add comment back

* add expire_lock to previously async access to expireset

* add flash option to create-cluster

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()));
     }
 }
@@ -3253,6 +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())
@@ -3377,6 +3387,7 @@ void redisDbPersistentData::prefetchKeysAsync(client *c, parsed_command &command
                     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);

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/object.cpp

@@ -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 */
@@ -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/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

@@ -5172,7 +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. */

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