Fix string-conversion issue in internal_repo_rocksdb/repo/utilities/persistent_cache/volatile_tier_impl.cc +5 (#14296)

Summary:
Pull Request resolved: https://github.com/facebook/rocksdb/pull/14296

This could is triggering `-Wstring-conversion`, which presents as:
```
warning: implicit conversion turns string literal into bool: A to B
```
This is often a bug and what was intended. The most frequent cause is the code was:
```
void foo(bool) { ... }
void foo(std::string) { ... }
foo("this gets interpreted as a bool");
```

It is also possible the issue is innocuous as part of an assert:
```
assert(!"this string is true, so the assertion is false");
EXPECT_FALSE("this string is true, so the expect fails");
```
in these cases the use is to "cute", so we modify the code to make it more obvious.
```
assert(false && "the compiler recognizes and doesn't complain about this pattern");
FAIL() << "much more obvious";
```

Reviewed By: anand1976

Differential Revision: D92013593

fbshipit-source-id: 0b4e00339bef3f76fc5b9ad35e2383c5e4f828f9

.github/actions/cache-getdeps-downloads/action.yml

@@ -0,0 +1,21 @@
+name: cache-getdeps-downloads
+description: Cache getdeps downloads to avoid unreliable mirrors and speed up builds
+outputs:
+  cache-hit:
+    description: Whether the cache was hit
+    value: ${{ steps.cache-downloads.outputs.cache-hit }}
+runs:
+  using: composite
+  steps:
+  - name: Cache getdeps downloads
+    id: cache-downloads
+    uses: actions/cache@v4
+    with:
+      # Use a fixed path that we control - folly.mk will sync with getdeps downloads dir
+      path: /tmp/rocksdb-getdeps-cache
+      # Use a rolling cache key - the cache accumulates downloads over time
+      # The key includes a weekly timestamp to ensure periodic refresh
+      key: getdeps-downloads-${{ runner.os }}-${{ runner.arch }}-week-${{ github.run_id }}
+      restore-keys: |
+        getdeps-downloads-${{ runner.os }}-${{ runner.arch }}-week-
+        getdeps-downloads-${{ runner.os }}-${{ runner.arch }}-

.github/workflows/nightly.yml

@@ -54,6 +54,7 @@ jobs:
     steps:
     - uses: actions/checkout@v4.1.0
     - uses: "./.github/actions/pre-steps"
+    - uses: "./.github/actions/cache-getdeps-downloads"
     - uses: "./.github/actions/setup-folly"
     - uses: "./.github/actions/build-folly"
     - run: LIB_MODE=static USE_CLANG=1 USE_FOLLY=1 COMPILE_WITH_UBSAN=1 COMPILE_WITH_ASAN=1 make -j32 check
@@ -68,6 +69,7 @@ jobs:
     steps:
     - uses: actions/checkout@v4.1.0
     - uses: "./.github/actions/pre-steps"
+    - uses: "./.github/actions/cache-getdeps-downloads"
     - uses: "./.github/actions/setup-folly"
     - uses: "./.github/actions/cache-folly"
       id: cache-folly
@@ -86,6 +88,7 @@ jobs:
     steps:
     - uses: actions/checkout@v4.1.0
     - uses: "./.github/actions/pre-steps"
+    - uses: "./.github/actions/cache-getdeps-downloads"
     - uses: "./.github/actions/setup-folly"
     - run: "DEBUG_LEVEL=0 make -j20 build_folly"
     - run: "USE_FOLLY=1 LIB_MODE=static DEBUG_LEVEL=0 V=1 make -j20 release"
@@ -110,6 +113,20 @@ 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-linux-arm-crashtest:
+    if: ${{ github.repository_owner == 'facebook' }}
+    runs-on:
+      labels: 4-core-ubuntu-arm
+    steps:
+    - uses: actions/checkout@v4.1.0
+    - uses: "./.github/actions/pre-steps"
+    - run: sudo apt-get update && sudo apt-get install -y build-essential libgflags-dev libsnappy-dev zlib1g-dev libbz2-dev liblz4-dev libzstd-dev
+    - run: sudo mount -o remount,size=16G /dev/shm
+    - run: sudo dd bs=1048576 count=4096 if=/dev/zero of=/swapfile && sudo chmod 600 /swapfile && sudo mkswap /swapfile && sudo swapon /swapfile
+    - run: ulimit -S -n `ulimit -H -n` && make V=1 -j8 CRASH_TEST_EXT_ARGS='--duration=1800 --max_key=2500000' blackbox_crash_test_with_atomic_flush
+    - run: rm -rf /dev/shm/rocksdb.*
+    - run: ulimit -S -n `ulimit -H -n` && make V=1 -j8 CRASH_TEST_EXT_ARGS='--duration=1800 --max_key=2500000' blackbox_crash_test_with_multiops_wc_txn
+    - uses: "./.github/actions/post-steps"
   build-examples:
     if: ${{ github.repository_owner == 'facebook' }}
     runs-on:
@@ -148,6 +165,7 @@ jobs:
     steps:
     - uses: actions/checkout@v4.1.0
     - uses: "./.github/actions/pre-steps"
+    - uses: "./.github/actions/cache-getdeps-downloads"
     - uses: "./.github/actions/setup-folly"
     - run: "(mkdir build && cd build && cmake -DUSE_FOLLY_LITE=1 -DWITH_GFLAGS=1 -DCMAKE_CXX_FLAGS=-DGLOG_USE_GLOG_EXPORT .. && make VERBOSE=1 -j20 && ctest -j20)"
     - uses: "./.github/actions/post-steps"

.github/workflows/pr-jobs.yml

@@ -102,6 +102,7 @@ jobs:
     steps:
     - uses: actions/checkout@v4.1.0
     - uses: "./.github/actions/pre-steps"
+    - uses: "./.github/actions/cache-getdeps-downloads"
     - uses: "./.github/actions/setup-folly"
     - uses: "./.github/actions/cache-folly"
       id: cache-folly
@@ -120,6 +121,7 @@ jobs:
     steps:
     - uses: actions/checkout@v4.1.0
     - uses: "./.github/actions/pre-steps"
+    - uses: "./.github/actions/cache-getdeps-downloads"
     - uses: "./.github/actions/setup-folly"
     - run: USE_FOLLY_LITE=1 EXTRA_CXXFLAGS=-DGLOG_USE_GLOG_EXPORT V=1 make -j32 all
     - uses: "./.github/actions/post-steps"
@@ -133,6 +135,7 @@ jobs:
     steps:
     - uses: actions/checkout@v4.1.0
     - uses: "./.github/actions/pre-steps"
+    - uses: "./.github/actions/cache-getdeps-downloads"
     - uses: "./.github/actions/setup-folly"
     - uses: "./.github/actions/cache-folly"
       id: cache-folly

BUCK

@@ -268,6 +268,7 @@ cpp_library_wrapper(name="rocksdb_lib", srcs=[
         "util/dynamic_bloom.cc",
         "util/file_checksum_helper.cc",
         "util/hash.cc",
+        "util/io_dispatcher_imp.cc",
         "util/murmurhash.cc",
         "util/random.cc",
         "util/rate_limiter.cc",
@@ -5214,6 +5215,12 @@ cpp_unittest_wrapper(name="interval_test",
             extra_compiler_flags=[])
 
 
+cpp_unittest_wrapper(name="io_dispatcher_test",
+            srcs=["util/io_dispatcher_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"],

CMakeLists.txt

@@ -219,6 +219,10 @@ if(MSVC)
     set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /Zi /nologo /EHsc /GS /Gd /GR /GF /fp:precise /Zc:wchar_t /Zc:forScope /errorReport:queue")
     set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /FC /d2Zi+ /W4 /wd4127 /wd4996 /wd4100 /wd4324")
   endif()
+  if(CMAKE_BUILD_TYPE STREQUAL "Release")
+    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /DNDEBUG")
+    message(STATUS "Setting /DNDEBUG as CMAKE_BUILD_TYPE is set to ${CMAKE_BUILD_TYPE}")
+  endif()
 else()
   set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -W -Wextra -Wall -pthread")
   set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wsign-compare -Wshadow -Wno-unused-parameter -Wno-unused-variable -Woverloaded-virtual -Wnon-virtual-dtor -Wno-missing-field-initializers -Wno-strict-aliasing -Wno-invalid-offsetof")

HISTORY.md

@@ -1,6 +1,56 @@
 # Rocksdb Change Log
 > NOTE: Entries for next release do not go here. Follow instructions in `unreleased_history/README.txt`
 
+## 10.11.2 (02/25/2026)
+### Bug Fixes
+* Fix longstanding failures that can arise from reading and/or compacting old DB dirs with range deletions (likely from version < 5.19.0) in many newer versions.
+
+## 10.11.1 (02/06/2026)
+### Bug Fixes
+* Fixed a bug in PosixFS AbortIO that could cause it to ignore an IO completion and a subsequent Poll to hang
+
+## 10.11.0 (01/23/2026)
+### Public API Changes
+* New SetOptions API that allows setting options for multiple CFs, avoiding the need to reserialize OPTIONS file for each CF
+* Remove remaining pieces of Lua integration
+
+### Behavior Changes
+* The new default for `BlockBasedTableOptions::format_version` is 7, which has been supported since RocksDB 10.4.0 and is required in order to use CompressionManagers supporting custom compression types.
+
+### Bug Fixes
+* Fixed a small performance bug with `format_version=7` when decompressing formats other than Snappy and ZSTD.
+* Fixed an infinite compaction loop bug with User-Defined Timestamps (UDT) where bottommost files were repeatedly marked for compaction even though their timestamp could not be collapsed.
+* Bugfix for persisted UDT record sequence number zeroing logic.
+
+## 10.10.0 (12/16/2025)
+### Bug Fixes
+* Fixed a bug in best-efforts recovery that causes use-after-free crashes when accessing SST files that were cached during the recovery.
+* Fix resumable compaction incorrectly allowing resumption from a truncated range deletion that is not well handled currently.
+* Fixed a bug in `PosixRandomFileAccess` IO uring submission queue ownership & management. Fix eliminates the false positive 'Bad cqe data' IO errors in `PosixRandomFileAccess::MultiRead` when interleaved with `PosixRandomFileAccess::ReadAsync` on the same thread.
+
+## 10.9.0 (11/21/2025)
+### New Features
+* Added an auto-tuning feature for DB manifest file size that also (by default) improves the safety of existing configurations in case `max_manifest_file_size` is repeatedly exceeded. The new recommendation is to set `max_manifest_file_size` to something small like 1MB and tune `max_manifest_space_amp_pct` as needed to balance write amp and space amp in the manifest. Refer to comments on those options in `DBOptions` for details. Both options are (now) mutable.
+* Added a new API to support option migration for multiple column families
+* Added new option target_file_size_is_upper_bound  that makes most compaction output SST files come close to the target file size without exceeding it, rather than commonly exceeding it by some fraction (current behavior). For now the new behavior is off by default, but we expect to enable it by default in the future.
+* Add a new option allow_trivial_move in CompactionOptions to allow CompactFiles to perform trivial move if possible. By default the flag of allow_trivial_move is false, so it preserve the original behavior.
+
+### Public API Changes
+* To reduce risk of ODR violations or similar, `ROCKSDB_USING_THREAD_STATUS` has been removed from public headers and replaced with static `const bool ThreadStatus::kEnabled`. Some other uses of conditional compilation have been removed from public API headers to reduce risk of ODR violations or other issues.
+
+### Behavior Changes
+* PosixWritableFile now repositions the seek pointer to the new end of file after a call to Truncate.
+* Updated standalone range deletion L0 file compaction behavior to avoid compacting with any newer L0 files (which is expensive and not useful).
+
+### Bug Fixes
+* Fix a bug where compaction with range deletion can persist kTypeMaxValid in MANIFEST as file metadata. kTypeMaxValid is not supposed to be persisted and can change as new value types are introduced. This can cause a forward compatibility issue where older versions of RocksDB don't recognize kTypeMaxValid from newer versions. A new placeholder value type kTypeTruncatedRangeDeletionSentinel is also introduced to replace kTypeMaxValid when reading existing SST files' metadata from MANIFEST. This allows us to strengthen some checks to avoid using kTypeMaxValid in the future.
+* Fixed a bug where `DB::GetSortedWalFiles()` could hang when waiting for a purge operation that found nothing to do (potentially triggered by iterator release, flush, compaction, etc.).
+* Fixed a bug in MultiScan where `max_sequential_skip_in_iterations` could cause the iterator to seek backward to already-unpinned blocks when the same user key spans multiple data blocks, leading to assertion failures or seg fault.
+* Fixed a bug for `WAL_ttl_seconds > 0` use cases where the newest archived WAL files could be incorrectly deleted when the system clock moved backwards.
+
+### Performance Improvements
+* Added optimization that allowed for the asynchronous prefetching of all data outlined in a multiscan iterator. This optimization was applied to the level iterator, which prefetches all data through each of the block-based iterators.
+
 ## 10.8.0 (10/21/2025)
 ### New Features
 * Add kFSPrefetch to FSSupportedOps enum to allow file systems to indicate prefetch support capability, avoiding unnecessary prefetch system calls on file systems that don't support them.

Makefile

@@ -296,6 +296,28 @@ $(info $(shell $(CC) --version))
 $(info $(shell $(CXX) --version))
 endif
 
+# ccache support
+# Set USE_CCACHE=1 to enable ccache, or let it auto-detect
+ifndef USE_CCACHE
+  CCACHE := $(shell which ccache 2>/dev/null)
+  ifneq ($(CCACHE),)
+    USE_CCACHE := 1
+  else
+    USE_CCACHE := 0
+  endif
+endif
+
+ifeq ($(USE_CCACHE), 1)
+  CCACHE := $(shell which ccache 2>/dev/null)
+  ifneq ($(CCACHE),)
+    $(info Using ccache: $(CCACHE))
+    CC := $(CCACHE) $(CC)
+    CXX := $(CCACHE) $(CXX)
+  else
+    $(warning ccache requested but not found in PATH)
+  endif
+endif
+
 missing_make_config_paths := $(shell				\
 	grep "\./\S*\|/\S*" -o $(CURDIR)/make_config.mk | 	\
 	while read path;					\
@@ -368,8 +390,6 @@ ifdef COMPILE_WITH_TSAN
         # Turn off -pg when enabling TSAN testing, because that induces
         # a link failure.  TODO: find the root cause
 	PROFILING_FLAGS =
-	# LUA is not supported under TSAN
-	LUA_PATH =
 	# Limit keys for crash test under TSAN to avoid error:
 	# "ThreadSanitizer: DenseSlabAllocator overflow. Dying."
 	CRASH_TEST_EXT_ARGS += --max_key=1000000
@@ -486,32 +506,6 @@ ifndef DISABLE_WARNING_AS_ERROR
 endif
 
 
-ifdef LUA_PATH
-
-ifndef LUA_INCLUDE
-LUA_INCLUDE=$(LUA_PATH)/include
-endif
-
-LUA_INCLUDE_FILE=$(LUA_INCLUDE)/lualib.h
-
-ifeq ("$(wildcard $(LUA_INCLUDE_FILE))", "")
-# LUA_INCLUDE_FILE does not exist
-$(error Cannot find lualib.h under $(LUA_INCLUDE).  Try to specify both LUA_PATH and LUA_INCLUDE manually)
-endif
-LUA_FLAGS = -I$(LUA_INCLUDE) -DLUA -DLUA_COMPAT_ALL
-CFLAGS += $(LUA_FLAGS)
-CXXFLAGS += $(LUA_FLAGS)
-
-ifndef LUA_LIB
-LUA_LIB = $(LUA_PATH)/lib/liblua.a
-endif
-ifeq ("$(wildcard $(LUA_LIB))", "") # LUA_LIB does not exist
-$(error $(LUA_LIB) does not exist.  Try to specify both LUA_PATH and LUA_LIB manually)
-endif
-EXEC_LDFLAGS += $(LUA_LIB)
-
-endif
-
 ifeq ($(NO_THREEWAY_CRC32C), 1)
 	CXXFLAGS += -DNO_THREEWAY_CRC32C
 endif
@@ -582,8 +576,8 @@ ifneq ($(filter check-headers, $(MAKECMDGOALS)),)
 # TODO: add/support JNI headers
 	DEV_HEADER_DIRS := $(sort include/ $(dir $(ALL_SOURCES)))
 # Some headers like in port/ are platform-specific
-	DEV_HEADERS_TO_CHECK := $(shell $(FIND) $(DEV_HEADER_DIRS) -type f -name '*.h' | grep -E -v 'port/|plugin/|lua/|range_tree/|secondary_index/')
-	PUBLIC_HEADERS_TO_CHECK := $(shell $(FIND) include/ -type f -name '*.h' | grep -E -v 'lua/')
+	DEV_HEADERS_TO_CHECK := $(shell $(FIND) $(DEV_HEADER_DIRS) -type f -name '*.h' | grep -E -v 'port/|plugin/|range_tree/|secondary_index/')
+	PUBLIC_HEADERS_TO_CHECK := $(shell $(FIND) include/ -type f -name '*.h')
 else
 	DEV_HEADERS_TO_CHECK :=
 	PUBLIC_HEADERS_TO_CHECK :=
@@ -1928,6 +1922,9 @@ blob_source_test: $(OBJ_DIR)/db/blob/blob_source_test.o $(TEST_LIBRARY) $(LIBRAR
 blob_garbage_meter_test: $(OBJ_DIR)/db/blob/blob_garbage_meter_test.o $(TEST_LIBRARY) $(LIBRARY)
 	$(AM_LINK)
 
+io_dispatcher_test: $(OBJ_DIR)/util/io_dispatcher_test.o $(TEST_LIBRARY) $(LIBRARY)
+	$(AM_LINK)
+
 timer_test: $(OBJ_DIR)/util/timer_test.o $(TEST_LIBRARY) $(LIBRARY)
 	$(AM_LINK)
 

build_tools/build_detect_platform

@@ -831,7 +831,6 @@ fi
 if test -n "$WITH_JEMALLOC_FLAG"; then
   echo "WITH_JEMALLOC_FLAG=$WITH_JEMALLOC_FLAG" >> "$OUTPUT"
 fi
-echo "LUA_PATH=$LUA_PATH" >> "$OUTPUT"
 if test -n "$USE_FOLLY"; then
   echo "USE_FOLLY=$USE_FOLLY" >> "$OUTPUT"
 fi

build_tools/check-public-header.sh

@@ -8,7 +8,7 @@ BAD=""
 
 # Look for potential for ODR violations caused by public headers depending on
 # build parameters that could vary between RocksDB build and application build.
-# * Cases like LUA, ROCKSDB_NAMESPACE, and ROCKSDB_ASSERT_STATUS_CHECKED are
+# * Cases like ROCKSDB_NAMESPACE, and ROCKSDB_ASSERT_STATUS_CHECKED are
 #   intentional, hard to avoid. (We expect definitions to change and the user
 #   should also.)
 # * Cases like _WIN32, OS_WIN, and __cplusplus are essentially ODR-safe.
@@ -19,7 +19,7 @@ BAD=""
 #   that should not cause ODR violations can be exempted with the ODR-SAFE
 #   marker recognized here.
 
-grep -nHE '^#if' -- "$@" | grep -vE 'ROCKSDB_NAMESPACE|ROCKSDB_ASSERT_STATUS_CHECKED|LUA|_WIN32|OS_WIN|ODR-SAFE|__cplusplus|ROCKSDB_DLL|ROCKSDB_LIBRARY_EXPORTS'
+grep -nHE '^#if' -- "$@" | grep -vE 'ROCKSDB_NAMESPACE|ROCKSDB_ASSERT_STATUS_CHECKED|_WIN32|OS_WIN|ODR-SAFE|__cplusplus|ROCKSDB_DLL|ROCKSDB_LIBRARY_EXPORTS'
 if [ "$?" != "1" ]; then
   echo "^^^^^ #if in public API could cause an ODR violation."
   echo "      Add // ODR-SAFE if verified safe."

build_tools/dependencies_platform010.sh

@@ -19,4 +19,3 @@ BENCHMARK_BASE=/mnt/gvfs/third-party2/benchmark/780c7a0f9cf0967961e69ad08e61cddd
 KERNEL_HEADERS_BASE=/mnt/gvfs/third-party2/kernel-headers/624a2f8f6c93c3c1df8aa4a6255d8202631a6c80/fb/platform010/da39a3e
 BINUTILS_BASE=/mnt/gvfs/third-party2/binutils/39579e8603b48b3540f8b0633f43adf29acccb8b/2.37/centos8-native/da39a3e
 VALGRIND_BASE=/mnt/gvfs/third-party2/valgrind/cd9cc656d49ecb53797ce4d055e49fde29fd57ff/3.19.0/platform010/76ebdda
-LUA_BASE=/mnt/gvfs/third-party2/lua/363787fa5cac2a8aa20638909210443278fa138e/5.3.4/platform010/9079c97

build_tools/fbcode_config.sh

@@ -164,12 +164,4 @@ EXEC_LDFLAGS_SHARED="$SNAPPY_LIBS $ZLIB_LIBS $BZIP_LIBS $LZ4_LIBS $ZSTD_LIBS $GF
 
 VALGRIND_VER="$VALGRIND_BASE/bin/"
 
-LUA_PATH="$LUA_BASE"
-
-if test -z $PIC_BUILD; then
-  LUA_LIB=" $LUA_PATH/lib/liblua.a"
-else
-  LUA_LIB=" $LUA_PATH/lib/liblua_pic.a"
-fi
-
-export CC CXX AR CFLAGS CXXFLAGS EXEC_LDFLAGS EXEC_LDFLAGS_SHARED VALGRIND_VER JEMALLOC_LIB JEMALLOC_INCLUDE CLANG_ANALYZER CLANG_SCAN_BUILD LUA_PATH LUA_LIB
+export CC CXX AR CFLAGS CXXFLAGS EXEC_LDFLAGS EXEC_LDFLAGS_SHARED VALGRIND_VER JEMALLOC_LIB JEMALLOC_INCLUDE CLANG_ANALYZER CLANG_SCAN_BUILD

build_tools/fbcode_config_platform010.sh

@@ -172,4 +172,4 @@ EXEC_LDFLAGS_SHARED="$SNAPPY_LIBS $ZLIB_LIBS $BZIP_LIBS $LZ4_LIBS $ZSTD_LIBS $GF
 
 VALGRIND_VER="$VALGRIND_BASE/bin/"
 
-export CC CXX AR AS CFLAGS CXXFLAGS EXEC_LDFLAGS EXEC_LDFLAGS_SHARED VALGRIND_VER JEMALLOC_LIB JEMALLOC_INCLUDE CLANG_ANALYZER CLANG_SCAN_BUILD LUA_PATH LUA_LIB
+export CC CXX AR AS CFLAGS CXXFLAGS EXEC_LDFLAGS EXEC_LDFLAGS_SHARED VALGRIND_VER JEMALLOC_LIB JEMALLOC_INCLUDE CLANG_ANALYZER CLANG_SCAN_BUILD

build_tools/getdeps_fallback_mirror.py

@@ -0,0 +1,123 @@
+#!/usr/bin/env python3
+"""
+Pre-download packages with unreliable mirrors using fallback mirrors.
+Reads package info from folly's getdeps manifest files.
+"""
+import sys
+import os
+import hashlib
+import subprocess
+import configparser
+
+def sha256_file(path):
+    """Calculate SHA256 hash of a file."""
+    h = hashlib.sha256()
+    try:
+        with open(path, 'rb') as f:
+            for chunk in iter(lambda: f.read(65536), b''):
+                h.update(chunk)
+        return h.hexdigest()
+    except Exception:
+        return None
+
+def parse_manifest(manifest_path):
+    """Parse a getdeps manifest file to extract download info."""
+    config = configparser.ConfigParser()
+    try:
+        config.read(manifest_path)
+        if 'download' in config:
+            return {
+                'url': config['download'].get('url', ''),
+                'sha256': config['download'].get('sha256', ''),
+            }
+    except Exception:
+        pass
+    return None
+
+def get_fallback_mirrors(url):
+    """Get fallback mirror URLs for a given URL."""
+    # Fallback mirror patterns for known unreliable hosts
+    mirror_fallbacks = {
+        "ftp.gnu.org/gnu/": [
+            "https://mirrors.kernel.org/gnu/",
+            "https://ftpmirror.gnu.org/gnu/",
+            "https://ftp.gnu.org/gnu/",
+        ],
+        "ftpmirror.gnu.org/gnu/": [
+            "https://mirrors.kernel.org/gnu/",
+            "https://ftpmirror.gnu.org/gnu/",
+            "https://ftp.gnu.org/gnu/",
+        ],
+    }
+
+    for pattern, mirrors in mirror_fallbacks.items():
+        if pattern in url:
+            # Extract the path after the pattern
+            path_start = url.find(pattern) + len(pattern)
+            path = url[path_start:]
+            return [mirror + path for mirror in mirrors]
+    return [url]  # No fallback, use original
+
+def main():
+    if len(sys.argv) != 4:
+        print(f"Usage: {sys.argv[0]} <download_dir> <cache_dir> <manifests_dir>")
+        sys.exit(1)
+
+    download_dir, cache_dir, manifests_dir = sys.argv[1], sys.argv[2], sys.argv[3]
+
+    # Packages known to have unreliable mirrors
+    packages_to_check = ["autoconf", "automake", "libtool"]
+
+    for package in packages_to_check:
+        manifest_path = os.path.join(manifests_dir, package)
+        if not os.path.exists(manifest_path):
+            continue
+
+        info = parse_manifest(manifest_path)
+        if not info or not info['url'] or not info['sha256']:
+            continue
+
+        # Determine filename from URL
+        url = info['url']
+        expected_sha256 = info['sha256']
+        url_filename = os.path.basename(url)
+
+        # getdeps uses format: {package}-{filename}
+        filename = f"{package}-{url_filename}"
+        filepath = os.path.join(download_dir, filename)
+        cache_path = os.path.join(cache_dir, filename)
+
+        # Check if already valid
+        if os.path.exists(filepath) and sha256_file(filepath) == expected_sha256:
+            print(f"  {filename}: OK (already downloaded)")
+            continue
+
+        # Check cache
+        if os.path.exists(cache_path) and sha256_file(cache_path) == expected_sha256:
+            print(f"  {filename}: OK (from cache)")
+            subprocess.run(['cp', cache_path, filepath], check=True)
+            continue
+
+        # Try fallback mirrors
+        mirrors = get_fallback_mirrors(url)
+        downloaded = False
+        for mirror_url in mirrors:
+            print(f"  {filename}: trying {mirror_url}...")
+            try:
+                subprocess.run(['wget', '-q', '-O', filepath, mirror_url], check=True, timeout=120)
+                if sha256_file(filepath) == expected_sha256:
+                    print(f"  {filename}: OK (downloaded)")
+                    subprocess.run(['cp', filepath, cache_path], check=False)
+                    downloaded = True
+                    break
+                else:
+                    os.remove(filepath)
+            except Exception:
+                if os.path.exists(filepath):
+                    os.remove(filepath)
+
+        if not downloaded:
+            print(f"  {filename}: WARNING - all mirrors failed")
+
+if __name__ == "__main__":
+    main()

build_tools/update_dependencies.sh

@@ -101,6 +101,5 @@ get_lib_base benchmark  LATEST  platform010
 get_lib_base kernel-headers fb platform010
 get_lib_base binutils   LATEST centos8-native
 get_lib_base valgrind   LATEST platform010
-get_lib_base lua        5.3.4  platform010
 
 git diff $OUTPUT

cache/cache_test.cc

@@ -644,7 +644,7 @@ using TypedHandle = SharedCache::TypedHandle;
 
 TEST_P(CacheTest, SetCapacity) {
   if (IsHyperClock()) {
-    // TODO: update test & code for limited supoort
+    // TODO: update test & code for limited support
     ROCKSDB_GTEST_BYPASS(
         "HyperClockCache doesn't support arbitrary capacity "
         "adjustments.");

cache/clock_cache.h

@@ -317,40 +317,89 @@ struct ClockHandle : public ClockHandleBasicData {
   // | acquire counter      | release counter     | hit bit | state marker |
   // -----------------------------------------------------------------------
 
-  // For reading or updating counters in meta word.
-  static constexpr uint8_t kCounterNumBits = 30;
-  static constexpr uint64_t kCounterMask = (uint64_t{1} << kCounterNumBits) - 1;
+  struct SlotMeta : public BitFields<uint64_t, SlotMeta> {
+    // For reading or updating counters in meta word.
+    static constexpr uint8_t kCounterNumBits = 30;
+    // Number of times the a reference has been acquired (or attempted)
+    // since last reset by eviction processing
+    using AcquireCounter =
+        UnsignedBitField<SlotMeta, kCounterNumBits, NoPrevBitField>;
+    // Number of times the a reference has been released (or attempted)
+    // since last reset by eviction processing
+    using ReleaseCounter =
+        UnsignedBitField<SlotMeta, kCounterNumBits, AcquireCounter>;
+    // Metadata bit in support of secondary cache
+    using HitFlag = BoolBitField<SlotMeta, ReleaseCounter>;
+    // Occupied means any state other than empty
+    using OccupiedFlag = BoolBitField<SlotMeta, HitFlag>;
+    // Shareable means the entry is reference counted (visible or invisible)
+    // (only set if also occupied)
+    using ShareableFlag = BoolBitField<SlotMeta, OccupiedFlag>;
+    // Visible is only set if also shareable (invisible can't be found by
+    // Lookup)
+    using VisibleFlag = BoolBitField<SlotMeta, ShareableFlag>;
 
-  static constexpr uint8_t kAcquireCounterShift = 0;
-  static constexpr uint64_t kAcquireIncrement = uint64_t{1}
-                                                << kAcquireCounterShift;
-  static constexpr uint8_t kReleaseCounterShift = kCounterNumBits;
-  static constexpr uint64_t kReleaseIncrement = uint64_t{1}
-                                                << kReleaseCounterShift;
+    // Convenience functions
+    uint32_t GetAcquireCounter() const { return Get<AcquireCounter>(); }
+    void SetAcquireCounter(uint32_t val) { Set<AcquireCounter>(val); }
+    uint32_t GetReleaseCounter() const { return Get<ReleaseCounter>(); }
+    void SetReleaseCounter(uint32_t val) { Set<ReleaseCounter>(val); }
+    uint32_t GetRefcount() const {
+      return Get<AcquireCounter>() - Get<ReleaseCounter>();
+    }
+    bool GetHit() const { return Get<HitFlag>(); }
+    void SetHit(bool val) { Set<HitFlag>(val); }
 
-  // For setting the hit bit
-  static constexpr uint8_t kHitBitShift = 2U * kCounterNumBits;
-  static constexpr uint64_t kHitBitMask = uint64_t{1} << kHitBitShift;
+    // Some distinct states for the various state flags
+    bool IsEmpty() const {
+      bool rv = !Get<OccupiedFlag>();
+      if (rv) {
+        assert(!Get<ShareableFlag>());
+        assert(!Get<VisibleFlag>());
+      }
+      return rv;
+    }
 
-  // For reading or updating the state marker in meta word
-  static constexpr uint8_t kStateShift = kHitBitShift + 1;
+    bool IsUnderConstruction() const {
+      bool rv = Get<OccupiedFlag>() && !Get<ShareableFlag>();
+      if (rv) {
+        assert(!Get<VisibleFlag>());
+      }
+      return rv;
+    }
+    void SetUnderConstruction() {
+      Set<OccupiedFlag>(true);
+      Set<ShareableFlag>(false);
+      Set<VisibleFlag>(false);
+    }
 
-  // Bits contribution to state marker.
-  // Occupied means any state other than empty
-  static constexpr uint8_t kStateOccupiedBit = 0b100;
-  // Shareable means the entry is reference counted (visible or invisible)
-  // (only set if also occupied)
-  static constexpr uint8_t kStateShareableBit = 0b010;
-  // Visible is only set if also shareable
-  static constexpr uint8_t kStateVisibleBit = 0b001;
+    bool IsShareable() const { return Get<ShareableFlag>(); }
+    bool IsInvisible() const {
+      bool rv = Get<ShareableFlag>() && !Get<VisibleFlag>();
+      if (rv) {
+        assert(Get<OccupiedFlag>());
+      }
+      return rv;
+    }
+    void SetInvisible() {
+      Set<OccupiedFlag>(true);
+      Set<ShareableFlag>(true);
+      Set<VisibleFlag>(false);
+    }
 
-  // Complete state markers (not shifted into full word)
-  static constexpr uint8_t kStateEmpty = 0b000;
-  static constexpr uint8_t kStateConstruction = kStateOccupiedBit;
-  static constexpr uint8_t kStateInvisible =
-      kStateOccupiedBit | kStateShareableBit;
-  static constexpr uint8_t kStateVisible =
-      kStateOccupiedBit | kStateShareableBit | kStateVisibleBit;
+    bool IsVisible() const {
+      bool rv = Get<ShareableFlag>() && Get<VisibleFlag>();
+      if (rv) {
+        assert(Get<OccupiedFlag>());
+      }
+      return rv;
+    }
+    void SetVisible() {
+      Set<OccupiedFlag>(true);
+      Set<ShareableFlag>(true);
+      Set<VisibleFlag>(true);
+    }
+  };
 
   // Constants for initializing the countdown clock. (Countdown clock is only
   // in effect with zero refs, acquire counter == release counter, and in that
@@ -364,7 +413,7 @@ struct ClockHandle : public ClockHandleBasicData {
   // TODO: make these coundown values tuning parameters for eviction?
 
   // See above. Mutable for read reference counting.
-  mutable AcqRelAtomic<uint64_t> meta{};
+  mutable BitFieldsAtomic<SlotMeta> meta{};
 };  // struct ClockHandle
 
 class BaseClockTable {
@@ -431,9 +480,9 @@ class BaseClockTable {
   bool IsEvictionEffortExceeded(const BaseClockTable::EvictionData& data) const;
 #ifndef NDEBUG
   // Acquire N references
-  void TEST_RefN(ClockHandle& handle, size_t n);
+  void TEST_RefN(ClockHandle& handle, uint32_t n);
   // Helper for TEST_ReleaseN
-  void TEST_ReleaseNMinus1(ClockHandle* handle, size_t n);
+  void TEST_ReleaseNMinus1(ClockHandle* handle, uint32_t n);
 #endif
 
  private:  // fns
@@ -489,13 +538,13 @@ class BaseClockTable {
   // TODO: is this separation needed if we don't do background evictions?
   ALIGN_AS(CACHE_LINE_SIZE)
   // Number of elements in the table.
-  AcqRelAtomic<size_t> occupancy_{};
+  Atomic<size_t> occupancy_{};
 
   // Memory usage by entries tracked by the cache (including standalone)
-  AcqRelAtomic<size_t> usage_{};
+  Atomic<size_t> usage_{};
 
   // Part of usage by standalone entries (not in table)
-  AcqRelAtomic<size_t> standalone_usage_{};
+  Atomic<size_t> standalone_usage_{};
 
   // Maximum total charge of all elements stored in the table.
   // (Relaxed: eventual consistency/update is OK)
@@ -586,7 +635,7 @@ class FixedHyperClockTable : public BaseClockTable {
   bool GrowIfNeeded(size_t new_occupancy, InsertState& state);
 
   HandleImpl* DoInsert(const ClockHandleBasicData& proto,
-                       uint64_t initial_countdown, bool take_ref,
+                       uint32_t initial_countdown, bool take_ref,
                        InsertState& state);
 
   // Runs the clock eviction algorithm trying to reclaim at least
@@ -614,7 +663,7 @@ class FixedHyperClockTable : public BaseClockTable {
   }
 
   // Release N references
-  void TEST_ReleaseN(HandleImpl* handle, size_t n);
+  void TEST_ReleaseN(HandleImpl* handle, uint32_t n);
 #endif
 
   // The load factor p is a real number in (0, 1) such that at all
@@ -775,6 +824,7 @@ class AutoHyperClockTable : public BaseClockTable {
     // chain--specifically the next entry in the chain.
     // * The end of a chain is given a special "end" marker and refers back
     // to the head of the chain.
+    // These decorated pointers use the NextWithShift bit field struct below.
     //
     // Why do we need shift on each pointer? To make Lookup wait-free, we need
     // to be able to query a chain without missing anything, and preferably
@@ -794,47 +844,63 @@ class AutoHyperClockTable : public BaseClockTable {
     // it is normal to see "under construction" entries on the chain, and it
     // is not safe to read their hashed key without either a read reference
     // on the entry or a rewrite lock on the chain.
+    struct NextWithShift : public BitFields<uint64_t, NextWithShift> {
+      // The "shift" associated with this decorated pointer (see description
+      // above).
+      using Shift = UnsignedBitField<NextWithShift, 6, NoPrevBitField>;
+      // Marker for the end of a chain. Must also (a) point back to the head of
+      // the chain (with end marker removed), and (b) set the LockedFlag
+      // (below), so that attempting to lock an empty chain has no effect (not
+      // needed, as the lock is only needed for removals).
+      using EndFlag = BoolBitField<NextWithShift, Shift>;
+      // Marker that some thread owning writes to the chain structure (except
+      // for inserts), but only if not an "end" pointer. Also called the
+      // "rewrite lock."
+      using LockedFlag = BoolBitField<NextWithShift, EndFlag>;
+      // The "next" associated with this decorated pointer, which is an index
+      // into the table's array_ (see description above).
+      using Next = UnsignedBitField<NextWithShift, 56, LockedFlag>;
 
-    // Marker in a "with_shift" head pointer for some thread owning writes
-    // to the chain structure (except for inserts), but only if not an
-    // "end" pointer. Also called the "rewrite lock."
-    static constexpr uint64_t kHeadLocked = uint64_t{1} << 7;
+      bool IsLocked() const { return Get<LockedFlag>(); }
+      bool IsEnd() const {
+        // End flag should imply locked flag
+        assert(!Get<EndFlag>() || Get<LockedFlag>());
+        return Get<EndFlag>();
+      }
+      bool IsLockedNotEnd() const {
+        // NOTE: helping GCC to optimize this simpler code:
+        // return IsLocked() && !IsEnd();
+        constexpr U kEndFlag = U{1} << EndFlag::kBitOffset;
+        constexpr U kLockedFlag = U{1} << LockedFlag::kBitOffset;
+        return (underlying & (kEndFlag | kLockedFlag)) == kLockedFlag;
+      }
+      auto GetNext() const { return Get<Next>(); }
+      auto GetShift() const { return Get<Shift>(); }
 
-    // Marker in a "with_shift" pointer for the end of a chain. Must also
-    // point back to the head of the chain (with end marker removed).
-    // Also includes the "locked" bit so that attempting to lock an empty
-    // chain has no effect (not needed, as the lock is only needed for
-    // removals).
-    static constexpr uint64_t kNextEndFlags = (uint64_t{1} << 6) | kHeadLocked;
+      static NextWithShift Make(size_t next, int shift) {
+        return NextWithShift{}.With<Next>(next).With<Shift>(
+            static_cast<uint8_t>(shift));
+      }
 
-    static inline bool IsEnd(uint64_t next_with_shift) {
-      // Assuming certain values never used, suffices to check this one bit
-      constexpr auto kCheckBit = kNextEndFlags ^ kHeadLocked;
-      return next_with_shift & kCheckBit;
-    }
-
-    // Bottom bits to right shift away to get an array index from a
-    // "with_shift" pointer.
-    static constexpr int kNextShift = 8;
-
-    // A bit mask for the "shift" associated with each "with_shift" pointer.
-    // Always bottommost bits.
-    static constexpr int kShiftMask = 63;
+      static NextWithShift MakeEnd(size_t next, int shift) {
+        return Make(next, shift).With<EndFlag>(true).With<LockedFlag>(true);
+      }
+    };
 
     // A marker for head_next_with_shift that indicates this HandleImpl is
     // heap allocated (standalone) rather than in the table.
-    static constexpr uint64_t kStandaloneMarker = UINT64_MAX;
+    static constexpr NextWithShift kStandaloneMarker{UINT64_MAX};
 
     // A marker for head_next_with_shift indicating the head is not yet part
     // of the usable table, or for chain_next_with_shift indicating that the
     // entry is not present or is not yet part of a chain (must not be
     // "shareable" state).
-    static constexpr uint64_t kUnusedMarker = 0;
+    static constexpr NextWithShift kUnusedMarker{0};
 
     // See above. The head pointer is logically independent of the rest of
     // the entry, including the chain next pointer.
-    AcqRelAtomic<uint64_t> head_next_with_shift{kUnusedMarker};
-    AcqRelAtomic<uint64_t> chain_next_with_shift{kUnusedMarker};
+    BitFieldsAtomic<NextWithShift> head_next_with_shift{kUnusedMarker};
+    BitFieldsAtomic<NextWithShift> chain_next_with_shift{kUnusedMarker};
 
     // For supporting CreateStandalone and some fallback cases.
     inline bool IsStandalone() const {
@@ -880,7 +946,7 @@ class AutoHyperClockTable : public BaseClockTable {
   bool GrowIfNeeded(size_t new_occupancy, InsertState& state);
 
   HandleImpl* DoInsert(const ClockHandleBasicData& proto,
-                       uint64_t initial_countdown, bool take_ref,
+                       uint32_t initial_countdown, bool take_ref,
                        InsertState& state);
 
   // Runs the clock eviction algorithm trying to reclaim at least
@@ -908,7 +974,7 @@ class AutoHyperClockTable : public BaseClockTable {
   }
 
   // Release N references
-  void TEST_ReleaseN(HandleImpl* handle, size_t n);
+  void TEST_ReleaseN(HandleImpl* handle, uint32_t n);
 #endif
 
   // Maximum ratio of number of occupied slots to number of usable slots. The
@@ -990,7 +1056,7 @@ class AutoHyperClockTable : public BaseClockTable {
   // To maximize parallelization of Grow() operations, this field is only
   // updated opportunistically after Grow() operations and in DoInsert() where
   // it is found to be out-of-date. See CatchUpLengthInfoNoWait().
-  AcqRelAtomic<uint64_t> length_info_;
+  Atomic<uint64_t> length_info_;
 
   // An already-computed version of the usable length times the max load
   // factor. Could be slightly out of date but GrowIfNeeded()/Grow() handle
@@ -1113,8 +1179,8 @@ class ALIGN_AS(CACHE_LINE_SIZE) ClockCacheShard final : public CacheShardBase {
     return table_.TEST_MutableOccupancyLimit();
   }
   // Acquire/release N references
-  void TEST_RefN(HandleImpl* handle, size_t n);
-  void TEST_ReleaseN(HandleImpl* handle, size_t n);
+  void TEST_RefN(HandleImpl* handle, uint32_t n);
+  void TEST_ReleaseN(HandleImpl* handle, uint32_t n);
 #endif
 
  private:  // data

cache/lru_cache_test.cc

@@ -1503,7 +1503,7 @@ TEST_P(BasicSecondaryCacheTest, FullCapacityTest) {
         /*context*/ this, Cache::Priority::LOW);
     ASSERT_EQ(handle1, nullptr);
 
-    // k1 promotion can fail with strict_capacit_limit=true, but Lookup still
+    // k1 promotion can fail with strict_capacity_limit=true, but Lookup still
     // succeeds using a standalone handle
     handle1 = cache->Lookup(k1.AsSlice(), GetHelper(),
                             /*context*/ this, Cache::Priority::LOW);
@@ -1680,7 +1680,7 @@ TEST_P(DBSecondaryCacheTest, TestSecondaryCacheCorrectness2) {
   // After Flush is successful, RocksDB will do the paranoid check for the new
   // SST file. Meta blocks are always cached in the block cache and they
   // will not be evicted. When block_2 is cache miss and read out, it is
-  // inserted to the block cache. Thefore, block_1 is evicted from block
+  // inserted to the block cache. Therefore, block_1 is evicted from block
   // cache and successfully inserted to the secondary cache. Here are 2
   // lookups in the secondary cache for block_1 and block_2.
   ASSERT_EQ(secondary_cache->num_inserts(), 1u);
@@ -1721,7 +1721,7 @@ TEST_P(DBSecondaryCacheTest, TestSecondaryCacheCorrectness2) {
   v = Get(Key(0));
   ASSERT_EQ(1007, v.size());
   // This Get needs to access block_1, since block_1 is not in block cache
-  // there is one econdary cache lookup. Then, block_1 is cached in the
+  // there is one secondary cache lookup. Then, block_1 is cached in the
   // block cache.
   ASSERT_EQ(secondary_cache->num_inserts(), 2u);
   ASSERT_EQ(secondary_cache->num_lookups(), 5u);
@@ -1785,7 +1785,7 @@ TEST_P(DBSecondaryCacheTest, NoSecondaryCacheInsertion) {
   std::string v = Get(Key(0));
   ASSERT_EQ(1000, v.size());
   // Since the block cache is large enough, all the blocks are cached. we
-  // do not need to lookup the seondary cache.
+  // do not need to lookup the secondary cache.
   ASSERT_EQ(secondary_cache->num_inserts(), 0u);
   ASSERT_EQ(secondary_cache->num_lookups(), 2u);
 
@@ -2150,7 +2150,7 @@ TEST_P(DBSecondaryCacheTest, LRUCacheDumpLoadBasic) {
   ASSERT_OK(Flush());
   Compact("a", "z");
 
-  // do th eread for all the key value pairs, so all the blocks should be in
+  // do the read for all the key value pairs, so all the blocks should be in
   // cache
   uint32_t start_insert = cache->GetInsertCount();
   uint32_t start_lookup = cache->GetLookupcount();
@@ -2464,7 +2464,7 @@ TEST_P(DBSecondaryCacheTest, TestSecondaryCacheOptionBasic) {
   std::string v = Get(Key(0));
   ASSERT_EQ(1007, v.size());
 
-  // Check the data in first block. Cache miss, direclty read from SST file.
+  // Check the data in first block. Cache miss, directly read from SST file.
   ASSERT_EQ(secondary_cache->num_inserts(), 0u);
   ASSERT_EQ(secondary_cache->num_lookups(), 0u);
 
@@ -2598,7 +2598,7 @@ TEST_P(DBSecondaryCacheTest, TestSecondaryCacheOptionChange) {
 }
 
 // Two DB test. We create 2 DBs sharing the same block cache and secondary
-// cache. We diable the secondary cache option for DB2.
+// cache. We disable the secondary cache option for DB2.
 TEST_P(DBSecondaryCacheTest, TestSecondaryCacheOptionTwoDB) {
   if (IsHyperClock()) {
     ROCKSDB_GTEST_BYPASS("Test depends on LRUCache-specific behaviors");

crash_test.mk

@@ -8,7 +8,7 @@ DB_STRESS_CMD?=./db_stress
 include common.mk
 
 CRASHTEST_MAKE=$(MAKE) -f crash_test.mk
-CRASHTEST_PY=$(PYTHON) -u tools/db_crashtest.py --stress_cmd=$(DB_STRESS_CMD) --cleanup_cmd='$(DB_CLEANUP_CMD)'
+CRASHTEST_PY=$(PYTHON) -u tools/db_crashtest.py --stress_cmd=$(DB_STRESS_CMD) --cleanup_cmd='$(DB_CLEANUP_CMD)' --destroy_db_initially=1
 
 .PHONY: crash_test crash_test_with_atomic_flush crash_test_with_txn \
 	crash_test_with_wc_txn crash_test_with_wp_txn crash_test_with_wup_txn \
@@ -34,6 +34,7 @@ CRASHTEST_PY=$(PYTHON) -u tools/db_crashtest.py --stress_cmd=$(DB_STRESS_CMD) --
 	whitebox_crash_test_with_txn whitebox_crash_test_with_ts \
 	whitebox_crash_test_with_optimistic_txn \
 	whitebox_crash_test_with_tiered_storage \
+	crash_test_db_cleanup \
 
 crash_test: $(DB_STRESS_CMD)
 # Do not parallelize
@@ -161,6 +162,9 @@ whitebox_crash_test_with_optimistic_txn: $(DB_STRESS_CMD)
 	$(CRASHTEST_PY) --optimistic_txn whitebox --random_kill_odd \
       $(CRASH_TEST_KILL_ODD) $(CRASH_TEST_EXT_ARGS)
 
+crash_test_db_cleanup: $(DB_STRESS_CMD)
+	$(DB_STRESS_CMD) --delete_dir_and_exit=$(TEST_TMPDIR)
+
 # Old names DEPRECATED
 crash_test_with_txn: crash_test_with_wc_txn
 whitebox_crash_test_with_txn: whitebox_crash_test_with_wc_txn

db/blob/blob_file_builder.cc

@@ -67,6 +67,16 @@ BlobFileBuilder::BlobFileBuilder(
       min_blob_size_(mutable_cf_options->min_blob_size),
       blob_file_size_(mutable_cf_options->blob_file_size),
       blob_compression_type_(mutable_cf_options->blob_compression_type),
+      // TODO: support most CompressionOptions with a new CF option
+      // blob_compression_opts
+      // TODO with schema change: support custom compression manager and options
+      // such as max_compressed_bytes_per_kb
+      // NOTE: returns nullptr for kNoCompression
+      blob_compressor_(GetBuiltinV2CompressionManager()->GetCompressor(
+          CompressionOptions{}, blob_compression_type_)),
+      blob_compressor_wa_(blob_compressor_
+                              ? blob_compressor_->ObtainWorkingArea()
+                              : Compressor::ManagedWorkingArea{}),
       prepopulate_blob_cache_(mutable_cf_options->prepopulate_blob_cache),
       file_options_(file_options),
       write_options_(write_options),
@@ -113,7 +123,7 @@ Status BlobFileBuilder::Add(const Slice& key, const Slice& value,
   }
 
   Slice blob = value;
-  std::string compressed_blob;
+  GrowableBuffer compressed_blob;
 
   {
     const Status s = CompressBlobIfNeeded(&blob, &compressed_blob);
@@ -254,36 +264,27 @@ Status BlobFileBuilder::OpenBlobFileIfNeeded() {
 }
 
 Status BlobFileBuilder::CompressBlobIfNeeded(
-    Slice* blob, std::string* compressed_blob) const {
+    Slice* blob, GrowableBuffer* compressed_blob) const {
   assert(blob);
   assert(compressed_blob);
   assert(compressed_blob->empty());
   assert(immutable_options_);
 
-  if (blob_compression_type_ == kNoCompression) {
+  if (!blob_compressor_) {
+    assert(blob_compression_type_ == kNoCompression);
     return Status::OK();
   }
+  assert(blob_compression_type_ != kNoCompression);
 
-  // TODO: allow user CompressionOptions, including max_compressed_bytes_per_kb
-  CompressionOptions opts;
-  CompressionContext context(blob_compression_type_, opts);
+  // WART: always stored as compressed even when that increases the size.
 
-  CompressionInfo info(opts, context, CompressionDict::GetEmptyDict(),
-                       blob_compression_type_);
-
-  constexpr uint32_t compression_format_version = 2;
-
-  bool success = false;
-
-  {
-    StopWatch stop_watch(immutable_options_->clock, immutable_options_->stats,
-                         BLOB_DB_COMPRESSION_MICROS);
-    success = OLD_CompressData(*blob, info, compression_format_version,
-                               compressed_blob);
-  }
-
-  if (!success) {
-    return Status::Corruption("Error compressing blob");
+  Status s;
+  StopWatch stop_watch(immutable_options_->clock, immutable_options_->stats,
+                       BLOB_DB_COMPRESSION_MICROS);
+  s = LegacyForceBuiltinCompression(*blob_compressor_, &blob_compressor_wa_,
+                                    *blob, compressed_blob);
+  if (!s.ok()) {
+    return s;
   }
 
   *blob = Slice(*compressed_blob);

db/blob/blob_file_builder.h

@@ -10,12 +10,14 @@
 #include <string>
 #include <vector>
 
+#include "rocksdb/advanced_compression.h"
 #include "rocksdb/advanced_options.h"
 #include "rocksdb/compression_type.h"
 #include "rocksdb/env.h"
 #include "rocksdb/options.h"
 #include "rocksdb/rocksdb_namespace.h"
 #include "rocksdb/types.h"
+#include "util/aligned_buffer.h"
 
 namespace ROCKSDB_NAMESPACE {
 
@@ -76,7 +78,8 @@ class BlobFileBuilder {
  private:
   bool IsBlobFileOpen() const;
   Status OpenBlobFileIfNeeded();
-  Status CompressBlobIfNeeded(Slice* blob, std::string* compressed_blob) const;
+  Status CompressBlobIfNeeded(Slice* blob,
+                              GrowableBuffer* compressed_blob) const;
   Status WriteBlobToFile(const Slice& key, const Slice& blob,
                          uint64_t* blob_file_number, uint64_t* blob_offset);
   Status CloseBlobFile();
@@ -91,6 +94,8 @@ class BlobFileBuilder {
   uint64_t min_blob_size_;
   uint64_t blob_file_size_;
   CompressionType blob_compression_type_;
+  std::unique_ptr<Compressor> blob_compressor_;
+  mutable Compressor::ManagedWorkingArea blob_compressor_wa_;
   PrepopulateBlobCache prepopulate_blob_cache_;
   const FileOptions* file_options_;
   const WriteOptions* write_options_;

db/blob/blob_file_builder_test.cc

@@ -457,11 +457,12 @@ TEST_F(BlobFileBuilderTest, CompressionError) {
       nullptr /*IOTracer*/, nullptr /*BlobFileCompletionCallback*/,
       BlobFileCreationReason::kFlush, &blob_file_paths, &blob_file_additions);
 
-  SyncPoint::GetInstance()->SetCallBack("CompressData:TamperWithReturnValue",
-                                        [](void* arg) {
-                                          bool* ret = static_cast<bool*>(arg);
-                                          *ret = false;
-                                        });
+  SyncPoint::GetInstance()->SetCallBack(
+      "LegacyForceBuiltinCompression:TamperWithStatus", [](void* arg) {
+        Status* ret = static_cast<Status*>(arg);
+        ASSERT_OK(*ret);
+        *ret = Status::Corruption("Tampered result");
+      });
   SyncPoint::GetInstance()->EnableProcessing();
 
   constexpr char key[] = "1";
@@ -469,7 +470,7 @@ TEST_F(BlobFileBuilderTest, CompressionError) {
 
   std::string blob_index;
 
-  ASSERT_TRUE(builder.Add(key, value, &blob_index).IsCorruption());
+  ASSERT_EQ(builder.Add(key, value, &blob_index).code(), Status::kCorruption);
 
   SyncPoint::GetInstance()->DisableProcessing();
   SyncPoint::GetInstance()->ClearAllCallBacks();

db/blob/blob_file_reader.cc

@@ -17,10 +17,10 @@
 #include "rocksdb/file_system.h"
 #include "rocksdb/slice.h"
 #include "rocksdb/status.h"
+#include "table/format.h"
 #include "table/multiget_context.h"
 #include "test_util/sync_point.h"
 #include "util/compression.h"
-#include "util/crc32c.h"
 #include "util/stop_watch.h"
 
 namespace ROCKSDB_NAMESPACE {
@@ -69,9 +69,16 @@ Status BlobFileReader::Create(
     }
   }
 
-  blob_file_reader->reset(
-      new BlobFileReader(std::move(file_reader), file_size, compression_type,
-                         immutable_options.clock, statistics));
+  std::shared_ptr<Decompressor> decompressor;
+  if (compression_type != kNoCompression) {
+    // The blob format has always used compression format 2
+    decompressor = GetBuiltinV2CompressionManager()->GetDecompressorOptimizeFor(
+        compression_type);
+  }
+
+  blob_file_reader->reset(new BlobFileReader(
+      std::move(file_reader), file_size, compression_type,
+      std::move(decompressor), immutable_options.clock, statistics));
 
   return Status::OK();
 }
@@ -282,11 +289,13 @@ Status BlobFileReader::ReadFromFile(const RandomAccessFileReader* file_reader,
 
 BlobFileReader::BlobFileReader(
     std::unique_ptr<RandomAccessFileReader>&& file_reader, uint64_t file_size,
-    CompressionType compression_type, SystemClock* clock,
+    CompressionType compression_type,
+    std::shared_ptr<Decompressor> decompressor, SystemClock* clock,
     Statistics* statistics)
     : file_reader_(std::move(file_reader)),
       file_size_(file_size),
       compression_type_(compression_type),
+      decompressor_(std::move(decompressor)),
       clock_(clock),
       statistics_(statistics) {
   assert(file_reader_);
@@ -375,8 +384,9 @@ Status BlobFileReader::GetBlob(
   const Slice value_slice(record_slice.data() + adjustment, value_size);
 
   {
-    const Status s = UncompressBlobIfNeeded(
-        value_slice, compression_type, allocator, clock_, statistics_, result);
+    const Status s = UncompressBlobIfNeeded(value_slice, compression_type,
+                                            decompressor_.get(), allocator,
+                                            clock_, statistics_, result);
     if (!s.ok()) {
       return s;
     }
@@ -524,9 +534,9 @@ void BlobFileReader::MultiGetBlob(
 
     // Uncompress blob if needed
     Slice value_slice(record_slice.data() + adjustments[i], req->len);
-    *req->status =
-        UncompressBlobIfNeeded(value_slice, compression_type_, allocator,
-                               clock_, statistics_, &blob_reqs[i].second);
+    *req->status = UncompressBlobIfNeeded(
+        value_slice, compression_type_, decompressor_.get(), allocator, clock_,
+        statistics_, &blob_reqs[i].second);
     if (req->status->ok()) {
       total_bytes += record_slice.size();
     }
@@ -583,8 +593,8 @@ Status BlobFileReader::VerifyBlob(const Slice& record_slice,
 
 Status BlobFileReader::UncompressBlobIfNeeded(
     const Slice& value_slice, CompressionType compression_type,
-    MemoryAllocator* allocator, SystemClock* clock, Statistics* statistics,
-    std::unique_ptr<BlobContents>* result) {
+    Decompressor* decompressor, MemoryAllocator* allocator, SystemClock* clock,
+    Statistics* statistics, std::unique_ptr<BlobContents>* result) {
   assert(result);
 
   if (compression_type == kNoCompression) {
@@ -593,31 +603,33 @@ Status BlobFileReader::UncompressBlobIfNeeded(
     return Status::OK();
   }
 
-  UncompressionContext context(compression_type);
-  UncompressionInfo info(context, UncompressionDict::GetEmptyDict(),
-                         compression_type);
+  assert(decompressor);
 
-  size_t uncompressed_size = 0;
-  constexpr uint32_t compression_format_version = 2;
+  Decompressor::Args args;
+  args.compression_type = compression_type;
+  args.compressed_data = value_slice;
 
-  CacheAllocationPtr output;
+  Status s = decompressor->ExtractUncompressedSize(args);
+  if (!s.ok()) {
+    return Status::Corruption(s.ToString());
+  }
+
+  CacheAllocationPtr output = AllocateBlock(args.uncompressed_size, allocator);
 
   {
     PERF_TIMER_GUARD(blob_decompress_time);
     StopWatch stop_watch(clock, statistics, BLOB_DB_DECOMPRESSION_MICROS);
-    output = OLD_UncompressData(info, value_slice.data(), value_slice.size(),
-                                &uncompressed_size, compression_format_version,
-                                allocator);
+    s = decompressor->DecompressBlock(args, output.get());
   }
 
   TEST_SYNC_POINT_CALLBACK(
-      "BlobFileReader::UncompressBlobIfNeeded:TamperWithResult", &output);
+      "BlobFileReader::UncompressBlobIfNeeded:TamperWithResult", &s);
 
-  if (!output) {
-    return Status::Corruption("Unable to uncompress blob");
+  if (!s.ok()) {
+    return Status::Corruption(s.ToString());
   }
 
-  result->reset(new BlobContents(std::move(output), uncompressed_size));
+  result->reset(new BlobContents(std::move(output), args.uncompressed_size));
 
   return Status::OK();
 }

db/blob/blob_file_reader.h

@@ -10,6 +10,7 @@
 
 #include "db/blob/blob_read_request.h"
 #include "file/random_access_file_reader.h"
+#include "rocksdb/advanced_compression.h"
 #include "rocksdb/compression_type.h"
 #include "rocksdb/rocksdb_namespace.h"
 #include "util/autovector.h"
@@ -64,7 +65,8 @@ class BlobFileReader {
  private:
   BlobFileReader(std::unique_ptr<RandomAccessFileReader>&& file_reader,
                  uint64_t file_size, CompressionType compression_type,
-                 SystemClock* clock, Statistics* statistics);
+                 std::shared_ptr<Decompressor> decompressor, SystemClock* clock,
+                 Statistics* statistics);
 
   static Status OpenFile(const ImmutableOptions& immutable_options,
                          const FileOptions& file_opts,
@@ -96,6 +98,7 @@ class BlobFileReader {
 
   static Status UncompressBlobIfNeeded(const Slice& value_slice,
                                        CompressionType compression_type,
+                                       Decompressor* decompressor,
                                        MemoryAllocator* allocator,
                                        SystemClock* clock,
                                        Statistics* statistics,
@@ -104,6 +107,7 @@ class BlobFileReader {
   std::unique_ptr<RandomAccessFileReader> file_reader_;
   uint64_t file_size_;
   CompressionType compression_type_;
+  std::shared_ptr<Decompressor> decompressor_;
   SystemClock* clock_;
   Statistics* statistics_;
 };

db/blob/blob_file_reader_test.cc

@@ -65,7 +65,7 @@ void WriteBlobFile(const ImmutableOptions& immutable_options,
 
   ASSERT_OK(blob_log_writer.WriteHeader(WriteOptions(), header));
 
-  std::vector<std::string> compressed_blobs(num);
+  std::vector<GrowableBuffer> compressed_blobs(num);
   std::vector<Slice> blobs_to_write(num);
   if (kNoCompression == compression) {
     for (size_t i = 0; i < num; ++i) {
@@ -73,16 +73,13 @@ void WriteBlobFile(const ImmutableOptions& immutable_options,
       blob_sizes[i] = blobs[i].size();
     }
   } else {
-    CompressionOptions opts;
-    CompressionContext context(compression, opts);
-    CompressionInfo info(opts, context, CompressionDict::GetEmptyDict(),
-                         compression);
-
-    constexpr uint32_t compression_format_version = 2;
+    auto compressor =
+        GetBuiltinV2CompressionManager()->GetCompressor({}, compression);
 
     for (size_t i = 0; i < num; ++i) {
-      ASSERT_TRUE(OLD_CompressData(blobs[i], info, compression_format_version,
-                                   &compressed_blobs[i]));
+      ASSERT_OK(LegacyForceBuiltinCompression(*compressor,
+                                              /*working_area=*/nullptr,
+                                              blobs[i], &compressed_blobs[i]));
       blobs_to_write[i] = compressed_blobs[i];
       blob_sizes[i] = compressed_blobs[i].size();
     }
@@ -809,11 +806,10 @@ TEST_F(BlobFileReaderTest, UncompressionError) {
 
   SyncPoint::GetInstance()->SetCallBack(
       "BlobFileReader::UncompressBlobIfNeeded:TamperWithResult", [](void* arg) {
-        CacheAllocationPtr* const output =
-            static_cast<CacheAllocationPtr*>(arg);
-        assert(output);
+        auto* result = static_cast<Status*>(arg);
+        assert(result);
 
-        output->reset();
+        *result = Status::Corruption("Injected result");
       });
 
   SyncPoint::GetInstance()->EnableProcessing();
@@ -824,11 +820,12 @@ TEST_F(BlobFileReaderTest, UncompressionError) {
   std::unique_ptr<BlobContents> value;
   uint64_t bytes_read = 0;
 
-  ASSERT_TRUE(reader
-                  ->GetBlob(ReadOptions(), key, blob_offset, blob_size,
-                            kSnappyCompression, prefetch_buffer, allocator,
-                            &value, &bytes_read)
-                  .IsCorruption());
+  ASSERT_EQ(reader
+                ->GetBlob(ReadOptions(), key, blob_offset, blob_size,
+                          kSnappyCompression, prefetch_buffer, allocator,
+                          &value, &bytes_read)
+                .code(),
+            Status::Code::kCorruption);
   ASSERT_EQ(value, nullptr);
   ASSERT_EQ(bytes_read, 0);
 

db/blob/blob_source_test.cc

@@ -67,7 +67,7 @@ void WriteBlobFile(const ImmutableOptions& immutable_options,
 
   ASSERT_OK(blob_log_writer.WriteHeader(WriteOptions(), header));
 
-  std::vector<std::string> compressed_blobs(num);
+  std::vector<GrowableBuffer> compressed_blobs(num);
   std::vector<Slice> blobs_to_write(num);
   if (kNoCompression == compression) {
     for (size_t i = 0; i < num; ++i) {
@@ -75,16 +75,13 @@ void WriteBlobFile(const ImmutableOptions& immutable_options,
       blob_sizes[i] = blobs[i].size();
     }
   } else {
-    CompressionOptions opts;
-    CompressionContext context(compression, opts);
-    CompressionInfo info(opts, context, CompressionDict::GetEmptyDict(),
-                         compression);
-
-    constexpr uint32_t compression_format_version = 2;
+    auto compressor =
+        GetBuiltinV2CompressionManager()->GetCompressor({}, compression);
 
     for (size_t i = 0; i < num; ++i) {
-      ASSERT_TRUE(OLD_CompressData(blobs[i], info, compression_format_version,
-                                   &compressed_blobs[i]));
+      ASSERT_OK(LegacyForceBuiltinCompression(*compressor,
+                                              /*working_area=*/nullptr,
+                                              blobs[i], &compressed_blobs[i]));
       blobs_to_write[i] = compressed_blobs[i];
       blob_sizes[i] = compressed_blobs[i].size();
     }

db/builder.cc

@@ -56,6 +56,18 @@ TableBuilder* NewTableBuilder(const TableBuilderOptions& tboptions,
   return tboptions.moptions.table_factory->NewTableBuilder(tboptions, file);
 }
 
+void ExtractTimestampFromTableProperties(const TableProperties& tp,
+                                         FileMetaData* meta) {
+  auto min_ts_iter = tp.user_collected_properties.find("rocksdb.timestamp_min");
+  if (min_ts_iter != tp.user_collected_properties.end()) {
+    meta->min_timestamp = min_ts_iter->second;
+  }
+  auto max_ts_iter = tp.user_collected_properties.find("rocksdb.timestamp_max");
+  if (max_ts_iter != tp.user_collected_properties.end()) {
+    meta->max_timestamp = max_ts_iter->second;
+  }
+}
+
 Status BuildTable(
     const std::string& dbname, VersionSet* versions,
     const ImmutableDBOptions& db_options, const TableBuilderOptions& tboptions,
@@ -355,6 +367,7 @@ Status BuildTable(
       assert(meta->fd.GetFileSize() > 0);
       tp = builder
                ->GetTableProperties();  // refresh now that builder is finished
+      ExtractTimestampFromTableProperties(tp, meta);
       if (memtable_payload_bytes != nullptr &&
           memtable_garbage_bytes != nullptr) {
         const CompactionIterationStats& ci_stats = c_iter.iter_stats();

db/builder.h

@@ -41,6 +41,13 @@ class BlobFileCompletionCallback;
 TableBuilder* NewTableBuilder(const TableBuilderOptions& tboptions,
                               WritableFileWriter* file);
 
+// Extract min/max timestamps from table properties and populate FileMetaData.
+// This is used by both flush (BuildTable) and compaction (CompactionOutputs)
+// to populate timestamp range in FileMetaData from the TimestampTableProperties
+// collector output.
+void ExtractTimestampFromTableProperties(const TableProperties& tp,
+                                         FileMetaData* meta);
+
 // Build a Table file from the contents of *iter.  The generated file
 // will be named according to number specified in meta. On success, the rest of
 // *meta will be filled with metadata about the generated table.

db/c.cc

@@ -71,6 +71,11 @@ using ROCKSDB_NAMESPACE::CompactionFilterFactory;
 using ROCKSDB_NAMESPACE::CompactionJobInfo;
 using ROCKSDB_NAMESPACE::CompactionOptionsFIFO;
 using ROCKSDB_NAMESPACE::CompactionReason;
+using ROCKSDB_NAMESPACE::CompactionService;
+using ROCKSDB_NAMESPACE::CompactionServiceJobInfo;
+using ROCKSDB_NAMESPACE::CompactionServiceJobStatus;
+using ROCKSDB_NAMESPACE::CompactionServiceOptionsOverride;
+using ROCKSDB_NAMESPACE::CompactionServiceScheduleResponse;
 using ROCKSDB_NAMESPACE::CompactRangeOptions;
 using ROCKSDB_NAMESPACE::Comparator;
 using ROCKSDB_NAMESPACE::CompressionType;
@@ -84,10 +89,12 @@ using ROCKSDB_NAMESPACE::EnvOptions;
 using ROCKSDB_NAMESPACE::EventListener;
 using ROCKSDB_NAMESPACE::ExportImportFilesMetaData;
 using ROCKSDB_NAMESPACE::ExternalFileIngestionInfo;
+using ROCKSDB_NAMESPACE::FileChecksumGenFactory;
 using ROCKSDB_NAMESPACE::FileLock;
 using ROCKSDB_NAMESPACE::FilterPolicy;
 using ROCKSDB_NAMESPACE::FlushJobInfo;
 using ROCKSDB_NAMESPACE::FlushOptions;
+using ROCKSDB_NAMESPACE::GetFileChecksumGenCrc32cFactory;
 using ROCKSDB_NAMESPACE::HistogramData;
 using ROCKSDB_NAMESPACE::HyperClockCacheOptions;
 using ROCKSDB_NAMESPACE::ImportColumnFamilyOptions;
@@ -107,6 +114,8 @@ using ROCKSDB_NAMESPACE::NewCompactOnDeletionCollectorFactory;
 using ROCKSDB_NAMESPACE::NewGenericRateLimiter;
 using ROCKSDB_NAMESPACE::NewLRUCache;
 using ROCKSDB_NAMESPACE::NewRibbonFilterPolicy;
+using ROCKSDB_NAMESPACE::NewSstPartitionerFixedPrefixFactory;
+using ROCKSDB_NAMESPACE::OpenAndCompactOptions;
 using ROCKSDB_NAMESPACE::OptimisticTransactionDB;
 using ROCKSDB_NAMESPACE::OptimisticTransactionOptions;
 using ROCKSDB_NAMESPACE::Options;
@@ -127,9 +136,11 @@ using ROCKSDB_NAMESPACE::Snapshot;
 using ROCKSDB_NAMESPACE::SstFileManager;
 using ROCKSDB_NAMESPACE::SstFileMetaData;
 using ROCKSDB_NAMESPACE::SstFileWriter;
+using ROCKSDB_NAMESPACE::SstPartitionerFactory;
 using ROCKSDB_NAMESPACE::Status;
 using ROCKSDB_NAMESPACE::StderrLogger;
 using ROCKSDB_NAMESPACE::SubcompactionJobInfo;
+using ROCKSDB_NAMESPACE::TableFactory;
 using ROCKSDB_NAMESPACE::TablePropertiesCollectorFactory;
 using ROCKSDB_NAMESPACE::Transaction;
 using ROCKSDB_NAMESPACE::TransactionDB;
@@ -229,6 +240,15 @@ struct rocksdb_filelock_t {
 struct rocksdb_logger_t {
   std::shared_ptr<Logger> rep;
 };
+struct rocksdb_file_checksum_gen_factory_t {
+  std::shared_ptr<FileChecksumGenFactory> rep;
+};
+struct rocksdb_sst_partitioner_factory_t {
+  std::shared_ptr<SstPartitionerFactory> rep;
+};
+struct rocksdb_table_properties_collector_factory_t {
+  std::shared_ptr<TablePropertiesCollectorFactory> rep;
+};
 struct rocksdb_lru_cache_options_t {
   LRUCacheOptions rep;
 };
@@ -349,6 +369,22 @@ struct rocksdb_statistics_histogram_data_t {
   HistogramData rep;
 };
 
+struct rocksdb_compactionservice_scheduleresponse_t {
+  CompactionServiceScheduleResponse rep;
+};
+
+struct rocksdb_compactionservice_jobinfo_t {
+  CompactionServiceJobInfo rep;
+};
+
+struct rocksdb_compaction_service_options_override_t {
+  CompactionServiceOptionsOverride rep;
+};
+
+struct rocksdb_open_and_compact_options_t {
+  OpenAndCompactOptions rep;
+};
+
 struct rocksdb_compactionfilter_t : public CompactionFilter {
   void* state_;
   void (*destructor_)(void*);
@@ -643,6 +679,547 @@ static inline char* CopyString(const Slice& slice) {
   return result;
 }
 
+const char* rocksdb_compactionservice_jobinfo_t_get_db_name(
+    const rocksdb_compactionservice_jobinfo_t* info, size_t* len) {
+  *len = info->rep.db_name.size();
+  return info->rep.db_name.data();
+}
+
+const char* rocksdb_compactionservice_jobinfo_t_get_db_id(
+    const rocksdb_compactionservice_jobinfo_t* info, size_t* len) {
+  *len = info->rep.db_id.size();
+  return info->rep.db_id.data();
+}
+
+const char* rocksdb_compactionservice_jobinfo_t_get_db_session_id(
+    const rocksdb_compactionservice_jobinfo_t* info, size_t* len) {
+  *len = info->rep.db_session_id.size();
+  return info->rep.db_session_id.data();
+}
+
+const char* rocksdb_compactionservice_jobinfo_t_get_cf_name(
+    const rocksdb_compactionservice_jobinfo_t* info, size_t* len) {
+  *len = info->rep.cf_name.size();
+  return info->rep.cf_name.data();
+}
+
+uint32_t rocksdb_compactionservice_jobinfo_t_get_cf_id(
+    const rocksdb_compactionservice_jobinfo_t* info) {
+  return info->rep.cf_id;
+}
+
+uint64_t rocksdb_compactionservice_jobinfo_t_get_job_id(
+    const rocksdb_compactionservice_jobinfo_t* info) {
+  return info->rep.job_id;
+}
+
+int rocksdb_compactionservice_jobinfo_t_get_priority(
+    const rocksdb_compactionservice_jobinfo_t* info) {
+  return static_cast<int>(info->rep.priority);
+}
+
+int rocksdb_compactionservice_jobinfo_t_get_compaction_reason(
+    const rocksdb_compactionservice_jobinfo_t* info) {
+  return static_cast<int>(info->rep.compaction_reason);
+}
+
+int rocksdb_compactionservice_jobinfo_t_get_base_input_level(
+    const rocksdb_compactionservice_jobinfo_t* info) {
+  return info->rep.base_input_level;
+}
+
+int rocksdb_compactionservice_jobinfo_t_get_output_level(
+    const rocksdb_compactionservice_jobinfo_t* info) {
+  return info->rep.output_level;
+}
+
+unsigned char rocksdb_compactionservice_jobinfo_t_is_full_compaction(
+    const rocksdb_compactionservice_jobinfo_t* info) {
+  return info->rep.is_full_compaction;
+}
+
+unsigned char rocksdb_compactionservice_jobinfo_t_is_manual_compaction(
+    const rocksdb_compactionservice_jobinfo_t* info) {
+  return info->rep.is_manual_compaction;
+}
+
+unsigned char rocksdb_compactionservice_jobinfo_t_is_bottommost_level(
+    const rocksdb_compactionservice_jobinfo_t* info) {
+  return info->rep.bottommost_level;
+}
+
+// Helper function to validate compaction service job status
+static inline bool IsValidCompactionServiceJobStatus(int status) {
+  return status >= rocksdb_compactionservice_jobstatus_success &&
+         status <= rocksdb_compactionservice_jobstatus_use_local;
+}
+
+rocksdb_compactionservice_scheduleresponse_t*
+rocksdb_compactionservice_scheduleresponse_create(const char* scheduled_job_id,
+                                                  int status, char** errptr) {
+  // Validate status is in range [success=0, failure=1, aborted=2, use_local=3]
+  if (!IsValidCompactionServiceJobStatus(status)) {
+    SaveError(errptr,
+              Status::InvalidArgument("Invalid status value. Must be 0-3."));
+    return nullptr;
+  }
+
+  rocksdb_compactionservice_scheduleresponse_t* response =
+      new rocksdb_compactionservice_scheduleresponse_t{
+          CompactionServiceScheduleResponse(
+              scheduled_job_id ? std::string(scheduled_job_id) : "",
+              static_cast<CompactionServiceJobStatus>(status))};
+  return response;
+}
+
+rocksdb_compactionservice_scheduleresponse_t*
+rocksdb_compactionservice_scheduleresponse_create_with_status(int status,
+                                                              char** errptr) {
+  // Validate status is in range [success=0, failure=1, aborted=2, use_local=3]
+  if (!IsValidCompactionServiceJobStatus(status)) {
+    SaveError(errptr,
+              Status::InvalidArgument("Invalid status value. Must be 0-3."));
+    return nullptr;
+  }
+
+  rocksdb_compactionservice_scheduleresponse_t* response =
+      new rocksdb_compactionservice_scheduleresponse_t{
+          CompactionServiceScheduleResponse(
+              static_cast<CompactionServiceJobStatus>(status))};
+  return response;
+}
+
+void rocksdb_compactionservice_scheduleresponse_t_destroy(
+    rocksdb_compactionservice_scheduleresponse_t* response) {
+  if (response) {
+    delete response;
+  }
+}
+
+int rocksdb_compactionservice_scheduleresponse_getstatus(
+    const rocksdb_compactionservice_scheduleresponse_t* response) {
+  if (!response) {
+    return rocksdb_compactionservice_jobstatus_failure;
+  }
+  return static_cast<int>(response->rep.status);
+}
+
+const char* rocksdb_compactionservice_scheduleresponse_get_scheduled_job_id(
+    const rocksdb_compactionservice_scheduleresponse_t* response, size_t* len) {
+  if (!response || !len) {
+    if (len) {
+      *len = 0;
+    }
+    return "";
+  }
+  *len = response->rep.scheduled_job_id.size();
+  return response->rep.scheduled_job_id.data();
+}
+
+struct rocksdb_compactionservice_t : public CompactionService {
+  void* state_;
+  void (*destructor_)(void*);
+  rocksdb_compaction_service_schedule_cb schedule_;
+  std::string name_;
+  rocksdb_compaction_service_wait_cb wait_;
+  rocksdb_compaction_service_cancel_awaiting_jobs_cb cancel_awaiting_jobs_;
+  rocksdb_compaction_service_on_installation_cb on_installation_;
+
+  rocksdb_compactionservice_t(
+      void* state, void (*destructor)(void*),
+      rocksdb_compaction_service_schedule_cb
+          rocksdb_compaction_service_schedule_ptr,
+      const char* name, rocksdb_compaction_service_wait_cb wait,
+      rocksdb_compaction_service_cancel_awaiting_jobs_cb cancel_awaiting_jobs,
+      rocksdb_compaction_service_on_installation_cb on_installation)
+      : state_(state),
+        destructor_(destructor),
+        schedule_(rocksdb_compaction_service_schedule_ptr),
+        name_(name ? name : "CompactionService"),
+        wait_(wait),
+        cancel_awaiting_jobs_(cancel_awaiting_jobs),
+        on_installation_(on_installation) {}
+
+  ~rocksdb_compactionservice_t() override {
+    if (destructor_) {
+      (*destructor_)(state_);
+    }
+  }
+
+  const char* Name() const override { return name_.c_str(); }
+
+  CompactionServiceScheduleResponse Schedule(
+      const CompactionServiceJobInfo& info,
+      const std::string& compaction_service_input) override {
+    if (schedule_ == nullptr) {
+      return CompactionServiceScheduleResponse(
+          CompactionServiceJobStatus::kUseLocal);
+    }
+
+    rocksdb_compactionservice_scheduleresponse_t* c_response = (*schedule_)(
+        state_,
+        reinterpret_cast<const rocksdb_compactionservice_jobinfo_t*>(&info),
+        compaction_service_input.data(), compaction_service_input.size());
+
+    if (c_response == nullptr) {
+      return CompactionServiceScheduleResponse(
+          CompactionServiceJobStatus::kFailure);
+    }
+
+    CompactionServiceScheduleResponse response = std::move(c_response->rep);
+    delete c_response;
+    return response;
+  }
+
+  CompactionServiceJobStatus Wait(const std::string& scheduled_job_id,
+                                  std::string* result) override {
+    if (wait_ == nullptr) {
+      return CompactionServiceJobStatus::kUseLocal;
+    }
+
+    char* c_result = nullptr;
+    size_t result_len = 0;
+
+    int status =
+        (*wait_)(state_, scheduled_job_id.c_str(), &c_result, &result_len);
+
+    if (c_result != nullptr) {
+      if (result != nullptr) {
+        result->assign(c_result, result_len);
+      }
+      free(c_result);
+    }
+
+    return static_cast<CompactionServiceJobStatus>(status);
+  }
+
+  void CancelAwaitingJobs() override {
+    if (cancel_awaiting_jobs_ != nullptr) {
+      (*cancel_awaiting_jobs_)(state_);
+    }
+  }
+
+  void OnInstallation(const std::string& scheduled_job_id,
+                      CompactionServiceJobStatus status) override {
+    if (on_installation_ != nullptr) {
+      (*on_installation_)(state_, scheduled_job_id.c_str(),
+                          static_cast<int>(status));
+    }
+  }
+};
+
+rocksdb_compactionservice_t* rocksdb_compactionservice_create(
+    void* state, void (*destructor)(void*),
+    rocksdb_compaction_service_schedule_cb schedule, const char* name,
+    rocksdb_compaction_service_wait_cb wait,
+    rocksdb_compaction_service_cancel_awaiting_jobs_cb cancel_awaiting_jobs,
+    rocksdb_compaction_service_on_installation_cb on_installation) {
+  return new rocksdb_compactionservice_t(state, destructor, schedule, name,
+                                         wait, cancel_awaiting_jobs,
+                                         on_installation);
+}
+
+void rocksdb_options_set_compaction_service(
+    rocksdb_options_t* opt, rocksdb_compactionservice_t* service) {
+  if (!opt || !service) {
+    return;
+  }
+
+  opt->rep.compaction_service = std::shared_ptr<CompactionService>(service);
+}
+
+// CompactionServiceOptionsOverride functions
+rocksdb_compaction_service_options_override_t*
+rocksdb_compaction_service_options_override_create() {
+  return new rocksdb_compaction_service_options_override_t;
+}
+
+rocksdb_compaction_service_options_override_t*
+rocksdb_compaction_service_options_override_create_from_options(
+    rocksdb_options_t* options) {
+  if (!options) {
+    return nullptr;
+  }
+
+  rocksdb_compaction_service_options_override_t* override_opts =
+      new rocksdb_compaction_service_options_override_t;
+
+  // Copy all relevant options from rocksdb_options_t
+  override_opts->rep.env = options->rep.env;
+  override_opts->rep.file_checksum_gen_factory =
+      options->rep.file_checksum_gen_factory;
+  override_opts->rep.comparator = options->rep.comparator;
+  override_opts->rep.merge_operator = options->rep.merge_operator;
+  override_opts->rep.compaction_filter = options->rep.compaction_filter;
+  override_opts->rep.compaction_filter_factory =
+      options->rep.compaction_filter_factory;
+  override_opts->rep.prefix_extractor = options->rep.prefix_extractor;
+  override_opts->rep.table_factory = options->rep.table_factory;
+  override_opts->rep.sst_partitioner_factory =
+      options->rep.sst_partitioner_factory;
+  override_opts->rep.listeners = options->rep.listeners;
+  override_opts->rep.statistics = options->rep.statistics;
+  override_opts->rep.info_log = options->rep.info_log;
+  override_opts->rep.table_properties_collector_factories =
+      options->rep.table_properties_collector_factories;
+
+  return override_opts;
+}
+
+void rocksdb_compaction_service_options_override_destroy(
+    rocksdb_compaction_service_options_override_t* override_options) {
+  if (override_options) {
+    delete override_options;
+  }
+}
+
+void rocksdb_compaction_service_options_override_set_env(
+    rocksdb_compaction_service_options_override_t* override_options,
+    rocksdb_env_t* env) {
+  if (override_options && env) {
+    override_options->rep.env = env->rep;
+  }
+}
+
+void rocksdb_compaction_service_options_override_set_comparator(
+    rocksdb_compaction_service_options_override_t* override_options,
+    rocksdb_comparator_t* comparator) {
+  if (override_options && comparator) {
+    override_options->rep.comparator = comparator;
+  }
+}
+
+void rocksdb_compaction_service_options_override_set_merge_operator(
+    rocksdb_compaction_service_options_override_t* override_options,
+    rocksdb_mergeoperator_t* merge_operator) {
+  if (override_options && merge_operator) {
+    override_options->rep.merge_operator =
+        std::shared_ptr<MergeOperator>(merge_operator);
+  }
+}
+
+void rocksdb_compaction_service_options_override_set_compaction_filter(
+    rocksdb_compaction_service_options_override_t* override_options,
+    rocksdb_compactionfilter_t* compaction_filter) {
+  if (override_options && compaction_filter) {
+    override_options->rep.compaction_filter = compaction_filter;
+  }
+}
+
+void rocksdb_compaction_service_options_override_set_compaction_filter_factory(
+    rocksdb_compaction_service_options_override_t* override_options,
+    rocksdb_compactionfilterfactory_t* compaction_filter_factory) {
+  if (override_options && compaction_filter_factory) {
+    override_options->rep.compaction_filter_factory =
+        std::shared_ptr<CompactionFilterFactory>(compaction_filter_factory);
+  }
+}
+
+void rocksdb_compaction_service_options_override_set_prefix_extractor(
+    rocksdb_compaction_service_options_override_t* override_options,
+    rocksdb_slicetransform_t* prefix_extractor) {
+  if (override_options && prefix_extractor) {
+    override_options->rep.prefix_extractor =
+        std::shared_ptr<const SliceTransform>(prefix_extractor);
+  }
+}
+
+void rocksdb_compaction_service_options_override_set_block_based_table_factory(
+    rocksdb_compaction_service_options_override_t* override_options,
+    rocksdb_block_based_table_options_t* table_options) {
+  if (override_options && table_options) {
+    override_options->rep.table_factory = std::shared_ptr<TableFactory>(
+        NewBlockBasedTableFactory(table_options->rep));
+  }
+}
+
+void rocksdb_compaction_service_options_override_set_cuckoo_table_factory(
+    rocksdb_compaction_service_options_override_t* override_options,
+    rocksdb_cuckoo_table_options_t* table_options) {
+  if (override_options && table_options) {
+    override_options->rep.table_factory = std::shared_ptr<TableFactory>(
+        NewCuckooTableFactory(table_options->rep));
+  }
+}
+
+// Note: add_event_listener is defined later after rocksdb_eventlistener_t
+// struct
+
+void rocksdb_compaction_service_options_override_set_statistics(
+    rocksdb_compaction_service_options_override_t* override_options,
+    rocksdb_options_t* options) {
+  if (override_options && options) {
+    override_options->rep.statistics = options->rep.statistics;
+  }
+}
+
+void rocksdb_compaction_service_options_override_set_info_log(
+    rocksdb_compaction_service_options_override_t* override_options,
+    rocksdb_logger_t* logger) {
+  if (override_options && logger) {
+    override_options->rep.info_log = logger->rep;
+  }
+}
+
+void rocksdb_compaction_service_options_override_set_option(
+    rocksdb_compaction_service_options_override_t* override_options,
+    const char* key, const char* value) {
+  if (override_options && key && value) {
+    override_options->rep.options_map[std::string(key)] = std::string(value);
+  }
+}
+
+void rocksdb_compaction_service_options_override_set_file_checksum_gen_factory(
+    rocksdb_compaction_service_options_override_t* override_options,
+    rocksdb_file_checksum_gen_factory_t* factory) {
+  if (override_options && factory) {
+    override_options->rep.file_checksum_gen_factory = factory->rep;
+  }
+}
+
+void rocksdb_compaction_service_options_override_set_sst_partitioner_factory(
+    rocksdb_compaction_service_options_override_t* override_options,
+    rocksdb_sst_partitioner_factory_t* factory) {
+  if (override_options && factory) {
+    override_options->rep.sst_partitioner_factory = factory->rep;
+  }
+}
+
+void rocksdb_compaction_service_options_override_add_table_properties_collector_factory(
+    rocksdb_compaction_service_options_override_t* override_options,
+    rocksdb_table_properties_collector_factory_t* factory) {
+  if (override_options && factory) {
+    override_options->rep.table_properties_collector_factories.push_back(
+        factory->rep);
+  }
+}
+
+// Atomic bool management for cancellation
+unsigned char* rocksdb_open_and_compact_canceled_create() {
+  return reinterpret_cast<unsigned char*>(new std::atomic<bool>(false));
+}
+
+void rocksdb_open_and_compact_canceled_destroy(unsigned char* canceled) {
+  if (canceled) {
+    delete reinterpret_cast<std::atomic<bool>*>(canceled);
+  }
+}
+
+void rocksdb_open_and_compact_canceled_set(unsigned char* canceled,
+                                           unsigned char value) {
+  if (canceled) {
+    reinterpret_cast<std::atomic<bool>*>(canceled)->store(value != 0);
+  }
+}
+
+// OpenAndCompactOptions functions
+rocksdb_open_and_compact_options_t* rocksdb_open_and_compact_options_create() {
+  return new rocksdb_open_and_compact_options_t;
+}
+
+void rocksdb_open_and_compact_options_destroy(
+    rocksdb_open_and_compact_options_t* options) {
+  if (options) {
+    delete options;
+  }
+}
+
+void rocksdb_open_and_compact_options_set_canceled(
+    rocksdb_open_and_compact_options_t* options, unsigned char* canceled) {
+  if (options && canceled) {
+    options->rep.canceled = reinterpret_cast<std::atomic<bool>*>(canceled);
+  }
+}
+
+void rocksdb_open_and_compact_options_set_allow_resumption(
+    rocksdb_open_and_compact_options_t* options,
+    unsigned char allow_resumption) {
+  if (options) {
+    options->rep.allow_resumption = allow_resumption != 0;
+  }
+}
+
+// OpenAndCompact functions
+char* rocksdb_open_and_compact(
+    const char* db_path, const char* output_directory, const char* input,
+    size_t input_len, size_t* output_len,
+    const rocksdb_compaction_service_options_override_t* override_options,
+    char** errptr) {
+  if (!db_path || !output_directory || !input || !override_options) {
+    SaveError(errptr, Status::InvalidArgument("Invalid arguments"));
+    return nullptr;
+  }
+
+  std::string input_str(input, input_len);
+  std::string output_str;
+
+  Status s = DB::OpenAndCompact(db_path, output_directory, input_str,
+                                &output_str, override_options->rep);
+
+  if (!s.ok()) {
+    SaveError(errptr, s);
+    return nullptr;
+  }
+
+  // Allocate +1 for null terminator
+  char* result = static_cast<char*>(malloc(output_str.size() + 1));
+  if (!result) {
+    SaveError(errptr, Status::MemoryLimit("Failed to allocate output buffer"));
+    return nullptr;
+  }
+
+  memcpy(result, output_str.data(), output_str.size());
+  result[output_str.size()] = '\0';
+
+  // Only set output_len after successful allocation
+  if (output_len) {
+    *output_len = output_str.size();
+  }
+
+  return result;
+}
+
+char* rocksdb_open_and_compact_with_options(
+    const rocksdb_open_and_compact_options_t* options, const char* db_path,
+    const char* output_directory, const char* input, size_t input_len,
+    size_t* output_len,
+    const rocksdb_compaction_service_options_override_t* override_options,
+    char** errptr) {
+  if (!options || !db_path || !output_directory || !input ||
+      !override_options) {
+    SaveError(errptr, Status::InvalidArgument("Invalid arguments"));
+    return nullptr;
+  }
+
+  std::string input_str(input, input_len);
+  std::string output_str;
+
+  Status s = DB::OpenAndCompact(options->rep, db_path, output_directory,
+                                input_str, &output_str, override_options->rep);
+
+  if (!s.ok()) {
+    SaveError(errptr, s);
+    return nullptr;
+  }
+
+  // Allocate +1 for null terminator
+  char* result = static_cast<char*>(malloc(output_str.size() + 1));
+  if (!result) {
+    SaveError(errptr, Status::MemoryLimit("Failed to allocate output buffer"));
+    return nullptr;
+  }
+
+  memcpy(result, output_str.data(), output_str.size());
+  result[output_str.size()] = '\0';  // Null terminate
+
+  // Only set output_len after successful allocation
+  if (output_len) {
+    *output_len = output_str.size();
+  }
+
+  return result;
+}
+
 rocksdb_t* rocksdb_open(const rocksdb_options_t* options, const char* name,
                         char** errptr) {
   DB* db;
@@ -3186,6 +3763,11 @@ void rocksdb_block_based_options_set_unpartitioned_pinning_tier(
       static_cast<ROCKSDB_NAMESPACE::PinningTier>(v);
 }
 
+void rocksdb_block_based_options_set_block_align(
+    rocksdb_block_based_table_options_t* options, unsigned char v) {
+  options->rep.block_align = v;
+}
+
 /* FlushJobInfo */
 
 const char* rocksdb_flushjobinfo_cf_name(const rocksdb_flushjobinfo_t* info,
@@ -3541,6 +4123,15 @@ void rocksdb_options_add_eventlistener(rocksdb_options_t* opt,
   opt->rep.listeners.emplace_back(std::shared_ptr<EventListener>(t));
 }
 
+void rocksdb_compaction_service_options_override_add_event_listener(
+    rocksdb_compaction_service_options_override_t* override_options,
+    rocksdb_eventlistener_t* event_listener) {
+  if (override_options && event_listener) {
+    override_options->rep.listeners.emplace_back(
+        std::shared_ptr<EventListener>(event_listener));
+  }
+}
+
 rocksdb_cuckoo_table_options_t* rocksdb_cuckoo_options_create() {
   return new rocksdb_cuckoo_table_options_t;
 }
@@ -3752,6 +4343,65 @@ rocksdb_logger_t* rocksdb_logger_create_callback_logger(
 
 void rocksdb_logger_destroy(rocksdb_logger_t* logger) { delete logger; }
 
+/* File Checksum Gen Factory */
+
+rocksdb_file_checksum_gen_factory_t*
+rocksdb_file_checksum_gen_crc32c_factory_create() {
+  rocksdb_file_checksum_gen_factory_t* factory =
+      new rocksdb_file_checksum_gen_factory_t;
+  factory->rep = GetFileChecksumGenCrc32cFactory();
+  return factory;
+}
+
+void rocksdb_file_checksum_gen_factory_destroy(
+    rocksdb_file_checksum_gen_factory_t* factory) {
+  delete factory;
+}
+
+void rocksdb_options_set_file_checksum_gen_factory(
+    rocksdb_options_t* opt, rocksdb_file_checksum_gen_factory_t* factory) {
+  if (opt && factory) {
+    opt->rep.file_checksum_gen_factory = factory->rep;
+  }
+}
+
+/* SST Partitioner Factory */
+
+rocksdb_sst_partitioner_factory_t*
+rocksdb_sst_partitioner_fixed_prefix_factory_create(size_t prefix_len) {
+  rocksdb_sst_partitioner_factory_t* factory =
+      new rocksdb_sst_partitioner_factory_t;
+  factory->rep = NewSstPartitionerFixedPrefixFactory(prefix_len);
+  return factory;
+}
+
+void rocksdb_sst_partitioner_factory_destroy(
+    rocksdb_sst_partitioner_factory_t* factory) {
+  delete factory;
+}
+
+void rocksdb_options_set_sst_partitioner_factory(
+    rocksdb_options_t* opt, rocksdb_sst_partitioner_factory_t* factory) {
+  if (opt && factory) {
+    opt->rep.sst_partitioner_factory = factory->rep;
+  }
+}
+
+/* Table Properties Collector Factory */
+
+void rocksdb_table_properties_collector_factory_destroy(
+    rocksdb_table_properties_collector_factory_t* factory) {
+  delete factory;
+}
+
+void rocksdb_options_add_table_properties_collector_factory(
+    rocksdb_options_t* opt,
+    rocksdb_table_properties_collector_factory_t* factory) {
+  if (opt && factory) {
+    opt->rep.table_properties_collector_factories.push_back(factory->rep);
+  }
+}
+
 void rocksdb_options_set_env(rocksdb_options_t* opt, rocksdb_env_t* env) {
   opt->rep.env = (env ? env->rep : nullptr);
 }
@@ -7841,7 +8491,7 @@ rocksdb_memory_usage_t* rocksdb_approximate_memory_usage_create(
     dbs.push_back(db->rep);
   }
 
-  unordered_set<const Cache*> cache_set;
+  std::unordered_set<const Cache*> cache_set;
   for (auto cache : consumers->caches) {
     cache_set.insert(const_cast<const Cache*>(cache->rep.get()));
   }

db/c_test.c

@@ -722,6 +722,88 @@ static void LoadAndCheckLatestOptions(const char* db_name, rocksdb_env_t* env,
                                       num_column_families);
 }
 
+// Global state for tracking remote compaction calls
+typedef struct {
+  int schedule_called;
+  int wait_called;
+  int cancel_called;
+  char last_scheduled_job_id[256];
+  char last_db_name[256];
+} RemoteCompactionState;
+
+// Schedule callback - gets called when compaction is scheduled
+static rocksdb_compactionservice_scheduleresponse_t* RemoteCompactionSchedule(
+    void* state, const rocksdb_compactionservice_jobinfo_t* info,
+    const char* input, size_t input_len) {
+  (void)input;
+  (void)input_len;
+  RemoteCompactionState* rcs = (RemoteCompactionState*)state;
+  rcs->schedule_called++;
+
+  // Extract job info
+  size_t db_name_len;
+  const char* db_name =
+      rocksdb_compactionservice_jobinfo_t_get_db_name(info, &db_name_len);
+  memcpy(rcs->last_db_name, db_name, db_name_len);
+  rcs->last_db_name[db_name_len] = '\0';
+
+  // Generate a job ID
+  snprintf(rcs->last_scheduled_job_id, sizeof(rcs->last_scheduled_job_id),
+           "job-%d", rcs->schedule_called);
+
+  // Create response with success status
+  char* err = NULL;
+  rocksdb_compactionservice_scheduleresponse_t* response =
+      rocksdb_compactionservice_scheduleresponse_create(
+          rcs->last_scheduled_job_id,
+          rocksdb_compactionservice_jobstatus_success, &err);
+  if (err) {
+    free(err);
+  }
+  return response;
+}
+
+// Wait callback - simulates waiting for remote compaction to complete
+static int RemoteCompactionWait(void* state, const char* scheduled_job_id,
+                                char** result, size_t* result_len) {
+  RemoteCompactionState* rcs = (RemoteCompactionState*)state;
+  rcs->wait_called++;
+
+  if (strcmp(scheduled_job_id, rcs->last_scheduled_job_id) != 0) {
+    return rocksdb_compactionservice_jobstatus_failure;
+  }
+
+  // For testing purposes, return kUseLocal to cause RocksDB to fall back to
+  // local compaction. This tests the callback mechanism without needing a fully
+  // serialized result. In a real scenario, this would communicate with a remote
+  // worker that calls rocksdb_open_and_compact() and returns a properly
+  // serialized CompactionServiceResult
+  *result = NULL;
+  *result_len = 0;
+
+  return rocksdb_compactionservice_jobstatus_use_local;
+}
+
+// Cancel callback - cancels pending jobs
+static void RemoteCompactionCancel(void* state) {
+  RemoteCompactionState* rcs = (RemoteCompactionState*)state;
+  rcs->cancel_called++;
+}
+
+// Destructor callback
+static void RemoteCompactionDestroy(void* state) { (void)state; }
+
+// NULL schedule callback for testing failure handling
+static rocksdb_compactionservice_scheduleresponse_t* NullSchedule(
+    void* state, const rocksdb_compactionservice_jobinfo_t* info,
+    const char* input, size_t input_len) {
+  (void)state;
+  (void)info;
+  (void)input;
+  (void)input_len;
+  return NULL;  // Return NULL to simulate failure
+}
+
 int main(int argc, char** argv) {
   (void)argc;
   (void)argv;
@@ -4483,6 +4565,296 @@ int main(int argc, char** argv) {
     rocksdb_cache_destroy(lru);
   }
 
+  StartPhase("remote_compaction_service");
+  {
+    RemoteCompactionState remote_state = {0, 0, 0, "", ""};
+
+    // Create compaction service
+    rocksdb_compactionservice_t* service = rocksdb_compactionservice_create(
+        &remote_state,             // state
+        RemoteCompactionDestroy,   // destructor
+        RemoteCompactionSchedule,  // schedule callback
+        "TestRemoteCompaction",    // name
+        RemoteCompactionWait,      // wait callback
+        RemoteCompactionCancel,    // cancel_awaiting_jobs
+        NULL);                     // on_installation
+
+    // Create options with remote compaction
+    rocksdb_options_t* remote_options = rocksdb_options_create();
+    rocksdb_options_set_create_if_missing(remote_options, 1);
+    rocksdb_options_set_level0_file_num_compaction_trigger(remote_options, 2);
+    rocksdb_options_set_write_buffer_size(remote_options,
+                                          64 * 1024);  // 64KB buffer
+    rocksdb_options_set_max_bytes_for_level_base(remote_options,
+                                                 256 * 1024);  // 256KB
+    rocksdb_options_set_target_file_size_base(
+        remote_options, 64 * 1024);  // 64KB target file size
+    // Disable automatic compactions to test manual compaction only
+    rocksdb_options_set_disable_auto_compactions(remote_options, 1);
+    rocksdb_options_set_compaction_service(remote_options, service);
+
+    // Destroy old DB and create new one
+    rocksdb_close(db);
+    rocksdb_destroy_db(remote_options, dbname, &err);
+    CheckNoError(err);
+
+    db = rocksdb_open(remote_options, dbname, &err);
+    CheckNoError(err);
+
+    // Create multiple SST files to trigger compaction
+    rocksdb_flushoptions_t* flush_opts = rocksdb_flushoptions_create();
+    rocksdb_flushoptions_set_wait(flush_opts, 1);
+
+    // Write and flush multiple times to create multiple L0 files
+    // Write more data with larger values to ensure files are substantial
+    for (int batch = 0; batch < 5; batch++) {
+      for (int i = 0; i < 200; i++) {
+        char key[20], val[1000];
+        snprintf(key, sizeof(key), "key%d_%d", batch, i);
+        // Fill value with repeated data to make it larger
+        memset(val, 'a' + (batch % 26), sizeof(val) - 1);
+        val[sizeof(val) - 1] = '\0';
+        rocksdb_put(db, woptions, key, strlen(key), val, strlen(val), &err);
+        CheckNoError(err);
+      }
+      rocksdb_flush(db, flush_opts, &err);
+      CheckNoError(err);
+    }
+    rocksdb_flushoptions_destroy(flush_opts);
+
+    // Trigger manual compaction to invoke remote compaction service
+    rocksdb_compact_range(db, NULL, 0, NULL, 0);
+
+    rocksdb_wait_for_compact_options_t* wco =
+        rocksdb_wait_for_compact_options_create();
+    rocksdb_wait_for_compact(db, wco, &err);
+    CheckNoError(err);
+    rocksdb_wait_for_compact_options_destroy(wco);
+
+    // Verify that callbacks were actually called
+    CheckCondition(remote_state.schedule_called > 0);
+    CheckCondition(remote_state.wait_called > 0);
+    CheckCondition(strlen(remote_state.last_db_name) > 0);
+    CheckCondition(strstr(remote_state.last_db_name, "rocksdb_c_test") != NULL);
+
+    // Verify data is still accessible after remote compaction
+    // Just check a few keys to verify data integrity
+    for (int batch = 0; batch < 5; batch++) {
+      char key[20];
+      snprintf(key, sizeof(key), "key%d_0", batch);
+      size_t vallen;
+      char* val = rocksdb_get(db, roptions, key, strlen(key), &vallen, &err);
+      CheckNoError(err);
+      CheckCondition(val != NULL);
+      CheckCondition(vallen == 999);  // strlen of 1000-byte string
+      free(val);
+    }
+
+    // Test cancellation API directly
+    RemoteCompactionCancel(&remote_state);
+    CheckCondition(remote_state.cancel_called > 0);
+
+    // Cleanup
+    rocksdb_close(db);
+    rocksdb_destroy_db(remote_options, dbname, &err);
+    CheckNoError(err);
+    rocksdb_options_destroy(remote_options);
+
+    // Reopen DB with original options for subsequent tests
+    db = rocksdb_open(options, dbname, &err);
+    CheckNoError(err);
+  }
+
+  StartPhase("remote_compaction_scheduleresponse");
+  {
+    // Test scheduleresponse creation and getters
+    rocksdb_compactionservice_scheduleresponse_t* response;
+
+    // Test success response
+    err = NULL;
+    response = rocksdb_compactionservice_scheduleresponse_create(
+        "test-job-123", rocksdb_compactionservice_jobstatus_success, &err);
+    CheckNoError(err);
+    CheckCondition(response != NULL);
+    CheckCondition(
+        rocksdb_compactionservice_scheduleresponse_getstatus(response) ==
+        rocksdb_compactionservice_jobstatus_success);
+
+    size_t job_id_len;
+    const char* job_id =
+        rocksdb_compactionservice_scheduleresponse_get_scheduled_job_id(
+            response, &job_id_len);
+    CheckCondition(job_id_len == strlen("test-job-123"));
+    CheckCondition(memcmp(job_id, "test-job-123", job_id_len) == 0);
+    rocksdb_compactionservice_scheduleresponse_t_destroy(response);
+
+    // Test failure response
+    response = rocksdb_compactionservice_scheduleresponse_create_with_status(
+        rocksdb_compactionservice_jobstatus_failure, &err);
+    CheckCondition(response != NULL);
+    CheckCondition(
+        rocksdb_compactionservice_scheduleresponse_getstatus(response) ==
+        rocksdb_compactionservice_jobstatus_failure);
+    rocksdb_compactionservice_scheduleresponse_t_destroy(response);
+
+    response = rocksdb_compactionservice_scheduleresponse_create_with_status(
+        999, &err);
+    CheckCondition(response == NULL);  // Invalid status
+    if (err) {
+      Free(&err);
+    }
+  }
+
+  StartPhase("remote_compaction_options_override");
+  {
+    // Test CompactionServiceOptionsOverride API
+    rocksdb_compaction_service_options_override_t* override_opts =
+        rocksdb_compaction_service_options_override_create();
+    CheckCondition(override_opts != NULL);
+
+    // Set up override options
+    rocksdb_compaction_service_options_override_set_env(override_opts, env);
+    rocksdb_compaction_service_options_override_set_comparator(override_opts,
+                                                               cmp);
+
+    // Test file checksum gen factory
+    rocksdb_file_checksum_gen_factory_t* checksum_factory =
+        rocksdb_file_checksum_gen_crc32c_factory_create();
+    CheckCondition(checksum_factory != NULL);
+    rocksdb_compaction_service_options_override_set_file_checksum_gen_factory(
+        override_opts, checksum_factory);
+
+    // Test SST partitioner factory
+    rocksdb_sst_partitioner_factory_t* partitioner_factory =
+        rocksdb_sst_partitioner_fixed_prefix_factory_create(4);
+    CheckCondition(partitioner_factory != NULL);
+    rocksdb_compaction_service_options_override_set_sst_partitioner_factory(
+        override_opts, partitioner_factory);
+
+    // Test merge operator
+    rocksdb_compaction_service_options_override_set_merge_operator(
+        override_opts, NULL);
+
+    // Test compaction filter
+    rocksdb_compaction_service_options_override_set_compaction_filter(
+        override_opts, NULL);
+
+    // Test prefix extractor
+    rocksdb_compaction_service_options_override_set_prefix_extractor(
+        override_opts, NULL);
+
+    // Test table factory - block based
+    rocksdb_block_based_table_options_t* table_opts =
+        rocksdb_block_based_options_create();
+    rocksdb_compaction_service_options_override_set_block_based_table_factory(
+        override_opts, table_opts);
+    rocksdb_block_based_options_destroy(table_opts);
+
+    // Test statistics via options
+    rocksdb_options_t* stats_opts = rocksdb_options_create();
+    rocksdb_options_enable_statistics(stats_opts);
+    rocksdb_compaction_service_options_override_set_statistics(override_opts,
+                                                               stats_opts);
+    rocksdb_options_destroy(stats_opts);
+
+    // Test info log
+    rocksdb_logger_t* logger =
+        rocksdb_logger_create_stderr_logger(1, "test_prefix");
+    rocksdb_compaction_service_options_override_set_info_log(override_opts,
+                                                             logger);
+    rocksdb_logger_destroy(logger);
+
+    // Test options map
+    rocksdb_compaction_service_options_override_set_option(
+        override_opts, "max_bytes_for_level_base", "67108864");
+
+    // Cleanup
+    rocksdb_file_checksum_gen_factory_destroy(checksum_factory);
+    rocksdb_sst_partitioner_factory_destroy(partitioner_factory);
+    rocksdb_compaction_service_options_override_destroy(override_opts);
+  }
+
+  StartPhase("factory_options_on_regular_options");
+  {
+    // Test that the new factory types work with regular rocksdb_options_t
+    rocksdb_options_t* test_opts = rocksdb_options_create();
+
+    // Test file checksum gen factory on regular options
+    rocksdb_file_checksum_gen_factory_t* checksum_factory =
+        rocksdb_file_checksum_gen_crc32c_factory_create();
+    CheckCondition(checksum_factory != NULL);
+    rocksdb_options_set_file_checksum_gen_factory(test_opts, checksum_factory);
+
+    // Test SST partitioner factory on regular options
+    rocksdb_sst_partitioner_factory_t* partitioner_factory =
+        rocksdb_sst_partitioner_fixed_prefix_factory_create(8);
+    CheckCondition(partitioner_factory != NULL);
+    rocksdb_options_set_sst_partitioner_factory(test_opts, partitioner_factory);
+
+    // Cleanup
+    rocksdb_file_checksum_gen_factory_destroy(checksum_factory);
+    rocksdb_sst_partitioner_factory_destroy(partitioner_factory);
+    rocksdb_options_destroy(test_opts);
+  }
+
+  StartPhase("remote_compaction_null_callback_handling");
+  {
+    // Test that NULL callback returns are handled gracefully
+    // This simulates a failure in the remote compaction service
+    rocksdb_compactionservice_t* null_service =
+        rocksdb_compactionservice_create(NULL, NULL, NullSchedule,
+                                         "NullTestService", NULL, NULL, NULL);
+
+    rocksdb_options_t* null_opts = rocksdb_options_create();
+    rocksdb_options_set_create_if_missing(null_opts, 1);
+    rocksdb_options_set_compaction_service(null_opts, null_service);
+
+    const char* null_db = "rocksdb_c_test_null_service";
+
+    rocksdb_t* null_db_handle = rocksdb_open(null_opts, null_db, &err);
+    CheckNoError(err);
+
+    // Write data and trigger compaction
+    for (int i = 0; i < 100; i++) {
+      char key[20], val[50];
+      snprintf(key, sizeof(key), "key%d", i);
+      snprintf(val, sizeof(val), "val%d", i);
+      rocksdb_put(null_db_handle, woptions, key, strlen(key), val, strlen(val),
+                  &err);
+      CheckNoError(err);
+    }
+
+    // This should fall back to local compaction (not crash)
+    rocksdb_compact_range(null_db_handle, NULL, 0, NULL, 0);
+
+    // Data should still be readable
+    CheckGet(null_db_handle, roptions, "key50", "val50");
+
+    rocksdb_close(null_db_handle);
+    rocksdb_destroy_db(null_opts, null_db, &err);
+    rocksdb_options_destroy(null_opts);
+  }
+
+  StartPhase("remote_compaction_canceled_flag");
+  {
+    // Test atomic cancellation flag API
+    unsigned char* canceled = rocksdb_open_and_compact_canceled_create();
+    CheckCondition(canceled != NULL);
+
+    // Set cancellation
+    rocksdb_open_and_compact_canceled_set(canceled, 1);
+
+    // Use with OpenAndCompactOptions
+    rocksdb_open_and_compact_options_t* oac_opts =
+        rocksdb_open_and_compact_options_create();
+    rocksdb_open_and_compact_options_set_canceled(oac_opts, canceled);
+    rocksdb_open_and_compact_options_set_allow_resumption(oac_opts, 1);
+
+    // Cleanup
+    rocksdb_open_and_compact_options_destroy(oac_opts);
+    rocksdb_open_and_compact_canceled_destroy(canceled);
+  }
+
   StartPhase("sst_file_manager");
   {
     rocksdb_sst_file_manager_t* sst_file_manager;

db/column_family.cc

@@ -1248,7 +1248,7 @@ Compaction* ColumnFamilyData::PickCompaction(
   auto* result = compaction_picker_->PickCompaction(
       GetName(), mutable_options, mutable_db_options, existing_snapshots,
       snapshot_checker, current_->storage_info(), log_buffer,
-      require_max_output_level);
+      GetFullHistoryTsLow(), require_max_output_level);
   if (result != nullptr) {
     result->FinalizeInputInfo(current_);
   }
@@ -1336,7 +1336,7 @@ Compaction* ColumnFamilyData::CompactRange(
       GetName(), mutable_cf_options, mutable_db_options,
       current_->storage_info(), input_level, output_level,
       compact_range_options, begin, end, compaction_end, conflict,
-      max_file_num_to_ignore, trim_ts);
+      max_file_num_to_ignore, trim_ts, GetFullHistoryTsLow());
   if (result != nullptr) {
     result->FinalizeInputInfo(current_);
   }

db/compact_files_test.cc

@@ -433,8 +433,9 @@ TEST_F(CompactFilesTest, SentinelCompressionType) {
     ROCKSDB_NAMESPACE::TablePropertiesCollection all_tables_props;
     ASSERT_OK(db->GetPropertiesOfAllTables(&all_tables_props));
     for (const auto& name_and_table_props : all_tables_props) {
-      ASSERT_EQ(CompressionTypeToString(CompressionType::kZlibCompression),
-                name_and_table_props.second->compression_name);
+      // As of format_version 7, more elaborate information is encoded into the
+      // compression_name property
+      ASSERT_EQ("BuiltinV2;02;", name_and_table_props.second->compression_name);
     }
     delete db;
   }

db/compaction/compaction_iterator.cc

@@ -1310,14 +1310,14 @@ void CompactionIterator::PrepareOutput() {
             validity_info_.rep);
         assert(false);
       }
-      ikey_.sequence = 0;
-      last_key_seq_zeroed_ = true;
-      TEST_SYNC_POINT_CALLBACK("CompactionIterator::PrepareOutput:ZeroingSeq",
-                               &ikey_);
+
+      bool zeroed_seqno = false;
       if (!timestamp_size_) {
         current_key_.UpdateInternalKey(0, ikey_.type);
+        zeroed_seqno = true;
       } else if (full_history_ts_low_ && cmp_with_history_ts_low_ < 0) {
-        // We can also zero out timestamp for better compression.
+        // For UDT, the seqno and timestamp could only be zeroed out after the
+        // key is below history_ts_low_.
         // For the same user key (excluding timestamp), the timestamp-based
         // history can be collapsed to save some space if the timestamp is
         // older than *full_history_ts_low_.
@@ -1325,6 +1325,14 @@ void CompactionIterator::PrepareOutput() {
         const Slice ts_slice = kTsMin;
         ikey_.SetTimestamp(ts_slice);
         current_key_.UpdateInternalKey(0, ikey_.type, &ts_slice);
+        zeroed_seqno = true;
+      }
+
+      if (zeroed_seqno) {
+        ikey_.sequence = 0;
+        last_key_seq_zeroed_ = true;
+        TEST_SYNC_POINT_CALLBACK("CompactionIterator::PrepareOutput:ZeroingSeq",
+                                 &ikey_);
       }
     }
   }

db/compaction/compaction_job.cc

@@ -1480,11 +1480,11 @@ CompactionJob::CreateFileHandlers(SubcompactionState* sub_compact,
   const CompactionFileCloseFunc close_file_func =
       [this, sub_compact, start_user_key, end_user_key](
           const Status& status,
-          const ParsedInternalKey& prev_table_last_internal_key,
+          const ParsedInternalKey& prev_iter_output_internal_key,
           const Slice& next_table_min_key, const CompactionIterator* c_iter,
           CompactionOutputs& outputs) {
         return this->FinishCompactionOutputFile(
-            status, prev_table_last_internal_key, next_table_min_key,
+            status, prev_iter_output_internal_key, next_table_min_key,
             start_user_key, end_user_key, c_iter, sub_compact, outputs);
       };
 
@@ -1499,8 +1499,8 @@ Status CompactionJob::ProcessKeyValue(
   const uint64_t kRecordStatsEvery = 1000;
   [[maybe_unused]] const std::optional<const Slice> end = sub_compact->end;
 
-  IterKey last_output_key;
-  ParsedInternalKey last_output_ikey;
+  IterKey prev_iter_output_key;
+  ParsedInternalKey prev_iter_output_internal_key;
 
   TEST_SYNC_POINT_CALLBACK(
       "CompactionJob::ProcessKeyValueCompaction()::Processing",
@@ -1551,9 +1551,9 @@ Status CompactionJob::ProcessKeyValue(
     // and `close_file_func`.
     // TODO: it would be better to have the compaction file open/close moved
     // into `CompactionOutputs` which has the output file information.
-    status =
-        sub_compact->AddToOutput(*c_iter, use_proximal_output, open_file_func,
-                                 close_file_func, last_output_ikey);
+    status = sub_compact->AddToOutput(*c_iter, use_proximal_output,
+                                      open_file_func, close_file_func,
+                                      prev_iter_output_internal_key);
     if (!status.ok()) {
       break;
     }
@@ -1562,9 +1562,10 @@ Status CompactionJob::ProcessKeyValue(
                              static_cast<void*>(const_cast<std::atomic<bool>*>(
                                  &manual_compaction_canceled_)));
 
-    last_output_key.SetInternalKey(c_iter->key(), &last_output_ikey);
-    last_output_ikey.sequence = ikey.sequence;
-    last_output_ikey.type = ikey.type;
+    prev_iter_output_key.SetInternalKey(c_iter->key(),
+                                        &prev_iter_output_internal_key);
+    prev_iter_output_internal_key.sequence = ikey.sequence;
+    prev_iter_output_internal_key.type = ikey.type;
     c_iter->Next();
 
 #ifndef NDEBUG
@@ -1871,7 +1872,7 @@ void CompactionJob::RecordDroppedKeys(
 
 Status CompactionJob::FinishCompactionOutputFile(
     const Status& input_status,
-    const ParsedInternalKey& prev_table_last_internal_key,
+    const ParsedInternalKey& prev_iter_output_internal_key,
     const Slice& next_table_min_key, const Slice* comp_start_user_key,
     const Slice* comp_end_user_key, const CompactionIterator* c_iter,
     SubcompactionState* sub_compact, CompactionOutputs& outputs) {
@@ -2049,7 +2050,7 @@ Status CompactionJob::FinishCompactionOutputFile(
   }
 
   if (s.ok() && ShouldUpdateSubcompactionProgress(sub_compact, c_iter,
-                                                  prev_table_last_internal_key,
+                                                  prev_iter_output_internal_key,
                                                   next_table_min_key, meta)) {
     UpdateSubcompactionProgress(c_iter, next_table_min_key, sub_compact);
     s = PersistSubcompactionProgress(sub_compact);
@@ -2060,10 +2061,10 @@ Status CompactionJob::FinishCompactionOutputFile(
 
 bool CompactionJob::ShouldUpdateSubcompactionProgress(
     const SubcompactionState* sub_compact, const CompactionIterator* c_iter,
-    const ParsedInternalKey& prev_table_last_internal_key,
+    const ParsedInternalKey& prev_iter_output_internal_key,
     const Slice& next_table_min_internal_key, const FileMetaData* meta) const {
   const auto* cfd = sub_compact->compaction->column_family_data();
-  // No need to update when the output will not get persisted
+  // No need to update when the progress will not get persisted
   if (compaction_progress_writer_ == nullptr) {
     return false;
   }
@@ -2087,19 +2088,24 @@ bool CompactionJob::ShouldUpdateSubcompactionProgress(
   }
 
   // LIMITATION: Compaction progress persistence disabled for file boundaries
-  // contaning range deletions. Range deletions can span file boundaries, making
-  // it difficult (but possible) to ensure adjacent output tables have different
-  // user keys. See the last check for why different users keys of adjacent
-  // output tables are needed
+  // containing range deletions. Range deletions can span file boundaries,
+  // making it difficult to ensure adjacent output tables have different user
+  // keys. See the last check for why different users keys of adjacent output
+  // tables are needed
   const ValueType next_table_min_internal_key_type =
       ExtractValueType(next_table_min_internal_key);
-  const ValueType prev_table_last_internal_key_type =
-      prev_table_last_internal_key.user_key.empty()
+  const ValueType prev_iter_output_internal_key_type =
+      prev_iter_output_internal_key.user_key.empty()
           ? ValueType::kTypeValue
-          : prev_table_last_internal_key.type;
+          : prev_iter_output_internal_key.type;
 
-  if (next_table_min_internal_key_type == ValueType::kTypeRangeDeletion ||
-      prev_table_last_internal_key_type == ValueType::kTypeRangeDeletion) {
+  // Range deletes truncated to align with file boundaries may be output by the
+  // compaction iterator with `ValueType::kTypeMaxValid` instead of the original
+  // type.
+  if ((next_table_min_internal_key_type == ValueType::kTypeRangeDeletion ||
+       next_table_min_internal_key_type == ValueType::kTypeMaxValid) ||
+      (prev_iter_output_internal_key_type == ValueType::kTypeRangeDeletion ||
+       prev_iter_output_internal_key_type == ValueType::kTypeMaxValid)) {
     return false;
   }
 
@@ -2109,9 +2115,9 @@ bool CompactionJob::ShouldUpdateSubcompactionProgress(
   const Slice next_table_min_user_key =
       ExtractUserKey(next_table_min_internal_key);
   const Slice prev_table_last_user_key =
-      prev_table_last_internal_key.user_key.empty()
+      prev_iter_output_internal_key.user_key.empty()
           ? Slice()
-          : prev_table_last_internal_key.user_key;
+          : prev_iter_output_internal_key.user_key;
 
   if (cfd->user_comparator()->EqualWithoutTimestamp(next_table_min_user_key,
                                                     prev_table_last_user_key)) {
@@ -2432,6 +2438,32 @@ bool CompactionJob::UpdateInternalStatsFromInputFiles(
   bool has_error = false;
   const ReadOptions read_options(Env::IOActivity::kCompaction);
   const auto& input_table_properties = compaction->GetInputTableProperties();
+
+  // Check all input files for old block-based SST format_version. Why? Old
+  // block-based SST files from roughly version 5.0 to 5.18 could produce
+  // inaccurate num_entries counts due to the evolution of its handling along
+  // with num_range_deletions. We have to disable some paranoid checks when
+  // compacting files from such an old release. However, we don't have great
+  // information to identify those files, so we heuristically over-approximate
+  // that set of files using
+  // (a) format_version < 5, which will be true for any files from RocksDB <
+  // 6.6.0 and should not be true for any recent production files
+  // (b) to avoid including non-block-based SST files (which still use older
+  // format_version markers, and do not support DeleteRange), we also require
+  // the presence of the user property "rocksdb.block.based.table.index.type",
+  // which was added in RocksDB 2.8 and is always present in block-based tables.
+  for (const auto& tp_pair : input_table_properties) {
+    if (tp_pair.second && tp_pair.second->format_version < 5) {
+      // Check for block-based table by looking for its index type property
+      const auto& user_props = tp_pair.second->user_collected_properties;
+      if (user_props.find(BlockBasedTablePropertyNames::kIndexType) !=
+          user_props.end()) {
+        job_stats_->has_accurate_num_input_records = false;
+        break;
+      }
+    }
+  }
+
   for (int input_level = 0;
        input_level < static_cast<int>(compaction->num_input_levels());
        ++input_level) {
@@ -2676,7 +2708,10 @@ Status CompactionJob::ReadTablePropertiesDirectly(
     std::shared_ptr<const TableProperties>* tp) {
   std::unique_ptr<FSRandomAccessFile> file;
   std::string file_name = GetTableFileName(file_meta->fd.GetNumber());
-  Status s = ioptions.fs->NewRandomAccessFile(file_name, file_options_, &file,
+  FileOptions fopts = file_options_;
+  fopts.file_checksum = file_meta->file_checksum;
+  fopts.file_checksum_func_name = file_meta->file_checksum_func_name;
+  Status s = ioptions.fs->NewRandomAccessFile(file_name, fopts, &file,
                                               nullptr /* dbg */);
   if (!s.ok()) {
     return s;

db/compaction/compaction_job.h

@@ -250,6 +250,10 @@ class CompactionJob {
   // @param num_input_range_del if non-null, will be set to the number of range
   // deletion entries in this compaction input.
   //
+  // If any input file has potentially unreliable num_entries count (old SST
+  // files - details in implementation),
+  // job_stats_->has_accurate_num_input_records is set to false.
+  //
   // Returns true iff internal_stats_.output_level_stats.num_input_records and
   // num_input_range_del are calculated successfully.
   //
@@ -430,7 +434,7 @@ class CompactionJob {
 
   Status FinishCompactionOutputFile(
       const Status& input_status,
-      const ParsedInternalKey& prev_table_last_internal_key,
+      const ParsedInternalKey& prev_iter_output_internal_key,
       const Slice& next_table_min_key, const Slice* comp_start_user_key,
       const Slice* comp_end_user_key, const CompactionIterator* c_iter,
       SubcompactionState* sub_compact, CompactionOutputs& outputs);
@@ -545,7 +549,7 @@ class CompactionJob {
 
   bool ShouldUpdateSubcompactionProgress(
       const SubcompactionState* sub_compact, const CompactionIterator* c_iter,
-      const ParsedInternalKey& prev_table_last_internal_key,
+      const ParsedInternalKey& prev_iter_output_internal_key,
       const Slice& next_table_min_internal_key, const FileMetaData* meta) const;
 
   void UpdateSubcompactionProgress(const CompactionIterator* c_iter,

db/compaction/compaction_job_test.cc

@@ -2424,6 +2424,7 @@ class ResumableCompactionJobTest : public CompactionJobTestBase {
   bool enable_cancel_ = false;
   std::atomic<int> stop_count_{0};
   std::atomic<bool> cancel_{false};
+  SequenceNumber cancel_before_seqno = kMaxSequenceNumber;
 
   void SetUp() override {
     CompactionJobTestBase::SetUp();
@@ -2437,7 +2438,9 @@ class ResumableCompactionJobTest : public CompactionJobTestBase {
           if (enable_cancel_) {
             ParsedInternalKey parsed_key;
             if (ParseInternalKey(pair->second, &parsed_key, true).ok()) {
-              if (parsed_key.user_key == kCancelBeforeThisKey) {
+              if (parsed_key.user_key == kCancelBeforeThisKey &&
+                  (cancel_before_seqno == kMaxSequenceNumber ||
+                   parsed_key.sequence == cancel_before_seqno)) {
                 cancel_.store(true);
               }
             }
@@ -2665,7 +2668,7 @@ class ResumableCompactionJobTest : public CompactionJobTestBase {
       const std::initializer_list<mock::KVPair>& input_file_2,
       uint64_t last_sequence, const std::vector<uint64_t>& snapshots,
       const std::string& expected_next_key_to_compact,
-      const std::vector<std::string>& expected_input_keys, bool exists_progress,
+      const std::vector<std::string>& expected_input_keys,
       bool cancelled_past_mid_point = false) {
     std::shared_ptr<Statistics> stats = ROCKSDB_NAMESPACE::CreateDBStatistics();
 
@@ -2701,7 +2704,7 @@ class ResumableCompactionJobTest : public CompactionJobTestBase {
 
     // Resume compaction
     CompactionProgress compaction_progress;
-    if (exists_progress) {
+    if (expected_next_key_to_compact != "") {
       compaction_progress.push_back(
           ReadAndParseProgress(compaction_progress_file));
     }
@@ -2774,35 +2777,43 @@ TEST_F(ResumableCompactionJobTest, BasicProgressResume) {
       4U /* last_sequence */, {} /* snapshots */,
       kCancelBeforeThisKey /* expected_next_key_to_compact */,
       {"a", "b", "bb", kCancelBeforeThisKey} /* expected_input_keys */,
-      true /* exists_progress */, true /* cancelled_past_mid_point*/);
+      true /* cancelled_past_mid_point */);
 }
 
 TEST_F(ResumableCompactionJobTest, NoProgressResumeOnSameKey) {
   NewDB();
 
+  // `cancel_before_seqno` is set to 0U to force cancellation after
+  // `kCancelBeforeThisKey@1` instead of `kCancelBeforeThisKey@2`.
+  // The seqno is 0 because `kCancelBeforeThisKey@1` will have its sequence
+  // number zeroed during compaction while `kCancelBeforeThisKey@2` won't be
+  cancel_before_seqno = 0U;
   RunCancelAndResumeTest(
       {{KeyStr(kCancelBeforeThisKey, 1U, kTypeValue),
         "val1"}} /* input_file_1 */,
-      {{KeyStr(kCancelBeforeThisKey, 2U, kTypeValue),
-        "val2"}} /* input_file_2 */,
-      2U /* last_sequence */, {1U} /* snapshots */,
+      {{KeyStr(kCancelBeforeThisKey, 2U, kTypeValue), "val11"},
+       {KeyStr("d", 3U, kTypeValue), "val2"}} /* input_file_2 */,
+      3U /* last_sequence */, {1U} /* snapshots */,
       "" /* expected_next_key_to_compact */,
-      {kCancelBeforeThisKey, kCancelBeforeThisKey} /* expected_input_keys */,
-      false /* exists_progress */);
+      {kCancelBeforeThisKey, kCancelBeforeThisKey,
+       "d"} /* expected_input_keys */);
 }
 
 TEST_F(ResumableCompactionJobTest, NoProgressResumeOnDeleteRange) {
   NewDB();
 
   RunCancelAndResumeTest(
-      {{KeyStr(kCancelBeforeThisKey, 1U, kTypeValue),
-        "val1"}} /* input_file_1 */,
-      {{KeyStr(kCancelBeforeThisKey, 2U, kTypeRangeDeletion),
-        "val2"}} /* input_file_2 */,
-      2U /* last_sequence */, {1U} /* snapshots */,
-      "" /* expected_next_key_to_compact */,
-      {kCancelBeforeThisKey, kCancelBeforeThisKey} /* expected_input_keys */,
-      false /* exists_progress */);
+      {{KeyStr("a", 1U, kTypeValue), "val1"},
+       {KeyStr("b", 2U, kTypeValue), "val2"},
+       {KeyStr(kCancelBeforeThisKey, 3U, kTypeValue),
+        "val3"}} /* input_file_1 */,
+      {{KeyStr(kCancelBeforeThisKey, 4U, kTypeRangeDeletion),
+        "range_deletion_end_key"},
+       {KeyStr("d", 5U, kTypeValue), "val4"}} /* input_file_2 */,
+      5U /* last_sequence */, {3U} /* snapshots */,
+      "b" /* expected_next_key_to_compact */,
+      {"a", "b", kCancelBeforeThisKey, kCancelBeforeThisKey,
+       "d"} /* expected_input_keys */);
 }
 
 TEST_F(ResumableCompactionJobTest, NoProgressResumeOnMerge) {
@@ -2817,8 +2828,7 @@ TEST_F(ResumableCompactionJobTest, NoProgressResumeOnMerge) {
         "val4"}} /* input_file_2 */,
       4U /* last_sequence */, {} /* snapshots */,
       "bb" /* expected_next_key_to_compact */,
-      {"a", "b", "bb", kCancelBeforeThisKey} /* expected_input_keys */,
-      true /* exists_progress */);
+      {"a", "b", "bb", kCancelBeforeThisKey} /* expected_input_keys */);
 }
 
 TEST_F(ResumableCompactionJobTest, NoProgressResumeOnSingleDelete) {
@@ -2834,8 +2844,7 @@ TEST_F(ResumableCompactionJobTest, NoProgressResumeOnSingleDelete) {
       5U /* last_sequence */, {3U} /* snapshots */,
       "b" /* expected_next_key_to_compact */,
       {"a", "b", kCancelBeforeThisKey, kCancelBeforeThisKey,
-       "d"} /* expected_input_keys */,
-      true /* exists_progress */);
+       "d"} /* expected_input_keys */);
 }
 
 TEST_F(ResumableCompactionJobTest, NoProgressResumeOnDeletionAtBottom) {
@@ -2851,8 +2860,7 @@ TEST_F(ResumableCompactionJobTest, NoProgressResumeOnDeletionAtBottom) {
       5U /* last_sequence */, {3U} /* snapshots */,
       "b" /* expected_next_key_to_compact */,
       {"a", "b", kCancelBeforeThisKey, kCancelBeforeThisKey,
-       "d"} /* expected_input_keys */,
-      true /* exists_progress */);
+       "d"} /* expected_input_keys */);
 }
 }  // namespace ROCKSDB_NAMESPACE
 

db/compaction/compaction_outputs.cc

@@ -49,8 +49,10 @@ Status CompactionOutputs::Finish(
     meta->fd.file_size = current_bytes;
     meta->tail_size = builder_->GetTailSize();
     meta->marked_for_compaction = builder_->NeedCompact();
-    meta->user_defined_timestamps_persisted = static_cast<bool>(
-        builder_->GetTableProperties().user_defined_timestamps_persisted);
+    const TableProperties& tp = builder_->GetTableProperties();
+    meta->user_defined_timestamps_persisted =
+        static_cast<bool>(tp.user_defined_timestamps_persisted);
+    ExtractTimestampFromTableProperties(tp, meta);
   }
   current_output().finished = true;
   stats_.bytes_written += current_bytes;
@@ -364,7 +366,7 @@ Status CompactionOutputs::AddToOutput(
     const CompactionIterator& c_iter,
     const CompactionFileOpenFunc& open_file_func,
     const CompactionFileCloseFunc& close_file_func,
-    const ParsedInternalKey& prev_table_last_internal_key) {
+    const ParsedInternalKey& prev_iter_output_internal_key) {
   Status s;
   bool is_range_del = c_iter.IsDeleteRangeSentinelKey();
   if (is_range_del && compaction_->bottommost_level()) {
@@ -375,8 +377,8 @@ Status CompactionOutputs::AddToOutput(
   }
   const Slice& key = c_iter.key();
   if (ShouldStopBefore(c_iter) && HasBuilder()) {
-    s = close_file_func(c_iter.InputStatus(), prev_table_last_internal_key, key,
-                        &c_iter, *this);
+    s = close_file_func(c_iter.InputStatus(), prev_iter_output_internal_key,
+                        key, &c_iter, *this);
     if (!s.ok()) {
       return s;
     }

db/compaction/compaction_outputs.h

@@ -262,7 +262,7 @@ class CompactionOutputs {
   Status AddToOutput(const CompactionIterator& c_iter,
                      const CompactionFileOpenFunc& open_file_func,
                      const CompactionFileCloseFunc& close_file_func,
-                     const ParsedInternalKey& prev_table_last_internal_key);
+                     const ParsedInternalKey& prev_iter_output_internal_key);
 
   // Close the current output. `open_file_func` is needed for creating new file
   // for range-dels only output file.

db/compaction/compaction_picker.cc

@@ -611,7 +611,8 @@ Compaction* CompactionPicker::PickCompactionForCompactRange(
     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) {
+    uint64_t max_file_num_to_ignore, const std::string& trim_ts,
+    const std::string& full_history_ts_low) {
   // CompactionPickerFIFO has its own implementation of compact range
   assert(ioptions_.compaction_style != kCompactionStyleFIFO);
 
@@ -690,7 +691,8 @@ Compaction* CompactionPicker::PickCompactionForCompactRange(
         compact_range_options.blob_garbage_collection_age_cutoff);
 
     RegisterCompaction(c);
-    vstorage->ComputeCompactionScore(ioptions_, mutable_cf_options);
+    vstorage->ComputeCompactionScore(ioptions_, mutable_cf_options,
+                                     full_history_ts_low);
     return c;
   }
 
@@ -887,7 +889,8 @@ Compaction* CompactionPicker::PickCompactionForCompactRange(
   // takes running compactions into account (by skipping files that are already
   // being compacted). Since we just changed compaction score, we recalculate it
   // here
-  vstorage->ComputeCompactionScore(ioptions_, mutable_cf_options);
+  vstorage->ComputeCompactionScore(ioptions_, mutable_cf_options,
+                                   full_history_ts_low);
 
   return compaction;
 }

db/compaction/compaction_picker.h

@@ -65,7 +65,8 @@ class CompactionPicker {
       const MutableDBOptions& mutable_db_options,
       const std::vector<SequenceNumber>& existing_snapshots,
       const SnapshotChecker* snapshot_checker, VersionStorageInfo* vstorage,
-      LogBuffer* log_buffer, bool require_max_output_level) = 0;
+      LogBuffer* log_buffer, const std::string& full_history_ts_low,
+      bool require_max_output_level = false) = 0;
 
   // The returned Compaction might not include the whole requested range.
   // In that case, compaction_end will be set to the next key that needs
@@ -82,7 +83,8 @@ class CompactionPicker {
       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);
+      uint64_t max_file_num_to_ignore, const std::string& trim_ts,
+      const std::string& full_history_ts_low);
 
   // The maximum allowed output level.  Default value is NumberLevels() - 1.
   virtual int MaxOutputLevel() const { return NumberLevels() - 1; }
@@ -284,7 +286,8 @@ class NullCompactionPicker : public CompactionPicker {
       const std::vector<SequenceNumber>& /*existing_snapshots*/,
       const SnapshotChecker* /*snapshot_checker*/,
       VersionStorageInfo* /*vstorage*/, LogBuffer* /* log_buffer */,
-      bool /*require_max_output_level*/ = false) override {
+      const std::string& /*full_history_ts_low*/,
+      bool /*require_max_output_level*/) override {
     return nullptr;
   }
 
@@ -298,8 +301,8 @@ class NullCompactionPicker : public CompactionPicker {
       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 {
+      uint64_t /*max_file_num_to_ignore*/, const std::string& /*trim_ts*/,
+      const std::string& /*full_history_ts_low*/) override {
     return nullptr;
   }
 

db/compaction/compaction_picker_fifo.cc

@@ -419,12 +419,19 @@ Compaction* FIFOCompactionPicker::PickTemperatureChangeCompaction(
   return c;
 }
 
+// The full_history_ts_low parameter is used to control bottommost file marking
+// for compaction when user-defined timestamps (UDT) are enabled.
+
+// TODO leverage full_history_ts_low for FIFO compaction, by trigggerring
+// compaction early for data that has already expired to achieve the goal of TTL
+// enforced compliance.
 Compaction* FIFOCompactionPicker::PickCompaction(
     const std::string& cf_name, const MutableCFOptions& mutable_cf_options,
     const MutableDBOptions& mutable_db_options,
     const std::vector<SequenceNumber>& /* existing_snapshots */,
     const SnapshotChecker* /* snapshot_checker */, VersionStorageInfo* vstorage,
-    LogBuffer* log_buffer, bool /* require_max_output_level*/) {
+    LogBuffer* log_buffer, const std::string& /* full_history_ts_low */,
+    bool /* require_max_output_level*/) {
   Compaction* c = nullptr;
   if (mutable_cf_options.ttl > 0) {
     c = PickTTLCompaction(cf_name, mutable_cf_options, mutable_db_options,
@@ -449,7 +456,8 @@ Compaction* FIFOCompactionPicker::PickCompactionForCompactRange(
     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*/) {
+    uint64_t /*max_file_num_to_ignore*/, const std::string& /*trim_ts*/,
+    const std::string& full_history_ts_low) {
 #ifdef NDEBUG
   (void)input_level;
   (void)output_level;
@@ -458,10 +466,10 @@ Compaction* FIFOCompactionPicker::PickCompactionForCompactRange(
   assert(output_level == 0);
   *compaction_end = nullptr;
   LogBuffer log_buffer(InfoLogLevel::INFO_LEVEL, ioptions_.logger);
-  Compaction* c =
-      PickCompaction(cf_name, mutable_cf_options, mutable_db_options,
-                     /*existing_snapshots*/ {}, /*snapshot_checker*/ nullptr,
-                     vstorage, &log_buffer);
+  Compaction* c = PickCompaction(
+      cf_name, mutable_cf_options, mutable_db_options,
+      /*existing_snapshots*/ {}, /*snapshot_checker*/ nullptr, vstorage,
+      &log_buffer, full_history_ts_low, /* require_max_output_level */ false);
   log_buffer.FlushBufferToLog();
   return c;
 }

db/compaction/compaction_picker_fifo.h

@@ -24,6 +24,7 @@ class FIFOCompactionPicker : public CompactionPicker {
       const std::vector<SequenceNumber>& /* existing_snapshots */,
       const SnapshotChecker* /* snapshot_checker */,
       VersionStorageInfo* version, LogBuffer* log_buffer,
+      const std::string& /* full_history_ts_low */,
       bool /* require_max_output_level*/ = false) override;
 
   Compaction* PickCompactionForCompactRange(
@@ -33,7 +34,8 @@ class FIFOCompactionPicker : public CompactionPicker {
       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;
+      uint64_t max_file_num_to_ignore, const std::string& trim_ts,
+      const std::string& full_history_ts_low) override;
 
   // The maximum allowed output level.  Always returns 0.
   int MaxOutputLevel() const override { return 0; }

db/compaction/compaction_picker_level.cc

@@ -61,14 +61,16 @@ class LevelCompactionBuilder {
                          LogBuffer* log_buffer,
                          const MutableCFOptions& mutable_cf_options,
                          const ImmutableOptions& ioptions,
-                         const MutableDBOptions& mutable_db_options)
+                         const MutableDBOptions& mutable_db_options,
+                         const std::string& full_history_ts_low)
       : cf_name_(cf_name),
         vstorage_(vstorage),
         compaction_picker_(compaction_picker),
         log_buffer_(log_buffer),
         mutable_cf_options_(mutable_cf_options),
         ioptions_(ioptions),
-        mutable_db_options_(mutable_db_options) {}
+        mutable_db_options_(mutable_db_options),
+        full_history_ts_low_(full_history_ts_low) {}
 
   // Pick and return a compaction.
   Compaction* PickCompaction();
@@ -155,6 +157,7 @@ class LevelCompactionBuilder {
   const MutableCFOptions& mutable_cf_options_;
   const ImmutableOptions& ioptions_;
   const MutableDBOptions& mutable_db_options_;
+  const std::string& full_history_ts_low_;
   // Pick a path ID to place a newly generated file, with its level
   static uint32_t GetPathId(const ImmutableCFOptions& ioptions,
                             const MutableCFOptions& mutable_cf_options,
@@ -571,7 +574,8 @@ Compaction* LevelCompactionBuilder::GetCompaction() {
   // takes running compactions into account (by skipping files that are already
   // being compacted). Since we just changed compaction score, we recalculate it
   // here
-  vstorage_->ComputeCompactionScore(ioptions_, mutable_cf_options_);
+  vstorage_->ComputeCompactionScore(ioptions_, mutable_cf_options_,
+                                    full_history_ts_low_);
   return c;
 }
 
@@ -976,10 +980,11 @@ Compaction* LevelCompactionPicker::PickCompaction(
     const MutableDBOptions& mutable_db_options,
     const std::vector<SequenceNumber>& /*existing_snapshots */,
     const SnapshotChecker* /*snapshot_checker*/, VersionStorageInfo* vstorage,
-    LogBuffer* log_buffer, bool /* require_max_output_level*/) {
+    LogBuffer* log_buffer, const std::string& full_history_ts_low,
+    bool /* require_max_output_level*/) {
   LevelCompactionBuilder builder(cf_name, vstorage, this, log_buffer,
                                  mutable_cf_options, ioptions_,
-                                 mutable_db_options);
+                                 mutable_db_options, full_history_ts_low);
   return builder.PickCompaction();
 }
 }  // namespace ROCKSDB_NAMESPACE

db/compaction/compaction_picker_level.h

@@ -26,6 +26,7 @@ class LevelCompactionPicker : public CompactionPicker {
       const std::vector<SequenceNumber>& /* existing_snapshots */,
       const SnapshotChecker* /* snapshot_checker */,
       VersionStorageInfo* vstorage, LogBuffer* log_buffer,
+      const std::string& full_history_ts_low,
       bool /*require_max_output_level*/ = false) override;
 
   bool NeedsCompaction(const VersionStorageInfo* vstorage) const override;

db/compaction/compaction_picker_universal.cc

@@ -39,7 +39,7 @@ class UniversalCompactionBuilder {
       const std::vector<SequenceNumber>& existing_snapshots,
       const SnapshotChecker* snapshot_checker, VersionStorageInfo* vstorage,
       UniversalCompactionPicker* picker, LogBuffer* log_buffer,
-      bool require_max_output_level)
+      bool require_max_output_level, const std::string& full_history_ts_low)
       : ioptions_(ioptions),
         icmp_(icmp),
         cf_name_(cf_name),
@@ -50,7 +50,8 @@ class UniversalCompactionBuilder {
         log_buffer_(log_buffer),
         require_max_output_level_(require_max_output_level),
         allow_ingest_behind_(ioptions.cf_allow_ingest_behind ||
-                             ioptions.allow_ingest_behind) {
+                             ioptions.allow_ingest_behind),
+        full_history_ts_low_(full_history_ts_low) {
     assert(icmp_);
     const auto* ucmp = icmp_->user_comparator();
     assert(ucmp);
@@ -450,6 +451,7 @@ class UniversalCompactionBuilder {
   std::map<uint64_t, size_t> file_marked_for_compaction_to_sorted_run_index_;
   bool require_max_output_level_;
   bool allow_ingest_behind_;
+  const std::string& full_history_ts_low_;
 
   std::vector<UniversalCompactionBuilder::SortedRun> CalculateSortedRuns(
       const VersionStorageInfo& vstorage, int last_level,
@@ -595,16 +597,20 @@ bool UniversalCompactionPicker::NeedsCompaction(
   return false;
 }
 
+// TODO leverage full_history_ts_low in universal compaction picking. It could
+// help reduce the same infinite compaction loop issue found in level
+// compaction.
 Compaction* UniversalCompactionPicker::PickCompaction(
     const std::string& cf_name, const MutableCFOptions& mutable_cf_options,
     const MutableDBOptions& mutable_db_options,
     const std::vector<SequenceNumber>& existing_snapshots,
     const SnapshotChecker* snapshot_checker, VersionStorageInfo* vstorage,
-    LogBuffer* log_buffer, bool require_max_output_level) {
+    LogBuffer* log_buffer, const std::string& full_history_ts_low,
+    bool require_max_output_level) {
   UniversalCompactionBuilder builder(
       ioptions_, icmp_, cf_name, mutable_cf_options, mutable_db_options,
       existing_snapshots, snapshot_checker, vstorage, this, log_buffer,
-      require_max_output_level);
+      require_max_output_level, full_history_ts_low);
   return builder.PickCompaction();
 }
 
@@ -825,7 +831,8 @@ Compaction* UniversalCompactionBuilder::PickCompaction() {
   RecordInHistogram(ioptions_.stats, NUM_FILES_IN_SINGLE_COMPACTION, num_files);
 
   picker_->RegisterCompaction(c);
-  vstorage_->ComputeCompactionScore(ioptions_, mutable_cf_options_);
+  vstorage_->ComputeCompactionScore(ioptions_, mutable_cf_options_,
+                                    full_history_ts_low_);
 
   TEST_SYNC_POINT_CALLBACK("UniversalCompactionBuilder::PickCompaction:Return",
                            c);

db/compaction/compaction_picker_universal.h

@@ -26,7 +26,8 @@ class UniversalCompactionPicker : public CompactionPicker {
       const MutableDBOptions& mutable_db_options,
       const std::vector<SequenceNumber>& existing_snapshots,
       const SnapshotChecker* snapshot_checker, VersionStorageInfo* vstorage,
-      LogBuffer* log_buffer, bool require_max_output_level = false) override;
+      LogBuffer* log_buffer, const std::string& full_history_ts_low,
+      bool require_max_output_level = false) override;
   int MaxOutputLevel() const override { return NumberLevels() - 1; }
 
   bool NeedsCompaction(const VersionStorageInfo* vstorage) const override;

db/compaction/subcompaction_state.cc

@@ -109,12 +109,12 @@ Status SubcompactionState::AddToOutput(
     const CompactionIterator& iter, bool use_proximal_output,
     const CompactionFileOpenFunc& open_file_func,
     const CompactionFileCloseFunc& close_file_func,
-    const ParsedInternalKey& prev_table_last_internal_key) {
+    const ParsedInternalKey& prev_iter_output_internal_key) {
   // update target output
   current_outputs_ =
       use_proximal_output ? &proximal_level_outputs_ : &compaction_outputs_;
   return current_outputs_->AddToOutput(iter, open_file_func, close_file_func,
-                                       prev_table_last_internal_key);
+                                       prev_iter_output_internal_key);
 }
 
 }  // namespace ROCKSDB_NAMESPACE

db/compaction/subcompaction_state.h

@@ -221,7 +221,7 @@ class SubcompactionState {
   Status AddToOutput(const CompactionIterator& iter, bool use_proximal_output,
                      const CompactionFileOpenFunc& open_file_func,
                      const CompactionFileCloseFunc& close_file_func,
-                     const ParsedInternalKey& prev_table_last_internal_key);
+                     const ParsedInternalKey& prev_iter_output_internal_key);
 
   // Close all compaction output files, both output_to_proximal_level outputs
   // and normal outputs.

db/convenience.cc

@@ -65,7 +65,7 @@ Status VerifySstFileChecksum(const Options& options,
 }
 
 Status VerifySstFileChecksumInternal(const Options& options,
-                                     const EnvOptions& env_options,
+                                     const FileOptions& file_options,
                                      const ReadOptions& read_options,
                                      const std::string& file_path,
                                      const SequenceNumber& largest_seqno) {
@@ -74,8 +74,8 @@ Status VerifySstFileChecksumInternal(const Options& options,
   InternalKeyComparator internal_comparator(options.comparator);
   ImmutableOptions ioptions(options);
 
-  Status s = ioptions.fs->NewRandomAccessFile(
-      file_path, FileOptions(env_options), &file, nullptr);
+  Status s =
+      ioptions.fs->NewRandomAccessFile(file_path, file_options, &file, nullptr);
   if (s.ok()) {
     s = ioptions.fs->GetFileSize(file_path, IOOptions(), &file_size, nullptr);
   } else {
@@ -94,7 +94,7 @@ Status VerifySstFileChecksumInternal(const Options& options,
   const bool kImmortal = true;
   auto reader_options = TableReaderOptions(
       ioptions, options.prefix_extractor, options.compression_manager.get(),
-      env_options, internal_comparator, options.block_protection_bytes_per_key,
+      file_options, internal_comparator, options.block_protection_bytes_per_key,
       false /* skip_filters */, !kImmortal, false /* force_direct_prefetch */,
       -1 /* level */);
   reader_options.largest_seqno = largest_seqno;

db/convenience_impl.h

@@ -5,10 +5,11 @@
 
 #pragma once
 #include "rocksdb/db.h"
+#include "rocksdb/file_system.h"
 
 namespace ROCKSDB_NAMESPACE {
 Status VerifySstFileChecksumInternal(const Options& options,
-                                     const EnvOptions& env_options,
+                                     const FileOptions& file_options,
                                      const ReadOptions& read_options,
                                      const std::string& file_path,
                                      const SequenceNumber& largest_seqno = 0);

db/db_basic_test.cc

@@ -242,6 +242,43 @@ TEST_F(DBBasicTest, ReadOnlyDB) {
             Status::Code::kNotSupported);
 }
 
+TEST_F(DBBasicTest, ReadOnlyDBFlushWAL) {
+  // Test that FlushWAL returns NotSupported on read-only DB, and that
+  // GetLiveFilesStorageInfo works correctly even with manual_wal_flush=true.
+  // This is a regression test for a bug where GetLiveFilesStorageInfo would
+  // crash on read-only DBs with manual_wal_flush=true because FlushWAL
+  // accessed logs_.back() on an empty deque.
+  auto options = CurrentOptions();
+  options.manual_wal_flush = true;
+  DestroyAndReopen(options);
+  ASSERT_OK(Put("foo", "v1"));
+  ASSERT_OK(Put("bar", "v2"));
+  ASSERT_OK(Flush());
+  ASSERT_OK(Put("baz", "v3"));  // Unflushed data in WAL
+  Close();
+
+  // Reopen as read-only
+  ASSERT_OK(ReadOnlyReopen(options));
+  ASSERT_EQ("v1", Get("foo"));
+  ASSERT_EQ("v2", Get("bar"));
+  ASSERT_EQ("v3", Get("baz"));
+
+  // FlushWAL should return NotSupported (not crash)
+  ASSERT_EQ(db_->FlushWAL(/*sync=*/false).code(), Status::Code::kNotSupported);
+  ASSERT_EQ(db_->FlushWAL(/*sync=*/true).code(), Status::Code::kNotSupported);
+
+  // GetLiveFilesStorageInfo should succeed (previously crashed with
+  // manual_wal_flush=true because it called FlushWAL which accessed
+  // logs_.back() on empty deque)
+  LiveFilesStorageInfoOptions lfsi_opts;
+  lfsi_opts.wal_size_for_flush = 0;
+  std::vector<LiveFileStorageInfo> files;
+  ASSERT_OK(db_->GetLiveFilesStorageInfo(lfsi_opts, &files));
+  ASSERT_GT(files.size(), 0);
+
+  Close();
+}
+
 TEST_F(DBBasicTest, ReadOnlyDBWithWriteDBIdToManifestSet) {
   auto options = CurrentOptions();
   options.write_dbid_to_manifest = false;
@@ -3868,6 +3905,75 @@ TEST_F(DBBasicTest, SkipWALIfMissingTableFiles) {
   ASSERT_OK(iter->status());
 }
 
+TEST_F(DBBasicTest, BestEffortRecoveryFailureWithTableCacheUseAfterFree) {
+  Options options = CurrentOptions();
+  options.create_if_missing = true;
+  options.env = env_;
+  // Force multiple manifest files
+  options.max_manifest_file_size = 1;
+  options.max_manifest_space_amp_pct = 0;
+
+  DestroyAndReopen(options);
+
+  // Disable file deletions to preserve old manifest files for
+  // best-efforts recovery to succeed
+  ASSERT_OK(db_->DisableFileDeletions());
+
+  // Create multiple SST files to populate TableCache during
+  // best-efforts recovery
+  for (int i = 0; i < 10; i++) {
+    ASSERT_OK(Put("key" + std::to_string(i),
+                  std::string(1000, static_cast<char>('a' + i))));
+    ASSERT_OK(Flush());
+  }
+
+  // Verify we have multiple manifest files
+  std::vector<std::string> files;
+  ASSERT_OK(env_->GetChildren(dbname_, &files));
+  int manifest_count = 0;
+  for (const auto& file : files) {
+    if (file.find("MANIFEST") != std::string::npos) {
+      manifest_count++;
+    }
+  }
+  ASSERT_GE(manifest_count, 2);
+
+  // Inject corruption after TableCache is populated (count > 3), but only once
+  // (injected flag) to allow best-effort recovery to trigger retry and succeed.
+  // This coerce the bug: first recovery caches SSTs with reference to column
+  // family's options in table cache and retry deletes column family so the
+  // reference becomes dangling.
+  int count = 0;
+  bool injected = false;
+  SyncPoint::GetInstance()->SetCallBack(
+      "VersionBuilder::CheckConsistencyBeforeReturn", [&](void* arg) {
+        count++;
+        if (count > 3 && !injected) {
+          ASSERT_NE(nullptr, arg);
+          *(static_cast<Status*>(arg)) =
+              Status::Corruption("Injected corruption");
+          injected = true;
+        }
+      });
+  SyncPoint::GetInstance()->EnableProcessing();
+
+  options.best_efforts_recovery = true;
+
+  Status s = TryReopen(options);
+  ASSERT_OK(s);
+
+  SyncPoint::GetInstance()->DisableProcessing();
+  SyncPoint::GetInstance()->ClearAllCallBacks();
+
+  for (int i = 0; i < 10; i++) {
+    std::string value;
+    // Without the fix, ASAN detects use-after-free when accessing cached SST
+    // files that hold dangling references to deleted ioptions.
+    s = db_->Get(ReadOptions(), "key" + std::to_string(i), &value);
+    ASSERT_TRUE(s.ok() || s.IsNotFound());
+  }
+}
+
 TEST_F(DBBasicTest, DisableTrackWal) {
   // If WAL tracking was enabled, and then disabled during reopen,
   // the previously tracked WALs should be removed from MANIFEST.
@@ -5368,6 +5474,94 @@ INSTANTIATE_TEST_CASE_P(DBBasicTestDeadline, DBBasicTestDeadline,
                         ::testing::Values(std::make_tuple(true, false),
                                           std::make_tuple(false, true),
                                           std::make_tuple(true, true)));
+
+// FileSystemWrapper that captures FileOptions passed to NewRandomAccessFile
+// for .sst files, so we can verify file_checksum fields are populated.
+class ChecksumCapturingFS : public FileSystemWrapper {
+ public:
+  explicit ChecksumCapturingFS(const std::shared_ptr<FileSystem>& base)
+      : FileSystemWrapper(base) {}
+
+  static const char* kClassName() { return "ChecksumCapturingFS"; }
+  const char* Name() const override { return kClassName(); }
+
+  IOStatus NewRandomAccessFile(const std::string& fname,
+                               const FileOptions& opts,
+                               std::unique_ptr<FSRandomAccessFile>* result,
+                               IODebugContext* dbg) override {
+    if (fname.find(".sst") != std::string::npos) {
+      std::lock_guard<std::mutex> lock(mu_);
+      captured_file_checksum_ = opts.file_checksum;
+      captured_file_checksum_func_name_ = opts.file_checksum_func_name;
+      capture_count_++;
+    }
+    return target()->NewRandomAccessFile(fname, opts, result, dbg);
+  }
+
+  std::string GetCapturedFileChecksum() {
+    std::lock_guard<std::mutex> lock(mu_);
+    return captured_file_checksum_;
+  }
+
+  std::string GetCapturedFileChecksumFuncName() {
+    std::lock_guard<std::mutex> lock(mu_);
+    return captured_file_checksum_func_name_;
+  }
+
+  int GetCaptureCount() {
+    std::lock_guard<std::mutex> lock(mu_);
+    return capture_count_;
+  }
+
+  void Reset() {
+    std::lock_guard<std::mutex> lock(mu_);
+    captured_file_checksum_.clear();
+    captured_file_checksum_func_name_.clear();
+    capture_count_ = 0;
+  }
+
+ private:
+  std::mutex mu_;
+  std::string captured_file_checksum_;
+  std::string captured_file_checksum_func_name_;
+  int capture_count_ = 0;
+};
+
+TEST_F(DBBasicTest, FileChecksumInFileOptions) {
+  // Verify that file_checksum and file_checksum_func_name from FileMetaData
+  // are propagated through FileOptions when opening SST files.
+  auto capturing_fs =
+      std::make_shared<ChecksumCapturingFS>(env_->GetFileSystem());
+  std::unique_ptr<Env> env(new CompositeEnvWrapper(env_, capturing_fs));
+
+  Options options = GetDefaultOptions();
+  options.create_if_missing = true;
+  options.env = env.get();
+  options.file_checksum_gen_factory = GetFileChecksumGenCrc32cFactory();
+  DestroyAndReopen(options);
+
+  // Write data and flush to create an SST with a file checksum.
+  ASSERT_OK(Put("key1", "value1"));
+  ASSERT_OK(Flush());
+
+  // Reset captures, then reopen to trigger TableCache SST open.
+  capturing_fs->Reset();
+  Reopen(options);
+
+  // Read to trigger SST open through TableCache::GetTableReader.
+  ASSERT_EQ("value1", Get("key1"));
+
+  // Verify that checksum fields were populated.
+  ASSERT_GT(capturing_fs->GetCaptureCount(), 0);
+  ASSERT_FALSE(capturing_fs->GetCapturedFileChecksum().empty());
+  ASSERT_NE(capturing_fs->GetCapturedFileChecksumFuncName(),
+            capturing_fs->GetCapturedFileChecksum());
+  ASSERT_EQ(capturing_fs->GetCapturedFileChecksumFuncName(),
+            "FileChecksumCrc32c");
+
+  Close();
+}
+
 }  // namespace ROCKSDB_NAMESPACE
 
 int main(int argc, char** argv) {

db/db_flush_test.cc

@@ -101,7 +101,7 @@ TEST_F(DBFlushTest, SyncFail) {
   TEST_SYNC_POINT("DBFlushTest::SyncFail:2");
   fault_injection_env->SetFilesystemActive(true);
   // Now the background job will do the flush; wait for it.
-  // Returns the IO error happend during flush.
+  // Returns the IO error happened during flush.
   ASSERT_NOK(dbfull()->TEST_WaitForFlushMemTable());
   ASSERT_EQ("", FilesPerLevel());  // flush failed.
   Destroy(options);
@@ -518,11 +518,11 @@ TEST_F(DBFlushTest, StatisticsGarbageInsertAndDeletes) {
   // Note : one set of delete for KEY1, KEY2, KEY3 is written to
   // SSTable to propagate the delete operations to K-V pairs
   // that could have been inserted into the database during past Flush
-  // opeartions.
+  // operations.
   EXPECTED_MEMTABLE_GARBAGE_BYTES_AT_FLUSH -=
       KEY1.size() + KEY2.size() + KEY3.size() + 3 * sizeof(uint64_t);
 
-  // Additional useful paylaod.
+  // Additional useful payload.
   ASSERT_OK(Delete(KEY4));
   ASSERT_OK(Delete(KEY5));
   ASSERT_OK(Delete(KEY6));
@@ -614,7 +614,7 @@ TEST_F(DBFlushTest, StatisticsGarbageRangeDeletes) {
 
   // Note : one set of deleteRange for (KEY1, KEY2) and (KEY2, KEY3) is written
   // to SSTable to propagate the deleteRange operations to K-V pairs that could
-  // have been inserted into the database during past Flush opeartions.
+  // have been inserted into the database during past Flush operations.
   EXPECTED_MEMTABLE_GARBAGE_BYTES_AT_FLUSH -=
       (KEY1.size() + KEY2.size() + sizeof(uint64_t)) +
       (KEY2.size() + KEY3.size() + sizeof(uint64_t));
@@ -842,7 +842,7 @@ TEST_F(DBFlushTest, FixFlushReasonRaceFromConcurrentFlushes) {
     ASSERT_OK(Put(1, Key(idx), std::string(1, 'v')));
   }
 
-  // To coerce a manual flush happenning in the middle of GetLiveFiles's flush,
+  // To coerce a manual flush happening in the middle of GetLiveFiles's flush,
   // we need to pause background flush thread and enable it later.
   std::shared_ptr<test::SleepingBackgroundTask> sleeping_task =
       std::make_shared<test::SleepingBackgroundTask>();
@@ -851,7 +851,7 @@ TEST_F(DBFlushTest, FixFlushReasonRaceFromConcurrentFlushes) {
                  sleeping_task.get(), Env::Priority::HIGH);
   sleeping_task->WaitUntilSleeping();
 
-  // Coerce a manual flush happenning in the middle of GetLiveFiles's flush
+  // Coerce a manual flush happening in the middle of GetLiveFiles's flush
   bool get_live_files_paused_at_sync_point = false;
   ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->SetCallBack(
       "DBImpl::AtomicFlushMemTables:AfterScheduleFlush", [&](void* /* arg */) {
@@ -1428,7 +1428,7 @@ TEST_F(DBFlushTest, MemPurgeDeleteAndDeleteRange) {
   Close();
 }
 
-// Create a Compaction Fitler that will be invoked
+// Create a Compaction Filter that will be invoked
 // at flush time and will update the value of a KV pair
 // if the key string is "lower" than the filter_key_ string.
 class ConditionalUpdateFilter : public CompactionFilter {

db/db_impl/db_impl.cc

@@ -1177,23 +1177,38 @@ FSDirectory* DBImpl::GetDataDir(ColumnFamilyData* cfd, size_t path_id) const {
 }
 
 Status DBImpl::SetOptions(
-    ColumnFamilyHandle* column_family,
-    const std::unordered_map<std::string, std::string>& options_map) {
+    const std::unordered_map<ColumnFamilyHandle*,
+                             std::unordered_map<std::string, std::string>>&
+        column_families_opts_map) {
   // TODO: plumb Env::IOActivity, Env::IOPriority
   const ReadOptions read_options;
   const WriteOptions write_options;
 
-  auto* cfd =
-      static_cast_with_check<ColumnFamilyHandleImpl>(column_family)->cfd();
-  if (options_map.empty()) {
-    ROCKS_LOG_WARN(immutable_db_options_.info_log,
-                   "SetOptions() on column family [%s], empty input",
-                   cfd->GetName().c_str());
-    return Status::InvalidArgument("empty input");
+  if (column_families_opts_map.empty()) {
+    return Status::OK();
+  }
+
+  for (const auto& cf_opts : column_families_opts_map) {
+    if (cf_opts.second.empty()) {
+      ROCKS_LOG_WARN(immutable_db_options_.info_log,
+                     "SetOptions() on column family [%s], empty input",
+                     cf_opts.first->GetName().c_str());
+      return Status::InvalidArgument("empty input");
+    }
+  }
+
+  autovector<std::pair<ColumnFamilyData*,
+                       const std::unordered_map<std::string, std::string>*>>
+      column_family_datas;
+  for (const auto& cf_opts : column_families_opts_map) {
+    column_family_datas.push_back(
+        {static_cast_with_check<ColumnFamilyHandleImpl>(cf_opts.first)->cfd(),
+         &cf_opts.second});
   }
 
   InstrumentedMutexLock ol(&options_mutex_);
-  MutableCFOptions new_options_copy;  // For logging outside of DB mutex
+  autovector<MutableCFOptions>
+      new_options_copy;  // For logging outside of DB mutex
   Status s;
   Status persist_options_status;
   SuperVersionContext sv_context(/* create_superversion */ true);
@@ -1216,68 +1231,104 @@ Status DBImpl::SetOptions(
     // Thus aren't releasing the DB mutex from LogAndApply calling pre_cb,
     // through installing the new Version until the end of this block, after
     // installing the new SuperVersion.
-    auto pre_cb = [&]() -> Status {
-      Status cb_s = cfd->SetOptions(db_options, options_map);
-      if (cb_s.ok()) {
-        new_options_copy = cfd->GetLatestMutableCFOptions();
-      }
-      return cb_s;
-    };
     VersionEdit dummy_edit;
     dummy_edit.MarkNoManifestWriteDummy();
     TEST_SYNC_POINT_CALLBACK("DBImpl::SetOptions:dummy_edit", &dummy_edit);
-    s = versions_->LogAndApply(
-        cfd, read_options, write_options, &dummy_edit, &mutex_,
-        directories_.GetDbDir(), false /*new_descriptor_log=*/,
-        nullptr /*new_opts*/, {} /*manifest_wcb*/, pre_cb);
-    if (!versions_->io_status().ok()) {
-      assert(!s.ok());
-      error_handler_.SetBGError(versions_->io_status(),
-                                BackgroundErrorReason::kManifestWrite);
+    for (const auto& cfd_opts : column_family_datas) {
+      auto* cfd = cfd_opts.first;
+      const auto* options_map_ptr = cfd_opts.second;
+      auto pre_cb = [&]() -> Status {
+        Status cb_s = cfd->SetOptions(db_options, *options_map_ptr);
+        if (cb_s.ok()) {
+          new_options_copy.emplace_back(cfd->GetLatestMutableCFOptions());
+        }
+        return cb_s;
+      };
+
+      s = versions_->LogAndApply(
+          cfd, read_options, write_options, &dummy_edit, &mutex_,
+          directories_.GetDbDir(), false /*new_descriptor_log=*/,
+          nullptr /*new_opts*/, {} /*manifest_wcb*/, pre_cb);
+      if (!versions_->io_status().ok()) {
+        assert(!s.ok());
+        error_handler_.SetBGError(versions_->io_status(),
+                                  BackgroundErrorReason::kManifestWrite);
+      }
+      if (!s.ok()) {
+        break;
+      }
     }
 
     if (s.ok()) {
       // Trigger possible flush/compactions. This has to be before we persist
       // options to file, otherwise there will be a deadlock with writer
       // thread.
-      InstallSuperVersionForConfigChange(cfd, &sv_context);
+      for (const auto& cfd_opts : column_family_datas) {
+        InstallSuperVersionForConfigChange(cfd_opts.first, &sv_context);
+      }
       persist_options_status =
           WriteOptionsFile(write_options, true /*db_mutex_already_held*/);
       bg_cv_.SignalAll();
 
-      assert(new_options_copy == cfd->GetLatestMutableCFOptions());
-      assert(cfd->GetLatestMutableCFOptions() ==
-             cfd->GetCurrentMutableCFOptions());
-      assert(cfd->GetCurrentMutableCFOptions() ==
-             cfd->current()->GetMutableCFOptions());
+#ifndef NDEBUG
+      for (size_t i = 0; i < column_family_datas.size(); ++i) {
+        auto* cfd = column_family_datas[i].first;
+        assert(new_options_copy[i] == cfd->GetLatestMutableCFOptions());
+        assert(cfd->GetLatestMutableCFOptions() ==
+               cfd->GetCurrentMutableCFOptions());
+        assert(cfd->GetCurrentMutableCFOptions() ==
+               cfd->current()->GetMutableCFOptions());
+      }
+#endif
     }
   }
   sv_context.Clean();
 
-  if (s.ok() && (options_map.count("preserve_internal_time_seconds") > 0 ||
-                 options_map.count("preclude_last_level_data_seconds") > 0)) {
-    s = RegisterRecordSeqnoTimeWorker();
+  if (s.ok()) {
+    bool needs_seqno_worker = false;
+    for (const auto& cf_opts : column_families_opts_map) {
+      if (cf_opts.second.count("preserve_internal_time_seconds") > 0 ||
+          cf_opts.second.count("preclude_last_level_data_seconds") > 0) {
+        needs_seqno_worker = true;
+        break;
+      }
+    }
+    if (needs_seqno_worker) {
+      s = RegisterRecordSeqnoTimeWorker();
+    }
   }
 
-  ROCKS_LOG_INFO(
-      immutable_db_options_.info_log,
-      "SetOptions() on column family [%s], inputs:", cfd->GetName().c_str());
-  for (const auto& o : options_map) {
-    ROCKS_LOG_INFO(immutable_db_options_.info_log, "%s: %s\n", o.first.c_str(),
-                   o.second.c_str());
+  ROCKS_LOG_INFO(immutable_db_options_.info_log,
+                 "SetOptions() on [%zu] column families, inputs:",
+                 column_family_datas.size());
+  for (size_t i = 0; i < column_family_datas.size(); ++i) {
+    const auto* cfd = column_family_datas[i].first;
+    const auto* options_map_ptr = column_family_datas[i].second;
+    ROCKS_LOG_INFO(immutable_db_options_.info_log,
+                   "Set options on column family [%s] (%zu/%zu), inputs:",
+                   cfd->GetName().c_str(), i, column_family_datas.size());
+    for (const auto& o : *options_map_ptr) {
+      ROCKS_LOG_INFO(immutable_db_options_.info_log, "%s: %s\n",
+                     o.first.c_str(), o.second.c_str());
+    }
   }
   if (s.ok()) {
-    ROCKS_LOG_INFO(immutable_db_options_.info_log,
-                   "[%s] SetOptions() succeeded", cfd->GetName().c_str());
-    new_options_copy.Dump(immutable_db_options_.info_log.get());
+    for (size_t i = 0; i < column_family_datas.size(); ++i) {
+      const auto* cfd = column_family_datas[i].first;
+      ROCKS_LOG_INFO(immutable_db_options_.info_log,
+                     "Set options on column family [%s] (%zu/%zu) succeeded, "
+                     "updated CF options:",
+                     cfd->GetName().c_str(), i, column_family_datas.size());
+      new_options_copy[i].Dump(immutable_db_options_.info_log.get());
+    }
     if (!persist_options_status.ok()) {
       // NOTE: WriteOptionsFile already logs on failure
       s = persist_options_status;
     }
   } else {
     persist_options_status.PermitUncheckedError();  // less important
-    ROCKS_LOG_WARN(immutable_db_options_.info_log, "[%s] SetOptions() failed",
-                   cfd->GetName().c_str());
+    ROCKS_LOG_WARN(immutable_db_options_.info_log, "SetOptions() failed: %s",
+                   s.ToString().c_str());
   }
   LogFlush(immutable_db_options_.info_log);
   return s;
@@ -4351,7 +4402,8 @@ void DBImpl::ReleaseSnapshot(const Snapshot* s) {
       for (auto* cfd : *versions_->GetColumnFamilySet()) {
         if (!cfd->AllowIngestBehind()) {
           cfd->current()->storage_info()->UpdateOldestSnapshot(
-              oldest_snapshot, /*allow_ingest_behind=*/false);
+              oldest_snapshot, /*allow_ingest_behind=*/false,
+              cfd->ioptions().user_comparator, cfd->GetFullHistoryTsLow());
           if (!cfd->current()
                    ->storage_info()
                    ->BottommostFilesMarkedForCompaction()
@@ -4987,7 +5039,8 @@ Status DBImpl::DeleteFilesInRanges(ColumnFamilyHandle* column_family,
     }
     if (!deleted_files.empty()) {
       vstorage->ComputeCompactionScore(cfd->ioptions(),
-                                       cfd->GetLatestMutableCFOptions());
+                                       cfd->GetLatestMutableCFOptions(),
+                                       cfd->GetFullHistoryTsLow());
     }
     if (edit.GetDeletedFiles().empty()) {
       job_context.Clean();
@@ -6418,8 +6471,11 @@ Status DBImpl::VerifyChecksumInternal(const ReadOptions& read_options,
                                      fmeta->file_checksum_func_name, fname,
                                      read_options);
         } else {
+          FileOptions fopts = file_options_;
+          fopts.file_checksum = fmeta->file_checksum;
+          fopts.file_checksum_func_name = fmeta->file_checksum_func_name;
           s = ROCKSDB_NAMESPACE::VerifySstFileChecksumInternal(
-              opts, file_options_, read_options, fname, fd.largest_seqno);
+              opts, fopts, read_options, fname, fd.largest_seqno);
         }
         RecordTick(stats_, VERIFY_CHECKSUM_READ_BYTES,
                    IOSTATS(bytes_read) - prev_bytes_read);
@@ -6487,12 +6543,15 @@ Status DBImpl::VerifyFullFileChecksum(const std::string& file_checksum_expected,
   }
   std::string file_checksum;
   std::string func_name;
+  FileOptions fopts;
+  fopts.file_checksum = file_checksum_expected;
+  fopts.file_checksum_func_name = func_name_expected;
   s = ROCKSDB_NAMESPACE::GenerateOneFileChecksum(
       fs_.get(), fname, immutable_db_options_.file_checksum_gen_factory.get(),
       func_name_expected, &file_checksum, &func_name,
       read_options.readahead_size, immutable_db_options_.allow_mmap_reads,
       io_tracer_, immutable_db_options_.rate_limiter.get(), read_options,
-      immutable_db_options_.stats, immutable_db_options_.clock);
+      immutable_db_options_.stats, immutable_db_options_.clock, fopts);
   if (s.ok()) {
     assert(func_name_expected == func_name);
     if (file_checksum != file_checksum_expected) {
@@ -6851,7 +6910,8 @@ void DBImpl::TriggerPeriodicCompaction() {
       if (cfd->GetLatestCFOptions().periodic_compaction_seconds &&
           !cfd->queued_for_compaction()) {
         cfd->current()->storage_info()->ComputeCompactionScore(
-            cfd->ioptions(), cfd->GetLatestMutableCFOptions());
+            cfd->ioptions(), cfd->GetLatestMutableCFOptions(),
+            cfd->GetFullHistoryTsLow());
         EnqueuePendingCompaction(cfd);
         if (cfd->queued_for_compaction()) {
           ROCKS_LOG_INFO(immutable_db_options_.info_log,

db/db_impl/db_impl.h

@@ -458,8 +458,9 @@ class DBImpl : public DB {
 
   using DB::SetOptions;
   Status SetOptions(
-      ColumnFamilyHandle* column_family,
-      const std::unordered_map<std::string, std::string>& options_map) override;
+      const std::unordered_map<ColumnFamilyHandle*,
+                               std::unordered_map<std::string, std::string>>&
+          column_families_opts_map) override;
 
   Status SetDBOptions(
       const std::unordered_map<std::string, std::string>& options_map) override;

db/db_impl/db_impl_compaction_flush.cc

@@ -1449,7 +1449,8 @@ Status DBImpl::PerformTrivialMove(Compaction& c, LogBuffer* log_buffer,
                         f->file_creation_time, f->epoch_number,
                         f->file_checksum, f->file_checksum_func_name,
                         f->unique_id, f->compensated_range_deletion_size,
-                        f->tail_size, f->user_defined_timestamps_persisted);
+                        f->tail_size, f->user_defined_timestamps_persisted,
+                        f->min_timestamp, f->max_timestamp);
       moved_bytes += static_cast<size_t>(c.input(l, i)->fd.GetFileSize());
       ROCKS_LOG_BUFFER(
           log_buffer, "[%s] Moved #%" PRIu64 " to level-%d %" PRIu64 " bytes\n",
@@ -1645,8 +1646,8 @@ Status DBImpl::CompactFilesImpl(
   // takes running compactions into account (by skipping files that are already
   // being compacted). Since we just changed compaction score, we recalculate it
   // here.
-  version->storage_info()->ComputeCompactionScore(cfd->ioptions(),
-                                                  c->mutable_cf_options());
+  version->storage_info()->ComputeCompactionScore(
+      cfd->ioptions(), c->mutable_cf_options(), cfd->GetFullHistoryTsLow());
 
   compaction_job.Prepare(std::nullopt /*subcompact to be computed*/);
 
@@ -1971,7 +1972,8 @@ Status DBImpl::ReFitLevel(ColumnFamilyData* cfd, int level, int target_level) {
           f->oldest_ancester_time, f->file_creation_time, f->epoch_number,
           f->file_checksum, f->file_checksum_func_name, f->unique_id,
           f->compensated_range_deletion_size, f->tail_size,
-          f->user_defined_timestamps_persisted);
+          f->user_defined_timestamps_persisted, f->min_timestamp,
+          f->max_timestamp);
     }
     ROCKS_LOG_DEBUG(immutable_db_options_.info_log,
                     "[%s] Apply version edit:\n%s", cfd->GetName().c_str(),
@@ -2889,16 +2891,8 @@ Status DBImpl::WaitForFlushMemTables(
 
 Status DBImpl::EnableAutoCompaction(
     const std::vector<ColumnFamilyHandle*>& column_family_handles) {
-  Status s;
-  for (auto cf_ptr : column_family_handles) {
-    Status status =
-        this->SetOptions(cf_ptr, {{"disable_auto_compactions", "false"}});
-    if (!status.ok()) {
-      s = status;
-    }
-  }
-
-  return s;
+  return SetOptions(column_family_handles,
+                    {{"disable_auto_compactions", "false"}});
 }
 
 // NOTE: Calling DisableManualCompaction() may overwrite the
@@ -3839,7 +3833,8 @@ Status DBImpl::BackgroundCompaction(bool* made_progress,
               ->current()
               ->storage_info()
               ->ComputeCompactionScore(c->immutable_options(),
-                                       c->mutable_cf_options());
+                                       c->mutable_cf_options(),
+                                       cfd->GetFullHistoryTsLow());
           EnqueuePendingCompaction(cfd);
 
           c.reset();
@@ -4096,7 +4091,9 @@ Status DBImpl::BackgroundCompaction(bool* made_progress,
           in_file->unique_id,
           in_file->compensated_range_deletion_size,
           in_file->tail_size,
-          in_file->user_defined_timestamps_persisted};
+          in_file->user_defined_timestamps_persisted,
+          in_file->min_timestamp,
+          in_file->max_timestamp};
 
       out_files.push_back(std::move(out_file_metadata));
     }
@@ -4401,7 +4398,8 @@ Status DBImpl::BackgroundCompaction(bool* made_progress,
           ->current()
           ->storage_info()
           ->ComputeCompactionScore(c->immutable_options(),
-                                   c->mutable_cf_options());
+                                   c->mutable_cf_options(),
+                                   cfd->GetFullHistoryTsLow());
       EnqueuePendingCompaction(cfd);
     }
   }
@@ -4507,7 +4505,7 @@ Compaction* DBImpl::CreateIntendedCompactionForwardedToBottomPriorityPool(
                      nullptr /* snapshot_checker */, c->compaction_reason());
 
   cfd->compaction_picker()->RegisterCompaction(intended_compaction);
-  vstorage->ComputeCompactionScore(io, mo);
+  vstorage->ComputeCompactionScore(io, mo, cfd->GetFullHistoryTsLow());
   intended_compaction->FinalizeInputInfo(cfd->current());
 
   return intended_compaction;
@@ -4879,7 +4877,8 @@ void DBImpl::ResetBottomPriCompactionIntent(ColumnFamilyData* cfd,
                                             std::unique_ptr<Compaction>& c) {
   c->ReleaseCompactionFiles(Status::OK());
   cfd->current()->storage_info()->ComputeCompactionScore(
-      c->immutable_options(), c->mutable_cf_options());
+      c->immutable_options(), c->mutable_cf_options(),
+      cfd->GetFullHistoryTsLow());
   c.reset();
 }
 

db/db_impl/db_impl_experimental.cc

@@ -46,7 +46,8 @@ Status DBImpl::SuggestCompactRange(ColumnFamilyHandle* column_family,
     // Since we have some more files to compact, we should also recompute
     // compaction score
     vstorage->ComputeCompactionScore(cfd->ioptions(),
-                                     cfd->GetLatestMutableCFOptions());
+                                     cfd->GetLatestMutableCFOptions(),
+                                     cfd->GetFullHistoryTsLow());
     EnqueuePendingCompaction(cfd);
     MaybeScheduleFlushOrCompaction();
   }
@@ -143,7 +144,8 @@ Status DBImpl::PromoteL0(ColumnFamilyHandle* column_family, int target_level) {
                    f->file_creation_time, f->epoch_number, f->file_checksum,
                    f->file_checksum_func_name, f->unique_id,
                    f->compensated_range_deletion_size, f->tail_size,
-                   f->user_defined_timestamps_persisted);
+                   f->user_defined_timestamps_persisted, f->min_timestamp,
+                   f->max_timestamp);
     }
 
     status = versions_->LogAndApply(cfd, read_options, write_options, &edit,

db/db_impl/db_impl_open.cc

@@ -657,7 +657,8 @@ Status DBImpl::Recover(
                            f->file_creation_time, f->epoch_number,
                            f->file_checksum, f->file_checksum_func_name,
                            f->unique_id, f->compensated_range_deletion_size,
-                           f->tail_size, f->user_defined_timestamps_persisted);
+                           f->tail_size, f->user_defined_timestamps_persisted,
+                           f->min_timestamp, f->max_timestamp);
               ROCKS_LOG_WARN(immutable_db_options_.info_log,
                              "[%s] Moving #%" PRIu64
                              " from from_level-%d to from_level-%d %" PRIu64

db/db_impl/db_impl_readonly.h

@@ -121,6 +121,11 @@ class DBImplReadOnly : public DBImpl {
     return Status::NotSupported("Not supported operation in read only mode.");
   }
 
+  using DBImpl::FlushWAL;
+  Status FlushWAL(const FlushWALOptions& /*options*/) override {
+    return Status::NotSupported("Not supported operation in read only mode.");
+  }
+
   using DB::IngestExternalFile;
   Status IngestExternalFile(
       ColumnFamilyHandle* /*column_family*/,

db/db_impl/db_impl_secondary.cc

@@ -1102,11 +1102,6 @@ Status DBImplSecondary::InitializeCompactionWorkspace(
     return s;
   }
 
-  ROCKS_LOG_INFO(immutable_db_options_.info_log,
-                 "Initialized compaction workspace with %zu subcompaction "
-                 "progress to resume",
-                 compaction_progress_.size());
-
   return Status::OK();
 }
 
@@ -1219,6 +1214,11 @@ Status DBImplSecondary::PrepareCompactionProgressState() {
       return HandleInvalidOrNoCompactionProgress(compaction_progress_file_path,
                                                  scan_result);
     }
+
+    ROCKS_LOG_DEBUG(
+        immutable_db_options_.info_log,
+        "Loaded compaction progress with %zu subcompaction(s) from %s",
+        compaction_progress_.size(), compaction_progress_file_path.c_str());
     return s;
   } else {
     return HandleInvalidOrNoCompactionProgress(
@@ -1740,6 +1740,11 @@ Status DBImplSecondary::FinalizeCompactionProgressWriter(
     return HandleCompactionProgressWriterCreationFailure(
         "" /* temp_file_path */, final_file_path, compaction_progress_writer);
   }
+
+  ROCKS_LOG_DEBUG(immutable_db_options_.info_log,
+                  "Finalized compaction progress writer onto %s",
+                  final_file_path.c_str());
+
   return Status::OK();
 }
 }  // namespace ROCKSDB_NAMESPACE

db/db_impl/db_impl_secondary.h

@@ -216,9 +216,9 @@ class DBImplSecondary : public DBImpl {
 
   using DBImpl::SetOptions;
   Status SetOptions(
-      ColumnFamilyHandle* /*cfd*/,
-      const std::unordered_map<std::string, std::string>& /*options_map*/)
-      override {
+      const std::unordered_map<ColumnFamilyHandle*,
+                               std::unordered_map<std::string, std::string>>&
+      /*column_families_opts_map*/) override {
     // Currently not supported because changing certain options may cause
     // flush/compaction and/or write to MANIFEST.
     return Status::NotSupported("Not supported operation in secondary mode.");

db/db_iterator_test.cc

@@ -5027,6 +5027,152 @@ TEST_P(DBMultiScanIteratorTest, AsyncPrefetchWithExternalFileIngestion) {
   ASSERT_EQ(total_keys, 400);
   iter.reset();
 }
+
+TEST_P(DBMultiScanIteratorTest, StatisticsTest) {
+  // Test that multi scan statistics are properly recorded
+  auto options = CurrentOptions();
+  options.statistics = CreateDBStatistics();
+  // Use small block size to ensure multiple blocks
+  BlockBasedTableOptions table_options;
+  table_options.block_size = 256;
+  options.table_factory.reset(NewBlockBasedTableFactory(table_options));
+  DestroyAndReopen(options);
+
+  // Create data across multiple blocks
+  for (int i = 0; i < 100; ++i) {
+    std::stringstream ss;
+    ss << std::setw(3) << std::setfill('0') << i;
+    // Use larger values to ensure multiple blocks
+    ASSERT_OK(Put("k" + ss.str(), std::string(100, 'v')));
+  }
+  ASSERT_OK(Flush());
+
+  // Reset stats before multi scan
+  ASSERT_OK(options.statistics->Reset());
+
+  // Set up two scan ranges
+  std::vector<std::string> key_ranges({"k010", "k030", "k060", "k080"});
+  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);
+
+  // Iterate through all ranges
+  int count = 0;
+  try {
+    for (auto range : *iter) {
+      for (auto it : range) {
+        (void)it;
+        count++;
+      }
+    }
+  } catch (MultiScanException& ex) {
+    ASSERT_NOK(ex.status());
+    std::cerr << "Iterator returned status " << ex.what();
+    abort();
+  }
+  ASSERT_EQ(count, 40);  // 20 keys per range
+  iter.reset();
+
+  // Check statistics
+  // MULTISCAN_PREPARE_CALLS should be at least 1
+  ASSERT_GE(TestGetTickerCount(options, MULTISCAN_PREPARE_CALLS), 1);
+
+  // MULTISCAN_PREPARE_ERRORS should be 0
+  ASSERT_EQ(TestGetTickerCount(options, MULTISCAN_PREPARE_ERRORS), 0);
+
+  // MULTISCAN_SEEK_ERRORS should be 0
+  ASSERT_EQ(TestGetTickerCount(options, MULTISCAN_SEEK_ERRORS), 0);
+
+  // Blocks should be prefetched or from cache
+  uint64_t blocks_prefetched =
+      TestGetTickerCount(options, MULTISCAN_BLOCKS_PREFETCHED);
+  uint64_t blocks_from_cache =
+      TestGetTickerCount(options, MULTISCAN_BLOCKS_FROM_CACHE);
+  ASSERT_GT(blocks_prefetched + blocks_from_cache, 0);
+
+  // If blocks were prefetched, prefetch bytes and IO requests should be > 0
+  if (blocks_prefetched > 0) {
+    ASSERT_GT(TestGetTickerCount(options, MULTISCAN_PREFETCH_BYTES), 0);
+    uint64_t io_requests = TestGetTickerCount(options, MULTISCAN_IO_REQUESTS);
+    ASSERT_GT(io_requests, 0);
+    ASSERT_LE(io_requests, blocks_prefetched);
+  }
+
+  // Wasted blocks should be 0 since we iterated through everything
+  ASSERT_EQ(TestGetTickerCount(options, MULTISCAN_PREFETCH_BLOCKS_WASTED), 0);
+}
+
+TEST_P(DBMultiScanIteratorTest, StatisticsWastedBlocksTest) {
+  // Test that wasted blocks are tracked when iteration is abandoned early
+  auto options = CurrentOptions();
+  options.statistics = CreateDBStatistics();
+  // Use small block size to ensure multiple blocks
+  BlockBasedTableOptions table_options;
+  table_options.block_size = 256;
+  options.table_factory.reset(NewBlockBasedTableFactory(table_options));
+  DestroyAndReopen(options);
+
+  // Create data across multiple blocks
+  for (int i = 0; i < 100; ++i) {
+    std::stringstream ss;
+    ss << std::setw(3) << std::setfill('0') << i;
+    ASSERT_OK(Put("k" + ss.str(), std::string(100, 'v')));
+  }
+  ASSERT_OK(Flush());
+
+  // Reset stats before multi scan
+  ASSERT_OK(options.statistics->Reset());
+
+  // Set up a large scan range
+  ReadOptions ro;
+  ro.fill_cache = GetParam();
+  MultiScanArgs scan_options(BytewiseComparator());
+  scan_options.insert("k000", "k099");
+
+  ColumnFamilyHandle* cfh = dbfull()->DefaultColumnFamily();
+  std::unique_ptr<MultiScan> iter =
+      dbfull()->NewMultiScan(ro, cfh, scan_options);
+
+  // Only iterate through a few keys, then abandon
+  int count = 0;
+  try {
+    for (auto range : *iter) {
+      for (auto it : range) {
+        (void)it;
+        count++;
+        if (count >= 5) {
+          break;  // Abandon iteration early
+        }
+      }
+      if (count >= 5) {
+        break;
+      }
+    }
+  } catch (MultiScanException& ex) {
+    ASSERT_NOK(ex.status());
+    std::cerr << "Iterator returned status " << ex.what();
+    abort();
+  }
+  ASSERT_EQ(count, 5);
+
+  // Destroy iterator to trigger wasted blocks counting
+  iter.reset();
+
+  uint64_t blocks_prefetched =
+      TestGetTickerCount(options, MULTISCAN_BLOCKS_PREFETCHED);
+
+  // If blocks were prefetched, some should be wasted since we abandoned early
+  if (blocks_prefetched > 1) {
+    // We only read a few keys, so there should be wasted blocks
+    ASSERT_GT(TestGetTickerCount(options, MULTISCAN_PREFETCH_BLOCKS_WASTED), 0);
+  }
+}
 }  // namespace ROCKSDB_NAMESPACE
 
 int main(int argc, char** argv) {

db/db_options_test.cc

@@ -1689,6 +1689,46 @@ TEST_F(DBOptionsTest, SetOptionsNoManifestWrite) {
   ASSERT_EQ(Get("x"), "x");
 }
 
+TEST_F(DBOptionsTest, SetOptionsMultipleColumnFamilies) {
+  Options options;
+  options.create_if_missing = true;
+  options.env = CurrentOptions().env;
+  options.disable_auto_compactions = true;
+  Reopen(options);
+
+  // Create two additional column families
+  CreateColumnFamilies({"cf1", "cf2"}, options);
+  ReopenWithColumnFamilies({"default", "cf1", "cf2"}, options);
+
+  // Verify initial state - auto compaction should be disabled
+  ASSERT_TRUE(dbfull()->GetOptions(handles_[0]).disable_auto_compactions);
+  ASSERT_TRUE(dbfull()->GetOptions(handles_[1]).disable_auto_compactions);
+  ASSERT_TRUE(dbfull()->GetOptions(handles_[2]).disable_auto_compactions);
+
+  // Set options on multiple column families at once
+  ASSERT_OK(dbfull()->SetOptions({handles_[1], handles_[2]},
+                                 {{"disable_auto_compactions", "false"}}));
+
+  ASSERT_TRUE(
+      dbfull()->GetOptions(handles_[0]).disable_auto_compactions);  // unchanged
+  ASSERT_FALSE(
+      dbfull()->GetOptions(handles_[1]).disable_auto_compactions);  // changed
+  ASSERT_FALSE(
+      dbfull()->GetOptions(handles_[2]).disable_auto_compactions);  // changed
+
+  std::unordered_map<ColumnFamilyHandle*,
+                     std::unordered_map<std::string, std::string>>
+      options_map;
+  options_map[handles_[0]] = {{"disable_auto_compactions", "false"}};
+  options_map[handles_[1]] = {{"disable_auto_compactions", "true"}};
+  options_map[handles_[2]] = {{"disable_auto_compactions", "true"}};
+  ASSERT_OK(dbfull()->SetOptions(options_map));
+
+  ASSERT_FALSE(dbfull()->GetOptions(handles_[0]).disable_auto_compactions);
+  ASSERT_TRUE(dbfull()->GetOptions(handles_[1]).disable_auto_compactions);
+  ASSERT_TRUE(dbfull()->GetOptions(handles_[2]).disable_auto_compactions);
+}
+
 }  // namespace ROCKSDB_NAMESPACE
 
 int main(int argc, char** argv) {

db/db_properties_test.cc

@@ -1519,16 +1519,14 @@ TEST_F(DBPropertiesTest, NeedCompactHintPersistentTest) {
 
 // Excluded from RocksDB lite tests due to `GetPropertiesOfAllTables()` usage.
 TEST_F(DBPropertiesTest, BlockAddForCompressionSampling) {
-  // Sampled compression requires at least one of the following four types.
-  if (!Snappy_Supported() && !Zlib_Supported() && !LZ4_Supported() &&
-      !ZSTD_Supported()) {
-    return;
-  }
-
   Options options = CurrentOptions();
   options.disable_auto_compactions = true;
   options.table_properties_collector_factories.emplace_back(
       std::make_shared<BlockCountingTablePropertiesCollectorFactory>());
+  options.compression = kNoCompression;
+
+  bool fast_sampling_supported = Snappy_Supported() || LZ4_Supported();
+  bool slow_sampling_supported = ZSTD_Supported() || Zlib_Supported();
 
   for (bool sample_for_compression : {false, true}) {
     // For simplicity/determinism, sample 100% when enabled, or 0% when disabled
@@ -1542,10 +1540,11 @@ TEST_F(DBPropertiesTest, BlockAddForCompressionSampling) {
     // L1_0 ["a", "b"]
     //
     // L0_0 was created by flush. L1_0 was created by compaction. Each file
-    // contains one data block.
+    // contains one data block with enough data to be compressible.
     for (int i = 0; i < 3; ++i) {
-      ASSERT_OK(Put("a", "val"));
-      ASSERT_OK(Put("b", "val"));
+      for (int j = 0; j < 50; ++j) {
+        ASSERT_OK(Put(std::to_string(j), "thisismyvalue"));
+      }
       ASSERT_OK(Flush());
       if (i == 1) {
         ASSERT_OK(db_->CompactRange(CompactRangeOptions(), nullptr, nullptr));
@@ -1558,13 +1557,33 @@ TEST_F(DBPropertiesTest, BlockAddForCompressionSampling) {
     ASSERT_OK(db_->GetPropertiesOfAllTables(&file_to_props));
     ASSERT_EQ(2, file_to_props.size());
     for (const auto& file_and_props : file_to_props) {
-      auto& user_props = file_and_props.second->user_collected_properties;
+      auto& props = *file_and_props.second;
+      auto& user_props = props.user_collected_properties;
       ASSERT_TRUE(user_props.find(BlockCountingTablePropertiesCollector::
                                       kNumSampledBlocksPropertyName) !=
                   user_props.end());
       ASSERT_EQ(user_props.at(BlockCountingTablePropertiesCollector::
                                   kNumSampledBlocksPropertyName),
                 std::to_string(sample_for_compression ? 1 : 0));
+      if (sample_for_compression) {
+        EXPECT_GT(props.fast_compression_estimated_data_size, 0);
+        EXPECT_GT(props.slow_compression_estimated_data_size, 0);
+        if (fast_sampling_supported) {
+          EXPECT_LT(props.fast_compression_estimated_data_size,
+                    props.data_size);
+          if (slow_sampling_supported) {
+            EXPECT_LT(props.slow_compression_estimated_data_size,
+                      props.fast_compression_estimated_data_size);
+          }
+        }
+        if (slow_sampling_supported) {
+          EXPECT_LT(props.slow_compression_estimated_data_size,
+                    props.data_size);
+        }
+      } else {
+        EXPECT_EQ(props.fast_compression_estimated_data_size, 0);
+        EXPECT_EQ(props.slow_compression_estimated_data_size, 0);
+      }
     }
   }
 }

db/db_range_del_test.cc

@@ -3826,133 +3826,88 @@ TEST_F(DBRangeDelTest, RowCache) {
   ASSERT_OK(Put(Key(5), "foo"));
 }
 
-TEST_F(DBRangeDelTest, FileCutWithTruncatedRangeDelKey) {
-  // Test for a bug that used to generate files with meta.smallest
-  // containing kMaxValid.
-  //
-  // Setup:
-  // - Write Key(2), Key(3) and DeleteRange(Key(1), Key(4))
-  // - Flush to L0
-  // - Use SingleKeySstPartitioner to force each user key into its own file
-  // - Compact files from L0 to L1 will generate files
-  // File[0]:
-  //   smallest=[user_key=key000001, seq=4, type=15],
-  //   largest= [user_key=key000002, seq=72057594037927935, type=15]
-  // File[1]:
-  //   smallest=[user_key=key000002, seq=2, type=1],
-  //   largest= [user_key=key000003, seq=72057594037927935, type=15]
-  // File[2]:
-  //   smallest=[user_key=key000003, seq=3, type=1],
-  //   largest= [user_key=key000004, seq=72057594037927935, type=15]
-  // With range deletions truncated to each files key range.
-  //
-  // - Compact these files again into L2. RocksDB usede to set truncated
-  // range deletion start key to have value type kMaxValid. The range deletion
-  // start key is used in compaction file cutting decision.
-  // - Verify the file boundary keys after compaction have valid boundary keys
-  //
-  // Before the fix:
-  // File[0]:
-  //   smallest=[user_key=key000001, seq=4, type=15],
-  //   largest= [user_key=key000002, seq=72057594037927935, type=15]
-  // File[1]:
-  //   smallest=[user_key=key000002, seq=2, type=26],
-  //   largest= [user_key=key000003, seq=72057594037927935, type=15]
-  // File[2]:
-  //   smallest=[user_key=key000003, seq=3, type=26],
-  //   largest= [user_key=key000004, seq=72057594037927935, type=15]
-  //
-  // After the fix:
-  // File[0]:
-  //   smallest=[user_key=key000001, seq=4, type=15],
-  //   largest= [user_key=key000002, seq=72057594037927935, type=15]
-  // File[1]:
-  //   smallest=[user_key=key000002, seq=2, type=1],
-  //   largest= [user_key=key000003, seq=72057594037927935, type=15]
-  // File[2]:
-  //   smallest=[user_key=key000003, seq=3, type=1],
-  //   largest= [user_key=key000004, seq=72057594037927935, type=15]
+TEST_F(DBRangeDelTest, SeekForPrevTest) {
+  // open db
+  Options options = GetDefaultOptions();
+  options.create_if_missing = true;
+  options.compaction_style = kCompactionStyleUniversal;
 
-  Options options = CurrentOptions();
-  options.disable_auto_compactions = true;
+  // add SST partitioner, split sst file with prefix length 2
+  options.sst_partitioner_factory = NewSstPartitionerFixedPrefixFactory(2);
+  Reopen(options);
 
-  // Use partitioner that cuts before every new user key.
-  // Key(x) generates keys of length 9.
-  auto factory = std::shared_ptr<SstPartitionerFactory>(
-      NewSstPartitionerFixedPrefixFactory(10));
-  options.sst_partitioner_factory = factory;
+  // File uses SST partitioner, so it will be split into 3 files
+  // SST file 1: ka1, ka2
+  // SST file 2: kb1
+  // SST file 3: kc1, kc2
+  // Delete range covers from ka2 to kc2, which means record ka2 and kb1, kc1
+  // are covered by the delete range
 
-  DestroyAndReopen(options);
+  std::vector<std::pair<std::string, std::string>> kv = {{"ka1", "value_1"},
+                                                         {"ka2", "value_2"},
+                                                         {"kb1", "value_3"},
+                                                         {"kc1", "value_4"},
+                                                         {"kc2", "value_5"}};
+  for (auto& p : kv) {
+    ASSERT_OK(Put(p.first, p.second));
+  }
 
-  Random rnd(301);
-
-  // Create a file in a lower level so the compactions below are not
-  // bottommost compactions. Range deletion start keys are not considered
-  // in bottommost compaction.
-  ASSERT_OK(Put(Key(3), rnd.RandomBinaryString(100)));
   ASSERT_OK(Flush());
-  MoveFilesToLevel(6);
-  ASSERT_EQ(1, NumTableFilesAtLevel(6));
+  // Compact to Lmax, it should have seq 0 now.
+  ASSERT_OK(CompactRange(CompactRangeOptions(), nullptr, nullptr));
 
-  ASSERT_OK(Put(Key(2), rnd.RandomString(100)));
-  // Snapshots keep point keys alive.
-  ManagedSnapshot snapshot1(db_);
-  ASSERT_OK(Put(Key(3), rnd.RandomString(100)));
-  ManagedSnapshot snapshot2(db_);
-  ASSERT_OK(db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), Key(1),
-                             Key(4)));
-  ASSERT_OK(Flush());
-  ASSERT_EQ(1, NumTableFilesAtLevel(0));
-
-  ColumnFamilyMetaData cf_meta_l0;
-  db_->GetColumnFamilyMetaData(db_->DefaultColumnFamily(), &cf_meta_l0);
-  ASSERT_EQ(1, cf_meta_l0.levels[0].files.size());
-  std::vector<std::string> l0_filenames;
-  for (const auto& sst_file : cf_meta_l0.levels[0].files) {
-    l0_filenames.push_back(sst_file.name);
-  }
-
-  // Compact L0 files to L1
-  CompactionOptions compact_options_l0;
-  ASSERT_OK(db_->CompactFiles(compact_options_l0, l0_filenames, 1));
-  ASSERT_EQ(3, NumTableFilesAtLevel(1));
-
-  // Check L1 file metadata
-  std::vector<std::vector<FileMetaData>> files_l1;
-  dbfull()->TEST_GetFilesMetaData(db_->DefaultColumnFamily(), &files_l1);
-
-  for (const auto& file : files_l1[1]) {
-    ASSERT_LT(ExtractValueType(file.smallest.Encode()), kTypeMaxValid);
-    ASSERT_LT(ExtractValueType(file.largest.Encode()), kTypeMaxValid);
-  }
-
-  // Get file names from level 1
-  ColumnFamilyMetaData cf_meta;
-  db_->GetColumnFamilyMetaData(db_->DefaultColumnFamily(), &cf_meta);
-  std::vector<std::string> input_filenames;
-  for (const auto& sst_file : cf_meta.levels[1].files) {
-    input_filenames.push_back(sst_file.name);
-  }
-
-  // Compact files from L1 to L2
-  CompactionOptions compact_options;
-  ASSERT_OK(db_->CompactFiles(compact_options, input_filenames, 2));
-
-  // Check L2 file metadata
-  std::vector<std::vector<FileMetaData>> files;
-  dbfull()->TEST_GetFilesMetaData(db_->DefaultColumnFamily(), &files);
-
-  for (const auto& file : files[2]) {
-    ASSERT_LT(ExtractValueType(file.smallest.Encode()), kTypeMaxValid);
-    ASSERT_LT(ExtractValueType(file.largest.Encode()), kTypeMaxValid);
-  }
-
-  // // Verify iteration works correctly
-  std::unique_ptr<Iterator> iter{db_->NewIterator(ReadOptions())};
+  // Open an iterator and create a snapshot, so that keys are not deleted
+  // completely by delete range in SST
+  ReadOptions read_opts;
+  read_opts.snapshot = db_->GetSnapshot();
+  std::unique_ptr<Iterator> iter(db_->NewIterator(read_opts));
   iter->SeekToFirst();
+  // iterate all the keys and validate the value
+  for (int i = 0; iter->Valid(); iter->Next()) {
+    ASSERT_EQ(kv[i].first, iter->key().ToString());
+    ASSERT_EQ(kv[i].second, iter->value().ToString());
+    i++;
+  }
+
+  // use delete range to delete the record
+  ASSERT_OK(db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), "ka2",
+                             "kc2"));
+  // Flush
+  ASSERT_OK(Flush());
+  // Compact to Lmax
+  ASSERT_OK(CompactRange(CompactRangeOptions(), nullptr, nullptr));
+
+  // Close the iterator and release the snapshot.
   ASSERT_OK(iter->status());
-  ASSERT_FALSE(iter->Valid());
+  iter.reset();
+  db_->ReleaseSnapshot(read_opts.snapshot);
+
+  // create second iterator, seek each key and validate result
+  std::unique_ptr<Iterator> iter2(db_->NewIterator(ReadOptions()));
+  // Validate keys are deleted
+  iter2->SeekToFirst();
+  ASSERT_TRUE(iter2->Valid());
+  ASSERT_EQ("ka1", iter2->key().ToString());
+  iter2->Next();
+  ASSERT_TRUE(iter2->Valid());
+  ASSERT_EQ("kc2", iter2->key().ToString());
+  iter2->Next();
+  ASSERT_FALSE(iter2->Valid());
+
+  // Validate seek for prev result
+  for (auto& p : kv) {
+    iter2->SeekForPrev(p.first);
+    ASSERT_TRUE(iter2->Valid());
+    if (p.first == "kc2") {
+      ASSERT_EQ("kc2", iter2->key().ToString());
+    } else {
+      ASSERT_EQ("ka1", iter2->key().ToString());
+    }
+  }
+  ASSERT_OK(iter2->status());
+  iter2.reset();
 }
+
 }  // namespace ROCKSDB_NAMESPACE
 
 int main(int argc, char** argv) {

db/db_test2.cc

@@ -6544,6 +6544,9 @@ TEST_F(DBTest2, LastLevelStatistics) {
 
     DestroyAndReopen(options);
 
+    get_iostats_context()->Reset();
+    IOStatsContext* iostats = get_iostats_context();
+
     // generate 1 sst on level 0
     ASSERT_OK(Put("foo1", "bar"));
     ASSERT_OK(Put("bar", "bar"));
@@ -6644,9 +6647,87 @@ TEST_F(DBTest2, LastLevelStatistics) {
     // Control
     ASSERT_NE(options.statistics->getTickerCount(LAST_LEVEL_READ_COUNT),
               options.statistics->getTickerCount(NON_LAST_LEVEL_READ_COUNT));
+
+    // Control: unknown temperature iostats should be zero since files have
+    // explicit temperatures (mapped or written)
+    EXPECT_EQ(
+        iostats->file_io_stats_by_temperature.unknown_non_last_level_bytes_read,
+        0);
+    EXPECT_EQ(
+        iostats->file_io_stats_by_temperature.unknown_non_last_level_read_count,
+        0);
+    EXPECT_EQ(
+        iostats->file_io_stats_by_temperature.unknown_last_level_bytes_read, 0);
+    EXPECT_EQ(
+        iostats->file_io_stats_by_temperature.unknown_last_level_read_count, 0);
   }
 }
 
+// Test the iostats for files with Temperature::kUnknown that is not mapped
+// to another temperature. These stats are used to indicate which non-tiered
+// workloads are most promising for tiering (so this test doesn't set
+// temperatures).
+TEST_F(DBTest2, UnknownLastLevelStatistics) {
+  Options options = CurrentOptions();
+  options.statistics = CreateDBStatistics();
+  BlockBasedTableOptions bbto;
+  bbto.no_block_cache = true;
+  options.table_factory.reset(NewBlockBasedTableFactory(bbto));
+
+  DestroyAndReopen(options);
+
+  get_iostats_context()->Reset();
+  IOStatsContext* iostats = get_iostats_context();
+
+  // Generate 1 sst file on level 0 with kUnknown temperature
+  ASSERT_OK(Put("foo", "bar"));
+  ASSERT_OK(Flush());
+
+  // Read from the kUnknown file on non-last level
+  ASSERT_EQ("bar", Get("foo"));
+
+  // Verify unknown_non_last_level stats are populated
+  EXPECT_GT(
+      iostats->file_io_stats_by_temperature.unknown_non_last_level_bytes_read,
+      0);
+  EXPECT_GT(
+      iostats->file_io_stats_by_temperature.unknown_non_last_level_read_count,
+      0);
+  // No reads from last level yet
+  EXPECT_EQ(iostats->file_io_stats_by_temperature.unknown_last_level_bytes_read,
+            0);
+  EXPECT_EQ(iostats->file_io_stats_by_temperature.unknown_last_level_read_count,
+            0);
+
+  // Compact to the last level (level 6) explicitly using MoveFilesToLevel
+  ASSERT_OK(db_->CompactRange(CompactRangeOptions(), nullptr, nullptr));
+  MoveFilesToLevel(6);
+
+  // Reopen DB to ensure table cache is cleared and files are re-opened
+  // with correct is_last_level flag
+  Reopen(options);
+
+  // Reset iostats to measure only the following reads
+  get_iostats_context()->Reset();
+
+  // Read from the file now on last level (still kUnknown since
+  // last_level_temperature is not set)
+  ASSERT_EQ("bar", Get("foo"));
+
+  // Verify unknown_last_level stats are populated
+  EXPECT_GT(iostats->file_io_stats_by_temperature.unknown_last_level_bytes_read,
+            0);
+  EXPECT_GT(iostats->file_io_stats_by_temperature.unknown_last_level_read_count,
+            0);
+  // No new reads from non-last level
+  EXPECT_EQ(
+      iostats->file_io_stats_by_temperature.unknown_non_last_level_bytes_read,
+      0);
+  EXPECT_EQ(
+      iostats->file_io_stats_by_temperature.unknown_non_last_level_read_count,
+      0);
+}
+
 TEST_F(DBTest2, CheckpointFileTemperature) {
   class NoLinkTestFS : public FileTemperatureTestFS {
     using FileTemperatureTestFS::FileTemperatureTestFS;

db/db_wal_test.cc

@@ -1613,7 +1613,7 @@ TEST_F(DBWALTest, SyncWalPartialFailure) {
       return s;
     }
 
-    AcqRelAtomic<uint32_t> syncs_before_failure_{UINT32_MAX};
+    Atomic<uint32_t> syncs_before_failure_{UINT32_MAX};
 
    protected:
     class MyTestWritableFile : public FSWritableFileOwnerWrapper {

db/db_with_timestamp_compaction_test.cc

@@ -7,9 +7,13 @@
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file. See the AUTHORS file for names of contributors.
 
+#include <set>
+
+#include "db/column_family.h"
 #include "db/compaction/compaction.h"
 #include "db/db_test_util.h"
 #include "port/stack_trace.h"
+#include "rocksdb/sst_file_reader.h"
 #include "test_util/testutil.h"
 
 namespace ROCKSDB_NAMESPACE {
@@ -48,6 +52,122 @@ class TimestampCompatibleCompactionTest : public DBTestBase {
     }
     return value;
   }
+
+  // Helper to get all files with their level and timestamps
+  std::vector<std::tuple<int, std::string, std::string>>
+  GetAllFileTimestamps() {
+    std::vector<std::tuple<int, std::string, std::string>> results;
+    ColumnFamilyHandle* cfh = db_->DefaultColumnFamily();
+    auto* cfd = static_cast_with_check<ColumnFamilyHandleImpl>(cfh)->cfd();
+    auto* vstorage = cfd->current()->storage_info();
+
+    for (int level = 0; level < cfd->NumberLevels(); level++) {
+      for (auto* file : vstorage->LevelFiles(level)) {
+        results.emplace_back(level, file->min_timestamp, file->max_timestamp);
+      }
+    }
+    return results;
+  }
+
+  // Helper to compute overall min/max timestamps across all files
+  // Returns {min_ts, max_ts} as uint64_t values
+  // Asserts that all files have non-empty timestamps
+  std::pair<uint64_t, uint64_t> GetOverallTimestampRange() {
+    auto files = GetAllFileTimestamps();
+    EXPECT_GE(files.size(), 1U);
+
+    uint64_t overall_min = UINT64_MAX;
+    uint64_t overall_max = 0;
+    for (const auto& [level, min_ts, max_ts] : files) {
+      EXPECT_FALSE(min_ts.empty()) << "min_timestamp empty at level " << level;
+      EXPECT_FALSE(max_ts.empty()) << "max_timestamp empty at level " << level;
+
+      if (!min_ts.empty() && !max_ts.empty()) {
+        uint64_t file_min = DecodeFixed64(min_ts.data());
+        uint64_t file_max = DecodeFixed64(max_ts.data());
+        overall_min = std::min(overall_min, file_min);
+        overall_max = std::max(overall_max, file_max);
+      }
+    }
+    return {overall_min, overall_max};
+  }
+
+  // Helper to verify timestamp range matches expected values, including after
+  // reopen
+  void VerifyTimestampRangeWithPersistence(const Options& options,
+                                           uint64_t expected_min,
+                                           uint64_t expected_max) {
+    // Verify before reopen
+    auto [min_ts, max_ts] = GetOverallTimestampRange();
+    ASSERT_EQ(expected_min, min_ts);
+    ASSERT_EQ(expected_max, max_ts);
+
+    size_t file_count_before = GetAllFileTimestamps().size();
+
+    // Verify manifest persistence by reopening
+    Reopen(options);
+
+    // Verify after reopen
+    auto [reopened_min_ts, reopened_max_ts] = GetOverallTimestampRange();
+    ASSERT_EQ(expected_min, reopened_min_ts);
+    ASSERT_EQ(expected_max, reopened_max_ts);
+    ASSERT_EQ(file_count_before, GetAllFileTimestamps().size());
+  }
+
+  // Helper to create common options for UDT tests with level compaction
+  Options CreateTimestampOptions(bool disable_auto_compactions = false) {
+    Options options = CurrentOptions();
+    options.env = env_;
+    options.compaction_style = kCompactionStyleLevel;
+    options.num_levels = 4;
+    options.persist_user_defined_timestamps = true;
+    options.comparator = test::BytewiseComparatorWithU64TsWrapper();
+    options.disable_auto_compactions = disable_auto_compactions;
+    return options;
+  }
+
+  // Helper to write test data with alternating timestamps in a range
+  // Writes keys [start_key, end_key) with timestamps alternating between
+  // min_ts and max_ts
+  void WriteDataWithTimestampRange(int start_key, int end_key, uint64_t min_ts,
+                                   uint64_t max_ts) {
+    std::string ts_buf;
+    for (int i = start_key; i < end_key; i++) {
+      ts_buf.clear();
+      uint64_t ts = (i % 2 == 0) ? min_ts : max_ts;
+      PutFixed64(&ts_buf, ts);
+      ASSERT_OK(db_->Put(WriteOptions(), Key(i), ts_buf,
+                         "value" + std::to_string(i)));
+    }
+  }
+
+  // Helper to check if any file has the expected timestamp range
+  bool HasFileWithTimestampRange(uint64_t expected_min, uint64_t expected_max) {
+    auto file_timestamps = GetAllFileTimestamps();
+    for (const auto& [level, min_ts, max_ts] : file_timestamps) {
+      if (!min_ts.empty() && !max_ts.empty()) {
+        uint64_t file_min = DecodeFixed64(min_ts.data());
+        uint64_t file_max = DecodeFixed64(max_ts.data());
+        if (file_min == expected_min && file_max == expected_max) {
+          return true;
+        }
+      }
+    }
+    return false;
+  }
+
+  // Helper to verify data is readable with a given timestamp
+  void VerifyDataReadable(int key, const std::string& expected_value,
+                          uint64_t read_ts) {
+    std::string value;
+    std::string ts_buf;
+    PutFixed64(&ts_buf, read_ts);
+    ReadOptions read_opts;
+    Slice ts_slice(ts_buf);
+    read_opts.timestamp = &ts_slice;
+    ASSERT_OK(db_->Get(read_opts, Key(key), &value));
+    ASSERT_EQ(expected_value, value);
+  }
 };
 
 TEST_F(TimestampCompatibleCompactionTest, UserKeyCrossFileBoundary) {
@@ -344,6 +464,385 @@ TEST_F(TimestampCompatibleCompactionTest, EmptyCompactionOutput) {
   ASSERT_OK(db_->CompactRange(cro, nullptr, nullptr));
 }
 
+TEST_F(TimestampCompatibleCompactionTest, SeqnoZeroingWithUDT) {
+  // This test validates that seqno is only zeroed when the timestamp is older
+  // than full_history_ts_low_. Before the fix, seqno was incorrectly zeroed
+  // even when UDT was enabled but timestamp wasn't old enough.
+
+  Options options = CurrentOptions();
+  options.env = env_;
+  options.comparator = test::BytewiseComparatorWithU64TsWrapper();
+  options.disable_auto_compactions = true;
+  DestroyAndReopen(options);
+
+  // Track seqno zeroing events and which keys are zeroed
+  std::set<std::string> zeroed_keys;
+  SyncPoint::GetInstance()->SetCallBack(
+      "CompactionIterator::PrepareOutput:ZeroingSeq", [&](void* arg) {
+        auto* ikey = static_cast<ParsedInternalKey*>(arg);
+        ASSERT_EQ(0, ikey->sequence);
+        // Extract user key without timestamp (last 8 bytes)
+        Slice user_key_with_ts = ikey->user_key;
+        std::string user_key =
+            user_key_with_ts.ToString().substr(0, user_key_with_ts.size() - 8);
+        zeroed_keys.insert(user_key);
+      });
+  SyncPoint::GetInstance()->EnableProcessing();
+
+  // Case 1: Test that seqno is NOT zeroed when full_history_ts_low is not set
+  // Write a key with timestamp 100
+  std::string ts_str = Timestamp(100);
+  ASSERT_OK(db_->Put(WriteOptions(), "key1", ts_str, "value1"));
+  ASSERT_OK(Flush());
+
+  zeroed_keys.clear();
+  {
+    CompactRangeOptions cro;
+    cro.bottommost_level_compaction = BottommostLevelCompaction::kForce;
+    ASSERT_OK(db_->CompactRange(cro, nullptr, nullptr));
+  }
+  // With UDT enabled and no full_history_ts_low, seqno should NOT be zeroed
+  ASSERT_TRUE(zeroed_keys.empty());
+
+  // Case 2: Test that seqno IS zeroed when timestamp < full_history_ts_low
+  // Write a new key with timestamp 200
+  ts_str = Timestamp(200);
+  ASSERT_OK(db_->Put(WriteOptions(), "key2", ts_str, "value2"));
+  ASSERT_OK(Flush());
+
+  zeroed_keys.clear();
+  {
+    // Set full_history_ts_low to 300, so ts < 300 should be zeroed
+    std::string full_history_ts_low = Timestamp(300);
+    Slice ts_slice = full_history_ts_low;
+    CompactRangeOptions cro;
+    cro.full_history_ts_low = &ts_slice;
+    cro.bottommost_level_compaction = BottommostLevelCompaction::kForce;
+    ASSERT_OK(db_->CompactRange(cro, nullptr, nullptr));
+  }
+  // key1 (ts=100) and key2 (ts=200) both have ts < 300, so both should be
+  // zeroed
+  ASSERT_EQ(2u, zeroed_keys.size());
+  ASSERT_TRUE(zeroed_keys.count("key1") > 0);
+  ASSERT_TRUE(zeroed_keys.count("key2") > 0);
+
+  // Case 3: Write a new key with timestamp >= full_history_ts_low
+  // and verify it is NOT zeroed while old keys are re-zeroed
+  ts_str = Timestamp(500);
+  ASSERT_OK(db_->Put(WriteOptions(), "key3", ts_str, "value3"));
+  ASSERT_OK(Flush());
+
+  zeroed_keys.clear();
+  {
+    // Set full_history_ts_low to 400
+    // key1 (ts=100) and key2 (ts=200) have ts < 400, will be re-processed
+    // key3 (ts=500) has ts >= 400, should NOT be zeroed
+    std::string full_history_ts_low = Timestamp(400);
+    Slice ts_slice = full_history_ts_low;
+    CompactRangeOptions cro;
+    cro.full_history_ts_low = &ts_slice;
+    cro.bottommost_level_compaction = BottommostLevelCompaction::kForce;
+    ASSERT_OK(db_->CompactRange(cro, nullptr, nullptr));
+  }
+  // key3 should NOT appear in zeroed_keys since ts=500 >= 400
+  ASSERT_TRUE(zeroed_keys.count("key3") == 0);
+
+  SyncPoint::GetInstance()->DisableProcessing();
+  SyncPoint::GetInstance()->ClearAllCallBacks();
+
+  // Verify data is still readable
+  std::string value;
+  ts_str = Timestamp(600);
+  Slice read_ts = ts_str;
+  ReadOptions read_opts;
+  read_opts.timestamp = &read_ts;
+  ASSERT_OK(db_->Get(read_opts, "key1", &value));
+  ASSERT_EQ("value1", value);
+  ASSERT_OK(db_->Get(read_opts, "key2", &value));
+  ASSERT_EQ("value2", value);
+  ASSERT_OK(db_->Get(read_opts, "key3", &value));
+  ASSERT_EQ("value3", value);
+}
+
+// Test that files with max_timestamp >= full_history_ts_low are not marked
+// for bottommost compaction, which prevents infinite compaction loops.
+TEST_F(TimestampCompatibleCompactionTest,
+       BottommostCompactionRespectsFullHistoryTsLow) {
+  Options options = CreateTimestampOptions();
+  options.level0_file_num_compaction_trigger = 4;
+
+  DestroyAndReopen(options);
+
+  // Write some data with timestamps 100-199
+  std::string ts_buf;
+  for (int i = 0; i < 100; i++) {
+    ts_buf.clear();
+    PutFixed64(&ts_buf, 100 + i);
+    ASSERT_OK(
+        db_->Put(WriteOptions(), Key(i), ts_buf, "value" + std::to_string(i)));
+  }
+  ASSERT_OK(Flush());
+
+  // Compact to the bottommost level
+  CompactRangeOptions cro;
+  cro.bottommost_level_compaction = BottommostLevelCompaction::kForce;
+  ASSERT_OK(db_->CompactRange(cro, nullptr, nullptr));
+
+  // Set full_history_ts_low to 150 - files with max_ts >= 150 should NOT be
+  // marked for bottommost compaction since seqno cannot be zeroed
+  ts_buf.clear();
+  PutFixed64(&ts_buf, 150);
+  ASSERT_OK(db_->IncreaseFullHistoryTsLow(db_->DefaultColumnFamily(), ts_buf));
+
+  // Release a snapshot to potentially trigger bottommost file marking
+  // but files should NOT be marked because max_ts (199) >= full_history_ts_low
+  // (150)
+  const Snapshot* snap = db_->GetSnapshot();
+  db_->ReleaseSnapshot(snap);
+
+  // Wait for any scheduled compactions - should complete without infinite loop
+  // Use a reasonable timeout to detect infinite loops
+  WaitForCompactOptions wfc_options;
+  wfc_options.timeout = std::chrono::microseconds(5000000);  // 5 seconds
+  Status s = dbfull()->WaitForCompact(wfc_options);
+  // Should succeed without timeout (no infinite compaction loop)
+  ASSERT_TRUE(s.ok() || s.IsTimedOut());
+  if (s.IsTimedOut()) {
+    // If timeout, the fix is not working - this should not happen
+    FAIL() << "WaitForCompact timed out - possible infinite compaction loop";
+  }
+
+  // Now set full_history_ts_low beyond max timestamp in the file (200+)
+  // This should allow the file to be properly marked and compacted
+  ts_buf.clear();
+  PutFixed64(&ts_buf, 300);
+  ASSERT_OK(db_->IncreaseFullHistoryTsLow(db_->DefaultColumnFamily(), ts_buf));
+
+  // Trigger another snapshot release to potentially mark files
+  snap = db_->GetSnapshot();
+  db_->ReleaseSnapshot(snap);
+
+  // Now compaction should clean up the file.
+  ASSERT_OK(dbfull()->TEST_WaitForCompact());
+}
+
+// Test that files are NOT marked for bottommost compaction when UDT is enabled
+// and full_history_ts_low has never been set (empty).
+TEST_F(TimestampCompatibleCompactionTest,
+       BottommostCompactionSkipsWhenFullHistoryTsLowNotSet) {
+  Options options = CreateTimestampOptions();
+
+  DestroyAndReopen(options);
+
+  // Write some data with timestamps 100-199
+  std::string ts_buf;
+  for (int i = 0; i < 100; i++) {
+    ts_buf.clear();
+    PutFixed64(&ts_buf, 100 + i);
+    ASSERT_OK(
+        db_->Put(WriteOptions(), Key(i), ts_buf, "value" + std::to_string(i)));
+  }
+  ASSERT_OK(Flush());
+
+  // Compact to the bottommost level without setting full_history_ts_low
+  CompactRangeOptions cro;
+  cro.bottommost_level_compaction = BottommostLevelCompaction::kForce;
+  ASSERT_OK(db_->CompactRange(cro, nullptr, nullptr));
+
+  // Verify files have valid max_timestamp
+  auto file_timestamps = GetAllFileTimestamps();
+  ASSERT_GE(file_timestamps.size(), 1U);
+  for (const auto& [level, min_ts, max_ts] : file_timestamps) {
+    ASSERT_FALSE(max_ts.empty()) << "max_timestamp should not be empty";
+  }
+
+  // full_history_ts_low is NOT set (empty), so files should NOT be marked
+  // for bottommost compaction even after releasing a snapshot.
+  // This tests the branch: if (full_history_ts_low.empty()) { continue; }
+  const Snapshot* snap = db_->GetSnapshot();
+  db_->ReleaseSnapshot(snap);
+
+  // Wait for any scheduled compactions
+  ASSERT_OK(dbfull()->TEST_WaitForCompact());
+
+  // Now set full_history_ts_low to a value > max_timestamp (199) in the file
+  // This should allow the file to be properly marked and compacted
+  ts_buf.clear();
+  PutFixed64(&ts_buf, 300);
+  ASSERT_OK(db_->IncreaseFullHistoryTsLow(db_->DefaultColumnFamily(), ts_buf));
+
+  // Trigger another snapshot release to potentially mark files
+  snap = db_->GetSnapshot();
+  db_->ReleaseSnapshot(snap);
+
+  // Now compaction should be able to proceed since full_history_ts_low is set
+  ASSERT_OK(dbfull()->TEST_WaitForCompact());
+
+  // Verify data is still readable
+  VerifyDataReadable(0, "value0", 250);
+}
+
+// Test that ingested SST files created with UDT have their min/max timestamps
+// properly extracted from table properties and populated in FileMetaData.
+// This verifies the fix in external_sst_file_ingestion_job.cc that calls
+// ExtractTimestampFromTableProperties after creating FileMetaData.
+TEST_F(TimestampCompatibleCompactionTest,
+       IngestedFileTimestampsExtractedFromTableProperties) {
+  Options options = CreateTimestampOptions();
+
+  DestroyAndReopen(options);
+
+  // Create an SST file WITH timestamps using SstFileWriter
+  std::string sst_file = dbname_ + "/ingested_udt_file.sst";
+  const uint64_t kMinTs = 100;
+  const uint64_t kMaxTs = 200;
+
+  {
+    SstFileWriter sst_file_writer(EnvOptions(), options);
+    ASSERT_OK(sst_file_writer.Open(sst_file));
+
+    std::string ts_buf;
+    for (int i = 0; i < 10; i++) {
+      // Alternate between min and max timestamps
+      uint64_t ts = (i % 2 == 0) ? kMinTs : kMaxTs;
+      ts_buf.clear();
+      PutFixed64(&ts_buf, ts);
+      // SstFileWriter with UDT comparator requires key with timestamp
+      ASSERT_OK(
+          sst_file_writer.Put(Key(i), ts_buf, "value" + std::to_string(i)));
+    }
+    ASSERT_OK(sst_file_writer.Finish());
+  }
+
+  // Verify the SST file has timestamp properties before ingestion
+  {
+    std::unique_ptr<SstFileReader> reader(new SstFileReader(options));
+    ASSERT_OK(reader->Open(sst_file));
+    auto props = reader->GetTableProperties();
+    auto& user_collected = props->user_collected_properties;
+    ASSERT_TRUE(user_collected.find("rocksdb.timestamp_min") !=
+                user_collected.end())
+        << "SST file should have rocksdb.timestamp_min property";
+    ASSERT_TRUE(user_collected.find("rocksdb.timestamp_max") !=
+                user_collected.end())
+        << "SST file should have rocksdb.timestamp_max property";
+  }
+
+  // Ingest the SST file
+  IngestExternalFileOptions ifo;
+  ifo.move_files = false;
+  ASSERT_OK(db_->IngestExternalFile({sst_file}, ifo));
+
+  // Verify the ingested file has proper timestamps in FileMetaData
+  ASSERT_TRUE(HasFileWithTimestampRange(kMinTs, kMaxTs))
+      << "Ingested file should have min_timestamp=" << kMinTs
+      << " and max_timestamp=" << kMaxTs << " in FileMetaData";
+
+  // Verify timestamps persist after reopen
+  Reopen(options);
+
+  ASSERT_TRUE(HasFileWithTimestampRange(kMinTs, kMaxTs))
+      << "Ingested file timestamps should persist after reopen";
+
+  // Verify data is readable
+  VerifyDataReadable(0, "value0", kMaxTs);
+
+  // Clean up
+  ASSERT_OK(env_->DeleteFile(sst_file));
+}
+
+// Test that min/max timestamps are correctly tracked in FileMetaData and
+// persisted in the manifest during flush.
+TEST_F(TimestampCompatibleCompactionTest, TimestampRangePersistenceFlush) {
+  Options options = CreateTimestampOptions();
+
+  DestroyAndReopen(options);
+
+  // Expected timestamp range
+  const uint64_t kMinTs = 100;
+  const uint64_t kMaxTs = 200;
+
+  // Write data with specific timestamp range
+  WriteDataWithTimestampRange(0, 50, kMinTs, kMaxTs);
+  ASSERT_OK(Flush());
+
+  // First verify table properties have the timestamps
+  // (this confirms TimestampTablePropertiesCollector is working)
+  TablePropertiesCollection props;
+  ASSERT_OK(db_->GetPropertiesOfAllTables(&props));
+  ASSERT_EQ(1U, props.size());
+  for (const auto& item : props) {
+    auto& user_collected = item.second->user_collected_properties;
+    ASSERT_TRUE(user_collected.find("rocksdb.timestamp_min") !=
+                user_collected.end());
+    ASSERT_TRUE(user_collected.find("rocksdb.timestamp_max") !=
+                user_collected.end());
+    // Verify the collected timestamps match expected values
+    std::string collected_min_ts = user_collected.at("rocksdb.timestamp_min");
+    std::string collected_max_ts = user_collected.at("rocksdb.timestamp_max");
+    ASSERT_EQ(kMinTs, DecodeFixed64(collected_min_ts.data()));
+    ASSERT_EQ(kMaxTs, DecodeFixed64(collected_max_ts.data()));
+  }
+
+  // Verify FileMetaData timestamps and persistence through reopen
+  VerifyTimestampRangeWithPersistence(options, kMinTs, kMaxTs);
+
+  // Verify we can still read the data
+  VerifyDataReadable(0, "value0", kMaxTs);
+}
+
+// Test that min/max timestamps are correctly merged during compaction
+// and persisted in the manifest.
+TEST_F(TimestampCompatibleCompactionTest, TimestampRangePersistenceCompaction) {
+  Options options = CreateTimestampOptions(true /* disable_auto_compactions */);
+
+  DestroyAndReopen(options);
+
+  // Create multiple L0 files with different timestamp ranges
+  // File 1: timestamps 100-150
+  const uint64_t kFile1MinTs = 100;
+  const uint64_t kFile1MaxTs = 150;
+  WriteDataWithTimestampRange(0, 10, kFile1MinTs, kFile1MaxTs);
+  ASSERT_OK(Flush());
+
+  // File 2: timestamps 50-80 (earlier range)
+  const uint64_t kFile2MinTs = 50;
+  const uint64_t kFile2MaxTs = 80;
+  WriteDataWithTimestampRange(10, 20, kFile2MinTs, kFile2MaxTs);
+  ASSERT_OK(Flush());
+
+  // File 3: timestamps 200-300 (later range)
+  const uint64_t kFile3MinTs = 200;
+  const uint64_t kFile3MaxTs = 300;
+  WriteDataWithTimestampRange(20, 30, kFile3MinTs, kFile3MaxTs);
+  ASSERT_OK(Flush());
+
+  // Expected combined range: min=50, max=300
+  const uint64_t kExpectedMinTs = 50;
+  const uint64_t kExpectedMaxTs = 300;
+
+  // Verify we have 3 L0 files before compaction with valid timestamps
+  auto files_before = GetAllFileTimestamps();
+  ASSERT_EQ(3U, files_before.size());
+  for (const auto& [level, min_ts, max_ts] : files_before) {
+    ASSERT_EQ(0, level);  // All files should be in L0
+    ASSERT_FALSE(min_ts.empty());
+    ASSERT_FALSE(max_ts.empty());
+  }
+
+  // Trigger compaction
+  ASSERT_OK(db_->CompactRange(CompactRangeOptions(), nullptr, nullptr));
+  ASSERT_OK(dbfull()->TEST_WaitForCompact());
+
+  // Verify timestamp range and persistence through reopen
+  VerifyTimestampRangeWithPersistence(options, kExpectedMinTs, kExpectedMaxTs);
+
+  // Verify data is still readable
+  VerifyDataReadable(0, "value0", kExpectedMaxTs);
+  VerifyDataReadable(15, "value15", kExpectedMaxTs);
+  VerifyDataReadable(25, "value25", kExpectedMaxTs);
+}
+
 }  // namespace ROCKSDB_NAMESPACE
 
 int main(int argc, char** argv) {

db/db_write_test.cc

@@ -741,7 +741,7 @@ TEST_P(DBWriteTest, LockWALConcurrentRecursive) {
     ExternalSstFileInfo external_info;
     ASSERT_OK(sst_file_writer.Finish(&external_info));
   }
-  AcqRelAtomic<bool> parallel_ingest_completed{false};
+  Atomic<bool> parallel_ingest_completed{false};
   port::Thread parallel_ingest{[&]() {
     IngestExternalFileOptions ingest_opts;
     ingest_opts.move_files = true;  // faster than copy
@@ -750,7 +750,7 @@ TEST_P(DBWriteTest, LockWALConcurrentRecursive) {
     parallel_ingest_completed.Store(true);
   }};
 
-  AcqRelAtomic<bool> flush_completed{false};
+  Atomic<bool> flush_completed{false};
   port::Thread parallel_flush{[&]() {
     FlushOptions flush_opts;
     // NB: Flush with wait=false case is tested above in LockWALInEffect
@@ -762,7 +762,7 @@ TEST_P(DBWriteTest, LockWALConcurrentRecursive) {
     flush_completed.Store(true);
   }};
 
-  AcqRelAtomic<bool> parallel_put_completed{false};
+  Atomic<bool> parallel_put_completed{false};
   port::Thread parallel_put{[&]() {
     // This can make certain failure scenarios more likely:
     //   sleep(1);

db/dbformat.h

@@ -72,12 +72,6 @@ enum ValueType : unsigned char {
   kTypeColumnFamilyWideColumnEntity = 0x17,     // WAL only
   kTypeValuePreferredSeqno = 0x18,              // Value with a unix write time
   kTypeColumnFamilyValuePreferredSeqno = 0x19,  // WAL only
-  // Placeholder value type for legacy SST files with incorrectly persisted
-  // file boundaries. Prior to the fix, TruncatedRangeDelIterator assigned
-  // kTypeMaxValid to truncated range deletion keys, which was then
-  // incorrectly persisted to SST file metadata. This placeholder value allows
-  // reading such legacy files for without using kTypeMaxValid.
-  kTypeTruncatedRangeDeletionSentinel = 0x1A,
   kTypeMaxValid,    // Should be after the last valid type, only used for
                     // validation
   kMaxValue = 0x7F  // Not used for storing records.
@@ -124,11 +118,10 @@ inline bool IsValueType(ValueType t) {
 
 // Checks whether a type is from user operation
 // kTypeRangeDeletion is in meta block so this API is separated from above
-// kTypeTruncatedRangeDeletionSentinel is for legacy files with incorrectly
-// persisted file boundaries.
+// kTypeMaxValid can be from keys generated by
+// TruncatedRangeDelIterator::start_key()
 inline bool IsExtendedValueType(ValueType t) {
-  return IsValueType(t) || t == kTypeRangeDeletion ||
-         t == kTypeTruncatedRangeDeletionSentinel;
+  return IsValueType(t) || t == kTypeRangeDeletion || t == kTypeMaxValid;
 }
 
 // We leave eight bits empty at the bottom so a type and sequence#
@@ -187,7 +180,8 @@ inline size_t InternalKeyEncodingLength(const ParsedInternalKey& key) {
 // Pack a sequence number and a ValueType into a uint64_t
 inline uint64_t PackSequenceAndType(uint64_t seq, ValueType t) {
   assert(seq <= kMaxSequenceNumber);
-  assert(IsExtendedValueType(t));
+  // kTypeMaxValid is used in TruncatedRangeDelIterator, see its constructor.
+  assert(IsExtendedValueType(t) || t == kTypeMaxValid);
   return (seq << 8) | t;
 }
 

db/experimental.cc

@@ -158,15 +158,17 @@ Status UpdateManifestForFilesState(
               // Current state inconsistent with manifest
               ++files_updated;
               edit.DeleteFile(level, number);
-              edit.AddFile(
-                  level, number, lf->fd.GetPathId(), lf->fd.GetFileSize(),
-                  lf->smallest, lf->largest, lf->fd.smallest_seqno,
-                  lf->fd.largest_seqno, lf->marked_for_compaction, temp,
-                  lf->oldest_blob_file_number, lf->oldest_ancester_time,
-                  lf->file_creation_time, lf->epoch_number, lf->file_checksum,
-                  lf->file_checksum_func_name, lf->unique_id,
-                  lf->compensated_range_deletion_size, lf->tail_size,
-                  lf->user_defined_timestamps_persisted);
+              edit.AddFile(level, lf->fd.GetNumber(), lf->fd.GetPathId(),
+                           lf->fd.GetFileSize(), lf->smallest, lf->largest,
+                           lf->fd.smallest_seqno, lf->fd.largest_seqno,
+                           lf->marked_for_compaction, temp,
+                           lf->oldest_blob_file_number,
+                           lf->oldest_ancester_time, lf->file_creation_time,
+                           lf->epoch_number, lf->file_checksum,
+                           lf->file_checksum_func_name, lf->unique_id,
+                           lf->compensated_range_deletion_size, lf->tail_size,
+                           lf->user_defined_timestamps_persisted,
+                           lf->min_timestamp, lf->max_timestamp);
             }
           }
         } else {

db/external_sst_file_ingestion_job.cc

@@ -11,6 +11,7 @@
 #include <unordered_set>
 #include <vector>
 
+#include "db/builder.h"
 #include "db/db_impl/db_impl.h"
 #include "db/version_edit.h"
 #include "file/file_util.h"
@@ -299,6 +300,18 @@ Status ExternalSstFileIngestionJob::Prepare(
         // ingestion.
         // TODO: plumb Env::IOActivity
         ReadOptions ro;
+        // Pass user-provided checksums through FileOptions when available.
+        // The caller may not have provided checksums at all (empty vectors),
+        // so we guard with a bounds check.
+        FileOptions fopts;
+        if (i < files_checksums.size()) {
+          fopts.file_checksum = files_checksums[i];
+        }
+        if (i < files_checksum_func_names.size()) {
+          fopts.file_checksum_func_name = files_checksum_func_names[i];
+        } else {
+          fopts.file_checksum_func_name = kNoFileChecksumFuncName;
+        }
         IOStatus io_s = GenerateOneFileChecksum(
             fs_.get(), files_to_ingest_[i].internal_file_path,
             db_options_.file_checksum_gen_factory.get(),
@@ -307,7 +320,7 @@ Status ExternalSstFileIngestionJob::Prepare(
             ingestion_options_.verify_checksums_readahead_size,
             db_options_.allow_mmap_reads, io_tracer_,
             db_options_.rate_limiter.get(), ro, db_options_.stats,
-            db_options_.clock);
+            db_options_.clock, fopts);
         if (!io_s.ok()) {
           status = io_s;
           ROCKS_LOG_WARN(db_options_.info_log,
@@ -699,9 +712,13 @@ Status ExternalSstFileIngestionJob::AssignLevelsForOneBatch(
             ? kReservedEpochNumberForFileIngestedBehind
             : 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);
+        tail_size, file->user_defined_timestamps_persisted, "", "");
     f_metadata.temperature = file->file_temperature;
     f_metadata.marked_for_compaction = marked_for_compaction;
+    // Extract min/max timestamps from table properties for UDT support.
+    // This ensures ingested files have proper timestamp ranges in FileMetaData,
+    // similar to files created by flush and compaction.
+    ExtractTimestampFromTableProperties(file->table_properties, &f_metadata);
     edit_.AddFile(file->picked_level, f_metadata);
 
     *batch_uppermost_level =
@@ -1477,13 +1494,15 @@ IOStatus ExternalSstFileIngestionJob::GenerateChecksumForIngestedFile(
   // TODO: rate limit file reads for checksum calculation during file ingestion.
   // TODO: plumb Env::IOActivity
   ReadOptions ro;
+  FileOptions gen_fopts;
+  gen_fopts.file_checksum_func_name = kNoFileChecksumFuncName;
   IOStatus io_s = GenerateOneFileChecksum(
       fs_.get(), file_to_ingest->internal_file_path,
       db_options_.file_checksum_gen_factory.get(), requested_checksum_func_name,
       &file_checksum, &file_checksum_func_name,
       ingestion_options_.verify_checksums_readahead_size,
       db_options_.allow_mmap_reads, io_tracer_, db_options_.rate_limiter.get(),
-      ro, db_options_.stats, db_options_.clock);
+      ro, db_options_.stats, db_options_.clock, gen_fopts);
   if (!io_s.ok()) {
     ROCKS_LOG_WARN(
         db_options_.info_log, "Failed to generate checksum for %s: %s",

db/flush_job.cc

@@ -1095,7 +1095,8 @@ Status FlushJob::WriteLevel0Table() {
                    meta_.file_creation_time, meta_.epoch_number,
                    meta_.file_checksum, meta_.file_checksum_func_name,
                    meta_.unique_id, meta_.compensated_range_deletion_size,
-                   meta_.tail_size, meta_.user_defined_timestamps_persisted);
+                   meta_.tail_size, meta_.user_defined_timestamps_persisted,
+                   meta_.min_timestamp, meta_.max_timestamp);
     edit_->SetBlobFileAdditions(std::move(blob_file_additions));
   }
   // Piggyback FlushJobInfo on the first first flushed memtable.

db/import_column_family_job.cc

@@ -310,8 +310,10 @@ Status ImportColumnFamilyJob::GetIngestedFileInfo(
   std::unique_ptr<FSRandomAccessFile> sst_file;
   std::unique_ptr<RandomAccessFileReader> sst_file_reader;
 
-  status =
-      fs_->NewRandomAccessFile(external_file, env_options_, &sst_file, nullptr);
+  FileOptions fo{env_options_};
+  fo.file_checksum = file_meta.file_checksum;
+  fo.file_checksum_func_name = file_meta.file_checksum_func_name;
+  status = fs_->NewRandomAccessFile(external_file, fo, &sst_file, nullptr);
   if (!status.ok()) {
     return status;
   }

db/range_del_aggregator.cc

@@ -36,6 +36,7 @@ TruncatedRangeDelIterator::TruncatedRangeDelIterator(
     Status pik_status = ParseInternalKey(smallest->Encode(), &parsed_smallest,
                                          false /* log_err_key */);  // TODO
     pik_status.PermitUncheckedError();
+    parsed_smallest.type = kTypeMaxValid;
     assert(pik_status.ok());
     smallest_ = &parsed_smallest;
   }
@@ -70,6 +71,9 @@ TruncatedRangeDelIterator::TruncatedRangeDelIterator(
       // the truncated end key can cover the largest key in this sstable, reduce
       // its sequence number by 1.
       parsed_largest.sequence -= 1;
+      // This line is not needed for correctness, but it ensures that the
+      // truncated end key is not covering keys from the next SST file.
+      parsed_largest.type = kTypeMaxValid;
     }
     largest_ = &parsed_largest;
   }

db/range_del_aggregator_test.cc

@@ -89,9 +89,7 @@ void VerifyIterator(
   for (size_t i = 0; i < expected_range_dels.size(); i++, iter->Next()) {
     ASSERT_TRUE(iter->Valid());
     EXPECT_EQ(0, icmp.Compare(iter->start_key(), expected_range_dels[i].start));
-    EXPECT_EQ(0, icmp.Compare(iter->end_key(), expected_range_dels[i].end))
-        << iter->end_key().DebugString(false, false) << " "
-        << expected_range_dels[i].end.DebugString(false, false);
+    EXPECT_EQ(0, icmp.Compare(iter->end_key(), expected_range_dels[i].end));
     EXPECT_EQ(expected_range_dels[i].seq, iter->seq());
   }
   EXPECT_FALSE(iter->Valid());
@@ -307,28 +305,28 @@ TEST_F(RangeDelAggregatorTest, TruncatedIterPartiallyCutTombstones) {
 
   VerifyIterator(
       &iter, bytewise_icmp,
-      {{InternalValue("d", 7, kTypeValue), UncutEndpoint("e"), 10},
+      {{InternalValue("d", 7, kTypeMaxValid), UncutEndpoint("e"), 10},
        {InternalValue("e", 8, kTypeRangeDeletion), UncutEndpoint("g"), 8},
        {InternalValue("j", 4, kTypeRangeDeletion),
-        InternalValue("m", 8, kTypeValue), 4}});
+        InternalValue("m", 8, kTypeMaxValid), 4}});
 
   VerifySeek(
       &iter, bytewise_icmp,
-      {{"d", InternalValue("d", 7, kTypeValue), UncutEndpoint("e"), 10},
+      {{"d", InternalValue("d", 7, kTypeMaxValid), UncutEndpoint("e"), 10},
        {"e", InternalValue("e", 8, kTypeRangeDeletion), UncutEndpoint("g"), 8},
        {"ia", InternalValue("j", 4, kTypeRangeDeletion),
-        InternalValue("m", 8, kTypeValue), 4, false /* invalid */},
+        InternalValue("m", 8, kTypeMaxValid), 4, false /* invalid */},
        {"n", InternalValue("", 0, kTypeRangeDeletion), UncutEndpoint(""), 0,
         true /* invalid */},
-       {"", InternalValue("d", 7, kTypeValue), UncutEndpoint("e"), 10}});
+       {"", InternalValue("d", 7, kTypeMaxValid), UncutEndpoint("e"), 10}});
 
   VerifySeekForPrev(
       &iter, bytewise_icmp,
-      {{"d", InternalValue("d", 7, kTypeValue), UncutEndpoint("e"), 10},
+      {{"d", InternalValue("d", 7, kTypeMaxValid), UncutEndpoint("e"), 10},
        {"e", InternalValue("e", 8, kTypeRangeDeletion), UncutEndpoint("g"), 8},
        {"ia", InternalValue("e", 8, kTypeRangeDeletion), UncutEndpoint("g"), 8},
        {"n", InternalValue("j", 4, kTypeRangeDeletion),
-        InternalValue("m", 8, kTypeValue), 4, false /* invalid */},
+        InternalValue("m", 8, kTypeMaxValid), 4, false /* invalid */},
        {"", InternalValue("", 0, kTypeRangeDeletion), UncutEndpoint(""), 0,
         true /* invalid */}});
 }
@@ -347,21 +345,23 @@ TEST_F(RangeDelAggregatorTest, TruncatedIterFullyCutTombstones) {
   TruncatedRangeDelIterator iter(std::move(input_iter), &bytewise_icmp,
                                  &smallest, &largest);
 
-  VerifyIterator(&iter, bytewise_icmp,
-                 {{InternalValue("f", 7, kTypeValue), UncutEndpoint("g"), 8}});
+  VerifyIterator(
+      &iter, bytewise_icmp,
+      {{InternalValue("f", 7, kTypeMaxValid), UncutEndpoint("g"), 8}});
 
-  VerifySeek(&iter, bytewise_icmp,
-             {{"d", InternalValue("f", 7, kTypeValue), UncutEndpoint("g"), 8},
-              {"f", InternalValue("f", 7, kTypeValue), UncutEndpoint("g"), 8},
-              {"j", InternalValue("", 0, kTypeRangeDeletion), UncutEndpoint(""),
-               0, true /* invalid */}});
+  VerifySeek(
+      &iter, bytewise_icmp,
+      {{"d", InternalValue("f", 7, kTypeMaxValid), UncutEndpoint("g"), 8},
+       {"f", InternalValue("f", 7, kTypeMaxValid), UncutEndpoint("g"), 8},
+       {"j", InternalValue("", 0, kTypeRangeDeletion), UncutEndpoint(""), 0,
+        true /* invalid */}});
 
   VerifySeekForPrev(
       &iter, bytewise_icmp,
       {{"d", InternalValue("", 0, kTypeRangeDeletion), UncutEndpoint(""), 0,
         true /* invalid */},
-       {"f", InternalValue("f", 7, kTypeValue), UncutEndpoint("g"), 8},
-       {"j", InternalValue("f", 7, kTypeValue), UncutEndpoint("g"), 8}});
+       {"f", InternalValue("f", 7, kTypeMaxValid), UncutEndpoint("g"), 8},
+       {"j", InternalValue("f", 7, kTypeMaxValid), UncutEndpoint("g"), 8}});
 }
 
 TEST_F(RangeDelAggregatorTest, SingleIterInAggregator) {

db/repair.cc

@@ -704,17 +704,17 @@ class Repairer {
       VersionEdit dummy_edit;
       for (const auto* table : cf_id_and_tables.second) {
         // TODO(opt): separate out into multiple levels
+        const auto& meta = table->meta;
         dummy_edit.AddFile(
-            0, table->meta.fd.GetNumber(), table->meta.fd.GetPathId(),
-            table->meta.fd.GetFileSize(), table->meta.smallest,
-            table->meta.largest, table->meta.fd.smallest_seqno,
-            table->meta.fd.largest_seqno, table->meta.marked_for_compaction,
-            table->meta.temperature, table->meta.oldest_blob_file_number,
-            table->meta.oldest_ancester_time, table->meta.file_creation_time,
-            table->meta.epoch_number, table->meta.file_checksum,
-            table->meta.file_checksum_func_name, table->meta.unique_id,
-            table->meta.compensated_range_deletion_size, table->meta.tail_size,
-            table->meta.user_defined_timestamps_persisted);
+            0, meta.fd.GetNumber(), meta.fd.GetPathId(), meta.fd.GetFileSize(),
+            meta.smallest, meta.largest, meta.fd.smallest_seqno,
+            meta.fd.largest_seqno, meta.marked_for_compaction, meta.temperature,
+            meta.oldest_blob_file_number, meta.oldest_ancester_time,
+            meta.file_creation_time, meta.epoch_number, meta.file_checksum,
+            meta.file_checksum_func_name, meta.unique_id,
+            meta.compensated_range_deletion_size, meta.tail_size,
+            meta.user_defined_timestamps_persisted, meta.min_timestamp,
+            meta.max_timestamp);
       }
       s = dummy_version_builder.Apply(&dummy_edit);
       if (s.ok()) {

db/table_cache.cc

@@ -98,6 +98,8 @@ Status TableCache::GetTableReader(
   std::unique_ptr<FSRandomAccessFile> file;
   FileOptions fopts = file_options;
   fopts.temperature = file_temperature;
+  fopts.file_checksum = file_meta.file_checksum;
+  fopts.file_checksum_func_name = file_meta.file_checksum_func_name;
   Status s = PrepareIOFromReadOptions(ro, ioptions_.clock, fopts.io_options);
   TEST_SYNC_POINT_CALLBACK("TableCache::GetTableReader:BeforeOpenFile",
                            const_cast<Status*>(&s));
@@ -113,8 +115,7 @@ Status TableCache::GetTableReader(
     Status temp_s =
         PrepareIOFromReadOptions(ro, ioptions_.clock, fopts.io_options);
     if (temp_s.ok()) {
-      temp_s = ioptions_.fs->NewRandomAccessFile(fname, file_options, &file,
-                                                 nullptr);
+      temp_s = ioptions_.fs->NewRandomAccessFile(fname, fopts, &file, nullptr);
     }
     if (temp_s.ok()) {
       RecordTick(ioptions_.stats, NO_FILE_OPENS);

db/version_builder_test.cc

@@ -76,7 +76,8 @@ class VersionBuilderTest : public testing::Test {
         oldest_blob_file_number, kUnknownOldestAncesterTime,
         kUnknownFileCreationTime, epoch_number, kUnknownFileChecksum,
         kUnknownFileChecksumFuncName, kNullUniqueId64x2, 0, 0,
-        /* user_defined_timestamps_persisted */ true);
+        /* user_defined_timestamps_persisted */ true, /* min timestamp */ "",
+        /* max timestamp */ "");
     f->compensated_file_size = file_size;
     f->num_entries = num_entries;
     f->num_deletions = num_deletions;

db/version_edit.cc

@@ -30,7 +30,6 @@ uint64_t PackFileNumberAndPathId(uint64_t number, uint64_t path_id) {
 Status FileMetaData::UpdateBoundaries(const Slice& key, const Slice& value,
                                       SequenceNumber seqno,
                                       ValueType value_type) {
-  assert(value_type < kTypeMaxValid);
   if (value_type == kTypeBlobIndex) {
     BlobIndex blob_index;
     const Status s = blob_index.DecodeFrom(value);
@@ -305,6 +304,15 @@ void VersionEdit::EncodeToNewFile4(const FileMetaData& f, int level,
     char p = static_cast<char>(0);
     PutLengthPrefixedSlice(dst, Slice(&p, 1));
   }
+  // Encode min/max timestamp if they are non-empty
+  if (!f.min_timestamp.empty()) {
+    PutVarint32(dst, NewFileCustomTag::kMinTimestamp);
+    PutLengthPrefixedSlice(dst, Slice(f.min_timestamp));
+  }
+  if (!f.max_timestamp.empty()) {
+    PutVarint32(dst, NewFileCustomTag::kMaxTimestamp);
+    PutLengthPrefixedSlice(dst, Slice(f.max_timestamp));
+  }
   TEST_SYNC_POINT_CALLBACK("VersionEdit::EncodeTo:NewFile4:CustomizeFields",
                            dst);
 
@@ -444,6 +452,12 @@ const char* VersionEdit::DecodeNewFile4From(Slice* input, int& max_level,
           }
           f.user_defined_timestamps_persisted = (field[0] == 1);
           break;
+        case kMinTimestamp:
+          f.min_timestamp = field.ToString();
+          break;
+        case kMaxTimestamp:
+          f.max_timestamp = field.ToString();
+          break;
         default:
           if ((custom_tag & kCustomTagNonSafeIgnoreMask) != 0) {
             // Should not proceed if cannot understand it

db/version_edit.h

@@ -276,6 +276,14 @@ struct FileMetaData {
   // false, it's explicitly written to Manifest.
   bool user_defined_timestamps_persisted = true;
 
+  // Minimum user-defined timestamp in the file. Empty if no UDT or unknown.
+  // This is populated from the table properties "rocksdb.timestamp_min".
+  std::string min_timestamp;
+
+  // Maximum user-defined timestamp in the file. Empty if no UDT or unknown.
+  // This is populated from the table properties "rocksdb.timestamp_max".
+  std::string max_timestamp;
+
   FileMetaData() = default;
 
   FileMetaData(uint64_t file, uint32_t file_path_id, uint64_t file_size,
@@ -288,7 +296,9 @@ struct FileMetaData {
                const std::string& _file_checksum_func_name,
                UniqueId64x2 _unique_id,
                const uint64_t _compensated_range_deletion_size,
-               uint64_t _tail_size, bool _user_defined_timestamps_persisted)
+               uint64_t _tail_size, bool _user_defined_timestamps_persisted,
+               const std::string& _min_timestamp,
+               const std::string& _max_timestamp)
       : fd(file, file_path_id, file_size, smallest_seq, largest_seq),
         smallest(smallest_key),
         largest(largest_key),
@@ -303,7 +313,9 @@ struct FileMetaData {
         file_checksum_func_name(_file_checksum_func_name),
         unique_id(std::move(_unique_id)),
         tail_size(_tail_size),
-        user_defined_timestamps_persisted(_user_defined_timestamps_persisted) {
+        user_defined_timestamps_persisted(_user_defined_timestamps_persisted),
+        min_timestamp(_min_timestamp),
+        max_timestamp(_max_timestamp) {
     TEST_SYNC_POINT_CALLBACK("FileMetaData::FileMetaData", this);
   }
 
@@ -317,9 +329,6 @@ struct FileMetaData {
   void UpdateBoundariesForRange(const InternalKey& start,
                                 const InternalKey& end, SequenceNumber seqno,
                                 const InternalKeyComparator& icmp) {
-    assert(ExtractValueType(start.Encode()) < kTypeMaxValid);
-    assert(ExtractValueType(end.Encode()) < kTypeMaxValid);
-
     if (smallest.size() == 0 || icmp.Compare(start, smallest) < 0) {
       smallest = start;
     }
@@ -389,7 +398,8 @@ struct FileMetaData {
     usage += sizeof(*this);
 #endif  // ROCKSDB_MALLOC_USABLE_SIZE
     usage += smallest.size() + largest.size() + file_checksum.size() +
-             file_checksum_func_name.size();
+             file_checksum_func_name.size() + min_timestamp.size() +
+             max_timestamp.size();
     return usage;
   }
 
@@ -560,16 +570,19 @@ struct SubcompactionProgress {
       Slice key_slice(next_internal_key_to_compact);
       if (ParseInternalKey(key_slice, &parsed_key, false /* log_err_key */)
               .ok()) {
-        oss << "user_key=\"" << parsed_key.user_key.ToString(false /* hex */)
-            << "\" (hex:" << parsed_key.user_key.ToString(true /* hex */)
-            << ")";
+        oss << "user_key(hex)=" << parsed_key.user_key.ToString(true /* hex */);
         oss << ", seq=";
         if (parsed_key.sequence == kMaxSequenceNumber) {
           oss << "kMaxSequenceNumber";
         } else {
           oss << parsed_key.sequence;
         }
-        oss << ", type=" << static_cast<int>(parsed_key.type);
+        oss << ", type=";
+        if (parsed_key.type == kValueTypeForSeek) {
+          oss << "kValueTypeForSeek";
+        } else {
+          oss << static_cast<int>(parsed_key.type);
+        }
       } else {
         oss << "raw=" << key_slice.ToString(true /* hex */);
       }
@@ -737,17 +750,19 @@ class VersionEdit {
                const std::string& file_checksum_func_name,
                const UniqueId64x2& unique_id,
                const uint64_t compensated_range_deletion_size,
-               uint64_t tail_size, bool user_defined_timestamps_persisted) {
+               uint64_t tail_size, bool user_defined_timestamps_persisted,
+               const std::string& min_timestamp = "",
+               const std::string& max_timestamp = "") {
     assert(smallest_seqno <= largest_seqno);
     new_files_.emplace_back(
         level,
-        FileMetaData(file, file_path_id, file_size, smallest, largest,
-                     smallest_seqno, largest_seqno, marked_for_compaction,
-                     temperature, oldest_blob_file_number, oldest_ancester_time,
-                     file_creation_time, epoch_number, file_checksum,
-                     file_checksum_func_name, unique_id,
-                     compensated_range_deletion_size, tail_size,
-                     user_defined_timestamps_persisted));
+        FileMetaData(
+            file, file_path_id, file_size, smallest, largest, smallest_seqno,
+            largest_seqno, marked_for_compaction, temperature,
+            oldest_blob_file_number, oldest_ancester_time, file_creation_time,
+            epoch_number, file_checksum, file_checksum_func_name, unique_id,
+            compensated_range_deletion_size, tail_size,
+            user_defined_timestamps_persisted, min_timestamp, max_timestamp));
     files_to_quarantine_.push_back(file);
     if (!HasLastSequence() || largest_seqno > GetLastSequence()) {
       SetLastSequence(largest_seqno);

db/version_set.cc

@@ -1820,8 +1820,10 @@ Status Version::GetTableProperties(const ReadOptions& read_options,
     file_name = TableFileName(ioptions.cf_paths, file_meta->fd.GetNumber(),
                               file_meta->fd.GetPathId());
   }
-  s = ioptions.fs->NewRandomAccessFile(file_name, file_options_, &file,
-                                       nullptr);
+  FileOptions fopts = file_options_;
+  fopts.file_checksum = file_meta->file_checksum;
+  fopts.file_checksum_func_name = file_meta->file_checksum_func_name;
+  s = ioptions.fs->NewRandomAccessFile(file_name, fopts, &file, nullptr);
   if (!s.ok()) {
     return s;
   }
@@ -3444,7 +3446,7 @@ bool Version::MaybeInitializeFileMetaData(const ReadOptions& read_options,
   // Ensure new invariants on old files
   file_meta->num_deletions =
       std::max(tp->num_deletions, tp->num_range_deletions);
-  file_meta->num_entries = std::max(tp->num_entries, tp->num_deletions);
+  file_meta->num_entries = std::max(tp->num_entries, file_meta->num_deletions);
   return true;
 }
 
@@ -3737,7 +3739,8 @@ bool ShouldChangeFileTemperature(const ImmutableOptions& ioptions,
 
 void VersionStorageInfo::ComputeCompactionScore(
     const ImmutableOptions& immutable_options,
-    const MutableCFOptions& mutable_cf_options) {
+    const MutableCFOptions& mutable_cf_options,
+    const std::string& full_history_ts_low) {
   double total_downcompact_bytes = 0.0;
   // Historically, score is defined as actual bytes in a level divided by
   // the level's target size, and 1.0 is the threshold for triggering
@@ -3936,7 +3939,8 @@ void VersionStorageInfo::ComputeCompactionScore(
   ComputeFilesMarkedForCompaction(max_output_level);
   ComputeBottommostFilesMarkedForCompaction(
       immutable_options.cf_allow_ingest_behind ||
-      immutable_options.allow_ingest_behind);
+          immutable_options.allow_ingest_behind,
+      immutable_options.user_comparator, full_history_ts_low);
   ComputeExpiredTtlFiles(immutable_options, mutable_cf_options.ttl);
   ComputeFilesMarkedForPeriodicCompaction(
       immutable_options, mutable_cf_options.periodic_compaction_seconds,
@@ -4527,17 +4531,20 @@ void VersionStorageInfo::GenerateFileLocationIndex() {
   }
 }
 
-void VersionStorageInfo::UpdateOldestSnapshot(SequenceNumber seqnum,
-                                              bool allow_ingest_behind) {
+void VersionStorageInfo::UpdateOldestSnapshot(
+    SequenceNumber seqnum, bool allow_ingest_behind, const Comparator* ucmp,
+    const std::string& full_history_ts_low) {
   assert(seqnum >= oldest_snapshot_seqnum_);
   oldest_snapshot_seqnum_ = seqnum;
   if (oldest_snapshot_seqnum_ > bottommost_files_mark_threshold_) {
-    ComputeBottommostFilesMarkedForCompaction(allow_ingest_behind);
+    ComputeBottommostFilesMarkedForCompaction(allow_ingest_behind, ucmp,
+                                              full_history_ts_low);
   }
 }
 
 void VersionStorageInfo::ComputeBottommostFilesMarkedForCompaction(
-    bool allow_ingest_behind) {
+    bool allow_ingest_behind, const Comparator* ucmp,
+    const std::string& full_history_ts_low) {
   bottommost_files_marked_for_compaction_.clear();
   bottommost_files_mark_threshold_ = kMaxSequenceNumber;
   if (allow_ingest_behind) {
@@ -4558,12 +4565,39 @@ void VersionStorageInfo::ComputeBottommostFilesMarkedForCompaction(
         current_time - static_cast<int64_t>(bottommost_file_compaction_delay_);
   }
 
+  // For UDT, we need to check if the file's max timestamp is below
+  // full_history_ts_low. If not, the compaction won't be able to collapse the
+  // timestamp to clean up the tombstone , so marking the file would be futile
+  // and could cause an infinite compaction loop.
+  const bool has_udt = ucmp && ucmp->timestamp_size() > 0;
+
   for (auto& level_and_file : bottommost_files_) {
     if (!level_and_file.second->being_compacted &&
         level_and_file.second->fd.largest_seqno != 0) {
       // largest_seqno might be nonzero due to containing the final key in an
       // earlier compaction, whose seqnum we didn't zero out.
       if (level_and_file.second->fd.largest_seqno < oldest_snapshot_seqnum_) {
+        if (has_udt) {
+          const std::string& max_ts = level_and_file.second->max_timestamp;
+          // If max_timestamp is empty, the file could come from very old
+          // version which does not have timestamp. In that case, we should pick
+          // the file for compaction. After compaction, the file will have
+          // max_timestamp set propertly.
+          if (!max_ts.empty()) {
+            // If full_history_ts_low is empty, it means it was never set, which
+            // means its value is 0. Therefore, it would be always smaller than
+            // max_timestamp
+            if (full_history_ts_low.empty()) {
+              continue;
+            }
+            // If max timestamp >= full_history_ts_low, skip this file
+            if (ucmp->CompareTimestamp(Slice(max_ts), full_history_ts_low) >=
+                0) {
+              continue;
+            }
+          }
+        }
+
         if (!needs_delay) {
           bottommost_files_marked_for_compaction_.push_back(level_and_file);
         } else if (creation_time_ub > 0) {
@@ -5571,6 +5605,15 @@ void VersionSet::Reset() {
   if (column_family_set_) {
     WriteBufferManager* wbm = column_family_set_->write_buffer_manager();
     WriteController* wc = column_family_set_->write_controller();
+
+    // Clear TableCache to prevent use-after-free: Reset() deletes old
+    // ColumnFamilySet but reuses table_cache_, which may contain
+    // BlockBasedTable entries with dangling references to deleted CFD's
+    // ioptions.
+    if (table_cache_) {
+      table_cache_->EraseUnRefEntries();
+    }
+
     // db_id becomes the source of truth after DBImpl::Recover():
     // https://github.com/facebook/rocksdb/blob/v7.3.1/db/db_impl/db_impl_open.cc#L527
     // Note: we may not be able to recover db_id from MANIFEST if
@@ -5630,7 +5673,8 @@ void VersionSet::AppendVersion(ColumnFamilyData* column_family_data,
   // compute new compaction score
   v->storage_info()->ComputeCompactionScore(
       column_family_data->ioptions(),
-      column_family_data->GetLatestMutableCFOptions());
+      column_family_data->GetLatestMutableCFOptions(),
+      column_family_data->GetFullHistoryTsLow());
 
   // Mark v finalized
   v->storage_info_.SetFinalized();
@@ -7093,7 +7137,6 @@ Status VersionSet::WriteCurrentStateToManifest(
 
         for (const auto& f : level_files) {
           assert(f);
-
           edit.AddFile(level, f->fd.GetNumber(), f->fd.GetPathId(),
                        f->fd.GetFileSize(), f->smallest, f->largest,
                        f->fd.smallest_seqno, f->fd.largest_seqno,
@@ -7102,7 +7145,8 @@ Status VersionSet::WriteCurrentStateToManifest(
                        f->file_creation_time, f->epoch_number, f->file_checksum,
                        f->file_checksum_func_name, f->unique_id,
                        f->compensated_range_deletion_size, f->tail_size,
-                       f->user_defined_timestamps_persisted);
+                       f->user_defined_timestamps_persisted, f->min_timestamp,
+                       f->max_timestamp);
         }
       }
 

db/version_set.h

@@ -200,7 +200,8 @@ class VersionStorageInfo {
   // REQUIRES: db_mutex held!!
   // TODO find a better way to pass compaction_options_fifo.
   void ComputeCompactionScore(const ImmutableOptions& immutable_options,
-                              const MutableCFOptions& mutable_cf_options);
+                              const MutableCFOptions& mutable_cf_options,
+                              const std::string& full_history_ts_low);
 
   // Estimate est_comp_needed_bytes_
   void EstimateCompactionBytesNeeded(
@@ -230,8 +231,15 @@ class VersionStorageInfo {
   // oldest snapshot changes as that is when bottom-level files can become
   // eligible for compaction.
   //
+  // For columns with User Defined Timestamps (UDT), also checks that the
+  // file's largest timestamp is below full_history_ts_low before marking,
+  // since compaction can only collapse timestamp when it is below this
+  // threshold.
+  //
   // REQUIRES: DB mutex held
-  void ComputeBottommostFilesMarkedForCompaction(bool allow_ingest_behind);
+  void ComputeBottommostFilesMarkedForCompaction(
+      bool allow_ingest_behind, const Comparator* ucmp,
+      const std::string& full_history_ts_low);
 
   // This computes files_marked_for_forced_blob_gc_ and is called by
   // ComputeCompactionScore()
@@ -248,7 +256,8 @@ class VersionStorageInfo {
   // files marked for compaction.
   // REQUIRES: DB mutex held
   void UpdateOldestSnapshot(SequenceNumber oldest_snapshot_seqnum,
-                            bool allow_ingest_behind);
+                            bool allow_ingest_behind, const Comparator* ucmp,
+                            const std::string& full_history_ts_low);
 
   int MaxInputLevel() const;
   int MaxOutputLevel(bool allow_ingest_behind) const;

db/version_set_test.cc

@@ -56,7 +56,8 @@ class GenerateLevelFilesBriefTest : public testing::Test {
         kInvalidBlobFileNumber, kUnknownOldestAncesterTime,
         kUnknownFileCreationTime, kUnknownEpochNumber, kUnknownFileChecksum,
         kUnknownFileChecksumFuncName, kNullUniqueId64x2, 0, 0,
-        /* user_defined_timestamps_persisted */ true);
+        /* user_defined_timestamps_persisted */ true, /* min timestamp */ "",
+        /* max timestamp */ "");
     files_.push_back(f);
   }
 
@@ -172,7 +173,8 @@ class VersionStorageInfoTestBase : public testing::Test {
         kUnknownOldestAncesterTime, kUnknownFileCreationTime,
         kUnknownEpochNumber, kUnknownFileChecksum, kUnknownFileChecksumFuncName,
         kNullUniqueId64x2, compensated_range_deletion_size, 0,
-        /* user_defined_timestamps_persisted */ true);
+        /* user_defined_timestamps_persisted */ true, /* min timestamp */ "",
+        /* max timestamp */ "");
     vstorage_.AddFile(level, f);
   }
 
@@ -391,7 +393,8 @@ TEST_F(VersionStorageInfoTest, MaxBytesForLevelDynamicWithLargeL0_1) {
   ASSERT_EQ(51450U, vstorage_.MaxBytesForLevel(3));
   ASSERT_EQ(257250U, vstorage_.MaxBytesForLevel(4));
 
-  vstorage_.ComputeCompactionScore(ioptions_, mutable_cf_options_);
+  vstorage_.ComputeCompactionScore(ioptions_, mutable_cf_options_,
+                                   /*full_history_ts_low=*/"");
   // Only L0 hits compaction.
   ASSERT_EQ(vstorage_.CompactionScoreLevel(0), 0);
 }
@@ -421,7 +424,8 @@ TEST_F(VersionStorageInfoTest, MaxBytesForLevelDynamicWithLargeL0_2) {
   ASSERT_EQ(51450U, vstorage_.MaxBytesForLevel(3));
   ASSERT_EQ(257250U, vstorage_.MaxBytesForLevel(4));
 
-  vstorage_.ComputeCompactionScore(ioptions_, mutable_cf_options_);
+  vstorage_.ComputeCompactionScore(ioptions_, mutable_cf_options_,
+                                   /*full_history_ts_low=*/"");
   // Although L2 and l3 have higher unadjusted compaction score, considering
   // a relatively large L0 being compacted down soon, L4 is picked up for
   // compaction.
@@ -453,7 +457,8 @@ TEST_F(VersionStorageInfoTest, MaxBytesForLevelDynamicWithLargeL0_3) {
   ASSERT_EQ(2, vstorage_.base_level());
   ASSERT_EQ(20000U, vstorage_.MaxBytesForLevel(2));
 
-  vstorage_.ComputeCompactionScore(ioptions_, mutable_cf_options_);
+  vstorage_.ComputeCompactionScore(ioptions_, mutable_cf_options_,
+                                   /*full_history_ts_low=*/"");
   // Although L2 has higher unadjusted compaction score, considering
   // a relatively large L0 being compacted down soon, L3 is picked up for
   // compaction.
@@ -483,7 +488,8 @@ TEST_F(VersionStorageInfoTest, DrainUnnecessaryLevel) {
   ASSERT_EQ(1, vstorage_.base_level());
   ASSERT_EQ(1000, vstorage_.MaxBytesForLevel(1));
   ASSERT_EQ(10100, vstorage_.MaxBytesForLevel(3));
-  vstorage_.ComputeCompactionScore(ioptions_, mutable_cf_options_);
+  vstorage_.ComputeCompactionScore(ioptions_, mutable_cf_options_,
+                                   /*full_history_ts_low=*/"");
 
   // Tests that levels 1 and 3 are eligible for compaction.
   // Levels 1 and 3 are much smaller than target size,
@@ -1346,7 +1352,8 @@ class VersionSetTestBase {
           Temperature::kUnknown, info.oldest_blob_file_number, 0, 0,
           info.epoch_number, kUnknownFileChecksum, kUnknownFileChecksumFuncName,
           kNullUniqueId64x2, 0, 0,
-          /* user_defined_timestamps_persisted */ true);
+          /* user_defined_timestamps_persisted */ true, /* min timestamp */ "",
+          /* max timestamp */ "");
       if (info.file_missing) {
         ASSERT_OK(fs_->DeleteFile(fname, IOOptions(), nullptr));
       }
@@ -3940,7 +3947,8 @@ TEST_F(VersionSetTestMissingFiles, ManifestFarBehindSst) {
         largest_ikey, 0, 0, false, Temperature::kUnknown, 0, 0, 0,
         file_num /* epoch_number */, kUnknownFileChecksum,
         kUnknownFileChecksumFuncName, kNullUniqueId64x2, 0, 0,
-        /* user_defined_timestamps_persisted */ true);
+        /* user_defined_timestamps_persisted */ true, /* min timestamp */ "",
+        /* max timestamp */ "");
     added_files.emplace_back(0, meta);
   }
   WriteFileAdditionAndDeletionToManifest(
@@ -4001,7 +4009,8 @@ TEST_F(VersionSetTestMissingFiles, ManifestAheadofSst) {
         largest_ikey, 0, 0, false, Temperature::kUnknown, 0, 0, 0,
         file_num /* epoch_number */, kUnknownFileChecksum,
         kUnknownFileChecksumFuncName, kNullUniqueId64x2, 0, 0,
-        /* user_defined_timestamps_persisted */ true);
+        /* user_defined_timestamps_persisted */ true, /* min timestamp */ "",
+        /* max timestamp */ "");
     added_files.emplace_back(0, meta);
   }
   WriteFileAdditionAndDeletionToManifest(

db_stress_tool/db_stress_common.cc

@@ -877,5 +877,24 @@ Status DestroyUnverifiedSubdir(const std::string& dirname) {
   return s;
 }
 
+Status DbStressDestroyDb(const std::string& db_path) {
+  Status s;
+  Options options;
+  // NOTE: using db_stress_listener_env in order to see obsolete MANIFEST files
+  options.env = db_stress_listener_env;
+  // Remove DB files in a principled way to avoid issues
+  if (FLAGS_use_blob_db) {
+    s = blob_db::DestroyBlobDB(db_path, options, blob_db::BlobDBOptions());
+  } else {
+    s = DestroyDB(db_path, options);
+  }
+  if (!s.ok()) {
+    return s;
+  }
+  // Remove everything else recursively, only reporting success if able to
+  // delete everything
+  return DestroyDir(db_stress_listener_env, db_path);
+}
+
 }  // namespace ROCKSDB_NAMESPACE
 #endif  // GFLAGS

db_stress_tool/db_stress_common.h

@@ -100,6 +100,8 @@ DECLARE_bool(enable_pipelined_write);
 DECLARE_bool(verify_before_write);
 DECLARE_bool(histogram);
 DECLARE_bool(destroy_db_initially);
+DECLARE_bool(destroy_db_and_exit);
+DECLARE_string(delete_dir_and_exit);
 DECLARE_bool(verbose);
 DECLARE_bool(progress_reports);
 DECLARE_uint64(db_write_buffer_size);
@@ -820,5 +822,10 @@ Status SaveFilesInDirectory(const std::string& src_dirname,
                             const std::string& dst_dirname);
 Status DestroyUnverifiedSubdir(const std::string& dirname);
 Status InitUnverifiedSubdir(const std::string& dirname);
+
+// Destroy the DB at the given path under the env configured for db_stress.
+// Handles both regular DB and BlobDB, and cleans and removes the entire dir.
+Status DbStressDestroyDb(const std::string& db_path);
+
 }  // namespace ROCKSDB_NAMESPACE
 #endif  // GFLAGS

db_stress_tool/db_stress_env_wrapper.h

@@ -9,8 +9,11 @@
 
 #ifdef GFLAGS
 #pragma once
+
 #include "db_stress_tool/db_stress_common.h"
+#include "file/filename.h"
 #include "monitoring/thread_status_util.h"
+#include "rocksdb/file_checksum.h"
 
 namespace ROCKSDB_NAMESPACE {
 namespace {
@@ -173,6 +176,35 @@ class DbStressFSWrapper : public FileSystemWrapper {
                                const FileOptions& file_opts,
                                std::unique_ptr<FSRandomAccessFile>* r,
                                IODebugContext* dbg) override {
+    // verify that file checksums are propagated through FileOptions
+    // for SST file opens.
+
+    std::string basename = f.substr(f.rfind('/') + 1);
+    uint64_t file_number;
+    FileType file_type;
+    if (ParseFileName(basename, &file_number, &file_type) &&
+        file_type == kTableFile) {
+      // file_checksum_func_name must always be populated to be sure each call
+      // site within RocksDB is intentional about populating the fields with the
+      // best available information:
+      //  - kNoFileChecksumFuncName: no checksum context available
+      //    (e.g., SstFileDumper, SstFileReader, checksum generation),
+      //    always paired with empty checksum
+      //  - kUnknownFileChecksumFuncName: file created without a
+      //    checksum factory (from MANIFEST), always paired with
+      //    empty checksum
+      //  - a real name (e.g., "FileChecksumCrc32c"): checksum exists
+      assert(!file_opts.file_checksum_func_name.empty());
+      if (file_opts.file_checksum_func_name == kUnknownFileChecksumFuncName ||
+          file_opts.file_checksum_func_name == kNoFileChecksumFuncName) {
+        // No checksum available — checksum value must be empty
+        assert(file_opts.file_checksum.empty());
+      } else {
+        // A real checksum function — checksum value must be present
+        assert(!file_opts.file_checksum.empty());
+      }
+    }
+
     std::unique_ptr<FSRandomAccessFile> file;
     IOStatus s = target()->NewRandomAccessFile(f, file_opts, &file, dbg);
     if (s.ok()) {

db_stress_tool/db_stress_gflags.cc

@@ -135,6 +135,14 @@ DEFINE_bool(histogram, false, "Print histogram of operation timings");
 DEFINE_bool(destroy_db_initially, true,
             "Destroys the database dir before start if this is true");
 
+DEFINE_bool(destroy_db_and_exit, false,
+            "Destroys the database dir and exits. Useful for cleanup without "
+            "running stress test. Other options are mostly ignored.");
+
+DEFINE_string(delete_dir_and_exit, "",
+              "Recursively deletes the specified directory and exits. "
+              "Useful for cleaning up TEST_TMPDIR after crash tests.");
+
 DEFINE_bool(verbose, false, "Verbose");
 
 DEFINE_bool(progress_reports, true,

db_stress_tool/db_stress_test_base.cc

@@ -77,22 +77,7 @@ StressTest::StressTest()
       secondary_db_(nullptr),
       is_db_stopped_(false) {
   if (FLAGS_destroy_db_initially) {
-    std::vector<std::string> files;
-    db_stress_env->GetChildren(FLAGS_db, &files);
-    for (unsigned int i = 0; i < files.size(); i++) {
-      if (Slice(files[i]).starts_with("heap-")) {
-        db_stress_env->DeleteFile(FLAGS_db + "/" + files[i]);
-      }
-    }
-
-    Options options;
-    options.env = db_stress_env;
-    // Remove files without preserving manfiest files
-    const Status s = !FLAGS_use_blob_db
-                         ? DestroyDB(FLAGS_db, options)
-                         : blob_db::DestroyBlobDB(FLAGS_db, options,
-                                                  blob_db::BlobDBOptions());
-
+    const Status s = DbStressDestroyDb(FLAGS_db);
     if (!s.ok()) {
       fprintf(stderr, "Cannot destroy original db: %s\n", s.ToString().c_str());
       exit(1);

db_stress_tool/db_stress_tool.cc

@@ -94,10 +94,39 @@ int db_stress_tool(int argc, char** argv) {
     raw_env = fault_env_guard.get();
   }
 
-  env_wrapper_guard = std::make_shared<CompositeEnvWrapper>(
-      raw_env, std::make_shared<DbStressFSWrapper>(raw_env->GetFileSystem()));
+  auto db_stress_fs =
+      std::make_shared<DbStressFSWrapper>(raw_env->GetFileSystem());
+  env_wrapper_guard =
+      std::make_shared<CompositeEnvWrapper>(raw_env, db_stress_fs);
   db_stress_env = env_wrapper_guard.get();
 
+  // Handle --destroy_db_and_exit early, before other option validation
+  if (FLAGS_destroy_db_and_exit) {
+    s = DbStressDestroyDb(FLAGS_db);
+    if (s.ok()) {
+      fprintf(stdout, "Successfully destroyed db at %s\n", FLAGS_db.c_str());
+      return 0;
+    } else {
+      fprintf(stderr, "Failed to destroy db at %s: %s\n", FLAGS_db.c_str(),
+              s.ToString().c_str());
+      return 1;
+    }
+  }
+
+  // Handle --delete_dir_and_exit early, before other option validation
+  if (!FLAGS_delete_dir_and_exit.empty()) {
+    s = DestroyDir(raw_env, FLAGS_delete_dir_and_exit);
+    if (s.ok()) {
+      fprintf(stdout, "Successfully deleted directory %s\n",
+              FLAGS_delete_dir_and_exit.c_str());
+      return 0;
+    } else {
+      fprintf(stderr, "Failed to delete directory %s: %s\n",
+              FLAGS_delete_dir_and_exit.c_str(), s.ToString().c_str());
+      return 1;
+    }
+  }
+
   FLAGS_rep_factory = StringToRepFactory(FLAGS_memtablerep.c_str());
 
   // The number of background threads should be at least as much the

db_stress_tool/no_batched_ops_stress.cc

@@ -1904,6 +1904,17 @@ class NonBatchedOpsStressTest : public StressTest {
     } while (!s.ok() && IsErrorInjectedAndRetryable(s) &&
              initial_wal_write_may_succeed);
 
+    if ((s.IsDeadlock() || s.IsTimedOut()) &&
+        (FLAGS_use_multiget || FLAGS_use_multi_get_entity)) {
+      // Deadlock or timeout is ok, when multi get is tested. Because multi get
+      // tests execute MaybeAddKeyToTxnForRYW function which writes to the
+      // same key space but does not acquire stress test level mutex. So it is
+      // possible RocksDB returns deadlock or timeout. Return OK() for these
+      // cases
+      pending_expected_value.Rollback();
+      return Status::OK();
+    }
+
     if (!s.ok()) {
       pending_expected_value.Rollback();
       if (IsErrorInjectedAndRetryable(s)) {

docs/_posts/2025-12-31-bit-fields-api.markdown

@@ -0,0 +1,279 @@
+---
+title: "BitFields API: Type-Safe Bit Packing for Lock-Free Data Structures"
+layout: post
+author: pdillinger
+category: blog
+---
+
+Modern concurrent data structures increasingly rely on [atomic operations](https://en.cppreference.com/w/cpp/atomic/atomic) to avoid the overhead of locking. A valuable but under-utilized technique for maximizing the effectiveness of atomic operations is [bit packing](https://en.wikipedia.org/wiki/Bit_field)---fitting multiple logical fields into a single atomic variable for algorithmic simplicity and efficiency. However, language support for bit packing does not guarantee dense packing, and manually managing bit manipulation quickly becomes error-prone, especially when dealing with complex state machines.
+
+To address this in RocksDB, we have developed a reusable **BitFields API**, a type-safe, zero-overhead abstraction for bit packing in C++. This works in conjunction with clean wrappers for `std::atomic` for powerful and relatively safe bit-packing of atomic data. For broader use, a [variant of the code](https://github.com/facebook/folly/pull/2549) has been proposed for adding to folly.
+
+## The Problem: Managing Packed Bit Fields
+
+Consider HyperClockCache, an essentially lock-free cache implementation in RocksDB, which was [refactored to use this BitFields API](https://github.com/facebook/rocksdb/pull/14154). It is a hash table built on *slots* that can each hold a cache entry and relevant metadata. For atomic simplicity and efficiency, all the essential metadata for each slot is packed into a single 64-bit value:
+- The reference count and eviction metadata are together encoded into *acquire* and *release* counters, 30 bits each.
+- The possible states of {*empty*, *under construction/destruction*, *occupied+visible*, and *occupied+invisible*} are encoded into three state bits (instead of two, for easier decoding and manipulation).
+- A *hit* bit is used for secondary cache integration.
+
+Traditionally, you might write code like this:
+
+```cpp
+// Old approach: manual bit manipulation
+constexpr uint64_t kAcquireCounterShift = 0;
+constexpr uint64_t kReleaseCounterShift = 30;
+constexpr uint64_t kCounterMask = 0x3FFFFFFF;
+constexpr uint64_t kHitBitShift = 60;
+constexpr uint64_t kOccupiedShift = 61;
+constexpr uint64_t kShareableShift = 62;
+constexpr uint64_t kVisibleShift = 63;
+constexpr uint64_t kStateShift = kOccupiedShift;
+
+std::atomic<uint64_t> meta_;
+
+bool IsUnderConstruction(uint64_t meta) const {
+    return (meta & (uint64_t{1} << kOccupiedShift)) && !(meta & (uint64_t{1} << kShareableShift));
+}
+
+// Getting fields
+uint64_t meta = meta_.load(std::memory_order_acquire);
+if (IsUnderConstruction(meta)) {
+  // ...
+} else if ((meta >> kVisibleShift) & 1) {
+  uint32_t refcount =
+      static_cast<uint32_t>(((meta >> kAcquireCounterShift) -
+                             (meta >> kReleaseCounterShift)) & kCounterMask);
+  // ...
+}
+
+
+// Setting fields
+
+// Set the hit bit (relaxed)
+meta_.fetch_or(uint64_t{1} << kHitBitShift, std::memory_order_relaxed);
+
+// Set both counters to `new_count` (as in eviction processing)
+uint64_t meta = meta_.load(std::memory_order_relaxed);
+uint64_t new_meta =
+    (meta & ((uint64_t{1} << kHitBitShift) | (uint64_t{7} << kStateShift))) |
+    (new_count << kReleaseCounterShift) |
+    (new_count << kAcquireCounterShift);
+bool success = meta_.compare_exchange_strong(meta, new_meta,
+                                             std::memory_order_acq_rel);
+
+// Increment acquire counter by initial_countdown
+old_meta = meta_.fetch_add((uint64_t{1} << kAcquireCounterShift) * initial_countdown,
+                           std::memory_order_acq_rel);
+```
+
+This approach has several problems:
+1. **Error-prone**: Easy to get masks and shifts wrong
+2. **Maintenance burden**: Changes to field sizes require updating multiple constants
+3. **Abstraction challenges**: Even if writing a full set of well-tested getters and setters to hide all the details, details can leak in to do things like update multiple fields in one non-CAS (compare-and-swap) atomic operation.
+
+## New Solution: BitFields API
+
+The BitFields API provides a declarative, type-safe way to define bit-packed structures. Here's how the same example looks with BitFields:
+
+```cpp
+// New approach: declarative bit fields. (Each field must reference the
+// previous, so that the declaration machinery is simply stateless.)
+struct SlotMeta : public BitFields<uint64_t, SlotMeta> {
+  using AcquireCounter = UnsignedBitField<SlotMeta, 30, NoPrevBitField>;
+  using ReleaseCounter = UnsignedBitField<SlotMeta, 30, AcquireCounter>;
+  using HitFlag = BoolBitField<SlotMeta, ReleaseCounter>;
+  using OccupiedFlag = BoolBitField<SlotMeta, HitFlag>;
+  using ShareableFlag = BoolBitField<SlotMeta, OccupiedFlag>;
+  using VisibleFlag = BoolBitField<SlotMeta, ShareableFlag>;
+
+  // Convenience helpers
+  bool IsUnderConstruction() const {
+    return Get<OccupiedFlag>() && !Get<ShareableFlag>();
+  }
+};
+
+BitFieldsAtomic<SlotMeta> meta_;
+
+// Getting fields
+SlotMeta state = meta_.Load();
+if (state.IsUnderConstruction()) {
+  // ...
+} else if (state.Get<SlotMeta::VisibleFlag>()) {
+  uint32_t refcount = state.Get<SlotMeta::AcquireCounter>() -
+                      state.Get<SlotMeta::ReleaseCounter>();
+  // ...
+}
+
+// Setting fields
+
+// Set the hit bit (relaxed)
+meta_.ApplyRelaxed(SlotMeta::HitFlag::SetTransform());
+
+// Set both counters to `new_count` (as in eviction processing)
+SlotMeta meta = meta_.LoadRelaxed();
+SlotMeta new_meta = meta;
+new_meta.Set<SlotMeta::ReleaseCounter>(new_count);
+new_meta.Set<SlotMeta::AcquireCounter>(new_count);
+meta_.CasStrongRelaxed(meta, new_meta);
+
+// Increment acquire counter by initial_countdown
+auto add_acquire =
+    AcquireCounter::PlusTransformPromiseNoOverflow(initial_countdown);
+meta_.Apply(add_acquire, &old_meta);
+
+// Bonus: Atomic multi-field updates without compare-exchange
+auto transform = AcquireCounter::PlusTransformPromiseNoOverflow(1) +
+                 ReleaseCounter::PlusTransformPromiseNoOverflow(1);
+meta_.Apply(transform);
+```
+
+## Key Features
+
+### Type Safety and Self-Documentation
+
+Each field has a specific type (`bool` for `BoolBitField`, appropriately-sized unsigned int for `UnsignedBitField`) and clear semantic meaning. The field definitions are self-documenting: you can immediately see how many bits each field occupies and in what order.
+
+### [Zero Overhead](https://en.cppreference.com/w/cpp/language/Zero-overhead_principle)
+
+Because of heavy use of templates and constexpr operations and the ability to satisfy multiple field reads or writes from a single atomic operation, we have seen no runtime overhead vs. hand-written bit manipulation, in RocksDB. In one case, we verified the assembly code was identical.
+
+[For folly's LifoSem](https://github.com/facebook/folly/pull/2550), there was one case where an optimization hack with detected overflow from one field to another couldn't be replicated as efficiently with the BitFields API because it would violate overflow checking. For that case I dove into the underlying representation to bypass the BitFields overflow check.
+
+### Atomic Operations with Transforms
+
+One of the most powerful features is the ability to combine multiple field updates into a single atomic operation using "transforms", if they are all either (a) some combination of addition and subtraction, (b) bitwise-and, or (c) bitwise-or. For example:
+
+```cpp
+// Clear several but not all fields atomically
+auto and_transform = Field1::AndTransform(0) +
+                 Field2::ClearTransform() +
+                 Field4::ClearTransform();
+atomic_bitfields.Apply(and_transform, &old_state, &new_state);
+...
+// Set more than one boolean field atomically
+auto or_transform = Field2::SetTransform() +
+                 Field4::SetTransform();
+atomic_bitfields.Apply(or_transform, &old_state, &new_state);
+...
+auto add_transform = Field1::PlusTransformPromiseNoOverflow(1) +
+                     Field3::MinusTransformPromiseNoUnderflow(1);
+atomic_bitfields.Apply(add_transform, &old_state, &new_state);
+```
+
+Each `Apply()` generates a single atomic operation (e.g., `fetch_add` or `fetch_or`) that updates all the specified fields, and optionally returns both the old and new values. This enables a number of hacks for atomic updates without CAS.
+
+### Overflow Protection
+
+The API includes built-in overflow detection in debug builds:
+
+```cpp
+// An assertion will fail in debug builds if the counter overflows
+auto transform = Counter::PlusTransformPromiseNoOverflow(value);
+atomic.Apply(transform);
+```
+
+For fields at the top of the underlying representation (where overflow doesn't affect other fields), overflow is explicitly ignored. (A compile time error is generated if you try to use `PlusTransformPromiseNoOverflow` on a field at the top of the representation or `PlusTransformIgnoreOverflow` on a field not at the top of the representation.)
+
+```cpp
+// For wraparound counters
+auto transform = Counter::PlusTransformIgnoreOverflow(value);
+```
+
+This capability is used in a folly data structure called LifoSem, which [I have proposed to refactor](https://github.com/facebook/folly/pull/2550) to a proposed BitFields API variant for folly.
+
+### Compare-and-Swap (CAS) Support
+
+The atomic wrappers provide full CAS support for lock-free algorithms:
+
+```cpp
+SlotMeta expected = current_state;
+SlotMeta desired = expected.With<Field1>(new_value).With<Field2>(true);
+if (meta_.CasStrong(expected, desired)) {
+  // Successfully updated
+  ...
+}
+```
+
+### Atomic wrappers
+
+The BitFields API includes two atomic wrappers: `RelaxedBitFieldsAtomic` and `BitFieldsAtomic`. However, RocksDB also has versions of these wrappers for regular `std::atomic` variables that help with memory ordering discipline: `RelaxedAtomic` and `Atomic` in `util/atomic.h`.
+
+These wrappers help in a couple of ways:
+* **Self-document intended memory order**: An atomic field generally has a single memory order that all or most operations should use, typically either `std::memory_order_relaxed` or `std::memory_order_acq_rel`.
+* **More intentional memory orders and atomic operations**: The standard library's implicit conversions and default memory ordering (`memory_order_seq_cst`) make it easy to accidentally use sequential consistency with acquire/release ordering or even relaxed, which could hurt performance, and tend to hide where atomic operations are actually happening (e.g. implicit vs. explicit load).
+
+For example, instead of writing:
+```cpp
+std::atomic<uint64_t> stat_counter;
+stat_counter++;  // Uses memory_order_seq_cst implicitly - maybe inefficient
+```
+
+You write:
+```cpp
+RelaxedAtomic<uint64_t> stat_counter;
+stat_counter.FetchAddRelaxed(1);  // Explicitly relaxed - appropriate for a diagnostic counter
+```
+
+Or for data providing synchronization:
+```cpp
+Atomic<size_t> refcount;
+refcount.FetchAdd(1);  // Standard acquire-release semantics for coordinating with other threads
+```
+
+These wrappers complement the BitFields atomic wrappers by providing the same ordering discipline for non-packed atomic variables throughout much of RocksDB, creating a more readable and less clunky approach to concurrent programming. Migrating remaining uses of `std::atomic` is an ongoing effort.
+
+## Real-World Usage in RocksDB
+
+The BitFields API was developed along with the revamped parallel compression in RocksDB, but with the intention to also clean up the HyperClockCache (HCC) implementation. With that migration complete, we can see the benefits. Specifically, **by packing more of the state machine into a single atomic value, the parallel algorithms became both simpler and more efficient.** Concurrent algorithms that could have blown up in their state space with elaborate interleavings between threads trying not to block each other, e.g. because of multi-step consensus on work assignments, were instead able to quickly and more easily make progress, e.g. with atomically clear work assignments.
+
+### Before: Manual Bit Manipulation
+
+The old HCC code was difficult to read and maintain. Many of the common read and update operations had manually written helper functions, but it was not practical to develop the full set of functions needed for rare cases. Consider this code that clears the "visible" flag on a slot when an entry is erased from subsequent lookups but might still be referenced:
+
+```cpp
+// Old HCC code, without atomic wrappers
+uint64_t old_meta =
+        h->meta.fetch_and(~(uint64_t{ClockHandle::kStateVisibleBit}
+                                   << ClockHandle::kStateShift), std::memory_order_acq_rel);
+// Apply update to local copy
+uint64_t new_meta = old_meta & ~(uint64_t{ClockHandle::kStateVisibleBit}
+                            << ClockHandle::kStateShift);
+
+// New HCC code
+SlotMeta old_meta, new_meta;
+h->meta.Apply(SlotMeta::VisibleFlag::ClearTransform(), &old_meta, &new_meta);
+```
+
+Or this assertion that the acquire and release counters are different:
+
+```cpp
+// Old HCC code
+uint64_t old_meta = ...;
+assert(((old_meta >> ClockHandle::kAcquireCounterShift) &
+        ClockHandle::kCounterMask) !=
+        ((old_meta >> ClockHandle::kReleaseCounterShift) &
+        ClockHandle::kCounterMask));
+
+// New HCC code without single-purpose helper functions
+SlotMeta old_meta = ...;
+assert(old_meta.Get<SlotMeta::AcquireCounter>() !=
+       old_meta.Get<SlotMeta::ReleaseCounter>());
+
+// New HCC code, with single-purpose helper functions
+SlotMeta old_meta = ...;
+assert(old_meta.GetAcquireCounter() != old_meta.GetReleaseCounter());
+```
+
+Some hand-written helper functions or using directives are still useful for brevity, but even without them all the bit manipulation details are hidden in the BitFields implementation.
+
+## Future Directions
+
+We hope the proposed folly version is accepted to make the BitFields API available for broader usage. Additionally, some quality-of-life improvements are likely possible, perhaps including easier declaration and usage syntax, hopefully without delving into boost-like macro hell. Better runtime and compile time checks might also be possible.
+
+## Conclusion
+
+The BitFields API demonstrates that zero-overhead abstractions can significantly improve code quality without sacrificing performance. By providing type safety, self-documentation, and convenience features around bit manipulation and atomic operations, it makes lock-free programming more accessible and maintainable. Bit-packed atomics are arguably essential for *slaying the complexity dragon* of efficient lock-free and low-lock algorithms, because they reduce explosion in algorithm states.
+
+For RocksDB specifically, the migration to BitFields has made the HyperClockCache implementation substantially easier to understand and modify, while maintaining the same high-performance characteristics. Combined with the recent [parallel compression revamp](/blog/2025/10/08/parallel-compression-revamp.html), these improvements showcase our ongoing commitment to writing clean, efficient, and maintainable code.
+
+The BitFields API is available in RocksDB's util/bit_fields.h and can be adapted for use in other projects requiring efficient, type-safe bit packing. For those building high-performance concurrent systems, it offers a compelling alternative to manual bit manipulation—proving that safe abstractions and peak performance are not mutually exclusive.

env/env_test.cc

@@ -41,6 +41,9 @@
 #include "env/env_chroot.h"
 #include "env/env_encryption_ctr.h"
 #include "env/fs_readonly.h"
+#if defined(ROCKSDB_IOURING_PRESENT)
+#include "env/io_posix.h"
+#endif
 #include "env/mock_env.h"
 #include "env/unique_id_gen.h"
 #include "logging/log_buffer.h"
@@ -1655,42 +1658,6 @@ void GenerateFilesAndRequest(Env* env, const std::string& fname,
   }
 }
 
-TEST_F(EnvPosixTest, MultiReadIOUringError) {
-  // In this test we don't do aligned read, so we can't do direct I/O.
-  EnvOptions soptions;
-  soptions.use_direct_reads = soptions.use_direct_writes = false;
-  std::string fname = test::PerThreadDBPath(env_, "testfile");
-
-  std::vector<std::string> scratches;
-  std::vector<ReadRequest> reqs;
-  GenerateFilesAndRequest(env_, fname, &reqs, &scratches);
-  // Query the data
-  std::unique_ptr<RandomAccessFile> file;
-  ASSERT_OK(env_->NewRandomAccessFile(fname, &file, soptions));
-
-  bool io_uring_wait_cqe_called = false;
-  SyncPoint::GetInstance()->SetCallBack(
-      "PosixRandomAccessFile::MultiRead:io_uring_wait_cqe:return",
-      [&](void* arg) {
-        if (!io_uring_wait_cqe_called) {
-          io_uring_wait_cqe_called = true;
-          ssize_t& ret = *(static_cast<ssize_t*>(arg));
-          ret = 1;
-        }
-      });
-  SyncPoint::GetInstance()->EnableProcessing();
-
-  Status s = file->MultiRead(reqs.data(), reqs.size());
-  if (io_uring_wait_cqe_called) {
-    ASSERT_NOK(s);
-  } else {
-    s.PermitUncheckedError();
-  }
-
-  SyncPoint::GetInstance()->DisableProcessing();
-  SyncPoint::GetInstance()->ClearAllCallBacks();
-}
-
 TEST_F(EnvPosixTest, MultiReadIOUringError2) {
   // In this test we don't do aligned read, so we can't do direct I/O.
   EnvOptions soptions;
@@ -1706,19 +1673,20 @@ TEST_F(EnvPosixTest, MultiReadIOUringError2) {
 
   bool io_uring_submit_and_wait_called = false;
   SyncPoint::GetInstance()->SetCallBack(
-      "PosixRandomAccessFile::MultiRead:io_uring_submit_and_wait:return1",
+      "PosixRandomAccessFile::MultiRead:io_uring_sq_ready:return1",
       [&](void* arg) {
         io_uring_submit_and_wait_called = true;
-        ssize_t* ret = static_cast<ssize_t*>(arg);
-        (*ret)--;
+        unsigned* ret = static_cast<unsigned*>(arg);
+        *ret = 1;
       });
   SyncPoint::GetInstance()->SetCallBack(
       "PosixRandomAccessFile::MultiRead:io_uring_submit_and_wait:return2",
       [&](void* arg) {
         struct io_uring* iu = static_cast<struct io_uring*>(arg);
         struct io_uring_cqe* cqe;
-        assert(io_uring_wait_cqe(iu, &cqe) == 0);
-        io_uring_cqe_seen(iu, cqe);
+        // CQ should be empty after drain - peek should fail
+        int ret = io_uring_peek_cqe(iu, &cqe);
+        assert(-EAGAIN == ret);  // No CQEs available
       });
   SyncPoint::GetInstance()->EnableProcessing();
 
@@ -3640,6 +3608,486 @@ TEST_F(TestAsyncRead, ReadAsync) {
   }
 }
 
+// Test ReadAsync -> MultiRead -> Poll with real io_uring (not mock).
+// This verifies that MultiRead doesn't interfere with async read buffers.
+TEST_F(TestAsyncRead, InterleavingIOUringOperations) {
+#if defined(ROCKSDB_IOURING_PRESENT)
+  // Use the real filesystem directly (not the mock ReadAsyncFS).
+  std::shared_ptr<FileSystem> fs = env_->GetFileSystem();
+  std::string fname = test::PerThreadDBPath(env_, "testfile_iouring");
+
+  constexpr size_t kSectorSize = 4096;
+  constexpr size_t kNumSectors = 8;
+
+  // 1. Create & write to a file.
+  {
+    std::unique_ptr<FSWritableFile> wfile;
+    ASSERT_OK(
+        fs->NewWritableFile(fname, FileOptions(), &wfile, nullptr /*dbg*/));
+
+    for (size_t i = 0; i < kNumSectors; ++i) {
+      auto data = NewAligned(kSectorSize * 8, static_cast<char>(i + 1));
+      Slice slice(data.get(), kSectorSize);
+      ASSERT_OK(wfile->Append(slice, IOOptions(), nullptr));
+    }
+    ASSERT_OK(wfile->Close(IOOptions(), nullptr));
+  }
+
+  // 2. Test interleaved ReadAsync and MultiRead operations.
+  {
+    std::unique_ptr<FSRandomAccessFile> file;
+    ASSERT_OK(fs->NewRandomAccessFile(fname, FileOptions(), &file, nullptr));
+
+    IOOptions opts;
+    std::vector<void*> io_handles(kNumSectors);
+    std::vector<FSReadRequest> async_reqs(kNumSectors);
+    std::vector<std::unique_ptr<char, Deleter>> async_data;
+    std::vector<size_t> vals;
+    IOHandleDeleter del_fn;
+
+    // Initialize async read requests.
+    for (size_t i = 0; i < kNumSectors; i++) {
+      async_reqs[i].offset = i * kSectorSize;
+      async_reqs[i].len = kSectorSize;
+      async_data.emplace_back(NewAligned(kSectorSize, 0));
+      async_reqs[i].scratch = async_data.back().get();
+      vals.push_back(i);
+    }
+
+    // Callback function for async reads.
+    std::function<void(FSReadRequest&, void*)> callback =
+        [&](FSReadRequest& req, void* cb_arg) {
+          assert(cb_arg != nullptr);
+          size_t i = *(reinterpret_cast<size_t*>(cb_arg));
+          async_reqs[i].offset = req.offset;
+          async_reqs[i].result = req.result;
+          async_reqs[i].status = req.status;
+        };
+
+    // Submit asynchronous read requests.
+    for (size_t i = 0; i < kNumSectors; i++) {
+      void* cb_arg = static_cast<void*>(&(vals[i]));
+      IOStatus s = file->ReadAsync(async_reqs[i], opts, callback, cb_arg,
+                                   &(io_handles[i]), &del_fn, nullptr);
+      if (s.IsNotSupported()) {
+        // io_uring not supported on this system, skip the test.
+        fprintf(stderr, "Skipping test - io_uring not supported: %s\n",
+                s.ToString().c_str());
+        for (size_t j = 0; j < i; j++) {
+          if (io_handles[j] != nullptr) {
+            del_fn(io_handles[j]);
+          }
+        }
+        return;
+      }
+      // For any other error, fail the test.
+      ASSERT_OK(s);
+    }
+
+    // Do a MultiRead on same sectors while async reads are submitted.
+    std::vector<FSReadRequest> multi_reqs(kNumSectors);
+    std::vector<std::unique_ptr<char, Deleter>> multi_data;
+    for (size_t i = 0; i < kNumSectors; i++) {
+      multi_reqs[i].offset = i * kSectorSize;
+      multi_reqs[i].len = kSectorSize;
+      multi_data.emplace_back(NewAligned(kSectorSize, 0));
+      multi_reqs[i].scratch = multi_data.back().get();
+    }
+    ASSERT_OK(file->MultiRead(multi_reqs.data(), kNumSectors, opts, nullptr));
+
+    // Check the status of MultiRead requests (should all succeed).
+    for (size_t i = 0; i < kNumSectors; i++) {
+      auto buf = NewAligned(kSectorSize * 8, static_cast<char>(i + 1));
+      Slice expected_data(buf.get(), kSectorSize);
+
+      ASSERT_EQ(multi_reqs[i].offset, i * kSectorSize);
+      ASSERT_OK(multi_reqs[i].status);
+      ASSERT_EQ(expected_data.ToString(), multi_reqs[i].result.ToString());
+    }
+
+    // Poll for the submitted async requests.
+    ASSERT_OK(fs->Poll(io_handles, kNumSectors));
+
+    // Check the status of async read requests (should all succeed).
+    for (size_t i = 0; i < kNumSectors; i++) {
+      auto buf = NewAligned(kSectorSize * 8, static_cast<char>(i + 1));
+      Slice expected_data(buf.get(), kSectorSize);
+
+      ASSERT_EQ(async_reqs[i].offset, i * kSectorSize);
+      ASSERT_OK(async_reqs[i].status);
+      ASSERT_EQ(expected_data.ToString(), async_reqs[i].result.ToString());
+    }
+
+    // Delete io_handles.
+    for (size_t i = 0; i < io_handles.size(); i++) {
+      del_fn(io_handles[i]);
+    }
+  }
+#else
+  fprintf(stderr, "Skipping test - ROCKSDB_IOURING_PRESENT not defined\n");
+#endif
+}
+
+// Helper function to run AbortIO test with parameterized read requests.
+// Each request is specified as {offset, length}.
+// use_direct_io: if true, opens the file with O_DIRECT to bypass page cache.
+// iterations: number of times to repeat the test (useful for race conditions).
+void TestAbortIOWithRequests(
+    Env* env, size_t file_size,
+    const std::vector<std::pair<uint64_t, size_t>>& read_specs,
+    bool use_direct_io = false, int iterations = 1) {
+#if defined(ROCKSDB_IOURING_PRESENT)
+  fprintf(stderr,
+          "TestAbortIOWithRequests: file_size=%zu, num_reads=%zu, "
+          "direct_io=%d, iterations=%d\n",
+          file_size, read_specs.size(), use_direct_io, iterations);
+  std::shared_ptr<FileSystem> fs = env->GetFileSystem();
+  std::string fname = test::PerThreadDBPath(env, "testfile_abortio");
+
+  // 1. Create test file once (content doesn't change between iterations)
+  {
+    std::unique_ptr<FSWritableFile> wfile;
+    FileOptions file_opts;
+    file_opts.use_direct_writes = true;
+    ASSERT_OK(fs->NewWritableFile(fname, file_opts, &wfile, nullptr));
+
+    // Query the file's required buffer alignment (logical block size)
+    // instead of hardcoding 4096, to support devices with different
+    // sector sizes.
+    size_t sector_size = wfile->GetRequiredBufferAlignment();
+
+    // Round up to full sectors for direct IO writes
+    size_t num_sectors = (file_size + sector_size - 1) / sector_size;
+    for (size_t i = 0; i < num_sectors; ++i) {
+      auto data = NewAligned(sector_size, static_cast<char>(i + 1));
+      Slice slice(data.get(), sector_size);
+      ASSERT_OK(wfile->Append(slice, IOOptions(), nullptr));
+    }
+
+    // Truncate to exact file size if not aligned to sector boundary
+    if (file_size % sector_size != 0) {
+      ASSERT_OK(wfile->Truncate(file_size, IOOptions(), nullptr));
+    }
+
+    ASSERT_OK(wfile->Close(IOOptions(), nullptr));
+  }
+
+  for (int iter = 0; iter < iterations; iter++) {
+    // 2. Submit ReadAsync requests and immediately abort
+    {
+      FileOptions file_opts;
+      file_opts.use_direct_reads = use_direct_io;
+      std::unique_ptr<FSRandomAccessFile> file;
+      ASSERT_OK(fs->NewRandomAccessFile(fname, file_opts, &file, nullptr));
+
+      const size_t num_reads = read_specs.size();
+      IOOptions opts;
+      std::vector<void*> io_handles(num_reads);
+      std::vector<FSReadRequest> reqs(num_reads);
+      std::vector<std::unique_ptr<char, Deleter>> data;
+      std::vector<size_t> vals;
+      IOHandleDeleter del_fn;
+      std::atomic<int> callbacks_invoked{0};
+
+      // Initialize read requests from specs
+      for (size_t i = 0; i < num_reads; i++) {
+        reqs[i].offset = read_specs[i].first;
+        reqs[i].len = read_specs[i].second;
+        data.emplace_back(NewAligned(reqs[i].len, 0));
+        reqs[i].scratch = data.back().get();
+        vals.push_back(i);
+      }
+
+      // Callback
+      std::function<void(FSReadRequest&, void*)> callback =
+          [&](FSReadRequest& req, void* cb_arg) {
+            size_t i = *(reinterpret_cast<size_t*>(cb_arg));
+            reqs[i].status = req.status;
+            callbacks_invoked++;
+          };
+
+      // Submit all ReadAsync requests
+      for (size_t i = 0; i < num_reads; i++) {
+        void* cb_arg = static_cast<void*>(&(vals[i]));
+        IOStatus s = file->ReadAsync(reqs[i], opts, callback, cb_arg,
+                                     &(io_handles[i]), &del_fn, nullptr);
+        if (s.IsNotSupported()) {
+          // io_uring not supported, clean up and skip
+          fprintf(stderr,
+                  "WARNING: io_uring not supported, skipping test: %s\n",
+                  s.ToString().c_str());
+          for (size_t j = 0; j < i; j++) {
+            if (io_handles[j]) {
+              del_fn(io_handles[j]);
+            }
+          }
+          ASSERT_OK(fs->DeleteFile(fname, IOOptions(), nullptr));
+          return;
+        }
+        ASSERT_OK(s);
+      }
+
+      // Immediately call AbortIO - this should NOT hang
+      ASSERT_OK(fs->AbortIO(io_handles));
+
+      // Verify all handles are finished and all callbacks were invoked.
+      // Since all handles are passed to AbortIO, every handle is guaranteed
+      // to be finalized (either completed or cancelled).
+      for (size_t i = 0; i < num_reads; i++) {
+        Posix_IOHandle* h = static_cast<Posix_IOHandle*>(io_handles[i]);
+        ASSERT_TRUE(h->is_finished);
+      }
+      ASSERT_EQ(callbacks_invoked.load(), static_cast<int>(num_reads));
+
+      // Clean up handles
+      for (size_t i = 0; i < num_reads; i++) {
+        if (io_handles[i]) {
+          del_fn(io_handles[i]);
+        }
+      }
+    }
+  }
+
+  ASSERT_OK(fs->DeleteFile(fname, IOOptions(), nullptr));
+
+  fprintf(stderr, "TestAbortIOWithRequests: completed %d iterations\n",
+          iterations);
+#else
+  fprintf(stderr,
+          "TestAbortIOWithRequests: SKIPPED (ROCKSDB_IOURING_PRESENT not "
+          "defined)\n");
+  (void)env;
+  (void)file_size;
+  (void)read_specs;
+  (void)use_direct_io;
+  (void)iterations;
+#endif
+}
+
+// Test overlapping reads at aligned offsets (multiples of 4KB)
+TEST_F(TestAsyncRead, AbortIOOverlappingAligned) {
+  // 4 reads of 16KB each, overlapping by 8KB, all at 4KB-aligned offsets
+  // Read 0: [0, 16KB), Read 1: [8KB, 24KB), Read 2: [16KB, 32KB), Read 3:
+  // [24KB, 40KB)
+  std::vector<std::pair<uint64_t, size_t>> specs = {
+      {0, 16384},
+      {8192, 16384},
+      {16384, 16384},
+      {24576, 16384},
+  };
+  TestAbortIOWithRequests(env_, 64 * 1024, specs);
+}
+
+// Test reads at unaligned offsets (not multiples of 4KB)
+TEST_F(TestAsyncRead, AbortIOUnalignedOffsets) {
+  // Reads starting at non-4KB-aligned offsets
+  std::vector<std::pair<uint64_t, size_t>> specs = {
+      {1000, 8192},    // starts at 1000 (unaligned)
+      {5000, 12288},   // starts at 5000 (unaligned), spans multiple sectors
+      {15000, 8192},   // starts at 15000 (unaligned)
+      {25500, 16384},  // starts at 25500 (unaligned)
+  };
+  TestAbortIOWithRequests(env_, 64 * 1024, specs);
+}
+
+// Test mix of aligned and unaligned, various sizes
+TEST_F(TestAsyncRead, AbortIOMixedOffsets) {
+  std::vector<std::pair<uint64_t, size_t>> specs = {
+      {0, 4096},       // aligned, 1 sector
+      {1500, 8192},    // unaligned, 2 sectors
+      {4096, 20480},   // aligned, 5 sectors
+      {7000, 4096},    // unaligned, spans 2 sectors
+      {16384, 32768},  // aligned, 8 sectors
+      {50000, 8192},   // unaligned
+  };
+  TestAbortIOWithRequests(env_, 128 * 1024, specs);
+}
+
+// Stress test with many concurrent handles
+TEST_F(TestAsyncRead, AbortIOStress) {
+  std::vector<std::pair<uint64_t, size_t>> specs;
+  // 16 overlapping reads with mixed alignment
+  for (int i = 0; i < 16; i++) {
+    uint64_t offset = i * 4000;          // Not aligned to 4KB
+    size_t len = 8192 + (i % 4) * 4096;  // 8KB to 20KB
+    specs.emplace_back(offset, len);
+  }
+  TestAbortIOWithRequests(env_, 256 * 1024, specs);
+}
+
+// Regression test for a fixed bug in AbortIO where out-of-order io_uring
+// completions could cause an infinite hang. The bug occurred when completions
+// for a different handle arrived while waiting for the current handle - the
+// code would consume those completions but not mark the handle as finished,
+// causing a hang when later iterating to that handle.
+//
+// Uses a large read (1MB) followed by a small read (4KB) with Direct I/O to
+// maximize the chance of out-of-order completions. Runs 100 iterations to
+// increase the likelihood of triggering the race condition.
+TEST_F(TestAsyncRead, AbortIOReversedHandles) {
+  // Request 0: LARGE (1MB) at offset 0
+  // Request 1: SMALL (4KB) at offset 1MB
+  std::vector<std::pair<uint64_t, size_t>> specs = {
+      {0, 1024 * 1024},     // 1MB read
+      {1024 * 1024, 4096},  // 4KB read at 1MB offset
+  };
+  // 2MB file, Direct I/O enabled, 100 iterations
+  TestAbortIOWithRequests(env_, 2 * 1024 * 1024, specs,
+                          /*use_direct_io=*/true, /*iterations=*/100);
+}
+
+// Test for bug fix: AbortIO with partial handles should correctly handle
+// completions for non-aborted handles.
+//
+// Previously, AbortIO would consume completions for non-aborted handles but
+// not set is_finished (since it expected req_count==2 for all handles).
+// This caused subsequent Poll calls to hang forever.
+//
+// The fix correctly detects handles not in the abort set and finalizes them
+// immediately when their completion arrives (at req_count==1).
+TEST_F(TestAsyncRead, AbortIOPartialHandlesBug) {
+#if defined(ROCKSDB_IOURING_PRESENT)
+  std::shared_ptr<FileSystem> fs = env_->GetFileSystem();
+  std::string fname = test::PerThreadDBPath(env_, "testfile_abortio_partial");
+
+  constexpr size_t kSectorSize = 4096;
+  constexpr size_t kFileSize = 2 * 1024 * 1024;  // 2MB
+
+  // 1. Create test file with direct I/O
+  {
+    std::unique_ptr<FSWritableFile> wfile;
+    FileOptions file_opts;
+    file_opts.use_direct_writes = true;
+    ASSERT_OK(fs->NewWritableFile(fname, file_opts, &wfile, nullptr));
+
+    size_t num_sectors = kFileSize / kSectorSize;
+    for (size_t i = 0; i < num_sectors; ++i) {
+      auto data = NewAligned(kSectorSize, static_cast<char>(i + 1));
+      Slice slice(data.get(), kSectorSize);
+      ASSERT_OK(wfile->Append(slice, IOOptions(), nullptr));
+    }
+    ASSERT_OK(wfile->Close(IOOptions(), nullptr));
+  }
+
+  // 2. Submit 3 ReadAsync requests, abort only the first one, then Poll the
+  // rest
+  {
+    FileOptions file_opts;
+    file_opts.use_direct_reads = true;
+    std::unique_ptr<FSRandomAccessFile> file;
+    ASSERT_OK(fs->NewRandomAccessFile(fname, file_opts, &file, nullptr));
+
+    IOOptions opts;
+    constexpr size_t kNumReads = 3;
+    std::vector<void*> io_handles(kNumReads);
+    std::vector<FSReadRequest> reqs(kNumReads);
+    std::vector<std::unique_ptr<char, Deleter>> data;
+    std::vector<size_t> vals;
+    IOHandleDeleter del_fn;
+    std::atomic<int> callbacks_invoked{0};
+
+    // H0: 1MB read, H1: 4KB read, H2: 4KB read
+    std::vector<std::pair<uint64_t, size_t>> read_specs = {
+        {0, 1024 * 1024},            // H0: 1MB at offset 0
+        {1024 * 1024, 4096},         // H1: 4KB at offset 1MB
+        {1024 * 1024 + 4096, 4096},  // H2: 4KB at offset 1MB+4KB
+    };
+
+    for (size_t i = 0; i < kNumReads; i++) {
+      reqs[i].offset = read_specs[i].first;
+      reqs[i].len = read_specs[i].second;
+      data.emplace_back(NewAligned(reqs[i].len, 0));
+      reqs[i].scratch = data.back().get();
+      vals.push_back(i);
+    }
+
+    std::function<void(FSReadRequest&, void*)> callback =
+        [&](FSReadRequest& req, void* cb_arg) {
+          size_t i = *(reinterpret_cast<size_t*>(cb_arg));
+          reqs[i].status = req.status;
+          callbacks_invoked++;
+        };
+
+    // Submit all ReadAsync requests
+    for (size_t i = 0; i < kNumReads; i++) {
+      void* cb_arg = static_cast<void*>(&(vals[i]));
+      IOStatus s = file->ReadAsync(reqs[i], opts, callback, cb_arg,
+                                   &(io_handles[i]), &del_fn, nullptr);
+      if (s.IsNotSupported()) {
+        // io_uring not supported, clean up and skip
+        for (size_t j = 0; j < i; j++) {
+          if (io_handles[j]) {
+            del_fn(io_handles[j]);
+          }
+        }
+        ASSERT_OK(fs->DeleteFile(fname, IOOptions(), nullptr));
+        return;
+      }
+      ASSERT_OK(s);
+    }
+
+    // Wait for reads to complete in io_uring (completions in queue but not
+    // consumed). 5 seconds should be plenty for direct I/O reads to complete.
+    std::this_thread::sleep_for(std::chrono::seconds(5));
+
+    // Abort ONLY H0 - this will consume all completions but should correctly
+    // finalize H1 and H2 (since they're not in the abort set).
+    std::vector<void*> abort_handles = {io_handles[0]};
+    ASSERT_OK(fs->AbortIO(abort_handles));
+
+    // Verify H0 is finished (aborted)
+    Posix_IOHandle* h0 = static_cast<Posix_IOHandle*>(io_handles[0]);
+    ASSERT_TRUE(h0->is_finished);
+    ASSERT_EQ(h0->req_count, 2u);  // original + cancel
+
+    // Note: H1 and H2 may or may not be finished at this point. AbortIO
+    // finalizes non-aborted handles whose CQEs arrive while waiting for
+    // aborted handles, but CQE ordering is non-deterministic. If H0's
+    // completions arrived first, H1/H2's CQEs are still in the queue.
+    // Poll handles either case correctly.
+
+    // Poll on H1, H2 - completes them if not already finalized by AbortIO
+    std::vector<void*> poll_handles = {io_handles[1], io_handles[2]};
+
+    // Use a watchdog to detect hang (regression test for the original bug
+    // where AbortIO consumed non-aborted CQEs without finalizing them)
+    std::atomic<bool> poll_completed{false};
+    std::thread watchdog([&]() {
+      for (int i = 0; i < 500; i++) {  // 5 seconds timeout
+        std::this_thread::sleep_for(std::chrono::milliseconds(10));
+        if (poll_completed) return;
+      }
+      // Bug regression: Poll hung
+      _exit(1);
+    });
+
+    fs->Poll(poll_handles, poll_handles.size());
+    poll_completed = true;
+    watchdog.join();
+
+    // After Poll, H1 and H2 must be finished
+    Posix_IOHandle* h1 = static_cast<Posix_IOHandle*>(io_handles[1]);
+    Posix_IOHandle* h2 = static_cast<Posix_IOHandle*>(io_handles[2]);
+    ASSERT_TRUE(h1->is_finished);
+    ASSERT_TRUE(h2->is_finished);
+
+    // Verify all callbacks were invoked
+    ASSERT_EQ(callbacks_invoked.load(), 3);
+
+    // Clean up handles
+    for (size_t i = 0; i < kNumReads; i++) {
+      if (io_handles[i]) {
+        del_fn(io_handles[i]);
+      }
+    }
+  }
+
+  ASSERT_OK(fs->DeleteFile(fname, IOOptions(), nullptr));
+#else
+  (void)env_;  // Suppress unused variable warning
+#endif
+}
+
 struct StaticDestructionTester {
   bool activated = false;
   ~StaticDestructionTester() {

env/fs_posix.cc

@@ -270,7 +270,10 @@ class PosixFileSystem : public FileSystem {
           options
 #if defined(ROCKSDB_IOURING_PRESENT)
           ,
-          !IsIOUringEnabled() ? nullptr : thread_local_io_urings_.get()
+          !IsIOUringEnabled() ? nullptr
+                              : thread_local_async_read_io_urings_.get(),
+          !IsIOUringEnabled() ? nullptr
+                              : thread_local_multi_read_io_urings_.get()
 #endif
               ));
     }
@@ -1087,8 +1090,9 @@ class PosixFileSystem : public FileSystem {
 #if defined(ROCKSDB_IOURING_PRESENT)
     // io_uring_queue_init.
     struct io_uring* iu = nullptr;
-    if (thread_local_io_urings_) {
-      iu = static_cast<struct io_uring*>(thread_local_io_urings_->Get());
+    if (thread_local_async_read_io_urings_) {
+      iu = static_cast<struct io_uring*>(
+          thread_local_async_read_io_urings_->Get());
     }
 
     // Init failed, platform doesn't support io_uring.
@@ -1125,25 +1129,7 @@ class PosixFileSystem : public FileSystem {
         // Reset cqe data to catch any stray reuse of it
         static_cast<struct io_uring_cqe*>(cqe)->user_data = 0xd5d5d5d5d5d5d5d5;
 
-        FSReadRequest req;
-        req.scratch = posix_handle->scratch;
-        req.offset = posix_handle->offset;
-        req.len = posix_handle->len;
-
-        size_t finished_len = 0;
-        size_t bytes_read = 0;
-        bool read_again = false;
-        UpdateResult(cqe, "", req.len, posix_handle->iov.iov_len,
-                     true /*async_read*/, posix_handle->use_direct_io,
-                     posix_handle->alignment, finished_len, &req, bytes_read,
-                     read_again);
-        posix_handle->is_finished = true;
-        io_uring_cqe_seen(iu, cqe);
-        posix_handle->cb(req, posix_handle->cb_arg);
-
-        (void)finished_len;
-        (void)bytes_read;
-        (void)read_again;
+        FinalizeAsyncRead(iu, cqe, posix_handle);
 
         if (static_cast<Posix_IOHandle*>(io_handles[i]) == posix_handle) {
           break;
@@ -1161,8 +1147,9 @@ class PosixFileSystem : public FileSystem {
 #if defined(ROCKSDB_IOURING_PRESENT)
     // io_uring_queue_init.
     struct io_uring* iu = nullptr;
-    if (thread_local_io_urings_) {
-      iu = static_cast<struct io_uring*>(thread_local_io_urings_->Get());
+    if (thread_local_async_read_io_urings_) {
+      iu = static_cast<struct io_uring*>(
+          thread_local_async_read_io_urings_->Get());
     }
 
     // Init failed, platform doesn't support io_uring.
@@ -1183,6 +1170,11 @@ class PosixFileSystem : public FileSystem {
         return IOStatus::IOError("");
       }
 
+      // Mark this handle as being aborted. This is used when processing
+      // completions to distinguish between aborted handles (expect 2
+      // completions: original + cancel) and non-aborted handles (expect 1).
+      posix_handle->is_being_aborted = true;
+
       // Prepare the cancel request.
       struct io_uring_sqe* sqe;
       sqe = io_uring_get_sqe(iu);
@@ -1229,6 +1221,14 @@ class PosixFileSystem : public FileSystem {
         }
         posix_handle->req_count++;
 
+        if (!posix_handle->is_being_aborted) {
+          // This is a completion for a handle NOT being aborted.
+          // It only has 1 outstanding request (the original read), so we
+          // should finalize it now.
+          FinalizeAsyncRead(iu, cqe, posix_handle);
+          continue;
+        }
+
         // Reset cqe data to catch any stray reuse of it
         static_cast<struct io_uring_cqe*>(cqe)->user_data = 0xd5d5d5d5d5d5d5d5;
         io_uring_cqe_seen(iu, cqe);
@@ -1242,16 +1242,23 @@ class PosixFileSystem : public FileSystem {
         // - And finally, if the request to cancel wasn't
         //   found, the cancel request is completed with -ENOENT.
         //
-        // Every handle has to wait for 2 requests completion: original one and
-        // the cancel request which is tracked by PosixHandle::req_count.
-        if (posix_handle->req_count == 2 &&
-            static_cast<Posix_IOHandle*>(io_handles[i]) == posix_handle) {
+        // Every handle being aborted has to wait for 2 requests completion:
+        // original one and the cancel request which is tracked by
+        // PosixHandle::req_count.
+        // Note: We must mark is_finished and invoke the callback for ANY handle
+        // that reaches req_count == 2, not just the one we're currently waiting
+        // for (io_handles[i]). Otherwise, if completions arrive out of order,
+        // we consume another handle's completions without marking it finished,
+        // causing an infinite hang when we later wait for that handle.
+        if (posix_handle->req_count == 2) {
           posix_handle->is_finished = true;
           FSReadRequest req;
           req.status = IOStatus::Aborted();
           posix_handle->cb(req, posix_handle->cb_arg);
 
-          break;
+          if (static_cast<Posix_IOHandle*>(io_handles[i]) == posix_handle) {
+            break;
+          }
         }
       }
     }
@@ -1267,7 +1274,7 @@ class PosixFileSystem : public FileSystem {
   void SupportedOps(int64_t& supported_ops) override {
     supported_ops = 0;
 #if defined(ROCKSDB_IOURING_PRESENT)
-    if (IsIOUringEnabled()) {
+    if (IsIOUringEnabled() && thread_local_async_read_io_urings_) {
       // Underlying FS supports async_io
       supported_ops |= (1 << FSSupportedOps::kAsyncIO);
     }
@@ -1277,7 +1284,8 @@ class PosixFileSystem : public FileSystem {
 
 #if defined(ROCKSDB_IOURING_PRESENT)
   // io_uring instance
-  std::unique_ptr<ThreadLocalPtr> thread_local_io_urings_;
+  std::unique_ptr<ThreadLocalPtr> thread_local_async_read_io_urings_;
+  std::unique_ptr<ThreadLocalPtr> thread_local_multi_read_io_urings_;
 #endif
 
   size_t page_size_;
@@ -1332,12 +1340,12 @@ PosixFileSystem::PosixFileSystem()
       page_size_(getpagesize()),
       allow_non_owner_access_(true) {
 #if defined(ROCKSDB_IOURING_PRESENT)
-  // Test whether IOUring is supported, and if it does, create a managing
-  // object for thread local point so that in the future thread-local
-  // io_uring can be created.
+  // Test whether IOUring is supported with the same flags that ReadAsync and
+  // MultiRead will use at runtime.
   struct io_uring* new_io_uring = CreateIOUring();
   if (new_io_uring != nullptr) {
-    thread_local_io_urings_.reset(new ThreadLocalPtr(DeleteIOUring));
+    thread_local_async_read_io_urings_.reset(new ThreadLocalPtr(DeleteIOUring));
+    thread_local_multi_read_io_urings_.reset(new ThreadLocalPtr(DeleteIOUring));
     delete new_io_uring;
   }
 #endif

env/io_posix.cc

@@ -589,7 +589,8 @@ PosixRandomAccessFile::PosixRandomAccessFile(
     const EnvOptions& options
 #if defined(ROCKSDB_IOURING_PRESENT)
     ,
-    ThreadLocalPtr* thread_local_io_urings
+    ThreadLocalPtr* thread_local_async_read_io_urings,
+    ThreadLocalPtr* thread_local_multi_read_io_urings
 #endif
     )
     : filename_(fname),
@@ -598,7 +599,8 @@ PosixRandomAccessFile::PosixRandomAccessFile(
       logical_sector_size_(logical_block_size)
 #if defined(ROCKSDB_IOURING_PRESENT)
       ,
-      thread_local_io_urings_(thread_local_io_urings)
+      thread_local_async_read_io_urings_(thread_local_async_read_io_urings),
+      thread_local_multi_read_io_urings_(thread_local_multi_read_io_urings)
 #endif
 {
   assert(!options.use_direct_reads || !options.use_mmap_reads);
@@ -659,6 +661,83 @@ IOStatus PosixRandomAccessFile::Read(uint64_t offset, size_t n,
   return s;
 }
 
+// MultiRead: Perform multiple concurrent read requests using io_uring.
+//
+// OVERVIEW:
+// This function batches multiple read requests and submits them concurrently
+// to io_uring for improved I/O performance. It operates synchronously from the
+// caller's perspective (blocks until all reads complete) but uses io_uring's
+// async capabilities internally for parallel I/O execution.
+//
+// IO_URING LIFECYCLE:
+// 1. Preparation Phase:
+//    - Allocate SQEs (Submission Queue Entries) for read requests
+//    - Limited by: min(pending_work, io_uring_sq_space_left(), kIoUringDepth -
+//    inflight)
+//    - Uses io_uring_sq_space_left() to query available SQ slots
+//    - Each SQE is tracked in wrap_cache for completion matching
+//
+// 2. Submission Phase:
+//    - Loop: while io_uring_sq_ready() > 0 (SQEs pending submission)
+//    - Call io_uring_submit_and_wait() to submit SQEs and wait for CQEs
+//    - Handles retryable errors (EINTR, EAGAIN) by continuing
+//    - Breaks on terminal errors (logs error, sets err variable)
+//
+// 3. Completion Phase:
+//    - Non-blocking CQE reaping via io_uring_for_each_cqe()
+//    - Matches CQEs to requests using user_data pointer
+//    - Processes results: updates bytes read, handles partial reads
+//    - Removes completed requests from wrap_cache
+//
+// 4. Loop Iteration:
+//    - Repeats until: all requests submitted AND all completions reaped
+//    - Termination condition: (num_reqs == reqs_off) &&
+//    resubmit_rq_list.empty() && wrap_cache.empty()
+//
+// ERROR HANDLING STRATEGY:
+// - Retryable submission errors (-EINTR, -EAGAIN): Retry submission
+// - Memory pressure (-ENOMEM): Mark memory_pressure_on_submission, attempt
+// recovery
+// - Terminal submission errors: Break, enter teardown path
+// - Retryable CQE errors (-EINTR, -EAGAIN): Add to resubmit_rq_list for retry
+// - Terminal CQE errors: Set ios to IOError, continue processing other CQEs
+// - Teardown path: If SQEs remain unsubmitted after error, reap submitted CQEs,
+//   destroy io_uring instance, return error
+//
+// PARTIAL READ HANDLING:
+// - Short reads (bytes_read < requested): Request added to resubmit_rq_list
+// - finished_len tracks cumulative bytes read across resubmissions
+// - iov.iov_base/iov_len adjusted on each resubmission attempt
+// - UpdateResult() determines if read should be retried based on:
+//   * Direct I/O alignment requirements
+//   * EOF detection
+//   * Error conditions
+//
+// RESUBMISSION LOGIC:
+// - resubmit_rq_list: Requests needing retry (short reads, EINTR/EAGAIN errors)
+// - Prioritized in SQE allocation loop: resubmits before new requests
+// - List cleared after SQE preparation
+// - Requests remain in wrap_cache across resubmissions until fully complete
+//
+// CONCURRENCY CONTROL:
+// - wrap_cache.size(): Tracks total inflight requests (SQ + CQ)
+// - io_uring_sq_ready(): Queries SQEs prepared but not yet submitted
+// - io_uring_sq_space_left(): Queries available SQ slots
+// - Max concurrency: kIoUringDepth (256)
+//
+// ACCOUNTING CORRECTNESS:
+// - Uses io_uring native APIs (io_uring_sq_ready, io_uring_sq_space_left)
+//   instead of manual counters for robustness
+// - wrap_cache is the authoritative source for inflight request tracking
+// - Re-query io_uring_sq_ready() after submission loop to detect
+//   unsubmitted SQEs (indicates submission errors)
+//
+// THREAD SAFETY:
+// - Uses thread-local io_uring instance (thread_local_multi_read_io_urings_)
+// - IORING_SETUP_SINGLE_ISSUER: Only one thread submits to this ring
+// - IORING_SETUP_DEFER_TASKRUN: Task work runs in submitting thread
+// - No cross-thread coordination required
+//
 IOStatus PosixRandomAccessFile::MultiRead(FSReadRequest* reqs, size_t num_reqs,
                                           const IOOptions& options,
                                           IODebugContext* dbg) {
@@ -672,12 +751,13 @@ IOStatus PosixRandomAccessFile::MultiRead(FSReadRequest* reqs, size_t num_reqs,
 
 #if defined(ROCKSDB_IOURING_PRESENT)
   struct io_uring* iu = nullptr;
-  if (thread_local_io_urings_) {
-    iu = static_cast<struct io_uring*>(thread_local_io_urings_->Get());
+  if (thread_local_multi_read_io_urings_) {
+    iu = static_cast<struct io_uring*>(
+        thread_local_multi_read_io_urings_->Get());
     if (iu == nullptr) {
       iu = CreateIOUring();
       if (iu != nullptr) {
-        thread_local_io_urings_->Reset(iu);
+        thread_local_multi_read_io_urings_->Reset(iu);
       }
     }
   }
@@ -688,8 +768,6 @@ IOStatus PosixRandomAccessFile::MultiRead(FSReadRequest* reqs, size_t num_reqs,
     return FSRandomAccessFile::MultiRead(reqs, num_reqs, options, dbg);
   }
 
-  IOStatus ios = IOStatus::OK();
-
   struct WrappedReadRequest {
     FSReadRequest* req;
     struct iovec iov;
@@ -698,118 +776,199 @@ IOStatus PosixRandomAccessFile::MultiRead(FSReadRequest* reqs, size_t num_reqs,
   };
 
   autovector<WrappedReadRequest, 32> req_wraps;
-  autovector<WrappedReadRequest*, 4> incomplete_rq_list;
+  autovector<WrappedReadRequest*, 4> resubmit_rq_list;
   std::unordered_set<WrappedReadRequest*> wrap_cache;
 
   for (size_t i = 0; i < num_reqs; i++) {
     req_wraps.emplace_back(&reqs[i]);
   }
 
+  IOStatus ios = IOStatus::OK();
   size_t reqs_off = 0;
-  while (num_reqs > reqs_off || !incomplete_rq_list.empty()) {
-    size_t this_reqs = (num_reqs - reqs_off) + incomplete_rq_list.size();
-
-    // If requests exceed depth, split it into batches
-    if (this_reqs > kIoUringDepth) {
-      this_reqs = kIoUringDepth;
-    }
-
-    assert(incomplete_rq_list.size() <= this_reqs);
-    for (size_t i = 0; i < this_reqs; i++) {
-      WrappedReadRequest* rep_to_submit;
-      if (i < incomplete_rq_list.size()) {
-        rep_to_submit = incomplete_rq_list[i];
+  while ((num_reqs > reqs_off) || !resubmit_rq_list.empty() ||
+         !wrap_cache.empty()) {
+    assert(resubmit_rq_list.size() + wrap_cache.size() <= kIoUringDepth);
+    // Total number of requests that still need to be submitted, includes:
+    //
+    //  1) requests NOT yet submitted (num_reqs - reqs_off)
+    //  2) requests on resubmission list (resubmit_rq_list)
+    //
+    // capped by min of the # of remaining entries in IO ring submission queue
+    // and the max IO ring depth less the inflight requests.
+    size_t new_sqe_reqs_count = std::min({
+        num_reqs - reqs_off + resubmit_rq_list.size(),
+        static_cast<size_t>(io_uring_sq_space_left(iu)),
+        kIoUringDepth - wrap_cache.size()  // queue depth less inflight requests
+    });
+    for (size_t i = 0; i < new_sqe_reqs_count; i++) {
+      WrappedReadRequest* req;
+      if (i < resubmit_rq_list.size()) {
+        req = resubmit_rq_list[i];
       } else {
-        rep_to_submit = &req_wraps[reqs_off++];
+        req = &req_wraps[reqs_off++];
       }
-      assert(rep_to_submit->req->len > rep_to_submit->finished_len);
-      rep_to_submit->iov.iov_base =
-          rep_to_submit->req->scratch + rep_to_submit->finished_len;
-      rep_to_submit->iov.iov_len =
-          rep_to_submit->req->len - rep_to_submit->finished_len;
+      assert(req->req->len > req->finished_len);
+      req->iov.iov_base = req->req->scratch + req->finished_len;
+      req->iov.iov_len = req->req->len - req->finished_len;
 
       struct io_uring_sqe* sqe;
       sqe = io_uring_get_sqe(iu);
-      io_uring_prep_readv(
-          sqe, fd_, &rep_to_submit->iov, 1,
-          rep_to_submit->req->offset + rep_to_submit->finished_len);
-      io_uring_sqe_set_data(sqe, rep_to_submit);
-      wrap_cache.emplace(rep_to_submit);
+      // NULL is unexpected as we do maintain proper ring accounting.
+      assert(sqe);
+      io_uring_prep_readv(sqe, fd_, &req->iov, 1,
+                          req->req->offset + req->finished_len);
+      io_uring_sqe_set_data(sqe, req);
+      wrap_cache.emplace(req);
     }
-    incomplete_rq_list.clear();
+    resubmit_rq_list.clear();
 
-    ssize_t ret =
-        io_uring_submit_and_wait(iu, static_cast<unsigned int>(this_reqs));
-    TEST_SYNC_POINT_CALLBACK(
-        "PosixRandomAccessFile::MultiRead:io_uring_submit_and_wait:return1",
-        &ret);
-    TEST_SYNC_POINT_CALLBACK(
-        "PosixRandomAccessFile::MultiRead:io_uring_submit_and_wait:return2",
-        iu);
-
-    if (static_cast<size_t>(ret) != this_reqs) {
-      fprintf(stderr, "ret = %ld this_reqs: %ld\n", (long)ret, (long)this_reqs);
-      // If error happens and we submitted fewer than expected, it is an
-      // exception case and we don't retry here. We should still consume
-      // what is is submitted in the ring.
-      for (ssize_t i = 0; i < ret; i++) {
-        struct io_uring_cqe* cqe = nullptr;
-        io_uring_wait_cqe(iu, &cqe);
-        if (cqe != nullptr) {
-          io_uring_cqe_seen(iu, cqe);
+    struct io_uring_cqe* cqe = nullptr;
+    unsigned head;
+    ssize_t err = 0;
+    bool memory_pressure_on_submission = false;
+    unsigned reqs_pending_submission;
+    unsigned reqs_submitted = 0;
+    while ((reqs_pending_submission = io_uring_sq_ready(iu))) {
+      // MultiRead is synchronous in nature. io_uring_submit_and_wait provides
+      // batching semantics (submit + best effort wait in one syscall), while
+      // io_uring_submit enables async producer/consumer semantics (submit
+      // only, requires separate reaping). We chose batching approach to
+      // reduce the volume of syscalls and context switches.
+      ssize_t ret = io_uring_submit_and_wait(iu, reqs_pending_submission);
+      if (ret < 0) {
+        if (-EINTR == ret || -EAGAIN == ret) {
+          // Submission failed due to rare, retryable syscall error. Try again.
+          continue;
         }
-      }
-      return IOStatus::IOError("io_uring_submit_and_wait() requested " +
-                               std::to_string(this_reqs) + " but returned " +
-                               std::to_string(ret));
-    }
-
-    for (size_t i = 0; i < this_reqs; i++) {
-      struct io_uring_cqe* cqe = nullptr;
-      WrappedReadRequest* req_wrap;
-
-      // We could use the peek variant here, but this seems safer in terms
-      // of our initial wait not reaping all completions
-      ret = io_uring_wait_cqe(iu, &cqe);
-      TEST_SYNC_POINT_CALLBACK(
-          "PosixRandomAccessFile::MultiRead:io_uring_wait_cqe:return", &ret);
-      if (ret) {
-        ios = IOStatus::IOError("io_uring_wait_cqe() returns " +
-                                std::to_string(ret));
-
-        if (cqe != nullptr) {
-          io_uring_cqe_seen(iu, cqe);
+        if (-ENOMEM == ret) {
+          fprintf(stderr,
+                  "PosixRandomAccessFile::MultiRead: io_uring_submit_and_wait "
+                  "experienced terse memory condition.\n");
+          // Best effort to reclaim resources in terse condition.
+          memory_pressure_on_submission = true;
+        } else {
+          fprintf(stderr,
+                  "PosixRandomAccessFile::MultiRead: "
+                  "io_uring_submit_and_wait returned terminal error: %zd.\n",
+                  ret);
+          err = ret;
         }
-        continue;
+        break;
       }
-
-      req_wrap = static_cast<WrappedReadRequest*>(io_uring_cqe_get_data(cqe));
-      // Reset cqe data to catch any stray reuse of it
-      static_cast<struct io_uring_cqe*>(cqe)->user_data = 0xd5d5d5d5d5d5d5d5;
-      // Check that we got a valid unique cqe data
-      auto wrap_check = wrap_cache.find(req_wrap);
-      if (wrap_check == wrap_cache.end()) {
+      if (0 == ret) {
+        // This scenario is unexpected for any modern kernel!
+        // We deliberately error out to avoid bugs around infinite loops.
         fprintf(stderr,
                 "PosixRandomAccessFile::MultiRead: "
-                "Bad cqe data from IO uring - %p\n",
-                req_wrap);
-        port::PrintStack();
-        ios = IOStatus::IOError("io_uring_cqe_get_data() returned " +
-                                std::to_string((uint64_t)req_wrap));
-        continue;
+                "io_uring_submit_and_wait returned 0 submissions!\n");
+        break;
       }
-      wrap_cache.erase(wrap_check);
+      reqs_submitted += static_cast<unsigned int>(ret);
+    };
+    reqs_pending_submission = io_uring_sq_ready(iu);
 
-      FSReadRequest* req = req_wrap->req;
-      size_t bytes_read = 0;
-      bool read_again = false;
-      UpdateResult(cqe, filename_, req->len, req_wrap->iov.iov_len,
-                   false /*async_read*/, use_direct_io(),
-                   GetRequiredBufferAlignment(), req_wrap->finished_len, req,
-                   bytes_read, read_again);
-      int32_t res = cqe->res;
-      if (res >= 0) {
-        if (bytes_read == 0) {
+    TEST_SYNC_POINT_CALLBACK(
+        "PosixRandomAccessFile::MultiRead:io_uring_sq_ready:return1",
+        &reqs_pending_submission);
+
+    // Error occurred or IO uring stopped submitting outstanding requests.
+    if (reqs_pending_submission && !memory_pressure_on_submission) {
+      // IO ring is initialized once in thread-local variable and then reused
+      // to handle the consecutive MultiRead API calls. Therefore, it's crucial
+      // to reap all the submitted requests.
+      //
+      // NOTE: Loop will run indefinitely until we reap all the completions!!!
+      size_t nr = 0;
+      assert(reqs_pending_submission <= wrap_cache.size());
+      size_t nr_await_cqe = wrap_cache.size() - reqs_pending_submission;
+      while (nr < nr_await_cqe) {
+        // blocking
+        io_uring_wait_cqes(iu, &cqe,
+                           static_cast<unsigned int>(nr_await_cqe - nr),
+                           nullptr, nullptr);
+        size_t reaped_cqe_count = 0;
+        io_uring_for_each_cqe(iu, head, cqe) { reaped_cqe_count++; }
+        if (reaped_cqe_count > 0) {
+          io_uring_cq_advance(iu, static_cast<unsigned int>(reaped_cqe_count));
+          nr += reaped_cqe_count;
+        }
+      }
+
+      TEST_SYNC_POINT_CALLBACK(
+          "PosixRandomAccessFile::MultiRead:io_uring_submit_and_wait:return2",
+          iu);
+
+      // While all the submitted completions have been reaped successfully,
+      // IO ring submission queue still contains at least one non-submitted
+      // request. Destroy io_uring (discards unsubmitted SQEs).
+      //
+      // NOTE: This is a rare scenario and should not happen in normal cases.
+      //       Hence, this should NOT materially impact the performance metrics.
+      io_uring_queue_exit(iu);
+      delete iu;
+      thread_local_multi_read_io_urings_->Reset(nullptr);
+
+      if (err < 0) {
+        return IOStatus::IOError(
+            "io_uring_submit_and_wait() failed with an error " +
+            std::to_string(err));
+      }
+      return IOStatus::IOError(
+          "io_uring_submit_and_wait() requested " +
+          std::to_string(reqs_submitted + reqs_pending_submission) +
+          " but returned " + std::to_string(reqs_submitted));
+    }
+
+    if ((0 == reqs_submitted) && wrap_cache.size() > reqs_pending_submission) {
+      // If no requests have been submitted and there is at least one request
+      // pending completion, wait for at least one completion to arrive.
+      // This is a guardrail to prevent the busy CPU loops.
+      //
+      // NOTE: it's not really a tight CPU-burning loop in the traditional sense
+      // as it's naturally throttled by the io_uring_submit_and_wait() syscall.
+      io_uring_wait_cqe(iu, &cqe);
+    }
+
+    unsigned int nr = 0;
+    io_uring_for_each_cqe(iu, head, cqe) {  // non-blocking
+      if (cqe->user_data) {  // non-discarded, valid user data only!
+        nr++;
+        WrappedReadRequest* req_wrap =
+            static_cast<WrappedReadRequest*>(io_uring_cqe_get_data(cqe));
+        // Reset cqe data to catch any stray reuse of it
+        static_cast<struct io_uring_cqe*>(cqe)->user_data = 0xd5d5d5d5d5d5d5d5;
+        // Check that we got a valid unique cqe data
+        auto wrap_check = wrap_cache.find(req_wrap);
+        if (wrap_check == wrap_cache.end()) {
+          fprintf(stderr,
+                  "PosixRandomAccessFile::MultiRead: "
+                  "Bad cqe data from IO uring - %p\n",
+                  req_wrap);
+          port::PrintStack();
+          ios = IOStatus::IOError("io_uring_cqe_get_data() returned " +
+                                  std::to_string((uint64_t)req_wrap));
+          continue;
+        }
+        wrap_cache.erase(wrap_check);
+        if (cqe->res < 0) {
+          if (-EINTR == cqe->res || -EAGAIN == cqe->res) {
+            resubmit_rq_list.push_back(req_wrap);
+          } else {
+            ios = IOStatus::IOError("io_uring_for_each_cqe() returns " +
+                                    std::to_string(cqe->res));
+          }
+          continue;
+        }
+        // cqe->res >= 0
+        FSReadRequest* req = req_wrap->req;
+        size_t bytes_read = 0;
+        bool read_again = false;
+        UpdateResult(cqe, filename_, req->len, req_wrap->iov.iov_len,
+                     false /*async_read*/, use_direct_io(),
+                     GetRequiredBufferAlignment(), req_wrap->finished_len, req,
+                     bytes_read, read_again);
+
+        if (0 == bytes_read) {
           if (read_again) {
             Slice tmp_slice;
             req->status =
@@ -819,14 +978,15 @@ IOStatus PosixRandomAccessFile::MultiRead(FSReadRequest* reqs, size_t num_reqs,
             req->result =
                 Slice(req->scratch, req_wrap->finished_len + tmp_slice.size());
           }
-          // else It means EOF so no need to do anything.
+          // else it means EOF so no need to do anything.
         } else if (bytes_read < req_wrap->iov.iov_len) {
-          incomplete_rq_list.push_back(req_wrap);
+          resubmit_rq_list.push_back(req_wrap);
         }
       }
-      io_uring_cqe_seen(iu, cqe);
     }
-    wrap_cache.clear();
+    if (nr > 0) {
+      io_uring_cq_advance(iu, nr);
+    }
   }
   return ios;
 #else
@@ -923,12 +1083,13 @@ IOStatus PosixRandomAccessFile::ReadAsync(
 #if defined(ROCKSDB_IOURING_PRESENT)
   // io_uring_queue_init.
   struct io_uring* iu = nullptr;
-  if (thread_local_io_urings_) {
-    iu = static_cast<struct io_uring*>(thread_local_io_urings_->Get());
+  if (thread_local_async_read_io_urings_) {
+    iu = static_cast<struct io_uring*>(
+        thread_local_async_read_io_urings_->Get());
     if (iu == nullptr) {
       iu = CreateIOUring();
       if (iu != nullptr) {
-        thread_local_io_urings_->Reset(iu);
+        thread_local_async_read_io_urings_->Reset(iu);
       }
     }
   }
@@ -966,11 +1127,35 @@ IOStatus PosixRandomAccessFile::ReadAsync(
   io_uring_sqe_set_data(sqe, posix_handle);
 
   // Step 4: io_uring_submit
-  ssize_t ret = io_uring_submit(iu);
-  if (ret < 0) {
-    fprintf(stderr, "io_uring_submit error: %ld\n", long(ret));
-    return IOStatus::IOError("io_uring_submit() requested but returned " +
-                             std::to_string(ret));
+  ssize_t ret;
+  do {
+    ret = io_uring_submit(iu);
+    if (ret < 0) {
+      if (-EINTR == ret || -EAGAIN == ret) {
+        // Submission failed due to transient error. Try again.
+        continue;
+      }
+      fprintf(stderr,
+              "PosixRandomAccessFile::ReadAsync: "
+              "io_uring_submit returned terminal error = %zd\n",
+              ret);
+      break;
+    }
+    if (0 == ret) {
+      // Unexpected. Will be reported as error.
+      break;
+    }
+  } while (ret < 1);
+  if (ret <= 0) {
+    return IOStatus::IOError(
+        "PosixRandomAccessFile::ReadAsync: io_uring_submit() returned " +
+        std::to_string(ret));
+  }
+  if (ret > 1) {
+    fprintf(stderr,
+            "PosixRandomAccessFile::ReadAsync: "
+            "io_uring_submit() returned = %zd\n",
+            ret);
   }
   return IOStatus::OK();
 #else

env/io_posix.h

@@ -11,6 +11,16 @@
 #if defined(ROCKSDB_IOURING_PRESENT)
 #include <liburing.h>
 #include <sys/uio.h>
+
+// Compatibility defines for io_uring flags that may not be present in older
+// kernel headers. These values are fixed and won't change, so it's safe to
+// define them even if the running kernel doesn't support them.
+#ifndef IORING_SETUP_SINGLE_ISSUER
+#define IORING_SETUP_SINGLE_ISSUER (1U << 12)
+#endif
+#ifndef IORING_SETUP_DEFER_TASKRUN
+#define IORING_SETUP_DEFER_TASKRUN (1U << 13)
+#endif
 #endif
 #include <unistd.h>
 
@@ -117,6 +127,7 @@ struct Posix_IOHandle {
         use_direct_io(_use_direct_io),
         alignment(_alignment),
         is_finished(false),
+        is_being_aborted(false),
         req_count(0) {}
 
   struct iovec iov;
@@ -129,6 +140,10 @@ struct Posix_IOHandle {
   bool use_direct_io;
   size_t alignment;
   bool is_finished;
+  // is_being_aborted is set by AbortIO when a cancel request is submitted.
+  // Used to distinguish between aborted handles (expect 2 completions) and
+  // non-aborted handles (expect 1 completion) when processing completions.
+  bool is_being_aborted;
   // req_count is used by AbortIO API to keep track of number of requests.
   uint32_t req_count;
 };
@@ -187,6 +202,27 @@ inline void UpdateResult(struct io_uring_cqe* cqe, const std::string& file_name,
   (void)len;
 #endif
 }
+
+// Finalize a completed async read request.
+// Processes the CQE result, marks the handle as finished, and invokes the
+// callback. This is shared between Poll and AbortIO (for non-aborted handles).
+inline void FinalizeAsyncRead(struct io_uring* iu, struct io_uring_cqe* cqe,
+                              Posix_IOHandle* posix_handle) {
+  FSReadRequest req;
+  req.scratch = posix_handle->scratch;
+  req.offset = posix_handle->offset;
+  req.len = posix_handle->len;
+
+  size_t finished_len = 0;
+  size_t bytes_read = 0;
+  bool read_again = false;
+  UpdateResult(cqe, "", req.len, posix_handle->iov.iov_len, true /*async_read*/,
+               posix_handle->use_direct_io, posix_handle->alignment,
+               finished_len, &req, bytes_read, read_again);
+  posix_handle->is_finished = true;
+  io_uring_cqe_seen(iu, cqe);
+  posix_handle->cb(req, posix_handle->cb_arg);
+}
 #endif
 
 #ifdef OS_LINUX
@@ -299,7 +335,10 @@ inline void DeleteIOUring(void* p) {
 
 inline struct io_uring* CreateIOUring() {
   struct io_uring* new_io_uring = new struct io_uring;
-  int ret = io_uring_queue_init(kIoUringDepth, new_io_uring, 0);
+  unsigned int flags = 0;
+  flags |= IORING_SETUP_SINGLE_ISSUER;
+  flags |= IORING_SETUP_DEFER_TASKRUN;
+  int ret = io_uring_queue_init(kIoUringDepth, new_io_uring, flags);
   if (ret) {
     delete new_io_uring;
     new_io_uring = nullptr;
@@ -315,7 +354,8 @@ class PosixRandomAccessFile : public FSRandomAccessFile {
   bool use_direct_io_;
   size_t logical_sector_size_;
 #if defined(ROCKSDB_IOURING_PRESENT)
-  ThreadLocalPtr* thread_local_io_urings_;
+  ThreadLocalPtr* thread_local_async_read_io_urings_;
+  ThreadLocalPtr* thread_local_multi_read_io_urings_;
 #endif
 
  public:
@@ -323,7 +363,8 @@ class PosixRandomAccessFile : public FSRandomAccessFile {
                         size_t logical_block_size, const EnvOptions& options
 #if defined(ROCKSDB_IOURING_PRESENT)
                         ,
-                        ThreadLocalPtr* thread_local_io_urings
+                        ThreadLocalPtr* thread_local_async_read_io_urings,
+                        ThreadLocalPtr* thread_local_multi_read_io_urings
 #endif
   );
   virtual ~PosixRandomAccessFile();

file/file_prefetch_buffer.cc

@@ -351,7 +351,7 @@ void FilePrefetchBuffer::ClearOutdatedData(uint64_t offset, size_t length) {
   assert(IsBufferQueueEmpty() || buf->IsOffsetInBuffer(offset));
 }
 
-void FilePrefetchBuffer::PollIfNeeded(uint64_t offset, size_t length) {
+Status FilePrefetchBuffer::PollIfNeeded(uint64_t offset, size_t length) {
   BufferInfo* buf = GetFirstBuffer();
 
   if (buf->async_read_in_progress_ && fs_ != nullptr) {
@@ -362,7 +362,16 @@ void FilePrefetchBuffer::PollIfNeeded(uint64_t offset, size_t length) {
       std::vector<void*> handles;
       handles.emplace_back(buf->io_handle_);
       StopWatch sw(clock_, stats_, POLL_WAIT_MICROS);
-      fs_->Poll(handles, 1).PermitUncheckedError();
+      IOStatus io_s = fs_->Poll(handles, 1);
+      // Allow tests to inject Poll errors
+      TEST_SYNC_POINT_CALLBACK("FilePrefetchBuffer::PollIfNeeded:IOStatus",
+                               &io_s);
+      if (!io_s.ok()) {
+        // On Poll failure, clean up the handle and abort.
+        // DestroyAndClearIOHandle also sets async_read_in_progress_ to false.
+        DestroyAndClearIOHandle(buf);
+        return io_s;
+      }
     }
 
     // Reset and Release io_handle after the Poll API as request has been
@@ -373,6 +382,7 @@ void FilePrefetchBuffer::PollIfNeeded(uint64_t offset, size_t length) {
   // Always call outdated data after Poll as Buffers might be out of sync w.r.t
   // offset and length.
   ClearOutdatedData(offset, length);
+  return Status::OK();
 }
 
 // ReadAheadSizeTuning API calls readaheadsize_cb_
@@ -511,7 +521,10 @@ Status FilePrefetchBuffer::HandleOverlappingAsyncData(
   // by Seek, but the next access is at another offset.
   if (buf->async_read_in_progress_ &&
       buf->IsOffsetInBufferWithAsyncProgress(offset)) {
-    PollIfNeeded(offset, length);
+    Status poll_status = PollIfNeeded(offset, length);
+    if (!poll_status.ok()) {
+      return poll_status;
+    }
   }
 
   if (IsBufferQueueEmpty() || NumBuffersAllocated() == 1) {
@@ -646,7 +659,10 @@ Status FilePrefetchBuffer::PrefetchInternal(const IOOptions& opts,
       return s;
     }
   } else {
-    PollIfNeeded(tmp_offset, tmp_length);
+    Status poll_status = PollIfNeeded(tmp_offset, tmp_length);
+    if (!poll_status.ok()) {
+      return poll_status;
+    }
   }
 
   AllocateBufferIfEmpty();

file/file_prefetch_buffer.h

@@ -93,8 +93,8 @@ struct BufferInfo {
   //
   // For example - if end offset of previous buffer was 100 and because of
   // readahead_size optimization, end_offset was trimmed to 60. Then for next
-  // prefetch call, start_offset should be intialized to 100 i.e  start_offset =
-  // buf->initial_end_offset_.
+  // prefetch call, start_offset should be initialized to 100 i.e  start_offset
+  // = buf->initial_end_offset_.
   uint64_t initial_end_offset_ = 0;
 
   bool IsDataBlockInBuffer(uint64_t offset, size_t length) {
@@ -155,7 +155,7 @@ enum class FilePrefetchBufferUsage {
 // When reusing the file system allocated buffer, overlap_buf_ is used if the
 // main buffer only contains part of the requested data. It is returned to
 // the caller after the remaining data is fetched.
-// If num_buffers_ > 1, then the data is prefetched asynchronosuly in the
+// If num_buffers_ > 1, then the data is prefetched asynchronously in the
 // buffers whenever the data is consumed from the buffers and that buffer is
 // freed.
 // If num_buffers > 1, then requested data can be overlapping between 2 buffers.
@@ -431,7 +431,7 @@ class FilePrefetchBuffer {
   void ClearOutdatedData(uint64_t offset, size_t len);
 
   // It calls Poll API to check for any pending asynchronous request.
-  void PollIfNeeded(uint64_t offset, size_t len);
+  Status PollIfNeeded(uint64_t offset, size_t len);
 
   Status PrefetchInternal(const IOOptions& opts, RandomAccessFileReader* reader,
                           uint64_t offset, size_t length, size_t readahead_size,

file/file_util.cc

@@ -178,7 +178,8 @@ IOStatus GenerateOneFileChecksum(
     std::string* file_checksum_func_name,
     size_t verify_checksums_readahead_size, bool /*allow_mmap_reads*/,
     std::shared_ptr<IOTracer>& io_tracer, RateLimiter* rate_limiter,
-    const ReadOptions& read_options, Statistics* stats, SystemClock* clock) {
+    const ReadOptions& read_options, Statistics* stats, SystemClock* clock,
+    const FileOptions& file_options) {
   if (checksum_factory == nullptr) {
     return IOStatus::InvalidArgument("Checksum factory is invalid");
   }
@@ -218,7 +219,12 @@ IOStatus GenerateOneFileChecksum(
   std::unique_ptr<RandomAccessFileReader> reader;
   {
     std::unique_ptr<FSRandomAccessFile> r_file;
-    io_s = fs->NewRandomAccessFile(file_path, FileOptions(), &r_file, nullptr);
+    FileOptions fopts = file_options;
+    if (fopts.file_checksum.empty()) {
+      // No expected checksum is known — this is a from-scratch computation.
+      fopts.file_checksum_func_name = kNoFileChecksumFuncName;
+    }
+    io_s = fs->NewRandomAccessFile(file_path, fopts, &r_file, nullptr);
     if (!io_s.ok()) {
       return io_s;
     }

file/file_util.h

@@ -83,7 +83,8 @@ IOStatus GenerateOneFileChecksum(
     std::string* file_checksum_func_name,
     size_t verify_checksums_readahead_size, bool allow_mmap_reads,
     std::shared_ptr<IOTracer>& io_tracer, RateLimiter* rate_limiter,
-    const ReadOptions& read_options, Statistics* stats, SystemClock* clock);
+    const ReadOptions& read_options, Statistics* stats, SystemClock* clock,
+    const FileOptions& file_options);
 
 inline IOStatus PrepareIOFromReadOptions(const ReadOptions& ro,
                                          SystemClock* clock, IOOptions& opts,