Merge branch 'develop' into xiaoping/add_test_case

src/balance/src/bnMain.c

@@ -237,22 +237,22 @@ static bool bnCheckVgroupReady(SVgObj *pVgroup, SVnodeGid *pRmVnode) {
   bool isReady = false;
   for (int32_t i = 0; i < pVgroup->numOfVnodes; ++i) {
     SVnodeGid *pVnode = pVgroup->vnodeGid + i;
+    SDnodeObj *pDnode = pVnode->pDnode;
     if (pVnode == pRmVnode) continue;
     int32_t vver = mnodeGetVgidVer(pVnode->vver);
 
-    mTrace("vgId:%d, check vgroup status, vindex:%d dnode:%d status:%s role:%s vver:%d, rmvver:%d" , pVgroup->vgId, i,
-           pVnode->dnodeId, dnodeStatus[pVnode->pDnode->status], syncRole[pVnode->role], vver, rmVnodeVer);
-    if (pVnode->pDnode->status == TAOS_DN_STATUS_DROPPING) continue;
-    if (pVnode->pDnode->status == TAOS_DN_STATUS_OFFLINE) continue;
+    mTrace("vgId:%d, check vgroup status, vindex:%d dnode:%d status:%s role:%s vver:%d, rmvver:%d", pVgroup->vgId, i,
+           pVnode->dnodeId, dnodeStatus[pDnode->status], syncRole[pVnode->role], vver, rmVnodeVer);
+    if (pDnode->status == TAOS_DN_STATUS_DROPPING) continue;
+    if (pDnode->status == TAOS_DN_STATUS_OFFLINE) continue;
     if (pVnode->role != TAOS_SYNC_ROLE_SLAVE && pVnode->role != TAOS_SYNC_ROLE_MASTER) continue;
 
     if (rmVnodeVer == 0 || vver >= rmVnodeVer) {
-      mInfo("vgId:%d, is ready for vindex:%d in dnode:%d status:%s role:%s vver:%d larger than rmvver:%d", pVgroup->vgId, i,
-             pVnode->dnodeId, dnodeStatus[pVnode->pDnode->status], syncRole[pVnode->role], vver, rmVnodeVer);
-    }
-
+      mInfo("vgId:%d, is ready for vindex:%d in dnode:%d status:%s role:%s vver:%d larger than rmvver:%d",
+            pVgroup->vgId, i, pVnode->dnodeId, dnodeStatus[pDnode->status], syncRole[pVnode->role], vver, rmVnodeVer);
       isReady = true;
     }
+  }
 
   return isReady;
 }

src/client/inc/tscUtil.h

@@ -20,9 +20,6 @@
 extern "C" {
 #endif
 
-/*
- * @date   2018/09/30
- */
 #include "exception.h"
 #include "os.h"
 #include "qExtbuffer.h"
@@ -216,7 +213,7 @@ STableMetaInfo* tscGetMetaInfo(SQueryInfo *pQueryInfo, int32_t tableIndex);
 SQueryInfo *tscGetQueryInfoDetail(SSqlCmd* pCmd, int32_t subClauseIndex);
 SQueryInfo *tscGetQueryInfoDetailSafely(SSqlCmd *pCmd, int32_t subClauseIndex);
 
-void tscClearTableMetaInfo(STableMetaInfo* pTableMetaInfo, bool removeFromCache);
+void tscClearTableMetaInfo(STableMetaInfo* pTableMetaInfo);
 
 STableMetaInfo* tscAddTableMetaInfo(SQueryInfo* pQueryInfo, const char* name, STableMeta* pTableMeta,
     SVgroupsInfo* vgroupList, SArray* pTagCols, SArray* pVgroupTables);
@@ -276,7 +273,7 @@ void tscPrintSelectClause(SSqlObj* pSql, int32_t subClauseIndex);
 bool hasMoreVnodesToTry(SSqlObj *pSql);
 bool hasMoreClauseToTry(SSqlObj* pSql);
 
-void tscFreeQueryInfo(SSqlCmd* pCmd, bool removeFromCache);
+void tscFreeQueryInfo(SSqlCmd* pCmd);
 
 void tscTryQueryNextVnode(SSqlObj *pSql, __async_cb_func_t fp);
 void tscAsyncQuerySingleRowForNextVnode(void *param, TAOS_RES *tres, int numOfRows);
@@ -290,6 +287,14 @@ int32_t doArithmeticCalculate(SQueryInfo* pQueryInfo, tFilePage* pOutput, int32_
 char*   serializeTagData(STagData* pTagData, char* pMsg);
 int32_t copyTagData(STagData* dst, const STagData* src);
 
+STableMeta* createSuperTableMeta(STableMetaMsg* pChild);
+uint32_t tscGetTableMetaSize(STableMeta* pTableMeta);
+CChildTableMeta* tscCreateChildMeta(STableMeta* pTableMeta);
+uint32_t tscGetTableMetaMaxSize();
+int32_t tscCreateTableMetaFromCChildMeta(STableMeta* pChild, const char* name);
+STableMeta* tscTableMetaClone(STableMeta* pTableMeta);
+
+
 void* malloc_throw(size_t size);
 void* calloc_throw(size_t nmemb, size_t size);
 char* strdup_throw(const char* str);

src/client/inc/tschemautil.h

@@ -105,7 +105,10 @@ SSchema tscGetTbnameColumnSchema();
  * @param size size of the table meta
  * @return
  */
-STableMeta* tscCreateTableMetaFromMsg(STableMetaMsg* pTableMetaMsg, size_t* size);
+STableMeta* tscCreateTableMetaFromMsg(STableMetaMsg* pTableMetaMsg);
+
+bool vgroupInfoIdentical(SNewVgroupInfo *pExisted, SVgroupMsg* src);
+SNewVgroupInfo createNewVgroupInfo(SVgroupMsg *pVgroupMsg);
 
 #ifdef __cplusplus
 }

src/client/inc/tsclient.h

@@ -56,23 +56,28 @@ typedef struct STableComInfo {
   int32_t rowSize;
 } STableComInfo;
 
-typedef struct SCorVgroupInfo {
-  int32_t  version;
+typedef struct SNewVgroupInfo {
+  int32_t    vgId;
   int8_t     inUse;
   int8_t     numOfEps;
-  SEpAddr1 epAddr[TSDB_MAX_REPLICA];
-} SCorVgroupInfo;
+  SEpAddrMsg ep[TSDB_MAX_REPLICA];
+} SNewVgroupInfo;
+
+typedef struct CChildTableMeta {
+  int32_t        vgId;
+  STableId       id;
+  uint8_t        tableType;
+  char           sTableName[TSDB_TABLE_FNAME_LEN];
+} CChildTableMeta;
 
 typedef struct STableMeta {
-  STableComInfo  tableInfo;
+  int32_t        vgId;
+  STableId       id;
   uint8_t        tableType;
+  char           sTableName[TSDB_TABLE_FNAME_LEN];
   int16_t        sversion;
   int16_t        tversion;
-  char           sTableId[TSDB_TABLE_FNAME_LEN];
-  int32_t        vgId;
-  SCorVgroupInfo corVgroupInfo;
-  STableId       id;
-//  union {int64_t stableUid; SSchema* schema;};
+  STableComInfo  tableInfo;
   SSchema        schema[];  // if the table is TSDB_CHILD_TABLE, schema is acquired by super table meta info
 } STableMeta;
 
@@ -171,7 +176,7 @@ typedef struct SParamInfo {
 } SParamInfo;
 
 typedef struct STableDataBlocks {
-  char        tableId[TSDB_TABLE_FNAME_LEN];
+  char        tableName[TSDB_TABLE_FNAME_LEN];
   int8_t      tsSource;     // where does the UNIX timestamp come from, server or client
   bool        ordered;      // if current rows are ordered or not
   int64_t     vgId;         // virtual group id
@@ -249,7 +254,7 @@ typedef struct {
   int8_t       submitSchema;   // submit block is built with table schema
   STagData     tagData;        // NOTE: pTagData->data is used as a variant length array
 
-  STableMeta **pTableMetaList; // all involved tableMeta list of current insert sql statement.
+  char       **pTableNameList; // all involved tableMeta list of current insert sql statement.
   int32_t      numOfTables;
 
   SHashObj    *pTableBlockHashList;     // data block for each table
@@ -386,7 +391,7 @@ void tscProcessMsgFromServer(SRpcMsg *rpcMsg, SRpcEpSet *pEpSet);
 int  tscProcessSql(SSqlObj *pSql);
 
 int  tscRenewTableMeta(SSqlObj *pSql, int32_t tableIndex);
-void tscQueueAsyncRes(SSqlObj *pSql);
+void tscAsyncResultOnError(SSqlObj *pSql);
 
 void tscQueueAsyncError(void(*fp), void *param, int32_t code);
 
@@ -400,7 +405,7 @@ void    tscRestoreSQLFuncForSTableQuery(SQueryInfo *pQueryInfo);
 int32_t tscCreateResPointerInfo(SSqlRes *pRes, SQueryInfo *pQueryInfo);
 void tscSetResRawPtr(SSqlRes* pRes, SQueryInfo* pQueryInfo);
 
-void tscResetSqlCmdObj(SSqlCmd *pCmd, bool removeFromCache);
+void tscResetSqlCmdObj(SSqlCmd *pCmd);
 
 /**
  * free query result of the sql object
@@ -414,7 +419,6 @@ void tscFreeSqlResult(SSqlObj *pSql);
  */
 void tscFreeSqlObj(SSqlObj *pSql);
 void tscFreeRegisteredSqlObj(void *pSql);
-void tscFreeTableMetaHelper(void *pTableMeta);
 
 void tscCloseTscObj(void *pObj);
 
@@ -479,7 +483,9 @@ static FORCE_INLINE void tscGetResultColumnChr(SSqlRes* pRes, SFieldInfo* pField
   }
 }
 
-extern SCacheObj *tscMetaCache;
+extern int32_t    sentinel;
+extern SHashObj  *tscVgroupMap;
+extern SHashObj  *tscTableMetaInfo;
 
 extern int   tscObjRef;
 extern void *tscTmr;

src/client/src/tscAsync.c

@@ -18,7 +18,6 @@
 
 #include "tnote.h"
 #include "trpc.h"
-#include "tcache.h"
 #include "tscLog.h"
 #include "tscSubquery.h"
 #include "tscLocalMerge.h"
@@ -57,7 +56,7 @@ void doAsyncQuery(STscObj* pObj, SSqlObj* pSql, __async_cb_func_t fp, void* para
   if (pSql->sqlstr == NULL) {
     tscError("%p failed to malloc sql string buffer", pSql);
     pSql->res.code = TSDB_CODE_TSC_OUT_OF_MEMORY;
-    tscQueueAsyncRes(pSql);
+    tscAsyncResultOnError(pSql);
     return;
   }
 
@@ -71,7 +70,7 @@ void doAsyncQuery(STscObj* pObj, SSqlObj* pSql, __async_cb_func_t fp, void* para
   
   if (code != TSDB_CODE_SUCCESS) {
     pSql->res.code = code;
-    tscQueueAsyncRes(pSql);
+    tscAsyncResultOnError(pSql);
     return;
   }
 
@@ -166,7 +165,7 @@ static void tscProcessAsyncRetrieveImpl(void *param, TAOS_RES *tres, int numOfRo
       pRes->code = numOfRows;
     }
 
-    tscQueueAsyncRes(pSql);
+    tscAsyncResultOnError(pSql);
     return;
   }
 
@@ -217,7 +216,7 @@ void taos_fetch_rows_a(TAOS_RES *taosa, __async_cb_func_t fp, void *param) {
     pRes->code = TSDB_CODE_TSC_INVALID_QHANDLE;
     pSql->param = param;
 
-    tscQueueAsyncRes(pSql);
+    tscAsyncResultOnError(pSql);
     return;
   }
 
@@ -280,7 +279,7 @@ void taos_fetch_row_a(TAOS_RES *taosa, void (*fp)(void *, TAOS_RES *, TAOS_ROW),
     pSql->param = param;
     pRes->code = TSDB_CODE_TSC_INVALID_QHANDLE;
 
-    tscQueueAsyncRes(pSql);
+    tscAsyncResultOnError(pSql);
     return;
   }
 
@@ -382,7 +381,7 @@ void tscQueueAsyncError(void(*fp), void *param, int32_t code) {
 }
 
 
-void tscQueueAsyncRes(SSqlObj *pSql) {
+void tscAsyncResultOnError(SSqlObj *pSql) {
   if (pSql == NULL || pSql->signature != pSql) {
     tscDebug("%p SqlObj is freed, not add into queue async res", pSql);
     return;
@@ -423,7 +422,7 @@ void tscTableMetaCallBack(void *param, TAOS_RES *res, int code) {
 
     // check if it is a sub-query of super table query first, if true, enter another routine
     if (TSDB_QUERY_HAS_TYPE(pQueryInfo->type, (TSDB_QUERY_TYPE_STABLE_SUBQUERY|TSDB_QUERY_TYPE_TAG_FILTER_QUERY))) {
-      tscDebug("%p update table meta in local cache, continue to process sql and send the corresponding query", pSql);
+      tscDebug("%p update local table meta, continue to process sql and send the corresponding query", pSql);
 
       STableMetaInfo* pTableMetaInfo = tscGetMetaInfo(pQueryInfo, 0);
       code = tscGetTableMeta(pSql, pTableMetaInfo);
@@ -440,7 +439,7 @@ void tscTableMetaCallBack(void *param, TAOS_RES *res, int code) {
       return;
     } else {  // continue to process normal async query
       if (pCmd->parseFinished) {
-        tscDebug("%p update table meta in local cache, continue to process sql and send corresponding query", pSql);
+        tscDebug("%p update local table meta, continue to process sql and send corresponding query", pSql);
 
         STableMetaInfo* pTableMetaInfo = tscGetTableMetaInfoFromCmd(pCmd, pCmd->clauseIndex, 0);
         code = tscGetTableMeta(pSql, pTableMetaInfo);
@@ -455,7 +454,7 @@ void tscTableMetaCallBack(void *param, TAOS_RES *res, int code) {
         if (pCmd->command == TSDB_SQL_SELECT) {
           tscDebug("%p redo parse sql string and proceed", pSql);
           pCmd->parseFinished = false;
-          tscResetSqlCmdObj(pCmd, false);
+          tscResetSqlCmdObj(pCmd);
 
           code = tsParseSql(pSql, true);
           if (code == TSDB_CODE_TSC_ACTION_IN_PROGRESS) {
@@ -532,6 +531,6 @@ void tscTableMetaCallBack(void *param, TAOS_RES *res, int code) {
   _error:
   if (code != TSDB_CODE_SUCCESS) {
     pSql->res.code = code;
-    tscQueueAsyncRes(pSql);
+    tscAsyncResultOnError(pSql);
   }
 }

src/client/src/tscLocal.c

@@ -17,7 +17,6 @@
 #include "taosmsg.h"
 
 #include "taosdef.h"
-#include "tcache.h"
 #include "tname.h"
 #include "tscLog.h"
 #include "tscUtil.h"
@@ -273,7 +272,7 @@ void tscSCreateCallBack(void *param, TAOS_RES *tres, int code) {
   if (pRes->code != TSDB_CODE_SUCCESS) {
     taos_free_result(pSql);  
     free(builder);
-    tscQueueAsyncRes(pParentSql);
+    tscAsyncResultOnError(pParentSql);
     return;
   }
 
@@ -291,7 +290,7 @@ void tscSCreateCallBack(void *param, TAOS_RES *tres, int code) {
     if (pRes->code == TSDB_CODE_SUCCESS) {
       (*pParentSql->fp)(pParentSql->param, pParentSql, code);  
     } else {
-      tscQueueAsyncRes(pParentSql);
+      tscAsyncResultOnError(pParentSql);
     }
   }
 }
@@ -571,7 +570,7 @@ static int32_t tscRebuildDDLForSubTable(SSqlObj *pSql, const char *tableName, ch
 
   char fullName[TSDB_TABLE_FNAME_LEN * 2] = {0};
   extractDBName(pTableMetaInfo->name, fullName);
-  extractTableName(pMeta->sTableId, param->sTableName);
+  extractTableName(pMeta->sTableName, param->sTableName);
   snprintf(fullName + strlen(fullName), TSDB_TABLE_FNAME_LEN - strlen(fullName),  ".%s", param->sTableName);
   extractTableName(pTableMetaInfo->name, param->buf);
 
@@ -901,7 +900,7 @@ int tscProcessLocalCmd(SSqlObj *pSql) {
   } else if (pCmd->command == TSDB_SQL_SHOW_CREATE_DATABASE) {
     pRes->code = tscProcessShowCreateDatabase(pSql); 
   } else if (pCmd->command == TSDB_SQL_RESET_CACHE) {
-    taosCacheEmpty(tscMetaCache);
+    taosHashEmpty(tscTableMetaInfo);
     pRes->code = TSDB_CODE_SUCCESS;
   } else if (pCmd->command == TSDB_SQL_SERV_VERSION) {
     pRes->code = tscProcessServerVer(pSql);
@@ -925,7 +924,7 @@ int tscProcessLocalCmd(SSqlObj *pSql) {
     (*pSql->fp)(pSql->param, pSql, code);
   } else if (code == TSDB_CODE_TSC_ACTION_IN_PROGRESS){
   } else {
-    tscQueueAsyncRes(pSql);
+    tscAsyncResultOnError(pSql);
   }
   return code;
 }

src/client/src/tscLocalMerge.c

@@ -89,7 +89,7 @@ static void tscInitSqlContext(SSqlCmd *pCmd, SLocalReducer *pReducer, tOrderDesc
     pCtx->startOffset = 0;
     pCtx->size = 1;
     pCtx->hasNull = true;
-    pCtx->currentStage = SECONDARY_STAGE_MERGE;
+    pCtx->currentStage = MERGE_STAGE;
 
     // for top/bottom function, the output of timestamp is the first column
     int32_t functionId = pExpr->functionId;
@@ -1067,7 +1067,7 @@ static void doExecuteSecondaryMerge(SSqlCmd *pCmd, SLocalReducer *pLocalReducer,
       pCtx->param[0].i64Key = pExpr->param[0].i64Key;
     }
 
-    pCtx->currentStage = SECONDARY_STAGE_MERGE;
+    pCtx->currentStage = MERGE_STAGE;
 
     if (needInit) {
       aAggs[pCtx->functionId].init(pCtx);
@@ -1080,7 +1080,7 @@ static void doExecuteSecondaryMerge(SSqlCmd *pCmd, SLocalReducer *pLocalReducer,
       continue;
     }
 
-    aAggs[functionId].distSecondaryMergeFunc(&pLocalReducer->pCtx[j]);
+    aAggs[functionId].mergeFunc(&pLocalReducer->pCtx[j]);
   }
 }
 
@@ -1647,7 +1647,7 @@ int32_t doArithmeticCalculate(SQueryInfo* pQueryInfo, tFilePage* pOutput, int32_
     // calculate the result from several other columns
     if (pSup->pArithExprInfo != NULL) {
       arithSup.pArithExpr = pSup->pArithExprInfo;
-      tExprTreeCalcTraverse(arithSup.pArithExpr->pExpr, (int32_t) pOutput->num, pbuf + pOutput->num*offset, &arithSup, TSDB_ORDER_ASC, getArithmeticInputSrc);
+      arithmeticTreeTraverse(arithSup.pArithExpr->pExpr, (int32_t) pOutput->num, pbuf + pOutput->num*offset, &arithSup, TSDB_ORDER_ASC, getArithmeticInputSrc);
     } else {
       SSqlExpr* pExpr = pSup->pSqlExpr;
       memcpy(pbuf + pOutput->num * offset, pExpr->offset * pOutput->num + pOutput->data, (size_t)(pExpr->resBytes * pOutput->num));

src/client/src/tscParseInsert.c

@@ -1339,7 +1339,7 @@ int tsParseSql(SSqlObj *pSql, bool initial) {
     if (sqlstr == NULL || pSql->parseRetry >= 1 || ret != TSDB_CODE_TSC_INVALID_SQL) {
       free(sqlstr);
     } else {
-      tscResetSqlCmdObj(pCmd, true);
+      tscResetSqlCmdObj(pCmd);
       free(pSql->sqlstr);
       pSql->sqlstr = sqlstr;
       pSql->parseRetry++;
@@ -1351,7 +1351,7 @@ int tsParseSql(SSqlObj *pSql, bool initial) {
     SSqlInfo SQLInfo = qSQLParse(pSql->sqlstr);
     ret = tscToSQLCmd(pSql, &SQLInfo);
     if (ret == TSDB_CODE_TSC_INVALID_SQL && pSql->parseRetry == 0 && SQLInfo.type == TSDB_SQL_NULL) {
-      tscResetSqlCmdObj(pCmd, true);
+      tscResetSqlCmdObj(pCmd);
       pSql->parseRetry++;
       ret = tscToSQLCmd(pSql, &SQLInfo);
     }
@@ -1429,7 +1429,7 @@ static void parseFileSendDataBlock(void *param, TAOS_RES *tres, int code) {
       fclose(fp);
 
       pParentSql->res.code = code;
-      tscQueueAsyncRes(pParentSql);
+      tscAsyncResultOnError(pParentSql);
       return;
     } while (0);
   }
@@ -1451,7 +1451,7 @@ static void parseFileSendDataBlock(void *param, TAOS_RES *tres, int code) {
   int32_t count = 0;
   int32_t maxRows = 0;
 
-  tfree(pCmd->pTableMetaList);
+  tfree(pCmd->pTableNameList);
   pCmd->pDataBlocks = tscDestroyBlockArrayList(pCmd->pDataBlocks);
 
   if (pCmd->pTableBlockHashList == NULL) {
@@ -1500,7 +1500,7 @@ static void parseFileSendDataBlock(void *param, TAOS_RES *tres, int code) {
     code = doPackSendDataBlock(pSql, count, pTableDataBlock);
     if (code != TSDB_CODE_SUCCESS) {
       pParentSql->res.code = code;
-      tscQueueAsyncRes(pParentSql);
+      tscAsyncResultOnError(pParentSql);
       return;
     }
 
@@ -1535,7 +1535,7 @@ void tscProcessMultiVnodesImportFromFile(SSqlObj *pSql) {
     tscError("%p failed to open file %s to load data from file, code:%s", pSql, pCmd->payload, tstrerror(pSql->res.code));
 
     tfree(pSupporter);
-    tscQueueAsyncRes(pSql);
+    tscAsyncResultOnError(pSql);
 
     return;
   }

src/client/src/tscSQLParser.c

@@ -910,7 +910,7 @@ int32_t tscSetTableFullName(STableMetaInfo* pTableMetaInfo, SStrToken* pzTableNa
    * that are corresponding to the old name for the new table name.
    */
   if (strlen(oldName) > 0 && strncasecmp(oldName, pTableMetaInfo->name, tListLen(pTableMetaInfo->name)) != 0) {
-    tscClearTableMetaInfo(pTableMetaInfo, false);
+    tscClearTableMetaInfo(pTableMetaInfo);
   }
 
   return TSDB_CODE_SUCCESS;
@@ -4019,6 +4019,7 @@ static int32_t setTableCondForSTableQuery(SSqlCmd* pCmd, SQueryInfo* pQueryInfo,
   if (pExpr->nSQLOptr == TK_LIKE) {
     char* str = taosStringBuilderGetResult(sb, NULL);
     pQueryInfo->tagCond.tbnameCond.cond = strdup(str);
+    pQueryInfo->tagCond.tbnameCond.len = (int32_t) strlen(str);
     return TSDB_CODE_SUCCESS;
   }
 
@@ -4068,6 +4069,7 @@ static int32_t setTableCondForSTableQuery(SSqlCmd* pCmd, SQueryInfo* pQueryInfo,
 
   char* str = taosStringBuilderGetResult(&sb1, NULL);
   pQueryInfo->tagCond.tbnameCond.cond = strdup(str);
+  pQueryInfo->tagCond.tbnameCond.len = (int32_t) strlen(str);
 
   taosStringBuilderDestroy(&sb1);
   tfree(segments);
@@ -6381,6 +6383,41 @@ int32_t doCheckForStream(SSqlObj* pSql, SSqlInfo* pInfo) {
   return TSDB_CODE_SUCCESS;
 }
 
+static int32_t checkQueryRangeForFill(SSqlCmd* pCmd, SQueryInfo* pQueryInfo) {
+  const char* msg3 = "start(end) time of query range required or time range too large";
+
+  if (pQueryInfo->interval.interval == 0) {
+    return TSDB_CODE_SUCCESS;
+  }
+
+    bool initialWindows = TSWINDOW_IS_EQUAL(pQueryInfo->window, TSWINDOW_INITIALIZER);
+    if (initialWindows) {
+      return invalidSqlErrMsg(tscGetErrorMsgPayload(pCmd), msg3);
+    }
+
+    int64_t timeRange = ABS(pQueryInfo->window.skey - pQueryInfo->window.ekey);
+
+    int64_t intervalRange = 0;
+    if (pQueryInfo->interval.intervalUnit == 'n' || pQueryInfo->interval.intervalUnit == 'y') {
+      int64_t f = 1;
+      if (pQueryInfo->interval.intervalUnit == 'n') {
+        f = 30L * MILLISECOND_PER_DAY;
+      } else if (pQueryInfo->interval.intervalUnit == 'y') {
+        f = 365L * MILLISECOND_PER_DAY;
+      }
+
+      intervalRange = pQueryInfo->interval.interval * f;
+    } else {
+      intervalRange = pQueryInfo->interval.interval;
+    }
+    // number of result is not greater than 10,000,000
+    if ((timeRange == 0) || (timeRange / intervalRange) >= MAX_INTERVAL_TIME_WINDOW) {
+      return invalidSqlErrMsg(tscGetErrorMsgPayload(pCmd), msg3);
+    }
+
+    return TSDB_CODE_SUCCESS;
+}
+
 int32_t doCheckForQuery(SSqlObj* pSql, SQuerySQL* pQuerySql, int32_t index) {
   assert(pQuerySql != NULL && (pQuerySql->from == NULL || taosArrayGetSize(pQuerySql->from) > 0));
 
@@ -6576,31 +6613,21 @@ int32_t doCheckForQuery(SSqlObj* pSql, SQuerySQL* pQuerySql, int32_t index) {
 
   tscFieldInfoUpdateOffset(pQueryInfo);
 
-  /*
-   * fill options are set at the end position, when all columns are set properly
-   * the columns may be increased due to group by operation
-   */
   if (pQuerySql->fillType != NULL) {
     if (pQueryInfo->interval.interval == 0 && (!tscIsPointInterpQuery(pQueryInfo))) {
       return invalidSqlErrMsg(tscGetErrorMsgPayload(pCmd), msg2);
     }
 
-    if (pQueryInfo->interval.interval > 0) {
-      bool initialWindows = TSWINDOW_IS_EQUAL(pQueryInfo->window, TSWINDOW_INITIALIZER);
-      if (initialWindows) {
-        return invalidSqlErrMsg(tscGetErrorMsgPayload(pCmd), msg3);
-      }
-
-      int64_t timeRange = ABS(pQueryInfo->window.skey - pQueryInfo->window.ekey);
-      // number of result is not greater than 10,000,000
-      if ((timeRange == 0) || (timeRange / pQueryInfo->interval.interval) > MAX_INTERVAL_TIME_WINDOW) {
-        return invalidSqlErrMsg(tscGetErrorMsgPayload(pCmd), msg3);
-      }
+    /*
+     * fill options are set at the end position, when all columns are set properly
+     * the columns may be increased due to group by operation
+     */
+    if ((code = checkQueryRangeForFill(pCmd, pQueryInfo)) != TSDB_CODE_SUCCESS) {
+      return code;
     }
 
-    int32_t ret = parseFillClause(pCmd, pQueryInfo, pQuerySql);
-    if (ret != TSDB_CODE_SUCCESS) {
-      return ret;
+    if ((code = parseFillClause(pCmd, pQueryInfo, pQuerySql)) != TSDB_CODE_SUCCESS) {
+      return code;
     }
   }
 

src/client/src/tscSchemaUtil.c

@@ -130,19 +130,8 @@ SSchema* tscGetColumnSchemaById(STableMeta* pTableMeta, int16_t colId) {
   return NULL;
 }
 
-static void tscInitCorVgroupInfo(SCorVgroupInfo *corVgroupInfo, SVgroupMsg *pVgroupMsg) {
-  corVgroupInfo->version = 0;
-  corVgroupInfo->inUse   = 0;
-  corVgroupInfo->numOfEps = pVgroupMsg->numOfEps;
-
-  for (int32_t i = 0; i < pVgroupMsg->numOfEps; i++) {
-    corVgroupInfo->epAddr[i].fqdn = strndup(pVgroupMsg->epAddr[i].fqdn, tListLen(pVgroupMsg->epAddr[0].fqdn));
-    corVgroupInfo->epAddr[i].port = pVgroupMsg->epAddr[i].port;
-  }
-}
-
-STableMeta* tscCreateTableMetaFromMsg(STableMetaMsg* pTableMetaMsg, size_t* size) {
-  assert(pTableMetaMsg != NULL);
+STableMeta* tscCreateTableMetaFromMsg(STableMetaMsg* pTableMetaMsg) {
+  assert(pTableMetaMsg != NULL && pTableMetaMsg->numOfColumns >= 2 && pTableMetaMsg->numOfTags >= 0);
   
   int32_t schemaSize = (pTableMetaMsg->numOfColumns + pTableMetaMsg->numOfTags) * sizeof(SSchema);
   STableMeta* pTableMeta = calloc(1, sizeof(STableMeta) + schemaSize);
@@ -159,11 +148,9 @@ STableMeta* tscCreateTableMetaFromMsg(STableMetaMsg* pTableMetaMsg, size_t* size
   pTableMeta->id.tid = pTableMetaMsg->tid;
   pTableMeta->id.uid = pTableMetaMsg->uid;
 
-  tscInitCorVgroupInfo(&pTableMeta->corVgroupInfo, &pTableMetaMsg->vgroup);
-
   pTableMeta->sversion = pTableMetaMsg->sversion;
   pTableMeta->tversion = pTableMetaMsg->tversion;
-  tstrncpy(pTableMeta->sTableId, pTableMetaMsg->sTableId, TSDB_TABLE_FNAME_LEN);
+  tstrncpy(pTableMeta->sTableName, pTableMetaMsg->sTableName, TSDB_TABLE_FNAME_LEN);
   
   memcpy(pTableMeta->schema, pTableMetaMsg->schema, schemaSize);
   
@@ -172,11 +159,42 @@ STableMeta* tscCreateTableMetaFromMsg(STableMetaMsg* pTableMetaMsg, size_t* size
     pTableMeta->tableInfo.rowSize += pTableMeta->schema[i].bytes;
   }
   
-  if (size != NULL) {
-    *size = sizeof(STableMeta) + schemaSize;
+  return pTableMeta;
+}
+
+bool vgroupInfoIdentical(SNewVgroupInfo *pExisted, SVgroupMsg* src) {
+  assert(pExisted != NULL && src != NULL);
+  if (pExisted->numOfEps != src->numOfEps) {
+    return false;
   }
 
-  return pTableMeta;
+  for(int32_t i = 0; i < pExisted->numOfEps; ++i) {
+    if (pExisted->ep[i].port != src->epAddr[i].port) {
+      return false;
+    }
+
+    if (strncmp(pExisted->ep[i].fqdn, src->epAddr[i].fqdn, tListLen(pExisted->ep[i].fqdn)) != 0) {
+      return false;
+    }
+  }
+
+  return true;
+}
+
+SNewVgroupInfo createNewVgroupInfo(SVgroupMsg *pVgroupMsg) {
+  assert(pVgroupMsg != NULL);
+
+  SNewVgroupInfo info = {0};
+  info.numOfEps = pVgroupMsg->numOfEps;
+  info.vgId     = pVgroupMsg->vgId;
+  info.inUse    = 0;
+
+  for(int32_t i = 0; i < pVgroupMsg->numOfEps; ++i) {
+    tstrncpy(info.ep[i].fqdn, pVgroupMsg->epAddr[i].fqdn, TSDB_FQDN_LEN);
+    info.ep[i].port = pVgroupMsg->epAddr[i].port;
+  }
+
+  return info;
 }
 
 // todo refactor

src/client/src/tscServer.c

@@ -14,7 +14,6 @@
  */
 
 #include "os.h"
-#include "tcache.h"
 #include "tcmdtype.h"
 #include "trpc.h"
 #include "tscLocalMerge.h"
@@ -86,6 +85,7 @@ bool tscEpSetIsEqual(SRpcEpSet *s1, SRpcEpSet *s2) {
    if (s1->numOfEps != s2->numOfEps || s1->inUse != s2->inUse) {
      return false;
    }
+
    for (int32_t i = 0; i < s1->numOfEps; i++) {
      if (s1->port[i] != s2->port[i] 
         || strncmp(s1->fqdn[i], s2->fqdn[i], TSDB_FQDN_LEN) != 0)
@@ -93,6 +93,7 @@ bool tscEpSetIsEqual(SRpcEpSet *s1, SRpcEpSet *s2) {
    }
    return true;
 }
+
 void tscUpdateMgmtEpSet(SSqlObj *pSql, SRpcEpSet *pEpSet) {
   // no need to update if equal
   SRpcCorEpSet *pCorEpSet = pSql->pTscObj->tscCorMgmtEpSet;
@@ -101,37 +102,44 @@ void tscUpdateMgmtEpSet(SSqlObj *pSql, SRpcEpSet *pEpSet) {
   taosCorEndWrite(&pCorEpSet->version);
 }
 
-static void tscDumpEpSetFromVgroupInfo(SRpcEpSet *pEpSet, SCorVgroupInfo *pVgroupInfo) {
+static void tscDumpEpSetFromVgroupInfo(SRpcEpSet *pEpSet, SNewVgroupInfo *pVgroupInfo) {
   if (pVgroupInfo == NULL) { return;}
-  taosCorBeginRead(&pVgroupInfo->version);
   int8_t inUse = pVgroupInfo->inUse;
   pEpSet->inUse = (inUse >= 0 && inUse < TSDB_MAX_REPLICA) ? inUse: 0; 
   pEpSet->numOfEps = pVgroupInfo->numOfEps;  
   for (int32_t i = 0; i < pVgroupInfo->numOfEps; ++i) {
-    tstrncpy(pEpSet->fqdn[i], pVgroupInfo->epAddr[i].fqdn, sizeof(pEpSet->fqdn[i]));
-    pEpSet->port[i] = pVgroupInfo->epAddr[i].port;
+    tstrncpy(pEpSet->fqdn[i], pVgroupInfo->ep[i].fqdn, sizeof(pEpSet->fqdn[i]));
+    pEpSet->port[i] = pVgroupInfo->ep[i].port;
   }
-  taosCorEndRead(&pVgroupInfo->version);
 }
 
 static void tscUpdateVgroupInfo(SSqlObj *pObj, SRpcEpSet *pEpSet) {
   SSqlCmd *pCmd = &pObj->cmd;
   STableMetaInfo *pTableMetaInfo = tscGetTableMetaInfoFromCmd(pCmd, pCmd->clauseIndex, 0);
-  if (pTableMetaInfo == NULL || pTableMetaInfo->pTableMeta == NULL) { return;}
-  SCorVgroupInfo *pVgroupInfo = &pTableMetaInfo->pTableMeta->corVgroupInfo;
+  if (pTableMetaInfo == NULL || pTableMetaInfo->pTableMeta == NULL) {
+    return;
+  }
 
-  taosCorBeginWrite(&pVgroupInfo->version);
-  tscDebug("before: Endpoint in use: %d", pVgroupInfo->inUse);
-  pVgroupInfo->inUse = pEpSet->inUse;
-  pVgroupInfo->numOfEps = pEpSet->numOfEps;
-  for (int32_t i = 0; i < pVgroupInfo->numOfEps; i++) {
-    tfree(pVgroupInfo->epAddr[i].fqdn);
-    pVgroupInfo->epAddr[i].fqdn = strndup(pEpSet->fqdn[i], tListLen(pEpSet->fqdn[i]));
-    pVgroupInfo->epAddr[i].port = pEpSet->port[i];
+  int32_t vgId = pTableMetaInfo->pTableMeta->vgId;
+  if (pTableMetaInfo->pTableMeta->tableType == TSDB_SUPER_TABLE) {
+    assert(vgId == 0);
+    return;
   }
 
-  tscDebug("after: EndPoint in use: %d", pVgroupInfo->inUse);
-  taosCorEndWrite(&pVgroupInfo->version);
+  SNewVgroupInfo vgroupInfo = {.vgId = -1};
+  taosHashGetClone(tscVgroupMap, &vgId, sizeof(vgId), NULL, &vgroupInfo, sizeof(SNewVgroupInfo));
+  assert(vgroupInfo.numOfEps > 0 && vgroupInfo.vgId > 0);
+
+  tscDebug("before: Endpoint in use:%d, numOfEps:%d", vgroupInfo.inUse, vgroupInfo.numOfEps);
+  vgroupInfo.inUse    = pEpSet->inUse;
+  vgroupInfo.numOfEps = pEpSet->numOfEps;
+  for (int32_t i = 0; i < vgroupInfo.numOfEps; i++) {
+    strncpy(vgroupInfo.ep[i].fqdn, pEpSet->fqdn[i], TSDB_FQDN_LEN);
+    vgroupInfo.ep[i].port = pEpSet->port[i];
+  }
+
+  tscDebug("after: EndPoint in use:%d, numOfEps:%d", vgroupInfo.inUse, vgroupInfo.numOfEps);
+  taosHashPut(tscVgroupMap, &vgId, sizeof(vgId), &vgroupInfo, sizeof(SNewVgroupInfo));
 }
 
 void tscProcessHeartBeatRsp(void *param, TAOS_RES *tres, int code) {
@@ -303,7 +311,7 @@ void tscProcessMsgFromServer(SRpcMsg *rpcMsg, SRpcEpSet *pEpSet) {
     return;
   }
 
-  if (pEpSet) {
+  if (pEpSet) { // todo update this
     if (!tscEpSetIsEqual(&pSql->epSet, pEpSet)) {
       if (pCmd->command < TSDB_SQL_MGMT) {
         tscUpdateVgroupInfo(pSql, pEpSet);
@@ -437,7 +445,7 @@ int doProcessSql(SSqlObj *pSql) {
   }
   
   if (pRes->code != TSDB_CODE_SUCCESS) {
-    tscQueueAsyncRes(pSql);
+    tscAsyncResultOnError(pSql);
     return pRes->code;
   }
 
@@ -446,7 +454,7 @@ int doProcessSql(SSqlObj *pSql) {
   // NOTE: if code is TSDB_CODE_SUCCESS, pSql may have been released here already by other threads.
   if (code != TSDB_CODE_SUCCESS) {
     pRes->code = code;
-    tscQueueAsyncRes(pSql);
+    tscAsyncResultOnError(pSql);
     return code;
   }
   
@@ -549,7 +557,10 @@ int tscBuildSubmitMsg(SSqlObj *pSql, SSqlInfo *pInfo) {
 
   // pSql->cmd.payloadLen is set during copying data into payload
   pSql->cmd.msgType = TSDB_MSG_TYPE_SUBMIT;
-  tscDumpEpSetFromVgroupInfo(&pSql->epSet, &pTableMeta->corVgroupInfo);
+
+  SNewVgroupInfo vgroupInfo = {0};
+  taosHashGetClone(tscVgroupMap, &pTableMeta->vgId, sizeof(pTableMeta->vgId), NULL, &vgroupInfo, sizeof(SNewVgroupInfo));
+  tscDumpEpSetFromVgroupInfo(&pSql->epSet, &vgroupInfo);
 
   tscDebug("%p build submit msg, vgId:%d numOfTables:%d numberOfEP:%d", pSql, pTableMeta->vgId, pSql->cmd.numOfTablesInSubmit,
       pSql->epSet.numOfEps);
@@ -559,9 +570,11 @@ int tscBuildSubmitMsg(SSqlObj *pSql, SSqlInfo *pInfo) {
 /*
  * for table query, simply return the size <= 1k
  */
-static int32_t tscEstimateQueryMsgSize(SSqlCmd *pCmd, int32_t clauseIndex) {
+static int32_t tscEstimateQueryMsgSize(SSqlObj *pSql, int32_t clauseIndex) {
   const static int32_t MIN_QUERY_MSG_PKT_SIZE = TSDB_MAX_BYTES_PER_ROW * 5;
-  SQueryInfo *         pQueryInfo = tscGetQueryInfoDetail(pCmd, clauseIndex);
+
+  SSqlCmd* pCmd = &pSql->cmd;
+  SQueryInfo *pQueryInfo = tscGetQueryInfoDetail(pCmd, clauseIndex);
 
   int32_t srcColListSize = (int32_t)(taosArrayGetSize(pQueryInfo->colList) * sizeof(SColumnInfo));
 
@@ -569,6 +582,8 @@ static int32_t tscEstimateQueryMsgSize(SSqlCmd *pCmd, int32_t clauseIndex) {
   int32_t exprSize = (int32_t)(sizeof(SSqlFuncMsg) * numOfExprs * 2);
 
   int32_t tsBufSize = (pQueryInfo->tsBuf != NULL) ? pQueryInfo->tsBuf->fileSize : 0;
+  int32_t sqlLen = (int32_t) strlen(pSql->sqlstr) + 1;
+
 
   int32_t tableSerialize = 0;
   STableMetaInfo *pTableMetaInfo = tscGetMetaInfo(pQueryInfo, 0);
@@ -585,7 +600,7 @@ static int32_t tscEstimateQueryMsgSize(SSqlCmd *pCmd, int32_t clauseIndex) {
   }
 
   return MIN_QUERY_MSG_PKT_SIZE + minMsgSize() + sizeof(SQueryTableMsg) + srcColListSize + exprSize + tsBufSize +
-         tableSerialize + 4096;
+         tableSerialize + sqlLen + 4096;
 }
 
 static char *doSerializeTableInfo(SQueryTableMsg* pQueryMsg, SSqlObj *pSql, char *pMsg) {
@@ -611,7 +626,10 @@ static char *doSerializeTableInfo(SQueryTableMsg* pQueryMsg, SSqlObj *pSql, char
       tscDebug("%p query on stable, vgIndex:%d, numOfVgroups:%d", pSql, index, pTableMetaInfo->vgroupList->numOfVgroups);
     } else {
       vgId = pTableMeta->vgId;
-      tscDumpEpSetFromVgroupInfo(&pSql->epSet, &pTableMeta->corVgroupInfo);
+
+      SNewVgroupInfo vgroupInfo = {0};
+      taosHashGetClone(tscVgroupMap, &pTableMeta->vgId, sizeof(pTableMeta->vgId), NULL, &vgroupInfo, sizeof(SNewVgroupInfo));
+      tscDumpEpSetFromVgroupInfo(&pSql->epSet, &vgroupInfo);
     }
 
     pSql->epSet.inUse = rand()%pSql->epSet.numOfEps;
@@ -662,7 +680,7 @@ static char *doSerializeTableInfo(SQueryTableMsg* pQueryMsg, SSqlObj *pSql, char
 int tscBuildQueryMsg(SSqlObj *pSql, SSqlInfo *pInfo) {
   SSqlCmd *pCmd = &pSql->cmd;
 
-  int32_t size = tscEstimateQueryMsgSize(pCmd, pCmd->clauseIndex);
+  int32_t size = tscEstimateQueryMsgSize(pSql, pCmd->clauseIndex);
 
   if (TSDB_CODE_SUCCESS != tscAllocPayload(pCmd, size)) {
     tscError("%p failed to malloc for query msg", pSql);
@@ -695,6 +713,7 @@ int tscBuildQueryMsg(SSqlObj *pSql, SSqlInfo *pInfo) {
   tstrncpy(pQueryMsg->version, version, tListLen(pQueryMsg->version));
 
   int32_t numOfTags = (int32_t)taosArrayGetSize(pTableMetaInfo->tagColList);
+  int32_t sqlLen = (int32_t) strlen(pSql->sqlstr);
 
   if (pQueryInfo->order.order == TSDB_ORDER_ASC) {
     pQueryMsg->window.skey = htobe64(pQueryInfo->window.skey);
@@ -718,9 +737,11 @@ int tscBuildQueryMsg(SSqlObj *pSql, SSqlInfo *pInfo) {
   pQueryMsg->interval.offsetUnit   = pQueryInfo->interval.offsetUnit;
   pQueryMsg->numOfGroupCols = htons(pQueryInfo->groupbyExpr.numOfGroupCols);
   pQueryMsg->tagNameRelType = htons(pQueryInfo->tagCond.relType);
+  pQueryMsg->tbnameCondLen  = htonl(pQueryInfo->tagCond.tbnameCond.len);
   pQueryMsg->numOfTags      = htonl(numOfTags);
   pQueryMsg->queryType      = htonl(pQueryInfo->type);
   pQueryMsg->vgroupLimit    = htobe64(pQueryInfo->vgroupLimit);
+  pQueryMsg->sqlstrLen      = htonl(sqlLen);
 
   size_t numOfOutput = tscSqlExprNumOfExprs(pQueryInfo);
   pQueryMsg->numOfOutput = htons((int16_t)numOfOutput);  // this is the stage one output column number
@@ -955,12 +976,10 @@ int tscBuildQueryMsg(SSqlObj *pSql, SSqlInfo *pInfo) {
     }
   }
 
-  if (pQueryInfo->tagCond.tbnameCond.cond == NULL) {
-    *pMsg = 0;
-    pMsg++;
-  } else {
-    strcpy(pMsg, pQueryInfo->tagCond.tbnameCond.cond);
-    pMsg += strlen(pQueryInfo->tagCond.tbnameCond.cond) + 1;
+  SCond* pCond = &pQueryInfo->tagCond.tbnameCond;
+  if (pCond->len > 0) {
+    strncpy(pMsg, pCond->cond, pCond->len);
+    pMsg += pCond->len;
   }
 
   // compressed ts block
@@ -981,6 +1000,9 @@ int tscBuildQueryMsg(SSqlObj *pSql, SSqlInfo *pInfo) {
     pQueryMsg->tsNumOfBlocks = htonl(pQueryMsg->tsNumOfBlocks);
   }
 
+  memcpy(pMsg, pSql->sqlstr, sqlLen);
+  pMsg += sqlLen;
+
   int32_t msgLen = (int32_t)(pMsg - pCmd->payload);
 
   tscDebug("%p msg built success, len:%d bytes", pSql, msgLen);
@@ -1447,10 +1469,14 @@ int tscBuildUpdateTagMsg(SSqlObj* pSql, SSqlInfo *pInfo) {
   SUpdateTableTagValMsg* pUpdateMsg = (SUpdateTableTagValMsg*) pCmd->payload;
   pCmd->payloadLen = htonl(pUpdateMsg->head.contLen);
 
-  SQueryInfo *    pQueryInfo = tscGetQueryInfoDetail(pCmd, 0);
-  STableMetaInfo *pTableMetaInfo = tscGetMetaInfo(pQueryInfo, 0);
+  SQueryInfo *pQueryInfo = tscGetQueryInfoDetail(pCmd, 0);
+  STableMeta *pTableMeta = tscGetMetaInfo(pQueryInfo, 0)->pTableMeta;
 
-  tscDumpEpSetFromVgroupInfo(&pSql->epSet, &pTableMetaInfo->pTableMeta->corVgroupInfo);
+  SNewVgroupInfo vgroupInfo = {.vgId = -1};
+  taosHashGetClone(tscVgroupMap, &pTableMeta->vgId, sizeof(pTableMeta->vgId), NULL, &vgroupInfo, sizeof(SNewVgroupInfo));
+  assert(vgroupInfo.vgId > 0);
+
+  tscDumpEpSetFromVgroupInfo(&pSql->epSet, &vgroupInfo);
 
   return TSDB_CODE_SUCCESS;
 }
@@ -1516,7 +1542,7 @@ static int tscLocalResultCommonBuilder(SSqlObj *pSql, int32_t numOfRes) {
     if (code == TSDB_CODE_SUCCESS) {
       (*pSql->fp)(pSql->param, pSql, pSql->res.numOfRows);
     } else {
-      tscQueueAsyncRes(pSql);
+      tscAsyncResultOnError(pSql);
     }
   }
 
@@ -1545,7 +1571,7 @@ int tscProcessRetrieveLocalMergeRsp(SSqlObj *pSql) {
 
   int32_t code = pRes->code;
   if (pRes->code != TSDB_CODE_SUCCESS) {
-    tscQueueAsyncRes(pSql);
+    tscAsyncResultOnError(pSql);
     return code;
   }
 
@@ -1564,7 +1590,7 @@ int tscProcessRetrieveLocalMergeRsp(SSqlObj *pSql) {
   if (pRes->code == TSDB_CODE_SUCCESS) {
     (*pSql->fp)(pSql->param, pSql, pRes->numOfRows);
   } else {
-    tscQueueAsyncRes(pSql);
+    tscAsyncResultOnError(pSql);
   }
 
   return code;
@@ -1808,19 +1834,46 @@ int tscProcessTableMetaRsp(SSqlObj *pSql) {
     pSchema++;
   }
 
-  size_t size = 0;
-  STableMeta* pTableMeta = tscCreateTableMetaFromMsg(pMetaMsg, &size);
+  STableMeta* pTableMeta = tscCreateTableMetaFromMsg(pMetaMsg);
   
-  // todo add one more function: taosAddDataIfNotExists();
   STableMetaInfo *pTableMetaInfo = tscGetTableMetaInfoFromCmd(&pSql->cmd, 0, 0);
   assert(pTableMetaInfo->pTableMeta == NULL);
 
-  pTableMetaInfo->pTableMeta = (STableMeta *) taosCachePut(tscMetaCache, pTableMetaInfo->name,
-      strlen(pTableMetaInfo->name), pTableMeta, size, tsTableMetaKeepTimer * 1000);
+  if (pTableMeta->tableType == TSDB_CHILD_TABLE) {
+    // check if super table hashmap or not
+    int32_t len = (int32_t) strnlen(pTableMeta->sTableName, TSDB_TABLE_FNAME_LEN);
 
-  if (pTableMetaInfo->pTableMeta == NULL) {
-    free(pTableMeta);
-    return TSDB_CODE_TSC_OUT_OF_MEMORY;
+    // super tableMeta data alreay exists, create it according to tableMeta and add it to hash map
+    STableMeta* pSupTableMeta = createSuperTableMeta(pMetaMsg);
+
+    uint32_t size = tscGetTableMetaSize(pSupTableMeta);
+    int32_t code = taosHashPut(tscTableMetaInfo, pTableMeta->sTableName, len, pSupTableMeta, size);
+    assert(code == TSDB_CODE_SUCCESS);
+
+    tfree(pSupTableMeta);
+
+    CChildTableMeta* cMeta = tscCreateChildMeta(pTableMeta);
+    taosHashPut(tscTableMetaInfo, pTableMetaInfo->name, strlen(pTableMetaInfo->name), cMeta, sizeof(CChildTableMeta));
+    tfree(cMeta);
+  } else {
+    uint32_t s = tscGetTableMetaSize(pTableMeta);
+    taosHashPut(tscTableMetaInfo, pTableMetaInfo->name, strlen(pTableMetaInfo->name), pTableMeta, s);
+  }
+
+  // update the vgroupInfo if needed
+  if (pTableMeta->vgId > 0) {
+    int32_t vgId = pTableMeta->vgId;
+    assert(pTableMeta->tableType != TSDB_SUPER_TABLE);
+
+    SNewVgroupInfo vgroupInfo = {.inUse = -1};
+    taosHashGetClone(tscVgroupMap, &vgId, sizeof(vgId), NULL, &vgroupInfo, sizeof(SNewVgroupInfo));
+
+    if (((vgroupInfo.inUse >= 0) && !vgroupInfoIdentical(&vgroupInfo, &pMetaMsg->vgroup)) ||
+        (vgroupInfo.inUse < 0)) {  // vgroup info exists, compare with it
+      vgroupInfo = createNewVgroupInfo(&pMetaMsg->vgroup);
+      taosHashPut(tscVgroupMap, &vgId, sizeof(vgId), &vgroupInfo, sizeof(vgroupInfo));
+      tscDebug("add new VgroupInfo, vgId:%d, total:%d", vgId, (int32_t) taosHashGetSize(tscVgroupMap));
+    }
   }
 
   tscDebug("%p recv table meta, uid:%"PRId64 ", tid:%d, name:%s", pSql, pTableMeta->id.uid, pTableMeta->id.tid, pTableMetaInfo->name);
@@ -1831,8 +1884,8 @@ int tscProcessTableMetaRsp(SSqlObj *pSql) {
 
 /**
  *  multi table meta rsp pkg format:
- *  | STaosRsp | ieType | SMultiTableInfoMsg | SMeterMeta0 | SSchema0 | SMeterMeta1 | SSchema1 | SMeterMeta2 | SSchema2
- *  |...... 1B        1B            4B
+ *  | STaosRsp | SMultiTableInfoMsg | SMeterMeta0 | SSchema0 | SMeterMeta1 | SSchema1 | SMeterMeta2 | SSchema2
+ *  |...... 1B                    4B
  **/
 int tscProcessMultiMeterMetaRsp(SSqlObj *pSql) {
 #if 0
@@ -1986,14 +2039,10 @@ int tscProcessSTableVgroupRsp(SSqlObj *pSql) {
   return pSql->res.code;
 }
 
-/*
- * current process do not use the cache at all
- */
 int tscProcessShowRsp(SSqlObj *pSql) {
   STableMetaMsg *pMetaMsg;
   SShowRsp *     pShow;
   SSchema *      pSchema;
-  char           key[20];
 
   SSqlRes *pRes = &pSql->res;
   SSqlCmd *pCmd = &pSql->cmd;
@@ -2018,20 +2067,10 @@ int tscProcessShowRsp(SSqlObj *pSql) {
     pSchema++;
   }
 
-  key[0] = pCmd->msgType + 'a';
-  strcpy(key + 1, "showlist");
-
-  if (pTableMetaInfo->pTableMeta != NULL) {
-    taosCacheRelease(tscMetaCache, (void *)&(pTableMetaInfo->pTableMeta), false);
-  }
-
-  size_t size = 0;
-  STableMeta* pTableMeta = tscCreateTableMetaFromMsg(pMetaMsg, &size);
+  tfree(pTableMetaInfo->pTableMeta);
+  pTableMetaInfo->pTableMeta = tscCreateTableMetaFromMsg(pMetaMsg);
 
-  pTableMetaInfo->pTableMeta = taosCachePut(tscMetaCache, key, strlen(key), (char *)pTableMeta, size,
-      tsTableMetaKeepTimer * 1000);
   SSchema *pTableSchema = tscGetTableSchema(pTableMetaInfo->pTableMeta);
-
   if (pQueryInfo->colList == NULL) {
     pQueryInfo->colList = taosArrayInit(4, POINTER_BYTES);
   }
@@ -2054,12 +2093,9 @@ int tscProcessShowRsp(SSqlObj *pSql) {
   
   pCmd->numOfCols = pQueryInfo->fieldsInfo.numOfOutput;
   tscFieldInfoUpdateOffset(pQueryInfo);
-  
-  tfree(pTableMeta);
   return 0;
 }
 
-// TODO multithread problem
 static void createHBObj(STscObj* pObj) {
   if (pObj->hbrid != 0) {
     return;
@@ -2141,51 +2177,34 @@ int tscProcessUseDbRsp(SSqlObj *pSql) {
 
 int tscProcessDropDbRsp(SSqlObj *pSql) {
   pSql->pTscObj->db[0] = 0;
-  taosCacheEmpty(tscMetaCache);
+  taosHashEmpty(tscTableMetaInfo);
   return 0;
 }
 
 int tscProcessDropTableRsp(SSqlObj *pSql) {
   STableMetaInfo *pTableMetaInfo = tscGetTableMetaInfoFromCmd(&pSql->cmd, 0, 0);
 
-  STableMeta *pTableMeta = taosCacheAcquireByKey(tscMetaCache, pTableMetaInfo->name, strlen(pTableMetaInfo->name));
-  if (pTableMeta == NULL) { /* not in cache, abort */
-    return 0;
-  }
+  //The cached tableMeta is expired in this case, so clean it in hash table
+  taosHashRemove(tscTableMetaInfo, pTableMetaInfo->name, strnlen(pTableMetaInfo->name, TSDB_TABLE_FNAME_LEN));
+  tscDebug("%p remove table meta after drop table:%s, numOfRemain:%d", pSql, pTableMetaInfo->name,
+           (int32_t) taosHashGetSize(tscTableMetaInfo));
 
-  /*
-   * 1. if a user drops one table, which is the only table in a vnode, remove operation will incur vnode to be removed.
-   * 2. Then, a user creates a new metric followed by a table with identical name of removed table but different schema,
-   * here the table will reside in a new vnode.
-   * The cached information is expired, however, we may have lost the ref of original meter. So, clear whole cache
-   * instead.
-   */
-  tscDebug("%p force release table meta after drop table:%s", pSql, pTableMetaInfo->name);
-  taosCacheRelease(tscMetaCache, (void **)&pTableMeta, true);
   assert(pTableMetaInfo->pTableMeta == NULL);
-
   return 0;
 }
 
 int tscProcessAlterTableMsgRsp(SSqlObj *pSql) {
   STableMetaInfo *pTableMetaInfo = tscGetTableMetaInfoFromCmd(&pSql->cmd, 0, 0);
 
-  STableMeta *pTableMeta = taosCacheAcquireByKey(tscMetaCache, pTableMetaInfo->name, strlen(pTableMetaInfo->name));
-  if (pTableMeta == NULL) { /* not in cache, abort */
-    return 0;
-  }
+  char* name = pTableMetaInfo->name;
+  tscDebug("%p remove tableMeta in hashMap after alter-table: %s", pSql, name);
 
-  tscDebug("%p force release metermeta in cache after alter-table: %s", pSql, pTableMetaInfo->name);
-  taosCacheRelease(tscMetaCache, (void **)&pTableMeta, true);
-
-  if (pTableMetaInfo->pTableMeta) {
   bool isSuperTable = UTIL_TABLE_IS_SUPER_TABLE(pTableMetaInfo);
-    taosCacheRelease(tscMetaCache, (void **)&(pTableMetaInfo->pTableMeta), true);
+  taosHashRemove(tscTableMetaInfo, name, strnlen(name, TSDB_TABLE_FNAME_LEN));
+  tfree(pTableMetaInfo->pTableMeta);
 
-    if (isSuperTable) {  // if it is a super table, reset whole query cache
-      tscDebug("%p reset query cache since table:%s is stable", pSql, pTableMetaInfo->name);
-      taosCacheEmpty(tscMetaCache);
-    }
+  if (isSuperTable) {  // if it is a super table, iterate the hashTable and remove all the childTableMeta
+    taosHashEmpty(tscTableMetaInfo);
   }
 
   return 0;
@@ -2195,6 +2214,7 @@ int tscProcessAlterDbMsgRsp(SSqlObj *pSql) {
   UNUSED(pSql);
   return 0;
 }
+
 int tscProcessShowCreateRsp(SSqlObj *pSql) {
   return tscLocalResultCommonBuilder(pSql, 1);
 }
@@ -2315,7 +2335,7 @@ static int32_t getTableMetaFromMnode(SSqlObj *pSql, STableMetaInfo *pTableMetaIn
 
   int32_t code = tscProcessSql(pNew);
   if (code == TSDB_CODE_SUCCESS) {
-    code = TSDB_CODE_TSC_ACTION_IN_PROGRESS;  // notify upper application that current process need to be terminated
+    code = TSDB_CODE_TSC_ACTION_IN_PROGRESS;  // notify application that current process needs to be terminated
   }
 
   return code;
@@ -2323,17 +2343,25 @@ static int32_t getTableMetaFromMnode(SSqlObj *pSql, STableMetaInfo *pTableMetaIn
 
 int32_t tscGetTableMeta(SSqlObj *pSql, STableMetaInfo *pTableMetaInfo) {
   assert(strlen(pTableMetaInfo->name) != 0);
+  tfree(pTableMetaInfo->pTableMeta);
 
-  // If this STableMetaInfo owns a table meta, release it first
-  if (pTableMetaInfo->pTableMeta != NULL) {
-    taosCacheRelease(tscMetaCache, (void **)&(pTableMetaInfo->pTableMeta), false);
-  }
+  uint32_t size = tscGetTableMetaMaxSize();
+  pTableMetaInfo->pTableMeta = calloc(1, size);
 
-  pTableMetaInfo->pTableMeta = (STableMeta *)taosCacheAcquireByKey(tscMetaCache, pTableMetaInfo->name, strlen(pTableMetaInfo->name));
-  if (pTableMetaInfo->pTableMeta != NULL) {
-    STableComInfo tinfo = tscGetTableInfo(pTableMetaInfo->pTableMeta);
-    tscDebug("%p retrieve table Meta from cache, the number of columns:%d, numOfTags:%d, %p", pSql, tinfo.numOfColumns,
-             tinfo.numOfTags, pTableMetaInfo->pTableMeta);
+  pTableMetaInfo->pTableMeta->tableInfo.numOfColumns  = -1;
+  int32_t len = (int32_t) strlen(pTableMetaInfo->name);
+
+  taosHashGetClone(tscTableMetaInfo, pTableMetaInfo->name, len, NULL, pTableMetaInfo->pTableMeta, -1);
+
+  // TODO resize the tableMeta
+  STableMeta* pMeta = pTableMetaInfo->pTableMeta;
+  if (pMeta->id.uid > 0) {
+    if (pMeta->tableType == TSDB_CHILD_TABLE) {
+      int32_t code = tscCreateTableMetaFromCChildMeta(pTableMetaInfo->pTableMeta, pTableMetaInfo->name);
+      if (code != TSDB_CODE_SUCCESS) {
+        return getTableMetaFromMnode(pSql, pTableMetaInfo);
+      }
+    }
 
     return TSDB_CODE_SUCCESS;
   }
@@ -2347,7 +2375,7 @@ int tscGetTableMetaEx(SSqlObj *pSql, STableMetaInfo *pTableMetaInfo, bool create
 }
 
 /**
- * retrieve table meta from mnode, and update the local table meta cache.
+ * retrieve table meta from mnode, and update the local table meta hashmap.
  * @param pSql          sql object
  * @param tableIndex    table index
  * @return              status code
@@ -2355,16 +2383,18 @@ int tscGetTableMetaEx(SSqlObj *pSql, STableMetaInfo *pTableMetaInfo, bool create
 int tscRenewTableMeta(SSqlObj *pSql, int32_t tableIndex) {
   SSqlCmd *pCmd = &pSql->cmd;
 
-  SQueryInfo *    pQueryInfo = tscGetQueryInfoDetail(pCmd, 0);
+  SQueryInfo     *pQueryInfo = tscGetQueryInfoDetail(pCmd, 0);
   STableMetaInfo *pTableMetaInfo = tscGetMetaInfo(pQueryInfo, tableIndex);
+  const char* name = pTableMetaInfo->name;
 
   STableMeta* pTableMeta = pTableMetaInfo->pTableMeta;
-  if (pTableMetaInfo->pTableMeta) {
-    tscDebug("%p update table meta, old meta numOfTags:%d, numOfCols:%d, uid:%" PRId64 ", addr:%p", pSql,
-             tscGetNumOfTags(pTableMeta), tscGetNumOfColumns(pTableMeta), pTableMeta->id.uid, pTableMeta);
+  if (pTableMeta) {
+    tscDebug("%p update table meta:%s, old meta numOfTags:%d, numOfCols:%d, uid:%" PRId64, pSql, name,
+        tscGetNumOfTags(pTableMeta), tscGetNumOfColumns(pTableMeta), pTableMeta->id.uid);
   }
 
-  taosCacheRelease(tscMetaCache, (void **)&(pTableMetaInfo->pTableMeta), true);
+  // remove stored tableMeta info in hash table
+  taosHashRemove(tscTableMetaInfo, name, strnlen(name, TSDB_TABLE_FNAME_LEN));
   return getTableMetaFromMnode(pSql, pTableMetaInfo);
 }
 
@@ -2405,7 +2435,7 @@ int tscGetSTableVgroupInfo(SSqlObj *pSql, int32_t clauseIndex) {
   SQueryInfo *pQueryInfo = tscGetQueryInfoDetail(pCmd, clauseIndex);
   for (int32_t i = 0; i < pQueryInfo->numOfTables; ++i) {
     STableMetaInfo *pMInfo = tscGetMetaInfo(pQueryInfo, i);
-    STableMeta *pTableMeta = taosCacheAcquireByData(tscMetaCache, pMInfo->pTableMeta);
+    STableMeta* pTableMeta = tscTableMetaClone(pMInfo->pTableMeta);
     tscAddTableMetaInfo(pNewQueryInfo, pMInfo->name, pTableMeta, NULL, pMInfo->tagColList, pMInfo->pVgroupTables);
   }
 

src/client/src/tscSql.c

@@ -709,7 +709,7 @@ static void tscKillSTableQuery(SSqlObj *pSql) {
       pSubObj->rpcRid = -1;
     }
 
-    tscQueueAsyncRes(pSubObj);
+    tscAsyncResultOnError(pSubObj);
     taosReleaseRef(tscObjRef, pSubObj->self);
   }
 
@@ -745,7 +745,7 @@ void taos_stop_query(TAOS_RES *res) {
         pSql->rpcRid = -1;
       }
 
-      tscQueueAsyncRes(pSql);
+      tscAsyncResultOnError(pSql);
     }
   }
 
@@ -909,7 +909,7 @@ int taos_validate_sql(TAOS *taos, const char *sql) {
 
 static int tscParseTblNameList(SSqlObj *pSql, const char *tblNameList, int32_t tblListLen) {
   // must before clean the sqlcmd object
-  tscResetSqlCmdObj(&pSql->cmd, false);
+  tscResetSqlCmdObj(&pSql->cmd);
 
   SSqlCmd *pCmd = &pSql->cmd;
 

src/client/src/tscStream.c

@@ -167,7 +167,9 @@ static void tscProcessStreamQueryCallback(void *param, TAOS_RES *tres, int numOf
              retryDelay);
 
     STableMetaInfo* pTableMetaInfo = tscGetTableMetaInfoFromCmd(&pStream->pSql->cmd, 0, 0);
-    taosCacheRelease(tscMetaCache, (void**)&(pTableMetaInfo->pTableMeta), true);
+
+    char* name = pTableMetaInfo->name;
+    taosHashRemove(tscTableMetaInfo, name, strnlen(name, TSDB_TABLE_FNAME_LEN));
     pTableMetaInfo->vgroupList = tscVgroupInfoClear(pTableMetaInfo->vgroupList);
 
     tscSetRetryTimer(pStream, pStream->pSql, retryDelay);
@@ -269,9 +271,8 @@ static void tscProcessStreamRetrieveResult(void *param, TAOS_RES *res, int numOf
     tscDebug("%p stream:%p, query on:%s, fetch result completed, fetched rows:%" PRId64, pSql, pStream, pTableMetaInfo->name,
              pStream->numOfRes);
 
-    // release the metric/meter meta information reference, so data in cache can be updated
+    tfree(pTableMetaInfo->pTableMeta);
 
-    taosCacheRelease(tscMetaCache, (void**)&(pTableMetaInfo->pTableMeta), false);
     tscFreeSqlResult(pSql);
     tfree(pSql->pSubs);
     pSql->subState.numOfSub = 0;

src/client/src/tscSubquery.c

@@ -779,7 +779,7 @@ static void tidTagRetrieveCallback(void* param, TAOS_RES* tres, int32_t numOfRow
     pParentSql->res.code = numOfRows;
     quitAllSubquery(pParentSql, pSupporter);
 
-    tscQueueAsyncRes(pParentSql);
+    tscAsyncResultOnError(pParentSql);
     return;
   }
 
@@ -796,7 +796,7 @@ static void tidTagRetrieveCallback(void* param, TAOS_RES* tres, int32_t numOfRow
       pParentSql->res.code = TAOS_SYSTEM_ERROR(errno);
       quitAllSubquery(pParentSql, pSupporter);
 
-      tscQueueAsyncRes(pParentSql);
+      tscAsyncResultOnError(pParentSql);
       return;
     }
 
@@ -845,7 +845,7 @@ static void tidTagRetrieveCallback(void* param, TAOS_RES* tres, int32_t numOfRow
   if (code != TSDB_CODE_SUCCESS) {
     freeJoinSubqueryObj(pParentSql);
     pParentSql->res.code = code;
-    tscQueueAsyncRes(pParentSql);
+    tscAsyncResultOnError(pParentSql);
 
     taosArrayDestroy(s1);
     taosArrayDestroy(s2);
@@ -916,7 +916,7 @@ static void tsCompRetrieveCallback(void* param, TAOS_RES* tres, int32_t numOfRow
     pParentSql->res.code = numOfRows;
     quitAllSubquery(pParentSql, pSupporter);
 
-    tscQueueAsyncRes(pParentSql);
+    tscAsyncResultOnError(pParentSql);
     return;
   }
 
@@ -930,7 +930,7 @@ static void tsCompRetrieveCallback(void* param, TAOS_RES* tres, int32_t numOfRow
       tscError("%p invalid ts comp file from vnode, abort subquery, file size:%d", pSql, numOfRows);
 
       pParentSql->res.code = TAOS_SYSTEM_ERROR(errno);
-      tscQueueAsyncRes(pParentSql);
+      tscAsyncResultOnError(pParentSql);
 
       return;
     }
@@ -1028,7 +1028,7 @@ static void joinRetrieveFinalResCallback(void* param, TAOS_RES* tres, int numOfR
     pParentSql->res.code = numOfRows;
     tscError("%p retrieve failed, index:%d, code:%s", pSql, pSupporter->subqueryIndex, tstrerror(numOfRows));
 
-    tscQueueAsyncRes(pParentSql);
+    tscAsyncResultOnError(pParentSql);
     return;
   }
 
@@ -1155,7 +1155,7 @@ void tscFetchDatablockForSubquery(SSqlObj* pSql) {
     if (pSql->res.code == TSDB_CODE_SUCCESS) {
       (*pSql->fp)(pSql->param, pSql, 0);
     } else {
-      tscQueueAsyncRes(pSql);
+      tscAsyncResultOnError(pSql);
     }
 
     return;
@@ -1233,7 +1233,7 @@ void tscFetchDatablockForSubquery(SSqlObj* pSql) {
     if (pSql->res.code == TSDB_CODE_SUCCESS) {
       (*pSql->fp)(pSql->param, pSql, 0);
     } else {
-      tscQueueAsyncRes(pSql);
+      tscAsyncResultOnError(pSql);
     }
 
     return;
@@ -1344,7 +1344,7 @@ void tscJoinQueryCallback(void* param, TAOS_RES* tres, int code) {
   if (pParentSql->res.code != TSDB_CODE_SUCCESS) {
     tscError("%p abort query due to other subquery failure. code:%d, global code:%d", pSql, code, pParentSql->res.code);
     quitAllSubquery(pParentSql, pSupporter);
-    tscQueueAsyncRes(pParentSql);
+    tscAsyncResultOnError(pParentSql);
 
     return;
   }
@@ -1357,7 +1357,7 @@ void tscJoinQueryCallback(void* param, TAOS_RES* tres, int code) {
     pParentSql->res.code = code;
 
     quitAllSubquery(pParentSql, pSupporter);
-    tscQueueAsyncRes(pParentSql);
+    tscAsyncResultOnError(pParentSql);
 
     return;
   }
@@ -1403,7 +1403,7 @@ void tscJoinQueryCallback(void* param, TAOS_RES* tres, int code) {
     if (pParentSql->res.code == TSDB_CODE_SUCCESS) {
       (*pParentSql->fp)(pParentSql->param, pParentSql, 0);
     } else {
-      tscQueueAsyncRes(pParentSql);
+      tscAsyncResultOnError(pParentSql);
     }
   }
 }
@@ -1612,7 +1612,7 @@ void tscHandleMasterJoinQuery(SSqlObj* pSql) {
 
   _error:
   pRes->code = code;
-  tscQueueAsyncRes(pSql);
+  tscAsyncResultOnError(pSql);
 }
 
 static void doCleanupSubqueries(SSqlObj *pSql, int32_t numOfSubs) {
@@ -1666,7 +1666,7 @@ int32_t tscHandleMasterSTableQuery(SSqlObj *pSql) {
   int32_t ret = tscLocalReducerEnvCreate(pSql, &pMemoryBuf, &pDesc, &pModel, &pFinalModel, nBufferSize);
   if (ret != 0) {
     pRes->code = TSDB_CODE_TSC_OUT_OF_MEMORY;
-    tscQueueAsyncRes(pSql);
+    tscAsyncResultOnError(pSql);
     tfree(pMemoryBuf);
     return ret;
   }
@@ -1680,7 +1680,7 @@ int32_t tscHandleMasterSTableQuery(SSqlObj *pSql) {
     pRes->code = TSDB_CODE_TSC_OUT_OF_MEMORY;
     tscLocalReducerEnvDestroy(pMemoryBuf, pDesc, pModel, pFinalModel,pState->numOfSub);
 
-    tscQueueAsyncRes(pSql);
+    tscAsyncResultOnError(pSql);
     return ret;
   }
 
@@ -1890,7 +1890,7 @@ void tscHandleSubqueryError(SRetrieveSupport *trsupport, SSqlObj *pSql, int numO
     (*pParentSql->fp)(pParentSql->param, pParentSql, pParentSql->res.code);
   } else {  // regular super table query
     if (pParentSql->res.code != TSDB_CODE_SUCCESS) {
-      tscQueueAsyncRes(pParentSql);
+      tscAsyncResultOnError(pParentSql);
     }
   }
 }
@@ -1968,7 +1968,7 @@ static void tscAllDataRetrievedFromDnode(SRetrieveSupport *trsupport, SSqlObj* p
   if (pParentSql->res.code == TSDB_CODE_SUCCESS) {
     (*pParentSql->fp)(pParentSql->param, pParentSql, 0);
   } else {
-    tscQueueAsyncRes(pParentSql);
+    tscAsyncResultOnError(pParentSql);
   }
 }
 
@@ -2220,7 +2220,7 @@ static void multiVnodeInsertFinalize(void* param, TAOS_RES* tres, int numOfRows)
     (*pParentObj->fp)(pParentObj->param, pParentObj, v);
   } else {
     if (!needRetryInsert(pParentObj, numOfSub)) {
-      tscQueueAsyncRes(pParentObj);
+      tscAsyncResultOnError(pParentObj);
       return;
     }
 
@@ -2231,7 +2231,7 @@ static void multiVnodeInsertFinalize(void* param, TAOS_RES* tres, int numOfRows)
         numOfFailed += 1;
 
         // clean up tableMeta in cache
-        tscFreeQueryInfo(&pSql->cmd, true);
+        tscFreeQueryInfo(&pSql->cmd);
         SQueryInfo* pQueryInfo = tscGetQueryInfoDetailSafely(&pSql->cmd, 0);
         STableMetaInfo* pMasterTableMetaInfo = tscGetTableMetaInfoFromCmd(&pParentObj->cmd, pSql->cmd.clauseIndex, 0);
         tscAddTableMetaInfo(pQueryInfo, pMasterTableMetaInfo->name, NULL, NULL, NULL, NULL);
@@ -2243,15 +2243,16 @@ static void multiVnodeInsertFinalize(void* param, TAOS_RES* tres, int numOfRows)
     tscError("%p Async insertion completed, total inserted:%d rows, numOfFailed:%d, numOfTotal:%d", pParentObj,
              pParentObj->res.numOfRows, numOfFailed, numOfSub);
 
-    tscDebug("%p cleanup %d tableMeta in cache", pParentObj, pParentObj->cmd.numOfTables);
+    tscDebug("%p cleanup %d tableMeta in hashTable", pParentObj, pParentObj->cmd.numOfTables);
     for(int32_t i = 0; i < pParentObj->cmd.numOfTables; ++i) {
-      taosCacheRelease(tscMetaCache, (void**)&(pParentObj->cmd.pTableMetaList[i]), true);
+      char* name = pParentObj->cmd.pTableNameList[i];
+      taosHashRemove(tscTableMetaInfo, name, strnlen(name, TSDB_TABLE_FNAME_LEN));
     }
 
     pParentObj->cmd.parseFinished = false;
     pParentObj->subState.numOfRemain = numOfFailed;
 
-    tscResetSqlCmdObj(&pParentObj->cmd, false);
+    tscResetSqlCmdObj(&pParentObj->cmd);
 
     // in case of insert, redo parsing the sql string and build new submit data block for two reasons:
     // 1. the table Id(tid & uid) may have been update, the submit block needs to be updated accordingly.
@@ -2264,7 +2265,7 @@ static void multiVnodeInsertFinalize(void* param, TAOS_RES* tres, int numOfRows)
 
     if (code != TSDB_CODE_SUCCESS) {
       pParentObj->res.code = code;
-      tscQueueAsyncRes(pParentObj);
+      tscAsyncResultOnError(pParentObj);
       return;
     }
 
@@ -2288,7 +2289,7 @@ int32_t tscHandleInsertRetry(SSqlObj* pParent, SSqlObj* pSql) {
   int32_t code = tscCopyDataBlockToPayload(pSql, pTableDataBlock);
 
   if ((pRes->code = code)!= TSDB_CODE_SUCCESS) {
-    tscQueueAsyncRes(pSql);
+    tscAsyncResultOnError(pSql);
     return code;  // here the pSql may have been released already.
   }
 
@@ -2481,7 +2482,7 @@ void tscBuildResFromSubqueries(SSqlObj *pSql) {
   SSqlRes* pRes = &pSql->res;
 
   if (pRes->code != TSDB_CODE_SUCCESS) {
-    tscQueueAsyncRes(pSql);
+    tscAsyncResultOnError(pSql);
     return;
   }
 
@@ -2496,7 +2497,7 @@ void tscBuildResFromSubqueries(SSqlObj *pSql) {
 
     if (pRes->tsrow == NULL || pRes->buffer == NULL || pRes->length == NULL) {
       pRes->code = TSDB_CODE_TSC_OUT_OF_MEMORY;
-      tscQueueAsyncRes(pSql);
+      tscAsyncResultOnError(pSql);
       return;
     }
 
@@ -2508,7 +2509,7 @@ void tscBuildResFromSubqueries(SSqlObj *pSql) {
   if (pRes->code == TSDB_CODE_SUCCESS) {
     (*pSql->fp)(pSql->param, pSql, pRes->numOfRows);
   } else {
-    tscQueueAsyncRes(pSql);
+    tscAsyncResultOnError(pSql);
   }
 }
 

src/client/src/tscSystem.c

@@ -31,15 +31,20 @@
 #include "tlocale.h"
 
 // global, not configurable
-SCacheObj *tscMetaCache;       // table meta cache
-SHashObj  *tscHashMap;         // hash map to keep the global vgroup info
-int     tscObjRef = -1;
+#define TSC_VAR_NOT_RELEASE 1
+#define TSC_VAR_RELEASED    0
+
+int32_t    sentinel = TSC_VAR_NOT_RELEASE;
+
+SHashObj  *tscVgroupMap;         // hash map to keep the global vgroup info
+SHashObj  *tscTableMetaInfo;     // table meta info
+int32_t    tscObjRef = -1;
 void      *tscTmr;
 void      *tscQhandle;
-void   *tscCheckDiskUsageTmr;
-int     tscRefId = -1;
-int     tscNumOfObj = 0;       // number of sqlObj in current process.
+int32_t    tscRefId = -1;
+int32_t    tscNumOfObj = 0;         // number of sqlObj in current process.
 
+static void  *tscCheckDiskUsageTmr;
 static pthread_once_t tscinit = PTHREAD_ONCE_INIT;
 
 void tscCheckDiskUsage(void *UNUSED_PARAM(para), void* UNUSED_PARAM(param)) {
@@ -129,11 +134,11 @@ void taos_init_imp(void) {
     taosTmrReset(tscCheckDiskUsage, 10, NULL, tscTmr, &tscCheckDiskUsageTmr);      
   }
 
-  int64_t refreshTime = 10; // 10 seconds by default
-  if (tscMetaCache == NULL) {
-    tscMetaCache = taosCacheInit(TSDB_DATA_TYPE_BINARY, refreshTime, false, tscFreeTableMetaHelper, "tableMeta");
+  if (tscTableMetaInfo == NULL) {
     tscObjRef  = taosOpenRef(40960, tscFreeRegisteredSqlObj);
-    tscHashMap = taosHashInit(1024, taosGetDefaultHashFunction(TSDB_DATA_TYPE_INT), true, HASH_ENTRY_LOCK);
+    tscVgroupMap = taosHashInit(256, taosGetDefaultHashFunction(TSDB_DATA_TYPE_INT), true, HASH_ENTRY_LOCK);
+    tscTableMetaInfo = taosHashInit(1024, taosGetDefaultHashFunction(TSDB_DATA_TYPE_BINARY), true, HASH_ENTRY_LOCK);
+    tscDebug("TableMeta:%p", tscTableMetaInfo);
   }
 
   tscRefId = taosOpenRef(200, tscCloseTscObj);
@@ -151,30 +156,38 @@ void taos_init() { pthread_once(&tscinit, taos_init_imp); }
 void taos_cleanup(void) {
   tscDebug("start to cleanup client environment");
 
-  void* m = tscMetaCache;
-  if (m != NULL && atomic_val_compare_exchange_ptr(&tscMetaCache, m, 0) == m) {
-    taosCacheCleanup(m);
+  if (atomic_val_compare_exchange_32(&sentinel, TSC_VAR_NOT_RELEASE, TSC_VAR_RELEASED) != TSC_VAR_NOT_RELEASE) {
+    return;
   }
 
-  int refId = atomic_exchange_32(&tscObjRef, -1);
-  if (refId != -1) {
-    taosCloseRef(refId);
-  }
+  taosHashCleanup(tscTableMetaInfo);
+  tscTableMetaInfo = NULL;
 
-  m = tscQhandle;
-  if (m != NULL && atomic_val_compare_exchange_ptr(&tscQhandle, m, 0) == m) {
-    taosCleanUpScheduler(m);
-  }
+  taosHashCleanup(tscVgroupMap);
+  tscVgroupMap = NULL;
+
+  int32_t id = tscObjRef;
+  tscObjRef = -1;
+  taosCloseRef(id);
+
+  void* p = tscQhandle;
+  tscQhandle = NULL;
+  taosCleanUpScheduler(p);
+
+  id = tscRefId;
+  tscRefId = -1;
+  taosCloseRef(id);
 
-  taosCloseRef(tscRefId);
   taosCleanupKeywordsTable();
   taosCloseLog();
-  if (tscEmbedded == 0) rpcCleanup();
 
-  m = tscTmr;
-  if (m != NULL && atomic_val_compare_exchange_ptr(&tscTmr, m, 0) == m) {
-    taosTmrCleanUp(m);
+  if (tscEmbedded == 0) {
+    rpcCleanup();
   }
+
+  p = tscTmr;
+  tscTmr = NULL;
+  taosTmrCleanUp(p);
 }
 
 static int taos_options_imp(TSDB_OPTION option, const char *pStr) {

src/client/src/tscUtil.c

@@ -18,7 +18,6 @@
 #include "os.h"
 #include "qAst.h"
 #include "taosmsg.h"
-#include "tcache.h"
 #include "tkey.h"
 #include "tmd5.h"
 #include "tscLocalMerge.h"
@@ -31,7 +30,7 @@
 #include "ttokendef.h"
 
 static void freeQueryInfoImpl(SQueryInfo* pQueryInfo);
-static void clearAllTableMetaInfo(SQueryInfo* pQueryInfo, const char* address, bool removeFromCache);
+static void clearAllTableMetaInfo(SQueryInfo* pQueryInfo);
 
 SCond* tsGetSTableQueryCond(STagCond* pTagCond, uint64_t uid) {
   if (pTagCond->pCond == NULL) {
@@ -379,17 +378,16 @@ static void tscDestroyResPointerInfo(SSqlRes* pRes) {
   pRes->data = NULL;  // pRes->data points to the buffer of pRsp, no need to free
 }
 
-void tscFreeQueryInfo(SSqlCmd* pCmd, bool removeFromCache) {
+void tscFreeQueryInfo(SSqlCmd* pCmd) {
   if (pCmd == NULL || pCmd->numOfClause == 0) {
     return;
   }
   
   for (int32_t i = 0; i < pCmd->numOfClause; ++i) {
-    char* addr = (char*)pCmd - offsetof(SSqlObj, cmd);
     SQueryInfo* pQueryInfo = tscGetQueryInfoDetail(pCmd, i);
     
     freeQueryInfoImpl(pQueryInfo);
-    clearAllTableMetaInfo(pQueryInfo, (const char*)addr, removeFromCache);
+    clearAllTableMetaInfo(pQueryInfo);
     tfree(pQueryInfo);
   }
   
@@ -397,7 +395,7 @@ void tscFreeQueryInfo(SSqlCmd* pCmd, bool removeFromCache) {
   tfree(pCmd->pQueryInfo);
 }
 
-void tscResetSqlCmdObj(SSqlCmd* pCmd, bool removeFromCache) {
+void tscResetSqlCmdObj(SSqlCmd* pCmd) {
   pCmd->command   = 0;
   pCmd->numOfCols = 0;
   pCmd->count     = 0;
@@ -407,17 +405,17 @@ void tscResetSqlCmdObj(SSqlCmd* pCmd, bool removeFromCache) {
   pCmd->autoCreated = 0;
 
   for(int32_t i = 0; i < pCmd->numOfTables; ++i) {
-    if (pCmd->pTableMetaList && pCmd->pTableMetaList[i]) {
-      taosCacheRelease(tscMetaCache, (void**)&(pCmd->pTableMetaList[i]), false);
+    if (pCmd->pTableNameList && pCmd->pTableNameList[i]) {
+      tfree(pCmd->pTableNameList[i]);
     }
   }
 
   pCmd->numOfTables = 0;
-  tfree(pCmd->pTableMetaList);
+  tfree(pCmd->pTableNameList);
 
   pCmd->pTableBlockHashList = tscDestroyBlockHashTable(pCmd->pTableBlockHashList);
   pCmd->pDataBlocks = tscDestroyBlockArrayList(pCmd->pDataBlocks);
-  tscFreeQueryInfo(pCmd, removeFromCache);
+  tscFreeQueryInfo(pCmd);
 }
 
 void tscFreeSqlResult(SSqlObj* pSql) {
@@ -468,17 +466,6 @@ void tscFreeRegisteredSqlObj(void *pSql) {
   tscDebug("%p free SqlObj, total in tscObj:%d, total:%d", pSql, num, total);
 }
 
-void tscFreeTableMetaHelper(void *pTableMeta) {
-  STableMeta* p = (STableMeta*) pTableMeta;
-
-  int32_t numOfEps1 = p->corVgroupInfo.numOfEps;
-  assert(numOfEps1 >= 0 && numOfEps1 <= TSDB_MAX_REPLICA);
-
-  for(int32_t i = 0; i < numOfEps1; ++i) {
-    tfree(p->corVgroupInfo.epAddr[i].fqdn);
-  }
-}
-
 void tscFreeSqlObj(SSqlObj* pSql) {
   if (pSql == NULL || pSql->signature != pSql) {
     return;
@@ -506,7 +493,7 @@ void tscFreeSqlObj(SSqlObj* pSql) {
   pSql->self = 0;
 
   tscFreeSqlResult(pSql);
-  tscResetSqlCmdObj(pCmd, false);
+  tscResetSqlCmdObj(pCmd);
 
   tfree(pCmd->tagData.data);
   pCmd->tagData.dataLen = 0;
@@ -529,7 +516,7 @@ void tscDestroyDataBlock(STableDataBlocks* pDataBlock) {
 
   // free the refcount for metermeta
   if (pDataBlock->pTableMeta != NULL) {
-    taosCacheRelease(tscMetaCache, (void**)&(pDataBlock->pTableMeta), false);
+    tfree(pDataBlock->pTableMeta);
   }
 
   tfree(pDataBlock);
@@ -600,15 +587,15 @@ int32_t tscCopyDataBlockToPayload(SSqlObj* pSql, STableDataBlocks* pDataBlock) {
 
   // set the correct table meta object, the table meta has been locked in pDataBlocks, so it must be in the cache
   if (pTableMetaInfo->pTableMeta != pDataBlock->pTableMeta) {
-    tstrncpy(pTableMetaInfo->name, pDataBlock->tableId, sizeof(pTableMetaInfo->name));
+    tstrncpy(pTableMetaInfo->name, pDataBlock->tableName, sizeof(pTableMetaInfo->name));
 
     if (pTableMetaInfo->pTableMeta != NULL) {
-      taosCacheRelease(tscMetaCache, (void**)&(pTableMetaInfo->pTableMeta), false);
+      tfree(pTableMetaInfo->pTableMeta);
     }
 
-    pTableMetaInfo->pTableMeta = taosCacheTransfer(tscMetaCache, (void**)&pDataBlock->pTableMeta);
+    pTableMetaInfo->pTableMeta = tscTableMetaClone(pDataBlock->pTableMeta);
   } else {
-    assert(strncmp(pTableMetaInfo->name, pDataBlock->tableId, tListLen(pDataBlock->tableId)) == 0);
+    assert(strncmp(pTableMetaInfo->name, pDataBlock->tableName, tListLen(pDataBlock->tableName)) == 0);
   }
 
   /*
@@ -671,14 +658,10 @@ int32_t tscCreateDataBlock(size_t initialSize, int32_t rowSize, int32_t startOff
   dataBuf->size = startOffset;
   dataBuf->tsSource = -1;
 
-  tstrncpy(dataBuf->tableId, name, sizeof(dataBuf->tableId));
+  tstrncpy(dataBuf->tableName, name, sizeof(dataBuf->tableName));
 
-  /*
-   * The table meta may be released since the table meta cache are completed clean by other thread
-   * due to operation such as drop database. So here we add the reference count directly instead of invoke
-   * taosGetDataFromCache, which may return NULL value.
-   */
-  dataBuf->pTableMeta = taosCacheAcquireByData(tscMetaCache, pTableMeta);
+  //Here we keep the tableMeta to avoid it to be remove by other threads.
+  dataBuf->pTableMeta = tscTableMetaClone(pTableMeta);
   assert(initialSize > 0 && pTableMeta != NULL && dataBuf->pTableMeta != NULL);
 
   *dataBlocks = dataBuf;
@@ -784,15 +767,15 @@ static int32_t getRowExpandSize(STableMeta* pTableMeta) {
   return result;
 }
 
-static void extractTableMeta(SSqlCmd* pCmd) {
+static void extractTableNameList(SSqlCmd* pCmd) {
   pCmd->numOfTables = (int32_t) taosHashGetSize(pCmd->pTableBlockHashList);
-  pCmd->pTableMetaList = calloc(pCmd->numOfTables, POINTER_BYTES);
+  pCmd->pTableNameList = calloc(pCmd->numOfTables, POINTER_BYTES);
 
   STableDataBlocks **p1 = taosHashIterate(pCmd->pTableBlockHashList, NULL);
   int32_t i = 0;
   while(p1) {
     STableDataBlocks* pBlocks = *p1;
-    pCmd->pTableMetaList[i++] = taosCacheTransfer(tscMetaCache, (void**) &pBlocks->pTableMeta);
+    pCmd->pTableNameList[i++] = strndup(pBlocks->tableName, TSDB_TABLE_FNAME_LEN);
     p1 = taosHashIterate(pCmd->pTableBlockHashList, p1);
   }
 
@@ -815,7 +798,7 @@ int32_t tscMergeTableDataBlocks(SSqlObj* pSql) {
     STableDataBlocks* dataBuf = NULL;
     
     int32_t ret = tscGetDataBlockFromList(pVnodeDataBlockHashList, pOneTableBlock->vgId, TSDB_PAYLOAD_SIZE,
-                                INSERT_HEAD_SIZE, 0, pOneTableBlock->tableId, pOneTableBlock->pTableMeta, &dataBuf, pVnodeDataBlockList);
+                                INSERT_HEAD_SIZE, 0, pOneTableBlock->tableName, pOneTableBlock->pTableMeta, &dataBuf, pVnodeDataBlockList);
     if (ret != TSDB_CODE_SUCCESS) {
       tscError("%p failed to prepare the data block buffer for merging table data, code:%d", pSql, ret);
       taosHashCleanup(pVnodeDataBlockHashList);
@@ -849,7 +832,7 @@ int32_t tscMergeTableDataBlocks(SSqlObj* pSql) {
     tscSortRemoveDataBlockDupRows(pOneTableBlock);
     char* ekey = (char*)pBlocks->data + pOneTableBlock->rowSize*(pBlocks->numOfRows-1);
     
-    tscDebug("%p tableId:%s, sid:%d rows:%d sversion:%d skey:%" PRId64 ", ekey:%" PRId64, pSql, pOneTableBlock->tableId,
+    tscDebug("%p name:%s, sid:%d rows:%d sversion:%d skey:%" PRId64 ", ekey:%" PRId64, pSql, pOneTableBlock->tableName,
         pBlocks->tid, pBlocks->numOfRows, pBlocks->sversion, GET_INT64_VAL(pBlocks->data), GET_INT64_VAL(ekey));
 
     int32_t len = pBlocks->numOfRows * (pOneTableBlock->rowSize + expandSize) + sizeof(STColumn) * tscGetNumOfColumns(pOneTableBlock->pTableMeta);
@@ -879,7 +862,7 @@ int32_t tscMergeTableDataBlocks(SSqlObj* pSql) {
     pOneTableBlock = *p;
   }
 
-  extractTableMeta(pCmd);
+  extractTableNameList(pCmd);
 
   // free the table data blocks;
   pCmd->pDataBlocks = pVnodeDataBlockList;
@@ -900,6 +883,7 @@ void tscCloseTscObj(void *param) {
     rpcClose(pObj->pDnodeConn);
     pObj->pDnodeConn = NULL;
   }
+
   tfree(pObj->tscCorMgmtEpSet);
   pthread_mutex_destroy(&pObj->mutex);
 
@@ -1528,6 +1512,7 @@ int32_t tscTagCondCopy(STagCond* dest, const STagCond* src) {
   }
 
   dest->tbnameCond.uid = src->tbnameCond.uid;
+  dest->tbnameCond.len = src->tbnameCond.len;
 
   memcpy(&dest->joinInfo, &src->joinInfo, sizeof(SJoinInfo));
   dest->relType = src->relType;
@@ -1823,14 +1808,12 @@ SArray* tscVgroupTableInfoClone(SArray* pVgroupTables) {
   return pa;
 }
 
-void clearAllTableMetaInfo(SQueryInfo* pQueryInfo, const char* address, bool removeFromCache) {
-  tscDebug("%p unref %d tables in the tableMeta cache", address, pQueryInfo->numOfTables);
-  
+void clearAllTableMetaInfo(SQueryInfo* pQueryInfo) {
   for(int32_t i = 0; i < pQueryInfo->numOfTables; ++i) {
     STableMetaInfo* pTableMetaInfo = tscGetMetaInfo(pQueryInfo, i);
 
     tscFreeVgroupTableInfo(pTableMetaInfo->pVgroupTables);
-    tscClearTableMetaInfo(pTableMetaInfo, removeFromCache);
+    tscClearTableMetaInfo(pTableMetaInfo);
     free(pTableMetaInfo);
   }
   
@@ -1884,14 +1867,12 @@ STableMetaInfo* tscAddEmptyMetaInfo(SQueryInfo* pQueryInfo) {
   return tscAddTableMetaInfo(pQueryInfo, NULL, NULL, NULL, NULL, NULL);
 }
 
-void tscClearTableMetaInfo(STableMetaInfo* pTableMetaInfo, bool removeFromCache) {
+void tscClearTableMetaInfo(STableMetaInfo* pTableMetaInfo) {
   if (pTableMetaInfo == NULL) {
     return;
   }
 
-  if (pTableMetaInfo->pTableMeta != NULL) {
-    taosCacheRelease(tscMetaCache, (void**)&(pTableMetaInfo->pTableMeta), removeFromCache);
-  }
+  tfree(pTableMetaInfo->pTableMeta);
 
   pTableMetaInfo->vgroupList = tscVgroupInfoClear(pTableMetaInfo->vgroupList);
   tscColumnListDestroy(pTableMetaInfo->tagColList);
@@ -2015,7 +1996,7 @@ SSqlObj* createSubqueryObj(SSqlObj* pSql, int16_t tableIndex, void (*fp)(), void
 
   pNew->pTscObj   = pSql->pTscObj;
   pNew->signature = pNew;
-  pNew->sqlstr    = NULL;
+  pNew->sqlstr    = strdup(pSql->sqlstr);
 
   SSqlCmd* pnCmd = &pNew->cmd;
   memcpy(pnCmd, pCmd, sizeof(SSqlCmd));
@@ -2031,7 +2012,7 @@ SSqlObj* createSubqueryObj(SSqlObj* pSql, int16_t tableIndex, void (*fp)(), void
 
   pnCmd->numOfTables = 0;
   pnCmd->parseFinished = 1;
-  pnCmd->pTableMetaList = NULL;
+  pnCmd->pTableNameList = NULL;
   pnCmd->pTableBlockHashList = NULL;
 
   if (tscAddSubqueryInfo(pnCmd) != TSDB_CODE_SUCCESS) {
@@ -2113,8 +2094,8 @@ SSqlObj* createSubqueryObj(SSqlObj* pSql, int16_t tableIndex, void (*fp)(), void
   char* name = pTableMetaInfo->name;
   STableMetaInfo* pFinalInfo = NULL;
 
-  if (pPrevSql == NULL) { // get by name may failed due to the cache cleanup
-    STableMeta* pTableMeta = taosCacheAcquireByData(tscMetaCache, pTableMetaInfo->pTableMeta);
+  if (pPrevSql == NULL) {
+    STableMeta* pTableMeta = tscTableMetaClone(pTableMetaInfo->pTableMeta);
     assert(pTableMeta != NULL);
 
     pFinalInfo = tscAddTableMetaInfo(pNewQueryInfo, name, pTableMeta, pTableMetaInfo->vgroupList,
@@ -2122,15 +2103,15 @@ SSqlObj* createSubqueryObj(SSqlObj* pSql, int16_t tableIndex, void (*fp)(), void
   } else {  // transfer the ownership of pTableMeta to the newly create sql object.
     STableMetaInfo* pPrevInfo = tscGetTableMetaInfoFromCmd(&pPrevSql->cmd, pPrevSql->cmd.clauseIndex, 0);
 
-    STableMeta*  pPrevTableMeta = taosCacheTransfer(tscMetaCache, (void**)&pPrevInfo->pTableMeta);
-    
+    STableMeta*  pPrevTableMeta = tscTableMetaClone(pPrevInfo->pTableMeta);
     SVgroupsInfo* pVgroupsInfo = pPrevInfo->vgroupList;
     pFinalInfo = tscAddTableMetaInfo(pNewQueryInfo, name, pPrevTableMeta, pVgroupsInfo, pTableMetaInfo->tagColList,
         pTableMetaInfo->pVgroupTables);
   }
 
+  // this case cannot be happened
   if (pFinalInfo->pTableMeta == NULL) {
-    tscError("%p new subquery failed since no tableMeta in cache, name:%s", pSql, name);
+    tscError("%p new subquery failed since no tableMeta, name:%s", pSql, name);
 
     if (pPrevSql != NULL) { // pass the previous error to client
       assert(pPrevSql->res.code != TSDB_CODE_SUCCESS);
@@ -2557,6 +2538,7 @@ void* tscVgroupInfoClear(SVgroupsInfo *vgroupList) {
     for(int32_t j = 0; j < pVgroupInfo->numOfEps; ++j) {
       tfree(pVgroupInfo->epAddr[j].fqdn);
     }
+
     for(int32_t j = pVgroupInfo->numOfEps; j < TSDB_MAX_REPLICA; j++) {
       assert( pVgroupInfo->epAddr[j].fqdn == NULL );
     }
@@ -2610,3 +2592,87 @@ int32_t copyTagData(STagData* dst, const STagData* src) {
 
   return 0;
 }
+
+STableMeta* createSuperTableMeta(STableMetaMsg* pChild) {
+  assert(pChild != NULL);
+  int32_t total = pChild->numOfColumns + pChild->numOfTags;
+
+  STableMeta* pTableMeta = calloc(1, sizeof(STableMeta) + sizeof(SSchema) * total);
+  pTableMeta->tableType = TSDB_SUPER_TABLE;
+  pTableMeta->tableInfo.numOfTags = pChild->numOfTags;
+  pTableMeta->tableInfo.numOfColumns = pChild->numOfColumns;
+  pTableMeta->tableInfo.precision = pChild->precision;
+
+  pTableMeta->id.tid = 0;
+  pTableMeta->id.uid = pChild->suid;
+  pTableMeta->tversion = pChild->tversion;
+  pTableMeta->sversion = pChild->sversion;
+
+  memcpy(pTableMeta->schema, pChild->schema, sizeof(SSchema) * total);
+
+  int32_t num = pTableMeta->tableInfo.numOfColumns;
+  for(int32_t i = 0; i < num; ++i) {
+    pTableMeta->tableInfo.rowSize += pTableMeta->schema[i].bytes;
+  }
+
+  return pTableMeta;
+}
+
+uint32_t tscGetTableMetaSize(STableMeta* pTableMeta) {
+  assert(pTableMeta != NULL);
+
+  int32_t totalCols = pTableMeta->tableInfo.numOfColumns + pTableMeta->tableInfo.numOfTags;
+  return sizeof(STableMeta) + totalCols * sizeof(SSchema);
+}
+
+CChildTableMeta* tscCreateChildMeta(STableMeta* pTableMeta) {
+  assert(pTableMeta != NULL);
+
+  CChildTableMeta* cMeta = calloc(1, sizeof(CChildTableMeta));
+  cMeta->tableType = TSDB_CHILD_TABLE;
+  cMeta->vgId = pTableMeta->vgId;
+  cMeta->id   = pTableMeta->id;
+  tstrncpy(cMeta->sTableName, pTableMeta->sTableName, TSDB_TABLE_FNAME_LEN);
+
+  return cMeta;
+}
+
+int32_t tscCreateTableMetaFromCChildMeta(STableMeta* pChild, const char* name) {
+  assert(pChild != NULL);
+
+  uint32_t size = tscGetTableMetaMaxSize();
+  STableMeta* p = calloc(1, size);
+
+  taosHashGetClone(tscTableMetaInfo, pChild->sTableName, strnlen(pChild->sTableName, TSDB_TABLE_FNAME_LEN), NULL, p, -1);
+  if (p->id.uid > 0) { // tableMeta exists, build child table meta and return
+    pChild->sversion = p->sversion;
+    pChild->tversion = p->tversion;
+
+    memcpy(&pChild->tableInfo, &p->tableInfo, sizeof(STableInfo));
+    int32_t total = pChild->tableInfo.numOfColumns + pChild->tableInfo.numOfTags;
+
+    memcpy(pChild->schema, p->schema, sizeof(SSchema) *total);
+
+    tfree(p);
+    return TSDB_CODE_SUCCESS;
+  } else { // super table has been removed, current tableMeta is also expired. remove it here
+    taosHashRemove(tscTableMetaInfo, name, strnlen(name, TSDB_TABLE_FNAME_LEN));
+
+    tfree(p);
+    return -1;
+  }
+}
+
+uint32_t tscGetTableMetaMaxSize() {
+  return sizeof(STableMeta) + TSDB_MAX_COLUMNS * sizeof(SSchema);
+}
+
+STableMeta* tscTableMetaClone(STableMeta* pTableMeta) {
+  assert(pTableMeta != NULL);
+  uint32_t size = tscGetTableMetaSize(pTableMeta);
+  STableMeta* p = calloc(1, size);
+  memcpy(p, pTableMeta, size);
+  return p;
+}
+
+

src/dnode/src/dnodeMWrite.c

@@ -184,7 +184,19 @@ void dnodeReprocessMWriteMsg(void *pMsg) {
     dDebug("msg:%p, app:%p type:%s is redirected for mnode not running, retry times:%d", pWrite, pWrite->rpcMsg.ahandle,
            taosMsg[pWrite->rpcMsg.msgType], pWrite->retry);
 
-    dnodeSendRedirectMsg(pMsg, true);
+    if (pWrite->pBatchMasterMsg) {
+      ++pWrite->pBatchMasterMsg->received;
+      if (pWrite->pBatchMasterMsg->successed + pWrite->pBatchMasterMsg->received
+	  >= pWrite->pBatchMasterMsg->expected) {
+	dnodeSendRedirectMsg(&pWrite->rpcMsg, true);
+	dnodeFreeMWriteMsg(pWrite);
+      }
+
+      mnodeDestroySubMsg(pWrite);
+
+      return;
+    }
+    dnodeSendRedirectMsg(&pWrite->rpcMsg, true);
     dnodeFreeMWriteMsg(pWrite);
   } else {
     dDebug("msg:%p, app:%p type:%s is reput into mwrite queue:%p, retry times:%d", pWrite, pWrite->rpcMsg.ahandle,

src/dnode/src/dnodeVnodes.c

@@ -73,7 +73,8 @@ static int32_t dnodeGetVnodeList(int32_t vnodeList[], int32_t *numOfVnodes) {
 
       if (*numOfVnodes >= TSDB_MAX_VNODES) {
         dError("vgId:%d, too many vnode directory in disk, exist:%d max:%d", vnode, *numOfVnodes, TSDB_MAX_VNODES);
-        continue;
+        closedir(dir);
+        return TSDB_CODE_DND_TOO_MANY_VNODES;
       } else {
         vnodeList[*numOfVnodes - 1] = vnode;
       }

src/inc/mnode.h

@@ -42,11 +42,12 @@ typedef struct SMnodeMsg {
   struct SVgObj *   pVgroup;
   struct STableObj *pTable;
   struct SSTableObj*pSTable;
+  struct SMnodeMsg *pBatchMasterMsg;
   SMnodeRsp rpcRsp;
-  int8_t    received;
-  int8_t    successed;
-  int8_t    expected;
-  int8_t    retry;
+  int16_t   received;
+  int16_t   successed;
+  int16_t   expected;
+  int16_t   retry;
   int32_t   incomingTs;
   int32_t   code;
   void *    pObj;
@@ -57,6 +58,7 @@ typedef struct SMnodeMsg {
 void *  mnodeCreateMsg(SRpcMsg *pRpcMsg);
 int32_t mnodeInitMsg(SMnodeMsg *pMsg);
 void    mnodeCleanupMsg(SMnodeMsg *pMsg);
+void    mnodeDestroySubMsg(SMnodeMsg *pSubMsg);
 
 int32_t mnodeInitSystem();
 int32_t mnodeStartSystem();

src/inc/taoserror.h

@@ -194,6 +194,7 @@ TAOS_DEFINE_ERROR(TSDB_CODE_DND_OUT_OF_MEMORY,            0, 0x0401, "Dnode out
 TAOS_DEFINE_ERROR(TSDB_CODE_DND_NO_WRITE_ACCESS,          0, 0x0402, "No permission for disk files in dnode")
 TAOS_DEFINE_ERROR(TSDB_CODE_DND_INVALID_MSG_LEN,          0, 0x0403, "Invalid message length")
 TAOS_DEFINE_ERROR(TSDB_CODE_DND_ACTION_IN_PROGRESS,       0, 0x0404, "Action in progress")
+TAOS_DEFINE_ERROR(TSDB_CODE_DND_TOO_MANY_VNODES,          0, 0x0405, "Too many vnode directories")
 
 // vnode 
 TAOS_DEFINE_ERROR(TSDB_CODE_VND_ACTION_IN_PROGRESS,       0, 0x0500, "Action in progress")

src/inc/taosmsg.h

@@ -476,6 +476,7 @@ typedef struct {
   int16_t     numOfCols;        // the number of columns will be load from vnode
   SInterval   interval;
   uint16_t    tagCondLen;       // tag length in current query
+  uint32_t    tbnameCondLen;    // table name filter condition string length
   int16_t     numOfGroupCols;   // num of group by columns
   int16_t     orderByIdx;
   int16_t     orderType;        // used in group by xx order by xxx
@@ -494,6 +495,7 @@ typedef struct {
   int32_t     tsNumOfBlocks;    // ts comp block numbers
   int32_t     tsOrder;          // ts comp block order
   int32_t     numOfTags;        // number of tags columns involved
+  int32_t     sqlstrLen;        // sql query string
   SColumnInfo colList[];
 } SQueryTableMsg;
 
@@ -725,7 +727,6 @@ typedef struct {
 typedef struct STableMetaMsg {
   int32_t       contLen;
   char          tableId[TSDB_TABLE_FNAME_LEN];   // table id
-  char          sTableId[TSDB_TABLE_FNAME_LEN];
   uint8_t       numOfTags;
   uint8_t       precision;
   uint8_t       tableType;
@@ -735,6 +736,9 @@ typedef struct STableMetaMsg {
   int32_t       tid;
   uint64_t      uid;
   SVgroupMsg    vgroup;
+
+  char          sTableName[TSDB_TABLE_FNAME_LEN];
+  uint64_t      suid;
   SSchema       schema[];
 } STableMetaMsg;
 

src/mnode/src/mnodeTable.c

@@ -26,6 +26,7 @@
 #include "tcompare.h"
 #include "tdataformat.h"
 #include "tgrant.h"
+#include "tqueue.h"
 #include "hash.h"
 #include "mnode.h"
 #include "dnode.h"
@@ -720,6 +721,133 @@ static void mnodeExtractTableName(char* tableId, char* name) {
   }
 }
 
+static SMnodeMsg *mnodeCreateSubMsg(SMnodeMsg *pBatchMasterMsg, int32_t contSize) {
+  SMnodeMsg *pSubMsg = taosAllocateQitem(sizeof(*pBatchMasterMsg) + contSize);
+  *pSubMsg = *pBatchMasterMsg;
+
+  //pSubMsg->pCont = (char *) pSubMsg + sizeof(SMnodeMsg);
+  pSubMsg->rpcMsg.pCont = pSubMsg->pCont;
+  pSubMsg->successed = 0;
+  pSubMsg->expected = 0;
+  SCMCreateTableMsg *pCM = pSubMsg->rpcMsg.pCont;
+  pCM->numOfTables = htonl(1);
+  pCM->contLen = htonl(contSize);
+
+  return pSubMsg;
+}
+
+void mnodeDestroySubMsg(SMnodeMsg *pSubMsg) {
+  if (pSubMsg) {
+    // pUser is retained in batch master msg
+    if (pSubMsg->pDb) mnodeDecDbRef(pSubMsg->pDb);
+    if (pSubMsg->pVgroup) mnodeDecVgroupRef(pSubMsg->pVgroup);
+    if (pSubMsg->pTable) mnodeDecTableRef(pSubMsg->pTable);
+    if (pSubMsg->pSTable) mnodeDecTableRef(pSubMsg->pSTable);
+    if (pSubMsg->pAcct) mnodeDecAcctRef(pSubMsg->pAcct);
+    if (pSubMsg->pDnode) mnodeDecDnodeRef(pSubMsg->pDnode);
+
+    taosFreeQitem(pSubMsg);
+  }
+}
+
+static int32_t mnodeValidateCreateTableMsg(SCreateTableMsg *pCreateTable, SMnodeMsg *pMsg) {
+  if (pMsg->pDb == NULL) pMsg->pDb = mnodeGetDb(pCreateTable->db);
+  if (pMsg->pDb == NULL) {
+    mError("msg:%p, app:%p table:%s, failed to create, db not selected", pMsg, pMsg->rpcMsg.ahandle, pCreateTable->tableId);
+    return TSDB_CODE_MND_DB_NOT_SELECTED;
+  }
+
+  if (pMsg->pDb->status != TSDB_DB_STATUS_READY) {
+    mError("db:%s, status:%d, in dropping", pMsg->pDb->name, pMsg->pDb->status);
+    return TSDB_CODE_MND_DB_IN_DROPPING;
+  }
+
+  if (pMsg->pTable == NULL) pMsg->pTable = mnodeGetTable(pCreateTable->tableId);
+  if (pMsg->pTable != NULL && pMsg->retry == 0) {
+    if (pCreateTable->getMeta) {
+      mDebug("msg:%p, app:%p table:%s, continue to get meta", pMsg, pMsg->rpcMsg.ahandle, pCreateTable->tableId);
+      return mnodeGetChildTableMeta(pMsg);
+    } else if (pCreateTable->igExists) {
+      mDebug("msg:%p, app:%p table:%s, is already exist", pMsg, pMsg->rpcMsg.ahandle, pCreateTable->tableId);
+      return TSDB_CODE_SUCCESS;
+    } else {
+      mError("msg:%p, app:%p table:%s, failed to create, table already exist", pMsg, pMsg->rpcMsg.ahandle,
+             pCreateTable->tableId);
+      return TSDB_CODE_MND_TABLE_ALREADY_EXIST;
+    }
+  }
+
+  if (pCreateTable->numOfTags != 0) {
+    mDebug("msg:%p, app:%p table:%s, create stable msg is received from thandle:%p", pMsg, pMsg->rpcMsg.ahandle,
+           pCreateTable->tableId, pMsg->rpcMsg.handle);
+    return mnodeProcessCreateSuperTableMsg(pMsg);
+  } else {
+    mDebug("msg:%p, app:%p table:%s, create ctable msg is received from thandle:%p", pMsg, pMsg->rpcMsg.ahandle,
+           pCreateTable->tableId, pMsg->rpcMsg.handle);
+    return mnodeProcessCreateChildTableMsg(pMsg);
+  }
+}
+
+static int32_t mnodeProcessBatchCreateTableMsg(SMnodeMsg *pMsg) {
+  if (pMsg->pBatchMasterMsg == NULL) { // batch master first round
+    pMsg->pBatchMasterMsg = pMsg;
+
+    SCMCreateTableMsg *pCreate = pMsg->rpcMsg.pCont;
+    int32_t numOfTables = htonl(pCreate->numOfTables);
+    int32_t contentLen = htonl(pCreate->contLen);
+    pMsg->expected = numOfTables;
+
+    int32_t code = TSDB_CODE_SUCCESS;
+    SCreateTableMsg *pCreateTable = (SCreateTableMsg*) ((char*) pCreate + sizeof(SCMCreateTableMsg));
+    for (SCreateTableMsg *p = pCreateTable; p < (SCreateTableMsg *) ((char *) pCreate + contentLen); p = (SCreateTableMsg *) ((char *) p + htonl(p->len))) {
+      SMnodeMsg *pSubMsg = mnodeCreateSubMsg(pMsg, sizeof(SCMCreateTableMsg) + htonl(p->len));
+      memcpy(pSubMsg->pCont + sizeof(SCMCreateTableMsg), p, htonl(p->len));
+      code = mnodeValidateCreateTableMsg(p, pSubMsg);
+
+      if (code == TSDB_CODE_SUCCESS || code == TSDB_CODE_MND_TABLE_ALREADY_EXIST) {
+	++pSubMsg->pBatchMasterMsg->successed;
+	mnodeDestroySubMsg(pSubMsg);
+	continue;
+      }
+
+      if (code != TSDB_CODE_MND_ACTION_IN_PROGRESS) {
+	mnodeDestroySubMsg(pSubMsg);
+	return code;
+      }
+    }
+
+    if (pMsg->successed >= pMsg->expected) {
+      return code;
+    } else {
+      return TSDB_CODE_MND_ACTION_IN_PROGRESS;
+    }
+  } else {
+    if (pMsg->pBatchMasterMsg != pMsg) { // batch sub replay
+      SCMCreateTableMsg *pCreate = pMsg->rpcMsg.pCont;
+      SCreateTableMsg *pCreateTable = (SCreateTableMsg*) ((char*) pCreate + sizeof(SCMCreateTableMsg));
+      int32_t code = mnodeValidateCreateTableMsg(pCreateTable, pMsg);
+      if (code == TSDB_CODE_SUCCESS || code == TSDB_CODE_MND_TABLE_ALREADY_EXIST) {
+	++pMsg->pBatchMasterMsg->successed;
+	mnodeDestroySubMsg(pMsg);
+      }
+
+      if (code != TSDB_CODE_MND_ACTION_IN_PROGRESS) {
+	mnodeDestroySubMsg(pMsg);
+	return code;
+      }
+
+      if (pMsg->pBatchMasterMsg->successed + pMsg->pBatchMasterMsg->received
+	  >= pMsg->pBatchMasterMsg->expected) {
+	return code;
+      } else {
+	return TSDB_CODE_MND_ACTION_IN_PROGRESS;
+      }
+    } else { // batch master replay, reprocess the whole batch
+      assert(0);
+    }
+  }
+}
+
 static int32_t mnodeProcessCreateTableMsg(SMnodeMsg *pMsg) {
   SCMCreateTableMsg *pCreate = pMsg->rpcMsg.pCont;
 
@@ -729,6 +857,11 @@ static int32_t mnodeProcessCreateTableMsg(SMnodeMsg *pMsg) {
     // todo return error
   }
 
+  // batch master msg first round or reprocessing and batch sub msg reprocessing
+  if (numOfTables > 1 || pMsg->pBatchMasterMsg != NULL) {
+    return mnodeProcessBatchCreateTableMsg(pMsg);
+  }
+
   SCreateTableMsg *p = (SCreateTableMsg*)((char*) pCreate + sizeof(SCMCreateTableMsg));
   if (pMsg->pDb == NULL) pMsg->pDb = mnodeGetDb(p->db);
   if (pMsg->pDb == NULL) {
@@ -1737,6 +1870,18 @@ static int32_t mnodeDoCreateChildTableCb(SMnodeMsg *pMsg, int32_t code) {
       mDebug("msg:%p, app:%p table:%s, created in dnode, thandle:%p", pMsg, pMsg->rpcMsg.ahandle, pTable->info.tableId,
              pMsg->rpcMsg.handle);
 
+      if (pMsg->pBatchMasterMsg) {
+	++pMsg->pBatchMasterMsg->successed;
+	if (pMsg->pBatchMasterMsg->successed + pMsg->pBatchMasterMsg->received
+	    >= pMsg->pBatchMasterMsg->expected) {
+	  dnodeSendRpcMWriteRsp(pMsg->pBatchMasterMsg, code);
+	}
+
+	mnodeDestroySubMsg(pMsg);
+
+	return TSDB_CODE_MND_ACTION_IN_PROGRESS;
+      }
+
       dnodeSendRpcMWriteRsp(pMsg, TSDB_CODE_SUCCESS);
     }
     return TSDB_CODE_MND_ACTION_IN_PROGRESS;
@@ -2171,11 +2316,12 @@ static int32_t mnodeDoGetChildTableMeta(SMnodeMsg *pMsg, STableMetaMsg *pMeta) {
   pMeta->precision = pDb->cfg.precision;
   pMeta->tableType = pTable->info.type;
   tstrncpy(pMeta->tableId, pTable->info.tableId, TSDB_TABLE_FNAME_LEN);
-  if (pTable->superTable != NULL) {
-    tstrncpy(pMeta->sTableId, pTable->superTable->info.tableId, TSDB_TABLE_FNAME_LEN);
-  }
 
-  if (pTable->info.type == TSDB_CHILD_TABLE && pTable->superTable != NULL) {
+  if (pTable->info.type == TSDB_CHILD_TABLE) {
+    assert(pTable->superTable != NULL);
+    tstrncpy(pMeta->sTableName, pTable->superTable->info.tableId, TSDB_TABLE_FNAME_LEN);
+
+    pMeta->suid         = pTable->superTable->uid;
     pMeta->sversion     = htons(pTable->superTable->sversion);
     pMeta->tversion     = htons(pTable->superTable->tversion);
     pMeta->numOfTags    = (int8_t)pTable->superTable->numOfTags;
@@ -2477,6 +2623,19 @@ static void mnodeProcessCreateChildTableRsp(SRpcMsg *rpcMsg) {
 
     mnodeSendDropChildTableMsg(pMsg, false);
     rpcMsg->code = TSDB_CODE_SUCCESS;
+
+    if (pMsg->pBatchMasterMsg) {
+      ++pMsg->pBatchMasterMsg->successed;
+      if (pMsg->pBatchMasterMsg->successed + pMsg->pBatchMasterMsg->received
+	  >= pMsg->pBatchMasterMsg->expected) {
+	dnodeSendRpcMWriteRsp(pMsg->pBatchMasterMsg, rpcMsg->code);
+      }
+
+      mnodeDestroySubMsg(pMsg);
+
+      return;
+    }
+
     dnodeSendRpcMWriteRsp(pMsg, rpcMsg->code);
     return;
   }
@@ -2494,6 +2653,19 @@ static void mnodeProcessCreateChildTableRsp(SRpcMsg *rpcMsg) {
     if (code != TSDB_CODE_SUCCESS && code != TSDB_CODE_MND_ACTION_IN_PROGRESS) {
       pMsg->pTable = NULL;
       mnodeDestroyChildTable(pTable);
+
+      if (pMsg->pBatchMasterMsg) {
+	++pMsg->pBatchMasterMsg->received;
+	if (pMsg->pBatchMasterMsg->successed + pMsg->pBatchMasterMsg->received
+	    >= pMsg->pBatchMasterMsg->expected) {
+	  dnodeSendRpcMWriteRsp(pMsg->pBatchMasterMsg, code);
+	}
+
+	mnodeDestroySubMsg(pMsg);
+
+	return;
+      }
+
       dnodeSendRpcMWriteRsp(pMsg, code);
     }
   } else {
@@ -2519,6 +2691,19 @@ static void mnodeProcessCreateChildTableRsp(SRpcMsg *rpcMsg) {
         //Avoid retry again in client
         rpcMsg->code = TSDB_CODE_MND_VGROUP_NOT_READY;
       }
+
+      if (pMsg->pBatchMasterMsg) {
+	++pMsg->pBatchMasterMsg->received;
+	if (pMsg->pBatchMasterMsg->successed + pMsg->pBatchMasterMsg->received
+	    >= pMsg->pBatchMasterMsg->expected) {
+	  dnodeSendRpcMWriteRsp(pMsg->pBatchMasterMsg, rpcMsg->code);
+	}
+
+	mnodeDestroySubMsg(pMsg);
+
+	return;
+      }
+
       dnodeSendRpcMWriteRsp(pMsg, rpcMsg->code);
     }
   }

src/plugins/http/src/httpResp.c

@@ -50,9 +50,16 @@ static void httpSendErrorRespImp(HttpContext *pContext, int32_t httpCode, char *
   char head[512] = {0};
   char body[512] = {0};
 
+  int8_t httpVersion = 0;
+  int8_t keepAlive = 0;
+  if (pContext->parser != NULL) {
+    httpVersion = pContext->parser->httpVersion;
+    keepAlive = pContext->parser->keepAlive;
+  }
+
   int32_t bodyLen = sprintf(body, httpRespTemplate[HTTP_RESPONSE_JSON_ERROR], errNo, desc);
-  int32_t headLen = sprintf(head, httpRespTemplate[HTTP_RESPONSE_ERROR], httpVersionStr[pContext->parser->httpVersion],
-                            httpCode, httpCodeStr, httpKeepAliveStr[pContext->parser->keepAlive], bodyLen);
+  int32_t headLen = sprintf(head, httpRespTemplate[HTTP_RESPONSE_ERROR], httpVersionStr[httpVersion], httpCode,
+                            httpCodeStr, httpKeepAliveStr[keepAlive], bodyLen);
 
   httpWriteBuf(pContext, head, headLen);
   httpWriteBuf(pContext, body, bodyLen);
@@ -164,9 +171,16 @@ void httpSendSuccResp(HttpContext *pContext, char *desc) {
   char head[1024] = {0};
   char body[1024] = {0};
 
+  int8_t httpVersion = 0;
+  int8_t keepAlive = 0;
+  if (pContext->parser != NULL) {
+    httpVersion = pContext->parser->httpVersion;
+    keepAlive = pContext->parser->keepAlive;
+  }
+
   int32_t bodyLen = sprintf(body, httpRespTemplate[HTTP_RESPONSE_JSON_OK], TSDB_CODE_SUCCESS, desc);
-  int32_t headLen = sprintf(head, httpRespTemplate[HTTP_RESPONSE_OK], httpVersionStr[pContext->parser->httpVersion],
-                            httpKeepAliveStr[pContext->parser->keepAlive], bodyLen);
+  int32_t headLen = sprintf(head, httpRespTemplate[HTTP_RESPONSE_OK], httpVersionStr[httpVersion],
+                            httpKeepAliveStr[keepAlive], bodyLen);
 
   httpWriteBuf(pContext, head, headLen);
   httpWriteBuf(pContext, body, bodyLen);
@@ -177,9 +191,16 @@ void httpSendOptionResp(HttpContext *pContext, char *desc) {
   char head[1024] = {0};
   char body[1024] = {0};
 
+  int8_t httpVersion = 0;
+  int8_t keepAlive = 0;
+  if (pContext->parser != NULL) {
+    httpVersion = pContext->parser->httpVersion;
+    keepAlive = pContext->parser->keepAlive;
+  }
+
   int32_t bodyLen = sprintf(body, httpRespTemplate[HTTP_RESPONSE_JSON_OK], TSDB_CODE_SUCCESS, desc);
-  int32_t headLen = sprintf(head, httpRespTemplate[HTTP_RESPONSE_OPTIONS], httpVersionStr[pContext->parser->httpVersion],
-                            httpKeepAliveStr[pContext->parser->keepAlive], bodyLen);
+  int32_t headLen = sprintf(head, httpRespTemplate[HTTP_RESPONSE_OPTIONS], httpVersionStr[httpVersion],
+                            httpKeepAliveStr[keepAlive], bodyLen);
 
   httpWriteBuf(pContext, head, headLen);
   httpWriteBuf(pContext, body, bodyLen);

src/query/inc/qSyntaxtreefunction.h → src/query/inc/qArithmeticOperator.h

@@ -23,7 +23,7 @@ extern "C" {
 typedef void (*_bi_consumer_fn_t)(void *left, void *right, int32_t numOfLeft, int32_t numOfRight, void *output,
                                   int32_t order);
 
-_bi_consumer_fn_t tGetBiConsumerFn(int32_t leftType, int32_t rightType, int32_t optr);
+_bi_consumer_fn_t getArithmeticOperatorFn(int32_t leftType, int32_t rightType, int32_t optr);
 
 #ifdef __cplusplus
 }

src/query/inc/qAst.h

@@ -74,9 +74,7 @@ typedef struct tExprNode {
   };
 } tExprNode;
 
-void tExprTreeTraverse(tExprNode *pExpr, SSkipList *pSkipList, SArray *result, SExprTraverseSupp *param);
-
-void tExprTreeCalcTraverse(tExprNode *pExprs, int32_t numOfRows, char *pOutput, void *param, int32_t order,
+void arithmeticTreeTraverse(tExprNode *pExprs, int32_t numOfRows, char *pOutput, void *param, int32_t order,
                                 char *(*cb)(void *, const char*, int32_t));
 
 tExprNode* exprTreeFromBinary(const void* data, size_t size);
@@ -87,6 +85,8 @@ void exprTreeToBinary(SBufferWriter* bw, tExprNode* pExprTree);
 void tExprNodeDestroy(tExprNode *pNode, void (*fp)(void *));
 void tExprTreeDestroy(tExprNode **pExprs, void (*fp)(void*));
 
+bool exprTreeApplayFilter(tExprNode *pExpr, const void *pItem, SExprTraverseSupp *param);
+
 #ifdef __cplusplus
 }
 #endif

src/query/inc/qExecutor.h

@@ -190,7 +190,7 @@ typedef struct SQueryRuntimeEnv {
   void*                pSecQueryHandle;  // another thread for
   bool                 stableQuery;      // super table query or not
   bool                 topBotQuery;      // TODO used bitwise flag
-  bool                 groupbyNormalCol; // denote if this is a groupby normal column query
+  bool                 groupbyColumn; // denote if this is a groupby normal column query
   bool                 hasTagResults;    // if there are tag values in final result or not
   bool                 timeWindowInterpo;// if the time window start/end required interpolation
   bool                 queryWindowIdentical; // all query time windows are identical for all tables in one group
@@ -204,6 +204,8 @@ typedef struct SQueryRuntimeEnv {
   int32_t*             rowCellInfoOffset;// offset value for each row result cell info
   char**               prevRow;
   char**               nextRow;
+
+  SArithmeticSupport  *sasArray;
 } SQueryRuntimeEnv;
 
 enum {
@@ -237,6 +239,7 @@ typedef struct SQInfo {
   int32_t          dataReady;   // denote if query result is ready or not
   void*            rspContext;  // response context
   int64_t          startExecTs; // start to exec timestamp
+  char*            sql;         // query sql string
 } SQInfo;
 
 #endif  // TDENGINE_QUERYEXECUTOR_H

src/query/inc/tsqlfunction.h

@@ -115,8 +115,7 @@ enum {
   MASTER_SCAN   = 0x0u,
   REVERSE_SCAN  = 0x1u,
   REPEAT_SCAN   = 0x2u,  //repeat scan belongs to the master scan
-  FIRST_STAGE_MERGE     = 0x10u,
-  SECONDARY_STAGE_MERGE = 0x20u,
+  MERGE_STAGE   = 0x20u,
 };
 
 #define QUERY_IS_STABLE_QUERY(type)      (((type)&TSDB_QUERY_TYPE_STABLE_QUERY) != 0)
@@ -191,8 +190,8 @@ typedef struct SQLFunctionCtx {
   int64_t      nStartQueryTimestamp;  // timestamp range of current query when function is executed on a specific data block
   int32_t      numOfParams;
   tVariant     param[4];      // input parameter, e.g., top(k, 20), the number of results for top query is kept in param */
-  int64_t *    ptsList;       // corresponding timestamp array list
-  void *       ptsOutputBuf;  // corresponding output buffer for timestamp of each result, e.g., top/bottom*/
+  int64_t     *ptsList;       // corresponding timestamp array list
+  void        *ptsOutputBuf;  // corresponding output buffer for timestamp of each result, e.g., top/bottom*/
   SQLPreAggVal preAggVals;
   tVariant     tag;
 
@@ -215,18 +214,12 @@ typedef struct SQLAggFuncElem {
   void (*xFunction)(SQLFunctionCtx *pCtx);                     // blocks version function
   void (*xFunctionF)(SQLFunctionCtx *pCtx, int32_t position);  // single-row function version
 
-  // some sql function require scan data twice or more, e.g.,stddev
+  // some sql function require scan data twice or more, e.g.,stddev, percentile
   void (*xNextStep)(SQLFunctionCtx *pCtx);
 
-  /*
-   * finalizer must be called after all xFunction has been executed to
-   * generated final result. Otherwise, the value in aOutputBuf is a intern result.
-   */
+  // finalizer must be called after all xFunction has been executed to generated final result.
   void (*xFinalize)(SQLFunctionCtx *pCtx);
-
-  void (*distMergeFunc)(SQLFunctionCtx *pCtx);
-
-  void (*distSecondaryMergeFunc)(SQLFunctionCtx *pCtx);
+  void (*mergeFunc)(SQLFunctionCtx *pCtx);
 
   int32_t (*dataReqFunc)(SQLFunctionCtx *pCtx, TSKEY start, TSKEY end, int32_t colId);
 } SQLAggFuncElem;

src/client/src/tscFunctionImpl.c → src/query/src/qAggMain.c

@@ -15,27 +15,26 @@
 
 #include "os.h"
 #include "qAst.h"
-#include "qExtbuffer.h"
 #include "qFill.h"
 #include "qHistogram.h"
 #include "qPercentile.h"
-#include "qSyntaxtreefunction.h"
 #include "qTsbuf.h"
 #include "taosdef.h"
 #include "taosmsg.h"
 #include "tscLog.h"
 #include "tscSubquery.h"
-#include "tscompression.h"
 #include "tsqlfunction.h"
-#include "tutil.h"
 #include "ttype.h"
 
-#define GET_INPUT_CHAR(x) (((char *)((x)->aInputElemBuf)) + ((x)->startOffset) * ((x)->inputBytes))
-#define GET_INPUT_CHAR_INDEX(x, y) (GET_INPUT_CHAR(x) + (y) * (x)->inputBytes)
+#define GET_INPUT_DATA_LIST(x) (((char *)((x)->aInputElemBuf)) + ((x)->startOffset) * ((x)->inputBytes))
+#define GET_INPUT_DATA(x, y) (GET_INPUT_DATA_LIST(x) + (y) * (x)->inputBytes)
+
+#define GET_TS_LIST(x)    ((TSKEY*)&((x)->ptsList[(x)->startOffset]))
+#define GET_TS_DATA(x, y) (GET_TS_LIST(x)[(y)])
 
 #define GET_TRUE_DATA_TYPE()                          \
   int32_t type = 0;                                   \
-  if (pCtx->currentStage == SECONDARY_STAGE_MERGE) {  \
+  if (pCtx->currentStage == MERGE_STAGE) {  \
     type = pCtx->outputType;                          \
     assert(pCtx->inputType == TSDB_DATA_TYPE_BINARY); \
   } else {                                            \
@@ -62,7 +61,7 @@
       }                                                                          \
       aAggs[TSDB_FUNC_TAG].xFunction(__ctx);                                     \
     }                                                                            \
-  } while (0);
+  } while (0)
 
 #define DO_UPDATE_TAG_COLUMNS_WITHOUT_TS(ctx)                   \
   do {                                                          \
@@ -395,7 +394,7 @@ static void count_function(SQLFunctionCtx *pCtx) {
   } else {
     if (pCtx->hasNull) {
       for (int32_t i = 0; i < pCtx->size; ++i) {
-        char *val = GET_INPUT_CHAR_INDEX(pCtx, i);
+        char *val = GET_INPUT_DATA(pCtx, i);
         if (isNull(val, pCtx->inputType)) {
           continue;
         }
@@ -420,7 +419,7 @@ static void count_function(SQLFunctionCtx *pCtx) {
 }
 
 static void count_function_f(SQLFunctionCtx *pCtx, int32_t index) {
-  char *pData = GET_INPUT_CHAR_INDEX(pCtx, index);
+  char *pData = GET_INPUT_DATA(pCtx, index);
   if (pCtx->hasNull && isNull(pData, pCtx->inputType)) {
     return;
   }
@@ -434,7 +433,7 @@ static void count_function_f(SQLFunctionCtx *pCtx, int32_t index) {
 }
 
 static void count_func_merge(SQLFunctionCtx *pCtx) {
-  int64_t *pData = (int64_t *)GET_INPUT_CHAR(pCtx);
+  int64_t *pData = (int64_t *)GET_INPUT_DATA_LIST(pCtx);
   for (int32_t i = 0; i < pCtx->size; ++i) {
     *((int64_t *)pCtx->aOutputBuf) += pData[i];
   }
@@ -498,7 +497,7 @@ int32_t no_data_info(SQLFunctionCtx *pCtx, TSKEY start, TSKEY end, int32_t colId
     if ((ctx)->hasNull && isNull((char *)&(list)[i], tsdbType)) { \
       continue;                                                   \
     }                                                             \
-    TSKEY key = (ctx)->ptsList[i];                                \
+    TSKEY key = GET_TS_DATA(ctx, i);                              \
     UPDATE_DATA(ctx, val, (list)[i], num, sign, key);             \
   }
 
@@ -525,7 +524,7 @@ static void do_sum(SQLFunctionCtx *pCtx) {
       *retVal += GET_DOUBLE_VAL((const char*)&(pCtx->preAggVals.statis.sum));
     }
   } else {  // computing based on the true data block
-    void *pData = GET_INPUT_CHAR(pCtx);
+    void *pData = GET_INPUT_DATA_LIST(pCtx);
     notNullElems = 0;
     
     if (pCtx->inputType >= TSDB_DATA_TYPE_TINYINT && pCtx->inputType <= TSDB_DATA_TYPE_BIGINT) {
@@ -558,7 +557,7 @@ static void do_sum(SQLFunctionCtx *pCtx) {
 }
 
 static void do_sum_f(SQLFunctionCtx *pCtx, int32_t index) {
-  void *pData = GET_INPUT_CHAR_INDEX(pCtx, index);
+  void *pData = GET_INPUT_DATA(pCtx, index);
   if (pCtx->hasNull && isNull(pData, pCtx->inputType)) {
     return;
   }
@@ -608,14 +607,14 @@ static void sum_function_f(SQLFunctionCtx *pCtx, int32_t index) {
   }
 }
 
-static int32_t sum_merge_impl(const SQLFunctionCtx *pCtx) {
+static void sum_func_merge(SQLFunctionCtx *pCtx) {
   int32_t notNullElems = 0;
 
   GET_TRUE_DATA_TYPE();
   assert(pCtx->stableQuery);
 
   for (int32_t i = 0; i < pCtx->size; ++i) {
-    char *    input = GET_INPUT_CHAR_INDEX(pCtx, i);
+    char *    input = GET_INPUT_DATA(pCtx, i);
     SSumInfo *pInput = (SSumInfo *)input;
     if (pInput->hasResult != DATA_SET_FLAG) {
       continue;
@@ -638,24 +637,6 @@ static int32_t sum_merge_impl(const SQLFunctionCtx *pCtx) {
     }
   }
 
-  return notNullElems;
-}
-
-static void sum_func_merge(SQLFunctionCtx *pCtx) {
-  int32_t notNullElems = sum_merge_impl(pCtx);
-  
-  SET_VAL(pCtx, notNullElems, 1);
-  SSumInfo *pSumInfo = (SSumInfo *)pCtx->aOutputBuf;
-  
-  if (notNullElems > 0) {
-    //    pCtx->numOfIteratedElems += notNullElems;
-    pSumInfo->hasResult = DATA_SET_FLAG;
-  }
-}
-
-static void sum_func_second_merge(SQLFunctionCtx *pCtx) {
-  int32_t notNullElems = sum_merge_impl(pCtx);
-  
   SET_VAL(pCtx, notNullElems, 1);
   SResultRowCellInfo *pResInfo = GET_RES_INFO(pCtx);
   
@@ -764,7 +745,7 @@ static void avg_function(SQLFunctionCtx *pCtx) {
       *pVal += GET_DOUBLE_VAL((const char *)&(pCtx->preAggVals.statis.sum));
     }
   } else {
-    void *pData = GET_INPUT_CHAR(pCtx);
+    void *pData = GET_INPUT_DATA_LIST(pCtx);
     
     if (pCtx->inputType == TSDB_DATA_TYPE_TINYINT) {
       LIST_ADD_N(*pVal, pCtx, pData, int8_t, notNullElems, pCtx->inputType);
@@ -799,7 +780,7 @@ static void avg_function(SQLFunctionCtx *pCtx) {
 }
 
 static void avg_function_f(SQLFunctionCtx *pCtx, int32_t index) {
-  void *pData = GET_INPUT_CHAR_INDEX(pCtx, index);
+  void *pData = GET_INPUT_DATA(pCtx, index);
   if (pCtx->hasNull && isNull(pData, pCtx->inputType)) {
     return;
   }
@@ -839,33 +820,9 @@ static void avg_function_f(SQLFunctionCtx *pCtx, int32_t index) {
 
 static void avg_func_merge(SQLFunctionCtx *pCtx) {
   SResultRowCellInfo *pResInfo = GET_RES_INFO(pCtx);
-  assert(pCtx->stableQuery);
-  
-  SAvgInfo *pAvgInfo = (SAvgInfo *)GET_ROWCELL_INTERBUF(pResInfo);
-  char *    input = GET_INPUT_CHAR(pCtx);
-  
-  for (int32_t i = 0; i < pCtx->size; ++i, input += pCtx->inputBytes) {
-    SAvgInfo *pInput = (SAvgInfo *)input;
-    if (pInput->num == 0) {  // current buffer is null
-      continue;
-    }
-    
-    pAvgInfo->sum += pInput->sum;
-    pAvgInfo->num += pInput->num;
-  }
-  
-  // if the data set hasResult is not set, the result is null
-  if (pAvgInfo->num > 0) {
-    pResInfo->hasResult = DATA_SET_FLAG;
-    memcpy(pCtx->aOutputBuf, GET_ROWCELL_INTERBUF(pResInfo), sizeof(SAvgInfo));
-  }
-}
-
-static void avg_func_second_merge(SQLFunctionCtx *pCtx) {
-  SResultRowCellInfo *pResInfo = GET_RES_INFO(pCtx);
   
   double *sum = (double*) pCtx->aOutputBuf;
-  char *  input = GET_INPUT_CHAR(pCtx);
+  char *  input = GET_INPUT_DATA_LIST(pCtx);
   
   for (int32_t i = 0; i < pCtx->size; ++i, input += pCtx->inputBytes) {
     SAvgInfo *pInput = (SAvgInfo *)input;
@@ -886,7 +843,7 @@ static void avg_func_second_merge(SQLFunctionCtx *pCtx) {
 static void avg_finalizer(SQLFunctionCtx *pCtx) {
   SResultRowCellInfo *pResInfo = GET_RES_INFO(pCtx);
   
-  if (pCtx->currentStage == SECONDARY_STAGE_MERGE) {
+  if (pCtx->currentStage == MERGE_STAGE) {
     assert(pCtx->inputType == TSDB_DATA_TYPE_BINARY);
     
     if (GET_INT64_VAL(GET_ROWCELL_INTERBUF(pResInfo)) <= 0) {
@@ -921,6 +878,10 @@ static void minMax_function(SQLFunctionCtx *pCtx, char *pOutput, int32_t isMin,
     *notNullElems = pCtx->size - pCtx->preAggVals.statis.numOfNull;
     assert(*notNullElems >= 0);
 
+    if (*notNullElems == 0) {
+      return;
+    }
+    
     if (*notNullElems == 0){
       return;
     }
@@ -948,6 +909,7 @@ static void minMax_function(SQLFunctionCtx *pCtx, char *pOutput, int32_t isMin,
         index = 0;
       }
 
+      // the index is the original position, not the relative position
       key = pCtx->ptsList[index];
     }
     
@@ -998,7 +960,9 @@ static void minMax_function(SQLFunctionCtx *pCtx, char *pOutput, int32_t isMin,
     return;
   }
   
-  void *p = GET_INPUT_CHAR(pCtx);
+  void  *p = GET_INPUT_DATA_LIST(pCtx);
+  TSKEY *tsList = GET_TS_LIST(pCtx);
+
   *notNullElems = 0;
   
   if (pCtx->inputType >= TSDB_DATA_TYPE_TINYINT && pCtx->inputType <= TSDB_DATA_TYPE_BIGINT) {
@@ -1017,7 +981,7 @@ static void minMax_function(SQLFunctionCtx *pCtx, char *pOutput, int32_t isMin,
         
         if ((*retVal < pData[i]) ^ isMin) {
           *retVal = pData[i];
-          TSKEY k = pCtx->ptsList[i];
+          TSKEY k = tsList[i];
           
           DO_UPDATE_TAG_COLUMNS(pCtx, k);
         }
@@ -1148,7 +1112,7 @@ static int32_t minmax_merge_impl(SQLFunctionCtx *pCtx, int32_t bytes, char *outp
   assert(pCtx->stableQuery);
   
   for (int32_t i = 0; i < pCtx->size; ++i) {
-    char *input = GET_INPUT_CHAR_INDEX(pCtx, i);
+    char *input = GET_INPUT_DATA(pCtx, i);
     if (input[bytes] != DATA_SET_FLAG) {
       continue;
     }
@@ -1202,17 +1166,6 @@ static int32_t minmax_merge_impl(SQLFunctionCtx *pCtx, int32_t bytes, char *outp
 }
 
 static void min_func_merge(SQLFunctionCtx *pCtx) {
-  int32_t notNullElems = minmax_merge_impl(pCtx, pCtx->inputBytes, pCtx->aOutputBuf, 1);
-  
-  SET_VAL(pCtx, notNullElems, 1);
-  
-  if (notNullElems > 0) {  // for super table query, SResultRowCellInfo is not used
-    char *flag = pCtx->aOutputBuf + pCtx->inputBytes;
-    *flag = DATA_SET_FLAG;
-  }
-}
-
-static void min_func_second_merge(SQLFunctionCtx *pCtx) {
   int32_t notNullElems = minmax_merge_impl(pCtx, pCtx->outputBytes, pCtx->aOutputBuf, 1);
   
   SET_VAL(pCtx, notNullElems, 1);
@@ -1224,16 +1177,6 @@ static void min_func_second_merge(SQLFunctionCtx *pCtx) {
 }
 
 static void max_func_merge(SQLFunctionCtx *pCtx) {
-  int32_t numOfElems = minmax_merge_impl(pCtx, pCtx->inputBytes, pCtx->aOutputBuf, 0);
-  
-  SET_VAL(pCtx, numOfElems, 1);
-  if (numOfElems > 0) {
-    char *flag = pCtx->aOutputBuf + pCtx->inputBytes;
-    *flag = DATA_SET_FLAG;
-  }
-}
-
-static void max_func_second_merge(SQLFunctionCtx *pCtx) {
   int32_t numOfElem = minmax_merge_impl(pCtx, pCtx->outputBytes, pCtx->aOutputBuf, 0);
   
   SET_VAL(pCtx, numOfElem, 1);
@@ -1245,8 +1188,8 @@ static void max_func_second_merge(SQLFunctionCtx *pCtx) {
 }
 
 static void minMax_function_f(SQLFunctionCtx *pCtx, int32_t index, int32_t isMin) {
-  char *pData = GET_INPUT_CHAR_INDEX(pCtx, index);
-  TSKEY key = pCtx->ptsList[index];
+  char *pData = GET_INPUT_DATA(pCtx, index);
+  TSKEY key   = GET_TS_DATA(pCtx, index);
   
   int32_t num = 0;
   if (pCtx->inputType == TSDB_DATA_TYPE_TINYINT) {
@@ -1285,7 +1228,7 @@ static void minMax_function_f(SQLFunctionCtx *pCtx, int32_t index, int32_t isMin
 }
 
 static void max_function_f(SQLFunctionCtx *pCtx, int32_t index) {
-  char *pData = GET_INPUT_CHAR_INDEX(pCtx, index);
+  char *pData = GET_INPUT_DATA(pCtx, index);
   if (pCtx->hasNull && isNull(pData, pCtx->inputType)) {
     return;
   }
@@ -1301,7 +1244,7 @@ static void max_function_f(SQLFunctionCtx *pCtx, int32_t index) {
 }
 
 static void min_function_f(SQLFunctionCtx *pCtx, int32_t index) {
-  char *pData = GET_INPUT_CHAR_INDEX(pCtx, index);
+  char *pData = GET_INPUT_DATA(pCtx, index);
   if (pCtx->hasNull && isNull(pData, pCtx->inputType)) {
     return;
   }
@@ -1334,7 +1277,7 @@ static void stddev_function(SQLFunctionCtx *pCtx) {
     double *retVal = &pStd->res;
     double  avg = pStd->avg;
     
-    void *pData = GET_INPUT_CHAR(pCtx);
+    void *pData = GET_INPUT_DATA_LIST(pCtx);
     
     switch (pCtx->inputType) {
       case TSDB_DATA_TYPE_INT: {
@@ -1385,7 +1328,7 @@ static void stddev_function_f(SQLFunctionCtx *pCtx, int32_t index) {
     avg_function_f(pCtx, index);
   } else {
     double avg = pStd->avg;
-    void * pData = GET_INPUT_CHAR_INDEX(pCtx, index);
+    void * pData = GET_INPUT_DATA(pCtx, index);
     
     if (pCtx->hasNull && isNull(pData, pCtx->inputType)) {
       return;
@@ -1491,14 +1434,14 @@ static void first_function(SQLFunctionCtx *pCtx) {
   
   // handle the null value
   for (int32_t i = 0; i < pCtx->size; ++i) {
-    char *data = GET_INPUT_CHAR_INDEX(pCtx, i);
+    char *data = GET_INPUT_DATA(pCtx, i);
     if (pCtx->hasNull && isNull(data, pCtx->inputType)) {
       continue;
     }
     
     memcpy(pCtx->aOutputBuf, data, pCtx->inputBytes);
     
-    TSKEY k = pCtx->ptsList[i];
+    TSKEY k = GET_TS_DATA(pCtx, i);
     DO_UPDATE_TAG_COLUMNS(pCtx, k);
     
     SResultRowCellInfo *pInfo = GET_RES_INFO(pCtx);
@@ -1517,7 +1460,7 @@ static void first_function_f(SQLFunctionCtx *pCtx, int32_t index) {
     return;
   }
   
-  void *pData = GET_INPUT_CHAR_INDEX(pCtx, index);
+  void *pData = GET_INPUT_DATA(pCtx, index);
   if (pCtx->hasNull && isNull(pData, pCtx->inputType)) {
     return;
   }
@@ -1525,7 +1468,7 @@ static void first_function_f(SQLFunctionCtx *pCtx, int32_t index) {
   SET_VAL(pCtx, 1, 1);
   memcpy(pCtx->aOutputBuf, pData, pCtx->inputBytes);
   
-  TSKEY ts = pCtx->ptsList[index];
+  TSKEY ts = GET_TS_DATA(pCtx, index);
   DO_UPDATE_TAG_COLUMNS(pCtx, ts);
   
   SResultRowCellInfo *pInfo = GET_RES_INFO(pCtx);
@@ -1534,7 +1477,7 @@ static void first_function_f(SQLFunctionCtx *pCtx, int32_t index) {
 }
 
 static void first_data_assign_impl(SQLFunctionCtx *pCtx, char *pData, int32_t index) {
-  int64_t *timestamp = pCtx->ptsList;
+  int64_t *timestamp = GET_TS_LIST(pCtx);
   
   SFirstLastInfo *pInfo = (SFirstLastInfo *)(pCtx->aOutputBuf + pCtx->inputBytes);
   
@@ -1565,7 +1508,7 @@ static void first_dist_function(SQLFunctionCtx *pCtx) {
 
   // find the first not null value
   for (int32_t i = 0; i < pCtx->size; ++i) {
-    char *data = GET_INPUT_CHAR_INDEX(pCtx, i);
+    char *data = GET_INPUT_DATA(pCtx, i);
     if (pCtx->hasNull && isNull(data, pCtx->inputType)) {
       continue;
     }
@@ -1583,7 +1526,7 @@ static void first_dist_function(SQLFunctionCtx *pCtx) {
 }
 
 static void first_dist_function_f(SQLFunctionCtx *pCtx, int32_t index) {
-  char *pData = GET_INPUT_CHAR_INDEX(pCtx, index);
+  char *pData = GET_INPUT_DATA(pCtx, index);
   if (pCtx->hasNull && isNull(pData, pCtx->inputType)) {
     return;
   }
@@ -1598,26 +1541,9 @@ static void first_dist_function_f(SQLFunctionCtx *pCtx, int32_t index) {
 }
 
 static void first_dist_func_merge(SQLFunctionCtx *pCtx) {
-  char *pData = GET_INPUT_CHAR(pCtx);
-  
-  assert(pCtx->size == 1 && pCtx->stableQuery);
-  
-  SFirstLastInfo *pInput = (SFirstLastInfo *)(pData + pCtx->inputBytes);
-  if (pInput->hasResult != DATA_SET_FLAG) {
-    return;
-  }
-  
-  SFirstLastInfo *pOutput = (SFirstLastInfo *)(pCtx->aOutputBuf + pCtx->inputBytes);
-  if (pOutput->hasResult != DATA_SET_FLAG || pInput->ts < pOutput->ts) {
-    memcpy(pCtx->aOutputBuf, pData, pCtx->inputBytes + sizeof(SFirstLastInfo));
-    DO_UPDATE_TAG_COLUMNS(pCtx, pInput->ts);
-  }
-}
-
-static void first_dist_func_second_merge(SQLFunctionCtx *pCtx) {
   assert(pCtx->stableQuery);
 
-  char *          pData = GET_INPUT_CHAR(pCtx);
+  char *          pData = GET_INPUT_DATA_LIST(pCtx);
   SFirstLastInfo *pInput = (SFirstLastInfo*) (pData + pCtx->outputBytes);
   if (pInput->hasResult != DATA_SET_FLAG) {
     return;
@@ -1652,7 +1578,7 @@ static void last_function(SQLFunctionCtx *pCtx) {
   int32_t notNullElems = 0;
   
   for (int32_t i = pCtx->size - 1; i >= 0; --i) {
-    char *data = GET_INPUT_CHAR_INDEX(pCtx, i);
+    char *data = GET_INPUT_DATA(pCtx, i);
     if (pCtx->hasNull && isNull(data, pCtx->inputType)) {
       if (!pCtx->requireNull) {
         continue; 
@@ -1660,7 +1586,7 @@ static void last_function(SQLFunctionCtx *pCtx) {
     }
     memcpy(pCtx->aOutputBuf, data, pCtx->inputBytes);
     
-    TSKEY ts = pCtx->ptsList[i];
+    TSKEY ts = GET_TS_DATA(pCtx, i);
     DO_UPDATE_TAG_COLUMNS(pCtx, ts);
     
     SResultRowCellInfo *pInfo = GET_RES_INFO(pCtx);
@@ -1675,7 +1601,7 @@ static void last_function(SQLFunctionCtx *pCtx) {
 }
 
 static void last_function_f(SQLFunctionCtx *pCtx, int32_t index) {
-  void *pData = GET_INPUT_CHAR_INDEX(pCtx, index);
+  void *pData = GET_INPUT_DATA(pCtx, index);
   if (pCtx->hasNull && isNull(pData, pCtx->inputType)) {
     return;
   }
@@ -1689,7 +1615,7 @@ static void last_function_f(SQLFunctionCtx *pCtx, int32_t index) {
     SET_VAL(pCtx, 1, 1);
     memcpy(pCtx->aOutputBuf, pData, pCtx->inputBytes);
 
-    TSKEY ts = pCtx->ptsList[index];
+    TSKEY ts = GET_TS_DATA(pCtx, index);
     DO_UPDATE_TAG_COLUMNS(pCtx, ts);
 
     SResultRowCellInfo *pResInfo = GET_RES_INFO(pCtx);
@@ -1697,7 +1623,7 @@ static void last_function_f(SQLFunctionCtx *pCtx, int32_t index) {
     pResInfo->complete = true;  // set query completed
   } else { // in case of ascending order check, all data needs to be checked
     SResultRowCellInfo* pResInfo = GET_RES_INFO(pCtx);
-    TSKEY ts = pCtx->ptsList[index];
+    TSKEY ts = GET_TS_DATA(pCtx, index);
 
     char* buf = GET_ROWCELL_INTERBUF(pResInfo);
     if (pResInfo->hasResult != DATA_SET_FLAG || (*(TSKEY*)buf) < ts) {
@@ -1711,7 +1637,7 @@ static void last_function_f(SQLFunctionCtx *pCtx, int32_t index) {
 }
 
 static void last_data_assign_impl(SQLFunctionCtx *pCtx, char *pData, int32_t index) {
-  int64_t *timestamp = pCtx->ptsList;
+  int64_t *timestamp = GET_TS_LIST(pCtx);
   
   SFirstLastInfo *pInfo = (SFirstLastInfo *)(pCtx->aOutputBuf + pCtx->inputBytes);
   
@@ -1745,7 +1671,7 @@ static void last_dist_function(SQLFunctionCtx *pCtx) {
   int32_t notNullElems = 0;
   
   for (int32_t i = pCtx->size - 1; i >= 0; --i) {
-    char *data = GET_INPUT_CHAR_INDEX(pCtx, i);
+    char *data = GET_INPUT_DATA(pCtx, i);
     if (pCtx->hasNull && isNull(data, pCtx->inputType)) {
       if (!pCtx->requireNull) {
         continue; 
@@ -1769,7 +1695,7 @@ static void last_dist_function_f(SQLFunctionCtx *pCtx, int32_t index) {
     return;
   }
   
-  char *pData = GET_INPUT_CHAR_INDEX(pCtx, index);
+  char *pData = GET_INPUT_DATA(pCtx, index);
   if (pCtx->hasNull && isNull(pData, pCtx->inputType)) {
     return;
   }
@@ -1787,32 +1713,13 @@ static void last_dist_function_f(SQLFunctionCtx *pCtx, int32_t index) {
   SET_VAL(pCtx, 1, 1);
 }
 
-static void last_dist_func_merge(SQLFunctionCtx *pCtx) {
-  char *pData = GET_INPUT_CHAR(pCtx);
-  
-  assert(pCtx->size == 1 && pCtx->stableQuery);
-  
-  // the input data is null
-  SFirstLastInfo *pInput = (SFirstLastInfo *)(pData + pCtx->inputBytes);
-  if (pInput->hasResult != DATA_SET_FLAG) {
-    return;
-  }
-  
-  SFirstLastInfo *pOutput = (SFirstLastInfo *)(pCtx->aOutputBuf + pCtx->inputBytes);
-  if (pOutput->hasResult != DATA_SET_FLAG || pOutput->ts < pInput->ts) {
-    memcpy(pCtx->aOutputBuf, pData, pCtx->inputBytes + sizeof(SFirstLastInfo));
-    
-    DO_UPDATE_TAG_COLUMNS(pCtx, pInput->ts);
-  }
-}
-
 /*
  * in the secondary merge(local reduce), the output is limited by the
  * final output size, so the main difference between last_dist_func_merge and second_merge
  * is: the output data format in computing
  */
-static void last_dist_func_second_merge(SQLFunctionCtx *pCtx) {
-  char *pData = GET_INPUT_CHAR(pCtx);
+static void last_dist_func_merge(SQLFunctionCtx *pCtx) {
+  char *pData = GET_INPUT_DATA_LIST(pCtx);
   
   SFirstLastInfo *pInput = (SFirstLastInfo*) (pData + pCtx->outputBytes);
   if (pInput->hasResult != DATA_SET_FLAG) {
@@ -1841,7 +1748,7 @@ static void last_dist_func_second_merge(SQLFunctionCtx *pCtx) {
  */
 static void last_row_function(SQLFunctionCtx *pCtx) {
   assert(pCtx->size >= 1);
-  char *pData = GET_INPUT_CHAR(pCtx);
+  char *pData = GET_INPUT_DATA_LIST(pCtx);
 
   // assign the last element in current data block
   assignVal(pCtx->aOutputBuf, pData + (pCtx->size - 1) * pCtx->inputBytes, pCtx->inputBytes, pCtx->inputType);
@@ -1852,12 +1759,13 @@ static void last_row_function(SQLFunctionCtx *pCtx) {
   // set the result to final result buffer in case of super table query
   if (pCtx->stableQuery) {
     SLastrowInfo *pInfo1 = (SLastrowInfo *)(pCtx->aOutputBuf + pCtx->inputBytes);
-    pInfo1->ts = pCtx->ptsList[pCtx->size - 1];
+    pInfo1->ts = GET_TS_DATA(pCtx, pCtx->size - 1);
     pInfo1->hasResult = DATA_SET_FLAG;
     
     DO_UPDATE_TAG_COLUMNS(pCtx, pInfo1->ts);
   } else {
-    DO_UPDATE_TAG_COLUMNS(pCtx, pCtx->ptsList[pCtx->size - 1]);
+    TSKEY ts = GET_TS_DATA(pCtx, pCtx->size - 1);
+    DO_UPDATE_TAG_COLUMNS(pCtx, ts);
   }
 
   SET_VAL(pCtx, pCtx->size, 1);
@@ -1888,7 +1796,7 @@ static void valuePairAssign(tValuePair *dst, int16_t type, const char *val, int6
   dst->timestamp = tsKey;
   
   int32_t size = 0;
-  if (stage == SECONDARY_STAGE_MERGE || stage == FIRST_STAGE_MERGE) {
+  if (stage == MERGE_STAGE) {
     memcpy(dst->pTags, pTags, (size_t)pTagInfo->tagsLen);
   } else {  // the tags are dumped from the ctx tag fields
     for (int32_t i = 0; i < pTagInfo->numOfTagCols; ++i) {
@@ -2151,7 +2059,7 @@ static STopBotInfo *getTopBotOutputInfo(SQLFunctionCtx *pCtx) {
   SResultRowCellInfo *pResInfo = GET_RES_INFO(pCtx);
 
   // only the first_stage_merge is directly written data into final output buffer
-  if (pCtx->stableQuery && pCtx->currentStage != SECONDARY_STAGE_MERGE) {
+  if (pCtx->stableQuery && pCtx->currentStage != MERGE_STAGE) {
     return (STopBotInfo*) pCtx->aOutputBuf;
   } else { // during normal table query and super table at the secondary_stage, result is written to intermediate buffer
     return GET_ROWCELL_INTERBUF(pResInfo);
@@ -2249,13 +2157,15 @@ static void top_function(SQLFunctionCtx *pCtx) {
   assert(pRes->num >= 0);
   
   for (int32_t i = 0; i < pCtx->size; ++i) {
-    char *data = GET_INPUT_CHAR_INDEX(pCtx, i);
+    char *data = GET_INPUT_DATA(pCtx, i);
+    TSKEY ts = GET_TS_DATA(pCtx, i);
+
     if (pCtx->hasNull && isNull(data, pCtx->inputType)) {
       continue;
     }
     
     notNullElems++;
-    do_top_function_add(pRes, (int32_t)pCtx->param[0].i64Key, data, pCtx->ptsList[i], pCtx->inputType, &pCtx->tagInfo, NULL, 0);
+    do_top_function_add(pRes, (int32_t)pCtx->param[0].i64Key, data, ts, pCtx->inputType, &pCtx->tagInfo, NULL, 0);
   }
   
   if (!pCtx->hasNull) {
@@ -2272,7 +2182,7 @@ static void top_function(SQLFunctionCtx *pCtx) {
 }
 
 static void top_function_f(SQLFunctionCtx *pCtx, int32_t index) {
-  char *pData = GET_INPUT_CHAR_INDEX(pCtx, index);
+  char *pData = GET_INPUT_DATA(pCtx, index);
   if (pCtx->hasNull && isNull(pData, pCtx->inputType)) {
     return;
   }
@@ -2281,36 +2191,16 @@ static void top_function_f(SQLFunctionCtx *pCtx, int32_t index) {
   assert(pRes->num >= 0);
   
   SET_VAL(pCtx, 1, 1);
-  do_top_function_add(pRes, (int32_t)pCtx->param[0].i64Key, pData, pCtx->ptsList[index], pCtx->inputType, &pCtx->tagInfo, NULL,
-                      0);
+  TSKEY ts = GET_TS_DATA(pCtx, index);
+
+  do_top_function_add(pRes, (int32_t)pCtx->param[0].i64Key, pData, ts, pCtx->inputType, &pCtx->tagInfo, NULL, 0);
   
   SResultRowCellInfo *pResInfo = GET_RES_INFO(pCtx);
   pResInfo->hasResult = DATA_SET_FLAG;
 }
 
 static void top_func_merge(SQLFunctionCtx *pCtx) {
-  char *input = GET_INPUT_CHAR(pCtx);
-  
-  STopBotInfo *pInput = (STopBotInfo *)input;
-  if (pInput->num <= 0) {
-    return;
-  }
-  
-  // remmap the input buffer may cause the struct pointer invalid, so rebuild the STopBotInfo is necessary
-  buildTopBotStruct(pInput, pCtx);
-  
-  assert(pCtx->stableQuery && pCtx->outputType == TSDB_DATA_TYPE_BINARY && pCtx->size == 1);
-  
-  STopBotInfo *pOutput = getTopBotOutputInfo(pCtx);
-  
-  for (int32_t i = 0; i < pInput->num; ++i) {
-    do_top_function_add(pOutput, (int32_t)pCtx->param[0].i64Key, &pInput->res[i]->v.i64Key, pInput->res[i]->timestamp,
-                        pCtx->inputType, &pCtx->tagInfo, pInput->res[i]->pTags, pCtx->currentStage);
-  }
-}
-
-static void top_func_second_merge(SQLFunctionCtx *pCtx) {
-  STopBotInfo *pInput = (STopBotInfo *)GET_INPUT_CHAR(pCtx);
+  STopBotInfo *pInput = (STopBotInfo *)GET_INPUT_DATA_LIST(pCtx);
   
   // construct the input data struct from binary data
   buildTopBotStruct(pInput, pCtx);
@@ -2338,14 +2228,15 @@ static void bottom_function(SQLFunctionCtx *pCtx) {
   STopBotInfo *pRes = getTopBotOutputInfo(pCtx);
   
   for (int32_t i = 0; i < pCtx->size; ++i) {
-    char *data = GET_INPUT_CHAR_INDEX(pCtx, i);
+    char *data = GET_INPUT_DATA(pCtx, i);
+    TSKEY ts = GET_TS_DATA(pCtx, i);
+
     if (pCtx->hasNull && isNull(data, pCtx->inputType)) {
       continue;
     }
     
     notNullElems++;
-    do_bottom_function_add(pRes, (int32_t)pCtx->param[0].i64Key, data, pCtx->ptsList[i], pCtx->inputType, &pCtx->tagInfo, NULL,
-                           0);
+    do_bottom_function_add(pRes, (int32_t)pCtx->param[0].i64Key, data, ts, pCtx->inputType, &pCtx->tagInfo, NULL, 0);
   }
   
   if (!pCtx->hasNull) {
@@ -2362,43 +2253,23 @@ static void bottom_function(SQLFunctionCtx *pCtx) {
 }
 
 static void bottom_function_f(SQLFunctionCtx *pCtx, int32_t index) {
-  char *pData = GET_INPUT_CHAR_INDEX(pCtx, index);
+  char *pData = GET_INPUT_DATA(pCtx, index);
+  TSKEY ts = GET_TS_DATA(pCtx, index);
+
   if (pCtx->hasNull && isNull(pData, pCtx->inputType)) {
     return;
   }
   
   STopBotInfo *pRes = getTopBotOutputInfo(pCtx);
   SET_VAL(pCtx, 1, 1);
-  do_bottom_function_add(pRes, (int32_t)pCtx->param[0].i64Key, pData, pCtx->ptsList[index], pCtx->inputType, &pCtx->tagInfo,
-                         NULL, 0);
+  do_bottom_function_add(pRes, (int32_t)pCtx->param[0].i64Key, pData, ts, pCtx->inputType, &pCtx->tagInfo, NULL, 0);
   
   SResultRowCellInfo *pResInfo = GET_RES_INFO(pCtx);
   pResInfo->hasResult = DATA_SET_FLAG;
 }
 
 static void bottom_func_merge(SQLFunctionCtx *pCtx) {
-  char *input = GET_INPUT_CHAR(pCtx);
-  
-  STopBotInfo *pInput = (STopBotInfo *)input;
-  if (pInput->num <= 0) {
-    return;
-  }
-  
-  // remmap the input buffer may cause the struct pointer invalid, so rebuild the STopBotInfo is necessary
-  buildTopBotStruct(pInput, pCtx);
-  
-  assert(pCtx->stableQuery && pCtx->outputType == TSDB_DATA_TYPE_BINARY && pCtx->size == 1);
-  
-  STopBotInfo *pOutput = getTopBotOutputInfo(pCtx);
-  
-  for (int32_t i = 0; i < pInput->num; ++i) {
-    do_bottom_function_add(pOutput, (int32_t)pCtx->param[0].i64Key, &pInput->res[i]->v.i64Key, pInput->res[i]->timestamp,
-                           pCtx->inputType, &pCtx->tagInfo, pInput->res[i]->pTags, pCtx->currentStage);
-  }
-}
-
-static void bottom_func_second_merge(SQLFunctionCtx *pCtx) {
-  STopBotInfo *pInput = (STopBotInfo *)GET_INPUT_CHAR(pCtx);
+  STopBotInfo *pInput = (STopBotInfo *)GET_INPUT_DATA_LIST(pCtx);
   
   // construct the input data struct from binary data
   buildTopBotStruct(pInput, pCtx);
@@ -2494,7 +2365,7 @@ static void percentile_function(SQLFunctionCtx *pCtx) {
       pInfo->numOfElems += (pCtx->size - pCtx->preAggVals.statis.numOfNull);
     } else {
       for (int32_t i = 0; i < pCtx->size; ++i) {
-        char *data = GET_INPUT_CHAR_INDEX(pCtx, i);
+        char *data = GET_INPUT_DATA(pCtx, i);
         if (pCtx->hasNull && isNull(data, pCtx->inputType)) {
           continue;
         }
@@ -2519,7 +2390,7 @@ static void percentile_function(SQLFunctionCtx *pCtx) {
 
   // the second stage, calculate the true percentile value
   for (int32_t i = 0; i < pCtx->size; ++i) {
-    char *data = GET_INPUT_CHAR_INDEX(pCtx, i);
+    char *data = GET_INPUT_DATA(pCtx, i);
     if (pCtx->hasNull && isNull(data, pCtx->inputType)) {
       continue;
     }
@@ -2533,7 +2404,7 @@ static void percentile_function(SQLFunctionCtx *pCtx) {
 }
 
 static void percentile_function_f(SQLFunctionCtx *pCtx, int32_t index) {
-  void *pData = GET_INPUT_CHAR_INDEX(pCtx, index);
+  void *pData = GET_INPUT_DATA(pCtx, index);
   if (pCtx->hasNull && isNull(pData, pCtx->inputType)) {
     return;
   }
@@ -2610,7 +2481,7 @@ static SAPercentileInfo *getAPerctInfo(SQLFunctionCtx *pCtx) {
   SResultRowCellInfo *pResInfo = GET_RES_INFO(pCtx);
   SAPercentileInfo* pInfo = NULL;
 
-  if (pCtx->stableQuery && pCtx->currentStage != SECONDARY_STAGE_MERGE) {
+  if (pCtx->stableQuery && pCtx->currentStage != MERGE_STAGE) {
     pInfo = (SAPercentileInfo*) pCtx->aOutputBuf;
   } else {
     pInfo = GET_ROWCELL_INTERBUF(pResInfo);
@@ -2641,7 +2512,7 @@ static void apercentile_function(SQLFunctionCtx *pCtx) {
   assert(pInfo->pHisto->elems != NULL);
   
   for (int32_t i = 0; i < pCtx->size; ++i) {
-    char *data = GET_INPUT_CHAR_INDEX(pCtx, i);
+    char *data = GET_INPUT_DATA(pCtx, i);
     if (pCtx->hasNull && isNull(data, pCtx->inputType)) {
       continue;
     }
@@ -2665,7 +2536,7 @@ static void apercentile_function(SQLFunctionCtx *pCtx) {
 }
 
 static void apercentile_function_f(SQLFunctionCtx *pCtx, int32_t index) {
-  void *pData = GET_INPUT_CHAR_INDEX(pCtx, index);
+  void *pData = GET_INPUT_DATA(pCtx, index);
   if (pCtx->hasNull && isNull(pData, pCtx->inputType)) {
     return;
   }
@@ -2683,42 +2554,7 @@ static void apercentile_function_f(SQLFunctionCtx *pCtx, int32_t index) {
 }
 
 static void apercentile_func_merge(SQLFunctionCtx *pCtx) {
-  SResultRowCellInfo *pResInfo = GET_RES_INFO(pCtx);
-  assert(pCtx->stableQuery);
-  
-  SAPercentileInfo *pInput = (SAPercentileInfo *)GET_INPUT_CHAR(pCtx);
-  
-  pInput->pHisto = (SHistogramInfo*) ((char *)pInput + sizeof(SAPercentileInfo));
-  pInput->pHisto->elems = (SHistBin*) ((char *)pInput->pHisto + sizeof(SHistogramInfo));
-  
-  if (pInput->pHisto->numOfElems <= 0) {
-    return;
-  }
-  
-  size_t size = sizeof(SHistogramInfo) + sizeof(SHistBin) * (MAX_HISTOGRAM_BIN + 1);
-  
-  SAPercentileInfo *pOutput = getAPerctInfo(pCtx);  //(SAPercentileInfo *)pCtx->aOutputBuf;
-  SHistogramInfo *  pHisto = pOutput->pHisto;
-  
-  if (pHisto->numOfElems <= 0) {
-    memcpy(pHisto, pInput->pHisto, size);
-    pHisto->elems = (SHistBin*) ((char *)pHisto + sizeof(SHistogramInfo));
-  } else {
-    pHisto->elems = (SHistBin*) ((char *)pHisto + sizeof(SHistogramInfo));
-    
-    SHistogramInfo *pRes = tHistogramMerge(pHisto, pInput->pHisto, MAX_HISTOGRAM_BIN);
-    memcpy(pHisto, pRes, sizeof(SHistogramInfo) + sizeof(SHistBin) * MAX_HISTOGRAM_BIN);
-    pHisto->elems = (SHistBin*) ((char *)pHisto + sizeof(SHistogramInfo));
-    
-    tHistogramDestroy(&pRes);
-  }
-  
-  SET_VAL(pCtx, 1, 1);
-  pResInfo->hasResult = DATA_SET_FLAG;
-}
-
-static void apercentile_func_second_merge(SQLFunctionCtx *pCtx) {
-  SAPercentileInfo *pInput = (SAPercentileInfo *)GET_INPUT_CHAR(pCtx);
+  SAPercentileInfo *pInput = (SAPercentileInfo *)GET_INPUT_DATA_LIST(pCtx);
   
   pInput->pHisto = (SHistogramInfo*) ((char *)pInput + sizeof(SAPercentileInfo));
   pInput->pHisto->elems = (SHistBin*) ((char *)pInput->pHisto + sizeof(SHistogramInfo));
@@ -2753,7 +2589,7 @@ static void apercentile_finalizer(SQLFunctionCtx *pCtx) {
   SResultRowCellInfo *     pResInfo = GET_RES_INFO(pCtx);
   SAPercentileInfo *pOutput = GET_ROWCELL_INTERBUF(pResInfo);
   
-  if (pCtx->currentStage == SECONDARY_STAGE_MERGE) {
+  if (pCtx->currentStage == MERGE_STAGE) {
     if (pResInfo->hasResult == DATA_SET_FLAG) {  // check for null
       assert(pOutput->pHisto->numOfElems > 0);
       
@@ -2821,7 +2657,7 @@ static void leastsquares_function(SQLFunctionCtx *pCtx) {
   double(*param)[3] = pInfo->mat;
   double x = pInfo->startVal;
   
-  void *pData = GET_INPUT_CHAR(pCtx);
+  void *pData = GET_INPUT_DATA_LIST(pCtx);
   
   int32_t numOfElem = 0;
   switch (pCtx->inputType) {
@@ -2881,7 +2717,7 @@ static void leastsquares_function(SQLFunctionCtx *pCtx) {
 }
 
 static void leastsquares_function_f(SQLFunctionCtx *pCtx, int32_t index) {
-  void *pData = GET_INPUT_CHAR_INDEX(pCtx, index);
+  void *pData = GET_INPUT_DATA(pCtx, index);
   if (pCtx->hasNull && isNull(pData, pCtx->inputType)) {
     return;
   }
@@ -2983,7 +2819,7 @@ static void col_project_function(SQLFunctionCtx *pCtx) {
 
   INC_INIT_VAL(pCtx, pCtx->size);
 
-  char *pData = GET_INPUT_CHAR(pCtx);
+  char *pData = GET_INPUT_DATA_LIST(pCtx);
   if (pCtx->order == TSDB_ORDER_ASC) {
     memcpy(pCtx->aOutputBuf, pData, (size_t) pCtx->size * pCtx->inputBytes);
   } else {
@@ -3008,7 +2844,7 @@ static void col_project_function_f(SQLFunctionCtx *pCtx, int32_t index) {
   }
 
   INC_INIT_VAL(pCtx, 1);
-  char *pData = GET_INPUT_CHAR_INDEX(pCtx, index);
+  char *pData = GET_INPUT_DATA(pCtx, index);
   memcpy(pCtx->aOutputBuf, pData, pCtx->inputBytes);
 
   pCtx->aOutputBuf += pCtx->inputBytes;
@@ -3062,7 +2898,7 @@ static void tag_function_f(SQLFunctionCtx *pCtx, int32_t index) {
 static void copy_function(SQLFunctionCtx *pCtx) {
   SET_VAL(pCtx, pCtx->size, 1);
   
-  char *pData = GET_INPUT_CHAR(pCtx);
+  char *pData = GET_INPUT_DATA_LIST(pCtx);
   assignVal(pCtx->aOutputBuf, pData, pCtx->inputBytes, pCtx->inputType);
 }
 
@@ -3082,7 +2918,7 @@ static bool diff_function_setup(SQLFunctionCtx *pCtx) {
 
 // TODO difference in date column
 static void diff_function(SQLFunctionCtx *pCtx) {
-  void *data = GET_INPUT_CHAR(pCtx);
+  void *data = GET_INPUT_DATA_LIST(pCtx);
   bool  isFirstBlock = (pCtx->param[1].nType == INITIAL_VALUE_NOT_ASSIGNED);
   
   int32_t notNullElems = 0;
@@ -3090,7 +2926,8 @@ static void diff_function(SQLFunctionCtx *pCtx) {
   int32_t step = GET_FORWARD_DIRECTION_FACTOR(pCtx->order);
   int32_t i = (pCtx->order == TSDB_ORDER_ASC) ? 0 : pCtx->size - 1;
   
-  TSKEY * pTimestamp = pCtx->ptsOutputBuf;
+  TSKEY* pTimestamp = pCtx->ptsOutputBuf;
+  TSKEY* tsList = GET_TS_LIST(pCtx);
 
   switch (pCtx->inputType) {
     case TSDB_DATA_TYPE_INT: {
@@ -3107,13 +2944,13 @@ static void diff_function(SQLFunctionCtx *pCtx) {
           pCtx->param[1].nType = pCtx->inputType;
         } else if ((i == 0 && pCtx->order == TSDB_ORDER_ASC) || (i == pCtx->size - 1 && pCtx->order == TSDB_ORDER_DESC)) {
           *pOutput = (int32_t)(pData[i] - pCtx->param[1].i64Key);
-          *pTimestamp = pCtx->ptsList[i];
+          *pTimestamp = tsList[i];
           
           pOutput += 1;
           pTimestamp += 1;
         } else {
           *pOutput = (int32_t)(pData[i] - pCtx->param[1].i64Key);  // direct previous may be null
-          *pTimestamp = pCtx->ptsList[i];
+          *pTimestamp = tsList[i];
           
           pOutput += 1;
           pTimestamp += 1;
@@ -3139,13 +2976,13 @@ static void diff_function(SQLFunctionCtx *pCtx) {
           pCtx->param[1].nType = pCtx->inputType;
         } else if ((i == 0 && pCtx->order == TSDB_ORDER_ASC) || (i == pCtx->size - 1 && pCtx->order == TSDB_ORDER_DESC)) {
           *pOutput = pData[i] - pCtx->param[1].i64Key;
-          *pTimestamp = pCtx->ptsList[i];
+          *pTimestamp = tsList[i];
           
           pOutput += 1;
           pTimestamp += 1;
         } else {
           *pOutput = pData[i] - pCtx->param[1].i64Key;
-          *pTimestamp = pCtx->ptsList[i];
+          *pTimestamp = tsList[i];
           
           pOutput += 1;
           pTimestamp += 1;
@@ -3171,12 +3008,12 @@ static void diff_function(SQLFunctionCtx *pCtx) {
           pCtx->param[1].nType = pCtx->inputType;
         } else if ((i == 0 && pCtx->order == TSDB_ORDER_ASC) || (i == pCtx->size - 1 && pCtx->order == TSDB_ORDER_DESC)) {
           *pOutput = pData[i] - pCtx->param[1].dKey;
-          *pTimestamp = pCtx->ptsList[i];
+          *pTimestamp = tsList[i];
           pOutput += 1;
           pTimestamp += 1;
         } else {
           *pOutput = pData[i] - pCtx->param[1].dKey;
-          *pTimestamp = pCtx->ptsList[i];
+          *pTimestamp = tsList[i];
           pOutput += 1;
           pTimestamp += 1;
         }
@@ -3201,13 +3038,13 @@ static void diff_function(SQLFunctionCtx *pCtx) {
           pCtx->param[1].nType = pCtx->inputType;
         } else if ((i == 0 && pCtx->order == TSDB_ORDER_ASC) || (i == pCtx->size - 1 && pCtx->order == TSDB_ORDER_DESC)) {
           *pOutput = (float)(pData[i] - pCtx->param[1].dKey);
-          *pTimestamp = pCtx->ptsList[i];
+          *pTimestamp = tsList[i];
           
           pOutput += 1;
           pTimestamp += 1;
         } else {
           *pOutput = (float)(pData[i] - pCtx->param[1].dKey);
-          *pTimestamp = pCtx->ptsList[i];
+          *pTimestamp = tsList[i];
           
           pOutput += 1;
           pTimestamp += 1;
@@ -3234,12 +3071,12 @@ static void diff_function(SQLFunctionCtx *pCtx) {
           pCtx->param[1].nType = pCtx->inputType;
         } else if ((i == 0 && pCtx->order == TSDB_ORDER_ASC) || (i == pCtx->size - 1 && pCtx->order == TSDB_ORDER_DESC)) {
           *pOutput = (int16_t)(pData[i] - pCtx->param[1].i64Key);
-          *pTimestamp = pCtx->ptsList[i];
+          *pTimestamp = tsList[i];
           pOutput += 1;
           pTimestamp += 1;
         } else {
           *pOutput = (int16_t)(pData[i] - pCtx->param[1].i64Key);
-          *pTimestamp = pCtx->ptsList[i];
+          *pTimestamp = tsList[i];
           
           pOutput += 1;
           pTimestamp += 1;
@@ -3265,13 +3102,13 @@ static void diff_function(SQLFunctionCtx *pCtx) {
           pCtx->param[1].nType = pCtx->inputType;
         } else if ((i == 0 && pCtx->order == TSDB_ORDER_ASC) || (i == pCtx->size - 1 && pCtx->order == TSDB_ORDER_DESC)) {
           *pOutput = (int8_t)(pData[i] - pCtx->param[1].i64Key);
-          *pTimestamp = pCtx->ptsList[i];
+          *pTimestamp = tsList[i];
           
           pOutput += 1;
           pTimestamp += 1;
         } else {
           *pOutput = (int8_t)(pData[i] - pCtx->param[1].i64Key);
-          *pTimestamp = pCtx->ptsList[i];
+          *pTimestamp = tsList[i];
           
           pOutput += 1;
           pTimestamp += 1;
@@ -3312,12 +3149,12 @@ static void diff_function(SQLFunctionCtx *pCtx) {
     } else {                                                                                 \
       *(type *)(ctx)->aOutputBuf = *(type *)(d) - (*(type *)(&(ctx)->param[1].i64Key));      \
       *(type *)(&(ctx)->param[1].i64Key) = *(type *)(d);                                     \
-      *(int64_t *)(ctx)->ptsOutputBuf = (ctx)->ptsList[index];                               \
+      *(int64_t *)(ctx)->ptsOutputBuf = GET_TS_DATA(ctx, index);                             \
     }                                                                                        \
   } while (0);
 
 static void diff_function_f(SQLFunctionCtx *pCtx, int32_t index) {
-  char *pData = GET_INPUT_CHAR_INDEX(pCtx, index);
+  char *pData = GET_INPUT_DATA(pCtx, index);
   if (pCtx->hasNull && isNull(pData, pCtx->inputType)) {
     return;
   }
@@ -3337,7 +3174,7 @@ static void diff_function_f(SQLFunctionCtx *pCtx, int32_t index) {
       } else {
         *(int32_t *)pCtx->aOutputBuf = *(int32_t *)pData - (int32_t)pCtx->param[1].i64Key;
         pCtx->param[1].i64Key = *(int32_t *)pData;
-        *(int64_t *)pCtx->ptsOutputBuf = pCtx->ptsList[index];
+        *(int64_t *)pCtx->ptsOutputBuf = GET_TS_DATA(pCtx, index);
       }
       break;
     };
@@ -3390,7 +3227,7 @@ static void arithmetic_function(SQLFunctionCtx *pCtx) {
   GET_RES_INFO(pCtx)->numOfRes += pCtx->size;
   SArithmeticSupport *sas = (SArithmeticSupport *)pCtx->param[1].pz;
   
-  tExprTreeCalcTraverse(sas->pArithExpr->pExpr, pCtx->size, pCtx->aOutputBuf, sas, pCtx->order, getArithColumnData);
+  arithmeticTreeTraverse(sas->pArithExpr->pExpr, pCtx->size, pCtx->aOutputBuf, sas, pCtx->order, getArithColumnData);
   
   pCtx->aOutputBuf += pCtx->outputBytes * pCtx->size;
   pCtx->param[1].pz = NULL;
@@ -3401,7 +3238,7 @@ static void arithmetic_function_f(SQLFunctionCtx *pCtx, int32_t index) {
   SArithmeticSupport *sas = (SArithmeticSupport *)pCtx->param[1].pz;
   
   sas->offset = index;
-  tExprTreeCalcTraverse(sas->pArithExpr->pExpr, 1, pCtx->aOutputBuf, sas, pCtx->order, getArithColumnData);
+  arithmeticTreeTraverse(sas->pArithExpr->pExpr, 1, pCtx->aOutputBuf, sas, pCtx->order, getArithColumnData);
   
   pCtx->aOutputBuf += pCtx->outputBytes;
 }
@@ -3432,7 +3269,7 @@ static bool spread_function_setup(SQLFunctionCtx *pCtx) {
   SSpreadInfo *pInfo = GET_ROWCELL_INTERBUF(GET_RES_INFO(pCtx));
   
   // this is the server-side setup function in client-side, the secondary merge do not need this procedure
-  if (pCtx->currentStage == SECONDARY_STAGE_MERGE) {
+  if (pCtx->currentStage == MERGE_STAGE) {
     pCtx->param[0].dKey = DBL_MAX;
     pCtx->param[3].dKey = -DBL_MAX;
   } else {
@@ -3481,7 +3318,7 @@ static void spread_function(SQLFunctionCtx *pCtx) {
     goto _spread_over;
   }
   
-  void *pData = GET_INPUT_CHAR(pCtx);
+  void *pData = GET_INPUT_DATA_LIST(pCtx);
   numOfElems = 0;
   
   if (pCtx->inputType == TSDB_DATA_TYPE_TINYINT) {
@@ -3517,7 +3354,7 @@ static void spread_function(SQLFunctionCtx *pCtx) {
 }
 
 static void spread_function_f(SQLFunctionCtx *pCtx, int32_t index) {
-  void *pData = GET_INPUT_CHAR_INDEX(pCtx, index);
+  void *pData = GET_INPUT_DATA(pCtx, index);
   if (pCtx->hasNull && isNull(pData, pCtx->inputType)) {
     return;
   }
@@ -3559,45 +3396,12 @@ static void spread_function_f(SQLFunctionCtx *pCtx, int32_t index) {
   }
 }
 
-void spread_func_merge(SQLFunctionCtx *pCtx) {
-  SResultRowCellInfo *pResInfo = GET_RES_INFO(pCtx);
-  assert(pCtx->stableQuery);
-  
-  SSpreadInfo *pResData = GET_ROWCELL_INTERBUF(pResInfo);
-  
-  int32_t notNullElems = 0;
-  for (int32_t i = 0; i < pCtx->size; ++i) {
-    SSpreadInfo *input = (SSpreadInfo *)GET_INPUT_CHAR_INDEX(pCtx, i);
-    
-    /* no assign tag, the value is null */
-    if (input->hasResult != DATA_SET_FLAG) {
-      continue;
-    }
-    
-    if (pResData->min > input->min) {
-      pResData->min = input->min;
-    }
-    
-    if (pResData->max < input->max) {
-      pResData->max = input->max;
-    }
-    
-    pResData->hasResult = DATA_SET_FLAG;
-    notNullElems++;
-  }
-  
-  if (notNullElems > 0) {
-    memcpy(pCtx->aOutputBuf, GET_ROWCELL_INTERBUF(pResInfo), sizeof(SSpreadInfo));
-    pResInfo->hasResult = DATA_SET_FLAG;
-  }
-}
-
 /*
  * here we set the result value back to the intermediate buffer, to apply the finalize the function
  * the final result is generated in spread_function_finalizer
  */
 void spread_func_sec_merge(SQLFunctionCtx *pCtx) {
-  SSpreadInfo *pData = (SSpreadInfo *)GET_INPUT_CHAR(pCtx);
+  SSpreadInfo *pData = (SSpreadInfo *)GET_INPUT_DATA_LIST(pCtx);
   if (pData->hasResult != DATA_SET_FLAG) {
     return;
   }
@@ -3620,7 +3424,7 @@ void spread_function_finalizer(SQLFunctionCtx *pCtx) {
    */
   SResultRowCellInfo *pResInfo = GET_RES_INFO(pCtx);
   
-  if (pCtx->currentStage == SECONDARY_STAGE_MERGE) {
+  if (pCtx->currentStage == MERGE_STAGE) {
     assert(pCtx->inputType == TSDB_DATA_TYPE_BINARY);
     
     if (pResInfo->hasResult != DATA_SET_FLAG) {
@@ -3675,26 +3479,25 @@ static double twa_get_area(SPoint1 s, SPoint1 e) {
   return val;
 }
 
-static int32_t twa_function_impl(SQLFunctionCtx* pCtx, int32_t tsIndex, int32_t index, int32_t size) {
+static int32_t twa_function_impl(SQLFunctionCtx* pCtx, int32_t index, int32_t size) {
   int32_t notNullElems = 0;
-  TSKEY *primaryKey = pCtx->ptsList;
-  
   SResultRowCellInfo *pResInfo = GET_RES_INFO(pCtx);
 
   STwaInfo *pInfo = GET_ROWCELL_INTERBUF(pResInfo);
+  TSKEY    *tsList = GET_TS_LIST(pCtx);
 
   int32_t i = index;
   int32_t step = GET_FORWARD_DIRECTION_FACTOR(pCtx->order);
   SPoint1* last = &pInfo->p;
 
   if (pCtx->start.key != INT64_MIN) {
-    assert((pCtx->start.key < primaryKey[tsIndex + i] && pCtx->order == TSDB_ORDER_ASC) ||
-               (pCtx->start.key > primaryKey[tsIndex + i] && pCtx->order == TSDB_ORDER_DESC));
+    assert((pCtx->start.key < tsList[i] && pCtx->order == TSDB_ORDER_ASC) ||
+               (pCtx->start.key > tsList[i] && pCtx->order == TSDB_ORDER_DESC));
 
     assert(last->key == INT64_MIN);
 
-    last->key = primaryKey[tsIndex + i];
-    GET_TYPED_DATA(last->val, double, pCtx->inputType, GET_INPUT_CHAR_INDEX(pCtx, index));
+    last->key = tsList[i];
+    GET_TYPED_DATA(last->val, double, pCtx->inputType, GET_INPUT_DATA(pCtx, index));
 
     pInfo->dOutput += twa_get_area(pCtx->start, *last);
 
@@ -3703,8 +3506,8 @@ static int32_t twa_function_impl(SQLFunctionCtx* pCtx, int32_t tsIndex, int32_t
     notNullElems++;
     i += step;
   } else if (pInfo->p.key == INT64_MIN) {
-    last->key = primaryKey[tsIndex + i];
-    GET_TYPED_DATA(last->val, double, pCtx->inputType, GET_INPUT_CHAR_INDEX(pCtx, index));
+    last->key = tsList[i];
+    GET_TYPED_DATA(last->val, double, pCtx->inputType, GET_INPUT_DATA(pCtx, index));
 
     pInfo->hasResult = DATA_SET_FLAG;
     pInfo->win.skey = last->key;
@@ -3715,78 +3518,78 @@ static int32_t twa_function_impl(SQLFunctionCtx* pCtx, int32_t tsIndex, int32_t
   // calculate the value of
   switch(pCtx->inputType) {
     case TSDB_DATA_TYPE_TINYINT: {
-      int8_t *val = (int8_t*) GET_INPUT_CHAR_INDEX(pCtx, 0);
+      int8_t *val = (int8_t*) GET_INPUT_DATA(pCtx, 0);
       for (; i < size && i >= 0; i += step) {
         if (pCtx->hasNull && isNull((const char*) &val[i], pCtx->inputType)) {
           continue;
         }
 
-        SPoint1 st = {.key = primaryKey[i + tsIndex], .val = val[i]};
+        SPoint1 st = {.key = tsList[i], .val = val[i]};
         pInfo->dOutput += twa_get_area(pInfo->p, st);
         pInfo->p = st;
       }
       break;
     }
     case TSDB_DATA_TYPE_SMALLINT: {
-      int16_t *val = (int16_t*) GET_INPUT_CHAR_INDEX(pCtx, 0);
+      int16_t *val = (int16_t*) GET_INPUT_DATA(pCtx, 0);
       for (; i < size && i >= 0; i += step) {
         if (pCtx->hasNull && isNull((const char*) &val[i], pCtx->inputType)) {
           continue;
         }
 
-        SPoint1 st = {.key = primaryKey[i + tsIndex], .val = val[i]};
+        SPoint1 st = {.key = tsList[i], .val = val[i]};
         pInfo->dOutput += twa_get_area(pInfo->p, st);
         pInfo->p = st;
       }
       break;
     }
     case TSDB_DATA_TYPE_INT: {
-      int32_t *val = (int32_t*) GET_INPUT_CHAR_INDEX(pCtx, 0);
+      int32_t *val = (int32_t*) GET_INPUT_DATA(pCtx, 0);
       for (; i < size && i >= 0; i += step) {
         if (pCtx->hasNull && isNull((const char*) &val[i], pCtx->inputType)) {
           continue;
         }
 
-        SPoint1 st = {.key = primaryKey[i + tsIndex], .val = val[i]};
+        SPoint1 st = {.key = tsList[i], .val = val[i]};
         pInfo->dOutput += twa_get_area(pInfo->p, st);
         pInfo->p = st;
       }
       break;
     }
     case TSDB_DATA_TYPE_BIGINT: {
-      int64_t *val = (int64_t*) GET_INPUT_CHAR_INDEX(pCtx, 0);
+      int64_t *val = (int64_t*) GET_INPUT_DATA(pCtx, 0);
       for (; i < size && i >= 0; i += step) {
         if (pCtx->hasNull && isNull((const char*) &val[i], pCtx->inputType)) {
           continue;
         }
 
-        SPoint1 st = {.key = primaryKey[i + tsIndex], .val = (double) val[i]};
+        SPoint1 st = {.key = tsList[i], .val = (double) val[i]};
         pInfo->dOutput += twa_get_area(pInfo->p, st);
         pInfo->p = st;
       }
       break;
     }
     case TSDB_DATA_TYPE_FLOAT: {
-      float *val = (float*) GET_INPUT_CHAR_INDEX(pCtx, 0);
+      float *val = (float*) GET_INPUT_DATA(pCtx, 0);
       for (; i < size && i >= 0; i += step) {
         if (pCtx->hasNull && isNull((const char*) &val[i], pCtx->inputType)) {
           continue;
         }
 
-        SPoint1 st = {.key = primaryKey[i + tsIndex], .val = val[i]};
+        SPoint1 st = {.key = tsList[i], .val = val[i]};
         pInfo->dOutput += twa_get_area(pInfo->p, st);
         pInfo->p = st;
       }
       break;
     }
     case TSDB_DATA_TYPE_DOUBLE: {
-      double *val = (double*) GET_INPUT_CHAR_INDEX(pCtx, 0);
+      double *val = (double*) GET_INPUT_DATA(pCtx, 0);
       for (; i < size && i >= 0; i += step) {
         if (pCtx->hasNull && isNull((const char*) &val[i], pCtx->inputType)) {
           continue;
         }
 
-        SPoint1 st = {.key = primaryKey[i + tsIndex], .val = val[i]};
+        SPoint1 st = {.key = tsList[i], .val = val[i]};
         pInfo->dOutput += twa_get_area(pInfo->p, st);
         pInfo->p = st;
       }
@@ -3806,7 +3609,7 @@ static int32_t twa_function_impl(SQLFunctionCtx* pCtx, int32_t tsIndex, int32_t
 }
 
 static void twa_function(SQLFunctionCtx *pCtx) {
-  void *data = GET_INPUT_CHAR(pCtx);
+  void *data = GET_INPUT_DATA_LIST(pCtx);
 
   SResultRowCellInfo *pResInfo = GET_RES_INFO(pCtx);
   STwaInfo *pInfo = GET_ROWCELL_INTERBUF(pResInfo);
@@ -3814,11 +3617,15 @@ static void twa_function(SQLFunctionCtx *pCtx) {
   // skip null value
   int32_t step = GET_FORWARD_DIRECTION_FACTOR(pCtx->order);
   int32_t i = (pCtx->order == TSDB_ORDER_ASC)? 0:(pCtx->size - 1);
-  while (pCtx->hasNull && i < pCtx->size && isNull((char *)data + pCtx->inputBytes * i, pCtx->inputType)) {
+  while (pCtx->hasNull && i < pCtx->size && i >= 0 && isNull((char *)data + pCtx->inputBytes * i, pCtx->inputType)) {
     i += step;
   }
 
-  int32_t notNullElems = twa_function_impl(pCtx, pCtx->startOffset, i, pCtx->size);
+  int32_t notNullElems = 0;
+  if (i >= 0 && i < pCtx->size) {
+    notNullElems = twa_function_impl(pCtx, i, pCtx->size);
+  }
+
   SET_VAL(pCtx, notNullElems, 1);
   
   if (notNullElems > 0) {
@@ -3830,135 +3637,15 @@ static void twa_function(SQLFunctionCtx *pCtx) {
   }
 }
 
-//TODO refactor
 static void twa_function_f(SQLFunctionCtx *pCtx, int32_t index) {
-  void *pData = GET_INPUT_CHAR_INDEX(pCtx, index);
+  void *pData = GET_INPUT_DATA(pCtx, index);
   if (pCtx->hasNull && isNull(pData, pCtx->inputType)) {
     return;
   }
 
-  int32_t notNullElems = 0;
-  TSKEY *primaryKey = pCtx->ptsList;
-
+  int32_t notNullElems = twa_function_impl(pCtx, index, 1);
   SResultRowCellInfo *pResInfo = GET_RES_INFO(pCtx);
 
-  STwaInfo *pInfo = GET_ROWCELL_INTERBUF(pResInfo);
-  int32_t i = pCtx->startOffset;
-  int32_t size = pCtx->size;
-
-  if (pCtx->start.key != INT64_MIN) {
-    assert(pInfo->p.key == INT64_MIN);
-
-    pInfo->p.key = primaryKey[index];
-    GET_TYPED_DATA(pInfo->p.val, double, pCtx->inputType, GET_INPUT_CHAR_INDEX(pCtx, index));
-
-    pInfo->dOutput += twa_get_area(pCtx->start, pInfo->p);
-
-    pInfo->hasResult = DATA_SET_FLAG;
-    pInfo->win.skey = pCtx->start.key;
-    notNullElems++;
-    i += 1;
-  } else if (pInfo->p.key == INT64_MIN) {
-    pInfo->p.key = primaryKey[index];
-    GET_TYPED_DATA(pInfo->p.val, double, pCtx->inputType, GET_INPUT_CHAR_INDEX(pCtx, index));
-
-    pInfo->hasResult = DATA_SET_FLAG;
-    pInfo->win.skey = pInfo->p.key;
-    notNullElems++;
-    i += 1;
-  }
-
-  // calculate the value of
-  switch(pCtx->inputType) {
-    case TSDB_DATA_TYPE_TINYINT: {
-      int8_t *val = (int8_t*) GET_INPUT_CHAR_INDEX(pCtx, index);
-      for (; i < size; i++) {
-        if (pCtx->hasNull && isNull((const char*) &val[i], pCtx->inputType)) {
-          continue;
-        }
-
-        SPoint1 st = {.key = primaryKey[i + index], .val = val[i]};
-        pInfo->dOutput += twa_get_area(pInfo->p, st);
-        pInfo->p = st;
-      }
-      break;
-    }
-    case TSDB_DATA_TYPE_SMALLINT: {
-      int16_t *val = (int16_t*) GET_INPUT_CHAR_INDEX(pCtx, index);
-      for (; i < size; i++) {
-        if (pCtx->hasNull && isNull((const char*) &val[i], pCtx->inputType)) {
-          continue;
-        }
-
-        SPoint1 st = {.key = primaryKey[i + index], .val = val[i]};
-        pInfo->dOutput += twa_get_area(pInfo->p, st);
-        pInfo->p = st;
-      }
-      break;
-    }
-    case TSDB_DATA_TYPE_INT: {
-      int32_t *val = (int32_t*) GET_INPUT_CHAR_INDEX(pCtx, index);
-      for (; i < size; i++) {
-        if (pCtx->hasNull && isNull((const char*) &val[i], pCtx->inputType)) {
-          continue;
-        }
-
-        SPoint1 st = {.key = primaryKey[i + index], .val = val[i]};
-        pInfo->dOutput += twa_get_area(pInfo->p, st);
-        pInfo->p = st;
-      }
-      break;
-    }
-    case TSDB_DATA_TYPE_BIGINT: {
-      int64_t *val = (int64_t*) GET_INPUT_CHAR_INDEX(pCtx, index);
-      for (; i < size; i++) {
-        if (pCtx->hasNull && isNull((const char*) &val[i], pCtx->inputType)) {
-          continue;
-        }
-
-        SPoint1 st = {.key = primaryKey[i + index], .val = (double) val[i]};
-        pInfo->dOutput += twa_get_area(pInfo->p, st);
-        pInfo->p = st;
-      }
-      break;
-    }
-    case TSDB_DATA_TYPE_FLOAT: {
-      float *val = (float*) GET_INPUT_CHAR_INDEX(pCtx, index);
-      for (; i < size; i++) {
-        if (pCtx->hasNull && isNull((const char*) &val[i], pCtx->inputType)) {
-          continue;
-        }
-
-        SPoint1 st = {.key = primaryKey[i + index], .val = val[i]};
-        pInfo->dOutput += twa_get_area(pInfo->p, st);//((val[i] + pInfo->p.val) / 2) * (primaryKey[i + index] - pInfo->p.key);
-        pInfo->p = st;
-      }
-      break;
-    }
-    case TSDB_DATA_TYPE_DOUBLE: {
-      double *val = (double*) GET_INPUT_CHAR_INDEX(pCtx, index);
-      for (; i < size; i++) {
-        if (pCtx->hasNull && isNull((const char*) &val[i], pCtx->inputType)) {
-          continue;
-        }
-
-        SPoint1 st = {.key = primaryKey[i + index], .val = val[i]};
-        pInfo->dOutput += twa_get_area(pInfo->p, st);//((val[i] + pInfo->p.val) / 2) * (primaryKey[i + index] - pInfo->p.key);
-        pInfo->p = st;
-      }
-      break;
-    }
-    default: assert(0);
-  }
-
-  // the last interpolated time window value
-  if (pCtx->end.key != INT64_MIN) {
-    pInfo->dOutput  += twa_get_area(pInfo->p, pCtx->end);//((pInfo->p.val + pCtx->end.val) / 2) * (pCtx->end.key - pInfo->p.key);
-    pInfo->p = pCtx->end;
-  }
-
-  pInfo->win.ekey  = pInfo->p.key;
-
   SET_VAL(pCtx, notNullElems, 1);
 
   if (notNullElems > 0) {
@@ -3970,34 +3657,6 @@ static void twa_function_f(SQLFunctionCtx *pCtx, int32_t index) {
   }
 }
 
-static void twa_func_merge(SQLFunctionCtx *pCtx) {
-  assert(pCtx->stableQuery);
-  
-  STwaInfo *pBuf = (STwaInfo *)pCtx->aOutputBuf;
-  char *    indicator = pCtx->aInputElemBuf;
-  
-  int32_t numOfNotNull = 0;
-  for (int32_t i = 0; i < pCtx->size; ++i, indicator += sizeof(STwaInfo)) {
-    STwaInfo *pInput = (STwaInfo*) indicator;
-    
-    if (pInput->hasResult != DATA_SET_FLAG) {
-      continue;
-    }
-    
-    numOfNotNull++;
-    pBuf->dOutput += pInput->dOutput;
-
-    pBuf->win = pInput->win;
-    pBuf->p   = pInput->p;
-  }
-  
-  SET_VAL(pCtx, numOfNotNull, 1);
-  
-  if (numOfNotNull > 0) {
-    pBuf->hasResult = DATA_SET_FLAG;
-  }
-}
-
 /*
  * To copy the input to interResBuf to avoid the input buffer space be over writen
  * by next input data. The TWA function only applies to each table, so no merge procedure
@@ -4040,8 +3699,10 @@ static void interp_function(SQLFunctionCtx *pCtx) {
   SResultRowCellInfo *pResInfo = GET_RES_INFO(pCtx);
   SInterpInfoDetail* pInfo = GET_ROWCELL_INTERBUF(pResInfo);
 
+  assert(pCtx->startOffset == 0);
+
   if (pCtx->size == 1) {
-    char *pData = GET_INPUT_CHAR(pCtx);
+    char *pData = GET_INPUT_DATA_LIST(pCtx);
     assignVal(pCtx->aOutputBuf, pData, pCtx->inputBytes, pCtx->inputType);
   } else {
     /*
@@ -4067,13 +3728,13 @@ static void interp_function(SQLFunctionCtx *pCtx) {
       } else if (pInfo->type == TSDB_FILL_SET_VALUE) {
         tVariantDump(&pCtx->param[1], pCtx->aOutputBuf, pCtx->inputType, true);
       } else if (pInfo->type == TSDB_FILL_PREV) {
-        char *data = GET_INPUT_CHAR_INDEX(pCtx, 0);
+        char *data = GET_INPUT_DATA(pCtx, 0);
         assignVal(pCtx->aOutputBuf, data, pCtx->outputBytes, pCtx->outputType);
   
         SET_VAL(pCtx, pCtx->size, 1);
       } else if (pInfo->type == TSDB_FILL_LINEAR) {
-        char *data1 = GET_INPUT_CHAR_INDEX(pCtx, 0);
-        char *data2 = GET_INPUT_CHAR_INDEX(pCtx, 1);
+        char *data1 = GET_INPUT_DATA(pCtx, 0);
+        char *data2 = GET_INPUT_DATA(pCtx, 1);
       
         TSKEY key1 = pCtx->ptsList[0];
         TSKEY key2 = pCtx->ptsList[1];
@@ -4135,14 +3796,14 @@ static void ts_comp_function(SQLFunctionCtx *pCtx) {
   SResultRowCellInfo *pResInfo = GET_RES_INFO(pCtx);
   STSBuf *     pTSbuf = ((STSCompInfo *)(GET_ROWCELL_INTERBUF(pResInfo)))->pTSBuf;
   
-  const char *input = GET_INPUT_CHAR(pCtx);
+  const char *input = GET_INPUT_DATA_LIST(pCtx);
   
   // primary ts must be existed, so no need to check its existance
   if (pCtx->order == TSDB_ORDER_ASC) {
     tsBufAppend(pTSbuf, (int32_t)pCtx->param[0].i64Key, &pCtx->tag, input, pCtx->size * TSDB_KEYSIZE);
   } else {
     for (int32_t i = pCtx->size - 1; i >= 0; --i) {
-      char *d = GET_INPUT_CHAR_INDEX(pCtx, i);
+      char *d = GET_INPUT_DATA(pCtx, i);
       tsBufAppend(pTSbuf, (int32_t)pCtx->param[0].i64Key, &pCtx->tag, d, (int32_t)TSDB_KEYSIZE);
     }
   }
@@ -4152,7 +3813,7 @@ static void ts_comp_function(SQLFunctionCtx *pCtx) {
 }
 
 static void ts_comp_function_f(SQLFunctionCtx *pCtx, int32_t index) {
-  void *pData = GET_INPUT_CHAR_INDEX(pCtx, index);
+  void *pData = GET_INPUT_DATA(pCtx, index);
   if (pCtx->hasNull && isNull(pData, pCtx->inputType)) {
     return;
   }
@@ -4214,7 +3875,6 @@ static double do_calc_rate(const SRateInfo* pRateInfo) {
   return resultVal;
 }
 
-
 static bool rate_function_setup(SQLFunctionCtx *pCtx) {
   if (!function_setup(pCtx)) {
     return false;
@@ -4236,18 +3896,17 @@ static bool rate_function_setup(SQLFunctionCtx *pCtx) {
   return true;
 }
 
-
 static void rate_function(SQLFunctionCtx *pCtx) {
+  SResultRowCellInfo *pResInfo = GET_RES_INFO(pCtx);
   
   int32_t    notNullElems = 0;
-  SResultRowCellInfo *pResInfo     = GET_RES_INFO(pCtx);
   SRateInfo *pRateInfo = (SRateInfo *)GET_ROWCELL_INTERBUF(pResInfo);
-  TSKEY       *primaryKey   = pCtx->ptsList;
+  TSKEY     *primaryKey = GET_TS_LIST(pCtx);
   
   tscDebug("%p rate_function() size:%d, hasNull:%d", pCtx, pCtx->size, pCtx->hasNull);
   
   for (int32_t i = 0; i < pCtx->size; ++i) {
-    char *pData = GET_INPUT_CHAR_INDEX(pCtx, i);
+    char *pData = GET_INPUT_DATA(pCtx, i);
     if (pCtx->hasNull && isNull(pData, pCtx->inputType)) {
       tscDebug("%p rate_function() index of null data:%d", pCtx, i);
       continue;
@@ -4295,7 +3954,7 @@ static void rate_function(SQLFunctionCtx *pCtx) {
 }
 
 static void rate_function_f(SQLFunctionCtx *pCtx, int32_t index) {
-  void *pData = GET_INPUT_CHAR_INDEX(pCtx, index);
+  void *pData = GET_INPUT_DATA(pCtx, index);
   if (pCtx->hasNull && isNull(pData, pCtx->inputType)) {
     return;
   }
@@ -4303,7 +3962,7 @@ static void rate_function_f(SQLFunctionCtx *pCtx, int32_t index) {
   // NOTE: keep the intermediate result into the interResultBuf
   SResultRowCellInfo *pResInfo   = GET_RES_INFO(pCtx);
   SRateInfo   *pRateInfo  = (SRateInfo *)GET_ROWCELL_INTERBUF(pResInfo);
-  TSKEY       *primaryKey = pCtx->ptsList;
+  TSKEY     *primaryKey = GET_TS_LIST(pCtx);
 
   int64_t v = 0;
   GET_TYPED_DATA(v, int64_t, pCtx->inputType, pData);
@@ -4336,42 +3995,6 @@ static void rate_function_f(SQLFunctionCtx *pCtx, int32_t index) {
   }
 }
 
-
-
-static void rate_func_merge(SQLFunctionCtx *pCtx) {
-  assert(pCtx->stableQuery);
-  
-  tscDebug("rate_func_merge() size:%d", pCtx->size);
-  
-  SRateInfo *pBuf      = (SRateInfo *)pCtx->aOutputBuf;
-  char      *indicator = pCtx->aInputElemBuf;
-  
-  assert(1 == pCtx->size);
-  
-  int32_t numOfNotNull = 0;
-  for (int32_t i = 0; i < pCtx->size; ++i, indicator += sizeof(SRateInfo)) {
-    SRateInfo *pInput = (SRateInfo *)indicator;
-    if (DATA_SET_FLAG != pInput->hasResult) {
-      continue;
-    }
-    
-    numOfNotNull++;
-    memcpy(pBuf, pInput, sizeof(SRateInfo));
-    tscDebug("%p rate_func_merge() isIRate:%d firstKey:%" PRId64 " lastKey:%" PRId64 " firstValue:%" PRId64 " lastValue:%" PRId64 " CorrectionValue:%" PRId64,
-           pCtx, pInput->isIRate, pInput->firstKey, pInput->lastKey, pInput->firstValue, pInput->lastValue, pInput->CorrectionValue);
-  }
-  
-  SET_VAL(pCtx, numOfNotNull, 1);
-  
-  if (numOfNotNull > 0) {
-    pBuf->hasResult = DATA_SET_FLAG;
-  }
-  
-  return;
-}
-
-
-
 static void rate_func_copy(SQLFunctionCtx *pCtx) {
   assert(pCtx->inputType == TSDB_DATA_TYPE_BINARY);
   
@@ -4380,12 +4003,10 @@ static void rate_func_copy(SQLFunctionCtx *pCtx) {
   pResInfo->hasResult = ((SRateInfo*)pCtx->aInputElemBuf)->hasResult;
   
   SRateInfo* pRateInfo = (SRateInfo*)pCtx->aInputElemBuf;
-  tscDebug("%p rate_func_second_merge() firstKey:%" PRId64 " lastKey:%" PRId64 " firstValue:%" PRId64 " lastValue:%" PRId64 " CorrectionValue:%" PRId64 " hasResult:%d",
+  tscDebug("%p rate_func_merge() firstKey:%" PRId64 " lastKey:%" PRId64 " firstValue:%" PRId64 " lastValue:%" PRId64 " CorrectionValue:%" PRId64 " hasResult:%d",
          pCtx, pRateInfo->firstKey, pRateInfo->lastKey, pRateInfo->firstValue, pRateInfo->lastValue, pRateInfo->CorrectionValue, pRateInfo->hasResult);
 }
 
-
-
 static void rate_finalizer(SQLFunctionCtx *pCtx) {
   SResultRowCellInfo *pResInfo  = GET_RES_INFO(pCtx);
   SRateInfo   *pRateInfo = (SRateInfo *)GET_ROWCELL_INTERBUF(pResInfo);
@@ -4409,13 +4030,12 @@ static void rate_finalizer(SQLFunctionCtx *pCtx) {
   doFinalizer(pCtx);
 }
 
-
 static void irate_function(SQLFunctionCtx *pCtx) {
   
   int32_t       notNullElems = 0;
   SResultRowCellInfo  *pResInfo     = GET_RES_INFO(pCtx);
   SRateInfo   *pRateInfo    = (SRateInfo *)GET_ROWCELL_INTERBUF(pResInfo);
-  TSKEY        *primaryKey   = pCtx->ptsList;
+  TSKEY     *primaryKey = GET_TS_LIST(pCtx);
 
   tscDebug("%p irate_function() size:%d, hasNull:%d", pCtx, pCtx->size, pCtx->hasNull);
   
@@ -4424,7 +4044,7 @@ static void irate_function(SQLFunctionCtx *pCtx) {
   }
   
   for (int32_t i = pCtx->size - 1; i >= 0; --i) {
-    char *pData = GET_INPUT_CHAR_INDEX(pCtx, i);
+    char *pData = GET_INPUT_DATA(pCtx, i);
     if (pCtx->hasNull && isNull(pData, pCtx->inputType)) {
       tscDebug("%p irate_function() index of null data:%d", pCtx, i);
       continue;
@@ -4467,7 +4087,7 @@ static void irate_function(SQLFunctionCtx *pCtx) {
 }
 
 static void irate_function_f(SQLFunctionCtx *pCtx, int32_t index) {
-  void *pData = GET_INPUT_CHAR_INDEX(pCtx, index);
+  void *pData = GET_INPUT_DATA(pCtx, index);
   if (pCtx->hasNull && isNull(pData, pCtx->inputType)) {
     return;
   }
@@ -4475,7 +4095,7 @@ static void irate_function_f(SQLFunctionCtx *pCtx, int32_t index) {
   // NOTE: keep the intermediate result into the interResultBuf
   SResultRowCellInfo  *pResInfo = GET_RES_INFO(pCtx);
   SRateInfo *pRateInfo  = (SRateInfo *)GET_ROWCELL_INTERBUF(pResInfo);
-  TSKEY        *primaryKey = pCtx->ptsList;
+  TSKEY     *primaryKey = GET_TS_LIST(pCtx);
   
   int64_t v = 0;
   GET_TYPED_DATA(v, int64_t, pCtx->inputType, pData);
@@ -4505,7 +4125,7 @@ static void do_sumrate_merge(SQLFunctionCtx *pCtx) {
   assert(pCtx->stableQuery);
   
   SRateInfo *pRateInfo = (SRateInfo *)GET_ROWCELL_INTERBUF(pResInfo);
-  char *    input = GET_INPUT_CHAR(pCtx);
+  char *    input = GET_INPUT_DATA_LIST(pCtx);
   
   for (int32_t i = 0; i < pCtx->size; ++i, input += pCtx->inputBytes) {
     SRateInfo *pInput = (SRateInfo *)input;
@@ -4537,11 +4157,6 @@ static void sumrate_func_merge(SQLFunctionCtx *pCtx) {
   do_sumrate_merge(pCtx);
 }
 
-static void sumrate_func_second_merge(SQLFunctionCtx *pCtx) {
-  tscDebug("%p sumrate_func_second_merge() process ...", pCtx);
-  do_sumrate_merge(pCtx);
-}
-
 static void sumrate_finalizer(SQLFunctionCtx *pCtx) {
   SResultRowCellInfo *pResInfo  = GET_RES_INFO(pCtx);
   SRateInfo   *pRateInfo = (SRateInfo *)GET_ROWCELL_INTERBUF(pResInfo);
@@ -4605,7 +4220,6 @@ SQLAggFuncElem aAggs[] = {{
                               no_next_step,
                               doFinalizer,
                               count_func_merge,
-                              count_func_merge,
                               count_load_data_info,
                           },
                           {
@@ -4620,7 +4234,6 @@ SQLAggFuncElem aAggs[] = {{
                               no_next_step,
                               function_finalizer,
                               sum_func_merge,
-                              sum_func_second_merge,
                               statisRequired,
                           },
                           {
@@ -4635,7 +4248,6 @@ SQLAggFuncElem aAggs[] = {{
                               no_next_step,
                               avg_finalizer,
                               avg_func_merge,
-                              avg_func_second_merge,
                               statisRequired,
                           },
                           {
@@ -4650,7 +4262,6 @@ SQLAggFuncElem aAggs[] = {{
                               no_next_step,
                               function_finalizer,
                               min_func_merge,
-                              min_func_second_merge,
                               statisRequired,
                           },
                           {
@@ -4665,7 +4276,6 @@ SQLAggFuncElem aAggs[] = {{
                               no_next_step,
                               function_finalizer,
                               max_func_merge,
-                              max_func_second_merge,
                               statisRequired,
                           },
                           {
@@ -4680,7 +4290,6 @@ SQLAggFuncElem aAggs[] = {{
                               stddev_next_step,
                               stddev_finalizer,
                               noop1,
-                              noop1,
                               dataBlockRequired,
                           },
                           {
@@ -4695,7 +4304,6 @@ SQLAggFuncElem aAggs[] = {{
                               percentile_next_step,
                               percentile_finalizer,
                               noop1,
-                              noop1,
                               dataBlockRequired,
                           },
                           {
@@ -4710,7 +4318,6 @@ SQLAggFuncElem aAggs[] = {{
                               no_next_step,
                               apercentile_finalizer,
                               apercentile_func_merge,
-                              apercentile_func_second_merge,
                               dataBlockRequired,
                           },
                           {
@@ -4725,7 +4332,6 @@ SQLAggFuncElem aAggs[] = {{
                               no_next_step,
                               function_finalizer,
                               noop1,
-                              noop1,
                               firstFuncRequired,
                           },
                           {
@@ -4740,7 +4346,6 @@ SQLAggFuncElem aAggs[] = {{
                               no_next_step,
                               function_finalizer,
                               noop1,
-                              noop1,
                               lastFuncRequired,
                           },
                           {
@@ -4755,8 +4360,7 @@ SQLAggFuncElem aAggs[] = {{
                               noop2,
                               no_next_step,
                               last_row_finalizer,
-                              noop1,
-                              last_dist_func_second_merge,
+                              last_dist_func_merge,
                               dataBlockRequired,
                           },
                           {
@@ -4772,7 +4376,6 @@ SQLAggFuncElem aAggs[] = {{
                               no_next_step,
                               top_bottom_func_finalizer,
                               top_func_merge,
-                              top_func_second_merge,
                               dataBlockRequired,
                           },
                           {
@@ -4788,7 +4391,6 @@ SQLAggFuncElem aAggs[] = {{
                               no_next_step,
                               top_bottom_func_finalizer,
                               bottom_func_merge,
-                              bottom_func_second_merge,
                               dataBlockRequired,
                           },
                           {
@@ -4802,7 +4404,6 @@ SQLAggFuncElem aAggs[] = {{
                               spread_function_f,
                               no_next_step,
                               spread_function_finalizer,
-                              spread_func_merge,
                               spread_func_sec_merge,
                               count_load_data_info,
                           },
@@ -4817,7 +4418,6 @@ SQLAggFuncElem aAggs[] = {{
                               twa_function_f,
                               no_next_step,
                               twa_function_finalizer,
-                              twa_func_merge,
                               twa_function_copy,
                               dataBlockRequired,
                           },
@@ -4833,7 +4433,6 @@ SQLAggFuncElem aAggs[] = {{
                               no_next_step,
                               leastsquares_finalizer,
                               noop1,
-                              noop1,
                               dataBlockRequired,
                           },
                           {
@@ -4848,7 +4447,6 @@ SQLAggFuncElem aAggs[] = {{
                               no_next_step,
                               doFinalizer,
                               copy_function,
-                              copy_function,
                               no_data_info,
                           },
                           {
@@ -4863,7 +4461,6 @@ SQLAggFuncElem aAggs[] = {{
                               no_next_step,
                               doFinalizer,
                               copy_function,
-                              copy_function,
                               dataBlockRequired,
                           },
                           {
@@ -4878,7 +4475,6 @@ SQLAggFuncElem aAggs[] = {{
                               no_next_step,
                               doFinalizer,
                               copy_function,
-                              copy_function,
                               no_data_info,
                           },
                           {
@@ -4893,7 +4489,6 @@ SQLAggFuncElem aAggs[] = {{
                               no_next_step,
                               ts_comp_finalize,
                               copy_function,
-                              copy_function,
                               dataBlockRequired,
                           },
                           {
@@ -4908,7 +4503,6 @@ SQLAggFuncElem aAggs[] = {{
                               no_next_step,
                               doFinalizer,
                               copy_function,
-                              copy_function,
                               no_data_info,
                           },
                           {
@@ -4923,7 +4517,6 @@ SQLAggFuncElem aAggs[] = {{
                               no_next_step,
                               doFinalizer,
                               copy_function,
-                              copy_function,
                               dataBlockRequired,
                           },
                           {
@@ -4938,7 +4531,6 @@ SQLAggFuncElem aAggs[] = {{
                               no_next_step,
                               doFinalizer,
                               copy_function,
-                              copy_function,
                               no_data_info,
                           },
                           {
@@ -4953,7 +4545,6 @@ SQLAggFuncElem aAggs[] = {{
                               no_next_step,
                               doFinalizer,
                               copy_function,
-                              copy_function,
                               dataBlockRequired,
                           },
                           {
@@ -4968,7 +4559,6 @@ SQLAggFuncElem aAggs[] = {{
                               no_next_step,
                               doFinalizer,
                               noop1,
-                              noop1,
                               dataBlockRequired,
                           },
     // distributed version used in two-stage aggregation processes
@@ -4984,7 +4574,6 @@ SQLAggFuncElem aAggs[] = {{
                               no_next_step,
                               function_finalizer,
                               first_dist_func_merge,
-                              first_dist_func_second_merge,
                               firstDistFuncRequired,
                           },
                           {
@@ -4999,7 +4588,6 @@ SQLAggFuncElem aAggs[] = {{
                               no_next_step,
                               function_finalizer,
                               last_dist_func_merge,
-                              last_dist_func_second_merge,
                               lastDistFuncRequired,
                           },
                           {
@@ -5013,7 +4601,6 @@ SQLAggFuncElem aAggs[] = {{
                               do_sum_f,  // todo filter handle
                               no_next_step,
                               doFinalizer,
-                              noop1,
                               copy_function,
                               dataBlockRequired,
                           },
@@ -5028,7 +4615,6 @@ SQLAggFuncElem aAggs[] = {{
                               rate_function_f,
                               no_next_step,
                               rate_finalizer,
-                              rate_func_merge,
                               rate_func_copy,
                               dataBlockRequired,
                           },
@@ -5043,7 +4629,6 @@ SQLAggFuncElem aAggs[] = {{
                               irate_function_f,
                               no_next_step,
                               rate_finalizer,
-                              rate_func_merge,
                               rate_func_copy,
                               dataBlockRequired,
                           },
@@ -5059,7 +4644,6 @@ SQLAggFuncElem aAggs[] = {{
                               no_next_step,
                               sumrate_finalizer,
                               sumrate_func_merge,
-                              sumrate_func_second_merge,
                               dataBlockRequired,
                           },
                           {
@@ -5074,7 +4658,6 @@ SQLAggFuncElem aAggs[] = {{
                               no_next_step,
                               sumrate_finalizer,
                               sumrate_func_merge,
-                              sumrate_func_second_merge,
                               dataBlockRequired,
                           },
                           {
@@ -5089,7 +4672,6 @@ SQLAggFuncElem aAggs[] = {{
                               no_next_step,
                               sumrate_finalizer,
                               sumrate_func_merge,
-                              sumrate_func_second_merge,
                               dataBlockRequired,
                           },
                           {
@@ -5104,7 +4686,6 @@ SQLAggFuncElem aAggs[] = {{
                               no_next_step,
                               sumrate_finalizer,
                               sumrate_func_merge,
-                              sumrate_func_second_merge,
                               dataBlockRequired,
                           },
                           {
@@ -5119,6 +4700,5 @@ SQLAggFuncElem aAggs[] = {{
                               no_next_step,
                               noop1,
                               noop1,
-                              noop1,
                               dataBlockRequired,
                           }};

src/query/src/qSyntaxtreefunction.c → src/query/src/qArithmeticOperator.c

@@ -15,7 +15,7 @@
 
 #include "os.h"
 
-#include "qSyntaxtreefunction.h"
+#include "qArithmeticOperator.h"
 #include "taosdef.h"
 #include "tutil.h"
 
@@ -1234,7 +1234,7 @@ _bi_consumer_fn_t rem_function_arraylist[8][10] = {
 
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////
 
-_bi_consumer_fn_t tGetBiConsumerFn(int32_t leftType, int32_t rightType, int32_t optr) {
+_bi_consumer_fn_t getArithmeticOperatorFn(int32_t leftType, int32_t rightType, int32_t optr) {
   switch (optr) {
     case TSDB_BINARY_OP_ADD:
       return add_function_arraylist[leftType][rightType];

src/query/src/qAst.c

@@ -16,29 +16,18 @@
 #include "os.h"
 
 #include "exception.h"
+#include "qArithmeticOperator.h"
 #include "qAst.h"
-#include "qSyntaxtreefunction.h"
 #include "taosdef.h"
 #include "taosmsg.h"
 #include "tarray.h"
 #include "tbuffer.h"
 #include "tcompare.h"
 #include "tname.h"
+#include "tschemautil.h"
 #include "tsdb.h"
 #include "tskiplist.h"
 #include "tsqlfunction.h"
-#include "tstoken.h"
-#include "tschemautil.h"
-
-typedef struct {
-  char*    v;
-  int32_t  optr;
-} SEndPoint;
-
-typedef struct {
-  SEndPoint* start;
-  SEndPoint* end;
-} SQueryCond;
 
 static uint8_t UNUSED_FUNC isQueryOnPrimaryKey(const char *primaryColumnName, const tExprNode *pLeft, const tExprNode *pRight) {
   if (pLeft->nodeType == TSQL_NODE_COL) {
@@ -53,6 +42,66 @@ static uint8_t UNUSED_FUNC isQueryOnPrimaryKey(const char *primaryColumnName, co
   }
 }
 
+static void reverseCopy(char* dest, const char* src, int16_t type, int32_t numOfRows) {
+  switch(type) {
+    case TSDB_DATA_TYPE_TINYINT: {
+      int8_t* p = (int8_t*) dest;
+      int8_t* pSrc = (int8_t*) src;
+
+      for(int32_t i = 0; i < numOfRows; ++i) {
+        p[i] = pSrc[numOfRows - i - 1];
+      }
+      break;
+    }
+    case TSDB_DATA_TYPE_SMALLINT: {
+      int16_t* p = (int16_t*) dest;
+      int16_t* pSrc = (int16_t*) src;
+
+      for(int32_t i = 0; i < numOfRows; ++i) {
+        p[i] = pSrc[numOfRows - i - 1];
+      }
+      break;
+    }
+    case TSDB_DATA_TYPE_INT: {
+      int32_t* p = (int32_t*) dest;
+      int32_t* pSrc = (int32_t*) src;
+
+      for(int32_t i = 0; i < numOfRows; ++i) {
+        p[i] = pSrc[numOfRows - i - 1];
+      }
+      break;
+    }
+    case TSDB_DATA_TYPE_BIGINT: {
+      int64_t* p = (int64_t*) dest;
+      int64_t* pSrc = (int64_t*) src;
+
+      for(int32_t i = 0; i < numOfRows; ++i) {
+        p[i] = pSrc[numOfRows - i - 1];
+      }
+      break;
+    }
+    case TSDB_DATA_TYPE_FLOAT: {
+      float* p = (float*) dest;
+      float* pSrc = (float*) src;
+
+      for(int32_t i = 0; i < numOfRows; ++i) {
+        p[i] = pSrc[numOfRows - i - 1];
+      }
+      break;
+    }
+    case TSDB_DATA_TYPE_DOUBLE: {
+      double* p = (double*) dest;
+      double* pSrc = (double*) src;
+
+      for(int32_t i = 0; i < numOfRows; ++i) {
+        p[i] = pSrc[numOfRows - i - 1];
+      }
+      break;
+    }
+    default: assert(0);
+  }
+}
+
 void tExprNodeDestroy(tExprNode *pNode, void (*fp)(void *)) {
   if (pNode == NULL) {
     return;
@@ -92,169 +141,25 @@ void tExprTreeDestroy(tExprNode **pExpr, void (*fp)(void *)) {
   *pExpr = NULL;
 }
 
-// todo check for malloc failure
-static int32_t setQueryCond(tQueryInfo *queryColInfo, SQueryCond* pCond) {
-  int32_t optr = queryColInfo->optr;
-  
-  if (optr == TSDB_RELATION_GREATER || optr == TSDB_RELATION_GREATER_EQUAL ||
-      optr == TSDB_RELATION_EQUAL || optr == TSDB_RELATION_NOT_EQUAL) {
-    pCond->start = calloc(1, sizeof(SEndPoint));
-    pCond->start->optr = queryColInfo->optr;
-    pCond->start->v = queryColInfo->q;
-  } else if (optr == TSDB_RELATION_LESS || optr == TSDB_RELATION_LESS_EQUAL) {
-    pCond->end = calloc(1, sizeof(SEndPoint));
-    pCond->end->optr = queryColInfo->optr;
-    pCond->end->v = queryColInfo->q;
-  } else if (optr == TSDB_RELATION_IN || optr == TSDB_RELATION_LIKE) {
-    assert(0);
-  }
-
-  return TSDB_CODE_SUCCESS;
-}
-
-static void tQueryIndexColumn(SSkipList* pSkipList, tQueryInfo* pQueryInfo, SArray* result) {
-  SSkipListIterator* iter = NULL;
-  
-  SQueryCond cond = {0};
-  if (setQueryCond(pQueryInfo, &cond) != TSDB_CODE_SUCCESS) {
-    //todo handle error
-  }
-
-  if (cond.start != NULL) {
-    iter = tSkipListCreateIterFromVal(pSkipList, (char*) cond.start->v, pSkipList->type, TSDB_ORDER_ASC);
-  } else {
-    iter = tSkipListCreateIterFromVal(pSkipList, (char*)(cond.end ? cond.end->v: NULL), pSkipList->type, TSDB_ORDER_DESC);
-  }
-  
-  if (cond.start != NULL) {
-    int32_t optr = cond.start->optr;
-    
-    if (optr == TSDB_RELATION_EQUAL) {   // equals
-      while(tSkipListIterNext(iter)) {
-        SSkipListNode* pNode = tSkipListIterGet(iter);
-
-        int32_t ret = pQueryInfo->compare(SL_GET_NODE_KEY(pSkipList, pNode), cond.start->v);
-        if (ret != 0) {
-          break;
-        }
-
-        STableKeyInfo info = {.pTable = (void*)SL_GET_NODE_DATA(pNode), .lastKey = TSKEY_INITIAL_VAL};
-        taosArrayPush(result, &info);
-      }
-    } else if (optr == TSDB_RELATION_GREATER || optr == TSDB_RELATION_GREATER_EQUAL) { // greater equal
-      bool comp = true;
-      int32_t ret = 0;
-      
-      while(tSkipListIterNext(iter)) {
-        SSkipListNode* pNode = tSkipListIterGet(iter);
-    
-        if (comp) {
-          ret = pQueryInfo->compare(SL_GET_NODE_KEY(pSkipList, pNode), cond.start->v);
-          assert(ret >= 0);
-        }
-        
-        if (ret == 0 && optr == TSDB_RELATION_GREATER) {
-          continue;
-        } else {
-          STableKeyInfo info = {.pTable = (void*)SL_GET_NODE_DATA(pNode), .lastKey = TSKEY_INITIAL_VAL};
-          taosArrayPush(result, &info);
-          comp = false;
-        }
-      }
-    } else if (optr == TSDB_RELATION_NOT_EQUAL) {   // not equal
-      bool comp = true;
-      
-      while(tSkipListIterNext(iter)) {
-        SSkipListNode* pNode = tSkipListIterGet(iter);
-        comp = comp && (pQueryInfo->compare(SL_GET_NODE_KEY(pSkipList, pNode), cond.start->v) == 0);
-        if (comp) {
-          continue;
-        }
-
-        STableKeyInfo info = {.pTable = (void*)SL_GET_NODE_DATA(pNode), .lastKey = TSKEY_INITIAL_VAL};
-        taosArrayPush(result, &info);
-      }
-      
-      tSkipListDestroyIter(iter);
-  
-      comp = true;
-      iter = tSkipListCreateIterFromVal(pSkipList, (char*) cond.start->v, pSkipList->type, TSDB_ORDER_DESC);
-      while(tSkipListIterNext(iter)) {
-        SSkipListNode* pNode = tSkipListIterGet(iter);
-        comp = comp && (pQueryInfo->compare(SL_GET_NODE_KEY(pSkipList, pNode), cond.start->v) == 0);
-      if (comp) {
-          continue;
-        }
-
-        STableKeyInfo info = {.pTable = (void*)SL_GET_NODE_DATA(pNode), .lastKey = TSKEY_INITIAL_VAL};
-        taosArrayPush(result, &info);
-      }
-  
-    } else {
-      assert(0);
-    }
-  } else {
-    int32_t optr = cond.end ? cond.end->optr : TSDB_RELATION_INVALID;
-    if (optr == TSDB_RELATION_LESS || optr == TSDB_RELATION_LESS_EQUAL) {
-      bool    comp = true;
-      int32_t ret = 0;
-
-      while (tSkipListIterNext(iter)) {
-        SSkipListNode *pNode = tSkipListIterGet(iter);
-
-        if (comp) {
-          ret = pQueryInfo->compare(SL_GET_NODE_KEY(pSkipList, pNode), cond.end->v);
-          assert(ret <= 0);
-        }
-
-        if (ret == 0 && optr == TSDB_RELATION_LESS) {
-          continue;
-        } else {
-          STableKeyInfo info = {.pTable = (void *)SL_GET_NODE_DATA(pNode), .lastKey = TSKEY_INITIAL_VAL};
-          taosArrayPush(result, &info);
-          comp = false;  // no need to compare anymore
-        }
-      }
-    } else {
-      assert(pQueryInfo->optr == TSDB_RELATION_ISNULL || pQueryInfo->optr == TSDB_RELATION_NOTNULL);
-
-      while (tSkipListIterNext(iter)) {
-        SSkipListNode *pNode = tSkipListIterGet(iter);
-
-        bool isnull = isNull(SL_GET_NODE_KEY(pSkipList, pNode), pQueryInfo->sch.type);
-        if ((pQueryInfo->optr == TSDB_RELATION_ISNULL && isnull) ||
-            (pQueryInfo->optr == TSDB_RELATION_NOTNULL && (!isnull))) {
-          STableKeyInfo info = {.pTable = (void *)SL_GET_NODE_DATA(pNode), .lastKey = TSKEY_INITIAL_VAL};
-          taosArrayPush(result, &info);
-        }
-      }
-    }
-  }
-
-  free(cond.start);
-  free(cond.end);
-  tSkipListDestroyIter(iter);
-}
-
-static bool filterItem(tExprNode *pExpr, const void *pItem, SExprTraverseSupp *param) {
+bool exprTreeApplayFilter(tExprNode *pExpr, const void *pItem, SExprTraverseSupp *param) {
   tExprNode *pLeft  = pExpr->_node.pLeft;
   tExprNode *pRight = pExpr->_node.pRight;
 
   //non-leaf nodes, recursively traverse the expression tree in the post-root order
   if (pLeft->nodeType == TSQL_NODE_EXPR && pRight->nodeType == TSQL_NODE_EXPR) {
     if (pExpr->_node.optr == TSDB_RELATION_OR) {  // or
-      if (filterItem(pLeft, pItem, param)) {
+      if (exprTreeApplayFilter(pLeft, pItem, param)) {
         return true;
       }
 
       // left child does not satisfy the query condition, try right child
-      return filterItem(pRight, pItem, param);
+      return exprTreeApplayFilter(pRight, pItem, param);
     } else {  // and
-      if (!filterItem(pLeft, pItem, param)) {
+      if (!exprTreeApplayFilter(pLeft, pItem, param)) {
         return false;
       }
 
-      return filterItem(pRight, pItem, param);
+      return exprTreeApplayFilter(pRight, pItem, param);
     }
   }
 
@@ -263,174 +168,7 @@ static bool filterItem(tExprNode *pExpr, const void *pItem, SExprTraverseSupp *p
   return param->nodeFilterFn(pItem, pExpr->_node.info);
 }
 
-static void tSQLBinaryTraverseOnSkipList(tExprNode *pExpr, SArray *pResult, SSkipList *pSkipList, SExprTraverseSupp *param ) {
-  SSkipListIterator* iter = tSkipListCreateIter(pSkipList);
-
-  while (tSkipListIterNext(iter)) {
-    SSkipListNode *pNode = tSkipListIterGet(iter);
-    if (filterItem(pExpr, pNode, param)) {
-      taosArrayPush(pResult, &(SL_GET_NODE_DATA(pNode)));
-    }
-  }
-  tSkipListDestroyIter(iter);
-}
-
-static void tQueryIndexlessColumn(SSkipList* pSkipList, tQueryInfo* pQueryInfo, SArray* res, __result_filter_fn_t filterFp) {
-  SSkipListIterator* iter = tSkipListCreateIter(pSkipList);
-
-  while (tSkipListIterNext(iter)) {
-    bool addToResult = false;
-
-    SSkipListNode *pNode = tSkipListIterGet(iter);
-    char *         pData = SL_GET_NODE_DATA(pNode);
-
-    tstr *name = (tstr*) tsdbGetTableName((void*) pData);
-
-    // todo speed up by using hash
-    if (pQueryInfo->sch.colId == TSDB_TBNAME_COLUMN_INDEX) {
-      if (pQueryInfo->optr == TSDB_RELATION_IN) {
-        addToResult = pQueryInfo->compare(name, pQueryInfo->q);
-      } else if (pQueryInfo->optr == TSDB_RELATION_LIKE) {
-        addToResult = !pQueryInfo->compare(name, pQueryInfo->q);
-      }
-    } else {
-      addToResult = filterFp(pNode, pQueryInfo);
-    }
-
-    if (addToResult) {
-      STableKeyInfo info = {.pTable = (void*)pData, .lastKey = TSKEY_INITIAL_VAL};
-      taosArrayPush(res, &info);
-    }
-  }
-
-  tSkipListDestroyIter(iter);
-}
-
-// post-root order traverse syntax tree
-void tExprTreeTraverse(tExprNode *pExpr, SSkipList *pSkipList, SArray *result, SExprTraverseSupp *param) {
-  if (pExpr == NULL) {
-    return;
-  }
-
-  tExprNode *pLeft  = pExpr->_node.pLeft;
-  tExprNode *pRight = pExpr->_node.pRight;
-
-  // column project
-  if (pLeft->nodeType != TSQL_NODE_EXPR && pRight->nodeType != TSQL_NODE_EXPR) {
-    assert(pLeft->nodeType == TSQL_NODE_COL && (pRight->nodeType == TSQL_NODE_VALUE || pRight->nodeType == TSQL_NODE_DUMMY));
-
-    param->setupInfoFn(pExpr, param->pExtInfo);
-
-    tQueryInfo *pQueryInfo = pExpr->_node.info;
-    if (pQueryInfo->indexed && pQueryInfo->optr != TSDB_RELATION_LIKE) {
-      tQueryIndexColumn(pSkipList, pQueryInfo, result);
-    } else {
-      tQueryIndexlessColumn(pSkipList, pQueryInfo, result, param->nodeFilterFn);
-    }
-
-    return;
-  }
-
-  // The value of hasPK is always 0.
-  uint8_t weight = pLeft->_node.hasPK + pRight->_node.hasPK;
-  assert(weight == 0 && pSkipList != NULL && taosArrayGetSize(result) == 0);
-
-  //apply the hierarchical expression to every node in skiplist for find the qualified nodes
-  tSQLBinaryTraverseOnSkipList(pExpr, result, pSkipList, param);
-
-#if 0
-  /*
-    * (weight == 1 && pExpr->nSQLBinaryOptr == TSDB_RELATION_AND) is handled here
-    *
-    * first, we filter results based on the skiplist index, which is the initial filter stage,
-    * then, we conduct the secondary filter operation based on the result from the initial filter stage.
-    */
-  assert(pExpr->_node.optr == TSDB_RELATION_AND);
-
-  tExprNode *pFirst = NULL;
-  tExprNode *pSecond = NULL;
-  if (pLeft->_node.hasPK == 1) {
-    pFirst = pLeft;
-    pSecond = pRight;
-  } else {
-    pFirst = pRight;
-    pSecond = pLeft;
-  }
-
-  assert(pFirst != pSecond && pFirst != NULL && pSecond != NULL);
-
-  // we filter the result based on the skiplist index in the first place
-  tExprTreeTraverse(pFirst, pSkipList, result, param);
-
-  /*
-    * recursively perform the filter operation based on the initial results,
-    * So, we do not set the skip list index as a parameter
-    */
-  tExprTreeTraverse(pSecond, NULL, result, param);
-#endif
-}
-
-static void reverseCopy(char* dest, const char* src, int16_t type, int32_t numOfRows) {
-  switch(type) {
-    case TSDB_DATA_TYPE_TINYINT: {
-      int8_t* p = (int8_t*) dest;
-      int8_t* pSrc = (int8_t*) src;
-
-      for(int32_t i = 0; i < numOfRows; ++i) {
-        p[i] = pSrc[numOfRows - i - 1];
-      }
-      break;
-    }
-    case TSDB_DATA_TYPE_SMALLINT: {
-      int16_t* p = (int16_t*) dest;
-      int16_t* pSrc = (int16_t*) src;
-
-      for(int32_t i = 0; i < numOfRows; ++i) {
-        p[i] = pSrc[numOfRows - i - 1];
-      }
-      break;
-    }
-    case TSDB_DATA_TYPE_INT: {
-      int32_t* p = (int32_t*) dest;
-      int32_t* pSrc = (int32_t*) src;
-
-      for(int32_t i = 0; i < numOfRows; ++i) {
-        p[i] = pSrc[numOfRows - i - 1];
-      }
-      break;
-    }
-    case TSDB_DATA_TYPE_BIGINT: {
-      int64_t* p = (int64_t*) dest;
-      int64_t* pSrc = (int64_t*) src;
-
-      for(int32_t i = 0; i < numOfRows; ++i) {
-        p[i] = pSrc[numOfRows - i - 1];
-      }
-      break;
-    }
-    case TSDB_DATA_TYPE_FLOAT: {
-      float* p = (float*) dest;
-      float* pSrc = (float*) src;
-
-      for(int32_t i = 0; i < numOfRows; ++i) {
-        p[i] = pSrc[numOfRows - i - 1];
-      }
-      break;
-    }
-    case TSDB_DATA_TYPE_DOUBLE: {
-      double* p = (double*) dest;
-      double* pSrc = (double*) src;
-
-      for(int32_t i = 0; i < numOfRows; ++i) {
-        p[i] = pSrc[numOfRows - i - 1];
-      }
-      break;
-    }
-    default: assert(0);
-  }
-}
-
-void tExprTreeCalcTraverse(tExprNode *pExprs, int32_t numOfRows, char *pOutput, void *param, int32_t order,
+void arithmeticTreeTraverse(tExprNode *pExprs, int32_t numOfRows, char *pOutput, void *param, int32_t order,
                                 char *(*getSourceDataBlock)(void *, const char*, int32_t)) {
   if (pExprs == NULL) {
     return;
@@ -442,7 +180,7 @@ void tExprTreeCalcTraverse(tExprNode *pExprs, int32_t numOfRows, char *pOutput,
   /* the left output has result from the left child syntax tree */
   char *pLeftOutput = (char*)malloc(sizeof(int64_t) * numOfRows);
   if (pLeft->nodeType == TSQL_NODE_EXPR) {
-    tExprTreeCalcTraverse(pLeft, numOfRows, pLeftOutput, param, order, getSourceDataBlock);
+    arithmeticTreeTraverse(pLeft, numOfRows, pLeftOutput, param, order, getSourceDataBlock);
   }
 
   /* the right output has result from the right child syntax tree */
@@ -450,7 +188,7 @@ void tExprTreeCalcTraverse(tExprNode *pExprs, int32_t numOfRows, char *pOutput,
   char *pdata = malloc(sizeof(int64_t) * numOfRows);
 
   if (pRight->nodeType == TSQL_NODE_EXPR) {
-    tExprTreeCalcTraverse(pRight, numOfRows, pRightOutput, param, order, getSourceDataBlock);
+    arithmeticTreeTraverse(pRight, numOfRows, pRightOutput, param, order, getSourceDataBlock);
   }
 
   if (pLeft->nodeType == TSQL_NODE_EXPR) {
@@ -459,11 +197,11 @@ void tExprTreeCalcTraverse(tExprNode *pExprs, int32_t numOfRows, char *pOutput,
        * exprLeft + exprRight
        * the type of returned value of one expression is always double float precious
        */
-      _bi_consumer_fn_t fp = tGetBiConsumerFn(TSDB_DATA_TYPE_DOUBLE, TSDB_DATA_TYPE_DOUBLE, pExprs->_node.optr);
+      _bi_consumer_fn_t fp = getArithmeticOperatorFn(TSDB_DATA_TYPE_DOUBLE, TSDB_DATA_TYPE_DOUBLE, pExprs->_node.optr);
       fp(pLeftOutput, pRightOutput, numOfRows, numOfRows, pOutput, TSDB_ORDER_ASC);
 
     } else if (pRight->nodeType == TSQL_NODE_COL) {  // exprLeft + columnRight
-      _bi_consumer_fn_t fp = tGetBiConsumerFn(TSDB_DATA_TYPE_DOUBLE, pRight->pSchema->type, pExprs->_node.optr);
+      _bi_consumer_fn_t fp = getArithmeticOperatorFn(TSDB_DATA_TYPE_DOUBLE, pRight->pSchema->type, pExprs->_node.optr);
 
       // set input buffer
       char *pInputData = getSourceDataBlock(param, pRight->pSchema->name, pRight->pSchema->colId);
@@ -475,14 +213,14 @@ void tExprTreeCalcTraverse(tExprNode *pExprs, int32_t numOfRows, char *pOutput,
       }
 
     } else if (pRight->nodeType == TSQL_NODE_VALUE) {  // exprLeft + 12
-      _bi_consumer_fn_t fp = tGetBiConsumerFn(TSDB_DATA_TYPE_DOUBLE, pRight->pVal->nType, pExprs->_node.optr);
+      _bi_consumer_fn_t fp = getArithmeticOperatorFn(TSDB_DATA_TYPE_DOUBLE, pRight->pVal->nType, pExprs->_node.optr);
       fp(pLeftOutput, &pRight->pVal->i64Key, numOfRows, 1, pOutput, TSDB_ORDER_ASC);
     }
   } else if (pLeft->nodeType == TSQL_NODE_COL) {
     // column data specified on left-hand-side
     char *pLeftInputData = getSourceDataBlock(param, pLeft->pSchema->name, pLeft->pSchema->colId);
     if (pRight->nodeType == TSQL_NODE_EXPR) {  // columnLeft + expr2
-      _bi_consumer_fn_t fp = tGetBiConsumerFn(pLeft->pSchema->type, TSDB_DATA_TYPE_DOUBLE, pExprs->_node.optr);
+      _bi_consumer_fn_t fp = getArithmeticOperatorFn(pLeft->pSchema->type, TSDB_DATA_TYPE_DOUBLE, pExprs->_node.optr);
 
       if (order == TSDB_ORDER_DESC) {
         reverseCopy(pdata, pLeftInputData, pLeft->pSchema->type, numOfRows);
@@ -494,12 +232,12 @@ void tExprTreeCalcTraverse(tExprNode *pExprs, int32_t numOfRows, char *pOutput,
     } else if (pRight->nodeType == TSQL_NODE_COL) {  // columnLeft + columnRight
       // column data specified on right-hand-side
       char *pRightInputData = getSourceDataBlock(param, pRight->pSchema->name, pRight->pSchema->colId);
-      _bi_consumer_fn_t fp = tGetBiConsumerFn(pLeft->pSchema->type, pRight->pSchema->type, pExprs->_node.optr);
+      _bi_consumer_fn_t fp = getArithmeticOperatorFn(pLeft->pSchema->type, pRight->pSchema->type, pExprs->_node.optr);
 
       // both columns are descending order, do not reverse the source data
       fp(pLeftInputData, pRightInputData, numOfRows, numOfRows, pOutput, order);
     } else if (pRight->nodeType == TSQL_NODE_VALUE) {  // columnLeft + 12
-      _bi_consumer_fn_t fp = tGetBiConsumerFn(pLeft->pSchema->type, pRight->pVal->nType, pExprs->_node.optr);
+      _bi_consumer_fn_t fp = getArithmeticOperatorFn(pLeft->pSchema->type, pRight->pVal->nType, pExprs->_node.optr);
 
       if (order == TSDB_ORDER_DESC) {
         reverseCopy(pdata, pLeftInputData, pLeft->pSchema->type, numOfRows);
@@ -511,13 +249,13 @@ void tExprTreeCalcTraverse(tExprNode *pExprs, int32_t numOfRows, char *pOutput,
   } else {
     // column data specified on left-hand-side
     if (pRight->nodeType == TSQL_NODE_EXPR) {  // 12 + expr2
-      _bi_consumer_fn_t fp = tGetBiConsumerFn(pLeft->pVal->nType, TSDB_DATA_TYPE_DOUBLE, pExprs->_node.optr);
+      _bi_consumer_fn_t fp = getArithmeticOperatorFn(pLeft->pVal->nType, TSDB_DATA_TYPE_DOUBLE, pExprs->_node.optr);
       fp(&pLeft->pVal->i64Key, pRightOutput, 1, numOfRows, pOutput, TSDB_ORDER_ASC);
 
     } else if (pRight->nodeType == TSQL_NODE_COL) {  // 12 + columnRight
       // column data specified on right-hand-side
       char *pRightInputData = getSourceDataBlock(param, pRight->pSchema->name, pRight->pSchema->colId);
-      _bi_consumer_fn_t fp = tGetBiConsumerFn(pLeft->pVal->nType, pRight->pSchema->type, pExprs->_node.optr);
+      _bi_consumer_fn_t fp = getArithmeticOperatorFn(pLeft->pVal->nType, pRight->pSchema->type, pExprs->_node.optr);
 
       if (order == TSDB_ORDER_DESC) {
         reverseCopy(pdata, pRightInputData, pRight->pSchema->type, numOfRows);
@@ -527,7 +265,7 @@ void tExprTreeCalcTraverse(tExprNode *pExprs, int32_t numOfRows, char *pOutput,
       }
 
     } else if (pRight->nodeType == TSQL_NODE_VALUE) {  // 12 + 12
-      _bi_consumer_fn_t fp = tGetBiConsumerFn(pLeft->pVal->nType, pRight->pVal->nType, pExprs->_node.optr);
+      _bi_consumer_fn_t fp = getArithmeticOperatorFn(pLeft->pVal->nType, pRight->pVal->nType, pExprs->_node.optr);
       fp(&pLeft->pVal->i64Key, &pRight->pVal->i64Key, 1, 1, pOutput, TSDB_ORDER_ASC);
     }
   }

src/query/src/qExecutor.c

@@ -86,7 +86,6 @@ typedef struct {
   int32_t     status;       // query status
   TSKEY       lastKey;      // the lastKey value before query executed
   STimeWindow w;            // whole query time window
-  STimeWindow curWindow;    // current query window
   int32_t     windowIndex;  // index of active time window result for interval query
   STSCursor   cur;
 } SQueryStatusInfo;
@@ -718,7 +717,7 @@ static FORCE_INLINE int32_t getForwardStepsInBlock(int32_t numOfRows, __block_se
   return forwardStep;
 }
 
-static void doUpdateResultRowIndex(SResultRowInfo*pResultRowInfo, TSKEY lastKey, bool ascQuery) {
+static void doUpdateResultRowIndex(SResultRowInfo*pResultRowInfo, TSKEY lastKey, bool ascQuery, bool timeWindowInterpo) {
   int64_t skey = TSKEY_INITIAL_VAL;
   int32_t i = 0;
   for (i = pResultRowInfo->size - 1; i >= 0; --i) {
@@ -728,12 +727,24 @@ static void doUpdateResultRowIndex(SResultRowInfo*pResultRowInfo, TSKEY lastKey,
     }
 
     // new closed result rows
+    if (timeWindowInterpo) {
+      if (pResult->endInterp && ((pResult->win.skey <= lastKey && ascQuery) || (pResult->win.skey >= lastKey && !ascQuery))) {
+        if (i > 0) { // the first time window, the startInterp is false.
+          assert(pResult->startInterp);
+        }
+
+        closeResultRow(pResultRowInfo, i);
+      } else {
+        skey = pResult->win.skey;
+      }
+    } else {
       if ((pResult->win.ekey <= lastKey && ascQuery) || (pResult->win.skey >= lastKey && !ascQuery)) {
         closeResultRow(pResultRowInfo, i);
       } else {
         skey = pResult->win.skey;
       }
     }
+  }
 
   // all result rows are closed, set the last one to be the skey
   if (skey == TSKEY_INITIAL_VAL) {
@@ -752,13 +763,13 @@ static void doUpdateResultRowIndex(SResultRowInfo*pResultRowInfo, TSKEY lastKey,
   }
 }
 
-static void updateResultRowIndex(SResultRowInfo* pResultRowInfo, STableQueryInfo* pTableQueryInfo, bool ascQuery) {
+static void updateResultRowIndex(SResultRowInfo* pResultRowInfo, STableQueryInfo* pTableQueryInfo, bool ascQuery, bool timeWindowInterpo) {
   if ((pTableQueryInfo->lastKey > pTableQueryInfo->win.ekey && ascQuery) || (pTableQueryInfo->lastKey < pTableQueryInfo->win.ekey && (!ascQuery))) {
     closeAllResultRows(pResultRowInfo);
     pResultRowInfo->curIndex = pResultRowInfo->size - 1;
   } else {
     int32_t step = ascQuery? 1:-1;
-    doUpdateResultRowIndex(pResultRowInfo, pTableQueryInfo->lastKey - step, ascQuery);
+    doUpdateResultRowIndex(pResultRowInfo, pTableQueryInfo->lastKey - step, ascQuery, timeWindowInterpo);
   }
 }
 
@@ -814,9 +825,6 @@ static void doBlockwiseApplyFunctions(SQueryRuntimeEnv *pRuntimeEnv, STimeWindow
     pCtx[k].startOffset = (QUERY_IS_ASC_QUERY(pQuery)) ? offset : offset - (forwardStep - 1);
 
     int32_t functionId = pQuery->pExpr1[k].base.functionId;
-    if ((aAggs[functionId].nStatus & TSDB_FUNCSTATE_SELECTIVITY) != 0) {
-      pCtx[k].ptsList = &tsCol[pCtx[k].startOffset];
-    }
 
     // not a whole block involved in query processing, statistics data can not be used
     // NOTE: the original value of isSet have been changed here
@@ -954,8 +962,7 @@ static void* getDataBlockImpl(SArray* pDataBlock, int32_t colId) {
   return NULL;
 }
 
-static char *getDataBlock(SQueryRuntimeEnv *pRuntimeEnv, SArithmeticSupport *sas, int32_t col, int32_t size,
-                    SArray *pDataBlock) {
+static char *getDataBlock(SQueryRuntimeEnv *pRuntimeEnv, SArithmeticSupport *sas, int32_t col, int32_t size, SArray *pDataBlock) {
   if (pDataBlock == NULL) {
     return NULL;
   }
@@ -966,15 +973,9 @@ static char *getDataBlock(SQueryRuntimeEnv *pRuntimeEnv, SArithmeticSupport *sas
   int32_t functionId = pQuery->pExpr1[col].base.functionId;
   if (functionId == TSDB_FUNC_ARITHM) {
     sas->pArithExpr = &pQuery->pExpr1[col];
-
     sas->offset     = (QUERY_IS_ASC_QUERY(pQuery))? pQuery->pos : pQuery->pos - (size - 1);
     sas->colList    = pQuery->colList;
     sas->numOfCols  = pQuery->numOfCols;
-    sas->data      = calloc(pQuery->numOfCols, POINTER_BYTES);
-
-    if (sas->data == NULL) {
-      longjmp(pRuntimeEnv->env, TSDB_CODE_QRY_OUT_OF_MEMORY);
-    }
 
     // here the pQuery->colList and sas->colList are identical
     int32_t numOfCols = (int32_t)taosArrayGetSize(pDataBlock);
@@ -1077,13 +1078,13 @@ static bool setTimeWindowInterpolationEndTs(SQueryRuntimeEnv* pRuntimeEnv, int32
   return true;
 }
 
-static void saveDataBlockLastRow(SQueryRuntimeEnv* pRuntimeEnv, SDataBlockInfo* pDataBlockInfo, SArray* pDataBlock) {
+static void saveDataBlockLastRow(SQueryRuntimeEnv* pRuntimeEnv, SDataBlockInfo* pDataBlockInfo, SArray* pDataBlock,
+    int32_t rowIndex) {
   if (pDataBlock == NULL) {
     return;
   }
 
   SQuery* pQuery = pRuntimeEnv->pQuery;
-  int32_t rowIndex = QUERY_IS_ASC_QUERY(pQuery)? pDataBlockInfo->rows-1:0;
   for (int32_t k = 0; k < pQuery->numOfCols; ++k) {
     SColumnInfoData *pColInfo = taosArrayGet(pDataBlock, k);
     memcpy(pRuntimeEnv->prevRow[k], ((char*)pColInfo->pData) + (pColInfo->info.bytes * rowIndex), pColInfo->info.bytes);
@@ -1166,15 +1167,10 @@ static void blockwiseApplyFunctions(SQueryRuntimeEnv *pRuntimeEnv, SDataStatis *
     tsCols = (TSKEY *)(pColInfo->pData);
   }
 
-  SArithmeticSupport *sasArray = calloc((size_t)pQuery->numOfOutput, sizeof(SArithmeticSupport));
-  if (sasArray == NULL) {
-    longjmp(pRuntimeEnv->env, TSDB_CODE_QRY_OUT_OF_MEMORY);
-  }
-
   SQInfo *pQInfo = GET_QINFO_ADDR(pRuntimeEnv);
   for (int32_t k = 0; k < pQuery->numOfOutput; ++k) {
-    char *dataBlock = getDataBlock(pRuntimeEnv, &sasArray[k], k, pDataBlockInfo->rows, pDataBlock);
-    setExecParams(pQuery, &pCtx[k], dataBlock, tsCols, pDataBlockInfo, pStatis, &sasArray[k], k, pQInfo->vgId);
+    char *dataBlock = getDataBlock(pRuntimeEnv, &pRuntimeEnv->sasArray[k], k, pDataBlockInfo->rows, pDataBlock);
+    setExecParams(pQuery, &pCtx[k], dataBlock, tsCols, pDataBlockInfo, pStatis, &pRuntimeEnv->sasArray[k], k, pQInfo->vgId);
   }
 
   int32_t step = GET_FORWARD_DIRECTION_FACTOR(pQuery->order.order);
@@ -1266,25 +1262,12 @@ static void blockwiseApplyFunctions(SQueryRuntimeEnv *pRuntimeEnv, SDataStatis *
 
   _end:
   if (pRuntimeEnv->timeWindowInterpo) {
-    saveDataBlockLastRow(pRuntimeEnv, pDataBlockInfo, pDataBlock);
-  }
-
-  for (int32_t i = 0; i < pQuery->numOfOutput; ++i) {
-    if (pQuery->pExpr1[i].base.functionId != TSDB_FUNC_ARITHM) {
-      continue;
-    }
-
-    tfree(sasArray[i].data);
+    int32_t rowIndex = QUERY_IS_ASC_QUERY(pQuery)? pDataBlockInfo->rows-1:0;
+    saveDataBlockLastRow(pRuntimeEnv, pDataBlockInfo, pDataBlock, rowIndex);
   }
-
-  tfree(sasArray);
 }
 
 static int32_t setGroupResultOutputBuf(SQueryRuntimeEnv *pRuntimeEnv, char *pData, int16_t type, int16_t bytes, int32_t groupIndex) {
-  if (isNull(pData, type)) {  // ignore the null value
-    return -1;
-  }
-
   SDiskbasedResultBuf *pResultBuf = pRuntimeEnv->pResultBuf;
 
   // not assign result buffer yet, add new result buffer, TODO remove it
@@ -1296,8 +1279,7 @@ static int32_t setGroupResultOutputBuf(SQueryRuntimeEnv *pRuntimeEnv, char *pDat
   } else if (type == TSDB_DATA_TYPE_FLOAT || type == TSDB_DATA_TYPE_DOUBLE) {
     SQInfo* pQInfo = GET_QINFO_ADDR(pRuntimeEnv);
     qError("QInfo:%p group by not supported on double/float columns, abort", pQInfo);
-
-    longjmp(pRuntimeEnv->env, TSDB_CODE_QRY_APP_ERROR);
+    return -1;
   }
 
   SResultRow *pResultRow = doPrepareResultRowFromKey(pRuntimeEnv, &pRuntimeEnv->windowResInfo, d, len, true, groupIndex);
@@ -1525,12 +1507,7 @@ static void rowwiseApplyFunctions(SQueryRuntimeEnv *pRuntimeEnv, SDataStatis *pS
   SColumnInfoData* pColumnInfoData = (SColumnInfoData *)taosArrayGet(pDataBlock, 0);
 
   TSKEY  *tsCols = (pColumnInfoData->info.type == TSDB_DATA_TYPE_TIMESTAMP)? (TSKEY*) pColumnInfoData->pData:NULL;
-  bool    groupbyColumnValue = pRuntimeEnv->groupbyNormalCol;
-
-  SArithmeticSupport *sasArray = calloc((size_t)pQuery->numOfOutput, sizeof(SArithmeticSupport));
-  if (sasArray == NULL) {
-    longjmp(pRuntimeEnv->env, TSDB_CODE_QRY_OUT_OF_MEMORY);
-  }
+  bool    groupbyColumnValue = pRuntimeEnv->groupbyColumn;
 
   int16_t type = 0;
   int16_t bytes = 0;
@@ -1542,8 +1519,8 @@ static void rowwiseApplyFunctions(SQueryRuntimeEnv *pRuntimeEnv, SDataStatis *pS
 
   SQInfo* pQInfo = GET_QINFO_ADDR(pRuntimeEnv);
   for (int32_t k = 0; k < pQuery->numOfOutput; ++k) {
-    char *dataBlock = getDataBlock(pRuntimeEnv, &sasArray[k], k, pDataBlockInfo->rows, pDataBlock);
-    setExecParams(pQuery, &pCtx[k], dataBlock, tsCols, pDataBlockInfo, pStatis, &sasArray[k], k, pQInfo->vgId);
+    char *dataBlock = getDataBlock(pRuntimeEnv, &pRuntimeEnv->sasArray[k], k, pDataBlockInfo->rows, pDataBlock);
+    setExecParams(pQuery, &pCtx[k], dataBlock, tsCols, pDataBlockInfo, pStatis, &pRuntimeEnv->sasArray[k], k, pQInfo->vgId);
     pCtx[k].size = 1;
   }
 
@@ -1628,10 +1605,9 @@ static void rowwiseApplyFunctions(SQueryRuntimeEnv *pRuntimeEnv, SDataStatis *pS
       }
 
       doRowwiseApplyFunctions(pRuntimeEnv, &win, offset);
-
-      STimeWindow nextWin = win;
       int32_t index = pWindowResInfo->curIndex;
 
+      STimeWindow nextWin = win;
       while (1) {
         getNextTimeWindow(pQuery, &nextWin);
         if ((nextWin.skey > pQuery->window.ekey && QUERY_IS_ASC_QUERY(pQuery)) ||
@@ -1653,15 +1629,19 @@ static void rowwiseApplyFunctions(SQueryRuntimeEnv *pRuntimeEnv, SDataStatis *pS
         doRowwiseApplyFunctions(pRuntimeEnv, &nextWin, offset);
       }
 
+      // restore the index, add the result row will move the index
       pWindowResInfo->curIndex = index;
     } else {  // other queries
       // decide which group this rows belongs to according to current state value
       if (groupbyColumnValue) {
         char *val = groupbyColumnData + bytes * offset;
+        if (isNull(val, type)) {  // ignore the null value
+          continue;
+        }
 
         int32_t ret = setGroupResultOutputBuf(pRuntimeEnv, val, type, bytes, item->groupIndex);
         if (ret != TSDB_CODE_SUCCESS) {  // null data, too many state code
-          continue;
+          longjmp(pRuntimeEnv->env, TSDB_CODE_QRY_APP_ERROR);
         }
       }
 
@@ -1686,27 +1666,20 @@ static void rowwiseApplyFunctions(SQueryRuntimeEnv *pRuntimeEnv, SDataStatis *pS
   }
 
   _end:
-  assert(offset >= 0);
-  if (tsCols != NULL) {
-    item->lastKey = tsCols[offset] + step;
-  } else {
-    item->lastKey = (QUERY_IS_ASC_QUERY(pQuery)? pDataBlockInfo->window.ekey:pDataBlockInfo->window.skey) + step;
+  assert(offset >= 0 && tsCols != NULL);
+  if (prevTs != INT64_MIN) {
+    assert(prevRowIndex >= 0);
+    item->lastKey = prevTs + step;
   }
 
-  if (pRuntimeEnv->pTsBuf != NULL) {
-    item->cur = tsBufGetCursor(pRuntimeEnv->pTsBuf);
-  }
-
-  // todo refactor: extract method
-  for(int32_t i = 0; i < pQuery->numOfOutput; ++i) {
-    if (pQuery->pExpr1[i].base.functionId != TSDB_FUNC_ARITHM) {
-      continue;
+  // In case of all rows in current block are not qualified
+  if (pRuntimeEnv->timeWindowInterpo && prevRowIndex != -1) {
+    saveDataBlockLastRow(pRuntimeEnv, pDataBlockInfo, pDataBlock, prevRowIndex);
   }
 
-    tfree(sasArray[i].data);
+  if (pRuntimeEnv->pTsBuf != NULL) {
+    item->cur = tsBufGetCursor(pRuntimeEnv->pTsBuf);
   }
-
-  free(sasArray);
 }
 
 static int32_t tableApplyFunctionsOnBlock(SQueryRuntimeEnv *pRuntimeEnv, SDataBlockInfo *pDataBlockInfo,
@@ -1716,7 +1689,7 @@ static int32_t tableApplyFunctionsOnBlock(SQueryRuntimeEnv *pRuntimeEnv, SDataBl
   STableQueryInfo* pTableQueryInfo = pQuery->current;
   SResultRowInfo*  pResultRowInfo = &pRuntimeEnv->windowResInfo;
 
-  if (pQuery->numOfFilterCols > 0 || pRuntimeEnv->pTsBuf != NULL || pRuntimeEnv->groupbyNormalCol) {
+  if (pQuery->numOfFilterCols > 0 || pRuntimeEnv->pTsBuf != NULL || pRuntimeEnv->groupbyColumn) {
     rowwiseApplyFunctions(pRuntimeEnv, pStatis, pDataBlockInfo, pResultRowInfo, pDataBlock);
   } else {
     blockwiseApplyFunctions(pRuntimeEnv, pStatis, pDataBlockInfo, pResultRowInfo, searchFn, pDataBlock);
@@ -1728,9 +1701,9 @@ static int32_t tableApplyFunctionsOnBlock(SQueryRuntimeEnv *pRuntimeEnv, SDataBl
 
   // interval query with limit applied
   int32_t numOfRes = 0;
-  if (QUERY_IS_INTERVAL_QUERY(pQuery) || pRuntimeEnv->groupbyNormalCol) {
+  if (QUERY_IS_INTERVAL_QUERY(pQuery) || pRuntimeEnv->groupbyColumn) {
     numOfRes = pResultRowInfo->size;
-    updateResultRowIndex(pResultRowInfo, pTableQueryInfo, QUERY_IS_ASC_QUERY(pQuery));
+    updateResultRowIndex(pResultRowInfo, pTableQueryInfo, QUERY_IS_ASC_QUERY(pQuery), pRuntimeEnv->timeWindowInterpo);
   } else { // projection query
     numOfRes = (int32_t) getNumOfResult(pRuntimeEnv);
 
@@ -1786,7 +1759,7 @@ void setExecParams(SQuery *pQuery, SQLFunctionCtx *pCtx, void* inputData, TSKEY
 
   uint32_t status = aAggs[functionId].nStatus;
   if (((status & (TSDB_FUNCSTATE_SELECTIVITY | TSDB_FUNCSTATE_NEED_TS)) != 0) && (tsCol != NULL)) {
-    pCtx->ptsList = &tsCol[pCtx->startOffset];
+    pCtx->ptsList = tsCol;
   }
 
   if (functionId >= TSDB_FUNC_FIRST_DST && functionId <= TSDB_FUNC_LAST_DST) {
@@ -1904,8 +1877,9 @@ static int32_t setupQueryRuntimeEnv(SQueryRuntimeEnv *pRuntimeEnv, int16_t order
   pRuntimeEnv->pCtx = (SQLFunctionCtx *)calloc(pQuery->numOfOutput, sizeof(SQLFunctionCtx));
   pRuntimeEnv->offset = calloc(pQuery->numOfOutput, sizeof(int16_t));
   pRuntimeEnv->rowCellInfoOffset = calloc(pQuery->numOfOutput, sizeof(int32_t));
+  pRuntimeEnv->sasArray = calloc(pQuery->numOfOutput, sizeof(SArithmeticSupport));
 
-  if (pRuntimeEnv->offset == NULL || pRuntimeEnv->pCtx == NULL || pRuntimeEnv->rowCellInfoOffset == NULL) {
+  if (pRuntimeEnv->offset == NULL || pRuntimeEnv->pCtx == NULL || pRuntimeEnv->rowCellInfoOffset == NULL || pRuntimeEnv->sasArray == NULL) {
     goto _clean;
   }
 
@@ -1981,18 +1955,24 @@ static int32_t setupQueryRuntimeEnv(SQueryRuntimeEnv *pRuntimeEnv, int16_t order
       pCtx->param[1].i64Key = pQuery->order.orderColId;
     }
 
+    if (functionId == TSDB_FUNC_ARITHM) {
+      pRuntimeEnv->sasArray[i].data = calloc(pQuery->numOfCols, POINTER_BYTES);
+      if (pRuntimeEnv->sasArray[i].data == NULL) {
+        goto _clean;
+      }
+    }
+
     if (i > 0) {
       pRuntimeEnv->offset[i] = pRuntimeEnv->offset[i - 1] + pRuntimeEnv->pCtx[i - 1].outputBytes;
       pRuntimeEnv->rowCellInfoOffset[i] = pRuntimeEnv->rowCellInfoOffset[i - 1] + sizeof(SResultRowCellInfo) + pQuery->pExpr1[i - 1].interBytes;
     }
-
   }
 
   *(int64_t*) pRuntimeEnv->prevRow[0] = INT64_MIN;
 
   // if it is group by normal column, do not set output buffer, the output buffer is pResult
   // fixed output query/multi-output query for normal table
-  if (!pRuntimeEnv->groupbyNormalCol && !pRuntimeEnv->stableQuery && !QUERY_IS_INTERVAL_QUERY(pRuntimeEnv->pQuery)) {
+  if (!pRuntimeEnv->groupbyColumn && !pRuntimeEnv->stableQuery && !QUERY_IS_INTERVAL_QUERY(pRuntimeEnv->pQuery)) {
     resetDefaultResInfoOutputBuf(pRuntimeEnv);
   }
 
@@ -2007,6 +1987,7 @@ _clean:
   tfree(pRuntimeEnv->pCtx);
   tfree(pRuntimeEnv->offset);
   tfree(pRuntimeEnv->rowCellInfoOffset);
+  tfree(pRuntimeEnv->sasArray);
 
   return TSDB_CODE_QRY_OUT_OF_MEMORY;
 }
@@ -2050,6 +2031,14 @@ static void teardownQueryRuntimeEnv(SQueryRuntimeEnv *pRuntimeEnv) {
     tfree(pRuntimeEnv->pCtx);
   }
 
+  if (pRuntimeEnv->sasArray != NULL) {
+    for(int32_t i = 0; i < pQuery->numOfOutput; ++i) {
+      tfree(pRuntimeEnv->sasArray[i].data);
+    }
+
+    tfree(pRuntimeEnv->sasArray);
+  }
+
   pRuntimeEnv->pFillInfo = taosDestroyFillInfo(pRuntimeEnv->pFillInfo);
 
   destroyResultBuf(pRuntimeEnv->pResultBuf);
@@ -2102,7 +2091,7 @@ static bool isFixedOutputQuery(SQueryRuntimeEnv* pRuntimeEnv) {
   }
 
   // Note:top/bottom query is fixed output query
-  if (pRuntimeEnv->topBotQuery || pRuntimeEnv->groupbyNormalCol) {
+  if (pRuntimeEnv->topBotQuery || pRuntimeEnv->groupbyColumn) {
     return true;
   }
 
@@ -2743,7 +2732,7 @@ static void ensureOutputBufferSimple(SQueryRuntimeEnv* pRuntimeEnv, int32_t capa
 static void ensureOutputBuffer(SQueryRuntimeEnv* pRuntimeEnv, SDataBlockInfo* pBlockInfo) {
   // in case of prj/diff query, ensure the output buffer is sufficient to accommodate the results of current block
   SQuery* pQuery = pRuntimeEnv->pQuery;
-  if (!QUERY_IS_INTERVAL_QUERY(pQuery) && !pRuntimeEnv->groupbyNormalCol && !isFixedOutputQuery(pRuntimeEnv) && !isTSCompQuery(pQuery)) {
+  if (!QUERY_IS_INTERVAL_QUERY(pQuery) && !pRuntimeEnv->groupbyColumn && !isFixedOutputQuery(pRuntimeEnv) && !isTSCompQuery(pQuery)) {
     SResultRec *pRec = &pQuery->rec;
 
     if (pQuery->rec.capacity - pQuery->rec.rows < pBlockInfo->rows) {
@@ -2967,50 +2956,6 @@ void setTagVal(SQueryRuntimeEnv *pRuntimeEnv, void *pTable, void *tsdb) {
   }
 }
 
-static UNUSED_FUNC void doMerge(SQueryRuntimeEnv *pRuntimeEnv, int64_t timestamp, SResultRow *pWindowRes, bool mergeFlag) {
-  SQuery *        pQuery = pRuntimeEnv->pQuery;
-  SQLFunctionCtx *pCtx = pRuntimeEnv->pCtx;
-
-  tFilePage *page = getResBufPage(pRuntimeEnv->pResultBuf, pWindowRes->pageId);
-
-  for (int32_t i = 0; i < pQuery->numOfOutput; ++i) {
-    int32_t functionId = pQuery->pExpr1[i].base.functionId;
-    if (!mergeFlag) {
-      pCtx[i].aOutputBuf = pCtx[i].aOutputBuf + pCtx[i].outputBytes;
-      pCtx[i].currentStage = FIRST_STAGE_MERGE;
-
-      RESET_RESULT_INFO(pCtx[i].resultInfo);
-      aAggs[functionId].init(&pCtx[i]);
-    }
-
-    pCtx[i].hasNull = true;
-    pCtx[i].nStartQueryTimestamp = timestamp;
-    pCtx[i].aInputElemBuf = getPosInResultPage(pRuntimeEnv, i, pWindowRes, page);
-
-    // in case of tag column, the tag information should be extracted from input buffer
-    if (functionId == TSDB_FUNC_TAG_DUMMY || functionId == TSDB_FUNC_TAG) {
-      tVariantDestroy(&pCtx[i].tag);
-
-      int32_t type = pCtx[i].outputType;
-      if (type == TSDB_DATA_TYPE_BINARY || type == TSDB_DATA_TYPE_NCHAR) {
-        tVariantCreateFromBinary(&pCtx[i].tag, varDataVal(pCtx[i].aInputElemBuf), varDataLen(pCtx[i].aInputElemBuf), type);
-      } else {
-        tVariantCreateFromBinary(&pCtx[i].tag, pCtx[i].aInputElemBuf, pCtx[i].inputBytes, pCtx[i].inputType);
-      }
-
-    }
-  }
-
-  for (int32_t i = 0; i < pQuery->numOfOutput; ++i) {
-    int32_t functionId = pQuery->pExpr1[i].base.functionId;
-    if (functionId == TSDB_FUNC_TAG_DUMMY) {
-      continue;
-    }
-
-    aAggs[functionId].distMergeFunc(&pCtx[i]);
-  }
-}
-
 static UNUSED_FUNC void printBinaryData(int32_t functionId, char *data, int32_t srcDataType) {
   if (functionId == TSDB_FUNC_FIRST_DST || functionId == TSDB_FUNC_LAST_DST) {
     switch (srcDataType) {
@@ -3170,8 +3115,8 @@ int32_t mergeGroupResult(SQInfo *pQInfo) {
     SArray *group = GET_TABLEGROUP(pQInfo, pQInfo->groupIndex);
 
     int32_t ret = mergeIntoGroupResultImpl(pGroupResInfo, group, pQInfo);
-    if (ret < 0) {
-      return -1;
+    if (ret != TSDB_CODE_SUCCESS) {
+      return ret;
     }
 
     // this group generates at least one result, return results
@@ -3266,7 +3211,7 @@ int32_t mergeIntoGroupResultImpl(SGroupResInfo* pGroupResInfo, SArray *pTableLis
   posList = calloc(size, sizeof(int32_t));
   pTableQueryInfoList = malloc(POINTER_BYTES * size);
 
-  if (pTableQueryInfoList == NULL || posList == NULL) {
+  if (pTableQueryInfoList == NULL || posList == NULL || pGroupResInfo->pRows == NULL) {
     qError("QInfo:%p failed alloc memory", pQInfo);
     code = TSDB_CODE_QRY_OUT_OF_MEMORY;
     goto _end;
@@ -3356,10 +3301,6 @@ int32_t mergeIntoGroupResultImpl(SGroupResInfo* pGroupResInfo, SArray *pTableLis
   tfree(posList);
   tfree(pTree);
 
-  if (code != TSDB_CODE_SUCCESS) {
-    longjmp(pRuntimeEnv->env, code);
-  }
-
   return code;
 }
 
@@ -3368,14 +3309,6 @@ static void updateTableQueryInfoForReverseScan(SQuery *pQuery, STableQueryInfo *
     return;
   }
 
-  // order has changed already
-  int32_t step = GET_FORWARD_DIRECTION_FACTOR(pQuery->order.order);
-  if (pTableQueryInfo->lastKey == pTableQueryInfo->win.skey) {
-    // do nothing, no results
-  } else {// NOTE: even win.skey != lastKey, the results may not generated.
-    pTableQueryInfo->win.ekey = pTableQueryInfo->lastKey + step;
-  }
-
   SWAP(pTableQueryInfo->win.skey, pTableQueryInfo->win.ekey, TSKEY);
   pTableQueryInfo->lastKey = pTableQueryInfo->win.skey;
 
@@ -3419,7 +3352,7 @@ void disableFuncInReverseScan(SQInfo *pQInfo) {
 
   // group by normal columns and interval query on normal table
   SResultRowInfo *pWindowResInfo = &pRuntimeEnv->windowResInfo;
-  if (pRuntimeEnv->groupbyNormalCol || QUERY_IS_INTERVAL_QUERY(pQuery)) {
+  if (pRuntimeEnv->groupbyColumn || QUERY_IS_INTERVAL_QUERY(pQuery)) {
     disableFuncInReverseScanImpl(pRuntimeEnv, pWindowResInfo, order);
   } else {  // for simple result of table query,
     for (int32_t j = 0; j < pQuery->numOfOutput; ++j) {  // todo refactor
@@ -3607,7 +3540,7 @@ bool needScanDataBlocksAgain(SQueryRuntimeEnv *pRuntimeEnv) {
   SQuery *pQuery = pRuntimeEnv->pQuery;
 
   bool toContinue = false;
-  if (pRuntimeEnv->groupbyNormalCol || QUERY_IS_INTERVAL_QUERY(pQuery)) {
+  if (pRuntimeEnv->groupbyColumn || QUERY_IS_INTERVAL_QUERY(pQuery)) {
     // for each group result, call the finalize function for each column
     SResultRowInfo *pWindowResInfo = &pRuntimeEnv->windowResInfo;
 
@@ -3658,10 +3591,6 @@ static SQueryStatusInfo getQueryStatusInfo(SQueryRuntimeEnv *pRuntimeEnv, TSKEY
   };
 
   TIME_WINDOW_COPY(info.w, pQuery->window);
-  TIME_WINDOW_COPY(info.curWindow, pTableQueryInfo->win);
-
-  info.curWindow.skey = start;
-
   return info;
 }
 
@@ -3677,9 +3606,7 @@ static void setEnvBeforeReverseScan(SQueryRuntimeEnv *pRuntimeEnv, SQueryStatusI
   }
 
   // reverse order time range
-  pQuery->window = pStatus->curWindow;
   SWAP(pQuery->window.skey, pQuery->window.ekey, TSKEY);
-
   SWITCH_ORDER(pQuery->order.order);
 
   if (QUERY_IS_ASC_QUERY(pQuery)) {
@@ -3745,9 +3672,7 @@ void scanOneTableDataBlocks(SQueryRuntimeEnv *pRuntimeEnv, TSKEY start) {
 
   // store the start query position
   SQueryStatusInfo qstatus = getQueryStatusInfo(pRuntimeEnv, start);
-
   SET_MASTER_SCAN_FLAG(pRuntimeEnv);
-  int32_t step = GET_FORWARD_DIRECTION_FACTOR(pQuery->order.order);
 
   while (1) {
     doScanAllDataBlocks(pRuntimeEnv);
@@ -3756,13 +3681,9 @@ void scanOneTableDataBlocks(SQueryRuntimeEnv *pRuntimeEnv, TSKEY start) {
       qstatus.status = pQuery->status;
 
       // do nothing if no data blocks are found qualified during scan
-      if (qstatus.lastKey != pTableQueryInfo->lastKey) {
-        qstatus.curWindow.ekey = pTableQueryInfo->lastKey - step;
-      } else { // the lastkey does not increase, which means no data checked yet
+      if (qstatus.lastKey == pTableQueryInfo->lastKey) {
         qDebug("QInfo:%p no results generated in this scan", pQInfo);
       }
-
-      qstatus.lastKey = pTableQueryInfo->lastKey;
     }
 
     if (!needScanDataBlocksAgain(pRuntimeEnv)) {
@@ -3778,7 +3699,7 @@ void scanOneTableDataBlocks(SQueryRuntimeEnv *pRuntimeEnv, TSKEY start) {
       tsdbCleanupQueryHandle(pRuntimeEnv->pSecQueryHandle);
     }
 
-    STsdbQueryCond cond = createTsdbQueryCond(pQuery, &qstatus.curWindow);
+    STsdbQueryCond cond = createTsdbQueryCond(pQuery, &pQuery->window);
     restoreTimeWindow(&pQInfo->tableGroupInfo, &cond);
     pRuntimeEnv->pSecQueryHandle = tsdbQueryTables(pQInfo->tsdb, &cond, &pQInfo->tableGroupInfo, pQInfo, &pQInfo->memRef);
     if (pRuntimeEnv->pSecQueryHandle == NULL) {
@@ -3814,10 +3735,10 @@ void scanOneTableDataBlocks(SQueryRuntimeEnv *pRuntimeEnv, TSKEY start) {
 void finalizeQueryResult(SQueryRuntimeEnv *pRuntimeEnv) {
   SQuery *pQuery = pRuntimeEnv->pQuery;
 
-  if (pRuntimeEnv->groupbyNormalCol || QUERY_IS_INTERVAL_QUERY(pQuery)) {
+  if (pRuntimeEnv->groupbyColumn || QUERY_IS_INTERVAL_QUERY(pQuery)) {
     // for each group result, call the finalize function for each column
     SResultRowInfo *pWindowResInfo = &pRuntimeEnv->windowResInfo;
-    if (pRuntimeEnv->groupbyNormalCol) {
+    if (pRuntimeEnv->groupbyColumn) {
       closeAllResultRows(pWindowResInfo);
     }
 
@@ -3871,7 +3792,7 @@ static STableQueryInfo *createTableQueryInfo(SQueryRuntimeEnv *pRuntimeEnv, void
   pTableQueryInfo->cur.vgroupIndex = -1;
 
   // set more initial size of interval/groupby query
-  if (QUERY_IS_INTERVAL_QUERY(pQuery) || pRuntimeEnv->groupbyNormalCol) {
+  if (QUERY_IS_INTERVAL_QUERY(pQuery) || pRuntimeEnv->groupbyColumn) {
     int32_t initialSize = 128;
     int32_t code = initResultRowInfo(&pTableQueryInfo->windowResInfo, initialSize, TSDB_DATA_TYPE_INT);
     if (code != TSDB_CODE_SUCCESS) {
@@ -4224,14 +4145,14 @@ static void stableApplyFunctionsOnBlock(SQueryRuntimeEnv *pRuntimeEnv, SDataBloc
   SResultRowInfo * pResultRowInfo = &pTableQueryInfo->windowResInfo;
   pQuery->pos = QUERY_IS_ASC_QUERY(pQuery)? 0 : pDataBlockInfo->rows - 1;
 
-  if (pQuery->numOfFilterCols > 0 || pRuntimeEnv->pTsBuf != NULL || pRuntimeEnv->groupbyNormalCol) {
+  if (pQuery->numOfFilterCols > 0 || pRuntimeEnv->pTsBuf != NULL || pRuntimeEnv->groupbyColumn) {
     rowwiseApplyFunctions(pRuntimeEnv, pStatis, pDataBlockInfo, pResultRowInfo, pDataBlock);
   } else {
     blockwiseApplyFunctions(pRuntimeEnv, pStatis, pDataBlockInfo, pResultRowInfo, searchFn, pDataBlock);
   }
 
   if (QUERY_IS_INTERVAL_QUERY(pQuery)) {
-    updateResultRowIndex(pResultRowInfo, pTableQueryInfo, QUERY_IS_ASC_QUERY(pQuery));
+    updateResultRowIndex(pResultRowInfo, pTableQueryInfo, QUERY_IS_ASC_QUERY(pQuery), pRuntimeEnv->timeWindowInterpo);
   }
 }
 
@@ -4267,7 +4188,7 @@ bool queryHasRemainResForTableQuery(SQueryRuntimeEnv* pRuntimeEnv) {
   } else {
     // there are results waiting for returned to client.
     if (Q_STATUS_EQUAL(pQuery->status, QUERY_COMPLETED) &&
-        (pRuntimeEnv->groupbyNormalCol || QUERY_IS_INTERVAL_QUERY(pQuery)) &&
+        (pRuntimeEnv->groupbyColumn || QUERY_IS_INTERVAL_QUERY(pQuery)) &&
         (pRuntimeEnv->windowResInfo.size > 0)) {
       return true;
     }
@@ -4521,6 +4442,7 @@ static TSKEY doSkipIntervalProcess(SQueryRuntimeEnv* pRuntimeEnv, STimeWindow* w
   } else {  // do nothing
     pQuery->window.skey      = tw.skey;
     pWindowResInfo->prevSKey = tw.skey;
+    pTableQueryInfo->lastKey = tw.skey;
 
     return tw.skey;
   }
@@ -4530,23 +4452,11 @@ static TSKEY doSkipIntervalProcess(SQueryRuntimeEnv* pRuntimeEnv, STimeWindow* w
 
 static bool skipTimeInterval(SQueryRuntimeEnv *pRuntimeEnv, TSKEY* start) {
   SQuery *pQuery = pRuntimeEnv->pQuery;
-
-  // get the first unclosed time window
-  bool assign = false;
-  for(int32_t i = 0; i < pRuntimeEnv->windowResInfo.size; ++i) {
-    if (pRuntimeEnv->windowResInfo.pResult[i]->closed) {
-      continue;
-    }
-
-    assign = true;
-    *start = pRuntimeEnv->windowResInfo.pResult[i]->win.skey;
-  }
-
-  if (!assign) {
-    *start = pQuery->current->lastKey;
-  }
-
+  if (QUERY_IS_ASC_QUERY(pQuery)) {
     assert(*start <= pQuery->current->lastKey);
+  } else {
+    assert(*start >= pQuery->current->lastKey);
+  }
 
   // if queried with value filter, do NOT forward query start position
   if (pQuery->limit.offset <= 0 || pQuery->numOfFilterCols > 0 || pRuntimeEnv->pTsBuf != NULL || pRuntimeEnv->pFillInfo != NULL) {
@@ -4561,6 +4471,7 @@ static bool skipTimeInterval(SQueryRuntimeEnv *pRuntimeEnv, TSKEY* start) {
   assert(pRuntimeEnv->windowResInfo.prevSKey == TSKEY_INITIAL_VAL);
 
   STimeWindow w = TSWINDOW_INITIALIZER;
+  bool ascQuery = QUERY_IS_ASC_QUERY(pQuery);
 
   SResultRowInfo *pWindowResInfo = &pRuntimeEnv->windowResInfo;
   STableQueryInfo *pTableQueryInfo = pQuery->current;
@@ -4585,19 +4496,25 @@ static bool skipTimeInterval(SQueryRuntimeEnv *pRuntimeEnv, TSKEY* start) {
     while (pQuery->limit.offset > 0) {
       STimeWindow tw = win;
 
-      if ((win.ekey <= blockInfo.window.ekey && QUERY_IS_ASC_QUERY(pQuery)) ||
-          (win.ekey >= blockInfo.window.skey && !QUERY_IS_ASC_QUERY(pQuery))) {
+      if ((win.ekey <= blockInfo.window.ekey && ascQuery) || (win.ekey >= blockInfo.window.skey && !ascQuery)) {
         pQuery->limit.offset -= 1;
         pWindowResInfo->prevSKey = win.skey;
+
+        // current time window is aligned with blockInfo.window.ekey
+        // restart it from next data block by set prevSKey to be TSKEY_INITIAL_VAL;
+        if ((win.ekey == blockInfo.window.ekey && ascQuery) || (win.ekey == blockInfo.window.skey && !ascQuery)) {
+          pWindowResInfo->prevSKey = TSKEY_INITIAL_VAL;
+        }
       }
 
-      // current window does not ended in current data block, try next data block
-      getNextTimeWindow(pQuery, &tw);
       if (pQuery->limit.offset == 0) {
         *start = doSkipIntervalProcess(pRuntimeEnv, &win, &blockInfo, pTableQueryInfo);
         return true;
       }
 
+      // current window does not ended in current data block, try next data block
+      getNextTimeWindow(pQuery, &tw);
+
       /*
        * If the next time window still starts from current data block,
        * load the primary timestamp column first, and then find the start position for the next queried time window.
@@ -4605,13 +4522,12 @@ static bool skipTimeInterval(SQueryRuntimeEnv *pRuntimeEnv, TSKEY* start) {
        * TODO: Optimize for this cases. All data blocks are not needed to be loaded, only if the first actually required
        * time window resides in current data block.
        */
-      if ((tw.skey <= blockInfo.window.ekey && QUERY_IS_ASC_QUERY(pQuery)) ||
-          (tw.ekey >= blockInfo.window.skey && !QUERY_IS_ASC_QUERY(pQuery))) {
-        SArray *         pDataBlock = tsdbRetrieveDataBlock(pRuntimeEnv->pQueryHandle, NULL);
+      if ((tw.skey <= blockInfo.window.ekey && ascQuery) || (tw.ekey >= blockInfo.window.skey && !ascQuery)) {
+
+        SArray *pDataBlock = tsdbRetrieveDataBlock(pRuntimeEnv->pQueryHandle, NULL);
         SColumnInfoData *pColInfoData = taosArrayGet(pDataBlock, 0);
 
-        if ((win.ekey > blockInfo.window.ekey && QUERY_IS_ASC_QUERY(pQuery)) ||
-            (win.ekey < blockInfo.window.skey && !QUERY_IS_ASC_QUERY(pQuery))) {
+        if ((win.ekey > blockInfo.window.ekey && ascQuery) || (win.ekey < blockInfo.window.skey && !ascQuery)) {
           pQuery->limit.offset -= 1;
         }
 
@@ -4754,7 +4670,7 @@ int32_t doInitQInfo(SQInfo *pQInfo, STSBuf *pTsBuf, void *tsdb, int32_t vgId, bo
   pRuntimeEnv->cur.vgroupIndex = -1;
   pRuntimeEnv->stableQuery = isSTableQuery;
   pRuntimeEnv->prevGroupId = INT32_MIN;
-  pRuntimeEnv->groupbyNormalCol = isGroupbyNormalCol(pQuery->pGroupbyExpr);
+  pRuntimeEnv->groupbyColumn = isGroupbyNormalCol(pQuery->pGroupbyExpr);
 
   if (pTsBuf != NULL) {
     int16_t order = (pQuery->order.order == pRuntimeEnv->pTsBuf->tsOrder) ? TSDB_ORDER_ASC : TSDB_ORDER_DESC;
@@ -4774,7 +4690,7 @@ int32_t doInitQInfo(SQInfo *pQInfo, STSBuf *pTsBuf, void *tsdb, int32_t vgId, bo
 
     if (!QUERY_IS_INTERVAL_QUERY(pQuery)) {
       int16_t type = TSDB_DATA_TYPE_NULL;
-      if (pRuntimeEnv->groupbyNormalCol) {  // group by columns not tags;
+      if (pRuntimeEnv->groupbyColumn) {  // group by columns not tags;
         type = getGroupbyColumnType(pQuery, pQuery->pGroupbyExpr);
       } else {
         type = TSDB_DATA_TYPE_INT;  // group id
@@ -4785,7 +4701,7 @@ int32_t doInitQInfo(SQInfo *pQInfo, STSBuf *pTsBuf, void *tsdb, int32_t vgId, bo
         return code;
       }
     }
-  } else if (pRuntimeEnv->groupbyNormalCol || QUERY_IS_INTERVAL_QUERY(pQuery) || (!isSTableQuery)) {
+  } else if (pRuntimeEnv->groupbyColumn || QUERY_IS_INTERVAL_QUERY(pQuery) || (!isSTableQuery)) {
     int32_t numOfResultRows = getInitialPageNum(pQInfo);
     getIntermediateBufInfo(pRuntimeEnv, &ps, &rowsize);
     code = createDiskbasedResultBuffer(&pRuntimeEnv->pResultBuf, rowsize, ps, TENMB, pQInfo);
@@ -4794,7 +4710,7 @@ int32_t doInitQInfo(SQInfo *pQInfo, STSBuf *pTsBuf, void *tsdb, int32_t vgId, bo
     }
 
     int16_t type = TSDB_DATA_TYPE_NULL;
-    if (pRuntimeEnv->groupbyNormalCol) {
+    if (pRuntimeEnv->groupbyColumn) {
       type = getGroupbyColumnType(pQuery, pQuery->pGroupbyExpr);
     } else {
       type = TSDB_DATA_TYPE_TIMESTAMP;
@@ -4900,7 +4816,7 @@ static int64_t scanMultiTableDataBlocks(SQInfo *pQInfo) {
     pQuery->current = *pTableQueryInfo;
     doTableQueryInfoTimeWindowCheck(pQuery, *pTableQueryInfo);
 
-    if (!pRuntimeEnv->groupbyNormalCol) {
+    if (!pRuntimeEnv->groupbyColumn) {
       setEnvForEachBlock(pQInfo, *pTableQueryInfo, &blockInfo);
     }
 
@@ -5158,7 +5074,7 @@ static void sequentialTableProcess(SQInfo *pQInfo) {
         break;
       }
     }
-  } else if (pRuntimeEnv->groupbyNormalCol) {  // group-by on normal columns query
+  } else if (pRuntimeEnv->groupbyColumn) {  // group-by on normal columns query
     while (pQInfo->groupIndex < numOfGroups) {
       SArray *group = taosArrayGetP(pQInfo->tableGroupInfo.pGroupList, pQInfo->groupIndex);
 
@@ -5488,7 +5404,7 @@ static void sequentialTableProcess(SQInfo *pQInfo) {
   }
 }
 
-static void doSaveContext(SQInfo *pQInfo) {
+static int32_t doSaveContext(SQInfo *pQInfo) {
   SQueryRuntimeEnv *pRuntimeEnv = &pQInfo->runtimeEnv;
   SQuery *          pQuery = pRuntimeEnv->pQuery;
 
@@ -5514,9 +5430,7 @@ static void doSaveContext(SQInfo *pQInfo) {
 
   pRuntimeEnv->prevGroupId = INT32_MIN;
   pRuntimeEnv->pSecQueryHandle = tsdbQueryTables(pQInfo->tsdb, &cond, &pQInfo->tableGroupInfo, pQInfo, &pQInfo->memRef);
-  if (pRuntimeEnv->pSecQueryHandle == NULL) {
-    longjmp(pRuntimeEnv->env, terrno);
-  }
+  return (pRuntimeEnv->pSecQueryHandle == NULL)? -1:0;
 }
 
 static void doRestoreContext(SQInfo *pQInfo) {
@@ -5555,7 +5469,7 @@ static void doCloseAllTimeWindowAfterScan(SQInfo *pQInfo) {
 
 static void multiTableQueryProcess(SQInfo *pQInfo) {
   SQueryRuntimeEnv *pRuntimeEnv = &pQInfo->runtimeEnv;
-  SQuery *          pQuery = pRuntimeEnv->pQuery;
+  SQuery           *pQuery = pRuntimeEnv->pQuery;
 
   if (pQInfo->groupIndex > 0) {
     /*
@@ -5589,12 +5503,14 @@ static void multiTableQueryProcess(SQInfo *pQInfo) {
   doCloseAllTimeWindowAfterScan(pQInfo);
 
   if (needReverseScan(pQuery)) {
-    doSaveContext(pQInfo);
-
+    int32_t code = doSaveContext(pQInfo);
+    if (code == TSDB_CODE_SUCCESS) {
       el = scanMultiTableDataBlocks(pQInfo);
       qDebug("QInfo:%p reversed scan completed, elapsed time: %" PRId64 "ms", pQInfo, el);
-
       doRestoreContext(pQInfo);
+    } else {
+      pQInfo->code = code;
+    }
   } else {
     qDebug("QInfo:%p no need to do reversed scan, query completed", pQInfo);
   }
@@ -5608,12 +5524,15 @@ static void multiTableQueryProcess(SQInfo *pQInfo) {
   }
 
   if (QUERY_IS_INTERVAL_QUERY(pQuery) || isSumAvgRateQuery(pQuery)) {
-    if (mergeGroupResult(pQInfo) == TSDB_CODE_SUCCESS) {
+    int32_t code = mergeGroupResult(pQInfo);
+    if (code == TSDB_CODE_SUCCESS) {
       copyResToQueryResultBuf(pQInfo, pQuery);
 
 #ifdef _DEBUG_VIEW
       displayInterResult(pQuery->sdata, pRuntimeEnv, pQuery->sdata[0]->num);
 #endif
+    } else {  // set the error code
+      pQInfo->code = code;
     }
   } else {  // not a interval query
     copyFromWindowResToSData(pQInfo, &pRuntimeEnv->windowResInfo);
@@ -5623,7 +5542,6 @@ static void multiTableQueryProcess(SQInfo *pQInfo) {
   qDebug("QInfo:%p points returned:%" PRId64 ", total:%" PRId64, pQInfo, pQuery->rec.rows, pQuery->rec.total + pQuery->rec.rows);
 }
 
-
 static char *getArithemicInputSrc(void *param, const char *name, int32_t colId) {
   SArithmeticSupport *pSupport = (SArithmeticSupport *) param;
   SExprInfo* pExprInfo = (SExprInfo*) pSupport->exprList;
@@ -5677,7 +5595,7 @@ static void doSecondaryArithmeticProcess(SQuery* pQuery) {
       }
     } else {
       arithSup.pArithExpr = pExpr;
-      tExprTreeCalcTraverse(arithSup.pArithExpr->pExpr, (int32_t)pQuery->rec.rows, data[i]->data, &arithSup, TSDB_ORDER_ASC,
+      arithmeticTreeTraverse(arithSup.pArithExpr->pExpr, (int32_t)pQuery->rec.rows, data[i]->data, &arithSup, TSDB_ORDER_ASC,
                             getArithemicInputSrc);
     }
   }
@@ -5700,7 +5618,7 @@ static void doSecondaryArithmeticProcess(SQuery* pQuery) {
  * select count(*)/top(field,k)/avg(field name) from table_name [where ts>now-1a];
  * select count(*) from table_name group by status_column;
  */
-static void tableFixedOutputProcess(SQInfo *pQInfo, STableQueryInfo* pTableInfo) {
+static void tableAggregationProcess(SQInfo *pQInfo, STableQueryInfo* pTableInfo) {
   SQueryRuntimeEnv *pRuntimeEnv = &pQInfo->runtimeEnv;
 
   SQuery *pQuery = pRuntimeEnv->pQuery;
@@ -5725,7 +5643,7 @@ static void tableFixedOutputProcess(SQInfo *pQInfo, STableQueryInfo* pTableInfo)
   limitResults(pRuntimeEnv);
 }
 
-static void tableMultiOutputProcess(SQInfo *pQInfo, STableQueryInfo* pTableInfo) {
+static void tableProjectionProcess(SQInfo *pQInfo, STableQueryInfo* pTableInfo) {
   SQueryRuntimeEnv *pRuntimeEnv = &pQInfo->runtimeEnv;
 
   SQuery *pQuery = pRuntimeEnv->pQuery;
@@ -5787,10 +5705,10 @@ static void tableIntervalProcess(SQInfo *pQInfo, STableQueryInfo* pTableInfo) {
   SQuery *pQuery = pRuntimeEnv->pQuery;
   pQuery->current = pTableInfo;
 
-  TSKEY newStartKey = TSKEY_INITIAL_VAL;
+  TSKEY newStartKey = QUERY_IS_ASC_QUERY(pQuery)? INT64_MIN:INT64_MAX;
 
   // skip blocks without load the actual data block from file if no filter condition present
-  if (!pRuntimeEnv->groupbyNormalCol) {
+  if (!pRuntimeEnv->groupbyColumn) {
     skipTimeInterval(pRuntimeEnv, &newStartKey);
     if (pQuery->limit.offset > 0 && pQuery->numOfFilterCols == 0 && pRuntimeEnv->pFillInfo == NULL) {
       setQueryStatus(pQuery, QUERY_COMPLETED);
@@ -5806,7 +5724,8 @@ static void tableIntervalProcess(SQInfo *pQInfo, STableQueryInfo* pTableInfo) {
   // skip offset result rows
   pQuery->rec.rows = 0;
 
-  if (pQuery->fillType == TSDB_FILL_NONE) {
+  // not fill or no result generated during this query
+  if (pQuery->fillType == TSDB_FILL_NONE || pRuntimeEnv->windowResInfo.size == 0) {
     // all data scanned, the group by normal column can return
     int32_t numOfClosed = numOfClosedResultRows(&pRuntimeEnv->windowResInfo);
     if (pQuery->limit.offset > numOfClosed) {
@@ -5886,13 +5805,13 @@ static void tableQueryImpl(SQInfo *pQInfo) {
   STableQueryInfo* item = taosArrayGetP(g, 0);
 
   // group by normal column, sliding window query, interval query are handled by interval query processor
-  if (QUERY_IS_INTERVAL_QUERY(pQuery) || pRuntimeEnv->groupbyNormalCol) {  // interval (down sampling operation)
+  if (QUERY_IS_INTERVAL_QUERY(pQuery) || pRuntimeEnv->groupbyColumn) {  // interval (down sampling operation)
     tableIntervalProcess(pQInfo, item);
   } else if (isFixedOutputQuery(pRuntimeEnv)) {
-    tableFixedOutputProcess(pQInfo, item);
+    tableAggregationProcess(pQInfo, item);
   } else {  // diff/add/multiply/subtract/division
     assert(pQuery->checkBuffer == 1);
-    tableMultiOutputProcess(pQInfo, item);
+    tableProjectionProcess(pQInfo, item);
   }
 
   // record the total elapsed time
@@ -5908,11 +5827,11 @@ static void stableQueryImpl(SQInfo *pQInfo) {
   int64_t st = taosGetTimestampUs();
 
   if (QUERY_IS_INTERVAL_QUERY(pQuery) ||
-      (isFixedOutputQuery(pRuntimeEnv) && (!isPointInterpoQuery(pQuery)) && (!pRuntimeEnv->groupbyNormalCol))) {
+      (isFixedOutputQuery(pRuntimeEnv) && (!isPointInterpoQuery(pQuery)) && (!pRuntimeEnv->groupbyColumn))) {
     multiTableQueryProcess(pQInfo);
   } else {
     assert((pQuery->checkBuffer == 1 && pQuery->interval.interval == 0) || isPointInterpoQuery(pQuery) ||
-            isFirstLastRowQuery(pQuery) || pRuntimeEnv->groupbyNormalCol);
+           pRuntimeEnv->groupbyColumn);
 
     sequentialTableProcess(pQInfo);
   }
@@ -5948,13 +5867,13 @@ static int32_t getColumnIndexInSource(SQueryTableMsg *pQueryMsg, SSqlFuncMsg *pE
       j += 1;
     }
   }
-  assert(0);
-  return -1;
+
+  return INT32_MIN;  // return a less than TSDB_TBNAME_COLUMN_INDEX value
 }
 
 bool validateExprColumnInfo(SQueryTableMsg *pQueryMsg, SSqlFuncMsg *pExprMsg, SColumnInfo* pTagCols) {
   int32_t j = getColumnIndexInSource(pQueryMsg, pExprMsg, pTagCols);
-  return j < pQueryMsg->numOfCols || j < pQueryMsg->numOfTags;
+  return j != INT32_MIN;
 }
 
 static bool validateQueryMsg(SQueryTableMsg *pQueryMsg) {
@@ -5981,12 +5900,14 @@ static bool validateQueryMsg(SQueryTableMsg *pQueryMsg) {
   return true;
 }
 
-static bool validateQuerySourceCols(SQueryTableMsg *pQueryMsg, SSqlFuncMsg** pExprMsg) {
+static bool validateQuerySourceCols(SQueryTableMsg *pQueryMsg, SSqlFuncMsg** pExprMsg, SColumnInfo* pTagCols) {
   int32_t numOfTotal = pQueryMsg->numOfCols + pQueryMsg->numOfTags;
   if (pQueryMsg->numOfCols < 0 || pQueryMsg->numOfTags < 0 || numOfTotal > TSDB_MAX_COLUMNS) {
     qError("qmsg:%p illegal value of numOfCols %d numOfTags:%d", pQueryMsg, pQueryMsg->numOfCols, pQueryMsg->numOfTags);
     return false;
-  } else if (numOfTotal == 0) {
+  }
+
+  if (numOfTotal == 0) {
     for(int32_t i = 0; i < pQueryMsg->numOfOutput; ++i) {
       SSqlFuncMsg* pFuncMsg = pExprMsg[i];
 
@@ -6000,6 +5921,12 @@ static bool validateQuerySourceCols(SQueryTableMsg *pQueryMsg, SSqlFuncMsg** pEx
     }
   }
 
+  for(int32_t i = 0; i < pQueryMsg->numOfOutput; ++i) {
+    if (!validateExprColumnInfo(pQueryMsg, pExprMsg[i], pTagCols)) {
+      return TSDB_CODE_QRY_INVALID_MSG;
+    }
+  }
+
   return true;
 }
 
@@ -6031,7 +5958,7 @@ static char *createTableIdList(SQueryTableMsg *pQueryMsg, char *pMsg, SArray **p
  * @return
  */
 static int32_t convertQueryMsg(SQueryTableMsg *pQueryMsg, SArray **pTableIdList, SSqlFuncMsg ***pExpr, SSqlFuncMsg ***pSecStageExpr,
-                               char **tagCond, char** tbnameCond, SColIndex **groupbyCols, SColumnInfo** tagCols) {
+                               char **tagCond, char** tbnameCond, SColIndex **groupbyCols, SColumnInfo** tagCols, char** sql) {
   int32_t code = TSDB_CODE_SUCCESS;
 
   if (taosCheckVersion(pQueryMsg->version, version, 3) != 0) {
@@ -6063,7 +5990,9 @@ static int32_t convertQueryMsg(SQueryTableMsg *pQueryMsg, SArray **pTableIdList,
   pQueryMsg->tsNumOfBlocks = htonl(pQueryMsg->tsNumOfBlocks);
   pQueryMsg->tsOrder = htonl(pQueryMsg->tsOrder);
   pQueryMsg->numOfTags = htonl(pQueryMsg->numOfTags);
+  pQueryMsg->tbnameCondLen = htonl(pQueryMsg->tbnameCondLen);
   pQueryMsg->secondStageOutput = htonl(pQueryMsg->secondStageOutput);
+  pQueryMsg->sqlstrLen = htonl(pQueryMsg->sqlstrLen);
 
   // query msg safety check
   if (!validateQueryMsg(pQueryMsg)) {
@@ -6158,20 +6087,11 @@ static int32_t convertQueryMsg(SQueryTableMsg *pQueryMsg, SArray **pTableIdList,
         code = TSDB_CODE_QRY_INVALID_MSG;
         goto _cleanup;
       }
-    } else {
-//      if (!validateExprColumnInfo(pQueryMsg, pExprMsg)) {
-//        return TSDB_CODE_QRY_INVALID_MSG;
-//      }
     }
 
     pExprMsg = (SSqlFuncMsg *)pMsg;
   }
 
-  if (!validateQuerySourceCols(pQueryMsg, *pExpr)) {
-    code = TSDB_CODE_QRY_INVALID_MSG;
-    goto _cleanup;
-  }
-
   if (pQueryMsg->secondStageOutput) {
     pExprMsg = (SSqlFuncMsg *)pMsg;
     *pSecStageExpr = calloc(pQueryMsg->secondStageOutput, POINTER_BYTES);
@@ -6205,10 +6125,6 @@ static int32_t convertQueryMsg(SQueryTableMsg *pQueryMsg, SArray **pTableIdList,
           code = TSDB_CODE_QRY_INVALID_MSG;
           goto _cleanup;
         }
-      } else {
-//      if (!validateExprColumnInfo(pQueryMsg, pExprMsg)) {
-//        return TSDB_CODE_QRY_INVALID_MSG;
-//      }
       }
 
       pExprMsg = (SSqlFuncMsg *)pMsg;
@@ -6287,17 +6203,22 @@ static int32_t convertQueryMsg(SQueryTableMsg *pQueryMsg, SArray **pTableIdList,
     pMsg += pQueryMsg->tagCondLen;
   }
 
-  if (*pMsg != 0) {
-    size_t len = strlen(pMsg) + 1;
-
-    *tbnameCond = malloc(len);
+  if (pQueryMsg->tbnameCondLen > 0) {
+    *tbnameCond = calloc(1, pQueryMsg->tbnameCondLen + 1);
     if (*tbnameCond == NULL) {
       code = TSDB_CODE_QRY_OUT_OF_MEMORY;
       goto _cleanup;
     }
 
-    strcpy(*tbnameCond, pMsg);
-    pMsg += len;
+    strncpy(*tbnameCond, pMsg, pQueryMsg->tbnameCondLen);
+    pMsg += pQueryMsg->tbnameCondLen;
+  }
+
+  *sql = strndup(pMsg, pQueryMsg->sqlstrLen);
+
+  if (!validateQuerySourceCols(pQueryMsg, *pExpr, *tagCols)) {
+    code = TSDB_CODE_QRY_INVALID_MSG;
+    goto _cleanup;
   }
 
   qDebug("qmsg:%p query %d tables, type:%d, qrange:%" PRId64 "-%" PRId64 ", numOfGroupbyTagCols:%d, order:%d, "
@@ -6306,6 +6227,7 @@ static int32_t convertQueryMsg(SQueryTableMsg *pQueryMsg, SArray **pTableIdList,
          pQueryMsg->order, pQueryMsg->numOfOutput, pQueryMsg->numOfCols, pQueryMsg->interval.interval,
          pQueryMsg->fillType, pQueryMsg->tsLen, pQueryMsg->tsNumOfBlocks, pQueryMsg->limit, pQueryMsg->offset);
 
+  qDebug("qmsg:%p, sql:%s", pQueryMsg, *sql);
   return TSDB_CODE_SUCCESS;
 
 _cleanup:
@@ -6316,6 +6238,7 @@ _cleanup:
   tfree(*groupbyCols);
   tfree(*tagCols);
   tfree(*tagCond);
+  tfree(*sql);
 
   return code;
 }
@@ -6388,7 +6311,15 @@ static int32_t createQueryFuncExprFromMsg(SQueryTableMsg *pQueryMsg, int32_t num
       }
     } else {
       int32_t j = getColumnIndexInSource(pQueryMsg, &pExprs[i].base, pTagCols);
-      assert(j < pQueryMsg->numOfCols || j < pQueryMsg->numOfTags);
+      if (TSDB_COL_IS_TAG(pExprs[i].base.colInfo.flag)) {
+        if (j < TSDB_TBNAME_COLUMN_INDEX || j >= pQueryMsg->numOfTags) {
+          return TSDB_CODE_QRY_INVALID_MSG;
+        }
+      } else {
+        if (j < PRIMARYKEY_TIMESTAMP_COL_INDEX || j >= pQueryMsg->numOfCols) {
+          return TSDB_CODE_QRY_INVALID_MSG;
+        }
+      }
 
       if (pExprs[i].base.colInfo.colId != TSDB_TBNAME_COLUMN_INDEX && j >= 0) {
         SColumnInfo* pCol = (TSDB_COL_IS_TAG(pExprs[i].base.colInfo.flag))? &pTagCols[j]:&pQueryMsg->colList[j];
@@ -6412,6 +6343,7 @@ static int32_t createQueryFuncExprFromMsg(SQueryTableMsg *pQueryMsg, int32_t num
     if (pExprs[i].base.functionId == TSDB_FUNC_TAG_DUMMY || pExprs[i].base.functionId == TSDB_FUNC_TS_DUMMY) {
       tagLen += pExprs[i].bytes;
     }
+
     assert(isValidDataType(pExprs[i].type));
   }
 
@@ -6614,7 +6546,7 @@ static void calResultBufSize(SQuery* pQuery) {
 }
 
 static SQInfo *createQInfoImpl(SQueryTableMsg *pQueryMsg, SSqlGroupbyExpr *pGroupbyExpr, SExprInfo *pExprs,
-                               SExprInfo *pSecExprs, STableGroupInfo *pTableGroupInfo, SColumnInfo* pTagCols, bool stableQuery) {
+                               SExprInfo *pSecExprs, STableGroupInfo *pTableGroupInfo, SColumnInfo* pTagCols, bool stableQuery, char* sql) {
   int16_t numOfCols = pQueryMsg->numOfCols;
   int16_t numOfOutput = pQueryMsg->numOfOutput;
 
@@ -6745,6 +6677,7 @@ static SQInfo *createQInfoImpl(SQueryTableMsg *pQueryMsg, SSqlGroupbyExpr *pGrou
   pQInfo->arrTableIdInfo = taosHashInit(pTableGroupInfo->numOfTables, taosGetDefaultHashFunction(TSDB_DATA_TYPE_INT), false, HASH_NO_LOCK);
   pQInfo->dataReady = QUERY_RESULT_NOT_READY;
   pQInfo->rspContext = NULL;
+  pQInfo->sql = sql;
   pthread_mutex_init(&pQInfo->lock, NULL);
   tsem_init(&pQInfo->ready, 0, 0);
 
@@ -6988,6 +6921,8 @@ static void freeQInfo(SQInfo *pQInfo) {
   doDestroyTableQueryInfo(&pQInfo->tableqinfoGroupInfo);
 
   tfree(pQInfo->pBuf);
+  tfree(pQInfo->sql);
+
   tsdbDestroyTableGroup(&pQInfo->tableGroupInfo);
   taosHashCleanup(pQInfo->arrTableIdInfo);
 
@@ -7086,6 +7021,7 @@ int32_t qCreateQueryInfo(void* tsdb, int32_t vgId, SQueryTableMsg* pQueryMsg, qi
 
   int32_t code = TSDB_CODE_SUCCESS;
 
+  char            *sql          = NULL;
   char            *tagCond      = NULL;
   char            *tbnameCond   = NULL;
   SArray          *pTableIdList = NULL;
@@ -7098,7 +7034,7 @@ int32_t qCreateQueryInfo(void* tsdb, int32_t vgId, SQueryTableMsg* pQueryMsg, qi
   SColumnInfo     *pTagColumnInfo = NULL;
   SSqlGroupbyExpr *pGroupbyExpr   = NULL;
 
-  code = convertQueryMsg(pQueryMsg, &pTableIdList, &pExprMsg, &pSecExprMsg, &tagCond, &tbnameCond, &pGroupColIndex, &pTagColumnInfo);
+  code = convertQueryMsg(pQueryMsg, &pTableIdList, &pExprMsg, &pSecExprMsg, &tagCond, &tbnameCond, &pGroupColIndex, &pTagColumnInfo, &sql);
   if (code != TSDB_CODE_SUCCESS) {
     goto _over;
   }
@@ -7182,8 +7118,9 @@ int32_t qCreateQueryInfo(void* tsdb, int32_t vgId, SQueryTableMsg* pQueryMsg, qi
     goto _over;
   }
 
-  (*pQInfo) = createQInfoImpl(pQueryMsg, pGroupbyExpr, pExprs, pSecExprs, &tableGroupInfo, pTagColumnInfo, isSTableQuery);
+  (*pQInfo) = createQInfoImpl(pQueryMsg, pGroupbyExpr, pExprs, pSecExprs, &tableGroupInfo, pTagColumnInfo, isSTableQuery, sql);
 
+  sql    = NULL;
   pExprs = NULL;
   pSecExprs = NULL;
   pGroupbyExpr = NULL;
@@ -7207,6 +7144,7 @@ _over:
   }
 
   free(pTagColumnInfo);
+  free(sql);
   free(pExprs);
   free(pSecExprs);
 

src/query/src/qHistogram.c

@@ -174,7 +174,7 @@ int32_t tHistogramAdd(SHistogramInfo** pHisto, double val) {
       }
 
       assert((*pHisto)->elems[idx].val > val);
-    } else {
+    } else if ((*pHisto)->numOfElems > 0) {
       assert((*pHisto)->elems[(*pHisto)->numOfEntries].val < val);
     }
 

src/sync/src/syncRetrieve.c

@@ -503,9 +503,10 @@ void *syncRetrieveData(void *param) {
   taosClose(pPeer->syncFd);
 
   // The ref is obtained in both the create thread and the current thread, so it is released twice
+  sInfo("%s, sync retrieve data over, sstatus:%s", pPeer->id, syncStatus[pPeer->sstatus]);
+
   syncReleasePeer(pPeer);
   syncReleasePeer(pPeer);
 
-  sInfo("%s, sync retrieve data over, sstatus:%s", pPeer->id, syncStatus[pPeer->sstatus]);
   return NULL;
 }

src/tsdb/src/tsdbRead.c

@@ -2645,13 +2645,12 @@ SArray* createTableGroup(SArray* pTableList, STSchema* pTagSchema, SColIndex* pC
   return pTableGroup;
 }
 
-static bool indexedNodeFilterFp(const void* pNode, void* param) {
+static bool tableFilterFp(const void* pNode, void* param) {
   tQueryInfo* pInfo = (tQueryInfo*) param;
 
   STable* pTable = (STable*)(SL_GET_NODE_DATA((SSkipListNode*)pNode));
 
   char* val = NULL;
-
   if (pInfo->sch.colId == TSDB_TBNAME_COLUMN_INDEX) {
     val = (char*) TABLE_NAME(pTable);
   } else {
@@ -2706,15 +2705,17 @@ static bool indexedNodeFilterFp(const void* pNode, void* param) {
   return true;
 }
 
+static void getTableListfromSkipList(tExprNode *pExpr, SSkipList *pSkipList, SArray *result, SExprTraverseSupp *param);
+
 static int32_t doQueryTableList(STable* pSTable, SArray* pRes, tExprNode* pExpr) {
   // query according to the expression tree
   SExprTraverseSupp supp = {
-      .nodeFilterFn = (__result_filter_fn_t) indexedNodeFilterFp,
+      .nodeFilterFn = (__result_filter_fn_t) tableFilterFp,
       .setupInfoFn = filterPrepare,
       .pExtInfo = pSTable->tagSchema,
       };
 
-  tExprTreeTraverse(pExpr, pSTable->pIndex, pRes, &supp);
+  getTableListfromSkipList(pExpr, pSTable->pIndex, pRes, &supp);
   tExprTreeDestroy(&pExpr, destroyHelper);
   return TSDB_CODE_SUCCESS;
 }
@@ -2956,3 +2957,235 @@ void tsdbDestroyTableGroup(STableGroupInfo *pGroupList) {
   taosArrayDestroy(pGroupList->pGroupList);
   pGroupList->numOfTables = 0;
 }
+
+static void applyFilterToSkipListNode(SSkipList *pSkipList, tExprNode *pExpr, SArray *pResult, SExprTraverseSupp *param) {
+  SSkipListIterator* iter = tSkipListCreateIter(pSkipList);
+
+  // Scan each node in the skiplist by using iterator
+  while (tSkipListIterNext(iter)) {
+    SSkipListNode *pNode = tSkipListIterGet(iter);
+    if (exprTreeApplayFilter(pExpr, pNode, param)) {
+      taosArrayPush(pResult, &(SL_GET_NODE_DATA(pNode)));
+    }
+  }
+
+  tSkipListDestroyIter(iter);
+}
+
+typedef struct {
+  char*    v;
+  int32_t  optr;
+} SEndPoint;
+
+typedef struct {
+  SEndPoint* start;
+  SEndPoint* end;
+} SQueryCond;
+
+// todo check for malloc failure
+static int32_t setQueryCond(tQueryInfo *queryColInfo, SQueryCond* pCond) {
+  int32_t optr = queryColInfo->optr;
+
+  if (optr == TSDB_RELATION_GREATER || optr == TSDB_RELATION_GREATER_EQUAL ||
+      optr == TSDB_RELATION_EQUAL || optr == TSDB_RELATION_NOT_EQUAL) {
+    pCond->start = calloc(1, sizeof(SEndPoint));
+    pCond->start->optr = queryColInfo->optr;
+    pCond->start->v = queryColInfo->q;
+  } else if (optr == TSDB_RELATION_LESS || optr == TSDB_RELATION_LESS_EQUAL) {
+    pCond->end = calloc(1, sizeof(SEndPoint));
+    pCond->end->optr = queryColInfo->optr;
+    pCond->end->v = queryColInfo->q;
+  } else if (optr == TSDB_RELATION_IN || optr == TSDB_RELATION_LIKE) {
+    assert(0);
+  }
+
+  return TSDB_CODE_SUCCESS;
+}
+
+static void queryIndexedColumn(SSkipList* pSkipList, tQueryInfo* pQueryInfo, SArray* result) {
+  SSkipListIterator* iter = NULL;
+
+  SQueryCond cond = {0};
+  if (setQueryCond(pQueryInfo, &cond) != TSDB_CODE_SUCCESS) {
+    //todo handle error
+  }
+
+  if (cond.start != NULL) {
+    iter = tSkipListCreateIterFromVal(pSkipList, (char*) cond.start->v, pSkipList->type, TSDB_ORDER_ASC);
+  } else {
+    iter = tSkipListCreateIterFromVal(pSkipList, (char*)(cond.end ? cond.end->v: NULL), pSkipList->type, TSDB_ORDER_DESC);
+  }
+
+  if (cond.start != NULL) {
+    int32_t optr = cond.start->optr;
+
+    if (optr == TSDB_RELATION_EQUAL) {   // equals
+      while(tSkipListIterNext(iter)) {
+        SSkipListNode* pNode = tSkipListIterGet(iter);
+
+        int32_t ret = pQueryInfo->compare(SL_GET_NODE_KEY(pSkipList, pNode), cond.start->v);
+        if (ret != 0) {
+          break;
+        }
+
+        STableKeyInfo info = {.pTable = (void*)SL_GET_NODE_DATA(pNode), .lastKey = TSKEY_INITIAL_VAL};
+        taosArrayPush(result, &info);
+      }
+    } else if (optr == TSDB_RELATION_GREATER || optr == TSDB_RELATION_GREATER_EQUAL) { // greater equal
+      bool comp = true;
+      int32_t ret = 0;
+
+      while(tSkipListIterNext(iter)) {
+        SSkipListNode* pNode = tSkipListIterGet(iter);
+
+        if (comp) {
+          ret = pQueryInfo->compare(SL_GET_NODE_KEY(pSkipList, pNode), cond.start->v);
+          assert(ret >= 0);
+        }
+
+        if (ret == 0 && optr == TSDB_RELATION_GREATER) {
+          continue;
+        } else {
+          STableKeyInfo info = {.pTable = (void*)SL_GET_NODE_DATA(pNode), .lastKey = TSKEY_INITIAL_VAL};
+          taosArrayPush(result, &info);
+          comp = false;
+        }
+      }
+    } else if (optr == TSDB_RELATION_NOT_EQUAL) {   // not equal
+      bool comp = true;
+
+      while(tSkipListIterNext(iter)) {
+        SSkipListNode* pNode = tSkipListIterGet(iter);
+        comp = comp && (pQueryInfo->compare(SL_GET_NODE_KEY(pSkipList, pNode), cond.start->v) == 0);
+        if (comp) {
+          continue;
+        }
+
+        STableKeyInfo info = {.pTable = (void*)SL_GET_NODE_DATA(pNode), .lastKey = TSKEY_INITIAL_VAL};
+        taosArrayPush(result, &info);
+      }
+
+      tSkipListDestroyIter(iter);
+
+      comp = true;
+      iter = tSkipListCreateIterFromVal(pSkipList, (char*) cond.start->v, pSkipList->type, TSDB_ORDER_DESC);
+      while(tSkipListIterNext(iter)) {
+        SSkipListNode* pNode = tSkipListIterGet(iter);
+        comp = comp && (pQueryInfo->compare(SL_GET_NODE_KEY(pSkipList, pNode), cond.start->v) == 0);
+        if (comp) {
+          continue;
+        }
+
+        STableKeyInfo info = {.pTable = (void*)SL_GET_NODE_DATA(pNode), .lastKey = TSKEY_INITIAL_VAL};
+        taosArrayPush(result, &info);
+      }
+
+    } else {
+      assert(0);
+    }
+  } else {
+    int32_t optr = cond.end ? cond.end->optr : TSDB_RELATION_INVALID;
+    if (optr == TSDB_RELATION_LESS || optr == TSDB_RELATION_LESS_EQUAL) {
+      bool    comp = true;
+      int32_t ret = 0;
+
+      while (tSkipListIterNext(iter)) {
+        SSkipListNode *pNode = tSkipListIterGet(iter);
+
+        if (comp) {
+          ret = pQueryInfo->compare(SL_GET_NODE_KEY(pSkipList, pNode), cond.end->v);
+          assert(ret <= 0);
+        }
+
+        if (ret == 0 && optr == TSDB_RELATION_LESS) {
+          continue;
+        } else {
+          STableKeyInfo info = {.pTable = (void *)SL_GET_NODE_DATA(pNode), .lastKey = TSKEY_INITIAL_VAL};
+          taosArrayPush(result, &info);
+          comp = false;  // no need to compare anymore
+        }
+      }
+    } else {
+      assert(pQueryInfo->optr == TSDB_RELATION_ISNULL || pQueryInfo->optr == TSDB_RELATION_NOTNULL);
+
+      while (tSkipListIterNext(iter)) {
+        SSkipListNode *pNode = tSkipListIterGet(iter);
+
+        bool isnull = isNull(SL_GET_NODE_KEY(pSkipList, pNode), pQueryInfo->sch.type);
+        if ((pQueryInfo->optr == TSDB_RELATION_ISNULL && isnull) ||
+            (pQueryInfo->optr == TSDB_RELATION_NOTNULL && (!isnull))) {
+          STableKeyInfo info = {.pTable = (void *)SL_GET_NODE_DATA(pNode), .lastKey = TSKEY_INITIAL_VAL};
+          taosArrayPush(result, &info);
+        }
+      }
+    }
+  }
+
+  free(cond.start);
+  free(cond.end);
+  tSkipListDestroyIter(iter);
+}
+
+static void queryIndexlessColumn(SSkipList* pSkipList, tQueryInfo* pQueryInfo, SArray* res, __result_filter_fn_t filterFp) {
+  SSkipListIterator* iter = tSkipListCreateIter(pSkipList);
+
+  while (tSkipListIterNext(iter)) {
+    bool addToResult = false;
+
+    SSkipListNode *pNode = tSkipListIterGet(iter);
+
+    char *pData = SL_GET_NODE_DATA(pNode);
+    tstr *name = (tstr*) tsdbGetTableName((void*) pData);
+
+    // todo speed up by using hash
+    if (pQueryInfo->sch.colId == TSDB_TBNAME_COLUMN_INDEX) {
+      if (pQueryInfo->optr == TSDB_RELATION_IN) {
+        addToResult = pQueryInfo->compare(name, pQueryInfo->q);
+      } else if (pQueryInfo->optr == TSDB_RELATION_LIKE) {
+        addToResult = !pQueryInfo->compare(name, pQueryInfo->q);
+      }
+    } else {
+      addToResult = filterFp(pNode, pQueryInfo);
+    }
+
+    if (addToResult) {
+      STableKeyInfo info = {.pTable = (void*)pData, .lastKey = TSKEY_INITIAL_VAL};
+      taosArrayPush(res, &info);
+    }
+  }
+
+  tSkipListDestroyIter(iter);
+}
+
+// Apply the filter expression to each node in the skiplist to acquire the qualified nodes in skip list
+void getTableListfromSkipList(tExprNode *pExpr, SSkipList *pSkipList, SArray *result, SExprTraverseSupp *param) {
+  if (pExpr == NULL) {
+    return;
+  }
+
+  tExprNode *pLeft  = pExpr->_node.pLeft;
+  tExprNode *pRight = pExpr->_node.pRight;
+
+  // column project
+  if (pLeft->nodeType != TSQL_NODE_EXPR && pRight->nodeType != TSQL_NODE_EXPR) {
+    assert(pLeft->nodeType == TSQL_NODE_COL && (pRight->nodeType == TSQL_NODE_VALUE || pRight->nodeType == TSQL_NODE_DUMMY));
+
+    param->setupInfoFn(pExpr, param->pExtInfo);
+
+    tQueryInfo *pQueryInfo = pExpr->_node.info;
+    if (pQueryInfo->indexed && pQueryInfo->optr != TSDB_RELATION_LIKE) {
+      queryIndexedColumn(pSkipList, pQueryInfo, result);
+    } else {
+      queryIndexlessColumn(pSkipList, pQueryInfo, result, param->nodeFilterFn);
+    }
+
+    return;
+  }
+
+  // The value of hasPK is always 0.
+  uint8_t weight = pLeft->_node.hasPK + pRight->_node.hasPK;
+  assert(weight == 0 && pSkipList != NULL && taosArrayGetSize(result) == 0);
+
+  //apply the hierarchical filter expression to every node in skiplist to find the qualified nodes
+  applyFilterToSkipListNode(pSkipList, pExpr, result, param);
+}

src/util/inc/hash.h

@@ -130,16 +130,14 @@ int32_t taosHashRemoveWithData(SHashObj *pHashObj, const void *key, size_t keyLe
 
 int32_t taosHashCondTraverse(SHashObj *pHashObj, bool (*fp)(void *, void *), void *param);
 
+void taosHashEmpty(SHashObj *pHashObj);
+
 /**
  * clean up hash table
  * @param handle
  */
 void taosHashCleanup(SHashObj *pHashObj);
 
-/*
-void *SHashMutableIterator* taosHashCreateIter(SHashObj *pHashObj, void *);
-*/
-
 /**
  *
  * @param pHashObj

src/util/src/hash.c

@@ -313,10 +313,10 @@ void* taosHashGetClone(SHashObj *pHashObj, const void *key, size_t keyLen, void
     }
 
     if (d != NULL) {
-      memcpy(d, GET_HASH_NODE_DATA(pNode), dsize);
-    } else {
-      data = GET_HASH_NODE_DATA(pNode);
+      memcpy(d, GET_HASH_NODE_DATA(pNode), pNode->dataLen);
     }
+
+    data = GET_HASH_NODE_DATA(pNode);
   }
 
   if (pHashObj->type == HASH_ENTRY_LOCK) {
@@ -472,16 +472,17 @@ int32_t taosHashCondTraverse(SHashObj *pHashObj, bool (*fp)(void *, void *), voi
   return 0;
 }
 
-void taosHashCleanup(SHashObj *pHashObj) {
+void taosHashEmpty(SHashObj *pHashObj) {
   if (pHashObj == NULL) {
     return;
   }
 
+  uDebug("hash:%p cleanup hash table", pHashObj);
+
   SHashNode *pNode, *pNext;
 
   __wr_lock(&pHashObj->lock, pHashObj->type);
 
-  if (pHashObj->hashList) {
   for (int32_t i = 0; i < pHashObj->capacity; ++i) {
     SHashEntry *pEntry = pHashObj->hashList[i];
     if (pEntry->num == 0) {
@@ -498,12 +499,22 @@ void taosHashCleanup(SHashObj *pHashObj) {
 
       pNode = pNext;
     }
-    }
 
-    free(pHashObj->hashList);
+    pEntry->num = 0;
+    pEntry->next = NULL;
   }
 
+  pHashObj->size = 0;
   __wr_unlock(&pHashObj->lock, pHashObj->type);
+}
+
+void taosHashCleanup(SHashObj *pHashObj) {
+  if (pHashObj == NULL) {
+    return;
+  }
+
+  taosHashEmpty(pHashObj);
+  tfree(pHashObj->hashList);
 
   // destroy mem block
   size_t memBlock = taosArrayGetSize(pHashObj->pMemBlock);

src/vnode/src/vnodeWrite.c

@@ -101,8 +101,7 @@ int32_t vnodeProcessWrite(void *vparam, void *wparam, int32_t qtype, void *rpara
   return syncCode;
 }
 
-static int32_t vnodeCheckWrite(void *vparam) {
-  SVnodeObj *pVnode = vparam;
+static int32_t vnodeCheckWrite(SVnodeObj *pVnode) {
   if (!(pVnode->accessState & TSDB_VN_WRITE_ACCCESS)) {
     vDebug("vgId:%d, no write auth, refCount:%d pVnode:%p", pVnode->vgId, pVnode->refCount, pVnode);
     return TSDB_CODE_VND_NO_WRITE_AUTH;
@@ -216,29 +215,21 @@ static int32_t vnodeProcessUpdateTagValMsg(SVnodeObj *pVnode, void *pCont, SRspR
   return TSDB_CODE_SUCCESS;
 }
 
-int32_t vnodeWriteToWQueue(void *vparam, void *wparam, int32_t qtype, void *rparam) {
-  SVnodeObj *pVnode = vparam;
-  SWalHead * pHead = wparam;
-  int32_t    code = 0;
-
-  if (qtype == TAOS_QTYPE_RPC) {
-    code = vnodeCheckWrite(pVnode);
-    if (code != TSDB_CODE_SUCCESS) return code;
-  }
-
+static SVWriteMsg *vnodeBuildVWriteMsg(SVnodeObj *pVnode, SWalHead *pHead, int32_t qtype, SRpcMsg *pRpcMsg) {
   if (pHead->len > TSDB_MAX_WAL_SIZE) {
     vError("vgId:%d, wal len:%d exceeds limit, hver:%" PRIu64, pVnode->vgId, pHead->len, pHead->version);
-    return TSDB_CODE_WAL_SIZE_LIMIT;
+    terrno = TSDB_CODE_WAL_SIZE_LIMIT;
+    return NULL;
   }
 
   int32_t size = sizeof(SVWriteMsg) + sizeof(SWalHead) + pHead->len;
   SVWriteMsg *pWrite = taosAllocateQitem(size);
   if (pWrite == NULL) {
-    return TSDB_CODE_VND_OUT_OF_MEMORY;
+    terrno = TSDB_CODE_VND_OUT_OF_MEMORY;
+    return NULL;
   }
 
-  if (rparam != NULL) {
-    SRpcMsg *pRpcMsg = rparam;
+  if (pRpcMsg != NULL) {
     pWrite->rpcMsg = *pRpcMsg;
   }
 
@@ -248,6 +239,21 @@ int32_t vnodeWriteToWQueue(void *vparam, void *wparam, int32_t qtype, void *rpar
 
   atomic_add_fetch_32(&pVnode->refCount, 1);
 
+  return pWrite;
+}
+
+static int32_t vnodeWriteToWQueueImp(SVWriteMsg *pWrite) {
+  SVnodeObj *pVnode = pWrite->pVnode;
+
+  if (pWrite->qtype == TAOS_QTYPE_RPC) {
+    int32_t code = vnodeCheckWrite(pVnode);
+    if (code != TSDB_CODE_SUCCESS) {
+      taosFreeQitem(pWrite);
+      vnodeRelease(pVnode);
+      return code;
+    }
+  }
+
   int32_t queued = atomic_add_fetch_32(&pVnode->queuedWMsg, 1);
   if (queued > MAX_QUEUED_MSG_NUM) {
     int32_t ms = (queued / MAX_QUEUED_MSG_NUM) * 10 + 3;
@@ -256,15 +262,25 @@ int32_t vnodeWriteToWQueue(void *vparam, void *wparam, int32_t qtype, void *rpar
     taosMsleep(ms);
   }
 
-  code = vnodePerformFlowCtrl(pWrite);
-  if (code != 0) return 0;
-
   vTrace("vgId:%d, write into vwqueue, refCount:%d queued:%d", pVnode->vgId, pVnode->refCount, pVnode->queuedWMsg);
 
-  taosWriteQitem(pVnode->wqueue, qtype, pWrite);
+  taosWriteQitem(pVnode->wqueue, pWrite->qtype, pWrite);
   return TSDB_CODE_SUCCESS;
 }
 
+int32_t vnodeWriteToWQueue(void *vparam, void *wparam, int32_t qtype, void *rparam) {
+  SVWriteMsg *pWrite = vnodeBuildVWriteMsg(vparam, wparam, qtype, rparam);
+  if (pWrite == NULL) {
+    assert(terrno != 0);
+    return terrno;
+  }
+
+  int32_t code = vnodePerformFlowCtrl(pWrite);
+  if (code != 0) return 0;
+
+  return vnodeWriteToWQueueImp(pWrite);
+}
+
 void vnodeFreeFromWQueue(void *vparam, SVWriteMsg *pWrite) {
   SVnodeObj *pVnode = vparam;
 
@@ -294,7 +310,10 @@ static void vnodeFlowCtrlMsgToWQueue(void *param, void *tmrId) {
       vDebug("vgId:%d, msg:%p, write into vwqueue after flowctrl, retry:%d", pVnode->vgId, pWrite,
              pWrite->processedCount);
       pWrite->processedCount = 0;
-      taosWriteQitem(pVnode->wqueue, pWrite->qtype, pWrite);
+      code = vnodeWriteToWQueueImp(pWrite);
+      if (code != 0) {
+        dnodeSendRpcVWriteRsp(pWrite->pVnode, pWrite, code);
+      }
     }
   }
 }

tests/pytest/stream/sys.py

@@ -28,6 +28,7 @@ class TDTestCase:
 
 
     def run(self):
+        time.sleep(5)
         tdSql.execute("use log")
 
         tdSql.execute("create table cpustrm as select count(*), avg(cpu_taosd), max(cpu_taosd), min(cpu_taosd), avg(cpu_system), max(cpu_cores), min(cpu_cores), last(cpu_cores) from log.dn1 interval(4s)")

tests/script/general/parser/first_last.sim

@@ -106,7 +106,7 @@ while $x < 5000
 endw
 
 system sh/exec.sh -n dnode1 -s stop -x SIGINT
-sleep 3000
+sleep 1000
 system sh/exec.sh -n dnode1 -s start
 print ================== server restart completed
 sql connect

tests/script/general/parser/function.sim

@@ -244,3 +244,120 @@ if $data00 != -2.000000000 then
   print expect -2.000000000, actual: $data00
   return -1
 endi
+
+sql create table tm1 (ts timestamp,  k int);
+sql insert into tm1 values('2020-10-30 18:11:56.680', -1000);
+sql insert into tm1 values('2020-11-19 18:11:45.773', NULL);
+sql insert into tm1 values('2020-12-09 18:11:17.098', NULL);
+sql insert into tm1 values('2020-12-20 18:11:49.412', 1);
+sql insert into tm1 values('2020-12-23 18:11:50.412', 2);
+sql insert into tm1 values('2020-12-28 18:11:52.412', 3);
+
+print =====================> td-2610
+sql select twa(k)from tm1 where ts>='2020-11-19 18:11:45.773' and ts<='2020-12-9 18:11:17.098'
+if $rows != 0 then
+ return -1
+endi
+
+print =====================> td-2609
+sql select apercentile(k, 50) from tm1 where ts>='2020-10-30 18:11:56.680' and ts<='2020-12-09 18:11:17.098'
+if $rows != 1 then
+  return -1
+endi
+
+if $data00 != -1000.000000000 then
+  return -1
+endi
+
+system sh/exec.sh -n dnode1 -s stop -x SIGINT
+sleep 1000
+system sh/exec.sh -n dnode1 -s start
+print ================== server restart completed
+sql connect
+sleep 500
+
+sql use m_func_db0
+
+print =====================> td-2583
+sql select min(k) from tm1 where ts>='2020-11-19 18:11:45.773' and ts<='2020-12-20 18:11:49.412'
+if $rows != 1 then
+  return -1
+endi
+
+if $data00 != 1 then
+  print expect 1, actual: $data00
+  return -1
+endi
+
+print =====================> td-2601
+sql select count(*) from tm1 where ts<='2020-6-1 00:00:00' and ts>='2020-1-1 00:00:00' interval(1n) fill(NULL)
+if $rows != 0 then
+  return -1
+endi
+
+print =====================> td-2615
+sql select last(ts) from tm1 interval(17a) limit 776 offset 3
+if $rows != 3 then
+  return -1
+endi
+
+sql select last(ts) from tm1 interval(17a) limit 1000 offset 4
+if $rows != 2 then
+  return -1
+endi
+
+sql select last(ts) from tm1 interval(17a) order by ts desc limit 1000 offset 0
+if $rows != 6 then
+  return -1
+endi
+
+print ==================> td-2624
+sql create table tm2(ts timestamp, k int, b binary(12));
+sql insert into tm2 values('2011-01-02 18:42:45.326',      -1,'abc');
+sql insert into tm2 values('2020-07-30 17:44:06.283',       0, null);
+sql insert into tm2 values('2020-07-30 17:44:19.578', 9999999, null);
+sql insert into tm2 values('2020-07-30 17:46:06.417',    NULL, null);
+sql insert into tm2 values('2020-11-09 18:42:25.538',       0, null);
+sql insert into tm2 values('2020-12-29 17:43:11.641',       0, null);
+sql insert into tm2 values('2020-12-29 18:43:17.129',       0, null);
+sql insert into tm2 values('2020-12-29 18:46:19.109',    NULL, null);
+sql insert into tm2 values('2021-01-03 18:40:40.065',       0, null);
+
+sql select twa(k),first(ts) from tm2 where k <50 interval(17s);
+if $rows != 6 then
+  return -1
+endi
+
+if $data00 != @11-01-02 18:42:42.000@ then
+  return -1
+endi
+
+if $data02 != @11-01-02 18:42:45.326@ then
+  return -1
+endi
+
+if $data10 != @20-07-30 17:43:59.000@ then
+  return -1
+endi
+
+if $data21 != 0.000000000 then
+  return -1
+endi
+
+sql select twa(k),first(ts) from tm2 where k <50 interval(17s) order by ts desc;
+if $rows != 6 then
+  return -1
+endi
+
+sql select twa(k),first(ts),count(k),first(k) from tm2 interval(17s) limit 20 offset 0;
+if $rows != 9 then
+  return -1
+endi
+
+if $data00 != @11-01-02 18:42:42.000@ then
+  return -1
+endi
+
+if $data10 != @20-07-30 17:43:59.000@ then
+  return -1
+endi

tests/script/unique/db/replica_reduce31.sim

@@ -99,9 +99,11 @@ print ========= step2 alter db
 sql_error alter database d1 replica 1
 sql_error alter database d2 replica 1
 sql_error alter database d3 replica 1
+sql_error alter database d4 replica 1
 sql alter database d1 replica 2
 sql alter database d2 replica 2
 sql alter database d3 replica 2
+sql alter database d4 replica 2
 
 $x = 0
 a2: 
@@ -129,9 +131,16 @@ if $data03 != 2 then
 	goto a2
 endi
 
+sql show d4.vgroups 
+print online vnodes $data03
+if $data03 != 2 then
+	goto a2
+endi
+
 sql alter database d1 replica 1
 sql alter database d2 replica 1
 sql alter database d3 replica 1
+sql alter database d4 replica 1
 
 $x = 0
 a1: 
@@ -159,6 +168,27 @@ if $data03 != 1 then
 	goto a1
 endi
 
+sql show d4.vgroups 
+print online vnodes $data03
+if $data03 != 1 then
+	goto a1
+endi
+
+sql show dnodes
+print $data00 $data01 $data02 $data03
+print $data10 $data11 $data12 $data13
+print $data20 $data21 $data22 $data23
+
+if $data02 != 0 then
+	goto a1
+endi
+if $data12 != 2 then
+	goto a1
+endi
+if $data22 != 2 then
+	goto a1
+endi
+
 print ========= step3
 sql reset query cache
 sleep 100
@@ -192,6 +222,7 @@ print ========= step4 alter db
 sql alter database d1 replica 2
 sql alter database d2 replica 2
 sql alter database d3 replica 2
+sql alter database d4 replica 2
 
 $x = 0
 step4: 
@@ -219,6 +250,12 @@ if $data03 != 2 then
 	goto step4
 endi
 
+sql show d4.vgroups 
+print online vnodes $data03
+if $data03 != 2 then
+	goto step4
+endi
+
 sql insert into d1.t1 values(now, 3)
 sql insert into d2.t2 values(now, 3)
 sql insert into d3.t3 values(now, 3)
@@ -287,3 +324,4 @@ sql select * from d4.t4
 system sh/exec.sh -n dnode1 -s stop  -x SIGINT
 system sh/exec.sh -n dnode2 -s stop  -x SIGINT
 system sh/exec.sh -n dnode3 -s stop  -x SIGINT
+system sh/exec.sh -n dnode4 -s stop  -x SIGINT
\ No newline at end of file