Merge remote-tracking branch 'internal/main' into merge_internal

* enable level_compaction_dynamic_level_bytes after flush, and flush expires

* update debug reload for flash

* update debug reload for flash complete

* missing forward declare

* commit existing changes then track changes for debug reload

* missing args

* commitChanges is conditional

Co-authored-by: John Sully <jsully@snapchat.com>

machamp_scripts/build.sh

@@ -2,7 +2,7 @@
 
 # make the build
 git submodule init && git submodule update
-make BUILD_TLS=yes -j$(nproc) KEYDB_CFLAGS='-Werror' KEYDB_CXXFLAGS='-Werror'
+make BUILD_TLS=yes ENABLE_FLASH=yes -j$(nproc) KEYDB_CFLAGS='-Werror' KEYDB_CXXFLAGS='-Werror'
 
 # gen-cert
 ./utils/gen-test-certs.sh

pkg/docker/Dockerfile

@@ -53,7 +53,7 @@ RUN set -eux; \
         grep -E '^ *createBoolConfig[(]"protected-mode",.*, *1 *,.*[)],$' ./src/config.cpp; \
         sed -ri 's!^( *createBoolConfig[(]"protected-mode",.*, *)1( *,.*[)],)$!\10\2!' ./src/config.cpp; \
         grep -E '^ *createBoolConfig[(]"protected-mode",.*, *0 *,.*[)],$' ./src/config.cpp; \
-        make -j$(nproc) BUILD_TLS=yes; \
+        make -j$(nproc) BUILD_TLS=yes ENABLE_FLASH=yes; \
         cd src; \
         strip keydb-cli keydb-benchmark keydb-check-rdb keydb-check-aof keydb-diagnostic-tool keydb-sentinel keydb-server; \
         mv keydb-server keydb-cli keydb-benchmark keydb-check-rdb keydb-check-aof keydb-diagnostic-tool keydb-sentinel /usr/local/bin/; \
@@ -83,6 +83,7 @@ RUN \
         sed -i 's/^\(logfile .*\)$/# \1/' /etc/keydb/keydb.conf; \
         sed -i 's/protected-mode yes/protected-mode no/g' /etc/keydb/keydb.conf; \
         sed -i 's/^\(bind .*\)$/# \1/' /etc/keydb/keydb.conf; \
+        cd /usr/local/bin; \
         ln -s keydb-cli redis-cli; \
         cd /etc/keydb; \
         ln -s keydb.conf redis.conf; \

pkg/docker/Dockerfile_Alpine

@@ -5,7 +5,7 @@ RUN mkdir -p /etc/keydb
 ARG BRANCH
 RUN set -eux; \
         \
-        apk add --no-cache su-exec; \
+        apk add --no-cache su-exec tini; \
         apk add --no-cache --virtual .build-deps \
                 coreutils \
                 gcc \
@@ -32,7 +32,7 @@ RUN set -eux; \
         grep -E '^ *createBoolConfig[(]"protected-mode",.*, *1 *,.*[)],$' ./src/config.cpp; \
         sed -ri 's!^( *createBoolConfig[(]"protected-mode",.*, *)1( *,.*[)],)$!\10\2!' ./src/config.cpp; \
         grep -E '^ *createBoolConfig[(]"protected-mode",.*, *0 *,.*[)],$' ./src/config.cpp; \
-        make -j$(nproc) BUILD_TLS=yes; \
+        make -j$(nproc) BUILD_TLS=yes ENABLE_FLASH=yes; \
         cd src; \
         strip keydb-cli keydb-benchmark keydb-check-rdb keydb-check-aof keydb-diagnostic-tool keydb-sentinel keydb-server; \
         mv keydb-server keydb-cli keydb-benchmark keydb-check-rdb keydb-check-aof keydb-diagnostic-tool keydb-sentinel /usr/local/bin/; \
@@ -54,6 +54,7 @@ RUN set -eux; \
         sed -i 's/^\(logfile .*\)$/# \1/' /etc/keydb/keydb.conf; \
         sed -i 's/protected-mode yes/protected-mode no/g' /etc/keydb/keydb.conf; \
         sed -i 's/^\(bind .*\)$/# \1/' /etc/keydb/keydb.conf; \
+        cd /usr/local/bin; \
         ln -s keydb-cli redis-cli; \
         cd /etc/keydb; \
         ln -s keydb.conf redis.conf; \
@@ -76,6 +77,6 @@ RUN set -eux; \
         chmod +x /usr/local/bin/docker-entrypoint.sh
 VOLUME /data
 WORKDIR /data
-ENTRYPOINT ["docker-entrypoint.sh"]
+ENTRYPOINT ["tini", "--", "docker-entrypoint.sh"]
 EXPOSE 6379
 CMD ["keydb-server", "/etc/keydb/keydb.conf"]

src/IStorage.h

@@ -1,6 +1,7 @@
 #pragma once
 #include <functional>
 #include "sds.h"
+#include <string>
 
 #define METADATA_DB_IDENTIFIER "c299fde0-6d42-4ec4-b939-34f680ffe39f"
 
@@ -43,6 +44,11 @@ public:
         endWriteBatch();
     }
 
+    virtual std::vector<std::string> getExpirationCandidates(unsigned int count) = 0;
+    virtual std::vector<std::string> getEvictionCandidates(unsigned int count) = 0;
+    virtual void setExpire(const char *key, size_t cchKey, long long expire) = 0;
+    virtual void removeExpire(const char *key, size_t cchKey, long long expire) = 0;
+
     virtual void beginWriteBatch() {} // NOP
     virtual void endWriteBatch() {} // NOP
 

src/StorageCache.cpp

@@ -84,19 +84,31 @@ void StorageCache::cacheKey(const char *rgch, size_t cch)
 
 bool StorageCache::erase(sds key)
 {
+    unsigned long long when = 0;
+    m_spstorage->retrieve(key, sdslen(key), [&when](const char *, size_t, const void * data, size_t cbdata) {
+        auto e = deserializeExpire((const char *)data, cbdata, nullptr);
+        if (e != nullptr)
+            when = e->when();
+    });
     bool result = m_spstorage->erase(key, sdslen(key));
     std::unique_lock<fastlock> ul(m_lock);
-    if (result && m_pdict != nullptr)
+    if (result)
     {
-        uint64_t hash = dictSdsHash(key);
-        dictEntry *de = dictFind(m_pdict, reinterpret_cast<void*>(hash));
-        serverAssert(de != nullptr);
-        de->v.s64--;
-        serverAssert(de->v.s64 >= 0);
-        if (de->v.s64 == 0) {
-            dictDelete(m_pdict, reinterpret_cast<void*>(hash));
-        } else {
-            m_collisionCount--;
+        if (m_pdict != nullptr)
+        {
+            uint64_t hash = dictSdsHash(key);
+            dictEntry *de = dictFind(m_pdict, reinterpret_cast<void*>(hash));
+            serverAssert(de != nullptr);
+            de->v.s64--;
+            serverAssert(de->v.s64 >= 0);
+            if (de->v.s64 == 0) {
+                dictDelete(m_pdict, reinterpret_cast<void*>(hash));
+            } else {
+                m_collisionCount--;
+            }
+        }
+        if (when != 0) {
+            m_spstorage->removeExpire(key, sdslen(key), when);
         }
     }
     return result;
@@ -111,6 +123,9 @@ void StorageCache::insert(sds key, const void *data, size_t cbdata, bool fOverwr
     }
     ul.unlock();
     m_spstorage->insert(key, sdslen(key), (void*)data, cbdata, fOverwrite);
+    auto e = deserializeExpire((const char *)data, cbdata, nullptr);
+    if (e != nullptr)
+        m_spstorage->setExpire(key, sdslen(key), e->when());
 }
 
 long _dictKeyIndex(dict *d, const void *key, uint64_t hash, dictEntry **existing);
@@ -119,13 +134,18 @@ void StorageCache::bulkInsert(char **rgkeys, size_t *rgcbkeys, char **rgvals, si
     std::vector<dictEntry*> vechashes;
     if (m_pdict != nullptr) {
         vechashes.reserve(celem);
+    }
 
-        for (size_t ielem = 0; ielem < celem; ++ielem) {
+    for (size_t ielem = 0; ielem < celem; ++ielem) {
+        if (m_pdict != nullptr) {
             dictEntry *de = (dictEntry*)zmalloc(sizeof(dictEntry));
             de->key = (void*)dictGenHashFunction(rgkeys[ielem], (int)rgcbkeys[ielem]);
             de->v.u64 = 1;
             vechashes.push_back(de);
         }
+        auto e = deserializeExpire(rgvals[ielem], rgcbvals[ielem], nullptr);
+        if (e != nullptr)
+            m_spstorage->setExpire(rgkeys[ielem], rgcbkeys[ielem], e->when());
     }
 
     std::unique_lock<fastlock> ul(m_lock);

src/StorageCache.h

@@ -51,11 +51,18 @@ public:
     bool enumerate(IStorage::callback fn) const { return m_spstorage->enumerate(fn); }
     bool enumerate_hashslot(IStorage::callback fn, unsigned int hashslot) const { return m_spstorage->enumerate_hashslot(fn, hashslot); }
 
+    std::vector<std::string> getExpirationCandidates(unsigned int count) { return m_spstorage->getExpirationCandidates(count); }
+    std::vector<std::string> getEvictionCandidates(unsigned int count) { return m_spstorage->getEvictionCandidates(count); }
+    void setExpire(const char *key, size_t cchKey, long long expire) { m_spstorage->setExpire(key, cchKey, expire); }
+    void removeExpire(const char *key, size_t cchKey, long long expire) { m_spstorage->removeExpire(key, cchKey, expire); }
+
     void beginWriteBatch();
     void endWriteBatch() { m_spstorage->endWriteBatch(); }
     void batch_lock() { return m_spstorage->batch_lock(); }
     void batch_unlock() { return m_spstorage->batch_unlock(); }
 
+    void flush() { m_spstorage->flush(); }
+
     size_t count() const;
 
     const StorageCache *clone();

src/ae.cpp

@@ -113,6 +113,8 @@ enum class AE_ASYNC_OP
     PostCppFunction,
     DeleteFileEvent,
     CreateFileEvent,
+    PostAsynDBFunction,
+
 };
 
 struct aeCommand
@@ -125,6 +127,8 @@ struct aeCommand
         aePostFunctionProc *proc;
         aeFileProc *fproc;
         std::function<void()> *pfn;
+        aePostFunctionTokenProc* tproc;
+
     };
     void *clientData;
 };
@@ -175,6 +179,17 @@ void aeProcessCmd(aeEventLoop *eventLoop, int fd, void *, int )
             (*cmd.pfn)();
 
             delete cmd.pfn;
+            break;
+        }
+        case AE_ASYNC_OP::PostAsynDBFunction:
+        {   //added to support async api IStorage
+           if (cmd.fLock && !ulock.owns_lock()) {
+                g_forkLock.releaseRead();
+                ulock.lock();
+                g_forkLock.acquireRead();
+            }  
+            ((aePostFunctionTokenProc*)cmd.tproc)(eventLoop,(StorageToken*)cmd.clientData);
+            break; 
         }
             break;
         }
@@ -256,6 +271,20 @@ int aePostFunction(aeEventLoop *eventLoop, aePostFunctionProc *proc, void *arg)
     return AE_OK;
 }
 
+int aePostFunction(aeEventLoop *eventLoop, aePostFunctionTokenProc *proc, StorageToken *token)
+{
+  //added to support async api IStorage
+    aeCommand cmd = {};
+    cmd.op = AE_ASYNC_OP::PostAsynDBFunction; 
+    cmd.tproc = proc;
+    cmd.clientData = (void*)token; 
+    cmd.fLock = true;
+    auto size = write(eventLoop->fdCmdWrite, &cmd, sizeof(cmd));
+    if (size != sizeof(cmd))
+        return AE_ERR;
+    return AE_OK;
+}
+
 int aePostFunction(aeEventLoop *eventLoop, std::function<void()> fn, bool fLock, bool fForceQueue)
 {
     if (eventLoop == g_eventLoopThisThread && !fForceQueue)
@@ -861,7 +890,7 @@ void aeReleaseForkLock()
 
 void aeForkLockInChild()
 {
-    g_forkLock.setNotify(false);
+    g_forkLock.setMulti(false);
 }
 
 int aeThreadOwnsLock()

src/ae.h

@@ -72,13 +72,15 @@ extern "C" {
 #define AE_NOTUSED(V) ((void) V)
 
 struct aeEventLoop;
-
 /* Types and data structures */
 typedef void aeFileProc(struct aeEventLoop *eventLoop, int fd, void *clientData, int mask);
 typedef int aeTimeProc(struct aeEventLoop *eventLoop, long long id, void *clientData);
 typedef void aeEventFinalizerProc(struct aeEventLoop *eventLoop, void *clientData);
 typedef void aeBeforeSleepProc(struct aeEventLoop *eventLoop);
 typedef void aePostFunctionProc(void *pvArgs);
+//added to support async api IStorage
+struct StorageToken;
+typedef void aePostFunctionTokenProc(struct aeEventLoop *el, struct StorageToken *token);
 
 /* File event structure */
 typedef struct aeFileEvent {
@@ -134,6 +136,8 @@ int aePostFunction(aeEventLoop *eventLoop, aePostFunctionProc *proc, void *arg);
 #ifdef __cplusplus
 }   // EXTERN C
 int aePostFunction(aeEventLoop *eventLoop, std::function<void()> fn, bool fLock = true, bool fForceQueue = false);
+//added to support async api IStorage
+int aePostFunction(aeEventLoop *eventLoop, aePostFunctionTokenProc *proc, StorageToken *token);
 extern "C" {
 #endif
 void aeDeleteEventLoop(aeEventLoop *eventLoop);

src/cluster.cpp

@@ -5991,7 +5991,9 @@ clusterNode *getNodeByQuery(client *c, struct redisCommand *cmd, robj **argv, in
                 /* If it is not the first key, make sure it is exactly
                  * the same key as the first we saw. */
                 if (!equalStringObjects(firstkey,thiskey)) {
-                    if (slot != thisslot) {
+                    clusterNode* nThisKey = g_pserver->cluster->slots[thisslot];
+
+                    if ((slot != thisslot) && (nThisKey != n || migrating_slot || importing_slot || g_pserver->cluster->migrating_slots_to[slot] != nullptr || g_pserver->cluster->importing_slots_from[slot] != nullptr)) {
                         /* Error: multiple keys from different slots. */
                         getKeysFreeResult(&result);
                         if (error_code)

src/config.cpp

@@ -2597,6 +2597,19 @@ static int updateMaxmemory(long long val, long long prev, const char **err) {
     return 1;
 }
 
+static int updateFlashMaxmemory(long long val, long long prev, const char **err) {
+    UNUSED(prev);
+    UNUSED(err);
+    if (val && g_pserver->m_pstorageFactory) {
+        size_t used = g_pserver->m_pstorageFactory->totalDiskspaceUsed();
+        if ((unsigned long long)val < used) {
+            serverLog(LL_WARNING,"WARNING: the new maxstorage value set via CONFIG SET (%llu) is smaller than the current storage usage (%zu). This will result in key eviction and/or the inability to accept new write commands depending on the maxmemory-policy.", g_pserver->maxstorage, used);
+        }
+        performEvictions(false /*fPreSnapshot*/);
+    }
+    return 1;
+}
+
 static int updateGoodSlaves(long long val, long long prev, const char **err) {
     UNUSED(val);
     UNUSED(prev);
@@ -2933,14 +2946,14 @@ standardConfig configs[] = {
     createLongLongConfig("latency-monitor-threshold", NULL, MODIFIABLE_CONFIG, 0, LLONG_MAX, g_pserver->latency_monitor_threshold, 0, INTEGER_CONFIG, NULL, NULL),
     createLongLongConfig("proto-max-bulk-len", NULL, MODIFIABLE_CONFIG, 1024*1024, LLONG_MAX, g_pserver->proto_max_bulk_len, 512ll*1024*1024, MEMORY_CONFIG, NULL, NULL), /* Bulk request max size */
     createLongLongConfig("stream-node-max-entries", NULL, MODIFIABLE_CONFIG, 0, LLONG_MAX, g_pserver->stream_node_max_entries, 100, INTEGER_CONFIG, NULL, NULL),
-    createLongLongConfig("repl-backlog-size", NULL, MODIFIABLE_CONFIG, 1, LLONG_MAX, g_pserver->repl_backlog_size, 1024*1024, MEMORY_CONFIG, NULL, updateReplBacklogSize), /* Default: 1mb */
+    createLongLongConfig("repl-backlog-size", NULL, MODIFIABLE_CONFIG, 1, LLONG_MAX, g_pserver->repl_backlog_config_size, 1024*1024, MEMORY_CONFIG, NULL, updateReplBacklogSize), /* Default: 1mb */
     createLongLongConfig("repl-backlog-disk-reserve", NULL, IMMUTABLE_CONFIG, 0, LLONG_MAX, cserver.repl_backlog_disk_size, 0, MEMORY_CONFIG, NULL, NULL),
     createLongLongConfig("max-snapshot-slip", NULL, MODIFIABLE_CONFIG, 0, 5000, g_pserver->snapshot_slip, 400, 0, NULL, NULL),
     createLongLongConfig("max-rand-count", NULL, MODIFIABLE_CONFIG, 0, LONG_MAX/2, g_pserver->rand_total_threshold, LONG_MAX/2, 0, NULL, NULL),
 
     /* Unsigned Long Long configs */
     createULongLongConfig("maxmemory", NULL, MODIFIABLE_CONFIG, 0, LLONG_MAX, g_pserver->maxmemory, 0, MEMORY_CONFIG, NULL, updateMaxmemory),
-    createULongLongConfig("maxstorage", NULL, MODIFIABLE_CONFIG, 0, LLONG_MAX, g_pserver->maxstorage, 0, MEMORY_CONFIG, NULL, NULL),
+    createULongLongConfig("maxstorage", NULL, MODIFIABLE_CONFIG, 0, LLONG_MAX, g_pserver->maxstorage, 0, MEMORY_CONFIG, NULL, updateFlashMaxmemory),
 
     /* Size_t configs */
     createSizeTConfig("hash-max-ziplist-entries", NULL, MODIFIABLE_CONFIG, 0, LONG_MAX, g_pserver->hash_max_ziplist_entries, 512, INTEGER_CONFIG, NULL, NULL),
@@ -2969,6 +2982,7 @@ standardConfig configs[] = {
     createSDSConfig("availability-zone", NULL, MODIFIABLE_CONFIG, 0, g_pserver->sdsAvailabilityZone, "", NULL, NULL),
     createIntConfig("overload-protect-percent", NULL, MODIFIABLE_CONFIG, 0, 200, g_pserver->overload_protect_threshold, 0, INTEGER_CONFIG, NULL, NULL),
     createIntConfig("force-eviction-percent", NULL, MODIFIABLE_CONFIG, 0, 100, g_pserver->force_eviction_percent, 0, INTEGER_CONFIG, NULL, NULL),
+    createBoolConfig("enable-async-rehash", NULL, MODIFIABLE_CONFIG, g_pserver->enable_async_rehash, 1, NULL, NULL),
 
 #ifdef USE_OPENSSL
     createIntConfig("tls-port", NULL, MODIFIABLE_CONFIG, 0, 65535, g_pserver->tls_port, 0, INTEGER_CONFIG, NULL, updateTLSPort), /* TCP port. */

src/db.cpp

@@ -2703,11 +2703,6 @@ void redisDbPersistentData::prepOverwriteForSnapshot(char *key)
         auto itr = m_pdbSnapshot->find_cached_threadsafe(key);
         if (itr.key() != nullptr)
         {
-            if (itr.val()->FExpires()) {
-                // Note: I'm sure we could handle this, but its too risky at the moment.
-                //  There are known bugs doing this with expires
-                return;
-            }
             sds keyNew = sdsdupshared(itr.key());
             if (dictAdd(m_pdictTombstone, keyNew, (void*)dictHashKey(m_pdict, key)) != DICT_OK)
                 sdsfree(keyNew);
@@ -2745,6 +2740,13 @@ void redisDbPersistentData::clear(void(callback)(void*))
     if (m_spstorage != nullptr)
         m_spstorage->clear(callback);
     dictEmpty(m_pdictTombstone,callback);
+
+    // To avoid issues with async rehash we completly free the old dict and create a fresh one
+    dictRelease(m_pdict);
+    dictRelease(m_pdictTombstone);
+    m_pdict = dictCreate(&dbDictType, this);
+    m_pdictTombstone = dictCreate(&dbTombstoneDictType, this);
+
     m_pdbSnapshot = nullptr;
     m_numexpires = 0;
 }
@@ -2880,10 +2882,10 @@ LNotFound:
             {
                 dictAdd(m_pdict, sdsNewKey, o);
                 
+                o->SetFExpires(spexpire != nullptr);
                 if (spexpire != nullptr) {
                     o->expire = std::move(*spexpire);
                 }
-                o->SetFExpires(spexpire != nullptr);
                 g_pserver->stat_storage_provider_read_hits++;
             } else {
                 sdsfree(sdsNewKey);
@@ -3254,8 +3256,8 @@ std::unique_ptr<expireEntry> deserializeExpire(const char *str, size_t cch, size
         if (subkey)
             sdsfree(subkey);
     }
-
-    *poffset = offset;
+    if (poffset != nullptr)
+        *poffset = offset;
     return spexpire;
 }
 

src/debug.cpp

@@ -79,6 +79,7 @@ void printCrashReport(void);
 void bugReportEnd(int killViaSignal, int sig);
 void logStackTrace(void *eip, int uplevel);
 void getTempFileName(char tmpfile[], int tmpfileNum);
+bool initializeStorageProvider(const char **err);
 
 /* ================================= Debugging ============================== */
 
@@ -308,7 +309,7 @@ void computeDatasetDigest(unsigned char *final) {
         mixDigest(final,&aux,sizeof(aux));
 
         /* Iterate this DB writing every entry */
-        db->iterate_threadsafe([final, db](const char *key, robj_roptr o)->bool {
+        db->iterate_threadsafe([final](const char *key, robj_roptr o)->bool {
             unsigned char digest[20];
             robj *keyobj;
 
@@ -535,30 +536,75 @@ NULL
                 return;
             }
         }
+        
+        if (g_pserver->m_pstorageFactory) {
+            protectClient(c);
 
-        /* The default behavior is to save the RDB file before loading
-         * it back. */
-        if (save) {
-            rdbSaveInfo rsi, *rsiptr;
-            rsiptr = rdbPopulateSaveInfo(&rsi);
-            if (rdbSave(nullptr, rsiptr) != C_OK) {
-                addReply(c,shared.err);
+            for (int idb = 0; idb < cserver.dbnum; ++idb) {
+                if (g_pserver->db[idb]->processChanges(false))
+                    g_pserver->db[idb]->commitChanges();
+                g_pserver->db[idb]->storageProviderDelete();
+            }
+            delete g_pserver->metadataDb;
+
+            if (flush) emptyDb(-1,EMPTYDB_NO_FLAGS,NULL);
+
+            delete g_pserver->m_pstorageFactory;
+
+            const char *err;
+            if (!initializeStorageProvider(&err))
+            {
+                serverLog(LL_WARNING, "Failed to initialize storage provider: %s",err);
+                exit(EXIT_FAILURE);
+            }
+
+            g_pserver->metadataDb = g_pserver->m_pstorageFactory->createMetadataDb();
+            for (int idb = 0; idb < cserver.dbnum; ++idb)
+            {
+                int dbid = idb;
+                std::string dbid_key = "db-" + std::to_string(idb);
+                g_pserver->metadataDb->retrieve(dbid_key.c_str(), dbid_key.length(), [&](const char *, size_t, const void *data, size_t){
+                    dbid = *(int*)data;
+                });
+                delete g_pserver->db[idb];
+                g_pserver->db[idb] = new (MALLOC_LOCAL) redisDb();
+                g_pserver->db[idb]->initialize(dbid);
+            }
+
+            moduleFireServerEvent(REDISMODULE_EVENT_LOADING, REDISMODULE_SUBEVENT_LOADING_FLASH_START, NULL);
+            for (int idb = 0; idb < cserver.dbnum; ++idb)
+            {
+                g_pserver->db[idb]->storageProviderInitialize();
+                g_pserver->db[idb]->trackChanges(false);
+            }
+            moduleFireServerEvent(REDISMODULE_EVENT_LOADING, REDISMODULE_SUBEVENT_LOADING_ENDED, NULL);
+            
+            unprotectClient(c);
+        } else {
+            /* The default behavior is to save the RDB file before loading
+            * it back. */
+            if (save) {
+                rdbSaveInfo rsi, *rsiptr;
+                rsiptr = rdbPopulateSaveInfo(&rsi);
+                if (rdbSave(nullptr, rsiptr) != C_OK) {
+                    addReply(c,shared.err);
+                    return;
+                }
+            }
+
+            /* The default behavior is to remove the current dataset from
+            * memory before loading the RDB file, however when MERGE is
+            * used together with NOFLUSH, we are able to merge two datasets. */
+            if (flush) emptyDb(-1,EMPTYDB_NO_FLAGS,NULL);
+
+            protectClient(c);
+            int ret = rdbLoadFile(g_pserver->rdb_filename,NULL,flags);
+            unprotectClient(c);
+            if (ret != C_OK) {
+                addReplyError(c,"Error trying to load the RDB dump");
                 return;
             }
         }
-
-        /* The default behavior is to remove the current dataset from
-         * memory before loading the RDB file, however when MERGE is
-         * used together with NOFLUSH, we are able to merge two datasets. */
-        if (flush) emptyDb(-1,EMPTYDB_NO_FLAGS,NULL);
-
-        protectClient(c);
-        int ret = rdbLoadFile(g_pserver->rdb_filename,NULL,flags);
-        unprotectClient(c);
-        if (ret != C_OK) {
-            addReplyError(c,"Error trying to load the RDB dump");
-            return;
-        }
         serverLog(LL_WARNING,"DB reloaded by DEBUG RELOAD");
         addReply(c,shared.ok);
     } else if (!strcasecmp(szFromObj(c->argv[1]),"loadaof")) {
@@ -932,6 +978,21 @@ NULL
         mallctl_string(c, c->argv+2, c->argc-2);
         return;
 #endif
+    } else if(!strcasecmp(szFromObj(c->argv[1]),"flush-storage") && c->argc == 2) {
+        if (g_pserver->m_pstorageFactory != nullptr) {
+            for (int i = 0; i < cserver.dbnum; i++) {
+                g_pserver->db[i]->getStorageCache()->flush();
+            }
+            addReply(c,shared.ok);
+        } else {
+            addReplyError(c, "Can't flush storage if no storage provider is set");
+        }
+    } else if (!strcasecmp(szFromObj(c->argv[1]),"get-storage-usage") && c->argc == 2) {
+        if (g_pserver->m_pstorageFactory != nullptr) {
+            addReplyLongLong(c, g_pserver->m_pstorageFactory->totalDiskspaceUsed());
+        } else {
+            addReplyLongLong(c, 0);
+        }
     } else {
         addReplySubcommandSyntaxError(c);
         return;

src/evict.cpp

@@ -100,6 +100,36 @@ unsigned long long estimateObjectIdleTime(robj_roptr o) {
     }
 }
 
+unsigned long long getIdle(robj *obj, const expireEntry *e) {
+    unsigned long long idle;
+    /* Calculate the idle time according to the policy. This is called
+        * idle just because the code initially handled LRU, but is in fact
+        * just a score where an higher score means better candidate. */
+    if (g_pserver->maxmemory_policy & MAXMEMORY_FLAG_LRU) {
+        idle = (obj != nullptr) ? estimateObjectIdleTime(obj) : 0;
+    } else if (g_pserver->maxmemory_policy & MAXMEMORY_FLAG_LFU) {
+        /* When we use an LRU policy, we sort the keys by idle time
+            * so that we expire keys starting from greater idle time.
+            * However when the policy is an LFU one, we have a frequency
+            * estimation, and we want to evict keys with lower frequency
+            * first. So inside the pool we put objects using the inverted
+            * frequency subtracting the actual frequency to the maximum
+            * frequency of 255. */
+        idle = 255-LFUDecrAndReturn(obj);
+    } else if (g_pserver->maxmemory_policy == MAXMEMORY_VOLATILE_TTL) {
+        /* In this case the sooner the expire the better. */
+        if (e != nullptr)
+            idle = ULLONG_MAX - e->when();
+        else
+            idle = 0;
+    } else if (g_pserver->maxmemory_policy & MAXMEMORY_FLAG_ALLKEYS) {
+        idle = ULLONG_MAX;
+    } else {
+        serverPanic("Unknown eviction policy in storage eviction");
+    }
+    return idle;
+}
+
 /* LRU approximation algorithm
  *
  * Redis uses an approximation of the LRU algorithm that runs in constant
@@ -137,28 +167,7 @@ void evictionPoolAlloc(void) {
 
 void processEvictionCandidate(int dbid, sds key, robj *o, const expireEntry *e, struct evictionPoolEntry *pool)
 {
-    unsigned long long idle;
-
-    /* Calculate the idle time according to the policy. This is called
-        * idle just because the code initially handled LRU, but is in fact
-        * just a score where an higher score means better candidate. */
-    if (g_pserver->maxmemory_policy & MAXMEMORY_FLAG_LRU) {
-        idle = (o != nullptr) ? estimateObjectIdleTime(o) : 0;
-    } else if (g_pserver->maxmemory_policy & MAXMEMORY_FLAG_LFU) {
-        /* When we use an LRU policy, we sort the keys by idle time
-            * so that we expire keys starting from greater idle time.
-            * However when the policy is an LFU one, we have a frequency
-            * estimation, and we want to evict keys with lower frequency
-            * first. So inside the pool we put objects using the inverted
-            * frequency subtracting the actual frequency to the maximum
-            * frequency of 255. */
-        idle = 255-LFUDecrAndReturn(o);
-    } else if (g_pserver->maxmemory_policy == MAXMEMORY_VOLATILE_TTL) {
-        /* In this case the sooner the expire the better. */
-        idle = ULLONG_MAX - e->when();
-    } else {
-        serverPanic("Unknown eviction policy in evictionPoolPopulate()");
-    }
+    unsigned long long idle = getIdle(o,e);
 
     /* Insert the element inside the pool.
         * First, find the first empty bucket or the first populated
@@ -600,6 +609,31 @@ static unsigned long evictionTimeLimitUs() {
     return ULONG_MAX;   /* No limit to eviction time */
 }
 
+void evict(redisDb *db, robj *keyobj) {
+    mstime_t eviction_latency;
+    propagateExpire(db,keyobj,g_pserver->lazyfree_lazy_eviction);
+    /* We compute the amount of memory freed by db*Delete() alone.
+    * It is possible that actually the memory needed to propagate
+    * the DEL in AOF and replication link is greater than the one
+    * we are freeing removing the key, but we can't account for
+    * that otherwise we would never exit the loop.
+    *
+    * AOF and Output buffer memory will be freed eventually so
+    * we only care about memory used by the key space. */
+    latencyStartMonitor(eviction_latency);
+    if (g_pserver->lazyfree_lazy_eviction)
+        dbAsyncDelete(db,keyobj);
+    else
+        dbSyncDelete(db,keyobj);
+    latencyEndMonitor(eviction_latency);
+    latencyAddSampleIfNeeded("eviction-del",eviction_latency);
+
+    signalModifiedKey(NULL,db,keyobj);
+    notifyKeyspaceEvent(NOTIFY_EVICTED, "evicted",
+        keyobj, db->id);
+    decrRefCount(keyobj);
+}
+
 static void updateSysAvailableMemory() {
     if (g_pserver->force_eviction_percent) {
         g_pserver->cron_malloc_stats.sys_available = getMemAvailable();
@@ -637,7 +671,7 @@ int performEvictions(bool fPreSnapshot) {
     int keys_freed = 0;
     size_t mem_reported, mem_tofree;
     long long mem_freed; /* May be negative */
-    mstime_t latency, eviction_latency;
+    mstime_t latency;
     long long delta;
     int slaves = listLength(g_pserver->slaves);
     const bool fEvictToStorage = !cserver.delete_on_evict && g_pserver->db[0]->FStorageProvider();
@@ -662,6 +696,43 @@ int performEvictions(bool fPreSnapshot) {
     monotime evictionTimer;
     elapsedStart(&evictionTimer);
 
+    if (g_pserver->maxstorage && g_pserver->m_pstorageFactory != nullptr) {
+        while (g_pserver->m_pstorageFactory->totalDiskspaceUsed() >= g_pserver->maxstorage && elapsedUs(evictionTimer) < eviction_time_limit_us) {
+            redisDb *db;
+            std::vector<std::string> evictionPool;
+            robj *bestkey = nullptr;
+            redisDb *bestdb = nullptr;
+            unsigned long long bestidle = 0;
+            for (int i = 0; i < cserver.dbnum; i++) {
+                db = g_pserver->db[i];
+                evictionPool = db->getStorageCache()->getEvictionCandidates(g_pserver->maxmemory_samples);
+                for (std::string key : evictionPool) {
+                    robj *keyobj = createStringObject(key.c_str(), key.size());
+                    robj *obj = db->find(szFromObj(keyobj));
+                    if (obj != nullptr) {
+                        expireEntry *e = db->getExpire(keyobj);
+                        unsigned long long idle = getIdle(obj, e);
+
+                        if (bestkey == nullptr || bestidle < idle) {
+                            if (bestkey != nullptr)
+                                decrRefCount(bestkey);
+                            incrRefCount(keyobj);
+                            bestkey = keyobj;
+                            bestidle = idle;
+                            bestdb = db;
+                        }
+                    }
+                    decrRefCount(keyobj);
+                }
+            }
+            if (bestkey) {
+                evict(bestdb, bestkey);
+            } else {
+                break; //could not find a key to evict so stop now
+            }
+        }
+    }
+
     if (g_pserver->maxstorage && g_pserver->m_pstorageFactory != nullptr && g_pserver->m_pstorageFactory->totalDiskspaceUsed() >= g_pserver->maxstorage)
         goto cant_free_storage;
 
@@ -776,7 +847,7 @@ int performEvictions(bool fPreSnapshot) {
                 if (db->removeCachedValue(bestkey, &deT)) {
                     mem_freed += splazy->addEntry(db->dictUnsafeKeyOnly(), deT);
                     ckeysFailed = 0;
-		    g_pserver->stat_evictedkeys++;
+                    g_pserver->stat_evictedkeys++;
                 }
                 else {
                     delta = 0;
@@ -788,30 +859,11 @@ int performEvictions(bool fPreSnapshot) {
             else
             {
                 robj *keyobj = createStringObject(bestkey,sdslen(bestkey));
-                propagateExpire(db,keyobj,g_pserver->lazyfree_lazy_eviction);
-                /* We compute the amount of memory freed by db*Delete() alone.
-                * It is possible that actually the memory needed to propagate
-                * the DEL in AOF and replication link is greater than the one
-                * we are freeing removing the key, but we can't account for
-                * that otherwise we would never exit the loop.
-                *
-                * AOF and Output buffer memory will be freed eventually so
-                * we only care about memory used by the key space. */
                 delta = (long long) zmalloc_used_memory();
-                latencyStartMonitor(eviction_latency);
-                if (g_pserver->lazyfree_lazy_eviction)
-                    dbAsyncDelete(db,keyobj);
-                else
-                    dbSyncDelete(db,keyobj);
-                latencyEndMonitor(eviction_latency);
-                latencyAddSampleIfNeeded("eviction-del",eviction_latency);
+                evict(db, keyobj);
                 delta -= (long long) zmalloc_used_memory();
                 mem_freed += delta;
                 g_pserver->stat_evictedkeys++;
-                signalModifiedKey(NULL,db,keyobj);
-                notifyKeyspaceEvent(NOTIFY_EVICTED, "evicted",
-                    keyobj, db->id);
-                decrRefCount(keyobj);
             }
             keys_freed++;
 

src/expire.cpp

@@ -396,126 +396,152 @@ void pexpireMemberAtCommand(client *c)
          * distribute the time evenly across DBs. */
         current_db++;
 
-        /* Continue to expire if at the end of the cycle there are still
-         * a big percentage of keys to expire, compared to the number of keys
-         * we scanned. The percentage, stored in config_cycle_acceptable_stale
-         * is not fixed, but depends on the Redis configured "expire effort". */
-        do {
-            unsigned long num, slots;
-            long long now, ttl_sum;
-            int ttl_samples;
-            iteration++;
+        if (g_pserver->m_pstorageFactory == nullptr) {
+            /* Continue to expire if at the end of the cycle there are still
+            * a big percentage of keys to expire, compared to the number of keys
+            * we scanned. The percentage, stored in config_cycle_acceptable_stale
+            * is not fixed, but depends on the Redis configured "expire effort". */
+            do {
+                unsigned long num, slots;
+                long long now, ttl_sum;
+                int ttl_samples;
+                iteration++;
 
-            /* If there is nothing to expire try next DB ASAP. */
-            if (db->expireSize() == 0) {
-                db->avg_ttl = 0;
-                break;
-            }
-            num = dictSize(db->m_pdict);
-            slots = dictSlots(db->m_pdict);
-            now = mstime();
-
-            /* When there are less than 1% filled slots, sampling the key
-             * space is expensive, so stop here waiting for better times...
-             * The dictionary will be resized asap. */
-            if (slots > DICT_HT_INITIAL_SIZE &&
-                (num*100/slots < 1)) break;
-
-            /* The main collection cycle. Sample random keys among keys
-             * with an expire set, checking for expired ones. */
-            expired = 0;
-            sampled = 0;
-            ttl_sum = 0;
-            ttl_samples = 0;
-
-            if (num > config_keys_per_loop)
-                num = config_keys_per_loop;
-
-            /* Here we access the low level representation of the hash table
-             * for speed concerns: this makes this code coupled with dict.c,
-             * but it hardly changed in ten years.
-             *
-             * Note that certain places of the hash table may be empty,
-             * so we want also a stop condition about the number of
-             * buckets that we scanned. However scanning for free buckets
-             * is very fast: we are in the cache line scanning a sequential
-             * array of NULL pointers, so we can scan a lot more buckets
-             * than keys in the same time. */
-            long max_buckets = num*20;
-            long checked_buckets = 0;
-
-            while (sampled < num && checked_buckets < max_buckets) {
-                for (int table = 0; table < 2; table++) {
-                    if (table == 1 && !dictIsRehashing(db->m_pdict)) break;
-
-                    unsigned long idx = db->expires_cursor;
-                    idx &= db->m_pdict->ht[table].sizemask;
-                    dictEntry *de = db->m_pdict->ht[table].table[idx];
-                    long long ttl;
-
-                    /* Scan the current bucket of the current table. */
-                    checked_buckets++;
-                    while(de) {
-                        /* Get the next entry now since this entry may get
-                         * deleted. */
-                        dictEntry *e = de;
-                        robj *o = (robj*)dictGetVal(de);
-                        de = de->next;
-                        if (!o->FExpires())
-                            continue;
-
-                        expireEntry *exp = &o->expire;
-
-                        serverAssert(exp->when() > 0);
-                        ttl = exp->when()-now;
-                        size_t tried = 0;
-                        if (exp->when() <= now) {
-                            if (activeExpireCycleExpire(db,(const char*)dictGetKey(e),*exp,now,tried)) expired++;
-                            serverAssert(ttl <= 0);
-                        } else {
-                            serverAssert(ttl > 0);
-                        }
-                        if (ttl > 0) {
-                            /* We want the average TTL of keys yet
-                             * not expired. */
-                            ttl_sum += ttl;
-                            ttl_samples++;
-                        }
-                        sampled++;
-                    }
-                }
-                db->expires_cursor++;
-            }
-            total_expired += expired;
-            total_sampled += sampled;
-
-            /* Update the average TTL stats for this database. */
-            if (ttl_samples) {
-                long long avg_ttl = ttl_sum/ttl_samples;
-
-                /* Do a simple running average with a few samples.
-                 * We just use the current estimate with a weight of 2%
-                 * and the previous estimate with a weight of 98%. */
-                if (db->avg_ttl == 0) db->avg_ttl = avg_ttl;
-                db->avg_ttl = (db->avg_ttl/50)*49 + (avg_ttl/50);
-            }
-
-            /* We can't block forever here even if there are many keys to
-             * expire. So after a given amount of milliseconds return to the
-             * caller waiting for the other active expire cycle. */
-            if ((iteration & 0xf) == 0) { /* check once every 16 iterations. */
-                elapsed = ustime()-start;
-                if (elapsed > timelimit) {
-                    timelimit_exit = 1;
-                    g_pserver->stat_expired_time_cap_reached_count++;
+                /* If there is nothing to expire try next DB ASAP. */
+                if (db->expireSize() == 0) {
+                    db->avg_ttl = 0;
                     break;
                 }
+                num = dictSize(db->m_pdict);
+                slots = dictSlots(db->m_pdict);
+                now = mstime();
+
+                /* When there are less than 1% filled slots, sampling the key
+                * space is expensive, so stop here waiting for better times...
+                * The dictionary will be resized asap. */
+                if (slots > DICT_HT_INITIAL_SIZE &&
+                    (num*100/slots < 1)) break;
+
+                /* The main collection cycle. Sample random keys among keys
+                * with an expire set, checking for expired ones. */
+                expired = 0;
+                sampled = 0;
+                ttl_sum = 0;
+                ttl_samples = 0;
+
+                if (num > config_keys_per_loop)
+                    num = config_keys_per_loop;
+
+                /* Here we access the low level representation of the hash table
+                * for speed concerns: this makes this code coupled with dict.c,
+                * but it hardly changed in ten years.
+                *
+                * Note that certain places of the hash table may be empty,
+                * so we want also a stop condition about the number of
+                * buckets that we scanned. However scanning for free buckets
+                * is very fast: we are in the cache line scanning a sequential
+                * array of NULL pointers, so we can scan a lot more buckets
+                * than keys in the same time. */
+                long max_buckets = num*20;
+                long checked_buckets = 0;
+
+                while (sampled < num && checked_buckets < max_buckets) {
+                    for (int table = 0; table < 2; table++) {
+                        if (table == 1 && !dictIsRehashing(db->m_pdict)) break;
+
+                        unsigned long idx = db->expires_cursor;
+                        idx &= db->m_pdict->ht[table].sizemask;
+                        dictEntry *de = db->m_pdict->ht[table].table[idx];
+                        long long ttl;
+
+                        /* Scan the current bucket of the current table. */
+                        checked_buckets++;
+                        while(de) {
+                            /* Get the next entry now since this entry may get
+                            * deleted. */
+                            dictEntry *e = de;
+                            robj *o = (robj*)dictGetVal(de);
+                            de = de->next;
+                            if (!o->FExpires())
+                                continue;
+
+                            expireEntry *exp = &o->expire;
+
+                            serverAssert(exp->when() > 0);
+                            ttl = exp->when()-now;
+                            size_t tried = 0;
+                            if (exp->when() <= now) {
+                                if (activeExpireCycleExpire(db,(const char*)dictGetKey(e),*exp,now,tried)) expired++;
+                                serverAssert(ttl <= 0);
+                            } else {
+                                serverAssert(ttl > 0);
+                            }
+                            if (ttl > 0) {
+                                /* We want the average TTL of keys yet
+                                * not expired. */
+                                ttl_sum += ttl;
+                                ttl_samples++;
+                            }
+                            sampled++;
+                        }
+                    }
+                    db->expires_cursor++;
+                }
+                total_expired += expired;
+                total_sampled += sampled;
+
+                /* Update the average TTL stats for this database. */
+                if (ttl_samples) {
+                    long long avg_ttl = ttl_sum/ttl_samples;
+
+                    /* Do a simple running average with a few samples.
+                    * We just use the current estimate with a weight of 2%
+                    * and the previous estimate with a weight of 98%. */
+                    if (db->avg_ttl == 0) db->avg_ttl = avg_ttl;
+                    db->avg_ttl = (db->avg_ttl/50)*49 + (avg_ttl/50);
+                }
+
+                /* We can't block forever here even if there are many keys to
+                * expire. So after a given amount of milliseconds return to the
+                * caller waiting for the other active expire cycle. */
+                if ((iteration & 0xf) == 0) { /* check once every 16 iterations. */
+                    elapsed = ustime()-start;
+                    if (elapsed > timelimit) {
+                        timelimit_exit = 1;
+                        g_pserver->stat_expired_time_cap_reached_count++;
+                        break;
+                    }
+                }
+                /* We don't repeat the cycle for the current database if there are
+                * an acceptable amount of stale keys (logically expired but yet
+                * not reclaimed). */
+            } while (sampled == 0 ||
+                    (expired*100/sampled) > config_cycle_acceptable_stale);
+        } else {
+            long prev_expired;
+            long long now = mstime();
+            size_t tried = 0;
+            std::vector<std::string> keys;
+            do {
+                prev_expired = total_expired;
+                keys = db->getStorageCache()->getExpirationCandidates(ACTIVE_EXPIRE_CYCLE_LOOKUPS_PER_LOOP);
+                for (std::string key : keys) {
+                    robj* keyobj = createStringObject(key.c_str(), key.size());
+                    db->find(szFromObj(keyobj));
+                    expireEntry *e = db->getExpire(keyobj);
+                    if (e != nullptr && e->when() < now)
+                        total_expired += activeExpireCycleExpire(db, szFromObj(keyobj), *e, now, tried);
+                    decrRefCount(keyobj);
+                }
+                total_sampled += keys.size();
+                elapsed = ustime()-start;
+            } while (keys.size() > 0 && (elapsed < timelimit) && (total_expired - prev_expired) > 0);
+
+            if (ustime()-start > timelimit) {
+                timelimit_exit = 1;
+                g_pserver->stat_expired_time_cap_reached_count++;
             }
-            /* We don't repeat the cycle for the current database if there are
-             * an acceptable amount of stale keys (logically expired but yet
-             * not reclaimed). */
-        } while (sampled == 0 ||
-                 (expired*100/sampled) > config_cycle_acceptable_stale);
+        }
     }
 
     elapsed = ustime()-start;

src/expire.h

@@ -160,6 +160,8 @@ public:
 
     expireEntry &operator=(expireEntry &&e)
     {
+        if (FFat())
+            delete pfatentry();
         s = e.s;
         e.s.m_whenAndPtrUnion = 0;
         e.s.fFat = false;
@@ -167,6 +169,8 @@ public:
     }
 
     expireEntry &operator=(expireEntry &e) {
+        if (FFat())
+            delete pfatentry();
         if (e.FFat()) {
             s.m_whenAndPtrUnion = reinterpret_cast<long long>(new expireEntryFat(*e.pfatentry()));
             s.fFat = true;

src/meminfo.cpp

@@ -19,6 +19,7 @@ static size_t getMemKey(std::string key) {
     }
     return 0;
 # else
+    (void)key;
     return 0;
 # endif
 }

src/module.cpp

@@ -690,8 +690,9 @@ void moduleHandlePropagationAfterCommandCallback(RedisModuleCtx *ctx) {
 }
 
 /* Free the context after the user function was called. */
-void moduleFreeContext(RedisModuleCtx *ctx) {
-    moduleHandlePropagationAfterCommandCallback(ctx);
+void moduleFreeContext(RedisModuleCtx *ctx, bool propogate) {
+    if (propogate)
+        moduleHandlePropagationAfterCommandCallback(ctx);
     autoMemoryCollect(ctx);
     poolAllocRelease(ctx);
     if (ctx->postponed_arrays) {
@@ -7057,7 +7058,7 @@ RedisModuleString *RM_DictPrev(RedisModuleCtx *ctx, RedisModuleDictIter *di, voi
 /* Compare the element currently pointed by the iterator to the specified
  * element given by key/keylen, according to the operator 'op' (the set of
  * valid operators are the same valid for RedisModule_DictIteratorStart).
- * If the comparision is successful the command returns REDISMODULE_OK
+ * If the comparison is successful the command returns REDISMODULE_OK
  * otherwise REDISMODULE_ERR is returned.
  *
  * This is useful when we want to just emit a lexicographical range, so

src/modules/keydb_modstatsd/Makefile

@@ -4,7 +4,7 @@ OBJECT_FILES := modmain.o
 MODSNAP_CXX_FLAGS := -std=gnu++14
 
 %.o: %.cpp
-	$(CXX) -o $@ -c $< $(MODULE_FLAGS) -I../../../deps/cpp-statsd-client/include $(MODSNAP_CXX_FLAGS)
+	$(CXX) -o $@ -c $< $(MODULE_FLAGS) -I../../../deps/cpp-statsd-client/include $(MODSNAP_CXX_FLAGS) -g
 
 modstatsd.so: $(OBJECT_FILES)
 	$(CXX) -shared $(OBJECT_FILES) -o modstatsd.so

src/modules/keydb_modstatsd/modmain.cpp

@@ -76,7 +76,7 @@ std::string m_strPrefix { "keydb" };
 const std::regex g_replica_or_db_info_regex { "^(slave|db)(\\d+)" };
 const char *g_string_counter_separator = "__";
 const uint64_t g_stats_buffer_size_bytes = 1600;
-utsname sysName;
+std::string nodeName;
 int unameResult;
 
 enum class StatsD_Type {
@@ -582,7 +582,7 @@ void event_cron_handler(struct RedisModuleCtx *ctx, RedisModuleEvent eid, uint64
         size_t startTime = ustime();
 
 #ifdef __linux__
-        /* Log CPU Usage */
+	/* Log CPU Usage */
 	static long long s_mscpuLast = 0;
 	struct rusage self_ru;
 	getrusage(RUSAGE_SELF, &self_ru);
@@ -591,7 +591,7 @@ void event_cron_handler(struct RedisModuleCtx *ctx, RedisModuleEvent eid, uint64
 			+ (self_ru.ru_stime.tv_sec * 1000) + (self_ru.ru_stime.tv_usec / 1000);
 
 
-        g_stats->gauge("cpu_load_perc", (double)(mscpuCur - s_mscpuLast) / ((curTime - lastTime)*1000), true /* prefixOnly */);   
+	g_stats->gauge("cpu_load_perc", ((double)(mscpuCur - s_mscpuLast) / ((curTime - lastTime)*1000))*100, false /* prefixOnly */);   
 	s_mscpuLast = mscpuCur;
 #endif
 
@@ -600,7 +600,7 @@ void event_cron_handler(struct RedisModuleCtx *ctx, RedisModuleEvent eid, uint64
 
         /* node name */
         if (unameResult == 0) {
-            g_stats->increment("node_name" + std::string(g_string_counter_separator) + sysName.nodename);
+            g_stats->increment("node_name" + std::string(g_string_counter_separator) + nodeName);
         }
 
         /* Log INFO Fields */
@@ -653,7 +653,7 @@ void event_cron_handler(struct RedisModuleCtx *ctx, RedisModuleEvent eid, uint64
         reply = RedisModule_Call(ctx, "dbsize", "");
         long long keys = RedisModule_CallReplyInteger(reply);
         RedisModule_FreeCallReply(reply);
-        g_stats->gauge("keys", keys);
+        g_stats->gauge("keys", keys, false /* prefixOnly */);
         RedisModule_Log(ctx, REDISMODULE_LOGLEVEL_DEBUG, "Emitting metric \"keys\": %llu", keys);
         g_stats->timing("emit_keys_metric_time_taken_us", ustime() - commandStartTime);
 
@@ -672,17 +672,20 @@ extern "C" int RedisModule_OnLoad(RedisModuleCtx *ctx, RedisModuleString **argv,
     RedisModule_AutoMemory(ctx);
     /* Use pod name if available*/
     const char *podName = getenv("POD_NAME");
+    utsname sysName;
     unameResult = uname(&sysName);
+    if (unameResult == 0) {
+        nodeName = std::string(sysName.nodename);
+        std::replace(nodeName.begin(), nodeName.end(), '.', '-');
+    }
     if (podName != nullptr) {
         m_strPrefix = podName;
+        std::replace(m_strPrefix.begin(), m_strPrefix.end(), '.', '-');
     }
-    else {
-        if (unameResult == 0) {
-            m_strPrefix = std::string(sysName.nodename);
-            unameResult = 1;
-        }
+    else if (unameResult == 0) {
+        m_strPrefix = nodeName;
+        unameResult = 1;
     }
-    std::replace(m_strPrefix.begin(), m_strPrefix.end(), '.', '-');
 
     for (int iarg = 0; iarg < argc; ++iarg) {
         size_t len = 0;

src/object.cpp

@@ -46,6 +46,7 @@ robj *createObject(int type, void *ptr) {
     char *oB = (char*)zcalloc(sizeof(robj)+mvccExtraBytes, MALLOC_SHARED);
     robj *o = reinterpret_cast<robj*>(oB + mvccExtraBytes);
     
+    new (o) redisObject;
     o->type = type;
     o->encoding = OBJ_ENCODING_RAW;
     o->m_ptr = ptr;
@@ -418,6 +419,7 @@ void decrRefCount(robj_roptr o) {
         case OBJ_NESTEDHASH: freeNestedHashObject(o); break;
         default: serverPanic("Unknown object type"); break;
         }
+        o->~redisObject();
         if (g_pserver->fActiveReplica) {
             zfree(reinterpret_cast<redisObjectExtended*>(o.unsafe_robjcast())-1);
         } else {

src/rdb.cpp

@@ -2542,7 +2542,7 @@ robj *rdbLoadObject(int rdbtype, rio *rdb, sds key, int *error, uint64_t mvcc_ts
          * encoding version in the lower 10 bits of the module ID. */
         void *ptr = mt->rdb_load(&io,moduleid&1023);
         if (io.ctx) {
-            moduleFreeContext(io.ctx);
+            moduleFreeContext(io.ctx, false /* propogate */);
             zfree(io.ctx);
         }
 

src/readwritelock.h

@@ -8,38 +8,46 @@ class readWriteLock {
     int m_readCount = 0;
     int m_writeCount = 0;
     bool m_writeWaiting = false;
-    bool m_notify = true;
+    bool m_multi = true;
 public:
     readWriteLock(const char *name) : m_readLock(name), m_writeLock(name) {}
 
     void acquireRead() {
-        std::unique_lock<fastlock> rm(m_readLock);
-        while (m_writeCount > 0 || m_writeWaiting)
-            m_cv.wait(rm);
+        std::unique_lock<fastlock> rm(m_readLock, std::defer_lock);
+        if (m_multi) {
+            rm.lock();
+            while (m_writeCount > 0 || m_writeWaiting)
+                m_cv.wait(rm);
+        }
         m_readCount++;
     }
     
     bool tryAcquireRead() {
         std::unique_lock<fastlock> rm(m_readLock, std::defer_lock);
-        if (!rm.try_lock())
-            return false;
-        if (m_writeCount > 0 || m_writeWaiting)
-            return false;
+        if (m_multi) {
+            if (!rm.try_lock())
+                return false;
+            if (m_writeCount > 0 || m_writeWaiting)
+                return false;
+        }
         m_readCount++;
         return true;
     }
 
     void acquireWrite(bool exclusive = true) {
-        std::unique_lock<fastlock> rm(m_readLock);
-        m_writeWaiting = true;
-        while (m_readCount > 0)
-            m_cv.wait(rm);
-        if (exclusive) {
-            /* Another thread might have the write lock while we have the read lock
-               but won't be able to release it until they can acquire the read lock
-               so release the read lock and try again instead of waiting to avoid deadlock */
-            while(!m_writeLock.try_lock())
+        std::unique_lock<fastlock> rm(m_readLock, std::defer_lock);
+        if (m_multi) {
+            rm.lock();
+            m_writeWaiting = true;
+            while (m_readCount > 0)
                 m_cv.wait(rm);
+            if (exclusive) {
+                /* Another thread might have the write lock while we have the read lock
+                but won't be able to release it until they can acquire the read lock
+                so release the read lock and try again instead of waiting to avoid deadlock */
+                while(!m_writeLock.try_lock())
+                    m_cv.wait(rm);
+            }
         }
         m_writeCount++;
         m_writeWaiting = false;
@@ -52,32 +60,38 @@ public:
 
     bool tryAcquireWrite(bool exclusive = true) {
         std::unique_lock<fastlock> rm(m_readLock, std::defer_lock);
-        if (!rm.try_lock())
-            return false;
-        if (m_readCount > 0)
-            return false;
-        if (exclusive)
-            if (!m_writeLock.try_lock())
+        if (m_multi) {
+            if (!rm.try_lock())
                 return false;
+            if (m_readCount > 0)
+                return false;
+            if (exclusive)
+                if (!m_writeLock.try_lock())
+                    return false;
+        }
         m_writeCount++;
         return true;
     }
 
     void releaseRead() {
-        std::unique_lock<fastlock> rm(m_readLock);
-        m_readCount--;
-        if (m_notify)
+        std::unique_lock<fastlock> rm(m_readLock, std::defer_lock);
+        if (m_multi) {
+            rm.lock();
             m_cv.notify_all();
+        }
+        m_readCount--;
     }
 
     void releaseWrite(bool exclusive = true) {
-        std::unique_lock<fastlock> rm(m_readLock);
+        std::unique_lock<fastlock> rm(m_readLock, std::defer_lock);
         serverAssert(m_writeCount > 0);
-        if (exclusive)
-            m_writeLock.unlock();
-        m_writeCount--;
-        if (m_notify)
+        if (m_multi) {
+            rm.lock();
+            if (exclusive)
+                m_writeLock.unlock();
             m_cv.notify_all();
+        }
+        m_writeCount--;
     }
 
     void downgradeWrite(bool exclusive = true) {
@@ -85,8 +99,8 @@ public:
         acquireRead();
     }
 
-    void setNotify(bool notify) {
-        m_notify = notify;
+    void setMulti(bool multi) {
+        m_multi = multi;
     }
 
     bool hasReader() {

src/replication.cpp

@@ -188,6 +188,8 @@ int bg_unlink(const char *filename) {
 /* ---------------------------------- MASTER -------------------------------- */
 
 bool createDiskBacklog() {
+    if (g_pserver->repl_backlog_disk != nullptr)
+        return true;    // already exists
     // Lets create some disk backed pages and add them here
     std::string path = "./repl-backlog-temp" + std::to_string(gettid());
 #if (defined __APPLE__ || defined __FreeBSD__)
@@ -3733,7 +3735,7 @@ retry_connect:
     }
 
     /* Prepare a suitable temp file for bulk transfer */
-    if (!useDisklessLoad()) {
+    if (!useDisklessLoad() && !mi->isRocksdbSnapshotRepl) {
         while(maxtries--) {
             auto dt = std::chrono::system_clock::now().time_since_epoch();
             auto dtMillisecond = std::chrono::duration_cast<std::chrono::milliseconds>(dt);

src/server.cpp

@@ -1870,6 +1870,9 @@ VersionCompareResult compareVersion(SymVer *pver)
     if ((symVerThis.major == 0 && symVerThis.minor == 0 && symVerThis.build == 0)
         || (pver->major == 0 && pver->minor == 0 && pver->build == 0))
         return VersionCompareResult::EqualVersion;
+
+    if (pver->major <= 6 && pver->minor <= 3 && pver->build <= 3)
+        return VersionCompareResult::IncompatibleVersion;
     
     for (int iver = 0; iver < 3; ++iver)
     {
@@ -2134,7 +2137,7 @@ void databasesCron(bool fMainThread) {
                 ::dict *dict = g_pserver->db[rehash_db]->dictUnsafeKeyOnly();
                 /* Are we async rehashing? And if so is it time to re-calibrate? */
                 /* The recalibration limit is a prime number to ensure balancing across threads */
-                if (rehashes_per_ms > 0 && async_rehashes < 131 && !cserver.active_defrag_enabled && cserver.cthreads > 1 && dictSize(dict) > 2048 && dictIsRehashing(dict) && !g_pserver->loading) {
+                if (g_pserver->enable_async_rehash && rehashes_per_ms > 0 && async_rehashes < 131 && !cserver.active_defrag_enabled && cserver.cthreads > 1 && dictSize(dict) > 2048 && dictIsRehashing(dict) && !g_pserver->loading && aeLockContention() > 1) {
                     serverTL->rehashCtl = dictRehashAsyncStart(dict, rehashes_per_ms * ((1000 / g_pserver->hz) / 10));  // Estimate 10% CPU time spent in lock contention
                     if (serverTL->rehashCtl)
                         ++async_rehashes;
@@ -5660,7 +5663,8 @@ sds genRedisInfoString(const char *section) {
             "lru_clock:%u\r\n"
             "executable:%s\r\n"
             "config_file:%s\r\n"
-            "availability_zone:%s\r\n",
+            "availability_zone:%s\r\n"
+            "features:%s\r\n",
             KEYDB_SET_VERSION,
             redisGitSHA1(),
             strtol(redisGitDirty(),NULL,10) > 0,
@@ -5687,7 +5691,8 @@ sds genRedisInfoString(const char *section) {
             lruclock,
             cserver.executable ? cserver.executable : "",
             cserver.configfile ? cserver.configfile : "",
-            g_pserver->sdsAvailabilityZone);
+            g_pserver->sdsAvailabilityZone,
+            "cluster_mget");
     }
 
     /* Clients */
@@ -7408,7 +7413,7 @@ static void validateConfiguration()
         exit(EXIT_FAILURE);
     }
 
-    g_pserver->repl_backlog_config_size = g_pserver->repl_backlog_size; // this is normally set in the update logic, but not on initial config
+    g_pserver->repl_backlog_size = g_pserver->repl_backlog_config_size; // this is normally set in the update logic, but not on initial config
 }
 
 int iAmMaster(void) {

src/server.h

@@ -966,6 +966,7 @@ struct redisObjectExtended {
 
 typedef struct redisObject {
     friend redisObject *createEmbeddedStringObject(const char *ptr, size_t len);
+    friend redisObject *createObject(int type, void *ptr);
 protected:
     redisObject() {}
 
@@ -2305,6 +2306,7 @@ struct redisServer {
     rax *errors;                /* Errors table */
     int activerehashing;        /* Incremental rehash in serverCron() */
     int active_defrag_running;  /* Active defragmentation running (holds current scan aggressiveness) */
+    int enable_async_rehash = 1;    /* Should we use the async rehash feature? */
     int cronloops;              /* Number of times the cron function run */
     char runid[CONFIG_RUN_ID_SIZE+1];  /* ID always different at every exec. */
     int sentinel_mode;          /* True if this instance is a Sentinel. */
@@ -2917,7 +2919,7 @@ int moduleGetCommandKeysViaAPI(struct redisCommand *cmd, robj **argv, int argc,
 moduleType *moduleTypeLookupModuleByID(uint64_t id);
 void moduleTypeNameByID(char *name, uint64_t moduleid);
 const char *moduleTypeModuleName(moduleType *mt);
-void moduleFreeContext(struct RedisModuleCtx *ctx);
+void moduleFreeContext(struct RedisModuleCtx *ctx, bool propogate = true);
 void unblockClientFromModule(client *c);
 void moduleHandleBlockedClients(int iel);
 void moduleBlockedClientTimedOut(client *c);

src/snapshot.cpp

@@ -587,8 +587,12 @@ bool redisDbPersistentDataSnapshot::iterate_threadsafe_core(std::function<bool(c
                 if (!fKeyOnly)
                 {
                     size_t offset = 0;
-                    deserializeExpire((const char*)data, cbData, &offset);
+                    std::unique_ptr<expireEntry> spexpire = deserializeExpire((const char*)data, cbData, &offset);
                     o = deserializeStoredObject(reinterpret_cast<const char*>(data)+offset, cbData-offset);
+                    o->SetFExpires(spexpire != nullptr);
+                    if (spexpire != nullptr) {
+                        o->expire = std::move(*spexpire);
+                    }
                 }
                 fContinue = fn(sdsKey, o);
                 if (o != nullptr)

src/storage/rocksdb.cpp

@@ -33,8 +33,8 @@ std::string prefixKey(const char *key, size_t cchKey)
     return FInternalKey(key, cchKey) ? std::string(key, cchKey) : getPrefix(keyHashSlot(key, cchKey)) + std::string(key, cchKey);
 }
 
-RocksDBStorageProvider::RocksDBStorageProvider(RocksDBStorageFactory *pfactory, std::shared_ptr<rocksdb::DB> &spdb, std::shared_ptr<rocksdb::ColumnFamilyHandle> &spcolfam, const rocksdb::Snapshot *psnapshot, size_t count)
-    : m_pfactory(pfactory), m_spdb(spdb), m_psnapshot(psnapshot), m_spcolfamily(spcolfam), m_count(count)
+RocksDBStorageProvider::RocksDBStorageProvider(RocksDBStorageFactory *pfactory, std::shared_ptr<rocksdb::DB> &spdb, std::shared_ptr<rocksdb::ColumnFamilyHandle> &spcolfam, std::shared_ptr<rocksdb::ColumnFamilyHandle> &spexpirecolfam, const rocksdb::Snapshot *psnapshot, size_t count)
+    : m_pfactory(pfactory), m_spdb(spdb), m_psnapshot(psnapshot), m_spcolfamily(spcolfam), m_spexpirecolfamily(spexpirecolfam), m_count(count)
 {
     m_readOptionsTemplate = rocksdb::ReadOptions();
     m_readOptionsTemplate.verify_checksums = false;
@@ -149,11 +149,22 @@ size_t RocksDBStorageProvider::clear()
 
     rocksdb::ColumnFamilyHandle *handle = nullptr;
     rocksdb::ColumnFamilyOptions cf_options(m_pfactory->RocksDbOptions());
+    cf_options.level_compaction_dynamic_level_bytes = true;
     m_spdb->CreateColumnFamily(cf_options, strName, &handle);
     m_spcolfamily = std::shared_ptr<rocksdb::ColumnFamilyHandle>(handle);
 
     if (!status.ok())
         throw status.ToString();
+   
+    status = m_spdb->DropColumnFamily(m_spexpirecolfamily.get());
+    strName = m_spexpirecolfamily->GetName();
+
+    m_spdb->CreateColumnFamily(cf_options, strName, &handle);
+    m_spexpirecolfamily = std::shared_ptr<rocksdb::ColumnFamilyHandle>(handle);
+
+    if (!status.ok())
+        throw status.ToString();
+
     m_count = 0;
     return celem;
 }
@@ -211,11 +222,79 @@ bool RocksDBStorageProvider::enumerate_hashslot(callback fn, unsigned int hashsl
     return full_iter;
 }
 
+void RocksDBStorageProvider::setExpire(const char *key, size_t cchKey, long long expire)
+{
+    rocksdb::Status status;
+    std::unique_lock<fastlock> l(m_lock);
+    std::string prefix((const char *)&expire,sizeof(long long));
+    std::string strKey(key, cchKey);
+    if (m_spbatch != nullptr)
+        status = m_spbatch->Put(m_spexpirecolfamily.get(), rocksdb::Slice(prefix + strKey), rocksdb::Slice(strKey));
+    else
+        status = m_spdb->Put(WriteOptions(), m_spexpirecolfamily.get(), rocksdb::Slice(prefix + strKey), rocksdb::Slice(strKey));
+    if (!status.ok())
+        throw status.ToString();
+}
+
+void RocksDBStorageProvider::removeExpire(const char *key, size_t cchKey, long long expire)
+{
+    rocksdb::Status status;
+    std::unique_lock<fastlock> l(m_lock);
+    std::string prefix((const char *)&expire,sizeof(long long));
+    std::string strKey(key, cchKey);
+    std::string fullKey = prefix + strKey;
+    if (!FExpireExists(fullKey))
+        return;
+    if (m_spbatch)
+        status = m_spbatch->Delete(m_spexpirecolfamily.get(), rocksdb::Slice(fullKey));
+    else
+        status = m_spdb->Delete(WriteOptions(), m_spexpirecolfamily.get(), rocksdb::Slice(fullKey));
+    if (!status.ok())
+        throw status.ToString();
+}
+
+std::vector<std::string> RocksDBStorageProvider::getExpirationCandidates(unsigned int count)
+{
+    std::vector<std::string> result;
+    std::unique_ptr<rocksdb::Iterator> it = std::unique_ptr<rocksdb::Iterator>(m_spdb->NewIterator(ReadOptions(), m_spexpirecolfamily.get()));
+    for (it->SeekToFirst(); it->Valid() && result.size() < count; it->Next()) {
+        if (FInternalKey(it->key().data(), it->key().size()))
+            continue;
+        result.emplace_back(it->value().data(), it->value().size());
+    }
+    return result;
+}
+
+std::string randomHashSlot() {
+    return getPrefix(genrand64_int63() % (1 << 16));
+}
+
+std::vector<std::string> RocksDBStorageProvider::getEvictionCandidates(unsigned int count)
+{
+    std::vector<std::string> result;
+    if (g_pserver->maxmemory_policy & MAXMEMORY_FLAG_ALLKEYS) {
+        std::unique_ptr<rocksdb::Iterator> it = std::unique_ptr<rocksdb::Iterator>(m_spdb->NewIterator(ReadOptions(), m_spcolfamily.get()));
+        for (it->Seek(randomHashSlot()); it->Valid() && result.size() < count; it->Next()) {
+            if (FInternalKey(it->key().data(), it->key().size()))
+                continue;
+            result.emplace_back(it->key().data() + 2, it->key().size() - 2);
+        }
+    } else {
+        std::unique_ptr<rocksdb::Iterator> it = std::unique_ptr<rocksdb::Iterator>(m_spdb->NewIterator(ReadOptions(), m_spexpirecolfamily.get()));
+        for (it->SeekToFirst(); it->Valid() && result.size() < count; it->Next()) {
+            if (FInternalKey(it->key().data(), it->key().size()))
+                continue;
+            result.emplace_back(it->value().data(), it->value().size());
+        }
+    }
+    return result;
+}
+
 const IStorage *RocksDBStorageProvider::clone() const
 {
     std::unique_lock<fastlock> l(m_lock);
     const rocksdb::Snapshot *psnapshot = const_cast<RocksDBStorageProvider*>(this)->m_spdb->GetSnapshot();
-    return new RocksDBStorageProvider(m_pfactory, const_cast<RocksDBStorageProvider*>(this)->m_spdb, const_cast<RocksDBStorageProvider*>(this)->m_spcolfamily, psnapshot, m_count);
+    return new RocksDBStorageProvider(m_pfactory, const_cast<RocksDBStorageProvider*>(this)->m_spdb, const_cast<RocksDBStorageProvider*>(this)->m_spcolfamily, const_cast<RocksDBStorageProvider*>(this)->m_spexpirecolfamily, psnapshot, m_count);
 }
 
 RocksDBStorageProvider::~RocksDBStorageProvider()
@@ -268,6 +347,7 @@ void RocksDBStorageProvider::batch_unlock()
 void RocksDBStorageProvider::flush()
 {
     m_spdb->SyncWAL();
+    m_spdb->Flush(rocksdb::FlushOptions());
 }
 
 bool RocksDBStorageProvider::FKeyExists(std::string& key) const
@@ -276,4 +356,12 @@ bool RocksDBStorageProvider::FKeyExists(std::string& key) const
     if (m_spbatch)
         return m_spbatch->GetFromBatchAndDB(m_spdb.get(), ReadOptions(), m_spcolfamily.get(), rocksdb::Slice(key), &slice).ok();
     return m_spdb->Get(ReadOptions(), m_spcolfamily.get(), rocksdb::Slice(key), &slice).ok();
+}
+
+bool RocksDBStorageProvider::FExpireExists(std::string& key) const
+{
+    rocksdb::PinnableSlice slice;
+    if (m_spbatch)
+        return m_spbatch->GetFromBatchAndDB(m_spdb.get(), ReadOptions(), m_spexpirecolfamily.get(), rocksdb::Slice(key), &slice).ok();
+    return m_spdb->Get(ReadOptions(), m_spexpirecolfamily.get(), rocksdb::Slice(key), &slice).ok();
 }

src/storage/rocksdb.h

@@ -10,6 +10,7 @@
 static const char count_key[] = INTERNAL_KEY_PREFIX "__keydb__count\1";
 static const char version_key[] = INTERNAL_KEY_PREFIX "__keydb__version\1";
 static const char meta_key[] = INTERNAL_KEY_PREFIX "__keydb__metadata\1";
+static const char last_expire_key[] = INTERNAL_KEY_PREFIX "__keydb__last_expire_time";
 class RocksDBStorageFactory;
 
 class RocksDBStorageProvider : public IStorage
@@ -19,12 +20,13 @@ class RocksDBStorageProvider : public IStorage
     std::unique_ptr<rocksdb::WriteBatchWithIndex> m_spbatch;
     const rocksdb::Snapshot *m_psnapshot = nullptr;
     std::shared_ptr<rocksdb::ColumnFamilyHandle> m_spcolfamily;
+    std::shared_ptr<rocksdb::ColumnFamilyHandle> m_spexpirecolfamily;
     rocksdb::ReadOptions m_readOptionsTemplate;
     size_t m_count = 0;
     mutable fastlock m_lock {"RocksDBStorageProvider"};
 
 public:
-    RocksDBStorageProvider(RocksDBStorageFactory *pfactory, std::shared_ptr<rocksdb::DB> &spdb, std::shared_ptr<rocksdb::ColumnFamilyHandle> &spcolfam, const rocksdb::Snapshot *psnapshot, size_t count);
+    RocksDBStorageProvider(RocksDBStorageFactory *pfactory, std::shared_ptr<rocksdb::DB> &spdb, std::shared_ptr<rocksdb::ColumnFamilyHandle> &spcolfam, std::shared_ptr<rocksdb::ColumnFamilyHandle> &spexpirecolfam, const rocksdb::Snapshot *psnapshot, size_t count);
     ~RocksDBStorageProvider();
 
     virtual void insert(const char *key, size_t cchKey, void *data, size_t cb, bool fOverwrite) override;
@@ -34,6 +36,11 @@ public:
     virtual bool enumerate(callback fn) const override;
     virtual bool enumerate_hashslot(callback fn, unsigned int hashslot) const override;
 
+    virtual std::vector<std::string> getExpirationCandidates(unsigned int count) override;
+    virtual std::vector<std::string> getEvictionCandidates(unsigned int count) override;
+    virtual void setExpire(const char *key, size_t cchKey, long long expire) override;
+    virtual void removeExpire(const char *key, size_t cchKey, long long expire) override;
+
     virtual const IStorage *clone() const override;
 
     virtual void beginWriteBatch() override;
@@ -50,6 +57,7 @@ public:
 
 protected:
     bool FKeyExists(std::string&) const;
+    bool FExpireExists(std::string&) const;
 
     const rocksdb::ReadOptions &ReadOptions() const { return m_readOptionsTemplate; }
     rocksdb::WriteOptions WriteOptions() const;

src/storage/rocksdbfactor_internal.h

@@ -5,6 +5,7 @@ class RocksDBStorageFactory : public IStorageFactory
 {
     std::shared_ptr<rocksdb::DB> m_spdb;    // Note: This must be first so it is deleted last
     std::vector<std::unique_ptr<rocksdb::ColumnFamilyHandle>> m_vecspcols;
+    std::vector<std::unique_ptr<rocksdb::ColumnFamilyHandle>> m_vecspexpirecols;
     std::shared_ptr<rocksdb::SstFileManager> m_pfilemanager;
     std::string m_path;
     bool m_fCreatedTempFolder = false;

src/storage/rocksdbfactory.cpp

@@ -61,8 +61,8 @@ RocksDBStorageFactory::RocksDBStorageFactory(const char *dbfile, int dbnum, cons
     auto status = rocksdb::DB::ListColumnFamilies(rocksdb::Options(), dbfile, &vecT);
     // RocksDB requires we know the count of col families before opening, if the user only wants to see less
     //  we still have to make room for all column family handles regardless
-    if (status.ok() && (int)vecT.size() > dbnum)
-        dbnum = (int)vecT.size();
+    if (status.ok() && (int)vecT.size()/2 > dbnum)
+        dbnum = (int)vecT.size()/2;
 
     std::vector<rocksdb::ColumnFamilyDescriptor> veccoldesc;
     veccoldesc.push_back(rocksdb::ColumnFamilyDescriptor(rocksdb::kDefaultColumnFamilyName, rocksdb::ColumnFamilyOptions()));  // ignore default col family
@@ -79,6 +79,7 @@ RocksDBStorageFactory::RocksDBStorageFactory(const char *dbfile, int dbnum, cons
         rocksdb::ColumnFamilyOptions cf_options(options);
         cf_options.level_compaction_dynamic_level_bytes = true;
         veccoldesc.push_back(rocksdb::ColumnFamilyDescriptor(std::to_string(idb), cf_options));
+        veccoldesc.push_back(rocksdb::ColumnFamilyDescriptor(std::to_string(idb) + "_expires", cf_options));
     }
 
     if (rgchConfig != nullptr)
@@ -100,23 +101,29 @@ RocksDBStorageFactory::RocksDBStorageFactory(const char *dbfile, int dbnum, cons
     m_spdb = std::shared_ptr<rocksdb::DB>(db);
     for (auto handle : handles)
     {
-        std::string strVersion;
-        auto status = m_spdb->Get(rocksdb::ReadOptions(), handle, rocksdb::Slice(version_key, sizeof(version_key)), &strVersion);
-        if (!status.ok())
-        {
-            setVersion(handle);
-        }
-        else
-        {
-            SymVer ver = parseVersion(strVersion.c_str());
-            auto cmp = compareVersion(&ver);
-            if (cmp == NewerVersion)
-                throw "Cannot load FLASH database created by newer version of KeyDB";
-            if (cmp == OlderVersion)
+        if (handle->GetName().size() > 7 && !strncmp(handle->GetName().substr(handle->GetName().size() - 7).c_str(), "expires", 7)) {
+            m_vecspexpirecols.emplace_back(handle);
+        } else {
+            std::string strVersion;
+            auto status = m_spdb->Get(rocksdb::ReadOptions(), handle, rocksdb::Slice(version_key, sizeof(version_key)), &strVersion);
+            if (!status.ok())
+            {
                 setVersion(handle);
+            }
+            else
+            {
+                SymVer ver = parseVersion(strVersion.c_str());
+                auto cmp = compareVersion(&ver);
+                if (cmp == NewerVersion)
+                    throw "Cannot load FLASH database created by newer version of KeyDB";
+                if (cmp == IncompatibleVersion)
+                    throw "Cannot load FLASH database from before 6.3.4";
+                if (cmp == OlderVersion)
+                    setVersion(handle);
+            }
+            m_vecspcols.emplace_back(handle);
         }
-        m_vecspcols.emplace_back(handle);
-    }   
+    }
 }
 
 RocksDBStorageFactory::~RocksDBStorageFactory()
@@ -156,6 +163,7 @@ IStorage *RocksDBStorageFactory::create(int db, key_load_iterator iter, void *pr
 {
     ++db;   // skip default col family
     std::shared_ptr<rocksdb::ColumnFamilyHandle> spcolfamily(m_vecspcols[db].release());
+    std::shared_ptr<rocksdb::ColumnFamilyHandle> spexpirecolfamily(m_vecspexpirecols[db].release());
     size_t count = 0;
     bool fUnclean = false;
     
@@ -192,7 +200,7 @@ IStorage *RocksDBStorageFactory::create(int db, key_load_iterator iter, void *pr
             ++count;
         }
     }
-    return new RocksDBStorageProvider(this, m_spdb, spcolfamily, nullptr, count);
+    return new RocksDBStorageProvider(this, m_spdb, spcolfamily, spexpirecolfamily, nullptr, count);
 }
 
 const char *RocksDBStorageFactory::name() const

src/storage/teststorageprovider.h

@@ -1,6 +1,7 @@
 #include "../IStorage.h"
 #include <string>
 #include <unordered_map>
+#include <vector>
 
 class TestStorageFactory : public IStorageFactory
 {
@@ -28,6 +29,11 @@ public:
     virtual bool enumerate_hashslot(callback fn, unsigned int hashslot) const override;
     virtual size_t count() const override;
 
+    virtual std::vector<std::string> getExpirationCandidates(unsigned int) override { return std::vector<std::string>(); }
+    virtual std::vector<std::string> getEvictionCandidates(unsigned int) override { return std::vector<std::string>(); }
+    virtual void setExpire(const char *, size_t, long long) override {}
+    virtual void removeExpire(const char *, size_t, long long) override {}
+
     virtual void flush() override;
 
     /* This is permitted to be a shallow clone */

src/version.h

@@ -7,6 +7,7 @@ enum VersionCompareResult
     EqualVersion,
     OlderVersion,
     NewerVersion,
+    IncompatibleVersion,
 };
 
 struct SymVer

tests/unit/flash.tcl

@@ -74,8 +74,11 @@ if {$::flash_enabled} {
             r set testkey foo ex 1
             r flushall cache
             assert_equal {1} [r dbsize]
-            after 1500
-            assert_equal {0} [r dbsize]
+            wait_for_condition 50 1000 {
+                [r dbsize] == 0
+            } else {
+                fail "key is not expired"
+            }
         }
 
         test { SUBKEY EXPIRE persists after cache flush } {
@@ -140,17 +143,17 @@ if {$::flash_enabled} {
                     r set $numkeys xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
                     incr numkeys
                     if {[s used_memory]+1024 >= $limit} {
-                    break
+                        break
                     }
                 }
                 # Add additional keys to force eviction
                 # should still be under the limit for maxmemory, however all keys set should still exist between flash and memory
                 # check same number of keys exist in addition to values of first and last keys
                 set err 0
-            set extra_keys [expr floor([expr ($limit * 0.4) / 1024])]
+                set extra_keys [expr floor([expr ($limit * 0.4) / 1024])]
                 for {set j 0} {$j < $extra_keys} {incr j} {
                     catch {
-                    r set p2$j xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
+                        r set p2$j xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
                     } err
                     assert {$err == {OK}}
                 }
@@ -165,7 +168,49 @@ if {$::flash_enabled} {
                 assert {[r get last] == {val}}
                 r flushall
             }
-        }
 
+            test "FLASH - is flash eviction working? (policy $policy)" {
+                # Get the current memory limit and calculate a new limit.
+                # Set limit to 100M.
+                set used [s used_memory]
+                set limit [expr {$used+60*1024*1024}]
+                r config set maxmemory $limit
+                r config set maxmemory-policy $policy
+                # Now add keys equivalent to 1024b until the limit is almost reached.
+                set numkeys 0
+                r set first val
+                while 1 {
+                    r set $numkeys xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
+                    incr numkeys
+                    if {[s used_memory]+1024 >= $limit} {
+                        break
+                    }
+                }
+                # Add additional keys to force eviction
+                # should still be under the limit for maxmemory, however all keys set should still exist between flash and memory
+                # check same number of keys exist in addition to values of first and last keys
+                set err 0
+                set extra_keys [expr floor([expr ($limit * 0.4) / 1024])]
+                for {set j 0} {$j < $extra_keys} {incr j} {
+                    catch {
+                        r set p2$j xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
+                    } err
+                    assert {$err == {OK}}
+                }
+                if {[log_file_matches [srv 0 stdout] "*Failed to evict*"]} {
+                    fail "Server did not evict cleanly (detected full flush)"
+                }
+                r set last val
+                r debug flush-storage
+                r config set maxstorage 1
+                r config set maxmemory 1
+                set dbsize [r dbsize]
+                # after setting maxstorage and memory below used amount we should evict from storage provider
+                assert {$dbsize < $numkeys+$extra_keys+2}
+                r config set maxstorage 0
+                r config set maxmemory 0
+                r flushall
+            }
+        }
     }
 }

tests/unit/moduleapi/load.tcl

@@ -1,7 +1,7 @@
 set testmodule [file normalize tests/modules/load.so]
 
 if {$::flash_enabled} {
-    start_server [list tags [list "modules"] overrides [list storage-provider {flash ./rocks.db.master} databases 256 loadmodule $testmodule]] {
+    start_server [list tags [list "modules"] overrides [list storage-provider {flash ./rocks.db.master.load.test} databases 256 loadmodule $testmodule]] {
         test "Module is notified of keys loaded from flash" {
             r flushall
             r set foo bar