Remove performance zapping test code

.vscode/settings.json

@@ -0,0 +1,58 @@
+{
+    "files.associations": {
+        "zmalloc.h": "c",
+        "stat.h": "c",
+        "array": "cpp",
+        "atomic": "cpp",
+        "*.tcc": "cpp",
+        "cctype": "cpp",
+        "chrono": "cpp",
+        "clocale": "cpp",
+        "cmath": "cpp",
+        "condition_variable": "cpp",
+        "cstdarg": "cpp",
+        "cstddef": "cpp",
+        "cstdint": "cpp",
+        "cstdio": "cpp",
+        "cstdlib": "cpp",
+        "cstring": "cpp",
+        "ctime": "cpp",
+        "cwchar": "cpp",
+        "cwctype": "cpp",
+        "deque": "cpp",
+        "list": "cpp",
+        "unordered_map": "cpp",
+        "vector": "cpp",
+        "exception": "cpp",
+        "fstream": "cpp",
+        "functional": "cpp",
+        "future": "cpp",
+        "initializer_list": "cpp",
+        "iomanip": "cpp",
+        "iosfwd": "cpp",
+        "iostream": "cpp",
+        "istream": "cpp",
+        "limits": "cpp",
+        "memory": "cpp",
+        "mutex": "cpp",
+        "new": "cpp",
+        "numeric": "cpp",
+        "optional": "cpp",
+        "ostream": "cpp",
+        "ratio": "cpp",
+        "scoped_allocator": "cpp",
+        "sstream": "cpp",
+        "stdexcept": "cpp",
+        "streambuf": "cpp",
+        "string_view": "cpp",
+        "system_error": "cpp",
+        "thread": "cpp",
+        "cinttypes": "cpp",
+        "tuple": "cpp",
+        "type_traits": "cpp",
+        "typeinfo": "cpp",
+        "utility": "cpp",
+        "set": "cpp",
+        "algorithm": "cpp"
+    }
+}

src/IStorage.h

@@ -0,0 +1,14 @@
+#pragma once
+#include <functional>
+
+class IStorage
+{
+public:
+    typedef std::function<void(const char *, size_t, const void *, size_t)> callback;
+
+    virtual void insert(const char *key, size_t cchKey, void *data, size_t cb) = 0;
+    virtual void erase(const char *key, size_t cchKey) = 0;
+    virtual void retrieve(const char *key, size_t cchKey, bool fDelete, callback fn) = 0;
+    virtual size_t clear() = 0;
+    virtual void enumerate(callback fn) = 0;
+};

src/aof.cpp

@@ -264,7 +264,7 @@ int startAppendOnly(void) {
             strerror(errno));
         return C_ERR;
     }
-    if (g_pserver->rdb_child_pid != -1) {
+    if (g_pserver->FRdbSaveInProgress()) {
         g_pserver->aof_rewrite_scheduled = 1;
         serverLog(LL_WARNING,"AOF was enabled but there is already a child process saving an RDB file on disk. An AOF background was scheduled to start when possible.");
     } else {
@@ -397,7 +397,7 @@ void flushAppendOnlyFile(int force) {
      * useful for graphing / monitoring purposes. */
     if (sync_in_progress) {
         latencyAddSampleIfNeeded("aof-write-pending-fsync",latency);
-    } else if (g_pserver->aof_child_pid != -1 || g_pserver->rdb_child_pid != -1) {
+    } else if (g_pserver->aof_child_pid != -1 || g_pserver->FRdbSaveInProgress()) {
         latencyAddSampleIfNeeded("aof-write-active-child",latency);
     } else {
         latencyAddSampleIfNeeded("aof-write-alone",latency);
@@ -494,7 +494,7 @@ try_fsync:
     /* Don't fsync if no-appendfsync-on-rewrite is set to yes and there are
      * children doing I/O in the background. */
     if (g_pserver->aof_no_fsync_on_rewrite &&
-        (g_pserver->aof_child_pid != -1 || g_pserver->rdb_child_pid != -1))
+        (g_pserver->aof_child_pid != -1 || g_pserver->FRdbSaveInProgress()))
             return;
 
     /* Perform the fsync if needed. */
@@ -1300,52 +1300,46 @@ ssize_t aofReadDiffFromParent(void) {
 
 int rewriteAppendOnlyFileRio(rio *aof) {
     dictIterator *di = NULL;
-    dictEntry *de;
     size_t processed = 0;
     int j;
 
     for (j = 0; j < cserver.dbnum; j++) {
         char selectcmd[] = "*2\r\n$6\r\nSELECT\r\n";
         redisDb *db = g_pserver->db+j;
-        dict *d = db->pdict;
-        if (dictSize(d) == 0) continue;
-        di = dictGetSafeIterator(d);
+        if (db->size() == 0) continue;
 
         /* SELECT the new DB */
         if (rioWrite(aof,selectcmd,sizeof(selectcmd)-1) == 0) goto werr;
         if (rioWriteBulkLongLong(aof,j) == 0) goto werr;
 
         /* Iterate this DB writing every entry */
-        while((de = dictNext(di)) != NULL) {
-            sds keystr;
-            robj key, *o;
+        bool fComplete = db->iterate([&](const char *keystr, robj *o)->bool{
+            robj key;
 
-            keystr = (sds)dictGetKey(de);
-            o = (robj*)dictGetVal(de);
-            initStaticStringObject(key,keystr);
+            initStaticStringObject(key,(sds)keystr);
 
-            expireEntry *pexpire = getExpire(db,&key);
+            expireEntry *pexpire = db->getExpire(&key);
 
             /* Save the key and associated value */
             if (o->type == OBJ_STRING) {
                 /* Emit a SET command */
                 char cmd[]="*3\r\n$3\r\nSET\r\n";
-                if (rioWrite(aof,cmd,sizeof(cmd)-1) == 0) goto werr;
+                if (rioWrite(aof,cmd,sizeof(cmd)-1) == 0) return false;
                 /* Key and value */
-                if (rioWriteBulkObject(aof,&key) == 0) goto werr;
-                if (rioWriteBulkObject(aof,o) == 0) goto werr;
+                if (rioWriteBulkObject(aof,&key) == 0) return false;
+                if (rioWriteBulkObject(aof,o) == 0) return false;
             } else if (o->type == OBJ_LIST) {
-                if (rewriteListObject(aof,&key,o) == 0) goto werr;
+                if (rewriteListObject(aof,&key,o) == 0) return false;
             } else if (o->type == OBJ_SET) {
-                if (rewriteSetObject(aof,&key,o) == 0) goto werr;
+                if (rewriteSetObject(aof,&key,o) == 0) return false;
             } else if (o->type == OBJ_ZSET) {
-                if (rewriteSortedSetObject(aof,&key,o) == 0) goto werr;
+                if (rewriteSortedSetObject(aof,&key,o) == 0) return false;
             } else if (o->type == OBJ_HASH) {
-                if (rewriteHashObject(aof,&key,o) == 0) goto werr;
+                if (rewriteHashObject(aof,&key,o) == 0) return false;
             } else if (o->type == OBJ_STREAM) {
-                if (rewriteStreamObject(aof,&key,o) == 0) goto werr;
+                if (rewriteStreamObject(aof,&key,o) == 0) return false;
             } else if (o->type == OBJ_MODULE) {
-                if (rewriteModuleObject(aof,&key,o) == 0) goto werr;
+                if (rewriteModuleObject(aof,&key,o) == 0) return false;
             } else {
                 serverPanic("Unknown object type");
             }
@@ -1355,17 +1349,17 @@ int rewriteAppendOnlyFileRio(rio *aof) {
                     if (subExpire.subkey() == nullptr)
                     {
                         char cmd[]="*3\r\n$9\r\nPEXPIREAT\r\n";
-                        if (rioWrite(aof,cmd,sizeof(cmd)-1) == 0) goto werr;
-                        if (rioWriteBulkObject(aof,&key) == 0) goto werr;
+                        if (rioWrite(aof,cmd,sizeof(cmd)-1) == 0) return false;
+                        if (rioWriteBulkObject(aof,&key) == 0) return false;
                     }
                     else
                     {
                         char cmd[]="*4\r\n$12\r\nEXPIREMEMBER\r\n";
-                        if (rioWrite(aof,cmd,sizeof(cmd)-1) == 0) goto werr;
-                        if (rioWriteBulkObject(aof,&key) == 0) goto werr;
-                        if (rioWrite(aof,subExpire.subkey(),sdslen(subExpire.subkey())) == 0) goto werr;
+                        if (rioWrite(aof,cmd,sizeof(cmd)-1) == 0) return false;
+                        if (rioWriteBulkObject(aof,&key) == 0) return false;
+                        if (rioWrite(aof,subExpire.subkey(),sdslen(subExpire.subkey())) == 0) return false;
                     }
-                    if (rioWriteBulkLongLong(aof,subExpire.when()) == 0) goto werr; // common
+                    if (rioWriteBulkLongLong(aof,subExpire.when()) == 0) return false; // common
                 }
             }
             /* Read some diff from the parent process from time to time. */
@@ -1373,9 +1367,10 @@ int rewriteAppendOnlyFileRio(rio *aof) {
                 processed = aof->processed_bytes;
                 aofReadDiffFromParent();
             }
-        }
-        dictReleaseIterator(di);
-        di = NULL;
+            return true;
+        });
+        if (!fComplete)
+            goto werr;
     }
     return C_OK;
 
@@ -1416,7 +1411,12 @@ int rewriteAppendOnlyFile(char *filename) {
 
     if (g_pserver->aof_use_rdb_preamble) {
         int error;
-        if (rdbSaveRio(&aof,&error,RDB_SAVE_AOF_PREAMBLE,NULL) == C_ERR) {
+        std::vector<redisDbPersistentData*> vecpdb;
+        for (int idb = 0; idb < cserver.dbnum; ++idb)
+        {
+            vecpdb.push_back(&g_pserver->db[idb]);
+        }
+        if (rdbSaveRio(&aof,vecpdb.data(),&error,RDB_SAVE_AOF_PREAMBLE,NULL) == C_ERR) {
             errno = error;
             goto werr;
         }
@@ -1593,7 +1593,7 @@ int rewriteAppendOnlyFileBackground(void) {
     pid_t childpid;
     long long start;
 
-    if (g_pserver->aof_child_pid != -1 || g_pserver->rdb_child_pid != -1) return C_ERR;
+    if (g_pserver->aof_child_pid != -1 || g_pserver->FRdbSaveInProgress()) return C_ERR;
     if (aofCreatePipes() != C_OK) return C_ERR;
     openChildInfoPipe();
     start = ustime();
@@ -1652,7 +1652,7 @@ int rewriteAppendOnlyFileBackground(void) {
 void bgrewriteaofCommand(client *c) {
     if (g_pserver->aof_child_pid != -1) {
         addReplyError(c,"Background append only file rewriting already in progress");
-    } else if (g_pserver->rdb_child_pid != -1) {
+    } else if (g_pserver->FRdbSaveInProgress()) {
         g_pserver->aof_rewrite_scheduled = 1;
         addReplyStatus(c,"Background append only file rewriting scheduled");
     } else if (rewriteAppendOnlyFileBackground() == C_OK) {

src/cluster.cpp

@@ -3910,7 +3910,7 @@ int verifyClusterConfigWithData(void) {
 
     /* Make sure we only have keys in DB0. */
     for (j = 1; j < cserver.dbnum; j++) {
-        if (dictSize(g_pserver->db[j].pdict)) return C_ERR;
+        if (g_pserver->db[j].size()) return C_ERR;
     }
 
     /* Check that all the slots we see populated memory have a corresponding
@@ -4286,7 +4286,7 @@ NULL
         clusterReplyMultiBulkSlots(c);
     } else if (!strcasecmp(szFromObj(c->argv[1]),"flushslots") && c->argc == 2) {
         /* CLUSTER FLUSHSLOTS */
-        if (dictSize(g_pserver->db[0].pdict) != 0) {
+        if (g_pserver->db[0].size() != 0) {
             addReplyError(c,"DB must be empty to perform CLUSTER FLUSHSLOTS.");
             return;
         }
@@ -4621,7 +4621,7 @@ NULL
          * slots nor keys to accept to replicate some other node.
          * Slaves can switch to another master without issues. */
         if (nodeIsMaster(myself) &&
-            (myself->numslots != 0 || dictSize(g_pserver->db[0].pdict) != 0)) {
+            (myself->numslots != 0 || g_pserver->db[0].size() != 0)) {
             addReplyError(c,
                 "To set a master the node must be empty and "
                 "without assigned slots.");
@@ -4778,7 +4778,7 @@ NULL
 
         /* Slaves can be reset while containing data, but not master nodes
          * that must be empty. */
-        if (nodeIsMaster(myself) && dictSize(c->db->pdict) != 0) {
+        if (nodeIsMaster(myself) && c->db->size() != 0) {
             addReplyError(c,"CLUSTER RESET can't be called with "
                             "master nodes containing keys");
             return;
@@ -4794,6 +4794,7 @@ NULL
 /* -----------------------------------------------------------------------------
  * DUMP, RESTORE and MIGRATE commands
  * -------------------------------------------------------------------------- */
+ssize_t rdbSaveAuxFieldStrStr(rio *rdb, const char *key, const char *val);
 
 /* Generates a DUMP-format representation of the object 'o', adding it to the
  * io stream pointed by 'rio'. This function can't fail. */
@@ -4806,6 +4807,9 @@ void createDumpPayload(rio *payload, robj_roptr o, robj *key) {
     rioInitWithBuffer(payload,sdsempty());
     serverAssert(rdbSaveObjectType(payload,o));
     serverAssert(rdbSaveObject(payload,o,key));
+    char szT[32];
+    snprintf(szT, 32, "%" PRIu64, o->mvcc_tstamp);
+    serverAssert(rdbSaveAuxFieldStrStr(payload,"mvcc-tstamp", szT) != -1);
 
     /* Write the footer, this is how it looks like:
      * ----------------+---------------------+---------------+
@@ -4941,6 +4945,21 @@ void restoreCommand(client *c) {
         addReplyError(c,"Bad data format");
         return;
     }
+    if (rdbLoadType(&payload) == RDB_OPCODE_AUX)
+    {
+        robj *auxkey, *auxval;
+        if ((auxkey = rdbLoadStringObject(&payload)) == NULL) goto eoferr;
+        if ((auxval = rdbLoadStringObject(&payload)) == NULL) {
+            decrRefCount(auxkey);
+            goto eoferr;
+        }
+        if (strcasecmp(szFromObj(auxkey), "mvcc-tstamp") == 0) {
+            obj->mvcc_tstamp = strtoull(szFromObj(auxval), nullptr, 10);
+        }
+        decrRefCount(auxkey);
+        decrRefCount(auxval);
+    }
+eoferr:
 
     /* Remove the old key if needed. */
     if (replace) dbDelete(c->db,c->argv[1]);
@@ -5194,7 +5213,7 @@ try_again:
     /* Create RESTORE payload and generate the protocol to call the command. */
     for (j = 0; j < num_keys; j++) {
         long long ttl = 0;
-        expireEntry *pexpire = getExpire(c->db,kv[j]);
+        expireEntry *pexpire = c->db->getExpire(kv[j]);
         long long expireat = -1;
         if (pexpire != nullptr)
             pexpire->FGetPrimaryExpire(&expireat);

src/compactvector.h

@@ -30,7 +30,16 @@ public:
         zfree(m_data);
     }
 
-    compactvector(compactvector &) noexcept = delete;
+    compactvector(const compactvector &src)
+    {
+        m_celem = src.m_celem;
+        m_max = src.m_max;
+        m_data = (T*)zmalloc(sizeof(T) * m_max, MALLOC_LOCAL);
+        for (size_t ielem = 0; ielem < m_celem; ++ielem)
+        {
+            new (m_data+ielem) T(src[ielem]);
+        }
+    }
 
     compactvector(compactvector &&src) noexcept
     {
@@ -42,7 +51,6 @@ public:
         src.m_max = 0;
     }
 
-    compactvector &operator=(const compactvector&) noexcept = delete;
     compactvector &operator=(compactvector &&src) noexcept
     {
         zfree(m_data);

src/db.cpp

@@ -41,7 +41,6 @@
 
 int keyIsExpired(redisDb *db, robj *key);
 int expireIfNeeded(redisDb *db, robj *key, robj *o);
-void dbOverwriteCore(redisDb *db, dictEntry *de, robj *key, robj *val, bool fUpdateMvcc, bool fRemoveExpire);
 
 /* Update LFU when an object is accessed.
  * Firstly, decrement the counter if the decrement time is reached.
@@ -57,8 +56,7 @@ void updateExpire(redisDb *db, sds key, robj *valOld, robj *valNew)
     serverAssert(valOld->FExpires());
     serverAssert(!valNew->FExpires());
     
-    auto itr = db->setexpire->find(key);
-    serverAssert(itr != db->setexpire->end());
+    serverAssert(db->FKeyExpires((const char*)key));
     
     valNew->SetFExpires(true);
     valOld->SetFExpires(false);
@@ -70,14 +68,13 @@ void updateExpire(redisDb *db, sds key, robj *valOld, robj *valNew)
  * implementations that should instead rely on lookupKeyRead(),
  * lookupKeyWrite() and lookupKeyReadWithFlags(). */
 static robj *lookupKey(redisDb *db, robj *key, int flags) {
-    dictEntry *de = dictFind(db->pdict,ptrFromObj(key));
-    if (de) {
-        robj *val = (robj*)dictGetVal(de);
-
+    auto itr = db->find(key);
+    if (itr) {
+        robj *val = itr.val();
         /* Update the access time for the ageing algorithm.
          * Don't do it if we have a saving child, as this will trigger
          * a copy on write madness. */
-        if (g_pserver->rdb_child_pid == -1 &&
+        if (!g_pserver->FRdbSaveInProgress() &&
             g_pserver->aof_child_pid == -1 &&
             !(flags & LOOKUP_NOTOUCH))
         {
@@ -90,6 +87,7 @@ static robj *lookupKey(redisDb *db, robj *key, int flags) {
 
         if (flags & LOOKUP_UPDATEMVCC) {
             val->mvcc_tstamp = getMvccTstamp();
+            db->trackkey(key);
         }
         return val;
     } else {
@@ -195,13 +193,14 @@ robj *lookupKeyWriteOrReply(client *c, robj *key, robj *reply) {
     return o;
 }
 
-int dbAddCore(redisDb *db, robj *key, robj *val) {
+bool dbAddCore(redisDb *db, robj *key, robj *val) {
     serverAssert(!val->FExpires());
     sds copy = sdsdup(szFromObj(key));
-    int retval = dictAdd(db->pdict, copy, val);
-    val->mvcc_tstamp = key->mvcc_tstamp = getMvccTstamp();
+    bool fInserted = db->insert(copy, val);
+    if (g_pserver->fActiveReplica)
+        val->mvcc_tstamp = key->mvcc_tstamp = getMvccTstamp();
 
-    if (retval == DICT_OK)
+    if (fInserted)
     {
         if (val->type == OBJ_LIST ||
             val->type == OBJ_ZSET)
@@ -213,7 +212,7 @@ int dbAddCore(redisDb *db, robj *key, robj *val) {
         sdsfree(copy);
     }
 
-    return retval;
+    return fInserted;
 }
 
 /* Add the key to the DB. It's up to the caller to increment the reference
@@ -222,23 +221,22 @@ int dbAddCore(redisDb *db, robj *key, robj *val) {
  * The program is aborted if the key already exists. */
 void dbAdd(redisDb *db, robj *key, robj *val)
 {
-    int retval = dbAddCore(db, key, val);
-    serverAssertWithInfo(NULL,key,retval == DICT_OK);
+    bool fInserted = dbAddCore(db, key, val);
+    serverAssertWithInfo(NULL,key,fInserted);
 }
 
-void dbOverwriteCore(redisDb *db, dictEntry *de, robj *key, robj *val, bool fUpdateMvcc, bool fRemoveExpire)
+void redisDb::dbOverwriteCore(redisDb::iter itr, robj *key, robj *val, bool fUpdateMvcc, bool fRemoveExpire)
 {
-    dictEntry auxentry = *de;
-    robj *old = (robj*)dictGetVal(de);
+    robj *old = itr.val();
 
     if (old->FExpires()) {
         if (fRemoveExpire) {
-            removeExpire(db, key);
+            ::removeExpire(this, key);
         }
         else {
             if (val->getrefcount(std::memory_order_relaxed) == OBJ_SHARED_REFCOUNT)
                 val = dupStringObject(val);
-            updateExpire(db, (sds)dictGetKey(de), old, val);
+            ::updateExpire(this, itr.key(), old, val);
         }
     }
 
@@ -251,14 +249,12 @@ void dbOverwriteCore(redisDb *db, dictEntry *de, robj *key, robj *val, bool fUpd
         val->mvcc_tstamp = getMvccTstamp();
     }
 
-    dictSetVal(db->pdict, de, val);
+    if (g_pserver->lazyfree_lazy_server_del)
+        freeObjAsync(itr.val());
+    else
+        decrRefCount(itr.val());
 
-    if (g_pserver->lazyfree_lazy_server_del) {
-        freeObjAsync(old);
-        dictSetVal(db->pdict, &auxentry, NULL);
-    }
-
-    dictFreeVal(db->pdict, &auxentry);
+    updateValue(itr, val);
 }
 
 /* Overwrite an existing key with a new value. Incrementing the reference
@@ -267,10 +263,10 @@ void dbOverwriteCore(redisDb *db, dictEntry *de, robj *key, robj *val, bool fUpd
  *
  * The program is aborted if the key was not already present. */
 void dbOverwrite(redisDb *db, robj *key, robj *val) {
-    dictEntry *de = dictFind(db->pdict,ptrFromObj(key));
+    auto itr = db->find(key);
 
-    serverAssertWithInfo(NULL,key,de != NULL);
-    dbOverwriteCore(db, de, key, val, !!g_pserver->fActiveReplica, false);
+    serverAssertWithInfo(NULL,key,itr != nullptr);
+    db->dbOverwriteCore(itr, key, val, !!g_pserver->fActiveReplica, false);
 }
 
 /* Insert a key, handling duplicate keys according to fReplace */
@@ -278,14 +274,14 @@ int dbMerge(redisDb *db, robj *key, robj *val, int fReplace)
 {
     if (fReplace)
     {
-        dictEntry *de = dictFind(db->pdict, ptrFromObj(key));
-        if (de == nullptr)
-            return (dbAddCore(db, key, val) == DICT_OK);
+        auto itr = db->find(key);
+        if (itr == nullptr)
+            return (dbAddCore(db, key, val) == true);
 
-        robj *old = (robj*)dictGetVal(de);
+        robj *old = itr.val();
         if (old->mvcc_tstamp <= val->mvcc_tstamp)
         {
-            dbOverwriteCore(db, de, key, val, false, true);
+            db->dbOverwriteCore(itr, key, val, false, true);
             return true;
         }
         
@@ -293,7 +289,7 @@ int dbMerge(redisDb *db, robj *key, robj *val, int fReplace)
     }
     else
     {
-        return (dbAddCore(db, key, val) == DICT_OK);
+        return (dbAddCore(db, key, val) == true);
     }
 }
 
@@ -306,18 +302,18 @@ int dbMerge(redisDb *db, robj *key, robj *val, int fReplace)
  *
  * All the new keys in the database should be created via this interface. */
 void setKey(redisDb *db, robj *key, robj *val) {
-    dictEntry *de = dictFind(db->pdict, ptrFromObj(key));
-    if (de == NULL) {
+    auto itr = db->find(key);
+    if (itr == NULL) {
         dbAdd(db,key,val);
     } else {
-        dbOverwriteCore(db,de,key,val,!!g_pserver->fActiveReplica,true);
+        db->dbOverwriteCore(itr,key,val,!!g_pserver->fActiveReplica,true);
     }
     incrRefCount(val);
     signalModifiedKey(db,key);
 }
 
 int dbExists(redisDb *db, robj *key) {
-    return dictFind(db->pdict,ptrFromObj(key)) != NULL;
+    return (db->find(key) != nullptr);
 }
 
 /* Return a random key, in form of a Redis object.
@@ -325,21 +321,20 @@ int dbExists(redisDb *db, robj *key) {
  *
  * The function makes sure to return keys not already expired. */
 robj *dbRandomKey(redisDb *db) {
-    dictEntry *de;
     int maxtries = 100;
-    int allvolatile = dictSize(db->pdict) == db->setexpire->size();
+    bool allvolatile = db->expireSize() == db->size();
 
     while(1) {
         sds key;
         robj *keyobj;
 
-        de = dictGetRandomKey(db->pdict);
-        if (de == NULL) return NULL;
+        auto itr = db->random();
+        if (itr == nullptr) return NULL;
 
-        key = (sds)dictGetKey(de);
+        key = itr.key();
         keyobj = createStringObject(key,sdslen(key));
 
-        if (((robj*)dictGetVal(de))->FExpires())
+        if (itr.val()->FExpires())
         {
             if (allvolatile && listLength(g_pserver->masters) && --maxtries == 0) {
                 /* If the DB is composed only of keys with an expire set,
@@ -354,7 +349,7 @@ robj *dbRandomKey(redisDb *db) {
             }
         }
             
-        if (((robj*)dictGetVal(de))->FExpires())
+        if (itr.val()->FExpires())
         {
              if (expireIfNeeded(db,keyobj)) {
                 decrRefCount(keyobj);
@@ -366,15 +361,16 @@ robj *dbRandomKey(redisDb *db) {
     }
 }
 
-/* Delete a key, value, and associated expiration entry if any, from the DB */
-int dbSyncDelete(redisDb *db, robj *key) {
+bool redisDbPersistentData::syncDelete(robj *key)
+{
     /* Deleting an entry from the expires dict will not free the sds of
      * the key, because it is shared with the main dictionary. */
 
-    dictEntry *de = dictFind(db->pdict, szFromObj(key));
-    if (de != nullptr && ((robj*)dictGetVal(de))->FExpires())
-        removeExpireCore(db, key, de);
-    if (dictDelete(db->pdict,ptrFromObj(key)) == DICT_OK) {
+    auto itr = find(szFromObj(key));
+    trackkey(szFromObj(key));
+    if (itr != nullptr && itr.val()->FExpires())
+        removeExpire(key, itr);
+    if (dictDelete(m_pdict,ptrFromObj(key)) == DICT_OK) {
         if (g_pserver->cluster_enabled) slotToKeyDel(key);
         return 1;
     } else {
@@ -382,6 +378,11 @@ int dbSyncDelete(redisDb *db, robj *key) {
     }
 }
 
+/* Delete a key, value, and associated expiration entry if any, from the DB */
+int dbSyncDelete(redisDb *db, robj *key) {
+    return db->syncDelete(key);
+}
+
 /* This is a wrapper whose behavior depends on the Redis lazy free
  * configuration. Deletes the key synchronously or asynchronously. */
 int dbDelete(redisDb *db, robj *key) {
@@ -427,7 +428,7 @@ robj *dbUnshareStringValue(redisDb *db, robj *key, robj *o) {
     return o;
 }
 
-/* Remove all keys from all the databases in a Redis g_pserver->
+/* Remove all keys from all the databases in a Redis DB
  * If callback is given the function is called from time to time to
  * signal that work is in progress.
  *
@@ -459,15 +460,7 @@ long long emptyDb(int dbnum, int flags, void(callback)(void*)) {
     }
 
     for (int j = startdb; j <= enddb; j++) {
-        removed += dictSize(g_pserver->db[j].pdict);
-        if (async) {
-            emptyDbAsync(&g_pserver->db[j]);
-        } else {
-            dictEmpty(g_pserver->db[j].pdict,callback);
-            delete g_pserver->db[j].setexpire;
-            g_pserver->db[j].setexpire = new (MALLOC_LOCAL) expireset();
-            g_pserver->db[j].expireitr = g_pserver->db[j].setexpire->end();
-        }
+        removed += g_pserver->db[j].clear(!!async, callback); 
     }
     if (g_pserver->cluster_enabled) {
         if (async) {
@@ -553,14 +546,14 @@ void flushallCommand(client *c) {
     signalFlushedDb(-1);
     g_pserver->dirty += emptyDb(-1,flags,NULL);
     addReply(c,shared.ok);
-    if (g_pserver->rdb_child_pid != -1) killRDBChild();
+    if (g_pserver->FRdbSaveInProgress()) killRDBChild();
     if (g_pserver->saveparamslen > 0) {
         /* Normally rdbSave() will reset dirty, but we don't want this here
          * as otherwise FLUSHALL will not be replicated nor put into the AOF. */
         int saved_dirty = g_pserver->dirty;
         rdbSaveInfo rsi, *rsiptr;
         rsiptr = rdbPopulateSaveInfo(&rsi);
-        rdbSave(rsiptr);
+        rdbSave(nullptr, rsiptr);
         g_pserver->dirty = saved_dirty;
     }
     g_pserver->dirty++;
@@ -635,22 +628,35 @@ void randomkeyCommand(client *c) {
     decrRefCount(key);
 }
 
+
+bool redisDbPersistentData::iterate(std::function<bool(const char*, robj*)> fn)
+{
+    dictIterator *di = dictGetSafeIterator(m_pdict);
+    dictEntry *de = nullptr;
+    bool fResult = true;
+    while((de = dictNext(di)) != nullptr)
+    {
+        ensure(de);
+        if (!fn((const char*)dictGetKey(de), (robj*)dictGetVal(de)))
+        {
+            fResult = false;
+            break;
+        }
+    }
+    dictReleaseIterator(di);
+    return fResult;
+}
+
 void keysCommand(client *c) {
-    dictIterator *di;
-    dictEntry *de;
     sds pattern = szFromObj(c->argv[1]);
     int plen = sdslen(pattern), allkeys;
     unsigned long numkeys = 0;
     void *replylen = addReplyDeferredLen(c);
 
-#ifdef MULTITHREADED_KEYS
     aeReleaseLock();
-#endif
 
-    di = dictGetSafeIterator(c->db->pdict);
     allkeys = (pattern[0] == '*' && pattern[1] == '\0');
-    while((de = dictNext(di)) != NULL) {
-        sds key = (sds)dictGetKey(de);
+    c->db->iterate([&](const char *key, robj *)->bool {
         robj *keyobj;
 
         if (allkeys || stringmatchlen(pattern,plen,key,sdslen(key),0)) {
@@ -661,17 +667,15 @@ void keysCommand(client *c) {
             }
             decrRefCount(keyobj);
         }
-    }
-    dictReleaseIterator(di);
+        return true;
+    });
     setDeferredArrayLen(c,replylen,numkeys);
     
-#ifdef MULTITHREADED_KEYS
     fastlock_unlock(&c->db->lock);  // we must release the DB lock before acquiring the AE lock to prevent deadlocks
     AeLocker lock;
     lock.arm(c);
     fastlock_lock(&c->db->lock);    // we still need the DB lock
     lock.release();
-#endif
 }
 
 /* This callback is used by scanGenericCommand in order to collect elements
@@ -799,7 +803,7 @@ void scanGenericCommand(client *c, robj_roptr o, unsigned long cursor) {
     /* Handle the case of a hash table. */
     ht = NULL;
     if (o == nullptr) {
-        ht = c->db->pdict;
+        ht = c->db->dictUnsafeKeyOnly();
     } else if (o->type == OBJ_SET && o->encoding == OBJ_ENCODING_HT) {
         ht = (dict*)ptrFromObj(o);
     } else if (o->type == OBJ_HASH && o->encoding == OBJ_ENCODING_HT) {
@@ -932,7 +936,7 @@ void scanCommand(client *c) {
 }
 
 void dbsizeCommand(client *c) {
-    addReplyLongLong(c,dictSize(c->db->pdict));
+    addReplyLongLong(c,c->db->size());
 }
 
 void lastsaveCommand(client *c) {
@@ -1015,7 +1019,7 @@ void renameGenericCommand(client *c, int nx) {
     std::unique_ptr<expireEntry> spexpire;
 
     {   // scope pexpireOld since it will be invalid soon
-    expireEntry *pexpireOld = getExpire(c->db,c->argv[1]);
+    expireEntry *pexpireOld = c->db->getExpire(c->argv[1]);
     if (pexpireOld != nullptr)
         spexpire = std::make_unique<expireEntry>(std::move(*pexpireOld));
     }
@@ -1093,7 +1097,7 @@ void moveCommand(client *c) {
 
     std::unique_ptr<expireEntry> spexpire;
     {   // scope pexpireOld
-    expireEntry *pexpireOld = getExpire(c->db,c->argv[1]);
+    expireEntry *pexpireOld = c->db->getExpire(c->argv[1]);
     if (pexpireOld != nullptr)
         spexpire = std::make_unique<expireEntry>(std::move(*pexpireOld));
     }
@@ -1152,17 +1156,14 @@ int dbSwapDatabases(int id1, int id2) {
     /* Swap hash tables. Note that we don't swap blocking_keys,
      * ready_keys and watched_keys, since we want clients to
      * remain in the same DB they were. */
-    db1->pdict = db2->pdict;
-    db1->setexpire = db2->setexpire;
-    db1->expireitr = db2->expireitr;
+    redisDbPersistentData::swap(db1, db2);
     db1->avg_ttl = db2->avg_ttl;
     db1->last_expire_set = db2->last_expire_set;
+    db1->expireitr = db2->expireitr;
 
-    db2->pdict = aux.pdict;
-    db2->setexpire = aux.setexpire;
-    db2->expireitr = aux.expireitr;
     db2->avg_ttl = aux.avg_ttl;
     db2->last_expire_set = aux.last_expire_set;
+    db2->expireitr = aux.expireitr;
 
     /* Now we need to handle clients blocked on lists: as an effect
      * of swapping the two DBs, a client that was waiting for list
@@ -1211,37 +1212,38 @@ void swapdbCommand(client *c) {
  * Expires API
  *----------------------------------------------------------------------------*/
 int removeExpire(redisDb *db, robj *key) {
-    dictEntry *de = dictFind(db->pdict,ptrFromObj(key));
-    return removeExpireCore(db, key, de);
+    auto itr = db->find(key);
+    return db->removeExpire(key, itr);
 }
-int removeExpireCore(redisDb *db, robj *key, dictEntry *de) {
+int redisDbPersistentData::removeExpire(robj *key, dict_iter itr) {
     /* An expire may only be removed if there is a corresponding entry in the
      * main dict. Otherwise, the key will never be freed. */
-    serverAssertWithInfo(NULL,key,de != NULL);
+    serverAssertWithInfo(NULL,key,itr != nullptr);
 
-    robj *val = (robj*)dictGetVal(de);
+    robj *val = itr.val();
     if (!val->FExpires())
         return 0;
 
-    auto itr = db->setexpire->find((sds)dictGetKey(de));
-    serverAssert(itr != db->setexpire->end());
-    serverAssert(itr->key() == (sds)dictGetKey(de));
-    db->setexpire->erase(itr);
+    trackkey(key);
+    auto itrExpire = m_setexpire->find(itr.key());
+    serverAssert(itrExpire != m_setexpire->end());
+    serverAssert(itrExpire->key() == itr.key());
+    m_setexpire->erase(itrExpire);
     val->SetFExpires(false);
     return 1;
 }
 
-int removeSubkeyExpire(redisDb *db, robj *key, robj *subkey) {
-    dictEntry *de = dictFind(db->pdict,ptrFromObj(key));
-    serverAssertWithInfo(NULL,key,de != NULL);
+int redisDbPersistentData::removeSubkeyExpire(robj *key, robj *subkey) {
+    auto de = find(szFromObj(key));
+    serverAssertWithInfo(NULL,key,de != nullptr);
     
-    robj *val = (robj*)dictGetVal(de);
+    robj *val = de.val();
     if (!val->FExpires())
         return 0;
     
-    auto itr = db->setexpire->find((sds)dictGetKey(de));
-    serverAssert(itr != db->setexpire->end());
-    serverAssert(itr->key() == (sds)dictGetKey(de));
+    auto itr = m_setexpire->find(de.key());
+    serverAssert(itr != m_setexpire->end());
+    serverAssert(itr->key() == de.key());
     if (!itr->FFat())
         return 0;
 
@@ -1259,7 +1261,7 @@ int removeSubkeyExpire(redisDb *db, robj *key, robj *subkey) {
     }
 
     if (itr->pfatentry()->size() == 0)
-        removeExpireCore(db, key, de);
+        this->removeExpire(key, de);
 
     return found;
 }
@@ -1269,49 +1271,23 @@ int removeSubkeyExpire(redisDb *db, robj *key, robj *subkey) {
  * to NULL. The 'when' parameter is the absolute unix time in milliseconds
  * after which the key will no longer be considered valid. */
 void setExpire(client *c, redisDb *db, robj *key, robj *subkey, long long when) {
-    dictEntry *kde;
-
     serverAssert(GlobalLocksAcquired());
 
-    /* Reuse the sds from the main dict in the expire dict */
-    kde = dictFind(db->pdict,ptrFromObj(key));
-    serverAssertWithInfo(NULL,key,kde != NULL);
-
-    if (((robj*)dictGetVal(kde))->getrefcount(std::memory_order_relaxed) == OBJ_SHARED_REFCOUNT)
-    {
-        // shared objects cannot have the expire bit set, create a real object
-        dictSetVal(db->pdict, kde, dupStringObject((robj*)dictGetVal(kde)));
-    }
-
     /* Update TTL stats (exponential moving average) */
     /*  Note: We never have to update this on expiry since we reduce it by the current elapsed time here */
     long long now = g_pserver->mstime;
     db->avg_ttl -= (now - db->last_expire_set); // reduce the TTL by the time that has elapsed
-    if (db->setexpire->empty())
+    if (db->expireSize() == 0)
         db->avg_ttl = 0;
     else
-        db->avg_ttl -= db->avg_ttl / db->setexpire->size(); // slide one entry out the window
+        db->avg_ttl -= db->avg_ttl / db->expireSize(); // slide one entry out the window
     if (db->avg_ttl < 0)
         db->avg_ttl = 0;    // TTLs are never negative
-    db->avg_ttl += (double)(when-now) / (db->setexpire->size()+1);    // add the new entry
+    db->avg_ttl += (double)(when-now) / (db->expireSize()+1);    // add the new entry
     db->last_expire_set = now;
 
     /* Update the expire set */
-    const char *szSubKey = (subkey != nullptr) ? szFromObj(subkey) : nullptr;
-    if (((robj*)dictGetVal(kde))->FExpires()) {
-        auto itr = db->setexpire->find((sds)dictGetKey(kde));
-        serverAssert(itr != db->setexpire->end());
-        expireEntry eNew(std::move(*itr));
-        eNew.update(szSubKey, when);
-        db->setexpire->erase(itr);
-        db->setexpire->insert(eNew);
-    }
-    else
-    {
-        expireEntry e((sds)dictGetKey(kde), szSubKey, when);
-        ((robj*)dictGetVal(kde))->SetFExpires(true);
-        db->setexpire->insert(e);
-    }
+    db->setExpire(key, subkey, when);
 
     int writable_slave = listLength(g_pserver->masters) && g_pserver->repl_slave_ro == 0;
     if (c && writable_slave && !(c->flags & CLIENT_MASTER))
@@ -1320,26 +1296,24 @@ void setExpire(client *c, redisDb *db, robj *key, robj *subkey, long long when)
 
 void setExpire(client *c, redisDb *db, robj *key, expireEntry &&e)
 {
-    dictEntry *kde;
-
     serverAssert(GlobalLocksAcquired());
 
     /* Reuse the sds from the main dict in the expire dict */
-    kde = dictFind(db->pdict,ptrFromObj(key));
+    auto kde = db->find(key);
     serverAssertWithInfo(NULL,key,kde != NULL);
 
-    if (((robj*)dictGetVal(kde))->getrefcount(std::memory_order_relaxed) == OBJ_SHARED_REFCOUNT)
+    if (kde.val()->getrefcount(std::memory_order_relaxed) == OBJ_SHARED_REFCOUNT)
     {
         // shared objects cannot have the expire bit set, create a real object
-        dictSetVal(db->pdict, kde, dupStringObject((robj*)dictGetVal(kde)));
+        db->updateValue(kde, dupStringObject(kde.val()));
     }
 
-    if (((robj*)dictGetVal(kde))->FExpires())
+    if (kde.val()->FExpires())
         removeExpire(db, key);
 
-    e.setKeyUnsafe((sds)dictGetKey(kde));
-    db->setexpire->insert(e);
-    ((robj*)dictGetVal(kde))->SetFExpires(true);
+    e.setKeyUnsafe(kde.key());
+    db->setExpire(std::move(e));
+    kde.val()->SetFExpires(true);
 
 
     int writable_slave = listLength(g_pserver->masters) && g_pserver->repl_slave_ro == 0;
@@ -1349,22 +1323,19 @@ void setExpire(client *c, redisDb *db, robj *key, expireEntry &&e)
 
 /* Return the expire time of the specified key, or null if no expire
  * is associated with this key (i.e. the key is non volatile) */
-expireEntry *getExpire(redisDb *db, robj_roptr key) {
-    dictEntry *de;
-
+expireEntry *redisDbPersistentData::getExpire(robj_roptr key) {
     /* No expire? return ASAP */
-    if (db->setexpire->size() == 0)
+    if (expireSize() == 0)
         return nullptr;
 
-    de = dictFind(db->pdict, ptrFromObj(key));
-    if (de == NULL)
+    auto itr = find(szFromObj(key));
+    if (itr == nullptr)
         return nullptr;
-    robj *obj = (robj*)dictGetVal(de);
-    if (!obj->FExpires())
+    if (!itr.val()->FExpires())
         return nullptr;
 
-    auto itr = db->setexpire->find((sds)dictGetKey(de));
-    return itr.operator->();
+    auto itrExpire = findExpire(itr.key());
+    return itrExpire.operator->();
 }
 
 /* Propagate expires into slaves and the AOF file.
@@ -1394,7 +1365,7 @@ void propagateExpire(redisDb *db, robj *key, int lazy) {
 
 /* Check if the key is expired. Note, this does not check subexpires */
 int keyIsExpired(redisDb *db, robj *key) {
-    expireEntry *pexpire = getExpire(db,key);
+    expireEntry *pexpire = db->getExpire(key);
 
     if (pexpire == nullptr) return 0; /* No expire for this key */
 
@@ -1826,3 +1797,304 @@ unsigned int delKeysInSlot(unsigned int hashslot) {
 unsigned int countKeysInSlot(unsigned int hashslot) {
     return g_pserver->cluster->slots_keys_count[hashslot];
 }
+
+void redisDbPersistentData::initialize()
+{
+    m_pdict = dictCreate(&dbDictType,this);
+    m_setexpire = new(MALLOC_LOCAL) expireset();
+    m_fAllChanged = false;
+    m_fTrackingChanges = 0;
+}
+
+void redisDb::initialize(int id)
+{
+    redisDbPersistentData::initialize();
+    this->expireitr = setexpire()->end();
+    this->blocking_keys = dictCreate(&keylistDictType,NULL);
+    this->ready_keys = dictCreate(&objectKeyPointerValueDictType,NULL);
+    this->watched_keys = dictCreate(&keylistDictType,NULL);
+    this->id = id;
+    this->avg_ttl = 0;
+    this->last_expire_set = 0;
+    this->defrag_later = listCreate();
+}
+
+bool redisDbPersistentData::insert(char *key, robj *o)
+{
+    int res = dictAdd(m_pdict, key, o);
+    if (res == DICT_OK)
+        trackkey(key);
+    return (res == DICT_OK);
+}
+
+void redisDbPersistentData::tryResize()
+{
+    if (htNeedsResize(m_pdict))
+        dictResize(m_pdict);
+}
+
+size_t redisDb::clear(bool fAsync, void(callback)(void*))
+{
+    size_t removed = size();
+    if (fAsync) {
+        redisDbPersistentData::emptyDbAsync();
+    } else {
+        redisDbPersistentData::clear(callback);
+    }
+    expireitr = setexpire()->end();
+    return removed;
+}
+
+void redisDbPersistentData::clear(void(callback)(void*))
+{
+    dictEmpty(m_pdict,callback);
+    if (m_fTrackingChanges)
+        m_fAllChanged = true;
+    delete m_setexpire;
+    m_setexpire = new (MALLOC_LOCAL) expireset();
+    if (m_pstorage != nullptr)
+        m_pstorage->clear();
+}
+
+/* static */ void redisDbPersistentData::swap(redisDbPersistentData *db1, redisDbPersistentData *db2)
+{
+    redisDbPersistentData aux = *db1;
+    db1->m_pdict = db2->m_pdict;
+    db1->m_fTrackingChanges = db2->m_fTrackingChanges;
+    db1->m_fAllChanged = db2->m_fAllChanged;
+    db1->m_setexpire = db2->m_setexpire;
+    db1->m_pstorage = db2->m_pstorage;
+    db1->m_spdbSnapshot = db2->m_spdbSnapshot;
+
+    db2->m_pdict = aux.m_pdict;
+    db2->m_fTrackingChanges = aux.m_fTrackingChanges;
+    db2->m_fAllChanged = aux.m_fAllChanged;
+    db2->m_setexpire = aux.m_setexpire;
+    db2->m_pstorage = aux.m_pstorage;
+    db2->m_spdbSnapshot = aux.m_spdbSnapshot;
+
+    db1->m_pdict->privdata = static_cast<redisDbPersistentData*>(db1);
+    db2->m_pdict->privdata = static_cast<redisDbPersistentData*>(db2);
+}
+
+void redisDbPersistentData::setExpire(robj *key, robj *subkey, long long when)
+{
+    /* Reuse the sds from the main dict in the expire dict */
+    dictEntry *kde = dictFind(m_pdict,ptrFromObj(key));
+    serverAssertWithInfo(NULL,key,kde != NULL);
+    trackkey(key);
+
+    if (((robj*)dictGetVal(kde))->getrefcount(std::memory_order_relaxed) == OBJ_SHARED_REFCOUNT)
+    {
+        // shared objects cannot have the expire bit set, create a real object
+        dictSetVal(m_pdict, kde, dupStringObject((robj*)dictGetVal(kde)));
+    }
+
+    const char *szSubKey = (subkey != nullptr) ? szFromObj(subkey) : nullptr;
+    if (((robj*)dictGetVal(kde))->FExpires()) {
+        auto itr = m_setexpire->find((sds)dictGetKey(kde));
+        serverAssert(itr != m_setexpire->end());
+        expireEntry eNew(std::move(*itr));
+        eNew.update(szSubKey, when);
+        m_setexpire->erase(itr);
+        m_setexpire->insert(eNew);
+    }
+    else
+    {
+        expireEntry e((sds)dictGetKey(kde), szSubKey, when);
+        ((robj*)dictGetVal(kde))->SetFExpires(true);
+        m_setexpire->insert(e);
+    }
+}
+
+void redisDbPersistentData::setExpire(expireEntry &&e)
+{
+    trackkey(e.key());
+    m_setexpire->insert(e);
+}
+
+bool redisDb::FKeyExpires(const char *key)
+{
+    return setexpireUnsafe()->find(key) != setexpire()->end();
+}
+
+void redisDbPersistentData::updateValue(dict_iter itr, robj *val)
+{
+    trackkey(itr.key());
+    dictSetVal(m_pdict, itr.de, val);
+}
+
+void redisDbPersistentData::ensure(const char *key)
+{
+    dictEntry *de = dictFind(m_pdict, key);
+    ensure(de);
+}
+
+void redisDbPersistentData::ensure(dictEntry *de)
+{
+    if (de != nullptr && dictGetVal(de) == nullptr)
+    {
+        if (m_spdbSnapshot != nullptr)
+        {
+            auto itr = m_spdbSnapshot->find((const char*)dictGetKey(de));
+            serverAssert(itr != m_spdbSnapshot->end());
+            if (itr.val()->getrefcount(std::memory_order_relaxed) == OBJ_SHARED_REFCOUNT)
+            {
+                dictSetVal(m_pdict, de, itr.val());
+            }
+            else
+            {
+                sds strT = serializeStoredObject(itr.val());
+                robj *objNew = deserializeStoredObject(strT, sdslen(strT));
+                sdsfree(strT);
+                dictSetVal(m_pdict, de, objNew);
+                serverAssert(objNew->getrefcount(std::memory_order_relaxed) == 1);
+                serverAssert(objNew->mvcc_tstamp == itr.val()->mvcc_tstamp);
+            }
+        }
+        else
+        {
+            serverAssert(m_pstorage != nullptr);
+            sds key = (sds)dictGetKey(de);
+            m_pstorage->retrieve(key, sdslen(key), true, [&](const char *, size_t, const void *data, size_t cb){
+                robj *o = deserializeStoredObject(data, cb);
+                serverAssert(o != nullptr);
+                dictSetVal(m_pdict, de, o);
+            });
+        }
+    }
+}
+
+void redisDbPersistentData::storeKey(const char *szKey, size_t cchKey, robj *o)
+{
+    sds temp = serializeStoredObject(o);
+    m_pstorage->insert(szKey, cchKey, temp, sdslen(temp));
+    sdsfree(temp);
+}
+
+void redisDbPersistentData::storeDatabase()
+{
+    dictIterator *di = dictGetIterator(m_pdict);
+    dictEntry *de;
+    while ((de = dictNext(di)) != NULL) {
+        sds key = (sds)dictGetKey(de);
+        robj *o = (robj*)dictGetVal(de);
+        storeKey(key, sdslen(key), o);
+    }
+    dictReleaseIterator(di);
+}
+
+void redisDbPersistentData::processChanges()
+{
+    --m_fTrackingChanges;
+    serverAssert(m_fTrackingChanges >= 0);
+
+    if (m_pstorage != nullptr)
+    {
+        if (m_fTrackingChanges == 0)
+        {
+            if (m_fAllChanged)
+            {
+                m_pstorage->clear();
+                storeDatabase();
+            }
+            else
+            {
+                for (auto &str : m_setchanged)
+                {
+                    sds sdsKey = sdsnewlen(str.data(), str.size());
+                    robj *o = find(sdsKey);
+                    if (o != nullptr)
+                    {
+                        storeKey(str.data(), str.size(), o);
+                    }
+                    else
+                    {
+                        m_pstorage->erase(str.data(), str.size());
+                    }
+                    sdsfree(sdsKey);
+                }
+            }
+        }
+    }
+    m_setchanged.clear();
+}
+
+std::shared_ptr<redisDbPersistentData> redisDbPersistentData::createSnapshot()
+{
+    serverAssert(GlobalLocksAcquired());
+    serverAssert(m_spdbSnapshot == nullptr);
+    auto spdb = std::make_shared<redisDbPersistentData>();
+    spdb->m_pdict = dictCreate(&dbDictType,spdb.get());
+    spdb->m_fAllChanged = false;
+    spdb->m_fTrackingChanges = 0;
+    spdb->m_pdict->rehashidx = m_pdict->rehashidx;
+    spdb->m_pdict->iterators++; // fake an iterator so it doesn't rehash
+    if (m_setexpire != nullptr)
+        spdb->m_setexpire = new (MALLOC_LOCAL) expireset(*m_setexpire);
+
+    for (unsigned iht = 0; iht < 2; ++iht)
+    {
+        spdb->m_pdict->ht[iht] = m_pdict->ht[iht];
+        if (m_pdict->ht[iht].size)
+            m_pdict->ht[iht].table = (dictEntry**)zcalloc(m_pdict->ht[iht].size*sizeof(dictEntry*), MALLOC_SHARED);
+        else
+            m_pdict->ht[iht].table = nullptr;
+
+        for (size_t idx = 0; idx < m_pdict->ht[iht].size; ++idx)
+        {
+            const dictEntry *deSrc = spdb->m_pdict->ht[iht].table[idx];
+            dictEntry **pdeDst = &m_pdict->ht[iht].table[idx];
+            while (deSrc != nullptr)
+            {
+                *pdeDst = (dictEntry*)zmalloc(sizeof(dictEntry), MALLOC_SHARED);
+                (*pdeDst)->key = deSrc->key;
+                (*pdeDst)->v.val = nullptr;
+                (*pdeDst)->next = nullptr;
+                pdeDst = &(*pdeDst)->next;
+                deSrc = deSrc->next;
+            }
+        }
+    }
+    
+    m_spdbSnapshot = std::move(spdb);
+    return m_spdbSnapshot;
+}
+
+void redisDbPersistentData::endSnapshot(const redisDbPersistentData *psnapshot)
+{
+    serverAssert(GlobalLocksAcquired());
+    serverAssert(m_spdbSnapshot.get() == psnapshot);
+
+    dictIterator *di = dictGetIterator(m_pdict);
+    dictEntry *de;
+    while ((de = dictNext(di)) != NULL)
+    {
+        dictEntry *deSnapshot = dictFind(m_spdbSnapshot->m_pdict, dictGetKey(de));
+        if (dictGetVal(de) == nullptr)
+        {
+            if (deSnapshot != nullptr)
+            {
+                de->v.val = deSnapshot->v.val;
+                deSnapshot->v.val = nullptr;
+            }
+        }
+        if (deSnapshot && (dictGetKey(deSnapshot) == dictGetKey(de)))
+        {
+            // The key is owned by the parent snapshot, so we modify the DB key dtor
+            //  to ensure the key is not free'd during the delete
+            m_spdbSnapshot->m_pdict->type = &dbSnapshotDictType;
+            dictDelete(m_spdbSnapshot->m_pdict, dictGetKey(de));
+            m_spdbSnapshot->m_pdict->type = &dbDictType;
+        }
+    }
+    dictReleaseIterator(di);
+    m_spdbSnapshot->m_pdict->iterators--;
+    m_spdbSnapshot = nullptr;
+}
+
+redisDbPersistentData::~redisDbPersistentData()
+{
+    dictRelease(m_pdict);
+    delete m_setexpire;
+}

src/debug.cpp

@@ -126,7 +126,7 @@ void mixStringObjectDigest(unsigned char *digest, robj_roptr o) {
 void xorObjectDigest(redisDb *db, robj_roptr keyobj, unsigned char *digest, robj_roptr o) {
     uint32_t aux = htonl(o->type);
     mixDigest(digest,&aux,sizeof(aux));
-    expireEntry *pexpire = getExpire(db,keyobj);
+    expireEntry *pexpire = db->getExpire(keyobj);
     long long expiretime = -1;
     char buf[128];
 
@@ -266,8 +266,6 @@ void xorObjectDigest(redisDb *db, robj_roptr keyobj, unsigned char *digest, robj
  * a different digest. */
 void computeDatasetDigest(unsigned char *final) {
     unsigned char digest[20];
-    dictIterator *di = NULL;
-    dictEntry *de;
     int j;
     uint32_t aux;
 
@@ -276,8 +274,7 @@ void computeDatasetDigest(unsigned char *final) {
     for (j = 0; j < cserver.dbnum; j++) {
         redisDb *db = g_pserver->db+j;
 
-        if (dictSize(db->pdict) == 0) continue;
-        di = dictGetSafeIterator(db->pdict);
+        if (db->size() == 0) continue;
 
         /* hash the DB id, so the same dataset moved in a different
          * DB will lead to a different digest */
@@ -285,24 +282,21 @@ void computeDatasetDigest(unsigned char *final) {
         mixDigest(final,&aux,sizeof(aux));
 
         /* Iterate this DB writing every entry */
-        while((de = dictNext(di)) != NULL) {
-            sds key;
-            robj *keyobj, *o;
+        db->iterate([&](const char *key, robj *o)->bool {
+            robj *keyobj;
 
             memset(digest,0,20); /* This key-val digest */
-            key = (sds)dictGetKey(de);
             keyobj = createStringObject(key,sdslen(key));
 
             mixDigest(digest,key,sdslen(key));
 
-            o = (robj*)dictGetVal(de);
             xorObjectDigest(db,keyobj,digest,o);
 
             /* We can finally xor the key-val digest to the final digest */
             xorDigest(final,digest,20);
             decrRefCount(keyobj);
-        }
-        dictReleaseIterator(di);
+            return true;
+        });
     }
 }
 
@@ -365,7 +359,7 @@ NULL
     } else if (!strcasecmp(szFromObj(c->argv[1]),"reload")) {
         rdbSaveInfo rsi, *rsiptr;
         rsiptr = rdbPopulateSaveInfo(&rsi);
-        if (rdbSave(rsiptr) != C_OK) {
+        if (rdbSave(nullptr, rsiptr) != C_OK) {
             addReply(c,shared.err);
             return;
         }
@@ -394,15 +388,14 @@ NULL
         serverLog(LL_WARNING,"Append Only File loaded by DEBUG LOADAOF");
         addReply(c,shared.ok);
     } else if (!strcasecmp(szFromObj(c->argv[1]),"object") && c->argc == 3) {
-        dictEntry *de;
         robj *val;
         const char *strenc;
 
-        if ((de = dictFind(c->db->pdict,ptrFromObj(c->argv[2]))) == NULL) {
+        val = c->db->find(c->argv[2]);
+        if (val == NULL) {
             addReply(c,shared.nokeyerr);
             return;
         }
-        val = (robj*)dictGetVal(de);
         strenc = strEncoding(val->encoding);
 
         char extra[138] = {0};
@@ -446,16 +439,14 @@ NULL
             strenc, rdbSavedObjectLen(val),
             val->lru, estimateObjectIdleTime(val)/1000, extra);
     } else if (!strcasecmp(szFromObj(c->argv[1]),"sdslen") && c->argc == 3) {
-        dictEntry *de;
-        robj *val;
-        sds key;
+        auto itr = c->db->find(c->argv[2]);
+        robj *val = itr.val();
+        const char *key = itr.key();
 
-        if ((de = dictFind(c->db->pdict,ptrFromObj(c->argv[2]))) == NULL) {
+        if (val == NULL) {
             addReply(c,shared.nokeyerr);
             return;
         }
-        val = (robj*)dictGetVal(de);
-        key = (sds)dictGetKey(de);
 
         if (val->type != OBJ_STRING || !sdsEncodedObject(val)) {
             addReplyError(c,"Not an sds encoded string.");
@@ -465,16 +456,16 @@ NULL
                 "val_sds_len:%lld, val_sds_avail:%lld, val_zmalloc: %lld",
                 (long long) sdslen(key),
                 (long long) sdsavail(key),
-                (long long) sdsZmallocSize(key),
+                (long long) sdsZmallocSize((sds)key),
                 (long long) sdslen(szFromObj(val)),
                 (long long) sdsavail(szFromObj(val)),
                 (long long) getStringObjectSdsUsedMemory(val));
         }
     } else if (!strcasecmp(szFromObj(c->argv[1]),"ziplist") && c->argc == 3) {
-        robj *o;
+        robj_roptr o;
 
         if ((o = objectCommandLookupOrReply(c,c->argv[2],shared.nokeyerr))
-                == NULL) return;
+                == nullptr) return;
 
         if (o->encoding != OBJ_ENCODING_ZIPLIST) {
             addReplyError(c,"Not an sds encoded string.");
@@ -490,7 +481,7 @@ NULL
 
         if (getLongFromObjectOrReply(c, c->argv[2], &keys, NULL) != C_OK)
             return;
-        dictExpand(c->db->pdict,keys);
+        c->db->expand(keys);
         for (j = 0; j < keys; j++) {
             long valsize = 0;
             snprintf(buf,sizeof(buf),"%s:%lu",
@@ -641,20 +632,20 @@ NULL
         }
 
         stats = sdscatprintf(stats,"[Dictionary HT]\n");
-        dictGetStats(buf,sizeof(buf),g_pserver->db[dbid].pdict);
+        g_pserver->db[dbid].getStats(buf,sizeof(buf));
         stats = sdscat(stats,buf);
 
         stats = sdscatprintf(stats,"[Expires set]\n");
-        g_pserver->db[dbid].setexpire->getstats(buf, sizeof(buf));
+        g_pserver->db[dbid].getExpireStats(buf, sizeof(buf));
         stats = sdscat(stats, buf);
 
         addReplyBulkSds(c,stats);
     } else if (!strcasecmp(szFromObj(c->argv[1]),"htstats-key") && c->argc == 3) {
-        robj *o;
+        robj_roptr o;
         dict *ht = NULL;
 
         if ((o = objectCommandLookupOrReply(c,c->argv[2],shared.nokeyerr))
-                == NULL) return;
+                == nullptr) return;
 
         /* Get the hash table reference from the object, if possible. */
         switch (o->encoding) {
@@ -1256,12 +1247,10 @@ void logCurrentClient(void) {
      * selected DB, and if so print info about the associated object. */
     if (cc->argc >= 1) {
         robj *val, *key;
-        dictEntry *de;
 
         key = getDecodedObject(cc->argv[1]);
-        de = dictFind(cc->db->pdict, ptrFromObj(key));
-        if (de) {
-            val = (robj*)dictGetVal(de);
+        val = cc->db->find(key);
+        if (val) {
             serverLog(LL_WARNING,"key '%s' found in DB containing the following object:", (char*)ptrFromObj(key));
             serverLogObjectDebugInfo(val);
         }

src/defrag.cpp

@@ -781,12 +781,14 @@ long defragKey(redisDb *db, dictEntry *de) {
     newsds = activeDefragSds(keysds);
     if (newsds)
         defragged++, de->key = newsds;
-    if (!db->setexpire->empty()) {
-        replaceSateliteOSetKeyPtr(*db->setexpire, keysds, newsds);
+    if (!db->setexpire()->empty()) {
+        replaceSateliteOSetKeyPtr(*const_cast<expireset*>(db->setexpire()), keysds, newsds);
     }
 
     /* Try to defrag robj and / or string value. */
     ob = (robj*)dictGetVal(de);
+    if (ob == nullptr)
+        return defragged;
     if ((newob = activeDefragStringOb(ob, &defragged))) {
         de->v.val = newob;
         ob = newob;
@@ -853,7 +855,7 @@ void defragScanCallback(void *privdata, const dictEntry *de) {
     g_pserver->stat_active_defrag_scanned++;
 }
 
-/* Defrag scan callback for each hash table bicket,
+/* Defrag scan callback for each hash table bucket,
  * used in order to defrag the dictEntry allocations. */
 void defragDictBucketCallback(void *privdata, dictEntry **bucketref) {
     UNUSED(privdata); /* NOTE: this function is also used by both activeDefragCycle and scanLaterHash, etc. don't use privdata */
@@ -902,9 +904,8 @@ long defragOtherGlobals() {
 
 /* returns 0 more work may or may not be needed (see non-zero cursor),
  * and 1 if time is up and more work is needed. */
-int defragLaterItem(dictEntry *de, unsigned long *cursor, long long endtime) {
-    if (de) {
-        robj *ob = (robj*)dictGetVal(de);
+int defragLaterItem(robj *ob, unsigned long *cursor, long long endtime) {
+    if (ob) {
         if (ob->type == OBJ_LIST) {
             g_pserver->stat_active_defrag_hits += scanLaterList(ob);
             *cursor = 0; /* list has no scan, we must finish it in one go */
@@ -959,11 +960,11 @@ int defragLaterStep(redisDb *db, long long endtime) {
         }
 
         /* each time we enter this function we need to fetch the key from the dict again (if it still exists) */
-        dictEntry *de = dictFind(db->pdict, current_key);
+        robj *o = db->find(current_key);
         key_defragged = g_pserver->stat_active_defrag_hits;
         do {
             int quit = 0;
-            if (defragLaterItem(de, &cursor, endtime))
+            if (defragLaterItem(o, &cursor, endtime))
                 quit = 1; /* time is up, we didn't finish all the work */
 
             /* Don't start a new BIG key in this loop, this is because the
@@ -1045,7 +1046,7 @@ void activeDefragCycle(void) {
     mstime_t latency;
     int quit = 0;
 
-    if (g_pserver->aof_child_pid!=-1 || g_pserver->rdb_child_pid!=-1)
+    if (g_pserver->aof_child_pid!=-1 || g_pserver->FRdbSaveInProgress())
         return; /* Defragging memory while there's a fork will just do damage. */
 
     /* Once a second, check if we the fragmentation justfies starting a scan
@@ -1112,7 +1113,8 @@ void activeDefragCycle(void) {
                 break; /* this will exit the function and we'll continue on the next cycle */
             }
 
-            cursor = dictScan(db->pdict, cursor, defragScanCallback, defragDictBucketCallback, db);
+            // we actually look at the objects too but defragScanCallback can handle missing values
+            cursor = dictScan(db->dictUnsafeKeyOnly(), cursor, defragScanCallback, defragDictBucketCallback, db);
 
             /* Once in 16 scan iterations, 512 pointer reallocations. or 64 keys
              * (if we have a lot of pointers in one hash bucket or rehasing),

src/evict.cpp

@@ -87,7 +87,7 @@ unsigned int LRU_CLOCK(void) {
 
 /* Given an object returns the min number of milliseconds the object was never
  * requested, using an approximated LRU algorithm. */
-unsigned long long estimateObjectIdleTime(robj *o) {
+unsigned long long estimateObjectIdleTime(robj_roptr o) {
     unsigned long long lruclock = LRU_CLOCK();
     if (lruclock >= o->lru) {
         return (lruclock - o->lru) * LRU_CLOCK_RESOLUTION;
@@ -252,19 +252,20 @@ struct visitFunctor
         return count < g_pserver->maxmemory_samples;
     }
 };
-void evictionPoolPopulate(int dbid, dict *dbdict, expireset *setexpire, struct evictionPoolEntry *pool)
+void evictionPoolPopulate(int dbid, redisDb *db, expireset *setexpire, struct evictionPoolEntry *pool)
 {
     if (setexpire != nullptr)
     {
-        visitFunctor visitor { dbid, dbdict, pool, 0 };
+        visitFunctor visitor { dbid, db->dictUnsafeKeyOnly(), pool, 0 };
         setexpire->random_visit(visitor);
     }
     else
     {
         dictEntry **samples = (dictEntry**)alloca(g_pserver->maxmemory_samples * sizeof(dictEntry*));
-        int count = dictGetSomeKeys(dbdict,samples,g_pserver->maxmemory_samples);
+        int count = dictGetSomeKeys(db->dictUnsafeKeyOnly(),samples,g_pserver->maxmemory_samples);
         for (int j = 0; j < count; j++) {
             robj *o = (robj*)dictGetVal(samples[j]);
+            serverAssert(o != nullptr); // BUG!!! We have to get the info we need here without permanently rehydrating the obj
             processEvictionCandidate(dbid, (sds)dictGetKey(samples[j]), o, nullptr, pool);
         }
     }
@@ -504,16 +505,16 @@ int freeMemoryIfNeeded(void) {
                     db = g_pserver->db+i;
                     if (g_pserver->maxmemory_policy & MAXMEMORY_FLAG_ALLKEYS)
                     {
-                        if ((keys = dictSize(db->pdict)) != 0) {
-                            evictionPoolPopulate(i, db->pdict, nullptr, pool);
+                        if ((keys = db->size()) != 0) {
+                            evictionPoolPopulate(i, db, nullptr, pool);
                             total_keys += keys;
                         }
                     }
                     else
                     {
-                        keys = db->setexpire->size();
+                        keys = db->expireSize();
                         if (keys != 0)
-                            evictionPoolPopulate(i, db->pdict, db->setexpire, pool);
+                            evictionPoolPopulate(i, db, db->setexpireUnsafe(), pool);
                         total_keys += keys;
                     }
                 }
@@ -525,9 +526,9 @@ int freeMemoryIfNeeded(void) {
                     bestdbid = pool[k].dbid;
                     sds key = nullptr;
 
-                    dictEntry *de = dictFind(g_pserver->db[pool[k].dbid].pdict,pool[k].key);
-                    if (de != nullptr && (g_pserver->maxmemory_policy & MAXMEMORY_FLAG_ALLKEYS || ((robj*)dictGetVal(de))->FExpires()))
-                        key = (sds)dictGetKey(de);
+                    auto itr = g_pserver->db[pool[k].dbid].find(pool[k].key);
+                    if (itr != nullptr && (g_pserver->maxmemory_policy & MAXMEMORY_FLAG_ALLKEYS || itr.val()->FExpires()))
+                        key = itr.key();
 
                     /* Remove the entry from the pool. */
                     if (pool[k].key != pool[k].cached)
@@ -559,18 +560,18 @@ int freeMemoryIfNeeded(void) {
                 db = g_pserver->db+j;
                 if (g_pserver->maxmemory_policy == MAXMEMORY_ALLKEYS_RANDOM)
                 {
-                    if (dictSize(db->pdict) != 0) {
-                        dictEntry *de = dictGetRandomKey(db->pdict);
-                        bestkey = (sds)dictGetKey(de);
+                    if (db->size() != 0) {
+                        auto itr = db->random();
+                        bestkey = itr.key();
                         bestdbid = j;
                         break;
                     }
                 }
                 else
                 {
-                    if (!db->setexpire->empty())
+                    if (db->expireSize())
                     {
-                        bestkey = (sds)db->setexpire->random_value().key();
+                        bestkey = (sds)db->random_expire().key();
                         bestdbid = j;
                         break;
                     }
@@ -581,16 +582,17 @@ int freeMemoryIfNeeded(void) {
         /* Finally remove the selected key. */
         if (bestkey) {
             db = g_pserver->db+bestdbid;
+
             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. */
+            * 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)
@@ -609,18 +611,18 @@ int freeMemoryIfNeeded(void) {
             keys_freed++;
 
             /* When the memory to free starts to be big enough, we may
-             * start spending so much time here that is impossible to
-             * deliver data to the slaves fast enough, so we force the
-             * transmission here inside the loop. */
+            * start spending so much time here that is impossible to
+            * deliver data to the slaves fast enough, so we force the
+            * transmission here inside the loop. */
             if (slaves) flushSlavesOutputBuffers();
 
             /* Normally our stop condition is the ability to release
-             * a fixed, pre-computed amount of memory. However when we
-             * are deleting objects in another thread, it's better to
-             * check, from time to time, if we already reached our target
-             * memory, since the "mem_freed" amount is computed only
-             * across the dbAsyncDelete() call, while the thread can
-             * release the memory all the time. */
+            * a fixed, pre-computed amount of memory. However when we
+            * are deleting objects in another thread, it's better to
+            * check, from time to time, if we already reached our target
+            * memory, since the "mem_freed" amount is computed only
+            * across the dbAsyncDelete() call, while the thread can
+            * release the memory all the time. */
             if (g_pserver->lazyfree_lazy_eviction && !(keys_freed % 16)) {
                 if (getMaxmemoryState(NULL,NULL,NULL,NULL) == C_OK) {
                     /* Let's satisfy our stop condition. */

src/expire.cpp

@@ -74,8 +74,7 @@ void activeExpireCycleExpire(redisDb *db, expireEntry &e, long long now) {
     }
 
     expireEntryFat *pfat = e.pfatentry();
-    dictEntry *de = dictFind(db->pdict, e.key());
-    robj *val = (robj*)dictGetVal(de);
+    robj *val = db->find(e.key());
     int deleted = 0;
     while (!pfat->FEmpty())
     {
@@ -157,7 +156,7 @@ void expireMemberCore(client *c, robj *key, robj *subkey, long long basetime, lo
 
     /* No key, return zero. */
     robj *val = lookupKeyWriteOrReply(c, key, shared.czero);
-    if (val == NULL) {
+    if (val == nullptr) {
         return;
     }
 
@@ -296,7 +295,7 @@ void activeExpireCycle(int type) {
         now = mstime();
 
         /* If there is nothing to expire try next DB ASAP. */
-        if (db->setexpire->empty())
+        if (db->setexpireUnsafe()->empty())
         {
             db->avg_ttl = 0;
             db->last_expire_set = now;
@@ -306,7 +305,7 @@ void activeExpireCycle(int type) {
         size_t expired = 0;
         size_t tried = 0;
         long long check = ACTIVE_EXPIRE_CYCLE_FAST_DURATION;    // assume a check is roughly 1us.  It isn't but good enough
-        db->expireitr = db->setexpire->enumerate(db->expireitr, now, [&](expireEntry &e) __attribute__((always_inline)) {
+        db->expireitr = db->setexpireUnsafe()->enumerate(db->expireitr, now, [&](expireEntry &e) __attribute__((always_inline)) {
             if (e.when() < now)
             {
                 activeExpireCycleExpire(db, e, now);
@@ -406,16 +405,16 @@ void expireSlaveKeys(void) {
                 redisDb *db = g_pserver->db+dbid;
 
                 // the expire is hashed based on the key pointer, so we need the point in the main db
-                dictEntry *deMain = dictFind(db->pdict, keyname);
-                auto itr = db->setexpire->end();
-                if (deMain != nullptr)
-                    itr = db->setexpire->find((sds)dictGetKey(deMain));
+                auto itrDB = db->find(keyname);
+                auto itrExpire = db->setexpire()->end();
+                if (itrDB != nullptr)
+                    itrExpire = db->setexpireUnsafe()->find(itrDB.key());
                 int expired = 0;
 
-                if (itr != db->setexpire->end())
+                if (itrExpire != db->setexpire()->end())
                 {
-                    if (itr->when() < start) {
-                        activeExpireCycleExpire(g_pserver->db+dbid,*itr,start);
+                    if (itrExpire->when() < start) {
+                        activeExpireCycleExpire(g_pserver->db+dbid,*itrExpire,start);
                         expired = 1;
                     }
                 }
@@ -424,7 +423,7 @@ void expireSlaveKeys(void) {
                  * corresponding bit in the new bitmap we set as value.
                  * At the end of the loop if the bitmap is zero, it means we
                  * no longer need to keep track of this key. */
-                if (itr != db->setexpire->end() && !expired) {
+                if (itrExpire != db->setexpire()->end() && !expired) {
                     noexpire++;
                     new_dbids |= (uint64_t)1 << dbid;
                 }
@@ -592,7 +591,7 @@ void ttlGenericCommand(client *c, int output_ms) {
 
     /* The key exists. Return -1 if it has no expire, or the actual
         * TTL value otherwise. */
-    expireEntry *pexpire = getExpire(c->db,c->argv[1]);
+    expireEntry *pexpire = c->db->getExpire(c->argv[1]);
 
     if (c->argc == 2) {
         // primary expire    
@@ -648,7 +647,7 @@ void persistCommand(client *c) {
                 addReply(c,shared.czero);
             }
         } else if (c->argc == 3) {
-            if (removeSubkeyExpire(c->db, c->argv[1], c->argv[2])) {
+            if (c->db->removeSubkeyExpire(c->argv[1], c->argv[2])) {
                 addReply(c,shared.cone);
                 g_pserver->dirty++;
             } else {

src/geohash_helper.cpp

@@ -46,13 +46,13 @@
 #define ECCENT (sqrt(1.0 - (RATIO *RATIO)))
 #define COM (0.5 * ECCENT)
 
-/// @brief The usual PI/180 constant
-const double DEG_TO_RAD = 0.017453292519943295769236907684886;
 /// @brief Earth's quatratic mean radius for WGS-84
 const double EARTH_RADIUS_IN_METERS = 6372797.560856;
 
 const double MERCATOR_MAX = 20037726.37;
+#if 0 //unused
 const double MERCATOR_MIN = -20037726.37;
+#endif
 
 static inline double deg_rad(double ang) { return ang * D_R; }
 static inline double rad_deg(double ang) { return ang / D_R; }

src/lazyfree.cpp

@@ -51,18 +51,18 @@ size_t lazyfreeGetFreeEffort(robj *obj) {
  * a lazy free list instead of being freed synchronously. The lazy free list
  * will be reclaimed in a different bio.c thread. */
 #define LAZYFREE_THRESHOLD 64
-int dbAsyncDelete(redisDb *db, robj *key) {
+bool redisDbPersistentData::asyncDelete(robj *key) {
     /* If the value is composed of a few allocations, to free in a lazy way
      * is actually just slower... So under a certain limit we just free
      * the object synchronously. */
-    dictEntry *de = dictUnlink(db->pdict,ptrFromObj(key));
+    dictEntry *de = dictUnlink(m_pdict,ptrFromObj(key));
     if (de) {
         robj *val = (robj*)dictGetVal(de);
         if (val->FExpires())
         {
             /* Deleting an entry from the expires dict will not free the sds of
              * the key, because it is shared with the main dictionary. */
-            removeExpireCore(db,key,de);
+            removeExpire(key,dict_iter(de));
         }
 
         size_t free_effort = lazyfreeGetFreeEffort(val);
@@ -78,21 +78,25 @@ int dbAsyncDelete(redisDb *db, robj *key) {
         if (free_effort > LAZYFREE_THRESHOLD && val->getrefcount(std::memory_order_relaxed) == 1) {
             atomicIncr(lazyfree_objects,1);
             bioCreateBackgroundJob(BIO_LAZY_FREE,val,NULL,NULL);
-            dictSetVal(db->pdict,de,NULL);
+            dictSetVal(m_pdict,de,NULL);
         }
     }
 
     /* Release the key-val pair, or just the key if we set the val
      * field to NULL in order to lazy free it later. */
     if (de) {
-        dictFreeUnlinkedEntry(db->pdict,de);
+        dictFreeUnlinkedEntry(m_pdict,de);
         if (g_pserver->cluster_enabled) slotToKeyDel(key);
-        return 1;
+        return true;
     } else {
-        return 0;
+        return false;
     }
 }
 
+int dbAsyncDelete(redisDb *db, robj *key) {
+    return db->asyncDelete(key);
+}
+
 /* Free an object, if the object is huge enough, free it in async way. */
 void freeObjAsync(robj *o) {
     size_t free_effort = lazyfreeGetFreeEffort(o);
@@ -107,12 +111,15 @@ void freeObjAsync(robj *o) {
 /* Empty a Redis DB asynchronously. What the function does actually is to
  * create a new empty set of hash tables and scheduling the old ones for
  * lazy freeing. */
-void emptyDbAsync(redisDb *db) {
-    dict *oldht1 = db->pdict;
-    auto *set = db->setexpire;
-    db->setexpire = new (MALLOC_LOCAL) expireset();
-    db->expireitr = db->setexpire->end();
-    db->pdict = dictCreate(&dbDictType,NULL);
+void redisDbPersistentData::emptyDbAsync() {
+    dict *oldht1 = m_pdict;
+    auto *set = m_setexpire;
+    m_setexpire = new (MALLOC_LOCAL) expireset();
+    m_pdict = dictCreate(&dbDictType,this);
+    if (m_pstorage != nullptr)
+        m_pstorage->clear();
+    if (m_fTrackingChanges)
+        m_fAllChanged = true;
     atomicIncr(lazyfree_objects,dictSize(oldht1));
     bioCreateBackgroundJob(BIO_LAZY_FREE,NULL,oldht1,set);
 }

src/module.cpp

@@ -1644,7 +1644,7 @@ int RM_UnlinkKey(RedisModuleKey *key) {
  * If no TTL is associated with the key or if the key is empty,
  * REDISMODULE_NO_EXPIRE is returned. */
 mstime_t RM_GetExpire(RedisModuleKey *key) {
-    expireEntry *pexpire = getExpire(key->db,key->key);
+    expireEntry *pexpire = key->db->getExpire(key->key);
     mstime_t expire = -1;
     if (pexpire != nullptr)
         pexpire->FGetPrimaryExpire(&expire);

src/multi.cpp

@@ -333,7 +333,7 @@ void touchWatchedKeysOnFlush(int dbid) {
              * key exists, mark the client as dirty, as the key will be
              * removed. */
             if (dbid == -1 || wk->db->id == dbid) {
-                if (dictFind(wk->db->pdict, ptrFromObj(wk->key)) != NULL)
+                if (wk->db->find(wk->key) != NULL)
                     c->flags |= CLIENT_DIRTY_CAS;
             }
         }

src/object.cpp

@@ -789,7 +789,7 @@ size_t streamRadixTreeMemoryUsage(rax *rax) {
  * case of aggregated data types where only "sample_size" elements
  * are checked and averaged to estimate the total size. */
 #define OBJ_COMPUTE_SIZE_DEF_SAMPLES 5 /* Default sample size. */
-size_t objectComputeSize(robj *o, size_t sample_size) {
+size_t objectComputeSize(robj_roptr o, size_t sample_size) {
     sds ele, ele2;
     dict *d;
     dictIterator *di;
@@ -800,7 +800,7 @@ size_t objectComputeSize(robj *o, size_t sample_size) {
         if(o->encoding == OBJ_ENCODING_INT) {
             asize = sizeof(*o);
         } else if(o->encoding == OBJ_ENCODING_RAW) {
-            asize = sdsAllocSize(szFromObj(o))+sizeof(*o);
+            asize = sdsAllocSize((sds)szFromObj(o))+sizeof(*o);
         } else if(o->encoding == OBJ_ENCODING_EMBSTR) {
             asize = sdslen(szFromObj(o))+2+sizeof(*o);
         } else {
@@ -1054,20 +1054,20 @@ struct redisMemOverhead *getMemoryOverheadData(void) {
 
     for (j = 0; j < cserver.dbnum; j++) {
         redisDb *db = g_pserver->db+j;
-        long long keyscount = dictSize(db->pdict);
+        long long keyscount = db->size();
         if (keyscount==0) continue;
 
         mh->total_keys += keyscount;
         mh->db = (decltype(mh->db))zrealloc(mh->db,sizeof(mh->db[0])*(mh->num_dbs+1), MALLOC_LOCAL);
         mh->db[mh->num_dbs].dbid = j;
 
-        mem = dictSize(db->pdict) * sizeof(dictEntry) +
-              dictSlots(db->pdict) * sizeof(dictEntry*) +
-              dictSize(db->pdict) * sizeof(robj);
+        mem = db->size() * sizeof(dictEntry) +
+              db->slots() * sizeof(dictEntry*) +
+              db->size() * sizeof(robj);
         mh->db[mh->num_dbs].overhead_ht_main = mem;
         mem_total+=mem;
 
-        mem = db->setexpire->bytes_used();
+        mem = db->setexpire()->bytes_used();
         mh->db[mh->num_dbs].overhead_ht_expires = mem;
         mem_total+=mem;
 
@@ -1246,15 +1246,12 @@ void objectSetLRUOrLFU(robj *val, long long lfu_freq, long long lru_idle,
 
 /* This is a helper function for the OBJECT command. We need to lookup keys
  * without any modification of LRU or other parameters. */
-robj *objectCommandLookup(client *c, robj *key) {
-    dictEntry *de;
-
-    if ((de = dictFind(c->db->pdict,ptrFromObj(key))) == NULL) return NULL;
-    return (robj*) dictGetVal(de);
+robj_roptr objectCommandLookup(client *c, robj *key) {
+    return c->db->find(key);
 }
 
-robj *objectCommandLookupOrReply(client *c, robj *key, robj *reply) {
-    robj *o = objectCommandLookup(c,key);
+robj_roptr objectCommandLookupOrReply(client *c, robj *key, robj *reply) {
+    robj_roptr o = objectCommandLookup(c,key);
 
     if (!o) addReply(c, reply);
     return o;
@@ -1263,7 +1260,7 @@ robj *objectCommandLookupOrReply(client *c, robj *key, robj *reply) {
 /* Object command allows to inspect the internals of an Redis Object.
  * Usage: OBJECT <refcount|encoding|idletime|freq> <key> */
 void objectCommand(client *c) {
-    robj *o;
+    robj_roptr o;
 
     if (c->argc == 2 && !strcasecmp(szFromObj(c->argv[1]),"help")) {
         const char *help[] = {
@@ -1276,15 +1273,15 @@ NULL
         addReplyHelp(c, help);
     } else if (!strcasecmp(szFromObj(c->argv[1]),"refcount") && c->argc == 3) {
         if ((o = objectCommandLookupOrReply(c,c->argv[2],shared.null[c->resp]))
-                == NULL) return;
+                == nullptr) return;
         addReplyLongLong(c,o->getrefcount(std::memory_order_relaxed));
     } else if (!strcasecmp(szFromObj(c->argv[1]),"encoding") && c->argc == 3) {
         if ((o = objectCommandLookupOrReply(c,c->argv[2],shared.null[c->resp]))
-                == NULL) return;
+                == nullptr) return;
         addReplyBulkCString(c,strEncoding(o->encoding));
     } else if (!strcasecmp(szFromObj(c->argv[1]),"idletime") && c->argc == 3) {
         if ((o = objectCommandLookupOrReply(c,c->argv[2],shared.null[c->resp]))
-                == NULL) return;
+                == nullptr) return;
         if (g_pserver->maxmemory_policy & MAXMEMORY_FLAG_LFU) {
             addReplyError(c,"An LFU maxmemory policy is selected, idle time not tracked. Please note that when switching between policies at runtime LRU and LFU data will take some time to adjust.");
             return;
@@ -1292,7 +1289,7 @@ NULL
         addReplyLongLong(c,estimateObjectIdleTime(o)/1000);
     } else if (!strcasecmp(szFromObj(c->argv[1]),"freq") && c->argc == 3) {
         if ((o = objectCommandLookupOrReply(c,c->argv[2],shared.null[c->resp]))
-                == NULL) return;
+                == nullptr) return;
         if (!(g_pserver->maxmemory_policy & MAXMEMORY_FLAG_LFU)) {
             addReplyError(c,"An LFU maxmemory policy is not selected, access frequency not tracked. Please note that when switching between policies at runtime LRU and LFU data will take some time to adjust.");
             return;
@@ -1301,10 +1298,10 @@ NULL
          * in case of the key has not been accessed for a long time,
          * because we update the access time only
          * when the key is read or overwritten. */
-        addReplyLongLong(c,LFUDecrAndReturn(o));
+        addReplyLongLong(c,LFUDecrAndReturn(o.unsafe_robjcast()));
     } else if (!strcasecmp(szFromObj(c->argv[1]), "lastmodified") && c->argc == 3) {
         if ((o = objectCommandLookupOrReply(c,c->argv[2],shared.null[c->resp]))
-                == NULL) return;
+                == nullptr) return;
         addReplyLongLong(c, (g_pserver->mstime - (o->mvcc_tstamp >> MVCC_MS_SHIFT)) / 1000);
     } else {
         addReplySubcommandSyntaxError(c);
@@ -1327,7 +1324,6 @@ NULL
         };
         addReplyHelp(c, help);
     } else if (!strcasecmp(szFromObj(c->argv[1]),"usage") && c->argc >= 3) {
-        dictEntry *de;
         long long samples = OBJ_COMPUTE_SIZE_DEF_SAMPLES;
         for (int j = 3; j < c->argc; j++) {
             if (!strcasecmp(szFromObj(c->argv[j]),"samples") &&
@@ -1346,12 +1342,14 @@ NULL
                 return;
             }
         }
-        if ((de = dictFind(c->db->pdict,ptrFromObj(c->argv[2]))) == NULL) {
+
+        auto itr = c->db->find(c->argv[2]);
+        if (itr == nullptr) {
             addReplyNull(c, shared.nullbulk);
             return;
         }
-        size_t usage = objectComputeSize((robj*)dictGetVal(de),samples);
-        usage += sdsAllocSize((sds)dictGetKey(de));
+        size_t usage = objectComputeSize(itr.val(),samples);
+        usage += sdsAllocSize(itr.key());
         usage += sizeof(dictEntry);
         addReplyLongLong(c,usage);
     } else if (!strcasecmp(szFromObj(c->argv[1]),"stats") && c->argc == 2) {
@@ -1495,3 +1493,110 @@ void redisObject::setrefcount(unsigned ref)
     serverAssert(!FExpires());
     refcount.store(ref, std::memory_order_relaxed); 
 }
+
+sds serializeStoredStringObject(sds str, robj_roptr o)
+{
+    str = sdscatlen(str, &(*o), sizeof(robj));
+    switch (o->encoding)
+    {
+    case OBJ_ENCODING_RAW:
+        str = sdscatsds(str, (sds)szFromObj(o));
+        break;
+
+    case OBJ_ENCODING_INT:
+        break;  //nop
+
+    case OBJ_ENCODING_EMBSTR:
+        size_t cb = zmalloc_size(o.unsafe_robjcast());
+        if (cb > sizeof(robj))
+        {
+            str = sdscatlen(str, o.unsafe_robjcast() + 1, cb - sizeof(robj));
+        }
+        break;
+    }
+        
+    return str;
+}
+
+robj *deserializeStoredStringObject(const char *data, size_t cb)
+{
+    const robj *oT = (const robj*)data;
+    robj *newObject = nullptr;
+    switch (oT->encoding)
+    {
+    case OBJ_ENCODING_INT:
+        serverAssert(cb == sizeof(robj));
+        [[fallthrough]];
+    case OBJ_ENCODING_EMBSTR:
+        newObject = (robj*)zmalloc(cb, MALLOC_LOCAL);
+        memcpy(newObject, data, cb);
+        newObject->setrefcount(1);
+        return newObject;
+
+    case OBJ_ENCODING_RAW:
+        newObject = (robj*)zmalloc(sizeof(robj), MALLOC_SHARED);
+        memcpy(newObject, data, sizeof(robj));
+        newObject->m_ptr = sdsnewlen(SDS_NOINIT,cb-sizeof(robj));
+        memcpy(newObject->m_ptr, data+sizeof(robj), cb-sizeof(robj));
+        newObject->setrefcount(1);
+        return newObject;
+    }
+    serverPanic("Unknown string object encoding from storage");
+    return nullptr;
+}
+
+robj *deserializeStoredObject(const void *data, size_t cb)
+{
+    switch (((char*)data)[0])
+    {
+        case RDB_TYPE_STRING:
+            return deserializeStoredStringObject(((char*)data)+1, cb-1);
+
+        default:
+            rio payload;
+            int type;
+            robj *obj;
+            rioInitWithConstBuffer(&payload,data,cb);
+            if (((type = rdbLoadObjectType(&payload)) == -1) ||
+                ((obj = rdbLoadObject(type,&payload,nullptr, OBJ_MVCC_INVALID)) == nullptr))
+            {
+                serverPanic("Bad data format");
+            }
+            if (rdbLoadType(&payload) == RDB_OPCODE_AUX)
+            {
+                robj *auxkey, *auxval;
+                if ((auxkey = rdbLoadStringObject(&payload)) == NULL) goto eoferr;
+                if ((auxval = rdbLoadStringObject(&payload)) == NULL) {
+                    decrRefCount(auxkey);
+                    goto eoferr;
+                }
+                if (strcasecmp(szFromObj(auxkey), "mvcc-tstamp") == 0) {
+                    obj->mvcc_tstamp = strtoull(szFromObj(auxval), nullptr, 10);
+                }
+                decrRefCount(auxkey);
+                decrRefCount(auxval);
+            }
+        eoferr:
+            return obj;
+    }
+    serverPanic("Unknown object type loading from storage");
+}
+
+sds serializeStoredObject(robj_roptr o)
+{
+    switch (o->type)
+    {
+        case OBJ_STRING:
+        {
+            sds sdsT = sdsnewlen(nullptr, 1);
+            sdsT[0] = RDB_TYPE_STRING;
+            return serializeStoredStringObject(sdsT, o);
+        }
+            
+        default:
+            rio rdb;
+            createDumpPayload(&rdb,o,nullptr);
+            return (sds)rdb.io.buffer.ptr;
+    }
+    serverPanic("Attempting to store unknown object type");
+}

src/rax.c

@@ -705,7 +705,7 @@ int raxGenericInsert(rax *rax, unsigned char *s, size_t len, void *data, void **
             nodesize = sizeof(raxNode)+trimmedlen+raxPadding(trimmedlen)+
                        sizeof(raxNode*);
             if (h->iskey && !h->isnull) nodesize += sizeof(void*);
-            trimmed = rax_malloc(nodesize);
+            trimmed = zcalloc(nodesize, MALLOC_LOCAL);//rax_malloc(nodesize);
         }
 
         if (postfixlen) {

src/rdb-s3.cpp

@@ -6,7 +6,7 @@ extern "C" {
 #include <sys/wait.h>
 
 /* Save the DB on disk. Return C_ERR on error, C_OK on success. */
-int rdbSaveS3(char *s3bucket, rdbSaveInfo *rsi)
+int rdbSaveS3(char *s3bucket, redisDbPersistentData **rgpdb, rdbSaveInfo *rsi)
 {
     int status = EXIT_FAILURE;
     int fd[2];
@@ -40,7 +40,7 @@ int rdbSaveS3(char *s3bucket, rdbSaveInfo *rsi)
             return C_ERR;
         }
 
-        if (rdbSaveFp(fp, rsi) != C_OK)
+        if (rdbSaveFp(fp, rgpdb, rsi) != C_OK)
         {
             fclose(fp);
             return C_ERR;

src/rdb.cpp

@@ -42,6 +42,9 @@
 #include <arpa/inet.h>
 #include <sys/stat.h>
 #include <sys/param.h>
+#include <thread>
+#include <future>
+#include "aelocker.h"
 
 #define rdbExitReportCorruptRDB(...) rdbCheckThenExit(__LINE__,__VA_ARGS__)
 
@@ -1103,21 +1106,21 @@ int rdbSaveInfoAuxFields(rio *rdb, int flags, rdbSaveInfo *rsi) {
     if (rsi) {
         if (rdbSaveAuxFieldStrInt(rdb,"repl-stream-db",rsi->repl_stream_db)
             == -1) return -1;
-        if (rdbSaveAuxFieldStrStr(rdb,"repl-id",g_pserver->replid)
+        if (rdbSaveAuxFieldStrStr(rdb,"repl-id",rsi->repl_id)
             == -1) return -1;
-        if (rdbSaveAuxFieldStrInt(rdb,"repl-offset",g_pserver->master_repl_offset)
+        if (rdbSaveAuxFieldStrInt(rdb,"repl-offset",rsi->master_repl_offset)
             == -1) return -1;
     }
     if (rdbSaveAuxFieldStrInt(rdb,"aof-preamble",aof_preamble) == -1) return -1;
     return 1;
 }
 
-int saveKey(rio *rdb, redisDb *db, int flags, size_t *processed, const char *keystr, robj *o)
+int saveKey(rio *rdb, redisDbPersistentData *db, int flags, size_t *processed, const char *keystr, robj *o)
 {    
     robj key;
 
     initStaticStringObject(key,(char*)keystr);
-    expireEntry *pexpire = getExpire(db, &key);
+    expireEntry *pexpire = db->getExpire(&key);
 
     if (rdbSaveKeyValuePair(rdb,&key,o,pexpire) == -1)
         return 0;
@@ -1142,9 +1145,9 @@ int saveKey(rio *rdb, redisDb *db, int flags, size_t *processed, const char *key
  * When the function returns C_ERR and if 'error' is not NULL, the
  * integer pointed by 'error' is set to the value of errno just after the I/O
  * error. */
-int rdbSaveRio(rio *rdb, int *error, int flags, rdbSaveInfo *rsi) {
-    dictIterator *di = NULL;
+int rdbSaveRio(rio *rdb, redisDbPersistentData **rgpdb, int *error, int flags, rdbSaveInfo *rsi) {
     dictEntry *de;
+    dictIterator *di = NULL;
     char magic[10];
     int j;
     uint64_t cksum;
@@ -1154,13 +1157,12 @@ int rdbSaveRio(rio *rdb, int *error, int flags, rdbSaveInfo *rsi) {
         rdb->update_cksum = rioGenericUpdateChecksum;
     snprintf(magic,sizeof(magic),"REDIS%04d",RDB_VERSION);
     if (rdbWriteRaw(rdb,magic,9) == -1) goto werr;
+    
     if (rdbSaveInfoAuxFields(rdb,flags,rsi) == -1) goto werr;
 
     for (j = 0; j < cserver.dbnum; j++) {
-        redisDb *db = g_pserver->db+j;
-        dict *d = db->pdict;
-        if (dictSize(d) == 0) continue;
-        di = dictGetSafeIterator(d);
+        redisDbPersistentData *db = rgpdb[j];
+        if (db->size() == 0) continue;
 
         /* Write the SELECT DB opcode */
         if (rdbSaveType(rdb,RDB_OPCODE_SELECTDB) == -1) goto werr;
@@ -1171,43 +1173,47 @@ int rdbSaveRio(rio *rdb, int *error, int flags, rdbSaveInfo *rsi) {
          * However this does not limit the actual size of the DB to load since
          * these sizes are just hints to resize the hash tables. */
         uint64_t db_size, expires_size;
-        db_size = dictSize(db->pdict);
-        expires_size = db->setexpire->size();
+        db_size = db->size();
+        expires_size = db->expireSize();
         if (rdbSaveType(rdb,RDB_OPCODE_RESIZEDB) == -1) goto werr;
         if (rdbSaveLen(rdb,db_size) == -1) goto werr;
         if (rdbSaveLen(rdb,expires_size) == -1) goto werr;
         
         /* Iterate this DB writing every entry */
         size_t ckeysExpired = 0;
-        while((de = dictNext(di)) != NULL) {
-            sds keystr = (sds)dictGetKey(de);
-            robj *o = (robj*)dictGetVal(de);
-
+        bool fSavedAll = db->iterate([&](const char *keystr, robj *o)->bool {
             if (o->FExpires())
                 ++ckeysExpired;
             
             if (!saveKey(rdb, db, flags, &processed, keystr, o))
-                goto werr;
-        }
-        serverAssert(ckeysExpired == db->setexpire->size());
-        dictReleaseIterator(di);
-        di = NULL; /* So that we don't release it again on error. */
+                return false;
+            return !g_pserver->rdbThreadVars.fRdbThreadCancel;
+        });
+        if (!fSavedAll)
+            goto werr;
+        serverAssert(ckeysExpired == db->expireSize());
     }
 
     /* If we are storing the replication information on disk, persist
      * the script cache as well: on successful PSYNC after a restart, we need
      * to be able to process any EVALSHA inside the replication backlog the
      * master will send us. */
+    {
+    AeLocker lock;
+    lock.arm(nullptr);
     if (rsi && dictSize(g_pserver->lua_scripts)) {
         di = dictGetIterator(g_pserver->lua_scripts);
         while((de = dictNext(di)) != NULL) {
             robj *body = (robj*)dictGetVal(de);
             if (rdbSaveAuxField(rdb,"lua",3,szFromObj(body),sdslen(szFromObj(body))) == -1)
                 goto werr;
+            if (g_pserver->rdbThreadVars.fRdbThreadCancel)
+                goto werr;
         }
         dictReleaseIterator(di);
         di = NULL; /* So that we don't release it again on error. */
     }
+    }   // AeLocker end scope
 
     /* EOF opcode */
     if (rdbSaveType(rdb,RDB_OPCODE_EOF) == -1) goto werr;
@@ -1233,7 +1239,7 @@ werr:
  * While the suffix is the 40 bytes hex string we announced in the prefix.
  * This way processes receiving the payload can understand when it ends
  * without doing any processing of the content. */
-int rdbSaveRioWithEOFMark(rio *rdb, int *error, rdbSaveInfo *rsi) {
+int rdbSaveRioWithEOFMark(rio *rdb, redisDbPersistentData **rgpdb, int *error, rdbSaveInfo *rsi) {
     char eofmark[RDB_EOF_MARK_SIZE];
 
     getRandomHexChars(eofmark,RDB_EOF_MARK_SIZE);
@@ -1241,7 +1247,7 @@ int rdbSaveRioWithEOFMark(rio *rdb, int *error, rdbSaveInfo *rsi) {
     if (rioWrite(rdb,"$EOF:",5) == 0) goto werr;
     if (rioWrite(rdb,eofmark,RDB_EOF_MARK_SIZE) == 0) goto werr;
     if (rioWrite(rdb,"\r\n",2) == 0) goto werr;
-    if (rdbSaveRio(rdb,error,RDB_SAVE_NONE,rsi) == C_ERR) goto werr;
+    if (rdbSaveRio(rdb,rgpdb,error,RDB_SAVE_NONE,rsi) == C_ERR) goto werr;
     if (rioWrite(rdb,eofmark,RDB_EOF_MARK_SIZE) == 0) goto werr;
     return C_OK;
 
@@ -1251,7 +1257,7 @@ werr: /* Write error. */
     return C_ERR;
 }
 
-int rdbSaveFp(FILE *fp, rdbSaveInfo *rsi)
+int rdbSaveFp(FILE *fp, redisDbPersistentData **rgpdb, rdbSaveInfo *rsi)
 {
     int error = 0;
     rio rdb;
@@ -1261,31 +1267,41 @@ int rdbSaveFp(FILE *fp, rdbSaveInfo *rsi)
     if (g_pserver->rdb_save_incremental_fsync)
         rioSetAutoSync(&rdb,REDIS_AUTOSYNC_BYTES);
 
-    if (rdbSaveRio(&rdb,&error,RDB_SAVE_NONE,rsi) == C_ERR) {
+    if (rdbSaveRio(&rdb,rgpdb,&error,RDB_SAVE_NONE,rsi) == C_ERR) {
         errno = error;
         return C_ERR;
     }
     return C_OK;
 }
 
-int rdbSave(rdbSaveInfo *rsi)
+int rdbSave(redisDbPersistentData **rgpdb, rdbSaveInfo *rsi)
 {
+    std::vector<redisDbPersistentData*> vecdb;
+    if (rgpdb == nullptr)
+    {
+        for (int idb = 0; idb < cserver.dbnum; ++idb)
+        {
+            vecdb.push_back(&g_pserver->db[idb]);
+        }
+        rgpdb = vecdb.data();
+    }
+
     int err = C_OK;
     if (g_pserver->rdb_filename != NULL)
-        err = rdbSaveFile(g_pserver->rdb_filename, rsi);
+        err = rdbSaveFile(g_pserver->rdb_filename, rgpdb, rsi);
 
     if (err == C_OK && g_pserver->rdb_s3bucketpath != NULL)
-        err = rdbSaveS3(g_pserver->rdb_s3bucketpath, rsi);
+        err = rdbSaveS3(g_pserver->rdb_s3bucketpath, rgpdb, rsi);
     return err;
 }
 
 /* Save the DB on disk. Return C_ERR on error, C_OK on success. */
-int rdbSaveFile(char *filename, rdbSaveInfo *rsi) {
+int rdbSaveFile(char *filename, redisDbPersistentData **rgpdb, rdbSaveInfo *rsi) {
     char tmpfile[256];
     char cwd[MAXPATHLEN]; /* Current working dir path for error messages. */
     FILE *fp;
 
-    snprintf(tmpfile,256,"temp-%d.rdb", (int) getpid());
+    snprintf(tmpfile,256,"temp-%d.rdb", g_pserver->rdbThreadVars.tmpfileNum);
     fp = fopen(tmpfile,"w");
     if (!fp) {
         char *cwdp = getcwd(cwd,MAXPATHLEN);
@@ -1298,7 +1314,7 @@ int rdbSaveFile(char *filename, rdbSaveInfo *rsi) {
         return C_ERR;
     }
 
-    if (rdbSaveFp(fp, rsi) == C_ERR){
+    if (rdbSaveFp(fp, rgpdb, rsi) == C_ERR){
         goto werr;
     }
 
@@ -1323,9 +1339,12 @@ int rdbSaveFile(char *filename, rdbSaveInfo *rsi) {
     }
 
     serverLog(LL_NOTICE,"DB saved on disk");
-    g_pserver->dirty = 0;
-    g_pserver->lastsave = time(NULL);
-    g_pserver->lastbgsave_status = C_OK;
+    if (serverTL != nullptr)
+    {
+        g_pserver->dirty = 0;
+        g_pserver->lastsave = time(NULL);
+        g_pserver->lastbgsave_status = C_OK;
+    }
     return C_OK;
 
 werr:
@@ -1335,63 +1354,99 @@ werr:
     return C_ERR;
 }
 
+struct rdbSaveThreadArgs
+{
+    rdbSaveInfo rsi;
+    redisDbPersistentData *rgpdb[1];    // NOTE: Variable Length
+};
+
+void *rdbSaveThread(void *vargs)
+{
+    rdbSaveThreadArgs *args = reinterpret_cast<rdbSaveThreadArgs*>(vargs);
+    serverAssert(serverTL == nullptr);
+    int retval = rdbSave(args->rgpdb, &args->rsi);
+    if (retval == C_OK) {
+        size_t private_dirty = zmalloc_get_private_dirty(-1);
+
+        if (private_dirty) {
+            serverLog(LL_NOTICE,
+                "RDB: %zu MB of memory used by copy-on-write",
+                private_dirty/(1024*1024));
+        }
+
+        g_pserver->child_info_data.cow_size = private_dirty;
+        sendChildInfo(CHILD_INFO_TYPE_RDB);
+    }
+
+    aeAcquireLock();
+    for (int idb = 0; idb < cserver.dbnum; ++idb)
+        g_pserver->db[idb].endSnapshot(args->rgpdb[idb]);
+    aeReleaseLock();
+    zfree(args);
+    return (retval == C_OK) ? (void*)0 : (void*)1;
+}
+
+int launchRdbSaveThread(pthread_t &child, rdbSaveInfo *rsi)
+{
+    rdbSaveThreadArgs *args = (rdbSaveThreadArgs*)zmalloc(sizeof(rdbSaveThreadArgs) + ((cserver.dbnum-1)*sizeof(redisDbPersistentData*)), MALLOC_LOCAL);
+    rdbSaveInfo rsiT = RDB_SAVE_INFO_INIT;
+    if (rsi == nullptr)
+        rsi = &rsiT;
+    memcpy(&args->rsi, rsi, sizeof(rdbSaveInfo));
+    memcpy(&args->rsi.repl_id, g_pserver->replid, sizeof(g_pserver->replid));
+    args->rsi.master_repl_offset = g_pserver->master_repl_offset;
+        
+    for (int idb = 0; idb < cserver.dbnum; ++idb)
+        args->rgpdb[idb] = g_pserver->db[idb].createSnapshot().get();
+
+    g_pserver->rdbThreadVars.tmpfileNum++;
+    g_pserver->rdbThreadVars.fRdbThreadCancel = false;
+    if (pthread_create(&child, NULL, rdbSaveThread, args)) {
+        for (int idb = 0; idb < cserver.dbnum; ++idb)
+            g_pserver->db[idb].endSnapshot(args->rgpdb[idb]);
+        zfree(args);
+        return C_ERR;
+    }
+    return C_OK;
+}
+
 int rdbSaveBackground(rdbSaveInfo *rsi) {
-    pid_t childpid;
+    pthread_t child;
     long long start;
 
-    if (g_pserver->aof_child_pid != -1 || g_pserver->rdb_child_pid != -1) return C_ERR;
+    if (g_pserver->aof_child_pid != -1 || g_pserver->FRdbSaveInProgress()) return C_ERR;
 
     g_pserver->dirty_before_bgsave = g_pserver->dirty;
     g_pserver->lastbgsave_try = time(NULL);
     openChildInfoPipe();
 
     start = ustime();
-    if ((childpid = fork()) == 0) {
-        int retval;
 
-        /* Child */
-        closeListeningSockets(0);
-        redisSetProcTitle("keydb-rdb-bgsave");
-        retval = rdbSave(rsi);
-        if (retval == C_OK) {
-            size_t private_dirty = zmalloc_get_private_dirty(-1);
-
-            if (private_dirty) {
-                serverLog(LL_NOTICE,
-                    "RDB: %zu MB of memory used by copy-on-write",
-                    private_dirty/(1024*1024));
-            }
-
-            g_pserver->child_info_data.cow_size = private_dirty;
-            sendChildInfo(CHILD_INFO_TYPE_RDB);
-        }
-        exitFromChild((retval == C_OK) ? 0 : 1);
-    } else {
-        /* Parent */
-        g_pserver->stat_fork_time = ustime()-start;
-        g_pserver->stat_fork_rate = (double) zmalloc_used_memory() * 1000000 / g_pserver->stat_fork_time / (1024*1024*1024); /* GB per second. */
-        latencyAddSampleIfNeeded("fork",g_pserver->stat_fork_time/1000);
-        if (childpid == -1) {
-            closeChildInfoPipe();
-            g_pserver->lastbgsave_status = C_ERR;
-            serverLog(LL_WARNING,"Can't save in background: fork: %s",
-                strerror(errno));
-            return C_ERR;
-        }
-        serverLog(LL_NOTICE,"Background saving started by pid %d",childpid);
-        g_pserver->rdb_save_time_start = time(NULL);
-        g_pserver->rdb_child_pid = childpid;
-        g_pserver->rdb_child_type = RDB_CHILD_TYPE_DISK;
-        updateDictResizePolicy();
-        return C_OK;
+    
+    g_pserver->stat_fork_time = ustime()-start;
+    g_pserver->stat_fork_rate = (double) zmalloc_used_memory() * 1000000 / g_pserver->stat_fork_time / (1024*1024*1024); /* GB per second. */
+    latencyAddSampleIfNeeded("fork",g_pserver->stat_fork_time/1000);
+    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));
+        return C_ERR;
     }
-    return C_OK; /* unreached */
+    serverLog(LL_NOTICE,"Background saving started");
+    g_pserver->rdb_save_time_start = time(NULL);
+    g_pserver->rdbThreadVars.fRdbThreadActive = true;
+    g_pserver->rdbThreadVars.rdb_child_thread = child;
+    g_pserver->rdb_child_type = RDB_CHILD_TYPE_DISK;
+    updateDictResizePolicy();
+
+    return C_OK;
 }
 
-void rdbRemoveTempFile(pid_t childpid) {
+void rdbRemoveTempFile(int tmpfileNum) {
     char tmpfile[256];
 
-    snprintf(tmpfile,sizeof(tmpfile),"temp-%d.rdb", (int) childpid);
+    snprintf(tmpfile,sizeof(tmpfile),"temp-%d.rdb", tmpfileNum);
     unlink(tmpfile);
 }
 
@@ -1998,7 +2053,7 @@ int rdbLoadRio(rio *rdb, rdbSaveInfo *rsi, int loading_aof) {
                 goto eoferr;
             if ((expires_size = rdbLoadLen(rdb,NULL)) == RDB_LENERR)
                 goto eoferr;
-            dictExpand(db->pdict,db_size);
+            db->expand(db_size);
             continue; /* Read next opcode. */
         } else if (type == RDB_OPCODE_AUX) {
             /* AUX: generic string-string fields. Use to add state to RDB
@@ -2236,7 +2291,7 @@ void backgroundSaveDoneHandlerDisk(int exitcode, int bysignal) {
         serverLog(LL_WARNING,
             "Background saving terminated by signal %d", bysignal);
         latencyStartMonitor(latency);
-        rdbRemoveTempFile(g_pserver->rdb_child_pid);
+        rdbRemoveTempFile(g_pserver->rdbThreadVars.tmpfileNum);
         latencyEndMonitor(latency);
         latencyAddSampleIfNeeded("rdb-unlink-temp-file",latency);
         /* SIGUSR1 is whitelisted, so we have a way to kill a child without
@@ -2244,7 +2299,7 @@ void backgroundSaveDoneHandlerDisk(int exitcode, int bysignal) {
         if (bysignal != SIGUSR1)
             g_pserver->lastbgsave_status = C_ERR;
     }
-    g_pserver->rdb_child_pid = -1;
+    g_pserver->rdbThreadVars.fRdbThreadActive = false;
     g_pserver->rdb_child_type = RDB_CHILD_TYPE_NONE;
     g_pserver->rdb_save_time_last = time(NULL)-g_pserver->rdb_save_time_start;
     g_pserver->rdb_save_time_start = -1;
@@ -2269,7 +2324,7 @@ void backgroundSaveDoneHandlerSocket(int exitcode, int bysignal) {
         serverLog(LL_WARNING,
             "Background transfer terminated by signal %d", bysignal);
     }
-    g_pserver->rdb_child_pid = -1;
+    g_pserver->rdbThreadVars.fRdbThreadActive = false;
     g_pserver->rdb_child_type = RDB_CHILD_TYPE_NONE;
     g_pserver->rdb_save_time_start = -1;
 
@@ -2368,26 +2423,114 @@ void backgroundSaveDoneHandler(int exitcode, int bysignal) {
  * the child did not exit for an error, but because we wanted), and performs
  * the cleanup needed. */
 void killRDBChild(void) {
-    kill(g_pserver->rdb_child_pid,SIGUSR1);
-    rdbRemoveTempFile(g_pserver->rdb_child_pid);
+    g_pserver->rdbThreadVars.fRdbThreadCancel = true;
+    void *rval;
+    pthread_join(g_pserver->rdbThreadVars.rdb_child_thread,&rval);
+    g_pserver->rdbThreadVars.fRdbThreadActive = false;
+    g_pserver->rdbThreadVars.fRdbThreadCancel = false;
+    rdbRemoveTempFile(g_pserver->rdbThreadVars.tmpfileNum);
     closeChildInfoPipe();
     updateDictResizePolicy();
 }
 
+struct rdbSaveSocketThreadArgs
+{
+    rdbSaveInfo rsi;
+    int *fds;
+    int numfds;
+    uint64_t *clientids;
+    redisDbPersistentData *rgpdb[1];
+};
+void *rdbSaveToSlavesSocketsThread(void *vargs)
+{
+    /* Child */
+    serverAssert(serverTL == nullptr);
+    rdbSaveSocketThreadArgs *args = (rdbSaveSocketThreadArgs*)vargs;
+    int retval;
+    rio slave_sockets;
+
+    rioInitWithFdset(&slave_sockets,args->fds,args->numfds);
+    zfree(args->fds);
+    args->fds = nullptr;
+
+    retval = rdbSaveRioWithEOFMark(&slave_sockets,args->rgpdb,NULL,&args->rsi);
+    if (retval == C_OK && rioFlush(&slave_sockets) == 0)
+        retval = C_ERR;
+
+    if (retval == C_OK) {
+        size_t private_dirty = zmalloc_get_private_dirty(-1);
+
+        if (private_dirty) {
+            serverLog(LL_NOTICE,
+                "RDB: %zu MB of memory used by copy-on-write",
+                private_dirty/(1024*1024));
+        }
+
+        g_pserver->child_info_data.cow_size = private_dirty;
+        sendChildInfo(CHILD_INFO_TYPE_RDB);
+
+        /* If we are returning OK, at least one replica was served
+            * with the RDB file as expected, so we need to send a report
+            * to the parent via the pipe. The format of the message is:
+            *
+            * <len> <replica[0].id> <replica[0].error> ...
+            *
+            * len, replica IDs, and replica errors, are all uint64_t integers,
+            * so basically the reply is composed of 64 bits for the len field
+            * plus 2 additional 64 bit integers for each entry, for a total
+            * of 'len' entries.
+            *
+            * The 'id' represents the replica's client ID, so that the master
+            * can match the report with a specific replica, and 'error' is
+            * set to 0 if the replication process terminated with a success
+            * or the error code if an error occurred. */
+        void *msg = zmalloc(sizeof(uint64_t)*(1+2*args->numfds), MALLOC_LOCAL);
+        uint64_t *len = (uint64_t*)msg;
+        uint64_t *ids = len+1;
+        int j, msglen;
+
+        *len = args->numfds;
+        for (j = 0; j < args->numfds; j++) {
+            *ids++ = args->clientids[j];
+            *ids++ = slave_sockets.io.fdset.state[j];
+        }
+
+        /* Write the message to the parent. If we have no good slaves or
+            * we are unable to transfer the message to the parent, we exit
+            * with an error so that the parent will abort the replication
+            * process with all the childre that were waiting. */
+        msglen = sizeof(uint64_t)*(1+2*args->numfds);
+        if (*len == 0 ||
+            write(g_pserver->rdb_pipe_write_result_to_parent,msg,msglen)
+            != msglen)
+        {
+            retval = C_ERR;
+        }
+        zfree(msg);
+    }
+    aeAcquireLock();
+    for (int idb = 0; idb < cserver.dbnum; ++idb)
+        g_pserver->db[idb].endSnapshot(args->rgpdb[idb]);
+    aeReleaseLock();
+    zfree(args->clientids);
+    zfree(args);
+    rioFreeFdset(&slave_sockets);
+
+    return (retval == C_OK) ? (void*)0 : (void*)1;
+}
+
 /* Spawn an RDB child that writes the RDB to the sockets of the slaves
  * that are currently in SLAVE_STATE_WAIT_BGSAVE_START state. */
 int rdbSaveToSlavesSockets(rdbSaveInfo *rsi) {
     serverAssert(GlobalLocksAcquired());
-    int *fds;
-    uint64_t *clientids;
-    int numfds;
     listNode *ln;
     listIter li;
-    pid_t childpid;
+    pthread_t child;
     long long start;
     int pipefds[2];
+    rdbSaveSocketThreadArgs *args = nullptr;
 
-    if (g_pserver->aof_child_pid != -1 || g_pserver->rdb_child_pid != -1) return C_ERR;
+    if (g_pserver->aof_child_pid != -1 || g_pserver->FRdbSaveInProgress()) return C_ERR;
 
     /* Before to fork, create a pipe that will be used in order to
      * send back to the parent the IDs of the slaves that successfully
@@ -2396,22 +2539,27 @@ int rdbSaveToSlavesSockets(rdbSaveInfo *rsi) {
     g_pserver->rdb_pipe_read_result_from_child = pipefds[0];
     g_pserver->rdb_pipe_write_result_to_parent = pipefds[1];
 
+    args = (rdbSaveSocketThreadArgs*)zmalloc(sizeof(rdbSaveSocketThreadArgs) + sizeof(redisDbPersistentData*)*(cserver.dbnum-1), MALLOC_LOCAL);
+
     /* Collect the file descriptors of the slaves we want to transfer
      * the RDB to, which are i WAIT_BGSAVE_START state. */
-    fds = (int*)zmalloc(sizeof(int)*listLength(g_pserver->slaves), MALLOC_LOCAL);
+    args->fds = (int*)zmalloc(sizeof(int)*listLength(g_pserver->slaves), MALLOC_LOCAL);
     /* We also allocate an array of corresponding client IDs. This will
      * be useful for the child process in order to build the report
      * (sent via unix pipe) that will be sent to the parent. */
-    clientids = (uint64_t*)zmalloc(sizeof(uint64_t)*listLength(g_pserver->slaves), MALLOC_LOCAL);
-    numfds = 0;
+    args->clientids = (uint64_t*)zmalloc(sizeof(uint64_t)*listLength(g_pserver->slaves), MALLOC_LOCAL);
+    args->numfds = 0;
+    memcpy(&args->rsi, rsi, sizeof(rdbSaveInfo));
+    memcpy(&args->rsi.repl_id, g_pserver->replid, sizeof(g_pserver->replid));
+    args->rsi.master_repl_offset = g_pserver->master_repl_offset;
 
     listRewind(g_pserver->slaves,&li);
     while((ln = listNext(&li))) {
         client *replica = (client*)ln->value;
 
         if (replica->replstate == SLAVE_STATE_WAIT_BGSAVE_START) {
-            clientids[numfds] = replica->id;
-            fds[numfds++] = replica->fd;
+            args->clientids[args->numfds] = replica->id;
+            args->fds[args->numfds++] = replica->fd;
             replicationSetupSlaveForFullResync(replica,getPsyncInitialOffset());
             /* Put the socket in blocking mode to simplify RDB transfer.
              * We'll restore it when the children returns (since duped socket
@@ -2424,126 +2572,63 @@ int rdbSaveToSlavesSockets(rdbSaveInfo *rsi) {
     /* Create the child process. */
     openChildInfoPipe();
     start = ustime();
-    if ((childpid = fork()) == 0) {
-        /* Child */
-        int retval;
-        rio slave_sockets;
 
-        rioInitWithFdset(&slave_sockets,fds,numfds);
-        zfree(fds);
+    for (int idb = 0; idb < cserver.dbnum; ++idb)
+        args->rgpdb[idb] = g_pserver->db[idb].createSnapshot().get();
 
-        closeListeningSockets(0);
-        redisSetProcTitle("keydb-rdb-to-slaves");
+    g_pserver->rdbThreadVars.tmpfileNum++;
+    g_pserver->rdbThreadVars.fRdbThreadCancel = false;
+    if (pthread_create(&child, nullptr, rdbSaveToSlavesSocketsThread, args)) {
+        serverLog(LL_WARNING,"Can't save in background: fork: %s",
+            strerror(errno));
 
-        retval = rdbSaveRioWithEOFMark(&slave_sockets,NULL,rsi);
-        if (retval == C_OK && rioFlush(&slave_sockets) == 0)
-            retval = C_ERR;
+        /* 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;
+            int j;
 
-        if (retval == C_OK) {
-            size_t private_dirty = zmalloc_get_private_dirty(-1);
-
-            if (private_dirty) {
-                serverLog(LL_NOTICE,
-                    "RDB: %zu MB of memory used by copy-on-write",
-                    private_dirty/(1024*1024));
-            }
-
-            g_pserver->child_info_data.cow_size = private_dirty;
-            sendChildInfo(CHILD_INFO_TYPE_RDB);
-
-            /* If we are returning OK, at least one replica was served
-             * with the RDB file as expected, so we need to send a report
-             * to the parent via the pipe. The format of the message is:
-             *
-             * <len> <replica[0].id> <replica[0].error> ...
-             *
-             * len, replica IDs, and replica errors, are all uint64_t integers,
-             * so basically the reply is composed of 64 bits for the len field
-             * plus 2 additional 64 bit integers for each entry, for a total
-             * of 'len' entries.
-             *
-             * The 'id' represents the replica's client ID, so that the master
-             * can match the report with a specific replica, and 'error' is
-             * set to 0 if the replication process terminated with a success
-             * or the error code if an error occurred. */
-            void *msg = zmalloc(sizeof(uint64_t)*(1+2*numfds), MALLOC_LOCAL);
-            uint64_t *len = (uint64_t*)msg;
-            uint64_t *ids = len+1;
-            int j, msglen;
-
-            *len = numfds;
-            for (j = 0; j < numfds; j++) {
-                *ids++ = clientids[j];
-                *ids++ = slave_sockets.io.fdset.state[j];
-            }
-
-            /* Write the message to the parent. If we have no good slaves or
-             * we are unable to transfer the message to the parent, we exit
-             * with an error so that the parent will abort the replication
-             * process with all the childre that were waiting. */
-            msglen = sizeof(uint64_t)*(1+2*numfds);
-            if (*len == 0 ||
-                write(g_pserver->rdb_pipe_write_result_to_parent,msg,msglen)
-                != msglen)
-            {
-                retval = C_ERR;
-            }
-            zfree(msg);
-        }
-        zfree(clientids);
-        rioFreeFdset(&slave_sockets);
-        exitFromChild((retval == C_OK) ? 0 : 1);
-    } else {
-        /* Parent */
-        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;
-                int j;
-
-                for (j = 0; j < numfds; j++) {
-                    if (replica->id == clientids[j]) {
-                        replica->replstate = SLAVE_STATE_WAIT_BGSAVE_START;
-                        break;
-                    }
+            for (j = 0; j < args->numfds; j++) {
+                if (replica->id == args->clientids[j]) {
+                    replica->replstate = SLAVE_STATE_WAIT_BGSAVE_START;
+                    break;
                 }
             }
-            close(pipefds[0]);
-            close(pipefds[1]);
-            closeChildInfoPipe();
-        } else {
-            g_pserver->stat_fork_time = ustime()-start;
-            g_pserver->stat_fork_rate = (double) zmalloc_used_memory() * 1000000 / g_pserver->stat_fork_time / (1024*1024*1024); /* GB per second. */
-            latencyAddSampleIfNeeded("fork",g_pserver->stat_fork_time/1000);
-
-            serverLog(LL_NOTICE,"Background RDB transfer started by pid %d",
-                childpid);
-            g_pserver->rdb_save_time_start = time(NULL);
-            g_pserver->rdb_child_pid = childpid;
-            g_pserver->rdb_child_type = RDB_CHILD_TYPE_SOCKET;
-            updateDictResizePolicy();
         }
-        zfree(clientids);
-        zfree(fds);
-        return (childpid == -1) ? C_ERR : C_OK;
+        close(pipefds[0]);
+        close(pipefds[1]);
+        closeChildInfoPipe();
+        for (int idb = 0; idb < cserver.dbnum; ++idb)
+            g_pserver->db[idb].endSnapshot(args->rgpdb[idb]);
+        zfree(args->clientids);
+        zfree(args->fds);
+        zfree(args);
+        return C_ERR;
     }
+
+    g_pserver->stat_fork_time = ustime()-start;
+    g_pserver->stat_fork_rate = (double) zmalloc_used_memory() * 1000000 / g_pserver->stat_fork_time / (1024*1024*1024); /* GB per second. */
+    latencyAddSampleIfNeeded("fork",g_pserver->stat_fork_time/1000);
+
+    serverLog(LL_NOTICE,"Background RDB transfer started");
+    g_pserver->rdb_save_time_start = time(NULL);
+    g_pserver->rdbThreadVars.rdb_child_thread = child;
+    g_pserver->rdbThreadVars.fRdbThreadActive = true;
+    g_pserver->rdb_child_type = RDB_CHILD_TYPE_SOCKET;
+    updateDictResizePolicy();
     return C_OK; /* Unreached. */
 }
 
 void saveCommand(client *c) {
-    if (g_pserver->rdb_child_pid != -1) {
+    if (g_pserver->FRdbSaveInProgress()) {
         addReplyError(c,"Background save already in progress");
         return;
     }
     rdbSaveInfo rsi, *rsiptr;
     rsiptr = rdbPopulateSaveInfo(&rsi);
-    if (rdbSave(rsiptr) == C_OK) {
+    if (rdbSave(nullptr, rsiptr) == C_OK) {
         addReply(c,shared.ok);
     } else {
         addReply(c,shared.err);
@@ -2568,7 +2653,7 @@ void bgsaveCommand(client *c) {
     rdbSaveInfo rsi, *rsiptr;
     rsiptr = rdbPopulateSaveInfo(&rsi);
 
-    if (g_pserver->rdb_child_pid != -1) {
+    if (g_pserver->FRdbSaveInProgress()) {
         addReplyError(c,"Background save already in progress");
     } else if (g_pserver->aof_child_pid != -1) {
         if (schedule) {
@@ -2600,6 +2685,9 @@ rdbSaveInfo *rdbPopulateSaveInfo(rdbSaveInfo *rsi) {
     rdbSaveInfo rsi_init = RDB_SAVE_INFO_INIT;
     *rsi = rsi_init;
 
+    memcpy(rsi->repl_id, g_pserver->replid, sizeof(g_pserver->replid));
+    rsi->master_repl_offset = g_pserver->master_repl_offset;
+
     /* If the instance is a master, we can populate the replication info
      * only when repl_backlog is not NULL. If the repl_backlog is NULL,
      * it means that the instance isn't in any replication chains. In this

src/rdb.h

@@ -139,11 +139,11 @@ int rdbLoad(rdbSaveInfo *rsi);
 int rdbLoadFile(const char *filename, rdbSaveInfo *rsi);
 int rdbSaveBackground(rdbSaveInfo *rsi);
 int rdbSaveToSlavesSockets(rdbSaveInfo *rsi);
-void rdbRemoveTempFile(pid_t childpid);
-int rdbSave(rdbSaveInfo *rsi);
-int rdbSaveFile(char *filename, rdbSaveInfo *rsi);
-int rdbSaveFp(FILE *pf, rdbSaveInfo *rsi);
-int rdbSaveS3(char *path, rdbSaveInfo *rsi);
+void rdbRemoveTempFile(int childpid);
+int rdbSave(redisDbPersistentData **rgpdb, rdbSaveInfo *rsi);
+int rdbSaveFile(char *filename, redisDbPersistentData **rgpdb, rdbSaveInfo *rsi);
+int rdbSaveFp(FILE *pf, redisDbPersistentData **rgpdb, rdbSaveInfo *rsi);
+int rdbSaveS3(char *path, redisDbPersistentData **rgpdb, rdbSaveInfo *rsi);
 int rdbLoadS3(char *path, rdbSaveInfo *rsi);
 ssize_t rdbSaveObject(rio *rdb, robj_roptr o, robj *key);
 size_t rdbSavedObjectLen(robj *o);

src/replication.cpp

@@ -858,7 +858,7 @@ void syncCommand(client *c) {
     }
 
     /* CASE 1: BGSAVE is in progress, with disk target. */
-    if (g_pserver->rdb_child_pid != -1 &&
+    if (g_pserver->FRdbSaveInProgress() &&
         g_pserver->rdb_child_type == RDB_CHILD_TYPE_DISK)
     {
         /* Ok a background save is in progress. Let's check if it is a good
@@ -889,7 +889,7 @@ void syncCommand(client *c) {
         }
 
     /* CASE 2: BGSAVE is in progress, with socket target. */
-    } else if (g_pserver->rdb_child_pid != -1 &&
+    } else if (g_pserver->FRdbSaveInProgress() &&
                g_pserver->rdb_child_type == RDB_CHILD_TYPE_SOCKET)
     {
         /* There is an RDB child process but it is writing directly to
@@ -1510,13 +1510,12 @@ void readSyncBulkPayload(aeEventLoop *el, int fd, void *privdata, int mask) {
         int aof_is_enabled = g_pserver->aof_state != AOF_OFF;
 
         /* Ensure background save doesn't overwrite synced data */
-        if (g_pserver->rdb_child_pid != -1) {
+        if (g_pserver->FRdbSaveInProgress()) {
             serverLog(LL_NOTICE,
                 "Replica is about to load the RDB file received from the "
                 "master, but there is a pending RDB child running. "
-                "Killing process %ld and removing its temp file to avoid "
-                "any race",
-                    (long) g_pserver->rdb_child_pid);
+                "Cancelling RDB the save and removing its temp file to avoid "
+                "any race");
             killRDBChild();
         }
 
@@ -3143,7 +3142,7 @@ void replicationCron(void) {
     * In case of diskless replication, we make sure to wait the specified
     * number of seconds (according to configuration) so that other slaves
     * have the time to arrive before we start streaming. */
-    if (g_pserver->rdb_child_pid == -1 && g_pserver->aof_child_pid == -1) {
+    if (!g_pserver->FRdbSaveInProgress() && g_pserver->aof_child_pid == -1) {
         time_t idle, max_idle = 0;
         int slaves_waiting = 0;
         int mincapa = -1;

src/rio.cpp

@@ -73,6 +73,14 @@ static size_t rioBufferRead(rio *r, void *buf, size_t len) {
     return 1;
 }
 
+static size_t rioConstBufferRead(rio *r, void *buf, size_t len) {
+     if (r->io.buffer.len-r->io.buffer.pos < (off_t)len)
+        return 0; /* not enough buffer to return len bytes. */
+    memcpy(buf,r->io.buffer.ptr+r->io.buffer.pos,len);
+    r->io.buffer.pos += len;
+    return 1;
+}
+
 /* Returns read/write position in buffer. */
 static off_t rioBufferTell(rio *r) {
     return r->io.buffer.pos;
@@ -97,12 +105,32 @@ static const rio rioBufferIO = {
     { { NULL, 0 } } /* union for io-specific vars */
 };
 
+static const rio rioConstBufferIO = {
+    rioConstBufferRead,
+    nullptr,
+    rioBufferTell,
+    rioBufferFlush,
+    NULL,           /* update_checksum */
+    0,              /* current checksum */
+    0,              /* bytes read or written */
+    0,              /* read/write chunk size */
+    { { NULL, 0 } } /* union for io-specific vars */
+};
+
 void rioInitWithBuffer(rio *r, sds s) {
     *r = rioBufferIO;
     r->io.buffer.ptr = s;
     r->io.buffer.pos = 0;
 }
 
+void rioInitWithConstBuffer(rio *r, const void *buf, size_t cb)
+{
+    *r = rioConstBufferIO;
+    r->io.buffer.ptr = (sds)buf;
+    r->io.buffer.pos = 0;
+    r->io.buffer.len = cb;
+}
+
 /* --------------------- Stdio file pointer implementation ------------------- */
 
 /* Returns 1 or 0 for success/failure. */

src/rio.h

@@ -70,6 +70,7 @@ struct _rio {
         struct {
             sds ptr;
             off_t pos;
+            off_t len;  // For const buffers only
         } buffer;
         /* Stdio file pointer target. */
         struct {
@@ -130,6 +131,7 @@ static inline int rioFlush(rio *r) {
 
 void rioInitWithFile(rio *r, FILE *fp);
 void rioInitWithBuffer(rio *r, sds s);
+void rioInitWithConstBuffer(rio *r, const void *rgch, size_t cch);
 void rioInitWithFdset(rio *r, int *fds, int numfds);
 
 void rioFreeFdset(rio *r);

src/semiorderedset.h

@@ -93,9 +93,9 @@ public:
         return end();
     }
 
-    setiter end()
+    setiter end() const
     {
-        setiter itr(this);
+        setiter itr(const_cast<semiorderedset<T,T_KEY,MEMMOVE_SAFE>*>(this));
         itr.idxPrimary = m_data.size();
         return itr;
     }

src/server.cpp

@@ -1179,6 +1179,21 @@ int dictSdsKeyCompare(void *privdata, const void *key1,
     return memcmp(key1, key2, l1) == 0;
 }
 
+void dictSdsNOPDestructor(void *, void *) {}
+
+void dictDbKeyDestructor(void *privdata, void *key)
+{
+    redisDbPersistentData *owner = (redisDbPersistentData*)privdata;
+    serverAssert(owner != nullptr);
+    if (owner->m_spdbSnapshot != nullptr)
+    {
+        dictEntry *deSnapshot = dictFind(owner->m_spdbSnapshot->m_pdict, key);
+        if (deSnapshot && (key == dictGetKey(deSnapshot)))
+            return; // don't free, it's now owned by the snapshot
+    }
+    sdsfree((sds)key);
+}
+
 /* A case insensitive version used for the command lookup table and other
  * places where case insensitive non binary-safe comparison is needed. */
 int dictSdsKeyCaseCompare(void *privdata, const void *key1,
@@ -1314,10 +1329,19 @@ dictType dbDictType = {
     NULL,                       /* key dup */
     NULL,                       /* val dup */
     dictSdsKeyCompare,          /* key compare */
-    dictSdsDestructor,          /* key destructor */
+    dictDbKeyDestructor,          /* key destructor */
     dictObjectDestructor   /* val destructor */
 };
 
+dictType dbSnapshotDictType = {
+    dictSdsHash,
+    NULL,
+    NULL,
+    dictSdsKeyCompare,
+    dictSdsNOPDestructor,
+    dictObjectDestructor,
+};
+
 /* g_pserver->lua_scripts sha (as sds string) -> scripts (as robj) cache. */
 dictType shaScriptObjectDictType = {
     dictSdsCaseHash,            /* hash function */
@@ -1439,8 +1463,7 @@ int htNeedsResize(dict *dict) {
 /* If the percentage of used slots in the HT reaches HASHTABLE_MIN_FILL
  * we resize the hash table to save memory */
 void tryResizeHashTables(int dbid) {
-    if (htNeedsResize(g_pserver->db[dbid].pdict))
-        dictResize(g_pserver->db[dbid].pdict);
+    g_pserver->db[dbid].tryResize();
 }
 
 /* Our hash table implementation performs rehashing incrementally while
@@ -1450,10 +1473,10 @@ void tryResizeHashTables(int dbid) {
  *
  * The function returns 1 if some rehashing was performed, otherwise 0
  * is returned. */
-int incrementallyRehash(int dbid) {
+int redisDbPersistentData::incrementallyRehash() {
     /* Keys dictionary */
-    if (dictIsRehashing(g_pserver->db[dbid].pdict)) {
-        dictRehashMilliseconds(g_pserver->db[dbid].pdict,1);
+    if (dictIsRehashing(m_pdict)) {
+        dictRehashMilliseconds(m_pdict,1);
         return 1; /* already used our millisecond for this loop... */
     }
     return 0;
@@ -1466,7 +1489,7 @@ int incrementallyRehash(int dbid) {
  * for dict.c to resize the hash tables accordingly to the fact we have o not
  * running childs. */
 void updateDictResizePolicy(void) {
-    if (g_pserver->rdb_child_pid == -1 && g_pserver->aof_child_pid == -1)
+    if (!g_pserver->FRdbSaveInProgress() && g_pserver->aof_child_pid == -1)
         dictEnableResize();
     else
         dictDisableResize();
@@ -1713,7 +1736,7 @@ void databasesCron(void) {
     /* Perform hash tables rehashing if needed, but only if there are no
      * other processes saving the DB on disk. Otherwise rehashing is bad
      * as will cause a lot of copy-on-write of memory pages. */
-    if (g_pserver->rdb_child_pid == -1 && g_pserver->aof_child_pid == -1) {
+    if (!g_pserver->FRdbSaveInProgress() && g_pserver->aof_child_pid == -1) {
         /* We use global counters so if we stop the computation at a given
          * DB we'll be able to start from the successive in the next
          * cron loop iteration. */
@@ -1734,7 +1757,7 @@ void databasesCron(void) {
         /* Rehash */
         if (g_pserver->activerehashing) {
             for (j = 0; j < dbs_per_call; j++) {
-                int work_done = incrementallyRehash(rehash_db);
+                int work_done = g_pserver->db[rehash_db].incrementallyRehash();
                 if (work_done) {
                     /* If the function did some work, stop here, we'll do
                      * more at the next cron loop. */
@@ -1893,9 +1916,9 @@ int serverCron(struct aeEventLoop *eventLoop, long long id, void *clientData) {
         for (j = 0; j < cserver.dbnum; j++) {
             long long size, used, vkeys;
 
-            size = dictSlots(g_pserver->db[j].pdict);
-            used = dictSize(g_pserver->db[j].pdict);
-            vkeys = g_pserver->db[j].setexpire->size();
+            size = g_pserver->db[j].slots();
+            used = g_pserver->db[j].size();
+            vkeys = g_pserver->db[j].expireSize();
             if (used || vkeys) {
                 serverLog(LL_VERBOSE,"DB %d: %lld keys (%lld volatile) in %lld slots HT.",j,used,vkeys,size);
                 /* dictPrintStats(g_pserver->dict); */
@@ -1922,20 +1945,37 @@ int serverCron(struct aeEventLoop *eventLoop, long long id, void *clientData) {
 
     /* Start a scheduled AOF rewrite if this was requested by the user while
      * a BGSAVE was in progress. */
-    if (g_pserver->rdb_child_pid == -1 && g_pserver->aof_child_pid == -1 &&
+    if (!g_pserver->FRdbSaveInProgress() && g_pserver->aof_child_pid == -1 &&
         g_pserver->aof_rewrite_scheduled)
     {
         rewriteAppendOnlyFileBackground();
     }
 
     /* Check if a background saving or AOF rewrite in progress terminated. */
-    if (g_pserver->rdb_child_pid != -1 || g_pserver->aof_child_pid != -1 ||
+    if (g_pserver->FRdbSaveInProgress() || g_pserver->aof_child_pid != -1 ||
         ldbPendingChildren())
     {
         int statloc;
         pid_t pid;
 
-        if ((pid = wait3(&statloc,WNOHANG,NULL)) != 0) {
+        if (g_pserver->FRdbSaveInProgress())
+        {
+            void *rval = nullptr;
+            if (pthread_tryjoin_np(g_pserver->rdbThreadVars.rdb_child_thread, &rval))
+            {
+                if (errno != EBUSY && errno != EAGAIN)
+                    serverLog(LL_WARNING, "Error joining the background RDB save thread: %s\n", strerror(errno));
+            }
+            else
+            {
+                int exitcode = (int)reinterpret_cast<ptrdiff_t>(rval);
+                backgroundSaveDoneHandler(exitcode,0);
+                if (exitcode == 0) receiveChildInfo();
+                updateDictResizePolicy();
+                closeChildInfoPipe();
+            }
+        }
+        else if ((pid = wait3(&statloc,WNOHANG,NULL)) != 0) {
             int exitcode = WEXITSTATUS(statloc);
             int bysignal = 0;
 
@@ -1943,13 +1983,9 @@ int serverCron(struct aeEventLoop *eventLoop, long long id, void *clientData) {
 
             if (pid == -1) {
                 serverLog(LL_WARNING,"wait3() returned an error: %s. "
-                    "rdb_child_pid = %d, aof_child_pid = %d",
+                    "aof_child_pid = %d",
                     strerror(errno),
-                    (int) g_pserver->rdb_child_pid,
                     (int) g_pserver->aof_child_pid);
-            } else if (pid == g_pserver->rdb_child_pid) {
-                backgroundSaveDoneHandler(exitcode,bysignal);
-                if (!bysignal && exitcode == 0) receiveChildInfo();
             } else if (pid == g_pserver->aof_child_pid) {
                 backgroundRewriteDoneHandler(exitcode,bysignal);
                 if (!bysignal && exitcode == 0) receiveChildInfo();
@@ -1990,7 +2026,7 @@ int serverCron(struct aeEventLoop *eventLoop, long long id, void *clientData) {
 
         /* Trigger an AOF rewrite if needed. */
         if (g_pserver->aof_state == AOF_ON &&
-            g_pserver->rdb_child_pid == -1 &&
+            !g_pserver->FRdbSaveInProgress() &&
             g_pserver->aof_child_pid == -1 &&
             g_pserver->aof_rewrite_perc &&
             g_pserver->aof_current_size > g_pserver->aof_rewrite_min_size)
@@ -2046,7 +2082,7 @@ int serverCron(struct aeEventLoop *eventLoop, long long id, void *clientData) {
      * Note: this code must be after the replicationCron() call above so
      * make sure when refactoring this file to keep this order. This is useful
      * because we want to give priority to RDB savings for replication. */
-    if (g_pserver->rdb_child_pid == -1 && g_pserver->aof_child_pid == -1 &&
+    if (!g_pserver->FRdbSaveInProgress() && g_pserver->aof_child_pid == -1 &&
         g_pserver->rdb_bgsave_scheduled &&
         (g_pserver->unixtime-g_pserver->lastbgsave_try > CONFIG_BGSAVE_RETRY_DELAY ||
          g_pserver->lastbgsave_status == C_OK))
@@ -2932,16 +2968,7 @@ void initServer(void) {
     /* Create the Redis databases, and initialize other internal state. */
     for (int j = 0; j < cserver.dbnum; j++) {
         new (&g_pserver->db[j]) redisDb;
-        g_pserver->db[j].pdict = dictCreate(&dbDictType,NULL);
-        g_pserver->db[j].setexpire = new(MALLOC_LOCAL) expireset();
-        g_pserver->db[j].expireitr = g_pserver->db[j].setexpire->end();
-        g_pserver->db[j].blocking_keys = dictCreate(&keylistDictType,NULL);
-        g_pserver->db[j].ready_keys = dictCreate(&objectKeyPointerValueDictType,NULL);
-        g_pserver->db[j].watched_keys = dictCreate(&keylistDictType,NULL);
-        g_pserver->db[j].id = j;
-        g_pserver->db[j].avg_ttl = 0;
-        g_pserver->db[j].last_expire_set = 0;
-        g_pserver->db[j].defrag_later = listCreate();
+        g_pserver->db[j].initialize(j);
     }
 
     /* Fixup Master Client Database */
@@ -2981,7 +3008,7 @@ void initServer(void) {
     listSetFreeMethod(g_pserver->pubsub_patterns,freePubsubPattern);
     listSetMatchMethod(g_pserver->pubsub_patterns,listMatchPubsubPattern);
     g_pserver->cronloops = 0;
-    g_pserver->rdb_child_pid = -1;
+    g_pserver->rdbThreadVars.fRdbThreadActive = false;
     g_pserver->aof_child_pid = -1;
     g_pserver->rdb_child_type = RDB_CHILD_TYPE_NONE;
     g_pserver->rdb_bgsave_scheduled = 0;
@@ -3707,7 +3734,11 @@ int processCommand(client *c, int callFlags) {
         addReply(c,shared.queued);
     } else {
         std::unique_lock<decltype(c->db->lock)> ulock(c->db->lock);
+        for (int idb = 0; idb < cserver.dbnum; ++idb)
+            g_pserver->db[idb].trackChanges();
         call(c,callFlags);
+        for (int idb = 0; idb < cserver.dbnum; ++idb)
+            g_pserver->db[idb].processChanges();
         c->woff = g_pserver->master_repl_offset;
         if (listLength(g_pserver->ready_keys))
             handleClientsBlockedOnKeys();
@@ -3748,7 +3779,7 @@ int prepareForShutdown(int flags) {
     /* Kill the saving child if there is a background saving in progress.
        We want to avoid race conditions, for instance our saving child may
        overwrite the synchronous saving did by SHUTDOWN. */
-    if (g_pserver->rdb_child_pid != -1) {
+    if (g_pserver->FRdbSaveInProgress()) {
         serverLog(LL_WARNING,"There is a child saving an .rdb. Killing it!");
         killRDBChild();
     }
@@ -3779,7 +3810,7 @@ int prepareForShutdown(int flags) {
         /* Snapshotting. Perform a SYNC SAVE and exit */
         rdbSaveInfo rsi, *rsiptr;
         rsiptr = rdbPopulateSaveInfo(&rsi);
-        if (rdbSave(rsiptr) != C_OK) {
+        if (rdbSave(nullptr, rsiptr) != C_OK) {
             /* Ooops.. error saving! The best we can do is to continue
              * operating. Note that if there was a background saving process,
              * in the next cron() Redis will be notified that the background
@@ -4262,12 +4293,12 @@ sds genRedisInfoString(const char *section) {
             "aof_last_cow_size:%zu\r\n",
             g_pserver->loading,
             g_pserver->dirty,
-            g_pserver->rdb_child_pid != -1,
+            g_pserver->FRdbSaveInProgress(),
             (intmax_t)g_pserver->lastsave,
             (g_pserver->lastbgsave_status == C_OK) ? "ok" : "err",
             (intmax_t)g_pserver->rdb_save_time_last,
-            (intmax_t)((g_pserver->rdb_child_pid == -1) ?
-                -1 : time(NULL)-g_pserver->rdb_save_time_start),
+            (intmax_t)(g_pserver->FRdbSaveInProgress() ?
+                time(NULL)-g_pserver->rdb_save_time_start : -1),
             g_pserver->stat_rdb_cow_bytes,
             g_pserver->aof_state != AOF_OFF,
             g_pserver->aof_child_pid != -1,
@@ -4583,8 +4614,8 @@ sds genRedisInfoString(const char *section) {
         for (j = 0; j < cserver.dbnum; j++) {
             long long keys, vkeys;
 
-            keys = dictSize(g_pserver->db[j].pdict);
-            vkeys = g_pserver->db[j].setexpire->size();
+            keys = g_pserver->db[j].size();
+            vkeys = g_pserver->db[j].expireSize();
 
             // Adjust TTL by the current time
             g_pserver->db[j].avg_ttl -= (g_pserver->mstime - g_pserver->db[j].last_expire_set);
@@ -4763,7 +4794,7 @@ static void sigShutdownHandler(int sig) {
      * on disk. */
     if (g_pserver->shutdown_asap && sig == SIGINT) {
         serverLogFromHandler(LL_WARNING, "You insist... exiting now.");
-        rdbRemoveTempFile(getpid());
+        rdbRemoveTempFile(g_pserver->rdbThreadVars.tmpfileNum);
         exit(1); /* Exit with an error since this was not a clean shutdown. */
     } else if (g_pserver->loading) {
         serverLogFromHandler(LL_WARNING, "Received shutdown signal during loading, exiting now.");

src/server.h

@@ -42,6 +42,7 @@
 #include <stdlib.h>
 #include <cmath>
 #include <string.h>
+#include <string>
 #include <time.h>
 #include <limits.h>
 #include <unistd.h>
@@ -54,6 +55,7 @@
 #include <vector>
 #include <algorithm>
 #include <memory>
+#include <set>
 #ifdef __cplusplus
 extern "C" {
 #include <lua.h>
@@ -89,6 +91,7 @@ typedef long long mstime_t; /* millisecond time type. */
 #include "sha1.h"
 #include "endianconv.h"
 #include "crc64.h"
+#include "IStorage.h"
 
 extern int g_fTestMode;
 
@@ -129,6 +132,11 @@ public:
         return m_ptr;
     }
 
+    const redisObject& operator*() const
+    {
+        return *m_ptr;
+    }
+
     bool operator!() const
     {
         return !m_ptr;
@@ -783,6 +791,16 @@ public:
             : when(when), spsubkey(subkey, sdsfree)
         {}
 
+        subexpireEntry(const subexpireEntry &e)
+            : when(e.when), spsubkey(nullptr, sdsfree)
+        {
+            if (e.spsubkey)
+                spsubkey = std::unique_ptr<const char, void(*)(const char*)>((const char*)sdsdup((sds)e.spsubkey.get()), sdsfree);
+        }
+
+        subexpireEntry(subexpireEntry &&e) = default;
+        subexpireEntry& operator=(subexpireEntry &&e) = default;
+
         bool operator<(long long when) const noexcept { return this->when < when; }
         bool operator<(const subexpireEntry &se) { return this->when < se.when; }
     };
@@ -795,6 +813,10 @@ public:
     expireEntryFat(sds keyPrimary)
         : m_keyPrimary(keyPrimary)
         {}
+        
+    expireEntryFat(const expireEntryFat &e) = default;
+    expireEntryFat(expireEntryFat &&e) = default;
+
     long long when() const noexcept { return m_vecexpireEntries.front().when; }
     const char *key() const noexcept { return m_keyPrimary; }
 
@@ -896,6 +918,14 @@ public:
         m_when = LLONG_MIN;
     }
 
+    expireEntry(const expireEntry &e)
+    {
+        u.m_key = e.u.m_key;
+        m_when = e.m_when;
+        if (e.FFat())
+            u.m_pfatentry = new (MALLOC_LOCAL) expireEntryFat(*e.u.m_pfatentry);
+    }
+
     expireEntry(expireEntry &&e)
     {
         u.m_key = e.u.m_key;
@@ -1039,17 +1069,183 @@ typedef struct clientReplyBlock {
 #endif
 } clientReplyBlock;
 
+struct dictEntry;
+class dict_const_iter
+{
+    friend struct redisDb;
+    friend class redisDbPersistentData;
+protected:
+    dictEntry *de;
+public:
+    explicit dict_const_iter(dictEntry *de)
+        : de(de)
+    {}
+
+    const char *key() const { return de ? (const char*)dictGetKey(de) : nullptr; }
+    robj_roptr val() const { return de ? (robj*)dictGetVal(de) : nullptr; }
+    const robj* operator->() const { return de ? (robj*)dictGetVal(de) : nullptr; }
+    operator robj_roptr() const { return de ? (robj*)dictGetVal(de) : nullptr; }
+
+    bool operator==(std::nullptr_t) const { return de == nullptr; }
+    bool operator!=(std::nullptr_t) const { return de != nullptr; }
+    bool operator==(const dict_const_iter &other) { return de == other.de; }
+};
+class dict_iter : public dict_const_iter
+{
+public:
+    explicit dict_iter(dictEntry *de)
+        : dict_const_iter(de)
+    {}
+    sds key() { return de ? (sds)dictGetKey(de) : nullptr; }
+    robj *val() { return de ? (robj*)dictGetVal(de) : nullptr; }
+    robj *operator->() { return de ? (robj*)dictGetVal(de) : nullptr; }
+    operator robj*() const { return de ? (robj*)dictGetVal(de) : nullptr; }
+};
+
+class redisDbPersistentData
+{
+    friend void dictDbKeyDestructor(void *privdata, void *key);
+public:
+    ~redisDbPersistentData();
+
+    static void swap(redisDbPersistentData *db1, redisDbPersistentData *db2);
+
+    size_t slots() const { return dictSlots(m_pdict); }
+    size_t size() const { return dictSize(m_pdict); }
+    void expand(uint64_t slots) { dictExpand(m_pdict, slots); }
+    
+    void trackkey(robj_roptr o)
+    {
+        trackkey(szFromObj(o));
+    }
+
+    void trackkey(const char *key)
+    {
+        if (m_fTrackingChanges && !m_fAllChanged)
+            m_setchanged.insert(std::string(key, sdslen(key)));
+    }
+
+    dict_iter find(const char *key) 
+    {
+        dictEntry *de = dictFind(m_pdict, key);
+        ensure(de);
+        return dict_iter(de);
+    }
+
+    dict_iter find(robj_roptr key)
+    {
+        return find(szFromObj(key));
+    }
+
+    dict_iter random()
+    {
+        dictEntry *de = dictGetRandomKey(m_pdict);
+        ensure(de);
+        return dict_iter(de);
+    }
+
+    const expireEntry &random_expire()
+    {
+        return m_setexpire->random_value();
+    }
+
+    auto findExpire(const char *key)
+    {
+        return m_setexpire->find(key);
+    }
+
+    dict_iter end()  { return dict_iter(nullptr); }
+    dict_const_iter end() const { return dict_const_iter(nullptr); }
+
+    void getStats(char *buf, size_t bufsize) { dictGetStats(buf, bufsize, m_pdict); }
+    void getExpireStats(char *buf, size_t bufsize) { m_setexpire->getstats(buf, bufsize); }
+
+    bool insert(char *k, robj *o);
+    void tryResize();
+    int incrementallyRehash();
+    void updateValue(dict_iter itr, robj *val);
+    bool syncDelete(robj *key);
+    bool asyncDelete(robj *key);
+    size_t expireSize() const { return m_setexpire->size(); }
+    int removeExpire(robj *key, dict_iter itr);
+    int removeSubkeyExpire(robj *key, robj *subkey);
+    void clear(void(callback)(void*));
+    void emptyDbAsync();
+    bool iterate(std::function<bool(const char*, robj*)> fn);
+    void setExpire(robj *key, robj *subkey, long long when);
+    void setExpire(expireEntry &&e);
+    expireEntry *getExpire(robj_roptr key);
+    void initialize();
+
+    void trackChanges() { m_fTrackingChanges++; }
+    void processChanges();
+
+    // This should only be used if you look at the key, we do not fixup
+    //  objects stored elsewhere
+    dict *dictUnsafeKeyOnly() { return m_pdict; }   
+
+    expireset *setexpireUnsafe() { return m_setexpire; }
+    const expireset *setexpire() { return m_setexpire; }
+
+    std::shared_ptr<redisDbPersistentData> createSnapshot();
+    void endSnapshot(const redisDbPersistentData *psnapshot);
+
+private:
+    void ensure(const char *key);
+    void ensure(dictEntry *de);
+    void storeDatabase();
+    void storeKey(const char *key, size_t cchKey, robj *o);
+
+    // Keyspace
+    dict *m_pdict = nullptr;                 /* The keyspace for this DB */
+    int m_fTrackingChanges = 0;     // Note: Stack based
+    bool m_fAllChanged = false;
+    std::set<std::string> m_setchanged;
+    IStorage *m_pstorage = nullptr;
+
+    // Expire
+    expireset *m_setexpire = nullptr;
+
+    std::shared_ptr<redisDbPersistentData> m_spdbSnapshot;
+};
+
 /* Redis database representation. There are multiple databases identified
  * by integers from 0 (the default database) up to the max configured
  * database. The database number is the 'id' field in the structure. */
-typedef struct redisDb {
-    redisDb() 
-        : expireitr(nullptr)
-    {};
-    dict *pdict;                 /* The keyspace for this DB */
-    expireset *setexpire;
-    expireset::setiter expireitr;
+typedef struct redisDb : public redisDbPersistentData 
+{
+    // Legacy C API, Do not add more
+    friend void tryResizeHashTables(int);
+    friend int dbSyncDelete(redisDb *db, robj *key);
+    friend int dbAsyncDelete(redisDb *db, robj *key);
+    friend long long emptyDb(int dbnum, int flags, void(callback)(void*));
+    friend void scanGenericCommand(struct client *c, robj_roptr o, unsigned long cursor);
+    friend int dbSwapDatabases(int id1, int id2);
+    friend int removeExpire(redisDb *db, robj *key);
+    friend void setExpire(struct client *c, redisDb *db, robj *key, robj *subkey, long long when);
+    friend void setExpire(client *c, redisDb *db, robj *key, expireEntry &&e);
+    friend void evictionPoolPopulate(int dbid, redisDb *db, expireset *setexpire, struct evictionPoolEntry *pool);
+    friend void activeDefragCycle(void);
+    friend int freeMemoryIfNeeded(void);
+    friend void activeExpireCycle(int);
+    friend void expireSlaveKeys(void);
 
+    typedef ::dict_const_iter const_iter;
+    typedef ::dict_iter iter;
+
+    redisDb()
+        : expireitr(nullptr)
+    {}
+    void initialize(int id);
+
+    void dbOverwriteCore(redisDb::iter itr, robj *key, robj *val, bool fUpdateMvcc, bool fRemoveExpire);
+
+
+    bool FKeyExpires(const char *key);
+    size_t clear(bool fAsync, void(callback)(void*));
+
+public:
+    expireset::setiter expireitr;
     dict *blocking_keys;        /* Keys with clients waiting for data (BLPOP)*/
     dict *ready_keys;           /* Blocked keys that received a PUSH */
     dict *watched_keys;         /* WATCHED keys for MULTI/EXEC CAS */
@@ -1392,6 +1588,9 @@ typedef struct rdbSaveInfo {
     char repl_id[CONFIG_RUN_ID_SIZE+1];     /* Replication ID. */
     long long repl_offset;                  /* Replication offset. */
     int fForceSetKey;
+
+    /* Used In Save */
+    long long master_repl_offset;
 } rdbSaveInfo;
 
 #define RDB_SAVE_INFO_INIT {-1,0,"000000000000000000000000000000",-1, TRUE}
@@ -1649,7 +1848,13 @@ struct redisServer {
     /* RDB persistence */
     long long dirty;                /* Changes to DB from the last save */
     long long dirty_before_bgsave;  /* Used to restore dirty on failed BGSAVE */
-    pid_t rdb_child_pid;            /* PID of RDB saving child */
+    struct _rdbThreadVars
+    {
+        bool fRdbThreadActive = false;
+        volatile bool fRdbThreadCancel = false;
+        pthread_t rdb_child_thread;     /* PID of RDB saving child */
+        int tmpfileNum = 0;
+    } rdbThreadVars;
     struct saveparam *saveparams;   /* Save points array for RDB */
     int saveparamslen;              /* Number of saving points */
     char *rdb_filename;             /* Name of RDB file */
@@ -1828,6 +2033,8 @@ struct redisServer {
 
     /* System hardware info */
     size_t system_memory_size;  /* Total memory in system as reported by OS */
+
+    bool FRdbSaveInProgress() const { return rdbThreadVars.fRdbThreadActive; }
 };
 
 typedef struct pubsubPattern {
@@ -1933,9 +2140,11 @@ extern dictType zsetDictType;
 extern dictType clusterNodesDictType;
 extern dictType clusterNodesBlackListDictType;
 extern dictType dbDictType;
+extern dictType dbSnapshotDictType;
 extern dictType shaScriptObjectDictType;
 extern double R_Zero, R_PosInf, R_NegInf, R_Nan;
 extern dictType hashDictType;
+extern dictType keylistDictType;
 extern dictType replScriptCacheDictType;
 extern dictType keyptrDictType;
 extern dictType modulesDictType;
@@ -2164,8 +2373,12 @@ const char *strEncoding(int encoding);
 int compareStringObjects(robj *a, robj *b);
 int collateStringObjects(robj *a, robj *b);
 int equalStringObjects(robj *a, robj *b);
-unsigned long long estimateObjectIdleTime(robj *o);
+unsigned long long estimateObjectIdleTime(robj_roptr o);
 void trimStringObjectIfNeeded(robj *o);
+
+robj *deserializeStoredObject(const void *data, size_t cb);
+sds serializeStoredObject(robj_roptr o);
+
 #define sdsEncodedObject(objptr) (objptr->encoding == OBJ_ENCODING_RAW || objptr->encoding == OBJ_ENCODING_EMBSTR)
 
 /* Synchronous I/O with timeout */
@@ -2218,7 +2431,7 @@ int writeCommandsDeniedByDiskError(void);
 
 /* RDB persistence */
 #include "rdb.h"
-int rdbSaveRio(rio *rdb, int *error, int flags, rdbSaveInfo *rsi);
+int rdbSaveRio(rio *rdb, redisDbPersistentData **rgpdb, int *error, int flags, rdbSaveInfo *rsi);
 void killRDBChild(void);
 
 /* AOF persistence */
@@ -2436,11 +2649,9 @@ int rewriteConfig(char *path);
 
 /* db.c -- Keyspace access API */
 int removeExpire(redisDb *db, robj *key);
-int removeExpireCore(redisDb *db, robj *key, dictEntry *de);
 int removeSubkeyExpire(redisDb *db, robj *key, robj *subkey);
 void propagateExpire(redisDb *db, robj *key, int lazy);
 int expireIfNeeded(redisDb *db, robj *key);
-expireEntry *getExpire(redisDb *db, robj_roptr key);
 void setExpire(client *c, redisDb *db, robj *key, robj *subkey, long long when);
 void setExpire(client *c, redisDb *db, robj *key, expireEntry &&entry);
 robj_roptr lookupKeyRead(redisDb *db, robj *key);
@@ -2448,8 +2659,8 @@ robj *lookupKeyWrite(redisDb *db, robj *key);
 robj_roptr lookupKeyReadOrReply(client *c, robj *key, robj *reply);
 robj *lookupKeyWriteOrReply(client *c, robj *key, robj *reply);
 robj_roptr lookupKeyReadWithFlags(redisDb *db, robj *key, int flags);
-robj *objectCommandLookup(client *c, robj *key);
-robj *objectCommandLookupOrReply(client *c, robj *key, robj *reply);
+robj_roptr objectCommandLookup(client *c, robj *key);
+robj_roptr objectCommandLookupOrReply(client *c, robj *key, robj *reply);
 void objectSetLRUOrLFU(robj *val, long long lfu_freq, long long lru_idle,
                        long long lru_clock);
 #define LOOKUP_NONE 0
@@ -2482,7 +2693,6 @@ void slotToKeyAdd(robj *key);
 void slotToKeyDel(robj *key);
 void slotToKeyFlush(void);
 int dbAsyncDelete(redisDb *db, robj *key);
-void emptyDbAsync(redisDb *db);
 void slotToKeyFlushAsync(void);
 size_t lazyfreeGetPendingObjectsCount(void);
 void freeObjAsync(robj *o);
@@ -2506,6 +2716,7 @@ void clusterPropagatePublish(robj *channel, robj *message);
 void migrateCloseTimedoutSockets(void);
 void clusterBeforeSleep(void);
 int clusterSendModuleMessageToTarget(const char *target, uint64_t module_id, uint8_t type, unsigned char *payload, uint32_t len);
+void createDumpPayload(rio *payload, robj_roptr o, robj *key);
 
 /* Sentinel */
 void initSentinelConfig(void);

src/t_set.cpp

@@ -219,9 +219,9 @@ int setTypeRandomElement(robj *setobj, sds *sdsele, int64_t *llele) {
     return setobj->encoding;
 }
 
-int setTypeRandomElement(robj_roptr setobj, const char **sdsele, const int64_t *llele)
+int setTypeRandomElement(robj_roptr setobj, const char **sdsele, int64_t *llele)
 {
-    return setTypeRandomElement(setobj.unsafe_robjcast(), (sds*)sdsele, const_cast<int64_t*>(llele));
+    return setTypeRandomElement(setobj.unsafe_robjcast(), (sds*)sdsele, llele);
 }
 
 unsigned long setTypeSize(robj_roptr subject) {

src/version.h

@@ -1,3 +1,3 @@
-#define KEYDB_REAL_VERSION "5.1.0"
+#define KEYDB_REAL_VERSION "0.0.0"
 extern const char *KEYDB_SET_VERSION;   // Unlike real version, this can be overriden by the config