other: merge 3.0

cmake/taostools_CMakeLists.txt.in

@@ -2,7 +2,7 @@
 # taos-tools
 ExternalProject_Add(taos-tools
         GIT_REPOSITORY https://github.com/taosdata/taos-tools.git
-        GIT_TAG cac24d3
+        GIT_TAG 4a4027c
         SOURCE_DIR "${TD_SOURCE_DIR}/tools/taos-tools"
         BINARY_DIR ""
         #BUILD_IN_SOURCE TRUE

include/client/taos.h

@@ -149,7 +149,7 @@ DLL_EXPORT TAOS      *taos_connect(const char *ip, const char *user, const char
 DLL_EXPORT TAOS *taos_connect_auth(const char *ip, const char *user, const char *auth, const char *db, uint16_t port);
 DLL_EXPORT void  taos_close(TAOS *taos);
 
-const char *taos_data_type(int type);
+DLL_EXPORT const char *taos_data_type(int type);
 
 DLL_EXPORT TAOS_STMT *taos_stmt_init(TAOS *taos);
 DLL_EXPORT TAOS_STMT *taos_stmt_init_with_reqid(TAOS *taos, int64_t reqid);

include/common/tmsg.h

@@ -70,6 +70,11 @@ static inline bool vnodeIsMsgBlock(tmsg_t type) {
   return (type == TDMT_VND_CREATE_TABLE) || (type == TDMT_VND_ALTER_TABLE) || (type == TDMT_VND_DROP_TABLE) ||
          (type == TDMT_VND_UPDATE_TAG_VAL);
 }
+
+static inline bool syncUtilUserCommit(tmsg_t msgType) {
+  return msgType != TDMT_SYNC_NOOP && msgType != TDMT_SYNC_LEADER_TRANSFER;
+}
+
 /* ------------------------ OTHER DEFINITIONS ------------------------ */
 // IE type
 #define TSDB_IE_TYPE_SEC         1

include/libs/sync/sync.h

@@ -47,6 +47,7 @@ extern "C" {
 
 #define SYNC_HEARTBEAT_SLOW_MS       1500
 #define SYNC_HEARTBEAT_REPLY_SLOW_MS 1500
+#define SYNC_SNAP_RESEND_MS          1000 * 60
 
 #define SYNC_MAX_BATCH_SIZE 1
 #define SYNC_INDEX_BEGIN    0

include/os/os.h

@@ -27,6 +27,7 @@ extern "C" {
 
 #if !defined(WINDOWS)
 #include <dirent.h>
+#include <execinfo.h>
 #include <libgen.h>
 #include <sched.h>
 #include <unistd.h>

include/os/osDef.h

@@ -120,12 +120,6 @@ void syslog(int unused, const char *format, ...);
 #define POINTER_SHIFT(p, b)      ((void *)((char *)(p) + (b)))
 #define POINTER_DISTANCE(p1, p2) ((char *)(p1) - (char *)(p2))
 
-#ifndef NDEBUG
-#define ASSERT(x) assert(x)
-#else
-#define ASSERT(x)
-#endif
-
 #ifndef UNUSED
 #define UNUSED(x) ((void)(x))
 #endif

include/os/osDir.h

@@ -62,6 +62,7 @@ int32_t taosRealPath(char *dirname, char *realPath, int32_t maxlen);
 bool    taosIsDir(const char *dirname);
 char   *taosDirName(char *dirname);
 char   *taosDirEntryBaseName(char *dirname);
+void    taosGetCwd(char *buf, int32_t len);
 
 TdDirPtr      taosOpenDir(const char *dirname);
 TdDirEntryPtr taosReadDir(TdDirPtr pDir);

include/os/osString.h

@@ -62,7 +62,7 @@ typedef int32_t TdUcs4;
 int32_t taosUcs4len(TdUcs4 *ucs4);
 int64_t taosStr2int64(const char *str);
 
-void    taosConvInit(void);
+int32_t taosConvInit(void);
 void    taosConvDestroy();
 int32_t taosUcs4ToMbs(TdUcs4 *ucs4, int32_t ucs4_max_len, char *mbs);
 bool    taosMbsToUcs4(const char *mbs, size_t mbs_len, TdUcs4 *ucs4, int32_t ucs4_max_len, int32_t *len);

include/os/osSystem.h

@@ -46,6 +46,29 @@ void    taosSetTerminalMode();
 int32_t taosGetOldTerminalMode();
 void    taosResetTerminalMode();
 
+#if !defined(WINDOWS)
+#define taosPrintTrace(flags, level, dflag)                                \
+  {                                                                        \
+    void*   array[100];                                                    \
+    int32_t size = backtrace(array, 100);                                  \
+    char**  strings = backtrace_symbols(array, size);                      \
+    if (strings != NULL) {                                                 \
+      taosPrintLog(flags, level, dflag, "obtained %d stack frames", size); \
+      for (int32_t i = 0; i < size; i++) {                                 \
+        taosPrintLog(flags, level, dflag, "frame:%d, %s", i, strings[i]);  \
+      }                                                                    \
+    }                                                                      \
+                                                                           \
+    taosMemoryFree(strings);                                               \
+  }
+#else
+#define taosPrintTrace(flags, level, dflag)                                                                \
+  {                                                                                                        \
+    taosPrintLog(flags, level, dflag,                                                                      \
+                 "backtrace not implemented on windows, so detailed stack information cannot be printed"); \
+    }
+#endif
+
 #ifdef __cplusplus
 }
 #endif

include/util/tcoding.h

@@ -17,6 +17,7 @@
 #define _TD_UTIL_CODING_H_
 
 #include "os.h"
+#include "tlog.h"
 
 #ifdef __cplusplus
 extern "C" {

include/util/tdef.h

@@ -308,8 +308,9 @@ typedef enum ELogicConditionType {
 #define TSDB_MIN_DURATION_PER_FILE      60  // unit minute
 #define TSDB_MAX_DURATION_PER_FILE      (3650 * 1440)
 #define TSDB_DEFAULT_DURATION_PER_FILE  (10 * 1440)
-#define TSDB_MIN_KEEP                   (1 * 1440)       // data in db to be reserved. unit minute
+#define TSDB_MIN_KEEP                   (1 * 1440)          // data in db to be reserved. unit minute
-#define TSDB_MAX_KEEP                   (365000 * 1440)  // data in db to be reserved.
+#define TSDB_MAX_KEEP                   (365000 * 1440)     // data in db to be reserved.
+#define TSDB_MAX_KEEP_NS                (365 * 292 * 1440)  // data in db to be reserved.
 #define TSDB_DEFAULT_KEEP               (3650 * 1440)    // ten years
 #define TSDB_MIN_MINROWS_FBLOCK         10
 #define TSDB_MAX_MINROWS_FBLOCK         1000

include/util/tlog.h

@@ -38,6 +38,7 @@ typedef void (*LogFp)(int64_t ts, ELogLevel level, const char *content);
 
 extern bool    tsLogEmbedded;
 extern bool    tsAsyncLog;
+extern bool    tsAssert;
 extern int32_t tsNumOfLogLines;
 extern int32_t tsLogKeepDays;
 extern LogFp   tsLogFp;
@@ -82,6 +83,10 @@ void taosPrintLongString(const char *flags, ELogLevel level, int32_t dflag, cons
 #endif
     ;
 
+bool taosAssert(bool condition, const char *file, int32_t line, const char *format, ...);
+#define ASSERTS(condition, ...) taosAssert(condition, __FILE__, __LINE__, __VA_ARGS__)
+#define ASSERT(condition)       ASSERTS(condition, "assert info not provided")
+
 // clang-format off
 #define uFatal(...) { if (uDebugFlag & DEBUG_FATAL) { taosPrintLog("UTL FATAL", DEBUG_FATAL, tsLogEmbedded ? 255 : uDebugFlag, __VA_ARGS__); }}
 #define uError(...) { if (uDebugFlag & DEBUG_ERROR) { taosPrintLog("UTL ERROR ", DEBUG_ERROR, tsLogEmbedded ? 255 : uDebugFlag, __VA_ARGS__); }}

packaging/tools/make_install.sh

@@ -497,27 +497,27 @@ function install_service_on_systemd() {
 
   taosd_service_config="${service_config_dir}/${serverName}.service"
 
-  ${csudo}bash -c "echo '[Unit]'                             >> ${taosd_service_config}"
+  ${csudo}bash -c "echo [Unit]                             >> ${taosd_service_config}"
-  ${csudo}bash -c "echo 'Description=${productName} server service' >> ${taosd_service_config}"
+  ${csudo}bash -c "echo Description=${productName} server service >> ${taosd_service_config}"
-  ${csudo}bash -c "echo 'After=network-online.target'        >> ${taosd_service_config}"
+  ${csudo}bash -c "echo After=network-online.target        >> ${taosd_service_config}"
-  ${csudo}bash -c "echo 'Wants=network-online.target'        >> ${taosd_service_config}"
+  ${csudo}bash -c "echo Wants=network-online.target        >> ${taosd_service_config}"
   ${csudo}bash -c "echo                                      >> ${taosd_service_config}"
-  ${csudo}bash -c "echo '[Service]'                          >> ${taosd_service_config}"
+  ${csudo}bash -c "echo [Service]                          >> ${taosd_service_config}"
-  ${csudo}bash -c "echo 'Type=simple'                        >> ${taosd_service_config}"
+  ${csudo}bash -c "echo Type=simple                        >> ${taosd_service_config}"
-  ${csudo}bash -c "echo 'ExecStart=/usr/bin/${serverName}'           >> ${taosd_service_config}"
+  ${csudo}bash -c "echo ExecStart=/usr/bin/${serverName}           >> ${taosd_service_config}"
-  ${csudo}bash -c "echo 'ExecStartPre=${installDir}/bin/startPre.sh'           >> ${taosd_service_config}"
+  ${csudo}bash -c "echo ExecStartPre=${installDir}/bin/startPre.sh           >> ${taosd_service_config}"
-  ${csudo}bash -c "echo 'TimeoutStopSec=1000000s'            >> ${taosd_service_config}"
+  ${csudo}bash -c "echo TimeoutStopSec=1000000s            >> ${taosd_service_config}"
-  ${csudo}bash -c "echo 'LimitNOFILE=infinity'               >> ${taosd_service_config}"
+  ${csudo}bash -c "echo LimitNOFILE=infinity               >> ${taosd_service_config}"
-  ${csudo}bash -c "echo 'LimitNPROC=infinity'                >> ${taosd_service_config}"
+  ${csudo}bash -c "echo LimitNPROC=infinity                >> ${taosd_service_config}"
-  ${csudo}bash -c "echo 'LimitCORE=infinity'                 >> ${taosd_service_config}"
+  ${csudo}bash -c "echo LimitCORE=infinity                 >> ${taosd_service_config}"
-  ${csudo}bash -c "echo 'TimeoutStartSec=0'                  >> ${taosd_service_config}"
+  ${csudo}bash -c "echo TimeoutStartSec=0                  >> ${taosd_service_config}"
-  ${csudo}bash -c "echo 'StandardOutput=null'                >> ${taosd_service_config}"
+  ${csudo}bash -c "echo StandardOutput=null                >> ${taosd_service_config}"
-  ${csudo}bash -c "echo 'Restart=always'                     >> ${taosd_service_config}"
+  ${csudo}bash -c "echo Restart=always                     >> ${taosd_service_config}"
-  ${csudo}bash -c "echo 'StartLimitBurst=3'                  >> ${taosd_service_config}"
+  ${csudo}bash -c "echo StartLimitBurst=3                  >> ${taosd_service_config}"
-  ${csudo}bash -c "echo 'StartLimitInterval=60s'             >> ${taosd_service_config}"
+  ${csudo}bash -c "echo StartLimitInterval=60s             >> ${taosd_service_config}"
   ${csudo}bash -c "echo                                      >> ${taosd_service_config}"
-  ${csudo}bash -c "echo '[Install]'                          >> ${taosd_service_config}"
+  ${csudo}bash -c "echo [Install]                          >> ${taosd_service_config}"
-  ${csudo}bash -c "echo 'WantedBy=multi-user.target'         >> ${taosd_service_config}"
+  ${csudo}bash -c "echo WantedBy=multi-user.target         >> ${taosd_service_config}"
   ${csudo}systemctl enable ${serverName}
 }
 

source/client/src/clientEnv.c

@@ -413,7 +413,9 @@ void taos_init_imp(void) {
 
   initQueryModuleMsgHandle();
 
-  taosConvInit();
+  if (taosConvInit() != 0) {
+    ASSERTS(0, "failed to init conv");
+  }
 
   rpcInit();
 

source/client/src/clientTmq.c

@@ -875,6 +875,7 @@ void tmqFreeImpl(void* handle) {
   tmq_t* tmq = (tmq_t*)handle;
 
   // TODO stop timer
+  tmqClearUnhandleMsg(tmq);
   if (tmq->mqueue) taosCloseQueue(tmq->mqueue);
   if (tmq->delayedTask) taosCloseQueue(tmq->delayedTask);
   if (tmq->qall) taosFreeQall(tmq->qall);
@@ -884,8 +885,7 @@ void tmqFreeImpl(void* handle) {
   int32_t sz = taosArrayGetSize(tmq->clientTopics);
   for (int32_t i = 0; i < sz; i++) {
     SMqClientTopic* pTopic = taosArrayGet(tmq->clientTopics, i);
-    if (pTopic->schema.nCols) taosMemoryFreeClear(pTopic->schema.pSchema);
+    taosMemoryFreeClear(pTopic->schema.pSchema);
-    int32_t vgSz = taosArrayGetSize(pTopic->vgs);
     taosArrayDestroy(pTopic->vgs);
   }
   taosArrayDestroy(tmq->clientTopics);
@@ -1304,7 +1304,6 @@ bool tmqUpdateEp(tmq_t* tmq, int32_t epoch, const SMqAskEpRsp* pRsp) {
     for (int32_t i = 0; i < sz; i++) {
       SMqClientTopic* pTopic = taosArrayGet(tmq->clientTopics, i);
       if (pTopic->schema.nCols) taosMemoryFreeClear(pTopic->schema.pSchema);
-      int32_t vgSz = taosArrayGetSize(pTopic->vgs);
       taosArrayDestroy(pTopic->vgs);
     }
     taosArrayDestroy(tmq->clientTopics);
@@ -1410,7 +1409,7 @@ int32_t tmqAskEp(tmq_t* tmq, bool async) {
     return -1;
   }
   void* pReq = taosMemoryCalloc(1, tlen);
-  if (tlen < 0) {
+  if (pReq == NULL) {
     tscError("failed to malloc askEpReq msg, size:%d", tlen);
     return -1;
   }
@@ -1738,7 +1737,7 @@ void* tmqHandleAllRsp(tmq_t* tmq, int64_t timeout, bool pollIfReset) {
         taosFreeQitem(pollRspWrapper);
         return pRsp;
       } else {
-        tscDebug("msg discard since epoch mismatch: msg epoch %d, consumer epoch %d\n",
+        tscDebug("msg discard since epoch mismatch: msg epoch %d, consumer epoch %d",
                  pollRspWrapper->taosxRsp.head.epoch, consumerEpoch);
         taosFreeQitem(pollRspWrapper);
       }

source/common/src/tglobal.c

@@ -333,6 +333,7 @@ static int32_t taosAddSystemCfg(SConfig *pCfg) {
   if (cfgAddTimezone(pCfg, "timezone", tsTimezoneStr) != 0) return -1;
   if (cfgAddLocale(pCfg, "locale", tsLocale) != 0) return -1;
   if (cfgAddCharset(pCfg, "charset", tsCharset) != 0) return -1;
+  if (cfgAddBool(pCfg, "assert", 1, 1) != 0) return -1;
   if (cfgAddBool(pCfg, "enableCoreFile", 1, 1) != 0) return -1;
   if (cfgAddFloat(pCfg, "numOfCores", tsNumOfCores, 1, 100000, 1) != 0) return -1;
 
@@ -693,6 +694,8 @@ static void taosSetSystemCfg(SConfig *pCfg) {
   bool enableCore = cfgGetItem(pCfg, "enableCoreFile")->bval;
   taosSetCoreDump(enableCore);
 
+  tsAssert = cfgGetItem(pCfg, "assert")->bval;
+
   // todo
   tsVersion = 30000000;
 }
@@ -788,6 +791,8 @@ int32_t taosSetCfg(SConfig *pCfg, char *name) {
     case 'a': {
       if (strcasecmp("asyncLog", name) == 0) {
         tsAsyncLog = cfgGetItem(pCfg, "asyncLog")->bval;
+      } else if (strcasecmp("assert", name) == 0) {
+        tsAssert = cfgGetItem(pCfg, "assert")->bval;
       }
       break;
     }

source/common/src/tmsg.c

@@ -28,6 +28,8 @@
 #undef TD_MSG_SEG_CODE_
 #include "tmsgdef.h"
 
+#include "tlog.h"
+
 int32_t tInitSubmitMsgIter(const SSubmitReq *pMsg, SSubmitMsgIter *pIter) {
   if (pMsg == NULL) {
     terrno = TSDB_CODE_TDB_SUBMIT_MSG_MSSED_UP;

source/common/src/trow.c

@@ -15,6 +15,7 @@
 
 #define _DEFAULT_SOURCE
 #include "trow.h"
+#include "tlog.h"
 
 const uint8_t tdVTypeByte[2][3] = {{
                                        // 2 bits

source/common/src/ttime.c

@@ -23,6 +23,8 @@
 #define _DEFAULT_SOURCE
 #include "ttime.h"
 
+#include "tlog.h"
+
 /*
  * mktime64 - Converts date to seconds.
  * Converts Gregorian date to seconds since 1970-01-01 00:00:00.

source/dnode/mgmt/exe/dmMain.c

@@ -17,6 +17,7 @@
 #include "dmMgmt.h"
 #include "mnode.h"
 #include "tconfig.h"
+#include "tglobal.h"
 
 // clang-format off
 #define DM_APOLLO_URL    "The apollo string to use when configuring the server, such as: -a 'jsonFile:./tests/cfg.json', cfg.json text can be '{\"fqdn\":\"td1\"}'."
@@ -45,9 +46,30 @@ static struct {
   SArray      *pArgs;  // SConfigPair
 } global = {0};
 
-static void dmStopDnode(int signum, void *info, void *ctx) { dmStop(); }
+static void dmSetDebugFlag(int32_t signum, void *sigInfo, void *context) { taosSetAllDebugFlag(143, true); }
+static void dmSetAssert(int32_t signum, void *sigInfo, void *context) { tsAssert = 1; }
+
+static void dmStopDnode(int signum, void *sigInfo, void *context) {
+  // taosIgnSignal(SIGUSR1);
+  // taosIgnSignal(SIGUSR2);
+  taosIgnSignal(SIGTERM);
+  taosIgnSignal(SIGHUP);
+  taosIgnSignal(SIGINT);
+  taosIgnSignal(SIGABRT);
+  taosIgnSignal(SIGBREAK);
+
+  dInfo("shut down signal is %d", signum);
+#ifndef WINDOWS
+  dInfo("sender PID:%d cmdline:%s", ((siginfo_t *)sigInfo)->si_pid,
+        taosGetCmdlineByPID(((siginfo_t *)sigInfo)->si_pid));
+#endif
+
+  dmStop();
+}
 
 static void dmSetSignalHandle() {
+  taosSetSignal(SIGUSR1, dmSetDebugFlag);
+  taosSetSignal(SIGUSR2, dmSetAssert);
   taosSetSignal(SIGTERM, dmStopDnode);
   taosSetSignal(SIGHUP, dmStopDnode);
   taosSetSignal(SIGINT, dmStopDnode);
@@ -105,6 +127,19 @@ static int32_t dmParseArgs(int32_t argc, char const *argv[]) {
   return 0;
 }
 
+static void dmPrintArgs(int32_t argc, char const *argv[]) {
+  char path[1024] = {0};
+  taosGetCwd(path, sizeof(path));
+
+  char    args[1024] = {0};
+  int32_t arglen = snprintf(args, sizeof(args), "%s", argv[0]);
+  for (int32_t i = 1; i < argc; ++i) {
+    arglen = arglen + snprintf(args + arglen, sizeof(args) - arglen, " %s", argv[i]);
+  }
+
+  dInfo("startup path:%s args:%s", path, args);
+}
+
 static void dmGenerateGrant() { mndGenerateMachineCode(); }
 
 static void dmPrintVersion() {
@@ -194,6 +229,8 @@ int mainWindows(int argc, char **argv) {
     return -1;
   }
 
+  dmPrintArgs(argc, argv);
+
   if (taosInitCfg(configDir, global.envCmd, global.envFile, global.apolloUrl, global.pArgs, 0) != 0) {
     dError("failed to start since read config error");
     taosCloseLog();
@@ -201,7 +238,12 @@ int mainWindows(int argc, char **argv) {
     return -1;
   }
 
-  taosConvInit();
+  if (taosConvInit() != 0) {
+    dError("failed to init conv");
+    taosCloseLog();
+    taosCleanupArgs();
+    return -1;
+  }
 
   if (global.dumpConfig) {
     dmDumpCfg();

source/dnode/mgmt/mgmt_dnode/src/dmHandle.c

@@ -103,7 +103,12 @@ void dmSendStatusReq(SDnodeMgmt *pMgmt) {
   tSerializeSStatusReq(pHead, contLen, &req);
   tFreeSStatusReq(&req);
 
-  SRpcMsg rpcMsg = {.pCont = pHead, .contLen = contLen, .msgType = TDMT_MND_STATUS, .info.ahandle = (void *)0x9527};
+  SRpcMsg rpcMsg = {.pCont = pHead,
+                    .contLen = contLen,
+                    .msgType = TDMT_MND_STATUS,
+                    .info.ahandle = (void *)0x9527,
+                    .info.refId = 0,
+                    .info.noResp = 0};
   SRpcMsg rpcRsp = {0};
 
   dTrace("send status req to mnode, dnodeVer:%" PRId64 " statusSeq:%d", req.dnodeVer, req.statusSeq);
@@ -150,7 +155,8 @@ static void dmGetServerRunStatus(SDnodeMgmt *pMgmt, SServerStatusRsp *pStatus) {
   SServerStatusRsp statusRsp = {0};
   SMonMloadInfo    minfo = {0};
   (*pMgmt->getMnodeLoadsFp)(&minfo);
-  if (minfo.isMnode && (minfo.load.syncState == TAOS_SYNC_STATE_ERROR || minfo.load.syncState == TAOS_SYNC_STATE_OFFLINE)) {
+  if (minfo.isMnode &&
+      (minfo.load.syncState == TAOS_SYNC_STATE_ERROR || minfo.load.syncState == TAOS_SYNC_STATE_OFFLINE)) {
     pStatus->statusCode = TSDB_SRV_STATUS_SERVICE_DEGRADED;
     snprintf(pStatus->details, sizeof(pStatus->details), "mnode sync state is %s", syncStr(minfo.load.syncState));
     return;

source/dnode/mgmt/mgmt_dnode/src/dmWorker.c

@@ -20,7 +20,9 @@ static void *dmStatusThreadFp(void *param) {
   SDnodeMgmt *pMgmt = param;
   int64_t     lastTime = taosGetTimestampMs();
   setThreadName("dnode-status");
-  
+
+  const static int16_t TRIM_FREQ = 30;
+  int32_t              trimCount = 0;
   while (1) {
     taosMsleep(200);
     if (pMgmt->pData->dropped || pMgmt->pData->stopped) break;
@@ -28,9 +30,13 @@ static void *dmStatusThreadFp(void *param) {
     int64_t curTime = taosGetTimestampMs();
     float   interval = (curTime - lastTime) / 1000.0f;
     if (interval >= tsStatusInterval) {
-      taosMemoryTrim(0); 
       dmSendStatusReq(pMgmt);
       lastTime = curTime;
+
+      trimCount = (trimCount + 1) % TRIM_FREQ;
+      if (trimCount == 0) {
+        taosMemoryTrim(0);
+      }
     }
   }
 

source/dnode/mgmt/mgmt_snode/src/smWorker.c

@@ -139,7 +139,7 @@ int32_t smPutMsgToQueue(SSnodeMgmt *pMgmt, EQueueType qtype, SRpcMsg *pRpc) {
 
   SSnode *pSnode = pMgmt->pSnode;
   if (pSnode == NULL) {
-    dError("snode: msg:%p failed to put into vnode queue since %s, type:%s qtype:%d", pMsg, terrstr(),
+    dError("msg:%p failed to put into snode queue since %s, type:%s qtype:%d", pMsg, terrstr(),
            TMSG_INFO(pMsg->msgType), qtype);
     taosFreeQitem(pMsg);
     rpcFreeCont(pRpc->pCont);
@@ -161,7 +161,8 @@ int32_t smPutMsgToQueue(SSnodeMgmt *pMgmt, EQueueType qtype, SRpcMsg *pRpc) {
       smPutNodeMsgToWriteQueue(pMgmt, pMsg);
       break;
     default:
-      ASSERT(0);
+      ASSERTS(0, "msg:%p failed to put into snode queue since %s, type:%s qtype:%d", pMsg, terrstr(),
+              TMSG_INFO(pMsg->msgType), qtype);
   }
   return 0;
 }

source/dnode/mgmt/node_mgmt/src/dmTransport.c

@@ -301,6 +301,7 @@ int32_t dmInitServer(SDnode *pDnode) {
   rpcInit.connType = TAOS_CONN_SERVER;
   rpcInit.idleTime = tsShellActivityTimer * 1000;
   rpcInit.parent = pDnode;
+  rpcInit.compressSize = tsCompressMsgSize;
 
   pTrans->serverRpc = rpcOpen(&rpcInit);
   if (pTrans->serverRpc == NULL) {

source/dnode/mnode/impl/src/mndConsumer.c

@@ -555,6 +555,12 @@ static int32_t mndProcessSubscribeReq(SRpcMsg *pMsg) {
     }
 
     if (mndCheckDbPrivilegeByName(pMnode, pMsg->info.conn.user, MND_OPER_READ_DB, pTopic->db) != 0) {
+      mndReleaseTopic(pMnode, pTopic);
+      goto SUBSCRIBE_OVER;
+    }
+
+    if (mndCheckTopicPrivilege(pMnode, pMsg->info.conn.user, MND_OPER_SUBSCRIBE, pTopic) != 0) {
+      mndReleaseTopic(pMnode, pTopic);
       goto SUBSCRIBE_OVER;
     }
 

source/dnode/mnode/impl/src/mndDb.c

@@ -825,7 +825,13 @@ static int32_t mndProcessAlterDbReq(SRpcMsg *pReq) {
   dbObj.cfgVersion++;
   dbObj.updateTime = taosGetTimestampMs();
   code = mndAlterDb(pMnode, pReq, pDb, &dbObj);
-  if (code == 0) code = TSDB_CODE_ACTION_IN_PROGRESS;
+
+  if (dbObj.cfg.replications != pDb->cfg.replications) {
+    // return quickly, operation executed asynchronously
+    mInfo("db:%s, alter db replica from %d to %d", pDb->name, pDb->cfg.replications, dbObj.cfg.replications);
+  } else {
+    if (code == 0) code = TSDB_CODE_ACTION_IN_PROGRESS;
+  }
 
 _OVER:
   if (code != 0 && code != TSDB_CODE_ACTION_IN_PROGRESS) {

source/dnode/mnode/impl/src/mndSync.c

@@ -119,7 +119,13 @@ int32_t mndProcessWriteMsg(const SSyncFSM *pFsm, SRpcMsg *pMsg, const SFsmCbMeta
 }
 
 int32_t mndSyncCommitMsg(const SSyncFSM *pFsm, SRpcMsg *pMsg, const SFsmCbMeta *pMeta) {
-  int32_t code = mndProcessWriteMsg(pFsm, pMsg, pMeta);
+  int32_t code = 0;
+  if (!syncUtilUserCommit(pMsg->msgType)) {
+    goto _out;
+  }
+  code = mndProcessWriteMsg(pFsm, pMsg, pMeta);
+
+_out:
   rpcFreeCont(pMsg->pCont);
   pMsg->pCont = NULL;
   return code;
@@ -143,9 +149,13 @@ void mndRestoreFinish(const SSyncFSM *pFsm) {
   SMnode *pMnode = pFsm->data;
 
   if (!pMnode->deploy) {
-    mInfo("vgId:1, sync restore finished, and will handle outstanding transactions");
+    if (!pMnode->restored) {
-    mndTransPullup(pMnode);
+      mInfo("vgId:1, sync restore finished, and will handle outstanding transactions");
-    mndSetRestored(pMnode, true);
+      mndTransPullup(pMnode);
+      mndSetRestored(pMnode, true);
+    } else {
+      mInfo("vgId:1, sync restore finished, repeat call");
+    }
   } else {
     mInfo("vgId:1, sync restore finished");
   }

source/dnode/vnode/inc/vnode.h

@@ -174,7 +174,7 @@ int32_t tsdbReaderOpen(SVnode *pVnode, SQueryTableDataCond *pCond, void *pTableL
 void         tsdbReaderClose(STsdbReader *pReader);
 bool         tsdbNextDataBlock(STsdbReader *pReader);
 void         tsdbRetrieveDataBlockInfo(const STsdbReader *pReader, int32_t *rows, uint64_t *uid, STimeWindow *pWindow);
-int32_t      tsdbRetrieveDatablockSMA(STsdbReader *pReader, SColumnDataAgg ***pBlockSMA, bool *allHave);
+int32_t      tsdbRetrieveDatablockSMA(STsdbReader *pReader, SSDataBlock* pDataBlock, bool *allHave);
 SSDataBlock *tsdbRetrieveDataBlock(STsdbReader *pTsdbReadHandle, SArray *pColumnIdList);
 int32_t      tsdbReaderReset(STsdbReader *pReader, SQueryTableDataCond *pCond);
 int32_t      tsdbGetFileBlocksDistInfo(STsdbReader *pReader, STableBlockDistInfo *pTableBlockInfo);

source/dnode/vnode/src/meta/metaCache.c

@@ -454,7 +454,7 @@ int32_t metaGetCachedTableUidList(SMeta* pMeta, tb_uid_t suid, const uint8_t* pK
 
       SListNode* pNode = NULL;
       while ((pNode = tdListNext(&iter)) != NULL) {
-        memcpy(pBuf + sizeof(suid), pNode->data, keyLen);
+        memcpy(&pBuf[1], pNode->data, keyLen);
 
         // check whether it is existed in LRU cache, and remove it from linked list if not.
         LRUHandle* pRes = taosLRUCacheLookup(pCache, pBuf, len);

source/dnode/vnode/src/meta/metaTable.c

@@ -549,8 +549,8 @@ int metaTtlDropTable(SMeta *pMeta, int64_t ttl, SArray *tbUids) {
 }
 
 static void metaBuildTtlIdxKey(STtlIdxKey *ttlKey, const SMetaEntry *pME) {
-  int64_t ttlDays;
+  int64_t ttlDays = 0;
-  int64_t ctime;
+  int64_t ctime = 0;
   if (pME->type == TSDB_CHILD_TABLE) {
     ctime = pME->ctbEntry.ctime;
     ttlDays = pME->ctbEntry.ttlDays;
@@ -1353,6 +1353,10 @@ static int metaUpdateTagIdx(SMeta *pMeta, const SMetaEntry *pCtbEntry) {
     goto end;
   }
 
+  if (stbEntry.stbEntry.schemaTag.pSchema == NULL) {
+    goto end;
+  }
+
   pTagColumn = &stbEntry.stbEntry.schemaTag.pSchema[0];
 
   STagVal tagVal = {.cid = pTagColumn->colId};

source/dnode/vnode/src/tsdb/tsdbCache.c

@@ -1025,7 +1025,10 @@ static int32_t nextRowIterOpen(CacheNextRowIter *pIter, tb_uid_t uid, STsdb *pTs
     SArray *pDelIdxArray = taosArrayInit(32, sizeof(SDelIdx));
 
     code = tsdbReadDelIdx(pDelFReader, pDelIdxArray);
-    if (code) goto _err;
+    if (code) {
+      tsdbDelFReaderClose(&pDelFReader);
+      goto _err;
+    }
 
     SDelIdx *delIdx = taosArraySearch(pDelIdxArray, &(SDelIdx){.suid = suid, .uid = uid}, tCmprDelIdx, TD_EQ);
 

source/dnode/vnode/src/tsdb/tsdbCommit.c

@@ -53,6 +53,7 @@ typedef struct {
   // --------------
   TSKEY   nextKey;  // reset by each table commit
   int32_t commitFid;
+  int32_t expLevel;
   TSKEY   minKey;
   TSKEY   maxKey;
   // commit file data
@@ -503,6 +504,7 @@ static int32_t tsdbCommitFileDataStart(SCommitter *pCommitter) {
 
   // memory
   pCommitter->commitFid = tsdbKeyFid(pCommitter->nextKey, pCommitter->minutes, pCommitter->precision);
+  pCommitter->expLevel = tsdbFidLevel(pCommitter->commitFid, &pCommitter->pTsdb->keepCfg, taosGetTimestampSec());
   tsdbFidKeyRange(pCommitter->commitFid, pCommitter->minutes, pCommitter->precision, &pCommitter->minKey,
                   &pCommitter->maxKey);
 #if 0
@@ -556,7 +558,10 @@ static int32_t tsdbCommitFileDataStart(SCommitter *pCommitter) {
     }
   } else {
     SDiskID did = {0};
-    tfsAllocDisk(pTsdb->pVnode->pTfs, 0, &did);
+    if (tfsAllocDisk(pTsdb->pVnode->pTfs, pCommitter->expLevel, &did) < 0) {
+      code = terrno;
+      TSDB_CHECK_CODE(code, lino, _exit);
+    }
     tfsMkdirRecurAt(pTsdb->pVnode->pTfs, pTsdb->path, did);
     wSet.diskId = did;
     wSet.nSttF = 1;

source/dnode/vnode/src/tsdb/tsdbFS.c

@@ -962,6 +962,7 @@ int32_t tsdbFSUpsertFSet(STsdbFS *pFS, SDFileSet *pSet) {
         }
       }
 
+      pDFileSet->diskId = pSet->diskId;
       goto _exit;
     }
   }

source/dnode/vnode/src/tsdb/tsdbRead.c

@@ -3128,7 +3128,7 @@ bool hasBeenDropped(const SArray* pDelList, int32_t* index, TSDBKEY* pKey, int32
         return false;
       } else if (pKey->ts == last->ts) {
         TSDBKEY* prev = taosArrayGet(pDelList, num - 2);
-        return (prev->version >= pKey->version);
+        return (prev->version >= pKey->version && prev->version <= pVerRange->maxVer && prev->version >= pVerRange->minVer);
       }
     } else {
       TSDBKEY* pCurrent = taosArrayGet(pDelList, *index);
@@ -4134,8 +4134,9 @@ static void doFillNullColSMA(SBlockLoadSuppInfo* pSup, int32_t numOfRows, int32_
   }
 }
 
-int32_t tsdbRetrieveDatablockSMA(STsdbReader* pReader, SColumnDataAgg*** pBlockSMA, bool* allHave) {
+int32_t tsdbRetrieveDatablockSMA(STsdbReader* pReader, SSDataBlock* pDataBlock, bool* allHave) {
   int32_t code = 0;
+  SColumnDataAgg ***pBlockSMA = &pDataBlock->pBlockAgg;
   *allHave = false;
 
   if (pReader->type == TIMEWINDOW_RANGE_EXTERNAL) {
@@ -4183,6 +4184,12 @@ int32_t tsdbRetrieveDatablockSMA(STsdbReader* pReader, SColumnDataAgg*** pBlockS
   int32_t i = 0, j = 0;
   size_t  size = taosArrayGetSize(pSup->pColAgg);
 
+  // ensure capacity
+  if(pDataBlock->pDataBlock) {
+     size_t colsNum = taosArrayGetSize(pDataBlock->pDataBlock);
+     taosArrayEnsureCap(pSup->pColAgg, colsNum);
+  }
+
   SSDataBlock* pResBlock = pReader->pResBlock;
   if (pResBlock->pBlockAgg == NULL) {
     size_t num = taosArrayGetSize(pResBlock->pDataBlock);

source/dnode/vnode/src/tsdb/tsdbSnapshot.c

@@ -155,7 +155,7 @@ static int32_t tsdbSnapReadOpenFile(STsdbSnapReader* pReader) {
 
         if (rowVer >= pReader->sver && rowVer <= pReader->ever) {
           pIter->rInfo.suid = pIter->bData.suid;
-          pIter->rInfo.uid = pIter->bData.uid;
+          pIter->rInfo.uid = pIter->bData.uid ? pIter->bData.uid : pIter->bData.aUid[pIter->iRow];
           pIter->rInfo.row = tsdbRowFromBlockData(&pIter->bData, pIter->iRow);
           goto _add_iter;
         }
@@ -179,16 +179,14 @@ _err:
   return code;
 }
 
-static SRowInfo* tsdbSnapGetRow(STsdbSnapReader* pReader) { return pReader->pIter ? &pReader->pIter->rInfo : NULL; }
-
 static int32_t tsdbSnapNextRow(STsdbSnapReader* pReader) {
   int32_t code = 0;
 
   if (pReader->pIter) {
-    SFDataIter* pIter = pReader->pIter;
+    SFDataIter* pIter = NULL;
-
     while (true) {
     _find_row:
+      pIter = pReader->pIter;
       for (pIter->iRow++; pIter->iRow < pIter->bData.nRow; pIter->iRow++) {
         int64_t rowVer = pIter->bData.aVersion[pIter->iRow];
 
@@ -224,6 +222,7 @@ static int32_t tsdbSnapNextRow(STsdbSnapReader* pReader) {
         }
 
         pReader->pIter = NULL;
+        break;
       } else if (pIter->type == SNAP_STT_FILE_ITER) {
         for (pIter->iSttBlk++; pIter->iSttBlk < taosArrayGetSize(pIter->aSttBlk); pIter->iSttBlk++) {
           SSttBlk* pSttBlk = (SSttBlk*)taosArrayGet(pIter->aSttBlk, pIter->iSttBlk);
@@ -238,6 +237,7 @@ static int32_t tsdbSnapNextRow(STsdbSnapReader* pReader) {
         }
 
         pReader->pIter = NULL;
+        break;
       } else {
         ASSERT(0);
       }
@@ -269,6 +269,20 @@ _err:
   return code;
 }
 
+static SRowInfo* tsdbSnapGetRow(STsdbSnapReader* pReader) {
+  if (pReader->pIter) {
+    return &pReader->pIter->rInfo;
+  } else {
+    tsdbSnapNextRow(pReader);
+
+    if (pReader->pIter) {
+      return &pReader->pIter->rInfo;
+    } else {
+      return NULL;
+    }
+  }
+}
+
 static int32_t tsdbSnapCmprData(STsdbSnapReader* pReader, uint8_t** ppData) {
   int32_t code = 0;
 
@@ -1356,7 +1370,7 @@ _exit:
       taosMemoryFree(pWriter);
     }
   } else {
-    tsdbDebug("vgId:%d, tsdb snapshot writer open for %s succeed", TD_VID(pTsdb->pVnode), pTsdb->path);
+    tsdbInfo("vgId:%d %s done", TD_VID(pTsdb->pVnode), __func__);
     *ppWriter = pWriter;
   }
   return code;
@@ -1421,7 +1435,7 @@ int32_t tsdbSnapWriterClose(STsdbSnapWriter** ppWriter, int8_t rollback) {
   for (int32_t iBuf = 0; iBuf < sizeof(pWriter->aBuf) / sizeof(uint8_t*); iBuf++) {
     tFree(pWriter->aBuf[iBuf]);
   }
-  tsdbInfo("vgId:%d, vnode snapshot tsdb writer close for %s", TD_VID(pWriter->pTsdb->pVnode), pWriter->pTsdb->path);
+  tsdbInfo("vgId:%d %s done", TD_VID(pWriter->pTsdb->pVnode), __func__);
   taosMemoryFree(pWriter);
   *ppWriter = NULL;
   return code;

source/dnode/vnode/src/tsdb/tsdbUtil.c

@@ -637,6 +637,7 @@ SColVal *tsdbRowIterNext(STSDBRowIter *pIter) {
     }
   } else {
     ASSERT(0);
+    return NULL;   // suppress error report by compiler
   }
 }
 

source/dnode/vnode/src/vnd/vnodeSvr.c

@@ -210,9 +210,13 @@ int32_t vnodeProcessWriteMsg(SVnode *pVnode, SRpcMsg *pMsg, int64_t version, SRp
          version);
 
   ASSERT(pVnode->state.applyTerm <= pMsg->info.conn.applyTerm);
+  ASSERT(pVnode->state.applied + 1 == version);
+
   pVnode->state.applied = version;
   pVnode->state.applyTerm = pMsg->info.conn.applyTerm;
 
+  if (!syncUtilUserCommit(pMsg->msgType)) goto _exit;
+
   // skip header
   pReq = POINTER_SHIFT(pMsg->pCont, sizeof(SMsgHead));
   len = pMsg->contLen - sizeof(SMsgHead);

source/libs/executor/src/scanoperator.c

@@ -224,7 +224,7 @@ static bool doFilterByBlockSMA(SFilterInfo* pFilterInfo, SColumnDataAgg** pColsA
 
 static bool doLoadBlockSMA(STableScanBase* pTableScanInfo, SSDataBlock* pBlock, SExecTaskInfo* pTaskInfo) {
   bool    allColumnsHaveAgg = true;
-  int32_t code = tsdbRetrieveDatablockSMA(pTableScanInfo->dataReader, &pBlock->pBlockAgg, &allColumnsHaveAgg);
+  int32_t code = tsdbRetrieveDatablockSMA(pTableScanInfo->dataReader, pBlock, &allColumnsHaveAgg);
   if (code != TSDB_CODE_SUCCESS) {
     T_LONG_JMP(pTaskInfo->env, code);
   }

source/libs/executor/src/timesliceoperator.c

@@ -95,6 +95,8 @@ static void doKeepLinearInfo(STimeSliceOperatorInfo* pSliceInfo, const SSDataBlo
     // TODO: optimize to ignore null values for linear interpolation.
     if (!pLinearInfo->isStartSet) {
       if (!colDataIsNull_s(pColInfoData, rowIndex)) {
+        ASSERT(IS_MATHABLE_TYPE(pColInfoData->info.type));
+
         pLinearInfo->start.key = *(int64_t*)colDataGetData(pTsCol, rowIndex);
         memcpy(pLinearInfo->start.val, colDataGetData(pColInfoData, rowIndex), pLinearInfo->bytes);
       }

source/libs/executor/src/timewindowoperator.c

@@ -863,19 +863,20 @@ static void removeResults(SArray* pWins, SHashObj* pUpdatedMap) {
 
 int32_t compareWinRes(void* pKey, void* data, int32_t index) {
   SArray*     res = (SArray*)data;
-  SWinKey*    pos = taosArrayGet(res, index);
+  SWinKey*    pDataPos = taosArrayGet(res, index);
-  SResKeyPos* pData = (SResKeyPos*)pKey;
+  SResKeyPos* pRKey = (SResKeyPos*)pKey;
-  if (*(int64_t*)pData->key == pos->ts) {
+  if (pRKey->groupId > pDataPos->groupId) {
-    if (pData->groupId > pos->groupId) {
-      return 1;
-    } else if (pData->groupId < pos->groupId) {
-      return -1;
-    }
-    return 0;
-  } else if (*(int64_t*)pData->key > pos->ts) {
     return 1;
+  } else if (pRKey->groupId < pDataPos->groupId) {
+    return -1;
   }
-  return -1;
+  
+  if (*(int64_t*)pRKey->key > pDataPos->ts) {
+    return 1;
+  } else if (*(int64_t*)pRKey->key < pDataPos->ts){
+    return -1;
+  }
+  return 0;
 }
 
 static void removeDeleteResults(SHashObj* pUpdatedMap, SArray* pDelWins) {
@@ -1400,19 +1401,21 @@ static int32_t getAllIntervalWindow(SSHashObj* pHashMap, SHashObj* resWins) {
 
 int32_t compareWinKey(void* pKey, void* data, int32_t index) {
   SArray*  res = (SArray*)data;
-  SWinKey* pos = taosArrayGet(res, index);
+  SWinKey* pDataPos = taosArrayGet(res, index);
-  SWinKey* pData = (SWinKey*)pKey;
+  SWinKey* pWKey = (SWinKey*)pKey;
-  if (pData->ts == pos->ts) {
+
-    if (pData->groupId > pos->groupId) {
+  if (pWKey->groupId > pDataPos->groupId) {
-      return 1;
-    } else if (pData->groupId < pos->groupId) {
-      return -1;
-    }
-    return 0;
-  } else if (pData->ts > pos->ts) {
     return 1;
+  } else if (pWKey->groupId < pDataPos->groupId) {
+    return -1;
   }
-  return -1;
+
+  if (pWKey->ts > pDataPos->ts) {
+    return 1;
+  } else if (pWKey->ts < pDataPos->ts) {
+    return -1;
+  }
+  return 0;
 }
 
 static int32_t closeStreamIntervalWindow(SSHashObj* pHashMap, STimeWindowAggSupp* pTwSup, SInterval* pInterval,

source/libs/executor/src/tlinearhash.c

@@ -17,6 +17,7 @@
 #include "taoserror.h"
 #include "tdef.h"
 #include "tpagedbuf.h"
+#include "tlog.h"
 
 #define LHASH_CAP_RATIO 0.85
 

source/libs/function/src/tpercentile.c

@@ -22,6 +22,7 @@
 #include "tpagedbuf.h"
 #include "tpercentile.h"
 #include "ttypes.h"
+#include "tlog.h"
 
 #define DEFAULT_NUM_OF_SLOT 1024
 
@@ -367,11 +368,13 @@ int32_t tMemBucketPut(tMemBucket *pBucket, const void *data, size_t size) {
         pSlot->info.data = NULL;
       }
 
-      SArray *pPageIdList = (SArray *)taosHashGet(pBucket->groupPagesMap, &groupId, sizeof(groupId));
+      SArray *pPageIdList;
-      if (pPageIdList == NULL) {
+      void *p = taosHashGet(pBucket->groupPagesMap, &groupId, sizeof(groupId));
-        SArray *pList = taosArrayInit(4, sizeof(int32_t));
+      if (p == NULL) {
-        taosHashPut(pBucket->groupPagesMap, &groupId, sizeof(groupId), &pList, POINTER_BYTES);
+        pPageIdList = taosArrayInit(4, sizeof(int32_t));
-        pPageIdList = pList;
+        taosHashPut(pBucket->groupPagesMap, &groupId, sizeof(groupId), &pPageIdList, POINTER_BYTES);
+      } else {
+        pPageIdList = *(SArray **)p;
       }
 
       pSlot->info.data = getNewBufPage(pBucket->pBuffer, &pageId);

source/libs/function/src/tudf.c

@@ -88,11 +88,13 @@ static int32_t udfSpawnUdfd(SUdfdData *pData) {
   }
 #ifdef WINDOWS
   if (strlen(path) == 0) {
-    strcat(path, "udfd.exe");
+    strcat(path, "C:\\TDengine");
-  } else {
-    strcat(path, "\\udfd.exe");
   }
+  strcat(path, "\\udfd.exe");
 #else
+  if (strlen(path) == 0) {
+    strcat(path, "/usr/bin");
+  }
   strcat(path, "/udfd");
 #endif
   char *argsUdfd[] = {path, "-c", configDir, NULL};

source/libs/parser/src/parTranslater.c

@@ -3879,12 +3879,17 @@ static int32_t checkDbKeepOption(STranslateContext* pCxt, SDatabaseOptions* pOpt
     pOptions->keep[2] = getBigintFromValueNode((SValueNode*)nodesListGetNode(pOptions->pKeep, 2));
   }
 
+  int64_t tsdbMaxKeep = TSDB_MAX_KEEP;
+  if (pOptions->precision == TSDB_TIME_PRECISION_NANO) {
+    tsdbMaxKeep = TSDB_MAX_KEEP_NS;
+  }
+
   if (pOptions->keep[0] < TSDB_MIN_KEEP || pOptions->keep[1] < TSDB_MIN_KEEP || pOptions->keep[2] < TSDB_MIN_KEEP ||
-      pOptions->keep[0] > TSDB_MAX_KEEP || pOptions->keep[1] > TSDB_MAX_KEEP || pOptions->keep[2] > TSDB_MAX_KEEP) {
+      pOptions->keep[0] > tsdbMaxKeep || pOptions->keep[1] > tsdbMaxKeep || pOptions->keep[2] > tsdbMaxKeep) {
     return generateSyntaxErrMsgExt(&pCxt->msgBuf, TSDB_CODE_PAR_INVALID_DB_OPTION,
                                    "Invalid option keep: %" PRId64 ", %" PRId64 ", %" PRId64 " valid range: [%dm, %dm]",
                                    pOptions->keep[0], pOptions->keep[1], pOptions->keep[2], TSDB_MIN_KEEP,
-                                   TSDB_MAX_KEEP);
+                                   tsdbMaxKeep);
   }
 
   if (!((pOptions->keep[0] <= pOptions->keep[1]) && (pOptions->keep[1] <= pOptions->keep[2]))) {
@@ -4036,7 +4041,10 @@ static int32_t checkDatabaseOptions(STranslateContext* pCxt, const char* pDbName
                               TSDB_MAX_MINROWS_FBLOCK);
   }
   if (TSDB_CODE_SUCCESS == code) {
-    code = checkDbKeepOption(pCxt, pOptions);
+    code = checkDbPrecisionOption(pCxt, pOptions);
+  }
+  if (TSDB_CODE_SUCCESS == code) {
+    code = checkDbKeepOption(pCxt, pOptions); // use precision
   }
   if (TSDB_CODE_SUCCESS == code) {
     code = checkDbRangeOption(pCxt, "pages", pOptions->pages, TSDB_MIN_PAGES_PER_VNODE, TSDB_MAX_PAGES_PER_VNODE);
@@ -4049,9 +4057,6 @@ static int32_t checkDatabaseOptions(STranslateContext* pCxt, const char* pDbName
     code = checkDbRangeOption(pCxt, "tsdbPagesize", pOptions->tsdbPageSize, TSDB_MIN_TSDB_PAGESIZE,
                               TSDB_MAX_TSDB_PAGESIZE);
   }
-  if (TSDB_CODE_SUCCESS == code) {
-    code = checkDbPrecisionOption(pCxt, pOptions);
-  }
   if (TSDB_CODE_SUCCESS == code) {
     code = checkDbEnumOption(pCxt, "replications", pOptions->replica, TSDB_MIN_DB_REPLICA, TSDB_MAX_DB_REPLICA);
   }

source/libs/sync/inc/syncSnapshot.h

@@ -44,6 +44,7 @@ typedef struct SSyncSnapshotSender {
   SyncTerm       term;
   int64_t        startTime;
   int64_t        endTime;
+  int64_t        lastSendTime;
   bool           finish;
 
   // init when create

source/libs/sync/inc/syncUtil.h

@@ -79,7 +79,6 @@ char*       syncUtilPrintBin2(char* ptr, uint32_t len);
 void        syncUtilMsgHtoN(void* msg);
 void        syncUtilMsgNtoH(void* msg);
 bool        syncUtilUserPreCommit(tmsg_t msgType);
-bool        syncUtilUserCommit(tmsg_t msgType);
 bool        syncUtilUserRollback(tmsg_t msgType);
 
 void syncPrintNodeLog(const char* flags, ELogLevel level, int32_t dflag, SSyncNode* pNode, const char* format, ...);

source/libs/sync/src/syncCommit.c

@@ -84,7 +84,7 @@ void syncOneReplicaAdvance(SSyncNode* pSyncNode) {
 }
 
 void syncMaybeAdvanceCommitIndex(SSyncNode* pSyncNode) {
-  ASSERT(false && "deprecated");
+  ASSERTS(false, "deprecated");
   if (pSyncNode == NULL) {
     sError("pSyncNode is NULL");
     return;

source/libs/sync/src/syncMain.c

@@ -151,7 +151,7 @@ int32_t syncReconfig(int64_t rid, SSyncCfg* pNewCfg) {
     }
 
     syncNodeStartHeartbeatTimer(pSyncNode);
-    syncNodeReplicate(pSyncNode);
+    //syncNodeReplicate(pSyncNode);
   }
 
   syncNodeRelease(pSyncNode);
@@ -791,9 +791,9 @@ static int32_t syncHbTimerStop(SSyncNode* pSyncNode, SSyncTimer* pSyncTimer) {
 }
 
 int32_t syncNodeLogStoreRestoreOnNeed(SSyncNode* pNode) {
-  ASSERT(pNode->pLogStore != NULL && "log store not created");
+  ASSERTS(pNode->pLogStore != NULL, "log store not created");
-  ASSERT(pNode->pFsm != NULL && "pFsm not registered");
+  ASSERTS(pNode->pFsm != NULL, "pFsm not registered");
-  ASSERT(pNode->pFsm->FpGetSnapshotInfo != NULL && "FpGetSnapshotInfo not registered");
+  ASSERTS(pNode->pFsm->FpGetSnapshotInfo != NULL, "FpGetSnapshotInfo not registered");
   SSnapshot snapshot;
   if (pNode->pFsm->FpGetSnapshotInfo(pNode->pFsm, &snapshot) < 0) {
     sError("vgId:%d, failed to get snapshot info since %s", pNode->vgId, terrstr());
@@ -1144,8 +1144,8 @@ void syncNodeMaybeUpdateCommitBySnapshot(SSyncNode* pSyncNode) {
 }
 
 int32_t syncNodeRestore(SSyncNode* pSyncNode) {
-  ASSERT(pSyncNode->pLogStore != NULL && "log store not created");
+  ASSERTS(pSyncNode->pLogStore != NULL, "log store not created");
-  ASSERT(pSyncNode->pLogBuf != NULL && "ring log buffer not created");
+  ASSERTS(pSyncNode->pLogBuf != NULL, "ring log buffer not created");
 
   SyncIndex lastVer = pSyncNode->pLogStore->syncLogLastIndex(pSyncNode->pLogStore);
   SyncIndex commitIndex = pSyncNode->pLogStore->syncLogCommitIndex(pSyncNode->pLogStore);
@@ -1839,7 +1839,8 @@ void syncNodeBecomeLeader(SSyncNode* pSyncNode, const char* debugStr) {
 #endif
 
   // close receiver
-  if (snapshotReceiverIsStart(pSyncNode->pNewNodeReceiver)) {
+  if (pSyncNode != NULL && pSyncNode->pNewNodeReceiver != NULL &&
+      snapshotReceiverIsStart(pSyncNode->pNewNodeReceiver)) {
     snapshotReceiverForceStop(pSyncNode->pNewNodeReceiver);
   }
 
@@ -2663,7 +2664,7 @@ int32_t syncNodeOnClientRequest(SSyncNode* ths, SRpcMsg* pMsg, SyncIndex* pRetIn
 
     int32_t code = syncNodeAppend(ths, pEntry);
     if (code < 0 && ths->vgId != 1 && vnodeIsMsgBlock(pEntry->originalRpcType)) {
-      ASSERT(false && "failed to append blocking msg");
+      ASSERTS(false, "failed to append blocking msg");
     }
     return code;
   }

source/libs/sync/src/syncPipeline.c

@@ -50,7 +50,7 @@ int32_t syncLogBufferAppend(SSyncLogBuffer* pBuf, SSyncNode* pNode, SSyncRaftEnt
 
   // initial log buffer with at least one item, e.g. commitIndex
   SSyncRaftEntry* pMatch = pBuf->entries[(index - 1 + pBuf->size) % pBuf->size].pItem;
-  ASSERT(pMatch != NULL && "no matched log entry");
+  ASSERTS(pMatch != NULL, "no matched log entry");
   ASSERT(pMatch->index + 1 == index);
 
   SSyncLogBufEntry tmp = {.pItem = pEntry, .prevLogIndex = pMatch->index, .prevLogTerm = pMatch->term};
@@ -86,14 +86,14 @@ SyncTerm syncLogReplMgrGetPrevLogTerm(SSyncLogReplMgr* pMgr, SSyncNode* pNode, S
 
   if (prevIndex >= pBuf->startIndex) {
     pEntry = pBuf->entries[(prevIndex + pBuf->size) % pBuf->size].pItem;
-    ASSERT(pEntry != NULL && "no log entry found");
+    ASSERTS(pEntry != NULL, "no log entry found");
     prevLogTerm = pEntry->term;
     return prevLogTerm;
   }
 
   if (pMgr && pMgr->startIndex <= prevIndex && prevIndex < pMgr->endIndex) {
     int64_t timeMs = pMgr->states[(prevIndex + pMgr->size) % pMgr->size].timeMs;
-    ASSERT(timeMs != 0 && "no log entry found");
+    ASSERTS(timeMs != 0, "no log entry found");
     prevLogTerm = pMgr->states[(prevIndex + pMgr->size) % pMgr->size].term;
     ASSERT(prevIndex == 0 || prevLogTerm != 0);
     return prevLogTerm;
@@ -141,9 +141,9 @@ int32_t syncLogValidateAlignmentOfCommit(SSyncNode* pNode, SyncIndex commitIndex
 }
 
 int32_t syncLogBufferInitWithoutLock(SSyncLogBuffer* pBuf, SSyncNode* pNode) {
-  ASSERT(pNode->pLogStore != NULL && "log store not created");
+  ASSERTS(pNode->pLogStore != NULL, "log store not created");
-  ASSERT(pNode->pFsm != NULL && "pFsm not registered");
+  ASSERTS(pNode->pFsm != NULL, "pFsm not registered");
-  ASSERT(pNode->pFsm->FpGetSnapshotInfo != NULL && "FpGetSnapshotInfo not registered");
+  ASSERTS(pNode->pFsm->FpGetSnapshotInfo != NULL, "FpGetSnapshotInfo not registered");
 
   SSnapshot snapshot;
   if (pNode->pFsm->FpGetSnapshotInfo(pNode->pFsm, &snapshot) < 0) {
@@ -437,7 +437,7 @@ _out:
 }
 
 int32_t syncLogFsmExecute(SSyncNode* pNode, SSyncFSM* pFsm, ESyncState role, SyncTerm term, SSyncRaftEntry* pEntry) {
-  ASSERT(pFsm->FpCommitCb != NULL && "No commit cb registered for the FSM");
+  ASSERTS(pFsm->FpCommitCb != NULL, "No commit cb registered for the FSM");
 
   if ((pNode->replicaNum == 1) && pNode->restoreFinish && pNode->vgId != 1) {
     return 0;
@@ -513,13 +513,8 @@ int32_t syncLogBufferCommit(SSyncLogBuffer* pBuf, SSyncNode* pNode, int64_t comm
     if (!syncUtilUserCommit(pEntry->originalRpcType)) {
       sInfo("vgId:%d, commit sync barrier. index: %" PRId64 ", term:%" PRId64 ", type: %s", vgId, pEntry->index,
             pEntry->term, TMSG_INFO(pEntry->originalRpcType));
-      pBuf->commitIndex = index;
-      if (!inBuf) {
-        syncEntryDestroy(pEntry);
-        pEntry = NULL;
-      }
-      continue;
     }
+
     if (syncLogFsmExecute(pNode, pFsm, role, term, pEntry) != 0) {
       sError("vgId:%d, failed to execute sync log entry. index:%" PRId64 ", term:%" PRId64
              ", role: %d, current term: %" PRId64,
@@ -905,7 +900,7 @@ int32_t syncNodeLogReplMgrInit(SSyncNode* pNode) {
     ASSERT(pNode->logReplMgrs[i] == NULL);
     pNode->logReplMgrs[i] = syncLogReplMgrCreate();
     pNode->logReplMgrs[i]->peerId = i;
-    ASSERT(pNode->logReplMgrs[i] != NULL && "Out of memory.");
+    ASSERTS(pNode->logReplMgrs[i] != NULL, "Out of memory.");
   }
   return 0;
 }

source/libs/sync/src/syncSnapshot.c

@@ -103,6 +103,7 @@ int32_t snapshotSenderStart(SSyncSnapshotSender *pSender) {
   pSender->sendingMS = 0;
   pSender->term = pSender->pSyncNode->pRaftStore->currentTerm;
   pSender->startTime = taosGetTimestampMs();
+  pSender->lastSendTime = pSender->startTime;
   pSender->finish = false;
 
   // build begin msg
@@ -201,6 +202,8 @@ int32_t snapshotSend(SSyncSnapshotSender *pSender) {
   syncNodeSendMsgById(&pMsg->destId, pSender->pSyncNode, &rpcMsg);
   syncLogSendSyncSnapshotSend(pSender->pSyncNode, pMsg, "");
 
+  pSender->lastSendTime = taosGetTimestampMs();
+
   // event log
   if (pSender->seq == SYNC_SNAPSHOT_SEQ_END) {
     sSTrace(pSender, "snapshot sender finish");
@@ -213,33 +216,36 @@ int32_t snapshotSend(SSyncSnapshotSender *pSender) {
 // send snapshot data from cache
 int32_t snapshotReSend(SSyncSnapshotSender *pSender) {
   // send current block data
+
+  // build msg
+  SRpcMsg rpcMsg = {0};
+  (void)syncBuildSnapshotSend(&rpcMsg, pSender->blockLen, pSender->pSyncNode->vgId);
+
+  SyncSnapshotSend *pMsg = rpcMsg.pCont;
+  pMsg->srcId = pSender->pSyncNode->myRaftId;
+  pMsg->destId = (pSender->pSyncNode->replicasId)[pSender->replicaIndex];
+  pMsg->term = pSender->pSyncNode->pRaftStore->currentTerm;
+  pMsg->beginIndex = pSender->snapshotParam.start;
+  pMsg->lastIndex = pSender->snapshot.lastApplyIndex;
+  pMsg->lastTerm = pSender->snapshot.lastApplyTerm;
+  pMsg->lastConfigIndex = pSender->snapshot.lastConfigIndex;
+  pMsg->lastConfig = pSender->lastConfig;
+  pMsg->seq = pSender->seq;
+
   if (pSender->pCurrentBlock != NULL && pSender->blockLen > 0) {
-    // build msg
-    SRpcMsg rpcMsg = {0};
-    (void)syncBuildSnapshotSend(&rpcMsg, pSender->blockLen, pSender->pSyncNode->vgId);
-
-    SyncSnapshotSend *pMsg = rpcMsg.pCont;
-    pMsg->srcId = pSender->pSyncNode->myRaftId;
-    pMsg->destId = (pSender->pSyncNode->replicasId)[pSender->replicaIndex];
-    pMsg->term = pSender->pSyncNode->pRaftStore->currentTerm;
-    pMsg->beginIndex = pSender->snapshotParam.start;
-    pMsg->lastIndex = pSender->snapshot.lastApplyIndex;
-    pMsg->lastTerm = pSender->snapshot.lastApplyTerm;
-    pMsg->lastConfigIndex = pSender->snapshot.lastConfigIndex;
-    pMsg->lastConfig = pSender->lastConfig;
-    pMsg->seq = pSender->seq;
-
     //  pMsg->privateTerm = pSender->privateTerm;
     memcpy(pMsg->data, pSender->pCurrentBlock, pSender->blockLen);
-
-    // send msg
-    syncNodeSendMsgById(&pMsg->destId, pSender->pSyncNode, &rpcMsg);
-    syncLogSendSyncSnapshotSend(pSender->pSyncNode, pMsg, "");
-
-    // event log
-    sSTrace(pSender, "snapshot sender resend");
   }
 
+  // send msg
+  syncNodeSendMsgById(&pMsg->destId, pSender->pSyncNode, &rpcMsg);
+  syncLogSendSyncSnapshotSend(pSender->pSyncNode, pMsg, "");
+
+  pSender->lastSendTime = taosGetTimestampMs();
+
+  // event log
+  sSTrace(pSender, "snapshot sender resend");
+
   return 0;
 }
 
@@ -339,6 +345,8 @@ bool snapshotReceiverIsStart(SSyncSnapshotReceiver *pReceiver) { return pReceive
 void snapshotReceiverForceStop(SSyncSnapshotReceiver *pReceiver) {
   // force close, abandon incomplete data
   if (pReceiver->pWriter != NULL) {
+    // event log
+    sRTrace(pReceiver, "snapshot receiver force stop");
     int32_t ret = pReceiver->pSyncNode->pFsm->FpSnapshotStopWrite(pReceiver->pSyncNode->pFsm, pReceiver->pWriter, false,
                                                                   &(pReceiver->snapshot));
     ASSERT(ret == 0);
@@ -348,7 +356,7 @@ void snapshotReceiverForceStop(SSyncSnapshotReceiver *pReceiver) {
   pReceiver->start = false;
 
   // event log
-  sRTrace(pReceiver, "snapshot receiver force stop");
+  // sRTrace(pReceiver, "snapshot receiver force stop");
 }
 
 int32_t snapshotReceiverStartWriter(SSyncSnapshotReceiver *pReceiver, SyncSnapshotSend *pBeginMsg) {
@@ -669,6 +677,7 @@ static int32_t syncNodeOnSnapshotEnd(SSyncNode *pSyncNode, SyncSnapshotSend *pMs
     sNTrace(pSyncNode, "snapshot receiver finish waitting for true time, now:%" PRId64 ", stime:%" PRId64, timeNow,
             pMsg->startTime);
     taosMsleep(10);
+    timeNow = taosGetTimestampMs();
   }
 
   int32_t code = snapshotReceiverFinish(pReceiver, pMsg);

source/libs/sync/src/syncTimeout.c

@@ -20,6 +20,7 @@
 #include "syncRaftLog.h"
 #include "syncReplication.h"
 #include "syncRespMgr.h"
+#include "syncSnapshot.h"
 #include "syncUtil.h"
 
 static void syncNodeCleanConfigIndex(SSyncNode* ths) {
@@ -70,6 +71,20 @@ static int32_t syncNodeTimerRoutine(SSyncNode* ths) {
   }
 
   int64_t timeNow = taosGetTimestampMs();
+
+  for (int i = 0; i < ths->peersNum; ++i) {
+    SSyncSnapshotSender* pSender = syncNodeGetSnapshotSender(ths, &(ths->peersId[i]));
+    if (pSender != NULL) {
+      if (ths->isStart && ths->state == TAOS_SYNC_STATE_LEADER && pSender->start &&
+          timeNow - pSender->lastSendTime > SYNC_SNAP_RESEND_MS) {
+        snapshotReSend(pSender);
+      } else {
+        sTrace("vgId:%d, do not resend: nstart%d, now:%" PRId64 ", lstsend:%" PRId64 ", diff:%" PRId64, ths->vgId,
+               ths->isStart, timeNow, pSender->lastSendTime, timeNow - pSender->lastSendTime);
+      }
+    }
+  }
+
   if (atomic_load_64(&ths->snapshottingIndex) != SYNC_INDEX_INVALID) {
     // end timeout wal snapshot
     if (timeNow - ths->snapshottingTime > SYNC_DEL_WAL_MS &&

source/libs/sync/src/syncUtil.c

@@ -160,8 +160,6 @@ void syncUtilMsgNtoH(void* msg) {
 
 bool syncUtilUserPreCommit(tmsg_t msgType) { return msgType != TDMT_SYNC_NOOP && msgType != TDMT_SYNC_LEADER_TRANSFER; }
 
-bool syncUtilUserCommit(tmsg_t msgType) { return msgType != TDMT_SYNC_NOOP && msgType != TDMT_SYNC_LEADER_TRANSFER; }
-
 bool syncUtilUserRollback(tmsg_t msgType) { return msgType != TDMT_SYNC_NOOP && msgType != TDMT_SYNC_LEADER_TRANSFER; }
 
 void syncCfg2SimpleStr(const SSyncCfg* pCfg, char* buf, int32_t bufLen) {
@@ -568,7 +566,7 @@ void syncLogSendSyncSnapshotSend(SSyncNode* pSyncNode, const SyncSnapshotSend* p
   syncUtilU642Addr(pMsg->destId.addr, host, sizeof(host), &port);
 
   sNTrace(pSyncNode,
-          "send sync-snapshot-send from %s:%d {term:%" PRId64 ", begin:%" PRId64 ", end:%" PRId64 ", lterm:%" PRId64
+          "send sync-snapshot-send to %s:%d {term:%" PRId64 ", begin:%" PRId64 ", end:%" PRId64 ", lterm:%" PRId64
           ", stime:%" PRId64 ", seq:%d}, %s",
           host, port, pMsg->term, pMsg->beginIndex, pMsg->lastIndex, pMsg->lastTerm, pMsg->startTime, pMsg->seq, s);
 }
@@ -595,7 +593,7 @@ void syncLogSendSyncSnapshotRsp(SSyncNode* pSyncNode, const SyncSnapshotRsp* pMs
   syncUtilU642Addr(pMsg->destId.addr, host, sizeof(host), &port);
 
   sNTrace(pSyncNode,
-          "send sync-snapshot-rsp from %s:%d {term:%" PRId64 ", begin:%" PRId64 ", lst:%" PRId64 ", lterm:%" PRId64
+          "send sync-snapshot-rsp to %s:%d {term:%" PRId64 ", begin:%" PRId64 ", lst:%" PRId64 ", lterm:%" PRId64
           ", stime:%" PRId64 ", ack:%d}, %s",
           host, port, pMsg->term, pMsg->snapBeginIndex, pMsg->lastIndex, pMsg->lastTerm, pMsg->startTime, pMsg->ack, s);
 }

source/libs/sync/test/syncAppendEntriesReplyTest.cpp

@@ -19,7 +19,7 @@ SyncAppendEntriesReply *createMsg() {
   pMsg->success = true;
   pMsg->matchIndex = 77;
   pMsg->term = 33;
-  pMsg->privateTerm = 44;
+  // pMsg->privateTerm = 44;
   pMsg->startTime = taosGetTimestampMs();
   return pMsg;
 }

source/libs/sync/test/syncTest.cpp

@@ -1,5 +1,5 @@
 #include "syncTest.h"
-#include <gtest/gtest.h>
+// #include <gtest/gtest.h>
 
 /*
 typedef enum {

source/libs/sync/test/sync_test_lib/inc/syncIO.h

@@ -81,6 +81,8 @@ int32_t syncIOQTimerStop();
 int32_t syncIOPingTimerStart();
 int32_t syncIOPingTimerStop();
 
+void syncEntryDestory(SSyncRaftEntry* pEntry);
+
 #ifdef __cplusplus
 }
 #endif

source/libs/sync/test/sync_test_lib/src/syncIO.c

@@ -469,3 +469,5 @@ static void syncIOTickPing(void *param, void *tmrId) {
 
   taosTmrReset(syncIOTickPing, io->pingTimerMS, io, io->timerMgr, &io->pingTimer);
 }
+
+void syncEntryDestory(SSyncRaftEntry* pEntry) {}

source/libs/sync/test/sync_test_lib/src/syncMessageDebug.c

@@ -1583,8 +1583,8 @@ cJSON* syncAppendEntriesReply2Json(const SyncAppendEntriesReply* pMsg) {
     cJSON_AddNumberToObject(pDestId, "vgId", pMsg->destId.vgId);
     cJSON_AddItemToObject(pRoot, "destId", pDestId);
 
-    snprintf(u64buf, sizeof(u64buf), "%" PRIu64, pMsg->privateTerm);
+    // snprintf(u64buf, sizeof(u64buf), "%" PRIu64, pMsg->privateTerm);
-    cJSON_AddStringToObject(pRoot, "privateTerm", u64buf);
+    // cJSON_AddStringToObject(pRoot, "privateTerm", u64buf);
 
     snprintf(u64buf, sizeof(u64buf), "%" PRIu64, pMsg->term);
     cJSON_AddStringToObject(pRoot, "term", u64buf);

source/libs/tdb/inc/tdb.h

@@ -81,6 +81,8 @@ void tdbFree(void *);
 
 typedef struct hashset_st *hashset_t;
 
+void hashset_destroy(hashset_t set);
+
 struct STxn {
   int     flags;
   int64_t txnId;

source/libs/tdb/src/db/tdbBtree.c

@@ -110,6 +110,7 @@ int tdbBtreeOpen(int keyLen, int valLen, SPager *pPager, char const *tbname, SPg
 
     ret = tdbBegin(pEnv, &txn, tdbDefaultMalloc, tdbDefaultFree, NULL, TDB_TXN_WRITE | TDB_TXN_READ_UNCOMMITTED);
     if (ret < 0) {
+      tdbOsFree(pBt);
       return -1;
     }
 
@@ -119,6 +120,7 @@ int tdbBtreeOpen(int keyLen, int valLen, SPager *pPager, char const *tbname, SPg
     ret = tdbPagerFetchPage(pPager, &pgno, &pPage, tdbBtreeInitPage, &zArg, txn);
     if (ret < 0) {
       tdbAbort(pEnv, txn);
+      tdbOsFree(pBt);
       return -1;
     }
 
@@ -126,6 +128,7 @@ int tdbBtreeOpen(int keyLen, int valLen, SPager *pPager, char const *tbname, SPg
     if (ret < 0) {
       tdbError("failed to write page since %s", terrstr());
       tdbAbort(pEnv, txn);
+      tdbOsFree(pBt);
       return -1;
     }
 
@@ -138,6 +141,7 @@ int tdbBtreeOpen(int keyLen, int valLen, SPager *pPager, char const *tbname, SPg
       ret = tdbTbInsert(pPager->pEnv->pMainDb, tbname, strlen(tbname) + 1, &pBt->info, sizeof(pBt->info), txn);
       if (ret < 0) {
         tdbAbort(pEnv, txn);
+        tdbOsFree(pBt);
         return -1;
       }
     }

source/libs/tdb/src/db/tdbPager.c

@@ -379,9 +379,6 @@ int tdbPagerPostCommit(SPager *pPager, TXN *pTxn) {
     return -1;
   }
 
-  if (pTxn->jPageSet) {
-    hashset_destroy(pTxn->jPageSet);
-  }
   // pPager->inTran = 0;
 
   return 0;
@@ -549,8 +546,6 @@ int tdbPagerAbort(SPager *pPager, TXN *pTxn) {
     return -1;
   }
 
-  hashset_destroy(pTxn->jPageSet);
-
   // pPager->inTran = 0;
 
   return 0;
@@ -922,6 +917,8 @@ int tdbPagerRestoreJournals(SPager *pPager, SBTree *pBt) {
     char *name = tdbDirEntryBaseName(tdbGetDirEntryName(pDirEntry));
     if (strncmp(TDB_MAINDB_NAME "-journal", name, 16) == 0) {
       if (tdbPagerRestore(pPager, pBt, name) < 0) {
+        tdbCloseDir(&pDir);
+
         tdbError("failed to restore file due to %s. jFileName:%s", strerror(errno), name);
         return -1;
       }
@@ -946,6 +943,8 @@ int tdbPagerRollback(SPager *pPager) {
 
     if (strncmp(TDB_MAINDB_NAME "-journal", name, 16) == 0) {
       if (tdbOsRemove(name) < 0 && errno != ENOENT) {
+        tdbCloseDir(&pDir);
+
         tdbError("failed to remove file due to %s. jFileName:%s", strerror(errno), name);
         terrno = TAOS_SYSTEM_ERROR(errno);
         return -1;

source/libs/tdb/src/db/tdbTxn.c

@@ -30,6 +30,11 @@ int tdbTxnOpen(TXN *pTxn, int64_t txnid, void *(*xMalloc)(void *, size_t), void
 
 int tdbTxnClose(TXN *pTxn) {
   if (pTxn) {
+    if (pTxn->jPageSet) {
+      hashset_destroy(pTxn->jPageSet);
+      pTxn->jPageSet = NULL;
+    }
+
     tdbOsFree(pTxn);
   }
 

source/libs/tdb/test/tdbExOVFLTest.cpp

@@ -8,6 +8,7 @@
 #include <string>
 #include <thread>
 #include <vector>
+#include "tlog.h"
 
 typedef struct SPoolMem {
   int64_t          size;

source/libs/tdb/test/tdbTest.cpp

@@ -8,6 +8,7 @@
 #include <string>
 #include <thread>
 #include <vector>
+#include "tlog.h"
 
 typedef struct SPoolMem {
   int64_t          size;

source/libs/transport/src/transCli.c

@@ -1052,9 +1052,6 @@ void cliHandleReq(SCliMsg* pMsg, SCliThrd* pThrd) {
 
   cliMayCvtFqdnToIp(&pCtx->epSet, &pThrd->cvtAddr);
 
-  char tbuf[256] = {0};
-  EPSET_DEBUG_STR(&pCtx->epSet, tbuf);
-
   if (!EPSET_IS_VALID(&pCtx->epSet)) {
     tError("invalid epset");
     destroyCmsg(pMsg);
@@ -1493,18 +1490,35 @@ FORCE_INLINE bool cliTryExtractEpSet(STransMsg* pResp, SEpSet* dst) {
 bool cliResetEpset(STransConnCtx* pCtx, STransMsg* pResp, bool hasEpSet) {
   bool noDelay = true;
   if (hasEpSet == false) {
-    // assert(pResp->contLen == 0);
     if (pResp->contLen == 0) {
       if (pCtx->epsetRetryCnt >= pCtx->epSet.numOfEps) {
         noDelay = false;
       } else {
         EPSET_FORWARD_INUSE(&pCtx->epSet);
       }
-    } else {
+    } else if (pResp->contLen != 0) {
-      if (pCtx->epsetRetryCnt >= pCtx->epSet.numOfEps) {
+      SEpSet  epSet;
-        noDelay = false;
+      int32_t valid = tDeserializeSEpSet(pResp->pCont, pResp->contLen, &epSet);
+      if (valid < 0) {
+        tDebug("get invalid epset, epset equal, continue");
+        if (pCtx->epsetRetryCnt >= pCtx->epSet.numOfEps) {
+          noDelay = false;
+        } else {
+          EPSET_FORWARD_INUSE(&pCtx->epSet);
+        }
       } else {
-        EPSET_FORWARD_INUSE(&pCtx->epSet);
+        if (!transEpSetIsEqual(&pCtx->epSet, &epSet)) {
+          tDebug("epset not equal, retry new epset");
+          pCtx->epSet = epSet;
+          noDelay = false;
+        } else {
+          if (pCtx->epsetRetryCnt >= pCtx->epSet.numOfEps) {
+            noDelay = false;
+          } else {
+            tDebug("epset equal, continue");
+            EPSET_FORWARD_INUSE(&pCtx->epSet);
+          }
+        }
       }
     }
   } else {
@@ -1584,7 +1598,7 @@ bool cliGenRetryRule(SCliConn* pConn, STransMsg* pResp, SCliMsg* pMsg) {
     addConnToPool(pThrd->pool, pConn);
   } else if (code == TSDB_CODE_SYN_RESTORING) {
     tTrace("code str %s, contlen:%d 0", tstrerror(code), pResp->contLen);
-    noDelay = cliResetEpset(pCtx, pResp, false);
+    noDelay = cliResetEpset(pCtx, pResp, true);
     addConnToPool(pThrd->pool, pConn);
     transFreeMsg(pResp->pCont);
   } else {

source/libs/transport/src/transSvr.c

@@ -1195,6 +1195,8 @@ void transCloseServer(void* arg) {
       sendQuitToWorkThrd(srv->pThreadObj[i]);
       destroyWorkThrd(srv->pThreadObj[i]);
     }
+  } else {
+    uv_loop_close(srv->loop);
   }
 
   taosMemoryFree(srv->pThreadObj);

source/os/src/osDir.c

@@ -163,7 +163,7 @@ int32_t taosMulMkDir(const char *dirname) {
       code = mkdir(temp, 0755);
 #endif
       if (code < 0 && errno != EEXIST) {
-        terrno = TAOS_SYSTEM_ERROR(errno);
+        // terrno = TAOS_SYSTEM_ERROR(errno);
         return code;
       }
       *pos = TD_DIRSEP[0];
@@ -179,7 +179,7 @@ int32_t taosMulMkDir(const char *dirname) {
     code = mkdir(temp, 0755);
 #endif
     if (code < 0 && errno != EEXIST) {
-      terrno = TAOS_SYSTEM_ERROR(errno);
+      // terrno = TAOS_SYSTEM_ERROR(errno);
       return code;
     }
   }
@@ -225,7 +225,7 @@ int32_t taosMulModeMkDir(const char *dirname, int mode) {
       code = mkdir(temp, mode);
 #endif
       if (code < 0 && errno != EEXIST) {
-        terrno = TAOS_SYSTEM_ERROR(errno);
+        // terrno = TAOS_SYSTEM_ERROR(errno);
         return code;
       }
       *pos = TD_DIRSEP[0];
@@ -241,7 +241,7 @@ int32_t taosMulModeMkDir(const char *dirname, int mode) {
     code = mkdir(temp, mode);
 #endif
     if (code < 0 && errno != EEXIST) {
-      terrno = TAOS_SYSTEM_ERROR(errno);
+      // terrno = TAOS_SYSTEM_ERROR(errno);
       return code;
     }
   }
@@ -497,3 +497,11 @@ int32_t taosCloseDir(TdDirPtr *ppDir) {
   return 0;
 #endif
 }
+
+void taosGetCwd(char *buf, int32_t len) {
+#if !defined(WINDOWS)
+  (void)getcwd(buf, len - 1);
+#else
+  strncpy(buf, "not implemented on windows", len -1);
+#endif
+}

source/os/src/osFile.c

@@ -313,7 +313,7 @@ TdFilePtr taosOpenFile(const char *path, int32_t tdFileOptions) {
     assert(!(tdFileOptions & TD_FILE_EXCL));
     fp = fopen(path, mode);
     if (fp == NULL) {
-      terrno = TAOS_SYSTEM_ERROR(errno);
+      // terrno = TAOS_SYSTEM_ERROR(errno);
       return NULL;
     }
   } else {
@@ -336,14 +336,14 @@ TdFilePtr taosOpenFile(const char *path, int32_t tdFileOptions) {
     fd = open(path, access, S_IRWXU | S_IRWXG | S_IRWXO);
 #endif
     if (fd == -1) {
-      terrno = TAOS_SYSTEM_ERROR(errno);
+      // terrno = TAOS_SYSTEM_ERROR(errno);
       return NULL;
     }
   }
 
   TdFilePtr pFile = (TdFilePtr)taosMemoryMalloc(sizeof(TdFile));
   if (pFile == NULL) {
-    terrno = TSDB_CODE_OUT_OF_MEMORY;
+    // terrno = TSDB_CODE_OUT_OF_MEMORY;
     if (fd >= 0) close(fd);
     if (fp != NULL) fclose(fp);
     return NULL;

source/os/src/osMemory.c

@@ -348,7 +348,7 @@ void taosMemoryTrim(int32_t size) {
 
 void* taosMemoryMallocAlign(uint32_t alignment, int64_t size) {
 #ifdef USE_TD_MEMORY
-  ASSERT(0);
+  assert(0);
 #else
 #if defined(LINUX)
   void* p = memalign(alignment, size);

source/os/src/osString.c

@@ -143,15 +143,17 @@ SConv  *gConv = NULL;
 int32_t convUsed = 0;
 int32_t gConvMaxNum = 0;
 
-void taosConvInit(void) {
+int32_t taosConvInit(void) {
   gConvMaxNum = 512;
   gConv = taosMemoryCalloc(gConvMaxNum, sizeof(SConv));
   for (int32_t i = 0; i < gConvMaxNum; ++i) {
     gConv[i].conv = iconv_open(DEFAULT_UNICODE_ENCODEC, tsCharset);
     if ((iconv_t)-1 == gConv[i].conv || (iconv_t)0 == gConv[i].conv) {
-      ASSERT(0);
+      return -1;
     }
   }
+
+  return 0;
 }
 
 void taosConvDestroy() {

source/os/src/osSysinfo.c

@@ -97,6 +97,7 @@ LONG WINAPI exceptionHandler(LPEXCEPTION_POINTERS exception);
 
 #include <errno.h>
 #include <libproc.h>
+#include <sys/sysctl.h>
 
 #else
 
@@ -275,34 +276,34 @@ int32_t taosGetEmail(char *email, int32_t maxLen) {
 #endif
 }
 
+
+
 int32_t taosGetOsReleaseName(char *releaseName, int32_t maxLen) {
 #ifdef WINDOWS
   snprintf(releaseName, maxLen, "Windows");
   return 0;
 #elif defined(_TD_DARWIN_64)
-  char    line[1024];
+  char osversion[32];
-  size_t  size = 0;
+  size_t osversion_len = sizeof(osversion) - 1;
-  int32_t code = -1;
+  int osversion_name[] = { CTL_KERN, KERN_OSRELEASE };
 
-  TdFilePtr pFile = taosOpenFile("/etc/os-release", TD_FILE_READ | TD_FILE_STREAM);
+  if (sysctl(osversion_name, 2, osversion, &osversion_len, NULL, 0) == -1) {
-  if (pFile == NULL) return false;
+    return -1;
-
-  while ((size = taosGetsFile(pFile, sizeof(line), line)) != -1) {
-    line[size - 1] = '\0';
-    if (strncmp(line, "PRETTY_NAME", 11) == 0) {
-      const char *p = strchr(line, '=') + 1;
-      if (*p == '"') {
-        p++;
-        line[size - 2] = 0;
-      }
-      tstrncpy(releaseName, p, maxLen);
-      code = 0;
-      break;
-    }
   }
 
-  taosCloseFile(&pFile);
+  uint32_t major, minor;
-  return code;
+  if (sscanf(osversion, "%u.%u", &major, &minor) != 2) {
+      return -1;
+  }
+  if (major >= 20) {
+      major -= 9; // macOS 11 and newer
+      sprintf(releaseName, "macOS %u.%u", major, minor);
+  } else {
+      major -= 4; // macOS 10.1.1 and newer
+      sprintf(releaseName, "macOS 10.%d.%d", major, minor);
+  }
+
+  return 0;
 #else
   char    line[1024];
   size_t  size = 0;
@@ -617,14 +618,14 @@ int32_t taosGetDiskSize(char *dataDir, SDiskSize *diskSize) {
     return 0;
   } else {
     // printf("failed to get disk size, dataDir:%s errno:%s", tsDataDir, strerror(errno));
-    terrno = TAOS_SYSTEM_ERROR(errno);
+    // terrno = TAOS_SYSTEM_ERROR(errno);
     return -1;
   }
 #elif defined(_TD_DARWIN_64)
   struct statvfs info;
   if (statvfs(dataDir, &info)) {
     // printf("failed to get disk size, dataDir:%s errno:%s", tsDataDir, strerror(errno));
-    terrno = TAOS_SYSTEM_ERROR(errno);
+    // terrno = TAOS_SYSTEM_ERROR(errno);
     return -1;
   } else {
     diskSize->total = info.f_blocks * info.f_frsize;
@@ -635,7 +636,7 @@ int32_t taosGetDiskSize(char *dataDir, SDiskSize *diskSize) {
 #else
   struct statvfs info;
   if (statvfs(dataDir, &info)) {
-    terrno = TAOS_SYSTEM_ERROR(errno);
+    // terrno = TAOS_SYSTEM_ERROR(errno);
     return -1;
   } else {
     diskSize->total = info.f_blocks * info.f_frsize;

source/util/src/talgo.c

@@ -15,6 +15,7 @@
 
 #define _DEFAULT_SOURCE
 #include "talgo.h"
+#include "tlog.h"
 
 #define doswap(__left, __right, __size, __buf) \
   do {                                         \

source/util/src/tjson.c

@@ -181,7 +181,7 @@ int32_t tjsonGetObjectValueString(const SJson* pJson, char** pValueString) {
 int32_t tjsonGetStringValue(const SJson* pJson, const char* pName, char* pVal) {
   char* p = cJSON_GetStringValue(tjsonGetObjectItem((cJSON*)pJson, pName));
   if (NULL == p) {
-    return TSDB_CODE_FAILED;
+    return TSDB_CODE_SUCCESS;
   }
   strcpy(pVal, p);
   return TSDB_CODE_SUCCESS;
@@ -190,7 +190,7 @@ int32_t tjsonGetStringValue(const SJson* pJson, const char* pName, char* pVal) {
 int32_t tjsonDupStringValue(const SJson* pJson, const char* pName, char** pVal) {
   char* p = cJSON_GetStringValue(tjsonGetObjectItem((cJSON*)pJson, pName));
   if (NULL == p) {
-    return TSDB_CODE_FAILED;
+    return TSDB_CODE_SUCCESS;
   }
   *pVal = strdup(p);
   return TSDB_CODE_SUCCESS;
@@ -199,7 +199,7 @@ int32_t tjsonDupStringValue(const SJson* pJson, const char* pName, char** pVal)
 int32_t tjsonGetBigIntValue(const SJson* pJson, const char* pName, int64_t* pVal) {
   char* p = cJSON_GetStringValue(tjsonGetObjectItem((cJSON*)pJson, pName));
   if (NULL == p) {
-    return TSDB_CODE_FAILED;
+    return TSDB_CODE_SUCCESS;
   }
 #ifdef WINDOWS
   sscanf(p, "%" PRId64, pVal);
@@ -233,7 +233,7 @@ int32_t tjsonGetTinyIntValue(const SJson* pJson, const char* pName, int8_t* pVal
 int32_t tjsonGetUBigIntValue(const SJson* pJson, const char* pName, uint64_t* pVal) {
   char* p = cJSON_GetStringValue(tjsonGetObjectItem((cJSON*)pJson, pName));
   if (NULL == p) {
-    return TSDB_CODE_FAILED;
+    return TSDB_CODE_SUCCESS;
   }
 #ifdef WINDOWS
   sscanf(p, "%" PRIu64, pVal);
@@ -259,6 +259,9 @@ int32_t tjsonGetUTinyIntValue(const SJson* pJson, const char* pName, uint8_t* pV
 
 int32_t tjsonGetBoolValue(const SJson* pJson, const char* pName, bool* pVal) {
   const SJson* pObject = tjsonGetObjectItem(pJson, pName);
+  if (NULL == pObject) {
+    return TSDB_CODE_SUCCESS;
+  }
   if (!cJSON_IsBool(pObject)) {
     return TSDB_CODE_FAILED;
   }
@@ -268,6 +271,9 @@ int32_t tjsonGetBoolValue(const SJson* pJson, const char* pName, bool* pVal) {
 
 int32_t tjsonGetDoubleValue(const SJson* pJson, const char* pName, double* pVal) {
   const SJson* pObject = tjsonGetObjectItem(pJson, pName);
+  if (NULL == pObject) {
+    return TSDB_CODE_SUCCESS;
+  }
   if (!cJSON_IsNumber(pObject)) {
     return TSDB_CODE_FAILED;
   }
@@ -282,7 +288,7 @@ SJson* tjsonGetArrayItem(const SJson* pJson, int32_t index) { return cJSON_GetAr
 int32_t tjsonToObject(const SJson* pJson, const char* pName, FToObject func, void* pObj) {
   SJson* pJsonObj = tjsonGetObjectItem(pJson, pName);
   if (NULL == pJsonObj) {
-    return TSDB_CODE_FAILED;
+    return TSDB_CODE_SUCCESS;
   }
   return func(pJsonObj, pObj);
 }
@@ -294,7 +300,7 @@ int32_t tjsonMakeObject(const SJson* pJson, const char* pName, FToObject func, v
 
   SJson* pJsonObj = tjsonGetObjectItem(pJson, pName);
   if (NULL == pJsonObj) {
-    return TSDB_CODE_FAILED;
+    return TSDB_CODE_SUCCESS;
   }
   *pObj = taosMemoryCalloc(1, objSize);
   if (NULL == *pObj) {

source/util/src/tlog.c

@@ -72,6 +72,7 @@ static int32_t  tsDaylightActive; /* Currently in daylight saving time. */
 
 bool    tsLogEmbedded = 0;
 bool    tsAsyncLog = true;
+bool    tsAssert = true;
 int32_t tsNumOfLogLines = 10000000;
 int32_t tsLogKeepDays = 0;
 LogFp   tsLogFp = NULL;
@@ -778,3 +779,37 @@ cmp_end:
 
   return ret;
 }
+
+bool taosAssert(bool condition, const char *file, int32_t line, const char *format, ...) {
+  if (condition) return false;
+
+  const char *flags = "UTL FATAL ";
+  ELogLevel   level = DEBUG_FATAL;
+  int32_t     dflag = 255;  // tsLogEmbedded ? 255 : uDebugFlag
+  char        buffer[LOG_MAX_LINE_BUFFER_SIZE];
+  int32_t     len = taosBuildLogHead(buffer, flags);
+
+  va_list argpointer;
+  va_start(argpointer, format);
+  len = len + vsnprintf(buffer + len, LOG_MAX_LINE_BUFFER_SIZE - len, format, argpointer);
+  va_end(argpointer);
+  buffer[len++] = '\n';
+  buffer[len] = 0;
+  taosPrintLogImp(1, 255, buffer, len);
+
+  taosPrintLog(flags, level, dflag, "tAssert at file %s:%d exit:%d", file, line, tsAssert);
+  taosPrintTrace(flags, level, dflag);
+
+  if (tsAssert) {
+    taosCloseLog();
+    taosMsleep(300);
+
+#ifdef NDEBUG
+    abort();
+#else
+    assert(0);
+#endif
+  }
+
+  return true;
+}

source/util/test/hashTest.cpp

@@ -6,6 +6,7 @@
 #include "taos.h"
 #include "taosdef.h"
 #include "thash.h"
+#include "tlog.h"
 
 namespace {
 

tests/parallel_test/cases.task

@@ -1022,9 +1022,9 @@
 ,,n,develop-test,python3 ./test.py -f 5-taos-tools/taosbenchmark/auto_create_table_json.py
 ,,n,develop-test,python3 ./test.py -f 5-taos-tools/taosbenchmark/custom_col_tag.py
 ,,n,develop-test,python3 ./test.py -f 5-taos-tools/taosbenchmark/default_json.py
-#,,n,develop-test,python3 ./test.py -f 5-taos-tools/taosbenchmark/demo.py
+,,n,develop-test,python3 ./test.py -f 5-taos-tools/taosbenchmark/demo.py
 ,,n,develop-test,python3 ./test.py -f 5-taos-tools/taosbenchmark/insert_alltypes_json.py
-#,,n,develop-test,python3 ./test.py -f 5-taos-tools/taosbenchmark/invalid_commandline.py
+,,n,develop-test,python3 ./test.py -f 5-taos-tools/taosbenchmark/invalid_commandline.py
 ,,n,develop-test,python3 ./test.py -f 5-taos-tools/taosbenchmark/json_tag.py
 ,,n,develop-test,python3 ./test.py -f 5-taos-tools/taosbenchmark/query_json.py
 ,,n,develop-test,python3 ./test.py -f 5-taos-tools/taosbenchmark/sample_csv_json.py

tests/script/tsim/db/alter_replica_13.sim

@@ -116,6 +116,25 @@ endi
 
 print ============= step4: alter database
 sql alter database db replica 3
+$wt = 0
+stepwt1:
+ 	$wt = $wt + 1
+ 	sleep 1000
+ 	if $wt == 200 then
+ 	  print ====> dnode not ready!
+ 		return -1
+ 	endi
+sql show transactions
+if $rows != 0 then
+  print wait 1 seconds to alter
+  goto stepwt1
+endi
+
+sql show db.vgroups
+print ---> $data00 $data01 $data02 $data03 $data04 $data05 $data06 $data07 $data08 $data09
+print ---> $data10 $data11 $data12 $data13 $data14 $data15 $data16 $data27 $data28 $data29
+print ---> $data10 $data11 $data12 $data13 $data14 $data15 $data26 $data37 $data38 $data39
+print ---> $data10 $data11 $data12 $data13 $data14 $data15 $data36 $data47 $data48 $data49
 
 $leaderIndex = 0
 

tests/script/tsim/db/alter_replica_31.sim

@@ -148,6 +148,26 @@ endi
 
 print ============= step3: alter database
 sql alter database db replica 1
+$wt = 0
+ stepwt1:
+ 	$wt = $wt + 1
+ 	sleep 1000
+ 	if $wt == 200 then
+ 	  print ====> dnode not ready!
+ 		return -1
+ 	endi
+sql show transactions
+if $rows != 0 then
+  print wait 1 seconds to alter
+  goto stepwt1
+endi
+
+sql show db.vgroups
+print ---> $data00 $data01 $data02 $data03 $data04 $data05 $data06 $data07 $data08 $data09
+print ---> $data10 $data11 $data12 $data13 $data14 $data15 $data16 $data27 $data28 $data29
+print ---> $data10 $data11 $data12 $data13 $data14 $data15 $data26 $data37 $data38 $data39
+print ---> $data10 $data11 $data12 $data13 $data14 $data15 $data36 $data47 $data48 $data49
+
 $hasleader = 0
 
 $x = 0

tools/shell/src/shellWebsocket.c

@@ -223,7 +223,7 @@ void shellRunSingleCommandWebsocketImp(char *command) {
       if (code == TSDB_CODE_WS_SEND_TIMEOUT || code == TSDB_CODE_WS_RECV_TIMEOUT) {
         fprintf(stderr, "Hint: use -t to increase the timeout in seconds\n");
       } else if (code == TSDB_CODE_WS_INTERNAL_ERRO || code == TSDB_CODE_WS_CLOSED) {
-        fprintf(stderr, "TDengine server is down, will try to reconnect\n");
+        fprintf(stderr, "TDengine server is disconnected, will try to reconnect\n");
         shell.ws_conn = NULL;
       }
       ws_free_result(res);

utils/test/c/sml_test.c

@@ -20,6 +20,7 @@
 #include <time.h>
 #include "taos.h"
 #include "types.h"
+#include "tlog.h"
 
 int smlProcess_influx_Test() {
   TAOS *taos = taos_connect("localhost", "root", "taosdata", NULL, 0);