT_OK(db_->Flush({}, live_write_cfh));
+      }
+    }
+    iter.reset();
+    db_->ReleaseSnapshot(snapshot);
+
+    // Compact temp_db to ensure zero sequence numbers
+    CompactRangeOptions cro;
+    cro.bottommost_level_compaction = BottommostLevelCompaction::kForce;
+    ASSERT_OK(temp_db->CompactRange(cro, nullptr, nullptr));
+    SCOPED_TRACE("Temp DB LSM: " +
+                 FilesPerLevel(temp_db->DefaultColumnFamily(), temp_db));
+
+    // Base data from snapshot
+    std::vector<std::string> sst_file_paths_zero_seqno;
+
+    // Collect SST file paths with zero sequence numbers
+    ASSERT_OK(temp_db->DisableFileDeletions());
+    ColumnFamilyMetaData cf_meta_temp_db;
+    temp_db->GetColumnFamilyMetaData(&cf_meta_temp_db);
+    for (const auto& level_meta : cf_meta_temp_db.levels) {
+      if (level_meta.level == 6) {
+        for (const auto& file_meta : level_meta.files) {
+          // Verify files have zero sequence numbers
+          ASSERT_EQ(0, file_meta.largest_seqno)
+              << "File " << file_meta.relative_filename
+              << " should have zero sequence number\n"
+              << debug_info.str();
+          sst_file_paths_zero_seqno.emplace_back(file_meta.directory + "/" +
+                                                 file_meta.relative_filename);
+        }
+      } else {
+        // All files should be in L6
+        ASSERT_EQ(0, level_meta.files.size()) << debug_info.str();
+      }
+    }
+
+    // Flush remaining catch up writes in memtable
+    ASSERT_OK(db_->Flush({}, live_write_cfh));
+    SCOPED_TRACE("LSM of live write cfh " + FilesPerLevel(live_write_cfh));
+    // Collect SST file paths with non-zero sequence numbers
+    ColumnFamilyMetaData live_write_cf_meta;
+    ASSERT_OK(db_->DisableFileDeletions());
+    db_->GetColumnFamilyMetaData(live_write_cfh, &live_write_cf_meta);
+
+    // Live writes after snapshot
+    std::vector<std::string> sst_file_paths_nonzero_seqno;
+    for (auto level_meta = live_write_cf_meta.levels.rbegin();
+         level_meta != live_write_cf_meta.levels.rend(); ++level_meta) {
+      // Reverse order is important for L0, where recent updates are ordered
+      // first
+      for (auto file_meta = level_meta->files.rbegin();
+           file_meta != level_meta->files.rend(); ++file_meta) {
+        sst_file_paths_nonzero_seqno.emplace_back(file_meta->directory + "/" +
+                                                  file_meta->relative_filename);
+        ASSERT_GT(file_meta->smallest_seqno, 0) << debug_info.str();
+      }
+      if (level_meta->level == 49) {
+        // Ingest behind does not compact to the last level
+        ASSERT_EQ(level_meta->files.size(), 0) << debug_info.str();
+      }
+    }
+
+    ASSERT_GT(sst_file_paths_zero_seqno.size(), 0) << debug_info.str();
+    ASSERT_GT(sst_file_paths_nonzero_seqno.size(), 0) << debug_info.str();
+
+    // Combine all SST file paths.
+    // File ingestion takes files from old to new.
+    std::vector<std::string> all_sst_files;
+    all_sst_files.insert(all_sst_files.end(), sst_file_paths_zero_seqno.begin(),
+                         sst_file_paths_zero_seqno.end());
+    all_sst_files.insert(all_sst_files.end(),
+                         sst_file_paths_nonzero_seqno.begin(),
+                         sst_file_paths_nonzero_seqno.end());
+    if (ingest_opts.fail_if_not_bottommost_level && options.num_levels > 1) {
+      // overlapping files will be ingested into different levels, including non
+      // Lmax
+      Status s =
+          db_->IngestExternalFile(target_cfh, all_sst_files, ingest_opts);
+      ASSERT_NOK(s);
+      ASSERT_TRUE(s.ToString().find("Files cannot be ingested to Lmax") !=
+                  std::string::npos);
+    } else {
+      ASSERT_OK(
+          db_->IngestExternalFile(target_cfh, all_sst_files, ingest_opts));
+
+      debug_info << "Zero seqno files: " << sst_file_paths_zero_seqno.size()
+                 << "\nNon-zero seqno files: "
+                 << sst_file_paths_nonzero_seqno.size() << "\n";
+
+      SCOPED_TRACE("Debug info:\n" + debug_info.str());
+      VerifyDBFromMap(expected, nullptr, false, nullptr, target_cfh, &deleted);
+    }
+
+    // clean up
+    ASSERT_OK(db_->EnableFileDeletions());
+    ASSERT_OK(temp_db->EnableFileDeletions());
+
+    // FIXME: Without this, the test triggers some data race between dropping
+    // CF and background compaction.
+    ASSERT_OK(db_->WaitForCompact({}));
+
+    ASSERT_OK(db_->DropColumnFamily(live_write_cfh));
+    ASSERT_OK(db_->DestroyColumnFamilyHandle(live_write_cfh));
+
+    ASSERT_OK(temp_db->Close());
+    delete temp_db;
+    ASSERT_OK(DestroyDB(temp_db_name, temp_db_opts));
+  } while (ChangeOptions(kSkipPlainTable | kSkipFIFOCompaction));
+}
+
 }  // namespace ROCKSDB_NAMESPACE

 int main(int argc, char** argv) {
@@ -502,8 +502,7 @@ Status FlushJob::MemPurge() {
        kMaxSequenceNumber, &job_context_->snapshot_seqs, earliest_snapshot_,
        job_context_->earliest_write_conflict_snapshot,
        job_context_->GetJobSnapshotSequence(), job_context_->snapshot_checker,
-        env, ShouldReportDetailedTime(env, ioptions.stats),
-        true /* internal key corruption is not ok */, range_del_agg.get(),
+        env, ShouldReportDetailedTime(env, ioptions.stats), range_del_agg.get(),
        nullptr, ioptions.allow_data_in_errors,
        ioptions.enforce_single_del_contracts,
        /*manual_compaction_canceled=*/kManualCompactionCanceledFalse,
@@ -95,7 +95,7 @@ bool Reader::ReadRecord(Slice* record, std::string* scratch,
  uint64_t prospective_record_offset = 0;

  Slice fragment;
-  while (true) {
+  for (;;) {
    uint64_t physical_record_offset = end_of_buffer_offset_ - buffer_.size();
    size_t drop_size = 0;
    const uint8_t record_type =
@@ -140,7 +140,7 @@ bool Reader::ReadRecord(Slice* record, std::string* scratch,
        prospective_record_offset = physical_record_offset;
        scratch->assign(fragment.data(), fragment.size());
        in_fragmented_record = true;
-        break;
+        break;  // switch

      case kMiddleType:
      case kRecyclableMiddleType:
@@ -153,7 +153,7 @@ bool Reader::ReadRecord(Slice* record, std::string* scratch,
          }
          scratch->append(fragment.data(), fragment.size());
        }
-        break;
+        break;  // switch

      case kLastType:
      case kRecyclableLastType:
@@ -171,7 +171,7 @@ bool Reader::ReadRecord(Slice* record, std::string* scratch,
          first_record_read_ = true;
          return true;
        }
-        break;
+        break;  // switch

      case kSetCompressionType: {
        if (compression_type_record_read_) {
@@ -193,7 +193,7 @@ bool Reader::ReadRecord(Slice* record, std::string* scratch,
        } else {
          InitCompression(compression_record);
        }
-        break;
+        break;  // switch
      }
      case kPredecessorWALInfoType:
      case kRecyclePredecessorWALInfoType: {
@@ -210,7 +210,7 @@ bool Reader::ReadRecord(Slice* record, std::string* scratch,
          MaybeVerifyPredecessorWALInfo(wal_recovery_mode, fragment,
                                        recorded_predecessor_wal_info);
        }
-        break;
+        break;  // switch
      }
      case kUserDefinedTimestampSizeType:
      case kRecyclableUserDefinedTimestampSizeType: {
@@ -235,7 +235,7 @@ bool Reader::ReadRecord(Slice* record, std::string* scratch,
            ReportCorruption(fragment.size(), s.getState());
          }
        }
-        break;
+        break;  // switch
      }

      case kBadHeader:
@@ -304,7 +304,7 @@ bool Reader::ReadRecord(Slice* record, std::string* scratch,
          in_fragmented_record = false;
          scratch->clear();
        }
-        break;
+        break;  // switch

      case kBadRecordLen:
        if (eof_) {
@@ -337,7 +337,7 @@ bool Reader::ReadRecord(Slice* record, std::string* scratch,
          in_fragmented_record = false;
          scratch->clear();
        }
-        break;
+        break;  // switch

      default: {
        if ((record_type & kRecordTypeSafeIgnoreMask) == 0) {
@@ -349,11 +349,11 @@ bool Reader::ReadRecord(Slice* record, std::string* scratch,
        }
        in_fragmented_record = false;
        scratch->clear();
-        break;
+        break;  // switch
      }
    }
  }
-  return false;
+  // unreachable
 }

 void Reader::MaybeVerifyPredecessorWALInfo(
@@ -380,8 +380,11 @@ void Reader::MaybeVerifyPredecessorWALInfo(
  } else {
    if (observed_predecessor_wal_info_.GetLogNumber() !=
        recorded_predecessor_log_number) {
-      std::string reason = "Missing WAL of log number " +
-                           std::to_string(recorded_predecessor_log_number);
+      std::string reason =
+          "Mismatched predecessor log number of WAL file " +
+          file_->file_name() + " Recorded " +
+          std::to_string(recorded_predecessor_log_number) + ". Observed " +
+          std::to_string(observed_predecessor_wal_info_.GetLogNumber());
      ReportCorruption(fragment.size(), reason.c_str(),
                       recorded_predecessor_log_number);
    } else if (observed_predecessor_wal_info_.GetLastSeqnoRecorded() !=
@@ -134,6 +134,16 @@ MemTable::MemTable(const InternalKeyComparator& cmp,
  auto new_cache = std::make_shared<FragmentedRangeTombstoneListCache>();
  size_t size = cached_range_tombstone_.Size();
  for (size_t i = 0; i < size; ++i) {
+#if defined(__cpp_lib_atomic_shared_ptr)
+    std::atomic<std::shared_ptr<FragmentedRangeTombstoneListCache>>*
+        local_cache_ref_ptr = cached_range_tombstone_.AccessAtCore(i);
+    auto new_local_cache_ref = std::make_shared<
+        const std::shared_ptr<FragmentedRangeTombstoneListCache>>(new_cache);
+    std::shared_ptr<FragmentedRangeTombstoneListCache> aliased_ptr(
+        new_local_cache_ref, new_cache.get());
+    local_cache_ref_ptr->store(std::move(aliased_ptr),
+                               std::memory_order_relaxed);
+#else
    std::shared_ptr<FragmentedRangeTombstoneListCache>* local_cache_ref_ptr =
        cached_range_tombstone_.AccessAtCore(i);
    auto new_local_cache_ref = std::make_shared<
@@ -143,6 +153,7 @@ MemTable::MemTable(const InternalKeyComparator& cmp,
        std::shared_ptr<FragmentedRangeTombstoneListCache>(new_local_cache_ref,
                                                           new_cache.get()),
        std::memory_order_relaxed);
+#endif
  }
  const Comparator* ucmp = cmp.user_comparator();
  assert(ucmp);
@@ -168,7 +179,7 @@ size_t MemTable::ApproximateMemoryUsage() {
    }
    total_usage += usage;
  }
-  approximate_memory_usage_.store(total_usage, std::memory_order_relaxed);
+  approximate_memory_usage_.StoreRelaxed(total_usage);
  // otherwise, return the actual usage
  return total_usage;
 }
@@ -182,12 +193,12 @@ bool MemTable::ShouldFlushNow() {
  // This is set if memtable_max_range_deletions is > 0,
  // and that many range deletions are done
  if (memtable_max_range_deletions_ > 0 &&
-      num_range_deletes_.load(std::memory_order_relaxed) >=
+      num_range_deletes_.LoadRelaxed() >=
          static_cast<uint64_t>(memtable_max_range_deletions_)) {
    return true;
  }

-  size_t write_buffer_size = write_buffer_size_.load(std::memory_order_relaxed);
+  size_t write_buffer_size = write_buffer_size_.LoadRelaxed();
  // In a lot of times, we cannot allocate arena blocks that exactly matches the
  // buffer size. Thus we have to decide if we should over-allocate or
  // under-allocate.
@@ -200,10 +211,10 @@ bool MemTable::ShouldFlushNow() {
  assert(range_del_table_->ApproximateMemoryUsage() == 0);
  // If arena still have room for new block allocation, we can safely say it
  // shouldn't flush.
-  auto allocated_memory = table_->ApproximateMemoryUsage() +
-                          arena_.MemoryAllocatedBytes();
+  auto allocated_memory =
+      table_->ApproximateMemoryUsage() + arena_.MemoryAllocatedBytes();

-  approximate_memory_usage_.store(allocated_memory, std::memory_order_relaxed);
+  approximate_memory_usage_.StoreRelaxed(allocated_memory);

  // if we can still allocate one more block without exceeding the
  // over-allocation ratio, then we should not flush.
@@ -745,7 +756,7 @@ FragmentedRangeTombstoneIterator* MemTable::NewRangeTombstoneIterator(
    const ReadOptions& read_options, SequenceNumber read_seq,
    bool immutable_memtable) {
  if (read_options.ignore_range_deletions ||
-      is_range_del_table_empty_.load(std::memory_order_relaxed)) {
+      is_range_del_table_empty_.LoadRelaxed()) {
    return nullptr;
  }
  return NewRangeTombstoneIteratorInternal(read_options, read_seq,
@@ -756,7 +767,7 @@ FragmentedRangeTombstoneIterator*
 MemTable::NewTimestampStrippingRangeTombstoneIterator(
    const ReadOptions& read_options, SequenceNumber read_seq, size_t ts_sz) {
  if (read_options.ignore_range_deletions ||
-      is_range_del_table_empty_.load(std::memory_order_relaxed)) {
+      is_range_del_table_empty_.LoadRelaxed()) {
    return nullptr;
  }
  if (!timestamp_stripping_fragmented_range_tombstone_list_) {
@@ -790,8 +801,13 @@ FragmentedRangeTombstoneIterator* MemTable::NewRangeTombstoneIteratorInternal(

  // takes current cache
  std::shared_ptr<FragmentedRangeTombstoneListCache> cache =
+#if defined(__cpp_lib_atomic_shared_ptr)
+      cached_range_tombstone_.Access()->load(std::memory_order_relaxed)
+#else
      std::atomic_load_explicit(cached_range_tombstone_.Access(),
-                                std::memory_order_relaxed);
+                                std::memory_order_relaxed)
+#endif
+      ;
  // construct fragmented tombstone list if necessary
  if (!cache->initialized.load(std::memory_order_acquire)) {
    cache->reader_mutex.lock();
@@ -815,7 +831,7 @@ void MemTable::ConstructFragmentedRangeTombstones() {
  // There should be no concurrent Construction.
  // We could also check fragmented_range_tombstone_list_ to avoid repeate
  // constructions. We just construct them here again to be safe.
-  if (!is_range_del_table_empty_.load(std::memory_order_relaxed)) {
+  if (!is_range_del_table_empty_.LoadRelaxed()) {
    // TODO: plumb Env::IOActivity, Env::IOPriority
    auto* unfragmented_iter = new MemTableIterator(
        MemTableIterator::kRangeDelEntries, *this, ReadOptions());
@@ -838,7 +854,7 @@ ReadOnlyMemTable::MemTableStats MemTable::ApproximateStats(
  if (entry_count == 0) {
    return {0, 0};
  }
-  uint64_t n = num_entries_.load(std::memory_order_relaxed);
+  uint64_t n = num_entries_.LoadRelaxed();
  if (n == 0) {
    return {0, 0};
  }
@@ -848,7 +864,7 @@ ReadOnlyMemTable::MemTableStats MemTable::ApproximateStats(
    // the inaccuracy.
    entry_count = n;
  }
-  uint64_t data_size = data_size_.load(std::memory_order_relaxed);
+  uint64_t data_size = data_size_.LoadRelaxed();
  return {entry_count * (data_size / n), entry_count};
 }

@@ -978,17 +994,14 @@ Status MemTable::Add(SequenceNumber s, ValueType type,

    // this is a bit ugly, but is the way to avoid locked instructions
    // when incrementing an atomic
-    num_entries_.store(num_entries_.load(std::memory_order_relaxed) + 1,
-                       std::memory_order_relaxed);
-    data_size_.store(data_size_.load(std::memory_order_relaxed) + encoded_len,
-                     std::memory_order_relaxed);
+    num_entries_.StoreRelaxed(num_entries_.LoadRelaxed() + 1);
+    data_size_.StoreRelaxed(data_size_.LoadRelaxed() + encoded_len);
    if (type == kTypeDeletion || type == kTypeSingleDeletion ||
        type == kTypeDeletionWithTimestamp) {
-      num_deletes_.store(num_deletes_.load(std::memory_order_relaxed) + 1,
-                         std::memory_order_relaxed);
+      num_deletes_.StoreRelaxed(num_deletes_.LoadRelaxed() + 1);
    } else if (type == kTypeRangeDeletion) {
-      uint64_t val = num_range_deletes_.load(std::memory_order_relaxed) + 1;
-      num_range_deletes_.store(val, std::memory_order_relaxed);
+      uint64_t val = num_range_deletes_.LoadRelaxed() + 1;
+      num_range_deletes_.StoreRelaxed(val);
    }

    if (bloom_filter_ && prefix_extractor_ &&
@@ -1059,6 +1072,16 @@ Status MemTable::Add(SequenceNumber s, ValueType type,
      range_del_mutex_.lock();
    }
    for (size_t i = 0; i < size; ++i) {
+#if defined(__cpp_lib_atomic_shared_ptr)
+      std::atomic<std::shared_ptr<FragmentedRangeTombstoneListCache>>*
+          local_cache_ref_ptr = cached_range_tombstone_.AccessAtCore(i);
+      auto new_local_cache_ref = std::make_shared<
+          const std::shared_ptr<FragmentedRangeTombstoneListCache>>(new_cache);
+      std::shared_ptr<FragmentedRangeTombstoneListCache> aliased_ptr(
+          new_local_cache_ref, new_cache.get());
+      local_cache_ref_ptr->store(std::move(aliased_ptr),
+                                 std::memory_order_relaxed);
+#else
      std::shared_ptr<FragmentedRangeTombstoneListCache>* local_cache_ref_ptr =
          cached_range_tombstone_.AccessAtCore(i);
      auto new_local_cache_ref = std::make_shared<
@@ -1073,12 +1096,13 @@ Status MemTable::Add(SequenceNumber s, ValueType type,
          std::shared_ptr<FragmentedRangeTombstoneListCache>(
              new_local_cache_ref, new_cache.get()),
          std::memory_order_relaxed);
+#endif
    }

    if (allow_concurrent) {
      range_del_mutex_.unlock();
    }
-    is_range_del_table_empty_.store(false, std::memory_order_relaxed);
+    is_range_del_table_empty_.StoreRelaxed(false);
  }
  UpdateOldestKeyTime();

@@ -1497,7 +1521,7 @@ void MemTable::MultiGet(const ReadOptions& read_options, MultiGetRange* range,
  // range tombstones. This is the simplest way to ensure range tombstones are
  // handled. TODO: allow Bloom checks where max_covering_tombstone_seq==0
  bool no_range_del = read_options.ignore_range_deletions ||
-                      is_range_del_table_empty_.load(std::memory_order_relaxed);
+                      is_range_del_table_empty_.LoadRelaxed();
  MultiGetRange temp_range(*range, range->begin(), range->end());
  if (bloom_filter_ && no_range_del) {
    bool whole_key =
@@ -8,7 +8,6 @@
 // found in the LICENSE file. See the AUTHORS file for names of contributors.

 #pragma once
-#include <atomic>
 #include <deque>
 #include <functional>
 #include <memory>
@@ -568,7 +567,7 @@ class MemTable final : public ReadOnlyMemTable {
  // As a cheap version of `ApproximateMemoryUsage()`, this function doesn't
  // require external synchronization. The value may be less accurate though
  size_t ApproximateMemoryUsageFast() const {
-    return approximate_memory_usage_.load(std::memory_order_relaxed);
+    return approximate_memory_usage_.LoadRelaxed();
  }

  size_t MemoryAllocatedBytes() const override {
@@ -689,16 +688,13 @@ class MemTable final : public ReadOnlyMemTable {
  // Used in concurrent memtable inserts.
  void BatchPostProcess(const MemTablePostProcessInfo& update_counters) {
    table_->BatchPostProcess();
-    num_entries_.fetch_add(update_counters.num_entries,
-                           std::memory_order_relaxed);
-    data_size_.fetch_add(update_counters.data_size, std::memory_order_relaxed);
+    num_entries_.FetchAddRelaxed(update_counters.num_entries);
+    data_size_.FetchAddRelaxed(update_counters.data_size);
    if (update_counters.num_deletes != 0) {
-      num_deletes_.fetch_add(update_counters.num_deletes,
-                             std::memory_order_relaxed);
+      num_deletes_.FetchAddRelaxed(update_counters.num_deletes);
    }
    if (update_counters.num_range_deletes > 0) {
-      num_range_deletes_.fetch_add(update_counters.num_range_deletes,
-                                   std::memory_order_relaxed);
+      num_range_deletes_.FetchAddRelaxed(update_counters.num_range_deletes);
      // noop for skip-list memtable
      // Besides correctness test in stress test, memtable flush record count
      // check will catch this if it were not noop.
@@ -707,35 +703,26 @@ class MemTable final : public ReadOnlyMemTable {
    UpdateFlushState();
  }

-  uint64_t NumEntries() const override {
-    return num_entries_.load(std::memory_order_relaxed);
-  }
+  uint64_t NumEntries() const override { return num_entries_.LoadRelaxed(); }

-  uint64_t NumDeletion() const override {
-    return num_deletes_.load(std::memory_order_relaxed);
-  }
+  uint64_t NumDeletion() const override { return num_deletes_.LoadRelaxed(); }

  uint64_t NumRangeDeletion() const override {
-    return num_range_deletes_.load(std::memory_order_relaxed);
+    return num_range_deletes_.LoadRelaxed();
  }

-  uint64_t GetDataSize() const override {
-    return data_size_.load(std::memory_order_relaxed);
-  }
+  uint64_t GetDataSize() const override { return data_size_.LoadRelaxed(); }

-  size_t write_buffer_size() const {
-    return write_buffer_size_.load(std::memory_order_relaxed);
-  }
+  size_t write_buffer_size() const { return write_buffer_size_.LoadRelaxed(); }

  // Dynamically change the memtable's capacity. If set below the current usage,
  // the next key added will trigger a flush. Can only increase size when
  // memtable prefix bloom is disabled, since we can't easily allocate more
-  // space.
+  // space. Non-atomic update ok because this is only called with DB mutex held.
  void UpdateWriteBufferSize(size_t new_write_buffer_size) {
    if (bloom_filter_ == nullptr ||
-        new_write_buffer_size < write_buffer_size_) {
-      write_buffer_size_.store(new_write_buffer_size,
-                               std::memory_order_relaxed);
+        new_write_buffer_size < write_buffer_size_.LoadRelaxed()) {
+      write_buffer_size_.StoreRelaxed(new_write_buffer_size);
    }
  }

@@ -827,7 +814,7 @@ class MemTable final : public ReadOnlyMemTable {

  bool IsFragmentedRangeTombstonesConstructed() const override {
    return fragmented_range_tombstone_list_.get() != nullptr ||
-           is_range_del_table_empty_;
+           is_range_del_table_empty_.LoadRelaxed();
  }

  //  Gets the newest user defined timestamps in the memtable. This should only
@@ -853,16 +840,22 @@ class MemTable final : public ReadOnlyMemTable {
  ConcurrentArena arena_;
  std::unique_ptr<MemTableRep> table_;
  std::unique_ptr<MemTableRep> range_del_table_;
-  std::atomic_bool is_range_del_table_empty_;
+  // This is OK to be relaxed access because consistency between table_ and
+  // range_del_table_ is provided by explicit multi-versioning with sequence
+  // numbers. It's ok for stale memory to say the range_del_table_ is empty when
+  // it's actually not because if it was relevant to our read (based on sequence
+  // number), the relaxed memory read would get a sufficiently updated value
+  // because of the ordering provided by LastPublishedSequence().
+  RelaxedAtomic<bool> is_range_del_table_empty_;

  // Total data size of all data inserted
-  std::atomic<uint64_t> data_size_;
-  std::atomic<uint64_t> num_entries_;
-  std::atomic<uint64_t> num_deletes_;
-  std::atomic<uint64_t> num_range_deletes_;
+  RelaxedAtomic<uint64_t> data_size_;
+  RelaxedAtomic<uint64_t> num_entries_;
+  RelaxedAtomic<uint64_t> num_deletes_;
+  RelaxedAtomic<uint64_t> num_range_deletes_;

  // Dynamically changeable memtable option
-  std::atomic<size_t> write_buffer_size_;
+  RelaxedAtomic<size_t> write_buffer_size_;

  // The sequence number of the kv that was inserted first
  std::atomic<SequenceNumber> first_seqno_;
@@ -898,7 +891,7 @@ class MemTable final : public ReadOnlyMemTable {

  // keep track of memory usage in table_, arena_, and range_del_table_.
  // Gets refreshed inside `ApproximateMemoryUsage()` or `ShouldFlushNow`
-  std::atomic<uint64_t> approximate_memory_usage_;
+  RelaxedAtomic<uint64_t> approximate_memory_usage_;

  // max range deletions in a memtable,  before automatic flushing, 0 for
  // unlimited.
@@ -949,9 +942,15 @@ class MemTable final : public ReadOnlyMemTable {

  // makes sure there is a single range tombstone writer to invalidate cache
  std::mutex range_del_mutex_;
+#if defined(__cpp_lib_atomic_shared_ptr)
+  CoreLocalArray<
+      std::atomic<std::shared_ptr<FragmentedRangeTombstoneListCache>>>
+      cached_range_tombstone_;
+#else
  CoreLocalArray<std::shared_ptr<FragmentedRangeTombstoneListCache>>
      cached_range_tombstone_;

+#endif
  void UpdateEntryChecksum(const ProtectionInfoKVOS64* kv_prot_info,
                           const Slice& key, const Slice& value, ValueType type,
                           SequenceNumber s, char* checksum_ptr);
@@ -497,6 +497,7 @@ Status MergeHelper::MergeUntil(InternalIterator* iter,
          ikey.sequence <= latest_snapshot_
              ? CompactionFilter::Decision::kKeep
              : FilterMerge(orig_ikey.user_key, value_slice);
+      // FIXME: should also check for kRemove here
      if (filter != CompactionFilter::Decision::kRemoveAndSkipUntil &&
          range_del_agg != nullptr &&
          range_del_agg->ShouldDelete(
@@ -250,7 +250,7 @@ class MergeHelper {
  // Parallel with keys_; stores the operands
  mutable MergeContext merge_context_;

-  StopWatchNano filter_timer_;
+  StopWatchNano<> filter_timer_;
  uint64_t total_filter_time_;
  Statistics* stats_;

@@ -0,0 +1,71 @@
+//  Copyright (c) Meta Platforms, Inc. and affiliates.
+//  This source code is licensed under both the GPLv2 (found in the
+//  COPYING file in the root directory) and Apache 2.0 License
+//  (found in the LICENSE.Apache file in the root directory).
+
+#include "rocksdb/db.h"
+
+namespace ROCKSDB_NAMESPACE {
+
+using MultiScanIterator = MultiScan::MultiScanIterator;
+
+MultiScan::MultiScan(const ReadOptions& read_options,
+                     const MultiScanArgs& scan_opts, DB* db,
+                     ColumnFamilyHandle* cfh)
+    : read_options_(read_options), scan_opts_(scan_opts), db_(db), cfh_(cfh) {
+  bool slow_path = false;
+  // Setup read_options with iterate_uuper_bound based on the first scan.
+  // Subsequent scans will update and allocate a new DB iterator as necessary
+  if (scan_opts.GetScanRanges()[0].range.limit) {
+    upper_bound_ = *scan_opts.GetScanRanges()[0].range.limit;
+    read_options_.iterate_upper_bound = &upper_bound_;
+  } else {
+    read_options_.iterate_upper_bound = nullptr;
+  }
+  for (const auto& opts : scan_opts.GetScanRanges()) {
+    // Check that all the ScanOptions either specify an upper bound or not. If
+    // its mixed we take the slow path which avoids calling Prepare: we have to
+    // reallocate the Iterator with updated read_options everytime we switch
+    // between upper bound or no upper bound, which complicates Prepare.
+    if (opts.range.limit.has_value() !=
+        scan_opts.GetScanRanges()[0].range.limit.has_value()) {
+      slow_path = true;
+      break;
+    }
+  }
+  db_iter_.reset(db->NewIterator(read_options_, cfh));
+  if (!slow_path) {
+    db_iter_->Prepare(scan_opts);
+  }
+}
+
+MultiScanIterator& MultiScanIterator::operator++() {
+  if (idx_ >= scan_opts_.size()) {
+    throw std::logic_error("Index out of range");
+  }
+  idx_++;
+  if (idx_ < scan_opts_.size()) {
+    // Check if we need to update read_options_
+    if (scan_opts_[idx_].range.limit.has_value() !=
+        (read_options_.iterate_upper_bound != nullptr)) {
+      if (scan_opts_[idx_].range.limit) {
+        *upper_bound_ = *scan_opts_[idx_].range.limit;
+        read_options_.iterate_upper_bound = upper_bound_;
+      } else {
+        read_options_.iterate_upper_bound = nullptr;
+      }
+      db_iter_.reset(db_->NewIterator(read_options_, cfh_));
+      scan_.Reset(db_iter_.get());
+    } else if (scan_opts_[idx_].range.limit) {
+      *upper_bound_ = *scan_opts_[idx_].range.limit;
+    }
+    db_iter_->Seek(*scan_opts_[idx_].range.start);
+    status_ = db_iter_->status();
+    if (!status_.ok()) {
+      throw MultiScanException(status_);
+    }
+  }
+  return *this;
+}
+
+}  // namespace ROCKSDB_NAMESPACE
@@ -26,6 +26,7 @@ static const std::map<PeriodicTaskType, uint64_t> kDefaultPeriodSeconds = {
    {PeriodicTaskType::kPersistStats, kInvalidPeriodSec},
    {PeriodicTaskType::kFlushInfoLog, 10},
    {PeriodicTaskType::kRecordSeqnoTime, kInvalidPeriodSec},
+    {PeriodicTaskType::kTriggerCompaction, 12 * 60 * 60}  // 12 hours
 };

 static const std::map<PeriodicTaskType, std::string> kPeriodicTaskTypeNames = {
@@ -33,16 +34,20 @@ static const std::map<PeriodicTaskType, std::string> kPeriodicTaskTypeNames = {
    {PeriodicTaskType::kPersistStats, "pst_st"},
    {PeriodicTaskType::kFlushInfoLog, "flush_info_log"},
    {PeriodicTaskType::kRecordSeqnoTime, "record_seq_time"},
+    {PeriodicTaskType::kTriggerCompaction, "trigger_compaction"},
 };

 Status PeriodicTaskScheduler::Register(PeriodicTaskType task_type,
-                                       const PeriodicTaskFunc& fn) {
-  return Register(task_type, fn, kDefaultPeriodSeconds.at(task_type));
+                                       const PeriodicTaskFunc& fn,
+                                       bool run_immediately) {
+  return Register(task_type, fn, kDefaultPeriodSeconds.at(task_type),
+                  run_immediately);
 }

 Status PeriodicTaskScheduler::Register(PeriodicTaskType task_type,
                                       const PeriodicTaskFunc& fn,
-                                       uint64_t repeat_period_seconds) {
+                                       uint64_t repeat_period_seconds,
+                                       bool run_immediately) {
  MutexLock l(&timer_mutex);
  static std::atomic<uint64_t> initial_delay(0);

@@ -65,10 +70,13 @@ Status PeriodicTaskScheduler::Register(PeriodicTaskType task_type,
  std::string unique_id =
      kPeriodicTaskTypeNames.at(task_type) + std::to_string(id_++);

-  bool succeeded = timer_->Add(
-      fn, unique_id,
-      (initial_delay.fetch_add(1) % repeat_period_seconds) * kMicrosInSecond,
-      repeat_period_seconds * kMicrosInSecond);
+  uint64_t initial_delay_micros =
+      (initial_delay.fetch_add(1) % repeat_period_seconds) * kMicrosInSecond;
+  if (!run_immediately) {
+    initial_delay_micros += repeat_period_seconds * kMicrosInSecond;
+  }
+  bool succeeded = timer_->Add(fn, unique_id, initial_delay_micros,
+                               repeat_period_seconds * kMicrosInSecond);
  if (!succeeded) {
    return Status::Aborted("Failed to register periodic task");
  }
@@ -21,6 +21,7 @@ enum class PeriodicTaskType : uint8_t {
  kPersistStats,
  kFlushInfoLog,
  kRecordSeqnoTime,
+  kTriggerCompaction,
  kMax,
 };

@@ -42,13 +43,16 @@ class PeriodicTaskScheduler {
  PeriodicTaskScheduler& operator=(PeriodicTaskScheduler&&) = delete;

  // Register a task with its default repeat period. Thread safe call.
-  Status Register(PeriodicTaskType task_type, const PeriodicTaskFunc& fn);
+  // @param run_immediately If true, the task will run soon after it's
+  // scheduled, instead of waiting for the repeat period.
+  Status Register(PeriodicTaskType task_type, const PeriodicTaskFunc& fn,
+                  bool run_immediately);

  // Register a task with specified repeat period. 0 is an invalid argument
  // (kInvalidPeriodSec). To stop the task, please use Unregister().
  // Thread safe call.
  Status Register(PeriodicTaskType task_type, const PeriodicTaskFunc& fn,
-                  uint64_t repeat_period_seconds);
+                  uint64_t repeat_period_seconds, bool run_immediately);

  // Unregister the task. Thread safe call.
  Status Unregister(PeriodicTaskType task_type);
@@ -56,6 +56,12 @@ TEST_F(PeriodicTaskSchedulerTest, Basic) {
  SyncPoint::GetInstance()->SetCallBack(
      "DBImpl::FlushInfoLog:StartRunning",
      [&](void*) { flush_info_log_counter++; });
+
+  int trigger_compaction_counter = 0;
+  SyncPoint::GetInstance()->SetCallBack(
+      "DBImpl::TriggerPeriodicCompaction:StartRunning",
+      [&](void*) { trigger_compaction_counter++; });
+
  SyncPoint::GetInstance()->EnableProcessing();

  Reopen(options);
@@ -70,7 +76,7 @@ TEST_F(PeriodicTaskSchedulerTest, Basic) {

  const PeriodicTaskScheduler& scheduler =
      dbfull()->TEST_GetPeriodicTaskScheduler();
-  ASSERT_EQ(3, scheduler.TEST_GetValidTaskNum());
+  ASSERT_EQ((int)PeriodicTaskType::kMax - 1, scheduler.TEST_GetValidTaskNum());

  ASSERT_EQ(1, dump_st_counter);
  ASSERT_EQ(1, pst_st_counter);
@@ -103,14 +109,14 @@ TEST_F(PeriodicTaskSchedulerTest, Basic) {
  ASSERT_EQ(3, pst_st_counter);
  ASSERT_EQ(4, flush_info_log_counter);

-  ASSERT_EQ(1u, scheduler.TEST_GetValidTaskNum());
+  ASSERT_EQ(2u, scheduler.TEST_GetValidTaskNum());

  // Re-enable one task
  ASSERT_OK(dbfull()->SetDBOptions({{"stats_dump_period_sec", "5"}}));
  ASSERT_EQ(5u, dbfull()->GetDBOptions().stats_dump_period_sec);
  ASSERT_EQ(0u, dbfull()->GetDBOptions().stats_persist_period_sec);

-  ASSERT_EQ(2, scheduler.TEST_GetValidTaskNum());
+  ASSERT_EQ(3, scheduler.TEST_GetValidTaskNum());

  dbfull()->TEST_WaitForPeriodicTaskRun(
      [&] { mock_clock_->MockSleepForSeconds(static_cast<int>(kPeriodSec)); });
@@ -118,6 +124,16 @@ TEST_F(PeriodicTaskSchedulerTest, Basic) {
  ASSERT_EQ(3, pst_st_counter);
  ASSERT_EQ(5, flush_info_log_counter);

+  ASSERT_EQ(0, trigger_compaction_counter);
+  dbfull()->TEST_WaitForPeriodicTaskRun([&] {
+    mock_clock_->MockSleepForSeconds(static_cast<int>(12 * 60 * 60));
+  });
+  ASSERT_EQ(1, trigger_compaction_counter);
+  dbfull()->TEST_WaitForPeriodicTaskRun([&] {
+    mock_clock_->MockSleepForSeconds(static_cast<int>(12 * 60 * 60));
+  });
+  ASSERT_EQ(2, trigger_compaction_counter);
+
  Close();
 }

@@ -150,7 +166,9 @@ TEST_F(PeriodicTaskSchedulerTest, MultiInstances) {
  auto dbi = static_cast_with_check<DBImpl>(dbs[kInstanceNum - 1]);

  const PeriodicTaskScheduler& scheduler = dbi->TEST_GetPeriodicTaskScheduler();
-  ASSERT_EQ(kInstanceNum * 3, scheduler.TEST_GetValidTaskNum());
+  // kRecordSeqnoTime is not registered since the feature is not enabled
+  ASSERT_EQ(kInstanceNum * ((int)PeriodicTaskType::kMax - 1),
+            scheduler.TEST_GetValidTaskNum());

  int expected_run = kInstanceNum;
  dbi->TEST_WaitForPeriodicTaskRun(
@@ -110,7 +110,7 @@ constexpr uint64_t kUnknownOldestAncesterTime = 0;
 constexpr uint64_t kUnknownNewestKeyTime = 0;
 constexpr uint64_t kUnknownFileCreationTime = 0;
 constexpr uint64_t kUnknownEpochNumber = 0;
-// If `Options::allow_ingest_behind` is true, this epoch number
+// If `Options::cf_allow_ingest_behind` is true, this epoch number
 // will be dedicated to files ingested behind.
 constexpr uint64_t kReservedEpochNumberForFileIngestedBehind = 1;

@@ -117,21 +117,43 @@ Status ListColumnFamiliesHandler::ApplyVersionEdit(
  return s;
 }

-Status FileChecksumRetriever::ApplyVersionEdit(VersionEdit& edit,
-                                               ColumnFamilyData** /*unused*/) {
-  for (const auto& deleted_file : edit.GetDeletedFiles()) {
-    Status s = file_checksum_list_.RemoveOneFileChecksum(deleted_file.second);
-    if (!s.ok()) {
-      return s;
+Status FileChecksumRetriever::FetchFileChecksumList(
+    FileChecksumList& file_checksum_list) {
+  Status s = Status::OK();
+  for (const auto& [cf, file_checksums] : cf_file_checksums_) {
+    [[maybe_unused]] const auto& _ = cf;
+    for (const auto& [file_number, info] : file_checksums) {
+      if (!(s = file_checksum_list.InsertOneFileChecksum(
+                file_number, info.first, info.second))
+               .ok()) {
+        break;
+      }
    }
  }
-  for (const auto& new_file : edit.GetNewFiles()) {
-    Status s = file_checksum_list_.InsertOneFileChecksum(
-        new_file.second.fd.GetNumber(), new_file.second.file_checksum,
-        new_file.second.file_checksum_func_name);
-    if (!s.ok()) {
-      return s;
+  return s;
+}
+
+Status FileChecksumRetriever::ApplyVersionEdit(VersionEdit& edit,
+                                               ColumnFamilyData** /*unused*/) {
+  uint32_t column_family_id = edit.GetColumnFamily();
+  if (edit.IsColumnFamilyDrop()) {
+    cf_file_checksums_.erase(column_family_id);
+  }
+  for (const auto& deleted_file : edit.GetDeletedFiles()) {
+    if (cf_file_checksums_.find(column_family_id) == cf_file_checksums_.end()) {
+      return Status::NotFound();
    }
+    if (cf_file_checksums_[column_family_id].find(deleted_file.second) ==
+        cf_file_checksums_[column_family_id].end()) {
+      return Status::NotFound();
+    }
+    cf_file_checksums_[column_family_id].erase(deleted_file.second);
+  }
+  for (const auto& new_file : edit.GetNewFiles()) {
+    cf_file_checksums_[column_family_id].emplace(
+        new_file.second.fd.GetNumber(),
+        std::make_pair(new_file.second.file_checksum,
+                       new_file.second.file_checksum_func_name));
  }
  for (const auto& new_blob_file : edit.GetBlobFileAdditions()) {
    std::string checksum_value = new_blob_file.GetChecksumValue();
@@ -141,11 +163,9 @@ Status FileChecksumRetriever::ApplyVersionEdit(VersionEdit& edit,
      checksum_value = kUnknownFileChecksum;
      checksum_method = kUnknownFileChecksumFuncName;
    }
-    Status s = file_checksum_list_.InsertOneFileChecksum(
-        new_blob_file.GetBlobFileNumber(), checksum_value, checksum_method);
-    if (!s.ok()) {
-      return s;
-    }
+    cf_file_checksums_[column_family_id].emplace(
+        new_blob_file.GetBlobFileNumber(),
+        std::make_pair(checksum_value, checksum_method));
  }
  return Status::OK();
 }
@@ -80,19 +80,42 @@ class ListColumnFamiliesHandler : public VersionEditHandlerBase {

 class FileChecksumRetriever : public VersionEditHandlerBase {
 public:
-  FileChecksumRetriever(const ReadOptions& read_options, uint64_t max_read_size,
-                        FileChecksumList& file_checksum_list)
-      : VersionEditHandlerBase(read_options, max_read_size),
-        file_checksum_list_(file_checksum_list) {}
+  FileChecksumRetriever(const ReadOptions& read_options, uint64_t max_read_size)
+      : VersionEditHandlerBase(read_options, max_read_size) {}

  ~FileChecksumRetriever() override {}

+  Status FetchFileChecksumList(FileChecksumList& file_checksum_list);
+
 protected:
  Status ApplyVersionEdit(VersionEdit& edit,
                          ColumnFamilyData** /*unused*/) override;

 private:
-  FileChecksumList& file_checksum_list_;
+  // Map from CF to file # to string pair, where first portion of the value
+  // is checksum, and second portion of the value is checksum function name.
+  //
+  // [column family id A]
+  //      |
+  //      |-- [file #1] -> [checksum #1, checksum function name #1]
+  //      |-- [file #2] -> [checksum #2, checksum function name #2]
+  //      |
+  //     ...
+  //      |
+  //      |-- [file #N] -> [checksum #N, checksum function name #N]
+  // [column family id B]
+  //      |
+  //      |-- [file #1] -> [checksum #1, checksum function name #1]
+  //      |
+  //     ...
+  //      |
+  //      |-- [file #M] -> [checksum #M, checksum function name #M]
+  //      |
+  //     ...
+  std::unordered_map<
+      uint32_t,
+      std::unordered_map<uint64_t, std::pair<std::string, std::string>>>
+      cf_file_checksums_;
 };

 using VersionBuilderUPtr = std::unique_ptr<BaseReferencedVersionBuilder>;
@@ -95,6 +95,8 @@ namespace ROCKSDB_NAMESPACE {

 namespace {

+using ScanOptionsMap = std::unordered_map<size_t, MultiScanArgs>;
+
 // Find File in LevelFilesBrief data structure
 // Within an index range defined by left and right
 int FindFileInRange(const InternalKeyComparator& icmp,
@@ -1099,10 +1101,52 @@ class LevelIterator final : public InternalIterator {
    read_seq_ = read_seq;
  }

-  void Prepare(const std::vector<ScanOptions>* scan_opts) override {
-    scan_opts_ = scan_opts;
-    if (file_iter_.iter()) {
-      file_iter_.Prepare(scan_opts_);
+  void Prepare(const MultiScanArgs* so) override {
+    // We assume here that scan_opts is sorted such that
+    // scan_opts[0].range.start < scan_opts[1].range.start, and non overlapping
+    if (so == nullptr) {
+      return;
+    }
+    scan_opts_ = so;
+
+    file_to_scan_opts_ = std::make_unique<ScanOptionsMap>();
+    for (size_t k = 0; k < scan_opts_->size(); k++) {
+      const ScanOptions& opt = scan_opts_->GetScanRanges().at(k);
+      auto start = opt.range.start;
+      auto end = opt.range.limit;
+
+      if (!start.has_value()) {
+        continue;
+      }
+
+      // We can capture this case in the future, but for now lets skip this.
+      if (!end.has_value()) {
+        continue;
+      }
+
+      InternalKey istart(start.value(), kMaxSequenceNumber, kValueTypeForSeek);
+      InternalKey iend(end.value(), 0, kValueTypeForSeekForPrev);
+
+      // TODO: This needs to be optimized, right now we iterate twice, which
+      // we dont need to. We can do this in N rather than 2N.
+      size_t fstart = FindFile(icomparator_, *flevel_, istart.Encode());
+      size_t fend = FindFile(icomparator_, *flevel_, iend.Encode());
+
+      // We need to check the relevant cases
+      // Cases:
+      // 1. [  S        E  ]
+      // 2. [  S  ]  [  E  ]
+      // 3. [  S  ] ...... [  E  ]
+      for (auto i = fstart; i <= fend; i++) {
+        if (i < flevel_->num_files) {
+          (*file_to_scan_opts_)[i].insert(start.value(), end.value(),
+                                          opt.property_bag);
+        }
+      }
+    }
+    // Propagate io colaescing threshold
+    for (auto& file_to_arg : *file_to_scan_opts_) {
+      file_to_arg.second.io_coalesce_threshold = so->io_coalesce_threshold;
    }
  }

@@ -1231,7 +1275,10 @@ class LevelIterator final : public InternalIterator {
  bool prefix_exhausted_ = false;
  // Whether next/prev key is a sentinel key.
  bool to_return_sentinel_ = false;
-  const std::vector<ScanOptions>* scan_opts_;
+  const MultiScanArgs* scan_opts_ = nullptr;
+
+  // Our stored scan_opts for each prefix
+  std::unique_ptr<ScanOptionsMap> file_to_scan_opts_ = nullptr;

  // Sets flags for if we should return the sentinel key next.
  // The condition for returning sentinel is reaching the end of current
@@ -1494,7 +1541,19 @@ bool LevelIterator::SkipEmptyFileForward() {
    // LevelIterator::Seek*, it should also call Seek* into the corresponding
    // range tombstone iterator.
    if (file_iter_.iter() != nullptr) {
-      file_iter_.SeekToFirst();
+      // If we are doing prepared scan opts then we should seek to the values
+      // specified by the scan opts
+      if (scan_opts_ && (*file_to_scan_opts_)[file_index_].size()) {
+        const ScanOptions& opts =
+            file_to_scan_opts_->at(file_index_).GetScanRanges().front();
+        if (opts.range.start.has_value()) {
+          InternalKey target(*opts.range.start.AsPtr(), kMaxSequenceNumber,
+                             kValueTypeForSeek);
+          file_iter_.Seek(target.Encode());
+        }
+      } else {
+        file_iter_.SeekToFirst();
+      }
      if (range_tombstone_iter_) {
        if (*range_tombstone_iter_) {
          (*range_tombstone_iter_)->SeekToFirst();
@@ -1542,10 +1601,14 @@ void LevelIterator::SetFileIterator(InternalIterator* iter) {
  }

  InternalIterator* old_iter = file_iter_.Set(iter);
-  // Since this is a new table iterator, no need to call Prepare() if
-  // scan_opts_ is null
  if (iter && scan_opts_) {
-    file_iter_.Prepare(scan_opts_);
+    if (file_to_scan_opts_.get() &&
+        file_to_scan_opts_->find(file_index_) != file_to_scan_opts_->end()) {
+      const MultiScanArgs& new_opts = file_to_scan_opts_->at(file_index_);
+      file_iter_.Prepare(&new_opts);
+    } else {
+      file_iter_.Prepare(scan_opts_);
+    }
  }

  // Update the read pattern for PrefetchBuffer.
@@ -1582,6 +1645,7 @@ void LevelIterator::InitFileIterator(size_t new_file_index) {
    }
  }
 }
+
 }  // anonymous namespace

 Status Version::GetTableProperties(const ReadOptions& read_options,
@@ -2741,9 +2805,10 @@ void Version::MultiGet(const ReadOptions& read_options, MultiGetRange* range,
          RecordTick(db_statistics_, MULTIGET_COROUTINE_COUNT,
                     mget_tasks.size());
          // Collect all results so far
-          std::vector<Status> statuses = folly::coro::blockingWait(
-              folly::coro::collectAllRange(std::move(mget_tasks))
-                  .scheduleOn(&range->context()->executor()));
+          std::vector<Status> statuses =
+              folly::coro::blockingWait(co_withExecutor(
+                  &range->context()->executor(),
+                  folly::coro::collectAllRange(std::move(mget_tasks))));
          if (s.ok()) {
            for (Status stat : statuses) {
              if (!stat.ok()) {
@@ -3028,9 +3093,10 @@ Status Version::MultiGetAsync(
        assert(waiting.size());
        RecordTick(db_statistics_, MULTIGET_COROUTINE_COUNT, mget_tasks.size());
        // Collect all results so far
-        std::vector<Status> statuses = folly::coro::blockingWait(
-            folly::coro::collectAllRange(std::move(mget_tasks))
-                .scheduleOn(&range->context()->executor()));
+        std::vector<Status> statuses =
+            folly::coro::blockingWait(co_withExecutor(
+                &range->context()->executor(),
+                folly::coro::collectAllRange(std::move(mget_tasks))));
        mget_tasks.clear();
        if (s.ok()) {
          for (Status stat : statuses) {
@@ -3466,7 +3532,9 @@ void VersionStorageInfo::ComputeCompactionScore(
  // maintaining it to be over 1.0, we scale the original score by 10x
  // if it is larger than 1.0.
  const double kScoreScale = 10.0;
-  int max_output_level = MaxOutputLevel(immutable_options.allow_ingest_behind);
+  int max_output_level =
+      MaxOutputLevel(immutable_options.cf_allow_ingest_behind ||
+                     immutable_options.allow_ingest_behind);
  for (int level = 0; level <= MaxInputLevel(); level++) {
    double score;
    if (level == 0) {
@@ -3507,8 +3575,13 @@ void VersionStorageInfo::ComputeCompactionScore(
      }

      if (compaction_style_ == kCompactionStyleFIFO) {
-        score = static_cast<double>(total_size) /
-                mutable_cf_options.compaction_options_fifo.max_table_files_size;
+        auto max_table_files_size =
+            mutable_cf_options.compaction_options_fifo.max_table_files_size;
+        if (max_table_files_size == 0) {
+          // avoid divide 0
+          max_table_files_size = 1;
+        }
+        score = static_cast<double>(total_size) / max_table_files_size;
        if (score < 1 &&
            mutable_cf_options.compaction_options_fifo.allow_compaction) {
          score = std::max(
@@ -3646,6 +3719,7 @@ void VersionStorageInfo::ComputeCompactionScore(
  }
  ComputeFilesMarkedForCompaction(max_output_level);
  ComputeBottommostFilesMarkedForCompaction(
+      immutable_options.cf_allow_ingest_behind ||
      immutable_options.allow_ingest_behind);
  ComputeExpiredTtlFiles(immutable_options, mutable_cf_options.ttl);
  ComputeFilesMarkedForPeriodicCompaction(
@@ -4643,8 +4717,7 @@ void VersionStorageInfo::RecoverEpochNumbers(ColumnFamilyData* cfd,
  if (restart_epoch) {
    cfd->ResetNextEpochNumber();

-    bool reserve_epoch_num_for_file_ingested_behind =
-        cfd->ioptions().allow_ingest_behind;
+    bool reserve_epoch_num_for_file_ingested_behind = cfd->AllowIngestBehind();
    if (reserve_epoch_num_for_file_ingested_behind) {
      uint64_t reserved_epoch_number = cfd->NewEpochNumber();
      assert(reserved_epoch_number ==
@@ -4652,7 +4725,8 @@ void VersionStorageInfo::RecoverEpochNumbers(ColumnFamilyData* cfd,
      ROCKS_LOG_INFO(cfd->ioptions().info_log.get(),
                     "[%s]CF has reserved epoch number %" PRIu64
                     " for files ingested "
-                     "behind since `Options::allow_ingest_behind` is true",
+                     "behind since `Options::allow_ingest_behind` or "
+                     "`Options::cf_allow_ingest_behind` is true",
                     cfd->GetName().c_str(), reserved_epoch_number);
    }
  }
@@ -6421,7 +6495,7 @@ Status VersionSet::ReduceNumberOfLevels(const std::string& dbname,
                      nullptr /*BlockCacheTracer*/, nullptr /*IOTracer*/,
                      /*db_id*/ "",
                      /*db_session_id*/ "", options->daily_offpeak_time_utc,
-                      /*error_handler_*/ nullptr, /*read_only=*/false);
+                      /*error_handler_*/ nullptr, /*unchanging=*/false);
  Status status;

  std::vector<ColumnFamilyDescriptor> dummy;
@@ -669,6 +669,8 @@ class VersionStorageInfo {

  // List of files per level, files in each level are arranged
  // in increasing order of keys
+  // In L0, files are ordered in decreasing epoch number, meaning
+  // more recent updates are ordered first.
  std::vector<FileMetaData*>* files_;

  // Map of all table files in version. Maps file number to (level, position on
@@ -2,6 +2,7 @@ add_executable(db_stress${ARTIFACT_SUFFIX}
  batched_ops_stress.cc
  cf_consistency_stress.cc
  db_stress.cc
+  db_stress_compression_manager.cc
  db_stress_common.cc
  db_stress_driver.cc
  db_stress_filters.cc
@@ -228,6 +228,66 @@ void CompressedCacheSetCapacityThread(void* v) {
  }
 }

+void RemoteCompactionWorkerThread(void* v) {
+  assert(FLAGS_remote_compaction_worker_threads > 0);
+  assert(FLAGS_remote_compaction_worker_interval > 0);
+  auto* thread = static_cast<ThreadState*>(v);
+  SharedState* shared = thread->shared;
+  StressTest* stress_test = shared->GetStressTest();
+  assert(stress_test != nullptr);
+  while (true) {
+    {
+      MutexLock l(shared->GetMutex());
+      if (shared->ShouldStopBgThread()) {
+        shared->IncBgThreadsFinished();
+        if (shared->BgThreadsFinished()) {
+          shared->GetCondVar()->SignalAll();
+        }
+        return;
+      }
+    }
+    std::string job_id;
+    CompactionServiceJobInfo job_info;
+    std::string serialized_input;
+    if (shared->DequeueRemoteCompaction(&job_id, &job_info,
+                                        &serialized_input)) {
+      auto options = stress_test->GetOptions(job_info.cf_id);
+      CompactionServiceOptionsOverride override_options{
+          .file_checksum_gen_factory = options.file_checksum_gen_factory,
+          .merge_operator = options.merge_operator,
+          .compaction_filter = options.compaction_filter,
+          .compaction_filter_factory = options.compaction_filter_factory,
+          .prefix_extractor = options.prefix_extractor,
+          .table_factory = options.table_factory,
+          .sst_partitioner_factory = options.sst_partitioner_factory,
+          .listeners = {},
+          .statistics = options.statistics,
+          .table_properties_collector_factories =
+              options.table_properties_collector_factories};
+      std::string tmp_output_dir = job_info.db_name + "/" + "tmp_output_" +
+                                   db_stress_env->GenerateUniqueId();
+      std::string serialized_output;
+      Status s = DB::OpenAndCompact(OpenAndCompactOptions{}, job_info.db_name,
+                                    tmp_output_dir, serialized_input,
+                                    &serialized_output, override_options);
+      if (!s.ok()) {
+        // Print in stdout instead of stderr to avoid stress test failure,
+        // because OpenAndCompact() failure doesn't necessarily mean
+        // primary db instance failure.
+        fprintf(stdout, "Failed to run OpenAndCompact(%s): %s\n",
+                job_info.db_name.c_str(), s.ToString().c_str());
+      }
+      // Add the output regardless of status, so that primary DB doesn't rely on
+      // the timeout to finish waiting. The actual failure from the
+      // deserialization can fail the compaction properly
+      shared->AddRemoteCompactionResult(job_id, serialized_output);
+    }
+    db_stress_env->SleepForMicroseconds(
+        thread->rand.Next() % FLAGS_remote_compaction_worker_interval * 1000 +
+        1);
+  }
+}
+
 void PrintKeyValue(int cf, uint64_t key, const char* value, size_t sz) {
  if (!FLAGS_verbose) {
    return;
@@ -422,12 +422,14 @@ DECLARE_string(file_temperature_age_thresholds);
 DECLARE_bool(allow_trivial_copy_when_change_temperature);
 DECLARE_uint32(commit_bypass_memtable_one_in);
 DECLARE_bool(track_and_verify_wals);
-DECLARE_bool(enable_remote_compaction);
+DECLARE_int32(remote_compaction_worker_threads);
+DECLARE_int32(remote_compaction_worker_interval);
 DECLARE_bool(auto_refresh_iterator_with_snapshot);
 DECLARE_uint32(memtable_op_scan_flush_trigger);
 DECLARE_uint32(memtable_avg_op_scan_flush_trigger);
 DECLARE_uint32(ingest_wbwi_one_in);
 DECLARE_bool(universal_reduce_file_locking);
+DECLARE_bool(use_multiscan);

 constexpr long KB = 1024;
 constexpr int kRandomValueMaxFactor = 3;
@@ -755,6 +757,8 @@ void PoolSizeChangeThread(void* v);

 void DbVerificationThread(void* v);

+void RemoteCompactionWorkerThread(void* v);
+
 void CompressedCacheSetCapacityThread(void* v);

 void TimestampedSnapshotsThread(void* v);
@@ -5,6 +5,8 @@

 #pragma once

+#include "db_stress_shared_state.h"
+#include "db_stress_tool/db_stress_common.h"
 #include "rocksdb/options.h"

 namespace ROCKSDB_NAMESPACE {
@@ -12,28 +14,82 @@ namespace ROCKSDB_NAMESPACE {
 // Service to simulate Remote Compaction in Stress Test
 class DbStressCompactionService : public CompactionService {
 public:
-  explicit DbStressCompactionService() {}
+  explicit DbStressCompactionService(SharedState* shared)
+      : shared_(shared), aborted_(false) {}

  static const char* kClassName() { return "DbStressCompactionService"; }

  const char* Name() const override { return kClassName(); }

+  static constexpr uint64_t kWaitIntervalInMicros = 10 * 1000;  // 10ms
+  static constexpr uint64_t kWaitTimeoutInMicros =
+      30 * 1000 * 1000;  // 30 seconds
+
  CompactionServiceScheduleResponse Schedule(
-      const CompactionServiceJobInfo& /*info*/,
-      const std::string& /*compaction_service_input*/) override {
+      const CompactionServiceJobInfo& info,
+      const std::string& compaction_service_input) override {
+    std::string job_id = info.db_id + "_" + info.db_session_id + "_" +
+                         std::to_string(info.job_id);
+    if (aborted_.load()) {
+      return CompactionServiceScheduleResponse(
+          job_id, CompactionServiceJobStatus::kUseLocal);
+    }
+    shared_->EnqueueRemoteCompaction(job_id, info, compaction_service_input);
    CompactionServiceScheduleResponse response(
-        "Implement Me", CompactionServiceJobStatus::kUseLocal);
+        job_id, CompactionServiceJobStatus::kSuccess);
    return response;
  }

-  CompactionServiceJobStatus Wait(const std::string& /*scheduled_job_id*/,
-                                  std::string* /*result*/) override {
-    // TODO - Implement
-    return CompactionServiceJobStatus::kUseLocal;
+  CompactionServiceJobStatus Wait(const std::string& scheduled_job_id,
+                                  std::string* result) override {
+    auto start = Env::Default()->NowMicros();
+    while (Env::Default()->NowMicros() - start < kWaitTimeoutInMicros) {
+      if (aborted_.load()) {
+        return CompactionServiceJobStatus::kUseLocal;
+      }
+      if (shared_->GetRemoteCompactionResult(scheduled_job_id, result).ok()) {
+        if (result && result->empty()) {
+          // Race: Remote worker aborted before client sets aborted_ = true
+          return CompactionServiceJobStatus::kUseLocal;
+        }
+        return CompactionServiceJobStatus::kSuccess;
+      }
+      Env::Default()->SleepForMicroseconds(kWaitIntervalInMicros);
+    }
+    return CompactionServiceJobStatus::kFailure;
  }

-  // TODO - Implement
-  void CancelAwaitingJobs() override {}
+  void OnInstallation(const std::string& scheduled_job_id,
+                      CompactionServiceJobStatus /*status*/) override {
+    // Clean up tmp directory
+    std::string serialized;
+    CompactionServiceResult result;
+    if (shared_->GetRemoteCompactionResult(scheduled_job_id, &serialized)
+            .ok()) {
+      if (CompactionServiceResult::Read(serialized, &result).ok()) {
+        std::vector<std::string> filenames;
+        Status s = Env::Default()->GetChildren(result.output_path, &filenames);
+        for (size_t i = 0; s.ok() && i < filenames.size(); ++i) {
+          s = Env::Default()->DeleteFile(result.output_path + "/" +
+                                         filenames[i]);
+          if (!s.ok()) {
+            // TODO - Handle clean up failure?
+            break;
+          }
+        }
+        if (s.ok()) {
+          Env::Default()->DeleteDir(result.output_path).PermitUncheckedError();
+        }
+      }
+      shared_->RemoveRemoteCompactionResult(scheduled_job_id);
+    }
+  }
+
+  void CancelAwaitingJobs() override { aborted_.store(true); }
+
+ private:
+  SharedState* shared_;
+  std::atomic_bool aborted_{false};
 };

 }  // namespace ROCKSDB_NAMESPACE
@@ -0,0 +1,28 @@
+//  Copyright (c) Meta Platforms, Inc. and affiliates.
+//  This source code is licensed under both the GPLv2 (found in the
+//  COPYING file in the root directory) and Apache 2.0 License
+//  (found in the LICENSE.Apache file in the root directory).
+#include "db_stress_compression_manager.h"
+
+#include "rocksdb/utilities/object_registry.h"
+
+namespace ROCKSDB_NAMESPACE {
+void DbStressCustomCompressionManager::Register() {
+  // We must register any compression managers with a custom
+  // CompatibilityName() so that if it was used in a past invocation but not
+  // the current invocation, we can still read the SST files requiring it.
+  static std::once_flag loaded;
+  std::call_once(loaded, [&]() {
+    TEST_AllowUnsupportedFormatVersion() = true;
+    auto& library = *ObjectLibrary::Default();
+    library.AddFactory<CompressionManager>(
+        DbStressCustomCompressionManager().CompatibilityName(),
+        [](const std::string& /*uri*/,
+           std::unique_ptr<CompressionManager>* guard,
+           std::string* /*errmsg*/) {
+          *guard = std::make_unique<DbStressCustomCompressionManager>();
+          return guard->get();
+        });
+  });
+}
+}  // namespace ROCKSDB_NAMESPACE
@@ -57,6 +57,8 @@ class DbStressCustomCompressionManager : public CompressionManager {
    return decomp;
  }

+  static void Register();
+
 protected:
  std::shared_ptr<CompressionManager> default_ =
      GetBuiltinV2CompressionManager();
@@ -102,6 +102,14 @@ bool RunStressTestImpl(SharedState* shared) {
    shared->IncBgThreads();
  }

+  uint32_t remote_compaction_worker_thread_count =
+      FLAGS_remote_compaction_worker_threads;
+  if (remote_compaction_worker_thread_count > 0) {
+    for (uint32_t i = 0; i < remote_compaction_worker_thread_count; i++) {
+      shared->IncBgThreads();
+    }
+  }
+
  std::vector<ThreadState*> threads(n);
  for (uint32_t i = 0; i < n; i++) {
    threads[i] = new ThreadState(i, shared);
@@ -126,6 +134,17 @@ bool RunStressTestImpl(SharedState* shared) {
                               &compressed_cache_set_capacity_thread);
  }

+  std::vector<ThreadState*> remote_compaction_worker_threads;
+  if (remote_compaction_worker_thread_count > 0) {
+    remote_compaction_worker_threads.reserve(
+        remote_compaction_worker_thread_count);
+    for (uint32_t i = 0; i < remote_compaction_worker_thread_count; i++) {
+      ThreadState* ts = new ThreadState(i, shared);
+      remote_compaction_worker_threads.push_back(ts);
+      db_stress_env->StartThread(RemoteCompactionWorkerThread, ts);
+    }
+  }
+
  // Each thread goes through the following states:
  // initializing -> wait for others to init -> read/populate/depopulate
  // wait for others to operate -> verify -> done
@@ -218,6 +237,7 @@ bool RunStressTestImpl(SharedState* shared) {
    delete threads[i];
    threads[i] = nullptr;
  }
+
  now = clock->NowMicros();
  if (!FLAGS_skip_verifydb && !FLAGS_test_batches_snapshots &&
      !shared->HasVerificationFailedYet()) {
@@ -232,7 +252,8 @@ bool RunStressTestImpl(SharedState* shared) {
  if (FLAGS_compaction_thread_pool_adjust_interval > 0 ||
      FLAGS_continuous_verification_interval > 0 ||
      FLAGS_compressed_secondary_cache_size > 0 ||
-      FLAGS_compressed_secondary_cache_ratio > 0.0) {
+      FLAGS_compressed_secondary_cache_ratio > 0.0 ||
+      remote_compaction_worker_thread_count > 0) {
    MutexLock l(shared->GetMutex());
    shared->SetShouldStopBgThread();
    while (!shared->BgThreadsFinished()) {
@@ -240,6 +261,15 @@ bool RunStressTestImpl(SharedState* shared) {
    }
  }

+  assert(remote_compaction_worker_threads.size() ==
+         remote_compaction_worker_thread_count);
+  if (remote_compaction_worker_thread_count > 0) {
+    for (ThreadState* thread_state : remote_compaction_worker_threads) {
+      delete thread_state;
+    }
+    remote_compaction_worker_threads.clear();
+  }
+
  if (shared->HasVerificationFailedYet()) {
    fprintf(stderr, "Verification failed :(\n");
    return false;
@@ -844,8 +844,14 @@ DEFINE_bool(track_and_verify_wals,
            ROCKSDB_NAMESPACE::Options().track_and_verify_wals,
            "See Options::track_and_verify_wals");

-DEFINE_bool(enable_remote_compaction, false,
-            "Enable (simulated) Remote Compaction");
+DEFINE_int32(
+    remote_compaction_worker_threads, 2,
+    "Remote Compaction Worker Thread count. If 0, remote compaction is "
+    "disabled");
+
+DEFINE_int32(remote_compaction_worker_interval, 10,
+             "Remote Compaction Worker Thread dequeue tasks every N "
+             "milliseconds. (Default: 10ms)");

 DEFINE_uint32(ingest_wbwi_one_in, 0,
              "If set, will call"
@@ -1495,4 +1501,8 @@ DEFINE_bool(
        .compaction_options_universal.reduce_file_locking,
    "Sets "
    "ColumnFamilyOptions().compaciton_options_universal.reduce_file_locking.");
+
+DEFINE_bool(use_multiscan, false,
+            "If set, use the batched MultiScan API for scans.");
+
 #endif  // GFLAGS
@@ -276,6 +276,53 @@ class SharedState {
    return expected_state_manager_->GetPersistedSeqno();
  }

+  void EnqueueRemoteCompaction(const std::string& job_id,
+                               const CompactionServiceJobInfo& job_info,
+                               const std::string& serialized_input) {
+    MutexLock l(&remote_compaction_queue_mu_);
+    remote_compaction_queue_.emplace(job_id, job_info, serialized_input);
+  }
+
+  bool DequeueRemoteCompaction(std::string* job_id,
+                               CompactionServiceJobInfo* job_info,
+                               std::string* serialized_input) {
+    assert(job_id);
+    assert(job_info);
+    assert(serialized_input);
+    MutexLock l(&remote_compaction_queue_mu_);
+    if (!remote_compaction_queue_.empty()) {
+      const auto [id, info, input] = remote_compaction_queue_.front();
+      *job_id = id;
+      *job_info = info;
+      *serialized_input = input;
+      remote_compaction_queue_.pop();
+      return true;
+    }
+    return false;
+  }
+
+  void AddRemoteCompactionResult(const std::string& job_id,
+                                 const std::string& result) {
+    MutexLock l(&remote_compaction_result_map_mu_);
+    remote_compaction_result_map_.emplace(job_id, result);
+  }
+
+  Status GetRemoteCompactionResult(const std::string& job_id,
+                                   std::string* result) {
+    MutexLock l(&remote_compaction_result_map_mu_);
+    if (remote_compaction_result_map_.find(job_id) !=
+        remote_compaction_result_map_.end()) {
+      *result = remote_compaction_result_map_.at(job_id);
+      return Status::OK();
+    }
+    return Status::NotFound();
+  }
+
+  void RemoveRemoteCompactionResult(const std::string& job_id) {
+    MutexLock l(&remote_compaction_result_map_mu_);
+    remote_compaction_result_map_.erase(job_id);
+  }
+
  // Prepare a Put that will be started but not finish yet
  // This is useful for crash-recovery testing when the process may crash
  // before updating the corresponding expected value
@@ -430,6 +477,16 @@ class SharedState {
  std::atomic<bool> verification_failure_;
  std::atomic<bool> should_stop_test_;

+  // Queue for the remote compaction. Tuple of job id, job info and serialized
+  // compaction_service_input
+  port::Mutex remote_compaction_queue_mu_;
+  std::queue<std::tuple<std::string, CompactionServiceJobInfo, std::string>>
+      remote_compaction_queue_;
+  // Result Map for the remote compaciton. Key is the scheduled_job_id and value
+  // is serialized compaction_service_result
+  port::Mutex remote_compaction_result_map_mu_;
+  std::unordered_map<std::string, std::string> remote_compaction_result_map_;
+
  // Keys that should not be overwritten
  const std::unordered_set<int64_t> no_overwrite_ids_;

@@ -793,6 +793,12 @@ Status StressTest::SetOptions(ThreadState* thread) {
  return db_->SetOptions(cfh, opts);
 }

+Options StressTest::GetOptions(int cf_id) {
+  auto cfh = column_families_[cf_id];
+  assert(cfh);
+  return db_->GetOptions(cfh);
+}
+
 void StressTest::ProcessRecoveredPreparedTxns(SharedState* shared) {
  assert(txn_db_);
  std::vector<Transaction*> recovered_prepared_trans;
@@ -876,6 +882,10 @@ Status StressTest::CommitTxn(Transaction& txn, ThreadState* thread) {
    return Status::InvalidArgument("CommitTxn when FLAGS_use_txn is not set");
  }
  Status s = Status::OK();
+  // We don't issue write to transaction's underlying WriteBatch in stress test
+  assert(txn.GetWriteBatch()->GetWriteBatch()->Count());
+  assert(txn.GetWriteBatch()->GetWBWIOpCount() ==
+         txn.GetWriteBatch()->GetWriteBatch()->Count());
  if (FLAGS_use_optimistic_txn) {
    assert(optimistic_txn_db_);
    s = txn.Commit();
@@ -1454,9 +1464,23 @@ void StressTest::OperateDb(ThreadState* thread) {
      } else if (prob_op < iterate_bound) {
        assert(delrange_bound <= prob_op);
        // OPERATION iterate
-        if (!FLAGS_skip_verifydb &&
-            thread->rand.OneInOpt(
-                FLAGS_verify_iterator_with_expected_state_one_in)) {
+        if (FLAGS_use_multiscan) {
+          int num_seeks = static_cast<int>(
+              std::min(static_cast<uint64_t>(thread->rand.Uniform(64)),
+                       static_cast<uint64_t>(FLAGS_ops_per_thread - i - 1)));
+          // Generate 2x num_seeks random keys, as each scan has a start key
+          // and an upper bound
+          rand_keys = GenerateNKeys(thread, num_seeks * 2, i);
+          i += num_seeks - 1;
+          ThreadStatusUtil::SetEnableTracking(FLAGS_enable_thread_tracking);
+          ThreadStatusUtil::SetThreadOperation(
+              ThreadStatus::OperationType::OP_DBITERATOR);
+          Status s;
+          s = TestMultiScan(thread, read_opts, rand_column_families, rand_keys);
+          ThreadStatusUtil::ResetThreadStatus();
+        } else if (!FLAGS_skip_verifydb &&
+                   thread->rand.OneInOpt(
+                       FLAGS_verify_iterator_with_expected_state_one_in)) {
          ThreadStatusUtil::SetEnableTracking(FLAGS_enable_thread_tracking);
          ThreadStatusUtil::SetThreadOperation(
              ThreadStatus::OperationType::OP_DBITERATOR);
@@ -1634,6 +1658,172 @@ Status StressTest::TestIterateAttributeGroups(
      verify_func);
 }

+Status StressTest::TestMultiScan(ThreadState* thread,
+                                 const ReadOptions& read_opts,
+                                 const std::vector<int>& rand_column_families,
+                                 const std::vector<int64_t>& rand_keys) {
+  size_t num_scans = rand_keys.size() / 2;
+  assert(!rand_column_families.empty());
+  assert(!rand_keys.empty());
+
+  ManagedSnapshot snapshot_guard(db_);
+
+  ReadOptions ro = read_opts;
+  ro.snapshot = snapshot_guard.snapshot();
+
+  std::string read_ts_str;
+  Slice read_ts_slice;
+  MaybeUseOlderTimestampForRangeScan(thread, read_ts_str, read_ts_slice, ro);
+
+  std::vector<std::string> start_key_strs;
+  std::vector<std::string> end_key_strs;
+  MultiScanArgs scan_opts;
+  start_key_strs.reserve(num_scans);
+  end_key_strs.reserve(num_scans);
+
+  // Will be initialized before Seek() below.
+  Slice ub;
+  ro.iterate_upper_bound = &ub;
+  for (size_t i = 0; i < num_scans * 2; i += 2) {
+    assert(rand_keys[i] <= rand_keys[i + 1]);
+    start_key_strs.emplace_back(Key(rand_keys[i]));
+    end_key_strs.emplace_back(Key(rand_keys[i + 1]));
+    scan_opts.insert(Slice(start_key_strs.back()), Slice(end_key_strs.back()));
+  }
+
+  std::string op_logs;
+  ro.pin_data = thread->rand.OneIn(2);
+  ro.background_purge_on_iterator_cleanup = thread->rand.OneIn(2);
+
+  assert(options_.prefix_extractor.get() == nullptr);
+
+  std::unique_ptr<Iterator> iter;
+  iter.reset(db_->NewIterator(ro, column_families_[rand_column_families[0]]));
+  iter->Prepare(scan_opts);
+
+  constexpr size_t kOpLogsLimit = 50000;
+
+  auto verify_func = [](Iterator* iterator) {
+    if (!VerifyWideColumns(iterator->value(), iterator->columns())) {
+      fprintf(stderr,
+              "Value and columns inconsistent for iterator: value: %s, "
+              "columns: %s\n",
+              iterator->value().ToString(/* hex */ true).c_str(),
+              WideColumnsToHex(iterator->columns()).c_str());
+      return false;
+    }
+    return true;
+  };
+
+  for (const ScanOptions& scan_opt : scan_opts.GetScanRanges()) {
+    if (op_logs.size() > kOpLogsLimit) {
+      // Shouldn't take too much memory for the history log. Clear it.
+      op_logs = "(cleared...)\n";
+    }
+
+    // Set up an iterator, perform the same operations without bounds and with
+    // total order seek, and compare the results. This is to identify bugs
+    // related to bounds, prefix extractor, or reseeking. Sometimes we are
+    // comparing iterators with the same set-up, and it doesn't hurt to check
+    // them to be equal.
+    //
+    // This `ReadOptions` is for validation purposes. Ignore
+    // `FLAGS_rate_limit_user_ops` to avoid slowing any validation.
+    ReadOptions cmp_ro;
+    cmp_ro.timestamp = ro.timestamp;
+    cmp_ro.iter_start_ts = ro.iter_start_ts;
+    cmp_ro.snapshot = snapshot_guard.snapshot();
+    cmp_ro.auto_refresh_iterator_with_snapshot =
+        ro.auto_refresh_iterator_with_snapshot;
+    cmp_ro.total_order_seek = true;
+
+    ColumnFamilyHandle* const cmp_cfh =
+        GetControlCfh(thread, rand_column_families[0]);
+    assert(cmp_cfh);
+
+    std::unique_ptr<Iterator> cmp_iter(db_->NewIterator(cmp_ro, cmp_cfh));
+
+    bool diverged = false;
+
+    assert(scan_opt.range.start);
+    assert(scan_opt.range.limit);
+    Slice key = scan_opt.range.start.value();
+    ub = scan_opt.range.limit.value();
+
+    LastIterateOp last_op;
+    iter->Seek(key);
+    cmp_iter->Seek(key);
+    last_op = kLastOpSeek;
+    op_logs += "S " + key.ToString(true) + " ";
+
+    if (iter->Valid() && ro.allow_unprepared_value) {
+      op_logs += "*";
+
+      if (!iter->PrepareValue()) {
+        assert(!iter->Valid());
+        assert(!iter->status().ok());
+      }
+    }
+
+    if (!iter->status().ok() && IsErrorInjectedAndRetryable(iter->status())) {
+      return iter->status();
+    } else if (!cmp_iter->status().ok() &&
+               IsErrorInjectedAndRetryable(cmp_iter->status())) {
+      return cmp_iter->status();
+    }
+
+    VerifyIterator(thread, cmp_cfh, ro, iter.get(), cmp_iter.get(), last_op,
+                   key, op_logs, verify_func, &diverged);
+
+    while (iter->Valid()) {
+      iter->Next();
+      if (!diverged) {
+        assert(cmp_iter->Valid());
+        cmp_iter->Next();
+      }
+      op_logs += "N";
+
+      if (iter->Valid() && ro.allow_unprepared_value) {
+        op_logs += "*";
+
+        if (!iter->PrepareValue()) {
+          assert(!iter->Valid());
+          assert(!iter->status().ok());
+        }
+      }
+
+      if (!iter->status().ok() && IsErrorInjectedAndRetryable(iter->status())) {
+        return iter->status();
+      } else if (!cmp_iter->status().ok() &&
+                 IsErrorInjectedAndRetryable(cmp_iter->status())) {
+        return cmp_iter->status();
+      }
+
+      VerifyIterator(thread, cmp_cfh, ro, iter.get(), cmp_iter.get(), last_op,
+                     key, op_logs, verify_func, &diverged);
+
+      if (diverged) {
+        const std::vector<ScanOptions>& scanoptions = scan_opts.GetScanRanges();
+        for (const auto& t : scanoptions) {
+          fprintf(stdout, "Multiscan options: %s to %s \n",
+                  t.range.start.value().ToString(true).c_str(),
+                  t.range.limit.value().ToString(true).c_str());
+        }
+        break;
+      }
+    }
+
+    thread->stats.AddIterations(1);
+
+    op_logs += "; ";
+    if (diverged) {
+      break;
+    }
+  }
+
+  return Status::OK();
+}
+
 template <typename IterType, typename NewIterFunc, typename VerifyFunc>
 Status StressTest::TestIterateImpl(ThreadState* thread,
                                   const ReadOptions& read_opts,
@@ -3413,24 +3603,8 @@ void StressTest::Open(SharedState* shared, bool reopen) {
    InitializeOptionsFromFlags(cache_, filter_policy_, options_);
  }
  InitializeOptionsGeneral(cache_, filter_policy_, sqfc_factory_, options_);
-  {
-    // We must register any compression managers with a custom
-    // CompatibilityName() so that if it was used in a past invocation but not
-    // the current invocation, we can still read the SST files requiring it.
-    static std::once_flag loaded;
-    std::call_once(loaded, [&]() {
-      TEST_AllowUnsupportedFormatVersion() = true;
-      auto& library = *ObjectLibrary::Default();
-      library.AddFactory<CompressionManager>(
-          DbStressCustomCompressionManager().CompatibilityName(),
-          [](const std::string& /*uri*/,
-             std::unique_ptr<CompressionManager>* guard,
-             std::string* /*errmsg*/) {
-            *guard = std::make_unique<DbStressCustomCompressionManager>();
-            return guard->get();
-          });
-    });
-  }
+  DbStressCustomCompressionManager::Register();
+
  if (!strcasecmp(FLAGS_compression_manager.c_str(), "custom")) {
    options_.compression_manager =
        std::make_shared<DbStressCustomCompressionManager>();
@@ -3462,6 +3636,27 @@ void StressTest::Open(SharedState* shared, bool reopen) {
            "memtablerep != prefix_hash\n");
  }

+  // Remote Compaction
+  if (FLAGS_remote_compaction_worker_threads > 0) {
+    // TODO(jaykorean) Remove this after fix - remote worker shouldn't recover
+    // from WAL
+    if (!FLAGS_disable_wal) {
+      fprintf(stderr,
+              "WAL is not compatible with Remote Compaction in Stress Test\n");
+      exit(1);
+    }
+    if ((options_.enable_blob_files ||
+         options_.enable_blob_garbage_collection ||
+         FLAGS_allow_setting_blob_options_dynamically)) {
+      fprintf(stderr,
+              "Integrated BlobDB is currently incompatible with Remote "
+              "Compaction\n");
+      exit(1);
+    }
+    options_.compaction_service =
+        std::make_shared<DbStressCompactionService>(shared);
+  }
+
  if ((options_.enable_blob_files || options_.enable_blob_garbage_collection ||
       FLAGS_allow_setting_blob_options_dynamically) &&
      FLAGS_best_efforts_recovery) {
@@ -4370,11 +4565,6 @@ void InitializeOptionsFromFlags(
  options.inplace_update_support = FLAGS_inplace_update_support;
  options.uncache_aggressiveness = FLAGS_uncache_aggressiveness;

-  // Remote Compaction
-  if (FLAGS_enable_remote_compaction) {
-    options.compaction_service = std::make_shared<DbStressCompactionService>();
-  }
-
  options.memtable_op_scan_flush_trigger = FLAGS_memtable_op_scan_flush_trigger;
  options.compaction_options_universal.reduce_file_locking =
      FLAGS_universal_reduce_file_locking;
@@ -53,6 +53,7 @@ class StressTest {
    Status s = db_->EnableAutoCompaction(column_families_);
    return s;
  }
+  Options GetOptions(int cf_id);
  void CleanUp();

 protected:
@@ -274,6 +275,10 @@ class StressTest {
    return Status::NotSupported();
  }

+  Status TestMultiScan(ThreadState* thread, const ReadOptions& read_opts,
+                       const std::vector<int>& rand_column_families,
+                       const std::vector<int64_t>& rand_keys);
+
  // Enum used by VerifyIterator() to identify the mode to validate.
  enum LastIterateOp {
    kLastOpSeek,
@@ -100,6 +100,10 @@ class CompositeRandomAccessFileWrapper : public RandomAccessFile {
    return target_->InvalidateCache(offset, length);
  }

+  Status GetFileSize(uint64_t* size) override {
+    return target_->GetFileSize(size);
+  }
+
 private:
  std::unique_ptr<FSRandomAccessFile> target_;
 };
@@ -186,6 +186,10 @@ class LegacyRandomAccessFileWrapper : public FSRandomAccessFile {
  IOStatus InvalidateCache(size_t offset, size_t length) override {
    return status_to_io_status(target_->InvalidateCache(offset, length));
  }
+  IOStatus GetFileSize(uint64_t* result) override {
+    auto status = target_->GetFileSize(result);
+    return status_to_io_status(std::move(status));
+  }

 private:
  std::unique_ptr<RandomAccessFile> target_;
@@ -665,17 +665,52 @@ class EncryptedFileSystemImpl : public EncryptedFileSystem {
                              std::unique_ptr<FSWritableFile>* result,
                              IODebugContext* dbg) override {
    result->reset();
-    if (options.use_mmap_writes) {
+    if (options.use_mmap_reads || options.use_mmap_writes) {
      return IOStatus::InvalidArgument();
    }
+
+    size_t prefix_length = 0;
+    std::unique_ptr<BlockAccessCipherStream> stream;
+
    // Open file using underlying Env implementation
    std::unique_ptr<FSWritableFile> underlying;
-    IOStatus status =
+    auto status =
        FileSystemWrapper::ReopenWritableFile(fname, options, &underlying, dbg);
    if (!status.ok()) {
      return status;
    }
-    return CreateWritableEncryptedFile(fname, underlying, options, result, dbg);
+
+    if (underlying->GetFileSize(options.io_options, dbg) != 0) {
+      // read the cipher stream from file for non-empty file
+      std::unique_ptr<FSRandomAccessFile> underlying_file_reader;
+      status = FileSystemWrapper::NewRandomAccessFile(
+          fname, options, &underlying_file_reader, dbg);
+      if (!status.ok()) {
+        return status;
+      }
+
+      status = CreateRandomReadCipherStream(
+          fname, underlying_file_reader, options, &prefix_length, &stream, dbg);
+
+      if (!status.ok()) {
+        return status;
+      }
+    } else {
+      // create cipher stream for new or empty file
+      status = CreateWritableCipherStream(fname, underlying, options,
+                                          &prefix_length, &stream, dbg);
+      if (!status.ok()) {
+        return status;
+      }
+    }
+
+    if (stream) {
+      result->reset(new EncryptedWritableFile(
+          std::move(underlying), std::move(stream), prefix_length));
+    } else {
+      result->reset(underlying.release());
+    }
+    return status;
  }

  IOStatus ReuseWritableFile(const std::string& fname,
@@ -3467,7 +3467,6 @@ class ReadAsyncRandomAccessFile : public FSRandomAccessFileOwnerWrapper {

 private:
  ReadAsyncFS& fs_;
-  std::unique_ptr<FSRandomAccessFile> file_;
  int counter = 0;
 };

@@ -3657,6 +3656,42 @@ TEST(EnvTestMisc, StaticDestruction) {
  static_destruction_tester.activated = true;
 }

+// Test GetFileSize API
+class TestGetFileSize : public testing::Test {
+ public:
+  TestGetFileSize() { env_ = Env::Default(); }
+  Env* env_;
+};
+
+// Validate GetFileSize API returns the right value.
+// Use the default implementation from env
+TEST_F(TestGetFileSize, GetFileSize) {
+  EnvOptions soptions;
+  auto fs = env_->GetFileSystem();
+
+  std::string fname = test::PerThreadDBPath(env_, "getFileSizeTestfile");
+
+  // randomize file size
+  auto rnd = Random::GetTLSInstance();
+  auto expectedFileSize = rnd->Uniform(256 * 1024) + 1;
+  auto content = rnd->RandomBinaryString(static_cast<int>(expectedFileSize));
+
+  ASSERT_OK(CreateFile(fs.get(), fname, content, false));
+
+  std::unique_ptr<FSRandomAccessFile> file;
+  ASSERT_OK(fs->NewRandomAccessFile(fname, FileOptions(), &file, nullptr));
+
+  uint64_t fileSizeFromFileSystemAPI;
+  ASSERT_OK(
+      fs->GetFileSize(fname, IOOptions(), &fileSizeFromFileSystemAPI, nullptr));
+  ASSERT_EQ(fileSizeFromFileSystemAPI, expectedFileSize);
+
+  uint64_t fileSizeFromFsRandomAccessFileAPI;
+  ASSERT_OK(file->GetFileSize(&fileSizeFromFsRandomAccessFileAPI));
+
+  ASSERT_EQ(fileSizeFromFsRandomAccessFileAPI, expectedFileSize);
+}
+
 }  // namespace ROCKSDB_NAMESPACE

 int main(int argc, char** argv) {
@@ -243,7 +243,7 @@ class PosixFileSystem : public FileSystem {
      // Use mmap when virtual address-space is plentiful.
      uint64_t size;
      IOOptions opts;
-      s = GetFileSize(fname, opts, &size, nullptr);
+      s = GetFileSizeOnOpenedFile(fd, fname, &size);
      if (s.ok()) {
        void* base = mmap(nullptr, size, PROT_READ, MAP_SHARED, fd, 0);
        if (base != MAP_FAILED) {
@@ -324,7 +324,7 @@ class PosixFileSystem : public FileSystem {
    }
    uint64_t initial_file_size = 0;
    if (reopen) {
-      s = GetFileSize(fname, IOOptions(), &initial_file_size, nullptr);
+      s = GetFileSizeOnOpenedFile(fd, fname, &initial_file_size);
      if (!s.ok()) {
        close(fd);
        return s;
@@ -509,7 +509,7 @@ class PosixFileSystem : public FileSystem {
    uint64_t size;
    if (status.ok()) {
      IOOptions opts;
-      status = GetFileSize(fname, opts, &size, nullptr);
+      status = GetFileSizeOnOpenedFile(fd, fname, &size);
    }
    void* base = nullptr;
    if (status.ok()) {
@@ -671,7 +671,7 @@ class PosixFileSystem : public FileSystem {

  IOStatus GetFileSize(const std::string& fname, const IOOptions& /*opts*/,
                       uint64_t* size, IODebugContext* /*dbg*/) override {
-    struct stat sbuf;
+    struct stat sbuf {};
    if (stat(fname.c_str(), &sbuf) != 0) {
      *size = 0;
      return IOError("while stat a file for size", fname, errno);
@@ -974,6 +974,22 @@ class PosixFileSystem : public FileSystem {
 private:
  bool forceMmapOff_ = false;  // do we override Env options?

+  // This is a faster API comparing to the public method that uses stat to get
+  // file size. However this API only works on opened file.
+  IOStatus GetFileSizeOnOpenedFile(const int fd, const std::string& name,
+                                   uint64_t* size) {
+    struct stat sb {};
+    *size = 0;
+    // Get file information using fstat
+    if (fstat(fd, &sb) == -1) {
+      return IOError(
+          "while fstat a file for size with fd " + std::to_string(fd), name,
+          errno);
+    }
+    *size = sb.st_size;
+    return IOStatus::OK();
+  }
+
 #ifdef OS_LINUX
  // Get the minimum "linux system limit" (i.e, the largest I/O size that the OS
  // can issue to block devices under a directory, also known as
@@ -1091,8 +1107,10 @@ class PosixFileSystem : public FileSystem {
        struct io_uring_cqe* cqe = nullptr;
        ssize_t ret = io_uring_wait_cqe(iu, &cqe);
        if (ret) {
-          // abort as it shouldn't be in indeterminate state and there is no
-          // good way currently to handle this error.
+          fprintf(stderr, "Poll: io_uring_wait_cqe failed: %ld", (long)ret);
+          if (ret == -EINTR || ret == -EAGAIN) {
+            continue;  // Retry
+          }
          abort();
        }

@@ -1194,8 +1212,10 @@ class PosixFileSystem : public FileSystem {
        struct io_uring_cqe* cqe = nullptr;
        ssize_t ret = io_uring_wait_cqe(iu, &cqe);
        if (ret) {
-          // abort as it shouldn't be in indeterminate state and there is no
-          // good way currently to handle this error.
+          fprintf(stderr, "AbortIO: io_uring_wait_cqe failed: %ld", (long)ret);
+          if (ret == -EINTR || ret == -EAGAIN) {
+            continue;  // Retry
+          }
          abort();
        }
        assert(cqe != nullptr);
@@ -607,6 +607,17 @@ PosixRandomAccessFile::PosixRandomAccessFile(

 PosixRandomAccessFile::~PosixRandomAccessFile() { close(fd_); }

+IOStatus PosixRandomAccessFile::GetFileSize(uint64_t* result) {
+  struct stat sbuf {};
+  if (fstat(fd_, &sbuf) != 0) {
+    *result = 0;
+    return IOError("While fstat with fd " + std::to_string(fd_), filename_,
+                   errno);
+  }
+  *result = sbuf.st_size;
+  return IOStatus::OK();
+}
+
 IOStatus PosixRandomAccessFile::Read(uint64_t offset, size_t n,
                                     const IOOptions& /*opts*/, Slice* result,
                                     char* scratch,
@@ -1056,6 +1067,11 @@ IOStatus PosixMmapReadableFile::InvalidateCache(size_t offset, size_t length) {
 #endif
 }

+IOStatus PosixMmapReadableFile::GetFileSize(uint64_t* result) {
+  *result = length_;
+  return IOStatus::OK();
+}
+
 /*
 * PosixMmapFile
 *
@@ -352,6 +352,8 @@ class PosixRandomAccessFile : public FSRandomAccessFile {
                             void* cb_arg, void** io_handle,
                             IOHandleDeleter* del_fn,
                             IODebugContext* dbg) override;
+
+  virtual IOStatus GetFileSize(uint64_t* result) override;
 };

 class PosixWritableFile : public FSWritableFile {
@@ -437,6 +439,7 @@ class PosixMmapReadableFile : public FSRandomAccessFile {
                char* scratch, IODebugContext* dbg) const override;
  void Hint(AccessPattern pattern) override;
  IOStatus InvalidateCache(size_t offset, size_t length) override;
+  virtual IOStatus GetFileSize(uint64_t* result) override;
 };

 class PosixMmapFile : public FSWritableFile {
@@ -322,6 +322,11 @@ class MockRandomAccessFile : public FSRandomAccessFile {
    }
  }

+  IOStatus GetFileSize(uint64_t* size) override {
+    *size = file_->Size();
+    return IOStatus::OK();
+  }
+
 private:
  MemFile* file_;
  bool use_direct_io_;
@@ -108,6 +108,10 @@ class ReadaheadRandomAccessFile : public FSRandomAccessFile {

  bool use_direct_io() const override { return file_->use_direct_io(); }

+  IOStatus GetFileSize(uint64_t* result) override {
+    return file_->GetFileSize(result);
+  }
+
 private:
  // Tries to read from buffer_ n bytes starting at offset. If anything was read
  // from the cache, it sets cached_len to the number of bytes actually read,
@@ -162,7 +162,6 @@ class SstFileManagerImpl : public SstFileManager {
  void Close();

  void SetStatisticsPtr(const std::shared_ptr<Statistics>& stats) override {
-    stats_ = stats;
    delete_scheduler_.SetStatisticsPtr(stats);
  }

@@ -216,7 +215,6 @@ class SstFileManagerImpl : public SstFileManager {
  std::list<ErrorHandler*> error_handler_list_;
  // Pointer to ErrorHandler instance that is currently processing recovery
  ErrorHandler* cur_instance_;
-  std::shared_ptr<Statistics> stats_;
 };

 }  // namespace ROCKSDB_NAMESPACE
@@ -19,6 +19,7 @@ namespace ROCKSDB_NAMESPACE {

 // TODO: alias/adapt for compression
 struct FilterBuildingContext;
+class Decompressor;

 // A Compressor represents a very specific but potentially adapting strategy for
 // compressing blocks, including the relevant algorithm(s), options, dictionary,
@@ -50,6 +51,10 @@ struct FilterBuildingContext;
 // a number of built-in CompressionTypes that ignore any dictionary block in
 // the file; therefore they cannot accommodate dictionary compression in the
 // future without a schema change / extension.)
+//
+// Exceptions MUST NOT propagate out of overridden functions into RocksDB,
+// because RocksDB is not exception-safe. This could cause undefined behavior
+// including data loss, unreported corruption, deadlocks, and more.
 class Compressor {
 public:
  Compressor() = default;
@@ -134,15 +139,15 @@ class Compressor {
  // EXTENSIBLE or reinterpret_cast-able by custom Compressor implementations
  struct WorkingArea {};

- protected:
  // To allow for flexible re-use / reclaimation, we have explicit Get and
  // Release functions, and usually wrap in a special RAII smart pointer.
  // For example, a WorkingArea could be saved/recycled in thread-local or
  // core-local storage, or heap managed, etc., though an explicit WorkingArea
  // is only advised for repeated compression (by a single thread).
+  // ReleaseWorkingArea() in not intended to be called directly, but used by
+  // ManagedWorkingArea.
  virtual void ReleaseWorkingArea(WorkingArea*) {}

- public:
  using ManagedWorkingArea =
      ManagedPtr<WorkingArea, Compressor, &Compressor::ReleaseWorkingArea>;

@@ -152,9 +157,11 @@ class Compressor {
    return {};
  }

-  // Compress `uncompressed_data` to `compressed_output`, which should be
-  // passed in empty. Note that the compressed output will be decompressed
-  // by the sequence Decompressor::ExtractUncompressedSize() followed by
+  // Compress `uncompressed_data` to buffer `compressed_output` of size
+  // `*compressed_output_size`, storing the final compressed size in
+  // `*compressed_output_size` and compression type in `*out_compression_type`.
+  // Note that the compressed output will be decompressed by the sequence
+  // Decompressor::ExtractUncompressedSize() followed by
  // Decompressor::DecompressBlock(), which must also be provided the same
  // CompressionType saved in `out_compression_type`. (In many configurations,
  // `compressed_output` will have a prefix storing the uncompressed_data size
@@ -166,28 +173,34 @@ class Compressor {
  // If return status is not OK, then some fatal condition has arisen. On OK
  // status, setting `*out_compression_type = kNoCompression` means compression
  // is declined and the caller should use the original uncompressed_data and
-  // ignore any result in `compressed_output`. Otherwise, compression has
-  // happened with results in `compressed_output` and `out_compression_type`,
-  // which are allowed to vary from call to call.
+  // ignore any result in `compressed_output`. In this case, setting
+  // *compressed_output_size to 0 suggests that compression was quickly
+  // "bypassed" and *compressed_output_size > 0 suggests that compression was
+  // attempted but rejected (e.g. insufficient compression ratio).
+  //
+  // On OK status and `*out_compression_type != kNoCompression`, compression has
+  // happened with results in `compressed_output`, `compressed_output_size`, and
+  // `out_compression_type`. The output compression type is allowed to vary from
+  // call to call but does not for compressors from BuiltinV2CompressionManager.
  //
  // The working area is optional and used to optimize repeated compression by
  // a single thread. ManagedWorkingArea is provided rather than just
  // WorkingArea so that it can be used only if the `owner` matches expectation.
  // This could be useful for a Compressor wrapping more than one alternative
  // underlying Compressor.
-  //
-  // TODO: instead of string, consider a buffer only large enough for max
-  // tolerable compressed size. Does that work for all existing algorithms?
-  // * Looks like Snappy doesn't support that. :(
-  //   * Except perhaps using the Sink interface
-  // * But looks like everything else should. :)
-  // Could save CPU by eliminating extra zero-ing and giving up quicker when
-  // ratio is insufficient.
-  virtual Status CompressBlock(Slice uncompressed_data,
-                               std::string* compressed_output,
+  virtual Status CompressBlock(Slice uncompressed_data, char* compressed_output,
+                               size_t* compressed_output_size,
                               CompressionType* out_compression_type,
                               ManagedWorkingArea* working_area) = 0;

+  // OPTIONAL: Return a decompressor that is optimized for output from this
+  // compressor.
+  virtual std::shared_ptr<Decompressor> GetOptimizedDecompressor() const {
+    // Default implementation: no optimization. Get a Decompressor from the
+    // CompressionManager.
+    return nullptr;
+  }
+
  // TODO: something to populate table properties based on settings, after all
  // or as WorkingAreas released. Maybe also update stats, or that could be in
  // thread-specific WorkingArea.
@@ -221,6 +234,10 @@ class Compressor {
 // decompressed into part of a single buffer allocated to hold a block's
 // uncompressed contents along with an in-memory object representation of the
 // block (to reduce fragmentation and other overheads of separate objects).
+//
+// Exceptions MUST NOT propagate out of overridden functions into RocksDB,
+// because RocksDB is not exception-safe. This could cause undefined behavior
+// including data loss, unreported corruption, deadlocks, and more.
 class Decompressor {
 public:
  Decompressor() = default;
@@ -278,6 +295,9 @@ class Decompressor {
  // supported for this kind of Decompressor. Corruption - dictionary is
  // malformed (though many implementations will accept any data as a
  // dictionary)
+  //
+  // RocksDB promises not to call this function with an empty dictionary slice
+  // (equivalent to no dictionary).
  virtual Status MaybeCloneForDict(const Slice& /*serialized_dict*/,
                                   std::unique_ptr<Decompressor>* /*out*/) {
    return Status::NotSupported(
@@ -339,6 +359,10 @@ class Decompressor {
 //   (because that would break backward compatibility, potential quiet
 //   corruption)
 // TODO: consider adding optional streaming compression support (low priority)
+//
+// Exceptions MUST NOT propagate out of overridden functions into RocksDB,
+// because RocksDB is not exception-safe. This could cause undefined behavior
+// including data loss, unreported corruption, deadlocks, and more.
 class CompressionManager
    : public std::enable_shared_from_this<CompressionManager>,
      public Customizable {
@@ -426,14 +450,6 @@ class CompressionManager
    // Safe default implementation
    return GetDecompressor();
  }
-
-  // Get a decompressor that is allowed to have support only for the
-  // CompressionTypes used by the given Compressor.
-  virtual std::shared_ptr<Decompressor> GetDecompressorForCompressor(
-      const Compressor& compressor) {
-    // Reasonable default implementation
-    return GetDecompressorOptimizeFor(compressor.GetPreferredCompressionType());
-  }
 };

 // ************************* Utility wrappers etc. *********************** //
@@ -466,11 +482,21 @@ class CompressorWrapper : public Compressor {
    return wrapped_->ObtainWorkingArea();
  }

-  Status CompressBlock(Slice uncompressed_data, std::string* compressed_output,
+  // NOTE: Don't need to override ReleaseWorkingArea() here because
+  // ManagedWorkingArea takes care of calling it on the Compressor that created
+  // the WorkingArea.
+
+  Status CompressBlock(Slice uncompressed_data, char* compressed_output,
+                       size_t* compressed_output_size,
                       CompressionType* out_compression_type,
                       ManagedWorkingArea* working_area) override {
    return wrapped_->CompressBlock(uncompressed_data, compressed_output,
-                                   out_compression_type, working_area);
+                                   compressed_output_size, out_compression_type,
+                                   working_area);
+  }
+
+  std::shared_ptr<Decompressor> GetOptimizedDecompressor() const override {
+    return wrapped_->GetOptimizedDecompressor();
  }

 protected:
@@ -491,6 +517,10 @@ class DecompressorWrapper : public Decompressor {
    wrapped_->ReleaseWorkingArea(wa);
  }

+  // NOTE: Don't need to override ReleaseWorkingArea() here because
+  // ManagedWorkingArea takes care of calling it on the Decompressor that
+  // created the WorkingArea.
+
  ManagedWorkingArea ObtainWorkingArea(CompressionType preferred) override {
    return wrapped_->ObtainWorkingArea(preferred);
  }
@@ -569,11 +599,6 @@ class CompressionManagerWrapper : public CompressionManager {
    return wrapped_->GetDecompressorForTypes(types_begin, types_end);
  }

-  std::shared_ptr<Decompressor> GetDecompressorForCompressor(
-      const Compressor& compressor) override {
-    return wrapped_->GetDecompressorForCompressor(compressor);
-  }
-
 protected:
  std::shared_ptr<CompressionManager> wrapped_;
 };
@@ -589,10 +614,14 @@ const std::shared_ptr<CompressionManager>& GetBuiltinV2CompressionManager();
 // lead to unexpected schema changes for user CompressionManagers building on
 // the built-in schema, in the unlikely/rare case of a new built-in schema.

-// Gets CompressionManager designed for the automated compression strategy.
+// Creates CompressionManager designed for the automated compression strategy.
 // This may include deciding to compress or not.
-// In future should be able to select compression algorithm based on the CPU
-// utilization and IO constraints.
+// EXPERIMENTAL
 std::shared_ptr<CompressionManagerWrapper> CreateAutoSkipCompressionManager(
-    std::shared_ptr<CompressionManager> wrapped);
+    std::shared_ptr<CompressionManager> wrapped = nullptr);
+// Creates CompressionManager designed for the CPU and IO cost aware compression
+// strategy
+// EXPERIMENTAL
+std::shared_ptr<CompressionManagerWrapper> CreateCostAwareCompressionManager(
+    std::shared_ptr<CompressionManager> wrapped = nullptr);
 }  // namespace ROCKSDB_NAMESPACE
@@ -526,7 +526,7 @@ struct AdvancedColumnFamilyOptions {
  // By doing it, we give max_bytes_for_level_multiplier a priority against
  // max_bytes_for_level_base, for a more predictable LSM tree shape. It is
  // useful to limit worse case space amplification.
-  // If `allow_ingest_behind=true` or `preclude_last_level_data_seconds > 0`,
+  // If `cf_allow_ingest_behind=true` or `preclude_last_level_data_seconds > 0`,
  // then the last level is reserved, and we will start filling LSM from the
  // second last level.
  //
@@ -1146,6 +1146,33 @@ struct AdvancedColumnFamilyOptions {
  // Dynamically changeable through the SetOptions() API.
  uint32_t memtable_avg_op_scan_flush_trigger = 0;

+  // If either DBOptions::allow_ingest_behind or this option is set to true,
+  // this column family will prepare for ingesting files to the last level
+  // (IngestExternalFiles() with ingest_behind=true). Users should set only
+  // this option since DBOptions::allow_ingest_behind is deprecated.
+  //
+  // Specifically, preparing a column family for ingesting files to the last
+  // level has the following effects:
+  // 1) Disables some internal optimizations around SST file compression.
+  // 2) Reserves the last level for ingested files only.
+  // 3) Compaction will not include any file from the last level.
+  // 4) Compaction will preserve necessary tombstones that can apply on
+  // top of ingested files.
+  //
+  // Note that only Universal Compaction supports cf_allow_ingest_behind.
+  // `num_levels` should be >= 3 if this option is turned on.
+  //
+  // Note that this option needs to be set to true before any write to the CF.
+  // It's recommended to set the option to true since CF creation. Otherwise,
+  // ingestion with ingest_behind = true might fail. Once file ingestions are
+  // done, the option should be flipped to false. Flipping this option to false
+  // allows the CF to disable the behavior changes detailed above and resume
+  // more efficient operation.
+  //
+  // Default: false
+  // Immutable.
+  bool cf_allow_ingest_behind = false;
+
  // Create ColumnFamilyOptions with default values for all fields
  AdvancedColumnFamilyOptions();
  // Create ColumnFamilyOptions from Options
@@ -1154,6 +1154,8 @@ rocksdb_block_based_options_set_unpartitioned_pinning_tier(
    rocksdb_block_based_table_options_t*, int);
 extern ROCKSDB_LIBRARY_API void rocksdb_options_set_write_buffer_manager(
    rocksdb_options_t* opt, rocksdb_write_buffer_manager_t* wbm);
+extern ROCKSDB_LIBRARY_API void rocksdb_options_set_sst_file_manager(
+    rocksdb_options_t* opt, rocksdb_sst_file_manager_t* sfm);

 /* Flush job info */

@@ -1171,6 +1173,10 @@ extern ROCKSDB_LIBRARY_API uint64_t
 rocksdb_flushjobinfo_smallest_seqno(const rocksdb_flushjobinfo_t*);
 extern ROCKSDB_LIBRARY_API void rocksdb_reset_status(
    rocksdb_status_ptr_t* status_ptr);
+extern ROCKSDB_LIBRARY_API uint32_t
+rocksdb_flushjobinfo_flush_reason(const rocksdb_flushjobinfo_t* info);
+extern ROCKSDB_LIBRARY_API void rocksdb_status_ptr_get_error(
+    rocksdb_status_ptr_t* status, char** errptr);

 /* Compaction job info */
 extern ROCKSDB_LIBRARY_API void rocksdb_compactionjobinfo_status(
@@ -1221,6 +1227,9 @@ extern ROCKSDB_LIBRARY_API int rocksdb_subcompactionjobinfo_base_input_level(
    const rocksdb_subcompactionjobinfo_t*);
 extern ROCKSDB_LIBRARY_API int rocksdb_subcompactionjobinfo_output_level(
    const rocksdb_subcompactionjobinfo_t*);
+extern ROCKSDB_LIBRARY_API uint32_t
+rocksdb_subcompactionjobinfo_compaction_reason(
+    const rocksdb_subcompactionjobinfo_t* info);

 /* External file ingestion info */
 extern ROCKSDB_LIBRARY_API const char*
@@ -1918,6 +1927,10 @@ extern ROCKSDB_LIBRARY_API void
 rocksdb_options_add_compact_on_deletion_collector_factory_del_ratio(
    rocksdb_options_t*, size_t window_size, size_t num_dels_trigger,
    double deletion_ratio);
+extern ROCKSDB_LIBRARY_API void
+rocksdb_options_add_compact_on_deletion_collector_factory_min_file_size(
+    rocksdb_options_t*, size_t window_size, size_t num_dels_trigger,
+    double deletion_ratio, uint64_t min_file_size);
 extern ROCKSDB_LIBRARY_API void rocksdb_options_set_manual_wal_flush(
    rocksdb_options_t* opt, unsigned char);
 extern ROCKSDB_LIBRARY_API unsigned char rocksdb_options_get_manual_wal_flush(