Merge remote-tracking branch 'origin/unstable' into mm-psync

Backport the time thread code from the enterprise version of KeyDB.

Also fix an issue with clustering where connections are potentially written to after being closed.

pkg/deb/debian/control

@@ -6,6 +6,7 @@ Build-Depends:
  debhelper (>= 9~),
  dpkg-dev (>= 1.17.5),
  systemd,
+ libsystemd-dev <!nocheck>,
  procps <!nocheck>,
  build-essential <!nocheck>,
  tcl <!nocheck>,

pkg/deb/debian/copyright

@@ -1,16 +1,94 @@
 Format: https://www.debian.org/doc/packaging-manuals/copyright-format/1.0/
 Upstream-Contact: John Sully <john@eqalpha.com>
-Upstream-Name: keydb-pro
-Source: https://github.com/JohnSully/KeyDB-Pro
+Upstream-Name: KeyDB
+Source: https://github.com/EQ-Alpha/KeyDB
 
 Files: *
-Copyright: © 2006-2014 Salvatore Sanfilippo <antirez@gmail.com>
-Copyright © 2019, John Sully
-License: Proprietary
+Copyright: 
+ © 2006-2014 Salvatore Sanfilippo <antirez@gmail.com>
+ © 2019-2021 John Sully
+ © 2019-2021 EQ Alpha Technology Ltd.
 
+License: BSD-3-clause
 
-Files: debian/*
-Copyright: © 2009 Chris Lamb <lamby@debian.org>
+Files:
+ src/rio.*
+ src/t_zset.c
+ src/ziplist.h
+ src/intset.*
+ src/redis-check-aof.c
+ deps/hiredis/*
+ deps/linenoise/*
+Copyright:
+ © 2009-2012 Pieter Noordhuis <pcnoordhuis@gmail.com>
+ © 2009-2012 Salvatore Sanfilippo <antirez@gmail.com>
+License: BSD-3-clause
+
+Files:
+ src/lzf.h
+ src/lzfP.h
+ src/lzf_d.c
+ src/lzf_c.c
+Copyright:
+ © 2000-2007 Marc Alexander Lehmann <schmorp@schmorp.de>
+ © 2009-2012 Salvatore Sanfilippo <antirez@gmail.com>
+License: BSD-2-clause
+
+Files: src/setproctitle.c
+Copyright:
+ © 2010 William Ahern
+ © 2013 Salvatore Sanfilippo
+ © 2013 Stam He
+License: BSD-3-clause
+
+Files: src/ae_evport.c
+Copyright: © 2012 Joyent, Inc.
+License: BSD-3-clause
+
+Files: src/ae_kqueue.c
+Copyright: © 2009 Harish Mallipeddi <harish.mallipeddi@gmail.com>
+License: BSD-3-clause
+
+Files: utils/install_server.sh
+Copyright: © 2011 Dvir Volk <dvirsk@gmail.com>
+License: BSD-3-clause
+
+Files: deps/jemalloc/*
+Copyright:
+ © 2002-2012 Jason Evans <jasone@canonware.com>
+ © 2007-2012 Mozilla Foundation
+ © 2009-2012 Facebook, Inc.
+License: BSD-3-clause
+
+Files: src/pqsort.*
+Copyright: © 1992-1993 The Regents of the University of California
+License: BSD-3-clause
+
+Files: deps/lua/*
+Copyright: © 1994-2012 Lua.org, PUC-Ri
+License: MIT
+ Permission is hereby granted, free of charge, to any person obtaining a copy
+ of this software and associated documentation files (the "Software"), to deal
+ in the Software without restriction, including without limitation the rights
+ to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+ copies of the Software, and to permit persons to whom the Software is
+ furnished to do so, subject to the following conditions:
+ .
+ The above copyright notice and this permission notice shall be included in
+ all copies or substantial portions of the Software.
+ .
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
+ THE SOFTWARE.
+
+Files: pkg/deb/debian/*
+Copyright:
+ © 2009 Chris Lamb <lamby@debian.org>
+ © 2020-2021 EQ Alpha Technology Ltd. <support@eqalpha.com>
 License: BSD-3-clause
 
 License: BSD-2-clause

pkg/deb/debian/keydb-tools.install

@@ -5,3 +5,4 @@ src/keydb-check-rdb	/usr/bin
 src/keydb-cli		/usr/bin
 src/keydb-server	/usr/bin
 src/keydb-sentinel	/usr/bin
+src/keydb-diagnostic-tool	/usr/bin

pkg/deb/debian/rules

@@ -3,6 +3,7 @@
 include /usr/share/dpkg/buildflags.mk
 
 export BUILD_TLS=yes
+export USE_SYSTEMD=yes
 export CFLAGS CPPFLAGS LDFLAGS
 export DEB_BUILD_MAINT_OPTIONS = hardening=+all
 export DEB_LDFLAGS_MAINT_APPEND = -ldl -latomic $(LUA_LDFLAGS)

pkg/deb/debian_dh9/control

@@ -6,6 +6,7 @@ Build-Depends:
  debhelper (>= 9~),
  dpkg-dev (>= 1.17.5),
  systemd,
+ libsystemd-dev <!nocheck>,
  procps <!nocheck>,
  build-essential <!nocheck>,
  tcl <!nocheck>,

pkg/deb/debian_dh9/copyright

@@ -1,11 +1,14 @@
 Format: https://www.debian.org/doc/packaging-manuals/copyright-format/1.0/
 Upstream-Contact: John Sully <john@eqalpha.com>
-Upstream-Name: keydb
-Source: https://github.com/JohnSully/KeyDB
+Upstream-Name: KeyDB
+Source: https://github.com/EQ-Alpha/KeyDB
 
 Files: *
-Copyright: © 2006-2014 Salvatore Sanfilippo <antirez@gmail.com>
-Copyright © 2019, John Sully
+Copyright: 
+ © 2006-2014 Salvatore Sanfilippo <antirez@gmail.com>
+ © 2019-2021 John Sully
+ © 2019-2021 EQ Alpha Technology Ltd.
+
 License: BSD-3-clause
 
 Files:
@@ -82,8 +85,10 @@ License: MIT
  OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
  THE SOFTWARE.
 
-Files: debian/*
-Copyright: © 2009 Chris Lamb <lamby@debian.org>
+Files: pkg/deb/debian_dh9/*
+Copyright: 
+ © 2009 Chris Lamb <lamby@debian.org>
+ © 2020-2021 EQ Alpha Technology Ltd. <support@eqalpha.com>
 License: BSD-3-clause
 
 License: BSD-2-clause

pkg/deb/debian_dh9/keydb-tools.install

@@ -5,3 +5,4 @@ src/keydb-check-aof	/usr/bin
 src/keydb-check-rdb	/usr/bin
 src/keydb-cli		/usr/bin
 src/keydb-sentinel	/usr/bin
+src/keydb-diagnostic-tool	/usr/bin

pkg/deb/debian_dh9/rules

@@ -9,6 +9,7 @@ include /usr/share/dpkg/buildflags.mk
 #LUA_LDFLAGS = $(addprefix -llua5.1-,$(LUA_LIBS_DEBIAN)) $(addprefix ../deps/lua/src/,$(LUA_OBJECTS))
 
 export BUILD_TLS=yes
+export USE_SYSTEMD=yes
 export CFLAGS CPPFLAGS LDFLAGS
 export DEB_BUILD_MAINT_OPTIONS = hardening=+all
 export DEB_LDFLAGS_MAINT_APPEND = -ldl -latomic $(LUA_LDFLAGS)

pkg/rpm/generate_rpms.sh

@@ -25,7 +25,13 @@ mkdir -p $DIR/keydb_build/keydb_rpm/var/log/keydb
 
 # move binaries to bin
 rm $DIR/keydb_build/keydb_rpm/usr/bin/*
-cp $DIR/../../src/keydb-* $DIR/keydb_build/keydb_rpm/usr/bin/
+cp $DIR/../../src/keydb-server $DIR/keydb_build/keydb_rpm/usr/bin/
+cp $DIR/../../src/keydb-sentinel $DIR/keydb_build/keydb_rpm/usr/bin/
+cp $DIR/../../src/keydb-cli $DIR/keydb_build/keydb_rpm/usr/bin/
+cp $DIR/../../src/keydb-benchmark $DIR/keydb_build/keydb_rpm/usr/bin/
+cp $DIR/../../src/keydb-check-aof $DIR/keydb_build/keydb_rpm/usr/bin/
+cp $DIR/../../src/keydb-check-rdb $DIR/keydb_build/keydb_rpm/usr/bin/
+cp $DIR/../../src/keydb-diagnostic-tool $DIR/keydb_build/keydb_rpm/usr/bin/
 
 # update spec file with build info
 sed -i '2d' $DIR/keydb_build/keydb.spec

src/acl.cpp

@@ -223,12 +223,18 @@ uint64_t ACLGetCommandCategoryFlagByName(const char *name) {
     return 0; /* No match. */
 }
 
-/* Method for passwords/pattern comparison used for the user->passwords list
+/* Method for pattern comparison used for the user->patterns list
  * so that we can search for items with listSearchKey(). */
 int ACLListMatchSds(void *a, void *b) {
     return sdscmp((sds)a,(sds)b) == 0;
 }
 
+/* Method for password comparison used for the user->passwords list
+ * Like the above, but uses a time independant compare for security reasons */
+int ACLListMatchSdsSecure(void *a, void* b) {
+    return time_independent_strcmp((sds)a,(sds)b) == 0;
+}
+
 /* Method to free list elements from ACL users password/patterns lists. */
 void ACLListFreeSds(const void *item) {
     sdsfree((sds)item);
@@ -253,7 +259,7 @@ user *ACLCreateUser(const char *name, size_t namelen) {
     u->passwords = listCreate();
     u->patterns = listCreate();
     u->channels = listCreate();
-    listSetMatchMethod(u->passwords,ACLListMatchSds);
+    listSetMatchMethod(u->passwords,ACLListMatchSdsSecure);
     listSetFreeMethod(u->passwords,ACLListFreeSds);
     listSetDupMethod(u->passwords,ACLListDupSds);
     listSetMatchMethod(u->patterns,ACLListMatchSds);

src/aof.cpp

@@ -879,7 +879,7 @@ int loadAppendOnlyFile(char *filename) {
     } else {
         /* RDB preamble. Pass loading the RDB functions. */
         rio rdb;
-        rdbSaveInfo rsi = RDB_SAVE_INFO_INIT;
+        rdbSaveInfo rsi;
 
         serverLog(LL_NOTICE,"Reading RDB preamble from AOF file...");
         if (fseek(fp,0,SEEK_SET) == -1) goto readerr;

src/bitops.cpp

@@ -405,6 +405,10 @@ void printBits(unsigned char *p, unsigned long count) {
 #define BITFIELDOP_SET 1
 #define BITFIELDOP_INCRBY 2
 
+/* Make sure we don't bit shift too many bits at a time.
+ * Even this amount is probably way too large. */
+#define BITOP_SHIFT_MAX (1<<30)
+
 /* This helper function used by GETBIT / SETBIT parses the bit offset argument
  * making sure an error is returned if it is negative or if it overflows
  * Redis 512 MB limit for the string value or more (server.proto_max_bulk_len).
@@ -642,6 +646,11 @@ void bitopCommand(client *c) {
             addReplyError(c, "BITOP SHIFT's last parameter must be an integer");
             return;
         }
+        if (shift > BITOP_SHIFT_MAX) {
+            addReplyError(c, "BITOP SHIFT value is too large.");
+            return;
+        }
+
         if (op == BITOP_RSHIFT)
             shift = -shift;
     }
@@ -682,7 +691,7 @@ void bitopCommand(client *c) {
 
     if (fShiftOp)
     {
-        long newlen = (long)maxlen + shift/CHAR_BIT;
+        long long newlen = (long long)maxlen + shift/CHAR_BIT;
         if (shift > 0 && (shift % CHAR_BIT) != 0)
             newlen++;
 
@@ -694,14 +703,14 @@ void bitopCommand(client *c) {
             res = (unsigned char*) sdsnewlen(NULL,newlen);
             if (shift >= 0)
             {   // left shift
-                long byteoffset = shift/CHAR_BIT;
+                long long byteoffset = shift/CHAR_BIT;
                 memset(res, 0, byteoffset);
-                long srcLen = newlen - byteoffset - ((shift % CHAR_BIT) ? 1 : 0);
+                long long srcLen = newlen - byteoffset - ((shift % CHAR_BIT) ? 1 : 0);
 
                 // now the bitshift+copy
                 unsigned bitshift = shift % CHAR_BIT;
                 unsigned char carry = 0;
-                for (long iSrc = 0; iSrc < srcLen; ++iSrc)
+                for (long long iSrc = 0; iSrc < srcLen; ++iSrc)
                 {
                     res[byteoffset+iSrc] = (src[0][iSrc] << bitshift) | carry;
                     carry = src[0][iSrc] >> (CHAR_BIT - bitshift);
@@ -711,14 +720,14 @@ void bitopCommand(client *c) {
             } 
             else 
             {   // right shift
-                long byteoffset = -shift/CHAR_BIT;
+                long long byteoffset = -shift/CHAR_BIT;
                 unsigned bitshift = -shift % CHAR_BIT;
                 if (bitshift)
                     ++byteoffset;
                 res[0] = (src[0][byteoffset] << (CHAR_BIT-bitshift));
                 if (byteoffset > 0)
                     res[0] |= (src[0][byteoffset-1] >> bitshift);
-                for (long idx = 1; idx < newlen; ++idx)
+                for (long long idx = 1; idx < newlen; ++idx)
                 {
                     res[idx] = (src[0][byteoffset+idx] << (CHAR_BIT-bitshift)) | (src[0][byteoffset+idx-1] >> bitshift);
                 }
@@ -800,8 +809,8 @@ void bitopCommand(client *c) {
                     }
                 }
             }
+            #endif
         }
-        #endif
 
         /* j is set to the next byte to process by the previous loop. */
         for (; j < maxlen; j++) {

src/cron.cpp

@@ -51,6 +51,7 @@ void cronCommand(client *c)
         
         if (c->argc < (6 + numkeys)) {
             addReplyError(c, "Missing arguments or numkeys is too big");
+            return;
         }
     }
 

src/module.cpp

@@ -4289,7 +4289,14 @@ RedisModuleCallReply *RM_Call(RedisModuleCtx *ctx, const char *cmdname, const ch
         if (!(flags & REDISMODULE_ARGV_NO_REPLICAS))
             call_flags |= CMD_CALL_PROPAGATE_REPL;
     }
+
+    {
+    aeAcquireLock();
+    std::unique_lock<fastlock> ul(c->lock);
     call(c,call_flags);
+    aeReleaseLock();
+    }
+
     g_pserver->replication_allowed = prev_replication_allowed;
 
     serverAssert((c->flags & CLIENT_BLOCKED) == 0);

src/networking.cpp

@@ -1324,8 +1324,7 @@ void acceptOnThread(connection *conn, int flags, char *cip)
         int res = aePostFunction(g_pserver->rgthreadvar[ielTarget].el, [conn, flags, ielTarget, szT, fBootLoad] {
             connMarshalThread(conn);
             acceptCommonHandler(conn,flags,szT,ielTarget);
-            if (!g_fTestMode && !fBootLoad)
-                rgacceptsInFlight[ielTarget].fetch_sub(1, std::memory_order_relaxed);
+            rgacceptsInFlight[ielTarget].fetch_sub(1, std::memory_order_relaxed);
             zfree(szT);
         });
 
@@ -1333,8 +1332,7 @@ void acceptOnThread(connection *conn, int flags, char *cip)
             return;
         // If res != AE_OK we can still try to accept on the local thread
     }
-    if (!g_fTestMode && !fBootLoad)
-        rgacceptsInFlight[ielTarget].fetch_sub(1, std::memory_order_relaxed);
+    rgacceptsInFlight[ielTarget].fetch_sub(1, std::memory_order_relaxed);
 
     aeAcquireLock();
     acceptCommonHandler(conn,flags,cip,ielCur);
@@ -1944,7 +1942,9 @@ void ProcessPendingAsyncWrites()
                 bool fResult = c->postFunction([](client *c) {
                     c->fPendingAsyncWriteHandler = false;
                     clientInstallWriteHandler(c);
+                    c->lock.unlock();
                     handleClientsWithPendingWrites(c->iel, g_pserver->aof_state);
+                    c->lock.lock();
                 }, false);
 
                 if (!fResult)

src/new.cpp

@@ -27,14 +27,17 @@ void *operator new(std::size_t size, const std::nothrow_t &) noexcept
     return zmalloc(size, MALLOC_LOCAL);
 }
 
+//need to do null checks for delete since the compiler can optimize out null checks in zfree
 void operator delete(void * p) noexcept
 {
-    zfree(p);
+    if (p != nullptr)
+        zfree(p);
 }
 
 void operator delete(void *p, std::size_t) noexcept
 {
-    zfree(p);
+    if (p != nullptr)
+        zfree(p);
 }
 
 #endif

src/rdb.cpp

@@ -1205,6 +1205,40 @@ int rdbSaveInfoAuxFields(rio *rdb, int rdbflags, rdbSaveInfo *rsi) {
             == -1) return -1;
         if (rdbSaveAuxFieldStrInt(rdb,"repl-offset",g_pserver->master_repl_offset)
             == -1) return -1;
+        if (g_pserver->fActiveReplica && listLength(g_pserver->masters) > 0) {
+            sdsstring val = sdsstring(sdsempty());
+            listNode *ln;
+            listIter li;
+            redisMaster* mi;
+            listRewind(g_pserver->masters,&li);
+            while ((ln = listNext(&li)) != NULL) {
+                mi = (redisMaster*)listNodeValue(ln);
+                if (!mi->master) {
+                    // If master client is not available, use info from master struct - better than nothing
+                    serverLog(LL_NOTICE, "saving master %s", mi->master_replid);
+                    if (mi->master_replid[0] == 0) {
+                        // if replid is null, there's no reason to save it
+                        continue;
+                    }
+                    val = val.catfmt("%s:%I:%s:%i;", mi->master_replid,
+                        mi->master_initial_offset,
+                        mi->masterhost,
+                        mi->masterport);
+                }
+                else {
+                    serverLog(LL_NOTICE, "saving master %s", mi->master->replid);
+                    if (mi->master->replid[0] == 0) {
+                        // if replid is null, there's no reason to save it
+                        continue;
+                    }
+                    val = val.catfmt("%s:%I:%s:%i;", mi->master->replid,
+                        mi->master->reploff,
+                        mi->masterhost,
+                        mi->masterport);
+                }
+            }
+            if (rdbSaveAuxFieldStrStr(rdb, "repl-masters",val.get()) == -1) return -1;
+        }
     }
     if (rdbSaveAuxFieldStrInt(rdb,"aof-preamble",aof_preamble) == -1) return -1;
     return 1;
@@ -2653,6 +2687,23 @@ int rdbLoadRio(rio *rdb, int rdbflags, rdbSaveInfo *rsi) {
                     memcpy(rsi->repl_id,ptrFromObj(auxval),CONFIG_RUN_ID_SIZE+1);
                     rsi->repl_id_is_set = 1;
                 }
+            } else if (!strcasecmp(szFromObj(auxkey),"repl-masters")) {
+                if (rsi) {
+                    struct redisMaster mi;
+                    char *masters = szFromObj(auxval);
+                    char *entry = strtok(masters, ":");
+                    while (entry != NULL) {
+                        memcpy(mi.master_replid, entry, sizeof(mi.master_replid));
+                        entry = strtok(NULL, ":");
+                        mi.master_initial_offset = atoi(entry);
+                        entry = strtok(NULL, ":");
+                        mi.masterhost = entry;
+                        entry = strtok(NULL, ";");
+                        mi.masterport = atoi(entry);
+                        entry = strtok(NULL, ":");
+                        rsi->addMaster(mi);
+                    }
+                }
             } else if (!strcasecmp(szFromObj(auxkey),"repl-offset")) {
                 if (rsi) rsi->repl_offset = strtoll(szFromObj(auxval),NULL,10);
             } else if (!strcasecmp(szFromObj(auxkey),"lua")) {
@@ -2817,11 +2868,11 @@ int rdbLoadRio(rio *rdb, int rdbflags, rdbSaveInfo *rsi) {
             bool fExpiredKey = iAmMaster() && !(rdbflags&RDBFLAGS_AOF_PREAMBLE) && expiretime != -1 && expiretime < now;
         
             if (fStaleMvccKey || fExpiredKey) {
-                if (fStaleMvccKey && !fExpiredKey && rsi != nullptr && rsi->mi != nullptr && rsi->mi->staleKeyMap != nullptr && lookupKeyRead(db, &keyobj) == nullptr) {
+                if (fStaleMvccKey && !fExpiredKey && rsi != nullptr && rsi->masters != nullptr && rsi->masters[0].staleKeyMap != nullptr && lookupKeyRead(db, &keyobj) == nullptr) {
                     // We have a key that we've already deleted and is not back in our database.
                     //  We'll need to inform the sending master of the delete if it is also a replica of us
                     robj_sharedptr objKeyDup(createStringObject(key, sdslen(key)));
-                    rsi->mi->staleKeyMap->operator[](db - g_pserver->db).push_back(objKeyDup);                
+                    rsi->masters[0].staleKeyMap->operator[](db - g_pserver->db).push_back(objKeyDup);                
                 }
                 fLastKeyExpired = true;
                 sdsfree(key);
@@ -3246,8 +3297,8 @@ void bgsaveCommand(client *c) {
  * is returned, and the RDB saving will not persist any replication related
  * information. */
 rdbSaveInfo *rdbPopulateSaveInfo(rdbSaveInfo *rsi) {
-    rdbSaveInfo rsi_init = RDB_SAVE_INFO_INIT;
-    *rsi = rsi_init;
+    rdbSaveInfo rsi_init;
+    *rsi = std::move(rsi_init);
 
     /* 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,

src/redis-benchmark.cpp

@@ -1984,7 +1984,6 @@ int main(int argc, const char **argv) {
     } while(config.loop);
     zfree(data);
 
-    zfree(data);
     if (config.redis_config != NULL) freeRedisConfig(config.redis_config);
 
     return 0;

src/replication.cpp

@@ -1987,7 +1987,7 @@ void readSyncBulkPayload(connection *conn) {
     ssize_t nread, readlen, nwritten;
     int use_diskless_load = useDisklessLoad();
     dbBackup *diskless_load_backup = NULL;
-    rdbSaveInfo rsi = RDB_SAVE_INFO_INIT;
+    rdbSaveInfo rsi;
     int empty_db_flags = g_pserver->repl_slave_lazy_flush ? EMPTYDB_ASYNC :
                                                         EMPTYDB_NO_FLAGS;
     off_t left;
@@ -2308,7 +2308,7 @@ void readSyncBulkPayload(connection *conn) {
                 mi->staleKeyMap->clear();
             else
                 mi->staleKeyMap = new (MALLOC_LOCAL) std::map<int, std::vector<robj_sharedptr>>();
-            rsi.mi = mi;
+            rsi.addMaster(*mi);
         }
         if (rdbLoadFile(rdb_filename,&rsi,RDBFLAGS_REPLICATION) != C_OK) {
             serverLog(LL_WARNING,
@@ -2359,7 +2359,7 @@ void readSyncBulkPayload(connection *conn) {
     {
         mergeReplicationId(mi->master->replid);
     }
-    else
+    else if (!g_pserver->fActiveReplica)
     {
         /* After a full resynchroniziation we use the replication ID and
         * offset of the master. The secondary ID / offset are cleared since
@@ -2555,7 +2555,7 @@ int slaveTryPartialResynchronization(redisMaster *mi, connection *conn, int read
          * client structure representing the master into g_pserver->master. */
         mi->master_initial_offset = -1;
 
-        if (mi->cached_master && !g_pserver->fActiveReplica) {
+        if (mi->cached_master) {
             psync_replid = mi->cached_master->replid;
             snprintf(psync_offset,sizeof(psync_offset),"%lld", mi->cached_master->reploff+1);
             serverLog(LL_NOTICE,"Trying a partial resynchronization (request %s:%s).", psync_replid, psync_offset);
@@ -2654,14 +2654,15 @@ int slaveTryPartialResynchronization(redisMaster *mi, connection *conn, int read
                     sizeof(g_pserver->replid2));
                 g_pserver->second_replid_offset = g_pserver->master_repl_offset+1;
 
-                /* Update the cached master ID and our own primary ID to the
-                 * new one. */
-                memcpy(g_pserver->replid,sznew,sizeof(g_pserver->replid));
-                memcpy(mi->cached_master->replid,sznew,sizeof(g_pserver->replid));
+                if (!g_pserver->fActiveReplica) {
+                    /* Update the cached master ID and our own primary ID to the
+                     * new one. */
+                    memcpy(g_pserver->replid,sznew,sizeof(g_pserver->replid));
+                    memcpy(mi->cached_master->replid,sznew,sizeof(g_pserver->replid));
 
-                /* Disconnect all the sub-slaves: they need to be notified. */
-                if (!g_pserver->fActiveReplica)
+                    /* Disconnect all the sub-slaves: they need to be notified. */
                     disconnectSlaves();
+                }
             }
         }
 
@@ -3022,18 +3023,6 @@ retry_connect:
         disconnectSlavesExcept(mi->master_uuid); /* Force our slaves to resync with us as well. */
         freeReplicationBacklog(); /* Don't allow our chained slaves to PSYNC. */
     }
-    else
-    {
-        if (listLength(g_pserver->slaves))
-        {
-            changeReplicationId();
-            clearReplicationId2();
-        }
-        else
-        {
-            freeReplicationBacklog(); /* Don't allow our chained slaves to PSYNC. */
-        }
-    }
 
     /* Fall back to SYNC if needed. Otherwise psync_result == PSYNC_FULLRESYNC
      * and the g_pserver->master_replid and master_initial_offset are
@@ -3679,6 +3668,26 @@ void replicationCacheMasterUsingMyself(redisMaster *mi) {
     mi->master = NULL;
 }
 
+/* This function is called when reloading master info from an RDB in Active Replica mode.
+ * It creates a cached master client using the info contained in the redisMaster struct.
+ *
+ * Assumes that the passed struct contains valid master info. */
+void replicationCacheMasterUsingMaster(redisMaster *mi) {
+    if (mi->cached_master) {
+        freeClient(mi->cached_master);
+    }
+
+    replicationCreateMasterClient(mi, NULL, -1);
+    std::lock_guard<decltype(mi->master->lock)> lock(mi->master->lock);
+
+    memcpy(mi->master->replid, mi->master_replid, sizeof(mi->master_replid));
+    mi->master->reploff = mi->master_initial_offset;
+
+    unlinkClient(mi->master);
+    mi->cached_master = mi->master;
+    mi->master = NULL;
+}
+
 /* Free a cached master, called when there are no longer the conditions for
  * a partial resync on reconnection. */
 void replicationDiscardCachedMaster(redisMaster *mi) {

src/sds.h

@@ -424,6 +424,21 @@ public:
         return *this;
     }
 
+    sdsstring &operator=(sdsstring &&other)
+    {
+        sds tmp = m_str;
+        m_str = other.m_str;
+        other.m_str = tmp;
+        return *this;
+    }
+
+    template<typename... Args>
+    sdsstring catfmt(const char *fmt, Args... args)
+    {
+        m_str = sdscatfmt(m_str, fmt, args...);
+        return *this;
+    }
+
     sds release() {
         sds sdsT = m_str;
         m_str = nullptr;

src/server.cpp

@@ -6300,7 +6300,7 @@ void loadDataFromDisk(void) {
         if (loadAppendOnlyFile(g_pserver->aof_filename) == C_OK)
             serverLog(LL_NOTICE,"DB loaded from append only file: %.3f seconds",(float)(ustime()-start)/1000000);
     } else if (g_pserver->rdb_filename != NULL || g_pserver->rdb_s3bucketpath != NULL) {
-        rdbSaveInfo rsi = RDB_SAVE_INFO_INIT;
+        rdbSaveInfo rsi;
         errno = 0; /* Prevent a stale value from affecting error checking */
         if (rdbLoad(&rsi,RDBFLAGS_NONE) == C_OK) {
             serverLog(LL_NOTICE,"DB loaded from disk: %.3f seconds",
@@ -6328,11 +6328,23 @@ void loadDataFromDisk(void) {
                 while ((ln = listNext(&li)))
                 {
                     redisMaster *mi = (redisMaster*)listNodeValue(ln);
-                    /* If we are a replica, create a cached master from this
-                    * information, in order to allow partial resynchronizations
-                    * with masters. */
-                    replicationCacheMasterUsingMyself(mi);
-                    selectDb(mi->cached_master,rsi.repl_stream_db);
+                    if (g_pserver->fActiveReplica) {
+                        for (size_t i = 0; i < rsi.numMasters(); i++) {
+                            if (!strcmp(mi->masterhost, rsi.masters[i].masterhost) && mi->masterport == rsi.masters[i].masterport) {
+                                memcpy(mi->master_replid, rsi.masters[i].master_replid, sizeof(mi->master_replid));
+                                mi->master_initial_offset = rsi.masters[i].master_initial_offset;
+                                replicationCacheMasterUsingMaster(mi);
+                                serverLog(LL_NOTICE, "Cached master recovered from RDB for %s:%d", mi->masterhost, mi->masterport);
+                            }
+                        }
+                    }
+                    else {
+                        /* If we are a replica, create a cached master from this
+                        * information, in order to allow partial resynchronizations
+                        * with masters. */
+                        replicationCacheMasterUsingMyself(mi);
+                        selectDb(mi->cached_master,rsi.repl_stream_db);
+                    }
                 }
             }
         } else if (errno != ENOENT) {
@@ -6562,7 +6574,11 @@ void *timeThreadMain(void*) {
             }
         }
         updateCachedTime();
+#if defined(__APPLE__)
+        nanosleep(&delay, nullptr);
+#else
         clock_nanosleep(CLOCK_MONOTONIC, 0, &delay, NULL);
+#endif
     }
 }
 
@@ -6898,7 +6914,7 @@ int main(int argc, char **argv) {
             __AFL_INIT();
 #endif
             rio rdb;
-            rdbSaveInfo rsi = RDB_SAVE_INFO_INIT;
+            rdbSaveInfo rsi;
             startLoadingFile(stdin, (char*)"stdin", 0);
             rioInitWithFile(&rdb,stdin);
             rdbLoadRio(&rdb,0,&rsi);

src/server.h

@@ -1389,6 +1389,38 @@ struct redisMemOverhead {
     } *db;
 };
 
+struct redisMaster {
+    char *masteruser;               /* AUTH with this user and masterauth with master */
+    char *masterauth;               /* AUTH with this password with master */
+    char *masterhost;               /* Hostname of master */
+    int masterport;                 /* Port of master */
+    client *cached_master;          /* Cached master to be reused for PSYNC. */
+    client *master;
+    /* The following two fields is where we store master PSYNC replid/offset
+     * while the PSYNC is in progress. At the end we'll copy the fields into
+     * the server->master client structure. */
+    char master_replid[CONFIG_RUN_ID_SIZE+1];  /* Master PSYNC runid. */
+    long long master_initial_offset;           /* Master PSYNC offset. */
+
+    bool isActive = false;
+    int repl_state;          /* Replication status if the instance is a replica */
+    off_t repl_transfer_size; /* Size of RDB to read from master during sync. */
+    off_t repl_transfer_read; /* Amount of RDB read from master during sync. */
+    off_t repl_transfer_last_fsync_off; /* Offset when we fsync-ed last time. */
+    connection *repl_transfer_s;     /* Slave -> Master SYNC socket */
+    int repl_transfer_fd;    /* Slave -> Master SYNC temp file descriptor */
+    char *repl_transfer_tmpfile; /* Slave-> master SYNC temp file name */
+    time_t repl_transfer_lastio; /* Unix time of the latest read, for timeout */
+    time_t repl_down_since; /* Unix time at which link with master went down */
+
+    unsigned char master_uuid[UUID_BINARY_LEN];  /* Used during sync with master, this is our master's UUID */
+                                                /* After we've connected with our master use the UUID in g_pserver->master */
+    uint64_t mvccLastSync;
+    /* During a handshake the server may have stale keys, we track these here to share once a reciprocal connection is made */
+    std::map<int, std::vector<robj_sharedptr>> *staleKeyMap;
+    int ielReplTransfer = -1;
+};
+
 /* This structure can be optionally passed to RDB save/load functions in
  * order to implement additional functionalities, by storing and loading
  * metadata to the RDB file.
@@ -1397,7 +1429,66 @@ struct redisMemOverhead {
  * replication in order to make sure that chained slaves (slaves of slaves)
  * select the correct DB and are able to accept the stream coming from the
  * top-level master. */
-typedef struct rdbSaveInfo {
+class rdbSaveInfo {
+public:
+    rdbSaveInfo() {
+        repl_stream_db = -1;
+        repl_id_is_set = 0;
+        memcpy(repl_id, "0000000000000000000000000000000000000000", sizeof(repl_id));
+        repl_offset = -1;
+        fForceSetKey = TRUE;
+        mvccMinThreshold = 0;
+        masters = nullptr;
+        masterCount = 0;
+    }
+    rdbSaveInfo(const rdbSaveInfo &other) {
+        repl_stream_db = other.repl_stream_db;
+        repl_id_is_set = other.repl_id_is_set;
+        memcpy(repl_id, other.repl_id, sizeof(repl_id));
+        repl_offset = other.repl_offset;
+        fForceSetKey = other.fForceSetKey;
+        mvccMinThreshold = other.mvccMinThreshold;
+        masters = (struct redisMaster*)malloc(sizeof(struct redisMaster) * other.masterCount);
+        memcpy(masters, other.masters, sizeof(struct redisMaster) * other.masterCount);
+        masterCount = other.masterCount;
+    }
+    rdbSaveInfo(rdbSaveInfo &&other) : rdbSaveInfo() {
+        swap(*this, other);
+    }
+    rdbSaveInfo &operator=(rdbSaveInfo other) {
+        swap(*this, other);
+        return *this;
+    }
+    ~rdbSaveInfo() {
+        free(masters);
+    }
+    friend void swap(rdbSaveInfo &first, rdbSaveInfo &second) {
+        std::swap(first.repl_stream_db, second.repl_stream_db);
+        std::swap(first.repl_id_is_set, second.repl_id_is_set);
+        std::swap(first.repl_id, second.repl_id);
+        std::swap(first.repl_offset, second.repl_offset);
+        std::swap(first.fForceSetKey, second.fForceSetKey);
+        std::swap(first.mvccMinThreshold, second.mvccMinThreshold);
+        std::swap(first.masters, second.masters);
+        std::swap(first.masterCount, second.masterCount);
+
+    }
+
+    void addMaster(const struct redisMaster &mi) {
+        masterCount++;
+        if (masters == nullptr) {
+            masters = (struct redisMaster*)malloc(sizeof(struct redisMaster));
+        }
+        else {
+            masters = (struct redisMaster*)realloc(masters, sizeof(struct redisMaster) * masterCount);
+        }
+        memcpy(masters + masterCount - 1, &mi, sizeof(struct redisMaster));
+    }
+
+    size_t numMasters() {
+        return masterCount;
+    }
+
     /* Used saving and loading. */
     int repl_stream_db;  /* DB to select in g_pserver->master client. */
 
@@ -1407,10 +1498,11 @@ typedef struct rdbSaveInfo {
     long long repl_offset;                  /* Replication offset. */
     int fForceSetKey;
     uint64_t mvccMinThreshold;
-    struct redisMaster *mi;
-} rdbSaveInfo;
+    struct redisMaster *masters;
 
-#define RDB_SAVE_INFO_INIT {-1,0,"0000000000000000000000000000000000000000",-1, TRUE, 0, nullptr}
+private:
+    size_t masterCount;
+};
 
 struct malloc_stats {
     size_t zmalloc_used;
@@ -1499,38 +1591,6 @@ struct redisServerThreadVars {
     std::vector<client*> vecclientsProcess;
 };
 
-struct redisMaster {
-    char *masteruser;               /* AUTH with this user and masterauth with master */
-    char *masterauth;               /* AUTH with this password with master */
-    char *masterhost;               /* Hostname of master */
-    int masterport;                 /* Port of master */
-    client *cached_master;          /* Cached master to be reused for PSYNC. */
-    client *master;
-    /* The following two fields is where we store master PSYNC replid/offset
-     * while the PSYNC is in progress. At the end we'll copy the fields into
-     * the server->master client structure. */
-    char master_replid[CONFIG_RUN_ID_SIZE+1];  /* Master PSYNC runid. */
-    long long master_initial_offset;           /* Master PSYNC offset. */
-
-    bool isActive = false;
-    int repl_state;          /* Replication status if the instance is a replica */
-    off_t repl_transfer_size; /* Size of RDB to read from master during sync. */
-    off_t repl_transfer_read; /* Amount of RDB read from master during sync. */
-    off_t repl_transfer_last_fsync_off; /* Offset when we fsync-ed last time. */
-    connection *repl_transfer_s;     /* Slave -> Master SYNC socket */
-    int repl_transfer_fd;    /* Slave -> Master SYNC temp file descriptor */
-    char *repl_transfer_tmpfile; /* Slave-> master SYNC temp file name */
-    time_t repl_transfer_lastio; /* Unix time of the latest read, for timeout */
-    time_t repl_down_since; /* Unix time at which link with master went down */
-
-    unsigned char master_uuid[UUID_BINARY_LEN];  /* Used during sync with master, this is our master's UUID */
-                                                /* After we've connected with our master use the UUID in g_pserver->master */
-    uint64_t mvccLastSync;
-    /* During a handshake the server may have stale keys, we track these here to share once a reciprocal connection is made */
-    std::map<int, std::vector<robj_sharedptr>> *staleKeyMap;
-    int ielReplTransfer = -1;
-};
-
 // Const vars are not changed after worker threads are launched
 struct redisServerConst {
     pid_t pid;                  /* Main process pid. */
@@ -2441,6 +2501,7 @@ void clearReplicationId2(void);
 void mergeReplicationId(const char *);
 void chopReplicationBacklog(void);
 void replicationCacheMasterUsingMyself(struct redisMaster *mi);
+void replicationCacheMasterUsingMaster(struct redisMaster *mi);
 void feedReplicationBacklog(const void *ptr, size_t len);
 void updateMasterAuth();
 void showLatestBacklog();