Merge branch 'unstable' into RELEASE_5

.vscode/settings.json

@@ -1,58 +0,0 @@
-{
-    "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/Makefile

@@ -82,6 +82,10 @@ ifneq (,$(findstring armv,$(uname_M)))
 endif
 endif
 
+ifneq (,$(findstring armv,$(uname_M)))
+	FINAL_LIBS+=-latomic
+endif	
+
 # Backwards compatibility for selecting an allocator
 ifeq ($(USE_TCMALLOC),yes)
 	MALLOC=tcmalloc
@@ -105,7 +109,7 @@ endif
 FINAL_CFLAGS=$(STD) $(WARN) $(OPT) $(DEBUG) $(CFLAGS) $(REDIS_CFLAGS)
 FINAL_CXXFLAGS=$(CXX_STD) $(WARN) $(OPT) $(DEBUG) $(CXXFLAGS) $(REDIS_CFLAGS)
 FINAL_LDFLAGS=$(LDFLAGS) $(REDIS_LDFLAGS) $(DEBUG)
-FINAL_LIBS=-lm -lcurl
+FINAL_LIBS+=-lm -lcurl
 DEBUG=-g -ggdb
 
 ifeq ($(uname_S),SunOS)

src/expire.cpp

@@ -135,6 +135,15 @@ void activeExpireCycleExpire(redisDb *db, expireEntry &e, long long now) {
 
     if (deleted)
     {
+        if (!pfat->FEmpty())
+        {
+            // We need to resort the expire entry since it may no longer be in the correct position
+            auto itr = db->setexpire->find(e.key());
+            expireEntry eT = std::move(e);
+            db->setexpire->erase(itr);
+            db->setexpire->insert(eT);
+        }
+
         robj objT;
         switch (val->type)
         {

src/fastlock.cpp

@@ -27,6 +27,7 @@
  * POSSIBILITY OF SUCH DAMAGE.
  */
 
+#include "fmacros.h"
 #include "fastlock.h"
 #include <unistd.h>
 #include <sys/syscall.h>
@@ -43,6 +44,7 @@
 #include <string.h>
 #include <stdarg.h>
 #include <stdio.h>
+#include "config.h"
 
 #ifdef __APPLE__
 #include <TargetConditionals.h>
@@ -60,6 +62,11 @@
 #define UNUSED(x) ((void)x)
 #endif
 
+#ifdef HAVE_BACKTRACE
+#include <ucontext.h>
+__attribute__((weak)) void logStackTrace(ucontext_t *) {}
+#endif
+
 extern int g_fInCrash;
 
 /****************************************************
@@ -149,6 +156,43 @@ __attribute__((weak)) void serverLog(int , const char *fmt, ...)
     printf("\n");
 }
 
+extern "C" pid_t gettid()
+{
+    static thread_local int pidCache = -1;
+#ifdef __linux__
+    if (pidCache == -1)
+        pidCache = syscall(SYS_gettid);
+#else
+	if (pidCache == -1) {
+		uint64_t tidT;
+		pthread_threadid_np(nullptr, &tidT);
+		assert(tidT < UINT_MAX);
+		pidCache = (int)tidT;
+	}
+#endif
+    return pidCache;
+}
+
+void printTrace()
+{
+#ifdef HAVE_BACKTRACE
+    serverLog(3 /*LL_WARNING*/, "printing backtrace for thread %d", gettid());
+    ucontext_t ctxt;
+    getcontext(&ctxt);
+    logStackTrace(&ctxt);
+#endif
+}
+
+
+#ifdef __linux__
+static int futex(volatile unsigned *uaddr, int futex_op, int val,
+    const struct timespec *timeout, int val3)
+{
+    return syscall(SYS_futex, uaddr, futex_op, val,
+                    timeout, uaddr, val3);
+}
+#endif
+
 class DeadlockDetector
 {
     std::map<pid_t, fastlock *> m_mapwait;
@@ -156,9 +200,19 @@ class DeadlockDetector
 public:
     void registerwait(fastlock *lock, pid_t thispid)
     {
+        static volatile bool fInDeadlock = false;
+
         if (lock == &m_lock || g_fInCrash)
             return;
         fastlock_lock(&m_lock);
+        
+        if (fInDeadlock)
+        {
+            printTrace();
+            fastlock_unlock(&m_lock);
+            return;
+        }
+
         m_mapwait.insert(std::make_pair(thispid, lock));
 
         // Detect cycles
@@ -184,6 +238,19 @@ public:
                     if (pidCheck == thispid)
                         break;
                 }
+                // Wake All sleeping threads so they can print their callstacks
+#ifdef HAVE_BACKTRACE
+#ifdef __linux__
+                int mask = -1;
+                fInDeadlock = true;
+                fastlock_unlock(&m_lock);
+                futex(&lock->m_ticket.u, FUTEX_WAKE_BITSET_PRIVATE, INT_MAX, nullptr, mask);
+                futex(&itr->second->m_ticket.u, FUTEX_WAKE_BITSET_PRIVATE, INT_MAX, nullptr, mask);
+                sleep(2);
+                fastlock_lock(&m_lock);
+                printTrace();
+#endif
+#endif
                 serverLog(3 /*LL_WARNING*/, "!!! KeyDB Will Now Crash !!!");
                 _serverPanic(__FILE__, __LINE__, "Deadlock detected");
             }
@@ -222,32 +289,6 @@ uint64_t fastlock_getlongwaitcount()
     return rval;
 }
 
-#ifdef __linux__
-static int futex(volatile unsigned *uaddr, int futex_op, int val,
-    const struct timespec *timeout, int val3)
-{
-    return syscall(SYS_futex, uaddr, futex_op, val,
-                    timeout, uaddr, val3);
-}
-#endif
-
-extern "C" pid_t gettid()
-{
-    static thread_local int pidCache = -1;
-#ifdef __linux__
-    if (pidCache == -1)
-        pidCache = syscall(SYS_gettid);
-#else
-	if (pidCache == -1) {
-		uint64_t tidT;
-		pthread_threadid_np(nullptr, &tidT);
-		assert(tidT < UINT_MAX);
-		pidCache = (int)tidT;
-	}
-#endif
-    return pidCache;
-}
-
 extern "C" void fastlock_sleep(fastlock *lock, pid_t pid, unsigned wake, unsigned mask)
 {
 #ifdef __linux__

src/networking.cpp

@@ -496,7 +496,7 @@ void addReplyErrorLengthCore(client *c, const char *s, size_t len, bool fAsync)
 
         if (c->querybuf && sdslen(c->querybuf)) {
             std::string str = escapeString(c->querybuf);
-            serverLog(LL_WARNING, "\tquerybuf: %s", str.c_str());
+            printf("\tquerybuf: %s\n", str.c_str());
         }
         c->master_error = 1;
     }
@@ -1633,7 +1633,10 @@ void sendReplyToClient(aeEventLoop *el, int fd, void *privdata, int mask) {
         c->lock.lock();
         ae.arm(c);
         if (c->flags & CLIENT_CLOSE_ASAP)
-            freeClient(c);
+        {
+            if (!freeClient(c))
+                c->lock.unlock();
+        }
     }
 }
 
@@ -3104,12 +3107,23 @@ int processEventsWhileBlocked(int iel) {
     int iterations = 4; /* See the function top-comment. */
     int count = 0;
 
-    client *c = serverTL->current_client;
-    if (c != nullptr)
+    std::vector<client*> vecclients;
+    listIter li;
+    listNode *ln;
+    listRewind(g_pserver->clients, &li);
+
+    // All client locks must be acquired *after* the global lock is reacquired to prevent deadlocks
+    //  so unlock here, and save them for reacquisition later
+    while ((ln = listNext(&li)) != nullptr)
     {
-        serverAssert(c->flags & CLIENT_PROTECTED);
-        c->lock.unlock();
+        client *c = (client*)listNodeValue(ln);
+        if (c->lock.fOwnLock()) {
+            serverAssert(c->flags & CLIENT_PROTECTED);  // If the client is not protected we have no gurantee they won't be free'd in the event loop
+            c->lock.unlock();
+            vecclients.push_back(c);
+        }
     }
+    
 
     aeReleaseLock();
     try
@@ -3126,18 +3140,18 @@ int processEventsWhileBlocked(int iel) {
     {
         // Caller expects us to be locked so fix and rethrow
         AeLocker locker;
-        if (c != nullptr)
-            c->lock.lock();
-        locker.arm(c);
+        locker.arm(nullptr);
         locker.release();
+        for (client *c : vecclients)
+            c->lock.lock();
         throw;
     }
     
     AeLocker locker;
-    if (c != nullptr)
-        c->lock.lock();
-    locker.arm(c);
+    locker.arm(nullptr);
     locker.release();
+    for (client *c : vecclients)
+        c->lock.lock();
     return count;
 }
 

src/rdb.cpp

@@ -2095,7 +2095,7 @@ int rdbLoadRio(rio *rdb, rdbSaveInfo *rsi, int loading_aof) {
                 incrRefCount(subexpireKey);
             } else if (!strcasecmp(szFromObj(auxkey), "keydb-subexpire-when")) {
                 if (key == nullptr || subexpireKey == nullptr) {
-                    serverLog(LL_WARNING, "Corrupt subexpire entry in RDB skipping.");
+                    serverLog(LL_WARNING, "Corrupt subexpire entry in RDB skipping. key: %s subkey: %s", key != nullptr ? szFromObj(key) : "(null)", subexpireKey != nullptr ? szFromObj(subexpireKey) : "(null)");
                 }
                 else {
                     setExpire(NULL, db, key, subexpireKey, strtoll(szFromObj(auxval), nullptr, 10));
@@ -2187,8 +2187,6 @@ int rdbLoadRio(rio *rdb, rdbSaveInfo *rsi, int loading_aof) {
                 decrRefCount(val);
                 val = nullptr;
             }
-            decrRefCount(key);
-            key = nullptr;
         }
 
         /* Reset the state that is key-specified and is populated by
@@ -2199,7 +2197,10 @@ int rdbLoadRio(rio *rdb, rdbSaveInfo *rsi, int loading_aof) {
     }
 
     if (key != nullptr)
+    {
         decrRefCount(key);
+        key = nullptr;
+    }
 
     if (subexpireKey != nullptr)
     {
@@ -2226,6 +2227,17 @@ int rdbLoadRio(rio *rdb, rdbSaveInfo *rsi, int loading_aof) {
     return C_OK;
 
 eoferr: /* unexpected end of file is handled here with a fatal exit */
+    if (key != nullptr)
+    {
+        decrRefCount(key);
+        key = nullptr;
+    }
+    if (subexpireKey != nullptr)
+    {
+        decrRefCount(subexpireKey);
+        subexpireKey = nullptr;
+    }
+
     serverLog(LL_WARNING,"Short read or OOM loading DB. Unrecoverable error, aborting now.");
     rdbExitReportCorruptRDB("Unexpected EOF reading RDB file");
     return C_ERR; /* Just to avoid warning */

src/replication.cpp

@@ -357,8 +357,8 @@ void replicationFeedSlaves(list *slaves, int dictid, robj **argv, int argc) {
     char szMvcc[128];
     incrementMvccTstamp();
     uint64_t mvccTstamp = getMvccTstamp();
-    int cchMvccNum = snprintf(szMvcc, sizeof(szMvcc), "%lu", mvccTstamp);
-    int cchMvcc = snprintf(szMvcc, sizeof(szMvcc), "$%d\r\n%lu\r\n", cchMvccNum, mvccTstamp);
+    int cchMvccNum = snprintf(szMvcc, sizeof(szMvcc), "%" PRIu64, mvccTstamp);
+    int cchMvcc = snprintf(szMvcc, sizeof(szMvcc), "$%d\r\n%" PRIu64 "\r\n", cchMvccNum, mvccTstamp);
     cchMvcc = std::min<int>(cchMvcc, sizeof(szMvcc));    // tricky snprintf
 
     /* Write the command to the replication backlog if any. */
@@ -2425,8 +2425,6 @@ void freeMasterInfo(redisMaster *mi)
 {
     zfree(mi->masterauth);
     zfree(mi->masteruser);
-    if (mi->clientFake)
-        freeClient(mi->clientFake);
     delete mi->staleKeyMap;
     zfree(mi);
 }
@@ -2484,11 +2482,6 @@ void replicationHandleMasterDisconnection(redisMaster *mi) {
         mi->master = NULL;
         mi->repl_state = REPL_STATE_CONNECT;
         mi->repl_down_since = g_pserver->unixtime;
-        if (mi->clientFake) {
-            freeClient(mi->clientFake);
-            mi->clientFake = nullptr;
-
-        }
         /* We lost connection with our master, don't disconnect slaves yet,
         * maybe we'll be able to PSYNC with our master later. We'll disconnect
         * the slaves only if we'll have to do a full resync with our master. */
@@ -3378,8 +3371,13 @@ bool FInReplicaReplay()
     return s_pstate != nullptr && s_pstate->nesting() > 0;
 }
 
+struct RemoteMasterState
+{
+    uint64_t mvcc = 0;
+    client *cFake = nullptr;
+};
 
-static std::unordered_map<std::string, uint64_t> g_mapmvcc;
+static std::unordered_map<std::string, RemoteMasterState> g_mapremote;
 
 void replicaReplayCommand(client *c)
 {
@@ -3455,12 +3453,15 @@ void replicaReplayCommand(client *c)
     if (!s_pstate->FPush())
         return;
 
-    redisMaster *mi = s_pstate->getMi(c);
-    client *cFake = mi->clientFake;
-    if (mi->clientFakeNesting != s_pstate->nesting())
-        cFake = nullptr;
-    serverAssert(mi != nullptr);
-    if (mvcc != 0 && g_mapmvcc[uuid] >= mvcc)
+    RemoteMasterState &remoteState = g_mapremote[uuid];
+    if (remoteState.cFake == nullptr)
+        remoteState.cFake = createClient(-1, c->iel);
+    else
+        remoteState.cFake->iel = c->iel;
+
+    client *cFake = remoteState.cFake;
+
+    if (mvcc != 0 && remoteState.mvcc >= mvcc)
     {
         s_pstate->Cancel();
         s_pstate->Pop();
@@ -3469,8 +3470,6 @@ void replicaReplayCommand(client *c)
 
     // OK We've recieved a command lets execute
     client *current_clientSave = serverTL->current_client;
-    if (cFake == nullptr)
-        cFake = createClient(-1, c->iel);
     cFake->lock.lock();
     cFake->authenticated = c->authenticated;
     cFake->puser = c->puser;
@@ -3483,13 +3482,15 @@ void replicaReplayCommand(client *c)
     bool fExec = ccmdPrev != serverTL->commandsExecuted;
     cFake->lock.unlock();
     if (cFake->master_error)
-        addReplyError(c, "Error in rreplay command, please check logs");
+    {
+        addReplyError(c, "Error in rreplay command, please check logs.");
+    }
     if (fExec || cFake->flags & CLIENT_MULTI)
     {
         addReply(c, shared.ok);
         selectDb(c, cFake->db->id);
-        if (mvcc > g_mapmvcc[uuid])
-            g_mapmvcc[uuid] = mvcc;
+        if (mvcc > remoteState.mvcc)
+            remoteState.mvcc = mvcc;
     }
     else
     {
@@ -3497,17 +3498,6 @@ void replicaReplayCommand(client *c)
         addReplyError(c, "command did not execute");
     }
     serverAssert(sdslen(cFake->querybuf) == 0);
-    if (cFake->flags & CLIENT_MULTI)
-    {
-        mi->clientFake = cFake;
-        mi->clientFakeNesting = s_pstate->nesting();
-    }
-    else
-    {
-        if (mi->clientFake == cFake)
-            mi->clientFake = nullptr;
-        freeClient(cFake);
-    }
     serverTL->current_client = current_clientSave;
 
     // call() will not propogate this for us, so we do so here

src/server.cpp

@@ -1039,6 +1039,7 @@ struct redisCommand redisCommandTable[] = {
 /* We use a private localtime implementation which is fork-safe. The logging
  * function of Redis may be called from other threads. */
 extern "C" void nolocks_localtime(struct tm *tmp, time_t t, time_t tz, int dst);
+extern "C" pid_t gettid();
 
 /* Low level logging. To use only for very big messages, otherwise
  * serverLog() is to prefer. */
@@ -1076,8 +1077,8 @@ void serverLogRaw(int level, const char *msg) {
         } else {
             role_char = (listLength(g_pserver->masters) ? 'S':'M'); /* Slave or Master. */
         }
-        fprintf(fp,"%d:%c %s %c %s\n",
-            (int)getpid(),role_char, buf,c[level],msg);
+        fprintf(fp,"%d:%d:%c %s %c %s\n",
+            (int)getpid(),(int)gettid(),role_char, buf,c[level],msg);
     }
     fflush(fp);
 
@@ -3592,12 +3593,12 @@ int processCommand(client *c, int callFlags) {
             return C_OK;
         }
     }
-
-    incrementMvccTstamp();
     
     if (!locker.isArmed())
         locker.arm(c);
 
+    incrementMvccTstamp();
+
     /* Handle the maxmemory directive.
      *
      * Note that we do not want to reclaim memory if we are here re-entering
@@ -5017,14 +5018,15 @@ void incrementMvccTstamp()
     msPrev >>= MVCC_MS_SHIFT;  // convert to milliseconds
 
     long long mst;
-    __atomic_load(&g_pserver->mstime, &mst, __ATOMIC_RELAXED);
+    __atomic_load(&g_pserver->mstime, &mst, __ATOMIC_ACQUIRE);
     if (msPrev >= (uint64_t)mst)  // we can be greater if the count overflows
     {
-        atomicIncr(g_pserver->mvcc_tstamp, 1);
+        __atomic_fetch_add(&g_pserver->mvcc_tstamp, 1, __ATOMIC_RELEASE);
     }
     else
     {
-        atomicSet(g_pserver->mvcc_tstamp, ((uint64_t)g_pserver->mstime) << MVCC_MS_SHIFT);
+        uint64_t val = ((uint64_t)g_pserver->mstime) << MVCC_MS_SHIFT;
+        __atomic_store(&g_pserver->mvcc_tstamp, &val, __ATOMIC_RELEASE);
     }
 }
 
@@ -5035,7 +5037,7 @@ void OnTerminate()
         The easiest way to achieve that is to acutally segfault, so we assert
         here.
     */
-    serverAssert(false);
+    serverPanic("std::teminate() called");
 }
 
 void *workerThreadMain(void *parg)

src/server.h

@@ -1541,8 +1541,6 @@ struct redisMaster {
     int masterport;                 /* Port of master */
     client *cached_master;          /* Cached master to be reused for PSYNC. */
     client *master;
-    client *clientFake;
-    int clientFakeNesting;
     /* 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. */

src/version.h

@@ -1,3 +1,3 @@
-#define KEYDB_REAL_VERSION "0.0.0"
+#define KEYDB_REAL_VERSION "5.3.3"
 extern const char *KEYDB_SET_VERSION;   // Unlike real version, this can be overriden by the config
 

tests/unit/expire.tcl

@@ -220,7 +220,7 @@ start_server {tags {"expire"}} {
         assert {$ttl <= 98 && $ttl > 90}
     }
 
-    test { EXPIREMEMBER works (set) } {
+    test {EXPIREMEMBER works (set)} {
         r flushall
         r sadd testkey foo bar baz
         r expiremember testkey foo 1
@@ -228,7 +228,7 @@ start_server {tags {"expire"}} {
         assert_equal {2} [r scard testkey]
     }
 
-    test { EXPIREMEMBER works (hash) } {
+    test {EXPIREMEMBER works (hash)} {
         r flushall
         r hset testkey foo bar
         r expiremember testkey foo 1
@@ -236,7 +236,7 @@ start_server {tags {"expire"}} {
         r exists testkey
     } {0}
 
-    test { EXPIREMEMBER works (zset) } {
+    test {EXPIREMEMBER works (zset)} {
         r flushall
         r zadd testkey 1 foo
         r zadd testkey 2 bar
@@ -246,10 +246,19 @@ start_server {tags {"expire"}} {
         assert_equal {1} [r zcard testkey]
     }
 
-    test { TTL for subkey expires works } {
+    test {TTL for subkey expires works} {
         r flushall
         r sadd testkey foo bar baz
         r expiremember testkey foo 10000
         assert [expr [r ttl testkey foo] > 0]
     }
+
+    test {Roundtrip for subkey expires works} {
+        r flushall
+        r sadd testkey foo bar baz
+        r expiremember testkey foo 10000
+        r save
+        r debug reload
+        assert [expr [r ttl testkey foo] > 0]
+    }
 }