don't count base repl buffer size as overhead

* restore fork impl of save to slave sockets

* default to fork bgsave

deps/Makefile

@@ -95,7 +95,7 @@ JEMALLOC_LDFLAGS= $(LDFLAGS)
 
 jemalloc: .make-prerequisites
 	@printf '%b %b\n' $(MAKECOLOR)MAKE$(ENDCOLOR) $(BINCOLOR)$@$(ENDCOLOR)
-	cd jemalloc && ./configure --with-version=5.2.1-0-g0 --with-lg-quantum=3 --with-jemalloc-prefix=je_ --disable-cxx CFLAGS="$(JEMALLOC_CFLAGS)" LDFLAGS="$(JEMALLOC_LDFLAGS)"
+	cd jemalloc && ./configure --with-version=5.2.1-0-g0 --with-lg-quantum=3 --disable-cxx CFLAGS="$(JEMALLOC_CFLAGS)" LDFLAGS="$(JEMALLOC_LDFLAGS)"
 	cd jemalloc && $(MAKE) CFLAGS="$(JEMALLOC_CFLAGS)" LDFLAGS="$(JEMALLOC_LDFLAGS)" lib/libjemalloc.a
 
 .PHONY: jemalloc

src/Makefile

@@ -25,7 +25,7 @@ NODEPS:=clean distclean
 
 # Default settings
 STD=-pedantic -DREDIS_STATIC=''
-CXX_STD=-std=c++14 -pedantic -fno-rtti -D__STDC_FORMAT_MACROS  
+CXX_STD=-std=c++17 -pedantic -fno-rtti -D__STDC_FORMAT_MACROS  
 ifneq (,$(findstring clang,$(CC)))
   STD+=-Wno-c11-extensions
 else

src/SnapshotPayloadParseState.cpp

@@ -140,7 +140,7 @@ void SnapshotPayloadParseState::flushQueuedKeys() {
     auto &insertsInFlightTmp = insertsInFlight; // C++ GRRRRRRRRRRRRRRRR, we don't want to capute "this" because that's dangerous
     if (current_database < cserver.dbnum) {
         g_pserver->asyncworkqueue->AddWorkFunction([idb, vecqueuedKeys = std::move(this->vecqueuedKeys), vecqueuedKeysCb = std::move(this->vecqueuedKeysCb), vecqueuedVals = std::move(this->vecqueuedVals), vecqueuedValsCb = std::move(this->vecqueuedValsCb), &insertsInFlightTmp, pallocator = m_spallocator.release()]() mutable {
-            g_pserver->db[idb]->bulkStorageInsert(vecqueuedKeys.data(), vecqueuedKeysCb.data(), vecqueuedVals.data(), vecqueuedValsCb.data(), vecqueuedKeys.size());
+            g_pserver->db[idb]->bulkDirectStorageInsert(vecqueuedKeys.data(), vecqueuedKeysCb.data(), vecqueuedVals.data(), vecqueuedValsCb.data(), vecqueuedKeys.size());
             --insertsInFlightTmp;
             delete pallocator;
         });

src/childinfo.cpp

@@ -42,6 +42,7 @@ typedef struct {
  * RDB / AOF saving process from the child to the parent (for instance
  * the amount of copy on write memory used) */
 void openChildInfoPipe(void) {
+    serverAssert(g_pserver->child_info_pipe[0] == -1);
     if (pipe(g_pserver->child_info_pipe) == -1) {
         /* On error our two file descriptors should be still set to -1,
          * but we call anyway closeChildInfoPipe() since can't hurt. */

src/config.cpp

@@ -35,6 +35,7 @@
 
 #include <fcntl.h>
 #include <sys/stat.h>
+#include <sys/wait.h>
 #ifdef __linux__
 #include <sys/sysinfo.h>
 #endif
@@ -2456,6 +2457,32 @@ static int isValidAOFfilename(char *val, const char **err) {
     return 1;
 }
 
+static int isValidS3Bucket(char *s3bucket, const char **err) {
+    int status = EXIT_FAILURE;
+    pid_t pid = fork();
+    if (pid < 0)
+    {
+        *err = "couldn't fork to call aws cli";
+        return 0;
+    }
+
+    if (pid == 0)
+    {
+        execlp("aws", "aws", "s3", "ls", s3bucket);
+        exit(EXIT_FAILURE);
+    }
+    else 
+    {
+        waitpid(pid, &status, 0);
+    }
+
+    if (status != EXIT_SUCCESS) {
+        *err = "could not access s3 bucket";
+        return 0;
+    }
+    return 1;
+}
+
 /* Validate specified string is a valid proc-title-template */
 static int isValidProcTitleTemplate(char *val, const char **err) {
     if (!validateProcTitleTemplate(val)) {
@@ -2779,7 +2806,7 @@ standardConfig configs[] = {
     createBoolConfig("appendonly", NULL, MODIFIABLE_CONFIG, g_pserver->aof_enabled, 0, NULL, updateAppendonly),
     createBoolConfig("cluster-allow-reads-when-down", NULL, MODIFIABLE_CONFIG, g_pserver->cluster_allow_reads_when_down, 0, NULL, NULL),
     createBoolConfig("delete-on-evict", NULL, MODIFIABLE_CONFIG, cserver.delete_on_evict, 0, NULL, NULL),
-    createBoolConfig("use-fork", NULL, IMMUTABLE_CONFIG, cserver.fForkBgSave, 0, NULL, NULL),
+    createBoolConfig("use-fork", NULL, IMMUTABLE_CONFIG, cserver.fForkBgSave, 1, NULL, NULL),
     createBoolConfig("io-threads-do-reads", NULL, IMMUTABLE_CONFIG, fDummy, 0, NULL, NULL),
     createBoolConfig("time-thread-priority", NULL, IMMUTABLE_CONFIG, cserver.time_thread_priority, 0, NULL, NULL),
     createBoolConfig("prefetch-enabled", NULL, MODIFIABLE_CONFIG, g_pserver->prefetch_enabled, 1, NULL, NULL),
@@ -2803,6 +2830,7 @@ standardConfig configs[] = {
     createStringConfig("cluster-announce-ip", NULL, MODIFIABLE_CONFIG, EMPTY_STRING_IS_NULL, g_pserver->cluster_announce_ip, NULL, NULL, NULL),
     createStringConfig("syslog-ident", NULL, IMMUTABLE_CONFIG, ALLOW_EMPTY_STRING, g_pserver->syslog_ident, "redis", NULL, NULL),
     createStringConfig("dbfilename", NULL, MODIFIABLE_CONFIG, ALLOW_EMPTY_STRING, g_pserver->rdb_filename, CONFIG_DEFAULT_RDB_FILENAME, isValidDBfilename, NULL),
+    createStringConfig("db-s3-object", NULL, MODIFIABLE_CONFIG, EMPTY_STRING_IS_NULL, g_pserver->rdb_s3bucketpath, NULL, isValidS3Bucket, NULL),
     createStringConfig("appendfilename", NULL, IMMUTABLE_CONFIG, ALLOW_EMPTY_STRING, g_pserver->aof_filename, "appendonly.aof", isValidAOFfilename, NULL),
     createStringConfig("server_cpulist", NULL, IMMUTABLE_CONFIG, EMPTY_STRING_IS_NULL, g_pserver->server_cpulist, NULL, NULL, NULL),
     createStringConfig("bio_cpulist", NULL, IMMUTABLE_CONFIG, EMPTY_STRING_IS_NULL, g_pserver->bio_cpulist, NULL, NULL, NULL),

src/db.cpp

@@ -1931,6 +1931,7 @@ void setExpire(client *c, redisDb *db, robj *key, expireEntry &&e)
  * is associated with this key (i.e. the key is non volatile) */
 expireEntry *redisDbPersistentDataSnapshot::getExpire(const char *key) {
     /* No expire? return ASAP */
+    std::unique_lock<fastlock> ul(g_expireLock);
     if (expireSize() == 0)
         return nullptr;
 
@@ -2607,12 +2608,15 @@ void clusterStorageLoadCallback(const char *rgchkey, size_t cch, void *)
     slotToKeyUpdateKeyCore(rgchkey, cch, true /*add*/);
 }
 
-void moduleLoadCallback(const char * rgchKey, size_t cchKey, void *data) {
-    if (g_pserver->cluster_enabled) {
-        clusterStorageLoadCallback(rgchKey, cchKey, data);
-    }
-    robj *keyobj = createEmbeddedStringObject(rgchKey, cchKey);
-    moduleNotifyKeyspaceEvent(NOTIFY_LOADED, "loaded", keyobj, *(int *)data);
+void moduleLoadCallback(const char * rgchKey, size_t, void *data) {
+    redisObjectStack keyobj;
+    initStaticStringObject(keyobj, const_cast<char *>(rgchKey));
+    moduleNotifyKeyspaceEvent(NOTIFY_LOADED, "loaded", &keyobj, *(int *)data);
+}
+
+void moduleClusterLoadCallback(const char * rgchKey, size_t cchKey, void *data) {
+    clusterStorageLoadCallback(rgchKey, cchKey, data);
+    moduleLoadCallback(rgchKey, cchKey, data);
 }
 
 void redisDb::initialize(int id)
@@ -2633,7 +2637,7 @@ void redisDb::storageProviderInitialize()
 {
     if (g_pserver->m_pstorageFactory != nullptr)
     {
-        IStorageFactory::key_load_iterator itr = moduleLoadCallback;
+        IStorageFactory::key_load_iterator itr = g_pserver->cluster_enabled ? moduleClusterLoadCallback : moduleLoadCallback;
         this->setStorageProvider(StorageCache::create(g_pserver->m_pstorageFactory, id, itr, &id));
     }
 }
@@ -2724,8 +2728,11 @@ void redisDbPersistentData::clear(void(callback)(void*))
         m_cnewKeysPending = 0;
         m_fAllChanged++;
     }
+    {
+    std::unique_lock<fastlock> ul(g_expireLock);
     delete m_setexpire;
     m_setexpire = new (MALLOC_LOCAL) expireset();
+    }
     if (m_spstorage != nullptr)
         m_spstorage->clear(callback);
     dictEmpty(m_pdictTombstone,callback);
@@ -2858,7 +2865,8 @@ LNotFound:
             {
                 dictAdd(m_pdict, sdsNewKey, o);
                 o->SetFExpires(spexpire != nullptr);
-
+                
+                std::unique_lock<fastlock> ul(g_expireLock);
                 if (spexpire != nullptr)
                 {
                     auto itr = m_setexpire->find(sdsKey);
@@ -2881,6 +2889,7 @@ LNotFound:
     if (*pde != nullptr && dictGetVal(*pde) != nullptr)
     {
         robj *o = (robj*)dictGetVal(*pde);
+        std::unique_lock<fastlock> ul(g_expireLock);
         serverAssert(o->FExpires() == (m_setexpire->find(sdsKey) != m_setexpire->end()));
     }
 }
@@ -3002,8 +3011,16 @@ void redisDbPersistentData::processChangesAsync(std::atomic<int> &pendingJobs)
     });
 }
 
-void redisDbPersistentData::bulkStorageInsert(char **rgKeys, size_t *rgcbKeys, char **rgVals, size_t *rgcbVals, size_t celem)
+/* This function is to bulk insert directly to storage provider bypassing in memory, assumes rgKeys and rgVals are not sds strings */
+void redisDbPersistentData::bulkDirectStorageInsert(char **rgKeys, size_t *rgcbKeys, char **rgVals, size_t *rgcbVals, size_t celem)
 {
+    if (g_pserver->cluster_enabled) {
+        aeAcquireLock();
+        for (size_t i = 0; i < celem; i++) {
+            slotToKeyUpdateKeyCore(rgKeys[i], rgcbKeys[i], 1);
+        }
+        aeReleaseLock();
+    }
     m_spstorage->bulkInsert(rgKeys, rgcbKeys, rgVals, rgcbVals, celem);
 }
 
@@ -3245,6 +3262,7 @@ std::unique_ptr<expireEntry> deserializeExpire(sds key, const char *str, size_t
 
 sds serializeStoredObjectAndExpire(redisDbPersistentData *db, const char *key, robj_roptr o)
 {
+    std::unique_lock<fastlock> ul(g_expireLock);
     auto itrExpire = db->setexpire()->find(key);
     const expireEntry *pexpire = nullptr;
     if (itrExpire != db->setexpire()->end())
@@ -3265,7 +3283,7 @@ int dbnumFromDb(redisDb *db)
     serverPanic("invalid database pointer");
 }
 
-bool redisDbPersistentData::prefetchKeysAsync(client *c, parsed_command &command, bool fExecOK)
+void redisDbPersistentData::prefetchKeysAsync(client *c, parsed_command &command)
 {
     if (m_spstorage == nullptr) {
 #if defined(__x86_64__) || defined(__i386__)
@@ -3277,7 +3295,7 @@ bool redisDbPersistentData::prefetchKeysAsync(client *c, parsed_command &command
             const char *cmd = szFromObj(command.argv[0]);
             if (!strcasecmp(cmd, "set") || !strcasecmp(cmd, "get")) {
                 if (c->db->m_spdbSnapshotHOLDER != nullptr)
-                    return false; // this is dangerous enough without a snapshot around
+                    return; // this is dangerous enough without a snapshot around
                 auto h = dictSdsHash(szFromObj(command.argv[1]));
                 for (int iht = 0; iht < 2; ++iht) {
                     auto hT = h & c->db->m_pdict->ht[iht].sizemask;
@@ -3297,7 +3315,7 @@ bool redisDbPersistentData::prefetchKeysAsync(client *c, parsed_command &command
             }
         }
 #endif
-        return false;
+        return;
     }
 
     AeLocker lock;
@@ -3307,10 +3325,10 @@ bool redisDbPersistentData::prefetchKeysAsync(client *c, parsed_command &command
     getKeysResult result = GETKEYS_RESULT_INIT;
     auto cmd = lookupCommand(szFromObj(command.argv[0]));
     if (cmd == nullptr)
-        return false; // Bad command? It's not for us to judge, just bail
+        return; // Bad command? It's not for us to judge, just bail
     
     if (command.argc < std::abs(cmd->arity))
-        return false; // Invalid number of args
+        return; // Invalid number of args
     
     int numkeys = getKeysFromCommand(cmd, command.argv, command.argc, &result);
     for (int ikey = 0; ikey < numkeys; ++ikey)
@@ -3343,7 +3361,6 @@ bool redisDbPersistentData::prefetchKeysAsync(client *c, parsed_command &command
         }
     }
 
-    bool fNoInsert = false;
     if (!vecInserts.empty()) {
         lock.arm(c);
         for (auto &tuple : vecInserts)
@@ -3359,19 +3376,18 @@ bool redisDbPersistentData::prefetchKeysAsync(client *c, parsed_command &command
                     // While unlocked this was already ensured
                     decrRefCount(o);
                     sdsfree(sharedKey);
-                    fNoInsert = true;
                 }
                 else
                 {
                     if (spexpire != nullptr) {
                         if (spexpire->when() < mstime()) {
-                            fNoInsert = true;
                             break;
                         }
                     }
                     dictAdd(m_pdict, sharedKey, o);
                     o->SetFExpires(spexpire != nullptr);
 
+                    std::unique_lock<fastlock> ul(g_expireLock);
                     if (spexpire != nullptr)
                     {
                         auto itr = m_setexpire->find(sharedKey);
@@ -3392,12 +3408,5 @@ bool redisDbPersistentData::prefetchKeysAsync(client *c, parsed_command &command
         lock.disarm();
     }
 
-    if (fExecOK && !fNoInsert && cmd->proc == getCommand && !vecInserts.empty()) {
-        robj *o = std::get<1>(vecInserts[0]);
-        if (o != nullptr) {
-            addReplyBulk(c, o);
-            return true;
-        }
-    }
-    return false;
-}
+    return;
+}

src/debug.cpp

@@ -341,10 +341,10 @@ void mallctl_int(client *c, robj **argv, int argc) {
     }
     size_t sz = sizeof(old);
     while (sz > 0) {
-        if ((ret=je_mallctl(szFromObj(argv[0]), &old, &sz, argc > 1? &val: NULL, argc > 1?sz: 0))) {
+        if ((ret=mallctl(szFromObj(argv[0]), &old, &sz, argc > 1? &val: NULL, argc > 1?sz: 0))) {
             if (ret == EPERM && argc > 1) {
                 /* if this option is write only, try just writing to it. */
-                if (!(ret=je_mallctl(szFromObj(argv[0]), NULL, 0, &val, sz))) {
+                if (!(ret=mallctl(szFromObj(argv[0]), NULL, 0, &val, sz))) {
                     addReply(c, shared.ok);
                     return;
                 }
@@ -375,7 +375,7 @@ void mallctl_string(client *c, robj **argv, int argc) {
     char *old;
     size_t sz = sizeof(old);
     /* for strings, it seems we need to first get the old value, before overriding it. */
-    if ((rret=je_mallctl(szFromObj(argv[0]), &old, &sz, NULL, 0))) {
+    if ((rret=mallctl(szFromObj(argv[0]), &old, &sz, NULL, 0))) {
         /* return error unless this option is write only. */
         if (!(rret == EPERM && argc > 1)) {
             addReplyErrorFormat(c,"%s", strerror(rret));
@@ -387,7 +387,7 @@ void mallctl_string(client *c, robj **argv, int argc) {
         char **valref = &val;
         if ((!strcmp(val,"VOID")))
             valref = NULL, sz = 0;
-        wret = je_mallctl(szFromObj(argv[0]), NULL, 0, valref, sz);
+        wret = mallctl(szFromObj(argv[0]), NULL, 0, valref, sz);
     }
     if (!rret)
         addReplyBulkCString(c, old);

src/evict.cpp

@@ -380,7 +380,7 @@ size_t freeMemoryGetNotCountedMemory(void) {
 
     /* also don't count the replication backlog memory
      * that's where the replication clients get their memory from */
-    overhead += g_pserver->repl_backlog_size;
+    overhead += g_pserver->repl_backlog_size - g_pserver->repl_backlog_config_size;
 
     if (g_pserver->aof_state != AOF_OFF) {
         overhead += sdsalloc(g_pserver->aof_buf)+aofRewriteBufferSize();
@@ -771,6 +771,7 @@ int performEvictions(bool fPreSnapshot) {
                 if (db->removeCachedValue(bestkey, &deT)) {
                     mem_freed += splazy->addEntry(db->dictUnsafeKeyOnly(), deT);
                     ckeysFailed = 0;
+		    g_pserver->stat_evictedkeys++;
                 }
                 else {
                     delta = 0;

src/expire.cpp

@@ -481,6 +481,7 @@ void expireSlaveKeys(void) {
     if (slaveKeysWithExpire == NULL ||
         dictSize(slaveKeysWithExpire) == 0) return;
 
+    std::unique_lock<fastlock> ul(g_expireLock);
     int cycles = 0, noexpire = 0;
     mstime_t start = mstime();
     while(1) {

src/networking.cpp

@@ -2595,9 +2595,7 @@ void parseClientCommandBuffer(client *c) {
             (g_pserver->m_pstorageFactory || aeLockContested(cserver.cthreads/2) || cserver.cthreads == 1) && !GlobalLocksAcquired()) {
             auto &query = c->vecqueuedcmd.back();
             if (query.argc > 0 && query.argc == query.argcMax) {
-                if (c->db->prefetchKeysAsync(c, query, c->vecqueuedcmd.size() == 1)) {
-                    c->vecqueuedcmd.erase(c->vecqueuedcmd.begin());
-                }
+                c->db->prefetchKeysAsync(c, query);
             }
         }
         c->reqtype = 0;

src/new.cpp

@@ -41,10 +41,3 @@ void operator delete(void *p, std::size_t) noexcept
 }
 
 #endif
-
-#if defined(USE_JEMALLOC)
-extern "C" size_t malloc_usable_size(void *ptr)
-{
-    return zmalloc_usable_size(ptr);
-}
-#endif

src/object.cpp

@@ -1541,7 +1541,7 @@ void memoryCommand(client *c) {
     } else if (!strcasecmp(szFromObj(c->argv[1]),"malloc-stats") && c->argc == 2) {
 #if defined(USE_JEMALLOC)
         sds info = sdsempty();
-        je_malloc_stats_print(inputCatSds, &info, NULL);
+        malloc_stats_print(inputCatSds, &info, NULL);
         addReplyVerbatim(c,info,sdslen(info),"txt");
         sdsfree(info);
 #else
@@ -1668,6 +1668,7 @@ robj *deserializeStoredObjectCore(const void *data, size_t cb)
 robj *deserializeStoredObject(const redisDbPersistentData *db, const char *key, const void *data, size_t cb)
 {
     robj *o = deserializeStoredObjectCore(data, cb);
+    std::unique_lock<fastlock> ul(g_expireLock);
     o->SetFExpires(db->setexpire()->exists(key));
     return o;
 }

src/rdb.cpp

@@ -1334,7 +1334,7 @@ int rdbSaveRio(rio *rdb, const redisDbPersistentDataSnapshot **rgpdb, int *error
     if (rdbSaveModulesAux(rdb, REDISMODULE_AUX_BEFORE_RDB) == -1) goto werr;
 
     for (j = 0; j < cserver.dbnum; j++) {
-        const redisDbPersistentDataSnapshot *db = rgpdb[j];
+        const redisDbPersistentDataSnapshot *db = rgpdb != nullptr ? rgpdb[j] : g_pserver->db[j];
         if (db->size() == 0) continue;
 
         /* Write the SELECT DB opcode */
@@ -1490,7 +1490,7 @@ int rdbSaveFile(char *filename, const redisDbPersistentDataSnapshot **rgpdb, rdb
     rio rdb;
     int error = 0;
 
-    snprintf(tmpfile,sizeof(tmpfile),"temp-%d-%d.rdb", getpid(), g_pserver->rdbThreadVars.tmpfileNum);
+    getTempFileName(tmpfile, g_pserver->rdbThreadVars.tmpfileNum);
     fp = fopen(tmpfile,"w");
     if (!fp) {
         char *cwdp = getcwd(cwd,MAXPATHLEN);
@@ -1660,12 +1660,14 @@ int launchRdbSaveThread(pthread_t &child, rdbSaveInfo *rsi)
         pthread_attr_t tattr;
         pthread_attr_init(&tattr);
         pthread_attr_setstacksize(&tattr, 1 << 23); // 8 MB
+        openChildInfoPipe();
         if (pthread_create(&child, &tattr, rdbSaveThread, args)) {
             pthread_attr_destroy(&tattr);
             for (int idb = 0; idb < cserver.dbnum; ++idb)
                 g_pserver->db[idb]->endSnapshot(args->rgpdb[idb]);
             args->~rdbSaveThreadArgs();
             zfree(args);
+            closeChildInfoPipe();
             return C_ERR;
         }
         pthread_attr_destroy(&tattr);
@@ -1683,13 +1685,11 @@ int rdbSaveBackground(rdbSaveInfo *rsi) {
 
     g_pserver->dirty_before_bgsave = g_pserver->dirty;
     g_pserver->lastbgsave_try = time(NULL);
-    openChildInfoPipe();
 
     start = ustime();
     latencyStartMonitor(g_pserver->rdb_save_latency);
 
     if (launchRdbSaveThread(child, rsi) != C_OK) {
-        closeChildInfoPipe();
         g_pserver->lastbgsave_status = C_ERR;
         serverLog(LL_WARNING,"Can't save in background: fork: %s",
             strerror(errno));
@@ -1715,7 +1715,7 @@ void getTempFileName(char tmpfile[], int tmpfileNum) {
     char tmpfileNumString[214];
 
     /* Generate temp rdb file name using aync-signal safe functions. */
-    int pid_len = ll2string(pid, sizeof(pid), getpid());
+    int pid_len = ll2string(pid, sizeof(pid), g_pserver->in_fork_child ? getppid() : getpid());
     int tmpfileNum_len = ll2string(tmpfileNumString, sizeof(tmpfileNumString), tmpfileNum);
     strcpy(tmpfile, "temp-");
     strncpy(tmpfile+5, pid, pid_len);
@@ -3832,57 +3832,130 @@ int rdbSaveToSlavesSockets(rdbSaveInfo *rsi) {
     }
 
     /* Create the child process. */
-    openChildInfoPipe();
+    if (cserver.fForkBgSave) {
+        pid_t childpid;
+        if ((childpid = redisFork(CHILD_TYPE_RDB)) == 0) {
+            /* Child */
+            int retval, dummy;
+            rio rdb;
 
-    for (int idb = 0; idb < cserver.dbnum; ++idb)
-        args->rgpdb[idb] = g_pserver->db[idb]->createSnapshot(getMvccTstamp(), false /*fOptional*/);
+            rioInitWithFd(&rdb,args->rdb_pipe_write);
 
-    g_pserver->rdbThreadVars.tmpfileNum++;
-    g_pserver->rdbThreadVars.fRdbThreadCancel = false;
-    pthread_attr_t tattr;
-    pthread_attr_init(&tattr);
-    pthread_attr_setstacksize(&tattr, 1 << 23); // 8 MB
-    if (pthread_create(&child, &tattr, rdbSaveToSlavesSocketsThread, args)) {
-        pthread_attr_destroy(&tattr);
-        serverLog(LL_WARNING,"Can't save in background: fork: %s",
-            strerror(errno));
+            redisSetProcTitle("keydb-rdb-to-slaves");
+            redisSetCpuAffinity(g_pserver->bgsave_cpulist);
 
-        /* Undo the state change. The caller will perform cleanup on
-            * all the slaves in BGSAVE_START state, but an early call to
-            * replicationSetupSlaveForFullResync() turned it into BGSAVE_END */
-        listRewind(g_pserver->slaves,&li);
-        while((ln = listNext(&li))) {
-            client *replica = (client*)ln->value;
-            if (replica->replstate == SLAVE_STATE_WAIT_BGSAVE_END) {
-                replica->replstate = SLAVE_STATE_WAIT_BGSAVE_START;
+            retval = rdbSaveRioWithEOFMark(&rdb,nullptr,nullptr,rsi);
+            if (retval == C_OK && rioFlush(&rdb) == 0)
+                retval = C_ERR;
+
+            if (retval == C_OK) {
+                sendChildCowInfo(CHILD_INFO_TYPE_RDB_COW_SIZE, "RDB");
             }
-        }
-        close(args->rdb_pipe_write);
-        close(g_pserver->rdb_pipe_read);
-        zfree(g_pserver->rdb_pipe_conns);
-        close(args->safe_to_exit_pipe);
-        g_pserver->rdb_pipe_conns = NULL;
-        g_pserver->rdb_pipe_numconns = 0;
-        g_pserver->rdb_pipe_numconns_writing = 0;
-        args->rsi.~rdbSaveInfo();
-        zfree(args);
-        closeChildInfoPipe();
-        return C_ERR;
-    }
-    pthread_attr_destroy(&tattr);
-    g_pserver->child_type = CHILD_TYPE_RDB;
 
-    serverLog(LL_NOTICE,"Background RDB transfer started");
-    g_pserver->rdb_save_time_start = time(NULL);
-    serverAssert(!g_pserver->rdbThreadVars.fRdbThreadActive);
-    g_pserver->rdbThreadVars.rdb_child_thread = child;
-    g_pserver->rdbThreadVars.fRdbThreadActive = true;
-    g_pserver->rdb_child_type = RDB_CHILD_TYPE_SOCKET;
-    aePostFunction(g_pserver->rgthreadvar[IDX_EVENT_LOOP_MAIN].el, []{
-        if (aeCreateFileEvent(serverTL->el, g_pserver->rdb_pipe_read, AE_READABLE, rdbPipeReadHandler,NULL) == AE_ERR) {
-            serverPanic("Unrecoverable error creating server.rdb_pipe_read file event.");
+            rioFreeFd(&rdb);
+            /* wake up the reader, tell it we're done. */
+            close(args->rdb_pipe_write);
+            close(g_pserver->rdb_child_exit_pipe); /* close write end so that we can detect the close on the parent. */
+            /* hold exit until the parent tells us it's safe. we're not expecting
+            * to read anything, just get the error when the pipe is closed. */
+            dummy = read(args->safe_to_exit_pipe, pipefds, 1);
+            UNUSED(dummy);
+            exitFromChild((retval == C_OK) ? 0 : 1);
+        } else {
+            /* Parent */
+            close(args->safe_to_exit_pipe);
+            if (childpid == -1) {
+                serverLog(LL_WARNING,"Can't save in background: fork: %s",
+                    strerror(errno));
+
+                /* Undo the state change. The caller will perform cleanup on
+                * all the slaves in BGSAVE_START state, but an early call to
+                * replicationSetupSlaveForFullResync() turned it into BGSAVE_END */
+                listRewind(g_pserver->slaves,&li);
+                while((ln = listNext(&li))) {
+                    client *replica = (client*)ln->value;
+                    if (replica->replstate == SLAVE_STATE_WAIT_BGSAVE_END) {
+                        replica->replstate = SLAVE_STATE_WAIT_BGSAVE_START;
+                    }
+                }
+                close(args->rdb_pipe_write);
+                close(g_pserver->rdb_pipe_read);
+                zfree(g_pserver->rdb_pipe_conns);
+                g_pserver->rdb_pipe_conns = NULL;
+                g_pserver->rdb_pipe_numconns = 0;
+                g_pserver->rdb_pipe_numconns_writing = 0;
+                args->rsi.~rdbSaveInfo();
+                zfree(args);
+            } else {
+                serverLog(LL_NOTICE,"Background RDB transfer started by pid %ld",
+                    (long)childpid);
+                g_pserver->rdb_save_time_start = time(NULL);
+                g_pserver->rdb_child_type = RDB_CHILD_TYPE_SOCKET;
+                g_pserver->rdbThreadVars.fRdbThreadActive = true;
+                updateDictResizePolicy();
+                close(args->rdb_pipe_write); /* close write in parent so that it can detect the close on the child. */
+                aePostFunction(g_pserver->rgthreadvar[IDX_EVENT_LOOP_MAIN].el, []{
+                    if (aeCreateFileEvent(serverTL->el, g_pserver->rdb_pipe_read, AE_READABLE, rdbPipeReadHandler, nullptr) == AE_ERR) {
+                        serverPanic("Unrecoverable error creating g_pserver->rdb_pipe_read file event.");
+                    }
+                });
+            }
+            return (childpid == -1) ? C_ERR : C_OK;
         }
-    });
+    }
+    else {
+        openChildInfoPipe();
+
+        for (int idb = 0; idb < cserver.dbnum; ++idb)
+            args->rgpdb[idb] = g_pserver->db[idb]->createSnapshot(getMvccTstamp(), false /*fOptional*/);
+
+        g_pserver->rdbThreadVars.tmpfileNum++;
+        g_pserver->rdbThreadVars.fRdbThreadCancel = false;
+        pthread_attr_t tattr;
+        pthread_attr_init(&tattr);
+        pthread_attr_setstacksize(&tattr, 1 << 23); // 8 MB
+        if (pthread_create(&child, &tattr, rdbSaveToSlavesSocketsThread, args)) {
+            pthread_attr_destroy(&tattr);
+            serverLog(LL_WARNING,"Can't save in background: fork: %s",
+                strerror(errno));
+
+            /* Undo the state change. The caller will perform cleanup on
+                * all the slaves in BGSAVE_START state, but an early call to
+                * replicationSetupSlaveForFullResync() turned it into BGSAVE_END */
+            listRewind(g_pserver->slaves,&li);
+            while((ln = listNext(&li))) {
+                client *replica = (client*)ln->value;
+                if (replica->replstate == SLAVE_STATE_WAIT_BGSAVE_END) {
+                    replica->replstate = SLAVE_STATE_WAIT_BGSAVE_START;
+                }
+            }
+            close(args->rdb_pipe_write);
+            close(g_pserver->rdb_pipe_read);
+            zfree(g_pserver->rdb_pipe_conns);
+            close(args->safe_to_exit_pipe);
+            g_pserver->rdb_pipe_conns = NULL;
+            g_pserver->rdb_pipe_numconns = 0;
+            g_pserver->rdb_pipe_numconns_writing = 0;
+            args->rsi.~rdbSaveInfo();
+            zfree(args);
+            closeChildInfoPipe();
+            return C_ERR;
+        }
+        pthread_attr_destroy(&tattr);
+        g_pserver->child_type = CHILD_TYPE_RDB;
+
+        serverLog(LL_NOTICE,"Background RDB transfer started");
+        g_pserver->rdb_save_time_start = time(NULL);
+        serverAssert(!g_pserver->rdbThreadVars.fRdbThreadActive);
+        g_pserver->rdbThreadVars.rdb_child_thread = child;
+        g_pserver->rdbThreadVars.fRdbThreadActive = true;
+        g_pserver->rdb_child_type = RDB_CHILD_TYPE_SOCKET;
+        aePostFunction(g_pserver->rgthreadvar[IDX_EVENT_LOOP_MAIN].el, []{
+            if (aeCreateFileEvent(serverTL->el, g_pserver->rdb_pipe_read, AE_READABLE, rdbPipeReadHandler, nullptr) == AE_ERR) {
+                serverPanic("Unrecoverable error creating server.rdb_pipe_read file event.");
+            }
+        });
+    }
 
     return C_OK; /* Unreached. */
 }

src/rdb.h

@@ -151,6 +151,7 @@ int rdbLoad(rdbSaveInfo *rsi, int rdbflags);
 int rdbLoadFile(const char *filename, rdbSaveInfo *rsi, int rdbflags);
 int rdbSaveBackground(rdbSaveInfo *rsi);
 int rdbSaveToSlavesSockets(rdbSaveInfo *rsi);
+void getTempFileName(char tmpfile[], int tmpfileNum);
 void rdbRemoveTempFile(pid_t childpid, int from_signal);
 int rdbSave(const redisDbPersistentDataSnapshot **rgpdb, rdbSaveInfo *rsi);
 int rdbSaveFile(char *filename, const redisDbPersistentDataSnapshot **rgpdb, rdbSaveInfo *rsi);

src/replication.cpp

@@ -3797,9 +3797,11 @@ int connectWithMaster(redisMaster *mi) {
     connSetPrivateData(mi->repl_transfer_s, mi);
     if (connConnect(mi->repl_transfer_s, mi->masterhost, mi->masterport,
                 NET_FIRST_BIND_ADDR, syncWithMaster) == C_ERR) {
-        int sev = g_pserver->enable_multimaster ? LL_NOTICE : LL_WARNING;   // with multimaster its not unheard of to intentiallionall have downed masters
-        serverLog(sev,"Unable to connect to MASTER: %s",
-                connGetLastError(mi->repl_transfer_s));
+        const char *err = "Unknown Error";
+        if (mi->repl_transfer_s->last_errno != 0)
+            err = connGetLastError(mi->repl_transfer_s);
+        int sev = g_pserver->enable_multimaster ? LL_NOTICE : LL_WARNING;   // with multimaster its not unheard of to intentionally have downed masters
+        serverLog(sev, "Unable to connect to MASTER: %s", err);
         connClose(mi->repl_transfer_s);
         mi->repl_transfer_s = NULL;
         return C_ERR;

src/scripting.cpp

@@ -1264,6 +1264,9 @@ void scriptingInit(int setup) {
 
     /* Finally set the table as 'redis' global var. */
     lua_setglobal(lua,"redis");
+    /* Set table as 'keydb' global var as well */
+    lua_getglobal(lua,"redis");
+    lua_setglobal(lua,"keydb");
 
     /* Replace math.random and math.randomseed with our implementations. */
     lua_getglobal(lua,"math");

src/server.cpp

@@ -2219,6 +2219,7 @@ void checkChildrenDone(void) {
             g_pserver->rdbThreadVars.fRdbThreadCancel = false;
             g_pserver->rdbThreadVars.fDone = false;
             if (exitcode == 0) receiveChildInfo();
+            closeChildInfoPipe();
         }
     }
     else if ((pid = waitpid(-1, &statloc, WNOHANG)) != 0) {
@@ -5171,7 +5172,7 @@ int prepareForShutdown(int flags) {
          * to unlink file actully) in background thread.
          * The temp rdb file fd may won't be closed when redis exits quickly,
          * but OS will close this fd when process exits. */
-        rdbRemoveTempFile(g_pserver->child_pid, 0);
+        rdbRemoveTempFile(g_pserver->rdbThreadVars.tmpfileNum, 0);
     }
 
     /* Kill module child if there is one. */

src/server.h

@@ -1204,12 +1204,12 @@ public:
     void disableKeyCache();
     bool keycacheIsEnabled();
 
-    bool prefetchKeysAsync(client *c, struct parsed_command &command, bool fExecOK);
+    void prefetchKeysAsync(client *c, struct parsed_command &command);
 
     bool FSnapshot() const { return m_spdbSnapshotHOLDER != nullptr; }
 
     std::unique_ptr<const StorageCache> CloneStorageCache() { return std::unique_ptr<const StorageCache>(m_spstorage->clone()); }
-    void bulkStorageInsert(char **rgKeys, size_t *rgcbKeys, char **rgVals, size_t *rgcbVals, size_t celem);
+    void bulkDirectStorageInsert(char **rgKeys, size_t *rgcbKeys, char **rgVals, size_t *rgcbVals, size_t celem);
 
     dict_iter find_cached_threadsafe(const char *key) const;
 
@@ -1370,7 +1370,7 @@ struct redisDb : public redisDbPersistentDataSnapshot
     using redisDbPersistentData::FRehashing;
     using redisDbPersistentData::FTrackingChanges;
     using redisDbPersistentData::CloneStorageCache;
-    using redisDbPersistentData::bulkStorageInsert;
+    using redisDbPersistentData::bulkDirectStorageInsert;
 
 public:
     const redisDbPersistentDataSnapshot *createSnapshot(uint64_t mvccCheckpoint, bool fOptional) {
@@ -3894,13 +3894,13 @@ void incrementMvccTstamp();
 
 #if __GNUC__ >= 7 && !defined(NO_DEPRECATE_FREE)
  [[deprecated]]
-void *calloc(size_t count, size_t size);
+void *calloc(size_t count, size_t size) noexcept;
  [[deprecated]]
-void free(void *ptr);
+void free(void *ptr) noexcept;
  [[deprecated]]
-void *malloc(size_t size);
+void *malloc(size_t size) noexcept;
  [[deprecated]]
-void *realloc(void *ptr, size_t size);
+void *realloc(void *ptr, size_t size) noexcept;
 #endif
 
 /* Debugging stuff */

src/zmalloc.cpp

@@ -80,12 +80,12 @@ static_assert((PREFIX_SIZE % 16) == 0, "Our prefix must be modulo 16-bytes or ou
 #define realloc(ptr,size, type) tc_realloc(ptr,size)
 #define free(ptr) tc_free(ptr)
 #elif defined(USE_JEMALLOC)
-#define malloc(size, type) je_malloc(size)
-#define calloc(count,size,type) je_calloc(count,size)
-#define realloc(ptr,size,type) je_realloc(ptr,size)
-#define free(ptr) je_free(ptr)
-#define mallocx(size,flags) je_mallocx(size,flags)
-#define dallocx(ptr,flags) je_dallocx(ptr,flags)
+#define malloc(size, type) malloc(size)
+#define calloc(count,size,type) calloc(count,size)
+#define realloc(ptr,size,type) realloc(ptr,size)
+#define free(ptr) free(ptr)
+#define mallocx(size,flags) mallocx(size,flags)
+#define dallocx(ptr,flags) dallocx(ptr,flags)
 #else
 #define malloc(size, type) malloc(size)
 #define calloc(count,size,type) calloc(count,size)
@@ -494,17 +494,17 @@ int zmalloc_get_allocator_info(size_t *allocated,
     *allocated = *resident = *active = 0;
     /* Update the statistics cached by mallctl. */
     sz = sizeof(epoch);
-    je_mallctl("epoch", &epoch, &sz, &epoch, sz);
+    mallctl("epoch", &epoch, &sz, &epoch, sz);
     sz = sizeof(size_t);
     /* Unlike RSS, this does not include RSS from shared libraries and other non
      * heap mappings. */
-    je_mallctl("stats.resident", resident, &sz, NULL, 0);
+    mallctl("stats.resident", resident, &sz, NULL, 0);
     /* Unlike resident, this doesn't not include the pages jemalloc reserves
      * for re-use (purge will clean that). */
-    je_mallctl("stats.active", active, &sz, NULL, 0);
+    mallctl("stats.active", active, &sz, NULL, 0);
     /* Unlike zmalloc_used_memory, this matches the stats.resident by taking
      * into account all allocations done by this process (not only zmalloc). */
-    je_mallctl("stats.allocated", allocated, &sz, NULL, 0);
+    mallctl("stats.allocated", allocated, &sz, NULL, 0);
     return 1;
 }
 
@@ -512,7 +512,7 @@ void set_jemalloc_bg_thread(int enable) {
     /* let jemalloc do purging asynchronously, required when there's no traffic 
      * after flushdb */
     char val = !!enable;
-    je_mallctl("background_thread", NULL, 0, &val, 1);
+    mallctl("background_thread", NULL, 0, &val, 1);
 }
 
 int jemalloc_purge() {
@@ -520,9 +520,9 @@ int jemalloc_purge() {
     char tmp[32];
     unsigned narenas = 0;
     size_t sz = sizeof(unsigned);
-    if (!je_mallctl("arenas.narenas", &narenas, &sz, NULL, 0)) {
+    if (!mallctl("arenas.narenas", &narenas, &sz, NULL, 0)) {
         snprintf(tmp, sizeof(tmp), "arena.%d.purge", narenas);
-        if (!je_mallctl(tmp, NULL, 0, NULL, 0))
+        if (!mallctl(tmp, NULL, 0, NULL, 0))
             return 0;
     }
     return -1;

src/zmalloc.h

@@ -55,7 +55,7 @@
 #include <jemalloc/jemalloc.h>
 #if (JEMALLOC_VERSION_MAJOR == 2 && JEMALLOC_VERSION_MINOR >= 1) || (JEMALLOC_VERSION_MAJOR > 2)
 #define HAVE_MALLOC_SIZE 1
-#define zmalloc_size(p) je_malloc_usable_size(p)
+#define zmalloc_size(p) malloc_usable_size(p)
 #else
 #error "Newer version of jemalloc required"
 #endif

tests/unit/introspection.tcl

@@ -162,15 +162,16 @@ start_server {tags {"introspection"}} {
             aof_rewrite_cpulist
             time-thread-priority
             bgsave_cpulist
-	        storage-cache-mode
-	        storage-provider-options
-	        use-fork
+	    storage-cache-mode
+	    storage-provider-options
+	    use-fork
             multi-master
             active-replica
             bind
             set-proc-title
             repl-backlog-disk-reserve
 	    tls-allowlist
+            db-s3-object
         }
 
         if {!$::tls} {

tests/unit/scripting.tcl

@@ -43,10 +43,15 @@ start_server {tags {"scripting"}} {
         r eval {return redis.call('get',KEYS[1])} 1 mykey
     } {myval}
 
-    test { EVAL - keys command works? } {
+    test {EVAL - keys command works? } {
         r eval {return redis.call('keys', 'test')} 0
     }
 
+    test {EVAL - KeyDB global works } {
+        r eval {return keydb.call('get', KEYS[1])} 1 mykey
+        assert_equal [r eval {return redis.call('get',KEYS[1])} 1 mykey] [r eval {return keydb.call('get', KEYS[1])} 1 mykey]
+    }
+
     test {EVALSHA - Can we call a SHA1 if already defined?} {
         r evalsha fd758d1589d044dd850a6f05d52f2eefd27f033f 1 mykey
     } {myval}

utils/create-cluster/create-cluster

@@ -26,7 +26,11 @@ then
     while [ $((PORT < ENDPORT)) != "0" ]; do
         PORT=$((PORT+1))
         echo "Starting $PORT"
-        $BIN_PATH/keydb-server --port $PORT  --protected-mode $PROTECTED_MODE --cluster-enabled yes --cluster-config-file nodes-${PORT}.conf --cluster-node-timeout $TIMEOUT --appendonly yes --appendfilename appendonly-${PORT}.aof --dbfilename dump-${PORT}.rdb --logfile ${PORT}.log --daemonize yes ${ADDITIONAL_OPTIONS}
+        if [ "$2" == "flash" ]
+        then
+            ADDITIONAL_OPTIONS="--save \"\" \"\" \"\" --semi-ordered-set-bucket-size 8 --client-output-buffer-limit replica 1 1 0 --maxmemory 100000000 --storage-provider flash ./$PORT.flash"
+        fi
+        $BIN_PATH/keydb-server --server-threads 4 --port $PORT  --protected-mode $PROTECTED_MODE --cluster-enabled yes --cluster-config-file nodes-${PORT}.conf --cluster-node-timeout $TIMEOUT --dbfilename dump-${PORT}.rdb --logfile ${PORT}.log --daemonize yes ${ADDITIONAL_OPTIONS}
     done
     exit 0
 fi
@@ -97,6 +101,8 @@ then
     rm -rf appendonly*.aof
     rm -rf dump*.rdb
     rm -rf nodes*.conf
+    rm -rf *.flash
+    rm -rf temp*.rdb
     exit 0
 fi