From a4fd2def98108357e1097afb955cd7177ac151d0 Mon Sep 17 00:00:00 2001
From: Benguang Zhao <bgzhao@taosdata.com>
Date: Mon, 17 Oct 2022 13:30:23 +0800
Subject: [PATCH 01/42] fix: resolve a typo in errormsg printing in
 walScanLogGetLastVer

---
 source/libs/wal/src/walMeta.c | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

diff --git a/source/libs/wal/src/walMeta.c b/source/libs/wal/src/walMeta.c
index 279f4dc656..85791fa682 100644
--- a/source/libs/wal/src/walMeta.c
+++ b/source/libs/wal/src/walMeta.c
@@ -123,8 +123,8 @@ static FORCE_INLINE int64_t walScanLogGetLastVer(SWal* pWal, int32_t fileIdx) {
       }
       SWalCkHead* logContent = (SWalCkHead*)candidate;
       if (walValidHeadCksum(logContent) != 0) {
-        wError("vgId:%d, failed to validate checksum of wal entry header. offset:% %" PRId64 ", file:%s",
-               ((char*)(logContent)-buf), fnameStr);
+        wWarn("vgId:%d, failed to validate checksum of wal entry header. offset:%" PRId64 ", file:%s", pWal->cfg.vgId,
+               offset + ((char*)(logContent)-buf), fnameStr);
         haystack = candidate + 1;
         if (firstTrial) {
           break;
@@ -162,8 +162,8 @@ static FORCE_INLINE int64_t walScanLogGetLastVer(SWal* pWal, int32_t fileIdx) {
       }
       if (walValidBodyCksum(logContent) != 0) {
         terrno = TSDB_CODE_WAL_CHKSUM_MISMATCH;
-        wError("vgId:%d, failed to validate checksum of wal entry body. offset:% %" PRId64 ", file:%s",
-               ((char*)(logContent)-buf), fnameStr);
+        wWarn("vgId:%d, failed to validate checksum of wal entry body. offset:%" PRId64 ", file:%s", pWal->cfg.vgId,
+               offset + ((char*)(logContent)-buf), fnameStr);
         haystack = candidate + 1;
         if (firstTrial) {
           break;

From 23a4a3767779d2c95b01787a5db25e88a9ca770e Mon Sep 17 00:00:00 2001
From: Benguang Zhao <bgzhao@taosdata.com>
Date: Mon, 17 Oct 2022 14:17:34 +0800
Subject: [PATCH 02/42] enh: not allowed to start when disk space unavailable

---
 source/dnode/mgmt/node_mgmt/src/dmEnv.c | 17 ++++++++++++++++-
 1 file changed, 16 insertions(+), 1 deletion(-)

diff --git a/source/dnode/mgmt/node_mgmt/src/dmEnv.c b/source/dnode/mgmt/node_mgmt/src/dmEnv.c
index 076826ebc2..becefcd6a0 100644
--- a/source/dnode/mgmt/node_mgmt/src/dmEnv.c
+++ b/source/dnode/mgmt/node_mgmt/src/dmEnv.c
@@ -51,6 +51,7 @@ static int32_t dmInitMonitor() {
 
 static bool dmCheckDiskSpace() {
   osUpdate();
+  // sufficiency
   if (!osDataSpaceSufficient()) {
     dWarn("free data disk size: %f GB, not sufficient, expected %f GB at least", (double)tsDataSpace.size.avail / 1024.0 / 1024.0 / 1024.0, (double)tsDataSpace.reserved / 1024.0 / 1024.0 / 1024.0);
   }
@@ -60,7 +61,21 @@ static bool dmCheckDiskSpace() {
   if (!osTempSpaceSufficient()) {
     dWarn("free temp disk size: %f GB, not sufficient, expected %f GB at least", (double)tsTempSpace.size.avail / 1024.0 / 1024.0 / 1024.0, (double)tsTempSpace.reserved / 1024.0 / 1024.0 / 1024.0);
   }
-  return true;
+  // availability
+  bool ret = true;
+  if (!osDataSpaceAvailable()) {
+    dError("data disk space unavailable, i.e. %s", tsDataDir);
+    ret = false;
+  }
+  if (!osLogSpaceAvailable()) {
+    dError("log disk space unavailable, i.e. %s", tsLogDir);
+    ret = false;
+  }
+  if (!osTempSpaceAvailable()) {
+    dError("temp disk space unavailable, i.e. %s", tsTempDir);
+    ret = false;
+  }
+  return ret;
 }
 
 static bool dmCheckDataDirVersion() {

From 8fe672feab8edac2bece5b3843a9dd13932ba4c7 Mon Sep 17 00:00:00 2001
From: Shengliang Guan <slguan@taosdata.com>
Date: Mon, 17 Oct 2022 14:34:51 +0800
Subject: [PATCH 03/42] fix: crash whle reconfig mnode will cause taosd reboot
 failure

---
 source/dnode/mnode/impl/src/mndMnode.c | 30 ++++++++++++++------------
 source/libs/sync/src/syncMain.c        | 18 ++++++++++++++++
 2 files changed, 34 insertions(+), 14 deletions(-)

diff --git a/source/dnode/mnode/impl/src/mndMnode.c b/source/dnode/mnode/impl/src/mndMnode.c
index c27241317c..63ce538536 100644
--- a/source/dnode/mnode/impl/src/mndMnode.c
+++ b/source/dnode/mnode/impl/src/mndMnode.c
@@ -394,7 +394,6 @@ static int32_t mndCreateMnode(SMnode *pMnode, SRpcMsg *pReq, SDnodeObj *pDnode,
   if (mndSetCreateMnodeRedoActions(pMnode, pTrans, pDnode, &mnodeObj) != 0) goto _OVER;
   if (mndSetCreateMnodeRedoLogs(pMnode, pTrans, &mnodeObj) != 0) goto _OVER;
   if (mndSetCreateMnodeCommitLogs(pMnode, pTrans, &mnodeObj) != 0) goto _OVER;
-  if (mndTransAppendNullLog(pTrans) != 0) goto _OVER;
   if (mndTransPrepare(pMnode, pTrans) != 0) goto _OVER;
 
   code = 0;
@@ -478,7 +477,6 @@ static int32_t mndSetDropMnodeCommitLogs(SMnode *pMnode, STrans *pTrans, SMnodeO
 static int32_t mndSetDropMnodeRedoActions(SMnode *pMnode, STrans *pTrans, SDnodeObj *pDnode, SMnodeObj *pObj) {
   SSdb          *pSdb = pMnode->pSdb;
   void          *pIter = NULL;
-  int32_t        numOfReplicas = 0;
   SDDropMnodeReq dropReq = {0};
   SEpSet         dropEpSet = {0};
 
@@ -505,9 +503,8 @@ static int32_t mndSetDropMnodeRedoActions(SMnode *pMnode, STrans *pTrans, SDnode
 
 int32_t mndSetDropMnodeInfoToTrans(SMnode *pMnode, STrans *pTrans, SMnodeObj *pObj) {
   if (pObj == NULL) return 0;
-  if (mndSetDropMnodeCommitLogs(pMnode, pTrans, pObj) != 0) return -1;
   if (mndSetDropMnodeRedoActions(pMnode, pTrans, pObj->pDnode, pObj) != 0) return -1;
-  if (mndTransAppendNullLog(pTrans) != 0) return -1;
+  if (mndSetDropMnodeCommitLogs(pMnode, pTrans, pObj) != 0) return -1;
   return 0;
 }
 
@@ -715,7 +712,8 @@ static void mndReloadSyncConfig(SMnode *pMnode) {
   SMnodeObj *pObj = NULL;
   ESdbStatus objStatus = 0;
   void      *pIter = NULL;
-  bool       hasUpdatingMnode = false;
+  int32_t    updatingMnodes = 0;
+  int32_t    readyMnodes = 0;
   SSyncCfg   cfg = {.myIndex = -1};
 
   while (1) {
@@ -723,7 +721,11 @@ static void mndReloadSyncConfig(SMnode *pMnode) {
     if (pIter == NULL) break;
     if (objStatus == SDB_STATUS_CREATING || objStatus == SDB_STATUS_DROPPING) {
       mInfo("vgId:1, has updating mnode:%d, status:%s", pObj->id, sdbStatusName(objStatus));
-      hasUpdatingMnode = true;
+      updatingMnodes++;
+    }
+    if (objStatus == SDB_STATUS_READY) {
+      mInfo("vgId:1, has ready mnode:%d, status:%s", pObj->id, sdbStatusName(objStatus));
+      readyMnodes++;
     }
 
     if (objStatus == SDB_STATUS_READY || objStatus == SDB_STATUS_CREATING) {
@@ -739,18 +741,18 @@ static void mndReloadSyncConfig(SMnode *pMnode) {
     sdbReleaseLock(pSdb, pObj, false);
   }
 
+  if (readyMnodes <= 0 || updatingMnodes <= 0) {
+    mInfo("vgId:1, mnode sync not reconfig since readyMnodes:%d updatingMnodes:%d", readyMnodes, updatingMnodes);
+    return;
+  }
+
   if (cfg.myIndex == -1) {
-    mInfo("vgId:1, mnode not reload since selfIndex is -1");
+    mInfo("vgId:1, mnode sync not reconfig since selfIndex is -1");
     return;
   }
 
-  if (!mndGetRestored(pMnode)) {
-    mInfo("vgId:1, mnode not reload since restore not finished");
-    return;
-  }
-
-  if (hasUpdatingMnode) {
-    mInfo("vgId:1, start to reload mnode sync, replica:%d myIndex:%d", cfg.replicaNum, cfg.myIndex);
+  if (updatingMnodes > 0) {
+    mInfo("vgId:1, mnode sync reconfig, replica:%d myIndex:%d", cfg.replicaNum, cfg.myIndex);
     for (int32_t i = 0; i < cfg.replicaNum; ++i) {
       SNodeInfo *pNode = &cfg.nodeInfo[i];
       mInfo("vgId:1, index:%d, fqdn:%s port:%d", i, pNode->nodeFqdn, pNode->nodePort);
diff --git a/source/libs/sync/src/syncMain.c b/source/libs/sync/src/syncMain.c
index 907eb21f4c..b255d83933 100644
--- a/source/libs/sync/src/syncMain.c
+++ b/source/libs/sync/src/syncMain.c
@@ -1891,8 +1891,26 @@ inline bool syncNodeInConfig(SSyncNode* pSyncNode, const SSyncCfg* config) {
   return b1;
 }
 
+static bool syncIsConfigChanged(const SSyncCfg* pOldCfg, const SSyncCfg* pNewCfg) {
+  if (pOldCfg->replicaNum != pNewCfg->replicaNum) return true;
+  if (pOldCfg->myIndex != pNewCfg->myIndex) return true;
+  for (int32_t i = 0; i < pOldCfg->replicaNum; ++i) {
+    const SNodeInfo* pOldInfo = &pOldCfg->nodeInfo[i];
+    const SNodeInfo* pNewInfo = &pNewCfg->nodeInfo[i];
+    if (strcmp(pOldInfo->nodeFqdn, pNewInfo->nodeFqdn) != 0) return true;
+    if (pOldInfo->nodePort != pNewInfo->nodePort) return true;
+  }
+
+  return false;
+}
+
 void syncNodeDoConfigChange(SSyncNode* pSyncNode, SSyncCfg* pNewConfig, SyncIndex lastConfigChangeIndex) {
   SSyncCfg oldConfig = pSyncNode->pRaftCfg->cfg;
+  if (!syncIsConfigChanged(&oldConfig, pNewConfig)) {
+    sInfo("vgId:1, sync not reconfig since not changed");
+    return;
+  }
+
   pSyncNode->pRaftCfg->cfg = *pNewConfig;
   pSyncNode->pRaftCfg->lastConfigIndex = lastConfigChangeIndex;
 

From 0029ddd232c2ebb338996b10538c3ecdb7a0f9be Mon Sep 17 00:00:00 2001
From: wangmm0220 <wangmm0220@gmail.com>
Date: Mon, 17 Oct 2022 17:48:30 +0800
Subject: [PATCH 04/42] feat:add new interface for schemaless

---
 include/client/taos.h         |   1 +
 source/client/src/clientSml.c | 220 ++++++++++++++++++++++------------
 2 files changed, 145 insertions(+), 76 deletions(-)

diff --git a/include/client/taos.h b/include/client/taos.h
index 270b647a77..1182a9c2f7 100644
--- a/include/client/taos.h
+++ b/include/client/taos.h
@@ -198,6 +198,7 @@ DLL_EXPORT const void *taos_get_raw_block(TAOS_RES *res);
 
 DLL_EXPORT int       taos_load_table_info(TAOS *taos, const char *tableNameList);
 DLL_EXPORT TAOS_RES *taos_schemaless_insert(TAOS *taos, char *lines[], int numLines, int protocol, int precision);
+DLL_EXPORT TAOS_RES *taos_schemaless_insert_raw(TAOS* taos, char* lines, int len, int32_t *totalRows, int protocol, int precision);
 
 /* --------------------------TMQ INTERFACE------------------------------- */
 
diff --git a/source/client/src/clientSml.c b/source/client/src/clientSml.c
index 24da5f7b70..a7b0aff42d 100644
--- a/source/client/src/clientSml.c
+++ b/source/client/src/clientSml.c
@@ -28,8 +28,8 @@
 #define QUOTE '"'
 #define SLASH '\\'
 
-#define JUMP_SPACE(sql)  \
-  while (*sql != '\0') { \
+#define JUMP_SPACE(sql, sqlEnd)  \
+  while (sql < sqlEnd) { \
     if (*sql == SPACE)   \
       sql++;             \
     else                 \
@@ -915,16 +915,17 @@ static int32_t smlParseValue(SSmlKv *pVal, SSmlMsgBuf *msg) {
   return TSDB_CODE_TSC_INVALID_VALUE;
 }
 
-static int32_t smlParseInfluxString(const char *sql, SSmlLineInfo *elements, SSmlMsgBuf *msg) {
+static int32_t smlParseInfluxString(const char *sql, const char *sqlEnd, SSmlLineInfo *elements, SSmlMsgBuf *msg) {
   if (!sql) return TSDB_CODE_SML_INVALID_DATA;
-  JUMP_SPACE(sql)
+  JUMP_SPACE(sql, sqlEnd)
   if (*sql == COMMA) return TSDB_CODE_SML_INVALID_DATA;
   elements->measure = sql;
 
   // parse measure
-  while (*sql != '\0') {
+  while (sql < sqlEnd) {
     if ((sql != elements->measure) && IS_SLASH_LETTER(sql)) {
-      MOVE_FORWARD_ONE(sql, strlen(sql) + 1);
+      MOVE_FORWARD_ONE(sql, sqlEnd - sql);
+      sqlEnd--;
       continue;
     }
     if (IS_COMMA(sql)) {
@@ -948,7 +949,7 @@ static int32_t smlParseInfluxString(const char *sql, SSmlLineInfo *elements, SSm
   } else {
     if (*sql == COMMA) sql++;
     elements->tags = sql;
-    while (*sql != '\0') {
+    while (sql < sqlEnd) {
       if (IS_SPACE(sql)) {
         break;
       }
@@ -959,10 +960,10 @@ static int32_t smlParseInfluxString(const char *sql, SSmlLineInfo *elements, SSm
   elements->measureTagsLen = sql - elements->measure;
 
   // parse cols
-  JUMP_SPACE(sql)
+  JUMP_SPACE(sql, sqlEnd)
   elements->cols = sql;
   bool isInQuote = false;
-  while (*sql != '\0') {
+  while (sql < sqlEnd) {
     if (IS_QUOTE(sql)) {
       isInQuote = !isInQuote;
     }
@@ -982,10 +983,10 @@ static int32_t smlParseInfluxString(const char *sql, SSmlLineInfo *elements, SSm
   }
 
   // parse timestamp
-  JUMP_SPACE(sql)
+  JUMP_SPACE(sql, sqlEnd)
   elements->timestamp = sql;
-  while (*sql != '\0') {
-    if (*sql == SPACE) {
+  while (sql < sqlEnd) {
+    if (isspace(*sql)) {
       break;
     }
     sql++;
@@ -995,8 +996,8 @@ static int32_t smlParseInfluxString(const char *sql, SSmlLineInfo *elements, SSm
   return TSDB_CODE_SUCCESS;
 }
 
-static void smlParseTelnetElement(const char **sql, const char **data, int32_t *len) {
-  while (**sql != '\0') {
+static void smlParseTelnetElement(const char **sql, const char *sqlEnd, const char **data, int32_t *len) {
+  while (*sql < sqlEnd) {
     if (**sql != SPACE && !(*data)) {
       *data = *sql;
     } else if (**sql == SPACE && *data) {
@@ -1007,19 +1008,19 @@ static void smlParseTelnetElement(const char **sql, const char **data, int32_t *
   }
 }
 
-static int32_t smlParseTelnetTags(const char *data, SArray *cols, char *childTableName, SHashObj *dumplicateKey,
+static int32_t smlParseTelnetTags(const char *data, const char *sqlEnd, SArray *cols, char *childTableName, SHashObj *dumplicateKey,
                                   SSmlMsgBuf *msg) {
   const char *sql = data;
   size_t      childTableNameLen = strlen(tsSmlChildTableName);
-  while (*sql != '\0') {
-    JUMP_SPACE(sql)
+  while (sql < sqlEnd) {
+    JUMP_SPACE(sql, sqlEnd)
     if (*sql == '\0') break;
 
     const char *key = sql;
     int32_t     keyLen = 0;
 
     // parse key
-    while (*sql != '\0') {
+    while (sql < sqlEnd) {
       if (*sql == SPACE) {
         smlBuildInvalidDataMsg(msg, "invalid data", sql);
         return TSDB_CODE_SML_INVALID_DATA;
@@ -1044,7 +1045,7 @@ static int32_t smlParseTelnetTags(const char *data, SArray *cols, char *childTab
     // parse value
     const char *value = sql;
     int32_t     valueLen = 0;
-    while (*sql != '\0') {
+    while (sql < sqlEnd) {
       // parse value
       if (*sql == SPACE) {
         break;
@@ -1089,11 +1090,11 @@ static int32_t smlParseTelnetTags(const char *data, SArray *cols, char *childTab
 }
 
 // format: <metric> <timestamp> <value> <tagk_1>=<tagv_1>[ <tagk_n>=<tagv_n>]
-static int32_t smlParseTelnetString(SSmlHandle *info, const char *sql, SSmlTableInfo *tinfo, SArray *cols) {
+static int32_t smlParseTelnetString(SSmlHandle *info, const char *sql, const char *sqlEnd, SSmlTableInfo *tinfo, SArray *cols) {
   if (!sql) return TSDB_CODE_SML_INVALID_DATA;
 
   // parse metric
-  smlParseTelnetElement(&sql, &tinfo->sTableName, &tinfo->sTableNameLen);
+  smlParseTelnetElement(&sql, sqlEnd, &tinfo->sTableName, &tinfo->sTableNameLen);
   if (!(tinfo->sTableName) || IS_INVALID_TABLE_LEN(tinfo->sTableNameLen)) {
     smlBuildInvalidDataMsg(&info->msgBuf, "invalid data", sql);
     return TSDB_CODE_TSC_INVALID_TABLE_ID_LENGTH;
@@ -1102,7 +1103,7 @@ static int32_t smlParseTelnetString(SSmlHandle *info, const char *sql, SSmlTable
   // parse timestamp
   const char *timestamp = NULL;
   int32_t     tLen = 0;
-  smlParseTelnetElement(&sql, &timestamp, &tLen);
+  smlParseTelnetElement(&sql, sqlEnd, &timestamp, &tLen);
   if (!timestamp || tLen == 0) {
     smlBuildInvalidDataMsg(&info->msgBuf, "invalid timestamp", sql);
     return TSDB_CODE_SML_INVALID_DATA;
@@ -1117,7 +1118,7 @@ static int32_t smlParseTelnetString(SSmlHandle *info, const char *sql, SSmlTable
   // parse value
   const char *value = NULL;
   int32_t     valueLen = 0;
-  smlParseTelnetElement(&sql, &value, &valueLen);
+  smlParseTelnetElement(&sql, sqlEnd, &value, &valueLen);
   if (!value || valueLen == 0) {
     smlBuildInvalidDataMsg(&info->msgBuf, "invalid value", sql);
     return TSDB_CODE_TSC_INVALID_VALUE;
@@ -1135,7 +1136,7 @@ static int32_t smlParseTelnetString(SSmlHandle *info, const char *sql, SSmlTable
   }
 
   // parse tags
-  ret = smlParseTelnetTags(sql, tinfo->tags, tinfo->childTableName, info->dumplicateKey, &info->msgBuf);
+  ret = smlParseTelnetTags(sql, sqlEnd, tinfo->tags, tinfo->childTableName, info->dumplicateKey, &info->msgBuf);
   if (ret != TSDB_CODE_SUCCESS) {
     smlBuildInvalidDataMsg(&info->msgBuf, "invalid data", sql);
     return ret;
@@ -2061,11 +2062,11 @@ static int32_t smlParseJSONString(SSmlHandle *info, cJSON *root, SSmlTableInfo *
 }
 /************* TSDB_SML_JSON_PROTOCOL function end **************/
 
-static int32_t smlParseInfluxLine(SSmlHandle *info, const char *sql) {
+static int32_t smlParseInfluxLine(SSmlHandle *info, const char *sql, const int len) {
   SSmlLineInfo elements = {0};
   uDebug("SML:0x%" PRIx64 " smlParseInfluxLine sql:%s, hello", info->id, sql);
 
-  int ret = smlParseInfluxString(sql, &elements, &info->msgBuf);
+  int ret = smlParseInfluxString(sql, sql + len, &elements, &info->msgBuf);
   if (ret != TSDB_CODE_SUCCESS) {
     uError("SML:0x%" PRIx64 " smlParseInfluxLine failed", info->id);
     return ret;
@@ -2170,7 +2171,7 @@ static int32_t smlParseInfluxLine(SSmlHandle *info, const char *sql) {
   return TSDB_CODE_SUCCESS;
 }
 
-static int32_t smlParseTelnetLine(SSmlHandle *info, void *data) {
+static int32_t smlParseTelnetLine(SSmlHandle *info, void *data, const int len) {
   int            ret = TSDB_CODE_SUCCESS;
   SSmlTableInfo *tinfo = smlBuildTableInfo();
   if (!tinfo) {
@@ -2184,7 +2185,7 @@ static int32_t smlParseTelnetLine(SSmlHandle *info, void *data) {
   }
 
   if (info->protocol == TSDB_SML_TELNET_PROTOCOL) {
-    ret = smlParseTelnetString(info, (const char *)data, tinfo, cols);
+    ret = smlParseTelnetString(info, (const char *)data, data + len, tinfo, cols);
   } else if (info->protocol == TSDB_SML_JSON_PROTOCOL) {
     ret = smlParseJSONString(info, (cJSON *)data, tinfo, cols);
   } else {
@@ -2275,7 +2276,7 @@ static int32_t smlParseJSON(SSmlHandle *info, char *payload) {
 
   for (int32_t i = 0; i < payloadNum; ++i) {
     cJSON *dataPoint = (payloadNum == 1 && cJSON_IsObject(root)) ? root : cJSON_GetArrayItem(root, i);
-    ret = smlParseTelnetLine(info, dataPoint);
+    ret = smlParseTelnetLine(info, dataPoint, -1);
     if (ret != TSDB_CODE_SUCCESS) {
       uError("SML:0x%" PRIx64 " Invalid JSON Payload", info->id);
       goto end;
@@ -2364,10 +2365,14 @@ static void smlPrintStatisticInfo(SSmlHandle *info) {
          info->cost.endTime - info->cost.insertRpcTime, info->cost.endTime - info->cost.parseTime);
 }
 
-static int32_t smlParseLine(SSmlHandle *info, char *lines[], int numLines) {
+static int32_t smlParseLine(SSmlHandle *info, char *lines[], char* rawLine, char* rawLineEnd, int numLines) {
   int32_t code = TSDB_CODE_SUCCESS;
   if (info->protocol == TSDB_SML_JSON_PROTOCOL) {
-    code = smlParseJSON(info, *lines);
+    if(lines){
+      code = smlParseJSON(info, *lines);
+    }else if(rawLine){
+      code = smlParseJSON(info, rawLine);
+    }
     if (code != TSDB_CODE_SUCCESS) {
       uError("SML:0x%" PRIx64 " smlParseJSON failed:%s", info->id, *lines);
       return code;
@@ -2376,10 +2381,25 @@ static int32_t smlParseLine(SSmlHandle *info, char *lines[], int numLines) {
   }
 
   for (int32_t i = 0; i < numLines; ++i) {
+    char *tmp = NULL;
+    int  len  = 0;
+    if(lines){
+      tmp = lines[i];
+      len = strlen(tmp);
+    }else if(rawLine){
+      tmp = rawLine;
+      while(rawLine < rawLineEnd){
+        if(*(rawLine++) == '\n'){
+          break;
+        }
+        len++;
+      }
+    }
+
     if (info->protocol == TSDB_SML_LINE_PROTOCOL) {
-      code = smlParseInfluxLine(info, lines[i]);
+      code = smlParseInfluxLine(info, tmp, len);
     } else if (info->protocol == TSDB_SML_TELNET_PROTOCOL) {
-      code = smlParseTelnetLine(info, lines[i]);
+      code = smlParseTelnetLine(info, tmp, len);
     } else {
       ASSERT(0);
     }
@@ -2391,13 +2411,13 @@ static int32_t smlParseLine(SSmlHandle *info, char *lines[], int numLines) {
   return code;
 }
 
-static int smlProcess(SSmlHandle *info, char *lines[], int numLines) {
+static int smlProcess(SSmlHandle *info, char *lines[], char* rawLine, char* rawLineEnd, int numLines) {
   int32_t code = TSDB_CODE_SUCCESS;
   int32_t retryNum = 0;
 
   info->cost.parseTime = taosGetTimestampUs();
 
-  code = smlParseLine(info, lines, numLines);
+  code = smlParseLine(info, lines, rawLine, rawLineEnd, numLines);
   if (code != 0) {
     uError("SML:0x%" PRIx64 " smlParseLine error : %s", info->id, tstrerror(code));
     return code;
@@ -2490,39 +2510,8 @@ static void smlInsertCallback(void *param, void *res, int32_t code) {
   smlDestroyInfo(info);
 }
 
-/**
- * taos_schemaless_insert() parse and insert data points into database according to
- * different protocol.
- *
- * @param $lines input array may contain multiple lines, each line indicates a data point.
- *               If protocol=2 is used input array should contain single JSON
- *               string(e.g. char *lines[] = {"$JSON_string"}). If need to insert
- *               multiple data points in JSON format, should include them in $JSON_string
- *               as a JSON array.
- * @param $numLines indicates how many data points in $lines.
- *                  If protocol = 2 is used this param will be ignored as $lines should
- *                  contain single JSON string.
- * @param $protocol indicates which protocol to use for parsing:
- *                  0 - influxDB line protocol
- *                  1 - OpenTSDB telnet line protocol
- *                  2 - OpenTSDB JSON format protocol
- * @return return zero for successful insertion. Otherwise return none-zero error code of
- *         failure reason.
- *
- */
-
-TAOS_RES *taos_schemaless_insert(TAOS *taos, char *lines[], int numLines, int protocol, int precision) {
-  if (NULL == taos) {
-    terrno = TSDB_CODE_TSC_DISCONNECTED;
-    return NULL;
-  }
-
-  SRequestObj *request = (SRequestObj *)createRequest(*(int64_t *)taos, TSDB_SQL_INSERT);
-  if (!request) {
-    uError("SML:taos_schemaless_insert error request is null");
-    return NULL;
-  }
 
+TAOS_RES *taos_schemaless_insert_inner(SRequestObj *request, char *lines[], char *rawLine, char *rawLineEnd, int numLines, int protocol, int precision) {
   int      batchs = 0;
   STscObj *pTscObj = request->pTscObj;
 
@@ -2546,12 +2535,6 @@ TAOS_RES *taos_schemaless_insert(TAOS *taos, char *lines[], int numLines, int pr
     goto end;
   }
 
-  if (!lines) {
-    request->code = TSDB_CODE_SML_INVALID_DATA;
-    smlBuildInvalidDataMsg(&msg, "lines is null", NULL);
-    goto end;
-  }
-
   if (protocol < TSDB_SML_LINE_PROTOCOL || protocol > TSDB_SML_JSON_PROTOCOL) {
     request->code = TSDB_CODE_SML_INVALID_PROTOCOL_TYPE;
     smlBuildInvalidDataMsg(&msg, "protocol invalidate", NULL);
@@ -2602,15 +2585,28 @@ TAOS_RES *taos_schemaless_insert(TAOS *taos, char *lines[], int numLines, int pr
     info->affectedRows = perBatch;
     info->pRequest->body.queryFp = smlInsertCallback;
     info->pRequest->body.param = info;
-    int32_t code = smlProcess(info, lines, perBatch);
-    lines += perBatch;
+    int32_t code = smlProcess(info, lines, rawLine, rawLineEnd, perBatch);
+    if(lines){
+      lines += perBatch;
+    }
+    if(rawLine){
+      int num = 0;
+      while(rawLine < rawLineEnd){
+        if(*(rawLine++) == '\n'){
+          num++;
+        }
+        if(num == perBatch){
+          break;
+        }
+      }
+    }
     if (code != TSDB_CODE_SUCCESS) {
       info->pRequest->body.queryFp(info, req, code);
     }
   }
   tsem_wait(&params.sem);
 
-end:
+  end:
   taosThreadSpinDestroy(&params.lock);
   tsem_destroy(&params.sem);
   //  ((STscObj *)taos)->schemalessType = 0;
@@ -2618,3 +2614,75 @@ end:
   uDebug("resultend:%s", request->msgBuf);
   return (TAOS_RES *)request;
 }
+
+/**
+ * taos_schemaless_insert() parse and insert data points into database according to
+ * different protocol.
+ *
+ * @param $lines input array may contain multiple lines, each line indicates a data point.
+ *               If protocol=2 is used input array should contain single JSON
+ *               string(e.g. char *lines[] = {"$JSON_string"}). If need to insert
+ *               multiple data points in JSON format, should include them in $JSON_string
+ *               as a JSON array.
+ * @param $numLines indicates how many data points in $lines.
+ *                  If protocol = 2 is used this param will be ignored as $lines should
+ *                  contain single JSON string.
+ * @param $protocol indicates which protocol to use for parsing:
+ *                  0 - influxDB line protocol
+ *                  1 - OpenTSDB telnet line protocol
+ *                  2 - OpenTSDB JSON format protocol
+ * @return return zero for successful insertion. Otherwise return none-zero error code of
+ *         failure reason.
+ *
+ */
+
+TAOS_RES *taos_schemaless_insert(TAOS *taos, char *lines[], int numLines, int protocol, int precision) {
+  if (NULL == taos) {
+    terrno = TSDB_CODE_TSC_DISCONNECTED;
+    return NULL;
+  }
+
+  SRequestObj *request = (SRequestObj *)createRequest(*(int64_t *)taos, TSDB_SQL_INSERT);
+  if (!request) {
+    uError("SML:taos_schemaless_insert error request is null");
+    return NULL;
+  }
+
+  if (!lines || numLines <= 0) {
+    SSmlMsgBuf msg = {ERROR_MSG_BUF_DEFAULT_SIZE, request->msgBuf};
+    request->code = TSDB_CODE_SML_INVALID_DATA;
+    smlBuildInvalidDataMsg(&msg, "lines is null", NULL);
+    return (TAOS_RES *)request;
+  }
+
+  return taos_schemaless_insert_inner(request, lines, NULL, NULL, numLines, protocol, precision);
+}
+
+TAOS_RES *taos_schemaless_insert_raw(TAOS* taos, char* lines, int len, int32_t *totalRows, int protocol, int precision){
+  if (NULL == taos) {
+    terrno = TSDB_CODE_TSC_DISCONNECTED;
+    return NULL;
+  }
+
+  SRequestObj *request = (SRequestObj *)createRequest(*(int64_t *)taos, TSDB_SQL_INSERT);
+  if (!request) {
+    uError("SML:taos_schemaless_insert error request is null");
+    return NULL;
+  }
+
+  if (!lines || len <= 0) {
+    SSmlMsgBuf msg = {ERROR_MSG_BUF_DEFAULT_SIZE, request->msgBuf};
+    request->code = TSDB_CODE_SML_INVALID_DATA;
+    smlBuildInvalidDataMsg(&msg, "lines is null", NULL);
+    return (TAOS_RES *)request;
+  }
+
+  int numLines = 0;
+  for(int i = 0; i < len; i++){
+    if(lines[i] == '\n' || i == len - 1){
+      numLines++;
+    }
+  }
+  *totalRows = numLines;
+  return taos_schemaless_insert_inner(request, NULL, lines, lines + len, numLines, protocol, precision);
+}

From b7222bc4ead1237d3f75eb7406d7baae272de7f0 Mon Sep 17 00:00:00 2001
From: wangmm0220 <wangmm0220@gmail.com>
Date: Mon, 17 Oct 2022 19:21:11 +0800
Subject: [PATCH 05/42] feat:add new interface for schemaless

---
 source/client/src/clientSml.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/source/client/src/clientSml.c b/source/client/src/clientSml.c
index a7b0aff42d..82f543f0dd 100644
--- a/source/client/src/clientSml.c
+++ b/source/client/src/clientSml.c
@@ -2648,7 +2648,7 @@ TAOS_RES *taos_schemaless_insert(TAOS *taos, char *lines[], int numLines, int pr
     return NULL;
   }
 
-  if (!lines || numLines <= 0) {
+  if (!lines) {
     SSmlMsgBuf msg = {ERROR_MSG_BUF_DEFAULT_SIZE, request->msgBuf};
     request->code = TSDB_CODE_SML_INVALID_DATA;
     smlBuildInvalidDataMsg(&msg, "lines is null", NULL);

From a0cd578668792352695545797a3ca4ebbca3b123 Mon Sep 17 00:00:00 2001
From: wangmm0220 <wangmm0220@gmail.com>
Date: Mon, 17 Oct 2022 19:37:34 +0800
Subject: [PATCH 06/42] feat:add new interface for schemaless

---
 source/client/src/clientSml.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/source/client/src/clientSml.c b/source/client/src/clientSml.c
index 55d117a3c8..fe1946d77d 100644
--- a/source/client/src/clientSml.c
+++ b/source/client/src/clientSml.c
@@ -2199,7 +2199,7 @@ static int32_t smlParseTelnetLine(SSmlHandle *info, void *data, const int len) {
   }
 
   if (info->protocol == TSDB_SML_TELNET_PROTOCOL) {
-    ret = smlParseTelnetString(info, (const char *)data, data + len, tinfo, cols);
+    ret = smlParseTelnetString(info, (const char *)data, POINTER_SHIFT(data, len), tinfo, cols);
   } else if (info->protocol == TSDB_SML_JSON_PROTOCOL) {
     ret = smlParseJSONString(info, (cJSON *)data, tinfo, cols);
   } else {

From 00080750fa99ca4a388a59ae2e2a94da0ab4ac81 Mon Sep 17 00:00:00 2001
From: Shengliang Guan <slguan@taosdata.com>
Date: Mon, 17 Oct 2022 20:40:11 +0800
Subject: [PATCH 07/42] fix: mnode may deadlock while drop another mnode if it
 crashed

---
 source/dnode/mnode/sdb/inc/sdb.h     |  4 ++
 source/dnode/mnode/sdb/src/sdb.c     | 20 +++++++
 source/dnode/mnode/sdb/src/sdbFile.c |  7 +--
 source/dnode/mnode/sdb/src/sdbHash.c | 83 +++++++++++++---------------
 4 files changed, 65 insertions(+), 49 deletions(-)

diff --git a/source/dnode/mnode/sdb/inc/sdb.h b/source/dnode/mnode/sdb/inc/sdb.h
index be93bd2f59..a2373a55ce 100644
--- a/source/dnode/mnode/sdb/inc/sdb.h
+++ b/source/dnode/mnode/sdb/inc/sdb.h
@@ -403,6 +403,10 @@ const char *sdbStatusName(ESdbStatus status);
 void        sdbPrintOper(SSdb *pSdb, SSdbRow *pRow, const char *oper);
 int32_t     sdbGetIdFromRaw(SSdb *pSdb, SSdbRaw *pRaw);
 
+void sdbWriteLock(SSdb *pSdb, int32_t type);
+void sdbReadLock(SSdb *pSdb, int32_t type);
+void sdbUnLock(SSdb *pSdb, int32_t type);
+
 #ifdef __cplusplus
 }
 #endif
diff --git a/source/dnode/mnode/sdb/src/sdb.c b/source/dnode/mnode/sdb/src/sdb.c
index 5393c42da3..e73fd28e71 100644
--- a/source/dnode/mnode/sdb/src/sdb.c
+++ b/source/dnode/mnode/sdb/src/sdb.c
@@ -181,3 +181,23 @@ void sdbGetCommitInfo(SSdb *pSdb, int64_t *index, int64_t *term, int64_t *config
          pSdb->applyIndex, pSdb->applyTerm, pSdb->applyConfig, *index, *term, *config);
 #endif
 }
+
+void sdbWriteLock(SSdb *pSdb, int32_t type) {
+  TdThreadRwlock *pLock = &pSdb->locks[type];
+  // mTrace("sdb table:%d start write lock:%p", type, pLock);
+  taosThreadRwlockWrlock(pLock);
+  // mTrace("sdb table:%d stop write lock:%p", type, pLock);
+}
+
+void sdbReadLock(SSdb *pSdb, int32_t type) {
+  TdThreadRwlock *pLock = &pSdb->locks[type];
+  // mTrace("sdb table:%d start read lock:%p", type, pLock);
+  taosThreadRwlockRdlock(pLock);
+  // mTrace("sdb table:%d stop read lock:%p", type, pLock);
+}
+
+void sdbUnLock(SSdb *pSdb, int32_t type) {
+  TdThreadRwlock *pLock = &pSdb->locks[type];
+  // mTrace("sdb table:%d unlock:%p", type, pLock);
+  taosThreadRwlockUnlock(pLock);
+}
diff --git a/source/dnode/mnode/sdb/src/sdbFile.c b/source/dnode/mnode/sdb/src/sdbFile.c
index 18dbea150b..5eedcc545a 100644
--- a/source/dnode/mnode/sdb/src/sdbFile.c
+++ b/source/dnode/mnode/sdb/src/sdbFile.c
@@ -363,9 +363,8 @@ static int32_t sdbWriteFileImp(SSdb *pSdb) {
 
     mInfo("write %s to sdb file, total %d rows", sdbTableName(i), sdbGetSize(pSdb, i));
 
-    SHashObj       *hash = pSdb->hashObjs[i];
-    TdThreadRwlock *pLock = &pSdb->locks[i];
-    taosThreadRwlockWrlock(pLock);
+    SHashObj *hash = pSdb->hashObjs[i];
+    sdbWriteLock(pSdb, i);
 
     SSdbRow **ppRow = taosHashIterate(hash, NULL);
     while (ppRow != NULL) {
@@ -410,7 +409,7 @@ static int32_t sdbWriteFileImp(SSdb *pSdb) {
       sdbFreeRaw(pRaw);
       ppRow = taosHashIterate(hash, ppRow);
     }
-    taosThreadRwlockUnlock(pLock);
+    sdbUnLock(pSdb, i);
   }
 
   if (code == 0) {
diff --git a/source/dnode/mnode/sdb/src/sdbHash.c b/source/dnode/mnode/sdb/src/sdbHash.c
index 0fee0fbca1..1e56eb87bf 100644
--- a/source/dnode/mnode/sdb/src/sdbHash.c
+++ b/source/dnode/mnode/sdb/src/sdbHash.c
@@ -133,12 +133,12 @@ static int32_t sdbGetkeySize(SSdb *pSdb, ESdbType type, const void *pKey) {
 }
 
 static int32_t sdbInsertRow(SSdb *pSdb, SHashObj *hash, SSdbRaw *pRaw, SSdbRow *pRow, int32_t keySize) {
-  TdThreadRwlock *pLock = &pSdb->locks[pRow->type];
-  taosThreadRwlockWrlock(pLock);
+  int32_t type = pRow->type;
+  sdbWriteLock(pSdb, type);
 
   SSdbRow *pOldRow = taosHashGet(hash, pRow->pObj, keySize);
   if (pOldRow != NULL) {
-    taosThreadRwlockUnlock(pLock);
+    sdbUnLock(pSdb, type);
     sdbFreeRow(pSdb, pRow, false);
     terrno = TSDB_CODE_SDB_OBJ_ALREADY_THERE;
     return terrno;
@@ -149,7 +149,7 @@ static int32_t sdbInsertRow(SSdb *pSdb, SHashObj *hash, SSdbRaw *pRaw, SSdbRow *
   sdbPrintOper(pSdb, pRow, "insert");
 
   if (taosHashPut(hash, pRow->pObj, keySize, &pRow, sizeof(void *)) != 0) {
-    taosThreadRwlockUnlock(pLock);
+    sdbUnLock(pSdb, type);
     sdbFreeRow(pSdb, pRow, false);
     terrno = TSDB_CODE_OUT_OF_MEMORY;
     return terrno;
@@ -164,12 +164,12 @@ static int32_t sdbInsertRow(SSdb *pSdb, SHashObj *hash, SSdbRaw *pRaw, SSdbRow *
       taosHashRemove(hash, pRow->pObj, keySize);
       sdbFreeRow(pSdb, pRow, false);
       terrno = code;
-      taosThreadRwlockUnlock(pLock);
+      sdbUnLock(pSdb, type);
       return terrno;
     }
   }
 
-  taosThreadRwlockUnlock(pLock);
+  sdbUnLock(pSdb, type);
 
   if (pSdb->keyTypes[pRow->type] == SDB_KEY_INT32) {
     pSdb->maxId[pRow->type] = TMAX(pSdb->maxId[pRow->type], *((int32_t *)pRow->pObj));
@@ -183,26 +183,27 @@ static int32_t sdbInsertRow(SSdb *pSdb, SHashObj *hash, SSdbRaw *pRaw, SSdbRow *
 }
 
 static int32_t sdbUpdateRow(SSdb *pSdb, SHashObj *hash, SSdbRaw *pRaw, SSdbRow *pNewRow, int32_t keySize) {
-  TdThreadRwlock *pLock = &pSdb->locks[pNewRow->type];
-  taosThreadRwlockWrlock(pLock);
+  int32_t type = pNewRow->type;
+  sdbWriteLock(pSdb, type);
 
   SSdbRow **ppOldRow = taosHashGet(hash, pNewRow->pObj, keySize);
   if (ppOldRow == NULL || *ppOldRow == NULL) {
-    taosThreadRwlockUnlock(pLock);
+    sdbUnLock(pSdb, type);
     return sdbInsertRow(pSdb, hash, pRaw, pNewRow, keySize);
   }
 
   SSdbRow *pOldRow = *ppOldRow;
   pOldRow->status = pRaw->status;
   sdbPrintOper(pSdb, pOldRow, "update");
+  sdbUnLock(pSdb, type);
 
   int32_t     code = 0;
-  SdbUpdateFp updateFp = pSdb->updateFps[pNewRow->type];
+  SdbUpdateFp updateFp = pSdb->updateFps[type];
   if (updateFp != NULL) {
     code = (*updateFp)(pSdb, pOldRow->pObj, pNewRow->pObj);
   }
 
-  taosThreadRwlockUnlock(pLock);
+  sdbUnLock(pSdb, type);
   sdbFreeRow(pSdb, pNewRow, false);
 
   pSdb->tableVer[pOldRow->type]++;
@@ -210,12 +211,12 @@ static int32_t sdbUpdateRow(SSdb *pSdb, SHashObj *hash, SSdbRaw *pRaw, SSdbRow *
 }
 
 static int32_t sdbDeleteRow(SSdb *pSdb, SHashObj *hash, SSdbRaw *pRaw, SSdbRow *pRow, int32_t keySize) {
-  TdThreadRwlock *pLock = &pSdb->locks[pRow->type];
-  taosThreadRwlockWrlock(pLock);
+  int32_t type = pRow->type;
+  sdbWriteLock(pSdb, type);
 
   SSdbRow **ppOldRow = taosHashGet(hash, pRow->pObj, keySize);
   if (ppOldRow == NULL || *ppOldRow == NULL) {
-    taosThreadRwlockUnlock(pLock);
+    sdbUnLock(pSdb, type);
     sdbFreeRow(pSdb, pRow, false);
     terrno = TSDB_CODE_SDB_OBJ_NOT_THERE;
     return terrno;
@@ -228,7 +229,7 @@ static int32_t sdbDeleteRow(SSdb *pSdb, SHashObj *hash, SSdbRaw *pRaw, SSdbRow *
 
   taosHashRemove(hash, pOldRow->pObj, keySize);
   pSdb->tableVer[pOldRow->type]++;
-  taosThreadRwlockUnlock(pLock);
+  sdbUnLock(pSdb, type);
 
   sdbFreeRow(pSdb, pRow, false);
 
@@ -282,12 +283,11 @@ void *sdbAcquire(SSdb *pSdb, ESdbType type, const void *pKey) {
   void   *pRet = NULL;
   int32_t keySize = sdbGetkeySize(pSdb, type, pKey);
 
-  TdThreadRwlock *pLock = &pSdb->locks[type];
-  taosThreadRwlockRdlock(pLock);
+  sdbReadLock(pSdb, type);
 
   SSdbRow **ppRow = taosHashGet(hash, pKey, keySize);
   if (ppRow == NULL || *ppRow == NULL) {
-    taosThreadRwlockUnlock(pLock);
+    sdbUnLock(pSdb, type);
     terrno = TSDB_CODE_SDB_OBJ_NOT_THERE;
     return NULL;
   }
@@ -310,13 +310,13 @@ void *sdbAcquire(SSdb *pSdb, ESdbType type, const void *pKey) {
       break;
   }
 
-  taosThreadRwlockUnlock(pLock);
+  sdbUnLock(pSdb, type);
   return pRet;
 }
 
 static void sdbCheckRow(SSdb *pSdb, SSdbRow *pRow) {
-  TdThreadRwlock *pLock = &pSdb->locks[pRow->type];
-  taosThreadRwlockWrlock(pLock);
+  int32_t type = pRow->type;
+  sdbWriteLock(pSdb, type);
 
   int32_t ref = atomic_sub_fetch_32(&pRow->refCount, 1);
   sdbPrintOper(pSdb, pRow, "check");
@@ -324,7 +324,7 @@ static void sdbCheckRow(SSdb *pSdb, SSdbRow *pRow) {
     sdbFreeRow(pSdb, pRow, true);
   }
 
-  taosThreadRwlockUnlock(pLock);
+  sdbUnLock(pSdb, type);
 }
 
 void sdbReleaseLock(SSdb *pSdb, void *pObj, bool lock) {
@@ -333,9 +333,9 @@ void sdbReleaseLock(SSdb *pSdb, void *pObj, bool lock) {
   SSdbRow *pRow = (SSdbRow *)((char *)pObj - sizeof(SSdbRow));
   if (pRow->type >= SDB_MAX) return;
 
-  TdThreadRwlock *pLock = &pSdb->locks[pRow->type];
+  int32_t type = pRow->type;
   if (lock) {
-    taosThreadRwlockWrlock(pLock);
+    sdbWriteLock(pSdb, type);
   }
 
   int32_t ref = atomic_sub_fetch_32(&pRow->refCount, 1);
@@ -345,7 +345,7 @@ void sdbReleaseLock(SSdb *pSdb, void *pObj, bool lock) {
   }
 
   if (lock) {
-    taosThreadRwlockUnlock(pLock);
+    sdbUnLock(pSdb, type);
   }
 }
 
@@ -357,8 +357,7 @@ void *sdbFetch(SSdb *pSdb, ESdbType type, void *pIter, void **ppObj) {
   SHashObj *hash = sdbGetHash(pSdb, type);
   if (hash == NULL) return NULL;
 
-  TdThreadRwlock *pLock = &pSdb->locks[type];
-  taosThreadRwlockRdlock(pLock);
+  sdbReadLock(pSdb, type);
 
   SSdbRow **ppRow = taosHashIterate(hash, pIter);
   while (ppRow != NULL) {
@@ -373,7 +372,7 @@ void *sdbFetch(SSdb *pSdb, ESdbType type, void *pIter, void **ppObj) {
     *ppObj = pRow->pObj;
     break;
   }
-  taosThreadRwlockUnlock(pLock);
+  sdbUnLock(pSdb, type);
 
   return ppRow;
 }
@@ -384,9 +383,8 @@ void *sdbFetchAll(SSdb *pSdb, ESdbType type, void *pIter, void **ppObj, ESdbStat
   SHashObj *hash = sdbGetHash(pSdb, type);
   if (hash == NULL) return NULL;
 
-  TdThreadRwlock *pLock = &pSdb->locks[type];
   if (lock) {
-    taosThreadRwlockRdlock(pLock);
+    sdbReadLock(pSdb, type);
   }
 
   SSdbRow **ppRow = taosHashIterate(hash, pIter);
@@ -404,7 +402,7 @@ void *sdbFetchAll(SSdb *pSdb, ESdbType type, void *pIter, void **ppObj, ESdbStat
     break;
   }
   if (lock) {
-    taosThreadRwlockUnlock(pLock);
+    sdbUnLock(pSdb, type);
   }
 
   return ppRow;
@@ -416,18 +414,17 @@ void sdbCancelFetch(SSdb *pSdb, void *pIter) {
   SHashObj *hash = sdbGetHash(pSdb, pRow->type);
   if (hash == NULL) return;
 
-  TdThreadRwlock *pLock = &pSdb->locks[pRow->type];
-  taosThreadRwlockRdlock(pLock);
+  int32_t type = pRow->type;
+  sdbReadLock(pSdb, type);
   taosHashCancelIterate(hash, pIter);
-  taosThreadRwlockUnlock(pLock);
+  sdbUnLock(pSdb, type);
 }
 
 void sdbTraverse(SSdb *pSdb, ESdbType type, sdbTraverseFp fp, void *p1, void *p2, void *p3) {
   SHashObj *hash = sdbGetHash(pSdb, type);
   if (hash == NULL) return;
 
-  TdThreadRwlock *pLock = &pSdb->locks[type];
-  taosThreadRwlockRdlock(pLock);
+  sdbReadLock(pSdb, type);
 
   SSdbRow **ppRow = taosHashIterate(hash, NULL);
   while (ppRow != NULL) {
@@ -443,17 +440,16 @@ void sdbTraverse(SSdb *pSdb, ESdbType type, sdbTraverseFp fp, void *p1, void *p2
     ppRow = taosHashIterate(hash, ppRow);
   }
 
-  taosThreadRwlockUnlock(pLock);
+  sdbUnLock(pSdb, type);
 }
 
 int32_t sdbGetSize(SSdb *pSdb, ESdbType type) {
   SHashObj *hash = sdbGetHash(pSdb, type);
   if (hash == NULL) return 0;
 
-  TdThreadRwlock *pLock = &pSdb->locks[type];
-  taosThreadRwlockRdlock(pLock);
+  sdbReadLock(pSdb, type);
   int32_t size = taosHashGetSize(hash);
-  taosThreadRwlockUnlock(pLock);
+  sdbUnLock(pSdb, type);
 
   return size;
 }
@@ -465,9 +461,7 @@ int32_t sdbGetMaxId(SSdb *pSdb, ESdbType type) {
   if (pSdb->keyTypes[type] != SDB_KEY_INT32) return -1;
 
   int32_t maxId = 0;
-
-  TdThreadRwlock *pLock = &pSdb->locks[type];
-  taosThreadRwlockRdlock(pLock);
+  sdbReadLock(pSdb, type);
 
   SSdbRow **ppRow = taosHashIterate(hash, NULL);
   while (ppRow != NULL) {
@@ -477,8 +471,7 @@ int32_t sdbGetMaxId(SSdb *pSdb, ESdbType type) {
     ppRow = taosHashIterate(hash, ppRow);
   }
 
-  taosThreadRwlockUnlock(pLock);
-
+  sdbUnLock(pSdb, type);
   maxId = TMAX(maxId, pSdb->maxId[type]);
   return maxId + 1;
 }

From 2d9c9e472c9b4e918b66b400a1da105707c259ec Mon Sep 17 00:00:00 2001
From: Shengliang Guan <slguan@taosdata.com>
Date: Mon, 17 Oct 2022 21:38:39 +0800
Subject: [PATCH 08/42] test: comment out unstable case

---
 tests/system-test/fulltest.sh | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/tests/system-test/fulltest.sh b/tests/system-test/fulltest.sh
index 39adfc9d53..4c0ef5a527 100755
--- a/tests/system-test/fulltest.sh
+++ b/tests/system-test/fulltest.sh
@@ -248,7 +248,7 @@ python3 ./test.py -f 6-cluster/5dnode3mnodeRestartDnodeInsertDataAsync.py -N 5 -
 
 python3 ./test.py -f 6-cluster/5dnode3mnodeAdd1Ddnoe.py -N 6 -M 3 -C 5
 # BUG python3 ./test.py -f 6-cluster/5dnode3mnodeStopInsert.py
-python3 ./test.py -f 6-cluster/5dnode3mnodeDrop.py -N 5
+# python3 ./test.py -f 6-cluster/5dnode3mnodeDrop.py -N 5
 python3 test.py -f 6-cluster/5dnode3mnodeStopConnect.py -N 5 -M 3
 
 python3 ./test.py -f 6-cluster/5dnode3mnodeRecreateMnode.py  -N 5 -M 3

From 3190ef65a3c43521e62cd015bd1fae6f1de9f3e5 Mon Sep 17 00:00:00 2001
From: Benguang Zhao <bgzhao@taosdata.com>
Date: Mon, 17 Oct 2022 21:13:52 +0800
Subject: [PATCH 09/42] fix: deal with misaligned idx entries in
 walCheckAndRepairIdxFile

---
 source/libs/wal/src/walMeta.c | 6 ++++++
 1 file changed, 6 insertions(+)

diff --git a/source/libs/wal/src/walMeta.c b/source/libs/wal/src/walMeta.c
index 85791fa682..fa22805df2 100644
--- a/source/libs/wal/src/walMeta.c
+++ b/source/libs/wal/src/walMeta.c
@@ -481,6 +481,10 @@ int walCheckAndRepairIdxFile(SWal* pWal, int32_t fileIdx) {
       continue;
     }
 
+    if (offset != (idxEntry.ver - pFileInfo->firstVer) * sizeof(SWalIdxEntry)) {
+      continue;
+    }
+
     if (walReadLogHead(pLogFile, idxEntry.offset, &ckHead) < 0) {
       wWarn("vgId:%d, failed to read log file since %s. file:%s, offset:%" PRId64 ", idx entry ver:%" PRId64 "",
             pWal->cfg.vgId, terrstr(), fLogNameStr, idxEntry.offset, idxEntry.ver);
@@ -493,6 +497,8 @@ int walCheckAndRepairIdxFile(SWal* pWal, int32_t fileIdx) {
   }
   offset += sizeof(SWalIdxEntry);
 
+  ASSERT(offset == (idxEntry.ver - pFileInfo->firstVer + 1) * sizeof(SWalIdxEntry));
+
   // ftruncate idx file
   if (offset < fileSize) {
     if (taosFtruncateFile(pIdxFile, offset) < 0) {

From 23e18bff76781825aed80efc4cfe90450c6ab74b Mon Sep 17 00:00:00 2001
From: Benguang Zhao <bgzhao@taosdata.com>
Date: Tue, 18 Oct 2022 01:31:19 +0800
Subject: [PATCH 10/42] fix: fsync fatal log messages

---
 source/util/src/tlog.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

diff --git a/source/util/src/tlog.c b/source/util/src/tlog.c
index 2e2300ba14..9d97cf7ab2 100644
--- a/source/util/src/tlog.c
+++ b/source/util/src/tlog.c
@@ -443,10 +443,13 @@ static inline int32_t taosBuildLogHead(char *buffer, const char *flags) {
 static inline void taosPrintLogImp(ELogLevel level, int32_t dflag, const char *buffer, int32_t len) {
   if ((dflag & DEBUG_FILE) && tsLogObj.logHandle && tsLogObj.logHandle->pFile != NULL && osLogSpaceAvailable()) {
     taosUpdateLogNums(level);
-    if (tsAsyncLog) {
+    if (tsAsyncLog && level != DEBUG_FATAL) {
       taosPushLogBuffer(tsLogObj.logHandle, buffer, len);
     } else {
       taosWriteFile(tsLogObj.logHandle->pFile, buffer, len);
+      if (level == DEBUG_FATAL) {
+        taosFsyncFile(tsLogObj.logHandle->pFile);
+      }
     }
 
     if (tsLogObj.maxLines > 0) {

From f0427c93223193fc03ea8261726b42a8159e863c Mon Sep 17 00:00:00 2001
From: Benguang Zhao <bgzhao@taosdata.com>
Date: Tue, 18 Oct 2022 01:37:55 +0800
Subject: [PATCH 11/42] fix: ftrunc on appending WAL log failure for recovery
 in walWriteImpl

---
 source/libs/wal/src/walWrite.c | 41 ++++++++++++++++++++++++----------
 1 file changed, 29 insertions(+), 12 deletions(-)

diff --git a/source/libs/wal/src/walWrite.c b/source/libs/wal/src/walWrite.c
index 1e5c63fa8d..7da88ec95d 100644
--- a/source/libs/wal/src/walWrite.c
+++ b/source/libs/wal/src/walWrite.c
@@ -408,25 +408,33 @@ END:
 
 static int32_t walWriteIndex(SWal *pWal, int64_t ver, int64_t offset) {
   SWalIdxEntry entry = {.ver = ver, .offset = offset};
-  int64_t      idxOffset = taosLSeekFile(pWal->pIdxFile, 0, SEEK_END);
+  SWalFileInfo *pFileInfo = walGetCurFileInfo(pWal);
+  ASSERT(pFileInfo != NULL);
+  ASSERT(pFileInfo->firstVer >= 0);
+  int64_t idxOffset = (entry.ver - pFileInfo->firstVer) * sizeof(SWalIdxEntry);
   wDebug("vgId:%d, write index, index:%" PRId64 ", offset:%" PRId64 ", at %" PRId64, pWal->cfg.vgId, ver, offset,
          idxOffset);
+
   int64_t size = taosWriteFile(pWal->pIdxFile, &entry, sizeof(SWalIdxEntry));
   if (size != sizeof(SWalIdxEntry)) {
+    wError("vgId:%d, failed to write idx entry due to %s. ver:%lld", pWal->cfg.vgId, strerror(errno), ver);
     terrno = TAOS_SYSTEM_ERROR(errno);
-    // TODO truncate
     return -1;
   }
   return 0;
 }
 
-// TODO  gurantee atomicity by truncate failed writing
 static FORCE_INLINE int32_t walWriteImpl(SWal *pWal, int64_t index, tmsg_t msgType, SWalSyncInfo syncMeta,
                                          const void *body, int32_t bodyLen) {
   int64_t code = 0;
 
   int64_t offset = walGetCurFileOffset(pWal);
+  SWalFileInfo *pFileInfo = walGetCurFileInfo(pWal);
+  ASSERT(pFileInfo != NULL);
 
+  if (pFileInfo->firstVer == -1) {
+    pFileInfo->firstVer = index;
+  }
   pWal->writeHead.head.version = index;
   pWal->writeHead.head.bodyLen = bodyLen;
   pWal->writeHead.head.msgType = msgType;
@@ -437,11 +445,9 @@ static FORCE_INLINE int32_t walWriteImpl(SWal *pWal, int64_t index, tmsg_t msgTy
 
   pWal->writeHead.cksumHead = walCalcHeadCksum(&pWal->writeHead);
   pWal->writeHead.cksumBody = walCalcBodyCksum(body, bodyLen);
-
   wDebug("vgId:%d, wal write log %ld, msgType: %s", pWal->cfg.vgId, index, TMSG_INFO(msgType));
 
   if (taosWriteFile(pWal->pLogFile, &pWal->writeHead, sizeof(SWalCkHead)) != sizeof(SWalCkHead)) {
-    // TODO ftruncate
     terrno = TAOS_SYSTEM_ERROR(errno);
     wError("vgId:%d, file:%" PRId64 ".log, failed to write since %s", pWal->cfg.vgId, walGetLastFileFirstVer(pWal),
            strerror(errno));
@@ -450,7 +456,6 @@ static FORCE_INLINE int32_t walWriteImpl(SWal *pWal, int64_t index, tmsg_t msgTy
   }
 
   if (taosWriteFile(pWal->pLogFile, (char *)body, bodyLen) != bodyLen) {
-    // TODO ftruncate
     terrno = TAOS_SYSTEM_ERROR(errno);
     wError("vgId:%d, file:%" PRId64 ".log, failed to write since %s", pWal->cfg.vgId, walGetLastFileFirstVer(pWal),
            strerror(errno));
@@ -460,7 +465,6 @@ static FORCE_INLINE int32_t walWriteImpl(SWal *pWal, int64_t index, tmsg_t msgTy
 
   code = walWriteIndex(pWal, index, offset);
   if (code < 0) {
-    // TODO ftruncate
     goto END;
   }
 
@@ -468,14 +472,27 @@ static FORCE_INLINE int32_t walWriteImpl(SWal *pWal, int64_t index, tmsg_t msgTy
   if (pWal->vers.firstVer == -1) pWal->vers.firstVer = index;
   pWal->vers.lastVer = index;
   pWal->totSize += sizeof(SWalCkHead) + bodyLen;
-  if (walGetCurFileInfo(pWal)->firstVer == -1) {
-    walGetCurFileInfo(pWal)->firstVer = index;
-  }
-  walGetCurFileInfo(pWal)->lastVer = index;
-  walGetCurFileInfo(pWal)->fileSize += sizeof(SWalCkHead) + bodyLen;
+  pFileInfo->lastVer = index;
+  pFileInfo->fileSize += sizeof(SWalCkHead) + bodyLen;
 
   return 0;
+
 END:
+  // recover in a reverse order
+  if (taosFtruncateFile(pWal->pLogFile, offset) < 0) {
+    wFatal("vgId:%d, failed to ftruncate logfile to offset:%lld during recovery due to %s", pWal->cfg.vgId, offset,
+           strerror(errno));
+    terrno = TAOS_SYSTEM_ERROR(errno);
+    ASSERT(0 && "failed to recover from error");
+  }
+
+  int64_t idxOffset = (index - pFileInfo->firstVer) * sizeof(SWalIdxEntry);
+  if (taosFtruncateFile(pWal->pIdxFile, idxOffset) < 0) {
+    wFatal("vgId:%d, failed to ftruncate idxfile to offset:%lld during recovery due to %s", pWal->cfg.vgId, idxOffset,
+           strerror(errno));
+    terrno = TAOS_SYSTEM_ERROR(errno);
+    ASSERT(0 && "failed to recover from error");
+  }
   return -1;
 }
 

From af9aa9d78331061da00d37fdd95ee776d860d1a9 Mon Sep 17 00:00:00 2001
From: Benguang Zhao <bgzhao@taosdata.com>
Date: Tue, 18 Oct 2022 02:12:05 +0800
Subject: [PATCH 12/42] enh: assert on alignment of idx entries in
 walWriteIndex

---
 source/libs/wal/src/walWrite.c | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/source/libs/wal/src/walWrite.c b/source/libs/wal/src/walWrite.c
index 7da88ec95d..c088145ff6 100644
--- a/source/libs/wal/src/walWrite.c
+++ b/source/libs/wal/src/walWrite.c
@@ -421,6 +421,8 @@ static int32_t walWriteIndex(SWal *pWal, int64_t ver, int64_t offset) {
     terrno = TAOS_SYSTEM_ERROR(errno);
     return -1;
   }
+
+  ASSERT(taosLSeekFile(pWal->pIdxFile, 0, SEEK_END) == idxOffset + sizeof(SWalIdxEntry) && "Offset of idx entries misaligned");
   return 0;
 }
 

From cf2b6ccbfb5544070f379040a3f01a3a782d7e1d Mon Sep 17 00:00:00 2001
From: Benguang Zhao <bgzhao@taosdata.com>
Date: Tue, 18 Oct 2022 09:58:02 +0800
Subject: [PATCH 13/42] enh: write idx ahead of writing its log entry in WAL

---
 source/libs/wal/src/walWrite.c | 10 +++++-----
 1 file changed, 5 insertions(+), 5 deletions(-)

diff --git a/source/libs/wal/src/walWrite.c b/source/libs/wal/src/walWrite.c
index c088145ff6..aa6d52174d 100644
--- a/source/libs/wal/src/walWrite.c
+++ b/source/libs/wal/src/walWrite.c
@@ -449,6 +449,11 @@ static FORCE_INLINE int32_t walWriteImpl(SWal *pWal, int64_t index, tmsg_t msgTy
   pWal->writeHead.cksumBody = walCalcBodyCksum(body, bodyLen);
   wDebug("vgId:%d, wal write log %ld, msgType: %s", pWal->cfg.vgId, index, TMSG_INFO(msgType));
 
+  code = walWriteIndex(pWal, index, offset);
+  if (code < 0) {
+    goto END;
+  }
+
   if (taosWriteFile(pWal->pLogFile, &pWal->writeHead, sizeof(SWalCkHead)) != sizeof(SWalCkHead)) {
     terrno = TAOS_SYSTEM_ERROR(errno);
     wError("vgId:%d, file:%" PRId64 ".log, failed to write since %s", pWal->cfg.vgId, walGetLastFileFirstVer(pWal),
@@ -465,11 +470,6 @@ static FORCE_INLINE int32_t walWriteImpl(SWal *pWal, int64_t index, tmsg_t msgTy
     goto END;
   }
 
-  code = walWriteIndex(pWal, index, offset);
-  if (code < 0) {
-    goto END;
-  }
-
   // set status
   if (pWal->vers.firstVer == -1) pWal->vers.firstVer = index;
   pWal->vers.lastVer = index;

From 9fe6d23462a41390493bf3bb81181a2302cc38d6 Mon Sep 17 00:00:00 2001
From: wangmm0220 <wangmm0220@gmail.com>
Date: Tue, 18 Oct 2022 11:49:03 +0800
Subject: [PATCH 14/42] feat:add new interface for schemaless

---
 source/client/src/clientSml.c  | 16 +++++--
 source/client/test/smlTest.cpp | 28 +++++------
 utils/test/c/sml_test.c        | 85 ++++++++++++++++++++++------------
 3 files changed, 82 insertions(+), 47 deletions(-)

diff --git a/source/client/src/clientSml.c b/source/client/src/clientSml.c
index fe1946d77d..9241b3c5d4 100644
--- a/source/client/src/clientSml.c
+++ b/source/client/src/clientSml.c
@@ -79,7 +79,7 @@
 #define NCHAR_ADD_LEN  3  // L"nchar"   3 means L" "
 
 #define MAX_RETRY_TIMES 5
-#define LINE_BATCH      2000
+#define LINE_BATCH      2
 //=================================================================================================
 typedef TSDB_SML_PROTOCOL_TYPE SMLProtocolType;
 
@@ -2199,7 +2199,7 @@ static int32_t smlParseTelnetLine(SSmlHandle *info, void *data, const int len) {
   }
 
   if (info->protocol == TSDB_SML_TELNET_PROTOCOL) {
-    ret = smlParseTelnetString(info, (const char *)data, POINTER_SHIFT(data, len), tinfo, cols);
+    ret = smlParseTelnetString(info, (const char *)data, (char*)data + len, tinfo, cols);
   } else if (info->protocol == TSDB_SML_JSON_PROTOCOL) {
     ret = smlParseJSONString(info, (cJSON *)data, tinfo, cols);
   } else {
@@ -2408,6 +2408,9 @@ static int32_t smlParseLine(SSmlHandle *info, char *lines[], char* rawLine, char
         }
         len++;
       }
+      if(info->protocol == TSDB_SML_LINE_PROTOCOL && tmp[0] == '#'){ // this line is comment
+        continue;
+      }
     }
 
     if (info->protocol == TSDB_SML_LINE_PROTOCOL) {
@@ -2418,7 +2421,7 @@ static int32_t smlParseLine(SSmlHandle *info, char *lines[], char* rawLine, char
       ASSERT(0);
     }
     if (code != TSDB_CODE_SUCCESS) {
-      uError("SML:0x%" PRIx64 " smlParseLine failed. line %d : %s", info->id, i, lines[i]);
+      uError("SML:0x%" PRIx64 " smlParseLine failed. line %d : %s", info->id, i, tmp);
       return code;
     }
   }
@@ -2692,11 +2695,16 @@ TAOS_RES *taos_schemaless_insert_raw(TAOS* taos, char* lines, int len, int32_t *
   }
 
   int numLines = 0;
+  *totalRows = 0;
+  char *tmp = lines;
   for(int i = 0; i < len; i++){
     if(lines[i] == '\n' || i == len - 1){
       numLines++;
+      if(tmp[0] != '#' || protocol != TSDB_SML_LINE_PROTOCOL){    //ignore comment
+        (*totalRows)++;
+      }
+      tmp = lines + i + 1;
     }
   }
-  *totalRows = numLines;
   return taos_schemaless_insert_inner(request, NULL, lines, lines + len, numLines, protocol, precision);
 }
diff --git a/source/client/test/smlTest.cpp b/source/client/test/smlTest.cpp
index ab4442eca7..54778e87a7 100644
--- a/source/client/test/smlTest.cpp
+++ b/source/client/test/smlTest.cpp
@@ -44,7 +44,7 @@ TEST(testCase, smlParseInfluxString_Test) {
   char *tmp = "\\,st,t1=3,t2=4,t3=t3 c1=3i64,c3=\"passit hello,c1=2\",c2=false,c4=4f64 1626006833639000000    ,32,c=3";
   char *sql = (char *)taosMemoryCalloc(256, 1);
   memcpy(sql, tmp, strlen(tmp) + 1);
-  int ret = smlParseInfluxString(sql, &elements, &msgBuf);
+  int ret = smlParseInfluxString(sql, sql + strlen(sql), &elements, &msgBuf);
   ASSERT_EQ(ret, 0);
   ASSERT_EQ(elements.measure, sql);
   ASSERT_EQ(elements.measureLen, strlen(",st"));
@@ -63,14 +63,14 @@ TEST(testCase, smlParseInfluxString_Test) {
   tmp = "st,t1=3,t2=4,t3=t3 c1=3i64,c3=\"passit hello,c1=2,c2=false,c4=4f64 1626006833639000000";
   memcpy(sql, tmp, strlen(tmp) + 1);
   memset(&elements, 0, sizeof(SSmlLineInfo));
-  ret = smlParseInfluxString(sql, &elements, &msgBuf);
+  ret = smlParseInfluxString(sql, sql + strlen(sql), &elements, &msgBuf);
   ASSERT_NE(ret, 0);
 
   // case 3  false
   tmp = "st, t1=3,t2=4,t3=t3 c1=3i64,c3=\"passit hello,c1=2,c2=false,c4=4f64 1626006833639000000";
   memcpy(sql, tmp, strlen(tmp) + 1);
   memset(&elements, 0, sizeof(SSmlLineInfo));
-  ret = smlParseInfluxString(sql, &elements, &msgBuf);
+  ret = smlParseInfluxString(sql, sql + strlen(sql), &elements, &msgBuf);
   ASSERT_EQ(ret, 0);
   ASSERT_EQ(elements.cols, sql + elements.measureTagsLen + 1);
   ASSERT_EQ(elements.colsLen, strlen("t1=3,t2=4,t3=t3"));
@@ -79,7 +79,7 @@ TEST(testCase, smlParseInfluxString_Test) {
   tmp = "st, c1=3i64,c3=\"passit hello,c1=2\",c2=false,c4=4f64 1626006833639000000";
   memcpy(sql, tmp, strlen(tmp) + 1);
   memset(&elements, 0, sizeof(SSmlLineInfo));
-  ret = smlParseInfluxString(sql, &elements, &msgBuf);
+  ret = smlParseInfluxString(sql, sql + strlen(sql), &elements, &msgBuf);
   ASSERT_EQ(ret, 0);
   ASSERT_EQ(elements.measure, sql);
   ASSERT_EQ(elements.measureLen, strlen("st"));
@@ -98,7 +98,7 @@ TEST(testCase, smlParseInfluxString_Test) {
   tmp = " st   c1=3i64,c3=\"passit hello,c1=2\",c2=false,c4=4f64  1626006833639000000 ";
   memcpy(sql, tmp, strlen(tmp) + 1);
   memset(&elements, 0, sizeof(SSmlLineInfo));
-  ret = smlParseInfluxString(sql, &elements, &msgBuf);
+  ret = smlParseInfluxString(sql, sql + strlen(sql), &elements, &msgBuf);
   ASSERT_EQ(ret, 0);
   ASSERT_EQ(elements.measure, sql + 1);
   ASSERT_EQ(elements.measureLen, strlen("st"));
@@ -116,21 +116,21 @@ TEST(testCase, smlParseInfluxString_Test) {
   tmp = " st   c1=3i64,c3=\"passit hello,c1=2\",c2=false,c4=4f64   ";
   memcpy(sql, tmp, strlen(tmp) + 1);
   memset(&elements, 0, sizeof(SSmlLineInfo));
-  ret = smlParseInfluxString(sql, &elements, &msgBuf);
+  ret = smlParseInfluxString(sql, sql + strlen(sql), &elements, &msgBuf);
   ASSERT_EQ(ret, 0);
 
   // case 7
   tmp = " st   ,   ";
   memcpy(sql, tmp, strlen(tmp) + 1);
   memset(&elements, 0, sizeof(SSmlLineInfo));
-  ret = smlParseInfluxString(sql, &elements, &msgBuf);
+  ret = smlParseInfluxString(sql, sql + strlen(sql), &elements, &msgBuf);
   ASSERT_EQ(ret, 0);
 
   // case 8 false
   tmp = ", st   ,   ";
   memcpy(sql, tmp, strlen(tmp) + 1);
   memset(&elements, 0, sizeof(SSmlLineInfo));
-  ret = smlParseInfluxString(sql, &elements, &msgBuf);
+  ret = smlParseInfluxString(sql, sql + strlen(sql), &elements, &msgBuf);
   ASSERT_NE(ret, 0);
   taosMemoryFree(sql);
 }
@@ -542,7 +542,7 @@ TEST(testCase, smlParseTelnetLine_error_Test) {
       "sys.procs.running 1479496100 42 host= web01",
   };
   for (int i = 0; i < sizeof(sql) / sizeof(sql[0]); i++) {
-    int ret = smlParseTelnetLine(info, (void *)sql[i]);
+    int ret = smlParseTelnetLine(info, (void *)sql[i], strlen(sql[i]));
     ASSERT_NE(ret, 0);
   }
 
@@ -561,7 +561,7 @@ TEST(testCase, smlParseTelnetLine_diff_type_Test) {
 
   int ret = TSDB_CODE_SUCCESS;
   for (int i = 0; i < sizeof(sql) / sizeof(sql[0]); i++) {
-    ret = smlParseTelnetLine(info, (void *)sql[i]);
+    ret = smlParseTelnetLine(info, (void *)sql[i], strlen(sql[i]));
     if (ret != TSDB_CODE_SUCCESS) break;
   }
   ASSERT_NE(ret, 0);
@@ -617,7 +617,7 @@ TEST(testCase, smlParseTelnetLine_json_error_Test) {
 
   int ret = TSDB_CODE_SUCCESS;
   for (int i = 0; i < sizeof(sql) / sizeof(sql[0]); i++) {
-    ret = smlParseTelnetLine(info, (void *)sql[i]);
+    ret = smlParseTelnetLine(info, (void *)sql[i], strlen(sql[i]));
     ASSERT_NE(ret, 0);
   }
 
@@ -653,7 +653,7 @@ TEST(testCase, smlParseTelnetLine_diff_json_type1_Test) {
 
   int ret = TSDB_CODE_SUCCESS;
   for (int i = 0; i < sizeof(sql) / sizeof(sql[0]); i++) {
-    ret = smlParseTelnetLine(info, (void *)sql[i]);
+    ret = smlParseTelnetLine(info, (void *)sql[i], strlen(sql[i]));
     if (ret != TSDB_CODE_SUCCESS) break;
   }
   ASSERT_NE(ret, 0);
@@ -688,7 +688,7 @@ TEST(testCase, smlParseTelnetLine_diff_json_type2_Test) {
   };
   int ret = TSDB_CODE_SUCCESS;
   for (int i = 0; i < sizeof(sql) / sizeof(sql[0]); i++) {
-    ret = smlParseTelnetLine(info, (void *)sql[i]);
+    ret = smlParseTelnetLine(info, (void *)sql[i], strlen(sql[i]));
     if (ret != TSDB_CODE_SUCCESS) break;
   }
   ASSERT_NE(ret, 0);
@@ -1002,7 +1002,7 @@ TEST(testCase, sml_col_4096_Test) {
 
   int ret = TSDB_CODE_SUCCESS;
   for (int i = 0; i < sizeof(sql) / sizeof(sql[0]); i++) {
-    ret = smlParseInfluxLine(info, sql[i]);
+    ret = smlParseInfluxLine(info, sql[i], strlen(sql[i]));
     if (ret != TSDB_CODE_SUCCESS) break;
   }
   ASSERT_NE(ret, 0);
diff --git a/utils/test/c/sml_test.c b/utils/test/c/sml_test.c
index b6d8d75ba0..09079d3a26 100644
--- a/utils/test/c/sml_test.c
+++ b/utils/test/c/sml_test.c
@@ -1155,37 +1155,64 @@ int smlProcess_18784_Test() {
   return code;
 }
 
+int sml_19221_Test() {
+  TAOS *taos = taos_connect("localhost", "root", "taosdata", NULL, 0);
+
+  TAOS_RES *pRes = taos_query(taos, "create database if not exists sml_db schemaless 1");
+  taos_free_result(pRes);
+
+  const char *sql[] = {
+      "qelhxo,id=pnnqhsa,t0=t,t1=127i8 c11=L\"ncharColValue\",c0=t,c1=127i8 1626006833639000000\nqelhxo,id=pnnhsa,t0=t,t1=127i8 c11=L\"ncharColValue\",c0=t,c1=127i8 1626006833639000000\nqelhxo,id=pnqhsa,t0=t,t1=127i8 c11=L\"ncharColValue\",c0=t,c1=127i8 1626006833639000000",
+  };
+
+  pRes = taos_query(taos, "use sml_db");
+  taos_free_result(pRes);
+
+  void* tmp = taosMemoryCalloc(256, 1);
+  memcpy(tmp, sql[0], strlen(sql[0]));
+  *(char*)(tmp+44) = 0;
+  int32_t totalRows = 0;
+  pRes = taos_schemaless_insert_raw(taos, tmp, strlen(sql[0]), &totalRows, TSDB_SML_LINE_PROTOCOL, TSDB_SML_TIMESTAMP_NANO_SECONDS);
+
+  printf("%s result:%s\n", __FUNCTION__, taos_errstr(pRes));
+  int code = taos_errno(pRes);
+  taos_free_result(pRes);
+  return code;
+}
+
 int main(int argc, char *argv[]) {
   int ret = 0;
-  ret = smlProcess_influx_Test();
-  ASSERT(!ret);
-  ret = smlProcess_telnet_Test();
-  ASSERT(!ret);
-  ret = smlProcess_json1_Test();
-  ASSERT(!ret);
-  ret = smlProcess_json2_Test();
-  ASSERT(!ret);
-  ret = smlProcess_json3_Test();
-  ASSERT(!ret);
-  ret = smlProcess_json4_Test();
-  ASSERT(!ret);
-  ret = sml_TD15662_Test();
-  ASSERT(!ret);
-  ret = sml_TD15742_Test();
-  ASSERT(!ret);
-  ret = sml_16384_Test();
-  ASSERT(!ret);
-  ret = sml_oom_Test();
-  ASSERT(!ret);
-  ret = sml_16368_Test();
-  ASSERT(!ret);
-  ret = sml_dup_time_Test();
-  ASSERT(!ret);
-  ret = sml_16960_Test();
-  ASSERT(!ret);
-  ret = sml_add_tag_col_Test();
-  ASSERT(!ret);
-  ret = smlProcess_18784_Test();
+//  ret = smlProcess_influx_Test();
+//  ASSERT(!ret);
+//  ret = smlProcess_telnet_Test();
+//  ASSERT(!ret);
+//  ret = smlProcess_json1_Test();
+//  ASSERT(!ret);
+//  ret = smlProcess_json2_Test();
+//  ASSERT(!ret);
+//  ret = smlProcess_json3_Test();
+//  ASSERT(!ret);
+//  ret = smlProcess_json4_Test();
+//  ASSERT(!ret);
+//  ret = sml_TD15662_Test();
+//  ASSERT(!ret);
+//  ret = sml_TD15742_Test();
+//  ASSERT(!ret);
+//  ret = sml_16384_Test();
+//  ASSERT(!ret);
+//  ret = sml_oom_Test();
+//  ASSERT(!ret);
+//  ret = sml_16368_Test();
+//  ASSERT(!ret);
+//  ret = sml_dup_time_Test();
+//  ASSERT(!ret);
+//  ret = sml_16960_Test();
+//  ASSERT(!ret);
+//  ret = sml_add_tag_col_Test();
+//  ASSERT(!ret);
+//  ret = smlProcess_18784_Test();
+//  ASSERT(!ret);
+  ret = sml_19221_Test();
   ASSERT(!ret);
   return ret;
 }

From ab7a00abbe5d7833bc2891b1bb1aec0ffe36806d Mon Sep 17 00:00:00 2001
From: dapan1121 <wpan@taosdata.com>
Date: Tue, 18 Oct 2022 11:54:30 +0800
Subject: [PATCH 15/42] fix: fix case when crash issue

---
 source/libs/executor/src/timewindowoperator.c |  3 +++
 source/libs/scalar/src/scalar.c               | 14 ++++++++++++++
 source/libs/scalar/src/sclvector.c            |  2 +-
 3 files changed, 18 insertions(+), 1 deletion(-)

diff --git a/source/libs/executor/src/timewindowoperator.c b/source/libs/executor/src/timewindowoperator.c
index cbbd874bb9..55a8f16698 100644
--- a/source/libs/executor/src/timewindowoperator.c
+++ b/source/libs/executor/src/timewindowoperator.c
@@ -1724,6 +1724,9 @@ static void destroyStateWindowOperatorInfo(void* param) {
   cleanupBasicInfo(&pInfo->binfo);
   taosMemoryFreeClear(pInfo->stateKey.pData);
   cleanupExprSupp(&pInfo->scalarSup);
+  colDataDestroy(&pInfo->twAggSup.timeWindowData);
+  cleanupAggSup(&pInfo->aggSup);
+  cleanupGroupResInfo(&pInfo->groupResInfo);
 
   taosMemoryFreeClear(param);
 }
diff --git a/source/libs/scalar/src/scalar.c b/source/libs/scalar/src/scalar.c
index 1f09fd4799..cc1949f17d 100644
--- a/source/libs/scalar/src/scalar.c
+++ b/source/libs/scalar/src/scalar.c
@@ -629,6 +629,8 @@ _return:
   
   sclFreeParam(pWhen);
   sclFreeParam(pThen);
+  taosMemoryFree(pWhen);
+  taosMemoryFree(pThen);
 
   SCL_RET(code);
 }
@@ -664,6 +666,8 @@ int32_t sclWalkWhenList(SScalarCtx *ctx, SNodeList* pList, struct SListCell* pCe
 
     sclFreeParam(pWhen);
     sclFreeParam(pThen);
+    taosMemoryFreeClear(pWhen);
+    taosMemoryFreeClear(pThen);
   }
 
   if (pElse) {
@@ -688,6 +692,8 @@ _return:
 
   sclFreeParam(pWhen);
   sclFreeParam(pThen);
+  taosMemoryFree(pWhen);
+  taosMemoryFree(pThen);
 
   SCL_RET(code);
 }
@@ -929,6 +935,10 @@ int32_t sclExecCaseWhen(SCaseWhenNode *node, SScalarCtx *ctx, SScalarParam *outp
   sclFreeParam(&comp);
   sclFreeParam(pWhen);
   sclFreeParam(pThen);
+  taosMemoryFree(pCase);
+  taosMemoryFree(pElse);
+  taosMemoryFree(pWhen);
+  taosMemoryFree(pThen);
 
   return TSDB_CODE_SUCCESS;
 
@@ -940,6 +950,10 @@ _return:
   sclFreeParam(pWhen);
   sclFreeParam(pThen);
   sclFreeParam(output);
+  taosMemoryFree(pCase);
+  taosMemoryFree(pElse);
+  taosMemoryFree(pWhen);
+  taosMemoryFree(pThen);
 
   SCL_RET(code);
 }
diff --git a/source/libs/scalar/src/sclvector.c b/source/libs/scalar/src/sclvector.c
index f467721248..91039ed577 100644
--- a/source/libs/scalar/src/sclvector.c
+++ b/source/libs/scalar/src/sclvector.c
@@ -654,7 +654,7 @@ int32_t vectorConvertSingleColImpl(const SScalarParam* pIn, SScalarParam* pOut,
     return TSDB_CODE_APP_ERROR;
   }
 
-  int32_t rstart = startIndex >= 0 ? startIndex : 0;
+  int32_t rstart = (startIndex >= 0 && startIndex < pIn->numOfRows) ? startIndex : 0;
   int32_t rend = numOfRows > 0 ? rstart + numOfRows - 1 : rstart + pIn->numOfRows - 1;
   SSclVectorConvCtx cCtx = {pIn, pOut, rstart, rend, pInputCol->info.type, pOutputCol->info.type};
 

From f76261dcb567fcdd89e561a9a62fde0a5c3f5666 Mon Sep 17 00:00:00 2001
From: Shengliang Guan <slguan@taosdata.com>
Date: Tue, 18 Oct 2022 12:31:15 +0800
Subject: [PATCH 16/42] fix: become leader if replica num is 1

---
 source/dnode/mnode/impl/src/mndMain.c  | 19 ++++++++++++-------
 source/dnode/mnode/impl/src/mndMnode.c |  6 ++++++
 source/libs/sync/src/syncElection.c    | 13 +++++++++++++
 source/libs/sync/src/syncMain.c        |  2 +-
 4 files changed, 32 insertions(+), 8 deletions(-)

diff --git a/source/dnode/mnode/impl/src/mndMain.c b/source/dnode/mnode/impl/src/mndMain.c
index 5fb23b045c..8314601df3 100644
--- a/source/dnode/mnode/impl/src/mndMain.c
+++ b/source/dnode/mnode/impl/src/mndMain.c
@@ -603,22 +603,27 @@ static int32_t mndCheckMnodeState(SRpcMsg *pMsg) {
     return 0;
   }
   if (mndAcquireRpc(pMsg->info.node) == 0) return 0;
+
+  SMnode     *pMnode = pMsg->info.node;
+  const char *role = syncGetMyRoleStr(pMnode->syncMgmt.sync);
+  bool        restored = syncIsRestoreFinish(pMnode->syncMgmt.sync);
   if (pMsg->msgType == TDMT_MND_MQ_TIMER || pMsg->msgType == TDMT_MND_TELEM_TIMER ||
       pMsg->msgType == TDMT_MND_TRANS_TIMER || pMsg->msgType == TDMT_MND_TTL_TIMER ||
       pMsg->msgType == TDMT_MND_UPTIME_TIMER) {
+    mTrace("timer not process since mnode restored:%d stopped:%d, sync restored:%d role:%s ", pMnode->restored,
+           pMnode->stopped, restored, role);
     return -1;
   }
 
-  SEpSet  epSet = {0};
-  SMnode *pMnode = pMsg->info.node;
+  const STraceId *trace = &pMsg->info.traceId;
+  SEpSet          epSet = {0};
   mndGetMnodeEpSet(pMnode, &epSet);
 
-  const STraceId *trace = &pMsg->info.traceId;
   mDebug(
-      "msg:%p, failed to check mnode state since %s, mnode restored:%d stopped:%d, sync restored:%d role:%s type:%s "
-      "numOfEps:%d inUse:%d",
-      pMsg, terrstr(), pMnode->restored, pMnode->stopped, syncIsRestoreFinish(pMnode->syncMgmt.sync),
-      syncGetMyRoleStr(pMnode->syncMgmt.sync), TMSG_INFO(pMsg->msgType), epSet.numOfEps, epSet.inUse);
+      "msg:%p, type:%s failed to process since %s, mnode restored:%d stopped:%d, sync restored:%d "
+      "role:%s, redirect numOfEps:%d inUse:%d",
+      pMsg, TMSG_INFO(pMsg->msgType), terrstr(), pMnode->restored, pMnode->stopped, restored, role, epSet.numOfEps,
+      epSet.inUse);
 
   if (epSet.numOfEps > 0) {
     for (int32_t i = 0; i < epSet.numOfEps; ++i) {
diff --git a/source/dnode/mnode/impl/src/mndMnode.c b/source/dnode/mnode/impl/src/mndMnode.c
index 63ce538536..f25d436aee 100644
--- a/source/dnode/mnode/impl/src/mndMnode.c
+++ b/source/dnode/mnode/impl/src/mndMnode.c
@@ -747,7 +747,13 @@ static void mndReloadSyncConfig(SMnode *pMnode) {
   }
 
   if (cfg.myIndex == -1) {
+#if 1
     mInfo("vgId:1, mnode sync not reconfig since selfIndex is -1");
+#else
+    // cannot reconfig because the leader may fail to elect after reboot
+    mInfo("vgId:1, mnode sync not reconfig since selfIndex is -1, do sync stop oper");
+    syncStop(pMnode->syncMgmt.sync);
+#endif
     return;
   }
 
diff --git a/source/libs/sync/src/syncElection.c b/source/libs/sync/src/syncElection.c
index 375f2e5730..3f13249ce6 100644
--- a/source/libs/sync/src/syncElection.c
+++ b/source/libs/sync/src/syncElection.c
@@ -96,6 +96,19 @@ int32_t syncNodeElect(SSyncNode* pSyncNode) {
     syncNodeCandidate2Leader(pSyncNode);
     pSyncNode->pVotesGranted->toLeader = true;
     return ret;
+  } 
+
+  if (pSyncNode->replicaNum == 1) {
+    // only myself, to leader
+    voteGrantedUpdate(pSyncNode->pVotesGranted, pSyncNode);
+    votesRespondUpdate(pSyncNode->pVotesRespond, pSyncNode);
+
+    pSyncNode->quorum = syncUtilQuorum(pSyncNode->pRaftCfg->cfg.replicaNum);
+
+    syncNodeCandidate2Leader(pSyncNode);
+    pSyncNode->pVotesGranted->toLeader = true;
+    return ret;
+
   }
 
   switch (pSyncNode->pRaftCfg->snapshotStrategy) {
diff --git a/source/libs/sync/src/syncMain.c b/source/libs/sync/src/syncMain.c
index b255d83933..83bd7a5f38 100644
--- a/source/libs/sync/src/syncMain.c
+++ b/source/libs/sync/src/syncMain.c
@@ -2282,7 +2282,7 @@ void syncNodeBecomeLeader(SSyncNode* pSyncNode, const char* debugStr) {
 
 void syncNodeCandidate2Leader(SSyncNode* pSyncNode) {
   ASSERT(pSyncNode->state == TAOS_SYNC_STATE_CANDIDATE);
-  ASSERT(voteGrantedMajority(pSyncNode->pVotesGranted));
+  //ASSERT(voteGrantedMajority(pSyncNode->pVotesGranted));
   syncNodeBecomeLeader(pSyncNode, "candidate to leader");
 
   syncNodeLog2("==state change syncNodeCandidate2Leader==", pSyncNode);

From 7b84b217c2cba993bead5af81c658162c51cda95 Mon Sep 17 00:00:00 2001
From: wangmm0220 <wangmm0220@gmail.com>
Date: Tue, 18 Oct 2022 13:46:42 +0800
Subject: [PATCH 17/42] opti:add auto compile

---
 source/client/src/clientSml.c |  2 +-
 utils/test/c/sml_test.c       | 65 ++++++++++++++++++-----------------
 2 files changed, 34 insertions(+), 33 deletions(-)

diff --git a/source/client/src/clientSml.c b/source/client/src/clientSml.c
index 9241b3c5d4..f2739c9ab8 100644
--- a/source/client/src/clientSml.c
+++ b/source/client/src/clientSml.c
@@ -79,7 +79,7 @@
 #define NCHAR_ADD_LEN  3  // L"nchar"   3 means L" "
 
 #define MAX_RETRY_TIMES 5
-#define LINE_BATCH      2
+#define LINE_BATCH      2000
 //=================================================================================================
 typedef TSDB_SML_PROTOCOL_TYPE SMLProtocolType;
 
diff --git a/utils/test/c/sml_test.c b/utils/test/c/sml_test.c
index 09079d3a26..1066054b7d 100644
--- a/utils/test/c/sml_test.c
+++ b/utils/test/c/sml_test.c
@@ -1162,18 +1162,19 @@ int sml_19221_Test() {
   taos_free_result(pRes);
 
   const char *sql[] = {
-      "qelhxo,id=pnnqhsa,t0=t,t1=127i8 c11=L\"ncharColValue\",c0=t,c1=127i8 1626006833639000000\nqelhxo,id=pnnhsa,t0=t,t1=127i8 c11=L\"ncharColValue\",c0=t,c1=127i8 1626006833639000000\nqelhxo,id=pnqhsa,t0=t,t1=127i8 c11=L\"ncharColValue\",c0=t,c1=127i8 1626006833639000000",
+      "qelhxo,id=pnnqhsa,t0=t,t1=127i8 c11=L\"ncharColValue\",c0=t,c1=127i8 1626006833639000000\nqelhxo,id=pnnhsa,t0=t,t1=127i8 c11=L\"ncharColValue\",c0=t,c1=127i8 1626006833639000000\n#comment\nqelhxo,id=pnqhsa,t0=t,t1=127i8 c11=L\"ncharColValue\",c0=t,c1=127i8 1626006833639000000",
   };
 
   pRes = taos_query(taos, "use sml_db");
   taos_free_result(pRes);
 
-  void* tmp = taosMemoryCalloc(256, 1);
+  char* tmp = (char*)taosMemoryCalloc(1024, 1);
   memcpy(tmp, sql[0], strlen(sql[0]));
   *(char*)(tmp+44) = 0;
   int32_t totalRows = 0;
   pRes = taos_schemaless_insert_raw(taos, tmp, strlen(sql[0]), &totalRows, TSDB_SML_LINE_PROTOCOL, TSDB_SML_TIMESTAMP_NANO_SECONDS);
 
+  ASSERT(totalRows == 3);
   printf("%s result:%s\n", __FUNCTION__, taos_errstr(pRes));
   int code = taos_errno(pRes);
   taos_free_result(pRes);
@@ -1182,36 +1183,36 @@ int sml_19221_Test() {
 
 int main(int argc, char *argv[]) {
   int ret = 0;
-//  ret = smlProcess_influx_Test();
-//  ASSERT(!ret);
-//  ret = smlProcess_telnet_Test();
-//  ASSERT(!ret);
-//  ret = smlProcess_json1_Test();
-//  ASSERT(!ret);
-//  ret = smlProcess_json2_Test();
-//  ASSERT(!ret);
-//  ret = smlProcess_json3_Test();
-//  ASSERT(!ret);
-//  ret = smlProcess_json4_Test();
-//  ASSERT(!ret);
-//  ret = sml_TD15662_Test();
-//  ASSERT(!ret);
-//  ret = sml_TD15742_Test();
-//  ASSERT(!ret);
-//  ret = sml_16384_Test();
-//  ASSERT(!ret);
-//  ret = sml_oom_Test();
-//  ASSERT(!ret);
-//  ret = sml_16368_Test();
-//  ASSERT(!ret);
-//  ret = sml_dup_time_Test();
-//  ASSERT(!ret);
-//  ret = sml_16960_Test();
-//  ASSERT(!ret);
-//  ret = sml_add_tag_col_Test();
-//  ASSERT(!ret);
-//  ret = smlProcess_18784_Test();
-//  ASSERT(!ret);
+  ret = smlProcess_influx_Test();
+  ASSERT(!ret);
+  ret = smlProcess_telnet_Test();
+  ASSERT(!ret);
+  ret = smlProcess_json1_Test();
+  ASSERT(!ret);
+  ret = smlProcess_json2_Test();
+  ASSERT(!ret);
+  ret = smlProcess_json3_Test();
+  ASSERT(!ret);
+  ret = smlProcess_json4_Test();
+  ASSERT(!ret);
+  ret = sml_TD15662_Test();
+  ASSERT(!ret);
+  ret = sml_TD15742_Test();
+  ASSERT(!ret);
+  ret = sml_16384_Test();
+  ASSERT(!ret);
+  ret = sml_oom_Test();
+  ASSERT(!ret);
+  ret = sml_16368_Test();
+  ASSERT(!ret);
+  ret = sml_dup_time_Test();
+  ASSERT(!ret);
+  ret = sml_16960_Test();
+  ASSERT(!ret);
+  ret = sml_add_tag_col_Test();
+  ASSERT(!ret);
+  ret = smlProcess_18784_Test();
+  ASSERT(!ret);
   ret = sml_19221_Test();
   ASSERT(!ret);
   return ret;

From 029abe73963bd672b97f6218c641bb1cfce713b7 Mon Sep 17 00:00:00 2001
From: Minglei Jin <mljin@taosdata.com>
Date: Tue, 18 Oct 2022 14:32:18 +0800
Subject: [PATCH 18/42] fix: coverity issues

---
 source/dnode/vnode/src/meta/metaTable.c | 2 +-
 source/libs/tdb/src/db/tdbBtree.c       | 1 +
 source/libs/tdb/src/db/tdbDb.c          | 6 +++---
 source/libs/tdb/src/db/tdbPCache.c      | 2 +-
 source/libs/tdb/src/db/tdbPager.c       | 4 ++--
 5 files changed, 8 insertions(+), 7 deletions(-)

diff --git a/source/dnode/vnode/src/meta/metaTable.c b/source/dnode/vnode/src/meta/metaTable.c
index 2b7982d381..c9a277ae1f 100644
--- a/source/dnode/vnode/src/meta/metaTable.c
+++ b/source/dnode/vnode/src/meta/metaTable.c
@@ -51,7 +51,7 @@ static int metaUpdateMetaRsp(tb_uid_t uid, char *tbName, SSchemaWrapper *pSchema
     return -1;
   }
 
-  strncpy(pMetaRsp->tbName, tbName, TSDB_TABLE_NAME_LEN);
+  tstrncpy(pMetaRsp->tbName, tbName, TSDB_TABLE_NAME_LEN);
   pMetaRsp->numOfColumns = pSchema->nCols;
   pMetaRsp->tableType = TSDB_NORMAL_TABLE;
   pMetaRsp->sversion = pSchema->version;
diff --git a/source/libs/tdb/src/db/tdbBtree.c b/source/libs/tdb/src/db/tdbBtree.c
index 9e17f50dce..8f11fb88a2 100644
--- a/source/libs/tdb/src/db/tdbBtree.c
+++ b/source/libs/tdb/src/db/tdbBtree.c
@@ -1086,6 +1086,7 @@ static int tdbBtreeEncodePayload(SPage *pPage, SCell *pCell, int nHeader, const
             // fetch next ofp, a new ofp and make it dirty
             ret = tdbFetchOvflPage(&pgno, &nextOfp, pTxn, pBt);
             if (ret < 0) {
+              tdbFree(pBuf);
               return -1;
             }
           }
diff --git a/source/libs/tdb/src/db/tdbDb.c b/source/libs/tdb/src/db/tdbDb.c
index 6c01348bc2..a2803d3ecf 100644
--- a/source/libs/tdb/src/db/tdbDb.c
+++ b/source/libs/tdb/src/db/tdbDb.c
@@ -106,7 +106,7 @@ int32_t tdbBegin(TDB *pDb, TXN *pTxn) {
   for (pPager = pDb->pgrList; pPager; pPager = pPager->pNext) {
     ret = tdbPagerBegin(pPager, pTxn);
     if (ret < 0) {
-      tdbError("failed to begin pager since %s. dbName:%s, txnId:%d", tstrerror(terrno), pDb->dbName, pTxn->txnId);
+      tdbError("failed to begin pager since %s. dbName:%s, txnId:%ld", tstrerror(terrno), pDb->dbName, pTxn->txnId);
       return -1;
     }
   }
@@ -121,7 +121,7 @@ int32_t tdbCommit(TDB *pDb, TXN *pTxn) {
   for (pPager = pDb->pgrList; pPager; pPager = pPager->pNext) {
     ret = tdbPagerCommit(pPager, pTxn);
     if (ret < 0) {
-      tdbError("failed to commit pager since %s. dbName:%s, txnId:%d", tstrerror(terrno), pDb->dbName, pTxn->txnId);
+      tdbError("failed to commit pager since %s. dbName:%s, txnId:%ld", tstrerror(terrno), pDb->dbName, pTxn->txnId);
       return -1;
     }
   }
@@ -136,7 +136,7 @@ int32_t tdbAbort(TDB *pDb, TXN *pTxn) {
   for (pPager = pDb->pgrList; pPager; pPager = pPager->pNext) {
     ret = tdbPagerAbort(pPager, pTxn);
     if (ret < 0) {
-      tdbError("failed to abort pager since %s. dbName:%s, txnId:%d", tstrerror(terrno), pDb->dbName, pTxn->txnId);
+      tdbError("failed to abort pager since %s. dbName:%s, txnId:%ld", tstrerror(terrno), pDb->dbName, pTxn->txnId);
       return -1;
     }
   }
diff --git a/source/libs/tdb/src/db/tdbPCache.c b/source/libs/tdb/src/db/tdbPCache.c
index c73ddce74c..732a57639f 100644
--- a/source/libs/tdb/src/db/tdbPCache.c
+++ b/source/libs/tdb/src/db/tdbPCache.c
@@ -268,7 +268,7 @@ static SPage *tdbPCacheFetchImpl(SPCache *pCache, const SPgid *pPgid, TXN *pTxn)
   // 4. Try a create new page
   if (!pPage) {
     ret = tdbPageCreate(pCache->szPage, &pPage, pTxn->xMalloc, pTxn->xArg);
-    if (ret < 0 && pPage != NULL) {
+    if (ret < 0 || pPage == NULL) {
       // TODO
       ASSERT(0);
       return NULL;
diff --git a/source/libs/tdb/src/db/tdbPager.c b/source/libs/tdb/src/db/tdbPager.c
index 57ad2a783a..90332617cd 100644
--- a/source/libs/tdb/src/db/tdbPager.c
+++ b/source/libs/tdb/src/db/tdbPager.c
@@ -541,7 +541,7 @@ static int tdbPagerWritePageToJournal(SPager *pPager, SPage *pPage) {
 
   ret = tdbOsWrite(pPager->jfd, pPage->pData, pPage->pageSize);
   if (ret < 0) {
-    tdbError("failed to write page data due to %s. file:%s, pageSize:%ld", strerror(errno), pPager->jFileName,
+    tdbError("failed to write page data due to %s. file:%s, pageSize:%d", strerror(errno), pPager->jFileName,
              pPage->pageSize);
     terrno = TAOS_SYSTEM_ERROR(errno);
     return -1;
@@ -570,7 +570,7 @@ static int tdbPagerWritePageToDB(SPager *pPager, SPage *pPage) {
 
   ret = tdbOsWrite(pPager->fd, pPage->pData, pPage->pageSize);
   if (ret < 0) {
-    tdbError("failed to write page data due to %s. file:%s, pageSize:%ld", strerror(errno), pPager->dbFileName,
+    tdbError("failed to write page data due to %s. file:%s, pageSize:%d", strerror(errno), pPager->dbFileName,
              pPage->pageSize);
     terrno = TAOS_SYSTEM_ERROR(errno);
     return -1;

From bce4fc9cf86a89afd5242e862690fc7a7e4956c7 Mon Sep 17 00:00:00 2001
From: wangmm0220 <wangmm0220@gmail.com>
Date: Tue, 18 Oct 2022 14:43:06 +0800
Subject: [PATCH 19/42] feat:add new interface for schemaless

---
 utils/test/c/sml_test.c | 30 ++++++++++++++++++++++++++++++
 1 file changed, 30 insertions(+)

diff --git a/utils/test/c/sml_test.c b/utils/test/c/sml_test.c
index 1066054b7d..f7fd59cd7b 100644
--- a/utils/test/c/sml_test.c
+++ b/utils/test/c/sml_test.c
@@ -63,6 +63,7 @@ int smlProcess_influx_Test() {
   printf("%s result:%s\n", __FUNCTION__, taos_errstr(pRes));
   int code = taos_errno(pRes);
   taos_free_result(pRes);
+  taos_close(taos);
 
   return code;
 }
@@ -86,6 +87,8 @@ int smlProcess_telnet_Test() {
   printf("%s result:%s\n", __FUNCTION__, taos_errstr(pRes));
   int code = taos_errno(pRes);
   taos_free_result(pRes);
+  taos_close(taos);
+
   return code;
 }
 
@@ -125,6 +128,8 @@ int smlProcess_json1_Test() {
   printf("%s result:%s\n", __FUNCTION__, taos_errstr(pRes));
   int code = taos_errno(pRes);
   taos_free_result(pRes);
+  taos_close(taos);
+
   return code;
 }
 
@@ -165,6 +170,8 @@ int smlProcess_json2_Test() {
   printf("%s result:%s\n", __FUNCTION__, taos_errstr(pRes));
   int code = taos_errno(pRes);
   taos_free_result(pRes);
+  taos_close(taos);
+
   return code;
 }
 
@@ -233,6 +240,8 @@ int smlProcess_json3_Test() {
   printf("%s result:%s\n", __FUNCTION__, taos_errstr(pRes));
   int code = taos_errno(pRes);
   taos_free_result(pRes);
+  taos_close(taos);
+
   return code;
 }
 
@@ -292,6 +301,8 @@ int smlProcess_json4_Test() {
   printf("%s result:%s\n", __FUNCTION__, taos_errstr(pRes));
   int code = taos_errno(pRes);
   taos_free_result(pRes);
+  taos_close(taos);
+
   return code;
 }
 
@@ -313,6 +324,8 @@ int sml_TD15662_Test() {
   printf("%s result:%s\n", __FUNCTION__, taos_errstr(pRes));
   int code = taos_errno(pRes);
   taos_free_result(pRes);
+  taos_close(taos);
+
   return code;
 }
 
@@ -333,6 +346,8 @@ int sml_TD15742_Test() {
   printf("%s result:%s\n", __FUNCTION__, taos_errstr(pRes));
   int code = taos_errno(pRes);
   taos_free_result(pRes);
+  taos_close(taos);
+
   return code;
 }
 
@@ -362,6 +377,8 @@ int sml_16384_Test() {
   printf("%s result:%s\n", __FUNCTION__, taos_errstr(pRes));
   code = taos_errno(pRes);
   taos_free_result(pRes);
+  taos_close(taos);
+
   return code;
 }
 
@@ -781,6 +798,8 @@ int sml_oom_Test() {
   printf("%s result:%s\n", __FUNCTION__, taos_errstr(pRes));
   int code = taos_errno(pRes);
   taos_free_result(pRes);
+  taos_close(taos);
+
   return code;
 }
 
@@ -825,6 +844,8 @@ int sml_16368_Test() {
   printf("%s result:%s\n", __FUNCTION__, taos_errstr(pRes));
   int code = taos_errno(pRes);
   taos_free_result(pRes);
+  taos_close(taos);
+
   return code;
 }
 
@@ -862,6 +883,8 @@ int sml_dup_time_Test() {
   printf("%s result:%s\n", __FUNCTION__, taos_errstr(pRes));
   int code = taos_errno(pRes);
   taos_free_result(pRes);
+  taos_close(taos);
+
   return code;
 }
 
@@ -1068,6 +1091,8 @@ int sml_16960_Test() {
   printf("%s result:%s\n", __FUNCTION__, taos_errstr(pRes));
   int code = taos_errno(pRes);
   taos_free_result(pRes);
+  taos_close(taos);
+
   return code;
 }
 
@@ -1097,6 +1122,7 @@ int sml_add_tag_col_Test() {
   printf("%s result:%s\n", __FUNCTION__, taos_errstr(pRes));
   code = taos_errno(pRes);
   taos_free_result(pRes);
+  taos_close(taos);
 
   return code;
 }
@@ -1151,6 +1177,7 @@ int smlProcess_18784_Test() {
     rowIndex++;
   }
   taos_free_result(pRes);
+  taos_close(taos);
 
   return code;
 }
@@ -1178,6 +1205,9 @@ int sml_19221_Test() {
   printf("%s result:%s\n", __FUNCTION__, taos_errstr(pRes));
   int code = taos_errno(pRes);
   taos_free_result(pRes);
+  taos_close(taos);
+  taosMemoryFree(tmp);
+
   return code;
 }
 

From 0f6ba60da4dcae5ee23e082a68f94b72cd9c52ea Mon Sep 17 00:00:00 2001
From: Shengliang Guan <slguan@taosdata.com>
Date: Tue, 18 Oct 2022 14:49:43 +0800
Subject: [PATCH 20/42] fix: wrong unlock commands

---
 source/dnode/mnode/sdb/src/sdbHash.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/source/dnode/mnode/sdb/src/sdbHash.c b/source/dnode/mnode/sdb/src/sdbHash.c
index 1e56eb87bf..85e937fcc3 100644
--- a/source/dnode/mnode/sdb/src/sdbHash.c
+++ b/source/dnode/mnode/sdb/src/sdbHash.c
@@ -203,7 +203,7 @@ static int32_t sdbUpdateRow(SSdb *pSdb, SHashObj *hash, SSdbRaw *pRaw, SSdbRow *
     code = (*updateFp)(pSdb, pOldRow->pObj, pNewRow->pObj);
   }
 
-  sdbUnLock(pSdb, type);
+  // sdbUnLock(pSdb, type);
   sdbFreeRow(pSdb, pNewRow, false);
 
   pSdb->tableVer[pOldRow->type]++;

From 0cbaa03fcc4cc6cc3ac836f89215d3ad4b929db0 Mon Sep 17 00:00:00 2001
From: Benguang Zhao <bgzhao@taosdata.com>
Date: Tue, 18 Oct 2022 14:51:19 +0800
Subject: [PATCH 21/42] fix: osDefaultInit on all platform in testing Init at
 sut.cpp

---
 source/dnode/mgmt/test/sut/src/sut.cpp | 2 --
 1 file changed, 2 deletions(-)

diff --git a/source/dnode/mgmt/test/sut/src/sut.cpp b/source/dnode/mgmt/test/sut/src/sut.cpp
index 699203e8c1..a4d2e46881 100644
--- a/source/dnode/mgmt/test/sut/src/sut.cpp
+++ b/source/dnode/mgmt/test/sut/src/sut.cpp
@@ -43,9 +43,7 @@ void Testbase::InitLog(const char* path) {
 }
 
 void Testbase::Init(const char* path, int16_t port) {
-#ifdef _TD_DARWIN_64
   osDefaultInit();
-#endif
   tsServerPort = port;
   strcpy(tsLocalFqdn, "localhost");
   snprintf(tsLocalEp, TSDB_EP_LEN, "%s:%u", tsLocalFqdn, tsServerPort);

From eaa2a344e1d78fbcb69a64f1710881cfaa989f99 Mon Sep 17 00:00:00 2001
From: Xiaoyu Wang <xiaoyuwang@taosdata.com>
Date: Tue, 18 Oct 2022 15:01:52 +0800
Subject: [PATCH 22/42] fix: plan problem with 'partition by tag group by tag'
 statement

---
 source/libs/planner/src/planOptimizer.c     | 31 +++++++++++++++++----
 source/libs/planner/test/planPartByTest.cpp |  2 ++
 2 files changed, 27 insertions(+), 6 deletions(-)

diff --git a/source/libs/planner/src/planOptimizer.c b/source/libs/planner/src/planOptimizer.c
index d6110d9fcd..ccae1f54be 100644
--- a/source/libs/planner/src/planOptimizer.c
+++ b/source/libs/planner/src/planOptimizer.c
@@ -124,8 +124,8 @@ static void optSetParentOrder(SLogicNode* pNode, EOrder order) {
 
 EDealRes scanPathOptHaveNormalColImpl(SNode* pNode, void* pContext) {
   if (QUERY_NODE_COLUMN == nodeType(pNode)) {
-     *((bool*)pContext) =
-         (COLUMN_TYPE_TAG != ((SColumnNode*)pNode)->colType && COLUMN_TYPE_TBNAME != ((SColumnNode*)pNode)->colType);
+    *((bool*)pContext) =
+        (COLUMN_TYPE_TAG != ((SColumnNode*)pNode)->colType && COLUMN_TYPE_TBNAME != ((SColumnNode*)pNode)->colType);
     return *((bool*)pContext) ? DEAL_RES_END : DEAL_RES_IGNORE_CHILD;
   }
   return DEAL_RES_CONTINUE;
@@ -1520,11 +1520,15 @@ static bool partTagsHasIndefRowsSelectFunc(SNodeList* pFuncs) {
   return false;
 }
 
-static int32_t partTagsRewriteGroupTagsToFuncs(SNodeList* pGroupTags, SNodeList* pAggFuncs) {
+static int32_t partTagsRewriteGroupTagsToFuncs(SNodeList* pGroupTags, int32_t start, SNodeList* pAggFuncs) {
   bool    hasIndefRowsSelectFunc = partTagsHasIndefRowsSelectFunc(pAggFuncs);
   int32_t code = TSDB_CODE_SUCCESS;
+  int32_t index = 0;
   SNode*  pNode = NULL;
   FOREACH(pNode, pGroupTags) {
+    if (index++ < start) {
+      continue;
+    }
     if (hasIndefRowsSelectFunc) {
       code = nodesListStrictAppend(pAggFuncs, partTagsCreateWrapperFunc("_select_value", pNode));
     } else {
@@ -1559,20 +1563,35 @@ static int32_t partTagsOptimize(SOptimizeContext* pCxt, SLogicSubplan* pLogicSub
     }
   } else {
     SAggLogicNode* pAgg = (SAggLogicNode*)pNode;
+    int32_t        start = -1;
     SNode*         pGroupKey = NULL;
     FOREACH(pGroupKey, pAgg->pGroupKeys) {
-      code = nodesListMakeStrictAppend(
-          &pScan->pGroupTags, nodesCloneNode(nodesListGetNode(((SGroupingSetNode*)pGroupKey)->pParameterList, 0)));
+      SNode* pGroupExpr = nodesListGetNode(((SGroupingSetNode*)pGroupKey)->pParameterList, 0);
+      if (NULL != pScan->pGroupTags) {
+        SNode* pGroupTag = NULL;
+        FOREACH(pGroupTag, pScan->pGroupTags) {
+          if (nodesEqualNode(pGroupTag, pGroupExpr)) {
+            continue;
+          }
+        }
+      }
+      if (start < 0) {
+        start = LIST_LENGTH(pScan->pGroupTags);
+      }
+      code = nodesListMakeStrictAppend(&pScan->pGroupTags, nodesCloneNode(pGroupExpr));
       if (TSDB_CODE_SUCCESS != code) {
         break;
       }
     }
     NODES_DESTORY_LIST(pAgg->pGroupKeys);
-    code = partTagsRewriteGroupTagsToFuncs(pScan->pGroupTags, pAgg->pAggFuncs);
+    if (TSDB_CODE_SUCCESS == code && start >= 0) {
+      code = partTagsRewriteGroupTagsToFuncs(pScan->pGroupTags, start, pAgg->pAggFuncs);
+    }
   }
   if (TSDB_CODE_SUCCESS == code) {
     code = partTagsOptRebuildTbanme(pScan->pGroupTags);
   }
+  pCxt->optimized = true;
   return code;
 }
 
diff --git a/source/libs/planner/test/planPartByTest.cpp b/source/libs/planner/test/planPartByTest.cpp
index 48ab1e1ac2..3d0467dffe 100644
--- a/source/libs/planner/test/planPartByTest.cpp
+++ b/source/libs/planner/test/planPartByTest.cpp
@@ -61,6 +61,8 @@ TEST_F(PlanPartitionByTest, withGroupBy) {
   run("SELECT COUNT(*) FROM t1 PARTITION BY c1 GROUP BY c2");
 
   run("SELECT TBNAME, c1 FROM st1 PARTITION BY TBNAME GROUP BY c1");
+
+  run("SELECT COUNT(*) FROM t1 PARTITION BY TBNAME GROUP BY TBNAME");
 }
 
 TEST_F(PlanPartitionByTest, withTimeLineFunc) {

From 41866ed397e6e6c3096cf1be3099585964f56469 Mon Sep 17 00:00:00 2001
From: wangmm0220 <wangmm0220@gmail.com>
Date: Tue, 18 Oct 2022 15:13:52 +0800
Subject: [PATCH 23/42] fix:defeats in converity scan

---
 source/client/src/clientSml.c |  10 ++-
 utils/test/c/sml_test.c       | 141 ++--------------------------------
 utils/test/c/tmqDemo.c        |  10 +--
 utils/test/c/tmqSim.c         |   2 +-
 4 files changed, 21 insertions(+), 142 deletions(-)

diff --git a/source/client/src/clientSml.c b/source/client/src/clientSml.c
index 5c37822222..904ad32d0a 100644
--- a/source/client/src/clientSml.c
+++ b/source/client/src/clientSml.c
@@ -1371,8 +1371,14 @@ static int32_t smlKvTimeArrayCompare(const void *key1, const void *key2) {
 static int32_t smlKvTimeHashCompare(const void *key1, const void *key2) {
   SHashObj *s1 = *(SHashObj **)key1;
   SHashObj *s2 = *(SHashObj **)key2;
-  SSmlKv   *kv1 = *(SSmlKv **)taosHashGet(s1, TS, TS_LEN);
-  SSmlKv   *kv2 = *(SSmlKv **)taosHashGet(s2, TS, TS_LEN);
+  SSmlKv   **kv1pp = (SSmlKv **)taosHashGet(s1, TS, TS_LEN);
+  SSmlKv   **kv2pp = (SSmlKv **)taosHashGet(s2, TS, TS_LEN);
+  if(!kv1pp || !kv2pp){
+    uError("smlKvTimeHashCompare kv is null");
+    return -1;
+  }
+  SSmlKv   *kv1 = *kv1pp;
+  SSmlKv   *kv2 = *kv2pp;
   if(!kv1 || kv1->type != TSDB_DATA_TYPE_TIMESTAMP){
     uError("smlKvTimeHashCompare kv1");
     return -1;
diff --git a/utils/test/c/sml_test.c b/utils/test/c/sml_test.c
index b6d8d75ba0..a9dbce62e5 100644
--- a/utils/test/c/sml_test.c
+++ b/utils/test/c/sml_test.c
@@ -99,27 +99,8 @@ int smlProcess_json1_Test() {
   taos_free_result(pRes);
 
   const char *sql[] = {
-      "["
-      "    {"
-      "        \"metric\": \"sys.cpu.nice\","
-      "        \"timestamp\": 0,"
-      "        \"value\": 18,"
-      "        \"tags\": {"
-      "           \"host\": \"web01\","
-      "           \"id\": \"t1\","
-      "           \"dc\": \"lga\""
-      "        }"
-      "    },"
-      "    {"
-      "        \"metric\": \"sys.cpu.nice\","
-      "        \"timestamp\": 1662344042,"
-      "        \"value\": 9,"
-      "        \"tags\": {"
-      "           \"host\": \"web02\","
-      "           \"dc\": \"lga\""
-      "        }"
-      "    }"
-      "]",};
+      "[{\"metric\":\"sys.cpu.nice\",\"timestamp\":0,\"value\":18,\"tags\":{\"host\":\"web01\",\"id\":\"t1\",\"dc\":\"lga\"}},{\"metric\":\"sys.cpu.nice\",\"timestamp\":1662344042,\"value\":9,\"tags\":{\"host\":\"web02\",\"dc\":\"lga\"}}]"
+  };
   pRes = taos_schemaless_insert(taos, (char **)sql, sizeof(sql) / sizeof(sql[0]), TSDB_SML_JSON_PROTOCOL,
                                 TSDB_SML_TIMESTAMP_NANO_SECONDS);
   printf("%s result:%s\n", __FUNCTION__, taos_errstr(pRes));
@@ -138,28 +119,8 @@ int smlProcess_json2_Test() {
   taos_free_result(pRes);
 
   const char *sql[] = {
-      "{"
-      "    \"metric\": \"meter_current0\","
-      "    \"timestamp\": {"
-      "        \"value\"  : 1662344042,"
-      "        \"type\"   : \"s\""
-      "    },"
-      "    \"value\": {"
-      "         \"value\" : 10.3,"
-      "         \"type\"  : \"i64\""
-      "    },"
-      "    \"tags\": {"
-      "       \"groupid\": { "
-      "           \"value\" : 2,"
-      "           \"type\"  : \"bigint\""
-      "       },"
-      "       \"location\": { "
-      "           \"value\" : \"北京\","
-      "           \"type\"  : \"binary\""
-      "       },"
-      "       \"id\": \"d1001\""
-      "    }"
-      "}",};
+      "{\"metric\":\"meter_current0\",\"timestamp\":{\"value\":1662344042,\"type\":\"s\"},\"value\":{\"value\":10.3,\"type\":\"i64\"},\"tags\":{\"groupid\":{\"value\":2,\"type\":\"bigint\"},\"location\":{\"value\":\"北京\",\"type\":\"binary\"},\"id\":\"d1001\"}}"
+  };
   pRes = taos_schemaless_insert(taos, (char **)sql, sizeof(sql) / sizeof(sql[0]), TSDB_SML_JSON_PROTOCOL,
                                 TSDB_SML_TIMESTAMP_NANO_SECONDS);
   printf("%s result:%s\n", __FUNCTION__, taos_errstr(pRes));
@@ -178,56 +139,7 @@ int smlProcess_json3_Test() {
   taos_free_result(pRes);
 
   const char *sql[] = {
-      "{"
-      "    \"metric\": \"meter_current1\","
-      "    \"timestamp\": {"
-      "        \"value\"  : 1662344042,"
-      "        \"type\"   : \"s\""
-      "    },"
-      "    \"value\": {"
-      "         \"value\" : 10.3,"
-      "         \"type\"  : \"i64\""
-      "    },"
-      "    \"tags\": {"
-      "       \"t1\": { "
-      "           \"value\" : 2,"
-      "           \"type\"  : \"bigint\""
-      "       },"
-      "       \"t2\": { "
-      "           \"value\" : 2,"
-      "           \"type\"  : \"int\""
-      "       },"
-      "       \"t3\": { "
-      "           \"value\" : 2,"
-      "           \"type\"  : \"i16\""
-      "       },"
-      "       \"t4\": { "
-      "           \"value\" : 2,"
-      "           \"type\"  : \"i8\""
-      "       },"
-      "       \"t5\": { "
-      "           \"value\" : 2,"
-      "           \"type\"  : \"f32\""
-      "       },"
-      "       \"t6\": { "
-      "           \"value\" : 2,"
-      "           \"type\"  : \"double\""
-      "       },"
-      "       \"t7\": { "
-      "           \"value\" : \"8323\","
-      "           \"type\"  : \"binary\""
-      "       },"
-      "       \"t8\": { "
-      "           \"value\" : \"北京\","
-      "           \"type\"  : \"nchar\""
-      "       },"
-      "       \"t9\": { "
-      "           \"value\" : true,"
-      "           \"type\"  : \"bool\""
-      "       },"
-      "       \"id\": \"d1001\""
-      "    }"
-      "}",};
+      "{\"metric\":\"meter_current1\",\"timestamp\":{\"value\":1662344042,\"type\":\"s\"},\"value\":{\"value\":10.3,\"type\":\"i64\"},\"tags\":{\"t1\":{\"value\":2,\"type\":\"bigint\"},\"t2\":{\"value\":2,\"type\":\"int\"},\"t3\":{\"value\":2,\"type\":\"i16\"},\"t4\":{\"value\":2,\"type\":\"i8\"},\"t5\":{\"value\":2,\"type\":\"f32\"},\"t6\":{\"value\":2,\"type\":\"double\"},\"t7\":{\"value\":\"8323\",\"type\":\"binary\"},\"t8\":{\"value\":\"北京\",\"type\":\"nchar\"},\"t9\":{\"value\":true,\"type\":\"bool\"},\"id\":\"d1001\"}}"};
   pRes = taos_schemaless_insert(taos, (char **)sql, sizeof(sql) / sizeof(sql[0]), TSDB_SML_JSON_PROTOCOL,
                                 TSDB_SML_TIMESTAMP_NANO_SECONDS);
   printf("%s result:%s\n", __FUNCTION__, taos_errstr(pRes));
@@ -246,47 +158,8 @@ int smlProcess_json4_Test() {
   taos_free_result(pRes);
 
   const char *sql[] = {
-      "{"
-      "    \"metric\": \"meter_current2\","
-      "    \"timestamp\": {"
-      "        \"value\"  : 1662344042000,"
-      "        \"type\"   : \"ms\""
-      "    },"
-      "    \"value\": \"ni\","
-      "    \"tags\": {"
-      "       \"t1\": { "
-      "           \"value\" : 20,"
-      "           \"type\"  : \"i64\""
-      "       },"
-      "       \"t2\": { "
-      "           \"value\" : 25,"
-      "           \"type\"  : \"i32\""
-      "       },"
-      "       \"t3\": { "
-      "           \"value\" : 2,"
-      "           \"type\"  : \"smallint\""
-      "       },"
-      "       \"t4\": { "
-      "           \"value\" : 2,"
-      "           \"type\"  : \"tinyint\""
-      "       },"
-      "       \"t5\": { "
-      "           \"value\" : 2,"
-      "           \"type\"  : \"float\""
-      "       },"
-      "       \"t6\": { "
-      "           \"value\" : 0.2,"
-      "           \"type\"  : \"f64\""
-      "       },"
-      "       \"t7\": \"nsj\","
-      "       \"t8\": { "
-      "           \"value\" : \"北京\","
-      "           \"type\"  : \"nchar\""
-      "       },"
-      "       \"t9\": false,"
-      "       \"id\": \"d1001\""
-      "    }"
-      "}",};
+      "{\"metric\":\"meter_current2\",\"timestamp\":{\"value\":1662344042000,\"type\":\"ms\"},\"value\":\"ni\",\"tags\":{\"t1\":{\"value\":20,\"type\":\"i64\"},\"t2\":{\"value\":25,\"type\":\"i32\"},\"t3\":{\"value\":2,\"type\":\"smallint\"},\"t4\":{\"value\":2,\"type\":\"tinyint\"},\"t5\":{\"value\":2,\"type\":\"float\"},\"t6\":{\"value\":0.2,\"type\":\"f64\"},\"t7\":\"nsj\",\"t8\":{\"value\":\"北京\",\"type\":\"nchar\"},\"t9\":false,\"id\":\"d1001\"}}"
+  };
   pRes = taos_schemaless_insert(taos, (char **)sql, sizeof(sql) / sizeof(sql[0]), TSDB_SML_JSON_PROTOCOL,
                                 TSDB_SML_TIMESTAMP_NANO_SECONDS);
   printf("%s result:%s\n", __FUNCTION__, taos_errstr(pRes));
diff --git a/utils/test/c/tmqDemo.c b/utils/test/c/tmqDemo.c
index 784b45c92b..d105b50579 100644
--- a/utils/test/c/tmqDemo.c
+++ b/utils/test/c/tmqDemo.c
@@ -130,15 +130,15 @@ void parseArgument(int32_t argc, char* argv[]) {
       printHelp();
       exit(0);
     } else if (strcmp(argv[i], "-d") == 0) {
-      strcpy(g_stConfInfo.dbName, argv[++i]);
+      tstrncpy(g_stConfInfo.dbName, argv[++i], sizeof(g_stConfInfo.dbName));
     } else if (strcmp(argv[i], "-c") == 0) {
-      strcpy(configDir, argv[++i]);
+      tstrncpy(configDir, argv[++i], PATH_MAX);
     } else if (strcmp(argv[i], "-s") == 0) {
-      strcpy(g_stConfInfo.stbName, argv[++i]);
+      tstrncpy(g_stConfInfo.stbName, argv[++i], sizeof(g_stConfInfo.stbName));
     } else if (strcmp(argv[i], "-w") == 0) {
-      strcpy(g_stConfInfo.vnodeWalPath, argv[++i]);
+      tstrncpy(g_stConfInfo.vnodeWalPath, argv[++i], sizeof(g_stConfInfo.vnodeWalPath));
     } else if (strcmp(argv[i], "-f") == 0) {
-      strcpy(g_stConfInfo.resultFileName, argv[++i]);
+      tstrncpy(g_stConfInfo.resultFileName, argv[++i], sizeof(g_stConfInfo.resultFileName));
     } else if (strcmp(argv[i], "-t") == 0) {
       g_stConfInfo.numOfThreads = atoi(argv[++i]);
     } else if (strcmp(argv[i], "-n") == 0) {
diff --git a/utils/test/c/tmqSim.c b/utils/test/c/tmqSim.c
index d36fe0855a..25e83363ee 100644
--- a/utils/test/c/tmqSim.c
+++ b/utils/test/c/tmqSim.c
@@ -949,7 +949,7 @@ void parseConsumeInfo() {
     token = strtok(g_stConfInfo.stThreads[i].topicString, delim);
     while (token != NULL) {
       // printf("%s\n", token );
-      strcpy(g_stConfInfo.stThreads[i].topics[g_stConfInfo.stThreads[i].numOfTopic], token);
+      tstrncpy(g_stConfInfo.stThreads[i].topics[g_stConfInfo.stThreads[i].numOfTopic], token, sizeof(g_stConfInfo.stThreads[i].topics[g_stConfInfo.stThreads[i].numOfTopic]));
       ltrim(g_stConfInfo.stThreads[i].topics[g_stConfInfo.stThreads[i].numOfTopic]);
       // printf("%s\n", g_stConfInfo.topics[g_stConfInfo.numOfTopic]);
       g_stConfInfo.stThreads[i].numOfTopic++;

From 9078f6157c763e6976ce836229efd450edb593a4 Mon Sep 17 00:00:00 2001
From: afwerar <1296468573@qq.com>
Date: Tue, 18 Oct 2022 15:37:00 +0800
Subject: [PATCH 24/42] fix: shell input Chinese char error

---
 tools/shell/src/shellCommand.c | 7 ++-----
 1 file changed, 2 insertions(+), 5 deletions(-)

diff --git a/tools/shell/src/shellCommand.c b/tools/shell/src/shellCommand.c
index 0b26d685fd..c34b22b691 100644
--- a/tools/shell/src/shellCommand.c
+++ b/tools/shell/src/shellCommand.c
@@ -101,11 +101,8 @@ void shellInsertChar(SShellCmd *cmd, char *c, int32_t size) {
   /* update the values */
   cmd->commandSize += size;
   cmd->cursorOffset += size;
-  for (int i = 0; i < size; i++) {
-    taosMbToWchar(&wc, c + i, size);
-    cmd->screenOffset += taosWcharWidth(wc);
-    cmd->endOffset    += taosWcharWidth(wc);
-  }  
+  cmd->screenOffset += taosWcharWidth(wc);
+  cmd->endOffset    += taosWcharWidth(wc);
 #ifdef WINDOWS
 #else
   shellShowOnScreen(cmd);

From 0f5f0d7cfbe8c68e18fa1a7dc946d6539330b21d Mon Sep 17 00:00:00 2001
From: Cary Xu <klxu@taosdata.com>
Date: Tue, 18 Oct 2022 15:42:28 +0800
Subject: [PATCH 25/42] chore: remove codes unused currently for sma

---
 source/dnode/vnode/src/inc/sma.h       |   2 +
 source/dnode/vnode/src/sma/smaCommit.c |   8 +
 source/dnode/vnode/src/sma/smaFS2.c    | 720 +++++++++++++++++++++++++
 source/dnode/vnode/src/sma/smaRollup.c | 420 +--------------
 source/dnode/vnode/src/sma/smaUtil.c   |   7 +-
 5 files changed, 740 insertions(+), 417 deletions(-)
 create mode 100644 source/dnode/vnode/src/sma/smaFS2.c

diff --git a/source/dnode/vnode/src/inc/sma.h b/source/dnode/vnode/src/inc/sma.h
index dade85b12d..50e3c3a9c1 100644
--- a/source/dnode/vnode/src/inc/sma.h
+++ b/source/dnode/vnode/src/inc/sma.h
@@ -258,6 +258,7 @@ enum {
   TD_FTYPE_RSMA_QTASKINFO = 0,
 };
 
+#if 0
 struct STFile {
   uint8_t   state;
   STFInfo   info;
@@ -287,6 +288,7 @@ int32_t tdUpdateTFileHeader(STFile *pTFile);
 void    tdUpdateTFileMagic(STFile *pTFile, void *pCksm);
 void    tdCloseTFile(STFile *pTFile);
 void    tdDestroyTFile(STFile *pTFile);
+#endif
 
 void tdGetVndFileName(int32_t vgId, const char *pdname, const char *dname, const char *fname, int64_t version,
                       char *outputName);
diff --git a/source/dnode/vnode/src/sma/smaCommit.c b/source/dnode/vnode/src/sma/smaCommit.c
index db5f4c55b9..6168a00815 100644
--- a/source/dnode/vnode/src/sma/smaCommit.c
+++ b/source/dnode/vnode/src/sma/smaCommit.c
@@ -17,14 +17,17 @@
 
 extern SSmaMgmt smaMgmt;
 
+#if 0
 static int32_t tdProcessRSmaSyncPreCommitImpl(SSma *pSma);
 static int32_t tdProcessRSmaSyncCommitImpl(SSma *pSma);
 static int32_t tdProcessRSmaSyncPostCommitImpl(SSma *pSma);
+#endif
 static int32_t tdProcessRSmaAsyncPreCommitImpl(SSma *pSma);
 static int32_t tdProcessRSmaAsyncCommitImpl(SSma *pSma);
 static int32_t tdProcessRSmaAsyncPostCommitImpl(SSma *pSma);
 static int32_t tdUpdateQTaskInfoFiles(SSma *pSma, SRSmaStat *pRSmaStat);
 
+#if 0
 /**
  * @brief Only applicable to Rollup SMA
  *
@@ -48,6 +51,7 @@ int32_t smaSyncCommit(SSma *pSma) { return tdProcessRSmaSyncCommitImpl(pSma); }
  * @return int32_t
  */
 int32_t smaSyncPostCommit(SSma *pSma) { return tdProcessRSmaSyncPostCommitImpl(pSma); }
+#endif
 
 /**
  * @brief Only applicable to Rollup SMA
@@ -108,6 +112,7 @@ int32_t smaBegin(SSma *pSma) {
   return TSDB_CODE_SUCCESS;
 }
 
+#if 0
 /**
  * @brief pre-commit for rollup sma(sync commit).
  *  1) set trigger stat of rsma timer TASK_TRIGGER_STAT_PAUSED.
@@ -169,6 +174,7 @@ static int32_t tdProcessRSmaSyncCommitImpl(SSma *pSma) {
 #endif
   return TSDB_CODE_SUCCESS;
 }
+#endif
 
 // SQTaskFile ======================================================
 
@@ -230,6 +236,7 @@ static int32_t tdUpdateQTaskInfoFiles(SSma *pSma, SRSmaStat *pStat) {
   return TSDB_CODE_SUCCESS;
 }
 
+#if 0
 /**
  * @brief post-commit for rollup sma
  *  1) clean up the outdated qtaskinfo files
@@ -249,6 +256,7 @@ static int32_t tdProcessRSmaSyncPostCommitImpl(SSma *pSma) {
 
   return TSDB_CODE_SUCCESS;
 }
+#endif
 
 /**
  * @brief Rsma async commit implementation(only do some necessary light weighted task)
diff --git a/source/dnode/vnode/src/sma/smaFS2.c b/source/dnode/vnode/src/sma/smaFS2.c
new file mode 100644
index 0000000000..9dced9c23b
--- /dev/null
+++ b/source/dnode/vnode/src/sma/smaFS2.c
@@ -0,0 +1,720 @@
+/*
+ * Copyright (c) 2019 TAOS Data, Inc. <jhtao@taosdata.com>
+ *
+ * This program is free software: you can use, redistribute, and/or modify
+ * it under the terms of the GNU Affero General Public License, version 3
+ * or later ("AGPL"), as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.
+ *
+ * You should have received a copy of the GNU Affero General Public License
+ * along with this program. If not, see <http://www.gnu.org/licenses/>.
+ */
+
+#include "sma.h"
+
+// =================================================================================================
+static int32_t tdRSmaEncodeFS(uint8_t *p, SRSmaFS *pFS);
+static int32_t tPutQTaskFile(uint8_t *p, SQTaskFile *pFile);
+static int32_t tPutQTaskFileSet(uint8_t *p, SQTaskFileSet *pSet);
+
+static int32_t tdFetchQTaskInfoFiles(SSma *pSma, int64_t version, SArray **output);
+static int32_t tdQTaskInfCmprFn1(const void *p1, const void *p2);
+static int32_t tdQTaskInfCmprFn2(const void *p1, const void *p2);
+
+// =================================================================================================
+static int32_t tdRSmaEncodeFS(uint8_t *p, SRSmaFS *pFS) {
+  int32_t  n = 0;
+  uint32_t nSet = taosArrayGetSize(pFS->aQTaskFSet);
+
+  // version
+  n += tPutU8(p ? p + n : p, 0);
+
+  // SArray<SQTaskFileSet>
+  n += tPutU32v(p ? p + n : p, nSet);
+  for (uint32_t i = 0; i < nSet; ++i) {
+    n += tPutQTaskFileSet(p ? p + n : p, (SQTaskFileSet *)taosArrayGet(pFS->aQTaskFSet, i));
+  }
+
+  return n;
+}
+
+static int32_t tPutQTaskFileSet(uint8_t *p, SQTaskFileSet *pSet) {
+  int32_t n = 0;
+  int32_t nQTaskF = taosArrayGetSize(pSet->aQTaskF);
+
+  n += tPutI64v(p ? p + n : p, pSet->suid);
+  n += tPutI32v(p ? p + n : p, nQTaskF);
+
+  // SQTaskFile
+  for (int32_t i = 0; i < nQTaskF; ++i) {
+    n += tPutQTaskFile(p ? p + n : p, taosArrayGet(pSet->aQTaskF, i));
+  }
+
+  return n;
+}
+
+static int32_t tPutQTaskFile(uint8_t *p, SQTaskFile *pFile) {
+  int32_t n = 0;
+
+  n += tPutI8(p ? p + n : p, pFile->level);
+  n += tPutI64v(p ? p + n : p, pFile->version);
+  n += tPutI64v(p ? p + n : p, pFile->mtime);
+
+  return n;
+}
+
+static int32_t tGetQTaskFileSet(uint8_t *p, SQTaskFileSet *pSet) {
+  int32_t n = 0;
+  int32_t nQTaskF = 0;
+
+  n += tGetI64v(p + n, &pSet->suid);
+  n += tGetI32v(p + n, &nQTaskF);
+
+  ASSERT(nQTaskF >= 0);
+
+  pSet->aQTaskF = taosArrayInit(nQTaskF, sizeof(SQTaskFile));
+  if (!pSet->aQTaskF) {
+    terrno = TSDB_CODE_OUT_OF_MEMORY;
+    return -1;
+  }
+
+  for (int32_t i = 0; i < nQTaskF; ++i) {
+    SQTaskFile *pFile = taosArrayGet(pSet->aQTaskF, i);
+    pFile->nRef = 1;
+    n += tGetQTaskFile(p + n, pFile);
+  }
+
+  return n;
+}
+
+static int32_t tGetQTaskFile(uint8_t *p, SQTaskFile *pFile) {
+  int32_t n = 0;
+
+  n += tGetI8(p + n, &pFile->level);
+  n += tGetI64v(p + n, &pFile->version);
+  n += tGetI64v(p + n, &pFile->mtime);
+
+  return n;
+}
+
+static int32_t tdRSmaGnrtFS(SSma *pSma, SRSmaFS *pFS, char *fname) {
+  int32_t   code = 0;
+  int64_t   n;
+  int64_t   size;
+  uint8_t  *pData = NULL;
+  TdFilePtr pFD = NULL;
+
+  // to binary
+  size = tdRSmaEncodeFS(NULL, pFS) + sizeof(TSCKSUM);
+  pData = taosMemoryMalloc(size);
+  if (pData == NULL) {
+    code = TSDB_CODE_OUT_OF_MEMORY;
+    goto _err;
+  }
+  n = tdRSmaEncodeFS(pData, pFS);
+  ASSERT(n + sizeof(TSCKSUM) == size);
+  taosCalcChecksumAppend(0, pData, size);
+
+  // create and write
+  pFD = taosOpenFile(fname, TD_FILE_WRITE | TD_FILE_CREATE | TD_FILE_TRUNC);
+  if (pFD == NULL) {
+    code = TAOS_SYSTEM_ERROR(errno);
+    goto _err;
+  }
+
+  n = taosWriteFile(pFD, pData, size);
+  if (n < 0) {
+    code = TAOS_SYSTEM_ERROR(errno);
+    goto _err;
+  }
+
+  if (taosFsyncFile(pFD) < 0) {
+    code = TAOS_SYSTEM_ERROR(errno);
+    goto _err;
+  }
+
+  taosCloseFile(&pFD);
+
+  if (pData) taosMemoryFree(pData);
+  return code;
+
+_err:
+  smaError("vgId:%d, rsma gnrt fs failed since %s", SMA_VID(pSma), tstrerror(code));
+  if (pData) taosMemoryFree(pData);
+  return code;
+}
+
+int32_t tdRSmaFSCommit1(SSma *pSma, SRSmaFS *pFSNew) {
+  int32_t code = 0;
+  char    tfname[TSDB_FILENAME_LEN];
+  char    fname[TSDB_FILENAME_LEN];
+
+  snprintf(tfname, TSDB_FILENAME_LEN - 1, "%s%s%s%sRSMAFS.t", tfsGetPrimaryPath(pSma->pVnode->pTfs), TD_DIRSEP,
+           VNODE_RSMA_DIR, TD_DIRSEP);
+  snprintf(fname, TSDB_FILENAME_LEN - 1, "%s%s%s%sRSMAFS", tfsGetPrimaryPath(pSma->pVnode->pTfs), TD_DIRSEP,
+           VNODE_RSMA_DIR, TD_DIRSEP);
+
+  // gnrt CURRENT.t
+  code = tdRSmaGnrtFS(pSma, pFSNew, tfname);
+  if (code) goto _err;
+
+  // rename
+  code = taosRenameFile(tfname, fname);
+  if (code) {
+    code = TAOS_SYSTEM_ERROR(code);
+    goto _err;
+  }
+
+  return code;
+
+_err:
+  smaError("vgId:%d, rsma fs commit phase 1 failed since %s", SMA_VID(pSma), tstrerror(code));
+  return code;
+}
+
+int32_t tdRSmaFSMergeQTaskFSet(SQTaskFileSet *pSetOld, SQTaskFileSet *pSetNew) {
+  SQTaskFileSet *
+
+}
+
+int32_t tdRSmaFSCommit2(SSma *pSma, SRSmaFS *pFSNew) {
+  SSmaEnv   *pEnv = SMA_RSMA_ENV(pSma);
+  SRSmaStat *pStat = NULL;
+  SRSmaInfo *pRSmaInfo = NULL;
+  SRSmaFS   *pRSmaFS = NULL;
+  int32_t    code = 0;
+  int32_t    nRef;
+  char       fname[TSDB_FILENAME_LEN];
+
+  if (!pEnv) {
+    terrno = TSDB_CODE_RSMA_INVALID_ENV;
+    return TSDB_CODE_FAILED;
+  }
+
+  pStat = (SRSmaStat *)SMA_ENV_STAT(pEnv);
+  if (!pStat || !(pRSmaFS = RSMA_FS(pStat))) {
+    terrno = TSDB_CODE_RSMA_INVALID_STAT;
+    return TSDB_CODE_FAILED;
+  }
+
+  // data
+  int32_t iOld = 0;
+  int32_t iNew = 0;
+  while (true) {
+    int32_t       nOld = taosArrayGetSize(pRSmaFS->aQTaskFSet);
+    int32_t       nNew = taosArrayGetSize(pFSNew->aQTaskFSet);
+    SQTaskFileSet fSet;
+
+    if (iOld >= nOld && iNew >= nNew) break;
+
+    SQTaskFileSet *pSetOld = (iOld < nOld) ? taosArrayGet(pRSmaFS->aQTaskFSet, iOld) : NULL;
+    SQTaskFileSet *pSetNew = (iNew < nNew) ? taosArrayGet(pFSNew->aQTaskFSet, iNew) : NULL;
+
+    if (pSetOld && pSetNew) {
+      if (pSetOld->suid == pSetNew->suid) {
+        goto _merge_old_and_new;
+      } else if (pSetOld->suid < pSetNew->suid) {
+        goto _remove_old;
+      } else {
+        goto _add_new;
+      }
+    } else if (pSetOld) {
+      goto _remove_old;
+    } else {
+      goto _add_new;
+    }
+
+  _merge_old_and_new:
+
+
+
+    // head
+    fSet.pHeadF = pSetOld->pHeadF;
+    if ((!sameDisk) || (pSetOld->pHeadF->commitID != pSetNew->pHeadF->commitID)) {
+      pSetOld->pHeadF = (SHeadFile *)taosMemoryMalloc(sizeof(SHeadFile));
+      if (pSetOld->pHeadF == NULL) {
+        code = TSDB_CODE_OUT_OF_MEMORY;
+        goto _err;
+      }
+      *pSetOld->pHeadF = *pSetNew->pHeadF;
+      pSetOld->pHeadF->nRef = 1;
+
+      nRef = atomic_sub_fetch_32(&fSet.pHeadF->nRef, 1);
+      if (nRef == 0) {
+        tsdbHeadFileName(pTsdb, pSetOld->diskId, pSetOld->fid, fSet.pHeadF, fname);
+        taosRemoveFile(fname);
+        taosMemoryFree(fSet.pHeadF);
+      }
+    } else {
+      ASSERT(fSet.pHeadF->size == pSetNew->pHeadF->size);
+      ASSERT(fSet.pHeadF->offset == pSetNew->pHeadF->offset);
+    }
+
+    // data
+    fSet.pDataF = pSetOld->pDataF;
+    if ((!sameDisk) || (pSetOld->pDataF->commitID != pSetNew->pDataF->commitID)) {
+      pSetOld->pDataF = (SDataFile *)taosMemoryMalloc(sizeof(SDataFile));
+      if (pSetOld->pDataF == NULL) {
+        code = TSDB_CODE_OUT_OF_MEMORY;
+        goto _err;
+      }
+      *pSetOld->pDataF = *pSetNew->pDataF;
+      pSetOld->pDataF->nRef = 1;
+
+      nRef = atomic_sub_fetch_32(&fSet.pDataF->nRef, 1);
+      if (nRef == 0) {
+        tsdbDataFileName(pTsdb, pSetOld->diskId, pSetOld->fid, fSet.pDataF, fname);
+        taosRemoveFile(fname);
+        taosMemoryFree(fSet.pDataF);
+      }
+    } else {
+      ASSERT(pSetOld->pDataF->size <= pSetNew->pDataF->size);
+      pSetOld->pDataF->size = pSetNew->pDataF->size;
+    }
+
+    // sma
+    fSet.pSmaF = pSetOld->pSmaF;
+    if ((!sameDisk) || (pSetOld->pSmaF->commitID != pSetNew->pSmaF->commitID)) {
+      pSetOld->pSmaF = (SSmaFile *)taosMemoryMalloc(sizeof(SSmaFile));
+      if (pSetOld->pSmaF == NULL) {
+        code = TSDB_CODE_OUT_OF_MEMORY;
+        goto _err;
+      }
+      *pSetOld->pSmaF = *pSetNew->pSmaF;
+      pSetOld->pSmaF->nRef = 1;
+
+      nRef = atomic_sub_fetch_32(&fSet.pSmaF->nRef, 1);
+      if (nRef == 0) {
+        tsdbSmaFileName(pTsdb, pSetOld->diskId, pSetOld->fid, fSet.pSmaF, fname);
+        taosRemoveFile(fname);
+        taosMemoryFree(fSet.pSmaF);
+      }
+    } else {
+      ASSERT(pSetOld->pSmaF->size <= pSetNew->pSmaF->size);
+      pSetOld->pSmaF->size = pSetNew->pSmaF->size;
+    }
+
+    // stt
+    if (sameDisk) {
+      if (pSetNew->nSttF > pSetOld->nSttF) {
+        ASSERT(pSetNew->nSttF = pSetOld->nSttF + 1);
+        pSetOld->aSttF[pSetOld->nSttF] = (SSttFile *)taosMemoryMalloc(sizeof(SSttFile));
+        if (pSetOld->aSttF[pSetOld->nSttF] == NULL) {
+          code = TSDB_CODE_OUT_OF_MEMORY;
+          goto _err;
+        }
+        *pSetOld->aSttF[pSetOld->nSttF] = *pSetNew->aSttF[pSetOld->nSttF];
+        pSetOld->aSttF[pSetOld->nSttF]->nRef = 1;
+        pSetOld->nSttF++;
+      } else if (pSetNew->nSttF < pSetOld->nSttF) {
+        ASSERT(pSetNew->nSttF == 1);
+        for (int32_t iStt = 0; iStt < pSetOld->nSttF; iStt++) {
+          SSttFile *pSttFile = pSetOld->aSttF[iStt];
+          nRef = atomic_sub_fetch_32(&pSttFile->nRef, 1);
+          if (nRef == 0) {
+            tsdbSttFileName(pTsdb, pSetOld->diskId, pSetOld->fid, pSttFile, fname);
+            taosRemoveFile(fname);
+            taosMemoryFree(pSttFile);
+          }
+          pSetOld->aSttF[iStt] = NULL;
+        }
+
+        pSetOld->nSttF = 1;
+        pSetOld->aSttF[0] = (SSttFile *)taosMemoryMalloc(sizeof(SSttFile));
+        if (pSetOld->aSttF[0] == NULL) {
+          code = TSDB_CODE_OUT_OF_MEMORY;
+          goto _err;
+        }
+        *pSetOld->aSttF[0] = *pSetNew->aSttF[0];
+        pSetOld->aSttF[0]->nRef = 1;
+      } else {
+        for (int32_t iStt = 0; iStt < pSetOld->nSttF; iStt++) {
+          if (pSetOld->aSttF[iStt]->commitID != pSetNew->aSttF[iStt]->commitID) {
+            SSttFile *pSttFile = pSetOld->aSttF[iStt];
+            nRef = atomic_sub_fetch_32(&pSttFile->nRef, 1);
+            if (nRef == 0) {
+              tsdbSttFileName(pTsdb, pSetOld->diskId, pSetOld->fid, pSttFile, fname);
+              taosRemoveFile(fname);
+              taosMemoryFree(pSttFile);
+            }
+
+            pSetOld->aSttF[iStt] = (SSttFile *)taosMemoryMalloc(sizeof(SSttFile));
+            if (pSetOld->aSttF[iStt] == NULL) {
+              code = TSDB_CODE_OUT_OF_MEMORY;
+              goto _err;
+            }
+            *pSetOld->aSttF[iStt] = *pSetNew->aSttF[iStt];
+            pSetOld->aSttF[iStt]->nRef = 1;
+          } else {
+            ASSERT(pSetOld->aSttF[iStt]->size == pSetOld->aSttF[iStt]->size);
+            ASSERT(pSetOld->aSttF[iStt]->offset == pSetOld->aSttF[iStt]->offset);
+          }
+        }
+      }
+    } else {
+      ASSERT(pSetOld->nSttF == pSetNew->nSttF);
+      for (int32_t iStt = 0; iStt < pSetOld->nSttF; iStt++) {
+        SSttFile *pSttFile = pSetOld->aSttF[iStt];
+        nRef = atomic_sub_fetch_32(&pSttFile->nRef, 1);
+        if (nRef == 0) {
+          tsdbSttFileName(pTsdb, pSetOld->diskId, pSetOld->fid, pSttFile, fname);
+          taosRemoveFile(fname);
+          taosMemoryFree(pSttFile);
+        }
+
+        pSetOld->aSttF[iStt] = (SSttFile *)taosMemoryMalloc(sizeof(SSttFile));
+        if (pSetOld->aSttF[iStt] == NULL) {
+          code = TSDB_CODE_OUT_OF_MEMORY;
+          goto _err;
+        }
+        *pSetOld->aSttF[iStt] = *pSetNew->aSttF[iStt];
+        pSetOld->aSttF[iStt]->nRef = 1;
+      }
+    }
+
+    if (!sameDisk) {
+      pSetOld->diskId = pSetNew->diskId;
+    }
+
+    iOld++;
+    iNew++;
+    continue;
+
+  _remove_old:
+    nRef = atomic_sub_fetch_32(&pSetOld->pHeadF->nRef, 1);
+    if (nRef == 0) {
+      tsdbHeadFileName(pTsdb, pSetOld->diskId, pSetOld->fid, pSetOld->pHeadF, fname);
+      taosRemoveFile(fname);
+      taosMemoryFree(pSetOld->pHeadF);
+    }
+
+    nRef = atomic_sub_fetch_32(&pSetOld->pDataF->nRef, 1);
+    if (nRef == 0) {
+      tsdbDataFileName(pTsdb, pSetOld->diskId, pSetOld->fid, pSetOld->pDataF, fname);
+      taosRemoveFile(fname);
+      taosMemoryFree(pSetOld->pDataF);
+    }
+
+    nRef = atomic_sub_fetch_32(&pSetOld->pSmaF->nRef, 1);
+    if (nRef == 0) {
+      tsdbSmaFileName(pTsdb, pSetOld->diskId, pSetOld->fid, pSetOld->pSmaF, fname);
+      taosRemoveFile(fname);
+      taosMemoryFree(pSetOld->pSmaF);
+    }
+
+    for (int8_t iStt = 0; iStt < pSetOld->nSttF; iStt++) {
+      nRef = atomic_sub_fetch_32(&pSetOld->aSttF[iStt]->nRef, 1);
+      if (nRef == 0) {
+        tsdbSttFileName(pTsdb, pSetOld->diskId, pSetOld->fid, pSetOld->aSttF[iStt], fname);
+        taosRemoveFile(fname);
+        taosMemoryFree(pSetOld->aSttF[iStt]);
+      }
+    }
+
+    taosArrayRemove(pTsdb->fs.aDFileSet, iOld);
+    continue;
+
+  _add_new:
+    fSet = (SDFileSet){.diskId = pSetNew->diskId, .fid = pSetNew->fid, .nSttF = 1};
+
+    // head
+    fSet.pHeadF = (SHeadFile *)taosMemoryMalloc(sizeof(SHeadFile));
+    if (fSet.pHeadF == NULL) {
+      code = TSDB_CODE_OUT_OF_MEMORY;
+      goto _err;
+    }
+    *fSet.pHeadF = *pSetNew->pHeadF;
+    fSet.pHeadF->nRef = 1;
+
+    // data
+    fSet.pDataF = (SDataFile *)taosMemoryMalloc(sizeof(SDataFile));
+    if (fSet.pDataF == NULL) {
+      code = TSDB_CODE_OUT_OF_MEMORY;
+      goto _err;
+    }
+    *fSet.pDataF = *pSetNew->pDataF;
+    fSet.pDataF->nRef = 1;
+
+    // sma
+    fSet.pSmaF = (SSmaFile *)taosMemoryMalloc(sizeof(SSmaFile));
+    if (fSet.pSmaF == NULL) {
+      code = TSDB_CODE_OUT_OF_MEMORY;
+      goto _err;
+    }
+    *fSet.pSmaF = *pSetNew->pSmaF;
+    fSet.pSmaF->nRef = 1;
+
+    // stt
+    ASSERT(pSetNew->nSttF == 1);
+    fSet.aSttF[0] = (SSttFile *)taosMemoryMalloc(sizeof(SSttFile));
+    if (fSet.aSttF[0] == NULL) {
+      code = TSDB_CODE_OUT_OF_MEMORY;
+      goto _err;
+    }
+    *fSet.aSttF[0] = *pSetNew->aSttF[0];
+    fSet.aSttF[0]->nRef = 1;
+
+    if (taosArrayInsert(pTsdb->fs.aDFileSet, iOld, &fSet) == NULL) {
+      code = TSDB_CODE_OUT_OF_MEMORY;
+      goto _err;
+    }
+    iOld++;
+    iNew++;
+    continue;
+  }
+
+  return code;
+
+_err:
+  tsdbError("vgId:%d, tsdb fs commit phase 2 failed since %s", TD_VID(pTsdb->pVnode), tstrerror(code));
+  return code;
+}
+
+/**
+ * @brief Open RSma FS from qTaskInfo files
+ *
+ * @param pSma
+ * @param version
+ * @return int32_t
+ */
+int32_t tdRSmaFSOpen(SSma *pSma, int64_t version) {
+  SVnode    *pVnode = pSma->pVnode;
+  int64_t    commitID = pVnode->state.commitID;
+  SSmaEnv   *pEnv = SMA_RSMA_ENV(pSma);
+  SRSmaStat *pStat = NULL;
+  SArray    *output = NULL;
+
+  terrno = TSDB_CODE_SUCCESS;
+
+  if (!pEnv) {
+    return TSDB_CODE_SUCCESS;
+  }
+
+  if (tdFetchQTaskInfoFiles(pSma, version, &output) < 0) {
+    goto _end;
+  }
+
+  pStat = (SRSmaStat *)SMA_ENV_STAT(pEnv);
+
+  for (int32_t i = 0; i < taosArrayGetSize(output); ++i) {
+    int32_t vid = 0;
+    int64_t version = -1;
+    sscanf((const char *)taosArrayGetP(output, i), "v%dqinf.v%" PRIi64, &vid, &version);
+    SQTaskFile qTaskFile = {.version = version, .nRef = 1};
+    if ((terrno = tdRSmaFSUpsertQTaskFile(RSMA_FS(pStat), &qTaskFile)) < 0) {
+      goto _end;
+    }
+    smaInfo("vgId:%d, open fs, version:%" PRIi64 ", ref:%d", TD_VID(pVnode), qTaskFile.version, qTaskFile.nRef);
+  }
+
+_end:
+  for (int32_t i = 0; i < taosArrayGetSize(output); ++i) {
+    void *ptr = taosArrayGetP(output, i);
+    taosMemoryFreeClear(ptr);
+  }
+  taosArrayDestroy(output);
+
+  if (terrno != TSDB_CODE_SUCCESS) {
+    smaError("vgId:%d, open rsma fs failed since %s", TD_VID(pVnode), terrstr());
+    return TSDB_CODE_FAILED;
+  }
+  return TSDB_CODE_SUCCESS;
+}
+
+void tdRSmaFSClose(SRSmaFS *fs) { taosArrayDestroy(fs->aQTaskFSet); }
+
+static int32_t tdQTaskInfCmprFn1(const void *p1, const void *p2) {
+  if (*(int64_t *)p1 < ((SQTaskFile *)p2)->version) {
+    return -1;
+  } else if (*(int64_t *)p1 > ((SQTaskFile *)p2)->version) {
+    return 1;
+  }
+  return 0;
+}
+
+int32_t tdRSmaFSRef(SSma *pSma, SRSmaStat *pStat, int64_t version) {
+  SArray     *aQTaskFSet = RSMA_FS(pStat)->aQTaskFSet;
+  SQTaskFile *pTaskF = NULL;
+  int32_t     oldVal = 0;
+
+  taosRLockLatch(RSMA_FS_LOCK(pStat));
+  if ((pTaskF = taosArraySearch(aQTaskFSet, &version, tdQTaskInfCmprFn1, TD_EQ))) {
+    oldVal = atomic_fetch_add_32(&pTaskF->nRef, 1);
+    ASSERT(oldVal > 0);
+  }
+  taosRUnLockLatch(RSMA_FS_LOCK(pStat));
+  return oldVal;
+}
+
+int64_t tdRSmaFSMaxVer(SSma *pSma, SRSmaStat *pStat) {
+  SArray *aQTaskFSet = RSMA_FS(pStat)->aQTaskFSet;
+  int64_t version = -1;
+
+  taosRLockLatch(RSMA_FS_LOCK(pStat));
+  if (taosArrayGetSize(aQTaskFSet) > 0) {
+    version = ((SQTaskFile *)taosArrayGetLast(aQTaskFSet))->version;
+  }
+  taosRUnLockLatch(RSMA_FS_LOCK(pStat));
+  return version;
+}
+
+void tdRSmaFSUnRef(SSma *pSma, SRSmaStat *pStat, int64_t version) {
+  SVnode     *pVnode = pSma->pVnode;
+  SArray     *aQTaskFSet = RSMA_FS(pStat)->aQTaskFSet;
+  char        qTaskFullName[TSDB_FILENAME_LEN];
+  SQTaskFile *pTaskF = NULL;
+  int32_t     idx = -1;
+
+  taosWLockLatch(RSMA_FS_LOCK(pStat));
+  if ((idx = taosArraySearchIdx(aQTaskFSet, &version, tdQTaskInfCmprFn1, TD_EQ)) >= 0) {
+    ASSERT(idx < taosArrayGetSize(aQTaskFSet));
+    pTaskF = taosArrayGet(aQTaskFSet, idx);
+    if (atomic_sub_fetch_32(&pTaskF->nRef, 1) <= 0) {
+      tdRSmaQTaskInfoGetFullName(TD_VID(pVnode), pTaskF->version, tfsGetPrimaryPath(pVnode->pTfs), qTaskFullName);
+      if (taosRemoveFile(qTaskFullName) < 0) {
+        smaWarn("vgId:%d, failed to remove %s since %s", TD_VID(pVnode), qTaskFullName,
+                tstrerror(TAOS_SYSTEM_ERROR(errno)));
+      } else {
+        smaDebug("vgId:%d, success to remove %s", TD_VID(pVnode), qTaskFullName);
+      }
+      taosArrayRemove(aQTaskFSet, idx);
+    }
+  }
+  taosWUnLockLatch(RSMA_FS_LOCK(pStat));
+}
+
+/**
+ * @brief Fetch qtaskfiles LE than version
+ *
+ * @param pSma
+ * @param version
+ * @param output
+ * @return int32_t
+ */
+static int32_t tdFetchQTaskInfoFiles(SSma *pSma, int64_t version, SArray **output) {
+  SVnode       *pVnode = pSma->pVnode;
+  TdDirPtr      pDir = NULL;
+  TdDirEntryPtr pDirEntry = NULL;
+  char          dir[TSDB_FILENAME_LEN];
+  const char   *pattern = "v[0-9]+qinf\\.v([0-9]+)?$";
+  regex_t       regex;
+  int           code = 0;
+
+  terrno = TSDB_CODE_SUCCESS;
+
+  tdGetVndDirName(TD_VID(pVnode), tfsGetPrimaryPath(pVnode->pTfs), VNODE_RSMA_DIR, true, dir);
+
+  if (!taosCheckExistFile(dir)) {
+    smaDebug("vgId:%d, fetch qtask files, no need as dir %s not exist", TD_VID(pVnode), dir);
+    return TSDB_CODE_SUCCESS;
+  }
+
+  // Resource allocation and init
+  if ((code = regcomp(&regex, pattern, REG_EXTENDED)) != 0) {
+    terrno = TSDB_CODE_RSMA_REGEX_MATCH;
+    char errbuf[128];
+    regerror(code, &regex, errbuf, sizeof(errbuf));
+    smaWarn("vgId:%d, fetch qtask files, regcomp for %s failed since %s", TD_VID(pVnode), dir, errbuf);
+    return TSDB_CODE_FAILED;
+  }
+
+  if (!(pDir = taosOpenDir(dir))) {
+    regfree(&regex);
+    terrno = TAOS_SYSTEM_ERROR(errno);
+    smaError("vgId:%d, fetch qtask files, open dir %s failed since %s", TD_VID(pVnode), dir, terrstr());
+    return TSDB_CODE_FAILED;
+  }
+
+  int32_t    dirLen = strlen(dir);
+  char      *dirEnd = POINTER_SHIFT(dir, dirLen);
+  regmatch_t regMatch[2];
+  while ((pDirEntry = taosReadDir(pDir))) {
+    char *entryName = taosGetDirEntryName(pDirEntry);
+    if (!entryName) {
+      continue;
+    }
+
+    code = regexec(&regex, entryName, 2, regMatch, 0);
+
+    if (code == 0) {
+      // match
+      smaInfo("vgId:%d, fetch qtask files, max ver:%" PRIi64 ", %s found", TD_VID(pVnode), version, entryName);
+
+      int64_t ver = -1;
+      sscanf((const char *)POINTER_SHIFT(entryName, regMatch[1].rm_so), "%" PRIi64, &ver);
+      if ((ver <= version) && (ver > -1)) {
+        if (!(*output)) {
+          if (!(*output = taosArrayInit(1, POINTER_BYTES))) {
+            terrno = TSDB_CODE_OUT_OF_MEMORY;
+            goto _end;
+          }
+        }
+        char *entryDup = strdup(entryName);
+        if (!entryDup) {
+          terrno = TSDB_CODE_OUT_OF_MEMORY;
+          goto _end;
+        }
+        if (!taosArrayPush(*output, &entryDup)) {
+          terrno = TSDB_CODE_OUT_OF_MEMORY;
+          goto _end;
+        }
+      } else {
+      }
+    } else if (code == REG_NOMATCH) {
+      // not match
+      smaTrace("vgId:%d, fetch qtask files, not match %s", TD_VID(pVnode), entryName);
+      continue;
+    } else {
+      // has other error
+      char errbuf[128];
+      regerror(code, &regex, errbuf, sizeof(errbuf));
+      smaWarn("vgId:%d, fetch qtask files, regexec failed since %s", TD_VID(pVnode), errbuf);
+      terrno = TSDB_CODE_RSMA_REGEX_MATCH;
+      goto _end;
+    }
+  }
+_end:
+  taosCloseDir(&pDir);
+  regfree(&regex);
+  return terrno == 0 ? TSDB_CODE_SUCCESS : TSDB_CODE_FAILED;
+}
+
+static int32_t tdQTaskFileCmprFn2(const void *p1, const void *p2) {
+  if (((SQTaskFile *)p1)->version < ((SQTaskFile *)p2)->version) {
+    return -1;
+  } else if (((SQTaskFile *)p1)->version > ((SQTaskFile *)p2)->version) {
+    return 1;
+  }
+
+  return 0;
+}
+
+int32_t tdRSmaFSUpsertQTaskFile(SRSmaFS *pFS, SQTaskFile *qTaskFile) {
+  int32_t code = 0;
+  int32_t idx = taosArraySearchIdx(pFS->aQTaskFSet, qTaskFile, tdQTaskFileCmprFn2, TD_GE);
+
+  if (idx < 0) {
+    idx = taosArrayGetSize(pFS->aQTaskFSet);
+  } else {
+    SQTaskFile *pTaskF = (SQTaskFile *)taosArrayGet(pFS->aQTaskFSet, idx);
+    int32_t     c = tdQTaskFileCmprFn2(pTaskF, qTaskFile);
+    if (c == 0) {
+      pTaskF->nRef = qTaskFile->nRef;
+      pTaskF->level = qTaskFile->level;
+      pTaskF->version = qTaskFile->version;
+      pTaskF->mtime = qTaskFile->mtime;
+      goto _exit;
+    }
+  }
+
+  if (taosArrayInsert(pFS->aQTaskFSet, idx, qTaskFile) == NULL) {
+    code = TSDB_CODE_OUT_OF_MEMORY;
+    goto _exit;
+  }
+
+_exit:
+  return code;
+}
\ No newline at end of file
diff --git a/source/dnode/vnode/src/sma/smaRollup.c b/source/dnode/vnode/src/sma/smaRollup.c
index 722a3f479e..77c5955098 100644
--- a/source/dnode/vnode/src/sma/smaRollup.c
+++ b/source/dnode/vnode/src/sma/smaRollup.c
@@ -15,8 +15,6 @@
 
 #include "sma.h"
 
-#define RSMA_QTASKINFO_BUFSIZE     (32768)                                               // size
-#define RSMA_QTASKINFO_HEAD_LEN    (sizeof(int32_t) + sizeof(int8_t) + sizeof(int64_t))  // len + type + suid
 #define RSMA_QTASKEXEC_SMOOTH_SIZE (100)                                                 // cnt
 #define RSMA_SUBMIT_BATCH_SIZE     (1024)                                                // cnt
 #define RSMA_FETCH_DELAY_MAX       (120000)                                              // ms
@@ -48,23 +46,10 @@ static int32_t    tdRSmaExecAndSubmitResult(SSma *pSma, qTaskInfo_t taskInfo, SR
 static void       tdRSmaFetchTrigger(void *param, void *tmrId);
 static int32_t    tdRSmaInfoClone(SSma *pSma, SRSmaInfo *pInfo);
 static void       tdRSmaQTaskInfoFree(qTaskInfo_t *taskHandle, int32_t vgId, int32_t level);
-static int32_t    tdRSmaQTaskInfoIterInit(SRSmaQTaskInfoIter *pIter, STFile *pTFile);
-static int32_t    tdRSmaQTaskInfoIterNextBlock(SRSmaQTaskInfoIter *pIter, bool *isFinish);
-static int32_t    tdRSmaQTaskInfoRestore(SSma *pSma, int8_t type, SRSmaQTaskInfoIter *pIter);
-static int32_t    tdRSmaQTaskInfoItemRestore(SSma *pSma, const SRSmaQTaskInfoItem *infoItem);
 static int32_t    tdRSmaRestoreQTaskInfoInit(SSma *pSma, int64_t *nTables);
 static int32_t    tdRSmaRestoreQTaskInfoReload(SSma *pSma, int8_t type, int64_t qTaskFileVer);
 static int32_t    tdRSmaRestoreTSDataReload(SSma *pSma);
 
-static SRSmaInfo *tdGetRSmaInfoByItem(SRSmaInfoItem *pItem) {
-  // adapt accordingly if definition of SRSmaInfo update
-  SRSmaInfo *pResult = NULL;
-  ASSERT(pItem->level == TSDB_RETENTION_L1 || pItem->level == TSDB_RETENTION_L2);
-  pResult = (SRSmaInfo *)POINTER_SHIFT(pItem, -(sizeof(SRSmaInfoItem) * (pItem->level - 1) + RSMA_INFO_HEAD_LEN));
-  ASSERT(pResult->pTSchema->numOfCols > 1);
-  return pResult;
-}
-
 struct SRSmaQTaskInfoItem {
   int32_t len;
   int8_t  type;
@@ -104,12 +89,6 @@ void tdRSmaQTaskInfoGetFullPathEx(int32_t vgId, tb_uid_t suid, int8_t level, con
   snprintf(outputName + rsmaLen, TSDB_FILENAME_LEN - rsmaLen, "%" PRIi64 "%s%" PRIi8, suid, TD_DIRSEP, level);
 }
 
-static FORCE_INLINE int32_t tdRSmaQTaskInfoContLen(int32_t lenWithHead) {
-  return lenWithHead - RSMA_QTASKINFO_HEAD_LEN;
-}
-
-static FORCE_INLINE void tdRSmaQTaskInfoIterDestroy(SRSmaQTaskInfoIter *pIter) { taosMemoryFreeClear(pIter->pBuf); }
-
 static void tdRSmaQTaskInfoFree(qTaskInfo_t *taskHandle, int32_t vgId, int32_t level) {
   // Note: free/kill may in RC
   if (!taskHandle || !(*taskHandle)) return;
@@ -803,6 +782,7 @@ static int32_t tdExecuteRSmaImplAsync(SSma *pSma, const void *pMsg, int32_t inpu
   return TSDB_CODE_SUCCESS;
 }
 
+#if 0
 static int32_t tdRsmaPrintSubmitReq(SSma *pSma, SSubmitReq *pReq) {
   SSubmitMsgIter msgIter = {0};
   SSubmitBlkIter blkIter = {0};
@@ -820,6 +800,7 @@ static int32_t tdRsmaPrintSubmitReq(SSma *pSma, SSubmitReq *pReq) {
   }
   return 0;
 }
+#endif
 
 /**
  * @brief sync mode
@@ -1189,65 +1170,6 @@ _err:
   return TSDB_CODE_FAILED;
 }
 
-static int32_t tdRSmaRestoreQTaskInfoReload(SSma *pSma, int8_t type, int64_t qTaskFileVer) {
-  SVnode *pVnode = pSma->pVnode;
-  STFile  tFile = {0};
-  char    qTaskInfoFName[TSDB_FILENAME_LEN] = {0};
-
-  tdRSmaQTaskInfoGetFileName(TD_VID(pVnode), qTaskFileVer, qTaskInfoFName);
-  if (tdInitTFile(&tFile, tfsGetPrimaryPath(pVnode->pTfs), qTaskInfoFName) < 0) {
-    goto _err;
-  }
-
-  if (!taosCheckExistFile(TD_TFILE_FULL_NAME(&tFile))) {
-    if (qTaskFileVer > 0) {
-      smaWarn("vgId:%d, restore rsma task %" PRIi8 " for version %" PRIi64 ", not start as %s not exist",
-              TD_VID(pVnode), type, qTaskFileVer, TD_TFILE_FULL_NAME(&tFile));
-    } else {
-      smaDebug("vgId:%d, restore rsma task %" PRIi8 " for version %" PRIi64 ", no need as %s not exist", TD_VID(pVnode),
-               type, qTaskFileVer, TD_TFILE_FULL_NAME(&tFile));
-    }
-    return TSDB_CODE_SUCCESS;
-  }
-
-  if (tdOpenTFile(&tFile, TD_FILE_READ) < 0) {
-    goto _err;
-  }
-
-  STFInfo tFileInfo = {0};
-  if (tdLoadTFileHeader(&tFile, &tFileInfo) < 0) {
-    goto _err;
-  }
-
-  SRSmaQTaskInfoIter fIter = {0};
-  if (tdRSmaQTaskInfoIterInit(&fIter, &tFile) < 0) {
-    tdRSmaQTaskInfoIterDestroy(&fIter);
-    tdCloseTFile(&tFile);
-    tdDestroyTFile(&tFile);
-    goto _err;
-  }
-
-  if (tdRSmaQTaskInfoRestore(pSma, type, &fIter) < 0) {
-    tdRSmaQTaskInfoIterDestroy(&fIter);
-    tdCloseTFile(&tFile);
-    tdDestroyTFile(&tFile);
-    goto _err;
-  }
-
-  tdRSmaQTaskInfoIterDestroy(&fIter);
-  tdCloseTFile(&tFile);
-  tdDestroyTFile(&tFile);
-
-  // restored successfully from committed or sync
-  smaInfo("vgId:%d, restore rsma task %" PRIi8 " for version %" PRIi64 ", qtaskinfo reload succeed", TD_VID(pVnode),
-          type, qTaskFileVer);
-  return TSDB_CODE_SUCCESS;
-_err:
-  smaError("vgId:%d, restore rsma task %" PRIi8 " for version %" PRIi64 ", qtaskinfo reload failed since %s",
-           TD_VID(pVnode), type, qTaskFileVer, terrstr());
-  return TSDB_CODE_FAILED;
-}
-
 /**
  * @brief reload ts data from checkpoint
  *
@@ -1270,19 +1192,12 @@ int32_t tdRSmaProcessRestoreImpl(SSma *pSma, int8_t type, int64_t qtaskFileVer)
     return TSDB_CODE_SUCCESS;
   }
 
-#if 0
-  // step 2: retrieve qtaskinfo items from the persistence file(rsma/qtaskinfo) and restore
-  if (tdRSmaRestoreQTaskInfoReload(pSma, type, qtaskFileVer) < 0) {
-    goto _err;
-  }
-#endif
-
-  // step 3: reload ts data from checkpoint
+  // step 2: reload ts data from checkpoint
   if (tdRSmaRestoreTSDataReload(pSma) < 0) {
     goto _err;
   }
 
-  // step 4: open SRSmaFS for qTaskFiles
+  // step 3: open SRSmaFS for qTaskFiles
   if (tdRSmaFSOpen(pSma, qtaskFileVer) < 0) {
     goto _err;
   }
@@ -1295,191 +1210,6 @@ _err:
   return TSDB_CODE_FAILED;
 }
 
-/**
- * @brief Restore from SRSmaQTaskInfoItem
- *
- * @param pSma
- * @param pItem
- * @return int32_t
- */
-static int32_t tdRSmaQTaskInfoItemRestore(SSma *pSma, const SRSmaQTaskInfoItem *pItem) {
-  SRSmaInfo *pRSmaInfo = NULL;
-  void      *qTaskInfo = NULL;
-
-  pRSmaInfo = tdAcquireRSmaInfoBySuid(pSma, pItem->suid);
-  if (!pRSmaInfo) {
-    smaDebug("vgId:%d, no restore as no rsma info for table:%" PRIu64, SMA_VID(pSma), pItem->suid);
-    return TSDB_CODE_SUCCESS;
-  }
-
-  if (pItem->type == TSDB_RETENTION_L1) {
-    qTaskInfo = RSMA_INFO_QTASK(pRSmaInfo, 0);
-  } else if (pItem->type == TSDB_RETENTION_L2) {
-    qTaskInfo = RSMA_INFO_QTASK(pRSmaInfo, 1);
-  } else {
-    ASSERT(0);
-  }
-
-  if (!qTaskInfo) {
-    tdReleaseRSmaInfo(pSma, pRSmaInfo);
-    smaDebug("vgId:%d, no restore as NULL rsma qTaskInfo for table:%" PRIu64, SMA_VID(pSma), pItem->suid);
-    return TSDB_CODE_SUCCESS;
-  }
-
-  if (qDeserializeTaskStatus(qTaskInfo, pItem->qTaskInfo, pItem->len) < 0) {
-    tdReleaseRSmaInfo(pSma, pRSmaInfo);
-    smaError("vgId:%d, restore rsma task failed for table:%" PRIi64 " level %d since %s", SMA_VID(pSma), pItem->suid,
-             pItem->type, terrstr());
-    return TSDB_CODE_FAILED;
-  }
-  smaDebug("vgId:%d, restore rsma task success for table:%" PRIi64 " level %d", SMA_VID(pSma), pItem->suid,
-           pItem->type);
-
-  tdReleaseRSmaInfo(pSma, pRSmaInfo);
-  return TSDB_CODE_SUCCESS;
-}
-
-static int32_t tdRSmaQTaskInfoIterInit(SRSmaQTaskInfoIter *pIter, STFile *pTFile) {
-  memset(pIter, 0, sizeof(*pIter));
-  pIter->pTFile = pTFile;
-  pIter->offset = TD_FILE_HEAD_SIZE;
-
-  if (tdGetTFileSize(pTFile, &pIter->fsize) < 0) {
-    return TSDB_CODE_FAILED;
-  }
-
-  if ((pIter->fsize - TD_FILE_HEAD_SIZE) < RSMA_QTASKINFO_BUFSIZE) {
-    pIter->nAlloc = pIter->fsize - TD_FILE_HEAD_SIZE;
-  } else {
-    pIter->nAlloc = RSMA_QTASKINFO_BUFSIZE;
-  }
-
-  if (pIter->nAlloc < TD_FILE_HEAD_SIZE) {
-    pIter->nAlloc = TD_FILE_HEAD_SIZE;
-  }
-
-  pIter->pBuf = taosMemoryMalloc(pIter->nAlloc);
-  if (!pIter->pBuf) {
-    terrno = TSDB_CODE_OUT_OF_MEMORY;
-    return TSDB_CODE_FAILED;
-  }
-  pIter->qBuf = pIter->pBuf;
-
-  return TSDB_CODE_SUCCESS;
-}
-
-static int32_t tdRSmaQTaskInfoIterNextBlock(SRSmaQTaskInfoIter *pIter, bool *isFinish) {
-  STFile *pTFile = pIter->pTFile;
-  int64_t nBytes = RSMA_QTASKINFO_BUFSIZE;
-
-  if (pIter->offset >= pIter->fsize) {
-    *isFinish = true;
-    return TSDB_CODE_SUCCESS;
-  }
-
-  if ((pIter->fsize - pIter->offset) < RSMA_QTASKINFO_BUFSIZE) {
-    nBytes = pIter->fsize - pIter->offset;
-  }
-
-  if (tdSeekTFile(pTFile, pIter->offset, SEEK_SET) < 0) {
-    return TSDB_CODE_FAILED;
-  }
-
-  if (tdReadTFile(pTFile, pIter->pBuf, nBytes) != nBytes) {
-    return TSDB_CODE_FAILED;
-  }
-
-  int32_t infoLen = 0;
-  taosDecodeFixedI32(pIter->pBuf, &infoLen);
-  if (infoLen > nBytes) {
-    if (infoLen <= RSMA_QTASKINFO_BUFSIZE) {
-      terrno = TSDB_CODE_RSMA_FILE_CORRUPTED;
-      smaError("iterate rsma qtaskinfo file %s failed since %s", TD_TFILE_FULL_NAME(pIter->pTFile), terrstr());
-      return TSDB_CODE_FAILED;
-    }
-    if (pIter->nAlloc < infoLen) {
-      pIter->nAlloc = infoLen;
-      void *pBuf = taosMemoryRealloc(pIter->pBuf, infoLen);
-      if (!pBuf) {
-        terrno = TSDB_CODE_OUT_OF_MEMORY;
-        return TSDB_CODE_FAILED;
-      }
-      pIter->pBuf = pBuf;
-    }
-
-    nBytes = infoLen;
-
-    if (tdSeekTFile(pTFile, pIter->offset, SEEK_SET) < 0) {
-      return TSDB_CODE_FAILED;
-    }
-
-    if (tdReadTFile(pTFile, pIter->pBuf, nBytes) != nBytes) {
-      return TSDB_CODE_FAILED;
-    }
-  }
-
-  pIter->qBuf = pIter->pBuf;
-  pIter->offset += nBytes;
-  pIter->nBytes = nBytes;
-  pIter->nBufPos = 0;
-
-  return TSDB_CODE_SUCCESS;
-}
-
-static int32_t tdRSmaQTaskInfoRestore(SSma *pSma, int8_t type, SRSmaQTaskInfoIter *pIter) {
-  while (1) {
-    // block iter
-    bool isFinish = false;
-    if (tdRSmaQTaskInfoIterNextBlock(pIter, &isFinish) < 0) {
-      return TSDB_CODE_FAILED;
-    }
-    if (isFinish) {
-      return TSDB_CODE_SUCCESS;
-    }
-
-    // consume the block
-    int32_t qTaskInfoLenWithHead = 0;
-    pIter->qBuf = taosDecodeFixedI32(pIter->qBuf, &qTaskInfoLenWithHead);
-    if (qTaskInfoLenWithHead < RSMA_QTASKINFO_HEAD_LEN) {
-      terrno = TSDB_CODE_TDB_FILE_CORRUPTED;
-      smaError("vgId:%d, restore rsma task %" PRIi8 " from qtaskinfo file %s failed since %s", SMA_VID(pSma), type,
-               TD_TFILE_FULL_NAME(pIter->pTFile), terrstr());
-      return TSDB_CODE_FAILED;
-    }
-
-    while (1) {
-      if ((pIter->nBufPos + qTaskInfoLenWithHead) <= pIter->nBytes) {
-        SRSmaQTaskInfoItem infoItem = {0};
-        pIter->qBuf = taosDecodeFixedI8(pIter->qBuf, &infoItem.type);
-        pIter->qBuf = taosDecodeFixedI64(pIter->qBuf, &infoItem.suid);
-        infoItem.qTaskInfo = pIter->qBuf;
-        infoItem.len = tdRSmaQTaskInfoContLen(qTaskInfoLenWithHead);
-        // do the restore job
-        smaDebug("vgId:%d, restore rsma task %" PRIi8 " from qtaskinfo file %s offset:%" PRIi64 "\n", SMA_VID(pSma),
-                 type, TD_TFILE_FULL_NAME(pIter->pTFile), pIter->offset - pIter->nBytes + pIter->nBufPos);
-        tdRSmaQTaskInfoItemRestore(pSma, &infoItem);
-
-        pIter->qBuf = POINTER_SHIFT(pIter->qBuf, infoItem.len);
-        pIter->nBufPos += qTaskInfoLenWithHead;
-
-        if ((pIter->nBufPos + RSMA_QTASKINFO_HEAD_LEN) >= pIter->nBytes) {
-          // prepare and load next block in the file
-          pIter->offset -= (pIter->nBytes - pIter->nBufPos);
-          break;
-        }
-
-        pIter->qBuf = taosDecodeFixedI32(pIter->qBuf, &qTaskInfoLenWithHead);
-        continue;
-      }
-      // prepare and load next block in the file
-      pIter->offset -= (pIter->nBytes - pIter->nBufPos);
-      break;
-    }
-  }
-
-  return TSDB_CODE_SUCCESS;
-}
-
 int32_t tdRSmaPersistExecImpl(SRSmaStat *pRSmaStat, SHashObj *pInfoHash) {
   SSma   *pSma = pRSmaStat->pSma;
   SVnode *pVnode = pSma->pVnode;
@@ -1523,148 +1253,6 @@ _err:
   return TSDB_CODE_FAILED;
 }
 
-#if 0
-int32_t tdRSmaPersistExecImpl(SRSmaStat *pRSmaStat, SHashObj *pInfoHash) {
-  SSma   *pSma = pRSmaStat->pSma;
-  SVnode *pVnode = pSma->pVnode;
-  int32_t vid = SMA_VID(pSma);
-  int64_t toffset = 0;
-  bool    isFileCreated = false;
-
-  if (taosHashGetSize(pInfoHash) <= 0) {
-    return TSDB_CODE_SUCCESS;
-  }
-
-  void *infoHash = taosHashIterate(pInfoHash, NULL);
-  if (!infoHash) {
-    return TSDB_CODE_SUCCESS;
-  }
-
-  int64_t fsMaxVer = tdRSmaFSMaxVer(pSma, pRSmaStat);
-  if (pRSmaStat->commitAppliedVer <= fsMaxVer) {
-    smaDebug("vgId:%d, rsma persist, no need as applied %" PRIi64 " not larger than fsMaxVer %" PRIi64, vid,
-             pRSmaStat->commitAppliedVer, fsMaxVer);
-    return TSDB_CODE_SUCCESS;
-  }
-
-  STFile tFile = {0};
-#if 0
-  if (pRSmaStat->commitAppliedVer > 0) {
-    char qTaskInfoFName[TSDB_FILENAME_LEN];
-    tdRSmaQTaskInfoGetFileName(vid, pRSmaStat->commitAppliedVer, qTaskInfoFName);
-    if (tdInitTFile(&tFile, tfsGetPrimaryPath(pVnode->pTfs), qTaskInfoFName) < 0) {
-      smaError("vgId:%d, rsma persist, init %s failed since %s", vid, qTaskInfoFName, terrstr());
-      goto _err;
-    }
-    if (tdCreateTFile(&tFile, true, TD_FTYPE_RSMA_QTASKINFO) < 0) {
-      smaError("vgId:%d, rsma persist, create %s failed since %s", vid, TD_TFILE_FULL_NAME(&tFile), terrstr());
-      goto _err;
-    }
-    smaDebug("vgId:%d, rsma, serialize qTaskInfo, file %s created", vid, TD_TFILE_FULL_NAME(&tFile));
-
-    isFileCreated = true;
-  }
-#endif
-
-  while (infoHash) {
-    SRSmaInfo *pRSmaInfo = *(SRSmaInfo **)infoHash;
-
-    if (RSMA_INFO_IS_DEL(pRSmaInfo)) {
-      infoHash = taosHashIterate(pInfoHash, infoHash);
-      continue;
-    }
-
-    for (int32_t i = 0; i < TSDB_RETENTION_L2; ++i) {
-#if 0
-      qTaskInfo_t taskInfo = RSMA_INFO_IQTASK(pRSmaInfo, i);
-#endif
-      qTaskInfo_t taskInfo = RSMA_INFO_QTASK(pRSmaInfo, i);
-      if (!taskInfo) {
-        smaDebug("vgId:%d, rsma, table %" PRIi64 " level %d qTaskInfo is NULL", vid, pRSmaInfo->suid, i + 1);
-        continue;
-      }
-
-      char   *pOutput = NULL;
-      int32_t len = 0;
-      int8_t  type = (int8_t)(i + 1);
-      if (qSerializeTaskStatus(taskInfo, &pOutput, &len) < 0) {
-        smaError("vgId:%d, rsma, table %" PRIi64 " level %d serialize qTaskInfo failed since %s", vid, pRSmaInfo->suid,
-                 i + 1, terrstr());
-        goto _err;
-      }
-      if (!pOutput || len <= 0) {
-        smaDebug("vgId:%d, rsma, table %" PRIi64
-                 " level %d serialize qTaskInfo success but no output(len %d), not persist",
-                 vid, pRSmaInfo->suid, i + 1, len);
-        taosMemoryFreeClear(pOutput);
-        continue;
-      }
-
-      smaDebug("vgId:%d, rsma, table %" PRIi64 " level %d serialize qTaskInfo success with len %d, need persist", vid,
-               pRSmaInfo->suid, i + 1, len);
-
-      if (!isFileCreated) {
-        char qTaskInfoFName[TSDB_FILENAME_LEN];
-        tdRSmaQTaskInfoGetFileName(vid, pRSmaStat->commitAppliedVer, qTaskInfoFName);
-        if (tdInitTFile(&tFile, tfsGetPrimaryPath(pVnode->pTfs), qTaskInfoFName) < 0) {
-          smaError("vgId:%d, rsma persist, init %s failed since %s", vid, qTaskInfoFName, terrstr());
-          goto _err;
-        }
-        if (tdCreateTFile(&tFile, true, TD_FTYPE_RSMA_QTASKINFO) < 0) {
-          smaError("vgId:%d, rsma persist, create %s failed since %s", vid, TD_TFILE_FULL_NAME(&tFile), terrstr());
-          goto _err;
-        }
-        smaDebug("vgId:%d, rsma, table %" PRIi64 " serialize qTaskInfo, file %s created", vid, pRSmaInfo->suid,
-                 TD_TFILE_FULL_NAME(&tFile));
-
-        isFileCreated = true;
-      }
-
-      char    tmpBuf[RSMA_QTASKINFO_HEAD_LEN] = {0};
-      void   *pTmpBuf = &tmpBuf;
-      int32_t headLen = 0;
-      headLen += taosEncodeFixedI32(&pTmpBuf, len + RSMA_QTASKINFO_HEAD_LEN);
-      headLen += taosEncodeFixedI8(&pTmpBuf, type);
-      headLen += taosEncodeFixedI64(&pTmpBuf, pRSmaInfo->suid);
-
-      ASSERT(headLen <= RSMA_QTASKINFO_HEAD_LEN);
-      tdAppendTFile(&tFile, (void *)&tmpBuf, headLen, &toffset);
-      smaDebug("vgId:%d, rsma, table %" PRIi64 " level %d head part(len:%d) appended to offset:%" PRIi64, vid,
-               pRSmaInfo->suid, i + 1, headLen, toffset);
-      tdAppendTFile(&tFile, pOutput, len, &toffset);
-      smaDebug("vgId:%d, rsma, table %" PRIi64 " level %d body part len:%d appended to offset:%" PRIi64, vid,
-               pRSmaInfo->suid, i + 1, len, toffset);
-
-      taosMemoryFree(pOutput);
-    }
-
-    infoHash = taosHashIterate(pInfoHash, infoHash);
-  }
-
-  if (isFileCreated) {
-    if (tdUpdateTFileHeader(&tFile) < 0) {
-      smaError("vgId:%d, rsma, failed to update tfile %s header since %s", vid, TD_TFILE_FULL_NAME(&tFile),
-               tstrerror(terrno));
-      goto _err;
-    } else {
-      smaDebug("vgId:%d, rsma, succeed to update tfile %s header", vid, TD_TFILE_FULL_NAME(&tFile));
-    }
-
-    tdCloseTFile(&tFile);
-    tdDestroyTFile(&tFile);
-  }
-  return TSDB_CODE_SUCCESS;
-_err:
-  smaError("vgId:%d, rsma persist failed since %s", vid, terrstr());
-  if (isFileCreated) {
-    tdRemoveTFile(&tFile);
-    tdDestroyTFile(&tFile);
-  }
-  return TSDB_CODE_FAILED;
-}
-
-#endif
-
 /**
  * @brief trigger to get rsma result in async mode
  *
diff --git a/source/dnode/vnode/src/sma/smaUtil.c b/source/dnode/vnode/src/sma/smaUtil.c
index 6d7b7df1ee..4d09d690d6 100644
--- a/source/dnode/vnode/src/sma/smaUtil.c
+++ b/source/dnode/vnode/src/sma/smaUtil.c
@@ -16,7 +16,7 @@
 #include "sma.h"
 
 // smaFileUtil ================
-
+#if 0
 #define TD_FILE_STATE_OK  0
 #define TD_FILE_STATE_BAD 1
 
@@ -182,6 +182,8 @@ void tdCloseTFile(STFile *pTFile) {
 
 void tdDestroyTFile(STFile *pTFile) { taosMemoryFreeClear(TD_TFILE_FULL_NAME(pTFile)); }
 
+#endif
+
 void tdGetVndFileName(int32_t vgId, const char *pdname, const char *dname, const char *fname, int64_t version,
                       char *outputName) {
   if (version < 0) {
@@ -221,6 +223,7 @@ void tdGetVndDirName(int32_t vgId, const char *pdname, const char *dname, bool e
   }
 }
 
+#if 0
 int32_t tdInitTFile(STFile *pTFile, const char *dname, const char *fname) {
   TD_TFILE_SET_STATE(pTFile, TD_FILE_STATE_OK);
   TD_TFILE_SET_CLOSED(pTFile);
@@ -286,6 +289,8 @@ int32_t tdRemoveTFile(STFile *pTFile) {
   return 0;
 }
 
+#endif
+
 // smaXXXUtil ================
 void *tdAcquireSmaRef(int32_t rsetId, int64_t refId) {
   void *pResult = taosAcquireRef(rsetId, refId);

From 4c6623d3ddbea2ed6a2e5e776a3a49861ef6e038 Mon Sep 17 00:00:00 2001
From: Cary Xu <klxu@taosdata.com>
Date: Tue, 18 Oct 2022 15:47:36 +0800
Subject: [PATCH 26/42] chore: remove obsolete file

---
 source/dnode/vnode/src/sma/smaFS2.c | 720 ----------------------------
 1 file changed, 720 deletions(-)
 delete mode 100644 source/dnode/vnode/src/sma/smaFS2.c

diff --git a/source/dnode/vnode/src/sma/smaFS2.c b/source/dnode/vnode/src/sma/smaFS2.c
deleted file mode 100644
index 9dced9c23b..0000000000
--- a/source/dnode/vnode/src/sma/smaFS2.c
+++ /dev/null
@@ -1,720 +0,0 @@
-/*
- * Copyright (c) 2019 TAOS Data, Inc. <jhtao@taosdata.com>
- *
- * This program is free software: you can use, redistribute, and/or modify
- * it under the terms of the GNU Affero General Public License, version 3
- * or later ("AGPL"), as published by the Free Software Foundation.
- *
- * This program is distributed in the hope that it will be useful, but WITHOUT
- * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
- * FITNESS FOR A PARTICULAR PURPOSE.
- *
- * You should have received a copy of the GNU Affero General Public License
- * along with this program. If not, see <http://www.gnu.org/licenses/>.
- */
-
-#include "sma.h"
-
-// =================================================================================================
-static int32_t tdRSmaEncodeFS(uint8_t *p, SRSmaFS *pFS);
-static int32_t tPutQTaskFile(uint8_t *p, SQTaskFile *pFile);
-static int32_t tPutQTaskFileSet(uint8_t *p, SQTaskFileSet *pSet);
-
-static int32_t tdFetchQTaskInfoFiles(SSma *pSma, int64_t version, SArray **output);
-static int32_t tdQTaskInfCmprFn1(const void *p1, const void *p2);
-static int32_t tdQTaskInfCmprFn2(const void *p1, const void *p2);
-
-// =================================================================================================
-static int32_t tdRSmaEncodeFS(uint8_t *p, SRSmaFS *pFS) {
-  int32_t  n = 0;
-  uint32_t nSet = taosArrayGetSize(pFS->aQTaskFSet);
-
-  // version
-  n += tPutU8(p ? p + n : p, 0);
-
-  // SArray<SQTaskFileSet>
-  n += tPutU32v(p ? p + n : p, nSet);
-  for (uint32_t i = 0; i < nSet; ++i) {
-    n += tPutQTaskFileSet(p ? p + n : p, (SQTaskFileSet *)taosArrayGet(pFS->aQTaskFSet, i));
-  }
-
-  return n;
-}
-
-static int32_t tPutQTaskFileSet(uint8_t *p, SQTaskFileSet *pSet) {
-  int32_t n = 0;
-  int32_t nQTaskF = taosArrayGetSize(pSet->aQTaskF);
-
-  n += tPutI64v(p ? p + n : p, pSet->suid);
-  n += tPutI32v(p ? p + n : p, nQTaskF);
-
-  // SQTaskFile
-  for (int32_t i = 0; i < nQTaskF; ++i) {
-    n += tPutQTaskFile(p ? p + n : p, taosArrayGet(pSet->aQTaskF, i));
-  }
-
-  return n;
-}
-
-static int32_t tPutQTaskFile(uint8_t *p, SQTaskFile *pFile) {
-  int32_t n = 0;
-
-  n += tPutI8(p ? p + n : p, pFile->level);
-  n += tPutI64v(p ? p + n : p, pFile->version);
-  n += tPutI64v(p ? p + n : p, pFile->mtime);
-
-  return n;
-}
-
-static int32_t tGetQTaskFileSet(uint8_t *p, SQTaskFileSet *pSet) {
-  int32_t n = 0;
-  int32_t nQTaskF = 0;
-
-  n += tGetI64v(p + n, &pSet->suid);
-  n += tGetI32v(p + n, &nQTaskF);
-
-  ASSERT(nQTaskF >= 0);
-
-  pSet->aQTaskF = taosArrayInit(nQTaskF, sizeof(SQTaskFile));
-  if (!pSet->aQTaskF) {
-    terrno = TSDB_CODE_OUT_OF_MEMORY;
-    return -1;
-  }
-
-  for (int32_t i = 0; i < nQTaskF; ++i) {
-    SQTaskFile *pFile = taosArrayGet(pSet->aQTaskF, i);
-    pFile->nRef = 1;
-    n += tGetQTaskFile(p + n, pFile);
-  }
-
-  return n;
-}
-
-static int32_t tGetQTaskFile(uint8_t *p, SQTaskFile *pFile) {
-  int32_t n = 0;
-
-  n += tGetI8(p + n, &pFile->level);
-  n += tGetI64v(p + n, &pFile->version);
-  n += tGetI64v(p + n, &pFile->mtime);
-
-  return n;
-}
-
-static int32_t tdRSmaGnrtFS(SSma *pSma, SRSmaFS *pFS, char *fname) {
-  int32_t   code = 0;
-  int64_t   n;
-  int64_t   size;
-  uint8_t  *pData = NULL;
-  TdFilePtr pFD = NULL;
-
-  // to binary
-  size = tdRSmaEncodeFS(NULL, pFS) + sizeof(TSCKSUM);
-  pData = taosMemoryMalloc(size);
-  if (pData == NULL) {
-    code = TSDB_CODE_OUT_OF_MEMORY;
-    goto _err;
-  }
-  n = tdRSmaEncodeFS(pData, pFS);
-  ASSERT(n + sizeof(TSCKSUM) == size);
-  taosCalcChecksumAppend(0, pData, size);
-
-  // create and write
-  pFD = taosOpenFile(fname, TD_FILE_WRITE | TD_FILE_CREATE | TD_FILE_TRUNC);
-  if (pFD == NULL) {
-    code = TAOS_SYSTEM_ERROR(errno);
-    goto _err;
-  }
-
-  n = taosWriteFile(pFD, pData, size);
-  if (n < 0) {
-    code = TAOS_SYSTEM_ERROR(errno);
-    goto _err;
-  }
-
-  if (taosFsyncFile(pFD) < 0) {
-    code = TAOS_SYSTEM_ERROR(errno);
-    goto _err;
-  }
-
-  taosCloseFile(&pFD);
-
-  if (pData) taosMemoryFree(pData);
-  return code;
-
-_err:
-  smaError("vgId:%d, rsma gnrt fs failed since %s", SMA_VID(pSma), tstrerror(code));
-  if (pData) taosMemoryFree(pData);
-  return code;
-}
-
-int32_t tdRSmaFSCommit1(SSma *pSma, SRSmaFS *pFSNew) {
-  int32_t code = 0;
-  char    tfname[TSDB_FILENAME_LEN];
-  char    fname[TSDB_FILENAME_LEN];
-
-  snprintf(tfname, TSDB_FILENAME_LEN - 1, "%s%s%s%sRSMAFS.t", tfsGetPrimaryPath(pSma->pVnode->pTfs), TD_DIRSEP,
-           VNODE_RSMA_DIR, TD_DIRSEP);
-  snprintf(fname, TSDB_FILENAME_LEN - 1, "%s%s%s%sRSMAFS", tfsGetPrimaryPath(pSma->pVnode->pTfs), TD_DIRSEP,
-           VNODE_RSMA_DIR, TD_DIRSEP);
-
-  // gnrt CURRENT.t
-  code = tdRSmaGnrtFS(pSma, pFSNew, tfname);
-  if (code) goto _err;
-
-  // rename
-  code = taosRenameFile(tfname, fname);
-  if (code) {
-    code = TAOS_SYSTEM_ERROR(code);
-    goto _err;
-  }
-
-  return code;
-
-_err:
-  smaError("vgId:%d, rsma fs commit phase 1 failed since %s", SMA_VID(pSma), tstrerror(code));
-  return code;
-}
-
-int32_t tdRSmaFSMergeQTaskFSet(SQTaskFileSet *pSetOld, SQTaskFileSet *pSetNew) {
-  SQTaskFileSet *
-
-}
-
-int32_t tdRSmaFSCommit2(SSma *pSma, SRSmaFS *pFSNew) {
-  SSmaEnv   *pEnv = SMA_RSMA_ENV(pSma);
-  SRSmaStat *pStat = NULL;
-  SRSmaInfo *pRSmaInfo = NULL;
-  SRSmaFS   *pRSmaFS = NULL;
-  int32_t    code = 0;
-  int32_t    nRef;
-  char       fname[TSDB_FILENAME_LEN];
-
-  if (!pEnv) {
-    terrno = TSDB_CODE_RSMA_INVALID_ENV;
-    return TSDB_CODE_FAILED;
-  }
-
-  pStat = (SRSmaStat *)SMA_ENV_STAT(pEnv);
-  if (!pStat || !(pRSmaFS = RSMA_FS(pStat))) {
-    terrno = TSDB_CODE_RSMA_INVALID_STAT;
-    return TSDB_CODE_FAILED;
-  }
-
-  // data
-  int32_t iOld = 0;
-  int32_t iNew = 0;
-  while (true) {
-    int32_t       nOld = taosArrayGetSize(pRSmaFS->aQTaskFSet);
-    int32_t       nNew = taosArrayGetSize(pFSNew->aQTaskFSet);
-    SQTaskFileSet fSet;
-
-    if (iOld >= nOld && iNew >= nNew) break;
-
-    SQTaskFileSet *pSetOld = (iOld < nOld) ? taosArrayGet(pRSmaFS->aQTaskFSet, iOld) : NULL;
-    SQTaskFileSet *pSetNew = (iNew < nNew) ? taosArrayGet(pFSNew->aQTaskFSet, iNew) : NULL;
-
-    if (pSetOld && pSetNew) {
-      if (pSetOld->suid == pSetNew->suid) {
-        goto _merge_old_and_new;
-      } else if (pSetOld->suid < pSetNew->suid) {
-        goto _remove_old;
-      } else {
-        goto _add_new;
-      }
-    } else if (pSetOld) {
-      goto _remove_old;
-    } else {
-      goto _add_new;
-    }
-
-  _merge_old_and_new:
-
-
-
-    // head
-    fSet.pHeadF = pSetOld->pHeadF;
-    if ((!sameDisk) || (pSetOld->pHeadF->commitID != pSetNew->pHeadF->commitID)) {
-      pSetOld->pHeadF = (SHeadFile *)taosMemoryMalloc(sizeof(SHeadFile));
-      if (pSetOld->pHeadF == NULL) {
-        code = TSDB_CODE_OUT_OF_MEMORY;
-        goto _err;
-      }
-      *pSetOld->pHeadF = *pSetNew->pHeadF;
-      pSetOld->pHeadF->nRef = 1;
-
-      nRef = atomic_sub_fetch_32(&fSet.pHeadF->nRef, 1);
-      if (nRef == 0) {
-        tsdbHeadFileName(pTsdb, pSetOld->diskId, pSetOld->fid, fSet.pHeadF, fname);
-        taosRemoveFile(fname);
-        taosMemoryFree(fSet.pHeadF);
-      }
-    } else {
-      ASSERT(fSet.pHeadF->size == pSetNew->pHeadF->size);
-      ASSERT(fSet.pHeadF->offset == pSetNew->pHeadF->offset);
-    }
-
-    // data
-    fSet.pDataF = pSetOld->pDataF;
-    if ((!sameDisk) || (pSetOld->pDataF->commitID != pSetNew->pDataF->commitID)) {
-      pSetOld->pDataF = (SDataFile *)taosMemoryMalloc(sizeof(SDataFile));
-      if (pSetOld->pDataF == NULL) {
-        code = TSDB_CODE_OUT_OF_MEMORY;
-        goto _err;
-      }
-      *pSetOld->pDataF = *pSetNew->pDataF;
-      pSetOld->pDataF->nRef = 1;
-
-      nRef = atomic_sub_fetch_32(&fSet.pDataF->nRef, 1);
-      if (nRef == 0) {
-        tsdbDataFileName(pTsdb, pSetOld->diskId, pSetOld->fid, fSet.pDataF, fname);
-        taosRemoveFile(fname);
-        taosMemoryFree(fSet.pDataF);
-      }
-    } else {
-      ASSERT(pSetOld->pDataF->size <= pSetNew->pDataF->size);
-      pSetOld->pDataF->size = pSetNew->pDataF->size;
-    }
-
-    // sma
-    fSet.pSmaF = pSetOld->pSmaF;
-    if ((!sameDisk) || (pSetOld->pSmaF->commitID != pSetNew->pSmaF->commitID)) {
-      pSetOld->pSmaF = (SSmaFile *)taosMemoryMalloc(sizeof(SSmaFile));
-      if (pSetOld->pSmaF == NULL) {
-        code = TSDB_CODE_OUT_OF_MEMORY;
-        goto _err;
-      }
-      *pSetOld->pSmaF = *pSetNew->pSmaF;
-      pSetOld->pSmaF->nRef = 1;
-
-      nRef = atomic_sub_fetch_32(&fSet.pSmaF->nRef, 1);
-      if (nRef == 0) {
-        tsdbSmaFileName(pTsdb, pSetOld->diskId, pSetOld->fid, fSet.pSmaF, fname);
-        taosRemoveFile(fname);
-        taosMemoryFree(fSet.pSmaF);
-      }
-    } else {
-      ASSERT(pSetOld->pSmaF->size <= pSetNew->pSmaF->size);
-      pSetOld->pSmaF->size = pSetNew->pSmaF->size;
-    }
-
-    // stt
-    if (sameDisk) {
-      if (pSetNew->nSttF > pSetOld->nSttF) {
-        ASSERT(pSetNew->nSttF = pSetOld->nSttF + 1);
-        pSetOld->aSttF[pSetOld->nSttF] = (SSttFile *)taosMemoryMalloc(sizeof(SSttFile));
-        if (pSetOld->aSttF[pSetOld->nSttF] == NULL) {
-          code = TSDB_CODE_OUT_OF_MEMORY;
-          goto _err;
-        }
-        *pSetOld->aSttF[pSetOld->nSttF] = *pSetNew->aSttF[pSetOld->nSttF];
-        pSetOld->aSttF[pSetOld->nSttF]->nRef = 1;
-        pSetOld->nSttF++;
-      } else if (pSetNew->nSttF < pSetOld->nSttF) {
-        ASSERT(pSetNew->nSttF == 1);
-        for (int32_t iStt = 0; iStt < pSetOld->nSttF; iStt++) {
-          SSttFile *pSttFile = pSetOld->aSttF[iStt];
-          nRef = atomic_sub_fetch_32(&pSttFile->nRef, 1);
-          if (nRef == 0) {
-            tsdbSttFileName(pTsdb, pSetOld->diskId, pSetOld->fid, pSttFile, fname);
-            taosRemoveFile(fname);
-            taosMemoryFree(pSttFile);
-          }
-          pSetOld->aSttF[iStt] = NULL;
-        }
-
-        pSetOld->nSttF = 1;
-        pSetOld->aSttF[0] = (SSttFile *)taosMemoryMalloc(sizeof(SSttFile));
-        if (pSetOld->aSttF[0] == NULL) {
-          code = TSDB_CODE_OUT_OF_MEMORY;
-          goto _err;
-        }
-        *pSetOld->aSttF[0] = *pSetNew->aSttF[0];
-        pSetOld->aSttF[0]->nRef = 1;
-      } else {
-        for (int32_t iStt = 0; iStt < pSetOld->nSttF; iStt++) {
-          if (pSetOld->aSttF[iStt]->commitID != pSetNew->aSttF[iStt]->commitID) {
-            SSttFile *pSttFile = pSetOld->aSttF[iStt];
-            nRef = atomic_sub_fetch_32(&pSttFile->nRef, 1);
-            if (nRef == 0) {
-              tsdbSttFileName(pTsdb, pSetOld->diskId, pSetOld->fid, pSttFile, fname);
-              taosRemoveFile(fname);
-              taosMemoryFree(pSttFile);
-            }
-
-            pSetOld->aSttF[iStt] = (SSttFile *)taosMemoryMalloc(sizeof(SSttFile));
-            if (pSetOld->aSttF[iStt] == NULL) {
-              code = TSDB_CODE_OUT_OF_MEMORY;
-              goto _err;
-            }
-            *pSetOld->aSttF[iStt] = *pSetNew->aSttF[iStt];
-            pSetOld->aSttF[iStt]->nRef = 1;
-          } else {
-            ASSERT(pSetOld->aSttF[iStt]->size == pSetOld->aSttF[iStt]->size);
-            ASSERT(pSetOld->aSttF[iStt]->offset == pSetOld->aSttF[iStt]->offset);
-          }
-        }
-      }
-    } else {
-      ASSERT(pSetOld->nSttF == pSetNew->nSttF);
-      for (int32_t iStt = 0; iStt < pSetOld->nSttF; iStt++) {
-        SSttFile *pSttFile = pSetOld->aSttF[iStt];
-        nRef = atomic_sub_fetch_32(&pSttFile->nRef, 1);
-        if (nRef == 0) {
-          tsdbSttFileName(pTsdb, pSetOld->diskId, pSetOld->fid, pSttFile, fname);
-          taosRemoveFile(fname);
-          taosMemoryFree(pSttFile);
-        }
-
-        pSetOld->aSttF[iStt] = (SSttFile *)taosMemoryMalloc(sizeof(SSttFile));
-        if (pSetOld->aSttF[iStt] == NULL) {
-          code = TSDB_CODE_OUT_OF_MEMORY;
-          goto _err;
-        }
-        *pSetOld->aSttF[iStt] = *pSetNew->aSttF[iStt];
-        pSetOld->aSttF[iStt]->nRef = 1;
-      }
-    }
-
-    if (!sameDisk) {
-      pSetOld->diskId = pSetNew->diskId;
-    }
-
-    iOld++;
-    iNew++;
-    continue;
-
-  _remove_old:
-    nRef = atomic_sub_fetch_32(&pSetOld->pHeadF->nRef, 1);
-    if (nRef == 0) {
-      tsdbHeadFileName(pTsdb, pSetOld->diskId, pSetOld->fid, pSetOld->pHeadF, fname);
-      taosRemoveFile(fname);
-      taosMemoryFree(pSetOld->pHeadF);
-    }
-
-    nRef = atomic_sub_fetch_32(&pSetOld->pDataF->nRef, 1);
-    if (nRef == 0) {
-      tsdbDataFileName(pTsdb, pSetOld->diskId, pSetOld->fid, pSetOld->pDataF, fname);
-      taosRemoveFile(fname);
-      taosMemoryFree(pSetOld->pDataF);
-    }
-
-    nRef = atomic_sub_fetch_32(&pSetOld->pSmaF->nRef, 1);
-    if (nRef == 0) {
-      tsdbSmaFileName(pTsdb, pSetOld->diskId, pSetOld->fid, pSetOld->pSmaF, fname);
-      taosRemoveFile(fname);
-      taosMemoryFree(pSetOld->pSmaF);
-    }
-
-    for (int8_t iStt = 0; iStt < pSetOld->nSttF; iStt++) {
-      nRef = atomic_sub_fetch_32(&pSetOld->aSttF[iStt]->nRef, 1);
-      if (nRef == 0) {
-        tsdbSttFileName(pTsdb, pSetOld->diskId, pSetOld->fid, pSetOld->aSttF[iStt], fname);
-        taosRemoveFile(fname);
-        taosMemoryFree(pSetOld->aSttF[iStt]);
-      }
-    }
-
-    taosArrayRemove(pTsdb->fs.aDFileSet, iOld);
-    continue;
-
-  _add_new:
-    fSet = (SDFileSet){.diskId = pSetNew->diskId, .fid = pSetNew->fid, .nSttF = 1};
-
-    // head
-    fSet.pHeadF = (SHeadFile *)taosMemoryMalloc(sizeof(SHeadFile));
-    if (fSet.pHeadF == NULL) {
-      code = TSDB_CODE_OUT_OF_MEMORY;
-      goto _err;
-    }
-    *fSet.pHeadF = *pSetNew->pHeadF;
-    fSet.pHeadF->nRef = 1;
-
-    // data
-    fSet.pDataF = (SDataFile *)taosMemoryMalloc(sizeof(SDataFile));
-    if (fSet.pDataF == NULL) {
-      code = TSDB_CODE_OUT_OF_MEMORY;
-      goto _err;
-    }
-    *fSet.pDataF = *pSetNew->pDataF;
-    fSet.pDataF->nRef = 1;
-
-    // sma
-    fSet.pSmaF = (SSmaFile *)taosMemoryMalloc(sizeof(SSmaFile));
-    if (fSet.pSmaF == NULL) {
-      code = TSDB_CODE_OUT_OF_MEMORY;
-      goto _err;
-    }
-    *fSet.pSmaF = *pSetNew->pSmaF;
-    fSet.pSmaF->nRef = 1;
-
-    // stt
-    ASSERT(pSetNew->nSttF == 1);
-    fSet.aSttF[0] = (SSttFile *)taosMemoryMalloc(sizeof(SSttFile));
-    if (fSet.aSttF[0] == NULL) {
-      code = TSDB_CODE_OUT_OF_MEMORY;
-      goto _err;
-    }
-    *fSet.aSttF[0] = *pSetNew->aSttF[0];
-    fSet.aSttF[0]->nRef = 1;
-
-    if (taosArrayInsert(pTsdb->fs.aDFileSet, iOld, &fSet) == NULL) {
-      code = TSDB_CODE_OUT_OF_MEMORY;
-      goto _err;
-    }
-    iOld++;
-    iNew++;
-    continue;
-  }
-
-  return code;
-
-_err:
-  tsdbError("vgId:%d, tsdb fs commit phase 2 failed since %s", TD_VID(pTsdb->pVnode), tstrerror(code));
-  return code;
-}
-
-/**
- * @brief Open RSma FS from qTaskInfo files
- *
- * @param pSma
- * @param version
- * @return int32_t
- */
-int32_t tdRSmaFSOpen(SSma *pSma, int64_t version) {
-  SVnode    *pVnode = pSma->pVnode;
-  int64_t    commitID = pVnode->state.commitID;
-  SSmaEnv   *pEnv = SMA_RSMA_ENV(pSma);
-  SRSmaStat *pStat = NULL;
-  SArray    *output = NULL;
-
-  terrno = TSDB_CODE_SUCCESS;
-
-  if (!pEnv) {
-    return TSDB_CODE_SUCCESS;
-  }
-
-  if (tdFetchQTaskInfoFiles(pSma, version, &output) < 0) {
-    goto _end;
-  }
-
-  pStat = (SRSmaStat *)SMA_ENV_STAT(pEnv);
-
-  for (int32_t i = 0; i < taosArrayGetSize(output); ++i) {
-    int32_t vid = 0;
-    int64_t version = -1;
-    sscanf((const char *)taosArrayGetP(output, i), "v%dqinf.v%" PRIi64, &vid, &version);
-    SQTaskFile qTaskFile = {.version = version, .nRef = 1};
-    if ((terrno = tdRSmaFSUpsertQTaskFile(RSMA_FS(pStat), &qTaskFile)) < 0) {
-      goto _end;
-    }
-    smaInfo("vgId:%d, open fs, version:%" PRIi64 ", ref:%d", TD_VID(pVnode), qTaskFile.version, qTaskFile.nRef);
-  }
-
-_end:
-  for (int32_t i = 0; i < taosArrayGetSize(output); ++i) {
-    void *ptr = taosArrayGetP(output, i);
-    taosMemoryFreeClear(ptr);
-  }
-  taosArrayDestroy(output);
-
-  if (terrno != TSDB_CODE_SUCCESS) {
-    smaError("vgId:%d, open rsma fs failed since %s", TD_VID(pVnode), terrstr());
-    return TSDB_CODE_FAILED;
-  }
-  return TSDB_CODE_SUCCESS;
-}
-
-void tdRSmaFSClose(SRSmaFS *fs) { taosArrayDestroy(fs->aQTaskFSet); }
-
-static int32_t tdQTaskInfCmprFn1(const void *p1, const void *p2) {
-  if (*(int64_t *)p1 < ((SQTaskFile *)p2)->version) {
-    return -1;
-  } else if (*(int64_t *)p1 > ((SQTaskFile *)p2)->version) {
-    return 1;
-  }
-  return 0;
-}
-
-int32_t tdRSmaFSRef(SSma *pSma, SRSmaStat *pStat, int64_t version) {
-  SArray     *aQTaskFSet = RSMA_FS(pStat)->aQTaskFSet;
-  SQTaskFile *pTaskF = NULL;
-  int32_t     oldVal = 0;
-
-  taosRLockLatch(RSMA_FS_LOCK(pStat));
-  if ((pTaskF = taosArraySearch(aQTaskFSet, &version, tdQTaskInfCmprFn1, TD_EQ))) {
-    oldVal = atomic_fetch_add_32(&pTaskF->nRef, 1);
-    ASSERT(oldVal > 0);
-  }
-  taosRUnLockLatch(RSMA_FS_LOCK(pStat));
-  return oldVal;
-}
-
-int64_t tdRSmaFSMaxVer(SSma *pSma, SRSmaStat *pStat) {
-  SArray *aQTaskFSet = RSMA_FS(pStat)->aQTaskFSet;
-  int64_t version = -1;
-
-  taosRLockLatch(RSMA_FS_LOCK(pStat));
-  if (taosArrayGetSize(aQTaskFSet) > 0) {
-    version = ((SQTaskFile *)taosArrayGetLast(aQTaskFSet))->version;
-  }
-  taosRUnLockLatch(RSMA_FS_LOCK(pStat));
-  return version;
-}
-
-void tdRSmaFSUnRef(SSma *pSma, SRSmaStat *pStat, int64_t version) {
-  SVnode     *pVnode = pSma->pVnode;
-  SArray     *aQTaskFSet = RSMA_FS(pStat)->aQTaskFSet;
-  char        qTaskFullName[TSDB_FILENAME_LEN];
-  SQTaskFile *pTaskF = NULL;
-  int32_t     idx = -1;
-
-  taosWLockLatch(RSMA_FS_LOCK(pStat));
-  if ((idx = taosArraySearchIdx(aQTaskFSet, &version, tdQTaskInfCmprFn1, TD_EQ)) >= 0) {
-    ASSERT(idx < taosArrayGetSize(aQTaskFSet));
-    pTaskF = taosArrayGet(aQTaskFSet, idx);
-    if (atomic_sub_fetch_32(&pTaskF->nRef, 1) <= 0) {
-      tdRSmaQTaskInfoGetFullName(TD_VID(pVnode), pTaskF->version, tfsGetPrimaryPath(pVnode->pTfs), qTaskFullName);
-      if (taosRemoveFile(qTaskFullName) < 0) {
-        smaWarn("vgId:%d, failed to remove %s since %s", TD_VID(pVnode), qTaskFullName,
-                tstrerror(TAOS_SYSTEM_ERROR(errno)));
-      } else {
-        smaDebug("vgId:%d, success to remove %s", TD_VID(pVnode), qTaskFullName);
-      }
-      taosArrayRemove(aQTaskFSet, idx);
-    }
-  }
-  taosWUnLockLatch(RSMA_FS_LOCK(pStat));
-}
-
-/**
- * @brief Fetch qtaskfiles LE than version
- *
- * @param pSma
- * @param version
- * @param output
- * @return int32_t
- */
-static int32_t tdFetchQTaskInfoFiles(SSma *pSma, int64_t version, SArray **output) {
-  SVnode       *pVnode = pSma->pVnode;
-  TdDirPtr      pDir = NULL;
-  TdDirEntryPtr pDirEntry = NULL;
-  char          dir[TSDB_FILENAME_LEN];
-  const char   *pattern = "v[0-9]+qinf\\.v([0-9]+)?$";
-  regex_t       regex;
-  int           code = 0;
-
-  terrno = TSDB_CODE_SUCCESS;
-
-  tdGetVndDirName(TD_VID(pVnode), tfsGetPrimaryPath(pVnode->pTfs), VNODE_RSMA_DIR, true, dir);
-
-  if (!taosCheckExistFile(dir)) {
-    smaDebug("vgId:%d, fetch qtask files, no need as dir %s not exist", TD_VID(pVnode), dir);
-    return TSDB_CODE_SUCCESS;
-  }
-
-  // Resource allocation and init
-  if ((code = regcomp(&regex, pattern, REG_EXTENDED)) != 0) {
-    terrno = TSDB_CODE_RSMA_REGEX_MATCH;
-    char errbuf[128];
-    regerror(code, &regex, errbuf, sizeof(errbuf));
-    smaWarn("vgId:%d, fetch qtask files, regcomp for %s failed since %s", TD_VID(pVnode), dir, errbuf);
-    return TSDB_CODE_FAILED;
-  }
-
-  if (!(pDir = taosOpenDir(dir))) {
-    regfree(&regex);
-    terrno = TAOS_SYSTEM_ERROR(errno);
-    smaError("vgId:%d, fetch qtask files, open dir %s failed since %s", TD_VID(pVnode), dir, terrstr());
-    return TSDB_CODE_FAILED;
-  }
-
-  int32_t    dirLen = strlen(dir);
-  char      *dirEnd = POINTER_SHIFT(dir, dirLen);
-  regmatch_t regMatch[2];
-  while ((pDirEntry = taosReadDir(pDir))) {
-    char *entryName = taosGetDirEntryName(pDirEntry);
-    if (!entryName) {
-      continue;
-    }
-
-    code = regexec(&regex, entryName, 2, regMatch, 0);
-
-    if (code == 0) {
-      // match
-      smaInfo("vgId:%d, fetch qtask files, max ver:%" PRIi64 ", %s found", TD_VID(pVnode), version, entryName);
-
-      int64_t ver = -1;
-      sscanf((const char *)POINTER_SHIFT(entryName, regMatch[1].rm_so), "%" PRIi64, &ver);
-      if ((ver <= version) && (ver > -1)) {
-        if (!(*output)) {
-          if (!(*output = taosArrayInit(1, POINTER_BYTES))) {
-            terrno = TSDB_CODE_OUT_OF_MEMORY;
-            goto _end;
-          }
-        }
-        char *entryDup = strdup(entryName);
-        if (!entryDup) {
-          terrno = TSDB_CODE_OUT_OF_MEMORY;
-          goto _end;
-        }
-        if (!taosArrayPush(*output, &entryDup)) {
-          terrno = TSDB_CODE_OUT_OF_MEMORY;
-          goto _end;
-        }
-      } else {
-      }
-    } else if (code == REG_NOMATCH) {
-      // not match
-      smaTrace("vgId:%d, fetch qtask files, not match %s", TD_VID(pVnode), entryName);
-      continue;
-    } else {
-      // has other error
-      char errbuf[128];
-      regerror(code, &regex, errbuf, sizeof(errbuf));
-      smaWarn("vgId:%d, fetch qtask files, regexec failed since %s", TD_VID(pVnode), errbuf);
-      terrno = TSDB_CODE_RSMA_REGEX_MATCH;
-      goto _end;
-    }
-  }
-_end:
-  taosCloseDir(&pDir);
-  regfree(&regex);
-  return terrno == 0 ? TSDB_CODE_SUCCESS : TSDB_CODE_FAILED;
-}
-
-static int32_t tdQTaskFileCmprFn2(const void *p1, const void *p2) {
-  if (((SQTaskFile *)p1)->version < ((SQTaskFile *)p2)->version) {
-    return -1;
-  } else if (((SQTaskFile *)p1)->version > ((SQTaskFile *)p2)->version) {
-    return 1;
-  }
-
-  return 0;
-}
-
-int32_t tdRSmaFSUpsertQTaskFile(SRSmaFS *pFS, SQTaskFile *qTaskFile) {
-  int32_t code = 0;
-  int32_t idx = taosArraySearchIdx(pFS->aQTaskFSet, qTaskFile, tdQTaskFileCmprFn2, TD_GE);
-
-  if (idx < 0) {
-    idx = taosArrayGetSize(pFS->aQTaskFSet);
-  } else {
-    SQTaskFile *pTaskF = (SQTaskFile *)taosArrayGet(pFS->aQTaskFSet, idx);
-    int32_t     c = tdQTaskFileCmprFn2(pTaskF, qTaskFile);
-    if (c == 0) {
-      pTaskF->nRef = qTaskFile->nRef;
-      pTaskF->level = qTaskFile->level;
-      pTaskF->version = qTaskFile->version;
-      pTaskF->mtime = qTaskFile->mtime;
-      goto _exit;
-    }
-  }
-
-  if (taosArrayInsert(pFS->aQTaskFSet, idx, qTaskFile) == NULL) {
-    code = TSDB_CODE_OUT_OF_MEMORY;
-    goto _exit;
-  }
-
-_exit:
-  return code;
-}
\ No newline at end of file

From 2525a5cf1fb6567f5508a18f0bb7bc7d488fbe42 Mon Sep 17 00:00:00 2001
From: Benguang Zhao <bgzhao@taosdata.com>
Date: Tue, 18 Oct 2022 16:06:52 +0800
Subject: [PATCH 27/42] enh: report proper errorcode on disk unavailability

---
 source/dnode/mgmt/node_mgmt/src/dmEnv.c | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/source/dnode/mgmt/node_mgmt/src/dmEnv.c b/source/dnode/mgmt/node_mgmt/src/dmEnv.c
index becefcd6a0..5a7f149bc6 100644
--- a/source/dnode/mgmt/node_mgmt/src/dmEnv.c
+++ b/source/dnode/mgmt/node_mgmt/src/dmEnv.c
@@ -65,14 +65,17 @@ static bool dmCheckDiskSpace() {
   bool ret = true;
   if (!osDataSpaceAvailable()) {
     dError("data disk space unavailable, i.e. %s", tsDataDir);
+    terrno = TSDB_CODE_VND_NO_DISKSPACE;
     ret = false;
   }
   if (!osLogSpaceAvailable()) {
     dError("log disk space unavailable, i.e. %s", tsLogDir);
+    terrno = TSDB_CODE_VND_NO_DISKSPACE;
     ret = false;
   }
   if (!osTempSpaceAvailable()) {
     dError("temp disk space unavailable, i.e. %s", tsTempDir);
+    terrno = TSDB_CODE_VND_NO_DISKSPACE;
     ret = false;
   }
   return ret;

From 0fe14f0b167d09bbb2526430212c9e53422a8b5d Mon Sep 17 00:00:00 2001
From: dapan1121 <wpan@taosdata.com>
Date: Tue, 18 Oct 2022 16:33:27 +0800
Subject: [PATCH 28/42] fix: fix coverity issues

---
 include/util/tdef.h                       |  1 +
 source/client/src/clientImpl.c            | 16 ++++++---
 source/client/src/clientMain.c            |  1 -
 source/client/src/clientMsgHandler.c      |  7 ++--
 source/client/src/clientStmt.c            |  2 +-
 source/dnode/mnode/impl/src/mndQuery.c    | 25 +++++++++++++
 source/dnode/mnode/impl/src/mndStb.c      |  9 +++--
 source/dnode/vnode/src/vnd/vnodeQuery.c   |  5 +++
 source/libs/catalog/src/catalog.c         |  8 +++--
 source/libs/catalog/src/ctgAsync.c        |  4 +--
 source/libs/catalog/src/ctgCache.c        | 43 ++++++++++++-----------
 source/libs/command/src/command.c         |  7 +++-
 source/libs/executor/src/dataDispatcher.c |  8 ++++-
 source/libs/executor/src/dataInserter.c   |  2 +-
 source/libs/qcom/src/queryUtil.c          |  3 +-
 source/libs/qcom/src/querymsg.c           | 12 +++----
 source/libs/qworker/src/qwUtil.c          |  6 +++-
 source/libs/qworker/src/qworker.c         |  1 -
 source/libs/scheduler/src/schTask.c       |  3 +-
 19 files changed, 113 insertions(+), 50 deletions(-)

diff --git a/include/util/tdef.h b/include/util/tdef.h
index d9bb558b74..2103dc928e 100644
--- a/include/util/tdef.h
+++ b/include/util/tdef.h
@@ -498,6 +498,7 @@ enum {
 #define MAX_NUM_STR_SIZE 40
 
 #define MAX_META_MSG_IN_BATCH 1048576
+#define MAX_META_BATCH_RSP_SIZE (1 * 1048576 * 1024)
 
 #ifdef __cplusplus
 }
diff --git a/source/client/src/clientImpl.c b/source/client/src/clientImpl.c
index b494140da5..272f78a980 100644
--- a/source/client/src/clientImpl.c
+++ b/source/client/src/clientImpl.c
@@ -199,7 +199,7 @@ int32_t buildRequest(uint64_t connId, const char* sql, int sqlLen, void* param,
   if (tsQueryUseNodeAllocator && !qIsInsertValuesSql((*pRequest)->sqlstr, (*pRequest)->sqlLen)) {
     if (TSDB_CODE_SUCCESS !=
         nodesCreateAllocator((*pRequest)->requestId, tsQueryNodeChunkSize, &((*pRequest)->allocatorRefId))) {
-      tscError("%d failed to create node allocator, reqId:0x%" PRIx64 ", conn:%d, %s", (*pRequest)->self,
+      tscError("%d failed to create node allocator, reqId:0x%" PRIx64 ", conn:%" PRId64 ", %s", (*pRequest)->self,
                (*pRequest)->requestId, pTscObj->id, sql);
 
       destroyRequest(*pRequest);
@@ -955,7 +955,12 @@ SRequestObj* launchQueryImpl(SRequestObj* pRequest, SQuery* pQuery, bool keepQue
   switch (pQuery->execMode) {
     case QUERY_EXEC_MODE_LOCAL:
       if (!pRequest->validateOnly) {
-        code = execLocalCmd(pRequest, pQuery);
+        if (NULL == pQuery->pRoot) {
+          terrno = TSDB_CODE_INVALID_PARA;
+          code = terrno;
+        } else {
+          code = execLocalCmd(pRequest, pQuery);
+        }
       }
       break;
     case QUERY_EXEC_MODE_RPC:
@@ -997,7 +1002,7 @@ SRequestObj* launchQueryImpl(SRequestObj* pRequest, SQuery* pQuery, bool keepQue
 
   handleQueryExecRsp(pRequest);
 
-  if (NULL != pRequest && TSDB_CODE_SUCCESS != code) {
+  if (TSDB_CODE_SUCCESS != code) {
     pRequest->code = terrno;
   }
 
@@ -2253,7 +2258,10 @@ void syncQueryFn(void* param, void* res, int32_t code) {
 void taosAsyncQueryImpl(uint64_t connId, const char* sql, __taos_async_fn_t fp, void* param, bool validateOnly) {
   if (sql == NULL || NULL == fp) {
     terrno = TSDB_CODE_INVALID_PARA;
-    fp(param, NULL, terrno);
+    if (fp) {
+      fp(param, NULL, terrno);
+    }
+    
     return;
   }
 
diff --git a/source/client/src/clientMain.c b/source/client/src/clientMain.c
index 6126817ece..286fb85373 100644
--- a/source/client/src/clientMain.c
+++ b/source/client/src/clientMain.c
@@ -944,7 +944,6 @@ void taos_fetch_rows_a(TAOS_RES *res, __taos_async_fn_t fp, void *param) {
   if (pResultInfo->completed) {
     // it is a local executed query, no need to do async fetch
     if (QUERY_EXEC_MODE_LOCAL == pRequest->body.execMode) {
-      ASSERT(pResultInfo->numOfRows >= 0);
       if (pResultInfo->localResultFetched) {
         pResultInfo->numOfRows = 0;
         pResultInfo->current = 0;
diff --git a/source/client/src/clientMsgHandler.c b/source/client/src/clientMsgHandler.c
index cdf977bea3..0a1135886e 100644
--- a/source/client/src/clientMsgHandler.c
+++ b/source/client/src/clientMsgHandler.c
@@ -288,8 +288,10 @@ int32_t processDropDbRsp(void* param, SDataBuf* pMsg, int32_t code) {
     tDeserializeSDropDbRsp(pMsg->pData, pMsg->len, &dropdbRsp);
 
     struct SCatalog* pCatalog = NULL;
-    catalogGetHandle(pRequest->pTscObj->pAppInfo->clusterId, &pCatalog);
-    catalogRemoveDB(pCatalog, dropdbRsp.db, dropdbRsp.uid);
+    int32_t code = catalogGetHandle(pRequest->pTscObj->pAppInfo->clusterId, &pCatalog);
+    if (TSDB_CODE_SUCCESS == code) {
+      catalogRemoveDB(pCatalog, dropdbRsp.db, dropdbRsp.uid);
+    }
   }
 
   taosMemoryFree(pMsg->pData);
@@ -393,6 +395,7 @@ static int32_t buildShowVariablesRsp(SArray* pVars, SRetrieveTableRsp** pRsp) {
   size_t rspSize = sizeof(SRetrieveTableRsp) + blockGetEncodeSize(pBlock);
   *pRsp = taosMemoryCalloc(1, rspSize);
   if (NULL == *pRsp) {
+    blockDataDestroy(pBlock);
     return TSDB_CODE_OUT_OF_MEMORY;
   }
 
diff --git a/source/client/src/clientStmt.c b/source/client/src/clientStmt.c
index bf3fd00f14..81d0d616c9 100644
--- a/source/client/src/clientStmt.c
+++ b/source/client/src/clientStmt.c
@@ -152,7 +152,7 @@ int32_t stmtUpdateBindInfo(TAOS_STMT* stmt, STableMeta* pTableMeta, void* tags,
   pStmt->bInfo.tbType = pTableMeta->tableType;
   pStmt->bInfo.boundTags = tags;
   pStmt->bInfo.tagsCached = false;
-  strcpy(pStmt->bInfo.stbFName, sTableName);
+  tstrncpy(pStmt->bInfo.stbFName, sTableName, sizeof(pStmt->bInfo.stbFName));
 
   return TSDB_CODE_SUCCESS;
 }
diff --git a/source/dnode/mnode/impl/src/mndQuery.c b/source/dnode/mnode/impl/src/mndQuery.c
index 2e01fadbae..3e4c8005b4 100644
--- a/source/dnode/mnode/impl/src/mndQuery.c
+++ b/source/dnode/mnode/impl/src/mndQuery.c
@@ -90,14 +90,39 @@ int32_t mndProcessBatchMetaMsg(SRpcMsg *pMsg) {
   }
 
   for (int32_t i = 0; i < msgNum; ++i) {
+    if (offset >= pMsg->contLen) {
+      mError("offset %d is bigger than contLen %d", offset, pMsg->contLen);
+      terrno = TSDB_CODE_MSG_NOT_PROCESSED;
+      taosArrayDestroy(batchRsp);
+      return -1;
+    }
+
     req.msgIdx = ntohl(*(int32_t *)((char *)pMsg->pCont + offset));
     offset += sizeof(req.msgIdx);
+    if (offset >= pMsg->contLen) {
+      mError("offset %d is bigger than contLen %d", offset, pMsg->contLen);
+      terrno = TSDB_CODE_MSG_NOT_PROCESSED;
+      taosArrayDestroy(batchRsp);
+      return -1;
+    }
 
     req.msgType = ntohl(*(int32_t *)((char *)pMsg->pCont + offset));
     offset += sizeof(req.msgType);
+    if (offset >= pMsg->contLen) {
+      mError("offset %d is bigger than contLen %d", offset, pMsg->contLen);
+      terrno = TSDB_CODE_MSG_NOT_PROCESSED;
+      taosArrayDestroy(batchRsp);
+      return -1;
+    }
 
     req.msgLen = ntohl(*(int32_t *)((char *)pMsg->pCont + offset));
     offset += sizeof(req.msgLen);
+    if (offset >= pMsg->contLen) {
+      mError("offset %d is bigger than contLen %d", offset, pMsg->contLen);
+      terrno = TSDB_CODE_MSG_NOT_PROCESSED;
+      taosArrayDestroy(batchRsp);
+      return -1;
+    }
 
     req.msg = (char *)pMsg->pCont + offset;
     offset += req.msgLen;
diff --git a/source/dnode/mnode/impl/src/mndStb.c b/source/dnode/mnode/impl/src/mndStb.c
index 10cfb0a660..8a3179b2a9 100644
--- a/source/dnode/mnode/impl/src/mndStb.c
+++ b/source/dnode/mnode/impl/src/mndStb.c
@@ -2553,12 +2553,17 @@ static int32_t mndRetrieveStb(SRpcMsg *pReq, SShowObj *pShow, SSDataBlock *pBloc
 
     char    rollup[160 + VARSTR_HEADER_SIZE] = {0};
     int32_t rollupNum = (int32_t)taosArrayGetSize(pStb->pFuncs);
+    char   *sep = ", ";
+    int32_t sepLen = strlen(sep);
+    int32_t rollupLen = sizeof(rollup) - 2;
     for (int32_t i = 0; i < rollupNum; ++i) {
       char *funcName = taosArrayGet(pStb->pFuncs, i);
       if (i) {
-        strcat(varDataVal(rollup), ", ");
+        strncat(varDataVal(rollup), sep, rollupLen);
+        rollupLen -= sepLen;
       }
-      strcat(varDataVal(rollup), funcName);
+      strncat(varDataVal(rollup), funcName, rollupLen);
+      rollupLen -= strlen(funcName);
     }
     varDataSetLen(rollup, strlen(varDataVal(rollup)));
 
diff --git a/source/dnode/vnode/src/vnd/vnodeQuery.c b/source/dnode/vnode/src/vnd/vnodeQuery.c
index 8c1f858cbb..64df3aa1eb 100644
--- a/source/dnode/vnode/src/vnd/vnodeQuery.c
+++ b/source/dnode/vnode/src/vnd/vnodeQuery.c
@@ -330,6 +330,11 @@ int32_t vnodeGetBatchMeta(SVnode *pVnode, SRpcMsg *pMsg) {
   rspSize += sizeof(int32_t);
   offset = 0;
 
+  if (rspSize > MAX_META_BATCH_RSP_SIZE) {
+    code = TSDB_CODE_INVALID_MSG_LEN;
+    goto _exit;
+  }
+  
   pRsp = rpcMallocCont(rspSize);
   if (pRsp == NULL) {
     code = TSDB_CODE_OUT_OF_MEMORY;
diff --git a/source/libs/catalog/src/catalog.c b/source/libs/catalog/src/catalog.c
index b960103d94..218a86ed5c 100644
--- a/source/libs/catalog/src/catalog.c
+++ b/source/libs/catalog/src/catalog.c
@@ -302,9 +302,11 @@ int32_t ctgUpdateTbMeta(SCatalog* pCtg, STableMetaRsp* rspMsg, bool syncOp) {
 
 _return:
 
-  taosMemoryFreeClear(output->tbMeta);
-  taosMemoryFreeClear(output);
-
+  if (output) {
+    taosMemoryFreeClear(output->tbMeta);
+    taosMemoryFreeClear(output);
+  }
+  
   CTG_RET(code);
 }
 
diff --git a/source/libs/catalog/src/ctgAsync.c b/source/libs/catalog/src/ctgAsync.c
index 1d2e3640a1..93d36bc4b3 100644
--- a/source/libs/catalog/src/ctgAsync.c
+++ b/source/libs/catalog/src/ctgAsync.c
@@ -252,7 +252,7 @@ int32_t ctgInitGetIndexTask(SCtgJob* pJob, int32_t taskIdx, void* param) {
 
   SCtgIndexCtx* ctx = task.taskCtx;
 
-  strcpy(ctx->indexFName, name);
+  tstrncpy(ctx->indexFName, name, sizeof(ctx->indexFName));
 
   taosArrayPush(pJob->pTasks, &task);
 
@@ -277,7 +277,7 @@ int32_t ctgInitGetUdfTask(SCtgJob* pJob, int32_t taskIdx, void* param) {
 
   SCtgUdfCtx* ctx = task.taskCtx;
 
-  strcpy(ctx->udfName, name);
+  tstrncpy(ctx->udfName, name, sizeof(ctx->udfName));
 
   taosArrayPush(pJob->pTasks, &task);
 
diff --git a/source/libs/catalog/src/ctgCache.c b/source/libs/catalog/src/ctgCache.c
index 89571ebf55..fe9a131ab8 100644
--- a/source/libs/catalog/src/ctgCache.c
+++ b/source/libs/catalog/src/ctgCache.c
@@ -660,7 +660,7 @@ int32_t ctgDropDbCacheEnqueue(SCatalog *pCtg, const char *dbFName, int64_t dbId)
   }
 
   msg->pCtg = pCtg;
-  strncpy(msg->dbFName, dbFName, sizeof(msg->dbFName));
+  tstrncpy(msg->dbFName, dbFName, sizeof(msg->dbFName));
   msg->dbId = dbId;
 
   op->data = msg;
@@ -693,7 +693,7 @@ int32_t ctgDropDbVgroupEnqueue(SCatalog *pCtg, const char *dbFName, bool syncOp)
   }
 
   msg->pCtg = pCtg;
-  strncpy(msg->dbFName, dbFName, sizeof(msg->dbFName));
+  tstrncpy(msg->dbFName, dbFName, sizeof(msg->dbFName));
 
   op->data = msg;
 
@@ -721,8 +721,8 @@ int32_t ctgDropStbMetaEnqueue(SCatalog *pCtg, const char *dbFName, int64_t dbId,
   }
 
   msg->pCtg = pCtg;
-  strncpy(msg->dbFName, dbFName, sizeof(msg->dbFName));
-  strncpy(msg->stbName, stbName, sizeof(msg->stbName));
+  tstrncpy(msg->dbFName, dbFName, sizeof(msg->dbFName));
+  tstrncpy(msg->stbName, stbName, sizeof(msg->stbName));
   msg->dbId = dbId;
   msg->suid = suid;
 
@@ -751,8 +751,8 @@ int32_t ctgDropTbMetaEnqueue(SCatalog *pCtg, const char *dbFName, int64_t dbId,
   }
 
   msg->pCtg = pCtg;
-  strncpy(msg->dbFName, dbFName, sizeof(msg->dbFName));
-  strncpy(msg->tbName, tbName, sizeof(msg->tbName));
+  tstrncpy(msg->dbFName, dbFName, sizeof(msg->dbFName));
+  tstrncpy(msg->tbName, tbName, sizeof(msg->tbName));
   msg->dbId = dbId;
 
   op->data = msg;
@@ -785,7 +785,7 @@ int32_t ctgUpdateVgroupEnqueue(SCatalog *pCtg, const char *dbFName, int64_t dbId
     dbFName = p + 1;
   }
 
-  strncpy(msg->dbFName, dbFName, sizeof(msg->dbFName));
+  tstrncpy(msg->dbFName, dbFName, sizeof(msg->dbFName));
   msg->pCtg = pCtg;
   msg->dbId = dbId;
   msg->dbInfo = dbInfo;
@@ -817,7 +817,8 @@ int32_t ctgUpdateTbMetaEnqueue(SCatalog *pCtg, STableMetaOutput *output, bool sy
 
   char *p = strchr(output->dbFName, '.');
   if (p && IS_SYS_DBNAME(p + 1)) {
-    memmove(output->dbFName, p + 1, strlen(p + 1));
+    int32_t len = strlen(p + 1);
+    memmove(output->dbFName, p + 1, len >= TSDB_DB_FNAME_LEN ? TSDB_DB_FNAME_LEN - 1 : len);
   }
 
   msg->pCtg = pCtg;
@@ -852,7 +853,7 @@ int32_t ctgUpdateVgEpsetEnqueue(SCatalog *pCtg, char *dbFName, int32_t vgId, SEp
   }
 
   msg->pCtg = pCtg;
-  strcpy(msg->dbFName, dbFName);
+  tstrncpy(msg->dbFName, dbFName, sizeof(msg->dbFName));
   msg->vgId = vgId;
   msg->epSet = *pEpSet;
 
@@ -1215,7 +1216,7 @@ int32_t ctgAddNewDBCache(SCatalog *pCtg, const char *dbFName, uint64_t dbId) {
   CTG_CACHE_STAT_INC(numOfDb, 1);
 
   SDbVgVersion vgVersion = {.dbId = newDBCache.dbId, .vgVersion = -1};
-  strncpy(vgVersion.dbFName, dbFName, sizeof(vgVersion.dbFName));
+  tstrncpy(vgVersion.dbFName, dbFName, sizeof(vgVersion.dbFName));
 
   ctgDebug("db added to cache, dbFName:%s, dbId:0x%" PRIx64, dbFName, dbId);
 
@@ -1331,8 +1332,8 @@ int32_t ctgUpdateRentStbVersion(SCatalog *pCtg, char *dbFName, char *tbName, uin
     metaRent.smaVer = pCache->pIndex->version;
   }
 
-  strcpy(metaRent.dbFName, dbFName);
-  strcpy(metaRent.stbName, tbName);
+  tstrncpy(metaRent.dbFName, dbFName, sizeof(metaRent.dbFName));
+  tstrncpy(metaRent.stbName, tbName, sizeof(metaRent.stbName));
 
   CTG_ERR_RET(ctgMetaRentUpdate(&pCtg->stbRent, &metaRent, metaRent.suid, sizeof(SSTableVersion),
                                 ctgStbVersionSortCompare, ctgStbVersionSearchCompare));
@@ -1416,8 +1417,10 @@ int32_t ctgWriteTbMetaToCache(SCatalog *pCtg, SCtgDBCache *dbCache, char *dbFNam
   ctgDebug("stb 0x%" PRIx64 " updated to cache, dbFName:%s, tbName:%s, tbType:%d", meta->suid, dbFName, tbName,
            meta->tableType);
 
-  CTG_ERR_RET(ctgUpdateRentStbVersion(pCtg, dbFName, tbName, dbId, meta->suid, pCache));
-
+  if (pCache) {
+    CTG_ERR_RET(ctgUpdateRentStbVersion(pCtg, dbFName, tbName, dbId, meta->suid, pCache));
+  }
+  
   return TSDB_CODE_SUCCESS;
 }
 
@@ -1584,7 +1587,7 @@ int32_t ctgOpUpdateVgroup(SCtgCacheOperation *operation) {
 
   dbCache = NULL;
 
-  strncpy(vgVersion.dbFName, dbFName, sizeof(vgVersion.dbFName));
+  tstrncpy(vgVersion.dbFName, dbFName, sizeof(vgVersion.dbFName));
   CTG_ERR_JRET(ctgMetaRentUpdate(&msg->pCtg->dbRent, &vgVersion, vgVersion.dbId, sizeof(SDbVgVersion),
                                  ctgDbVgVersionSortCompare, ctgDbVgVersionSearchCompare));
 
@@ -1674,9 +1677,9 @@ int32_t ctgOpUpdateTbMeta(SCtgCacheOperation *operation) {
 
   if (CTG_IS_META_TABLE(pMeta->metaType) || CTG_IS_META_BOTH(pMeta->metaType)) {
     int32_t metaSize = CTG_META_SIZE(pMeta->tbMeta);
-    CTG_ERR_JRET(
-        ctgWriteTbMetaToCache(pCtg, dbCache, pMeta->dbFName, pMeta->dbId, pMeta->tbName, pMeta->tbMeta, metaSize));
+    code = ctgWriteTbMetaToCache(pCtg, dbCache, pMeta->dbFName, pMeta->dbId, pMeta->tbName, pMeta->tbMeta, metaSize);
     pMeta->tbMeta = NULL;
+    CTG_ERR_JRET(code);
   }
 
   if (CTG_IS_META_CTABLE(pMeta->metaType) || CTG_IS_META_BOTH(pMeta->metaType)) {
@@ -1691,10 +1694,8 @@ int32_t ctgOpUpdateTbMeta(SCtgCacheOperation *operation) {
 
 _return:
 
-  if (pMeta) {
-    taosMemoryFreeClear(pMeta->tbMeta);
-    taosMemoryFreeClear(pMeta);
-  }
+  taosMemoryFreeClear(pMeta->tbMeta);
+  taosMemoryFreeClear(pMeta);
 
   taosMemoryFreeClear(msg);
 
diff --git a/source/libs/command/src/command.c b/source/libs/command/src/command.c
index a9134de537..b4229880b6 100644
--- a/source/libs/command/src/command.c
+++ b/source/libs/command/src/command.c
@@ -317,7 +317,12 @@ int32_t appendTagValues(char* buf, int32_t* len, STableCfg* pCfg) {
   SArray* pTagVals = NULL;
   STag*   pTag = (STag*)pCfg->pTags;
 
-  if (pCfg->pTags && tTagIsJson(pTag)) {
+  if (NULL == pCfg->pTags || pCfg->numOfTags <= 0) {
+    qError("tag missed in table cfg, pointer:%p, numOfTags:%d", pCfg->pTags, pCfg->numOfTags);
+    return TSDB_CODE_APP_ERROR;
+  }
+
+  if (tTagIsJson(pTag)) {
     char* pJson = parseTagDatatoJson(pTag);
     if (pJson) {
       *len += sprintf(buf + VARSTR_HEADER_SIZE + *len, "%s", pJson);
diff --git a/source/libs/executor/src/dataDispatcher.c b/source/libs/executor/src/dataDispatcher.c
index 8f8e486b3a..9a27110345 100644
--- a/source/libs/executor/src/dataDispatcher.c
+++ b/source/libs/executor/src/dataDispatcher.c
@@ -143,9 +143,15 @@ static int32_t getStatus(SDataDispatchHandle* pDispatcher) {
 static int32_t putDataBlock(SDataSinkHandle* pHandle, const SInputData* pInput, bool* pContinue) {
   SDataDispatchHandle* pDispatcher = (SDataDispatchHandle*)pHandle;
   SDataDispatchBuf*    pBuf = taosAllocateQitem(sizeof(SDataDispatchBuf), DEF_QITEM);
-  if (NULL == pBuf || !allocBuf(pDispatcher, pInput, pBuf)) {
+  if (NULL == pBuf) {
     return TSDB_CODE_QRY_OUT_OF_MEMORY;
   }
+
+  if (!allocBuf(pDispatcher, pInput, pBuf)) {
+    taosFreeQitem(pBuf);
+    return TSDB_CODE_QRY_OUT_OF_MEMORY;
+  }
+  
   toDataCacheEntry(pDispatcher, pInput, pBuf);
   taosWriteQitem(pDispatcher->pDataBlocks, pBuf);
   *pContinue = (DS_BUF_LOW == updateStatus(pDispatcher) ? true : false);
diff --git a/source/libs/executor/src/dataInserter.c b/source/libs/executor/src/dataInserter.c
index 4c4ba59fa9..ed455e5e75 100644
--- a/source/libs/executor/src/dataInserter.c
+++ b/source/libs/executor/src/dataInserter.c
@@ -323,7 +323,7 @@ int32_t createDataInserter(SDataSinkManager* pManager, const SDataSinkNode* pDat
   int32_t code =
       tsdbGetTableSchema(inserter->pParam->readHandle->vnode, pInserterNode->tableId, &inserter->pSchema, &suid);
   if (code) {
-    destroyDataSinker((SDataSinkHandle*)pInserterNode);
+    destroyDataSinker((SDataSinkHandle*)inserter);
     return code;
   }
 
diff --git a/source/libs/qcom/src/queryUtil.c b/source/libs/qcom/src/queryUtil.c
index a917397d02..618f597d72 100644
--- a/source/libs/qcom/src/queryUtil.c
+++ b/source/libs/qcom/src/queryUtil.c
@@ -357,8 +357,7 @@ char* parseTagDatatoJson(void* p) {
   for (int j = 0; j < nCols; ++j) {
     STagVal* pTagVal = (STagVal*)taosArrayGet(pTagVals, j);
     // json key  encode by binary
-    memset(tagJsonKey, 0, sizeof(tagJsonKey));
-    memcpy(tagJsonKey, pTagVal->pKey, strlen(pTagVal->pKey));
+    tstrncpy(tagJsonKey, pTagVal->pKey, sizeof(tagJsonKey));
     // json value
     char type = pTagVal->type;
     if (type == TSDB_DATA_TYPE_NULL) {
diff --git a/source/libs/qcom/src/querymsg.c b/source/libs/qcom/src/querymsg.c
index a6f26088de..fadd07a9f3 100644
--- a/source/libs/qcom/src/querymsg.c
+++ b/source/libs/qcom/src/querymsg.c
@@ -173,7 +173,7 @@ int32_t queryBuildGetDBCfgMsg(void *input, char **msg, int32_t msgSize, int32_t
   }
 
   SDbCfgReq dbCfgReq = {0};
-  strcpy(dbCfgReq.db, input);
+  strncpy(dbCfgReq.db, input, sizeof(dbCfgReq.db) - 1);
 
   int32_t bufLen = tSerializeSDbCfgReq(NULL, 0, &dbCfgReq);
   void   *pBuf = (*mallcFp)(bufLen);
@@ -191,7 +191,7 @@ int32_t queryBuildGetIndexMsg(void *input, char **msg, int32_t msgSize, int32_t
   }
 
   SUserIndexReq indexReq = {0};
-  strcpy(indexReq.indexFName, input);
+  strncpy(indexReq.indexFName, input, sizeof(indexReq.indexFName) - 1);
 
   int32_t bufLen = tSerializeSUserIndexReq(NULL, 0, &indexReq);
   void   *pBuf = (*mallcFp)(bufLen);
@@ -233,7 +233,7 @@ int32_t queryBuildGetUserAuthMsg(void *input, char **msg, int32_t msgSize, int32
   }
 
   SGetUserAuthReq req = {0};
-  strncpy(req.user, input, sizeof(req.user));
+  strncpy(req.user, input, sizeof(req.user) - 1);
 
   int32_t bufLen = tSerializeSGetUserAuthReq(NULL, 0, &req);
   void   *pBuf = (*mallcFp)(bufLen);
@@ -251,7 +251,7 @@ int32_t queryBuildGetTbIndexMsg(void *input, char **msg, int32_t msgSize, int32_
   }
 
   STableIndexReq indexReq = {0};
-  strcpy(indexReq.tbFName, input);
+  strncpy(indexReq.tbFName, input, sizeof(indexReq.tbFName) - 1);
 
   int32_t bufLen = tSerializeSTableIndexReq(NULL, 0, &indexReq);
   void   *pBuf = (*mallcFp)(bufLen);
@@ -271,8 +271,8 @@ int32_t queryBuildGetTbCfgMsg(void *input, char **msg, int32_t msgSize, int32_t
   SBuildTableInput *pInput = input;
   STableCfgReq      cfgReq = {0};
   cfgReq.header.vgId = pInput->vgId;
-  strncpy(cfgReq.dbFName, pInput->dbFName, sizeof(cfgReq.dbFName));
-  strncpy(cfgReq.tbName, pInput->tbName, sizeof(cfgReq.tbName));
+  strncpy(cfgReq.dbFName, pInput->dbFName, sizeof(cfgReq.dbFName) - 1);
+  strncpy(cfgReq.tbName, pInput->tbName, sizeof(cfgReq.tbName) - 1);
 
   int32_t bufLen = tSerializeSTableCfgReq(NULL, 0, &cfgReq);
   void   *pBuf = (*mallcFp)(bufLen);
diff --git a/source/libs/qworker/src/qwUtil.c b/source/libs/qworker/src/qwUtil.c
index 3038b87930..e9ded9b269 100644
--- a/source/libs/qworker/src/qwUtil.c
+++ b/source/libs/qworker/src/qwUtil.c
@@ -412,7 +412,7 @@ void qwSetHbParam(int64_t refId, SQWHbParam **pParam) {
   while (true) {
     paramIdx = atomic_load_32(&gQwMgmt.paramIdx);
     if (paramIdx == tListLen(gQwMgmt.param)) {
-      newParamIdx = 0;
+      newParamIdx = 1;
     } else {
       newParamIdx = paramIdx + 1;
     }
@@ -422,6 +422,10 @@ void qwSetHbParam(int64_t refId, SQWHbParam **pParam) {
     }
   }
 
+  if (paramIdx == tListLen(gQwMgmt.param)) {
+    paramIdx = 0;
+  }
+
   gQwMgmt.param[paramIdx].qwrId = gQwMgmt.qwRef;
   gQwMgmt.param[paramIdx].refId = refId;
 
diff --git a/source/libs/qworker/src/qworker.c b/source/libs/qworker/src/qworker.c
index 0ff40f3b9a..3df16563e2 100644
--- a/source/libs/qworker/src/qworker.c
+++ b/source/libs/qworker/src/qworker.c
@@ -398,7 +398,6 @@ int32_t qwHandlePrePhaseEvents(QW_FPARAMS_DEF, int8_t phase, SQWPhaseInput *inpu
       if (QW_EVENT_PROCESSED(ctx, QW_EVENT_DROP)) {
         QW_TASK_ELOG("task already dropped at wrong phase %s", qwPhaseStr(phase));
         QW_ERR_JRET(TSDB_CODE_QRY_TASK_STATUS_ERROR);
-        break;
       }
 
       if (QW_EVENT_RECEIVED(ctx, QW_EVENT_DROP)) {
diff --git a/source/libs/scheduler/src/schTask.c b/source/libs/scheduler/src/schTask.c
index c641b88152..b585373b0a 100644
--- a/source/libs/scheduler/src/schTask.c
+++ b/source/libs/scheduler/src/schTask.c
@@ -430,7 +430,8 @@ int32_t schHandleRedirect(SSchJob *pJob, SSchTask *pTask, SDataBuf *pData, int32
   if (SCH_IS_DATA_BIND_TASK(pTask)) {
     if (NULL == pData->pEpSet) {
       SCH_TASK_ELOG("no epset updated while got error %s", tstrerror(rspCode));
-      SCH_ERR_JRET(rspCode);
+      code = rspCode;
+      goto _return;
     }
   }
 

From ca2012a521ca4bab5c1253e586e2001efb559da8 Mon Sep 17 00:00:00 2001
From: Liu Jicong <liujicong@qq.com>
Date: Tue, 18 Oct 2022 16:56:07 +0800
Subject: [PATCH 29/42] refactor: tmq commit

---
 source/client/src/clientTmq.c      | 51 ++++++++++++++++--------------
 source/dnode/vnode/src/inc/tq.h    |  6 ++--
 source/dnode/vnode/src/tq/tq.c     |  4 +--
 source/dnode/vnode/src/tq/tqExec.c |  2 +-
 source/dnode/vnode/src/tq/tqSink.c |  4 +--
 5 files changed, 36 insertions(+), 31 deletions(-)

diff --git a/source/client/src/clientTmq.c b/source/client/src/clientTmq.c
index 838b851d6d..0bc33739c6 100644
--- a/source/client/src/clientTmq.c
+++ b/source/client/src/clientTmq.c
@@ -210,6 +210,8 @@ typedef struct {
 typedef struct {
   SMqCommitCbParamSet* params;
   STqOffset*           pOffset;
+  /*char                 topicName[TSDB_TOPIC_FNAME_LEN];*/
+  /*int32_t              vgId;*/
 } SMqCommitCbParam;
 
 tmq_conf_t* tmq_conf_new() {
@@ -407,6 +409,14 @@ int32_t tmqCommitDone(SMqCommitCbParamSet* pParamSet) {
   return 0;
 }
 
+static void tmqCommitRspCountDown(SMqCommitCbParamSet* pParamSet) {
+  int32_t waitingRspNum = atomic_sub_fetch_32(&pParamSet->waitingRspNum, 1);
+  ASSERT(waitingRspNum >= 0);
+  if (waitingRspNum == 0) {
+    tmqCommitDone(pParamSet);
+  }
+}
+
 int32_t tmqCommitCb(void* param, SDataBuf* pBuf, int32_t code) {
   SMqCommitCbParam*    pParam = (SMqCommitCbParam*)param;
   SMqCommitCbParamSet* pParamSet = (SMqCommitCbParamSet*)pParam->params;
@@ -420,18 +430,13 @@ int32_t tmqCommitCb(void* param, SDataBuf* pBuf, int32_t code) {
 #endif
 
   taosMemoryFree(pParam->pOffset);
-  if (pBuf->pData) taosMemoryFree(pBuf->pData);
+  taosMemoryFree(pBuf->pData);
 
   /*tscDebug("receive offset commit cb of %s on vgId:%d, offset is %" PRId64, pParam->pOffset->subKey, pParam->->vgId,
    * pOffset->version);*/
 
-  // count down waiting rsp
-  int32_t waitingRspNum = atomic_sub_fetch_32(&pParamSet->waitingRspNum, 1);
-  ASSERT(waitingRspNum >= 0);
+  tmqCommitRspCountDown(pParamSet);
 
-  if (waitingRspNum == 0) {
-    tmqCommitDone(pParamSet);
-  }
   return 0;
 }
 
@@ -591,14 +596,10 @@ FAIL:
   return 0;
 }
 
-int32_t tmqCommitInner(tmq_t* tmq, const TAOS_RES* msg, int8_t automatic, int8_t async, tmq_commit_cb* userCb,
-                       void* userParam) {
+static int32_t tmqCommitConsumerImpl(tmq_t* tmq, int8_t automatic, int8_t async, tmq_commit_cb* userCb,
+                                     void* userParam) {
   int32_t code = -1;
 
-  if (msg != NULL) {
-    return tmqCommitMsgImpl(tmq, msg, async, userCb, userParam);
-  }
-
   SMqCommitCbParamSet* pParamSet = taosMemoryCalloc(1, sizeof(SMqCommitCbParamSet));
   if (pParamSet == NULL) {
     code = TSDB_CODE_OUT_OF_MEMORY;
@@ -646,33 +647,37 @@ int32_t tmqCommitInner(tmq_t* tmq, const TAOS_RES* msg, int8_t automatic, int8_t
     }
   }
 
+  // no request is sent
   if (pParamSet->totalRspNum == 0) {
     tsem_destroy(&pParamSet->rspSem);
     taosMemoryFree(pParamSet);
     return 0;
   }
 
-  int32_t waitingRspNum = atomic_sub_fetch_32(&pParamSet->waitingRspNum, 1);
-  ASSERT(waitingRspNum >= 0);
-  if (waitingRspNum == 0) {
-    tmqCommitDone(pParamSet);
-  }
+  // count down since waiting rsp num init as 1
+  tmqCommitRspCountDown(pParamSet);
 
   if (!async) {
     tsem_wait(&pParamSet->rspSem);
     code = pParamSet->rspErr;
     tsem_destroy(&pParamSet->rspSem);
     taosMemoryFree(pParamSet);
-  }
-
 #if 0
-  if (!async) {
     taosArrayDestroyP(pParamSet->successfulOffsets, taosMemoryFree);
     taosArrayDestroyP(pParamSet->failedOffsets, taosMemoryFree);
-  }
 #endif
+  }
 
-  return 0;
+  return code;
+}
+
+int32_t tmqCommitInner(tmq_t* tmq, const TAOS_RES* msg, int8_t automatic, int8_t async, tmq_commit_cb* userCb,
+                       void* userParam) {
+  if (msg) {
+    return tmqCommitMsgImpl(tmq, msg, async, userCb, userParam);
+  } else {
+    return tmqCommitConsumerImpl(tmq, automatic, async, userCb, userParam);
+  }
 }
 
 void tmqAssignAskEpTask(void* param, void* tmrId) {
diff --git a/source/dnode/vnode/src/inc/tq.h b/source/dnode/vnode/src/inc/tq.h
index 69e6cdce9f..7f418439a8 100644
--- a/source/dnode/vnode/src/inc/tq.h
+++ b/source/dnode/vnode/src/inc/tq.h
@@ -149,7 +149,7 @@ int32_t tEncodeSTqHandle(SEncoder* pEncoder, const STqHandle* pHandle);
 int32_t tDecodeSTqHandle(SDecoder* pDecoder, STqHandle* pHandle);
 
 // tqRead
-int32_t tqScan(STQ* pTq, const STqHandle* pHandle, STaosxRsp* pRsp, SMqMetaRsp* pMetaRsp, STqOffsetVal* offset);
+int32_t tqScanTaosx(STQ* pTq, const STqHandle* pHandle, STaosxRsp* pRsp, SMqMetaRsp* pMetaRsp, STqOffsetVal* offset);
 int32_t tqScanData(STQ* pTq, const STqHandle* pHandle, SMqDataRsp* pRsp, STqOffsetVal* pOffset);
 int64_t tqFetchLog(STQ* pTq, STqHandle* pHandle, int64_t* fetchOffset, SWalCkHead** pHeadWithCkSum);
 
@@ -181,8 +181,8 @@ int32_t         tqOffsetDelete(STqOffsetStore* pStore, const char* subscribeKey)
 int32_t         tqOffsetCommitFile(STqOffsetStore* pStore);
 
 // tqSink
-void tqTableSink(SStreamTask* pTask, void* vnode, int64_t ver, void* data);
-void tqTableSink1(SStreamTask* pTask, void* vnode, int64_t ver, void* data);
+void tqSinkToTableMerge(SStreamTask* pTask, void* vnode, int64_t ver, void* data);
+void tqSinkToTablePipeline(SStreamTask* pTask, void* vnode, int64_t ver, void* data);
 
 // tqOffset
 char*   tqOffsetBuildFName(const char* path, int32_t ver);
diff --git a/source/dnode/vnode/src/tq/tq.c b/source/dnode/vnode/src/tq/tq.c
index 5e1cc15063..6d71496008 100644
--- a/source/dnode/vnode/src/tq/tq.c
+++ b/source/dnode/vnode/src/tq/tq.c
@@ -595,7 +595,7 @@ int32_t tqProcessPollReq(STQ* pTq, SRpcMsg* pMsg) {
   tqInitTaosxRsp(&taosxRsp, pReq);
 
   if (fetchOffsetNew.type != TMQ_OFFSET__LOG) {
-    tqScan(pTq, pHandle, &taosxRsp, &metaRsp, &fetchOffsetNew);
+    tqScanTaosx(pTq, pHandle, &taosxRsp, &metaRsp, &fetchOffsetNew);
 
     if (metaRsp.metaRspLen > 0) {
       if (tqSendMetaPollRsp(pTq, pMsg, pReq, &metaRsp) < 0) {
@@ -924,7 +924,7 @@ int32_t tqExpandTask(STQ* pTq, SStreamTask* pTask) {
     pTask->smaSink.smaSink = smaHandleRes;
   } else if (pTask->outputType == TASK_OUTPUT__TABLE) {
     pTask->tbSink.vnode = pTq->pVnode;
-    pTask->tbSink.tbSinkFunc = tqTableSink1;
+    pTask->tbSink.tbSinkFunc = tqSinkToTablePipeline;
 
     ASSERT(pTask->tbSink.pSchemaWrapper);
     ASSERT(pTask->tbSink.pSchemaWrapper->pSchema);
diff --git a/source/dnode/vnode/src/tq/tqExec.c b/source/dnode/vnode/src/tq/tqExec.c
index 58d051bec1..305ee82982 100644
--- a/source/dnode/vnode/src/tq/tqExec.c
+++ b/source/dnode/vnode/src/tq/tqExec.c
@@ -123,7 +123,7 @@ int32_t tqScanData(STQ* pTq, const STqHandle* pHandle, SMqDataRsp* pRsp, STqOffs
   return 0;
 }
 
-int32_t tqScan(STQ* pTq, const STqHandle* pHandle, STaosxRsp* pRsp, SMqMetaRsp* pMetaRsp, STqOffsetVal* pOffset) {
+int32_t tqScanTaosx(STQ* pTq, const STqHandle* pHandle, STaosxRsp* pRsp, SMqMetaRsp* pMetaRsp, STqOffsetVal* pOffset) {
   const STqExecHandle* pExec = &pHandle->execHandle;
   qTaskInfo_t          task = pExec->task;
 
diff --git a/source/dnode/vnode/src/tq/tqSink.c b/source/dnode/vnode/src/tq/tqSink.c
index ab195b1eb3..79d7e27642 100644
--- a/source/dnode/vnode/src/tq/tqSink.c
+++ b/source/dnode/vnode/src/tq/tqSink.c
@@ -284,7 +284,7 @@ SSubmitReq* tqBlockToSubmit(SVnode* pVnode, const SArray* pBlocks, const STSchem
   return ret;
 }
 
-void tqTableSink1(SStreamTask* pTask, void* vnode, int64_t ver, void* data) {
+void tqSinkToTablePipeline(SStreamTask* pTask, void* vnode, int64_t ver, void* data) {
   const SArray*   pBlocks = (const SArray*)data;
   SVnode*         pVnode = (SVnode*)vnode;
   int64_t         suid = pTask->tbSink.stbUid;
@@ -528,7 +528,7 @@ void tqTableSink1(SStreamTask* pTask, void* vnode, int64_t ver, void* data) {
   taosArrayDestroy(tagArray);
 }
 
-void tqTableSink(SStreamTask* pTask, void* vnode, int64_t ver, void* data) {
+void tqSinkToTableMerge(SStreamTask* pTask, void* vnode, int64_t ver, void* data) {
   const SArray*   pRes = (const SArray*)data;
   SVnode*         pVnode = (SVnode*)vnode;
   SBatchDeleteReq deleteReq = {0};

From 9e0feee081e1c6e77c877cd8926cf007d1c4bae0 Mon Sep 17 00:00:00 2001
From: Shengliang Guan <slguan@taosdata.com>
Date: Tue, 18 Oct 2022 17:06:03 +0800
Subject: [PATCH 30/42] enh: sync propose return at once if replica is 1

---
 source/dnode/mnode/impl/src/mndMnode.c | 1 +
 source/dnode/mnode/impl/src/mndSync.c  | 8 ++++++++
 source/libs/sync/src/syncMain.c        | 6 ++++--
 3 files changed, 13 insertions(+), 2 deletions(-)

diff --git a/source/dnode/mnode/impl/src/mndMnode.c b/source/dnode/mnode/impl/src/mndMnode.c
index f25d436aee..e23b895ebd 100644
--- a/source/dnode/mnode/impl/src/mndMnode.c
+++ b/source/dnode/mnode/impl/src/mndMnode.c
@@ -745,6 +745,7 @@ static void mndReloadSyncConfig(SMnode *pMnode) {
     mInfo("vgId:1, mnode sync not reconfig since readyMnodes:%d updatingMnodes:%d", readyMnodes, updatingMnodes);
     return;
   }
+    // ASSERT(0);
 
   if (cfg.myIndex == -1) {
 #if 1
diff --git a/source/dnode/mnode/impl/src/mndSync.c b/source/dnode/mnode/impl/src/mndSync.c
index 866a3fa8b9..ac95dd2795 100644
--- a/source/dnode/mnode/impl/src/mndSync.c
+++ b/source/dnode/mnode/impl/src/mndSync.c
@@ -293,6 +293,14 @@ int32_t mndSyncPropose(SMnode *pMnode, SSdbRaw *pRaw, int32_t transId) {
 
   if (code == 0) {
     tsem_wait(&pMgmt->syncSem);
+  } else if (code > 0) {
+    mInfo("trans:%d, confirm at once since replica is 1, continue execute", transId);
+    taosWLockLatch(&pMgmt->lock);
+    pMgmt->transId = 0;
+    taosWUnLockLatch(&pMgmt->lock);
+    sdbWriteWithoutFree(pMnode->pSdb, pRaw);
+    sdbSetApplyInfo(pMnode->pSdb, req.info.conn.applyIndex, req.info.conn.applyTerm, SYNC_INDEX_INVALID);
+    code = 0;
   } else if (code == -1 && terrno == TSDB_CODE_SYN_NOT_LEADER) {
     terrno = TSDB_CODE_APP_NOT_READY;
   } else if (code == -1 && terrno == TSDB_CODE_SYN_INTERNAL_ERROR) {
diff --git a/source/libs/sync/src/syncMain.c b/source/libs/sync/src/syncMain.c
index 83bd7a5f38..6e4eaa0aaf 100644
--- a/source/libs/sync/src/syncMain.c
+++ b/source/libs/sync/src/syncMain.c
@@ -3032,7 +3032,8 @@ static int32_t syncNodeProposeConfigChangeFinish(SSyncNode* ths, SyncReconfigFin
 }
 
 bool syncNodeIsOptimizedOneReplica(SSyncNode* ths, SRpcMsg* pMsg) {
-  return (ths->replicaNum == 1 && syncUtilUserCommit(pMsg->msgType) && ths->vgId != 1);
+  return (ths->replicaNum == 1 && syncUtilUserCommit(pMsg->msgType));
+  // return (ths->replicaNum == 1 && syncUtilUserCommit(pMsg->msgType) && ths->vgId != 1);
 }
 
 int32_t syncNodeCommit(SSyncNode* ths, SyncIndex beginIndex, SyncIndex endIndex, uint64_t flag) {
@@ -3084,7 +3085,8 @@ int32_t syncNodeCommit(SSyncNode* ths, SyncIndex beginIndex, SyncIndex endIndex,
         // user commit
         if ((ths->pFsm->FpCommitCb != NULL) && syncUtilUserCommit(pEntry->originalRpcType)) {
           bool internalExecute = true;
-          if ((ths->replicaNum == 1) && ths->restoreFinish && ths->vgId != 1) {
+          if ((ths->replicaNum == 1) && ths->restoreFinish) {
+            // if ((ths->replicaNum == 1) && ths->restoreFinish && ths->vgId != 1) {
             internalExecute = false;
           }
 

From aa906a0d3b46f56cf2f6ea9aef13c1b9c4272a8d Mon Sep 17 00:00:00 2001
From: plum-lihui <huili@taosdata.com>
Date: Tue, 18 Oct 2022 17:39:58 +0800
Subject: [PATCH 31/42] test: modify timestamp for insert success anywhen

---
 tests/script/tsim/parser/like.sim | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/tests/script/tsim/parser/like.sim b/tests/script/tsim/parser/like.sim
index 40dcec4080..5cac026b57 100644
--- a/tests/script/tsim/parser/like.sim
+++ b/tests/script/tsim/parser/like.sim
@@ -17,9 +17,9 @@ sql create table $table1 (ts timestamp, b binary(20))
 sql create table $table2 (ts timestamp, b binary(20))
 
 sql insert into $table1 values(now,    "table_name")
-sql insert into $table1 values(now-1m, "tablexname")
+sql insert into $table1 values(now-3m, "tablexname")
 sql insert into $table1 values(now-2m, "tablexxx")
-sql insert into $table1 values(now-2m, "table")
+sql insert into $table1 values(now-1m, "table")
 
 sql select b from $table1
 if $rows != 4 then

From 0f5e02b0b40aef533fc08fb95d2ee9b1d9e40ca8 Mon Sep 17 00:00:00 2001
From: Ganlin Zhao <ganlinzhao@hotmail.com>
Date: Tue, 18 Oct 2022 17:52:40 +0800
Subject: [PATCH 32/42] fix(query): mode function output one random result
 instead of NULL if there're multiple highest frequency of occurrence.

TD-19619
---
 source/libs/function/src/builtinsimpl.c | 6 ++----
 1 file changed, 2 insertions(+), 4 deletions(-)

diff --git a/source/libs/function/src/builtinsimpl.c b/source/libs/function/src/builtinsimpl.c
index 7077a9b780..6a428af1ed 100644
--- a/source/libs/function/src/builtinsimpl.c
+++ b/source/libs/function/src/builtinsimpl.c
@@ -5356,16 +5356,14 @@ int32_t modeFinalize(SqlFunctionCtx* pCtx, SSDataBlock* pBlock) {
   int32_t maxCount = 0;
   for (int32_t i = 0; i < pInfo->numOfPoints; ++i) {
     SModeItem* pItem = (SModeItem*)(pInfo->pItems + i * (sizeof(SModeItem) + pInfo->colBytes));
-    if (pItem->count > maxCount) {
+    if (pItem->count >= maxCount) {
       maxCount = pItem->count;
       resIndex = i;
-    } else if (pItem->count == maxCount) {
-      resIndex = -1;
     }
   }
 
   SModeItem* pResItem = (SModeItem*)(pInfo->pItems + resIndex * (sizeof(SModeItem) + pInfo->colBytes));
-  colDataAppend(pCol, currentRow, pResItem->data, (resIndex == -1) ? true : false);
+  colDataAppend(pCol, currentRow, pResItem->data, false);
 
   return pResInfo->numOfRes;
 }

From 25c2cec6028effa60312433b6af1b7da766362af Mon Sep 17 00:00:00 2001
From: Ganlin Zhao <ganlinzhao@hotmail.com>
Date: Tue, 18 Oct 2022 17:58:17 +0800
Subject: [PATCH 33/42] change mode en docs

---
 docs/en/12-taos-sql/10-function.md | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/docs/en/12-taos-sql/10-function.md b/docs/en/12-taos-sql/10-function.md
index 2b39c4b9e5..48bc5e743e 100644
--- a/docs/en/12-taos-sql/10-function.md
+++ b/docs/en/12-taos-sql/10-function.md
@@ -945,7 +945,7 @@ MIN(expr)
 MODE(expr)
 ```
 
-**Description**:The value which has the highest frequency of occurrence. NULL is returned if there are multiple values which have highest frequency of occurrence.
+**Description**:The value which has the highest frequency of occurrence. One random value is returned if there are multiple values which have highest frequency of occurrence.
 
 **Return value type**: Same as the input data
 

From 3ec1490fe2050e769f496c05347558d75e1ac3cc Mon Sep 17 00:00:00 2001
From: Ganlin Zhao <ganlinzhao@hotmail.com>
Date: Tue, 18 Oct 2022 18:00:53 +0800
Subject: [PATCH 34/42] change mode function zh docs

---
 docs/zh/12-taos-sql/10-function.md | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/docs/zh/12-taos-sql/10-function.md b/docs/zh/12-taos-sql/10-function.md
index 9346c3f763..e7ed74631d 100644
--- a/docs/zh/12-taos-sql/10-function.md
+++ b/docs/zh/12-taos-sql/10-function.md
@@ -946,7 +946,7 @@ MIN(expr)
 MODE(expr)
 ```
 
-**功能说明**：返回出现频率最高的值，若存在多个频率相同的最高值，输出NULL。
+**功能说明**：返回出现频率最高的值，若存在多个频率相同的最高值，则随机输出其中某个值。
 
 **返回数据类型**：与输入数据类型一致。
 

From 797ed80f707c5ab3558176858e9167c63e45885f Mon Sep 17 00:00:00 2001
From: Shengliang Guan <slguan@taosdata.com>
Date: Tue, 18 Oct 2022 18:08:49 +0800
Subject: [PATCH 35/42] enh: comment out unused codes in tsim

---
 utils/tsim/src/simExe.c   |  8 ++++++++
 utils/tsim/src/simParse.c | 12 ++++++++++++
 2 files changed, 20 insertions(+)

diff --git a/utils/tsim/src/simExe.c b/utils/tsim/src/simExe.c
index 4fe5fdf672..6f693cf015 100644
--- a/utils/tsim/src/simExe.c
+++ b/utils/tsim/src/simExe.c
@@ -37,6 +37,7 @@ void simLogSql(char *sql, bool useSharp) {
   taosFsyncFile(pFile);
 }
 
+#if 0
 char *simParseArbitratorName(char *varName) {
   static char hostName[140];
 #ifdef WINDOWS
@@ -47,6 +48,7 @@ char *simParseArbitratorName(char *varName) {
 #endif
   return hostName;
 }
+#endif
 
 char *simParseHostName(char *varName) {
   static char hostName[140];
@@ -102,9 +104,11 @@ char *simGetVariable(SScript *script, char *varName, int32_t varLen) {
     return simParseHostName(varName);
   }
 
+#if 0
   if (strncmp(varName, "arbitrator", 10) == 0) {
     return simParseArbitratorName(varName);
   }
+#endif
 
   if (strncmp(varName, "error", varLen) == 0) return script->error;
 
@@ -883,6 +887,7 @@ bool simExecuteSqlSlowCmd(SScript *script, char *rest) {
   return simExecuteSqlImpCmd(script, rest, isSlow);
 }
 
+#if 0
 bool simExecuteRestfulCmd(SScript *script, char *rest) {
   TdFilePtr pFile = NULL;
   char      filename[256];
@@ -924,6 +929,7 @@ bool simExecuteRestfulCmd(SScript *script, char *rest) {
 
   return simExecuteSystemCmd(script, cmd);
 }
+#endif
 
 bool simExecuteSqlErrorCmd(SScript *script, char *rest) {
   char      buf[3000];
@@ -981,6 +987,7 @@ bool simExecuteSqlErrorCmd(SScript *script, char *rest) {
   return false;
 }
 
+#if 0
 bool simExecuteLineInsertCmd(SScript *script, char *rest) {
   char buf[TSDB_MAX_BINARY_LEN] = {0};
 
@@ -1037,3 +1044,4 @@ bool simExecuteLineInsertErrorCmd(SScript *script, char *rest) {
     return true;
   }
 }
+#endif
\ No newline at end of file
diff --git a/utils/tsim/src/simParse.c b/utils/tsim/src/simParse.c
index 14be10bc8a..ae523edfc4 100644
--- a/utils/tsim/src/simParse.c
+++ b/utils/tsim/src/simParse.c
@@ -501,6 +501,7 @@ bool simParseEndwCmd(char *rest, SCommand *pCmd, int32_t lineNum) {
   return true;
 }
 
+#if 0
 bool simParseSwitchCmd(char *rest, SCommand *pCmd, int32_t lineNum) {
   char   *token;
   int32_t tokenLen;
@@ -647,6 +648,7 @@ bool simParseContinueCmd(char *rest, SCommand *pCmd, int32_t lineNum) {
   numOfLines++;
   return true;
 }
+#endif
 
 bool simParsePrintCmd(char *rest, SCommand *pCmd, int32_t lineNum) {
   int32_t expLen;
@@ -715,6 +717,7 @@ bool simParseSqlErrorCmd(char *rest, SCommand *pCmd, int32_t lineNum) {
   return true;
 }
 
+#if 0
 bool simParseSqlSlowCmd(char *rest, SCommand *pCmd, int32_t lineNum) {
   simParseSqlCmd(rest, pCmd, lineNum);
   cmdLine[numOfLines - 1].cmdno = SIM_CMD_SQL_SLOW;
@@ -726,6 +729,7 @@ bool simParseRestfulCmd(char *rest, SCommand *pCmd, int32_t lineNum) {
   cmdLine[numOfLines - 1].cmdno = SIM_CMD_RESTFUL;
   return true;
 }
+#endif
 
 bool simParseSystemCmd(char *rest, SCommand *pCmd, int32_t lineNum) {
   int32_t expLen;
@@ -838,6 +842,7 @@ bool simParseRunBackCmd(char *rest, SCommand *pCmd, int32_t lineNum) {
   return true;
 }
 
+#if 0
 bool simParseLineInsertCmd(char *rest, SCommand *pCmd, int32_t lineNum) {
   int32_t expLen;
 
@@ -869,6 +874,7 @@ bool simParseLineInsertErrorCmd(char *rest, SCommand *pCmd, int32_t lineNum) {
   numOfLines++;
   return true;
 }
+#endif
 
 void simInitsimCmdList() {
   int32_t cmdno;
@@ -930,6 +936,7 @@ void simInitsimCmdList() {
   simCmdList[cmdno].executeCmd = NULL;
   simAddCmdIntoHash(&(simCmdList[cmdno]));
 
+#if 0
   cmdno = SIM_CMD_SWITCH;
   simCmdList[cmdno].cmdno = cmdno;
   strcpy(simCmdList[cmdno].name, "switch");
@@ -977,6 +984,7 @@ void simInitsimCmdList() {
   simCmdList[cmdno].parseCmd = simParseEndsCmd;
   simCmdList[cmdno].executeCmd = NULL;
   simAddCmdIntoHash(&(simCmdList[cmdno]));
+#endif
 
   cmdno = SIM_CMD_SLEEP;
   simCmdList[cmdno].cmdno = cmdno;
@@ -1050,6 +1058,7 @@ void simInitsimCmdList() {
   simCmdList[cmdno].executeCmd = simExecuteSqlErrorCmd;
   simAddCmdIntoHash(&(simCmdList[cmdno]));
 
+#if 0
   cmdno = SIM_CMD_SQL_SLOW;
   simCmdList[cmdno].cmdno = cmdno;
   strcpy(simCmdList[cmdno].name, "sql_slow");
@@ -1065,6 +1074,7 @@ void simInitsimCmdList() {
   simCmdList[cmdno].parseCmd = simParseRestfulCmd;
   simCmdList[cmdno].executeCmd = simExecuteRestfulCmd;
   simAddCmdIntoHash(&(simCmdList[cmdno]));
+#endif
 
   /* test is only an internal command */
   cmdno = SIM_CMD_TEST;
@@ -1082,6 +1092,7 @@ void simInitsimCmdList() {
   simCmdList[cmdno].executeCmd = simExecuteReturnCmd;
   simAddCmdIntoHash(&(simCmdList[cmdno]));
 
+#if 0
   cmdno = SIM_CMD_LINE_INSERT;
   simCmdList[cmdno].cmdno = cmdno;
   strcpy(simCmdList[cmdno].name, "line_insert");
@@ -1097,4 +1108,5 @@ void simInitsimCmdList() {
   simCmdList[cmdno].parseCmd = simParseLineInsertErrorCmd;
   simCmdList[cmdno].executeCmd = simExecuteLineInsertErrorCmd;
   simAddCmdIntoHash(&(simCmdList[cmdno]));
+#endif
 }

From abbfa0b8e57c923ba95411ce1773d8695f635d23 Mon Sep 17 00:00:00 2001
From: afwerar <1296468573@qq.com>
Date: Tue, 18 Oct 2022 18:13:00 +0800
Subject: [PATCH 36/42] docs: update mac docs

---
 docs/en/14-reference/03-connector/_macos_install.mdx | 2 +-
 docs/zh/08-connector/_macos_install.mdx              | 2 +-
 2 files changed, 2 insertions(+), 2 deletions(-)

diff --git a/docs/en/14-reference/03-connector/_macos_install.mdx b/docs/en/14-reference/03-connector/_macos_install.mdx
index 8813d65869..effabbbebe 100644
--- a/docs/en/14-reference/03-connector/_macos_install.mdx
+++ b/docs/en/14-reference/03-connector/_macos_install.mdx
@@ -2,7 +2,7 @@ import PkgListV3 from "/components/PkgListV3";
 
 1. Download the client installation package
 
-  <PkgListV3 type={8} sys="MacOS" />
+  <PkgListV3 type={8} sys="macOS" />
 
    [All Downloads](../../releases/tdengine)
 
diff --git a/docs/zh/08-connector/_macos_install.mdx b/docs/zh/08-connector/_macos_install.mdx
index 4925ffabcd..081134dfee 100644
--- a/docs/zh/08-connector/_macos_install.mdx
+++ b/docs/zh/08-connector/_macos_install.mdx
@@ -2,7 +2,7 @@ import PkgListV3 from "/components/PkgListV3";
 
 1. 下载客户端安装包
 
-  <PkgListV3 type={8} sys="MacOS" />
+   <PkgListV3 type={8} sys="macOS" />
 
    [所有下载](../../releases/tdengine)
 

From 8efd7e8763727d77d76c154fc9d66ad38b87d856 Mon Sep 17 00:00:00 2001
From: 54liuyao <54liuyao@163.com>
Date: Tue, 18 Oct 2022 17:44:00 +0800
Subject: [PATCH 37/42] feat(stream):replace disk buff

---
 include/common/tcommon.h                      |   11 +-
 include/libs/stream/streamState.h             |   16 +
 source/common/src/tdatablock.c                |    7 +-
 source/libs/executor/inc/executorimpl.h       |   91 +-
 source/libs/executor/src/executorimpl.c       |  116 +-
 source/libs/executor/src/scanoperator.c       |   19 +-
 source/libs/executor/src/tfill.c              |    4 +-
 source/libs/executor/src/timewindowoperator.c | 1564 +++++++----------
 source/libs/function/src/builtinsimpl.c       |   11 +-
 source/libs/stream/src/streamState.c          |  297 +++-
 tests/script/jenkins/basic.txt                |    8 +-
 tests/script/tsim/stream/session0.sim         |    4 +-
 tests/script/tsim/stream/session1.sim         |    6 +-
 tests/script/tsim/stream/state0.sim           |    2 +-
 tests/script/tsim/stream/triggerSession0.sim  |    9 +-
 15 files changed, 1074 insertions(+), 1091 deletions(-)

diff --git a/include/common/tcommon.h b/include/common/tcommon.h
index eaddf4e983..bb2519bea3 100644
--- a/include/common/tcommon.h
+++ b/include/common/tcommon.h
@@ -44,12 +44,17 @@ enum {
 )
 // clang-format on
 
-typedef struct {
+typedef struct SWinKey {
   uint64_t groupId;
   TSKEY    ts;
 } SWinKey;
 
-static inline int sWinKeyCmprImpl(const void* pKey1, const void* pKey2) {
+typedef struct SSessionKey {
+  STimeWindow win;
+  uint64_t    groupId;
+} SSessionKey;
+
+static inline int winKeyCmprImpl(const void* pKey1, const void* pKey2) {
   SWinKey* pWin1 = (SWinKey*)pKey1;
   SWinKey* pWin2 = (SWinKey*)pKey2;
 
@@ -69,7 +74,7 @@ static inline int sWinKeyCmprImpl(const void* pKey1, const void* pKey2) {
 }
 
 static inline int winKeyCmpr(const void* pKey1, int kLen1, const void* pKey2, int kLen2) {
-  return sWinKeyCmprImpl(pKey1, pKey2);
+  return winKeyCmprImpl(pKey1, pKey2);
 }
 
 typedef struct {
diff --git a/include/libs/stream/streamState.h b/include/libs/stream/streamState.h
index 9445a0baa5..ec68e8ed71 100644
--- a/include/libs/stream/streamState.h
+++ b/include/libs/stream/streamState.h
@@ -25,6 +25,8 @@ extern "C" {
 
 typedef struct SStreamTask SStreamTask;
 
+typedef bool (*state_key_cmpr_fn)(void* pKey1, void* pKey2);
+
 // incremental state storage
 typedef struct {
   SStreamTask* pOwner;
@@ -32,6 +34,7 @@ typedef struct {
   TTB*         pStateDb;
   TTB*         pFuncStateDb;
   TTB*         pFillStateDb;  // todo refactor
+  TTB*         pSessionStateDb;
   TXN          txn;
   int32_t      number;
 } SStreamState;
@@ -57,6 +60,19 @@ int32_t streamStateDel(SStreamState* pState, const SWinKey* key);
 int32_t streamStateClear(SStreamState* pState);
 void    streamStateSetNumber(SStreamState* pState, int32_t number);
 
+int32_t streamStateSessionAddIfNotExist(SStreamState* pState, SSessionKey* key, void** pVal, int32_t* pVLen);
+int32_t streamStateSessionPut(SStreamState* pState, const SSessionKey* key, const void* value, int32_t vLen);
+int32_t streamStateSessionGet(SStreamState* pState, SSessionKey* key, void** pVal, int32_t* pVLen);
+int32_t streamStateSessionDel(SStreamState* pState, const SSessionKey* key);
+int32_t streamStateSessionClear(SStreamState* pState);
+int32_t streamStateSessionGetKVByCur(SStreamStateCur* pCur, SSessionKey* pKey, const void** pVal, int32_t* pVLen);
+int32_t streamStateStateAddIfNotExist(SStreamState* pState, SSessionKey* key, char* pKeyData, int32_t keyDataLen,
+                                      state_key_cmpr_fn fn, void** pVal, int32_t* pVLen);
+
+SStreamStateCur* streamStateSessionSeekKeyNext(SStreamState* pState, const SSessionKey* key);
+SStreamStateCur* streamStateSessionSeekKeyPrev(SStreamState* pState, const SSessionKey* key);
+SStreamStateCur* streamStateSessionGetCur(SStreamState* pState, const SSessionKey* key);
+
 int32_t streamStateFillPut(SStreamState* pState, const SWinKey* key, const void* value, int32_t vLen);
 int32_t streamStateFillGet(SStreamState* pState, const SWinKey* key, void** pVal, int32_t* pVLen);
 int32_t streamStateFillDel(SStreamState* pState, const SWinKey* key);
diff --git a/source/common/src/tdatablock.c b/source/common/src/tdatablock.c
index 1028a899b6..62706db980 100644
--- a/source/common/src/tdatablock.c
+++ b/source/common/src/tdatablock.c
@@ -1892,12 +1892,13 @@ char* dumpBlockData(SSDataBlock* pDataBlock, const char* flag, char** pDataBuf)
 
     for (int32_t k = 0; k < colNum; k++) {
       SColumnInfoData* pColInfoData = taosArrayGet(pDataBlock->pDataBlock, k);
-      void*            var = POINTER_SHIFT(pColInfoData->pData, j * pColInfoData->info.bytes);
       if (colDataIsNull(pColInfoData, rows, j, NULL) || !pColInfoData->pData) {
         len += snprintf(dumpBuf + len, size - len, " %15s |", "NULL");
         if (len >= size - 1) return dumpBuf;
         continue;
       }
+
+      void* var = colDataGetData(pColInfoData, j);
       switch (pColInfoData->info.type) {
         case TSDB_DATA_TYPE_TIMESTAMP:
           memset(pBuf, 0, sizeof(pBuf));
@@ -1926,8 +1927,8 @@ char* dumpBlockData(SSDataBlock* pDataBlock, const char* flag, char** pDataBuf)
           if (len >= size - 1) return dumpBuf;
           break;
         case TSDB_DATA_TYPE_DOUBLE:
-          len += snprintf(dumpBuf + len, size - len, " %15lf |", *(double*)var);
-          if (len >= size - 1) return dumpBuf;
+          // len += snprintf(dumpBuf + len, size - len, " %15lf |", *(double*)var);
+          // if (len >= size - 1) return dumpBuf;
           break;
         case TSDB_DATA_TYPE_BOOL:
           len += snprintf(dumpBuf + len, size - len, " %15d |", *(bool*)var);
diff --git a/source/libs/executor/inc/executorimpl.h b/source/libs/executor/inc/executorimpl.h
index fc996a6003..47450328be 100644
--- a/source/libs/executor/inc/executorimpl.h
+++ b/source/libs/executor/inc/executorimpl.h
@@ -53,6 +53,11 @@ typedef int32_t (*__block_search_fn_t)(char* data, int32_t num, int64_t key, int
 
 #define NEEDTO_COMPRESS_QUERY(size) ((size) > tsCompressColData ? 1 : 0)
 
+#define IS_VALID_SESSION_WIN(winInfo)        ((winInfo).sessionWin.win.skey > 0)
+#define SET_SESSION_WIN_INVALID(winInfo)     ((winInfo).sessionWin.win.skey = INT64_MIN)
+#define IS_INVALID_SESSION_WIN_KEY(winKey)   ((winKey).win.skey <= 0)
+#define SET_SESSION_WIN_KEY_INVALID(pWinKey) ((pWinKey)->win.skey = INT64_MIN)
+
 enum {
   // when this task starts to execute, this status will set
   TASK_NOT_COMPLETED = 0x1u,
@@ -434,15 +439,15 @@ typedef struct SCatchSupporter {
 } SCatchSupporter;
 
 typedef struct SStreamAggSupporter {
-  SHashObj*      pResultRows;
-  SArray*        pCurWins;
-  int32_t        valueSize;
-  int32_t        keySize;
-  char*          pKeyBuf;        // window key buffer
-  SDiskbasedBuf* pResultBuf;     // query result buffer based on blocked-wised disk file
-  int32_t        resultRowSize;  // the result buffer size for each result row, with the meta data size for each row
-  int32_t        currentPageId;  // buffer page that is active
-  SSDataBlock*   pScanBlock;
+  int32_t         resultRowSize;  // the result buffer size for each result row, with the meta data size for each row
+  SSDataBlock*    pScanBlock;
+  SStreamState*   pState;
+  int64_t         gap;            // stream session window gap
+  SqlFunctionCtx* pDummyCtx;      // for combine
+  SSHashObj*      pResultRows;
+  int32_t         stateKeySize;
+  int16_t         stateKeyType;
+  SDiskbasedBuf*  pResultBuf;
 } SStreamAggSupporter;
 
 typedef struct SWindowSupporter {
@@ -736,42 +741,54 @@ typedef struct SSessionAggOperatorInfo {
 } SSessionAggOperatorInfo;
 
 typedef struct SResultWindowInfo {
-  SResultRowPosition pos;
-  STimeWindow        win;
-  uint64_t           groupId;
+  void*              pOutputBuf;
+  SSessionKey        sessionWin;
   bool               isOutput;
-  bool               isClosed;
 } SResultWindowInfo;
 
 typedef struct SStateWindowInfo {
   SResultWindowInfo winInfo;
-  SStateKeys        stateKey;
+  SStateKeys*       pStateKey;
 } SStateWindowInfo;
 
 typedef struct SStreamSessionAggOperatorInfo {
   SOptrBasicInfo      binfo;
   SStreamAggSupporter streamAggSup;
-  SExprSupp           scalarSupp;  // supporter for perform scalar function
+  SExprSupp           scalarSupp;      // supporter for perform scalar function
   SGroupResInfo       groupResInfo;
-  int64_t             gap;             // session window gap
   int32_t             primaryTsIndex;  // primary timestamp slot id
   int32_t             endTsIndex;      // window end timestamp slot id
   int32_t             order;           // current SSDataBlock scan order
   STimeWindowAggSupp  twAggSup;
-  SSDataBlock*        pWinBlock;   // window result
-  SqlFunctionCtx*     pDummyCtx;   // for combine
-  SSDataBlock*        pDelRes;     // delete result
-  SSDataBlock*        pUpdateRes;  // update window
+  SSDataBlock*        pWinBlock;       // window result
+  SSDataBlock*        pDelRes;         // delete result
+  SSDataBlock*        pUpdateRes;      // update window
   bool                returnUpdate;
-  SHashObj*           pStDeleted;
+  SSHashObj*          pStDeleted;
   void*               pDelIterator;
-  SArray*             pChildren;  // cache for children's result; final stream operator
-  SPhysiNode*         pPhyNode;   // create new child
+  SArray*             pChildren;       // cache for children's result; final stream operator
+  SPhysiNode*         pPhyNode;        // create new child
   bool                isFinal;
   bool                ignoreExpiredData;
   SHashObj*           pGroupIdTbNameMap;
 } SStreamSessionAggOperatorInfo;
 
+typedef struct SStreamStateAggOperatorInfo {
+  SOptrBasicInfo      binfo;
+  SStreamAggSupporter streamAggSup;
+  SExprSupp           scalarSupp;      // supporter for perform scalar function
+  SGroupResInfo       groupResInfo;
+  int32_t             primaryTsIndex;  // primary timestamp slot id
+  STimeWindowAggSupp  twAggSup;
+  SColumn             stateCol;
+  SSDataBlock*        pDelRes;
+  SSHashObj*          pSeDeleted;
+  void*               pDelIterator;
+  SArray*             pChildren;       // cache for children's result;
+  bool                ignoreExpiredData;
+  SHashObj*           pGroupIdTbNameMap;
+} SStreamStateAggOperatorInfo;
+
 typedef struct SStreamPartitionOperatorInfo {
   SOptrBasicInfo        binfo;
   SPartitionBySupporter partitionSup;
@@ -834,24 +851,6 @@ typedef struct SStateWindowOperatorInfo {
   const SNode* pCondition;
 } SStateWindowOperatorInfo;
 
-typedef struct SStreamStateAggOperatorInfo {
-  SOptrBasicInfo      binfo;
-  SStreamAggSupporter streamAggSup;
-  SExprSupp           scalarSupp;  // supporter for perform scalar function
-  SGroupResInfo       groupResInfo;
-  int32_t             primaryTsIndex;  // primary timestamp slot id
-  int32_t             order;           // current SSDataBlock scan order
-  STimeWindowAggSupp  twAggSup;
-  SColumn             stateCol;
-  SqlFunctionCtx*     pDummyCtx;  // for combine
-  SSDataBlock*        pDelRes;
-  SHashObj*           pSeDeleted;
-  void*               pDelIterator;
-  SArray*             pChildren;  // cache for children's result;
-  bool                ignoreExpiredData;
-  SHashObj*           pGroupIdTbNameMap;
-} SStreamStateAggOperatorInfo;
-
 typedef struct SSortOperatorInfo {
   SOptrBasicInfo binfo;
   uint32_t       sortBufSize;  // max buffer size for in-memory sort
@@ -1064,13 +1063,8 @@ STimeWindow getActiveTimeWindow(SDiskbasedBuf* pBuf, SResultRowInfo* pResultRowI
 int32_t getNumOfRowsInTimeWindow(SDataBlockInfo* pDataBlockInfo, TSKEY* pPrimaryColumn, int32_t startPos, TSKEY ekey,
                                  __block_search_fn_t searchFn, STableQueryInfo* item, int32_t order);
 int32_t binarySearchForKey(char* pValue, int num, TSKEY key, int order);
-int32_t initStreamAggSupporter(SStreamAggSupporter* pSup, const char* pKey, SqlFunctionCtx* pCtx, int32_t numOfOutput,
-                               int32_t size);
 SResultRow*        getNewResultRow(SDiskbasedBuf* pResultBuf, int32_t* currentPageId, int32_t interBufSize);
-SResultWindowInfo* getSessionTimeWindow(SStreamAggSupporter* pAggSup, TSKEY startTs, TSKEY endTs, uint64_t groupId,
-                                        int64_t gap, int32_t* pIndex);
-SResultWindowInfo* getCurSessionWindow(SStreamAggSupporter* pAggSup, TSKEY startTs, TSKEY endTs, uint64_t groupId,
-                                       int64_t gap, int32_t* pIndex);
+void getCurSessionWindow(SStreamAggSupporter* pAggSup, TSKEY startTs, TSKEY endTs, uint64_t groupId, SSessionKey* pKey);
 bool               isInTimeWindow(STimeWindow* pWin, TSKEY ts, int64_t gap);
 bool               functionNeedToExecute(SqlFunctionCtx* pCtx);
 bool               isOverdue(TSKEY ts, STimeWindowAggSupp* pSup);
@@ -1100,6 +1094,9 @@ int32_t generateGroupIdMap(STableListInfo* pTableListInfo, SReadHandle* pHandle,
 void*   destroySqlFunctionCtx(SqlFunctionCtx* pCtx, int32_t numOfOutput);
 int32_t buildDataBlockFromGroupRes(SOperatorInfo* pOperator, SStreamState* pState, SSDataBlock* pBlock, SExprSupp* pSup,
                                    SGroupResInfo* pGroupResInfo);
+int32_t saveSessionDiscBuf(SStreamState* pState, SSessionKey* key, void* buf, int32_t size);
+int32_t buildSessionResultDataBlock(SExecTaskInfo* pTaskInfo, SStreamState* pState, SSDataBlock* pBlock,
+                                    SExprSupp* pSup, SGroupResInfo* pGroupResInfo);
 int32_t setOutputBuf(SStreamState* pState, STimeWindow* win, SResultRow** pResult, int64_t tableGroupId, SqlFunctionCtx* pCtx,
                      int32_t numOfOutput, int32_t* rowEntryInfoOffset, SAggSupporter* pAggSup);
 int32_t releaseOutputBuf(SStreamState* pState, SWinKey* pKey, SResultRow* pResult);
diff --git a/source/libs/executor/src/executorimpl.c b/source/libs/executor/src/executorimpl.c
index a7e98527f4..641bcb0c5b 100644
--- a/source/libs/executor/src/executorimpl.c
+++ b/source/libs/executor/src/executorimpl.c
@@ -4192,42 +4192,6 @@ int32_t getOperatorExplainExecInfo(SOperatorInfo* operatorInfo, SArray* pExecInf
   return TSDB_CODE_SUCCESS;
 }
 
-int32_t initStreamAggSupporter(SStreamAggSupporter* pSup, const char* pKey, SqlFunctionCtx* pCtx, int32_t numOfOutput,
-                               int32_t size) {
-  pSup->currentPageId = -1;
-  pSup->resultRowSize = getResultRowSize(pCtx, numOfOutput);
-  pSup->keySize = sizeof(int64_t) + sizeof(TSKEY);
-  pSup->pKeyBuf = taosMemoryCalloc(1, pSup->keySize);
-  _hash_fn_t hashFn = taosGetDefaultHashFunction(TSDB_DATA_TYPE_BINARY);
-  pSup->pResultRows = taosHashInit(1024, hashFn, false, HASH_NO_LOCK);
-  if (pSup->pKeyBuf == NULL || pSup->pResultRows == NULL) {
-    return TSDB_CODE_OUT_OF_MEMORY;
-  }
-  pSup->valueSize = size;
-
-  pSup->pScanBlock = createSpecialDataBlock(STREAM_CLEAR);
-  int32_t pageSize = 4096;
-  while (pageSize < pSup->resultRowSize * 4) {
-    pageSize <<= 1u;
-  }
-  // at least four pages need to be in buffer
-  int32_t bufSize = 4096 * 256;
-  if (bufSize <= pageSize) {
-    bufSize = pageSize * 4;
-  }
-  if (!osTempSpaceAvailable()) {
-    terrno = TSDB_CODE_NO_AVAIL_DISK;
-    qError("Init stream agg supporter failed since %s", terrstr(terrno));
-    return terrno;
-  }
-  int32_t code = createDiskbasedBuf(&pSup->pResultBuf, pageSize, bufSize, pKey, tsTempDir);
-  for (int32_t i = 0; i < numOfOutput; ++i) {
-    pCtx[i].saveHandle.pBuf = pSup->pResultBuf;
-  }
-
-  return code;
-}
-
 int32_t setOutputBuf(SStreamState* pState, STimeWindow* win, SResultRow** pResult, int64_t tableGroupId,
                      SqlFunctionCtx* pCtx, int32_t numOfOutput, int32_t* rowEntryInfoOffset, SAggSupporter* pAggSup) {
   SWinKey key = {
@@ -4237,7 +4201,6 @@ int32_t setOutputBuf(SStreamState* pState, STimeWindow* win, SResultRow** pResul
   char*   value = NULL;
   int32_t size = pAggSup->resultRowSize;
 
-  tSimpleHashPut(pAggSup->pResultRowHashTable, &key, sizeof(SWinKey), NULL, 0);
   if (streamStateAddIfNotExist(pState, &key, (void**)&value, &size) < 0) {
     return TSDB_CODE_QRY_OUT_OF_MEMORY;
   }
@@ -4342,3 +4305,82 @@ int32_t buildDataBlockFromGroupRes(SOperatorInfo* pOperator, SStreamState* pStat
   blockDataUpdateTsWindow(pBlock, 0);
   return TSDB_CODE_SUCCESS;
 }
+
+int32_t saveSessionDiscBuf(SStreamState* pState, SSessionKey* key, void* buf, int32_t size) {
+  streamStateSessionPut(pState, key, (const void*)buf, size);
+  releaseOutputBuf(pState, NULL, (SResultRow*)buf);
+  return TSDB_CODE_SUCCESS;
+}
+
+int32_t buildSessionResultDataBlock(SExecTaskInfo* pTaskInfo, SStreamState* pState, SSDataBlock* pBlock,
+                                    SExprSupp* pSup, SGroupResInfo* pGroupResInfo) {
+  SExprInfo*      pExprInfo = pSup->pExprInfo;
+  int32_t         numOfExprs = pSup->numOfExprs;
+  int32_t*        rowEntryOffset = pSup->rowEntryInfoOffset;
+  SqlFunctionCtx* pCtx = pSup->pCtx;
+
+  int32_t numOfRows = getNumOfTotalRes(pGroupResInfo);
+
+  for (int32_t i = pGroupResInfo->index; i < numOfRows; i += 1) {
+    SSessionKey* pKey = taosArrayGet(pGroupResInfo->pRows, i);
+    int32_t      size = 0;
+    void*        pVal = NULL;
+    int32_t      code = streamStateSessionGet(pState, pKey, &pVal, &size);
+    ASSERT(code == 0);
+    SResultRow* pRow = (SResultRow*)pVal;
+    doUpdateNumOfRows(pCtx, pRow, numOfExprs, rowEntryOffset);
+    // no results, continue to check the next one
+    if (pRow->numOfRows == 0) {
+      pGroupResInfo->index += 1;
+      releaseOutputBuf(pState, NULL, pRow);
+      continue;
+    }
+
+    if (pBlock->info.groupId == 0) {
+      pBlock->info.groupId = pKey->groupId;
+    } else {
+      // current value belongs to different group, it can't be packed into one datablock
+      if (pBlock->info.groupId != pKey->groupId) {
+        releaseOutputBuf(pState, NULL, pRow);
+        break;
+      }
+    }
+
+    if (pBlock->info.rows + pRow->numOfRows > pBlock->info.capacity) {
+      ASSERT(pBlock->info.rows > 0);
+      releaseOutputBuf(pState, NULL, pRow);
+      break;
+    }
+
+    pGroupResInfo->index += 1;
+
+    for (int32_t j = 0; j < numOfExprs; ++j) {
+      int32_t slotId = pExprInfo[j].base.resSchema.slotId;
+
+      pCtx[j].resultInfo = getResultEntryInfo(pRow, j, rowEntryOffset);
+      if (pCtx[j].fpSet.finalize) {
+        int32_t code1 = pCtx[j].fpSet.finalize(&pCtx[j], pBlock);
+        if (TAOS_FAILED(code1)) {
+          qError("%s build result data block error, code %s", GET_TASKID(pTaskInfo), tstrerror(code1));
+          T_LONG_JMP(pTaskInfo->env, code1);
+        }
+      } else if (strcmp(pCtx[j].pExpr->pExpr->_function.functionName, "_select_value") == 0) {
+        // do nothing, todo refactor
+      } else {
+        // expand the result into multiple rows. E.g., _wstart, top(k, 20)
+        // the _wstart needs to copy to 20 following rows, since the results of top-k expands to 20 different rows.
+        SColumnInfoData* pColInfoData = taosArrayGet(pBlock->pDataBlock, slotId);
+        char*            in = GET_ROWCELL_INTERBUF(pCtx[j].resultInfo);
+        for (int32_t k = 0; k < pRow->numOfRows; ++k) {
+          colDataAppend(pColInfoData, pBlock->info.rows + k, in, pCtx[j].resultInfo->isNullRes);
+        }
+      }
+    }
+
+    pBlock->info.rows += pRow->numOfRows;
+    // saveSessionDiscBuf(pState, pKey, pVal, size);
+    releaseOutputBuf(pState, NULL, pRow);
+  }
+  blockDataUpdateTsWindow(pBlock, 0);
+  return TSDB_CODE_SUCCESS;
+}
\ No newline at end of file
diff --git a/source/libs/executor/src/scanoperator.c b/source/libs/executor/src/scanoperator.c
index 8e1858d8c0..7385e9bb92 100644
--- a/source/libs/executor/src/scanoperator.c
+++ b/source/libs/executor/src/scanoperator.c
@@ -1190,23 +1190,22 @@ static int32_t generateSessionScanRange(SStreamScanInfo* pInfo, SSDataBlock* pSr
   SColumnInfoData* pDestGpCol = taosArrayGet(pDestBlock->pDataBlock, GROUPID_COLUMN_INDEX);
   SColumnInfoData* pDestCalStartTsCol = taosArrayGet(pDestBlock->pDataBlock, CALCULATE_START_TS_COLUMN_INDEX);
   SColumnInfoData* pDestCalEndTsCol = taosArrayGet(pDestBlock->pDataBlock, CALCULATE_END_TS_COLUMN_INDEX);
-  int32_t          dummy = 0;
   int64_t          version = pSrcBlock->info.version - 1;
   for (int32_t i = 0; i < pSrcBlock->info.rows; i++) {
     uint64_t groupId = getGroupIdByData(pInfo, uidCol[i], startData[i], version);
     // gap must be 0.
-    SResultWindowInfo* pStartWin =
-        getCurSessionWindow(pInfo->windowSup.pStreamAggSup, startData[i], endData[i], groupId, 0, &dummy);
-    if (!pStartWin) {
+    SSessionKey startWin = {0};
+    getCurSessionWindow(pInfo->windowSup.pStreamAggSup, startData[i], endData[i], groupId, &startWin);
+    if (IS_INVALID_SESSION_WIN_KEY(startWin)) {
       // window has been closed.
       continue;
     }
-    SResultWindowInfo* pEndWin =
-        getCurSessionWindow(pInfo->windowSup.pStreamAggSup, endData[i], endData[i], groupId, 0, &dummy);
-    ASSERT(pEndWin);
-    TSKEY ts = INT64_MIN;
-    colDataAppend(pDestStartCol, i, (const char*)&pStartWin->win.skey, false);
-    colDataAppend(pDestEndCol, i, (const char*)&pEndWin->win.ekey, false);
+    SSessionKey endWin = {0};
+    getCurSessionWindow(pInfo->windowSup.pStreamAggSup, endData[i], endData[i], groupId, &endWin);
+    ASSERT(!IS_INVALID_SESSION_WIN_KEY(endWin));
+    colDataAppend(pDestStartCol, i, (const char*)&startWin.win.skey, false);
+    colDataAppend(pDestEndCol, i, (const char*)&endWin.win.ekey, false);
+
     colDataAppendNULL(pDestUidCol, i);
     colDataAppend(pDestGpCol, i, (const char*)&groupId, false);
     colDataAppendNULL(pDestCalStartTsCol, i);
diff --git a/source/libs/executor/src/tfill.c b/source/libs/executor/src/tfill.c
index 8b716531e5..e199cf8406 100644
--- a/source/libs/executor/src/tfill.c
+++ b/source/libs/executor/src/tfill.c
@@ -693,7 +693,7 @@ void* destroyStreamFillSupporter(SStreamFillSupporter* pFillSup) {
   pFillSup->pAllColInfo = destroyFillColumnInfo(pFillSup->pAllColInfo, pFillSup->numOfFillCols, pFillSup->numOfAllCols);
   tSimpleHashCleanup(pFillSup->pResMap);
   pFillSup->pResMap = NULL;
-  streamStateReleaseBuf(NULL, NULL, pFillSup->cur.pRowVal);
+  releaseOutputBuf(NULL, NULL, (SResultRow*)pFillSup->cur.pRowVal);
   pFillSup->cur.pRowVal = NULL;
 
   taosMemoryFree(pFillSup);
@@ -736,7 +736,7 @@ static void resetFillWindow(SResultRowData* pRowData) {
 
 void resetPrevAndNextWindow(SStreamFillSupporter* pFillSup, SStreamState* pState) {
   resetFillWindow(&pFillSup->prev);
-  streamStateReleaseBuf(NULL, NULL, pFillSup->cur.pRowVal);
+  releaseOutputBuf(NULL, NULL, (SResultRow*)pFillSup->cur.pRowVal);
   resetFillWindow(&pFillSup->cur);
   resetFillWindow(&pFillSup->next);
   resetFillWindow(&pFillSup->nextNext);
diff --git a/source/libs/executor/src/timewindowoperator.c b/source/libs/executor/src/timewindowoperator.c
index 1f2b394b9c..76fddc3982 100644
--- a/source/libs/executor/src/timewindowoperator.c
+++ b/source/libs/executor/src/timewindowoperator.c
@@ -38,8 +38,6 @@ typedef struct SOpenWindowInfo {
   uint64_t           groupId;
 } SOpenWindowInfo;
 
-static SSDataBlock* doStreamSessionAgg(SOperatorInfo* pOperator);
-
 static int64_t* extractTsCol(SSDataBlock* pBlock, const SIntervalAggOperatorInfo* pInfo);
 
 static SResultRowPosition addToOpenWindowList(SResultRowInfo* pResultRowInfo, const SResultRow* pResult,
@@ -828,31 +826,9 @@ int32_t compareResKey(void* pKey, void* data, int32_t index) {
   return -1;
 }
 
-static int32_t saveResult(int64_t ts, int32_t pageId, int32_t offset, uint64_t groupId, SArray* pUpdated) {
-  int32_t size = taosArrayGetSize(pUpdated);
-  SWinKey data = {.ts = ts, .groupId = groupId};
-  int32_t index = binarySearchCom(pUpdated, size, &data, TSDB_ORDER_DESC, compareResKey);
-  if (index == -1) {
-    index = 0;
-  } else {
-    if (compareResKey(&data, pUpdated, index) > 0) {
-      index++;
-    } else {
-      return TSDB_CODE_SUCCESS;
-    }
-  }
-
-  SResKeyPos* newPos = taosMemoryMalloc(sizeof(SResKeyPos) + sizeof(uint64_t));
-  if (newPos == NULL) {
-    return TSDB_CODE_OUT_OF_MEMORY;
-  }
-  newPos->groupId = groupId;
-  newPos->pos = (SResultRowPosition){.pageId = pageId, .offset = offset};
-  *(int64_t*)newPos->key = ts;
-  if (taosArrayInsert(pUpdated, index, &newPos) == NULL) {
-    return TSDB_CODE_OUT_OF_MEMORY;
-  }
-  return TSDB_CODE_SUCCESS;
+static int32_t saveResult(SResultWindowInfo winInfo, SSHashObj* pStUpdated) {
+  winInfo.sessionWin.win.ekey = winInfo.sessionWin.win.skey;
+  return tSimpleHashPut(pStUpdated, &winInfo.sessionWin, sizeof(SSessionKey), &winInfo, sizeof(SResultWindowInfo));
 }
 
 static int32_t saveWinResult(int64_t ts, int32_t pageId, int32_t offset, uint64_t groupId, SHashObj* pUpdatedMap) {
@@ -874,10 +850,6 @@ static int32_t saveWinResultInfo(TSKEY ts, uint64_t groupId, SHashObj* pUpdatedM
   return saveWinResult(ts, -1, -1, groupId, pUpdatedMap);
 }
 
-static int32_t saveResultRow(SResultRow* result, uint64_t groupId, SArray* pUpdated) {
-  return saveResult(result->win.skey, result->pageId, result->offset, groupId, pUpdated);
-}
-
 static void removeResults(SArray* pWins, SHashObj* pUpdatedMap) {
   int32_t size = taosArrayGetSize(pWins);
   for (int32_t i = 0; i < size; i++) {
@@ -915,8 +887,8 @@ int32_t compareWinRes(void* pKey, void* data, int32_t index) {
 }
 
 static void removeDeleteResults(SHashObj* pUpdatedMap, SArray* pDelWins) {
-  taosArraySort(pDelWins, sWinKeyCmprImpl);
-  taosArrayRemoveDuplicate(pDelWins, sWinKeyCmprImpl, NULL);
+  taosArraySort(pDelWins, winKeyCmprImpl);
+  taosArrayRemoveDuplicate(pDelWins, winKeyCmprImpl, NULL);
   int32_t delSize = taosArrayGetSize(pDelWins);
   if (taosHashGetSize(pUpdatedMap) == 0 || delSize == 0) {
     return;
@@ -1385,19 +1357,6 @@ static void doClearWindowImpl(SResultRowPosition* p1, SDiskbasedBuf* pResultBuf,
   releaseBufPage(pResultBuf, bufPage);
 }
 
-static bool doClearWindow(SAggSupporter* pAggSup, SExprSupp* pSup, char* pData, int16_t bytes, uint64_t groupId,
-                          int32_t numOfOutput) {
-  SET_RES_WINDOW_KEY(pAggSup->keyBuf, pData, bytes, groupId);
-  SResultRowPosition* p1 =
-      (SResultRowPosition*)tSimpleHashGet(pAggSup->pResultRowHashTable, pAggSup->keyBuf, GET_RES_WINDOW_KEY_LEN(bytes));
-  if (!p1) {
-    // window has been closed
-    return false;
-  }
-  doClearWindowImpl(p1, pAggSup->pResultBuf, pSup, numOfOutput);
-  return true;
-}
-
 static bool doDeleteWindow(SOperatorInfo* pOperator, TSKEY ts, uint64_t groupId) {
   SStreamIntervalOperatorInfo* pInfo = pOperator->info;
   SWinKey                      key = {.ts = ts, .groupId = groupId};
@@ -1455,73 +1414,6 @@ static void doDeleteWindows(SOperatorInfo* pOperator, SInterval* pInterval, SSDa
   }
 }
 
-bool doDeleteIntervalWindow(SAggSupporter* pAggSup, TSKEY ts, uint64_t groupId) {
-  size_t bytes = sizeof(TSKEY);
-  SET_RES_WINDOW_KEY(pAggSup->keyBuf, &ts, bytes, groupId);
-  SResultRowPosition* p1 =
-      (SResultRowPosition*)tSimpleHashGet(pAggSup->pResultRowHashTable, pAggSup->keyBuf, GET_RES_WINDOW_KEY_LEN(bytes));
-  if (!p1) {
-    // window has been closed
-    return false;
-  }
-  tSimpleHashRemove(pAggSup->pResultRowHashTable, pAggSup->keyBuf, GET_RES_WINDOW_KEY_LEN(bytes));
-  return true;
-}
-
-static void doDeleteSpecifyIntervalWindow(SAggSupporter* pAggSup, STimeWindowAggSupp* pTwSup, SSDataBlock* pBlock,
-                                          SArray* pDelWins, SInterval* pInterval, SHashObj* pUpdatedMap) {
-  SColumnInfoData* pStartCol = taosArrayGet(pBlock->pDataBlock, START_TS_COLUMN_INDEX);
-  TSKEY*           tsStarts = (TSKEY*)pStartCol->pData;
-  SColumnInfoData* pEndCol = taosArrayGet(pBlock->pDataBlock, END_TS_COLUMN_INDEX);
-  TSKEY*           tsEnds = (TSKEY*)pEndCol->pData;
-  SColumnInfoData* pGroupCol = taosArrayGet(pBlock->pDataBlock, GROUPID_COLUMN_INDEX);
-  uint64_t*        groupIds = (uint64_t*)pGroupCol->pData;
-  int64_t          numOfWin = tSimpleHashGetSize(pAggSup->pResultRowHashTable);
-  for (int32_t i = 0; i < pBlock->info.rows; i++) {
-    TSKEY          startTs = TMAX(tsStarts[i], pTwSup->minTs);
-    TSKEY          endTs = TMIN(tsEnds[i], pTwSup->maxTs);
-    SResultRowInfo dumyInfo = {0};
-    dumyInfo.cur.pageId = -1;
-    STimeWindow win = getActiveTimeWindow(NULL, &dumyInfo, startTs, pInterval, TSDB_ORDER_ASC);
-    do {
-      doDeleteIntervalWindow(pAggSup, win.skey, groupIds[i]);
-      SWinKey winRes = {.ts = win.skey, .groupId = groupIds[i]};
-      if (pDelWins) {
-        taosArrayPush(pDelWins, &winRes);
-      }
-      if (pUpdatedMap) {
-        taosHashRemove(pUpdatedMap, &winRes, sizeof(SWinKey));
-      }
-      getNextTimeWindow(pInterval, pInterval->precision, TSDB_ORDER_ASC, &win);
-    } while (win.skey <= endTs);
-  }
-}
-
-static void doClearWindows(SAggSupporter* pAggSup, SExprSupp* pSup1, SInterval* pInterval, int32_t numOfOutput,
-                           SSDataBlock* pBlock, SArray* pUpWins) {
-  SColumnInfoData* pStartTsCol = taosArrayGet(pBlock->pDataBlock, START_TS_COLUMN_INDEX);
-  TSKEY*           startTsCols = (TSKEY*)pStartTsCol->pData;
-  SColumnInfoData* pEndTsCol = taosArrayGet(pBlock->pDataBlock, END_TS_COLUMN_INDEX);
-  TSKEY*           endTsCols = (TSKEY*)pEndTsCol->pData;
-  SColumnInfoData* pGpCol = taosArrayGet(pBlock->pDataBlock, GROUPID_COLUMN_INDEX);
-  uint64_t*        pGpDatas = (uint64_t*)pGpCol->pData;
-  for (int32_t i = 0; i < pBlock->info.rows; i++) {
-    SResultRowInfo dumyInfo = {0};
-    dumyInfo.cur.pageId = -1;
-
-    STimeWindow win = getActiveTimeWindow(NULL, &dumyInfo, startTsCols[i], pInterval, TSDB_ORDER_ASC);
-    while (win.ekey <= endTsCols[i]) {
-      uint64_t winGpId = pGpDatas[i];
-      bool     res = doClearWindow(pAggSup, pSup1, (char*)&win.skey, sizeof(TSKEY), winGpId, numOfOutput);
-      if (pUpWins && res) {
-        SWinKey winRes = {.ts = win.skey, .groupId = winGpId};
-        taosArrayPush(pUpWins, &winRes);
-      }
-      getNextTimeWindow(pInterval, pInterval->precision, TSDB_ORDER_ASC, &win);
-    }
-  }
-}
-
 static int32_t getAllIntervalWindow(SSHashObj* pHashMap, SHashObj* resWins) {
   void*   pIte = NULL;
   size_t  keyLen = 0;
@@ -2923,11 +2815,12 @@ bool hasIntervalWindow(SStreamState* pState, SWinKey* pKey) {
   return TSDB_CODE_SUCCESS == streamStateGet(pState, pKey, NULL, 0);
 }
 
-static void rebuildIntervalWindow(SOperatorInfo* pOperator, SExprSupp* pSup, SArray* pWinArray, SHashObj* pUpdatedMap) {
+static void rebuildIntervalWindow(SOperatorInfo* pOperator, SArray* pWinArray, SHashObj* pUpdatedMap) {
   SStreamIntervalOperatorInfo* pInfo = pOperator->info;
   SExecTaskInfo*               pTaskInfo = pOperator->pTaskInfo;
   int32_t                      size = taosArrayGetSize(pWinArray);
   int32_t                      numOfOutput = pOperator->exprSupp.numOfExprs;
+  SExprSupp*                   pSup = &pOperator->exprSupp;
   if (!pInfo->pChildren) {
     return;
   }
@@ -2935,8 +2828,7 @@ static void rebuildIntervalWindow(SOperatorInfo* pOperator, SExprSupp* pSup, SAr
     SWinKey*    pWinRes = taosArrayGet(pWinArray, i);
     SResultRow* pCurResult = NULL;
     STimeWindow parentWin = getFinalTimeWindow(pWinRes->ts, &pInfo->interval);
-    if (isDeletedStreamWindow(&parentWin, pWinRes->groupId, pInfo->pState, &pInfo->twAggSup) &&
-        isCloseWindow(&parentWin, &pInfo->twAggSup)) {
+    if (isDeletedStreamWindow(&parentWin, pWinRes->groupId, pInfo->pState, &pInfo->twAggSup)) {
       continue;
     }
 
@@ -3223,6 +3115,14 @@ static void doStreamIntervalAggImpl(SOperatorInfo* pOperatorInfo, SSDataBlock* p
     if (pInfo->twAggSup.calTrigger == STREAM_TRIGGER_AT_ONCE && pUpdatedMap) {
       saveWinResultInfo(pResult->win.skey, groupId, pUpdatedMap);
     }
+
+    if (pInfo->twAggSup.calTrigger == STREAM_TRIGGER_WINDOW_CLOSE) {
+      SWinKey key = {
+          .ts = pResult->win.skey,
+          .groupId = groupId,
+      };
+      tSimpleHashPut(pInfo->aggSup.pResultRowHashTable, &key, sizeof(SWinKey), NULL, 0);
+    }
     updateTimeWindowInfo(&pInfo->twAggSup.timeWindowData, &nextWin, true);
     doApplyFunctions(pTaskInfo, pSup->pCtx, &pInfo->twAggSup.timeWindowData, startPos, forwardRows,
                      pSDataBlock->info.rows, numOfOutput);
@@ -3338,7 +3238,7 @@ static SSDataBlock* doStreamFinalIntervalAgg(SOperatorInfo* pOperator) {
         SStreamIntervalOperatorInfo* pChildInfo = pChildOp->info;
         SExprSupp*                   pChildSup = &pChildOp->exprSupp;
         doDeleteWindows(pChildOp, &pChildInfo->interval, pBlock, NULL, NULL);
-        rebuildIntervalWindow(pOperator, pSup, delWins, pUpdatedMap);
+        rebuildIntervalWindow(pOperator, delWins, pUpdatedMap);
         addRetriveWindow(delWins, pInfo);
         taosArrayAddAll(pInfo->pDelWins, delWins);
         taosArrayDestroy(delWins);
@@ -3564,42 +3464,18 @@ _error:
 }
 
 void destroyStreamAggSupporter(SStreamAggSupporter* pSup) {
-  taosMemoryFreeClear(pSup->pKeyBuf);
-  void** pIte = NULL;
-  while ((pIte = taosHashIterate(pSup->pResultRows, pIte)) != NULL) {
-    SArray* pWins = (SArray*)(*pIte);
-    taosArrayDestroy(pWins);
-  }
-  taosHashCleanup(pSup->pResultRows);
-  destroyDiskbasedBuf(pSup->pResultBuf);
-  blockDataDestroy(pSup->pScanBlock);
-}
-
-void destroyStateWinInfo(void* ptr) {
-  if (ptr == NULL) {
-    return;
-  }
-  SStateWindowInfo* pWin = (SStateWindowInfo*)ptr;
-  taosMemoryFreeClear(pWin->stateKey.pData);
-}
-
-void destroyStateStreamAggSupporter(SStreamAggSupporter* pSup) {
-  taosMemoryFreeClear(pSup->pKeyBuf);
-  void** pIte = NULL;
-  while ((pIte = taosHashIterate(pSup->pResultRows, pIte)) != NULL) {
-    SArray* pWins = (SArray*)(*pIte);
-    taosArrayDestroyEx(pWins, (FDelete)destroyStateWinInfo);
-  }
-  taosHashCleanup(pSup->pResultRows);
+  tSimpleHashCleanup(pSup->pResultRows);
   destroyDiskbasedBuf(pSup->pResultBuf);
   blockDataDestroy(pSup->pScanBlock);
+  taosMemoryFreeClear(pSup->pState);
+  taosMemoryFreeClear(pSup->pDummyCtx);
 }
 
 void destroyStreamSessionAggOperatorInfo(void* param) {
   SStreamSessionAggOperatorInfo* pInfo = (SStreamSessionAggOperatorInfo*)param;
   cleanupBasicInfo(&pInfo->binfo);
   destroyStreamAggSupporter(&pInfo->streamAggSup);
-  cleanupGroupResInfo(&pInfo->groupResInfo);
+
   if (pInfo->pChildren != NULL) {
     int32_t size = taosArrayGetSize(pInfo->pChildren);
     for (int32_t i = 0; i < size; i++) {
@@ -3613,8 +3489,7 @@ void destroyStreamSessionAggOperatorInfo(void* param) {
   blockDataDestroy(pInfo->pDelRes);
   blockDataDestroy(pInfo->pWinBlock);
   blockDataDestroy(pInfo->pUpdateRes);
-  destroySqlFunctionCtx(pInfo->pDummyCtx, 0);
-  taosHashCleanup(pInfo->pStDeleted);
+  tSimpleHashCleanup(pInfo->pStDeleted);
   taosHashCleanup(pInfo->pGroupIdTbNameMap);
 
   taosMemoryFreeClear(param);
@@ -3643,129 +3518,62 @@ void initDummyFunction(SqlFunctionCtx* pDummy, SqlFunctionCtx* pCtx, int32_t num
   }
 }
 
-void initDownStream(SOperatorInfo* downstream, SStreamAggSupporter* pAggSup, int64_t gap, int64_t waterMark,
-                    uint16_t type, int32_t tsColIndex) {
+void initDownStream(SOperatorInfo* downstream, SStreamAggSupporter* pAggSup, int64_t waterMark, uint16_t type,
+                    int32_t tsColIndex) {
   if (downstream->operatorType == QUERY_NODE_PHYSICAL_PLAN_STREAM_PARTITION) {
     SStreamPartitionOperatorInfo* pScanInfo = downstream->info;
     pScanInfo->tsColIndex = tsColIndex;
   }
 
   if (downstream->operatorType != QUERY_NODE_PHYSICAL_PLAN_STREAM_SCAN) {
-    initDownStream(downstream->pDownstream[0], pAggSup, gap, waterMark, type, tsColIndex);
+    initDownStream(downstream->pDownstream[0], pAggSup, waterMark, type, tsColIndex);
     return;
   }
   SStreamScanInfo* pScanInfo = downstream->info;
-  pScanInfo->windowSup = (SWindowSupporter){.pStreamAggSup = pAggSup, .gap = gap, .parentType = type};
+  pScanInfo->windowSup = (SWindowSupporter){.pStreamAggSup = pAggSup, .gap = pAggSup->gap, .parentType = type};
   pScanInfo->pUpdateInfo = updateInfoInit(60000, TSDB_TIME_PRECISION_MILLI, waterMark);
 }
 
-int32_t initSessionAggSupporter(SStreamAggSupporter* pSup, const char* pKey, SqlFunctionCtx* pCtx,
-                                int32_t numOfOutput) {
-  return initStreamAggSupporter(pSup, pKey, pCtx, numOfOutput, sizeof(SResultWindowInfo));
-}
-
-SOperatorInfo* createStreamSessionAggOperatorInfo(SOperatorInfo* downstream, SPhysiNode* pPhyNode,
-                                                  SExecTaskInfo* pTaskInfo) {
-  SSessionWinodwPhysiNode*       pSessionNode = (SSessionWinodwPhysiNode*)pPhyNode;
-  int32_t                        numOfCols = 0;
-  int32_t                        code = TSDB_CODE_OUT_OF_MEMORY;
-
-  SStreamSessionAggOperatorInfo* pInfo = taosMemoryCalloc(1, sizeof(SStreamSessionAggOperatorInfo));
-  SOperatorInfo*                 pOperator = taosMemoryCalloc(1, sizeof(SOperatorInfo));
-  if (pInfo == NULL || pOperator == NULL) {
-    goto _error;
+int32_t initStreamAggSupporter(SStreamAggSupporter* pSup, SqlFunctionCtx* pCtx, int32_t numOfOutput, int64_t gap,
+                               SStreamState* pState, int32_t keySize, int16_t keyType) {
+  pSup->resultRowSize = keySize + getResultRowSize(pCtx, numOfOutput);
+  pSup->pScanBlock = createSpecialDataBlock(STREAM_CLEAR);
+  pSup->gap = gap;
+  pSup->stateKeySize = keySize;
+  pSup->stateKeyType = keyType;
+  pSup->pDummyCtx = (SqlFunctionCtx*)taosMemoryCalloc(numOfOutput, sizeof(SqlFunctionCtx));
+  if (pSup->pDummyCtx == NULL) {
+    return TSDB_CODE_OUT_OF_MEMORY;
   }
 
-  pOperator->pTaskInfo = pTaskInfo;
+  initDummyFunction(pSup->pDummyCtx, pCtx, numOfOutput);
+  pSup->pState = taosMemoryCalloc(1, sizeof(SStreamState));
+  *(pSup->pState) = *pState;
+  streamStateSetNumber(pSup->pState, -1);
 
-  initResultSizeInfo(&pOperator->resultInfo, 4096);
-  if (pSessionNode->window.pExprs != NULL) {
-    int32_t    numOfScalar = 0;
-    SExprInfo* pScalarExprInfo = createExprInfo(pSessionNode->window.pExprs, NULL, &numOfScalar);
-    code = initExprSupp(&pInfo->scalarSupp, pScalarExprInfo, numOfScalar);
-    if (code != TSDB_CODE_SUCCESS) {
-      goto _error;
-    }
-  }
-
-  SExprSupp* pSup = &pOperator->exprSupp;
-
-  SExprInfo*   pExprInfo = createExprInfo(pSessionNode->window.pFuncs, NULL, &numOfCols);
-  SSDataBlock* pResBlock = createResDataBlock(pPhyNode->pOutputDataBlockDesc);
-  code = initBasicInfoEx(&pInfo->binfo, pSup, pExprInfo, numOfCols, pResBlock);
-  if (code != TSDB_CODE_SUCCESS) {
-    goto _error;
-  }
-
-  code = initSessionAggSupporter(&pInfo->streamAggSup, "StreamSessionAggOperatorInfo", pSup->pCtx, numOfCols);
-  if (code != TSDB_CODE_SUCCESS) {
-    goto _error;
-  }
-
-  pInfo->pDummyCtx = (SqlFunctionCtx*)taosMemoryCalloc(numOfCols, sizeof(SqlFunctionCtx));
-  if (pInfo->pDummyCtx == NULL) {
-    goto _error;
-  }
-
-  initDummyFunction(pInfo->pDummyCtx, pSup->pCtx, numOfCols);
-  pInfo->twAggSup = (STimeWindowAggSupp){
-      .waterMark = pSessionNode->window.watermark,
-      .calTrigger = pSessionNode->window.triggerType,
-      .maxTs = INT64_MIN,
-      .minTs = INT64_MAX,
-  };
-
-  initResultRowInfo(&pInfo->binfo.resultRowInfo);
-  initExecTimeWindowInfo(&pInfo->twAggSup.timeWindowData, &pTaskInfo->window);
-
-  pInfo->primaryTsIndex = ((SColumnNode*)pSessionNode->window.pTspk)->slotId;
-  if (pSessionNode->window.pTsEnd) {
-    pInfo->endTsIndex = ((SColumnNode*)pSessionNode->window.pTsEnd)->slotId;
-  }
-  pInfo->gap = pSessionNode->gap;
-  pInfo->binfo.pRes = pResBlock;
-  pInfo->order = TSDB_ORDER_ASC;
   _hash_fn_t hashFn = taosGetDefaultHashFunction(TSDB_DATA_TYPE_BINARY);
-  pInfo->pStDeleted = taosHashInit(64, hashFn, true, HASH_NO_LOCK);
-  pInfo->pDelIterator = NULL;
-  pInfo->pDelRes = createSpecialDataBlock(STREAM_DELETE_RESULT);
-  pInfo->pChildren = NULL;
-  pInfo->isFinal = false;
-  pInfo->pPhyNode = pPhyNode;
-  pInfo->ignoreExpiredData = pSessionNode->window.igExpired;
+  pSup->pResultRows = tSimpleHashInit(32, hashFn);
 
-  pInfo->pGroupIdTbNameMap =
-      taosHashInit(1024, taosGetDefaultHashFunction(TSDB_DATA_TYPE_UBIGINT), false, HASH_NO_LOCK);
-
-  pOperator->name = "StreamSessionWindowAggOperator";
-  pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_STREAM_SESSION;
-  pOperator->blocking = true;
-  pOperator->status = OP_NOT_OPENED;
-  pOperator->info = pInfo;
-  pOperator->fpSet =
-      createOperatorFpSet(operatorDummyOpenFn, doStreamSessionAgg, NULL, NULL, destroyStreamSessionAggOperatorInfo,
-                          aggEncodeResultRow, aggDecodeResultRow, NULL);
-  if (downstream) {
-    initDownStream(downstream, &pInfo->streamAggSup, pInfo->gap, pInfo->twAggSup.waterMark, pOperator->operatorType,
-                   pInfo->primaryTsIndex);
-    code = appendDownstream(pOperator, &downstream, 1);
+  int32_t pageSize = 4096;
+  while (pageSize < pSup->resultRowSize * 4) {
+    pageSize <<= 1u;
   }
-  return pOperator;
-
-_error:
-  if (pInfo != NULL) {
-    destroyStreamSessionAggOperatorInfo(pInfo);
+  // at least four pages need to be in buffer
+  int32_t bufSize = 4096 * 256;
+  if (bufSize <= pageSize) {
+    bufSize = pageSize * 4;
+  }
+  if (!osTempSpaceAvailable()) {
+    terrno = TSDB_CODE_NO_AVAIL_DISK;
+    qError("Init stream agg supporter failed since %s", terrstr(terrno));
+    return terrno;
+  }
+  int32_t code = createDiskbasedBuf(&pSup->pResultBuf, pageSize, bufSize, "function", tsTempDir);
+  for (int32_t i = 0; i < numOfOutput; ++i) {
+    pCtx[i].saveHandle.pBuf = pSup->pResultBuf;
   }
 
-  taosMemoryFreeClear(pOperator);
-  pTaskInfo->code = code;
-  return NULL;
-}
-
-int64_t getSessionWindowEndkey(void* data, int32_t index) {
-  SArray*            pWinInfos = (SArray*)data;
-  SResultWindowInfo* pWin = taosArrayGet(pWinInfos, index);
-  return pWin->win.ekey;
+  return TSDB_CODE_SUCCESS;
 }
 
 bool isInTimeWindow(STimeWindow* pWin, TSKEY ts, int64_t gap) {
@@ -3775,305 +3583,243 @@ bool isInTimeWindow(STimeWindow* pWin, TSKEY ts, int64_t gap) {
   return false;
 }
 
-bool isInWindow(SResultWindowInfo* pWinInfo, TSKEY ts, int64_t gap) { return isInTimeWindow(&pWinInfo->win, ts, gap); }
-
-static SResultWindowInfo* insertNewSessionWindow(SArray* pWinInfos, TSKEY startTs, TSKEY endTs, int32_t index) {
-  SResultWindowInfo win = {
-      .pos.offset = -1, .pos.pageId = -1, .win.skey = startTs, .win.ekey = endTs, .isOutput = false};
-  return taosArrayInsert(pWinInfos, index, &win);
+bool isInWindow(SResultWindowInfo* pWinInfo, TSKEY ts, int64_t gap) {
+  return isInTimeWindow(&pWinInfo->sessionWin.win, ts, gap);
 }
 
-static SResultWindowInfo* addNewSessionWindow(SArray* pWinInfos, TSKEY startTs, TSKEY endTs) {
-  SResultWindowInfo win = {
-      .pos.offset = -1, .pos.pageId = -1, .win.skey = startTs, .win.ekey = endTs, .isOutput = false};
-  return taosArrayPush(pWinInfos, &win);
+void getCurSessionWindow(SStreamAggSupporter* pAggSup, TSKEY startTs, TSKEY endTs, uint64_t groupId,
+                         SSessionKey* pKey) {
+  pKey->win.skey = startTs;
+  pKey->win.ekey = endTs;
+  pKey->groupId = groupId;
+  SStreamStateCur* pCur = streamStateSessionGetCur(pAggSup->pState, pKey);
+  int32_t          code = streamStateSessionGetKVByCur(pCur, pKey, NULL, 0);
+  streamStateFreeCur(pCur);
+  if (code != TSDB_CODE_SUCCESS) {
+    SET_SESSION_WIN_KEY_INVALID(pKey);
+  }
 }
 
-SArray* getWinInfos(SStreamAggSupporter* pAggSup, uint64_t groupId) {
-  void**  ite = taosHashGet(pAggSup->pResultRows, &groupId, sizeof(uint64_t));
-  SArray* pWinInfos = NULL;
-  if (ite == NULL) {
-    pWinInfos = taosArrayInit(1024, pAggSup->valueSize);
-    taosHashPut(pAggSup->pResultRows, &groupId, sizeof(uint64_t), &pWinInfos, sizeof(void*));
+bool isInvalidSessionWin(SResultWindowInfo* pWinInfo) { return pWinInfo->sessionWin.win.skey == 0; }
+
+void setSessionOutputBuf(SStreamAggSupporter* pAggSup, TSKEY startTs, TSKEY endTs, uint64_t groupId,
+                         SResultWindowInfo* pCurWin) {
+  pCurWin->sessionWin.groupId = groupId;
+  pCurWin->sessionWin.win.skey = startTs - pAggSup->gap;
+  pCurWin->sessionWin.win.ekey = endTs + pAggSup->gap;
+  int32_t size = pAggSup->resultRowSize;
+  int32_t code = streamStateSessionAddIfNotExist(pAggSup->pState, &pCurWin->sessionWin, &pCurWin->pOutputBuf, &size);
+  if (code == TSDB_CODE_SUCCESS) {
+    pCurWin->isOutput = true;
   } else {
-    pWinInfos = *ite;
+    pCurWin->sessionWin.win.skey = startTs;
+    pCurWin->sessionWin.win.ekey = endTs;
   }
-  return pWinInfos;
 }
 
-// don't add new window
-SResultWindowInfo* getCurSessionWindow(SStreamAggSupporter* pAggSup, TSKEY startTs, TSKEY endTs, uint64_t groupId,
-                                       int64_t gap, int32_t* pIndex) {
-  STimeWindow searchWin = {.skey = startTs, .ekey = endTs};
-  SArray*     pWinInfos = getWinInfos(pAggSup, groupId);
-  pAggSup->pCurWins = pWinInfos;
-
-  int32_t size = taosArrayGetSize(pWinInfos);
-  if (size == 0) {
-    return NULL;
+int32_t getSessionWinBuf(SStreamAggSupporter* pAggSup, SStreamStateCur* pCur, SResultWindowInfo* pWinInfo) {
+  int32_t size = 0;
+  int32_t code = streamStateSessionGetKVByCur(pCur, &pWinInfo->sessionWin, (const void**)&pWinInfo->pOutputBuf, &size);
+  if (code != TSDB_CODE_SUCCESS) {
+    return code;
   }
-  // find the first position which is smaller than the key
-  int32_t            index = binarySearch(pWinInfos, size, startTs, TSDB_ORDER_DESC, getSessionWindowEndkey);
-  SResultWindowInfo* pWin = NULL;
-  if (index >= 0) {
-    pWin = taosArrayGet(pWinInfos, index);
-    if (isInWindow(pWin, startTs, gap) || isInTimeWindow(&searchWin, pWin->win.skey, gap)) {
-      *pIndex = index;
-      return pWin;
-    }
-  }
-
-  if (index + 1 < size) {
-    pWin = taosArrayGet(pWinInfos, index + 1);
-    if (isInWindow(pWin, startTs, gap) || isInTimeWindow(&searchWin, pWin->win.skey, gap)) {
-      *pIndex = index + 1;
-      return pWin;
-    } else if (endTs != INT64_MIN && isInWindow(pWin, endTs, gap)) {
-      *pIndex = index + 1;
-      return pWin;
-    }
-  }
-
-  return NULL;
+  streamStateCurNext(pAggSup->pState, pCur);
+  return TSDB_CODE_SUCCESS;
+}
+void saveDeleteInfo(SArray* pWins, SSessionKey key) {
+  // key.win.ekey = key.win.skey;
+  taosArrayPush(pWins, &key);
 }
 
-SResultWindowInfo* getSessionTimeWindow(SStreamAggSupporter* pAggSup, TSKEY startTs, TSKEY endTs, uint64_t groupId,
-                                        int64_t gap, int32_t* pIndex) {
-  SArray* pWinInfos = getWinInfos(pAggSup, groupId);
-  pAggSup->pCurWins = pWinInfos;
+void saveDeleteRes(SSHashObj* pStDelete, SSessionKey key) {
+  key.win.ekey = key.win.skey;
+  tSimpleHashPut(pStDelete, &key, sizeof(SSessionKey), NULL, 0);
+}
 
-  int32_t size = taosArrayGetSize(pWinInfos);
-  if (size == 0) {
-    *pIndex = 0;
-    return addNewSessionWindow(pWinInfos, startTs, endTs);
-  }
-  // find the first position which is smaller than the key
-  int32_t            index = binarySearch(pWinInfos, size, startTs, TSDB_ORDER_DESC, getSessionWindowEndkey);
-  SResultWindowInfo* pWin = NULL;
-  if (index >= 0) {
-    pWin = taosArrayGet(pWinInfos, index);
-    if (isInWindow(pWin, startTs, gap)) {
-      *pIndex = index;
-      return pWin;
-    }
-  }
+static void removeSessionResult(SSHashObj* pHashMap, SSHashObj* pResMap, SSessionKey key) {
+  key.win.ekey = key.win.skey;
+  tSimpleHashRemove(pHashMap, &key, sizeof(SSessionKey));
+  tSimpleHashRemove(pResMap, &key, sizeof(SSessionKey));
+}
 
-  if (index + 1 < size) {
-    pWin = taosArrayGet(pWinInfos, index + 1);
-    if (isInWindow(pWin, startTs, gap)) {
-      *pIndex = index + 1;
-      return pWin;
-    } else if (endTs != INT64_MIN && isInWindow(pWin, endTs, gap)) {
-      *pIndex = index;
-      return pWin;
-    }
+static void removeSessionResults(SSHashObj* pHashMap, SArray* pWins) {
+  if (tSimpleHashGetSize(pHashMap) == 0) {
+    return;
   }
-
-  if (index == size - 1) {
-    *pIndex = taosArrayGetSize(pWinInfos);
-    return addNewSessionWindow(pWinInfos, startTs, endTs);
+  int32_t size = taosArrayGetSize(pWins);
+  for (int32_t i = 0; i < size; i++) {
+    SSessionKey* pWin = taosArrayGet(pWins, i);
+    if (!pWin) continue;
+    SSessionKey key = *pWin;
+    key.win.ekey = key.win.skey;
+    tSimpleHashRemove(pHashMap, &key, sizeof(SSessionKey));
   }
-  *pIndex = index + 1;
-  return insertNewSessionWindow(pWinInfos, startTs, endTs, index + 1);
 }
 
 int32_t updateSessionWindowInfo(SResultWindowInfo* pWinInfo, TSKEY* pStartTs, TSKEY* pEndTs, uint64_t groupId,
-                                int32_t rows, int32_t start, int64_t gap, SHashObj* pStDeleted) {
+                                int32_t rows, int32_t start, int64_t gap, SSHashObj* pResultRows, SSHashObj* pStUpdated,
+                                SSHashObj* pStDeleted) {
   for (int32_t i = start; i < rows; ++i) {
     if (!isInWindow(pWinInfo, pStartTs[i], gap) && (!pEndTs || !isInWindow(pWinInfo, pEndTs[i], gap))) {
       return i - start;
     }
-    if (pWinInfo->win.skey > pStartTs[i]) {
+    if (pWinInfo->sessionWin.win.skey > pStartTs[i]) {
       if (pStDeleted && pWinInfo->isOutput) {
-        SWinKey res = {.ts = pWinInfo->win.skey, .groupId = groupId};
-        taosHashPut(pStDeleted, &res, sizeof(SWinKey), &res, sizeof(SWinKey));
-        pWinInfo->isOutput = false;
+        saveDeleteRes(pStDeleted, pWinInfo->sessionWin);
       }
-      pWinInfo->win.skey = pStartTs[i];
+      removeSessionResult(pStUpdated, pResultRows, pWinInfo->sessionWin);
+      pWinInfo->sessionWin.win.skey = pStartTs[i];
     }
-    pWinInfo->win.ekey = TMAX(pWinInfo->win.ekey, pStartTs[i]);
+    pWinInfo->sessionWin.win.ekey = TMAX(pWinInfo->sessionWin.win.ekey, pStartTs[i]);
     if (pEndTs) {
-      pWinInfo->win.ekey = TMAX(pWinInfo->win.ekey, pEndTs[i]);
+      pWinInfo->sessionWin.win.ekey = TMAX(pWinInfo->sessionWin.win.ekey, pEndTs[i]);
     }
   }
   return rows - start;
 }
 
-static int32_t setWindowOutputBuf(SResultWindowInfo* pWinInfo, SResultRow** pResult, SqlFunctionCtx* pCtx,
-                                  uint64_t groupId, int32_t numOfOutput, int32_t* rowEntryInfoOffset,
-                                  SStreamAggSupporter* pAggSup, SExecTaskInfo* pTaskInfo) {
-  assert(pWinInfo->win.skey <= pWinInfo->win.ekey);
-  // too many time window in query
-  int32_t size = taosArrayGetSize(pAggSup->pCurWins);
-  if (pTaskInfo->execModel == OPTR_EXEC_MODEL_BATCH && size > MAX_INTERVAL_TIME_WINDOW) {
-    T_LONG_JMP(pTaskInfo->env, TSDB_CODE_QRY_TOO_MANY_TIMEWINDOW);
-  }
-
-  if (pWinInfo->pos.pageId == -1) {
-    *pResult = getNewResultRow(pAggSup->pResultBuf, &pAggSup->currentPageId, pAggSup->resultRowSize);
-    if (*pResult == NULL) {
-      return TSDB_CODE_OUT_OF_MEMORY;
-    }
-
-    // add a new result set for a new group
-    pWinInfo->pos.pageId = (*pResult)->pageId;
-    pWinInfo->pos.offset = (*pResult)->offset;
-  } else {
-    *pResult = getResultRowByPos(pAggSup->pResultBuf, &pWinInfo->pos, true);
-    if (!(*pResult)) {
-      qError("getResultRowByPos return NULL, TID:%s", GET_TASKID(pTaskInfo));
-      return TSDB_CODE_FAILED;
-    }
-  }
-
+static int32_t initSessionOutputBuf(SResultWindowInfo* pWinInfo, SResultRow** pResult, SqlFunctionCtx* pCtx,
+                                    int32_t numOfOutput, int32_t* rowEntryInfoOffset) {
+  ASSERT(pWinInfo->sessionWin.win.skey <= pWinInfo->sessionWin.win.ekey);
+  *pResult = (SResultRow*)pWinInfo->pOutputBuf;
   // set time window for current result
-  (*pResult)->win = pWinInfo->win;
+  (*pResult)->win = pWinInfo->sessionWin.win;
   setResultRowInitCtx(*pResult, pCtx, numOfOutput, rowEntryInfoOffset);
   return TSDB_CODE_SUCCESS;
 }
 
-static int32_t doOneWindowAggImpl(int32_t tsColId, SOptrBasicInfo* pBinfo, SStreamAggSupporter* pAggSup,
-                                  SColumnInfoData* pTimeWindowData, SSDataBlock* pSDataBlock,
-                                  SResultWindowInfo* pCurWin, SResultRow** pResult, int32_t startIndex, int32_t winRows,
-                                  int32_t numOutput, SOperatorInfo* pOperator) {
+static int32_t doOneWindowAggImpl(SColumnInfoData* pTimeWindowData, SResultWindowInfo* pCurWin, SResultRow** pResult,
+                                  int32_t startIndex, int32_t winRows, int32_t rows, int32_t numOutput,
+                                  SOperatorInfo* pOperator) {
   SExprSupp*     pSup = &pOperator->exprSupp;
   SExecTaskInfo* pTaskInfo = pOperator->pTaskInfo;
-
-  SColumnInfoData* pColDataInfo = taosArrayGet(pSDataBlock->pDataBlock, tsColId);
-  TSKEY*           tsCols = (int64_t*)pColDataInfo->pData;
-  int32_t          code = setWindowOutputBuf(pCurWin, pResult, pSup->pCtx, pSDataBlock->info.groupId, numOutput,
-                                             pSup->rowEntryInfoOffset, pAggSup, pTaskInfo);
+  int32_t        code = initSessionOutputBuf(pCurWin, pResult, pSup->pCtx, numOutput, pSup->rowEntryInfoOffset);
   if (code != TSDB_CODE_SUCCESS || (*pResult) == NULL) {
     return TSDB_CODE_QRY_OUT_OF_MEMORY;
   }
-  updateTimeWindowInfo(pTimeWindowData, &pCurWin->win, false);
-  doApplyFunctions(pTaskInfo, pSup->pCtx, pTimeWindowData, startIndex, winRows, pSDataBlock->info.rows, numOutput);
-  SFilePage* bufPage = getBufPage(pAggSup->pResultBuf, pCurWin->pos.pageId);
-  setBufPageDirty(bufPage, true);
-  releaseBufPage(pAggSup->pResultBuf, bufPage);
+  updateTimeWindowInfo(pTimeWindowData, &pCurWin->sessionWin.win, false);
+  doApplyFunctions(pTaskInfo, pSup->pCtx, pTimeWindowData, startIndex, winRows, rows, numOutput);
   return TSDB_CODE_SUCCESS;
 }
 
-static int32_t doOneWindowAgg(SStreamSessionAggOperatorInfo* pInfo, SSDataBlock* pSDataBlock,
-                              SResultWindowInfo* pCurWin, SResultRow** pResult, int32_t startIndex, int32_t winRows,
-                              int32_t numOutput, SOperatorInfo* pOperator) {
-  return doOneWindowAggImpl(pInfo->primaryTsIndex, &pInfo->binfo, &pInfo->streamAggSup, &pInfo->twAggSup.timeWindowData,
-                            pSDataBlock, pCurWin, pResult, startIndex, winRows, numOutput, pOperator);
+static bool doDeleteSessionWindow(SStreamAggSupporter* pAggSup, SSessionKey* pKey) {
+  streamStateSessionDel(pAggSup->pState, pKey);
+  tSimpleHashRemove(pAggSup->pResultRows, pKey, sizeof(SSessionKey));
+  return true;
 }
 
-static int32_t doOneStateWindowAgg(SStreamStateAggOperatorInfo* pInfo, SSDataBlock* pSDataBlock,
-                                   SResultWindowInfo* pCurWin, SResultRow** pResult, int32_t startIndex,
-                                   int32_t winRows, int32_t numOutput, SOperatorInfo* pOperator) {
-  return doOneWindowAggImpl(pInfo->primaryTsIndex, &pInfo->binfo, &pInfo->streamAggSup, &pInfo->twAggSup.timeWindowData,
-                            pSDataBlock, pCurWin, pResult, startIndex, winRows, numOutput, pOperator);
-}
-
-int32_t getNumCompactWindow(SArray* pWinInfos, int32_t startIndex, int64_t gap) {
-  SResultWindowInfo* pCurWin = taosArrayGet(pWinInfos, startIndex);
-  int32_t            size = taosArrayGetSize(pWinInfos);
-  // Just look for the window behind StartIndex
-  for (int32_t i = startIndex + 1; i < size; i++) {
-    SResultWindowInfo* pWinInfo = taosArrayGet(pWinInfos, i);
-    if (!isInWindow(pCurWin, pWinInfo->win.skey, gap)) {
-      return i - startIndex - 1;
-    }
+static int32_t setSessionWinOutputInfo(SSHashObj* pStUpdated, SResultWindowInfo* pWinInfo) {
+  void* pVal = tSimpleHashGet(pStUpdated, &pWinInfo->sessionWin, sizeof(SSessionKey));
+  if (pVal) {
+    SResultWindowInfo* pWin = pVal;
+    pWinInfo->isOutput = pWin->isOutput;
   }
-
-  return size - startIndex - 1;
+  return TSDB_CODE_SUCCESS;
 }
 
-void compactTimeWindow(SStreamSessionAggOperatorInfo* pInfo, int32_t startIndex, int32_t num, uint64_t groupId,
-                       int32_t numOfOutput, SHashObj* pStUpdated, SHashObj* pStDeleted, SOperatorInfo* pOperator) {
-  SExprSupp*     pSup = &pOperator->exprSupp;
-  SExecTaskInfo* pTaskInfo = pOperator->pTaskInfo;
-
-  SResultWindowInfo* pCurWin = taosArrayGet(pInfo->streamAggSup.pCurWins, startIndex);
-  SResultRow*        pCurResult = NULL;
-  setWindowOutputBuf(pCurWin, &pCurResult, pSup->pCtx, groupId, numOfOutput, pSup->rowEntryInfoOffset,
-                     &pInfo->streamAggSup, pTaskInfo);
-  num += startIndex + 1;
-  ASSERT(num <= taosArrayGetSize(pInfo->streamAggSup.pCurWins));
-  // Just look for the window behind StartIndex
-  for (int32_t i = startIndex + 1; i < num; i++) {
-    SResultWindowInfo* pWinInfo = taosArrayGet(pInfo->streamAggSup.pCurWins, i);
-    SResultRow*        pWinResult = NULL;
-    setWindowOutputBuf(pWinInfo, &pWinResult, pInfo->pDummyCtx, groupId, numOfOutput, pSup->rowEntryInfoOffset,
-                       &pInfo->streamAggSup, pTaskInfo);
-    pCurWin->win.ekey = TMAX(pCurWin->win.ekey, pWinInfo->win.ekey);
-    updateTimeWindowInfo(&pInfo->twAggSup.timeWindowData, &pCurWin->win, true);
-    compactFunctions(pSup->pCtx, pInfo->pDummyCtx, numOfOutput, pTaskInfo, &pInfo->twAggSup.timeWindowData);
-    taosHashRemove(pStUpdated, &pWinInfo->pos, sizeof(SResultRowPosition));
-    if (pWinInfo->isOutput && pStDeleted) {
-      SWinKey res = {.ts = pWinInfo->win.skey, .groupId = groupId};
-      taosHashPut(pStDeleted, &res, sizeof(SWinKey), &res, sizeof(SWinKey));
-      pWinInfo->isOutput = false;
-    }
-    taosArrayRemove(pInfo->streamAggSup.pCurWins, i);
-    SFilePage* tmpPage = getBufPage(pInfo->streamAggSup.pResultBuf, pWinInfo->pos.pageId);
-    releaseBufPage(pInfo->streamAggSup.pResultBuf, tmpPage);
+SStreamStateCur* getNextSessionWinInfo(SStreamAggSupporter* pAggSup, SSHashObj* pStUpdated, SResultWindowInfo* pCurWin,
+                                       SResultWindowInfo* pNextWin) {
+  SStreamStateCur* pCur = streamStateSessionSeekKeyNext(pAggSup->pState, &pCurWin->sessionWin);
+  pNextWin->isOutput = true;
+  setSessionWinOutputInfo(pStUpdated, pNextWin);
+  int32_t size = 0;
+  pNextWin->sessionWin = pCurWin->sessionWin;
+  int32_t code = streamStateSessionGetKVByCur(pCur, &pNextWin->sessionWin, (const void**)&pNextWin->pOutputBuf, &size);
+  if (code != TSDB_CODE_SUCCESS) {
+    SET_SESSION_WIN_INVALID(*pNextWin);
   }
-  SFilePage* bufPage = getBufPage(pInfo->streamAggSup.pResultBuf, pCurWin->pos.pageId);
-  ASSERT(num > 0);
-  setBufPageDirty(bufPage, true);
-  releaseBufPage(pInfo->streamAggSup.pResultBuf, bufPage);
+  return pCur;
 }
 
-static void doStreamSessionAggImpl(SOperatorInfo* pOperator, SSDataBlock* pSDataBlock, SHashObj* pStUpdated,
-                                   SHashObj* pStDeleted, bool hasEndTs) {
+static void compactSessionWindow(SOperatorInfo* pOperator, SResultWindowInfo* pCurWin, SSHashObj* pStUpdated,
+                                 SSHashObj* pStDeleted) {
+  SExprSupp*                     pSup = &pOperator->exprSupp;
+  SExecTaskInfo*                 pTaskInfo = pOperator->pTaskInfo;
+  SStreamSessionAggOperatorInfo* pInfo = pOperator->info;
+  SResultRow*                    pCurResult = NULL;
+  int32_t                        numOfOutput = pOperator->exprSupp.numOfExprs;
+  SStreamAggSupporter*           pAggSup = &pInfo->streamAggSup;
+  initSessionOutputBuf(pCurWin, &pCurResult, pSup->pCtx, numOfOutput, pSup->rowEntryInfoOffset);
+  // Just look for the window behind StartIndex
+  while (1) {
+    SResultWindowInfo winInfo = {0};
+    SStreamStateCur*  pCur = getNextSessionWinInfo(pAggSup, pStUpdated, pCurWin, &winInfo);
+    if (!IS_VALID_SESSION_WIN(winInfo) || !isInWindow(pCurWin, winInfo.sessionWin.win.skey, pAggSup->gap)) {
+      streamStateFreeCur(pCur);
+      break;
+    }
+    SResultRow* pWinResult = NULL;
+    initSessionOutputBuf(&winInfo, &pWinResult, pAggSup->pDummyCtx, numOfOutput, pSup->rowEntryInfoOffset);
+    pCurWin->sessionWin.win.ekey = TMAX(pCurWin->sessionWin.win.ekey, winInfo.sessionWin.win.ekey);
+    updateTimeWindowInfo(&pInfo->twAggSup.timeWindowData, &pCurWin->sessionWin.win, true);
+    compactFunctions(pSup->pCtx, pAggSup->pDummyCtx, numOfOutput, pTaskInfo, &pInfo->twAggSup.timeWindowData);
+    tSimpleHashRemove(pStUpdated, &winInfo.sessionWin, sizeof(SSessionKey));
+    if (winInfo.isOutput && pStDeleted) {
+      saveDeleteRes(pStDeleted, winInfo.sessionWin);
+    }
+    removeSessionResult(pStUpdated, pAggSup->pResultRows, winInfo.sessionWin);
+    doDeleteSessionWindow(pAggSup, &winInfo.sessionWin);
+    streamStateFreeCur(pCur);
+  }
+}
+
+int32_t saveSessionOutputBuf(SStreamAggSupporter* pAggSup, SResultWindowInfo* pWinInfo) {
+  saveSessionDiscBuf(pAggSup->pState, &pWinInfo->sessionWin, pWinInfo->pOutputBuf, pAggSup->resultRowSize);
+  return TSDB_CODE_SUCCESS;
+}
+
+static void doStreamSessionAggImpl(SOperatorInfo* pOperator, SSDataBlock* pSDataBlock, SSHashObj* pStUpdated,
+                                   SSHashObj* pStDeleted, bool hasEndTs) {
   SExecTaskInfo*                 pTaskInfo = pOperator->pTaskInfo;
   SStreamSessionAggOperatorInfo* pInfo = pOperator->info;
-  bool                           masterScan = true;
   int32_t                        numOfOutput = pOperator->exprSupp.numOfExprs;
   uint64_t                       groupId = pSDataBlock->info.groupId;
-  int64_t                        gap = pInfo->gap;
   int64_t                        code = TSDB_CODE_SUCCESS;
+  SResultRow*                    pResult = NULL;
+  int32_t                        rows = pSDataBlock->info.rows;
+  int32_t                        winRows = 0;
 
-  int32_t     step = 1;
-  bool        ascScan = true;
-  TSKEY*      startTsCols = NULL;
-  TSKEY*      endTsCols = NULL;
-  SResultRow* pResult = NULL;
-  int32_t     winRows = 0;
-
-  ASSERT(pSDataBlock->pDataBlock);
   SColumnInfoData* pStartTsCol = taosArrayGet(pSDataBlock->pDataBlock, pInfo->primaryTsIndex);
-  startTsCols = (int64_t*)pStartTsCol->pData;
+  TSKEY*           startTsCols = (int64_t*)pStartTsCol->pData;
   SColumnInfoData* pEndTsCol = NULL;
   if (hasEndTs) {
     pEndTsCol = taosArrayGet(pSDataBlock->pDataBlock, pInfo->endTsIndex);
   } else {
     pEndTsCol = taosArrayGet(pSDataBlock->pDataBlock, pInfo->primaryTsIndex);
   }
-  endTsCols = (int64_t*)pEndTsCol->pData;
 
+  TSKEY*               endTsCols = (int64_t*)pEndTsCol->pData;
   SStreamAggSupporter* pAggSup = &pInfo->streamAggSup;
-  for (int32_t i = 0; i < pSDataBlock->info.rows;) {
+  for (int32_t i = 0; i < rows;) {
     if (pInfo->ignoreExpiredData && isOverdue(endTsCols[i], &pInfo->twAggSup)) {
       i++;
       continue;
     }
-    int32_t            winIndex = 0;
-    SResultWindowInfo* pCurWin = getSessionTimeWindow(pAggSup, startTsCols[i], endTsCols[i], groupId, gap, &winIndex);
-    winRows = updateSessionWindowInfo(pCurWin, startTsCols, endTsCols, groupId, pSDataBlock->info.rows, i, pInfo->gap,
-                                      pStDeleted);
-    code = doOneWindowAgg(pInfo, pSDataBlock, pCurWin, &pResult, i, winRows, numOfOutput, pOperator);
+    SResultWindowInfo winInfo = {0};
+    setSessionOutputBuf(pAggSup, startTsCols[i], endTsCols[i], groupId, &winInfo);
+    setSessionWinOutputInfo(pStUpdated, &winInfo);
+    winRows = updateSessionWindowInfo(&winInfo, startTsCols, endTsCols, groupId, rows, i, pAggSup->gap,
+                                      pAggSup->pResultRows, pStUpdated, pStDeleted);
+    code = doOneWindowAggImpl(&pInfo->twAggSup.timeWindowData, &winInfo, &pResult, i, winRows, rows, numOfOutput,
+                              pOperator);
     if (code != TSDB_CODE_SUCCESS || pResult == NULL) {
       T_LONG_JMP(pTaskInfo->env, TSDB_CODE_QRY_OUT_OF_MEMORY);
     }
+    compactSessionWindow(pOperator, &winInfo, pStUpdated, pStDeleted);
+    saveSessionOutputBuf(pAggSup, &winInfo);
 
-    int32_t winNum = getNumCompactWindow(pAggSup->pCurWins, winIndex, gap);
-    if (winNum > 0) {
-      compactTimeWindow(pInfo, winIndex, winNum, groupId, numOfOutput, pStUpdated, pStDeleted, pOperator);
-    }
-    pCurWin->isClosed = false;
     if (pInfo->twAggSup.calTrigger == STREAM_TRIGGER_AT_ONCE && pStUpdated) {
-      SWinKey value = {.ts = pCurWin->win.skey, .groupId = groupId};
-      code = taosHashPut(pStUpdated, &pCurWin->pos, sizeof(SResultRowPosition), &value, sizeof(SWinKey));
+      code = saveResult(winInfo, pStUpdated);
       if (code != TSDB_CODE_SUCCESS) {
         T_LONG_JMP(pTaskInfo->env, TSDB_CODE_QRY_OUT_OF_MEMORY);
       }
-      pCurWin->isOutput = true;
     }
+    if (pInfo->twAggSup.calTrigger == STREAM_TRIGGER_WINDOW_CLOSE) {
+      SSessionKey key = winInfo.sessionWin;
+      key.win.ekey = key.win.skey;
+      tSimpleHashPut(pAggSup->pResultRows, &key, sizeof(SSessionKey), &winInfo, sizeof(SResultWindowInfo));
+    }
+
     i += winRows;
   }
 }
@@ -4087,8 +3833,7 @@ void deleteWindow(SArray* pWinInfos, int32_t index, FDelete fp) {
   taosArrayRemove(pWinInfos, index);
 }
 
-static void doDeleteTimeWindows(SStreamAggSupporter* pAggSup, SSDataBlock* pBlock, int64_t gap, SArray* result,
-                                FDelete fp) {
+static void doDeleteTimeWindows(SStreamAggSupporter* pAggSup, SSDataBlock* pBlock, SArray* result) {
   SColumnInfoData* pStartTsCol = taosArrayGet(pBlock->pDataBlock, START_TS_COLUMN_INDEX);
   TSKEY*           startDatas = (TSKEY*)pStartTsCol->pData;
   SColumnInfoData* pEndTsCol = taosArrayGet(pBlock->pDataBlock, END_TS_COLUMN_INDEX);
@@ -4096,85 +3841,70 @@ static void doDeleteTimeWindows(SStreamAggSupporter* pAggSup, SSDataBlock* pBloc
   SColumnInfoData* pGroupCol = taosArrayGet(pBlock->pDataBlock, GROUPID_COLUMN_INDEX);
   uint64_t*        gpDatas = (uint64_t*)pGroupCol->pData;
   for (int32_t i = 0; i < pBlock->info.rows; i++) {
-    int32_t            winIndex = 0;
-    SResultWindowInfo* pCurWin = getCurSessionWindow(pAggSup, startDatas[i], endDatas[i], gpDatas[i], gap, &winIndex);
-    if (!pCurWin) {
-      continue;
-    }
-
-    do {
-      SResultWindowInfo delWin = *pCurWin;
-      deleteWindow(pAggSup->pCurWins, winIndex, fp);
-      if (result) {
-        delWin.groupId = gpDatas[i];
-        taosArrayPush(result, &delWin);
-      }
-      if (winIndex >= taosArrayGetSize(pAggSup->pCurWins)) {
+    while (1) {
+      SSessionKey curWin = {0};
+      getCurSessionWindow(pAggSup, startDatas[i], endDatas[i], gpDatas[i], &curWin);
+      if (IS_INVALID_SESSION_WIN_KEY(curWin)) {
         break;
       }
-      pCurWin = taosArrayGet(pAggSup->pCurWins, winIndex);
-    } while (pCurWin->win.skey <= endDatas[i]);
-  }
-}
-
-static void doClearSessionWindows(SStreamAggSupporter* pAggSup, SExprSupp* pSup, SSDataBlock* pBlock, int32_t tsIndex,
-                                  int32_t numOfOutput, int64_t gap, SArray* result) {
-  SColumnInfoData* pColDataInfo = taosArrayGet(pBlock->pDataBlock, tsIndex);
-  TSKEY*           tsCols = (TSKEY*)pColDataInfo->pData;
-  SColumnInfoData* pGpDataInfo = taosArrayGet(pBlock->pDataBlock, GROUPID_COLUMN_INDEX);
-  uint64_t*        gpCols = (uint64_t*)pGpDataInfo->pData;
-  int32_t          step = 0;
-  for (int32_t i = 0; i < pBlock->info.rows; i += step) {
-    int32_t            winIndex = 0;
-    SResultWindowInfo* pCurWin = getCurSessionWindow(pAggSup, tsCols[i], INT64_MIN, gpCols[i], gap, &winIndex);
-    if (!pCurWin || pCurWin->pos.pageId == -1) {
-      // window has been closed.
-      step = 1;
-      continue;
-    }
-    step = updateSessionWindowInfo(pCurWin, tsCols, NULL, 0, pBlock->info.rows, i, gap, NULL);
-    ASSERT(isInWindow(pCurWin, tsCols[i], gap));
-    doClearWindowImpl(&pCurWin->pos, pAggSup->pResultBuf, pSup, numOfOutput);
-    if (result) {
-      pCurWin->groupId = gpCols[i];
-      taosArrayPush(result, pCurWin);
+      doDeleteSessionWindow(pAggSup, &curWin);
+      if (result) {
+        saveDeleteInfo(result, curWin);
+      }
     }
   }
 }
 
-static int32_t copyUpdateResult(SHashObj* pStUpdated, SArray* pUpdated) {
-  void*  pData = NULL;
-  size_t keyLen = 0;
-  while ((pData = taosHashIterate(pStUpdated, pData)) != NULL) {
-    void* key = taosHashGetKey(pData, &keyLen);
-    ASSERT(keyLen == sizeof(SResultRowPosition));
-    SResKeyPos* pos = taosMemoryMalloc(sizeof(SResKeyPos) + sizeof(uint64_t));
-    if (pos == NULL) {
-      return TSDB_CODE_QRY_OUT_OF_MEMORY;
-    }
-    pos->groupId = ((SWinKey*)pData)->groupId;
-    pos->pos = *(SResultRowPosition*)key;
-    *(int64_t*)pos->key = ((SWinKey*)pData)->ts;
-    taosArrayPush(pUpdated, &pos);
+static inline int32_t sessionKeyCompareAsc(const void* pKey1, const void* pKey2) {
+  SSessionKey* pWin1 = (SSessionKey*)pKey1;
+  SSessionKey* pWin2 = (SSessionKey*)pKey2;
+
+  if (pWin1->groupId > pWin2->groupId) {
+    return 1;
+  } else if (pWin1->groupId < pWin2->groupId) {
+    return -1;
   }
-  taosArraySort(pUpdated, resultrowComparAsc);
+
+  if (pWin1->win.skey > pWin2->win.skey) {
+    return 1;
+  } else if (pWin1->win.skey < pWin2->win.skey) {
+    return -1;
+  }
+
+  return 0;
+}
+
+static int32_t copyUpdateResult(SSHashObj* pStUpdated, SArray* pUpdated) {
+  void*   pIte = NULL;
+  size_t  keyLen = 0;
+  int32_t iter = 0;
+  while ((pIte = tSimpleHashIterate(pStUpdated, pIte, &iter)) != NULL) {
+    void* key = tSimpleHashGetKey(pIte, &keyLen);
+    ASSERT(keyLen == sizeof(SSessionKey));
+    taosArrayPush(pUpdated, key);
+  }
+  taosArraySort(pUpdated, sessionKeyCompareAsc);
   return TSDB_CODE_SUCCESS;
 }
 
-void doBuildDeleteDataBlock(SHashObj* pStDeleted, SSDataBlock* pBlock, void** Ite) {
+void doBuildDeleteDataBlock(SSHashObj* pStDeleted, SSDataBlock* pBlock, void** Ite) {
   blockDataCleanup(pBlock);
-  int32_t size = taosHashGetSize(pStDeleted);
+  int32_t size = tSimpleHashGetSize(pStDeleted);
   if (size == 0) {
     return;
   }
   blockDataEnsureCapacity(pBlock, size);
-  size_t keyLen = 0;
-  while (((*Ite) = taosHashIterate(pStDeleted, *Ite)) != NULL) {
-    SWinKey*         res = *Ite;
+  size_t  keyLen = 0;
+  int32_t iter = 0;
+  while (((*Ite) = tSimpleHashIterate(pStDeleted, *Ite, &iter)) != NULL) {
+    if (pBlock->info.rows + 1 > pBlock->info.capacity) {
+      break;
+    }
+    SSessionKey*     res = tSimpleHashGetKey(*Ite, &keyLen);
     SColumnInfoData* pStartTsCol = taosArrayGet(pBlock->pDataBlock, START_TS_COLUMN_INDEX);
-    colDataAppend(pStartTsCol, pBlock->info.rows, (const char*)&res->ts, false);
+    colDataAppend(pStartTsCol, pBlock->info.rows, (const char*)&res->win.skey, false);
     SColumnInfoData* pEndTsCol = taosArrayGet(pBlock->pDataBlock, END_TS_COLUMN_INDEX);
-    colDataAppend(pEndTsCol, pBlock->info.rows, (const char*)&res->ts, false);
+    colDataAppend(pEndTsCol, pBlock->info.rows, (const char*)&res->win.skey, false);
     SColumnInfoData* pUidCol = taosArrayGet(pBlock->pDataBlock, UID_COLUMN_INDEX);
     colDataAppendNULL(pUidCol, pBlock->info.rows);
     SColumnInfoData* pGpCol = taosArrayGet(pBlock->pDataBlock, GROUPID_COLUMN_INDEX);
@@ -4184,167 +3914,137 @@ void doBuildDeleteDataBlock(SHashObj* pStDeleted, SSDataBlock* pBlock, void** It
     SColumnInfoData* pCalEdCol = taosArrayGet(pBlock->pDataBlock, CALCULATE_END_TS_COLUMN_INDEX);
     colDataAppendNULL(pCalEdCol, pBlock->info.rows);
     pBlock->info.rows += 1;
-    if (pBlock->info.rows + 1 >= pBlock->info.capacity) {
-      break;
-    }
   }
   if ((*Ite) == NULL) {
-    taosHashClear(pStDeleted);
+    tSimpleHashClear(pStDeleted);
   }
 }
 
-static void rebuildTimeWindow(SStreamSessionAggOperatorInfo* pInfo, SArray* pWinArray, int32_t numOfOutput,
-                              SOperatorInfo* pOperator, SHashObj* pStUpdated) {
-  SExprSupp*     pSup = &pOperator->exprSupp;
-  SExecTaskInfo* pTaskInfo = pOperator->pTaskInfo;
-  int32_t        size = taosArrayGetSize(pWinArray);
+static void rebuildSessionWindow(SOperatorInfo* pOperator, SArray* pWinArray, SSHashObj* pStUpdated) {
+  SExprSupp*                     pSup = &pOperator->exprSupp;
+  SExecTaskInfo*                 pTaskInfo = pOperator->pTaskInfo;
+  int32_t                        size = taosArrayGetSize(pWinArray);
+  SStreamSessionAggOperatorInfo* pInfo = pOperator->info;
+  SStreamAggSupporter*           pAggSup = &pInfo->streamAggSup;
+  int32_t                        numOfOutput = pSup->numOfExprs;
+  int32_t                        numOfChildren = taosArrayGetSize(pInfo->pChildren);
   ASSERT(pInfo->pChildren);
 
   for (int32_t i = 0; i < size; i++) {
-    SResultWindowInfo* pParentWin = taosArrayGet(pWinArray, i);
-    uint64_t           groupId = pParentWin->groupId;
-    int32_t            numOfChildren = taosArrayGetSize(pInfo->pChildren);
+    SSessionKey*      pWinKey = taosArrayGet(pWinArray, i);
+    int32_t           num = 0;
+    SResultWindowInfo parentWin = {0};
     for (int32_t j = 0; j < numOfChildren; j++) {
       SOperatorInfo*                 pChild = taosArrayGetP(pInfo->pChildren, j);
       SStreamSessionAggOperatorInfo* pChInfo = pChild->info;
-      SArray*                        pChWins = getWinInfos(&pChInfo->streamAggSup, groupId);
-      int32_t                        chWinSize = taosArrayGetSize(pChWins);
-      int32_t index = binarySearch(pChWins, chWinSize, pParentWin->win.skey, TSDB_ORDER_DESC, getSessionWindowEndkey);
-      if (index < 0) {
-        index = 0;
-      }
-      for (int32_t k = index; k < chWinSize; k++) {
-        SResultWindowInfo* pChWin = taosArrayGet(pChWins, k);
-        if (pParentWin->win.skey <= pChWin->win.skey && pChWin->win.ekey <= pParentWin->win.ekey) {
-          int32_t            winIndex = 0;
-          SResultWindowInfo* pNewParWin =
-              getSessionTimeWindow(&pInfo->streamAggSup, pChWin->win.skey, pChWin->win.ekey, groupId, 0, &winIndex);
-          SResultRow* pPareResult = NULL;
-          setWindowOutputBuf(pNewParWin, &pPareResult, pSup->pCtx, groupId, numOfOutput, pSup->rowEntryInfoOffset,
-                             &pInfo->streamAggSup, pTaskInfo);
-          SResultRow* pChResult = NULL;
-          setWindowOutputBuf(pChWin, &pChResult, pChild->exprSupp.pCtx, groupId, numOfOutput,
-                             pChild->exprSupp.rowEntryInfoOffset, &pChInfo->streamAggSup, pTaskInfo);
-          updateTimeWindowInfo(&pInfo->twAggSup.timeWindowData, &pNewParWin->win, true);
-          compactFunctions(pSup->pCtx, pChild->exprSupp.pCtx, numOfOutput, pTaskInfo, &pInfo->twAggSup.timeWindowData);
-
-          int32_t winNum = getNumCompactWindow(pInfo->streamAggSup.pCurWins, winIndex, pInfo->gap);
-          if (winNum > 0) {
-            compactTimeWindow(pInfo, winIndex, winNum, groupId, numOfOutput, pStUpdated, NULL, pOperator);
+      SStreamAggSupporter*           pChAggSup = &pChInfo->streamAggSup;
+      SStreamStateCur*               pCur = streamStateSessionGetCur(pChAggSup->pState, pWinKey);
+      SResultRow*                    pResult = NULL;
+      SResultRow*                    pChResult = NULL;
+      while (1) {
+        SResultWindowInfo childWin = {0};
+        childWin.sessionWin = *pWinKey;
+        int32_t code = getSessionWinBuf(pChAggSup, pCur, &childWin);
+        if (code == TSDB_CODE_SUCCESS && pWinKey->win.skey <= childWin.sessionWin.win.skey &&
+            childWin.sessionWin.win.ekey <= pWinKey->win.ekey) {
+          if (num == 0) {
+            setSessionOutputBuf(pAggSup, pWinKey->win.skey, pWinKey->win.ekey, pWinKey->groupId, &parentWin);
+            code = initSessionOutputBuf(&parentWin, &pResult, pSup->pCtx, numOfOutput, pSup->rowEntryInfoOffset);
+            if (code != TSDB_CODE_SUCCESS || pResult == NULL) {
+              break;
+            }
           }
-
-          SFilePage* bufPage = getBufPage(pChInfo->streamAggSup.pResultBuf, pChWin->pos.pageId);
-          releaseBufPage(pChInfo->streamAggSup.pResultBuf, bufPage);
-
-          bufPage = getBufPage(pInfo->streamAggSup.pResultBuf, pNewParWin->pos.pageId);
-          setBufPageDirty(bufPage, true);
-          releaseBufPage(pInfo->streamAggSup.pResultBuf, bufPage);
-          SWinKey value = {.ts = pNewParWin->win.skey, .groupId = groupId};
-          taosHashPut(pStUpdated, &pNewParWin->pos, sizeof(SResultRowPosition), &value, sizeof(SWinKey));
-        } else if (!pChWin->isClosed) {
+          num++;
+          updateTimeWindowInfo(&pInfo->twAggSup.timeWindowData, &parentWin.sessionWin.win, true);
+          initSessionOutputBuf(&childWin, &pChResult, pChild->exprSupp.pCtx, numOfOutput,
+                               pChild->exprSupp.rowEntryInfoOffset);
+          compactFunctions(pSup->pCtx, pChild->exprSupp.pCtx, numOfOutput, pTaskInfo, &pInfo->twAggSup.timeWindowData);
+          compactSessionWindow(pOperator, &parentWin, pStUpdated, NULL);
+          saveResult(parentWin, pStUpdated);
+        } else {
           break;
         }
       }
+      streamStateFreeCur(pCur);
+    }
+    if (num > 0) {
+      saveSessionOutputBuf(pAggSup, &parentWin);
     }
   }
 }
 
-typedef SResultWindowInfo* (*__get_win_info_)(void*);
-SResultWindowInfo* getResWinForSession(void* pData) { return (SResultWindowInfo*)pData; }
-SResultWindowInfo* getResWinForState(void* pData) { return &((SStateWindowInfo*)pData)->winInfo; }
-
-int32_t closeSessionWindow(SHashObj* pHashMap, STimeWindowAggSupp* pTwSup, SArray* pClosed, __get_win_info_ fn,
-                           bool delete, FDelete fp) {
-  // Todo(liuyao) save window to tdb
-  void** pIte = NULL;
-  size_t keyLen = 0;
-  while ((pIte = taosHashIterate(pHashMap, pIte)) != NULL) {
-    uint64_t* pGroupId = taosHashGetKey(pIte, &keyLen);
-    SArray*   pWins = (SArray*)(*pIte);
-    int32_t   size = taosArrayGetSize(pWins);
-    for (int32_t i = 0; i < size; i++) {
-      void*              pWin = taosArrayGet(pWins, i);
-      SResultWindowInfo* pSeWin = fn(pWin);
-      if (isCloseWindow(&pSeWin->win, pTwSup)) {
-        if (!pSeWin->isClosed) {
-          pSeWin->isClosed = true;
-          if (pTwSup->calTrigger == STREAM_TRIGGER_WINDOW_CLOSE && pClosed) {
-            int32_t code = saveResult(pSeWin->win.skey, pSeWin->pos.pageId, pSeWin->pos.offset, *pGroupId, pClosed);
-            if (code != TSDB_CODE_SUCCESS) {
-              return code;
-            }
-            pSeWin->isOutput = true;
-          }
-          if (delete) {
-            deleteWindow(pWins, i, fp);
-            i--;
-            size = taosArrayGetSize(pWins);
-          }
+int32_t closeSessionWindow(SSHashObj* pHashMap, STimeWindowAggSupp* pTwSup, SSHashObj* pClosed) {
+  void*   pIte = NULL;
+  size_t  keyLen = 0;
+  int32_t iter = 0;
+  while ((pIte = tSimpleHashIterate(pHashMap, pIte, &iter)) != NULL) {
+    SResultWindowInfo* pWinInfo = pIte;
+    if (isCloseWindow(&pWinInfo->sessionWin.win, pTwSup)) {
+      if (pTwSup->calTrigger == STREAM_TRIGGER_WINDOW_CLOSE && pClosed) {
+        int32_t code = saveResult(*pWinInfo, pClosed);
+        if (code != TSDB_CODE_SUCCESS) {
+          return code;
         }
-        continue;
       }
-      break;
+      tSimpleHashIterateRemove(pHashMap, &pWinInfo->sessionWin, sizeof(SSessionKey), &pIte, &iter);
     }
   }
   return TSDB_CODE_SUCCESS;
 }
 
-static void closeChildSessionWindow(SArray* pChildren, TSKEY maxTs, bool delete, FDelete fp) {
+static void closeChildSessionWindow(SArray* pChildren, TSKEY maxTs) {
   int32_t size = taosArrayGetSize(pChildren);
   for (int32_t i = 0; i < size; i++) {
     SOperatorInfo*                 pChildOp = taosArrayGetP(pChildren, i);
     SStreamSessionAggOperatorInfo* pChInfo = pChildOp->info;
     pChInfo->twAggSup.maxTs = TMAX(pChInfo->twAggSup.maxTs, maxTs);
-    closeSessionWindow(pChInfo->streamAggSup.pResultRows, &pChInfo->twAggSup, NULL, getResWinForSession, delete, fp);
+    closeSessionWindow(pChInfo->streamAggSup.pResultRows, &pChInfo->twAggSup, NULL);
   }
 }
 
-int32_t getAllSessionWindow(SHashObj* pHashMap, SArray* pClosed, __get_win_info_ fn) {
-  void** pIte = NULL;
-  while ((pIte = taosHashIterate(pHashMap, pIte)) != NULL) {
-    SArray* pWins = (SArray*)(*pIte);
-    int32_t size = taosArrayGetSize(pWins);
-    for (int32_t i = 0; i < size; i++) {
-      void*              pWin = taosArrayGet(pWins, i);
-      SResultWindowInfo* pSeWin = fn(pWin);
-      if (!pSeWin->isClosed) {
-        int32_t code = saveResult(pSeWin->win.skey, pSeWin->pos.pageId, pSeWin->pos.offset, 0, pClosed);
-        pSeWin->isOutput = true;
-      }
-    }
+int32_t getAllSessionWindow(SSHashObj* pHashMap, SSHashObj* pStUpdated) {
+  void*   pIte = NULL;
+  int32_t iter = 0;
+  while ((pIte = tSimpleHashIterate(pHashMap, pIte, &iter)) != NULL) {
+    SResultWindowInfo* pWinInfo = *(void**)pIte;
+    saveResult(*pWinInfo, pStUpdated);
   }
   return TSDB_CODE_SUCCESS;
 }
 
-static void copyDeleteWindowInfo(SArray* pResWins, SHashObj* pStDeleted) {
+static void copyDeleteWindowInfo(SArray* pResWins, SSHashObj* pStDeleted) {
   int32_t size = taosArrayGetSize(pResWins);
   for (int32_t i = 0; i < size; i++) {
-    SResultWindowInfo* pWinInfo = taosArrayGet(pResWins, i);
-    SWinKey            res = {.ts = pWinInfo->win.skey, .groupId = pWinInfo->groupId};
-    taosHashPut(pStDeleted, &res, sizeof(SWinKey), &res, sizeof(SWinKey));
+    SSessionKey* pWinKey = taosArrayGet(pResWins, i);
+    if (!pWinKey) continue;
+    SSessionKey winInfo = *pWinKey;
+    winInfo.win.ekey = winInfo.win.skey;
+    tSimpleHashPut(pStDeleted, &winInfo, sizeof(SSessionKey), NULL, 0);
   }
 }
 
-static void removeSessionResults(SHashObj* pHashMap, SArray* pWins) {
-  int32_t size = taosArrayGetSize(pWins);
-  for (int32_t i = 0; i < size; i++) {
-    SResultWindowInfo* pWin = taosArrayGet(pWins, i);
-    taosHashRemove(pHashMap, &pWin->pos, sizeof(SResultRowPosition));
-  }
+void initGroupResInfoFromArrayList(SGroupResInfo* pGroupResInfo, SArray* pArrayList) {
+  pGroupResInfo->pRows = pArrayList;
+  pGroupResInfo->index = 0;
+  ASSERT(pGroupResInfo->index <= getNumOfTotalRes(pGroupResInfo));
 }
 
-static void removeSessionDeleteResults(SArray* update, SHashObj* pStDeleted) {
-  int32_t size = taosHashGetSize(pStDeleted);
-  if (size == 0) {
+void doBuildSessionResult(SOperatorInfo* pOperator, SStreamState* pState, SGroupResInfo* pGroupResInfo,
+                          SSDataBlock* pBlock) {
+  SExecTaskInfo* pTaskInfo = pOperator->pTaskInfo;
+  // set output datablock version
+  pBlock->info.version = pTaskInfo->version;
+
+  blockDataCleanup(pBlock);
+  if (!hasRemainResults(pGroupResInfo)) {
+    taosArrayDestroy(pGroupResInfo->pRows);
+    pGroupResInfo->pRows = NULL;
     return;
   }
 
-  int32_t num = taosArrayGetSize(update);
-  for (int32_t i = 0; i < num; i++) {
-    SResKeyPos* pos = taosArrayGetP(update, i);
-    SWinKey     winKey = {.ts = *(int64_t*)pos->key, .groupId = pos->groupId};
-    taosHashRemove(pStDeleted, &winKey, sizeof(SWinKey));
-  }
+  // clear the existed group id
+  pBlock->info.groupId = 0;
+  buildSessionResultDataBlock(pTaskInfo, pState, pBlock, &pOperator->exprSupp, pGroupResInfo);
 }
 
 static SSDataBlock* doStreamSessionAgg(SOperatorInfo* pOperator) {
@@ -4352,6 +4052,7 @@ static SSDataBlock* doStreamSessionAgg(SOperatorInfo* pOperator) {
   SStreamSessionAggOperatorInfo* pInfo = pOperator->info;
   SOptrBasicInfo*                pBInfo = &pInfo->binfo;
   TSKEY                          maxTs = INT64_MIN;
+  SStreamAggSupporter*           pAggSup = &pInfo->streamAggSup;
   if (pOperator->status == OP_EXEC_DONE) {
     return NULL;
   } else if (pOperator->status == OP_RES_TO_RETURN) {
@@ -4360,18 +4061,20 @@ static SSDataBlock* doStreamSessionAgg(SOperatorInfo* pOperator) {
       printDataBlock(pInfo->pDelRes, IS_FINAL_OP(pInfo) ? "final session" : "single session");
       return pInfo->pDelRes;
     }
-    doBuildStreamResBlock(pOperator, pBInfo, &pInfo->groupResInfo, pInfo->streamAggSup.pResultBuf);
-    if (pBInfo->pRes->info.rows == 0 || !hasRemainResults(&pInfo->groupResInfo)) {
-      doSetOperatorCompleted(pOperator);
+    doBuildSessionResult(pOperator, pAggSup->pState, &pInfo->groupResInfo, pBInfo->pRes);
+    if (pBInfo->pRes->info.rows > 0) {
+      printDataBlock(pBInfo->pRes, IS_FINAL_OP(pInfo) ? "final session" : "single session");
+      return pBInfo->pRes;
     }
-    printDataBlock(pBInfo->pRes, IS_FINAL_OP(pInfo) ? "final session" : "single session");
-    return pBInfo->pRes->info.rows == 0 ? NULL : pBInfo->pRes;
+
+    doSetOperatorCompleted(pOperator);
+    return NULL;
   }
 
   _hash_fn_t     hashFn = taosGetDefaultHashFunction(TSDB_DATA_TYPE_BINARY);
-  SHashObj*      pStUpdated = taosHashInit(64, hashFn, true, HASH_NO_LOCK);
+  SSHashObj*     pStUpdated = tSimpleHashInit(64, hashFn);
   SOperatorInfo* downstream = pOperator->pDownstream[0];
-  SArray*        pUpdated = taosArrayInit(16, POINTER_BYTES);  // SResKeyPos
+  SArray*        pUpdated = taosArrayInit(16, sizeof(SSessionKey));  // SResKeyPos
   while (1) {
     SSDataBlock* pBlock = downstream->fpSet.getNextFn(downstream);
     if (pBlock == NULL) {
@@ -4385,38 +4088,25 @@ static SSDataBlock* doStreamSessionAgg(SOperatorInfo* pOperator) {
       /*printf("\n\n put tbname %s\n\n", pBlock->info.parTbName);*/
     }
 
-    if (pBlock->info.type == STREAM_CLEAR) {
-      SArray* pWins = taosArrayInit(16, sizeof(SResultWindowInfo));
-      doClearSessionWindows(&pInfo->streamAggSup, &pOperator->exprSupp, pBlock, START_TS_COLUMN_INDEX,
-                            pOperator->exprSupp.numOfExprs, 0, pWins);
-      if (IS_FINAL_OP(pInfo)) {
-        int32_t                        childIndex = getChildIndex(pBlock);
-        SOperatorInfo*                 pChildOp = taosArrayGetP(pInfo->pChildren, childIndex);
-        SStreamSessionAggOperatorInfo* pChildInfo = pChildOp->info;
-        doClearSessionWindows(&pChildInfo->streamAggSup, &pChildOp->exprSupp, pBlock, START_TS_COLUMN_INDEX,
-                              pChildOp->exprSupp.numOfExprs, 0, NULL);
-        rebuildTimeWindow(pInfo, pWins, pOperator->exprSupp.numOfExprs, pOperator, pStUpdated);
-      }
-      taosArrayDestroy(pWins);
-      continue;
-    } else if (pBlock->info.type == STREAM_DELETE_DATA || pBlock->info.type == STREAM_DELETE_RESULT) {
-      SArray* pWins = taosArrayInit(16, sizeof(SResultWindowInfo));
+    if (pBlock->info.type == STREAM_DELETE_DATA || pBlock->info.type == STREAM_DELETE_RESULT ||
+        pBlock->info.type == STREAM_CLEAR) {
+      SArray* pWins = taosArrayInit(16, sizeof(SSessionKey));
       // gap must be 0
-      doDeleteTimeWindows(&pInfo->streamAggSup, pBlock, 0, pWins, NULL);
+      doDeleteTimeWindows(pAggSup, pBlock, pWins);
+      removeSessionResults(pStUpdated, pWins);
       if (IS_FINAL_OP(pInfo)) {
         int32_t                        childIndex = getChildIndex(pBlock);
         SOperatorInfo*                 pChildOp = taosArrayGetP(pInfo->pChildren, childIndex);
         SStreamSessionAggOperatorInfo* pChildInfo = pChildOp->info;
         // gap must be 0
-        doDeleteTimeWindows(&pChildInfo->streamAggSup, pBlock, 0, NULL, NULL);
-        rebuildTimeWindow(pInfo, pWins, pOperator->exprSupp.numOfExprs, pOperator, pStUpdated);
+        doDeleteTimeWindows(&pChildInfo->streamAggSup, pBlock, NULL);
+        rebuildSessionWindow(pOperator, pWins, pStUpdated);
       }
       copyDeleteWindowInfo(pWins, pInfo->pStDeleted);
-      removeSessionResults(pStUpdated, pWins);
       taosArrayDestroy(pWins);
       continue;
     } else if (pBlock->info.type == STREAM_GET_ALL) {
-      getAllSessionWindow(pInfo->streamAggSup.pResultRows, pUpdated, getResWinForSession);
+      getAllSessionWindow(pAggSup->pResultRows, pStUpdated);
       continue;
     }
 
@@ -4424,7 +4114,6 @@ static SSDataBlock* doStreamSessionAgg(SOperatorInfo* pOperator) {
       SExprSupp* pExprSup = &pInfo->scalarSupp;
       projectApplyFunctions(pExprSup->pExprInfo, pBlock, pBlock, pExprSup->pCtx, pExprSup->numOfExprs, NULL);
     }
-    // the pDataBlock are always the same one, no need to call this again
     setInputDataBlock(pOperator, pSup->pCtx, pBlock, TSDB_ORDER_ASC, MAIN_SCAN, true);
     doStreamSessionAggImpl(pOperator, pBlock, pStUpdated, pInfo->pStDeleted, IS_FINAL_OP(pInfo));
     if (IS_FINAL_OP(pInfo)) {
@@ -4451,39 +4140,121 @@ static SSDataBlock* doStreamSessionAgg(SOperatorInfo* pOperator) {
   // restore the value
   pOperator->status = OP_RES_TO_RETURN;
 
-  closeSessionWindow(pInfo->streamAggSup.pResultRows, &pInfo->twAggSup, pUpdated, getResWinForSession,
-                     pInfo->ignoreExpiredData, NULL);
-  closeChildSessionWindow(pInfo->pChildren, pInfo->twAggSup.maxTs, pInfo->ignoreExpiredData, NULL);
+  closeSessionWindow(pAggSup->pResultRows, &pInfo->twAggSup, pStUpdated);
+  closeChildSessionWindow(pInfo->pChildren, pInfo->twAggSup.maxTs);
   copyUpdateResult(pStUpdated, pUpdated);
-  removeSessionDeleteResults(pUpdated, pInfo->pStDeleted);
-  taosHashCleanup(pStUpdated);
-
-  finalizeUpdatedResult(pSup->numOfExprs, pInfo->streamAggSup.pResultBuf, pUpdated, pSup->rowEntryInfoOffset);
-  initMultiResInfoFromArrayList(&pInfo->groupResInfo, pUpdated);
+  removeSessionResults(pInfo->pStDeleted, pUpdated);
+  tSimpleHashCleanup(pStUpdated);
+  initGroupResInfoFromArrayList(&pInfo->groupResInfo, pUpdated);
   blockDataEnsureCapacity(pInfo->binfo.pRes, pOperator->resultInfo.capacity);
+
   doBuildDeleteDataBlock(pInfo->pStDeleted, pInfo->pDelRes, &pInfo->pDelIterator);
   if (pInfo->pDelRes->info.rows > 0) {
     printDataBlock(pInfo->pDelRes, IS_FINAL_OP(pInfo) ? "final session" : "single session");
     return pInfo->pDelRes;
   }
-  doBuildStreamResBlock(pOperator, &pInfo->binfo, &pInfo->groupResInfo, pInfo->streamAggSup.pResultBuf);
-  printDataBlock(pBInfo->pRes, IS_FINAL_OP(pInfo) ? "final session" : "single session");
-  return pBInfo->pRes->info.rows == 0 ? NULL : pBInfo->pRes;
+
+  doBuildSessionResult(pOperator, pAggSup->pState, &pInfo->groupResInfo, pBInfo->pRes);
+  if (pBInfo->pRes->info.rows > 0) {
+    printDataBlock(pBInfo->pRes, IS_FINAL_OP(pInfo) ? "final session" : "single session");
+    return pBInfo->pRes;
+  }
+
+  doSetOperatorCompleted(pOperator);
+  return NULL;
+}
+
+SOperatorInfo* createStreamSessionAggOperatorInfo(SOperatorInfo* downstream, SPhysiNode* pPhyNode,
+                                                  SExecTaskInfo* pTaskInfo) {
+  SSessionWinodwPhysiNode*       pSessionNode = (SSessionWinodwPhysiNode*)pPhyNode;
+  int32_t                        numOfCols = 0;
+  int32_t                        code = TSDB_CODE_OUT_OF_MEMORY;
+  SStreamSessionAggOperatorInfo* pInfo = taosMemoryCalloc(1, sizeof(SStreamSessionAggOperatorInfo));
+  SOperatorInfo*                 pOperator = taosMemoryCalloc(1, sizeof(SOperatorInfo));
+  if (pInfo == NULL || pOperator == NULL) {
+    goto _error;
+  }
+
+  pOperator->pTaskInfo = pTaskInfo;
+
+  initResultSizeInfo(&pOperator->resultInfo, 4096);
+  if (pSessionNode->window.pExprs != NULL) {
+    int32_t    numOfScalar = 0;
+    SExprInfo* pScalarExprInfo = createExprInfo(pSessionNode->window.pExprs, NULL, &numOfScalar);
+    code = initExprSupp(&pInfo->scalarSupp, pScalarExprInfo, numOfScalar);
+    if (code != TSDB_CODE_SUCCESS) {
+      goto _error;
+    }
+  }
+  SExprSupp* pSup = &pOperator->exprSupp;
+
+  SExprInfo*   pExprInfo = createExprInfo(pSessionNode->window.pFuncs, NULL, &numOfCols);
+  SSDataBlock* pResBlock = createResDataBlock(pPhyNode->pOutputDataBlockDesc);
+  code = initBasicInfoEx(&pInfo->binfo, pSup, pExprInfo, numOfCols, pResBlock);
+  if (code != TSDB_CODE_SUCCESS) {
+    goto _error;
+  }
+
+  code = initStreamAggSupporter(&pInfo->streamAggSup, pSup->pCtx, numOfCols, pSessionNode->gap,
+                                pTaskInfo->streamInfo.pState, 0, 0);
+  if (code != TSDB_CODE_SUCCESS) {
+    goto _error;
+  }
+
+  pInfo->twAggSup = (STimeWindowAggSupp){
+      .waterMark = pSessionNode->window.watermark,
+      .calTrigger = pSessionNode->window.triggerType,
+      .maxTs = INT64_MIN,
+      .minTs = INT64_MAX,
+  };
+
+  initExecTimeWindowInfo(&pInfo->twAggSup.timeWindowData, &pTaskInfo->window);
+
+  pInfo->primaryTsIndex = ((SColumnNode*)pSessionNode->window.pTspk)->slotId;
+  if (pSessionNode->window.pTsEnd) {
+    pInfo->endTsIndex = ((SColumnNode*)pSessionNode->window.pTsEnd)->slotId;
+  }
+  pInfo->binfo.pRes = pResBlock;
+  pInfo->order = TSDB_ORDER_ASC;
+  _hash_fn_t hashFn = taosGetDefaultHashFunction(TSDB_DATA_TYPE_BINARY);
+  pInfo->pStDeleted = tSimpleHashInit(64, hashFn);
+  pInfo->pDelIterator = NULL;
+  pInfo->pDelRes = createSpecialDataBlock(STREAM_DELETE_RESULT);
+  pInfo->pChildren = NULL;
+  pInfo->isFinal = false;
+  pInfo->pPhyNode = pPhyNode;
+  pInfo->ignoreExpiredData = pSessionNode->window.igExpired;
+  pInfo->pGroupIdTbNameMap =
+      taosHashInit(1024, taosGetDefaultHashFunction(TSDB_DATA_TYPE_UBIGINT), false, HASH_NO_LOCK);
+
+  pOperator->name = "StreamSessionWindowAggOperator";
+  pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_STREAM_SESSION;
+  pOperator->blocking = true;
+  pOperator->status = OP_NOT_OPENED;
+  pOperator->info = pInfo;
+  pOperator->fpSet =
+      createOperatorFpSet(operatorDummyOpenFn, doStreamSessionAgg, NULL, NULL, destroyStreamSessionAggOperatorInfo,
+                          aggEncodeResultRow, aggDecodeResultRow, NULL);
+  if (downstream) {
+    initDownStream(downstream, &pInfo->streamAggSup, pInfo->twAggSup.waterMark, pOperator->operatorType,
+                   pInfo->primaryTsIndex);
+    code = appendDownstream(pOperator, &downstream, 1);
+  }
+  return pOperator;
+
+_error:
+  if (pInfo != NULL) {
+    destroyStreamSessionAggOperatorInfo(pInfo);
+  }
+
+  taosMemoryFreeClear(pOperator);
+  pTaskInfo->code = code;
+  return NULL;
 }
 
 static void clearStreamSessionOperator(SStreamSessionAggOperatorInfo* pInfo) {
-  void** pIte = NULL;
-  while ((pIte = taosHashIterate(pInfo->streamAggSup.pResultRows, pIte)) != NULL) {
-    SArray* pWins = (SArray*)(*pIte);
-    int32_t size = taosArrayGetSize(pWins);
-    for (int32_t i = 0; i < size; i++) {
-      SResultWindowInfo* pWin = (SResultWindowInfo*)taosArrayGet(pWins, i);
-      pWin->pos.pageId = -1;
-      pWin->pos.offset = -1;
-    }
-  }
-  clearDiskbasedBuf(pInfo->streamAggSup.pResultBuf);
-  pInfo->streamAggSup.currentPageId = -1;
+  tSimpleHashClear(pInfo->streamAggSup.pResultRows);
+  streamStateSessionClear(pInfo->streamAggSup.pState);
 }
 
 static SSDataBlock* doStreamSessionSemiAgg(SOperatorInfo* pOperator) {
@@ -4491,13 +4262,14 @@ static SSDataBlock* doStreamSessionSemiAgg(SOperatorInfo* pOperator) {
   SOptrBasicInfo*                pBInfo = &pInfo->binfo;
   TSKEY                          maxTs = INT64_MIN;
   SExprSupp*                     pSup = &pOperator->exprSupp;
+  SStreamAggSupporter*           pAggSup = &pInfo->streamAggSup;
 
   if (pOperator->status == OP_EXEC_DONE) {
     return NULL;
   }
 
   {
-    doBuildStreamResBlock(pOperator, pBInfo, &pInfo->groupResInfo, pInfo->streamAggSup.pResultBuf);
+    doBuildSessionResult(pOperator, pAggSup->pState, &pInfo->groupResInfo, pBInfo->pRes);
     if (pBInfo->pRes->info.rows > 0) {
       printDataBlock(pBInfo->pRes, "semi session");
       return pBInfo->pRes;
@@ -4505,30 +4277,23 @@ static SSDataBlock* doStreamSessionSemiAgg(SOperatorInfo* pOperator) {
 
     doBuildDeleteDataBlock(pInfo->pStDeleted, pInfo->pDelRes, &pInfo->pDelIterator);
     if (pInfo->pDelRes->info.rows > 0) {
-      printDataBlock(pInfo->pDelRes, "semi session");
+      printDataBlock(pInfo->pDelRes, "semi session delete");
       return pInfo->pDelRes;
     }
 
-    if (pInfo->pUpdateRes->info.rows > 0 && pInfo->returnUpdate) {
-      pInfo->returnUpdate = false;
-      // process the rest of the data
-      printDataBlock(pInfo->pUpdateRes, "semi session");
-      return pInfo->pUpdateRes;
-    }
-
     if (pOperator->status == OP_RES_TO_RETURN) {
       clearFunctionContext(&pOperator->exprSupp);
       // semi interval operator clear disk buffer
       clearStreamSessionOperator(pInfo);
-      pOperator->status = OP_EXEC_DONE;
+      doSetOperatorCompleted(pOperator);
       return NULL;
     }
   }
 
   _hash_fn_t     hashFn = taosGetDefaultHashFunction(TSDB_DATA_TYPE_BINARY);
-  SHashObj*      pStUpdated = taosHashInit(64, hashFn, true, HASH_NO_LOCK);
+  SSHashObj*     pStUpdated = tSimpleHashInit(64, hashFn);
   SOperatorInfo* downstream = pOperator->pDownstream[0];
-  SArray*        pUpdated = taosArrayInit(16, POINTER_BYTES);
+  SArray*        pUpdated = taosArrayInit(16, sizeof(SSessionKey));
   while (1) {
     SSDataBlock* pBlock = downstream->fpSet.getNextFn(downstream);
     if (pBlock == NULL) {
@@ -4544,24 +4309,17 @@ static SSDataBlock* doStreamSessionSemiAgg(SOperatorInfo* pOperator) {
       /*printf("\n\n put tbname %s\n\n", pBlock->info.parTbName);*/
     }
 
-    if (pBlock->info.type == STREAM_CLEAR) {
-      SArray* pWins = taosArrayInit(16, sizeof(SResultWindowInfo));
-      doClearSessionWindows(&pInfo->streamAggSup, pSup, pBlock, START_TS_COLUMN_INDEX, pSup->numOfExprs, 0, pWins);
-      removeSessionResults(pStUpdated, pWins);
-      taosArrayDestroy(pWins);
-      copyDataBlock(pInfo->pUpdateRes, pBlock);
-      pInfo->returnUpdate = true;
-      break;
-    } else if (pBlock->info.type == STREAM_DELETE_DATA || pBlock->info.type == STREAM_DELETE_RESULT) {
+    if (pBlock->info.type == STREAM_DELETE_DATA || pBlock->info.type == STREAM_DELETE_RESULT ||
+        pBlock->info.type == STREAM_CLEAR) {
       // gap must be 0
       SArray* pWins = taosArrayInit(16, sizeof(SResultWindowInfo));
-      doDeleteTimeWindows(&pInfo->streamAggSup, pBlock, 0, pWins, NULL);
-      copyDeleteWindowInfo(pWins, pInfo->pStDeleted);
+      doDeleteTimeWindows(&pInfo->streamAggSup, pBlock, pWins);
       removeSessionResults(pStUpdated, pWins);
+      copyDeleteWindowInfo(pWins, pInfo->pStDeleted);
       taosArrayDestroy(pWins);
       break;
     } else if (pBlock->info.type == STREAM_GET_ALL) {
-      getAllSessionWindow(pInfo->streamAggSup.pResultRows, pUpdated, getResWinForSession);
+      getAllSessionWindow(pInfo->streamAggSup.pResultRows, pStUpdated);
       continue;
     }
 
@@ -4579,15 +4337,13 @@ static SSDataBlock* doStreamSessionSemiAgg(SOperatorInfo* pOperator) {
   pBInfo->pRes->info.watermark = pInfo->twAggSup.maxTs;
 
   copyUpdateResult(pStUpdated, pUpdated);
-  removeSessionDeleteResults(pUpdated, pInfo->pStDeleted);
-  taosHashCleanup(pStUpdated);
+  removeSessionResults(pInfo->pStDeleted, pUpdated);
+  tSimpleHashCleanup(pStUpdated);
 
-  finalizeUpdatedResult(pOperator->exprSupp.numOfExprs, pInfo->streamAggSup.pResultBuf, pUpdated,
-                        pSup->rowEntryInfoOffset);
-  initMultiResInfoFromArrayList(&pInfo->groupResInfo, pUpdated);
+  initGroupResInfoFromArrayList(&pInfo->groupResInfo, pUpdated);
   blockDataEnsureCapacity(pBInfo->pRes, pOperator->resultInfo.capacity);
 
-  doBuildStreamResBlock(pOperator, pBInfo, &pInfo->groupResInfo, pInfo->streamAggSup.pResultBuf);
+  doBuildSessionResult(pOperator, pAggSup->pState, &pInfo->groupResInfo, pBInfo->pRes);
   if (pBInfo->pRes->info.rows > 0) {
     printDataBlock(pBInfo->pRes, "semi session");
     return pBInfo->pRes;
@@ -4595,18 +4351,14 @@ static SSDataBlock* doStreamSessionSemiAgg(SOperatorInfo* pOperator) {
 
   doBuildDeleteDataBlock(pInfo->pStDeleted, pInfo->pDelRes, &pInfo->pDelIterator);
   if (pInfo->pDelRes->info.rows > 0) {
-    printDataBlock(pInfo->pDelRes, "semi session");
+    printDataBlock(pInfo->pDelRes, "semi session delete");
     return pInfo->pDelRes;
   }
 
-  if (pInfo->pUpdateRes->info.rows > 0 && pInfo->returnUpdate) {
-    pInfo->returnUpdate = false;
-    // process the rest of the data
-    printDataBlock(pInfo->pUpdateRes, "semi session");
-    return pInfo->pUpdateRes;
-  }
-
-  pOperator->status = OP_EXEC_DONE;
+  clearFunctionContext(&pOperator->exprSupp);
+  // semi interval operator clear disk buffer
+  clearStreamSessionOperator(pInfo);
+  doSetOperatorCompleted(pOperator);
   return NULL;
 }
 
@@ -4639,11 +4391,14 @@ SOperatorInfo* createStreamFinalSessionAggOperatorInfo(SOperatorInfo* downstream
   if (numOfChild > 0) {
     pInfo->pChildren = taosArrayInit(numOfChild, sizeof(void*));
     for (int32_t i = 0; i < numOfChild; i++) {
-      SOperatorInfo* pChild = createStreamFinalSessionAggOperatorInfo(NULL, pPhyNode, pTaskInfo, 0);
-      if (pChild == NULL) {
+      SOperatorInfo* pChildOp = createStreamFinalSessionAggOperatorInfo(NULL, pPhyNode, pTaskInfo, 0);
+      if (pChildOp == NULL) {
         goto _error;
       }
-      taosArrayPush(pInfo->pChildren, &pChild);
+      SStreamSessionAggOperatorInfo* pChInfo = pChildOp->info;
+      pChInfo->twAggSup.calTrigger = STREAM_TRIGGER_AT_ONCE;
+      streamStateSetNumber(pChInfo->streamAggSup.pState, i);
+      taosArrayPush(pInfo->pChildren, &pChildOp);
     }
   }
   return pOperator;
@@ -4652,7 +4407,6 @@ _error:
   if (pInfo != NULL) {
     destroyStreamSessionAggOperatorInfo(pInfo);
   }
-
   taosMemoryFreeClear(pOperator);
   pTaskInfo->code = code;
   return NULL;
@@ -4661,7 +4415,7 @@ _error:
 void destroyStreamStateOperatorInfo(void* param) {
   SStreamStateAggOperatorInfo* pInfo = (SStreamStateAggOperatorInfo*)param;
   cleanupBasicInfo(&pInfo->binfo);
-  destroyStateStreamAggSupporter(&pInfo->streamAggSup);
+  destroyStreamAggSupporter(&pInfo->streamAggSup);
   cleanupGroupResInfo(&pInfo->groupResInfo);
   if (pInfo->pChildren != NULL) {
     int32_t size = taosArrayGetSize(pInfo->pChildren);
@@ -4674,151 +4428,75 @@ void destroyStreamStateOperatorInfo(void* param) {
   }
   colDataDestroy(&pInfo->twAggSup.timeWindowData);
   blockDataDestroy(pInfo->pDelRes);
-  taosHashCleanup(pInfo->pSeDeleted);
+  tSimpleHashCleanup(pInfo->pSeDeleted);
   taosHashCleanup(pInfo->pGroupIdTbNameMap);
-  destroySqlFunctionCtx(pInfo->pDummyCtx, 0);
-
   taosMemoryFreeClear(param);
 }
 
-int64_t getStateWinTsKey(void* data, int32_t index) {
-  SStateWindowInfo* pStateWin = taosArrayGet(data, index);
-  return pStateWin->winInfo.win.ekey;
-}
-
-SStateWindowInfo* addNewStateWindow(SArray* pWinInfos, TSKEY ts, char* pKeyData, SColumn* pCol) {
-  SStateWindowInfo win = {
-      .stateKey.bytes = pCol->bytes,
-      .stateKey.type = pCol->type,
-      .stateKey.pData = taosMemoryCalloc(1, pCol->bytes),
-      .winInfo.pos.offset = -1,
-      .winInfo.pos.pageId = -1,
-      .winInfo.win.skey = ts,
-      .winInfo.win.ekey = ts,
-      .winInfo.isOutput = false,
-      .winInfo.isClosed = false,
-  };
-  if (IS_VAR_DATA_TYPE(win.stateKey.type)) {
-    varDataCopy(win.stateKey.pData, pKeyData);
-  } else {
-    memcpy(win.stateKey.pData, pKeyData, win.stateKey.bytes);
-  }
-  return taosArrayPush(pWinInfos, &win);
-}
-
-SStateWindowInfo* insertNewStateWindow(SArray* pWinInfos, TSKEY ts, char* pKeyData, int32_t index, SColumn* pCol) {
-  SStateWindowInfo win = {
-      .stateKey.bytes = pCol->bytes,
-      .stateKey.type = pCol->type,
-      .stateKey.pData = taosMemoryCalloc(1, pCol->bytes),
-      .winInfo.pos.offset = -1,
-      .winInfo.pos.pageId = -1,
-      .winInfo.win.skey = ts,
-      .winInfo.win.ekey = ts,
-      .winInfo.isOutput = false,
-      .winInfo.isClosed = false,
-  };
-  if (IS_VAR_DATA_TYPE(win.stateKey.type)) {
-    varDataCopy(win.stateKey.pData, pKeyData);
-  } else {
-    memcpy(win.stateKey.pData, pKeyData, win.stateKey.bytes);
-  }
-  return taosArrayInsert(pWinInfos, index, &win);
-}
-
 bool isTsInWindow(SStateWindowInfo* pWin, TSKEY ts) {
-  if (pWin->winInfo.win.skey <= ts && ts <= pWin->winInfo.win.ekey) {
+  if (pWin->winInfo.sessionWin.win.skey <= ts && ts <= pWin->winInfo.sessionWin.win.ekey) {
     return true;
   }
   return false;
 }
 
 bool isEqualStateKey(SStateWindowInfo* pWin, char* pKeyData) {
-  return pKeyData && compareVal(pKeyData, &pWin->stateKey);
+  return pKeyData && compareVal(pKeyData, pWin->pStateKey);
 }
 
-SStateWindowInfo* getStateWindowByTs(SStreamAggSupporter* pAggSup, TSKEY ts, uint64_t groupId, int32_t* pIndex) {
-  SArray* pWinInfos = getWinInfos(pAggSup, groupId);
-  pAggSup->pCurWins = pWinInfos;
-  int32_t           size = taosArrayGetSize(pWinInfos);
-  int32_t           index = binarySearch(pWinInfos, size, ts, TSDB_ORDER_DESC, getStateWinTsKey);
-  SStateWindowInfo* pWin = NULL;
-  if (index >= 0) {
-    pWin = taosArrayGet(pWinInfos, index);
-    if (isTsInWindow(pWin, ts)) {
-      *pIndex = index;
-      return pWin;
-    }
-  }
-
-  if (index + 1 < size) {
-    pWin = taosArrayGet(pWinInfos, index + 1);
-    if (isTsInWindow(pWin, ts)) {
-      *pIndex = index + 1;
-      return pWin;
-    }
-  }
-  *pIndex = 0;
-  return NULL;
+bool compareStateKey(void* data, void* key) {
+  SStateKeys* stateKey = (SStateKeys*)key;
+  stateKey->pData = (char*)key + sizeof(SStateKeys);
+  return compareVal(data, stateKey);
 }
 
-SStateWindowInfo* getStateWindow(SStreamAggSupporter* pAggSup, TSKEY ts, uint64_t groupId, char* pKeyData,
-                                 SColumn* pCol, int32_t* pIndex) {
-  SArray* pWinInfos = getWinInfos(pAggSup, groupId);
-  pAggSup->pCurWins = pWinInfos;
-  int32_t size = taosArrayGetSize(pWinInfos);
-  if (size == 0) {
-    *pIndex = 0;
-    return addNewStateWindow(pWinInfos, ts, pKeyData, pCol);
-  }
-  int32_t           index = binarySearch(pWinInfos, size, ts, TSDB_ORDER_DESC, getStateWinTsKey);
-  SStateWindowInfo* pWin = NULL;
-  if (index >= 0) {
-    pWin = taosArrayGet(pWinInfos, index);
-    if (isTsInWindow(pWin, ts)) {
-      *pIndex = index;
-      return pWin;
+void setStateOutputBuf(SStreamAggSupporter* pAggSup, TSKEY ts, uint64_t groupId, char* pKeyData,
+                       SStateWindowInfo* pCurWin, SStateWindowInfo* pNextWin) {
+  int32_t size = pAggSup->resultRowSize;
+  pCurWin->winInfo.sessionWin.groupId = groupId;
+  pCurWin->winInfo.sessionWin.win.skey = ts;
+  pCurWin->winInfo.sessionWin.win.ekey = ts;
+  int32_t code =
+      streamStateStateAddIfNotExist(pAggSup->pState, &pCurWin->winInfo.sessionWin, pKeyData, pAggSup->stateKeySize,
+                                    compareStateKey, &pCurWin->winInfo.pOutputBuf, &size);
+  pCurWin->pStateKey =
+      (SStateKeys*)((char*)pCurWin->winInfo.pOutputBuf + (pAggSup->resultRowSize - pAggSup->stateKeySize));
+  pCurWin->pStateKey->bytes = pAggSup->stateKeySize - sizeof(SStateKeys);
+  pCurWin->pStateKey->type = pAggSup->stateKeyType;
+  pCurWin->pStateKey->pData = (char*)pCurWin->pStateKey + sizeof(SStateKeys);
+  pCurWin->pStateKey->isNull = false;
+
+  if (code == TSDB_CODE_SUCCESS) {
+    pCurWin->winInfo.isOutput = true;
+  } else {
+    if (IS_VAR_DATA_TYPE(pAggSup->stateKeyType)) {
+      varDataCopy(pCurWin->pStateKey->pData, pKeyData);
+    } else {
+      memcpy(pCurWin->pStateKey->pData, pKeyData, pCurWin->pStateKey->bytes);
     }
   }
 
-  if (index + 1 < size) {
-    pWin = taosArrayGet(pWinInfos, index + 1);
-    if (isTsInWindow(pWin, ts) || isEqualStateKey(pWin, pKeyData)) {
-      *pIndex = index + 1;
-      return pWin;
-    }
+  pNextWin->winInfo.sessionWin = pCurWin->winInfo.sessionWin;
+  pNextWin->winInfo.pOutputBuf = NULL;
+  SStreamStateCur* pCur = streamStateSessionSeekKeyNext(pAggSup->pState, &pCurWin->winInfo.sessionWin);
+  code = streamStateSessionGetKVByCur(pCur, &pNextWin->winInfo.sessionWin, NULL, 0);
+  if (code != TSDB_CODE_SUCCESS) {
+    SET_SESSION_WIN_INVALID(pNextWin->winInfo);
   }
-
-  if (index >= 0) {
-    pWin = taosArrayGet(pWinInfos, index);
-    if (isEqualStateKey(pWin, pKeyData)) {
-      *pIndex = index;
-      return pWin;
-    }
-  }
-
-  if (index == size - 1) {
-    *pIndex = taosArrayGetSize(pWinInfos);
-    return addNewStateWindow(pWinInfos, ts, pKeyData, pCol);
-  }
-  *pIndex = index + 1;
-  return insertNewStateWindow(pWinInfos, ts, pKeyData, index + 1, pCol);
+  streamStateFreeCur(pCur);
 }
 
-int32_t updateStateWindowInfo(SArray* pWinInfos, int32_t winIndex, TSKEY* pTs, uint64_t groupId,
+int32_t updateStateWindowInfo(SStateWindowInfo* pWinInfo, SStateWindowInfo* pNextWin, TSKEY* pTs, uint64_t groupId,
                               SColumnInfoData* pKeyCol, int32_t rows, int32_t start, bool* allEqual,
-                              SHashObj* pSeDeleted) {
+                              SSHashObj* pResultRows, SSHashObj* pSeUpdated, SSHashObj* pSeDeleted) {
   *allEqual = true;
-  SStateWindowInfo* pWinInfo = taosArrayGet(pWinInfos, winIndex);
   for (int32_t i = start; i < rows; ++i) {
     char* pKeyData = colDataGetData(pKeyCol, i);
     if (!isTsInWindow(pWinInfo, pTs[i])) {
       if (isEqualStateKey(pWinInfo, pKeyData)) {
-        int32_t size = taosArrayGetSize(pWinInfos);
-        if (winIndex + 1 < size) {
-          SStateWindowInfo* pNextWin = taosArrayGet(pWinInfos, winIndex + 1);
+        if (IS_VALID_SESSION_WIN(pNextWin->winInfo)) {
           // ts belongs to the next window
-          if (pTs[i] >= pNextWin->winInfo.win.skey) {
+          if (pTs[i] >= pNextWin->winInfo.sessionWin.win.skey) {
             return i - start;
           }
         }
@@ -4826,15 +4504,15 @@ int32_t updateStateWindowInfo(SArray* pWinInfos, int32_t winIndex, TSKEY* pTs, u
         return i - start;
       }
     }
-    if (pWinInfo->winInfo.win.skey > pTs[i]) {
+
+    if (pWinInfo->winInfo.sessionWin.win.skey > pTs[i]) {
       if (pSeDeleted && pWinInfo->winInfo.isOutput) {
-        SWinKey res = {.ts = pWinInfo->winInfo.win.skey, .groupId = groupId};
-        taosHashPut(pSeDeleted, &res, sizeof(SWinKey), &res, sizeof(SWinKey));
-        pWinInfo->winInfo.isOutput = false;
+        saveDeleteRes(pSeDeleted, pWinInfo->winInfo.sessionWin);
       }
-      pWinInfo->winInfo.win.skey = pTs[i];
+      removeSessionResult(pSeUpdated, pResultRows, pWinInfo->winInfo.sessionWin);
+      pWinInfo->winInfo.sessionWin.win.skey = pTs[i];
     }
-    pWinInfo->winInfo.win.ekey = TMAX(pWinInfo->winInfo.win.ekey, pTs[i]);
+    pWinInfo->winInfo.sessionWin.win.ekey = TMAX(pWinInfo->winInfo.sessionWin.win.ekey, pTs[i]);
     if (!isEqualStateKey(pWinInfo, pKeyData)) {
       *allEqual = false;
     }
@@ -4842,36 +4520,13 @@ int32_t updateStateWindowInfo(SArray* pWinInfos, int32_t winIndex, TSKEY* pTs, u
   return rows - start;
 }
 
-static void doClearStateWindows(SStreamAggSupporter* pAggSup, SSDataBlock* pBlock, SHashObj* pSeUpdated,
-                                SHashObj* pSeDeleted) {
-  SColumnInfoData* pTsColInfo = taosArrayGet(pBlock->pDataBlock, START_TS_COLUMN_INDEX);
-  SColumnInfoData* pGroupColInfo = taosArrayGet(pBlock->pDataBlock, GROUPID_COLUMN_INDEX);
-  TSKEY*           tsCol = (TSKEY*)pTsColInfo->pData;
-  bool             allEqual = false;
-  int32_t          step = 1;
-  uint64_t*        gpCol = (uint64_t*)pGroupColInfo->pData;
-  for (int32_t i = 0; i < pBlock->info.rows; i += step) {
-    int32_t           winIndex = 0;
-    SStateWindowInfo* pCurWin = getStateWindowByTs(pAggSup, tsCol[i], gpCol[i], &winIndex);
-    if (!pCurWin) {
-      continue;
-    }
-    updateSessionWindowInfo(&pCurWin->winInfo, tsCol, NULL, 0, pBlock->info.rows, i, 0, NULL);
-    taosHashRemove(pSeUpdated, &pCurWin->winInfo.pos, sizeof(SResultRowPosition));
-    deleteWindow(pAggSup->pCurWins, winIndex, destroyStateWinInfo);
-  }
-}
-
-static void doStreamStateAggImpl(SOperatorInfo* pOperator, SSDataBlock* pSDataBlock, SHashObj* pSeUpdated,
-                                 SHashObj* pStDeleted) {
+static void doStreamStateAggImpl(SOperatorInfo* pOperator, SSDataBlock* pSDataBlock, SSHashObj* pSeUpdated,
+                                 SSHashObj* pStDeleted) {
   SExecTaskInfo*               pTaskInfo = pOperator->pTaskInfo;
   SStreamStateAggOperatorInfo* pInfo = pOperator->info;
-  bool                         masterScan = true;
   int32_t                      numOfOutput = pOperator->exprSupp.numOfExprs;
   int64_t                      groupId = pSDataBlock->info.groupId;
   int64_t                      code = TSDB_CODE_SUCCESS;
-  int32_t                      step = 1;
-  bool                         ascScan = true;
   TSKEY*                       tsCols = NULL;
   SResultRow*                  pResult = NULL;
   int32_t                      winRows = 0;
@@ -4883,39 +4538,44 @@ static void doStreamStateAggImpl(SOperatorInfo* pOperator, SSDataBlock* pSDataBl
   }
 
   SStreamAggSupporter* pAggSup = &pInfo->streamAggSup;
-  blockDataEnsureCapacity(pAggSup->pScanBlock, pSDataBlock->info.rows);
+  int32_t              rows = pSDataBlock->info.rows;
+  blockDataEnsureCapacity(pAggSup->pScanBlock, rows);
   SColumnInfoData* pKeyColInfo = taosArrayGet(pSDataBlock->pDataBlock, pInfo->stateCol.slotId);
-  for (int32_t i = 0; i < pSDataBlock->info.rows; i += winRows) {
+  for (int32_t i = 0; i < rows; i += winRows) {
     if (pInfo->ignoreExpiredData && isOverdue(tsCols[i], &pInfo->twAggSup)) {
       i++;
       continue;
     }
-    char*             pKeyData = colDataGetData(pKeyColInfo, i);
-    int32_t           winIndex = 0;
-    bool              allEqual = true;
-    SStateWindowInfo* pCurWin = getStateWindow(pAggSup, tsCols[i], groupId, pKeyData, &pInfo->stateCol, &winIndex);
-    winRows = updateStateWindowInfo(pAggSup->pCurWins, winIndex, tsCols, groupId, pKeyColInfo, pSDataBlock->info.rows,
-                                    i, &allEqual, pStDeleted);
+    char*            pKeyData = colDataGetData(pKeyColInfo, i);
+    int32_t          winIndex = 0;
+    bool             allEqual = true;
+    SStateWindowInfo curWin = {0};
+    SStateWindowInfo nextWin = {0};
+    setStateOutputBuf(pAggSup, tsCols[i], groupId, pKeyData, &curWin, &nextWin);
+    setSessionWinOutputInfo(pSeUpdated, &curWin.winInfo);
+    winRows = updateStateWindowInfo(&curWin, &nextWin, tsCols, groupId, pKeyColInfo, rows, i, &allEqual,
+                                    pAggSup->pResultRows, pSeUpdated, pStDeleted);
     if (!allEqual) {
       uint64_t uid = 0;
-      appendOneRowToStreamSpecialBlock(pAggSup->pScanBlock, &pCurWin->winInfo.win.skey, &pCurWin->winInfo.win.ekey,
-                                       &uid, &groupId, NULL);
-      taosHashRemove(pSeUpdated, &pCurWin->winInfo.pos, sizeof(SResultRowPosition));
-      deleteWindow(pAggSup->pCurWins, winIndex, destroyStateWinInfo);
+      appendOneRowToStreamSpecialBlock(pAggSup->pScanBlock, &curWin.winInfo.sessionWin.win.skey,
+                                       &curWin.winInfo.sessionWin.win.ekey, &uid, &groupId, NULL);
+      tSimpleHashRemove(pSeUpdated, &curWin.winInfo.sessionWin, sizeof(SSessionKey));
+      doDeleteSessionWindow(pAggSup, &curWin.winInfo.sessionWin);
+      releaseOutputBuf(pAggSup->pState, NULL, (SResultRow*)curWin.winInfo.pOutputBuf);
       continue;
     }
-    code = doOneStateWindowAgg(pInfo, pSDataBlock, &pCurWin->winInfo, &pResult, i, winRows, numOfOutput, pOperator);
+    code = doOneWindowAggImpl(&pInfo->twAggSup.timeWindowData, &curWin.winInfo, &pResult, i, winRows, rows, numOfOutput,
+                              pOperator);
     if (code != TSDB_CODE_SUCCESS || pResult == NULL) {
       T_LONG_JMP(pTaskInfo->env, TSDB_CODE_QRY_OUT_OF_MEMORY);
     }
-    pCurWin->winInfo.isClosed = false;
+    saveSessionOutputBuf(pAggSup, &curWin.winInfo);
+
     if (pInfo->twAggSup.calTrigger == STREAM_TRIGGER_AT_ONCE) {
-      SWinKey value = {.ts = pCurWin->winInfo.win.skey, .groupId = groupId};
-      code = taosHashPut(pSeUpdated, &pCurWin->winInfo.pos, sizeof(SResultRowPosition), &value, sizeof(SWinKey));
+      code = saveResult(curWin.winInfo, pSeUpdated);
       if (code != TSDB_CODE_SUCCESS) {
         T_LONG_JMP(pTaskInfo->env, TSDB_CODE_QRY_OUT_OF_MEMORY);
       }
-      pCurWin->winInfo.isOutput = true;
     }
   }
 }
@@ -4932,21 +4592,24 @@ static SSDataBlock* doStreamStateAgg(SOperatorInfo* pOperator) {
   if (pOperator->status == OP_RES_TO_RETURN) {
     doBuildDeleteDataBlock(pInfo->pSeDeleted, pInfo->pDelRes, &pInfo->pDelIterator);
     if (pInfo->pDelRes->info.rows > 0) {
-      printDataBlock(pInfo->pDelRes, "single state");
+      printDataBlock(pInfo->pDelRes, "single state delete");
       return pInfo->pDelRes;
     }
-    doBuildStreamResBlock(pOperator, pBInfo, &pInfo->groupResInfo, pInfo->streamAggSup.pResultBuf);
-    if (pBInfo->pRes->info.rows == 0 || !hasRemainResults(&pInfo->groupResInfo)) {
-      doSetOperatorCompleted(pOperator);
+
+    doBuildSessionResult(pOperator, pInfo->streamAggSup.pState, &pInfo->groupResInfo, pBInfo->pRes);
+    if (pBInfo->pRes->info.rows > 0) {
+      printDataBlock(pBInfo->pRes, "single state");
+      return pBInfo->pRes;
     }
-    printDataBlock(pBInfo->pRes, "single state");
-    return pBInfo->pRes->info.rows == 0 ? NULL : pBInfo->pRes;
+
+    doSetOperatorCompleted(pOperator);
+    return NULL;
   }
 
   _hash_fn_t     hashFn = taosGetDefaultHashFunction(TSDB_DATA_TYPE_BINARY);
-  SHashObj*      pSeUpdated = taosHashInit(64, hashFn, true, HASH_NO_LOCK);
+  SSHashObj*     pSeUpdated = tSimpleHashInit(64, hashFn);
   SOperatorInfo* downstream = pOperator->pDownstream[0];
-  SArray*        pUpdated = taosArrayInit(16, POINTER_BYTES);
+  SArray*        pUpdated = taosArrayInit(16, sizeof(SSessionKey));
   while (1) {
     SSDataBlock* pBlock = downstream->fpSet.getNextFn(downstream);
     if (pBlock == NULL) {
@@ -4954,18 +4617,16 @@ static SSDataBlock* doStreamStateAgg(SOperatorInfo* pOperator) {
     }
     printDataBlock(pBlock, "single state recv");
 
-    if (pBlock->info.type == STREAM_CLEAR) {
-      doClearStateWindows(&pInfo->streamAggSup, pBlock, pSeUpdated, pInfo->pSeDeleted);
-      continue;
-    } else if (pBlock->info.type == STREAM_DELETE_DATA || pBlock->info.type == STREAM_DELETE_RESULT) {
-      SArray* pWins = taosArrayInit(16, sizeof(SResultWindowInfo));
-      doDeleteTimeWindows(&pInfo->streamAggSup, pBlock, 0, pWins, destroyStateWinInfo);
-      copyDeleteWindowInfo(pWins, pInfo->pSeDeleted);
+    if (pBlock->info.type == STREAM_DELETE_DATA || pBlock->info.type == STREAM_DELETE_RESULT ||
+        pBlock->info.type == STREAM_CLEAR) {
+      SArray* pWins = taosArrayInit(16, sizeof(SSessionKey));
+      doDeleteTimeWindows(&pInfo->streamAggSup, pBlock, pWins);
       removeSessionResults(pSeUpdated, pWins);
+      copyDeleteWindowInfo(pWins, pInfo->pSeDeleted);
       taosArrayDestroy(pWins);
       continue;
     } else if (pBlock->info.type == STREAM_GET_ALL) {
-      getAllSessionWindow(pInfo->streamAggSup.pResultRows, pUpdated, getResWinForState);
+      getAllSessionWindow(pInfo->streamAggSup.pResultRows, pSeUpdated);
       continue;
     }
 
@@ -4982,28 +4643,27 @@ static SSDataBlock* doStreamStateAgg(SOperatorInfo* pOperator) {
   // restore the value
   pOperator->status = OP_RES_TO_RETURN;
 
-  closeSessionWindow(pInfo->streamAggSup.pResultRows, &pInfo->twAggSup, pUpdated, getResWinForState,
-                     pInfo->ignoreExpiredData, destroyStateWinInfo);
-  // closeChildSessionWindow(pInfo->pChildren, pInfo->twAggSup.maxTs, pInfo->ignoreExpiredData, destroyStateWinInfo);
+  closeSessionWindow(pInfo->streamAggSup.pResultRows, &pInfo->twAggSup, pSeUpdated);
   copyUpdateResult(pSeUpdated, pUpdated);
-  taosHashCleanup(pSeUpdated);
+  removeSessionResults(pInfo->pSeDeleted, pUpdated);
+  tSimpleHashCleanup(pSeUpdated);
 
-  finalizeUpdatedResult(pOperator->exprSupp.numOfExprs, pInfo->streamAggSup.pResultBuf, pUpdated,
-                        pSup->rowEntryInfoOffset);
-  initMultiResInfoFromArrayList(&pInfo->groupResInfo, pUpdated);
+  initGroupResInfoFromArrayList(&pInfo->groupResInfo, pUpdated);
   blockDataEnsureCapacity(pInfo->binfo.pRes, pOperator->resultInfo.capacity);
+
   doBuildDeleteDataBlock(pInfo->pSeDeleted, pInfo->pDelRes, &pInfo->pDelIterator);
   if (pInfo->pDelRes->info.rows > 0) {
-    printDataBlock(pInfo->pDelRes, "single state");
+    printDataBlock(pInfo->pDelRes, "single state delete");
     return pInfo->pDelRes;
   }
-  doBuildStreamResBlock(pOperator, &pInfo->binfo, &pInfo->groupResInfo, pInfo->streamAggSup.pResultBuf);
-  printDataBlock(pBInfo->pRes, "single state");
-  return pBInfo->pRes->info.rows == 0 ? NULL : pBInfo->pRes;
-}
 
-int32_t initStateAggSupporter(SStreamAggSupporter* pSup, const char* pKey, SqlFunctionCtx* pCtx, int32_t numOfOutput) {
-  return initStreamAggSupporter(pSup, pKey, pCtx, numOfOutput, sizeof(SStateWindowInfo));
+  doBuildSessionResult(pOperator, pInfo->streamAggSup.pState, &pInfo->groupResInfo, pBInfo->pRes);
+  if (pBInfo->pRes->info.rows > 0) {
+    printDataBlock(pBInfo->pRes, "single state");
+    return pBInfo->pRes;
+  }
+  doSetOperatorCompleted(pOperator);
+  return NULL;
 }
 
 SOperatorInfo* createStreamStateAggOperatorInfo(SOperatorInfo* downstream, SPhysiNode* pPhyNode,
@@ -5020,11 +4680,6 @@ SOperatorInfo* createStreamStateAggOperatorInfo(SOperatorInfo* downstream, SPhys
     goto _error;
   }
 
-  SExprSupp* pSup = &pOperator->exprSupp;
-
-  int32_t    numOfCols = 0;
-  SExprInfo* pExprInfo = createExprInfo(pStateNode->window.pFuncs, NULL, &numOfCols);
-
   pInfo->stateCol = extractColumnFromColumnNode(pColNode);
   initResultSizeInfo(&pOperator->resultInfo, 4096);
   if (pStateNode->window.pExprs != NULL) {
@@ -5036,7 +4691,6 @@ SOperatorInfo* createStreamStateAggOperatorInfo(SOperatorInfo* downstream, SPhys
     }
   }
 
-  initResultRowInfo(&pInfo->binfo.resultRowInfo);
   pInfo->twAggSup = (STimeWindowAggSupp){
       .waterMark = pStateNode->window.watermark,
       .calTrigger = pStateNode->window.triggerType,
@@ -5045,27 +4699,25 @@ SOperatorInfo* createStreamStateAggOperatorInfo(SOperatorInfo* downstream, SPhys
   };
   initExecTimeWindowInfo(&pInfo->twAggSup.timeWindowData, &pTaskInfo->window);
 
+  SExprSupp*   pSup = &pOperator->exprSupp;
+  int32_t      numOfCols = 0;
+  SExprInfo*   pExprInfo = createExprInfo(pStateNode->window.pFuncs, NULL, &numOfCols);
   SSDataBlock* pResBlock = createResDataBlock(pPhyNode->pOutputDataBlockDesc);
   code = initBasicInfoEx(&pInfo->binfo, pSup, pExprInfo, numOfCols, pResBlock);
   if (code != TSDB_CODE_SUCCESS) {
     goto _error;
   }
-
-  code = initStateAggSupporter(&pInfo->streamAggSup, "StreamStateAggOperatorInfo", pSup->pCtx, numOfCols);
+  int32_t keySize = sizeof(SStateKeys) + pColNode->node.resType.bytes;
+  int16_t type = pColNode->node.resType.type;
+  code = initStreamAggSupporter(&pInfo->streamAggSup, pSup->pCtx, numOfCols, 0, pTaskInfo->streamInfo.pState, keySize,
+                                type);
   if (code != TSDB_CODE_SUCCESS) {
     goto _error;
   }
 
-  pInfo->pDummyCtx = (SqlFunctionCtx*)taosMemoryCalloc(numOfCols, sizeof(SqlFunctionCtx));
-  if (pInfo->pDummyCtx == NULL) {
-    goto _error;
-  }
-
-  initDummyFunction(pInfo->pDummyCtx, pSup->pCtx, numOfCols);
   pInfo->primaryTsIndex = tsSlotId;
-  pInfo->order = TSDB_ORDER_ASC;
   _hash_fn_t hashFn = taosGetDefaultHashFunction(TSDB_DATA_TYPE_BINARY);
-  pInfo->pSeDeleted = taosHashInit(64, hashFn, true, HASH_NO_LOCK);
+  pInfo->pSeDeleted = tSimpleHashInit(64, hashFn);
   pInfo->pDelIterator = NULL;
   pInfo->pDelRes = createSpecialDataBlock(STREAM_DELETE_RESULT);
   pInfo->pChildren = NULL;
@@ -5082,7 +4734,7 @@ SOperatorInfo* createStreamStateAggOperatorInfo(SOperatorInfo* downstream, SPhys
   pOperator->info = pInfo;
   pOperator->fpSet = createOperatorFpSet(operatorDummyOpenFn, doStreamStateAgg, NULL, NULL,
                                          destroyStreamStateOperatorInfo, aggEncodeResultRow, aggDecodeResultRow, NULL);
-  initDownStream(downstream, &pInfo->streamAggSup, 0, pInfo->twAggSup.waterMark, pOperator->operatorType,
+  initDownStream(downstream, &pInfo->streamAggSup, pInfo->twAggSup.waterMark, pOperator->operatorType,
                  pInfo->primaryTsIndex);
   code = appendDownstream(pOperator, &downstream, 1);
   if (code != TSDB_CODE_SUCCESS) {
@@ -5619,8 +5271,8 @@ SOperatorInfo* createMergeIntervalOperatorInfo(SOperatorInfo* downstream, SMerge
     goto _error;
   }
 
-  int32_t      num = 0;
-  SExprInfo*   pExprInfo = createExprInfo(pIntervalPhyNode->window.pFuncs, NULL, &num);
+  int32_t    num = 0;
+  SExprInfo* pExprInfo = createExprInfo(pIntervalPhyNode->window.pFuncs, NULL, &num);
 
   SInterval interval = {.interval = pIntervalPhyNode->interval,
                         .sliding = pIntervalPhyNode->sliding,
diff --git a/source/libs/function/src/builtinsimpl.c b/source/libs/function/src/builtinsimpl.c
index 7077a9b780..49df3aa136 100644
--- a/source/libs/function/src/builtinsimpl.c
+++ b/source/libs/function/src/builtinsimpl.c
@@ -2527,6 +2527,8 @@ int32_t apercentileFunction(SqlFunctionCtx* pCtx) {
 
   int32_t start = pInput->startRowIndex;
   if (pInfo->algo == APERCT_ALGO_TDIGEST) {
+    buildTDigestInfo(pInfo);
+    tdigestAutoFill(pInfo->pTDigest, COMPRESSION);
     for (int32_t i = start; i < pInput->numOfRows + start; ++i) {
       if (colDataIsNull_f(pCol->nullbitmap, i)) {
         continue;
@@ -2540,12 +2542,11 @@ int32_t apercentileFunction(SqlFunctionCtx* pCtx) {
       tdigestAdd(pInfo->pTDigest, v, w);
     }
   } else {
-    qDebug("%s before add %d elements into histogram, total:%d, numOfEntry:%d, pHisto:%p, elems: %p", __FUNCTION__,
-           numOfElems, pInfo->pHisto->numOfElems, pInfo->pHisto->numOfEntries, pInfo->pHisto, pInfo->pHisto->elems);
-
     // might be a race condition here that pHisto can be overwritten or setup function
     // has not been called, need to relink the buffer pHisto points to.
     buildHistogramInfo(pInfo);
+    qDebug("%s before add %d elements into histogram, total:%d, numOfEntry:%d, pHisto:%p, elems: %p", __FUNCTION__,
+           numOfElems, pInfo->pHisto->numOfElems, pInfo->pHisto->numOfEntries, pInfo->pHisto, pInfo->pHisto->elems);
     for (int32_t i = start; i < pInput->numOfRows + start; ++i) {
       if (colDataIsNull_f(pCol->nullbitmap, i)) {
         continue;
@@ -2579,8 +2580,9 @@ static void apercentileTransferInfo(SAPercentileInfo* pInput, SAPercentileInfo*
 
     buildTDigestInfo(pOutput);
     TDigest* pTDigest = pOutput->pTDigest;
+    tdigestAutoFill(pTDigest, COMPRESSION);
 
-    if (pTDigest->num_centroids <= 0) {
+    if (pTDigest->num_centroids <= 0 && pTDigest->num_buffered_pts == 0) {
       memcpy(pTDigest, pInput->pTDigest, (size_t)TDIGEST_SIZE(COMPRESSION));
       tdigestAutoFill(pTDigest, COMPRESSION);
     } else {
@@ -2652,6 +2654,7 @@ int32_t apercentileFinalize(SqlFunctionCtx* pCtx, SSDataBlock* pBlock) {
 
   if (pInfo->algo == APERCT_ALGO_TDIGEST) {
     buildTDigestInfo(pInfo);
+    tdigestAutoFill(pInfo->pTDigest, COMPRESSION);
     if (pInfo->pTDigest->size > 0) {
       pInfo->result = tdigestQuantile(pInfo->pTDigest, pInfo->percent / 100);
     } else {  // no need to free
diff --git a/source/libs/stream/src/streamState.c b/source/libs/stream/src/streamState.c
index c0ed135d74..6991f00bb7 100644
--- a/source/libs/stream/src/streamState.c
+++ b/source/libs/stream/src/streamState.c
@@ -24,6 +24,40 @@ typedef struct SStateKey {
   int64_t opNum;
 } SStateKey;
 
+typedef struct SStateSessionKey {
+  SSessionKey key;
+  int64_t     opNum;
+} SStateSessionKey;
+
+static inline int sessionKeyCmpr(const SSessionKey* pWin1, const SSessionKey* pWin2) {
+  if (pWin1->groupId > pWin2->groupId) {
+    return 1;
+  } else if (pWin1->groupId < pWin2->groupId) {
+    return -1;
+  }
+
+  if (pWin1->win.skey > pWin2->win.ekey) {
+    return 1;
+  } else if (pWin1->win.ekey < pWin2->win.skey) {
+    return -1;
+  }
+
+  return 0;
+}
+
+static inline int stateSessionKeyCmpr(const void* pKey1, int kLen1, const void* pKey2, int kLen2) {
+  SStateSessionKey* pWin1 = (SStateSessionKey*)pKey1;
+  SStateSessionKey* pWin2 = (SStateSessionKey*)pKey2;
+
+  if (pWin1->opNum > pWin2->opNum) {
+    return 1;
+  } else if (pWin1->opNum < pWin2->opNum) {
+    return -1;
+  }
+
+  return sessionKeyCmpr(&pWin1->key, &pWin2->key);
+}
+
 static inline int stateKeyCmpr(const void* pKey1, int kLen1, const void* pKey2, int kLen2) {
   SStateKey* pWin1 = (SStateKey*)pKey1;
   SStateKey* pWin2 = (SStateKey*)pKey2;
@@ -79,6 +113,11 @@ SStreamState* streamStateOpen(char* path, SStreamTask* pTask, bool specPath, int
     goto _err;
   }
 
+  if (tdbTbOpen("session.state.db", sizeof(SStateSessionKey), -1, stateSessionKeyCmpr, pState->db,
+                &pState->pSessionStateDb) < 0) {
+    goto _err;
+  }
+
   if (tdbTbOpen("func.state.db", sizeof(STupleKey), -1, STupleKeyCmpr, pState->db, &pState->pFuncStateDb) < 0) {
     goto _err;
   }
@@ -95,6 +134,7 @@ _err:
   tdbTbClose(pState->pStateDb);
   tdbTbClose(pState->pFuncStateDb);
   tdbTbClose(pState->pFillStateDb);
+  tdbTbClose(pState->pSessionStateDb);
   tdbClose(pState->db);
   taosMemoryFree(pState);
   return NULL;
@@ -105,6 +145,7 @@ void streamStateClose(SStreamState* pState) {
   tdbTbClose(pState->pStateDb);
   tdbTbClose(pState->pFuncStateDb);
   tdbTbClose(pState->pFillStateDb);
+  tdbTbClose(pState->pSessionStateDb);
   tdbClose(pState->db);
 
   taosMemoryFree(pState);
@@ -241,11 +282,11 @@ SStreamStateCur* streamStateGetCur(SStreamState* pState, const SWinKey* key) {
   if (pCur == NULL) return NULL;
   tdbTbcOpen(pState->pStateDb, &pCur->pCur, NULL);
 
-  int32_t   c;
+  int32_t   c = 0;
   SStateKey sKey = {.key = *key, .opNum = pState->number};
   tdbTbcMoveTo(pCur->pCur, &sKey, sizeof(SStateKey), &c);
   if (c != 0) {
-    taosMemoryFree(pCur);
+    streamStateFreeCur(pCur);
     return NULL;
   }
   pCur->number = pState->number;
@@ -257,7 +298,7 @@ SStreamStateCur* streamStateFillGetCur(SStreamState* pState, const SWinKey* key)
   if (pCur == NULL) return NULL;
   tdbTbcOpen(pState->pFillStateDb, &pCur->pCur, NULL);
 
-  int32_t c;
+  int32_t c = 0;
   tdbTbcMoveTo(pCur->pCur, key, sizeof(SWinKey), &c);
   if (c != 0) {
     streamStateFreeCur(pCur);
@@ -348,21 +389,21 @@ SStreamStateCur* streamStateSeekKeyNext(SStreamState* pState, const SWinKey* key
   }
   pCur->number = pState->number;
   if (tdbTbcOpen(pState->pStateDb, &pCur->pCur, NULL) < 0) {
-    taosMemoryFree(pCur);
+    streamStateFreeCur(pCur);
     return NULL;
   }
 
   SStateKey sKey = {.key = *key, .opNum = pState->number};
-  int32_t   c;
+  int32_t   c = 0;
   if (tdbTbcMoveTo(pCur->pCur, &sKey, sizeof(SStateKey), &c) < 0) {
     tdbTbcClose(pCur->pCur);
-    taosMemoryFree(pCur);
+    streamStateFreeCur(pCur);
     return NULL;
   }
   if (c > 0) return pCur;
 
   if (tdbTbcMoveToNext(pCur->pCur) < 0) {
-    taosMemoryFree(pCur);
+    streamStateFreeCur(pCur);
     return NULL;
   }
 
@@ -375,20 +416,20 @@ SStreamStateCur* streamStateFillSeekKeyNext(SStreamState* pState, const SWinKey*
     return NULL;
   }
   if (tdbTbcOpen(pState->pFillStateDb, &pCur->pCur, NULL) < 0) {
-    taosMemoryFree(pCur);
+    streamStateFreeCur(pCur);
     return NULL;
   }
 
-  int32_t c;
+  int32_t c = 0;
   if (tdbTbcMoveTo(pCur->pCur, key, sizeof(SWinKey), &c) < 0) {
     tdbTbcClose(pCur->pCur);
-    taosMemoryFree(pCur);
+    streamStateFreeCur(pCur);
     return NULL;
   }
   if (c > 0) return pCur;
 
   if (tdbTbcMoveToNext(pCur->pCur) < 0) {
-    taosMemoryFree(pCur);
+    streamStateFreeCur(pCur);
     return NULL;
   }
 
@@ -401,20 +442,20 @@ SStreamStateCur* streamStateFillSeekKeyPrev(SStreamState* pState, const SWinKey*
     return NULL;
   }
   if (tdbTbcOpen(pState->pFillStateDb, &pCur->pCur, NULL) < 0) {
-    taosMemoryFree(pCur);
+    streamStateFreeCur(pCur);
     return NULL;
   }
 
-  int32_t c;
+  int32_t c = 0;
   if (tdbTbcMoveTo(pCur->pCur, key, sizeof(SWinKey), &c) < 0) {
     tdbTbcClose(pCur->pCur);
-    taosMemoryFree(pCur);
+    streamStateFreeCur(pCur);
     return NULL;
   }
   if (c < 0) return pCur;
 
   if (tdbTbcMoveToPrev(pCur->pCur) < 0) {
-    taosMemoryFree(pCur);
+    streamStateFreeCur(pCur);
     return NULL;
   }
 
@@ -445,3 +486,229 @@ void streamStateFreeCur(SStreamStateCur* pCur) {
 }
 
 void streamFreeVal(void* val) { tdbFree(val); }
+
+int32_t streamStateSessionPut(SStreamState* pState, const SSessionKey* key, const void* value, int32_t vLen) {
+  SStateSessionKey sKey = {.key = *key, .opNum = pState->number};
+  return tdbTbUpsert(pState->pSessionStateDb, &sKey, sizeof(SStateSessionKey), value, vLen, &pState->txn);
+}
+
+SStreamStateCur* streamStateSessionGetRanomCur(SStreamState* pState, const SSessionKey* key) {
+  SStreamStateCur* pCur = taosMemoryCalloc(1, sizeof(SStreamStateCur));
+  if (pCur == NULL) return NULL;
+  tdbTbcOpen(pState->pSessionStateDb, &pCur->pCur, NULL);
+
+  int32_t          c = 0;
+  SStateSessionKey sKey = {.key = *key, .opNum = pState->number};
+  tdbTbcMoveTo(pCur->pCur, &sKey, sizeof(SStateSessionKey), &c);
+  if (c != 0) {
+    streamStateFreeCur(pCur);
+    return NULL;
+  }
+  pCur->number = pState->number;
+  return pCur;
+}
+
+int32_t streamStateSessionGet(SStreamState* pState, SSessionKey* key, void** pVal, int32_t* pVLen) {
+  SStreamStateCur* pCur = streamStateSessionGetRanomCur(pState, key);
+  void*            tmp = NULL;
+  if (streamStateSessionGetKVByCur(pCur, key, (const void**)&tmp, pVLen) == 0) {
+    *pVal = tdbRealloc(NULL, *pVLen);
+    memcpy(*pVal, tmp, *pVLen);
+    streamStateFreeCur(pCur);
+    return 0;
+  }
+  streamStateFreeCur(pCur);
+  return -1;
+}
+
+int32_t streamStateSessionDel(SStreamState* pState, const SSessionKey* key) {
+  SStateSessionKey sKey = {.key = *key, .opNum = pState->number};
+  return tdbTbDelete(pState->pSessionStateDb, &sKey, sizeof(SStateSessionKey), &pState->txn);
+}
+
+SStreamStateCur* streamStateSessionSeekKeyPrev(SStreamState* pState, const SSessionKey* key) {
+  SStreamStateCur* pCur = taosMemoryCalloc(1, sizeof(SStreamStateCur));
+  if (pCur == NULL) {
+    return NULL;
+  }
+  pCur->number = pState->number;
+  if (tdbTbcOpen(pState->pSessionStateDb, &pCur->pCur, NULL) < 0) {
+    streamStateFreeCur(pCur);
+    return NULL;
+  }
+
+  SStateSessionKey sKey = {.key = *key, .opNum = pState->number};
+  int32_t          c = 0;
+  if (tdbTbcMoveTo(pCur->pCur, &sKey, sizeof(SStateSessionKey), &c) < 0) {
+    tdbTbcClose(pCur->pCur);
+    streamStateFreeCur(pCur);
+    return NULL;
+  }
+  if (c > 0) return pCur;
+
+  if (tdbTbcMoveToPrev(pCur->pCur) < 0) {
+    streamStateFreeCur(pCur);
+    return NULL;
+  }
+
+  return pCur;
+}
+
+SStreamStateCur* streamStateSessionSeekKeyNext(SStreamState* pState, const SSessionKey* key) {
+  SStreamStateCur* pCur = taosMemoryCalloc(1, sizeof(SStreamStateCur));
+  if (pCur == NULL) {
+    return NULL;
+  }
+  pCur->number = pState->number;
+  if (tdbTbcOpen(pState->pSessionStateDb, &pCur->pCur, NULL) < 0) {
+    streamStateFreeCur(pCur);
+    return NULL;
+  }
+
+  SStateSessionKey sKey = {.key = *key, .opNum = pState->number};
+  int32_t          c = 0;
+  if (tdbTbcMoveTo(pCur->pCur, &sKey, sizeof(SStateSessionKey), &c) < 0) {
+    tdbTbcClose(pCur->pCur);
+    streamStateFreeCur(pCur);
+    return NULL;
+  }
+  if (c > 0) return pCur;
+
+  if (tdbTbcMoveToNext(pCur->pCur) < 0) {
+    streamStateFreeCur(pCur);
+    return NULL;
+  }
+
+  return pCur;
+}
+
+int32_t streamStateSessionGetKVByCur(SStreamStateCur* pCur, SSessionKey* pKey, const void** pVal, int32_t* pVLen) {
+  if (!pCur) {
+    return -1;
+  }
+  const SStateSessionKey* pKTmp = NULL;
+  int32_t                 kLen;
+  if (tdbTbcGet(pCur->pCur, (const void**)&pKTmp, &kLen, pVal, pVLen) < 0) {
+    return -1;
+  }
+  if (pKTmp->opNum != pCur->number) {
+    return -1;
+  }
+  if (pKey->groupId != 0 && pKey->groupId != pKTmp->key.groupId) {
+    return -1;
+  }
+  *pKey = pKTmp->key;
+  return 0;
+}
+
+int32_t streamStateSessionClear(SStreamState* pState) {
+  SSessionKey key = {.win.skey = 0, .win.ekey = 0, .groupId = 0};
+  streamStateSessionPut(pState, &key, NULL, 0);
+  SStreamStateCur* pCur = streamStateSessionSeekKeyNext(pState, &key);
+  while (1) {
+    SSessionKey delKey = {0};
+    void*       buf = NULL;
+    int32_t     size = 0;
+    int32_t     code = streamStateSessionGetKVByCur(pCur, &delKey, buf, &size);
+    if (code == 0) {
+      memset(buf, 0, size);
+      streamStateSessionPut(pState, &delKey, buf, size);
+    } else {
+      break;
+    }
+    streamStateCurNext(pState, pCur);
+  }
+  streamStateFreeCur(pCur);
+  streamStateSessionDel(pState, &key);
+  return 0;
+}
+
+SStreamStateCur* streamStateSessionGetCur(SStreamState* pState, const SSessionKey* key) {
+  SStreamStateCur* pCur = streamStateSessionGetRanomCur(pState, key);
+  SSessionKey      resKey = *key;
+  while (1) {
+    streamStateCurPrev(pState, pCur);
+    SSessionKey tmpKey = *key;
+    int32_t     code = streamStateSessionGetKVByCur(pCur, &tmpKey, NULL, 0);
+    if (code == TSDB_CODE_SUCCESS && sessionKeyCmpr(key, &tmpKey) == 0) {
+      resKey = tmpKey;
+    } else {
+      break;
+    }
+  }
+  streamStateFreeCur(pCur);
+  return streamStateSessionGetRanomCur(pState, &resKey);
+}
+
+int32_t streamStateSessionAddIfNotExist(SStreamState* pState, SSessionKey* key, void** pVal, int32_t* pVLen) {
+  // todo refactor
+  SStreamStateCur* pCur = streamStateSessionGetCur(pState, key);
+  int32_t          size = *pVLen;
+  void*            tmp = NULL;
+  *pVal = tdbRealloc(NULL, size);
+  memset(*pVal, 0, size);
+  if (streamStateSessionGetKVByCur(pCur, key, (const void**)&tmp, pVLen) == 0) {
+    memcpy(*pVal, tmp, *pVLen);
+    streamStateFreeCur(pCur);
+    return 0;
+  }
+  streamStateFreeCur(pCur);
+  return 1;
+}
+
+int32_t streamStateStateAddIfNotExist(SStreamState* pState, SSessionKey* key, char* pKeyData, int32_t keyDataLen,
+                                      state_key_cmpr_fn fn, void** pVal, int32_t* pVLen) {
+  // todo refactor
+  int32_t     res = TSDB_CODE_SUCCESS;
+  SSessionKey tmpKey = *key;
+  int32_t     valSize = *pVLen;
+  void*       tmp = tdbRealloc(NULL, valSize);
+  if (!tmp) {
+    return -1;
+  }
+
+  SStreamStateCur* pCur = streamStateSessionGetRanomCur(pState, key);
+  int32_t          code = streamStateSessionGetKVByCur(pCur, key, (const void**)pVal, pVLen);
+  if (code == TSDB_CODE_SUCCESS) {
+    memcpy(tmp, *pVal, valSize);
+    *pVal = tmp;
+    streamStateFreeCur(pCur);
+    return res;
+  }
+  streamStateFreeCur(pCur);
+
+  streamStateSessionPut(pState, key, NULL, 0);
+  pCur = streamStateSessionGetRanomCur(pState, key);
+  streamStateCurPrev(pState, pCur);
+  code = streamStateSessionGetKVByCur(pCur, key, (const void**)pVal, pVLen);
+  if (code == TSDB_CODE_SUCCESS) {
+    void* stateKey = (char*)(*pVal) + (valSize - keyDataLen);
+    if (fn(pKeyData, stateKey) == true) {
+      memcpy(tmp, *pVal, valSize);
+      goto _end;
+    }
+  }
+
+  streamStateFreeCur(pCur);
+  *key = tmpKey;
+  pCur = streamStateSessionSeekKeyNext(pState, key);
+  code = streamStateSessionGetKVByCur(pCur, key, (const void**)pVal, pVLen);
+  if (code == TSDB_CODE_SUCCESS) {
+    void* stateKey = (char*)(*pVal) + (valSize - keyDataLen);
+    if (fn(pKeyData, stateKey) == true) {
+      memcpy(tmp, *pVal, valSize);
+      goto _end;
+    }
+  }
+
+  *key = tmpKey;
+  res = 1;
+  memset(tmp, 0, valSize);
+
+_end:
+
+  *pVal = tmp;
+  streamStateSessionDel(pState, &tmpKey);
+  streamStateFreeCur(pCur);
+  return res;
+}
diff --git a/tests/script/jenkins/basic.txt b/tests/script/jenkins/basic.txt
index d8f4a36261..47611a7712 100644
--- a/tests/script/jenkins/basic.txt
+++ b/tests/script/jenkins/basic.txt
@@ -249,12 +249,18 @@
 ./test.sh -f tsim/stream/windowClose.sim
 ./test.sh -f tsim/stream/ignoreExpiredData.sim
 ./test.sh -f tsim/stream/sliding.sim
-#./test.sh -f tsim/stream/partitionbyColumnInterval.sim
+./test.sh -f tsim/stream/partitionbyColumnInterval.sim
 #./test.sh -f tsim/stream/partitionbyColumnSession.sim
 #./test.sh -f tsim/stream/partitionbyColumnState.sim
 #./test.sh -f tsim/stream/deleteInterval.sim
 #./test.sh -f tsim/stream/deleteSession.sim
 #./test.sh -f tsim/stream/deleteState.sim
+#./test.sh -f tsim/stream/fillIntervalDelete0.sim
+#./test.sh -f tsim/stream/fillIntervalDelete1.sim
+./test.sh -f tsim/stream/fillIntervalLinear.sim
+#./test.sh -f tsim/stream/fillIntervalPartitionBy.sim
+./test.sh -f tsim/stream/fillIntervalPrevNext.sim
+./test.sh -f tsim/stream/fillIntervalValue.sim
 
 # ---- transaction ----
 ./test.sh -f tsim/trans/lossdata1.sim
diff --git a/tests/script/tsim/stream/session0.sim b/tests/script/tsim/stream/session0.sim
index 3e0af354d8..afae8ef5da 100644
--- a/tests/script/tsim/stream/session0.sim
+++ b/tests/script/tsim/stream/session0.sim
@@ -216,12 +216,12 @@ if $data02 != 3.274823935 then
   goto loop2
 endi
 
-if $data03 != 1.800000000 then
+if $data03 != 1.500000000 then
   print ======$data03
   return -1
 endi
 
-if $data04 != 3.350000000 then
+if $data04 != 3.500000000 then
   print ======$data04
   return -1
 endi
diff --git a/tests/script/tsim/stream/session1.sim b/tests/script/tsim/stream/session1.sim
index ab173c5929..ee6eefde26 100644
--- a/tests/script/tsim/stream/session1.sim
+++ b/tests/script/tsim/stream/session1.sim
@@ -5,15 +5,15 @@ sleep 50
 sql connect
 
 print =============== create database
-sql create database test vgroups 1
-sql select * from information_schema.ins_databases
+sql create database test vgroups 1;
+sql select * from information_schema.ins_databases;
 if $rows != 3 then
   return -1
 endi
 
 print $data00 $data01 $data02
 
-sql use test
+sql use test;
 
 
 sql create table t1(ts timestamp, a int, b int , c int, d double,id int);
diff --git a/tests/script/tsim/stream/state0.sim b/tests/script/tsim/stream/state0.sim
index 877a2877b9..60a50f9560 100644
--- a/tests/script/tsim/stream/state0.sim
+++ b/tests/script/tsim/stream/state0.sim
@@ -349,7 +349,7 @@ endi
 
 if $rows != 3 then
   print ====loop4=rows=$rows
-#  goto loop4
+  goto loop4
 endi
 
 # row 0
diff --git a/tests/script/tsim/stream/triggerSession0.sim b/tests/script/tsim/stream/triggerSession0.sim
index 2ea689ef78..1bef439884 100644
--- a/tests/script/tsim/stream/triggerSession0.sim
+++ b/tests/script/tsim/stream/triggerSession0.sim
@@ -13,7 +13,7 @@ endi
 
 print $data00 $data01 $data02
 
-sql use test
+sql use test;
 sql create table t2(ts timestamp, a int, b int , c int, d double);
 sql create stream streams2 trigger window_close into streamt2 as select  _wstart, count(*) c1, count(d) c2 , sum(a) c3 , max(b)  c4, min(c) c5 from t2 session(ts, 10s);
 
@@ -58,16 +58,11 @@ endi
 sql insert into t2 values(1648791233002,1,2,3,1.0);
 sleep 300
 sql select * from streamt2;
-if $rows != 1 then
+if $rows != 0 then
   print ======$rows
   return -1
 endi
 
-if $data01 != 6 then
-  print ======$data01
-  return -1
-endi
-
 sql insert into t2 values(1648791253003,1,2,3,1.0);
 sleep 300
 sql select * from streamt2;

From 2d1f7354a94b9fd6e447c2bf73f7741d837d31a6 Mon Sep 17 00:00:00 2001
From: Ganlin Zhao <ganlinzhao@hotmail.com>
Date: Tue, 18 Oct 2022 19:11:49 +0800
Subject: [PATCH 38/42] fix null value handling

---
 source/libs/function/src/builtinsimpl.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/source/libs/function/src/builtinsimpl.c b/source/libs/function/src/builtinsimpl.c
index 6a428af1ed..0ab8fb0550 100644
--- a/source/libs/function/src/builtinsimpl.c
+++ b/source/libs/function/src/builtinsimpl.c
@@ -5363,7 +5363,7 @@ int32_t modeFinalize(SqlFunctionCtx* pCtx, SSDataBlock* pBlock) {
   }
 
   SModeItem* pResItem = (SModeItem*)(pInfo->pItems + resIndex * (sizeof(SModeItem) + pInfo->colBytes));
-  colDataAppend(pCol, currentRow, pResItem->data, false);
+  colDataAppend(pCol, currentRow, pResItem->data, (maxCount == 0) ? true : false);
 
   return pResInfo->numOfRes;
 }

From c77bc2b9b906f02e2eba6c7c397f30e397497e85 Mon Sep 17 00:00:00 2001
From: Xuefeng Tan <1172915550@qq.com>
Date: Tue, 18 Oct 2022 19:12:35 +0800
Subject: [PATCH 39/42] doc : driver-go support WebSocket-based bulk pulling
 (#17435)

* doc : driver-go support WebSocket-based bulk pulling

* doc : update English translation

* docs: remove preparation section

Co-authored-by: Shuduo Sang <sangshuduo@gmail.com>
---
 docs/en/14-reference/03-connector/05-go.mdx | 58 ++++++++++++++++++++-
 docs/zh/08-connector/20-go.mdx              | 55 +++++++++++++++++++
 2 files changed, 111 insertions(+), 2 deletions(-)

diff --git a/docs/en/14-reference/03-connector/05-go.mdx b/docs/en/14-reference/03-connector/05-go.mdx
index f00e635af9..a33b302a92 100644
--- a/docs/en/14-reference/03-connector/05-go.mdx
+++ b/docs/en/14-reference/03-connector/05-go.mdx
@@ -7,7 +7,6 @@ title: TDengine Go Connector
 import Tabs from '@theme/Tabs';
 import TabItem from '@theme/TabItem';
 
-import Preparition from "./_preparation.mdx"
 import GoInsert from "../../07-develop/03-insert-data/_go_sql.mdx"
 import GoInfluxLine from "../../07-develop/03-insert-data/_go_line.mdx"
 import GoOpenTSDBTelnet from "../../07-develop/03-insert-data/_go_opts_telnet.mdx"
@@ -176,6 +175,37 @@ func main() {
 }
 ```
 </TabItem>
+<TabItem value="WebSocket" label="WebSocket connection">
+
+_taosRestful_ implements Go's `database/sql/driver` interface via `http client`. You can use the [`database/sql`](https://golang.org/pkg/database/sql/) interface by simply introducing the driver (driver-go minimum version 3.0.2).
+
+Use `taosWS` as `driverName` and use a correct [DSN](#DSN) as `dataSourceName` with the following parameters supported by the DSN.
+
+* `writeTimeout` The timeout to send data via WebSocket.
+* `readTimeout` The timeout to receive response data via WebSocket.
+
+For example:
+
+```go
+package main
+
+import (
+    "database/sql"
+    "fmt"
+
+    _ "github.com/taosdata/driver-go/v3/taosWS"
+)
+
+func main() {
+    var taosUri = "root:taosdata@ws(localhost:6041)/"
+    taos, err := sql.Open("taosWS", taosUri)
+    if err != nil {
+        fmt.Println("failed to connect TDengine, err:", err)
+        return
+    }
+}
+```
+</TabItem>
 </Tabs>
 
 ## Usage examples
@@ -331,7 +361,7 @@ Creates consumer group.
 
 * `func (c *Consumer) Subscribe(topics []string) error`
 
-Subscribes to a topic.
+Subscribes to topics.
 
 * `func (c *Consumer) Poll(timeout time.Duration) (*Result, error)`
 
@@ -409,6 +439,30 @@ Close consumer.
 
   Closes the parameter binding.
 
+### Subscribe via WebSocket
+
+* `func NewConsumer(config *Config) (*Consumer, error)`
+
+   Creates consumer group.
+
+* `func (c *Consumer) Subscribe(topic []string) error`
+
+   Subscribes to topics.
+
+* `func (c *Consumer) Poll(timeout time.Duration) (*Result, error)`
+
+   Polling information.
+
+* `func (c *Consumer) Commit(messageID uint64) error`
+
+   Commit information.
+
+* `func (c *Consumer) Close() error`
+
+   Close consumer.
+
+For a complete example see [GitHub sample file](https://github.com/taosdata/driver-go/blob/3.0/examples/tmqoverws/main.go)
+
 ## API Reference
 
 Full API see [driver-go documentation](https://pkg.go.dev/github.com/taosdata/driver-go/v3)
diff --git a/docs/zh/08-connector/20-go.mdx b/docs/zh/08-connector/20-go.mdx
index 515d1b030b..7a6058db3c 100644
--- a/docs/zh/08-connector/20-go.mdx
+++ b/docs/zh/08-connector/20-go.mdx
@@ -177,6 +177,37 @@ func main() {
 }
 ```
 </TabItem>
+<TabItem value="WebSocket" label="WebSocket 连接">
+
+_taosWS_ 通过 `WebSocket` 实现了 Go 的 `database/sql/driver` 接口。只需要引入驱动（driver-go 最低版本 3.0.2）就可以使用[`database/sql`](https://golang.org/pkg/database/sql/)的接口。
+
+使用 `taosWS` 作为 `driverName` 并且使用一个正确的 [DSN](#DSN) 作为 `dataSourceName`，DSN 支持的参数：
+
+* `writeTimeout` 通过 WebSocket 发送数据的超时时间。
+* `readTimeout` 通过 WebSocket 接收响应数据的超时时间。
+
+示例：
+
+```go
+package main
+
+import (
+    "database/sql"
+    "fmt"
+
+    _ "github.com/taosdata/driver-go/v3/taosWS"
+)
+
+func main() {
+    var taosUri = "root:taosdata@ws(localhost:6041)/"
+    taos, err := sql.Open("taosWS", taosUri)
+    if err != nil {
+        fmt.Println("failed to connect TDengine, err:", err)
+        return
+    }
+}
+```
+</TabItem>
 </Tabs>
 
 ## 使用示例
@@ -410,6 +441,30 @@ func main() {
 
   结束参数绑定。
 
+### 通过 WebSocket 订阅
+
+* `func NewConsumer(config *Config) (*Consumer, error)`
+
+  创建消费者。
+
+* `func (c *Consumer) Subscribe(topic []string) error`
+
+  订阅主题。
+
+* `func (c *Consumer) Poll(timeout time.Duration) (*Result, error)`
+
+  轮询消息。
+
+* `func (c *Consumer) Commit(messageID uint64) error`
+
+  提交消息。
+
+* `func (c *Consumer) Close() error`
+
+  关闭消费者。
+
+完整订阅示例参见 [GitHub 示例文件](https://github.com/taosdata/driver-go/blob/3.0/examples/tmqoverws/main.go)
+
 ## API 参考
 
 全部 API 见 [driver-go 文档](https://pkg.go.dev/github.com/taosdata/driver-go/v3)

From 4292ef6e7c253cc5b241bbfc5b6fcca9403bb76c Mon Sep 17 00:00:00 2001
From: chenhaoran <haoran920c@163.com>
Date: Tue, 18 Oct 2022 12:26:05 +0000
Subject: [PATCH 40/42] test:add md5sum file of ci libtaos

---
 tests/parallel_test/run_case.sh | 4 ++++
 1 file changed, 4 insertions(+)

diff --git a/tests/parallel_test/run_case.sh b/tests/parallel_test/run_case.sh
index 58dcb87345..e0b905375a 100755
--- a/tests/parallel_test/run_case.sh
+++ b/tests/parallel_test/run_case.sh
@@ -67,9 +67,13 @@ mkdir -p /var/lib/taos
 cd $CONTAINER_TESTDIR/tests/$exec_dir
 ulimit -c unlimited
 
+md5sum /usr/lib/libtaos.so.1
+md5sum /home/TDinternal/debug/build/lib/libtaos.so
 $TIMEOUT_CMD $cmd
 RET=$?
 echo "cmd exit code: $RET"
+md5sum /usr/lib/libtaos.so.1
+md5sum /home/TDinternal/debug/build/lib/libtaos.so
 
 if [ $RET -ne 0 ]; then
     pwd

From 4a96af8037c29296676344ab777f05546af14e32 Mon Sep 17 00:00:00 2001
From: chenhaoran <haoran920c@163.com>
Date: Tue, 18 Oct 2022 14:04:49 +0000
Subject: [PATCH 41/42] test:add md5sum file of ci libtaos

---
 tests/parallel_test/run_container.sh | 5 +++++
 1 file changed, 5 insertions(+)

diff --git a/tests/parallel_test/run_container.sh b/tests/parallel_test/run_container.sh
index bb57f238f0..f58aaaf29d 100755
--- a/tests/parallel_test/run_container.sh
+++ b/tests/parallel_test/run_container.sh
@@ -68,12 +68,16 @@ if [ $ent -ne 0 ]; then
     CONTAINER_TESTDIR=/home/TDinternal/community
     SIM_DIR=/home/TDinternal/sim
     REP_MOUNT_PARAM="$INTERNAL_REPDIR:/home/TDinternal"
+    REP_MOUNT_LIB="$INTERNAL_REPDIR/debug/build/lib:/home/TDinternal/debug/build/lib:ro"
+
 else
     # community edition
     REPDIR=$WORKDIR/TDengine
     CONTAINER_TESTDIR=/home/TDengine
     SIM_DIR=/home/TDengine/sim
     REP_MOUNT_PARAM="$REPDIR:/home/TDengine"
+    REP_MOUNT_LIB="$REPDIR/debug/build/lib:/home/TDengine/debug/build/lib:ro"
+
 fi
 
 ulimit -c unlimited
@@ -103,6 +107,7 @@ coredump_dir=`cat /proc/sys/kernel/core_pattern | xargs dirname`
 
 docker run \
     -v $REP_MOUNT_PARAM \
+    -v $REP_MOUNT_LIB \
     -v $MOUNT_DIR \
     -v ${SOURCEDIR}:/usr/local/src/ \
     -v "$TMP_DIR/thread_volume/$thread_no/sim:${SIM_DIR}" \

From 312294bff4bf028c85ff5e92019bee64f406c882 Mon Sep 17 00:00:00 2001
From: Shuduo Sang <sangshuduo@gmail.com>
Date: Wed, 19 Oct 2022 08:57:07 +0800
Subject: [PATCH 42/42] feat: taosbenchmark support partial col 9284147
 (#17456)

---
 cmake/taostools_CMakeLists.txt.in | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/cmake/taostools_CMakeLists.txt.in b/cmake/taostools_CMakeLists.txt.in
index ea4d48a2b1..49e94896cd 100644
--- a/cmake/taostools_CMakeLists.txt.in
+++ b/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 c64858f
+        GIT_TAG 9284147
         SOURCE_DIR "${TD_SOURCE_DIR}/tools/taos-tools"
         BINARY_DIR ""
         #BUILD_IN_SOURCE TRUE