merge 3.0

Jenkinsfile2

@@ -4,11 +4,6 @@ import jenkins.model.CauseOfInterruption
 node {
 }
 
-win_test_stage = 0
-linux_ready = 0
-linux_node_ip = ""
-linux_node_pass = ""
-
 def abortPreviousBuilds() {
     def currentJobName = env.JOB_NAME
     def currentBuildNumber = env.BUILD_NUMBER.toInteger()
@@ -84,6 +79,7 @@ def pre_test(){
             git pull >/dev/null
             git log -5
             echo "`date "+%Y%m%d-%H%M%S"` ${JOB_NAME}:${BRANCH_NAME}:${BUILD_ID}:${CHANGE_TARGET}" >>${WKDIR}/jenkins.log
+            echo "CHANGE_BRANCH:${CHANGE_BRANCH}" >>${WKDIR}/jenkins.log
             echo "community log: `git log -5`" >>${WKDIR}/jenkins.log
             git fetch origin +refs/pull/${CHANGE_ID}/merge
             git checkout -qf FETCH_HEAD
@@ -100,6 +96,7 @@ def pre_test(){
             git pull >/dev/null
             git log -5
             echo "`date "+%Y%m%d-%H%M%S"` ${JOB_NAME}:${BRANCH_NAME}:${BUILD_ID}:${CHANGE_TARGET}" >>${WKDIR}/jenkins.log
+            echo "CHANGE_BRANCH:${CHANGE_BRANCH}" >>${WKDIR}/jenkins.log
             echo "tdinternal log: `git log -5`" >>${WKDIR}/jenkins.log
             git fetch origin +refs/pull/${CHANGE_ID}/merge
             git checkout -qf FETCH_HEAD
@@ -289,7 +286,6 @@ def run_win_ctest() {
     '''
 }
 def run_win_test() {
-    echo "LINUX NODE: ${linux_node_ip} - ${linux_node_pass}"
     bat '''
         echo "windows test ..."
         cd %WIN_CONNECTOR_ROOT%
@@ -298,9 +294,8 @@ def run_win_test() {
         ls -l C:\\Windows\\System32\\taos.dll
         time /t
         cd %WIN_SYSTEM_TEST_ROOT%
-        echo "node: ''' + linux_node_ip + ''':''' + linux_node_pass + '''"
         echo "testing ..."
-        test-all.bat  "{\\\"host\\\":\\\"''' + linux_node_ip + '''\\\",\\\"port\\\":22,\\\"user\\\":\\\"root\\\",\\\"password\\\":\\\"''' + linux_node_pass + '''\\\",\\\"path\\\":\\\"/var/lib/jenkins/workspace/TDinternal\\\"}"
+        test-all.bat ci
         time /t
     '''
 }
@@ -331,17 +326,9 @@ pipeline {
                                 pre_test_win()
                                 pre_test_build_win()
                                 run_win_ctest()
-                                script {
-                                    while(linux_ready == 0) {
-                                        sleep(8)
-                                    }
-                                }
                                 run_win_test()
                             }
                         }
-                        script {
-                            win_test_stage = 1
-                        }
                     }
                 }
                 stage('linux test') {
@@ -351,17 +338,6 @@ pipeline {
                         changeRequest()
                     }
                     steps {
-                        script {
-                            linux_node_ip = sh (
-                                script: 'jq .ip /home/node_info.json | sed "s/\\\"//g"',
-                                returnStdout: true
-                            ).trim()
-                            linux_node_pass = sh (
-                                script: 'jq .password /home/node_info.json | sed "s/\\\"//g" |sed "s/\\!/^^^^^^^^\\!/g"',
-                                returnStdout: true
-                            ).trim()
-                            echo "${linux_node_ip}:${linux_node_pass}"
-                        }
                         catchError(buildResult: 'FAILURE', stageResult: 'FAILURE') {
                             timeout(time: 120, unit: 'MINUTES'){
                                 pre_test()
@@ -418,38 +394,9 @@ pipeline {
                                         cd ${WKC}/packaging
                                         ./release.sh -v cluster -n 3.0.0.100 -s static
                                     '''
-                                    sh '''
-                                        echo "install ..."
-                                        cd ${WKC}/release
-                                        tar xzf TDengine-enterprise-server-3.0.0.100-Linux-x64.tar.gz
-                                        cd TDengine-enterprise-server-3.0.0.100
-                                        service taosd stop || :
-                                        rm -rf /var/lib/taos
-                                        ./install.sh -e no
-                                    '''
-                                    sh '''
-                                        echo "checking ..."
-                                        which taos
-                                        which taosd
-                                        rm -rf ${WK}/debug
-                                        mv ${WKC}/debug ${WK}/
-                                    '''
-                                    sh '''
-                                        echo "install taospy ..."
-                                        cd ${WKPY}
-                                        pip3 install .
-                                    '''
                                 }
                             }
                         }
-                        script {
-                            linux_ready = 1
-                        }
-                        script {
-                            while(win_test_stage == 0){
-                                sleep(12)
-                            }
-                        }
                     }
                 }
             }

include/common/tcommon.h

@@ -42,11 +42,13 @@ enum {
 typedef enum EStreamType {
   STREAM_NORMAL = 1,
   STREAM_INVERT,
-  STREAM_REPROCESS,
+  STREAM_CLEAR,
   STREAM_INVALID,
   STREAM_GET_ALL,
   STREAM_DELETE,
   STREAM_RETRIEVE,
+  STREAM_PUSH_DATA,
+  STREAM_PUSH_EMPTY,
 } EStreamType;
 
 typedef struct {

include/common/tdatablock.h

@@ -224,6 +224,7 @@ size_t blockDataGetCapacityInRow(const SSDataBlock* pBlock, size_t pageSize);
 int32_t blockDataTrimFirstNRows(SSDataBlock* pBlock, size_t n);
 
 int32_t      assignOneDataBlock(SSDataBlock* dst, const SSDataBlock* src);
+int32_t copyDataBlock(SSDataBlock* dst, const SSDataBlock* src);
 SSDataBlock* createOneDataBlock(const SSDataBlock* pDataBlock, bool copyData);
 SSDataBlock* createDataBlock();
 int32_t      blockDataAppendColInfo(SSDataBlock* pBlock, SColumnInfoData* pColInfoData);
@@ -236,6 +237,8 @@ void        blockCompressEncode(const SSDataBlock* pBlock, char* data, int32_t*
 const char* blockCompressDecode(SSDataBlock* pBlock, int32_t numOfCols, int32_t numOfRows, const char* pData);
 
 void blockDebugShowData(const SArray* dataBlocks, const char* flag);
+// for debug
+char* dumpBlockData(SSDataBlock* pDataBlock, const char* flag, char** dumpBuf);
 
 int32_t buildSubmitReqFromDataBlock(SSubmitReq** pReq, const SArray* pDataBlocks, STSchema* pTSchema, int32_t vgId,
                                     tb_uid_t suid);

include/libs/qcom/query.h

@@ -16,6 +16,7 @@
 #ifndef _TD_QUERY_H_
 #define _TD_QUERY_H_
 
+// clang-foramt off
 #ifdef __cplusplus
 extern "C" {
 #endif
@@ -206,13 +207,15 @@ int32_t queryCreateTableMetaFromMsg(STableMetaRsp* msg, bool isSuperTable, STabl
 char*   jobTaskStatusStr(int32_t status);
 
 SSchema createSchema(int8_t type, int32_t bytes, col_id_t colId, const char* name);
+
 void    destroyQueryExecRes(SQueryExecRes* pRes);
-int32_t dataConverToStr(char *str, int type, void *buf, int32_t bufSize, int32_t *len);
+int32_t dataConverToStr(char* str, int type, void* buf, int32_t bufSize, int32_t* len);
 char*   parseTagDatatoJson(void* p);
 int32_t cloneTableMeta(STableMeta* pSrc, STableMeta** pDst);
 int32_t cloneDbVgInfo(SDBVgInfo* pSrc, SDBVgInfo** pDst);
 
-extern int32_t (*queryBuildMsg[TDMT_MAX])(void *input, char **msg, int32_t msgSize, int32_t *msgLen, void*(*mallocFp)(int32_t));
+extern int32_t (*queryBuildMsg[TDMT_MAX])(void* input, char** msg, int32_t msgSize, int32_t* msgLen,
+                                          void* (*mallocFp)(int32_t));
 extern int32_t (*queryProcessMsgRsp[TDMT_MAX])(void* output, char* msg, int32_t msgSize);
 
 #define SET_META_TYPE_NULL(t)       (t) = META_TYPE_NULL_TABLE
@@ -231,11 +234,13 @@ extern int32_t (*queryProcessMsgRsp[TDMT_MAX])(void* output, char* msg, int32_t
 #define NEED_CLIENT_HANDLE_ERROR(_code)                                          \
   (NEED_CLIENT_RM_TBLMETA_ERROR(_code) || NEED_CLIENT_REFRESH_VG_ERROR(_code) || \
    NEED_CLIENT_REFRESH_TBLMETA_ERROR(_code))
-#define NEED_CLIENT_RM_TBLMETA_REQ(_type) ((_type) == TDMT_VND_CREATE_TABLE || (_type) == TDMT_VND_CREATE_STB \
+#define NEED_CLIENT_RM_TBLMETA_REQ(_type)                                                                  \
-  || (_type) == TDMT_VND_DROP_TABLE || (_type) == TDMT_VND_DROP_STB)
+  ((_type) == TDMT_VND_CREATE_TABLE || (_type) == TDMT_VND_CREATE_STB || (_type) == TDMT_VND_DROP_TABLE || \
+   (_type) == TDMT_VND_DROP_STB)
 
 #define NEED_SCHEDULER_RETRY_ERROR(_code)                                           \
-  ((_code) == TSDB_CODE_RPC_REDIRECT || (_code) == TSDB_CODE_RPC_NETWORK_UNAVAIL || (_code) == TSDB_CODE_SCH_TIMEOUT_ERROR)
+  ((_code) == TSDB_CODE_RPC_REDIRECT || (_code) == TSDB_CODE_RPC_NETWORK_UNAVAIL || \
+   (_code) == TSDB_CODE_SCH_TIMEOUT_ERROR)
 
 #define REQUEST_TOTAL_EXEC_TIMES 2
 
@@ -312,3 +317,4 @@ extern int32_t (*queryProcessMsgRsp[TDMT_MAX])(void* output, char* msg, int32_t
 #endif
 
 #endif /*_TD_QUERY_H_*/
+       // clang-foramt on

include/libs/stream/tstream.h

@@ -58,6 +58,7 @@ enum {
 enum {
   STREAM_INPUT__DATA_SUBMIT = 1,
   STREAM_INPUT__DATA_BLOCK,
+  STREAM_INPUT__DATA_RETRIEVE,
   STREAM_INPUT__TRIGGER,
   STREAM_INPUT__CHECKPOINT,
   STREAM_INPUT__DROP,
@@ -318,7 +319,7 @@ static FORCE_INLINE int32_t streamTaskInput(SStreamTask* pTask, SStreamQueueItem
       return -1;
     }
     taosWriteQitem(pTask->inputQueue->queue, pSubmitClone);
-  } else if (pItem->type == STREAM_INPUT__DATA_BLOCK) {
+  } else if (pItem->type == STREAM_INPUT__DATA_BLOCK || pItem->type == STREAM_INPUT__DATA_RETRIEVE) {
     taosWriteQitem(pTask->inputQueue->queue, pItem);
   } else if (pItem->type == STREAM_INPUT__CHECKPOINT) {
     taosWriteQitem(pTask->inputQueue->queue, pItem);

include/libs/stream/tstreamUpdate.h

@@ -32,12 +32,15 @@ typedef struct SUpdateInfo {
   int64_t interval;
   int64_t watermark;
   TSKEY minTS;
+  SScalableBf* pCloseWinSBF;
 } SUpdateInfo;
 
 SUpdateInfo *updateInfoInitP(SInterval* pInterval, int64_t watermark);
 SUpdateInfo *updateInfoInit(int64_t interval, int32_t precision, int64_t watermark);
 bool updateInfoIsUpdated(SUpdateInfo *pInfo, tb_uid_t tableId, TSKEY ts);
 void updateInfoDestroy(SUpdateInfo *pInfo);
+void updateInfoAddCloseWindowSBF(SUpdateInfo *pInfo);
+void updateInfoDestoryColseWinSBF(SUpdateInfo *pInfo);
 
 #ifdef __cplusplus
 }

include/libs/transport/trpc.h

@@ -34,10 +34,8 @@ extern int32_t tsRpcHeadSize;
 typedef struct {
   uint32_t clientIp;
   uint16_t clientPort;
-  union {
-    char    user[TSDB_USER_LEN];
   int64_t  applyIndex;
-  };
+  char     user[TSDB_USER_LEN];
 } SRpcConnInfo;
 
 typedef struct SRpcHandleInfo {

include/os/osSocket.h

@@ -157,7 +157,10 @@ int32_t taosNonblockwrite(TdSocketPtr pSocket, char *ptr, int32_t nbytes);
 int64_t taosCopyFds(TdSocketPtr pSrcSocket, TdSocketPtr pDestSocket, int64_t len);
 void    taosWinSocketInit();
 
-int taosCreateSocketWithTimeOutOpt(uint32_t conn_timeout_sec);
+/*
+ * set timeout(ms)
+ */
+int32_t taosCreateSocketWithTimeout(uint32_t timeout);
 
 TdSocketPtr       taosOpenUdpSocket(uint32_t localIp, uint16_t localPort);
 TdSocketPtr       taosOpenTcpClientSocket(uint32_t ip, uint16_t port, uint32_t localIp);

include/util/ttrace.h

@@ -47,7 +47,9 @@ typedef struct STraceId {
 
 #define TRACE_TO_STR(traceId, buf)                              \
   do {                                                          \
-    sprintf(buf, "0x%" PRIx64 ":0x%" PRIx64 "", traceId->rootId, traceId->msgId); \
+    int64_t rootId = (traceId) != NULL ? (traceId)->rootId : 0; \
+    int64_t msgId = (traceId) != NULL ? (traceId)->msgId : 0;   \
+    sprintf(buf, "0x%" PRIx64 ":0x%" PRIx64 "", rootId, msgId); \
   } while (0)
 
 #ifdef __cplusplus

source/client/src/clientEnv.c

@@ -13,11 +13,11 @@
  * along with this program. If not, see <http://www.gnu.org/licenses/>.
  */
 
-#include "os.h"
 #include "catalog.h"
-#include "functionMgt.h"
 #include "clientInt.h"
 #include "clientLog.h"
+#include "functionMgt.h"
+#include "os.h"
 #include "query.h"
 #include "scheduler.h"
 #include "tcache.h"
@@ -38,7 +38,7 @@ static TdThreadOnce tscinit = PTHREAD_ONCE_INIT;
 volatile int32_t    tscInitRes = 0;
 
 static void registerRequest(SRequestObj *pRequest) {
-  STscObj *pTscObj = acquireTscObj(*(int64_t*)pRequest->pTscObj->id);
+  STscObj *pTscObj = acquireTscObj(*(int64_t *)pRequest->pTscObj->id);
 
   assert(pTscObj != NULL);
 
@@ -54,14 +54,14 @@ static void registerRequest(SRequestObj *pRequest) {
     int32_t currentInst = atomic_add_fetch_64((int64_t *)&pSummary->currentRequests, 1);
     tscDebug("0x%" PRIx64 " new Request from connObj:0x%" PRIx64
              ", current:%d, app current:%d, total:%d, reqId:0x%" PRIx64,
-             pRequest->self, *(int64_t*)pRequest->pTscObj->id, num, currentInst, total, pRequest->requestId);
+             pRequest->self, *(int64_t *)pRequest->pTscObj->id, num, currentInst, total, pRequest->requestId);
   }
 }
 
 static void deregisterRequest(SRequestObj *pRequest) {
   assert(pRequest != NULL);
 
-  STscObj          *pTscObj = pRequest->pTscObj;
+  STscObj *           pTscObj = pRequest->pTscObj;
   SAppClusterSummary *pActivity = &pTscObj->pAppInfo->summary;
 
   int32_t currentInst = atomic_sub_fetch_64((int64_t *)&pActivity->currentRequests, 1);
@@ -70,8 +70,8 @@ static void deregisterRequest(SRequestObj *pRequest) {
   int64_t duration = taosGetTimestampUs() - pRequest->metric.start;
   tscDebug("0x%" PRIx64 " free Request from connObj: 0x%" PRIx64 ", reqId:0x%" PRIx64 " elapsed:%" PRIu64
            " ms, current:%d, app current:%d",
-           pRequest->self, *(int64_t*)pTscObj->id, pRequest->requestId, duration / 1000, num, currentInst);
+           pRequest->self, *(int64_t *)pTscObj->id, pRequest->requestId, duration / 1000, num, currentInst);
-  releaseTscObj(*(int64_t*)pTscObj->id);
+  releaseTscObj(*(int64_t *)pTscObj->id);
 }
 
 // todo close the transporter properly
@@ -80,12 +80,13 @@ void closeTransporter(STscObj *pTscObj) {
     return;
   }
 
-  tscDebug("free transporter:%p in connObj: 0x%" PRIx64, pTscObj->pAppInfo->pTransporter, *(int64_t*)pTscObj->id);
+  tscDebug("free transporter:%p in connObj: 0x%" PRIx64, pTscObj->pAppInfo->pTransporter, *(int64_t *)pTscObj->id);
   rpcClose(pTscObj->pAppInfo->pTransporter);
 }
 
 static bool clientRpcRfp(int32_t code) {
-  if (code == TSDB_CODE_RPC_REDIRECT) {
+  if (code == TSDB_CODE_RPC_REDIRECT || code == TSDB_CODE_RPC_NETWORK_UNAVAIL || code == TSDB_CODE_NODE_NOT_DEPLOYED ||
+      code == TSDB_CODE_SYN_NOT_LEADER || code == TSDB_CODE_APP_NOT_READY) {
     return true;
   } else {
     return false;
@@ -128,16 +129,17 @@ void closeAllRequests(SHashObj *pRequests) {
 void destroyTscObj(void *pObj) {
   STscObj *pTscObj = pObj;
 
-  SClientHbKey connKey = {.tscRid = *(int64_t*)pTscObj->id, .connType = pTscObj->connType};
+  SClientHbKey connKey = {.tscRid = *(int64_t *)pTscObj->id, .connType = pTscObj->connType};
   hbDeregisterConn(pTscObj->pAppInfo->pAppHbMgr, connKey);
   int64_t connNum = atomic_sub_fetch_64(&pTscObj->pAppInfo->numOfConns, 1);
   closeAllRequests(pTscObj->pRequests);
   schedulerStopQueryHb(pTscObj->pAppInfo->pTransporter);
   if (0 == connNum) {
     // TODO
-    //closeTransporter(pTscObj);
+    // closeTransporter(pTscObj);
   }
-  tscDebug("connObj 0x%" PRIx64 " destroyed, totalConn:%" PRId64, *(int64_t*)pTscObj->id, pTscObj->pAppInfo->numOfConns);
+  tscDebug("connObj 0x%" PRIx64 " destroyed, totalConn:%" PRId64, *(int64_t *)pTscObj->id,
+           pTscObj->pAppInfo->numOfConns);
   taosThreadMutexDestroy(&pTscObj->mutex);
   taosMemoryFreeClear(pTscObj);
 }
@@ -167,10 +169,10 @@ void *createTscObj(const char *user, const char *auth, const char *db, int32_t c
 
   taosThreadMutexInit(&pObj->mutex, NULL);
   pObj->id = taosMemoryMalloc(sizeof(int64_t));
-  *(int64_t*)pObj->id = taosAddRef(clientConnRefPool, pObj);
+  *(int64_t *)pObj->id = taosAddRef(clientConnRefPool, pObj);
   pObj->schemalessType = 1;
 
-  tscDebug("connObj created, 0x%" PRIx64, *(int64_t*)pObj->id);
+  tscDebug("connObj created, 0x%" PRIx64, *(int64_t *)pObj->id);
   return pObj;
 }
 
@@ -325,7 +327,7 @@ int taos_options_imp(TSDB_OPTION option, const char *str) {
     return 0;
   }
 
-  SConfig     *pCfg = taosGetCfg();
+  SConfig *    pCfg = taosGetCfg();
   SConfigItem *pItem = NULL;
 
   switch (option) {

source/client/src/clientMain.c

@@ -199,10 +199,7 @@ TAOS_FIELD *taos_fetch_fields(TAOS_RES *res) {
   return pResInfo->userFields;
 }
 
-
+TAOS_RES *taos_query(TAOS *taos, const char *sql) { return taosQueryImpl(taos, sql, false); }
-TAOS_RES *taos_query(TAOS *taos, const char *sql) {
-  return taosQueryImpl(taos, sql, false);
-}
 
 TAOS_ROW taos_fetch_row(TAOS_RES *res) {
   if (res == NULL) {
@@ -594,7 +591,7 @@ int *taos_get_column_data_offset(TAOS_RES *res, int columnIndex) {
 }
 
 int taos_validate_sql(TAOS *taos, const char *sql) {
-  TAOS_RES* pObj = taosQueryImpl(taos, sql, true);
+  TAOS_RES *pObj = taosQueryImpl(taos, sql, true);
 
   int code = taos_errno(pObj);
 
@@ -911,13 +908,13 @@ int taos_load_table_info(TAOS *taos, const char *tableNameList) {
     goto _return;
   }
 
-  SCatalog* pCtg = NULL;
+  SCatalog *pCtg = NULL;
   code = catalogGetHandle(pTscObj->pAppInfo->clusterId, &pCtg);
   if (code != TSDB_CODE_SUCCESS) {
     goto _return;
   }
 
-  char* sql = "taos_load_table_info";
+  char *sql = "taos_load_table_info";
   code = buildRequest(pTscObj, sql, strlen(sql), &pRequest);
   if (code != TSDB_CODE_SUCCESS) {
     terrno = code;
@@ -928,9 +925,8 @@ int taos_load_table_info(TAOS *taos, const char *tableNameList) {
   tsem_init(&param.sem, 0, 0);
   param.pRequest = pRequest;
 
-  SRequestConnInfo conn = {.pTrans = pTscObj->pAppInfo->pTransporter,
+  SRequestConnInfo conn = {
-                           .requestId = pRequest->requestId,
+      .pTrans = pTscObj->pAppInfo->pTransporter, .requestId = pRequest->requestId, .requestObjRefId = pRequest->self};
-                           .requestObjRefId = pRequest->self};
 
   conn.mgmtEps = getEpSet_s(&pTscObj->pAppInfo->mgmtEp);
 
@@ -951,7 +947,6 @@ _return:
   return code;
 }
 
-
 TAOS_STMT *taos_stmt_init(TAOS *taos) {
   STscObj *pObj = acquireTscObj(*(int64_t *)taos);
   if (NULL == pObj) {

source/common/src/tdatablock.c

@@ -1164,7 +1164,7 @@ int32_t colInfoDataEnsureCapacity(SColumnInfoData* pColumn, uint32_t numOfRows)
 
 int32_t blockDataEnsureCapacity(SSDataBlock* pDataBlock, uint32_t numOfRows) {
   int32_t code = 0;
-  ASSERT(numOfRows > 0);
+  // ASSERT(numOfRows > 0);
 
   if (numOfRows == 0) {
     return TSDB_CODE_SUCCESS;
@@ -1230,6 +1230,32 @@ int32_t assignOneDataBlock(SSDataBlock* dst, const SSDataBlock* src) {
   return 0;
 }
 
+int32_t copyDataBlock(SSDataBlock* dst, const SSDataBlock* src) {
+  ASSERT(src != NULL && dst != NULL);
+
+  blockDataCleanup(dst);
+  int32_t code = blockDataEnsureCapacity(dst, src->info.rows);
+  if (code != TSDB_CODE_SUCCESS) {
+    terrno = code;
+    return code;
+  }
+
+  size_t numOfCols = taosArrayGetSize(src->pDataBlock);
+  for (int32_t i = 0; i < numOfCols; ++i) {
+    SColumnInfoData* pDst = taosArrayGet(dst->pDataBlock, i);
+    SColumnInfoData* pSrc = taosArrayGet(src->pDataBlock, i);
+    if (pSrc->pData == NULL) {
+      continue;
+    }
+
+    colDataAssign(pDst, pSrc, src->info.rows, &src->info);
+  }
+
+  dst->info.rows = src->info.rows;
+  dst->info.window = src->info.window;
+  return TSDB_CODE_SUCCESS;
+}
+
 SSDataBlock* createOneDataBlock(const SSDataBlock* pDataBlock, bool copyData) {
   if (pDataBlock == NULL) {
     return NULL;
@@ -1627,6 +1653,57 @@ void blockDebugShowData(const SArray* dataBlocks, const char* flag) {
   }
 }
 
+// for debug
+char* dumpBlockData(SSDataBlock* pDataBlock, const char* flag, char** pDataBuf) {
+  int32_t size = 2048;
+  *pDataBuf = taosMemoryCalloc(size, 1);
+  char*   dumpBuf = *pDataBuf;
+  char    pBuf[128] = {0};
+  int32_t colNum = taosArrayGetSize(pDataBlock->pDataBlock);
+  int32_t rows = pDataBlock->info.rows;
+  int32_t len = 0;
+  len += snprintf(dumpBuf + len, size - len, "\n%s |block type %d |child id %d|\n", flag,
+                  (int32_t)pDataBlock->info.type, pDataBlock->info.childId);
+  for (int32_t j = 0; j < rows; j++) {
+    len += snprintf(dumpBuf + len, size - len, "%s |", flag);
+    for (int32_t k = 0; k < colNum; k++) {
+      SColumnInfoData* pColInfoData = taosArrayGet(pDataBlock->pDataBlock, k);
+      void*            var = POINTER_SHIFT(pColInfoData->pData, j * pColInfoData->info.bytes);
+      if (colDataIsNull(pColInfoData, rows, j, NULL)) {
+        len += snprintf(dumpBuf + len, size - len, " %15s |", "NULL");
+        continue;
+      }
+      switch (pColInfoData->info.type) {
+        case TSDB_DATA_TYPE_TIMESTAMP:
+          formatTimestamp(pBuf, *(uint64_t*)var, TSDB_TIME_PRECISION_MILLI);
+          len += snprintf(dumpBuf + len, size - len, " %25s |", pBuf);
+          break;
+        case TSDB_DATA_TYPE_INT:
+          len += snprintf(dumpBuf + len, size - len, " %15d |", *(int32_t*)var);
+          break;
+        case TSDB_DATA_TYPE_UINT:
+          len += snprintf(dumpBuf + len, size - len, " %15u |", *(uint32_t*)var);
+          break;
+        case TSDB_DATA_TYPE_BIGINT:
+          len += snprintf(dumpBuf + len, size - len, " %15ld |", *(int64_t*)var);
+          break;
+        case TSDB_DATA_TYPE_UBIGINT:
+          len += snprintf(dumpBuf + len, size - len, " %15lu |", *(uint64_t*)var);
+          break;
+        case TSDB_DATA_TYPE_FLOAT:
+          len += snprintf(dumpBuf + len, size - len, " %15f |", *(float*)var);
+          break;
+        case TSDB_DATA_TYPE_DOUBLE:
+          len += snprintf(dumpBuf + len, size - len, " %15lf |", *(double*)var);
+          break;
+      }
+    }
+    len += snprintf(dumpBuf + len, size - len, "\n");
+  }
+  len += snprintf(dumpBuf + len, size - len, "%s |end\n", flag);
+  return dumpBuf;
+}
+
 /**
  * @brief TODO: Assume that the final generated result it less than 3M
  *

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

@@ -247,8 +247,6 @@ int32_t dmAppendVariablesToBlock(SSDataBlock* pBlock, int32_t dnodeId) {
   return TSDB_CODE_SUCCESS;
 }
 
-
-
 int32_t dmProcessRetrieve(SDnodeMgmt *pMgmt, SRpcMsg *pMsg) {
   int32_t size = 0;
   int32_t rowsRead = 0;
@@ -259,6 +257,11 @@ int32_t dmProcessRetrieve(SDnodeMgmt *pMgmt, SRpcMsg *pMsg) {
     return -1;
   }
 
+  if (strcmp(retrieveReq.user, TSDB_DEFAULT_USER) != 0) {
+    terrno = TSDB_CODE_MND_NO_RIGHTS;
+    return -1;
+  }
+
   if (strcasecmp(retrieveReq.tb, TSDB_INS_TABLE_DNODE_VARIABLES)) {
     terrno = TSDB_CODE_INVALID_MSG;
     return -1;
@@ -309,7 +312,6 @@ int32_t dmProcessRetrieve(SDnodeMgmt *pMgmt, SRpcMsg *pMsg) {
   return TSDB_CODE_SUCCESS;
 }
 
-
 SArray *dmGetMsgHandles() {
   int32_t code = -1;
   SArray *pArray = taosArrayInit(16, sizeof(SMgmtHandle));

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

@@ -70,9 +70,9 @@ int32_t dmProcessNodeMsg(SMgmtWrapper *pWrapper, SRpcMsg *pMsg) {
 }
 
 static void dmProcessRpcMsg(SDnode *pDnode, SRpcMsg *pRpc, SEpSet *pEpSet) {
-  SDnodeTrans  *pTrans = &pDnode->trans;
+  SDnodeTrans * pTrans = &pDnode->trans;
   int32_t       code = -1;
-  SRpcMsg      *pMsg = NULL;
+  SRpcMsg *     pMsg = NULL;
   SMgmtWrapper *pWrapper = NULL;
   SDnodeHandle *pHandle = &pTrans->msgHandles[TMSG_INDEX(pRpc->msgType)];
 
@@ -194,11 +194,11 @@ int32_t dmInitMsgHandle(SDnode *pDnode) {
 
   for (EDndNodeType ntype = DNODE; ntype < NODE_END; ++ntype) {
     SMgmtWrapper *pWrapper = &pDnode->wrappers[ntype];
-    SArray       *pArray = (*pWrapper->func.getHandlesFp)();
+    SArray *      pArray = (*pWrapper->func.getHandlesFp)();
     if (pArray == NULL) return -1;
 
     for (int32_t i = 0; i < taosArrayGetSize(pArray); ++i) {
-      SMgmtHandle  *pMgmt = taosArrayGet(pArray, i);
+      SMgmtHandle * pMgmt = taosArrayGet(pArray, i);
       SDnodeHandle *pHandle = &pTrans->msgHandles[TMSG_INDEX(pMgmt->msgType)];
       if (pMgmt->needCheckVgId) {
         pHandle->needCheckVgId = pMgmt->needCheckVgId;
@@ -248,7 +248,14 @@ static inline void dmReleaseHandle(SRpcHandleInfo *pHandle, int8_t type) {
   }
 }
 
-static bool rpcRfp(int32_t code) { return code == TSDB_CODE_RPC_REDIRECT; }
+static bool rpcRfp(int32_t code) {
+  if (code == TSDB_CODE_RPC_REDIRECT || code == TSDB_CODE_RPC_NETWORK_UNAVAIL || code == TSDB_CODE_NODE_NOT_DEPLOYED ||
+      code == TSDB_CODE_SYN_NOT_LEADER || code == TSDB_CODE_APP_NOT_READY) {
+    return true;
+  } else {
+    return false;
+  }
+}
 
 int32_t dmInitClient(SDnode *pDnode) {
   SDnodeTrans *pTrans = &pDnode->trans;

source/dnode/mnode/impl/inc/mndPrivilege.h

@@ -57,6 +57,8 @@ typedef enum {
   MND_OPER_USE_DB,
   MND_OPER_WRITE_DB,
   MND_OPER_READ_DB,
+  MND_OPER_READ_OR_WRITE_DB,
+  MND_OPER_SHOW_VARIBALES,
 } EOperType;
 
 int32_t mndInitPrivilege(SMnode *pMnode);
@@ -64,7 +66,8 @@ void    mndCleanupPrivilege(SMnode *pMnode);
 
 int32_t mndCheckOperPrivilege(SMnode *pMnode, const char *user, EOperType operType);
 int32_t mndCheckDbPrivilege(SMnode *pMnode, const char *user, EOperType operType, SDbObj *pDb);
-int32_t mndCheckShowPrivilege(SMnode *pMnode, const char *user, int32_t showType);
+int32_t mndCheckDbPrivilegeByName(SMnode *pMnode, const char *user, EOperType operType, const char *dbname);
+int32_t mndCheckShowPrivilege(SMnode *pMnode, const char *user, EShowType showType, const char *dbname);
 int32_t mndCheckAlterUserPrivilege(SUserObj *pOperUser, SUserObj *pUser, SAlterUserReq *pAlter);
 
 #ifdef __cplusplus

source/dnode/mnode/impl/inc/mndStream.h

@@ -33,6 +33,10 @@ SSdbRow *mndStreamActionDecode(SSdbRaw *pRaw);
 
 int32_t mndDropStreamByDb(SMnode *pMnode, STrans *pTrans, SDbObj *pDb);
 int32_t mndPersistStream(SMnode *pMnode, STrans *pTrans, SStreamObj *pStream);
+// for sma
+// TODO refactor
+int32_t mndDropStreamTasks(SMnode *pMnode, STrans *pTrans, SStreamObj *pStream);
+int32_t mndPersistDropStreamLog(SMnode *pMnode, STrans *pTrans, SStreamObj *pStream);
 
 #ifdef __cplusplus
 }

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

@@ -431,6 +431,10 @@ static int32_t mndProcessSubscribeReq(SRpcMsg *pMsg) {
       goto SUBSCRIBE_OVER;
     }
 
+    if (mndCheckDbPrivilegeByName(pMnode, pMsg->info.conn.user, MND_OPER_READ_DB, pTopic->db) != 0) {
+      goto SUBSCRIBE_OVER;
+    }
+
 #if 0
     // ref topic to prevent drop
     // TODO make topic complete

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

@@ -15,9 +15,9 @@
 
 #define _DEFAULT_SOURCE
 #include "mndDb.h"
-#include "mndPrivilege.h"
 #include "mndDnode.h"
 #include "mndOffset.h"
+#include "mndPrivilege.h"
 #include "mndShow.h"
 #include "mndSma.h"
 #include "mndStb.h"
@@ -1336,7 +1336,7 @@ char *buildRetension(SArray *pRetension) {
 }
 
 static void dumpDbInfoData(SSDataBlock *pBlock, SDbObj *pDb, SShowObj *pShow, int32_t rows, int64_t numOfTables,
-                           bool sysDb, ESdbStatus objStatus) {
+                           bool sysDb, ESdbStatus objStatus, bool sysinfo) {
   int32_t cols = 0;
 
   int32_t     bytes = pShow->pMeta->pSchemas[cols].bytes;
@@ -1354,7 +1354,7 @@ static void dumpDbInfoData(SSDataBlock *pBlock, SDbObj *pDb, SShowObj *pShow, in
   char statusB[24] = {0};
   STR_WITH_SIZE_TO_VARSTR(statusB, status, strlen(status));
 
-  if (sysDb) {
+  if (sysDb || !sysinfo) {
     for (int32_t i = 0; i < pShow->numOfColumns; ++i) {
       SColumnInfoData *pColInfo = taosArrayGet(pBlock->pDataBlock, i);
       if (i == 0) {
@@ -1528,17 +1528,21 @@ static int32_t mndRetrieveDbs(SRpcMsg *pReq, SShowObj *pShow, SSDataBlock *pBloc
   SDbObj    *pDb = NULL;
   ESdbStatus objStatus = 0;
 
+  SUserObj *pUser = mndAcquireUser(pMnode, pReq->info.conn.user);
+  if (pUser == NULL) return 0;
+  bool sysinfo = pUser->sysInfo;
+
   // Append the information_schema database into the result.
   if (!pShow->sysDbRsp) {
     SDbObj infoschemaDb = {0};
     setInformationSchemaDbCfg(&infoschemaDb);
-    dumpDbInfoData(pBlock, &infoschemaDb, pShow, numOfRows, 14, true, 0);
+    dumpDbInfoData(pBlock, &infoschemaDb, pShow, numOfRows, 14, true, 0, 1);
 
     numOfRows += 1;
 
     SDbObj perfschemaDb = {0};
     setPerfSchemaDbCfg(&perfschemaDb);
-    dumpDbInfoData(pBlock, &perfschemaDb, pShow, numOfRows, 3, true, 0);
+    dumpDbInfoData(pBlock, &perfschemaDb, pShow, numOfRows, 3, true, 0, 1);
 
     numOfRows += 1;
     pShow->sysDbRsp = true;
@@ -1550,16 +1554,19 @@ static int32_t mndRetrieveDbs(SRpcMsg *pReq, SShowObj *pShow, SSDataBlock *pBloc
       break;
     }
 
+    if (mndCheckDbPrivilege(pMnode, pReq->info.conn.user, MND_OPER_READ_OR_WRITE_DB, pDb) == 0) {
       int32_t numOfTables = 0;
       sdbTraverse(pSdb, SDB_VGROUP, mndGetTablesOfDbFp, &numOfTables, NULL, NULL);
 
-    dumpDbInfoData(pBlock, pDb, pShow, numOfRows, numOfTables, false, objStatus);
+      dumpDbInfoData(pBlock, pDb, pShow, numOfRows, numOfTables, false, objStatus, sysinfo);
       numOfRows++;
+    }
+
     sdbRelease(pSdb, pDb);
   }
 
   pShow->numOfRows += numOfRows;
-
+  mndReleaseUser(pMnode, pUser);
   return numOfRows;
 }
 

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

@@ -560,6 +560,10 @@ static int32_t mndProcessShowVariablesReq(SRpcMsg *pReq) {
   SShowVariablesRsp rsp = {0};
   int32_t           code = -1;
 
+  if (mndCheckOperPrivilege(pReq->info.node, pReq->info.conn.user, MND_OPER_SHOW_VARIBALES) != 0) {
+    goto _OVER;
+  }
+
   rsp.variables = taosArrayInit(4, sizeof(SVariablesInfo));
   if (NULL == rsp.variables) {
     mError("failed to alloc SVariablesInfo array while process show variables req");

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

@@ -15,6 +15,7 @@
 
 #define _DEFAULT_SOURCE
 #include "mndPrivilege.h"
+#include "mndDb.h"
 #include "mndUser.h"
 
 int32_t mndInitPrivilege(SMnode *pMnode) { return 0; }
@@ -45,6 +46,7 @@ int32_t mndCheckOperPrivilege(SMnode *pMnode, const char *user, EOperType operTy
     case MND_OPER_CONNECT:
     case MND_OPER_CREATE_FUNC:
     case MND_OPER_DROP_FUNC:
+    case MND_OPER_SHOW_VARIBALES:
       break;
     default:
       terrno = TSDB_CODE_MND_NO_RIGHTS;
@@ -79,7 +81,7 @@ int32_t mndCheckAlterUserPrivilege(SUserObj *pOperUser, SUserObj *pUser, SAlterU
   return -1;
 }
 
-int32_t mndCheckShowPrivilege(SMnode *pMnode, const char *user, int32_t showType) {
+int32_t mndCheckShowPrivilege(SMnode *pMnode, const char *user, EShowType showType, const char *dbname) {
   int32_t   code = 0;
   SUserObj *pUser = mndAcquireUser(pMnode, user);
 
@@ -98,14 +100,34 @@ int32_t mndCheckShowPrivilege(SMnode *pMnode, const char *user, int32_t showType
     goto _OVER;
   }
 
-  if (!pUser->sysInfo) {
+  if (pUser->sysInfo) {
+    goto _OVER;
+  }
+
+  switch (showType) {
+    case TSDB_MGMT_TABLE_DB:
+    case TSDB_MGMT_TABLE_STB:
+    case TSDB_MGMT_TABLE_INDEX:
+    case TSDB_MGMT_TABLE_STREAMS:
+    case TSDB_MGMT_TABLE_CONSUMERS:
+    case TSDB_MGMT_TABLE_TOPICS:
+    case TSDB_MGMT_TABLE_SUBSCRIPTIONS:
+    case TSDB_MGMT_TABLE_FUNC:
+    case TSDB_MGMT_TABLE_QUERIES:
+    case TSDB_MGMT_TABLE_CONNS:
+    case TSDB_MGMT_TABLE_APPS:
+    case TSDB_MGMT_TABLE_TRANS:
+      code = 0;
+      break;
+    default:
       terrno = TSDB_CODE_MND_NO_RIGHTS;
       code = -1;
       goto _OVER;
   }
 
-  terrno = TSDB_CODE_MND_NO_RIGHTS;
+  if (showType == TSDB_MGMT_TABLE_STB || showType == TSDB_MGMT_TABLE_VGROUP || showType == TSDB_MGMT_TABLE_INDEX) {
-  code = -1;
+    code = mndCheckDbPrivilegeByName(pMnode, user, MND_OPER_READ_OR_WRITE_DB, dbname);
+  }
 
 _OVER:
   mndReleaseUser(pMnode, pUser);
@@ -133,19 +155,11 @@ int32_t mndCheckDbPrivilege(SMnode *pMnode, const char *user, EOperType operType
     if (pUser->sysInfo) goto _OVER;
   }
 
-  if (operType == MND_OPER_ALTER_DB) {
+  if (operType == MND_OPER_ALTER_DB || operType == MND_OPER_DROP_DB || operType == MND_OPER_COMPACT_DB) {
     if (strcmp(pUser->user, pDb->createUser) == 0 && pUser->sysInfo) goto _OVER;
   }
 
-  if (operType == MND_OPER_DROP_DB) {
+  if (operType == MND_OPER_USE_DB || operType == MND_OPER_READ_OR_WRITE_DB) {
-    if (strcmp(pUser->user, pDb->createUser) == 0 && pUser->sysInfo) goto _OVER;
-  }
-
-  if (operType == MND_OPER_COMPACT_DB) {
-    if (strcmp(pUser->user, pDb->createUser) == 0 && pUser->sysInfo) goto _OVER;
-  }
-
-  if (operType == MND_OPER_USE_DB) {
     if (strcmp(pUser->user, pDb->createUser) == 0) goto _OVER;
     if (taosHashGet(pUser->readDbs, pDb->name, strlen(pDb->name) + 1) != NULL) goto _OVER;
     if (taosHashGet(pUser->writeDbs, pDb->name, strlen(pDb->name) + 1) != NULL) goto _OVER;
@@ -168,3 +182,12 @@ _OVER:
   mndReleaseUser(pMnode, pUser);
   return code;
 }
+
+int32_t mndCheckDbPrivilegeByName(SMnode *pMnode, const char *user, EOperType operType, const char *dbname) {
+  SDbObj *pDb = mndAcquireDb(pMnode, dbname);
+  if (pDb == NULL) return -1;
+
+  int32_t code = mndCheckDbPrivilege(pMnode, user, operType, pDb);
+  mndReleaseDb(pMnode, pDb);
+  return code;
+}

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

@@ -254,6 +254,10 @@ static int32_t mndProcessConnectReq(SRpcMsg *pReq) {
               terrstr());
       goto _OVER;
     }
+
+    if (mndCheckDbPrivilege(pMnode, pReq->info.conn.user, MND_OPER_READ_OR_WRITE_DB, pDb) != 0) {
+      goto _OVER;
+    }
   }
 
   pConn = mndCreateConn(pMnode, pReq->info.conn.user, connReq.connType, pReq->info.conn.clientIp,

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

@@ -231,8 +231,14 @@ static int32_t mndProcessRetrieveSysTableReq(SRpcMsg *pReq) {
   }
 
   mDebug("show:0x%" PRIx64 ", start retrieve data, type:%d", pShow->id, pShow->type);
-
+  if (retrieveReq.user[0] != 0) {
-  // if (mndCheckShowPrivilege(pMnode, pReq->info.conn.user, pShow->type) != 0) return -1;
+    memcpy(pReq->info.conn.user, retrieveReq.user, TSDB_USER_LEN);
+  } else {
+    memcpy(pReq->info.conn.user, TSDB_DEFAULT_USER, TSDB_USER_LEN);
+  }
+  if (mndCheckShowPrivilege(pMnode, pReq->info.conn.user, pShow->type, retrieveReq.db) != 0) {
+    return -1;
+  }
 
   int32_t      numOfCols = pShow->pMeta->numOfColumns;
   SSDataBlock *pBlock = taosMemoryCalloc(1, sizeof(SSDataBlock));

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

@@ -15,11 +15,11 @@
 
 #define _DEFAULT_SOURCE
 #include "mndSma.h"
-#include "mndPrivilege.h"
 #include "mndDb.h"
 #include "mndDnode.h"
 #include "mndInfoSchema.h"
 #include "mndMnode.h"
+#include "mndPrivilege.h"
 #include "mndScheduler.h"
 #include "mndShow.h"
 #include "mndStb.h"
@@ -857,6 +857,23 @@ static int32_t mndDropSma(SMnode *pMnode, SRpcMsg *pReq, SDbObj *pDb, SSmaObj *p
   mDebug("trans:%d, used to drop sma:%s", pTrans->id, pSma->name);
   mndTransSetDbName(pTrans, pDb->name, NULL);
 
+  SStreamObj *pStream = mndAcquireStream(pMnode, pSma->name);
+  if (pStream == NULL || pStream->smaId != pSma->uid) {
+    sdbRelease(pMnode->pSdb, pStream);
+    goto _OVER;
+  } else {
+    if (mndDropStreamTasks(pMnode, pTrans, pStream) < 0) {
+      mError("stream:%s, failed to drop task since %s", pStream->name, terrstr());
+      sdbRelease(pMnode->pSdb, pStream);
+      goto _OVER;
+    }
+
+    // drop stream
+    if (mndPersistDropStreamLog(pMnode, pTrans, pStream) < 0) {
+      sdbRelease(pMnode->pSdb, pStream);
+      goto _OVER;
+    }
+  }
   if (mndSetDropSmaRedoLogs(pMnode, pTrans, pSma) != 0) goto _OVER;
   if (mndSetDropSmaVgroupRedoLogs(pMnode, pTrans, pVgroup) != 0) goto _OVER;
   if (mndSetDropSmaCommitLogs(pMnode, pTrans, pSma) != 0) goto _OVER;

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

@@ -15,12 +15,12 @@
 
 #define _DEFAULT_SOURCE
 #include "mndStb.h"
-#include "mndPrivilege.h"
 #include "mndDb.h"
 #include "mndDnode.h"
 #include "mndInfoSchema.h"
 #include "mndMnode.h"
 #include "mndPerfSchema.h"
+#include "mndPrivilege.h"
 #include "mndScheduler.h"
 #include "mndShow.h"
 #include "mndSma.h"
@@ -107,7 +107,7 @@ SSdbRaw *mndStbActionEncode(SStbObj *pStb) {
   int32_t funcNum = taosArrayGetSize(pStb->pFuncs);
   SDB_SET_INT32(pRaw, dataPos, funcNum, _OVER)
   for (int32_t i = 0; i < funcNum; ++i) {
-    char* func = taosArrayGet(pStb->pFuncs, i);
+    char *func = taosArrayGet(pStb->pFuncs, i);
     SDB_SET_BINARY(pRaw, dataPos, func, TSDB_FUNC_NAME_LEN, _OVER)
   }
 
@@ -1396,7 +1396,6 @@ static int32_t mndBuildStbCfgImp(SDbObj *pDb, SStbObj *pStb, const char *tbName,
   return 0;
 }
 
-
 static int32_t mndBuildStbSchema(SMnode *pMnode, const char *dbFName, const char *tbName, STableMetaRsp *pRsp,
                                  int32_t *smaVer) {
   char tbFName[TSDB_TABLE_FNAME_LEN] = {0};
@@ -1449,8 +1448,6 @@ static int32_t mndBuildStbCfg(SMnode *pMnode, const char *dbFName, const char *t
   return code;
 }
 
-
-
 static int32_t mndBuildSMAlterStbRsp(SDbObj *pDb, const SMAlterStbReq *pAlter, SStbObj *pObj, void **pCont,
                                      int32_t *pLen) {
   int32_t       ret;
@@ -1866,7 +1863,6 @@ _OVER:
   return code;
 }
 
-
 int32_t mndValidateStbInfo(SMnode *pMnode, SSTableVersion *pStbVersions, int32_t numOfStbs, void **ppRsp,
                            int32_t *pRspLen) {
   SSTbHbRsp hbRsp = {0};

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

@@ -14,10 +14,10 @@
  */
 
 #include "mndStream.h"
-#include "mndPrivilege.h"
 #include "mndDb.h"
 #include "mndDnode.h"
 #include "mndMnode.h"
+#include "mndPrivilege.h"
 #include "mndScheduler.h"
 #include "mndShow.h"
 #include "mndStb.h"
@@ -437,10 +437,6 @@ static int32_t mndCreateStbForStream(SMnode *pMnode, STrans *pTrans, const SStre
     goto _OVER;
   }
 
-  if (mndCheckDbPrivilege(pMnode, user, MND_OPER_WRITE_DB, pDb) != 0) {
-    goto _OVER;
-  }
-
   int32_t numOfStbs = -1;
   if (mndGetNumOfStbs(pMnode, pDb->name, &numOfStbs) != 0) {
     goto _OVER;
@@ -494,7 +490,7 @@ static int32_t mndPersistTaskDropReq(STrans *pTrans, SStreamTask *pTask) {
   return 0;
 }
 
-static int32_t mndDropStreamTasks(SMnode *pMnode, STrans *pTrans, SStreamObj *pStream) {
+int32_t mndDropStreamTasks(SMnode *pMnode, STrans *pTrans, SStreamObj *pStream) {
   int32_t lv = taosArrayGetSize(pStream->tasks);
   for (int32_t i = 0; i < lv; i++) {
     SArray *pTasks = taosArrayGetP(pStream->tasks, i);
@@ -542,19 +538,6 @@ static int32_t mndProcessCreateStreamReq(SRpcMsg *pReq) {
     goto _OVER;
   }
 
-  // TODO check read auth for source and write auth for target
-#if 0
-  pDb = mndAcquireDb(pMnode, createStreamReq.sourceDB);
-  if (pDb == NULL) {
-    terrno = TSDB_CODE_MND_DB_NOT_SELECTED;
-    goto _OVER;
-  }
-
-  if (mndCheckDbPrivilege(pMnode, pReq->info.conn.user, MND_OPER_WRITE_DB, pDb) != 0) {
-    goto _OVER;
-  }
-#endif
-
   // build stream obj from request
   SStreamObj streamObj = {0};
   if (mndBuildStreamObjFromCreateReq(pMnode, &streamObj, &createStreamReq) < 0) {
@@ -592,6 +575,16 @@ static int32_t mndProcessCreateStreamReq(SRpcMsg *pReq) {
     goto _OVER;
   }
 
+  if (mndCheckDbPrivilegeByName(pMnode, pReq->info.conn.user, MND_OPER_READ_DB, streamObj.sourceDb) != 0) {
+    mndTransDrop(pTrans);
+    goto _OVER;
+  }
+
+  if (mndCheckDbPrivilegeByName(pMnode, pReq->info.conn.user, MND_OPER_WRITE_DB, streamObj.targetDb) != 0) {
+    mndTransDrop(pTrans);
+    goto _OVER;
+  }
+
   // execute creation
   if (mndTransPrepare(pMnode, pTrans) != 0) {
     mError("trans:%d, failed to prepare since %s", pTrans->id, terrstr());
@@ -641,13 +634,9 @@ static int32_t mndProcessDropStreamReq(SRpcMsg *pReq) {
     }
   }
 
-#if 0
+  if (mndCheckDbPrivilegeByName(pMnode, pReq->info.conn.user, MND_OPER_WRITE_DB, pStream->targetDb) != 0) {
-  // todo check auth
+    return -1;
-  pUser = mndAcquireUser(pMnode, pReq->info.conn.user);
-  if (pUser == NULL) {
-    goto DROP_STREAM_OVER;
   }
-#endif
 
   STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_RETRY, TRN_CONFLICT_NOTHING, pReq);
   if (pTrans == NULL) {

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

@@ -14,12 +14,12 @@
  */
 
 #include "mndTopic.h"
-#include "mndPrivilege.h"
 #include "mndConsumer.h"
 #include "mndDb.h"
 #include "mndDnode.h"
 #include "mndMnode.h"
 #include "mndOffset.h"
+#include "mndPrivilege.h"
 #include "mndShow.h"
 #include "mndStb.h"
 #include "mndSubscribe.h"
@@ -401,6 +401,10 @@ static int32_t mndCreateTopic(SMnode *pMnode, SRpcMsg *pReq, SCMCreateTopicReq *
     }
   } else if (pCreate->subType == TOPIC_SUB_TYPE__TABLE) {
     SStbObj *pStb = mndAcquireStb(pMnode, pCreate->subStbName);
+    if (pStb == NULL) {
+      terrno = TSDB_CODE_MND_STB_NOT_EXIST;
+      return -1;
+    }
     topicObj.stbUid = pStb->uid;
   }
   /*} else if (pCreate->subType == TOPIC_SUB_TYPE__DB) {*/
@@ -480,7 +484,7 @@ static int32_t mndProcessCreateTopicReq(SRpcMsg *pReq) {
     goto _OVER;
   }
 
-  if (mndCheckDbPrivilege(pMnode, pReq->info.conn.user, MND_OPER_WRITE_DB, pDb) != 0) {
+  if (mndCheckDbPrivilege(pMnode, pReq->info.conn.user, MND_OPER_READ_DB, pDb) != 0) {
     goto _OVER;
   }
 
@@ -571,6 +575,10 @@ static int32_t mndProcessDropTopicReq(SRpcMsg *pReq) {
   }
 #endif
 
+  if (mndCheckDbPrivilegeByName(pMnode, pReq->info.conn.user, MND_OPER_READ_DB, pTopic->db) != 0) {
+    return -1;
+  }
+
   STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_ROLLBACK, TRN_CONFLICT_DB_INSIDE, pReq);
   mndTransSetDbName(pTrans, pTopic->db, NULL);
   if (pTrans == NULL) {

source/dnode/snode/src/snode.c

@@ -257,7 +257,7 @@ int32_t sndProcessSMsg(SSnode *pSnode, SRpcMsg *pMsg) {
     case TDMT_STREAM_TASK_RECOVER_RSP:
       return sndProcessTaskRecoverRsp(pSnode, pMsg);
     case TDMT_STREAM_RETRIEVE_RSP:
-      return sndProcessTaskRecoverRsp(pSnode, pMsg);
+      return sndProcessTaskRetrieveRsp(pSnode, pMsg);
     default:
       ASSERT(0);
   }

source/dnode/vnode/inc/vnode.h

@@ -116,7 +116,7 @@ typedef void *tsdbReaderT;
 #define BLOCK_LOAD_TABLE_SEQ_ORDER  2
 #define BLOCK_LOAD_TABLE_RR_ORDER   3
 
-int32_t      tsdbSetTableList(tsdbReaderT reader, SArray* tableList);
+int32_t     tsdbSetTableList(tsdbReaderT reader, SArray *tableList);
 tsdbReaderT tsdbReaderOpen(SVnode *pVnode, SQueryTableDataCond *pCond, SArray *tableList, uint64_t qId,
                            uint64_t taskId);
 tsdbReaderT tsdbQueryCacheLast(SVnode *pVnode, SQueryTableDataCond *pCond, STableListInfo *groupList, uint64_t qId,
@@ -150,8 +150,7 @@ int32_t tqReadHandleRemoveTbUidList(STqReadHandle *pHandle, const SArray *tbUidL
 int32_t tqReadHandleSetMsg(STqReadHandle *pHandle, SSubmitReq *pMsg, int64_t ver);
 bool    tqNextDataBlock(STqReadHandle *pHandle);
 bool    tqNextDataBlockFilterOut(STqReadHandle *pHandle, SHashObj *filterOutUids);
-int32_t tqRetrieveDataBlock(SSDataBlock* pBlock, STqReadHandle *pHandle, uint64_t *pGroupId, uint64_t *pUid,
+int32_t tqRetrieveDataBlock(SSDataBlock *pBlock, STqReadHandle *pHandle);
-                            int32_t *pNumOfRows);
 
 // sma
 int32_t smaGetTSmaDays(SVnodeCfg *pCfg, void *pCont, uint32_t contLen, int32_t *days);
@@ -196,6 +195,7 @@ struct SVnodeCfg {
 typedef struct {
   TSKEY    lastKey;
   uint64_t uid;
+  uint64_t groupId;
 } STableKeyInfo;
 
 struct SMetaEntry {

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

@@ -388,7 +388,7 @@ int metaTtlDropTable(SMeta *pMeta, int64_t ttl, SArray *tbUids) {
 }
 
 static void metaBuildTtlIdxKey(STtlIdxKey *ttlKey, const SMetaEntry *pME){
-  int32_t    ttlDays;
+  int64_t    ttlDays;
   int64_t    ctime;
   if (pME->type == TSDB_CHILD_TABLE) {
     ctime = pME->ctbEntry.ctime;

source/dnode/vnode/src/tq/tqExec.c

@@ -112,7 +112,7 @@ int32_t tqDataExec(STQ* pTq, STqExecHandle* pExec, SSubmitReq* pReq, SMqDataBlkR
     tqReadHandleSetMsg(pReader, pReq, 0);
     while (tqNextDataBlock(pReader)) {
       SSDataBlock block = {0};
-      if (tqRetrieveDataBlock(&block, pReader, &block.info.groupId, &block.info.uid, &block.info.rows) < 0) {
+      if (tqRetrieveDataBlock(&block, pReader) < 0) {
         if (terrno == TSDB_CODE_TQ_TABLE_SCHEMA_NOT_FOUND) continue;
         ASSERT(0);
       }
@@ -129,7 +129,7 @@ int32_t tqDataExec(STQ* pTq, STqExecHandle* pExec, SSubmitReq* pReq, SMqDataBlkR
     tqReadHandleSetMsg(pReader, pReq, 0);
     while (tqNextDataBlockFilterOut(pReader, pExec->execDb.pFilterOutTbUid)) {
       SSDataBlock block = {0};
-      if (tqRetrieveDataBlock(&block, pReader, &block.info.groupId, &block.info.uid, &block.info.rows) < 0) {
+      if (tqRetrieveDataBlock(&block, pReader) < 0) {
         if (terrno == TSDB_CODE_TQ_TABLE_SCHEMA_NOT_FOUND) continue;
         ASSERT(0);
       }

source/dnode/vnode/src/tq/tqRead.c

@@ -109,11 +109,15 @@ int32_t tqReadHandleSetMsg(STqReadHandle* pReadHandle, SSubmitReq* pMsg, int64_t
 }
 
 bool tqNextDataBlock(STqReadHandle* pHandle) {
+  if (pHandle->pMsg == NULL) return false;
   while (1) {
     if (tGetSubmitMsgNext(&pHandle->msgIter, &pHandle->pBlock) < 0) {
       return false;
     }
-    if (pHandle->pBlock == NULL) return false;
+    if (pHandle->pBlock == NULL) {
+      pHandle->pMsg = NULL;
+      return false;
+    }
 
     if (pHandle->tbIdHash == NULL) {
       return true;
@@ -142,10 +146,7 @@ bool tqNextDataBlockFilterOut(STqReadHandle* pHandle, SHashObj* filterOutUids) {
   return false;
 }
 
-int32_t tqRetrieveDataBlock(SSDataBlock* pBlock, STqReadHandle* pHandle, uint64_t* pGroupId, uint64_t* pUid,
+int32_t tqRetrieveDataBlock(SSDataBlock* pBlock, STqReadHandle* pHandle) {
-                            int32_t* pNumOfRows) {
-  *pUid = 0;
-
   // TODO: cache multiple schema
   int32_t sversion = htonl(pHandle->pBlock->sversion);
   if (pHandle->cachedSchemaSuid == 0 || pHandle->cachedSchemaVer != sversion ||
@@ -176,7 +177,6 @@ int32_t tqRetrieveDataBlock(SSDataBlock* pBlock, STqReadHandle* pHandle, uint64_
   STSchema*       pTschema = pHandle->pSchema;
   SSchemaWrapper* pSchemaWrapper = pHandle->pSchemaWrapper;
 
-  *pNumOfRows = pHandle->msgIter.numOfRows;
   int32_t colNumNeed = taosArrayGetSize(pHandle->pColIdList);
 
   if (colNumNeed == 0) {
@@ -217,22 +217,22 @@ int32_t tqRetrieveDataBlock(SSDataBlock* pBlock, STqReadHandle* pHandle, uint64_
     }
   }
 
-  if (blockDataEnsureCapacity(pBlock, *pNumOfRows) < 0) {
+  if (blockDataEnsureCapacity(pBlock, pHandle->msgIter.numOfRows) < 0) {
     goto FAIL;
   }
 
   int32_t colActual = blockDataGetNumOfCols(pBlock);
 
-  // TODO in stream shuffle case, fetch groupId
-  *pGroupId = 0;
-
   STSRowIter iter = {0};
   tdSTSRowIterInit(&iter, pTschema);
   STSRow* row;
   int32_t curRow = 0;
 
   tInitSubmitBlkIter(&pHandle->msgIter, pHandle->pBlock, &pHandle->blkIter);
-  *pUid = pHandle->msgIter.uid;  // set the uid of table for submit block
+
+  pBlock->info.groupId = 0;
+  pBlock->info.uid = pHandle->msgIter.uid;  // set the uid of table for submit block
+  pBlock->info.rows = pHandle->msgIter.numOfRows;
 
   while ((row = tGetSubmitBlkNext(&pHandle->blkIter)) != NULL) {
     tdSTSRowIterReset(&iter, row);

source/dnode/vnode/src/tq/tqSink.c

@@ -18,31 +18,21 @@
 SSubmitReq* tdBlockToSubmit(const SArray* pBlocks, const STSchema* pTSchema, bool createTb, int64_t suid,
                             const char* stbFullName, int32_t vgId) {
   SSubmitReq* ret = NULL;
+  SArray*     schemaReqs = NULL;
+  SArray*     schemaReqSz = NULL;
   SArray*     tagArray = taosArrayInit(1, sizeof(STagVal));
   if (!tagArray) {
     terrno = TSDB_CODE_OUT_OF_MEMORY;
     return NULL;
   }
 
-  // cal size
-  int32_t cap = sizeof(SSubmitReq);
   int32_t sz = taosArrayGetSize(pBlocks);
-  for (int32_t i = 0; i < sz; i++) {
-    SSDataBlock* pDataBlock = taosArrayGet(pBlocks, i);
-    int32_t      rows = pDataBlock->info.rows;
-    // TODO min
-    int32_t rowSize = pDataBlock->info.rowSize;
-    int32_t maxLen = TD_ROW_MAX_BYTES_FROM_SCHEMA(pTSchema);
-    int32_t schemaLen = 0;
 
   if (createTb) {
-      SVCreateTbReq createTbReq = {0};
+    schemaReqs = taosArrayInit(sz, sizeof(void*));
-      char*         cname = buildCtbNameByGroupId(stbFullName, pDataBlock->info.groupId);
+    schemaReqSz = taosArrayInit(sz, sizeof(int32_t));
-      createTbReq.name = cname;
+    for (int32_t i = 0; i < sz; i++) {
-      createTbReq.flags = 0;
+      SSDataBlock* pDataBlock = taosArrayGet(pBlocks, i);
-      createTbReq.type = TSDB_CHILD_TABLE;
-      createTbReq.ctb.suid = suid;
-
       STagVal      tagVal = {
                .cid = taosArrayGetSize(pDataBlock->pDataBlock) + 1,
                .type = TSDB_DATA_TYPE_UBIGINT,
@@ -53,22 +43,62 @@ SSubmitReq* tdBlockToSubmit(const SArray* pBlocks, const STSchema* pTSchema, boo
       taosArrayPush(tagArray, &tagVal);
       tTagNew(tagArray, 1, false, &pTag);
       if (pTag == NULL) {
-        tdDestroySVCreateTbReq(&createTbReq);
+        terrno = TSDB_CODE_OUT_OF_MEMORY;
         taosArrayDestroy(tagArray);
         return NULL;
       }
+
+      SVCreateTbReq createTbReq = {0};
+      createTbReq.name = buildCtbNameByGroupId(stbFullName, pDataBlock->info.groupId);
+      createTbReq.flags = 0;
+      createTbReq.type = TSDB_CHILD_TABLE;
+      createTbReq.ctb.suid = suid;
       createTbReq.ctb.pTag = (uint8_t*)pTag;
 
       int32_t code;
+      int32_t schemaLen;
       tEncodeSize(tEncodeSVCreateTbReq, &createTbReq, schemaLen, code);
-
-      tdDestroySVCreateTbReq(&createTbReq);
       if (code < 0) {
+        tdDestroySVCreateTbReq(&createTbReq);
         taosArrayDestroy(tagArray);
+        taosMemoryFreeClear(ret);
         return NULL;
       }
-    }
 
+      void* schemaStr = taosMemoryMalloc(schemaLen);
+      if (schemaStr == NULL) {
+        terrno = TSDB_CODE_OUT_OF_MEMORY;
+        return NULL;
+      }
+      taosArrayPush(schemaReqs, &schemaStr);
+      taosArrayPush(schemaReqSz, &schemaLen);
+
+      SEncoder encoder = {0};
+      tEncoderInit(&encoder, schemaStr, schemaLen);
+      code = tEncodeSVCreateTbReq(&encoder, &createTbReq);
+      if (code < 0) {
+        terrno = TSDB_CODE_OUT_OF_MEMORY;
+        return NULL;
+      }
+      tEncoderClear(&encoder);
+      tdDestroySVCreateTbReq(&createTbReq);
+    }
+  }
+  taosArrayDestroy(tagArray);
+
+  // cal size
+  int32_t cap = sizeof(SSubmitReq);
+  for (int32_t i = 0; i < sz; i++) {
+    SSDataBlock* pDataBlock = taosArrayGet(pBlocks, i);
+    int32_t      rows = pDataBlock->info.rows;
+    // TODO min
+    int32_t rowSize = pDataBlock->info.rowSize;
+    int32_t maxLen = TD_ROW_MAX_BYTES_FROM_SCHEMA(pTSchema);
+
+    int32_t schemaLen = 0;
+    if (createTb) {
+      schemaLen = *(int32_t*)taosArrayGet(schemaReqSz, i);
+    }
     cap += sizeof(SSubmitBlk) + schemaLen + rows * maxLen;
   }
 
@@ -99,55 +129,13 @@ SSubmitReq* tdBlockToSubmit(const SArray* pBlocks, const STSchema* pTSchema, boo
 
     int32_t schemaLen = 0;
     if (createTb) {
-      SVCreateTbReq createTbReq = {0};
+      schemaLen = *(int32_t*)taosArrayGet(schemaReqSz, i);
-      char*         cname = buildCtbNameByGroupId(stbFullName, pDataBlock->info.groupId);
+      void* schemaStr = taosArrayGetP(schemaReqs, i);
-      createTbReq.name = cname;
+      memcpy(blkSchema, schemaStr, schemaLen);
-      createTbReq.flags = 0;
-      createTbReq.type = TSDB_CHILD_TABLE;
-      createTbReq.ctb.suid = suid;
-
-      STagVal tagVal = {
-          .cid = taosArrayGetSize(pDataBlock->pDataBlock) + 1,
-          .type = TSDB_DATA_TYPE_UBIGINT,
-          .i64 = (int64_t)pDataBlock->info.groupId,
-      };
-      taosArrayClear(tagArray);
-      taosArrayPush(tagArray, &tagVal);
-      STag* pTag = NULL;
-      tTagNew(tagArray, 1, false, &pTag);
-      if (pTag == NULL) {
-        tdDestroySVCreateTbReq(&createTbReq);
-        taosArrayDestroy(tagArray);
-        taosMemoryFreeClear(ret);
-        return NULL;
-      }
-      createTbReq.ctb.pTag = (uint8_t*)pTag;
-
-      int32_t code;
-      tEncodeSize(tEncodeSVCreateTbReq, &createTbReq, schemaLen, code);
-      if (code < 0) {
-        tdDestroySVCreateTbReq(&createTbReq);
-        taosArrayDestroy(tagArray);
-        taosMemoryFreeClear(ret);
-        return NULL;
-      }
-
-      SEncoder encoder = {0};
-      tEncoderInit(&encoder, blkSchema, schemaLen);
-      code = tEncodeSVCreateTbReq(&encoder, &createTbReq);
-      tEncoderClear(&encoder);
-      tdDestroySVCreateTbReq(&createTbReq);
-
-      if (code < 0) {
-        taosArrayDestroy(tagArray);
-        taosMemoryFreeClear(ret);
-        return NULL;
-      }
     }
     blkHead->schemaLen = htonl(schemaLen);
 
     STSRow* rowData = POINTER_SHIFT(blkSchema, schemaLen);
-
     for (int32_t j = 0; j < rows; j++) {
       SRowBuilder rb = {0};
       tdSRowInit(&rb, pTSchema->version);
@@ -175,7 +163,10 @@ SSubmitReq* tdBlockToSubmit(const SArray* pBlocks, const STSchema* pTSchema, boo
   }
 
   ret->length = htonl(ret->length);
-  taosArrayDestroy(tagArray);
+
+  if (schemaReqs) taosArrayDestroyP(schemaReqs, taosMemoryFree);
+  taosArrayDestroy(schemaReqSz);
+
   return ret;
 }
 

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

@@ -2852,7 +2852,7 @@ int32_t tsdbGetAllTableList(SMeta* pMeta, uint64_t uid, SArray* list) {
       break;
     }
 
-    STableKeyInfo info = {.lastKey = TSKEY_INITIAL_VAL, uid = id};
+    STableKeyInfo info = {.lastKey = TSKEY_INITIAL_VAL, uid = id, .groupId = 0};
     taosArrayPush(list, &info);
   }
 

source/libs/executor/inc/executorimpl.h

@@ -293,6 +293,7 @@ typedef enum EStreamScanMode {
   STREAM_SCAN_FROM_RES,
   STREAM_SCAN_FROM_UPDATERES,
   STREAM_SCAN_FROM_DATAREADER,
+  STREAM_SCAN_FROM_DATAREADER_RETRIEVE,
 } EStreamScanMode;
 
 typedef struct SCatchSupporter {
@@ -348,7 +349,9 @@ typedef struct SStreamBlockScanInfo {
   SArray*        childIds;
   SessionWindowSupporter sessionSup;
   bool            assignBlockUid; // assign block uid to groupId, temporarily used for generating rollup SMA.
-  int32_t         scanWinIndex;
+  int32_t         scanWinIndex;   // for state operator
+  int32_t         pullDataResIndex;
+  SSDataBlock*    pPullDataRes;             // pull data SSDataBlock
 } SStreamBlockScanInfo;
 
 typedef struct SSysTableScanInfo {
@@ -427,8 +430,13 @@ typedef struct SStreamFinalIntervalOperatorInfo {
   STimeWindowAggSupp twAggSup;
   SArray*            pChildren;
   SSDataBlock*       pUpdateRes;
+  bool               returnUpdate;
   SPhysiNode*        pPhyNode;           // create new child
   bool               isFinal;
+  SHashObj*          pPullDataMap;
+  SArray*            pPullWins;          // SPullWindowInfo
+  int32_t            pullIndex;
+  SSDataBlock*       pPullDataRes;
 } SStreamFinalIntervalOperatorInfo;
 
 typedef struct SAggOperatorInfo {
@@ -851,6 +859,7 @@ SOperatorInfo* createTableMergeScanOperatorInfo(STableScanPhysiNode* pTableScanN
 void copyUpdateDataBlock(SSDataBlock* pDest, SSDataBlock* pSource, int32_t tsColIndex);
 
 int32_t generateGroupIdMap(STableListInfo* pTableListInfo, SReadHandle* pHandle, SNodeList* groupKey);
+SSDataBlock* createPullDataBlock();
 
 #ifdef __cplusplus
 }

source/libs/executor/src/executil.c

@@ -315,7 +315,7 @@ int32_t getTableList(void* metaHandle, SScanPhysiNode* pScanNode, STableListInfo
       }
 
       for (int i = 0; i < taosArrayGetSize(res); i++) {
-        STableKeyInfo info = {.lastKey = TSKEY_INITIAL_VAL, .uid = *(uint64_t*)taosArrayGet(res, i)};
+        STableKeyInfo info = {.lastKey = TSKEY_INITIAL_VAL, .uid = *(uint64_t*)taosArrayGet(res, i), .groupId = 0};
         taosArrayPush(pListInfo->pTableList, &info);
       }
       taosArrayDestroy(res);
@@ -336,7 +336,7 @@ int32_t getTableList(void* metaHandle, SScanPhysiNode* pScanNode, STableListInfo
       }
     }
   }else {  // Create one table group.
-    STableKeyInfo info = {.lastKey = 0, .uid = tableUid};
+    STableKeyInfo info = {.lastKey = 0, .uid = tableUid, .groupId = 0};
     taosArrayPush(pListInfo->pTableList, &info);
   }
   pListInfo->pGroupList = taosArrayInit(4, POINTER_BYTES);

source/libs/executor/src/executorimpl.c

@@ -4028,6 +4028,7 @@ int32_t generateGroupIdMap(STableListInfo* pTableListInfo, SReadHandle* pHandle,
     int32_t   len = (int32_t)(pStart - (char*)keyBuf);
     uint64_t groupId = calcGroupId(keyBuf, len);
     taosHashPut(pTableListInfo->map, &(info->uid), sizeof(uint64_t), &groupId, sizeof(uint64_t));
+    info->groupId = groupId;
     groupNum++;
 
     nodesClearList(groupNew);
@@ -4126,7 +4127,7 @@ SOperatorInfo* createOperatorTree(SPhysiNode* pPhyNode, SExecTaskInfo* pTaskInfo
           return NULL;
         }
       } else {  // Create one table group.
-        STableKeyInfo info = {.lastKey = 0, .uid = pBlockNode->uid};
+        STableKeyInfo info = {.lastKey = 0, .uid = pBlockNode->uid, .groupId = 0};
         taosArrayPush(pTableListInfo->pTableList, &info);
       }
 

source/libs/executor/src/scanoperator.c

@@ -507,20 +507,21 @@ static SSDataBlock* doTableScan(SOperatorInfo* pOperator) {
   STableScanInfo* pInfo = pOperator->info;
   SExecTaskInfo*  pTaskInfo = pOperator->pTaskInfo;
 
-  if(pInfo->currentGroupId == -1){
+  if (pInfo->currentGroupId == -1) {
     pInfo->currentGroupId++;
     if (pInfo->currentGroupId >= taosArrayGetSize(pTaskInfo->tableqinfoList.pGroupList)) {
       setTaskStatus(pTaskInfo, TASK_COMPLETED);
       return NULL;
     }
-    SArray *tableList = taosArrayGetP(pTaskInfo->tableqinfoList.pGroupList, pInfo->currentGroupId);
+    SArray* tableList = taosArrayGetP(pTaskInfo->tableqinfoList.pGroupList, pInfo->currentGroupId);
     tsdbCleanupReadHandle(pInfo->dataReader);
-    tsdbReaderT* pReader = tsdbReaderOpen(pInfo->readHandle.vnode, &pInfo->cond, tableList, pInfo->queryId, pInfo->taskId);
+    tsdbReaderT* pReader =
+        tsdbReaderOpen(pInfo->readHandle.vnode, &pInfo->cond, tableList, pInfo->queryId, pInfo->taskId);
     pInfo->dataReader = pReader;
   }
 
   SSDataBlock* result = doTableScanGroup(pOperator);
-  if(result){
+  if (result) {
     return result;
   }
 
@@ -530,7 +531,7 @@ static SSDataBlock* doTableScan(SOperatorInfo* pOperator) {
     return NULL;
   }
 
-  SArray *tableList = taosArrayGetP(pTaskInfo->tableqinfoList.pGroupList, pInfo->currentGroupId);
+  SArray* tableList = taosArrayGetP(pTaskInfo->tableqinfoList.pGroupList, pInfo->currentGroupId);
   tsdbSetTableList(pInfo->dataReader, tableList);
 
   tsdbResetReadHandle(pInfo->dataReader, &pInfo->cond, 0);
@@ -538,7 +539,7 @@ static SSDataBlock* doTableScan(SOperatorInfo* pOperator) {
   pInfo->scanTimes = 0;
 
   result = doTableScanGroup(pOperator);
-  if(result){
+  if (result) {
     return result;
   }
 
@@ -776,15 +777,14 @@ static bool isStateWindow(SStreamBlockScanInfo* pInfo) {
   return pInfo->sessionSup.parentType == QUERY_NODE_PHYSICAL_PLAN_STREAM_STATE;
 }
 
-static bool prepareDataScan(SStreamBlockScanInfo* pInfo) {
+static bool prepareDataScan(SStreamBlockScanInfo* pInfo, SSDataBlock* pSDB, int32_t tsColIndex, int32_t* pRowIndex) {
-  SSDataBlock* pSDB = pInfo->pUpdateRes;
   STimeWindow win = {
       .skey = INT64_MIN,
       .ekey = INT64_MAX,
   };
   bool needRead = false;
-  if (!isStateWindow(pInfo) && pInfo->updateResIndex < pSDB->info.rows) {
+  if (!isStateWindow(pInfo) && (*pRowIndex) < pSDB->info.rows) {
-    SColumnInfoData* pColDataInfo = taosArrayGet(pSDB->pDataBlock, pInfo->primaryTsIndex);
+    SColumnInfoData* pColDataInfo = taosArrayGet(pSDB->pDataBlock, tsColIndex);
     TSKEY*           tsCols = (TSKEY*)pColDataInfo->pData;
     SResultRowInfo   dumyInfo;
     dumyInfo.cur.pageId = -1;
@@ -793,15 +793,14 @@ static bool prepareDataScan(SStreamBlockScanInfo* pInfo) {
       int64_t              gap = pInfo->sessionSup.gap;
       int32_t              winIndex = 0;
       SResultWindowInfo*   pCurWin =
-          getSessionTimeWindow(pAggSup, tsCols[pInfo->updateResIndex], INT64_MIN, pSDB->info.groupId, gap, &winIndex);
+          getSessionTimeWindow(pAggSup, tsCols[(*pRowIndex)], INT64_MIN, pSDB->info.groupId, gap, &winIndex);
       win = pCurWin->win;
-      pInfo->updateResIndex +=
+      (*pRowIndex) += updateSessionWindowInfo(pCurWin, tsCols, NULL, pSDB->info.rows, (*pRowIndex), gap, NULL);
-          updateSessionWindowInfo(pCurWin, tsCols, NULL, pSDB->info.rows, pInfo->updateResIndex, gap, NULL);
     } else {
-      win = getActiveTimeWindow(NULL, &dumyInfo, tsCols[pInfo->updateResIndex], &pInfo->interval,
+      win =
-                                pInfo->interval.precision, NULL);
+          getActiveTimeWindow(NULL, &dumyInfo, tsCols[(*pRowIndex)], &pInfo->interval, pInfo->interval.precision, NULL);
-      pInfo->updateResIndex += getNumOfRowsInTimeWindow(&pSDB->info, tsCols, pInfo->updateResIndex, win.ekey,
+      (*pRowIndex) += getNumOfRowsInTimeWindow(&pSDB->info, tsCols, (*pRowIndex), win.ekey, binarySearchForKey, NULL,
-                                                        binarySearchForKey, NULL, TSDB_ORDER_ASC);
+                                               TSDB_ORDER_ASC);
     }
     needRead = true;
   } else if (isStateWindow(pInfo)) {
@@ -822,7 +821,10 @@ static bool prepareDataScan(SStreamBlockScanInfo* pInfo) {
   STableScanInfo* pTableScanInfo = pInfo->pSnapshotReadOp->info;
   pTableScanInfo->cond.twindows[0] = win;
   pTableScanInfo->curTWinIdx = 0;
-//  tsdbResetReadHandle(pTableScanInfo->dataReader, &pTableScanInfo->cond, 0);
+  //  tsdbResetReadHandle(pTableScanInfo->dataReader, &pTableScanInfo->cond, 0);
+  // if (!pTableScanInfo->dataReader) {
+  //   return false;
+  // }
   pTableScanInfo->scanTimes = 0;
   pTableScanInfo->currentGroupId = -1;
   return true;
@@ -862,12 +864,12 @@ static uint64_t getGroupId(SOperatorInfo* pOperator, SSDataBlock* pBlock, int32_
   */
 }
 
-static SSDataBlock* doDataScan(SStreamBlockScanInfo* pInfo) {
+static SSDataBlock* doDataScan(SStreamBlockScanInfo* pInfo, SSDataBlock* pSDB, int32_t tsColIndex, int32_t* pRowIndex) {
   while (1) {
     SSDataBlock* pResult = NULL;
     pResult = doTableScan(pInfo->pSnapshotReadOp);
     if (pResult == NULL) {
-      if (prepareDataScan(pInfo)) {
+      if (prepareDataScan(pInfo, pSDB, tsColIndex, pRowIndex)) {
         // scan next window data
         pResult = doTableScan(pInfo->pSnapshotReadOp);
       }
@@ -916,7 +918,7 @@ static void setUpdateData(SStreamBlockScanInfo* pInfo, SSDataBlock* pBlock, SSDa
     pUpdateBlock->info.rows = i;
     pInfo->tsArrayIndex += i;
     pUpdateBlock->info.groupId = pInfo->groupId;
-    pUpdateBlock->info.type = STREAM_REPROCESS;
+    pUpdateBlock->info.type = STREAM_CLEAR;
     blockDataUpdateTsWindow(pUpdateBlock, 0);
   }
   // all rows have same group id
@@ -970,6 +972,14 @@ static SSDataBlock* doStreamBlockScan(SOperatorInfo* pOperator) {
     int32_t      current = pInfo->validBlockIndex++;
     SSDataBlock* pBlock = taosArrayGetP(pInfo->pBlockLists, current);
     blockDataUpdateTsWindow(pBlock, 0);
+    if (pBlock->info.type == STREAM_RETRIEVE) {
+      pInfo->blockType = STREAM_DATA_TYPE_SUBMIT_BLOCK;
+      pInfo->scanMode = STREAM_SCAN_FROM_DATAREADER_RETRIEVE;
+      copyDataBlock(pInfo->pPullDataRes, pBlock);
+      pInfo->pullDataResIndex = 0;
+      prepareDataScan(pInfo, pInfo->pPullDataRes, 0, &pInfo->pullDataResIndex);
+      updateInfoAddCloseWindowSBF(pInfo->pUpdateInfo);
+    }
     return pBlock;
   } else if (pInfo->blockType == STREAM_DATA_TYPE_SUBMIT_BLOCK) {
     if (pInfo->scanMode == STREAM_SCAN_FROM_RES) {
@@ -979,28 +989,39 @@ static SSDataBlock* doStreamBlockScan(SOperatorInfo* pOperator) {
     } else if (pInfo->scanMode == STREAM_SCAN_FROM_UPDATERES) {
       pInfo->scanMode = STREAM_SCAN_FROM_DATAREADER;
       if (!isStateWindow(pInfo)) {
-        prepareDataScan(pInfo);
+        prepareDataScan(pInfo, pInfo->pUpdateRes, pInfo->primaryTsIndex, &pInfo->updateResIndex);
       }
       return pInfo->pUpdateRes;
+    } else if (pInfo->scanMode == STREAM_SCAN_FROM_DATAREADER_RETRIEVE) {
+      SSDataBlock* pSDB = doDataScan(pInfo, pInfo->pPullDataRes, 0, &pInfo->pullDataResIndex);
+      if (pSDB != NULL) {
+        getUpdateDataBlock(pInfo, true, pSDB, NULL);
+        pSDB->info.type = STREAM_PUSH_DATA;
+        return pSDB;
+      }
+      pInfo->scanMode = STREAM_SCAN_FROM_DATAREADER;
     } else {
       if (isStateWindow(pInfo) && taosArrayGetSize(pInfo->sessionSup.pStreamAggSup->pScanWindow) > 0) {
         pInfo->scanMode = STREAM_SCAN_FROM_DATAREADER;
         pInfo->updateResIndex = pInfo->pUpdateRes->info.rows;
-        prepareDataScan(pInfo);
+        prepareDataScan(pInfo, pInfo->pUpdateRes, pInfo->primaryTsIndex, &pInfo->updateResIndex);
       }
       if (pInfo->scanMode == STREAM_SCAN_FROM_DATAREADER) {
-        SSDataBlock* pSDB = doDataScan(pInfo);
+        SSDataBlock* pSDB = doDataScan(pInfo, pInfo->pUpdateRes, pInfo->primaryTsIndex, &pInfo->updateResIndex);
         if (pSDB == NULL) {
           setUpdateData(pInfo, pInfo->pRes, pInfo->pUpdateRes);
           if (pInfo->pUpdateRes->info.rows > 0) {
             if (!isStateWindow(pInfo)) {
-              prepareDataScan(pInfo);
+              // Todo(liuyao) mybe can delete this.
+              bool test = prepareDataScan(pInfo, pInfo->pUpdateRes, pInfo->primaryTsIndex, &pInfo->updateResIndex);
+              ASSERT(test == false);
             }
             return pInfo->pUpdateRes;
           } else {
             pInfo->scanMode = STREAM_SCAN_FROM_READERHANDLE;
           }
         } else {
+          pSDB->info.type = STREAM_NORMAL;
           getUpdateDataBlock(pInfo, true, pSDB, NULL);
           return pSDB;
         }
@@ -1012,12 +1033,13 @@ static SSDataBlock* doStreamBlockScan(SOperatorInfo* pOperator) {
 
     while (tqNextDataBlock(pInfo->streamBlockReader)) {
       SSDataBlock block = {0};
-      uint64_t    groupId = 0;
-      uint64_t    uid = 0;
-      int32_t     numOfRows = 0;
 
       // todo refactor
-      int32_t code = tqRetrieveDataBlock(&block, pInfo->streamBlockReader, &groupId, &uid, &numOfRows);
+      int32_t code = tqRetrieveDataBlock(&block, pInfo->streamBlockReader);
+
+      uint64_t groupId = block.info.groupId;
+      uint64_t uid = block.info.uid;
+      int32_t  numOfRows = block.info.rows;
 
       if (code != TSDB_CODE_SUCCESS || numOfRows == 0) {
         pTaskInfo->code = code;
@@ -1070,10 +1092,12 @@ static SSDataBlock* doStreamBlockScan(SOperatorInfo* pOperator) {
       taosArrayDestroy(block.pDataBlock);
       if (pInfo->pRes->pDataBlock == NULL) {
         // TODO add log
+        updateInfoDestoryColseWinSBF(pInfo->pUpdateInfo);
         pOperator->status = OP_EXEC_DONE;
         pTaskInfo->code = terrno;
         return NULL;
       }
+
       // currently only the tbname pseudo column
       if (pInfo->numOfPseudoExpr > 0) {
         addTagPseudoColumnData(&pInfo->readHandle, pInfo->pPseudoExpr, pInfo->numOfPseudoExpr, pInfo->pRes);
@@ -1091,12 +1115,13 @@ static SSDataBlock* doStreamBlockScan(SOperatorInfo* pOperator) {
     pOperator->resultInfo.totalRows += pBlockInfo->rows;
 
     if (pBlockInfo->rows == 0) {
+      updateInfoDestoryColseWinSBF(pInfo->pUpdateInfo);
       pOperator->status = OP_EXEC_DONE;
     } else if (pInfo->pUpdateInfo) {
       pInfo->tsArrayIndex = 0;
       getUpdateDataBlock(pInfo, true, pInfo->pRes, pInfo->pUpdateRes);
       if (pInfo->pUpdateRes->info.rows > 0) {
-        if (pInfo->pUpdateRes->info.type == STREAM_REPROCESS) {
+        if (pInfo->pUpdateRes->info.type == STREAM_CLEAR) {
           pInfo->updateResIndex = 0;
           pInfo->scanMode = STREAM_SCAN_FROM_UPDATERES;
         } else if (pInfo->pUpdateRes->info.type == STREAM_INVERT) {
@@ -1130,9 +1155,9 @@ static SArray* extractTableIdList(const STableListInfo* pTableGroupInfo) {
   return tableIdList;
 }
 
-SOperatorInfo* createStreamScanOperatorInfo(SReadHandle* pHandle,
+SOperatorInfo* createStreamScanOperatorInfo(SReadHandle* pHandle, STableScanPhysiNode* pTableScanNode,
-                                            STableScanPhysiNode* pTableScanNode, SExecTaskInfo* pTaskInfo,
+                                            SExecTaskInfo* pTaskInfo, STimeWindowAggSupp* pTwSup, uint64_t queryId,
-                                            STimeWindowAggSupp* pTwSup, uint64_t queryId, uint64_t taskId) {
+                                            uint64_t taskId) {
   SStreamBlockScanInfo* pInfo = taosMemoryCalloc(1, sizeof(SStreamBlockScanInfo));
   SOperatorInfo*        pOperator = taosMemoryCalloc(1, sizeof(SOperatorInfo));
 
@@ -1209,6 +1234,7 @@ SOperatorInfo* createStreamScanOperatorInfo(SReadHandle* pHandle,
   pInfo->scanMode = STREAM_SCAN_FROM_READERHANDLE;
   pInfo->sessionSup = (SessionWindowSupporter){.pStreamAggSup = NULL, .gap = -1};
   pInfo->groupId = 0;
+  pInfo->pPullDataRes = createPullDataBlock();
 
   pOperator->name = "StreamBlockScanOperator";
   pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_STREAM_SCAN;
@@ -1721,7 +1747,7 @@ SOperatorInfo* createSysTableScanOperatorInfo(void* readHandle, SSystemTableScan
   SArray* colList = extractColMatchInfo(pScanNode->pScanCols, pDescNode, &num, COL_MATCH_FROM_COL_ID);
 
   pInfo->accountId = pScanPhyNode->accountId;
-  pInfo->pUser       = taosMemoryStrDup((void*) pUser);
+  pInfo->pUser = taosMemoryStrDup((void*)pUser);
   pInfo->showRewrite = pScanPhyNode->showRewrite;
   pInfo->pRes = pResBlock;
   pInfo->pCondition = pScanNode->node.pConditions;
@@ -1965,7 +1991,10 @@ _error:
 
 typedef struct STableMergeScanInfo {
   STableListInfo* tableListInfo;
-  int32_t         currentGroupId;
+  int32_t         tableStartIndex;
+  int32_t         tableEndIndex;
+  bool            hasGroupId;
+  uint64_t        groupId;
 
   SArray*     dataReaders;  // array of tsdbReaderT*
   SReadHandle readHandle;
@@ -2030,6 +2059,22 @@ int32_t createScanTableListInfo(STableScanPhysiNode* pTableScanNode, SReadHandle
   return TSDB_CODE_SUCCESS;
 }
 
+int32_t createMultipleDataReaders(SQueryTableDataCond* pQueryCond, SReadHandle* pHandle, STableListInfo* pTableListInfo,
+                                  int32_t tableStartIdx, int32_t tableEndIdx, SArray* arrayReader, uint64_t queryId,
+                                  uint64_t taskId) {
+  for (int32_t i = tableStartIdx; i <= tableEndIdx; ++i) {
+    SArray* subTableList = taosArrayInit(1, sizeof(STableKeyInfo));
+    taosArrayPush(subTableList, taosArrayGet(pTableListInfo->pTableList, i));
+
+    tsdbReaderT* pReader = tsdbReaderOpen(pHandle->vnode, pQueryCond, subTableList, queryId, taskId);
+    taosArrayPush(arrayReader, &pReader);
+
+    taosArrayDestroy(subTableList);
+  }
+
+  return TSDB_CODE_SUCCESS;
+}
+
 // todo refactor
 static int32_t loadDataBlockFromOneTable(SOperatorInfo* pOperator, STableMergeScanInfo* pTableScanInfo,
                                          int32_t readerIdx, SSDataBlock* pBlock, uint32_t* status) {
@@ -2216,34 +2261,32 @@ SArray* generateSortByTsInfo(int32_t order) {
   return pList;
 }
 
-static int32_t createMultipleDataReaders(SQueryTableDataCond* pQueryCond, SReadHandle* pHandle, SArray* tableList, SArray* arrayReader, uint64_t queryId,
-                                  uint64_t taskId) {
-  for (int32_t i = 0; i < taosArrayGetSize(tableList); ++i) {
-    SArray* tmp = taosArrayInit(1, sizeof(STableKeyInfo));
-    taosArrayPush(tmp, taosArrayGet(tableList, i));
-
-    tsdbReaderT* pReader = tsdbReaderOpen(pHandle->vnode, pQueryCond, tmp, queryId, taskId);
-    taosArrayPush(arrayReader, &pReader);
-
-    taosArrayDestroy(tmp);
-  }
-
-  return TSDB_CODE_SUCCESS;
-}
-
 int32_t startGroupTableMergeScan(SOperatorInfo* pOperator) {
   STableMergeScanInfo* pInfo = pOperator->info;
   SExecTaskInfo*       pTaskInfo = pOperator->pTaskInfo;
 
-  SArray* tableList = taosArrayGetP(pInfo->tableListInfo->pGroupList, pInfo->currentGroupId);
+  {
+    size_t  tableListSize = taosArrayGetSize(pInfo->tableListInfo->pTableList);
+    int32_t i = pInfo->tableStartIndex + 1;
+    for (; i < tableListSize; ++i) {
+      STableKeyInfo* tableKeyInfo = taosArrayGet(pInfo->tableListInfo->pTableList, i);
+      if (tableKeyInfo->groupId != pInfo->groupId) {
+        break;
+      }
+    }
+    pInfo->tableEndIndex = i - 1;
+  }
 
-  createMultipleDataReaders(&pInfo->cond, &pInfo->readHandle, tableList,
+  int32_t tableStartIdx = pInfo->tableStartIndex;
+  int32_t tableEndIdx = pInfo->tableEndIndex;
+
+  STableListInfo* tableListInfo = pInfo->tableListInfo;
+  createMultipleDataReaders(&pInfo->cond, &pInfo->readHandle, tableListInfo, tableStartIdx, tableEndIdx,
                             pInfo->dataReaders, pInfo->queryId, pInfo->taskId);
 
   // todo the total available buffer should be determined by total capacity of buffer of this task.
   // the additional one is reserved for merge result
-  int32_t tableLen = taosArrayGetSize(tableList);
+  pInfo->sortBufSize = pInfo->bufPageSize * (tableEndIdx - tableStartIdx + 1 + 1);
-  pInfo->sortBufSize = pInfo->bufPageSize * ((tableLen==0?1:tableLen) + 1);
   int32_t numOfBufPage = pInfo->sortBufSize / pInfo->bufPageSize;
   pInfo->pSortHandle = tsortCreateSortHandle(pInfo->pSortInfo, SORT_MULTISOURCE_MERGE, pInfo->bufPageSize, numOfBufPage,
                                              pInfo->pSortInputBlock, pTaskInfo->id.str);
@@ -2330,42 +2373,37 @@ SSDataBlock* doTableMergeScan(SOperatorInfo* pOperator) {
   if (code != TSDB_CODE_SUCCESS) {
     longjmp(pTaskInfo->env, code);
   }
+  size_t tableListSize = taosArrayGetSize(pInfo->tableListInfo->pTableList);
+  if (!pInfo->hasGroupId) {
+    pInfo->hasGroupId = true;
 
-  if (pInfo->currentGroupId == -1) {
+    if (tableListSize == 0) {
-    pInfo->currentGroupId++;
-    if (pInfo->currentGroupId >= taosArrayGetSize(pInfo->tableListInfo->pGroupList)) {
       doSetOperatorCompleted(pOperator);
       return NULL;
     }
+    pInfo->tableStartIndex = 0;
+    pInfo->groupId = ((STableKeyInfo*)taosArrayGet(pInfo->tableListInfo->pTableList, pInfo->tableStartIndex))->groupId;
     startGroupTableMergeScan(pOperator);
   }
-  SSDataBlock* pBlock = getSortedTableMergeScanBlockData(pInfo->pSortHandle, pOperator->resultInfo.capacity, pOperator);
+  SSDataBlock* pBlock = NULL;
-  if (pBlock != NULL) {
+  while (pInfo->tableStartIndex < tableListSize) {
-    uint64_t* groupId = taosHashGet(pInfo->tableListInfo->map, &(pBlock->info.uid), sizeof(uint64_t));
-    if(groupId) pBlock->info.groupId = *groupId;
-
-    pOperator->resultInfo.totalRows += pBlock->info.rows;
-    return pBlock;
-  }
-
-  stopGroupTableMergeScan(pOperator);
-  pInfo->currentGroupId++;
-  if (pInfo->currentGroupId >= taosArrayGetSize(pInfo->tableListInfo->pGroupList)) {
-    doSetOperatorCompleted(pOperator);
-    return NULL;
-  }
-  startGroupTableMergeScan(pOperator);
-
     pBlock = getSortedTableMergeScanBlockData(pInfo->pSortHandle, pOperator->resultInfo.capacity, pOperator);
     if (pBlock != NULL) {
-    uint64_t* groupId = taosHashGet(pInfo->tableListInfo->map, &(pBlock->info.uid), sizeof(uint64_t));
+      pBlock->info.groupId = pInfo->groupId;
-    if(groupId) pBlock->info.groupId = *groupId;
-
       pOperator->resultInfo.totalRows += pBlock->info.rows;
       return pBlock;
-  }
+    } else {
-
+      stopGroupTableMergeScan(pOperator);
+      if (pInfo->tableEndIndex >= tableListSize - 1) {
         doSetOperatorCompleted(pOperator);
+        break;
+      }
+      pInfo->tableStartIndex = pInfo->tableEndIndex + 1;
+      pInfo->groupId =
+          ((STableKeyInfo*)taosArrayGet(pInfo->tableListInfo->pTableList, pInfo->tableStartIndex))->groupId;
+      startGroupTableMergeScan(pOperator);
+    }
+  }
 
   return pBlock;
 }
@@ -2403,6 +2441,12 @@ int32_t getTableMergeScanExplainExecInfo(SOperatorInfo* pOptr, void** pOptrExpla
   return TSDB_CODE_SUCCESS;
 }
 
+int32_t compareTableKeyInfoByGid(const void* p1, const void* p2) {
+  const STableKeyInfo* info1 = p1;
+  const STableKeyInfo* info2 = p2;
+  return info1->groupId - info2->groupId;
+}
+
 SOperatorInfo* createTableMergeScanOperatorInfo(STableScanPhysiNode* pTableScanNode, STableListInfo* pTableListInfo,
                                                 SReadHandle* readHandle, SExecTaskInfo* pTaskInfo, uint64_t queryId,
                                                 uint64_t taskId) {
@@ -2411,6 +2455,9 @@ SOperatorInfo* createTableMergeScanOperatorInfo(STableScanPhysiNode* pTableScanN
   if (pInfo == NULL || pOperator == NULL) {
     goto _error;
   }
+  if (pTableScanNode->pPartitionTags) {
+    taosArraySort(pTableListInfo->pTableList, compareTableKeyInfoByGid);
+  }
 
   SDataBlockDescNode* pDescNode = pTableScanNode->scan.node.pOutputDataBlockDesc;
 
@@ -2443,7 +2490,6 @@ SOperatorInfo* createTableMergeScanOperatorInfo(STableScanPhysiNode* pTableScanN
   pInfo->dataReaders = taosArrayInit(64, POINTER_BYTES);
   pInfo->queryId = queryId;
   pInfo->taskId = taskId;
-  pInfo->currentGroupId = -1;
 
   pInfo->sortSourceParams = taosArrayInit(64, sizeof(STableMergeScanSortSourceParam));
 

source/libs/executor/src/timewindowoperator.c

@@ -15,6 +15,7 @@
 #include "executorimpl.h"
 #include "function.h"
 #include "functionMgt.h"
+#include "tcompare.h"
 #include "tdatablock.h"
 #include "tfill.h"
 #include "ttime.h"
@@ -26,6 +27,16 @@ typedef enum SResultTsInterpType {
 
 #define IS_FINAL_OP(op) ((op)->isFinal)
 
+typedef struct SWinRes {
+  TSKEY    ts;
+  uint64_t groupId;
+} SWinRes;
+
+typedef struct SPullWindowInfo {
+  STimeWindow window;
+  uint64_t groupId;
+} SPullWindowInfo;
+
 static SSDataBlock* doStreamSessionAgg(SOperatorInfo* pOperator);
 
 static int64_t* extractTsCol(SSDataBlock* pBlock, const SIntervalAggOperatorInfo* pInfo);
@@ -684,11 +695,13 @@ static void doInterpUnclosedTimeWindow(SOperatorInfo* pOperatorInfo, int32_t num
 }
 
 void printDataBlock(SSDataBlock* pBlock, const char* flag) {
-  if (pBlock == NULL) return;
+  if (pBlock == NULL){
-  SArray* blocks = taosArrayInit(1, sizeof(SSDataBlock));
+    qDebug("======printDataBlock Block is Null");
-  taosArrayPush(blocks, pBlock);
+    return;
-  blockDebugShowData(blocks, flag);
+  }
-  taosArrayDestroy(blocks);
+  char *pBuf = NULL;
+  qDebug("%s", dumpBlockData(pBlock, flag, &pBuf));
+  taosMemoryFree(pBuf);
 }
 
 typedef int64_t (*__get_value_fn_t)(void* data, int32_t index);
@@ -1217,30 +1230,40 @@ void doClearWindowImpl(SResultRowPosition* p1, SDiskbasedBuf* pResultBuf, SExprS
   }
 }
 
-void doClearWindow(SAggSupporter* pAggSup, SExprSupp* pSup, char* pData, int16_t bytes, uint64_t groupId,
+bool doClearWindow(SAggSupporter* pAggSup, SExprSupp* pSup, char* pData, int16_t bytes, uint64_t groupId,
                    int32_t numOfOutput) {
   SET_RES_WINDOW_KEY(pAggSup->keyBuf, pData, bytes, groupId);
   SResultRowPosition* p1 =
       (SResultRowPosition*)taosHashGet(pAggSup->pResultRowHashTable, pAggSup->keyBuf, GET_RES_WINDOW_KEY_LEN(bytes));
   if (!p1) {
     // window has been closed
-    return;
+    return false;
   }
   doClearWindowImpl(p1, pAggSup->pResultBuf, pSup, numOfOutput);
+  return true;
 }
 
 static void doClearWindows(SAggSupporter* pAggSup, SExprSupp* pSup1, SInterval* pInterval, int32_t tsIndex,
                            int32_t numOfOutput, SSDataBlock* pBlock, SArray* pUpWins) {
-  SColumnInfoData* pColDataInfo = taosArrayGet(pBlock->pDataBlock, tsIndex);
+  SColumnInfoData* pTsCol = taosArrayGet(pBlock->pDataBlock, tsIndex);
-  TSKEY*           tsCols = (TSKEY*)pColDataInfo->pData;
+  TSKEY*           tsCols = (TSKEY*)pTsCol->pData;
+  uint64_t*        pGpDatas = NULL;
+  if (pBlock->info.type == STREAM_RETRIEVE) {
+      SColumnInfoData* pGpCol = taosArrayGet(pBlock->pDataBlock, 2);
+      pGpDatas = (uint64_t*)pGpCol->pData;
+  }
   int32_t          step = 0;
   for (int32_t i = 0; i < pBlock->info.rows; i += step) {
     SResultRowInfo dumyInfo;
     dumyInfo.cur.pageId = -1;
     STimeWindow win = getActiveTimeWindow(NULL, &dumyInfo, tsCols[i], pInterval, pInterval->precision, NULL);
     step = getNumOfRowsInTimeWindow(&pBlock->info, tsCols, i, win.ekey, binarySearchForKey, NULL, TSDB_ORDER_ASC);
-    doClearWindow(pAggSup, pSup1, (char*)&win.skey, sizeof(TKEY), pBlock->info.groupId, numOfOutput);
+    uint64_t groupId = pBlock->info.groupId;
-    if (pUpWins) {
+    if (pGpDatas) {
+      groupId = pGpDatas[i];
+    }
+    bool res = doClearWindow(pAggSup, pSup1, (char*)&win.skey, sizeof(TKEY), groupId, numOfOutput);
+    if (pUpWins && res) {
       taosArrayPush(pUpWins, &win);
     }
   }
@@ -1268,8 +1291,8 @@ bool isCloseWindow(STimeWindow* pWin, STimeWindowAggSupp* pSup) {
   return pSup->maxTs != INT64_MIN && pWin->ekey < pSup->maxTs - pSup->waterMark;
 }
 
-static int32_t closeIntervalWindow(SHashObj* pHashMap, STimeWindowAggSupp* pSup, SInterval* pInterval,
+static int32_t closeIntervalWindow(SHashObj* pHashMap, STimeWindowAggSupp* pSup,
-                                   SArray* closeWins) {
+    SInterval* pInterval, SHashObj* pPullDataMap, SArray* closeWins) {
   void*  pIte = NULL;
   size_t keyLen = 0;
   while ((pIte = taosHashIterate(pHashMap, pIte)) != NULL) {
@@ -1280,10 +1303,20 @@ static int32_t closeIntervalWindow(SHashObj* pHashMap, STimeWindowAggSupp* pSup,
     SResultRowInfo dumyInfo;
     dumyInfo.cur.pageId = -1;
     STimeWindow win = getActiveTimeWindow(NULL, &dumyInfo, ts, pInterval, pInterval->precision, NULL);
+    SWinRes winRe = {.ts = win.skey, .groupId = groupId,};
+    void* chIds = taosHashGet(pPullDataMap, &winRe, sizeof(SWinRes));
     if (isCloseWindow(&win, pSup)) {
-      char keyBuf[GET_RES_WINDOW_KEY_LEN(sizeof(TSKEY))];
+      if (chIds && pPullDataMap) {
-      SET_RES_WINDOW_KEY(keyBuf, &ts, sizeof(TSKEY), groupId);
+        SArray* chAy = *(SArray**) chIds;
-      taosHashRemove(pHashMap, keyBuf, keyLen);
+        int32_t size = taosArrayGetSize(chAy);
+        qInfo("======window %ld wait child size:%d", win.skey ,size);
+        for (int32_t i = 0; i < size; i++) {
+          qInfo("======window %ld wait chid id:%d", win.skey ,*(int32_t*)taosArrayGet(chAy, i));
+        }
+        continue;
+      } else if (pPullDataMap) {
+        qInfo("======close window %ld", win.skey);
+      }
       SResultRowPosition* pPos = (SResultRowPosition*)pIte;
       if (pSup->calTrigger == STREAM_TRIGGER_WINDOW_CLOSE) {
         int32_t code = saveResult(ts, pPos->pageId, pPos->offset, groupId, closeWins);
@@ -1291,11 +1324,25 @@ static int32_t closeIntervalWindow(SHashObj* pHashMap, STimeWindowAggSupp* pSup,
           return code;
         }
       }
+      char keyBuf[GET_RES_WINDOW_KEY_LEN(sizeof(TSKEY))];
+      SET_RES_WINDOW_KEY(keyBuf, &ts, sizeof(TSKEY), groupId);
+      taosHashRemove(pHashMap, keyBuf, keyLen);
     }
   }
   return TSDB_CODE_SUCCESS;
 }
 
+static void closeChildIntervalWindow(SArray* pChildren, TSKEY maxTs) {
+  int32_t size = taosArrayGetSize(pChildren);
+  for (int32_t i = 0; i < size; i++) {
+    SOperatorInfo* pChildOp = taosArrayGetP(pChildren, i);
+    SStreamFinalIntervalOperatorInfo* pChInfo = pChildOp->info;
+    ASSERT(pChInfo->twAggSup.calTrigger == STREAM_TRIGGER_AT_ONCE);
+    pChInfo->twAggSup.maxTs = TMAX(pChInfo->twAggSup.maxTs, maxTs);
+    closeIntervalWindow(pChInfo->aggSup.pResultRowHashTable, &pChInfo->twAggSup, &pChInfo->interval, NULL, NULL);
+  }
+}
+
 static SSDataBlock* doStreamIntervalAgg(SOperatorInfo* pOperator) {
   SIntervalAggOperatorInfo* pInfo = pOperator->info;
   SExecTaskInfo*            pTaskInfo = pOperator->pTaskInfo;
@@ -1324,7 +1371,7 @@ static SSDataBlock* doStreamIntervalAgg(SOperatorInfo* pOperator) {
       break;
     }
 
-    if (pBlock->info.type == STREAM_REPROCESS) {
+    if (pBlock->info.type == STREAM_CLEAR) {
       doClearWindows(&pInfo->aggSup, &pOperator->exprSupp, &pInfo->interval, 0, pOperator->exprSupp.numOfExprs, pBlock,
                      NULL);
       qDebug("%s clear existed time window results for updates checked", GET_TASKID(pTaskInfo));
@@ -1345,7 +1392,7 @@ static SSDataBlock* doStreamIntervalAgg(SOperatorInfo* pOperator) {
     pInfo->twAggSup.maxTs = TMAX(pInfo->twAggSup.maxTs, pBlock->info.window.ekey);
     hashIntervalAgg(pOperator, &pInfo->binfo.resultRowInfo, pBlock, MAIN_SCAN, pUpdated);
   }
-  closeIntervalWindow(pInfo->aggSup.pResultRowHashTable, &pInfo->twAggSup, &pInfo->interval, pUpdated);
+  closeIntervalWindow(pInfo->aggSup.pResultRowHashTable, &pInfo->twAggSup, &pInfo->interval, NULL, pUpdated);
 
   finalizeUpdatedResult(pOperator->exprSupp.numOfExprs, pInfo->aggSup.pResultBuf, pUpdated, pSup->rowEntryInfoOffset);
   initMultiResInfoFromArrayList(&pInfo->groupResInfo, pUpdated);
@@ -1373,6 +1420,11 @@ void destroyStreamFinalIntervalOperatorInfo(void* param, int32_t numOfOutput) {
   SStreamFinalIntervalOperatorInfo* pInfo = (SStreamFinalIntervalOperatorInfo*)param;
   cleanupBasicInfo(&pInfo->binfo);
   cleanupAggSup(&pInfo->aggSup);
+  //it should be empty.
+  taosHashCleanup(pInfo->pPullDataMap);
+  taosArrayDestroy(pInfo->pPullWins);
+  blockDataDestroy(pInfo->pPullDataRes);
+
   if (pInfo->pChildren) {
     int32_t size = taosArrayGetSize(pInfo->pChildren);
     for (int32_t i = 0; i < size; i++) {
@@ -2164,6 +2216,24 @@ bool isDeletedWindow(STimeWindow* pWin, uint64_t groupId, SAggSupporter* pSup) {
   return p1 == NULL;
 }
 
+int32_t getNexWindowPos(SInterval* pInterval, SDataBlockInfo* pBlockInfo, TSKEY* tsCols,
+    int32_t startPos, TSKEY eKey, STimeWindow* pNextWin) {
+  int32_t forwardRows = getNumOfRowsInTimeWindow(pBlockInfo, tsCols, startPos,
+      eKey, binarySearchForKey, NULL, TSDB_ORDER_ASC);
+  int32_t prevEndPos = forwardRows - 1 + startPos;
+  return getNextQualifiedWindow(pInterval, pNextWin, pBlockInfo, tsCols, prevEndPos, TSDB_ORDER_ASC);
+}
+
+void addPullWindow(SHashObj* pMap, SWinRes* pWinRes, int32_t size) {
+  SArray* childIds = taosArrayInit(8, sizeof(int32_t));
+  for (int32_t i = 0; i < size; i++) {
+    taosArrayPush(childIds, &i);
+  }
+  taosHashPut(pMap, pWinRes, sizeof(SWinRes), &childIds, sizeof(void*));
+}
+
+static int32_t getChildIndex(SSDataBlock* pBlock) { return pBlock->info.childId; }
+
 static void doHashInterval(SOperatorInfo* pOperatorInfo, SSDataBlock* pSDataBlock, uint64_t tableGroupId,
                            SArray* pUpdated) {
   SStreamFinalIntervalOperatorInfo* pInfo = (SStreamFinalIntervalOperatorInfo*)pOperatorInfo->info;
@@ -2177,35 +2247,59 @@ static void doHashInterval(SOperatorInfo* pOperatorInfo, SSDataBlock* pSDataBloc
   SResultRow*                       pResult = NULL;
   int32_t                           forwardRows = 0;
 
-  if (pSDataBlock->pDataBlock != NULL) {
+  ASSERT(pSDataBlock->pDataBlock != NULL);
   SColumnInfoData* pColDataInfo = taosArrayGet(pSDataBlock->pDataBlock, pInfo->primaryTsIndex);
   tsCols = (int64_t*)pColDataInfo->pData;
-  } else {
-    return;
-  }
 
   int32_t     startPos = ascScan ? 0 : (pSDataBlock->info.rows - 1);
   TSKEY       ts = getStartTsKey(&pSDataBlock->info.window, tsCols);
   STimeWindow nextWin = getActiveTimeWindow(pInfo->aggSup.pResultBuf, pResultRowInfo, ts, &pInfo->interval,
                                             pInfo->interval.precision, NULL);
   while (1) {
-    if (IS_FINAL_OP(pInfo) && isCloseWindow(&nextWin, &pInfo->twAggSup) &&
+    if (IS_FINAL_OP(pInfo) && isCloseWindow(&nextWin, &pInfo->twAggSup) && pInfo->pChildren) {
-        isDeletedWindow(&nextWin, tableGroupId, &pInfo->aggSup)) {
+      bool ignore = true;
-      SArray* pUpWins = taosArrayInit(8, sizeof(STimeWindow));
+      SWinRes winRes = {.ts = nextWin.skey, .groupId = tableGroupId,};
-      taosArrayPush(pUpWins, &nextWin);
+      void* chIds = taosHashGet(pInfo->pPullDataMap, &winRes, sizeof(SWinRes));
-      rebuildIntervalWindow(pInfo, pSup, pUpWins, pInfo->binfo.pRes->info.groupId, pSup->numOfExprs,
+      if (isDeletedWindow(&nextWin, tableGroupId, &pInfo->aggSup) && !chIds) {
-                            pOperatorInfo->pTaskInfo);
+        SPullWindowInfo pull = {.window = nextWin, .groupId = tableGroupId};
-      taosArrayDestroy(pUpWins);
+        // add pull data request
+        taosArrayPush(pInfo->pPullWins, &pull);
+        addPullWindow(pInfo->pPullDataMap, &winRes, taosArrayGetSize(pInfo->pChildren));
+      } else {
+        int32_t index = -1;
+        SArray* chArray = NULL;
+        if (chIds) {
+          chArray = *(void**) chIds;
+          int32_t chId = getChildIndex(pSDataBlock);
+          index = taosArraySearchIdx(chArray, &chId, compareInt32Val, TD_EQ);
         }
+        if (index != -1 && pSDataBlock->info.type == STREAM_PUSH_DATA) {
+          taosArrayRemove(chArray, index);
+          if (taosArrayGetSize(chArray) == 0) {
+            // pull data is over
+            taosHashRemove(pInfo->pPullDataMap, &winRes, sizeof(SWinRes));
+          }
+        }
+        if ( index == -1 || pSDataBlock->info.type == STREAM_PUSH_DATA) {
+          ignore = false;
+        }
+      }
+
+      if (ignore) {
+        startPos = getNexWindowPos(&pInfo->interval, &pSDataBlock->info, tsCols, startPos, nextWin.ekey, &nextWin);
+        if (startPos < 0) {
+          break;
+        }
+        continue;
+      }
+    }
+
     int32_t code = setTimeWindowOutputBuf(pResultRowInfo, &nextWin, true, &pResult, tableGroupId, pSup->pCtx,
                                           numOfOutput, pSup->rowEntryInfoOffset, &pInfo->aggSup, pTaskInfo);
     if (code != TSDB_CODE_SUCCESS || pResult == NULL) {
       longjmp(pTaskInfo->env, TSDB_CODE_QRY_OUT_OF_MEMORY);
     }
-    SResKeyPos* pos = taosMemoryMalloc(sizeof(SResKeyPos) + sizeof(uint64_t));
+
-    pos->groupId = tableGroupId;
-    pos->pos = (SResultRowPosition){.pageId = pResult->pageId, .offset = pResult->offset};
-    *(int64_t*)pos->key = pResult->win.skey;
     forwardRows = getNumOfRowsInTimeWindow(&pSDataBlock->info, tsCols, startPos, nextWin.ekey, binarySearchForKey, NULL,
                                            TSDB_ORDER_ASC);
     if (pInfo->twAggSup.calTrigger == STREAM_TRIGGER_AT_ONCE && pUpdated) {
@@ -2230,14 +2324,17 @@ static void clearStreamIntervalOperator(SStreamFinalIntervalOperatorInfo* pInfo)
   initResultRowInfo(&pInfo->binfo.resultRowInfo);
 }
 
-static void clearUpdateDataBlock(SSDataBlock* pBlock) {
+static void clearSpecialDataBlock(SSDataBlock* pBlock) {
+  if (pBlock->info.rows <= 0) {
+    return;
+  }
   blockDataCleanup(pBlock);
 }
 
 void copyUpdateDataBlock(SSDataBlock* pDest, SSDataBlock* pSource, int32_t tsColIndex) {
   // ASSERT(pDest->info.capacity >= pSource->info.rows);
   blockDataEnsureCapacity(pDest, pSource->info.rows);
-  clearUpdateDataBlock(pDest);
+  clearSpecialDataBlock(pDest);
   SColumnInfoData* pDestCol = taosArrayGet(pDest->pDataBlock, 0);
   SColumnInfoData* pSourceCol = taosArrayGet(pSource->pDataBlock, tsColIndex);
 
@@ -2254,7 +2351,63 @@ void copyUpdateDataBlock(SSDataBlock* pDest, SSDataBlock* pSource, int32_t tsCol
   blockDataUpdateTsWindow(pDest, 0);
 }
 
-static int32_t getChildIndex(SSDataBlock* pBlock) { return pBlock->info.childId; }
+static bool needBreak(SStreamFinalIntervalOperatorInfo* pInfo) {
+  int32_t size = taosArrayGetSize(pInfo->pPullWins);
+  if (pInfo->pullIndex < size) {
+    return true;
+  }
+  return false;
+}
+
+static void doBuildPullDataBlock(SArray* array, int32_t* pIndex, SSDataBlock* pBlock) {
+  clearSpecialDataBlock(pBlock);
+  int32_t size = taosArrayGetSize(array);
+  if (size - (*pIndex) == 0) {
+    return;
+  }
+  blockDataEnsureCapacity(pBlock, size - (*pIndex) );
+  ASSERT(3 <= taosArrayGetSize(pBlock->pDataBlock));
+  for (; (*pIndex) < size; (*pIndex)++) {
+    SPullWindowInfo* pWin = taosArrayGet(array, (*pIndex) );
+    SColumnInfoData* pStartTs = (SColumnInfoData*) taosArrayGet(pBlock->pDataBlock, 0);
+    colDataAppend(pStartTs, pBlock->info.rows, (const char*)&pWin->window.skey, false);
+
+    SColumnInfoData* pEndTs = (SColumnInfoData*) taosArrayGet(pBlock->pDataBlock, 1);
+    colDataAppend(pEndTs, pBlock->info.rows, (const char*)&pWin->window.ekey, false);
+
+    SColumnInfoData* pGroupId = (SColumnInfoData*) taosArrayGet(pBlock->pDataBlock, 2);
+    colDataAppend(pGroupId, pBlock->info.rows, (const char*)&pWin->groupId, false);
+    pBlock->info.rows++;
+  }
+  if ((*pIndex) == size) {
+    *pIndex = 0;
+    taosArrayClear(array);
+  }
+  blockDataUpdateTsWindow(pBlock, 0);
+}
+
+void processPushEmpty(SSDataBlock* pBlock, SHashObj* pMap) {
+  SColumnInfoData* pStartCol = taosArrayGet(pBlock->pDataBlock, 0);
+  TSKEY* tsData = (TSKEY*)pStartCol->pData;
+  SColumnInfoData* pGroupCol = taosArrayGet(pBlock->pDataBlock, 2);
+  uint64_t* groupIdData = (uint64_t*)pGroupCol->pData;
+  int32_t chId = getChildIndex(pBlock);
+  for (int32_t i = 0; i < pBlock->info.rows; i++) {
+    SWinRes winRes = {.ts = tsData[i], .groupId = groupIdData[i]};
+    void* chIds = taosHashGet(pMap, &winRes, sizeof(SWinRes));
+    if (chIds) {
+      SArray* chArray = *(SArray**) chIds;
+      int32_t index = taosArraySearchIdx(chArray, &chId, compareInt32Val, TD_EQ);
+      if (index != -1) {
+        taosArrayRemove(chArray, index);
+        if (taosArrayGetSize(chArray) == 0) {
+          // pull data is over
+          taosHashRemove(pMap, &winRes, sizeof(SWinRes));
+        }
+      }
+    }
+  }
+}
 
 static SSDataBlock* doStreamFinalIntervalAgg(SOperatorInfo* pOperator) {
   SStreamFinalIntervalOperatorInfo* pInfo = pOperator->info;
@@ -2270,28 +2423,50 @@ static SSDataBlock* doStreamFinalIntervalAgg(SOperatorInfo* pOperator) {
     doBuildResultDatablock(pOperator, &pInfo->binfo, &pInfo->groupResInfo, pInfo->aggSup.pResultBuf);
     if (pInfo->binfo.pRes->info.rows == 0) {
       pOperator->status = OP_EXEC_DONE;
-      if (IS_FINAL_OP(pInfo) || pInfo->pUpdateRes->info.rows == 0) {
       if (!IS_FINAL_OP(pInfo)) {
         // semi interval operator clear disk buffer
         clearStreamIntervalOperator(pInfo);
       }
       return NULL;
     }
+    printDataBlock(pInfo->binfo.pRes, IS_FINAL_OP(pInfo) ? "interval Final" : "interval  Semi");
+    return pInfo->binfo.pRes;
+  } else {
+    doBuildResultDatablock(pOperator, &pInfo->binfo, &pInfo->groupResInfo, pInfo->aggSup.pResultBuf);
+    if (pInfo->binfo.pRes->info.rows != 0) {
+      printDataBlock(pInfo->binfo.pRes, IS_FINAL_OP(pInfo) ? "interval Final" : "interval  Semi");
+      return pInfo->binfo.pRes;
+    }
+    if (pInfo->pUpdateRes->info.rows != 0 && pInfo->returnUpdate) {
+      pInfo->returnUpdate = false;
+      ASSERT(!IS_FINAL_OP(pInfo));
+      printDataBlock(pInfo->pUpdateRes, IS_FINAL_OP(pInfo) ? "interval Final" : "interval  Semi");
       // process the rest of the data
-      pOperator->status = OP_OPENED;
       return pInfo->pUpdateRes;
     }
-    return pInfo->binfo.pRes;
+    doBuildPullDataBlock(pInfo->pPullWins, &pInfo->pullIndex, pInfo->pPullDataRes);
+    if (pInfo->pPullDataRes->info.rows != 0) {
+      // process the rest of the data
+      ASSERT(IS_FINAL_OP(pInfo));
+      printDataBlock(pInfo->pPullDataRes, IS_FINAL_OP(pInfo) ? "interval Final" : "interval  Semi");
+      return pInfo->pPullDataRes;
+    }
   }
 
   while (1) {
     SSDataBlock* pBlock = downstream->fpSet.getNextFn(downstream);
     if (pBlock == NULL) {
-      clearUpdateDataBlock(pInfo->pUpdateRes);
+      clearSpecialDataBlock(pInfo->pUpdateRes);
+      pOperator->status = OP_RES_TO_RETURN;
+      qInfo("Stream Final Interval return data");
       break;
     }
+    printDataBlock(pBlock, IS_FINAL_OP(pInfo) ? "interval Final recv" : "interval  Semi recv");
+    maxTs = TMAX(maxTs, pBlock->info.window.ekey);
 
-    if (pBlock->info.type == STREAM_REPROCESS) {
+    if (pBlock->info.type == STREAM_NORMAL || pBlock->info.type == STREAM_PUSH_DATA || pBlock->info.type == STREAM_INVALID) {
+      pInfo->binfo.pRes->info.type = pBlock->info.type;
+    } else if (pBlock->info.type == STREAM_CLEAR) {
       SArray* pUpWins = taosArrayInit(8, sizeof(STimeWindow));
       doClearWindows(&pInfo->aggSup, pSup, &pInfo->interval, pInfo->primaryTsIndex, pOperator->exprSupp.numOfExprs,
                      pBlock, pUpWins);
@@ -2310,11 +2485,25 @@ static SSDataBlock* doStreamFinalIntervalAgg(SOperatorInfo* pOperator) {
       }
       removeResults(pUpWins, pUpdated);
       copyUpdateDataBlock(pInfo->pUpdateRes, pBlock, pInfo->primaryTsIndex);
+      pInfo->returnUpdate = true;
       taosArrayDestroy(pUpWins);
       break;
     } else if (pBlock->info.type == STREAM_GET_ALL && IS_FINAL_OP(pInfo)) {
       getAllIntervalWindow(pInfo->aggSup.pResultRowHashTable, pUpdated);
       continue;
+    } else if (pBlock->info.type == STREAM_RETRIEVE && !IS_FINAL_OP(pInfo)) {
+      SArray* pUpWins = taosArrayInit(8, sizeof(STimeWindow));
+      doClearWindows(&pInfo->aggSup, pSup, &pInfo->interval, 0, pOperator->exprSupp.numOfExprs,
+          pBlock, pUpWins);
+      removeResults(pUpWins, pUpdated);
+      taosArrayDestroy(pUpWins);
+      if (taosArrayGetSize(pUpdated) > 0) {
+        break;
+      }
+      continue;
+    } else if (pBlock->info.type == STREAM_PUSH_EMPTY && IS_FINAL_OP(pInfo)) {
+      processPushEmpty(pBlock, pInfo->pPullDataMap);
+      continue;
     }
 
     setInputDataBlock(pOperator, pSup->pCtx, pBlock, pInfo->order, MAIN_SCAN, true);
@@ -2334,31 +2523,70 @@ static SSDataBlock* doStreamFinalIntervalAgg(SOperatorInfo* pOperator) {
       SStreamFinalIntervalOperatorInfo* pChInfo = pChildOp->info;
       setInputDataBlock(pChildOp, pChildOp->exprSupp.pCtx, pBlock, pChInfo->order, MAIN_SCAN, true);
       doHashInterval(pChildOp, pBlock, pBlock->info.groupId, NULL);
-      pChInfo->twAggSup.maxTs = TMAX(pChInfo->twAggSup.maxTs, pBlock->info.window.ekey);
+
+      if (needBreak(pInfo)) {
+        break;
+      }
     }
-    maxTs = TMAX(maxTs, pBlock->info.window.ekey);
   }
 
   pInfo->twAggSup.maxTs = TMAX(pInfo->twAggSup.maxTs, maxTs);
   if (IS_FINAL_OP(pInfo)) {
-    closeIntervalWindow(pInfo->aggSup.pResultRowHashTable, &pInfo->twAggSup, &pInfo->interval, pUpdated);
+    closeIntervalWindow(pInfo->aggSup.pResultRowHashTable, &pInfo->twAggSup,
+        &pInfo->interval, pInfo->pPullDataMap, pUpdated);
+    closeChildIntervalWindow(pInfo->pChildren, pInfo->twAggSup.maxTs);
   }
 
   finalizeUpdatedResult(pOperator->exprSupp.numOfExprs, pInfo->aggSup.pResultBuf, pUpdated, pSup->rowEntryInfoOffset);
   initMultiResInfoFromArrayList(&pInfo->groupResInfo, pUpdated);
   blockDataEnsureCapacity(pInfo->binfo.pRes, pOperator->resultInfo.capacity);
   doBuildResultDatablock(pOperator, &pInfo->binfo, &pInfo->groupResInfo, pInfo->aggSup.pResultBuf);
-  pOperator->status = OP_RES_TO_RETURN;
+  if (pInfo->binfo.pRes->info.rows != 0) {
-  if (pInfo->binfo.pRes->info.rows == 0) {
+    printDataBlock(pInfo->binfo.pRes, IS_FINAL_OP(pInfo) ? "interval Final" : "interval  Semi");
-    pOperator->status = OP_EXEC_DONE;
+    return pInfo->binfo.pRes;
-    if (pInfo->pUpdateRes->info.rows == 0) {
-      return NULL;
   }
+
+  if (pInfo->pUpdateRes->info.rows != 0 && pInfo->returnUpdate) {
+    pInfo->returnUpdate = false;
+    ASSERT(!IS_FINAL_OP(pInfo));
+    printDataBlock(pInfo->pUpdateRes, IS_FINAL_OP(pInfo) ? "interval Final" : "interval  Semi");
     // process the rest of the data
-    pOperator->status = OP_OPENED;
     return pInfo->pUpdateRes;
   }
-  return pInfo->binfo.pRes;
+
+  doBuildPullDataBlock(pInfo->pPullWins, &pInfo->pullIndex, pInfo->pPullDataRes);
+  if (pInfo->pPullDataRes->info.rows != 0) {
+    // process the rest of the data
+    ASSERT(IS_FINAL_OP(pInfo));
+    printDataBlock(pInfo->pPullDataRes, IS_FINAL_OP(pInfo) ? "interval Final" : "interval  Semi");
+    return pInfo->pPullDataRes;
+  }
+  // ASSERT(false);
+  return NULL;
+}
+
+SSDataBlock* createPullDataBlock() {
+  SSDataBlock* pBlock = taosMemoryCalloc(1, sizeof(SSDataBlock));
+  pBlock->info.hasVarCol = false;
+  pBlock->info.groupId = 0;
+  pBlock->info.rows = 0;
+  pBlock->info.type = STREAM_RETRIEVE;
+  pBlock->info.rowSize = sizeof(TSKEY) + sizeof(TSKEY) + sizeof(uint64_t);
+
+  pBlock->pDataBlock = taosArrayInit(3, sizeof(SColumnInfoData));
+  SColumnInfoData infoData = {0};
+  infoData.info.type = TSDB_DATA_TYPE_TIMESTAMP;
+  infoData.info.bytes = sizeof(TSKEY);
+  // window start ts
+  taosArrayPush(pBlock->pDataBlock, &infoData);
+  // window end ts
+  taosArrayPush(pBlock->pDataBlock, &infoData);
+
+  infoData.info.type = TSDB_DATA_TYPE_UBIGINT;
+  infoData.info.bytes = sizeof(uint64_t);
+  taosArrayPush(pBlock->pDataBlock, &infoData);
+
+  return pBlock;
 }
 
 SOperatorInfo* createStreamFinalIntervalOperatorInfo(SOperatorInfo* downstream, SPhysiNode* pPhyNode,
@@ -2412,23 +2640,30 @@ SOperatorInfo* createStreamFinalIntervalOperatorInfo(SOperatorInfo* downstream,
       goto _error;
     }
   }
-  // semi interval operator does not catch result
   pInfo->pUpdateRes = createResDataBlock(pPhyNode->pOutputDataBlockDesc);
-  pInfo->pUpdateRes->info.type = STREAM_REPROCESS;
+  pInfo->pUpdateRes->info.type = STREAM_CLEAR;
   blockDataEnsureCapacity(pInfo->pUpdateRes, 128);
+  pInfo->returnUpdate = false;
+
   pInfo->pPhyNode = (SPhysiNode*)nodesCloneNode((SNode*)pPhyNode);
 
   if (pPhyNode->type == QUERY_NODE_PHYSICAL_PLAN_STREAM_FINAL_INTERVAL) {
     pInfo->isFinal = true;
     pOperator->name = "StreamFinalIntervalOperator";
   } else {
+    // semi interval operator does not catch result
     pInfo->isFinal = false;
     pOperator->name = "StreamSemiIntervalOperator";
   }
 
-  if (!IS_FINAL_OP(pInfo)) {
+  if (!IS_FINAL_OP(pInfo) || numOfChild == 0) {
     pInfo->twAggSup.calTrigger = STREAM_TRIGGER_AT_ONCE;
   }
+  pInfo->pPullWins = taosArrayInit(8, sizeof(SPullWindowInfo));
+  pInfo->pullIndex = 0;
+  _hash_fn_t hashFn = taosGetDefaultHashFunction(TSDB_DATA_TYPE_BINARY);
+  pInfo->pPullDataMap = taosHashInit(64, hashFn, false, HASH_NO_LOCK);
+  pInfo->pPullDataRes = createPullDataBlock();
 
   pOperator->operatorType = pPhyNode->type;
   pOperator->blocking = true;
@@ -2811,11 +3046,6 @@ void compactTimeWindow(SStreamSessionAggOperatorInfo* pInfo, int32_t startIndex,
   }
 }
 
-typedef struct SWinRes {
-  TSKEY    ts;
-  uint64_t groupId;
-} SWinRes;
-
 static void doStreamSessionAggImpl(SOperatorInfo* pOperator, SSDataBlock* pSDataBlock, SHashObj* pStUpdated,
                                    SHashObj* pStDeleted, bool hasEndTs) {
   SExecTaskInfo*                 pTaskInfo = pOperator->pTaskInfo;
@@ -3027,14 +3257,14 @@ static SSDataBlock* doStreamSessionAgg(SOperatorInfo* pOperator) {
   } else if (pOperator->status == OP_RES_TO_RETURN) {
     doBuildDeleteDataBlock(pInfo->pStDeleted, pInfo->pDelRes, &pInfo->pDelIterator);
     if (pInfo->pDelRes->info.rows > 0) {
-      /*printDataBlock(pInfo->pDelRes, "session del");*/
+      printDataBlock(pInfo->pDelRes, IS_FINAL_OP(pInfo)? "Final  Session" : "Single Session");
       return pInfo->pDelRes;
     }
     doBuildResultDatablock(pOperator, pBInfo, &pInfo->groupResInfo, pInfo->streamAggSup.pResultBuf);
     if (pBInfo->pRes->info.rows == 0 || !hasDataInGroupInfo(&pInfo->groupResInfo)) {
       doSetOperatorCompleted(pOperator);
     }
-    /*printDataBlock(pBInfo->pRes, "session insert");*/
+    printDataBlock(pBInfo->pRes, IS_FINAL_OP(pInfo)? "Final  Session" : "Single Session");
     return pBInfo->pRes->info.rows == 0 ? NULL : pBInfo->pRes;
   }
 
@@ -3048,7 +3278,7 @@ static SSDataBlock* doStreamSessionAgg(SOperatorInfo* pOperator) {
       break;
     }
 
-    if (pBlock->info.type == STREAM_REPROCESS) {
+    if (pBlock->info.type == STREAM_CLEAR) {
       SArray* pWins = taosArrayInit(16, sizeof(SResultWindowInfo));
       doClearSessionWindows(&pInfo->streamAggSup, &pOperator->exprSupp, pBlock, 0, pOperator->exprSupp.numOfExprs,
                             pInfo->gap, pWins);
@@ -3102,11 +3332,11 @@ static SSDataBlock* doStreamSessionAgg(SOperatorInfo* pOperator) {
   blockDataEnsureCapacity(pInfo->binfo.pRes, pOperator->resultInfo.capacity);
   doBuildDeleteDataBlock(pInfo->pStDeleted, pInfo->pDelRes, &pInfo->pDelIterator);
   if (pInfo->pDelRes->info.rows > 0) {
-    /*printDataBlock(pInfo->pDelRes, "session del");*/
+    printDataBlock(pInfo->pDelRes, IS_FINAL_OP(pInfo)? "Final  Session" : "Single Session");
     return pInfo->pDelRes;
   }
   doBuildResultDatablock(pOperator, &pInfo->binfo, &pInfo->groupResInfo, pInfo->streamAggSup.pResultBuf);
-  /*printDataBlock(pBInfo->pRes, "session insert");*/
+  printDataBlock(pBInfo->pRes, IS_FINAL_OP(pInfo)? "Final  Session" : "Single Session");
   return pBInfo->pRes->info.rows == 0 ? NULL : pBInfo->pRes;
 }
 
@@ -3169,11 +3399,11 @@ static SSDataBlock* doStreamSessionSemiAgg(SOperatorInfo* pOperator) {
   while (1) {
     SSDataBlock* pBlock = downstream->fpSet.getNextFn(downstream);
     if (pBlock == NULL) {
-      clearUpdateDataBlock(pInfo->pUpdateRes);
+      clearSpecialDataBlock(pInfo->pUpdateRes);
       break;
     }
 
-    if (pBlock->info.type == STREAM_REPROCESS) {
+    if (pBlock->info.type == STREAM_CLEAR) {
       SArray* pWins = taosArrayInit(16, sizeof(SResultWindowInfo));
       doClearSessionWindows(&pInfo->streamAggSup, pSup, pBlock, 0, pSup->numOfExprs, pInfo->gap, pWins);
       removeSessionResults(pStUpdated, pWins);
@@ -3236,7 +3466,7 @@ SOperatorInfo* createStreamFinalSessionAggOperatorInfo(SOperatorInfo* downstream
   } else {
     pInfo->isFinal = false;
     pInfo->pUpdateRes = createResDataBlock(pPhyNode->pOutputDataBlockDesc);
-    pInfo->pUpdateRes->info.type = STREAM_REPROCESS;
+    pInfo->pUpdateRes->info.type = STREAM_CLEAR;
     blockDataEnsureCapacity(pInfo->pUpdateRes, 128);
     pOperator->name = "StreamSessionSemiAggOperator";
     pOperator->fpSet =
@@ -3554,7 +3784,7 @@ static SSDataBlock* doStreamStateAgg(SOperatorInfo* pOperator) {
       break;
     }
 
-    if (pBlock->info.type == STREAM_REPROCESS) {
+    if (pBlock->info.type == STREAM_CLEAR) {
       doClearStateWindows(&pInfo->streamAggSup, pBlock, pInfo->primaryTsIndex, &pInfo->stateCol, pInfo->stateCol.slotId,
                           pSeUpdated, pInfo->pSeDeleted);
       continue;
@@ -3900,18 +4130,22 @@ _error:
 // merge interval operator
 typedef struct SMergeIntervalAggOperatorInfo {
   SIntervalAggOperatorInfo intervalAggOperatorInfo;
-
+  SList*       groupIntervals;
-  SHashObj*    groupIntervalHash;
+  SListIter   groupIntervalsIter;
-  void*        groupIntervalIter;
   bool         hasGroupId;
   uint64_t     groupId;
   SSDataBlock* prefetchedBlock;
   bool         inputBlocksFinished;
 } SMergeIntervalAggOperatorInfo;
 
+typedef struct SGroupTimeWindow {
+  uint64_t groupId;
+  STimeWindow window;
+} SGroupTimeWindow;
+
 void destroyMergeIntervalOperatorInfo(void* param, int32_t numOfOutput) {
   SMergeIntervalAggOperatorInfo* miaInfo = (SMergeIntervalAggOperatorInfo*)param;
-  taosHashCleanup(miaInfo->groupIntervalHash);
+  tdListFree(miaInfo->groupIntervals);
   destroyIntervalOperatorInfo(&miaInfo->intervalAggOperatorInfo, numOfOutput);
 }
 
@@ -3940,15 +4174,22 @@ static int32_t outputPrevIntervalResult(SOperatorInfo* pOperatorInfo, uint64_t t
   bool                           ascScan = (iaInfo->order == TSDB_ORDER_ASC);
   SExprSupp*                     pExprSup = &pOperatorInfo->exprSupp;
 
-  STimeWindow* prevWin = taosHashGet(miaInfo->groupIntervalHash, &tableGroupId, sizeof(tableGroupId));
+  SGroupTimeWindow groupTimeWindow = {.groupId = tableGroupId, .window = *newWin};
-  if (prevWin == NULL) {
+  tdListAppend(miaInfo->groupIntervals, &groupTimeWindow);
-    taosHashPut(miaInfo->groupIntervalHash, &tableGroupId, sizeof(tableGroupId), newWin, sizeof(STimeWindow));
-    return 0;
-  }
 
-  if ((ascScan && newWin->skey > prevWin->skey || (!ascScan) && newWin->skey < prevWin->skey)) {
+  SListIter iter = {0};
+  tdListInitIter(miaInfo->groupIntervals, &iter, TD_LIST_FORWARD);
+  SListNode* listNode = NULL;
+  while ((listNode = tdListNext(&iter)) != NULL) {
+    SGroupTimeWindow* prevGrpWin = (SGroupTimeWindow*)listNode->data;
+    if (prevGrpWin->groupId != tableGroupId ) {
+      continue;
+    }
+    STimeWindow* prevWin = &prevGrpWin->window;
+    if ((ascScan && newWin->skey > prevWin->ekey || (!ascScan) && newWin->skey < prevWin->ekey)) {
       finalizeWindowResult(pOperatorInfo, tableGroupId, prevWin, pResultBlock);
-    taosHashPut(miaInfo->groupIntervalHash, &tableGroupId, sizeof(tableGroupId), newWin, sizeof(STimeWindow));
+      tdListPopNode(miaInfo->groupIntervals, listNode);
+    }
   }
 
   return 0;
@@ -4075,6 +4316,7 @@ static SSDataBlock* doMergeIntervalAgg(SOperatorInfo* pOperator) {
       }
 
       if (pBlock == NULL) {
+        tdListInitIter(miaInfo->groupIntervals, &miaInfo->groupIntervalsIter, TD_LIST_FORWARD);
         miaInfo->inputBlocksFinished = true;
         break;
       }
@@ -4100,14 +4342,12 @@ static SSDataBlock* doMergeIntervalAgg(SOperatorInfo* pOperator) {
   }
 
   if (miaInfo->inputBlocksFinished) {
-    void* win = taosHashIterate(miaInfo->groupIntervalHash, miaInfo->groupIntervalIter);
+    SListNode* listNode = tdListNext(&miaInfo->groupIntervalsIter);
-    if (win != NULL) {
-      miaInfo->groupIntervalIter = win;
 
-      size_t    len = 0;
+    if (listNode != NULL) {
-      uint64_t* pTableGroupId = taosHashGetKey(win, &len);
+      SGroupTimeWindow* grpWin = (SGroupTimeWindow*)(listNode->data);
-      finalizeWindowResult(pOperator, *pTableGroupId, win, pRes);
+      finalizeWindowResult(pOperator, grpWin->groupId, &grpWin->window, pRes);
-      pRes->info.groupId = *pTableGroupId;
+      pRes->info.groupId = grpWin->groupId;
     }
   }
 
@@ -4129,8 +4369,7 @@ SOperatorInfo* createMergeIntervalOperatorInfo(SOperatorInfo* downstream, SExprI
     goto _error;
   }
 
-  miaInfo->groupIntervalHash = taosHashInit(128, taosGetDefaultHashFunction(TSDB_DATA_TYPE_UBIGINT), true, HASH_NO_LOCK);
+  miaInfo->groupIntervals = tdListNew(sizeof(SGroupTimeWindow));
-  miaInfo->groupIntervalIter = NULL;
 
   SIntervalAggOperatorInfo* iaInfo = &miaInfo->intervalAggOperatorInfo;
 

source/libs/function/src/builtins.c

@@ -1220,18 +1220,19 @@ static int32_t translateSubstr(SFunctionNode* pFunc, char* pErrBuf, int32_t len)
 
 static int32_t translateCast(SFunctionNode* pFunc, char* pErrBuf, int32_t len) {
   // The number of parameters has been limited by the syntax definition
-  uint8_t para1Type = ((SExprNode*)nodesListGetNode(pFunc->pParameterList, 0))->resType.type;
+  //uint8_t para1Type = ((SExprNode*)nodesListGetNode(pFunc->pParameterList, 0))->resType.type;
+
   // The function return type has been set during syntax parsing
   uint8_t para2Type = pFunc->node.resType.type;
-  if (para2Type != TSDB_DATA_TYPE_BIGINT && para2Type != TSDB_DATA_TYPE_UBIGINT &&
+  //if (para2Type != TSDB_DATA_TYPE_BIGINT && para2Type != TSDB_DATA_TYPE_UBIGINT &&
-      para2Type != TSDB_DATA_TYPE_VARCHAR && para2Type != TSDB_DATA_TYPE_NCHAR &&
+  //    para2Type != TSDB_DATA_TYPE_VARCHAR && para2Type != TSDB_DATA_TYPE_NCHAR &&
-      para2Type != TSDB_DATA_TYPE_TIMESTAMP) {
+  //    para2Type != TSDB_DATA_TYPE_TIMESTAMP) {
-    return invaildFuncParaTypeErrMsg(pErrBuf, len, pFunc->functionName);
+  //  return invaildFuncParaTypeErrMsg(pErrBuf, len, pFunc->functionName);
-  }
+  //}
-  if ((para2Type == TSDB_DATA_TYPE_TIMESTAMP && IS_VAR_DATA_TYPE(para1Type)) ||
+  //if ((para2Type == TSDB_DATA_TYPE_TIMESTAMP && IS_VAR_DATA_TYPE(para1Type)) ||
-      (para2Type == TSDB_DATA_TYPE_BINARY && para1Type == TSDB_DATA_TYPE_NCHAR)) {
+  //    (para2Type == TSDB_DATA_TYPE_BINARY && para1Type == TSDB_DATA_TYPE_NCHAR)) {
-    return invaildFuncParaTypeErrMsg(pErrBuf, len, pFunc->functionName);
+  //  return invaildFuncParaTypeErrMsg(pErrBuf, len, pFunc->functionName);
-  }
+  //}
 
   int32_t para2Bytes = pFunc->node.resType.bytes;
   if (IS_VAR_DATA_TYPE(para2Type)) {

source/libs/function/src/udfd.c

@@ -12,6 +12,8 @@
  * You should have received a copy of the GNU Affero General Public License
  * along with this program. If not, see <http://www.gnu.org/licenses/>.
  */
+
+// clang-format off
 #include "uv.h"
 #include "os.h"
 #include "fnLog.h"
@@ -25,6 +27,7 @@
 #include "tglobal.h"
 #include "tmsg.h"
 #include "trpc.h"
+// clang-foramt on
 
 typedef struct SUdfdContext {
   uv_loop_t * loop;
@@ -138,7 +141,7 @@ static void udfdCtrlReadCb(uv_stream_t *q, ssize_t nread, const uv_buf_t *buf);
 static int32_t udfdUvInit();
 static void    udfdCloseWalkCb(uv_handle_t *handle, void *arg);
 static int32_t udfdRun();
-static void udfdConnectMnodeThreadFunc(void* args);
+static void    udfdConnectMnodeThreadFunc(void *args);
 
 void udfdProcessRequest(uv_work_t *req) {
   SUvUdfWork *uvUdf = (SUvUdfWork *)(req->data);
@@ -401,11 +404,11 @@ void udfdProcessRpcRsp(void *parent, SRpcMsg *pMsg, SEpSet *pEpSet) {
     udf->bufSize = pFuncInfo->bufSize;
 
     char path[PATH_MAX] = {0};
-  #ifdef WINDOWS
+#ifdef WINDOWS
     snprintf(path, sizeof(path), "%s%s.dll", TD_TMP_DIR_PATH, pFuncInfo->name);
-  #else
+#else
     snprintf(path, sizeof(path), "%s/lib%s.so", TD_TMP_DIR_PATH, pFuncInfo->name);
-  #endif
+#endif
     TdFilePtr file =
         taosOpenFile(path, TD_FILE_CREATE | TD_FILE_WRITE | TD_FILE_READ | TD_FILE_TRUNC | TD_FILE_AUTO_DEL);
     if (file == NULL) {
@@ -544,7 +547,8 @@ int32_t udfdLoadUdf(char *udfName, SUdf *udf) {
   return 0;
 }
 static bool udfdRpcRfp(int32_t code) {
-  if (code == TSDB_CODE_RPC_REDIRECT) {
+  if (code == TSDB_CODE_RPC_REDIRECT || code == TSDB_CODE_RPC_NETWORK_UNAVAIL || code == TSDB_CODE_NODE_NOT_DEPLOYED ||
+      code == TSDB_CODE_SYN_NOT_LEADER || code == TSDB_CODE_APP_NOT_READY) {
     return true;
   } else {
     return false;
@@ -652,8 +656,7 @@ void udfdAllocBuffer(uv_handle_t *handle, size_t suggestedSize, uv_buf_t *buf) {
       buf->base = ctx->inputBuf;
       buf->len = ctx->inputCap;
     } else {
-      fnError("udfd can not allocate enough memory")
+      fnError("udfd can not allocate enough memory") buf->base = NULL;
-      buf->base = NULL;
       buf->len = 0;
     }
   } else {
@@ -664,8 +667,7 @@ void udfdAllocBuffer(uv_handle_t *handle, size_t suggestedSize, uv_buf_t *buf) {
       buf->base = ctx->inputBuf + ctx->inputLen;
       buf->len = ctx->inputCap - ctx->inputLen;
     } else {
-      fnError("udfd can not allocate enough memory")
+      fnError("udfd can not allocate enough memory") buf->base = NULL;
-      buf->base = NULL;
       buf->len = 0;
     }
   }
@@ -881,7 +883,7 @@ static int32_t udfdRun() {
   return 0;
 }
 
-void udfdConnectMnodeThreadFunc(void* args) {
+void udfdConnectMnodeThreadFunc(void *args) {
   int32_t retryMnodeTimes = 0;
   int32_t code = 0;
   while (retryMnodeTimes++ <= TSDB_MAX_REPLICA) {

source/libs/parser/src/parTranslater.c

@@ -5044,6 +5044,7 @@ typedef struct SVgroupCreateTableBatch {
 
 static void destroyCreateTbReq(SVCreateTbReq* pReq) {
   taosMemoryFreeClear(pReq->name);
+  taosMemoryFreeClear(pReq->comment);
   taosMemoryFreeClear(pReq->ntb.schemaRow.pSchema);
 }
 
@@ -5061,6 +5062,7 @@ static int32_t buildNormalTableBatchReq(int32_t acctId, const SCreateTableStmt*
   if (pStmt->pOptions->commentNull == false) {
     req.comment = strdup(pStmt->pOptions->comment);
     if (NULL == req.comment) {
+      destroyCreateTbReq(&req);
       return TSDB_CODE_OUT_OF_MEMORY;
     }
     req.commentLen = strlen(pStmt->pOptions->comment);
@@ -5132,6 +5134,7 @@ static void destroyCreateTbReqBatch(SVgroupCreateTableBatch* pTbBatch) {
   for (int32_t i = 0; i < size; ++i) {
     SVCreateTbReq* pTableReq = taosArrayGet(pTbBatch->req.pArray, i);
     taosMemoryFreeClear(pTableReq->name);
+    taosMemoryFreeClear(pTableReq->comment);
 
     if (pTableReq->type == TSDB_NORMAL_TABLE) {
       taosMemoryFreeClear(pTableReq->ntb.schemaRow.pSchema);

source/libs/planner/src/planOptimizer.c

@@ -1107,6 +1107,26 @@ static int32_t partTagsOptimize(SOptimizeContext* pCxt, SLogicSubplan* pLogicSub
   return code;
 }
 
+//=====================================================================================================================
+// eliminate project optimization
+static bool eliminateProjOptCheckProjColumnNames(SProjectLogicNode* pProjectNode) {
+  SHashObj* pProjColNameHash = taosHashInit(16, taosGetDefaultHashFunction(TSDB_DATA_TYPE_BINARY), true, HASH_NO_LOCK);
+  SNode*    pProjection;
+  FOREACH(pProjection, pProjectNode->pProjections) {
+    char*    projColumnName = ((SColumnNode*)pProjection)->colName;
+    int32_t* pExist = taosHashGet(pProjColNameHash, projColumnName, strlen(projColumnName));
+    if (NULL != pExist) {
+      taosHashCleanup(pProjColNameHash);
+      return false;
+    } else {
+      int32_t exist = 1;
+      taosHashPut(pProjColNameHash, projColumnName, strlen(projColumnName), &exist, sizeof(exist));
+    }
+  }
+  taosHashCleanup(pProjColNameHash);
+  return true;
+}
+
 static bool eliminateProjOptMayBeOptimized(SLogicNode* pNode) {
   // TODO: enable this optimization after new mechanising that map projection and targets of project node
   if (NULL != pNode->pParent) {
@@ -1122,28 +1142,15 @@ static bool eliminateProjOptMayBeOptimized(SLogicNode* pNode) {
     return false;
   }
 
-  SHashObj* pProjColNameHash = taosHashInit(16, taosGetDefaultHashFunction(TSDB_DATA_TYPE_BINARY), true, HASH_NO_LOCK);
   SNode* pProjection;
   FOREACH(pProjection, pProjectNode->pProjections) {
     SExprNode* pExprNode = (SExprNode*)pProjection;
     if (QUERY_NODE_COLUMN != nodeType(pExprNode)) {
-      taosHashCleanup(pProjColNameHash);
       return false;
     }
-
-    char*    projColumnName = ((SColumnNode*)pProjection)->colName;
-    int32_t* pExist = taosHashGet(pProjColNameHash, projColumnName, strlen(projColumnName));
-    if (NULL != pExist) {
-      taosHashCleanup(pProjColNameHash);
-      return false;
-    } else {
-      int32_t exist = 1;
-      taosHashPut(pProjColNameHash, projColumnName, strlen(projColumnName), &exist, sizeof(exist));
   }
-  }
-  taosHashCleanup(pProjColNameHash);
 
-  return true;
+  return eliminateProjOptCheckProjColumnNames(pProjectNode);
 }
 
 static int32_t eliminateProjOptimizeImpl(SOptimizeContext* pCxt, SLogicSubplan* pLogicSubplan,
@@ -1320,6 +1327,46 @@ static int32_t rewriteTailOptimize(SOptimizeContext* pCxt, SLogicSubplan* pLogic
   return rewriteTailOptimizeImpl(pCxt, pLogicSubplan, pIndef);
 }
 
+static bool eliminateSetOpMayBeOptimized(SLogicNode* pNode) {
+  SLogicNode* pParent = pNode->pParent;
+  if (NULL == pParent ||
+      QUERY_NODE_LOGIC_PLAN_AGG != nodeType(pParent) && QUERY_NODE_LOGIC_PLAN_PROJECT != nodeType(pParent) ||
+      LIST_LENGTH(pParent->pChildren) < 2) {
+    return false;
+  }
+  if (nodeType(pNode) != nodeType(pNode->pParent) || LIST_LENGTH(pNode->pChildren) < 2) {
+    return false;
+  }
+  return true;
+}
+
+static int32_t eliminateSetOpOptimizeImpl(SOptimizeContext* pCxt, SLogicSubplan* pLogicSubplan,
+                                          SLogicNode* pSetOpNode) {
+  SNode* pSibling;
+  FOREACH(pSibling, pSetOpNode->pParent->pChildren) {
+    if (nodesEqualNode(pSibling, (SNode*)pSetOpNode)) {
+      SNode* pChild;
+      FOREACH(pChild, pSetOpNode->pChildren) { ((SLogicNode*)pChild)->pParent = pSetOpNode->pParent; }
+      INSERT_LIST(pSetOpNode->pParent->pChildren, pSetOpNode->pChildren);
+
+      pSetOpNode->pChildren = NULL;
+      ERASE_NODE(pSetOpNode->pParent->pChildren);
+      return TSDB_CODE_SUCCESS;
+    }
+  }
+
+  return TSDB_CODE_PLAN_INTERNAL_ERROR;
+}
+
+static int32_t eliminateSetOpOptimize(SOptimizeContext* pCxt, SLogicSubplan* pLogicSubplan) {
+  SLogicNode* pSetOpNode = optFindPossibleNode(pLogicSubplan->pNode, eliminateSetOpMayBeOptimized);
+  if (NULL == pSetOpNode) {
+    return TSDB_CODE_SUCCESS;
+  }
+
+  return eliminateSetOpOptimizeImpl(pCxt, pLogicSubplan, pSetOpNode);
+}
+
 // clang-format off
 static const SOptimizeRule optimizeRuleSet[] = {
   {.pName = "OptimizeScanData",           .optimizeFunc = osdOptimize},
@@ -1328,6 +1375,7 @@ static const SOptimizeRule optimizeRuleSet[] = {
   {.pName = "SmaIndex",                   .optimizeFunc = smaOptimize},
   {.pName = "PartitionTags",              .optimizeFunc = partTagsOptimize},
   {.pName = "EliminateProject",           .optimizeFunc = eliminateProjOptimize},
+  {.pName = "EliminateSetOperator",       .optimizeFunc = eliminateSetOpOptimize},
   {.pName = "RewriteTail",                .optimizeFunc = rewriteTailOptimize}
 };
 // clang-format on

source/libs/qcom/src/queryUtil.c

@@ -19,6 +19,7 @@
 #include "tmsg.h"
 #include "trpc.h"
 #include "tsched.h"
+// clang-format off
 #include "cJSON.h"
 
 #define VALIDNUMOFCOLS(x) ((x) >= TSDB_MIN_COLUMNS && (x) <= TSDB_MAX_COLUMNS)
@@ -147,13 +148,15 @@ int32_t asyncSendMsgToServerExt(void* pTransporter, SEpSet* epSet, int64_t* pTra
   }
 
   memcpy(pMsg, pInfo->msgInfo.pData, pInfo->msgInfo.len);
-  SRpcMsg rpcMsg = {.msgType = pInfo->msgType,
+  SRpcMsg rpcMsg = {
+    .msgType = pInfo->msgType,
     .pCont = pMsg,
     .contLen = pInfo->msgInfo.len,
     .info.ahandle = (void*)pInfo,
     .info.handle = pInfo->msgInfo.handle,
     .info.persistHandle = persistHandle, 
-                    .code = 0};
+    .code = 0
+  };
   assert(pInfo->fp != NULL);
   TRACE_SET_ROOTID(&rpcMsg.info.traceId, pInfo->requestId);
   rpcSendRequestWithCtx(pTransporter, epSet, &rpcMsg, pTransporterId, rpcCtx);
@@ -221,8 +224,9 @@ void destroyQueryExecRes(SQueryExecRes* pRes) {
       qError("invalid exec result for request type %d", pRes->msgType);
   }
 }
+// clang-format on
 
-int32_t dataConverToStr(char *str, int type, void *buf, int32_t bufSize, int32_t *len) {
+int32_t dataConverToStr(char* str, int type, void* buf, int32_t bufSize, int32_t* len) {
   int32_t n = 0;
 
   switch (type) {
@@ -262,7 +266,7 @@ int32_t dataConverToStr(char *str, int type, void *buf, int32_t bufSize, int32_t
     case TSDB_DATA_TYPE_BINARY:
     case TSDB_DATA_TYPE_NCHAR:
       if (bufSize < 0) {
-//        tscError("invalid buf size");
+        //        tscError("invalid buf size");
         return TSDB_CODE_TSC_INVALID_VALUE;
       }
 
@@ -289,7 +293,7 @@ int32_t dataConverToStr(char *str, int type, void *buf, int32_t bufSize, int32_t
       break;
 
     default:
-//      tscError("unsupported type:%d", type);
+      //      tscError("unsupported type:%d", type);
       return TSDB_CODE_TSC_INVALID_VALUE;
   }
 
@@ -372,7 +376,6 @@ end:
   return string;
 }
 
-
 int32_t cloneTableMeta(STableMeta* pSrc, STableMeta** pDst) {
   if (NULL == pSrc) {
     *pDst = NULL;
@@ -400,13 +403,14 @@ int32_t cloneDbVgInfo(SDBVgInfo* pSrc, SDBVgInfo** pDst) {
   }
   memcpy(*pDst, pSrc, sizeof(*pSrc));
   if (pSrc->vgHash) {
-    (*pDst)->vgHash = taosHashInit(taosHashGetSize(pSrc->vgHash), taosGetDefaultHashFunction(TSDB_DATA_TYPE_INT), true, HASH_ENTRY_LOCK);
+    (*pDst)->vgHash = taosHashInit(taosHashGetSize(pSrc->vgHash), taosGetDefaultHashFunction(TSDB_DATA_TYPE_INT), true,
+                                   HASH_ENTRY_LOCK);
     if (NULL == (*pDst)->vgHash) {
       return TSDB_CODE_TSC_OUT_OF_MEMORY;
     }
 
     SVgroupInfo* vgInfo = NULL;
-    void *pIter = taosHashIterate(pSrc->vgHash, NULL);
+    void*        pIter = taosHashIterate(pSrc->vgHash, NULL);
     while (pIter) {
       vgInfo = pIter;
       int32_t* vgId = taosHashGetKey(pIter, NULL);
@@ -425,5 +429,3 @@ int32_t cloneDbVgInfo(SDBVgInfo* pSrc, SDBVgInfo** pDst) {
 
   return TSDB_CODE_SUCCESS;
 }
-
-

source/libs/scalar/src/sclfunc.c

@@ -735,6 +735,60 @@ int32_t castFunction(SScalarParam *pInput, int32_t inputNum, SScalarParam *pOutp
     char *input = colDataGetData(pInput[0].columnData, i);
 
     switch(outputType) {
+      case TSDB_DATA_TYPE_TINYINT: {
+        if (inputType == TSDB_DATA_TYPE_BINARY) {
+          *(int8_t *)output = taosStr2Int8(varDataVal(input), NULL, 10);
+        } else if (inputType == TSDB_DATA_TYPE_NCHAR) {
+          char *newBuf = taosMemoryCalloc(1, inputLen);
+          int32_t len  = taosUcs4ToMbs((TdUcs4 *)varDataVal(input), varDataLen(input), newBuf);
+          if (len < 0) {
+            taosMemoryFree(newBuf);
+            return TSDB_CODE_FAILED;
+          }
+          newBuf[len] = 0;
+          *(int8_t *)output = taosStr2Int8(newBuf, NULL, 10);
+          taosMemoryFree(newBuf);
+        } else {
+          GET_TYPED_DATA(*(int8_t *)output, int8_t, inputType, input);
+        }
+        break;
+      }
+      case TSDB_DATA_TYPE_SMALLINT: {
+        if (inputType == TSDB_DATA_TYPE_BINARY) {
+          *(int16_t *)output = taosStr2Int16(varDataVal(input), NULL, 10);
+        } else if (inputType == TSDB_DATA_TYPE_NCHAR) {
+          char *newBuf = taosMemoryCalloc(1, inputLen);
+          int32_t len  = taosUcs4ToMbs((TdUcs4 *)varDataVal(input), varDataLen(input), newBuf);
+          if (len < 0) {
+            taosMemoryFree(newBuf);
+            return TSDB_CODE_FAILED;
+          }
+          newBuf[len] = 0;
+          *(int16_t *)output = taosStr2Int16(newBuf, NULL, 10);
+          taosMemoryFree(newBuf);
+        } else {
+          GET_TYPED_DATA(*(int16_t *)output, int16_t, inputType, input);
+        }
+        break;
+      }
+      case TSDB_DATA_TYPE_INT: {
+        if (inputType == TSDB_DATA_TYPE_BINARY) {
+          *(int32_t *)output = taosStr2Int32(varDataVal(input), NULL, 10);
+        } else if (inputType == TSDB_DATA_TYPE_NCHAR) {
+          char *newBuf = taosMemoryCalloc(1, inputLen);
+          int32_t len  = taosUcs4ToMbs((TdUcs4 *)varDataVal(input), varDataLen(input), newBuf);
+          if (len < 0) {
+            taosMemoryFree(newBuf);
+            return TSDB_CODE_FAILED;
+          }
+          newBuf[len] = 0;
+          *(int32_t *)output = taosStr2Int32(newBuf, NULL, 10);
+          taosMemoryFree(newBuf);
+        } else {
+          GET_TYPED_DATA(*(int32_t *)output, int32_t, inputType, input);
+        }
+        break;
+      }
       case TSDB_DATA_TYPE_BIGINT: {
         if (inputType == TSDB_DATA_TYPE_BINARY) {
           *(int64_t *)output = taosStr2Int64(varDataVal(input), NULL, 10);
@@ -753,6 +807,60 @@ int32_t castFunction(SScalarParam *pInput, int32_t inputNum, SScalarParam *pOutp
         }
         break;
       }
+      case TSDB_DATA_TYPE_UTINYINT: {
+        if (inputType == TSDB_DATA_TYPE_BINARY) {
+          *(uint8_t *)output = taosStr2UInt8(varDataVal(input), NULL, 10);
+        } else if (inputType == TSDB_DATA_TYPE_NCHAR) {
+          char *newBuf = taosMemoryCalloc(1, inputLen);
+          int32_t len = taosUcs4ToMbs((TdUcs4 *)varDataVal(input), varDataLen(input), newBuf);
+          if (len < 0) {
+            taosMemoryFree(newBuf);
+            return TSDB_CODE_FAILED;
+          }
+          newBuf[len] = 0;
+          *(uint8_t *)output = taosStr2UInt8(newBuf, NULL, 10);
+          taosMemoryFree(newBuf);
+        } else {
+          GET_TYPED_DATA(*(uint8_t *)output, uint8_t, inputType, input);
+        }
+        break;
+      }
+      case TSDB_DATA_TYPE_USMALLINT: {
+        if (inputType == TSDB_DATA_TYPE_BINARY) {
+          *(uint16_t *)output = taosStr2UInt16(varDataVal(input), NULL, 10);
+        } else if (inputType == TSDB_DATA_TYPE_NCHAR) {
+          char *newBuf = taosMemoryCalloc(1, inputLen);
+          int32_t len = taosUcs4ToMbs((TdUcs4 *)varDataVal(input), varDataLen(input), newBuf);
+          if (len < 0) {
+            taosMemoryFree(newBuf);
+            return TSDB_CODE_FAILED;
+          }
+          newBuf[len] = 0;
+          *(uint16_t *)output = taosStr2UInt16(newBuf, NULL, 10);
+          taosMemoryFree(newBuf);
+        } else {
+          GET_TYPED_DATA(*(uint16_t *)output, uint16_t, inputType, input);
+        }
+        break;
+      }
+      case TSDB_DATA_TYPE_UINT: {
+        if (inputType == TSDB_DATA_TYPE_BINARY) {
+          *(uint32_t *)output = taosStr2UInt32(varDataVal(input), NULL, 10);
+        } else if (inputType == TSDB_DATA_TYPE_NCHAR) {
+          char *newBuf = taosMemoryCalloc(1, inputLen);
+          int32_t len = taosUcs4ToMbs((TdUcs4 *)varDataVal(input), varDataLen(input), newBuf);
+          if (len < 0) {
+            taosMemoryFree(newBuf);
+            return TSDB_CODE_FAILED;
+          }
+          newBuf[len] = 0;
+          *(uint32_t *)output = taosStr2UInt32(newBuf, NULL, 10);
+          taosMemoryFree(newBuf);
+        } else {
+          GET_TYPED_DATA(*(uint32_t *)output, uint32_t, inputType, input);
+        }
+        break;
+      }
       case TSDB_DATA_TYPE_UBIGINT: {
         if (inputType == TSDB_DATA_TYPE_BINARY) {
           *(uint64_t *)output = taosStr2UInt64(varDataVal(input), NULL, 10);
@@ -771,10 +879,64 @@ int32_t castFunction(SScalarParam *pInput, int32_t inputNum, SScalarParam *pOutp
         }
         break;
       }
+      case TSDB_DATA_TYPE_FLOAT: {
+        if (inputType == TSDB_DATA_TYPE_BINARY) {
+          *(float *)output = taosStr2Float(varDataVal(input), NULL);
+        } else if (inputType == TSDB_DATA_TYPE_NCHAR) {
+          char *newBuf = taosMemoryCalloc(1, inputLen);
+          int32_t len  = taosUcs4ToMbs((TdUcs4 *)varDataVal(input), varDataLen(input), newBuf);
+          if (len < 0) {
+            taosMemoryFree(newBuf);
+            return TSDB_CODE_FAILED;
+          }
+          newBuf[len] = 0;
+          *(float *)output = taosStr2Float(newBuf, NULL);
+          taosMemoryFree(newBuf);
+        } else {
+          GET_TYPED_DATA(*(float *)output, float, inputType, input);
+        }
+        break;
+      }
+      case TSDB_DATA_TYPE_DOUBLE: {
+        if (inputType == TSDB_DATA_TYPE_BINARY) {
+          *(double *)output = taosStr2Double(varDataVal(input), NULL);
+        } else if (inputType == TSDB_DATA_TYPE_NCHAR) {
+          char *newBuf = taosMemoryCalloc(1, inputLen);
+          int32_t len  = taosUcs4ToMbs((TdUcs4 *)varDataVal(input), varDataLen(input), newBuf);
+          if (len < 0) {
+            taosMemoryFree(newBuf);
+            return TSDB_CODE_FAILED;
+          }
+          newBuf[len] = 0;
+          *(double *)output = taosStr2Double(newBuf, NULL);
+          taosMemoryFree(newBuf);
+        } else {
+          GET_TYPED_DATA(*(double *)output, double, inputType, input);
+        }
+        break;
+      }
+      case TSDB_DATA_TYPE_BOOL: {
+        if (inputType == TSDB_DATA_TYPE_BINARY) {
+          *(bool *)output = taosStr2Int8(varDataVal(input), NULL, 10);
+        } else if (inputType == TSDB_DATA_TYPE_NCHAR) {
+          char *newBuf = taosMemoryCalloc(1, inputLen);
+          int32_t len  = taosUcs4ToMbs((TdUcs4 *)varDataVal(input), varDataLen(input), newBuf);
+          if (len < 0) {
+            taosMemoryFree(newBuf);
+            return TSDB_CODE_FAILED;
+          }
+          newBuf[len] = 0;
+          *(bool *)output = taosStr2Int8(newBuf, NULL, 10);
+          taosMemoryFree(newBuf);
+        } else {
+          GET_TYPED_DATA(*(bool *)output, bool, inputType, input);
+        }
+        break;
+      }
       case TSDB_DATA_TYPE_TIMESTAMP: {
         if (inputType == TSDB_DATA_TYPE_BINARY || inputType == TSDB_DATA_TYPE_NCHAR) {
-          //not support
+          //convert to 0
-          return TSDB_CODE_FAILED;
+          *(int64_t *)output = 0;
         } else {
           GET_TYPED_DATA(*(int64_t *)output, int64_t, inputType, input);
         }
@@ -789,8 +951,16 @@ int32_t castFunction(SScalarParam *pInput, int32_t inputNum, SScalarParam *pOutp
           len = sprintf(varDataVal(output), "%.*s", len, varDataVal(input));
           varDataSetLen(output, len);
         } else if (inputType == TSDB_DATA_TYPE_NCHAR) {
-          //not support
+          char *newBuf = taosMemoryCalloc(1, inputLen);
+          int32_t len = taosUcs4ToMbs((TdUcs4 *)varDataVal(input), varDataLen(input), newBuf);
+          if (len < 0) {
+            taosMemoryFree(newBuf);
             return TSDB_CODE_FAILED;
+          }
+          len = TMIN(len, outputLen - VARSTR_HEADER_SIZE);
+          memcpy(varDataVal(output), newBuf, len);
+          varDataSetLen(output, len);
+          taosMemoryFree(newBuf);
         } else {
           char tmp[400] = {0};
           NUM_TO_STRING(inputType, input, sizeof(tmp), tmp);

source/libs/scalar/src/sclvector.c

@@ -468,7 +468,7 @@ double getVectorDoubleValue_JSON(void *src, int32_t index){
 }
 
 void* ncharTobinary(void *buf){            // todo need to remove , if tobinary is nchar
-  int32_t inputLen = varDataLen(buf);
+  int32_t inputLen = varDataTLen(buf);
 
   void* t = taosMemoryCalloc(1, inputLen);
   int32_t len  = taosUcs4ToMbs((TdUcs4 *)varDataVal(buf), varDataLen(buf), varDataVal(t));

source/libs/stream/src/stream.c

@@ -111,7 +111,7 @@ int32_t streamTaskEnqueue(SStreamTask* pTask, SStreamDispatchReq* pReq, SRpcMsg*
 
   // enqueue
   if (pData != NULL) {
-    pData->type = STREAM_DATA_TYPE_SSDATA_BLOCK;
+    pData->type = STREAM_INPUT__DATA_BLOCK;
     pData->srcVgId = pReq->dataSrcVgId;
     // decode
     /*pData->blocks = pReq->data;*/
@@ -146,7 +146,7 @@ int32_t streamTaskEnqueueRetrieve(SStreamTask* pTask, SStreamRetrieveReq* pReq,
 
   // enqueue
   if (pData != NULL) {
-    pData->type = STREAM_DATA_TYPE_SSDATA_BLOCK;
+    pData->type = STREAM_INPUT__DATA_RETRIEVE;
     pData->srcVgId = 0;
     // decode
     /*pData->blocks = pReq->data;*/
@@ -170,7 +170,7 @@ int32_t streamTaskEnqueueRetrieve(SStreamTask* pTask, SStreamRetrieveReq* pReq,
   pCont->rspToTaskId = pReq->srcTaskId;
   pCont->rspFromTaskId = pReq->dstTaskId;
   pRsp->pCont = buf;
-  pRsp->contLen = sizeof(SMsgHead) + sizeof(SStreamDispatchRsp);
+  pRsp->contLen = sizeof(SMsgHead) + sizeof(SStreamRetrieveRsp);
   tmsgSendRsp(pRsp);
   return status == TASK_INPUT_STATUS__NORMAL ? 0 : -1;
 }

source/libs/stream/src/streamDispatch.c

@@ -115,6 +115,7 @@ int32_t streamBroadcastToChildren(SStreamTask* pTask, const SSDataBlock* pBlock)
       .srcNodeId = pTask->nodeId,
       .srcTaskId = pTask->taskId,
       .pRetrieve = pRetrieve,
+      .retrieveLen = dataStrLen,
   };
 
   int32_t sz = taosArrayGetSize(pTask->childEpInfo);
@@ -146,7 +147,7 @@ int32_t streamBroadcastToChildren(SStreamTask* pTask, const SSDataBlock* pBlock)
         .code = 0,
         .msgType = TDMT_STREAM_RETRIEVE,
         .pCont = buf,
-        .contLen = len,
+        .contLen = sizeof(SMsgHead) + len,
     };
 
     if (tmsgSendReq(&pEpInfo->epSet, &rpcMsg) < 0) {
@@ -300,7 +301,7 @@ int32_t streamDispatch(SStreamTask* pTask, SMsgCb* pMsgCb) {
     atomic_store_8(&pTask->outputStatus, TASK_OUTPUT_STATUS__NORMAL);
     return 0;
   }
-  ASSERT(pBlock->type == STREAM_DATA_TYPE_SSDATA_BLOCK);
+  ASSERT(pBlock->type == STREAM_INPUT__DATA_BLOCK);
 
   qInfo("stream continue dispatching: task %d", pTask->taskId);
 

source/libs/stream/src/streamExec.c

@@ -17,6 +17,7 @@
 
 static int32_t streamTaskExecImpl(SStreamTask* pTask, void* data, SArray* pRes) {
   void* exec = pTask->exec.executor;
+  bool  hasData = false;
 
   // set input
   SStreamQueueItem* pItem = (SStreamQueueItem*)data;
@@ -27,7 +28,7 @@ static int32_t streamTaskExecImpl(SStreamTask* pTask, void* data, SArray* pRes)
     ASSERT(pTask->isDataScan);
     SStreamDataSubmit* pSubmit = (SStreamDataSubmit*)data;
     qSetStreamInput(exec, pSubmit->data, STREAM_DATA_TYPE_SUBMIT_BLOCK, false);
-  } else if (pItem->type == STREAM_INPUT__DATA_BLOCK) {
+  } else if (pItem->type == STREAM_INPUT__DATA_BLOCK || pItem->type == STREAM_INPUT__DATA_RETRIEVE) {
     SStreamDataBlock* pBlock = (SStreamDataBlock*)data;
     SArray*           blocks = pBlock->blocks;
     qSetMultiStreamInput(exec, blocks->pData, blocks->size, STREAM_DATA_TYPE_SSDATA_BLOCK, false);
@@ -43,7 +44,21 @@ static int32_t streamTaskExecImpl(SStreamTask* pTask, void* data, SArray* pRes)
     if (qExecTask(exec, &output, &ts) < 0) {
       ASSERT(false);
     }
-    if (output == NULL) break;
+    if (output == NULL) {
+      if (pItem->type == STREAM_INPUT__DATA_RETRIEVE) {
+        //SSDataBlock block = {0};
+        //block.info.type = STREAM_PUSH_EMPTY;
+        //block.info.childId = pTask->selfChildId;
+        SStreamDataBlock* pRetrieveBlock = (SStreamDataBlock*)data;
+        ASSERT(taosArrayGetSize(pRetrieveBlock->blocks) == 1);
+        SSDataBlock* pBlock = createOneDataBlock(taosArrayGet(pRetrieveBlock->blocks, 0), true);
+        pBlock->info.type = STREAM_PUSH_EMPTY;
+        pBlock->info.childId = pTask->selfChildId;
+        taosArrayPush(pRes, pBlock);
+      }
+      break;
+    }
+    hasData = true;
 
     if (output->info.type == STREAM_RETRIEVE) {
       if (streamBroadcastToChildren(pTask, output) < 0) {
@@ -99,7 +114,7 @@ static SArray* streamExecForQall(SStreamTask* pTask, SArray* pRes) {
     if (type == STREAM_INPUT__TRIGGER) {
       blockDataDestroy(((SStreamTrigger*)data)->pBlock);
       taosFreeQitem(data);
-    } else if (type == STREAM_INPUT__DATA_BLOCK) {
+    } else if (type == STREAM_INPUT__DATA_BLOCK || type == STREAM_INPUT__DATA_RETRIEVE) {
       taosArrayDestroyEx(((SStreamDataBlock*)data)->blocks, (FDelete)tDeleteSSDataBlock);
       taosFreeQitem(data);
     } else if (type == STREAM_INPUT__DATA_SUBMIT) {

source/libs/stream/src/streamUpdate.c

@@ -119,6 +119,7 @@ SUpdateInfo *updateInfoInit(int64_t interval, int32_t precision, int64_t waterma
     taosArrayPush(pInfo->pTsBuckets, &dumy);
   }
   pInfo->numBuckets = DEFAULT_BUCKET_SIZE;
+  pInfo->pCloseWinSBF = NULL;
   return pInfo;
 }
 
@@ -154,6 +155,9 @@ bool updateInfoIsUpdated(SUpdateInfo *pInfo, tb_uid_t tableId, TSKEY ts) {
   TSKEY maxTs = *(TSKEY *)taosArrayGet(pInfo->pTsBuckets, index);
   if (ts < maxTs - pInfo->watermark) {
     // this window has been closed.
+    if (pInfo->pCloseWinSBF) {
+      return tScalableBfPut(pInfo->pCloseWinSBF, &ts, sizeof(TSKEY));
+    }
     return true;
   }
 
@@ -193,3 +197,19 @@ void updateInfoDestroy(SUpdateInfo *pInfo) {
   taosArrayDestroy(pInfo->pTsSBFs);
   taosMemoryFree(pInfo);
 }
+
+void updateInfoAddCloseWindowSBF(SUpdateInfo *pInfo) {
+  if (pInfo->pCloseWinSBF) {
+    return;
+  }
+  int64_t rows = adjustExpEntries(pInfo->interval * ROWS_PER_MILLISECOND);
+  pInfo->pCloseWinSBF = tScalableBfInit(rows, DEFAULT_FALSE_POSITIVE);
+}
+
+void updateInfoDestoryColseWinSBF(SUpdateInfo *pInfo) {
+  if (!pInfo || !pInfo->pCloseWinSBF) {
+    return;
+  }
+  tScalableBfDestroy(pInfo->pCloseWinSBF);
+  pInfo->pCloseWinSBF = NULL;
+}

source/libs/sync/src/syncMain.c

@@ -914,6 +914,9 @@ void syncNodeStart(SSyncNode* pSyncNode) {
     syncNodeBecomeLeader(pSyncNode, "one replica start");
 
     // Raft 3.6.2 Committing entries from previous terms
+    syncNodeAppendNoop(pSyncNode);
+    syncMaybeAdvanceCommitIndex(pSyncNode);
+
     return;
   }
 
@@ -1662,6 +1665,12 @@ void syncNodeDoConfigChange(SSyncNode* pSyncNode, SSyncCfg* pNewConfig, SyncInde
     // change isStandBy to normal (election timeout)
     if (pSyncNode->state == TAOS_SYNC_STATE_LEADER) {
       syncNodeBecomeLeader(pSyncNode, tmpbuf);
+
+      // Raft 3.6.2 Committing entries from previous terms
+      syncNodeReplicate(pSyncNode);
+      syncNodeAppendNoop(pSyncNode);
+      syncMaybeAdvanceCommitIndex(pSyncNode);
+
     } else {
       syncNodeBecomeFollower(pSyncNode, tmpbuf);
     }
@@ -1807,16 +1816,9 @@ void syncNodeBecomeLeader(SSyncNode* pSyncNode, const char* debugStr) {
   // stop elect timer
   syncNodeStopElectTimer(pSyncNode);
 
-  // start replicate right now!
-  syncNodeReplicate(pSyncNode);
-
   // start heartbeat timer
   syncNodeStartHeartbeatTimer(pSyncNode);
 
-  // append noop
-  syncNodeAppendNoop(pSyncNode);
-  syncMaybeAdvanceCommitIndex(pSyncNode);  // maybe only one replica
-
   // trace log
   do {
     int32_t debugStrLen = strlen(debugStr);
@@ -1841,9 +1843,9 @@ void syncNodeCandidate2Leader(SSyncNode* pSyncNode) {
   syncNodeLog2("==state change syncNodeCandidate2Leader==", pSyncNode);
 
   // Raft 3.6.2 Committing entries from previous terms
-
+  syncNodeReplicate(pSyncNode);
-  // do not use this
+  syncNodeAppendNoop(pSyncNode);
-  // syncNodeEqNoop(pSyncNode);
+  syncMaybeAdvanceCommitIndex(pSyncNode);
 }
 
 void syncNodeFollower2Candidate(SSyncNode* pSyncNode) {

source/libs/transport/inc/transComm.h

@@ -105,6 +105,13 @@ typedef SRpcCtxVal   STransCtxVal;
 typedef SRpcInfo     STrans;
 typedef SRpcConnInfo STransHandleInfo;
 
+// ref mgt
+// handle
+typedef struct SExHandle {
+  void*   handle;
+  int64_t refId;
+  void*   pThrd;
+} SExHandle;
 /*convet from fqdn to ip */
 typedef struct SCvtAddr {
   char ip[TSDB_FQDN_LEN];
@@ -118,9 +125,10 @@ typedef struct {
   void*  ahandle;   // handle provided by app
   tmsg_t msgType;   // message type
   int8_t connType;  // connection type cli/srv
-  int64_t rid;       // refId returned by taosAddRef
 
-  int8_t     retryCount;
+  int8_t retryCnt;
+  int8_t retryLimit;
+  // bool       setMaxRetry;
   STransCtx  appCtx;  //
   STransMsg* pRsp;    // for synchronous API
   tsem_t*    pSem;    // for synchronous API
@@ -239,6 +247,32 @@ int         transSendAsync(SAsyncPool* pool, queue* mq);
       }                                                       \
     }                                                         \
   } while (0)
+
+#define ASYNC_CHECK_HANDLE(exh1, id)                                                                               \
+  do {                                                                                                             \
+    if (id > 0) {                                                                                                  \
+      tTrace("handle step1");                                                                                      \
+      SExHandle* exh2 = transAcquireExHandle(refMgt, id);                                                          \
+      if (exh2 == NULL || id != exh2->refId) {                                                                     \
+        tTrace("handle %p except, may already freed, ignore msg, ref1: %" PRIu64 ", ref2 : %" PRIu64 "", exh1,     \
+               exh2 ? exh2->refId : 0, id);                                                                        \
+        goto _return1;                                                                                             \
+      }                                                                                                            \
+    } else if (id == 0) {                                                                                          \
+      tTrace("handle step2");                                                                                      \
+      SExHandle* exh2 = transAcquireExHandle(refMgt, id);                                                          \
+      if (exh2 == NULL || id == exh2->refId) {                                                                     \
+        tTrace("handle %p except, may already freed, ignore msg, ref1: %" PRIu64 ", ref2 : %" PRIu64 "", exh1, id, \
+               exh2 ? exh2->refId : 0);                                                                            \
+        goto _return1;                                                                                             \
+      } else {                                                                                                     \
+        id = exh1->refId;                                                                                          \
+      }                                                                                                            \
+    } else if (id < 0) {                                                                                           \
+      tTrace("handle step3");                                                                                      \
+      goto _return2;                                                                                               \
+    }                                                                                                              \
+  } while (0)
 int  transInitBuffer(SConnBuffer* buf);
 int  transClearBuffer(SConnBuffer* buf);
 int  transDestroyBuffer(SConnBuffer* buf);
@@ -349,21 +383,13 @@ void transDQDestroy(SDelayQueue* queue);
 
 int transDQSched(SDelayQueue* queue, void (*func)(void* arg), void* arg, uint64_t timeoutMs);
 
-void transPrintEpSet(SEpSet* pEpSet);
+// void transPrintEpSet(SEpSet* pEpSet);
 bool transEpSetIsEqual(SEpSet* a, SEpSet* b);
 /*
  * init global func
  */
 void transThreadOnce();
 
-// ref mgt
-// handle
-typedef struct SExHandle {
-  void*   handle;
-  int64_t refId;
-  void*   pThrd;
-} SExHandle;
-
 void       transInitEnv();
 int32_t    transOpenExHandleMgt(int size);
 void       transCloseExHandleMgt(int32_t mgt);

source/libs/transport/src/trans.c

@@ -79,6 +79,7 @@ void* rpcOpen(const SRpcInit* pInit) {
   return pRpc;
 }
 void rpcClose(void* arg) {
+  tInfo("start to close rpc");
   SRpcInfo* pRpc = (SRpcInfo*)arg;
   (*taosCloseHandle[pRpc->connType])(pRpc->tcphandle);
   transCloseExHandleMgt(pRpc->refMgt);

source/libs/transport/src/transCli.c

@@ -1,4 +1,4 @@
-/* * Copyright (c) 2019 TAOS Data, Inc. <jhtao@taosdata.com>
+/** Copyright (c) 2019 TAOS Data, Inc. <jhtao@taosdata.com>
  *
  * This program is free software: you can use, redistribute, and/or modify
  * it under the terms of the GNU Affero General Public License, version 3
@@ -25,7 +25,6 @@ typedef struct SCliConn {
   uv_write_t   writeReq;
 
   void* hostThrd;
-  int   hThrdIdx;
 
   SConnBuffer readBuf;
   STransQueue cliMsgs;
@@ -36,6 +35,7 @@ typedef struct SCliConn {
   bool       broken;  // link broken or not
   ConnStatus status;  //
 
+  int64_t  refId;
   char*    ip;
   uint32_t port;
 
@@ -54,7 +54,7 @@ typedef struct SCliMsg {
   int      sent;  //(0: no send, 1: alread sent)
 } SCliMsg;
 
-typedef struct SCliThrdObj {
+typedef struct SCliThrd {
   TdThread    thread;  // tid
   int64_t     pid;     // pid
   uv_loop_t*  loop;
@@ -72,13 +72,13 @@ typedef struct SCliThrdObj {
   SCvtAddr cvtAddr;
 
   bool quit;
-} SCliThrdObj;
+} SCliThrd;
 
 typedef struct SCliObj {
   char       label[TSDB_LABEL_LEN];
   int32_t    index;
   int        numOfThreads;
-  SCliThrdObj** pThreadObj;
+  SCliThrd** pThreadObj;
 } SCliObj;
 
 typedef struct SConnList {
@@ -106,11 +106,18 @@ static void cliAsyncCb(uv_async_t* handle);
 
 static int cliAppCb(SCliConn* pConn, STransMsg* pResp, SCliMsg* pMsg);
 
-static SCliConn* cliCreateConn(SCliThrdObj* thrd);
+static SCliConn* cliCreateConn(SCliThrd* thrd);
 static void      cliDestroyConn(SCliConn* pConn, bool clear /*clear tcp handle or not*/);
 static void      cliDestroy(uv_handle_t* handle);
 static void      cliSend(SCliConn* pConn);
 
+static bool cliIsEpsetUpdated(int32_t code, STransConnCtx* pCtx) {
+  if (code != 0) return false;
+  if (pCtx->retryCnt == 0) return false;
+  if (transEpSetIsEqual(&pCtx->epSet, &pCtx->origEpSet)) return false;
+  return true;
+}
+
 void cliMayCvtFqdnToIp(SEpSet* pEpSet, SCvtAddr* pCvtAddr);
 /*
  * set TCP connection timeout per-socket level
@@ -122,14 +129,14 @@ static void cliHandleResp(SCliConn* conn);
 static void cliHandleExcept(SCliConn* conn);
 
 // handle req from app
-static void cliHandleReq(SCliMsg* pMsg, SCliThrdObj* pThrd);
+static void cliHandleReq(SCliMsg* pMsg, SCliThrd* pThrd);
-static void cliHandleQuit(SCliMsg* pMsg, SCliThrdObj* pThrd);
+static void cliHandleQuit(SCliMsg* pMsg, SCliThrd* pThrd);
-static void cliHandleRelease(SCliMsg* pMsg, SCliThrdObj* pThrd);
+static void cliHandleRelease(SCliMsg* pMsg, SCliThrd* pThrd);
-static void cliHandleUpdate(SCliMsg* pMsg, SCliThrdObj* pThrd);
+static void cliHandleUpdate(SCliMsg* pMsg, SCliThrd* pThrd);
-static void (*cliAsyncHandle[])(SCliMsg* pMsg, SCliThrdObj* pThrd) = {cliHandleReq, cliHandleQuit, cliHandleRelease,
+static void (*cliAsyncHandle[])(SCliMsg* pMsg, SCliThrd* pThrd) = {cliHandleReq, cliHandleQuit, cliHandleRelease, NULL,
-                                                                      NULL, cliHandleUpdate};
+                                                                   cliHandleUpdate};
 
-static void cliSendQuit(SCliThrdObj* thrd);
+static void cliSendQuit(SCliThrd* thrd);
 static void destroyUserdata(STransMsg* userdata);
 
 static int cliRBChoseIdx(STrans* pTransInst);
@@ -137,8 +144,8 @@ static int cliRBChoseIdx(STrans* pTransInst);
 static void destroyCmsg(SCliMsg* cmsg);
 static void transDestroyConnCtx(STransConnCtx* ctx);
 // thread obj
-static SCliThrdObj* createThrdObj();
+static SCliThrd* createThrdObj();
-static void         destroyThrdObj(SCliThrdObj* pThrd);
+static void      destroyThrdObj(SCliThrd* pThrd);
 
 static void cliWalkCb(uv_handle_t* handle, void* arg);
 
@@ -154,7 +161,6 @@ static void cliReleaseUnfinishedMsg(SCliConn* conn) {
     destroyCmsg(pMsg);
   }
 }
-
 #define CLI_RELEASE_UV(loop)        \
   do {                              \
     uv_walk(loop, cliWalkCb, NULL); \
@@ -168,17 +174,23 @@ static void cliReleaseUnfinishedMsg(SCliConn* conn) {
     snprintf(key, sizeof(key), "%s:%d", ip, (int)port); \
   } while (0)
 
-#define CONN_HOST_THREAD_IDX(conn) (conn ? ((SCliConn*)conn)->hThrdIdx : -1)
+#define CONN_HOST_THREAD_IDX1(idx, exh, refId, pThrd) \
+  do {                                                \
+    if (exh == NULL) {                                \
+      idx = -1;                                       \
+    } else {                                          \
+      ASYNC_CHECK_HANDLE((exh), refId);               \
+      pThrd = (SCliThrd*)(exh)->pThrd;                \
+    }                                                 \
+  } while (0)
 #define CONN_PERSIST_TIME(para)    (para * 1000 * 10)
 #define CONN_GET_HOST_THREAD(conn) (conn ? ((SCliConn*)conn)->hostThrd : NULL)
-#define CONN_GET_INST_LABEL(conn)  (((STrans*)(((SCliThrdObj*)(conn)->hostThrd)->pTransInst))->label)
+#define CONN_GET_INST_LABEL(conn)  (((STrans*)(((SCliThrd*)(conn)->hostThrd)->pTransInst))->label)
 #define CONN_SHOULD_RELEASE(conn, head)                                                                            \
   do {                                                                                                             \
     if ((head)->release == 1 && (head->msgLen) == sizeof(*head)) {                                                 \
-      int      connStatus = conn->status;                                                                          \
       uint64_t ahandle = head->ahandle;                                                                            \
       CONN_GET_MSGCTX_BY_AHANDLE(conn, ahandle);                                                                   \
-      conn->status = ConnRelease;                                                                                  \
       transClearBuffer(&conn->readBuf);                                                                            \
       transFreeMsg(transContFromHead((char*)head));                                                                \
       tDebug("%s conn %p receive release request, ref: %d", CONN_GET_INST_LABEL(conn), conn, T_REF_VAL_GET(conn)); \
@@ -187,9 +199,7 @@ static void cliReleaseUnfinishedMsg(SCliConn* conn) {
       }                                                                                                            \
       destroyCmsg(pMsg);                                                                                           \
       cliReleaseUnfinishedMsg(conn);                                                                               \
-      if (connStatus != ConnInPool) {                                                                              \
+      addConnToPool(((SCliThrd*)conn->hostThrd)->pool, conn);                                                      \
-        addConnToPool(((SCliThrdObj*)conn->hostThrd)->pool, conn);                                                 \
-      }                                                                                                            \
       return;                                                                                                      \
     }                                                                                                              \
   } while (0)
@@ -255,8 +265,25 @@ static void cliReleaseUnfinishedMsg(SCliConn* conn) {
 #define REQUEST_PERSIS_HANDLE(msg)   ((msg)->info.persistHandle == 1)
 #define REQUEST_RELEASE_HANDLE(cmsg) ((cmsg)->type == Release)
 
+#define EPSET_GET_SIZE(epSet)       (epSet)->numOfEps
 #define EPSET_GET_INUSE_IP(epSet)   ((epSet)->eps[(epSet)->inUse].fqdn)
 #define EPSET_GET_INUSE_PORT(epSet) ((epSet)->eps[(epSet)->inUse].port)
+#define EPSET_FORWARD_INUSE(epSet)                               \
+  do {                                                           \
+    (epSet)->inUse = (++((epSet)->inUse)) % ((epSet)->numOfEps); \
+  } while (0)
+#define EPSET_DEBUG_STR(epSet, tbuf)                                                                                   \
+  do {                                                                                                                 \
+    int len = snprintf(tbuf, sizeof(tbuf), "epset:{");                                                                 \
+    for (int i = 0; i < (epSet)->numOfEps; i++) {                                                                      \
+      if (i == (epSet)->numOfEps - 1) {                                                                                \
+        len += snprintf(tbuf + len, sizeof(tbuf) - len, "%d. %s:%d", i, (epSet)->eps[i].fqdn, (epSet)->eps[i].port);   \
+      } else {                                                                                                         \
+        len += snprintf(tbuf + len, sizeof(tbuf) - len, "%d. %s:%d, ", i, (epSet)->eps[i].fqdn, (epSet)->eps[i].port); \
+      }                                                                                                                \
+    }                                                                                                                  \
+    len += snprintf(tbuf + len, sizeof(tbuf) - len, "}, inUse:%d", (epSet)->inUse);                                    \
+  } while (0);
 
 static void* cliWorkThread(void* arg);
 
@@ -271,7 +298,7 @@ _RETURN:
   return false;
 }
 void cliHandleResp(SCliConn* conn) {
-  SCliThrdObj* pThrd = conn->hostThrd;
+  SCliThrd* pThrd = conn->hostThrd;
   STrans*   pTransInst = pThrd->pTransInst;
 
   STransMsgHead* pHead = (STransMsgHead*)(conn->readBuf.buf);
@@ -292,17 +319,9 @@ void cliHandleResp(SCliConn* conn) {
 
   if (CONN_NO_PERSIST_BY_APP(conn)) {
     pMsg = transQueuePop(&conn->cliMsgs);
-    pCtx = pMsg ? pMsg->ctx : NULL;
+    pCtx = pMsg->ctx;
-    if (pMsg == NULL && !CONN_NO_PERSIST_BY_APP(conn)) {
+    transMsg.info.ahandle = pCtx->ahandle;
-      transMsg.info.ahandle = transCtxDumpVal(&conn->ctx, transMsg.msgType);
-      if (transMsg.info.ahandle == NULL) {
-        transMsg.info.ahandle = transCtxDumpBrokenlinkVal(&conn->ctx, (int32_t*)&(transMsg.msgType));
-      }
-      tDebug("%s conn %p construct ahandle %p, persist: 0", CONN_GET_INST_LABEL(conn), conn, transMsg.info.ahandle);
-    } else {
-      transMsg.info.ahandle = pCtx ? pCtx->ahandle : NULL;
     tDebug("%s conn %p get ahandle %p, persist: 0", CONN_GET_INST_LABEL(conn), conn, transMsg.info.ahandle);
-    }
   } else {
     uint64_t ahandle = (uint64_t)pHead->ahandle;
     CONN_GET_MSGCTX_BY_AHANDLE(conn, ahandle);
@@ -324,26 +343,22 @@ void cliHandleResp(SCliConn* conn) {
   }
   // buf's mem alread translated to transMsg.pCont
   transClearBuffer(&conn->readBuf);
-
   if (!CONN_NO_PERSIST_BY_APP(conn)) {
-    transMsg.info.handle = conn;
+    transMsg.info.handle = (void*)conn->refId;
     tDebug("%s conn %p ref by app", CONN_GET_INST_LABEL(conn), conn);
   }
-  // char buf[64] = {0};
+
-  // TRACE_TO_STR(&transMsg.info.traceId, buf);
   STraceId* trace = &transMsg.info.traceId;
-  tGTrace("conn %p %s received from %s:%d, local info: %s:%d, msg size: %d, code: %d", conn, TMSG_INFO(pHead->msgType),
+  tGTrace("%s conn %p %s received from %s:%d, local info: %s:%d, msg size: %d, code: %d", CONN_GET_INST_LABEL(conn),
-          taosInetNtoa(conn->addr.sin_addr), ntohs(conn->addr.sin_port), taosInetNtoa(conn->localAddr.sin_addr),
+          conn, TMSG_INFO(pHead->msgType), taosInetNtoa(conn->addr.sin_addr), ntohs(conn->addr.sin_port),
-          ntohs(conn->localAddr.sin_port), transMsg.contLen, transMsg.code);
+          taosInetNtoa(conn->localAddr.sin_addr), ntohs(conn->localAddr.sin_port), transMsg.contLen, transMsg.code);
 
   if (pCtx == NULL && CONN_NO_PERSIST_BY_APP(conn)) {
-    tDebug("%s except, server continue send while cli ignore it", CONN_GET_INST_LABEL(conn));
+    tDebug("%s except, conn %p read while cli ignore it", CONN_GET_INST_LABEL(conn), conn);
-    // transUnrefCliHandle(conn);
     return;
   }
   if (CONN_RELEASE_BY_SERVER(conn) && transMsg.info.ahandle == NULL) {
-    tDebug("%s except, server continue send while cli ignore it", CONN_GET_INST_LABEL(conn));
+    tDebug("%s except, conn %p read while cli ignore it", CONN_GET_INST_LABEL(conn), conn);
-    // transUnrefCliHandle(conn);
     return;
   }
 
@@ -375,7 +390,7 @@ void cliHandleExcept(SCliConn* pConn) {
       return;
     }
   }
-  SCliThrdObj* pThrd = pConn->hostThrd;
+  SCliThrd* pThrd = pConn->hostThrd;
   STrans*   pTransInst = pThrd->pTransInst;
   bool      once = false;
   do {
@@ -389,7 +404,6 @@ void cliHandleExcept(SCliConn* pConn) {
     transMsg.code = TSDB_CODE_RPC_NETWORK_UNAVAIL;
     transMsg.msgType = pMsg ? pMsg->msg.msgType + 1 : 0;
     transMsg.info.ahandle = NULL;
-    transMsg.info.handle = pConn;
 
     if (pMsg == NULL && !CONN_NO_PERSIST_BY_APP(pConn)) {
       transMsg.info.ahandle = transCtxDumpVal(&pConn->ctx, transMsg.msgType);
@@ -414,13 +428,13 @@ void cliHandleExcept(SCliConn* pConn) {
       return;
     }
     destroyCmsg(pMsg);
-    tTrace("%s conn %p start to destroy", CONN_GET_INST_LABEL(pConn), pConn);
+    tTrace("%s conn %p start to destroy, ref:%d", CONN_GET_INST_LABEL(pConn), pConn, T_REF_VAL_GET(pConn));
   } while (!transQueueEmpty(&pConn->cliMsgs));
   transUnrefCliHandle(pConn);
 }
 
 void cliTimeoutCb(uv_timer_t* handle) {
-  SCliThrdObj* pThrd = handle->data;
+  SCliThrd* pThrd = handle->data;
   STrans*   pTransInst = pThrd->pTransInst;
   int64_t   currentTime = pThrd->nextTimeout;
   tTrace("%s conn timeout, try to remove expire conn from conn pool", pTransInst->label);
@@ -487,10 +501,26 @@ static SCliConn* getConnFromPool(void* pool, char* ip, uint32_t port) {
   assert(h == &conn->conn);
   return conn;
 }
+static void allocConnRef(SCliConn* conn, bool update) {
+  if (update) {
+    transRemoveExHandle(refMgt, conn->refId);
+  }
+  SExHandle* exh = taosMemoryCalloc(1, sizeof(SExHandle));
+  exh->handle = conn;
+  exh->pThrd = conn->hostThrd;
+  exh->refId = transAddExHandle(refMgt, exh);
+  conn->refId = exh->refId;
+}
 static void addConnToPool(void* pool, SCliConn* conn) {
-  SCliThrdObj* thrd = conn->hostThrd;
+  if (conn->status == ConnInPool) {
+    // assert(0);
+    return;
+  }
+  SCliThrd* thrd = conn->hostThrd;
   CONN_HANDLE_THREAD_QUIT(thrd);
 
+  allocConnRef(conn, true);
+
   STrans* pTransInst = thrd->pTransInst;
   conn->expireTime = taosGetTimestampMs() + CONN_PERSIST_TIME(pTransInst->idleTime);
   transQueueClear(&conn->cliMsgs);
@@ -499,7 +529,7 @@ static void addConnToPool(void* pool, SCliConn* conn) {
 
   char key[128] = {0};
   CONN_CONSTRUCT_HASH_KEY(key, conn->ip, conn->port);
-  tTrace("%s conn %p added to conn pool, read buf cap: %d", CONN_GET_INST_LABEL(conn), conn, conn->readBuf.cap);
+  tTrace("%s conn %p added to conn pool, read buf cap:%d", CONN_GET_INST_LABEL(conn), conn, conn->readBuf.cap);
 
   SConnList* plist = taosHashGet((SHashObj*)pool, key, strlen(key));
   // list already create before
@@ -540,13 +570,14 @@ static void cliRecvCb(uv_stream_t* handle, ssize_t nread, const uv_buf_t* buf) {
     return;
   }
   if (nread < 0) {
-    tError("%s conn %p read error: %s", CONN_GET_INST_LABEL(conn), conn, uv_err_name(nread));
+    tError("%s conn %p read error: %s, ref: %d", CONN_GET_INST_LABEL(conn), conn, uv_err_name(nread),
+           T_REF_VAL_GET(conn));
     conn->broken = true;
     cliHandleExcept(conn);
   }
 }
 
-static SCliConn* cliCreateConn(SCliThrdObj* pThrd) {
+static SCliConn* cliCreateConn(SCliThrd* pThrd) {
   SCliConn* conn = taosMemoryCalloc(1, sizeof(SCliConn));
   // read/write stream handle
   conn->stream = (uv_stream_t*)taosMemoryMalloc(sizeof(uv_tcp_t));
@@ -562,11 +593,16 @@ static SCliConn* cliCreateConn(SCliThrdObj* pThrd) {
   conn->status = ConnNormal;
   conn->broken = 0;
   transRefCliHandle(conn);
+
+  allocConnRef(conn, false);
+
   return conn;
 }
 static void cliDestroyConn(SCliConn* conn, bool clear) {
   tTrace("%s conn %p remove from conn pool", CONN_GET_INST_LABEL(conn), conn);
   QUEUE_REMOVE(&conn->conn);
+  QUEUE_INIT(&conn->conn);
+  transRemoveExHandle(refMgt, conn->refId);
   if (clear) {
     uv_close((uv_handle_t*)conn->stream, cliDestroy);
   }
@@ -593,7 +629,7 @@ static bool cliHandleNoResp(SCliConn* conn) {
     }
     if (res == true) {
       if (cliMaySendCachedMsg(conn) == false) {
-        SCliThrdObj* thrd = conn->hostThrd;
+        SCliThrd* thrd = conn->hostThrd;
         addConnToPool(thrd->pool, conn);
       }
     }
@@ -629,7 +665,7 @@ void cliSend(SCliConn* pConn) {
 
   STransConnCtx* pCtx = pCliMsg->ctx;
 
-  SCliThrdObj* pThrd = pConn->hostThrd;
+  SCliThrd* pThrd = pConn->hostThrd;
   STrans*   pTransInst = pThrd->pTransInst;
 
   STransMsg* pMsg = (STransMsg*)(&pCliMsg->msg);
@@ -651,12 +687,10 @@ void cliSend(SCliConn* pConn) {
 
   uv_buf_t wb = uv_buf_init((char*)pHead, msgLen);
 
-  // char buf[64] = {0};
-  // TRACE_TO_STR(&pMsg->info.traceId, buf);
   STraceId* trace = &pMsg->info.traceId;
-  tGTrace("conn %p %s is sent to %s:%d, local info %s:%d", pConn, TMSG_INFO(pHead->msgType),
+  tGTrace("%s conn %p %s is sent to %s:%d, local info %s:%d", CONN_GET_INST_LABEL(pConn), pConn,
-          taosInetNtoa(pConn->addr.sin_addr), ntohs(pConn->addr.sin_port), taosInetNtoa(pConn->localAddr.sin_addr),
+          TMSG_INFO(pHead->msgType), taosInetNtoa(pConn->addr.sin_addr), ntohs(pConn->addr.sin_port),
-          ntohs(pConn->localAddr.sin_port));
+          taosInetNtoa(pConn->localAddr.sin_addr), ntohs(pConn->localAddr.sin_port));
 
   if (pHead->persist == 1) {
     CONN_SET_PERSIST_BY_APP(pConn);
@@ -664,7 +698,6 @@ void cliSend(SCliConn* pConn) {
 
   pConn->writeReq.data = pConn;
   uv_write(&pConn->writeReq, (uv_stream_t*)pConn->stream, &wb, 1, cliSendCb);
-
   return;
 _RETURN:
   return;
@@ -690,19 +723,24 @@ void cliConnCb(uv_connect_t* req, int status) {
   cliSend(pConn);
 }
 
-static void cliHandleQuit(SCliMsg* pMsg, SCliThrdObj* pThrd) {
+static void cliHandleQuit(SCliMsg* pMsg, SCliThrd* pThrd) {
+  pThrd->quit = true;
   tDebug("cli work thread %p start to quit", pThrd);
   destroyCmsg(pMsg);
   destroyConnPool(pThrd->pool);
   uv_timer_stop(&pThrd->timer);
   uv_walk(pThrd->loop, cliWalkCb, NULL);
 
-  pThrd->quit = true;
-
   // uv_stop(pThrd->loop);
 }
-static void cliHandleRelease(SCliMsg* pMsg, SCliThrdObj* pThrd) {
+static void cliHandleRelease(SCliMsg* pMsg, SCliThrd* pThrd) {
-  SCliConn* conn = pMsg->msg.info.handle;
+  int64_t    refId = (int64_t)(pMsg->msg.info.handle);
+  SExHandle* exh = transAcquireExHandle(refMgt, refId);
+  if (exh == NULL) {
+    tDebug("%" PRId64 " already release", refId);
+  }
+
+  SCliConn* conn = exh->handle;
   tDebug("%s conn %p start to release to inst", CONN_GET_INST_LABEL(conn), conn);
 
   if (T_REF_VAL_GET(conn) == 2) {
@@ -711,33 +749,37 @@ static void cliHandleRelease(SCliMsg* pMsg, SCliThrdObj* pThrd) {
       return;
     }
     cliSend(conn);
-  } else {
-    // conn already broken down
-    transUnrefCliHandle(conn);
   }
 }
-static void cliHandleUpdate(SCliMsg* pMsg, SCliThrdObj* pThrd) {
+static void cliHandleUpdate(SCliMsg* pMsg, SCliThrd* pThrd) {
   STransConnCtx* pCtx = pMsg->ctx;
-
   pThrd->cvtAddr = pCtx->cvtAddr;
   destroyCmsg(pMsg);
 }
 
-SCliConn* cliGetConn(SCliMsg* pMsg, SCliThrdObj* pThrd) {
+SCliConn* cliGetConn(SCliMsg* pMsg, SCliThrd* pThrd, bool* ignore) {
   SCliConn* conn = NULL;
-  if (pMsg->msg.info.handle != NULL) {
+  int64_t   refId = (int64_t)(pMsg->msg.info.handle);
-    conn = (SCliConn*)(pMsg->msg.info.handle);
+  if (refId != 0) {
-    if (conn != NULL) {
+    SExHandle* exh = transAcquireExHandle(refMgt, refId);
-      tTrace("%s conn %p reused", CONN_GET_INST_LABEL(conn), conn);
+    if (exh == NULL) {
-    }
+      *ignore = true;
+      destroyCmsg(pMsg);
+      return NULL;
+      // assert(0);
     } else {
+      conn = exh->handle;
+      transReleaseExHandle(refMgt, refId);
+    }
+    return conn;
+  };
+
   STransConnCtx* pCtx = pMsg->ctx;
   conn = getConnFromPool(pThrd->pool, EPSET_GET_INUSE_IP(&pCtx->epSet), EPSET_GET_INUSE_PORT(&pCtx->epSet));
   if (conn != NULL) {
-      tTrace("%s conn %p get from conn pool", CONN_GET_INST_LABEL(conn), conn);
+    tTrace("%s conn %p get from conn pool:%p", CONN_GET_INST_LABEL(conn), conn, pThrd->pool);
   } else {
-      tTrace("%s not found conn in conn pool %p", ((STrans*)pThrd->pTransInst)->label, pThrd->pool);
+    tTrace("%s not found conn in conn pool:%p", ((STrans*)pThrd->pTransInst)->label, pThrd->pool);
-    }
   }
   return conn;
 }
@@ -752,22 +794,19 @@ void cliMayCvtFqdnToIp(SEpSet* pEpSet, SCvtAddr* pCvtAddr) {
     }
   }
 }
-void cliHandleReq(SCliMsg* pMsg, SCliThrdObj* pThrd) {
+void cliHandleReq(SCliMsg* pMsg, SCliThrd* pThrd) {
-  uint64_t et = taosGetTimestampUs();
-  uint64_t el = et - pMsg->st;
-  // tTrace("%s cli msg tran time cost: %" PRIu64 "us", ((STrans*)pThrd->pTransInst)->label, el);
-
   STransConnCtx* pCtx = pMsg->ctx;
   STrans*        pTransInst = pThrd->pTransInst;
 
   cliMayCvtFqdnToIp(&pCtx->epSet, &pThrd->cvtAddr);
 
-  transPrintEpSet(&pCtx->epSet);
+  // transPrintEpSet(&pCtx->epSet);
-
+  bool      ignore = false;
-  SCliConn* conn = cliGetConn(pMsg, pThrd);
+  SCliConn* conn = cliGetConn(pMsg, pThrd, &ignore);
+  if (ignore == true) {
+    return;
+  }
   if (conn != NULL) {
-    conn->hThrdIdx = pCtx->hThrdIdx;
-
     transCtxMerge(&conn->ctx, &pCtx->appCtx);
     transQueuePush(&conn->cliMsgs, pMsg);
     cliSend(conn);
@@ -776,7 +815,6 @@ void cliHandleReq(SCliMsg* pMsg, SCliThrdObj* pThrd) {
     transCtxMerge(&conn->ctx, &pCtx->appCtx);
     transQueuePush(&conn->cliMsgs, pMsg);
 
-    conn->hThrdIdx = pCtx->hThrdIdx;
     conn->ip = strdup(EPSET_GET_INUSE_IP(&pCtx->epSet));
     conn->port = EPSET_GET_INUSE_PORT(&pCtx->epSet);
 
@@ -784,7 +822,7 @@ void cliHandleReq(SCliMsg* pMsg, SCliThrdObj* pThrd) {
     if (ret) {
       tError("%s conn %p failed to set conn option, errmsg %s", transLabel(pTransInst), conn, uv_err_name(ret));
     }
-    int fd = taosCreateSocketWithTimeOutOpt(TRANS_CONN_TIMEOUT);
+    int32_t fd = taosCreateSocketWithTimeout(TRANS_CONN_TIMEOUT);
     if (fd == -1) {
       tTrace("%s conn %p failed to create socket", transLabel(pTransInst), conn);
       cliHandleExcept(conn);
@@ -808,7 +846,7 @@ void cliHandleReq(SCliMsg* pMsg, SCliThrdObj* pThrd) {
 }
 static void cliAsyncCb(uv_async_t* handle) {
   SAsyncItem* item = handle->data;
-  SCliThrdObj* pThrd = item->pThrd;
+  SCliThrd*   pThrd = item->pThrd;
   SCliMsg*    pMsg = NULL;
 
   // batch process to avoid to lock/unlock frequently
@@ -835,7 +873,7 @@ static void cliAsyncCb(uv_async_t* handle) {
 }
 
 static void* cliWorkThread(void* arg) {
-  SCliThrdObj* pThrd = (SCliThrdObj*)arg;
+  SCliThrd* pThrd = (SCliThrd*)arg;
   pThrd->pid = taosGetSelfPthreadId();
   setThreadName("trans-cli-work");
   uv_run(pThrd->loop, UV_RUN_DEFAULT);
@@ -848,10 +886,10 @@ void* transInitClient(uint32_t ip, uint32_t port, char* label, int numOfThreads,
   STrans* pTransInst = shandle;
   memcpy(cli->label, label, strlen(label));
   cli->numOfThreads = numOfThreads;
-  cli->pThreadObj = (SCliThrdObj**)taosMemoryCalloc(cli->numOfThreads, sizeof(SCliThrdObj*));
+  cli->pThreadObj = (SCliThrd**)taosMemoryCalloc(cli->numOfThreads, sizeof(SCliThrd*));
 
   for (int i = 0; i < cli->numOfThreads; i++) {
-    SCliThrdObj* pThrd = createThrdObj();
+    SCliThrd* pThrd = createThrdObj();
     pThrd->nextTimeout = taosGetTimestampMs() + CONN_PERSIST_TIME(pTransInst->idleTime);
     pThrd->pTransInst = shandle;
 
@@ -885,8 +923,8 @@ static void destroyCmsg(SCliMsg* pMsg) {
   taosMemoryFree(pMsg);
 }
 
-static SCliThrdObj* createThrdObj() {
+static SCliThrd* createThrdObj() {
-  SCliThrdObj* pThrd = (SCliThrdObj*)taosMemoryCalloc(1, sizeof(SCliThrdObj));
+  SCliThrd* pThrd = (SCliThrd*)taosMemoryCalloc(1, sizeof(SCliThrd));
 
   QUEUE_INIT(&pThrd->msg);
   taosThreadMutexInit(&pThrd->msgMtx, NULL);
@@ -904,7 +942,7 @@ static SCliThrdObj* createThrdObj() {
   pThrd->quit = false;
   return pThrd;
 }
-static void destroyThrdObj(SCliThrdObj* pThrd) {
+static void destroyThrdObj(SCliThrd* pThrd) {
   if (pThrd == NULL) {
     return;
   }
@@ -925,7 +963,7 @@ static void transDestroyConnCtx(STransConnCtx* ctx) {
   taosMemoryFree(ctx);
 }
 
-void cliSendQuit(SCliThrdObj* thrd) {
+void cliSendQuit(SCliThrd* thrd) {
   // cli can stop gracefully
   SCliMsg* msg = taosMemoryCalloc(1, sizeof(SCliMsg));
   msg->type = Quit;
@@ -938,7 +976,10 @@ void cliWalkCb(uv_handle_t* handle, void* arg) {
 }
 
 int cliRBChoseIdx(STrans* pTransInst) {
-  int64_t index = pTransInst->index;
+  int8_t index = pTransInst->index;
+  if (pTransInst->numOfThreads == 0) {
+    return -1;
+  }
   if (pTransInst->index++ >= pTransInst->numOfThreads) {
     pTransInst->index = 0;
   }
@@ -946,15 +987,35 @@ int cliRBChoseIdx(STrans* pTransInst) {
 }
 static void doDelayTask(void* param) {
   STaskArg* arg = param;
-
   SCliMsg*  pMsg = arg->param1;
-  SCliThrdObj* pThrd = arg->param2;
+  SCliThrd* pThrd = arg->param2;
-  cliHandleReq(pMsg, pThrd);
-
   taosMemoryFree(arg);
+
+  cliHandleReq(pMsg, pThrd);
+}
+
+static void cliSchedMsgToNextNode(SCliMsg* pMsg, SCliThrd* pThrd) {
+  STransConnCtx* pCtx = pMsg->ctx;
+
+  STraceId* trace = &pMsg->msg.info.traceId;
+  char      tbuf[256] = {0};
+  EPSET_DEBUG_STR(&pCtx->epSet, tbuf);
+  tGTrace("%s retry on next node, use %s, retryCnt:%d, limit:%d", transLabel(pThrd->pTransInst), tbuf,
+          pCtx->retryCnt + 1, pCtx->retryLimit);
+
+  STaskArg* arg = taosMemoryMalloc(sizeof(STaskArg));
+  arg->param1 = pMsg;
+  arg->param2 = pThrd;
+  transDQSched(pThrd->delayQueue, doDelayTask, arg, TRANS_RETRY_INTERVAL);
+}
+
+void cliCompareAndSwap(int8_t* val, int8_t exp, int8_t newVal) {
+  if (*val != exp) {
+    *val = newVal;
+  }
 }
 int cliAppCb(SCliConn* pConn, STransMsg* pResp, SCliMsg* pMsg) {
-  SCliThrdObj* pThrd = pConn->hostThrd;
+  SCliThrd* pThrd = pConn->hostThrd;
   STrans*   pTransInst = pThrd->pTransInst;
 
   if (pMsg == NULL || pMsg->ctx == NULL) {
@@ -962,66 +1023,35 @@ int cliAppCb(SCliConn* pConn, STransMsg* pResp, SCliMsg* pMsg) {
     pTransInst->cfp(pTransInst->parent, pResp, NULL);
     return 0;
   }
-
-  STransConnCtx* pCtx = pMsg->ctx;
-  SEpSet*        pEpSet = &pCtx->epSet;
-
-  if (pCtx->retryCount == 0) {
-    pCtx->origEpSet = pCtx->epSet;
-  }
   /*
-   * upper layer handle retry if code equal TSDB_CODE_RPC_NETWORK_UNAVAIL
+   *  no retry
+   *  1. query conn
+   *  2. rpc thread already receive quit msg
    */
-  tmsg_t msgType = pCtx->msgType;
+  STransConnCtx* pCtx = pMsg->ctx;
-  if ((pTransInst->retry != NULL && pEpSet->numOfEps > 1 && (pTransInst->retry(pResp->code))) ||
+  int32_t        code = pResp->code;
-      (pResp->code == TSDB_CODE_RPC_NETWORK_UNAVAIL || pResp->code == TSDB_CODE_APP_NOT_READY ||
+  if (pTransInst->retry != NULL && pTransInst->retry(code)) {
-       pResp->code == TSDB_CODE_NODE_NOT_DEPLOYED || pResp->code == TSDB_CODE_SYN_NOT_LEADER)) {
     pMsg->sent = 0;
-    tTrace("try to send req to next node");
+    pCtx->retryCnt += 1;
-    pMsg->st = taosGetTimestampUs();
+    if (code == TSDB_CODE_RPC_NETWORK_UNAVAIL) {
-    pCtx->retryCount += 1;
+      cliCompareAndSwap(&pCtx->retryLimit, TRANS_RETRY_COUNT_LIMIT, EPSET_GET_SIZE(&pCtx->epSet) * 3);
-    if (pResp->code == TSDB_CODE_RPC_NETWORK_UNAVAIL && pCtx->setMaxRetry == false) {
+      if (pCtx->retryCnt < pCtx->retryLimit) {
-      if (pCtx->retryCount < pEpSet->numOfEps * 3) {
-        pEpSet->inUse = (++pEpSet->inUse) % pEpSet->numOfEps;
-        if (pThrd->quit == false) {
-          STaskArg* arg = taosMemoryMalloc(sizeof(STaskArg));
-          arg->param1 = pMsg;
-          arg->param2 = pThrd;
-          transDQSched(pThrd->delayQueue, doDelayTask, arg, TRANS_RETRY_INTERVAL);
-          transPrintEpSet(pEpSet);
-          tTrace("%s use local epset, inUse: %d, retry count:%d, limit: %d", pTransInst->label, pEpSet->inUse,
-                 pCtx->retryCount + 1, pEpSet->numOfEps * 3);
-
         transUnrefCliHandle(pConn);
+        EPSET_FORWARD_INUSE(&pCtx->epSet);
+        cliSchedMsgToNextNode(pMsg, pThrd);
         return -1;
       }
-      }
+    } else {
-    } else if (pCtx->retryCount < TRANS_RETRY_COUNT_LIMIT) {
+      cliCompareAndSwap(&pCtx->retryLimit, TRANS_RETRY_COUNT_LIMIT, TRANS_RETRY_COUNT_LIMIT);
-      pCtx->setMaxRetry = true;
+      if (pCtx->retryCnt < pCtx->retryLimit) {
+        addConnToPool(pThrd->pool, pConn);
         if (pResp->contLen == 0) {
-        pEpSet->inUse = (++pEpSet->inUse) % pEpSet->numOfEps;
+          EPSET_FORWARD_INUSE(&pCtx->epSet);
-        transPrintEpSet(&pCtx->epSet);
-        tTrace("%s use local epset, inUse: %d, retry count:%d, limit: %d", pTransInst->label, pEpSet->inUse,
-               pCtx->retryCount + 1, TRANS_RETRY_COUNT_LIMIT);
         } else {
-        SEpSet epSet = {0};
+          tDeserializeSEpSet(pResp->pCont, pResp->contLen, &pCtx->epSet);
-        tDeserializeSEpSet(pResp->pCont, pResp->contLen, &epSet);
-        pCtx->epSet = epSet;
-
-        transPrintEpSet(&pCtx->epSet);
-        tTrace("%s use remote epset, inUse: %d, retry count:%d, limit: %d", pTransInst->label, pEpSet->inUse,
-               pCtx->retryCount + 1, TRANS_RETRY_COUNT_LIMIT);
         }
-      if (pThrd->quit == false) {
+        transFreeMsg(pResp->pCont);
-        if (pResp->code != TSDB_CODE_RPC_NETWORK_UNAVAIL) {
+        cliSchedMsgToNextNode(pMsg, pThrd);
-          if (pConn->status != ConnInPool) addConnToPool(pThrd->pool, pConn);
-        } else {
-          transUnrefCliHandle(pConn);
-        }
-        STaskArg* arg = taosMemoryMalloc(sizeof(STaskArg));
-        arg->param1 = pMsg;
-        arg->param2 = pThrd;
-        transDQSched(pThrd->delayQueue, doDelayTask, arg, TRANS_RETRY_INTERVAL);
         return -1;
       }
     }
@@ -1029,20 +1059,20 @@ int cliAppCb(SCliConn* pConn, STransMsg* pResp, SCliMsg* pMsg) {
 
   STraceId* trace = &pResp->info.traceId;
   if (pCtx->pSem != NULL) {
-    tGTrace("conn %p(sync) handle resp", pConn);
+    tGTrace("%s conn %p(sync) handle resp", CONN_GET_INST_LABEL(pConn), pConn);
     if (pCtx->pRsp == NULL) {
-      tGTrace("conn %p(sync) failed to resp, ignore", pConn);
+      tGTrace("%s conn %p(sync) failed to resp, ignore", CONN_GET_INST_LABEL(pConn), pConn);
     } else {
       memcpy((char*)pCtx->pRsp, (char*)pResp, sizeof(*pResp));
     }
     tsem_post(pCtx->pSem);
     pCtx->pRsp = NULL;
   } else {
-    tGTrace("conn %p handle resp", pConn);
+    tGTrace("%s conn %p handle resp", CONN_GET_INST_LABEL(pConn), pConn);
-    if (pResp->code != 0 || pCtx->retryCount == 0 || transEpSetIsEqual(&pCtx->epSet, &pCtx->origEpSet)) {
+    if (!cliIsEpsetUpdated(code, pCtx)) {
       pTransInst->cfp(pTransInst->parent, pResp, NULL);
     } else {
-      pTransInst->cfp(pTransInst->parent, pResp, pEpSet);
+      pTransInst->cfp(pTransInst->parent, pResp, &pCtx->epSet);
     }
   }
   return 0;
@@ -1074,38 +1104,58 @@ void transUnrefCliHandle(void* handle) {
     return;
   }
   int ref = T_REF_DEC((SCliConn*)handle);
-  tTrace("%s conn %p ref %d", CONN_GET_INST_LABEL((SCliConn*)handle), handle, ref);
+  tTrace("%s conn %p ref:%d", CONN_GET_INST_LABEL((SCliConn*)handle), handle, ref);
   if (ref == 0) {
     cliDestroyConn((SCliConn*)handle, true);
   }
 }
+SCliThrd* transGetWorkThrdFromHandle(int64_t handle) {
+  SCliThrd*  pThrd = NULL;
+  SExHandle* exh = transAcquireExHandle(refMgt, handle);
+  if (exh == NULL) {
+    return NULL;
+  }
+  pThrd = exh->pThrd;
+  transReleaseExHandle(refMgt, handle);
+  return pThrd;
+}
+SCliThrd* transGetWorkThrd(STrans* trans, int64_t handle) {
+  if (handle == 0) {
+    int idx = cliRBChoseIdx(trans);
+    if (idx < 0) return NULL;
+    return ((SCliObj*)trans->tcphandle)->pThreadObj[idx];
+  }
+  return transGetWorkThrdFromHandle(handle);
+}
 void transReleaseCliHandle(void* handle) {
-  SCliThrdObj* thrd = CONN_GET_HOST_THREAD(handle);
+  int       idx = -1;
-  if (thrd == NULL) {
+  SCliThrd* pThrd = transGetWorkThrdFromHandle((int64_t)handle);
+  if (pThrd == NULL) {
     return;
   }
-
   STransMsg tmsg = {.info.handle = handle};
   SCliMsg*  cmsg = taosMemoryCalloc(1, sizeof(SCliMsg));
   cmsg->msg = tmsg;
   cmsg->type = Release;
 
-  transSendAsync(thrd->asyncPool, &cmsg->q);
+  transSendAsync(pThrd->asyncPool, &cmsg->q);
+  return;
 }
 
 void transSendRequest(void* shandle, const SEpSet* pEpSet, STransMsg* pReq, STransCtx* ctx) {
   STrans*   pTransInst = (STrans*)shandle;
-  int     idx = CONN_HOST_THREAD_IDX((SCliConn*)pReq->info.handle);
+  SCliThrd* pThrd = transGetWorkThrd(pTransInst, (int64_t)pReq->info.handle);
-  if (idx == -1) {
+  if (pThrd == NULL) {
-    idx = cliRBChoseIdx(pTransInst);
+    transFreeMsg(pReq->pCont);
+    return;
   }
+
   TRACE_SET_MSGID(&pReq->info.traceId, tGenIdPI64());
 
   STransConnCtx* pCtx = taosMemoryCalloc(1, sizeof(STransConnCtx));
   pCtx->epSet = *pEpSet;
   pCtx->ahandle = pReq->info.ahandle;
   pCtx->msgType = pReq->msgType;
-  pCtx->hThrdIdx = idx;
 
   if (ctx != NULL) {
     pCtx->appCtx = *ctx;
@@ -1118,19 +1168,19 @@ void transSendRequest(void* shandle, const SEpSet* pEpSet, STransMsg* pReq, STra
   cliMsg->st = taosGetTimestampUs();
   cliMsg->type = Normal;
 
-  SCliThrdObj* thrd = ((SCliObj*)pTransInst->tcphandle)->pThreadObj[idx];
-
   STraceId* trace = &pReq->info.traceId;
-  tGTrace("%s send request at thread:%08" PRId64 ", dst: %s:%d, app:%p", transLabel(pTransInst), thrd->pid,
+  tGTrace("%s send request at thread:%08" PRId64 ", dst: %s:%d, app:%p", transLabel(pTransInst), pThrd->pid,
           EPSET_GET_INUSE_IP(&pCtx->epSet), EPSET_GET_INUSE_PORT(&pCtx->epSet), pReq->info.ahandle);
-  ASSERT(transSendAsync(thrd->asyncPool, &(cliMsg->q)) == 0);
+  ASSERT(transSendAsync(pThrd->asyncPool, &(cliMsg->q)) == 0);
+  return;
 }
 
 void transSendRecv(void* shandle, const SEpSet* pEpSet, STransMsg* pReq, STransMsg* pRsp) {
   STrans*   pTransInst = (STrans*)shandle;
-  int     idx = CONN_HOST_THREAD_IDX(pReq->info.handle);
+  SCliThrd* pThrd = transGetWorkThrd(pTransInst, (int64_t)pReq->info.handle);
-  if (idx == -1) {
+  if (pThrd == NULL) {
-    idx = cliRBChoseIdx(pTransInst);
+    transFreeMsg(pReq->pCont);
+    return;
   }
   tsem_t* sem = taosMemoryCalloc(1, sizeof(tsem_t));
   tsem_init(sem, 0, 0);
@@ -1139,9 +1189,9 @@ void transSendRecv(void* shandle, const SEpSet* pEpSet, STransMsg* pReq, STransM
 
   STransConnCtx* pCtx = taosMemoryCalloc(1, sizeof(STransConnCtx));
   pCtx->epSet = *pEpSet;
+  pCtx->origEpSet = *pEpSet;
   pCtx->ahandle = pReq->info.ahandle;
   pCtx->msgType = pReq->msgType;
-  pCtx->hThrdIdx = idx;
   pCtx->pSem = sem;
   pCtx->pRsp = pRsp;
 
@@ -1151,22 +1201,21 @@ void transSendRecv(void* shandle, const SEpSet* pEpSet, STransMsg* pReq, STransM
   cliMsg->st = taosGetTimestampUs();
   cliMsg->type = Normal;
 
-  SCliThrdObj* thrd = ((SCliObj*)pTransInst->tcphandle)->pThreadObj[idx];
-
   STraceId* trace = &pReq->info.traceId;
-  tGTrace("%s send request at thread:%08" PRId64 ", dst: %s:%d, app:%p", transLabel(pTransInst), thrd->pid,
+  tGTrace("%s send request at thread:%08" PRId64 ", dst: %s:%d, app:%p", transLabel(pTransInst), pThrd->pid,
           EPSET_GET_INUSE_IP(&pCtx->epSet), EPSET_GET_INUSE_PORT(&pCtx->epSet), pReq->info.ahandle);
 
-  transSendAsync(thrd->asyncPool, &(cliMsg->q));
+  transSendAsync(pThrd->asyncPool, &(cliMsg->q));
   tsem_wait(sem);
   tsem_destroy(sem);
   taosMemoryFree(sem);
+  return;
 }
 /*
  *
  **/
-void transSetDefaultAddr(void* ahandle, const char* ip, const char* fqdn) {
+void transSetDefaultAddr(void* shandle, const char* ip, const char* fqdn) {
-  STrans* pTransInst = ahandle;
+  STrans* pTransInst = shandle;
 
   SCvtAddr cvtAddr = {0};
   if (ip != NULL && fqdn != NULL) {
@@ -1176,14 +1225,13 @@ void transSetDefaultAddr(void* ahandle, const char* ip, const char* fqdn) {
   }
   for (int i = 0; i < pTransInst->numOfThreads; i++) {
     STransConnCtx* pCtx = taosMemoryCalloc(1, sizeof(STransConnCtx));
-    pCtx->hThrdIdx = i;
     pCtx->cvtAddr = cvtAddr;
 
     SCliMsg* cliMsg = taosMemoryCalloc(1, sizeof(SCliMsg));
     cliMsg->ctx = pCtx;
     cliMsg->type = Update;
 
-    SCliThrdObj* thrd = ((SCliObj*)pTransInst->tcphandle)->pThreadObj[i];
+    SCliThrd* thrd = ((SCliObj*)pTransInst->tcphandle)->pThreadObj[i];
     tDebug("%s update epset at thread:%08" PRId64 "", pTransInst->label, thrd->pid);
 
     transSendAsync(thrd->asyncPool, &(cliMsg->q));

source/libs/transport/src/transComm.c

@@ -455,16 +455,16 @@ void transPrintEpSet(SEpSet* pEpSet) {
     return;
   }
   char buf[512] = {0};
-  int  len = snprintf(buf, sizeof(buf), "epset { ");
+  int  len = snprintf(buf, sizeof(buf), "epset:{");
   for (int i = 0; i < pEpSet->numOfEps; i++) {
     if (i == pEpSet->numOfEps - 1) {
-      len += snprintf(buf + len, sizeof(buf) - len, "%d. %s:%d ", i, pEpSet->eps[i].fqdn, pEpSet->eps[i].port);
+      len += snprintf(buf + len, sizeof(buf) - len, "%d. %s:%d", i, pEpSet->eps[i].fqdn, pEpSet->eps[i].port);
     } else {
       len += snprintf(buf + len, sizeof(buf) - len, "%d. %s:%d, ", i, pEpSet->eps[i].fqdn, pEpSet->eps[i].port);
     }
   }
   len += snprintf(buf + len, sizeof(buf) - len, "}");
-  tTrace("%s, inUse: %d", buf, pEpSet->inUse);
+  tTrace("%s, inUse:%d", buf, pEpSet->inUse);
 }
 bool transEpSetIsEqual(SEpSet* a, SEpSet* b) {
   if (a->numOfEps != b->numOfEps || a->inUse != b->inUse) {

source/libs/transport/src/transSvr.c

@@ -65,7 +65,7 @@ typedef struct SSvrMsg {
   STransMsgType type;
 } SSvrMsg;
 
-typedef struct SWorkThrdObj {
+typedef struct SWorkThrd {
   TdThread      thread;
   uv_connect_t  connect_req;
   uv_pipe_t*    pipe;
@@ -78,7 +78,7 @@ typedef struct SWorkThrdObj {
   queue conn;
   void* pTransInst;
   bool  quit;
-} SWorkThrdObj;
+} SWorkThrd;
 
 typedef struct SServerObj {
   TdThread   thread;
@@ -89,7 +89,7 @@ typedef struct SServerObj {
   int         workerIdx;
   int         numOfThreads;
   int         numOfWorkerReady;
-  SWorkThrdObj** pThreadObj;
+  SWorkThrd** pThreadObj;
 
   uv_pipe_t   pipeListen;
   uv_pipe_t** pipe;
@@ -133,13 +133,13 @@ static SSvrConn* createConn(void* hThrd);
 static void      destroyConn(SSvrConn* conn, bool clear /*clear handle or not*/);
 static void      destroyConnRegArg(SSvrConn* conn);
 
-static int reallocConnRefHandle(SSvrConn* conn);
+static int reallocConnRef(SSvrConn* conn);
 
-static void uvHandleQuit(SSvrMsg* msg, SWorkThrdObj* thrd);
+static void uvHandleQuit(SSvrMsg* msg, SWorkThrd* thrd);
-static void uvHandleRelease(SSvrMsg* msg, SWorkThrdObj* thrd);
+static void uvHandleRelease(SSvrMsg* msg, SWorkThrd* thrd);
-static void uvHandleResp(SSvrMsg* msg, SWorkThrdObj* thrd);
+static void uvHandleResp(SSvrMsg* msg, SWorkThrd* thrd);
-static void uvHandleRegister(SSvrMsg* msg, SWorkThrdObj* thrd);
+static void uvHandleRegister(SSvrMsg* msg, SWorkThrd* thrd);
-static void (*transAsyncHandle[])(SSvrMsg* msg, SWorkThrdObj* thrd) = {uvHandleResp, uvHandleQuit, uvHandleRelease,
+static void (*transAsyncHandle[])(SSvrMsg* msg, SWorkThrd* thrd) = {uvHandleResp, uvHandleQuit, uvHandleRelease,
                                                                     uvHandleRegister, NULL};
 
 static int32_t exHandlesMgt;
@@ -160,7 +160,7 @@ static void* transWorkerThread(void* arg);
 static void* transAcceptThread(void* arg);
 
 // add handle loop
-static bool addHandleToWorkloop(SWorkThrdObj* pThrd, char* pipeName);
+static bool addHandleToWorkloop(SWorkThrd* pThrd, char* pipeName);
 static bool addHandleToAcceptloop(void* arg);
 
 #define CONN_SHOULD_RELEASE(conn, head)                                               \
@@ -176,7 +176,7 @@ static bool addHandleToAcceptloop(void* arg);
       srvMsg->msg = tmsg;                                                             \
       srvMsg->type = Release;                                                         \
       srvMsg->pConn = conn;                                                           \
-      reallocConnRefHandle(conn);                                                     \
+      reallocConnRef(conn);                                                           \
       if (!transQueuePush(&conn->srvMsgs, srvMsg)) {                                  \
         return;                                                                       \
       }                                                                               \
@@ -206,32 +206,6 @@ static bool addHandleToAcceptloop(void* arg);
     }                                                   \
   } while (0)
 
-#define ASYNC_CHECK_HANDLE(exh1, refId)                                                                               \
-  do {                                                                                                                \
-    if (refId > 0) {                                                                                                  \
-      tTrace("handle step1");                                                                                         \
-      SExHandle* exh2 = transAcquireExHandle(refMgt, refId);                                                          \
-      if (exh2 == NULL || refId != exh2->refId) {                                                                     \
-        tTrace("handle %p except, may already freed, ignore msg, ref1: %" PRIu64 ", ref2 : %" PRIu64 "", exh1,        \
-               exh2 ? exh2->refId : 0, refId);                                                                        \
-        goto _return1;                                                                                                \
-      }                                                                                                               \
-    } else if (refId == 0) {                                                                                          \
-      tTrace("handle step2");                                                                                         \
-      SExHandle* exh2 = transAcquireExHandle(refMgt, refId);                                                          \
-      if (exh2 == NULL || refId != exh2->refId) {                                                                     \
-        tTrace("handle %p except, may already freed, ignore msg, ref1: %" PRIu64 ", ref2 : %" PRIu64 "", exh1, refId, \
-               exh2 ? exh2->refId : 0);                                                                               \
-        goto _return1;                                                                                                \
-      } else {                                                                                                        \
-        refId = exh1->refId;                                                                                          \
-      }                                                                                                               \
-    } else if (refId < 0) {                                                                                           \
-      tTrace("handle step3");                                                                                         \
-      goto _return2;                                                                                                  \
-    }                                                                                                                 \
-  } while (0)
-
 void uvAllocRecvBufferCb(uv_handle_t* handle, size_t suggested_size, uv_buf_t* buf) {
   SSvrConn*    conn = handle->data;
   SConnBuffer* pBuf = &conn->readBuf;
@@ -259,7 +233,7 @@ static void uvHandleReq(SSvrConn* pConn) {
   //  wreq->data = pConn;
   //  uv_read_stop((uv_stream_t*)pConn->pTcp);
   //  transRefSrvHandle(pConn);
-  //  uv_queue_work(((SWorkThrdObj*)pConn->hostThrd)->loop, wreq, uvWorkDoTask, uvWorkAfterTask);
+  //  uv_queue_work(((SWorkThrd*)pConn->hostThrd)->loop, wreq, uvWorkDoTask, uvWorkAfterTask);
 
   CONN_SHOULD_RELEASE(pConn, pHead);
 
@@ -379,7 +353,7 @@ void uvOnSendCb(uv_write_t* req, int status) {
       // if (msg->type == Release && conn->status != ConnNormal) {
       //  conn->status = ConnNormal;
       //  transUnrefSrvHandle(conn);
-      //  reallocConnRefHandle(conn);
+      //  reallocConnRef(conn);
       //  destroySmsg(msg);
       //  transQueueClear(&conn->srvMsgs);
       //  return;
@@ -448,6 +422,8 @@ static void uvPrepareSendData(SSvrMsg* smsg, uv_buf_t* wb) {
       transUnrefSrvHandle(pConn);
     } else {
       pHead->msgType = pMsg->msgType;
+      if (pHead->msgType == 0 && transMsgLenFromCont(pMsg->contLen) == sizeof(STransMsgHead))
+        pHead->msgType = pConn->inType + 1;
     }
   }
 
@@ -504,7 +480,7 @@ static void destroySmsg(SSvrMsg* smsg) {
   transFreeMsg(smsg->msg.pCont);
   taosMemoryFree(smsg);
 }
-static void destroyAllConn(SWorkThrdObj* pThrd) {
+static void destroyAllConn(SWorkThrd* pThrd) {
   tTrace("thread %p destroy all conn ", pThrd);
   while (!QUEUE_IS_EMPTY(&pThrd->conn)) {
     queue* h = QUEUE_HEAD(&pThrd->conn);
@@ -520,7 +496,7 @@ static void destroyAllConn(SWorkThrdObj* pThrd) {
 }
 void uvWorkerAsyncCb(uv_async_t* handle) {
   SAsyncItem* item = handle->data;
-  SWorkThrdObj* pThrd = item->pThrd;
+  SWorkThrd*  pThrd = item->pThrd;
   SSvrConn*   conn = NULL;
   queue       wq;
 
@@ -650,7 +626,7 @@ void uvOnConnectionCb(uv_stream_t* q, ssize_t nread, const uv_buf_t* buf) {
   assert(buf->base[0] == notify[0]);
   taosMemoryFree(buf->base);
 
-  SWorkThrdObj* pThrd = q->data;
+  SWorkThrd* pThrd = q->data;
 
   uv_pipe_t* pipe = (uv_pipe_t*)q;
   if (!uv_pipe_pending_count(pipe)) {
@@ -718,10 +694,10 @@ void uvOnPipeConnectionCb(uv_connect_t* connect, int status) {
   if (status != 0) {
     return;
   }
-  SWorkThrdObj* pThrd = container_of(connect, SWorkThrdObj, connect_req);
+  SWorkThrd* pThrd = container_of(connect, SWorkThrd, connect_req);
   uv_read_start((uv_stream_t*)pThrd->pipe, uvAllocConnBufferCb, uvOnConnectionCb);
 }
-static bool addHandleToWorkloop(SWorkThrdObj* pThrd, char* pipeName) {
+static bool addHandleToWorkloop(SWorkThrd* pThrd, char* pipeName) {
   pThrd->loop = (uv_loop_t*)taosMemoryMalloc(sizeof(uv_loop_t));
   if (0 != uv_loop_init(pThrd->loop)) {
     return false;
@@ -774,14 +750,14 @@ static bool addHandleToAcceptloop(void* arg) {
 }
 void* transWorkerThread(void* arg) {
   setThreadName("trans-worker");
-  SWorkThrdObj* pThrd = (SWorkThrdObj*)arg;
+  SWorkThrd* pThrd = (SWorkThrd*)arg;
   uv_run(pThrd->loop, UV_RUN_DEFAULT);
 
   return NULL;
 }
 
 static SSvrConn* createConn(void* hThrd) {
-  SWorkThrdObj* pThrd = hThrd;
+  SWorkThrd* pThrd = hThrd;
 
   SSvrConn* pConn = (SSvrConn*)taosMemoryCalloc(1, sizeof(SSvrConn));
   QUEUE_INIT(&pConn->queue);
@@ -826,7 +802,7 @@ static void destroyConnRegArg(SSvrConn* conn) {
     conn->regArg.init = 0;
   }
 }
-static int reallocConnRefHandle(SSvrConn* conn) {
+static int reallocConnRef(SSvrConn* conn) {
   transReleaseExHandle(refMgt, conn->refId);
   transRemoveExHandle(refMgt, conn->refId);
   // avoid app continue to send msg on invalid handle
@@ -844,7 +820,7 @@ static void uvDestroyConn(uv_handle_t* handle) {
   if (conn == NULL) {
     return;
   }
-  SWorkThrdObj* thrd = conn->hostThrd;
+  SWorkThrd* thrd = conn->hostThrd;
 
   transReleaseExHandle(refMgt, conn->refId);
   transRemoveExHandle(refMgt, conn->refId);
@@ -889,7 +865,7 @@ void* transInitServer(uint32_t ip, uint32_t port, char* label, int numOfThreads,
   srv->numOfThreads = numOfThreads;
   srv->workerIdx = 0;
   srv->numOfWorkerReady = 0;
-  srv->pThreadObj = (SWorkThrdObj**)taosMemoryCalloc(srv->numOfThreads, sizeof(SWorkThrdObj*));
+  srv->pThreadObj = (SWorkThrd**)taosMemoryCalloc(srv->numOfThreads, sizeof(SWorkThrd*));
   srv->pipe = (uv_pipe_t**)taosMemoryCalloc(srv->numOfThreads, sizeof(uv_pipe_t*));
   srv->ip = ip;
   srv->port = port;
@@ -914,7 +890,7 @@ void* transInitServer(uint32_t ip, uint32_t port, char* label, int numOfThreads,
   assert(0 == uv_listen((uv_stream_t*)&srv->pipeListen, SOMAXCONN, uvPipeListenCb));
 
   for (int i = 0; i < srv->numOfThreads; i++) {
-    SWorkThrdObj* thrd = (SWorkThrdObj*)taosMemoryCalloc(1, sizeof(SWorkThrdObj));
+    SWorkThrd* thrd = (SWorkThrd*)taosMemoryCalloc(1, sizeof(SWorkThrd));
     thrd->pTransInst = shandle;
     thrd->quit = false;
     srv->pThreadObj[i] = thrd;
@@ -959,7 +935,7 @@ End:
   return NULL;
 }
 
-void uvHandleQuit(SSvrMsg* msg, SWorkThrdObj* thrd) {
+void uvHandleQuit(SSvrMsg* msg, SWorkThrd* thrd) {
   thrd->quit = true;
   if (QUEUE_IS_EMPTY(&thrd->conn)) {
     uv_walk(thrd->loop, uvWalkCb, NULL);
@@ -968,10 +944,10 @@ void uvHandleQuit(SSvrMsg* msg, SWorkThrdObj* thrd) {
   }
   taosMemoryFree(msg);
 }
-void uvHandleRelease(SSvrMsg* msg, SWorkThrdObj* thrd) {
+void uvHandleRelease(SSvrMsg* msg, SWorkThrd* thrd) {
   SSvrConn* conn = msg->pConn;
   if (conn->status == ConnAcquire) {
-    reallocConnRefHandle(conn);
+    reallocConnRef(conn);
     if (!transQueuePush(&conn->srvMsgs, msg)) {
       return;
     }
@@ -982,12 +958,12 @@ void uvHandleRelease(SSvrMsg* msg, SWorkThrdObj* thrd) {
   }
   destroySmsg(msg);
 }
-void uvHandleResp(SSvrMsg* msg, SWorkThrdObj* thrd) {
+void uvHandleResp(SSvrMsg* msg, SWorkThrd* thrd) {
   // send msg to client
   tDebug("%s conn %p start to send resp (2/2)", transLabel(thrd->pTransInst), msg->pConn);
   uvStartSendResp(msg);
 }
-void uvHandleRegister(SSvrMsg* msg, SWorkThrdObj* thrd) {
+void uvHandleRegister(SSvrMsg* msg, SWorkThrd* thrd) {
   SSvrConn* conn = msg->pConn;
   tDebug("%s conn %p register brokenlink callback", transLabel(thrd->pTransInst), conn);
   if (conn->status == ConnAcquire) {
@@ -1008,7 +984,7 @@ void uvHandleRegister(SSvrMsg* msg, SWorkThrdObj* thrd) {
     taosMemoryFree(msg);
   }
 }
-void destroyWorkThrd(SWorkThrdObj* pThrd) {
+void destroyWorkThrd(SWorkThrd* pThrd) {
   if (pThrd == NULL) {
     return;
   }
@@ -1019,7 +995,7 @@ void destroyWorkThrd(SWorkThrdObj* pThrd) {
   taosMemoryFree(pThrd->loop);
   taosMemoryFree(pThrd);
 }
-void sendQuitToWorkThrd(SWorkThrdObj* pThrd) {
+void sendQuitToWorkThrd(SWorkThrd* pThrd) {
   SSvrMsg* msg = taosMemoryCalloc(1, sizeof(SSvrMsg));
   msg->type = Quit;
   tDebug("server send quit msg to work thread");
@@ -1055,8 +1031,6 @@ void transCloseServer(void* arg) {
 
   int ref = atomic_sub_fetch_32(&tranSSvrInst, 1);
   if (ref == 0) {
-    // TdThreadOnce tmpInit = PTHREAD_ONCE_INIT;
-    // memcpy(&transModuleInit, &tmpInit, sizeof(TdThreadOnce));
     transCloseExHandleMgt(refMgt);
   }
 }
@@ -1086,7 +1060,7 @@ void transReleaseSrvHandle(void* handle) {
 
   ASYNC_CHECK_HANDLE(exh, refId);
 
-  SWorkThrdObj* pThrd = exh->pThrd;
+  SWorkThrd* pThrd = exh->pThrd;
   ASYNC_ERR_JRET(pThrd);
 
   STransMsg tmsg = {.code = 0, .info.handle = exh, .info.ahandle = NULL, .info.refId = refId};
@@ -1116,7 +1090,7 @@ void transSendResponse(const STransMsg* msg) {
   STransMsg tmsg = *msg;
   tmsg.info.refId = refId;
 
-  SWorkThrdObj* pThrd = exh->pThrd;
+  SWorkThrd* pThrd = exh->pThrd;
   ASYNC_ERR_JRET(pThrd);
 
   SSvrMsg* m = taosMemoryCalloc(1, sizeof(SSvrMsg));
@@ -1146,7 +1120,7 @@ void transRegisterMsg(const STransMsg* msg) {
   STransMsg tmsg = *msg;
   tmsg.info.refId = refId;
 
-  SWorkThrdObj* pThrd = exh->pThrd;
+  SWorkThrd* pThrd = exh->pThrd;
   ASYNC_ERR_JRET(pThrd);
 
   SSvrMsg* m = taosMemoryCalloc(1, sizeof(SSvrMsg));

source/libs/wal/src/walRead.c

@@ -104,6 +104,8 @@ static int32_t walReadChangeFile(SWalReadHandle *pRead, int64_t fileFirstVer) {
     return -1;
   }
 
+  pRead->pReadLogTFile = pLogTFile;
+
   walBuildIdxName(pRead->pWal, fileFirstVer, fnameStr);
   TdFilePtr pIdxTFile = taosOpenFile(fnameStr, TD_FILE_READ);
   if (pIdxTFile == NULL) {
@@ -112,7 +114,6 @@ static int32_t walReadChangeFile(SWalReadHandle *pRead, int64_t fileFirstVer) {
     return -1;
   }
 
-  pRead->pReadLogTFile = pLogTFile;
   pRead->pReadIdxTFile = pIdxTFile;
   return 0;
 }

source/libs/wal/src/walWrite.c

@@ -42,6 +42,9 @@ int32_t walRestoreFromSnapshot(SWal *pWal, int64_t ver) {
       char          fnameStr[WAL_FILE_LEN];
       walBuildLogName(pWal, pFileInfo->firstVer, fnameStr);
       taosRemoveFile(fnameStr);
+
+      walBuildIdxName(pWal, pFileInfo->firstVer, fnameStr);
+      taosRemoveFile(fnameStr);
     }
   }
   walRemoveMeta(pWal);
@@ -105,7 +108,7 @@ int32_t walRollback(SWal *pWal, int64_t ver) {
   }
 
   walBuildIdxName(pWal, walGetCurFileFirstVer(pWal), fnameStr);
-  TdFilePtr pIdxTFile = taosOpenFile(fnameStr, TD_FILE_WRITE | TD_FILE_READ);
+  TdFilePtr pIdxTFile = taosOpenFile(fnameStr, TD_FILE_WRITE | TD_FILE_READ | TD_FILE_APPEND);
 
   if (pIdxTFile == NULL) {
     taosThreadMutexUnlock(&pWal->mutex);
@@ -126,7 +129,7 @@ int32_t walRollback(SWal *pWal, int64_t ver) {
   ASSERT(entry.ver == ver);
 
   walBuildLogName(pWal, walGetCurFileFirstVer(pWal), fnameStr);
-  TdFilePtr pLogTFile = taosOpenFile(fnameStr, TD_FILE_WRITE | TD_FILE_READ);
+  TdFilePtr pLogTFile = taosOpenFile(fnameStr, TD_FILE_WRITE | TD_FILE_READ | TD_FILE_APPEND);
   if (pLogTFile == NULL) {
     // TODO
     taosThreadMutexUnlock(&pWal->mutex);
@@ -307,8 +310,8 @@ int walRoll(SWal *pWal) {
 
 static int walWriteIndex(SWal *pWal, int64_t ver, int64_t offset) {
   SWalIdxEntry entry = {.ver = ver, .offset = offset};
-  /*int64_t      idxOffset = taosLSeekFile(pWal->pWriteIdxTFile, 0, SEEK_CUR);*/
+  int64_t      idxOffset = taosLSeekFile(pWal->pWriteIdxTFile, 0, SEEK_END);
-  /*wDebug("write index: ver: %ld, offset: %ld, at %ld", ver, offset, idxOffset);*/
+  wDebug("write index: ver: %ld, offset: %ld, at %ld", ver, offset, idxOffset);
   int64_t size = taosWriteFile(pWal->pWriteIdxTFile, &entry, sizeof(SWalIdxEntry));
   if (size != sizeof(SWalIdxEntry)) {
     terrno = TAOS_SYSTEM_ERROR(errno);

source/os/src/osFile.c

@@ -671,7 +671,7 @@ void taosFprintfFile(TdFilePtr pFile, const char *format, ...) {
   fflush(pFile->fp);
 }
 
-bool taosValidFile(TdFilePtr pFile) { return pFile != NULL; }
+bool taosValidFile(TdFilePtr pFile) { return pFile != NULL && pFile->fd > 0; }
 
 int32_t taosUmaskFile(int32_t maskVal) {
 #ifdef WINDOWS

source/os/src/osSocket.c

@@ -946,7 +946,7 @@ int32_t taosGetFqdn(char *fqdn) {
 #endif  // __APPLE__
   int32_t ret = getaddrinfo(hostname, NULL, &hints, &result);
   if (!result) {
-    fprintf(stderr,"failed to get fqdn, code:%d, reason:%s", ret, gai_strerror(ret));
+    fprintf(stderr, "failed to get fqdn, code:%d, reason:%s", ret, gai_strerror(ret));
     return -1;
   }
 
@@ -1073,7 +1073,7 @@ int32_t taosCloseEpoll(TdEpollPtr *ppEpoll) {
  * Set TCP connection timeout per-socket level.
  * ref [https://github.com/libuv/help/issues/54]
  */
-int taosCreateSocketWithTimeOutOpt(uint32_t conn_timeout_sec) {
+int32_t taosCreateSocketWithTimeout(uint32_t timeout) {
 #if defined(WINDOWS)
   SOCKET fd;
 #else
@@ -1083,11 +1083,11 @@ int taosCreateSocketWithTimeOutOpt(uint32_t conn_timeout_sec) {
     return -1;
   }
 #if defined(WINDOWS)
-  if (0 != setsockopt(fd, IPPROTO_TCP, TCP_MAXRT, (char *)&conn_timeout_sec, sizeof(conn_timeout_sec))) {
+  if (0 != setsockopt(fd, IPPROTO_TCP, TCP_MAXRT, (char *)&timeout, sizeof(timeout))) {
     return -1;
   }
 #else  // Linux like systems
-  uint32_t conn_timeout_ms = conn_timeout_sec * 1000;
+  uint32_t conn_timeout_ms = timeout * 1000;
   if (0 != setsockopt(fd, IPPROTO_TCP, TCP_USER_TIMEOUT, (char *)&conn_timeout_ms, sizeof(conn_timeout_ms))) {
     return -1;
   }

tests/pytest/util/cluster.py

@@ -24,7 +24,7 @@ class ClusterDnodes(TDDnodes):
 
 
 class ConfigureyCluster:
-    """This will create defined number of dnodes and create a cluset.
+    """This will create defined number of dnodes and create a cluster.
        at the same time, it will return TDDnodes list:  dnodes, """
     hostname= socket.gethostname()
 
@@ -85,8 +85,8 @@ class ConfigureyCluster:
             count+=1
             time.sleep(1)
         else:
-            tdLog.debug("create cluster with %d dnode but  check dnode not ready within 5s ! "%self.dnodeNums)
+            tdLog.exit("create cluster with %d dnode but  check dnode not ready within 5s ! "%self.dnodeNums)
-            return -1
+            
 
 
 cluster = ConfigureyCluster()

tests/pytest/util/dnodes.py

@@ -508,7 +508,6 @@ class TDDnode:
 
     def stoptaosd(self):
         if (not self.remoteIP == ""):
-            print("123")
             self.remoteExec(self.cfgDict, "tdDnodes.dnodes[%d].running=1\ntdDnodes.dnodes[%d].stop()"%(self.index-1,self.index-1))
             tdLog.info("stop dnode%d"%self.index)
             return
@@ -518,6 +517,9 @@ class TDDnode:
             toBeKilled = "valgrind.bin"
 
         if self.running != 0:
+            if platform.system().lower() == 'windows':
+                os.system("wmic process where \"name='taosd.exe' and CommandLine like '%%dnode%d%%'\" get processId | xargs echo | awk '{print $2}' | xargs taskkill -f -pid"%self.index)
+            else:
                 psCmd = "ps -ef|grep -w %s| grep dnode%d|grep -v grep | awk '{print $2}'" % (toBeKilled,self.index)
                 processID = subprocess.check_output(
                     psCmd, shell=True).decode("utf-8")

tests/script/test-all.bat

@@ -0,0 +1,58 @@
+@echo off
+SETLOCAL EnableDelayedExpansion
+for /F "tokens=1,2 delims=#" %%a in ('"prompt #$H#$E# & echo on & for %%b in (1) do     rem"') do (  set "DEL=%%a")
+set /a a=0
+echo Windows Taosd Full Test
+set /a exitNum=0
+rm -rf failed.txt
+set caseFile="jenkins\\basic.txt"
+if not "%2" == "" (
+    set caseFile="%2"
+)
+for /F "usebackq tokens=*" %%i in (!caseFile!) do (
+    set line=%%i
+    if "!line:~,9!" == "./test.sh" (
+        set /a a+=1
+        echo !a! Processing %%i
+        call :GetTimeSeconds !time!
+        set time1=!_timeTemp!
+        echo Start at !time!
+        call !line:./test.sh=wtest.bat! > result_!a!.txt 2>error_!a!.txt
+        if errorlevel 1 ( call :colorEcho 0c "failed" &echo. && set /a exitNum=8 && echo %%i >>failed.txt ) else ( call :colorEcho 0a "Success" &echo. )
+    )
+)
+exit !exitNum!
+
+:colorEcho
+set timeNow=%time%
+call :GetTimeSeconds %timeNow%
+set time2=%_timeTemp%
+set /a interTime=%time2% - %time1%
+echo End at %timeNow% , cast %interTime%s
+echo off
+<nul set /p ".=%DEL%" > "%~2"
+findstr /v /a:%1 /R "^$" "%~2" nul
+del "%~2" > nul 2>&1i
+goto :eof
+
+:GetTimeSeconds
+set tt=%1
+set tt=%tt:.= %
+set tt=%tt::= %
+set tt=%tt: 0= %
+set /a index=1
+for %%a in (%tt%) do (
+    if !index! EQU 1 (
+        set /a hh=%%a
+    )^
+    else if  !index! EQU 2 (
+        set /a mm=%%a
+ 
+    )^
+    else if  !index! EQU 3 (
+        set /a ss=%%a
+    )
+   set /a index=index+1
+)
+set /a _timeTemp=(%hh%*60+%mm%)*60+%ss%
+goto :eof

tests/script/tsim/stream/distributeInterval0.sim

@@ -10,6 +10,30 @@ sql create dnode $hostname2 port 7200
 
 system sh/exec.sh -n dnode2 -s start
 
+print ===== step1
+$x = 0
+step1:
+	$x = $x + 1
+	sleep 1000
+	if $x == 10 then
+	  print ====> dnode not ready!
+		return -1
+	endi
+sql show dnodes
+print ===> $data00 $data01 $data02 $data03 $data04 $data05
+print ===> $data10 $data11 $data12 $data13 $data14 $data15
+if $rows != 2 then
+  return -1
+endi
+if $data(1)[4] != ready then
+  goto step1
+endi
+if $data(2)[4] != ready then
+  goto step1
+endi
+
+print ===== step2
+
 sql create database test  vgroups 4;
 sql use test;
 sql create stable st(ts timestamp, a int, b int , c int, d double) tags(ta int,tb int,tc int);
@@ -80,17 +104,17 @@ endi
 
 if $data03 != 4 then
   print ======$data03
-  return -1
+  goto loop1
 endi
 
 if $data04 != 52 then
   print ======$data04
-  return -1
+  goto loop1
 endi
 
 if $data05 != 13 then
   print ======$data05
-  return -1
+  goto loop1
 endi
 
 # row 1
@@ -179,7 +203,7 @@ sql use test1;
 sql create stable st(ts timestamp, a int, b int , c int) tags(ta int,tb int,tc int);
 sql create table ts1 using st tags(1,1,1);
 sql create table ts2 using st tags(2,2,2);
-sql create stream stream_t2 trigger at_once  into streamtST1 as select  _wstartts, count(*) c1, count(a) c2 , sum(a) c3 ,  max(b)  c5, min(c) c6 from st interval(10s) ;
+sql create stream stream_t2 trigger at_once  watermark 20s into streamtST1 as select  _wstartts, count(*) c1, count(a) c2 , sum(a) c3 ,  max(b)  c5, min(c) c6 from st interval(10s) ;
 
 sql insert into ts1 values(1648791211000,1,2,3);
 sql insert into ts1 values(1648791222001,2,2,3);

tests/script/tsim/stream/partitionby.sim

@@ -74,7 +74,7 @@ sql create stable st(ts timestamp,a int,b int,c int,id int) tags(ta int,tb int,t
 sql create table ts1 using st tags(1,1,1);
 sql create table ts2 using st tags(2,2,2);
 
-sql create stream stream_t2 trigger at_once  into streamtST as select  _wstartts, count(*) c1, count(a) c2 , sum(a) c3 ,  max(b)  c5, min(c) c6, max(id) c7 from st partition by ta interval(10s) ;
+sql create stream stream_t2 trigger at_once  watermark 20s into streamtST as select  _wstartts, count(*) c1, count(a) c2 , sum(a) c3 ,  max(b)  c5, min(c) c6, max(id) c7 from st partition by ta interval(10s) ;
 sql insert into ts1 values(1648791211000,1,2,3,1);
 sql insert into ts1 values(1648791222001,2,2,3,2);
 sql insert into ts2 values(1648791211000,1,2,3,3);
@@ -83,16 +83,16 @@ sql insert into ts2 values(1648791222001,2,2,3,4);
 sql insert into ts2 values(1648791222002,2,2,3,5);
 sql insert into ts2 values(1648791222002,2,2,3,6);
 
-sql insert into ts1 values(1648791211000,1,2,3,1);
+sql insert into ts1 values(1648791211000,1,2,3,7);
-sql insert into ts1 values(1648791222001,2,2,3,2);
+sql insert into ts1 values(1648791222001,2,2,3,8);
-sql insert into ts2 values(1648791211000,1,2,3,3);
+sql insert into ts2 values(1648791211000,1,2,3,9);
-sql insert into ts2 values(1648791222001,2,2,3,4);
+sql insert into ts2 values(1648791222001,2,2,3,10);
 
 $loop_count = 0
 
 loop2:
 sleep 300
-sql select * from streamtST;
+sql select * from streamtST order by c7 asc;
 
 $loop_count = $loop_count + 1
 if $loop_count == 10 then
@@ -104,8 +104,18 @@ print =====data01=$data01
 goto loop2
 endi
 
-if $data02 != 1 then
+if $data11 != 1 then
-print =====data02=$data02
+print =====data11=$data11
+goto loop2
+endi
+
+if $data21 != 1 then
+print =====data21=$data21
+goto loop2
+endi
+
+if $data31 != 2 then
+print =====data31=$data31
 goto loop2
 endi
 
@@ -114,8 +124,18 @@ print =====data03=$data03
 goto loop2
 endi
 
-if $data04 != 2 then
+if $data13 != 2 then
-print =====data04=$data04
+print =====data13=$data13
+goto loop2
+endi
+
+if $data23 != 1 then
+print =====data23=$data23
+goto loop2
+endi
+
+if $data33 != 4 then
+print =====data33=$data33
 goto loop2
 endi
 

tests/script/tsim/stream/schedSnode.sim

@@ -79,17 +79,17 @@ endi
 
 if $data03 != 4 then
   print ======$data03
-  return -1
+  goto loop1
 endi
 
 if $data04 != 52 then
   print ======$data04
-  return -1
+  goto loop1
 endi
 
 if $data05 != 13 then
   print ======$data05
-  return -1
+  goto loop1
 endi
 
 # row 1

tests/script/tsim/user/privilege_sysinfo.sim

@@ -8,14 +8,12 @@ sql create user sysinfo0 pass 'taosdata'
 sql create user sysinfo1 pass 'taosdata'
 sql alter user sysinfo0 sysinfo 0
 sql alter user sysinfo1 sysinfo 1
+sql create database db
 
 print user sysinfo0 login
 sql close
 sql connect sysinfo0
 
-system sh/exec.sh -n dnode1 -s stop
-return
-
 print =============== check oper
 sql_error create user u1 pass 'u1'
 sql_error drop user sysinfo1
@@ -42,6 +40,24 @@ sql_error kill connection 1
 sql_error kill query 1
 
 print =============== check db
-sql_error create database db
+sql_error create database d1
+sql_error drop database db
+sql_error use db
+sql_error alter database db replica 1;
+sql_error show db.vgroups
+sql_error show db.stables
+sql_error show db.tables
+
+print =============== check show
+sql_error show users
+sql_error show cluster
+sql_error show dnodes
+sql_error show mnodes
+sql_error show snodes
+sql_error show qnodes
+sql_error show bnodes
+sql_error show grants
+sql_error show dnode 1 variables;
+sql show variables;
 
 system sh/exec.sh -n dnode1 -s stop -x SIGINT

tests/system-test/1-insert/table_param_ttl.py

@@ -0,0 +1,79 @@
+###################################################################
+#           Copyright (c) 2016 by TAOS Technologies, Inc.
+#                     All rights reserved.
+#
+#  This file is proprietary and confidential to TAOS Technologies.
+#  No part of this file may be reproduced, stored, transmitted,
+#  disclosed or used in any form or by any means other than as
+#  expressly provided by the written permission from Jianhui Tao
+#
+###################################################################
+
+# -*- coding: utf-8 -*-
+
+from util.log import *
+from util.cases import *
+from util.sql import *
+from util.common import *
+
+class TDTestCase:
+    updatecfgDict = {'ttlUnit':5,'ttlPushInterval':3}
+    def init(self, conn, logSql):
+        tdLog.debug("start to execute %s" % __file__)
+        tdSql.init(conn.cursor())      
+        self.ntbname = 'ntb'
+        self.stbname = 'stb'
+        self.tbnum = 10
+        self.ttl_param = 1
+        self.default_ttl = 100
+        self.modify_ttl = 1
+    def ttl_check_ntb(self):
+        tdSql.prepare()
+
+        for i in range(self.tbnum):
+            tdSql.execute(f'create table {self.ntbname}_{i} (ts timestamp,c0 int) ttl {self.ttl_param}')
+        tdSql.query(f'show tables')
+        tdSql.checkRows(self.tbnum)
+        sleep(self.updatecfgDict['ttlUnit']*self.ttl_param+self.updatecfgDict['ttlPushInterval'])
+        tdSql.query(f'show tables')
+        tdSql.checkRows(0)
+        for i in range(self.tbnum):
+            tdSql.execute(f'create table {self.ntbname}_{i} (ts timestamp,c0 int) ttl {self.default_ttl}')
+        for i in range(int(self.tbnum/2)):
+            tdSql.execute(f'alter table {self.ntbname}_{i} ttl {self.modify_ttl}')
+        sleep(self.updatecfgDict['ttlUnit']*self.modify_ttl+self.updatecfgDict['ttlPushInterval'])
+        tdSql.query(f'show tables')
+        tdSql.checkRows(self.tbnum - int(self.tbnum/2))
+        tdSql.execute('drop database db')
+    def ttl_check_ctb(self):
+        tdSql.prepare()
+        tdSql.execute(f'create table {self.stbname} (ts timestamp,c0 int) tags(t0 int)')
+
+        for i in range(self.tbnum):
+            tdSql.execute(f'create table {self.stbname}_{i} using {self.stbname} tags({i}) ttl {self.ttl_param}')
+        tdSql.query(f'show tables')
+        tdSql.checkRows(self.tbnum)
+        sleep(self.updatecfgDict['ttlUnit']*self.ttl_param+self.updatecfgDict['ttlPushInterval'])
+        tdSql.query(f'show tables')
+        tdSql.checkRows(0)
+        for i in range(self.tbnum):
+            tdSql.execute(f'create table {self.stbname}_{i} using {self.stbname} tags({i}) ttl {self.default_ttl}')
+        tdSql.query(f'show tables')
+        tdSql.checkRows(self.tbnum)
+        for i in range(int(self.tbnum/2)):
+            tdSql.execute(f'alter table {self.stbname}_{i} ttl {self.modify_ttl}')
+        sleep(self.updatecfgDict['ttlUnit']*self.modify_ttl+self.updatecfgDict['ttlPushInterval'])
+        tdSql.query(f'show tables')
+        tdSql.checkRows(self.tbnum - int(self.tbnum/2))
+        tdSql.execute('drop database db')
+
+    def run(self):
+        self.ttl_check_ntb()
+        self.ttl_check_ctb()
+
+    def stop(self):
+        tdSql.close()
+        tdLog.success("%s successfully executed" % __file__)
+
+tdCases.addWindows(__file__, TDTestCase())
+tdCases.addLinux(__file__, TDTestCase())

tests/system-test/2-query/bottom.py

@@ -17,99 +17,139 @@ from util.log import *
 from util.cases import *
 from util.sql import *
 from util.common import *
-
+from util.sqlset import *
 
 
 class TDTestCase:
     def init(self, conn, logSql):
         tdLog.debug("start to execute %s" % __file__)
         tdSql.init(conn.cursor())
-
+        self.dbname = 'db_test'
+        self.setsql = TDSetSql()
+        self.ntbname = 'ntb'
         self.rowNum = 10
         self.tbnum = 20
         self.ts = 1537146000000
         self.binary_str = 'taosdata'
         self.nchar_str = '涛思数据'
-    def bottom_check_base(self):
+        self.column_dict = {
-        tdSql.prepare()
+            'ts'  : 'timestamp',
-        tdSql.execute('''create table stb(ts timestamp, col1 tinyint, col2 smallint, col3 int, col4 bigint, col5 tinyint unsigned, col6 smallint unsigned, 
+            'col1': 'tinyint',
-                    col7 int unsigned, col8 bigint unsigned, col9 float, col10 double, col11 bool, col12 binary(20), col13 nchar(20)) tags(loc nchar(20))''')
+            'col2': 'smallint',
-        tdSql.execute("create table stb_1 using stb tags('beijing')")
+            'col3': 'int',
-        column_list = ['col1','col2','col3','col4','col5','col6','col7','col8']
+            'col4': 'bigint',
-        error_column_list = ['col11','col12','col13']
+            'col5': 'tinyint unsigned',
-        error_param_list = [0,101]
+            'col6': 'smallint unsigned',
+            'col7': 'int unsigned',
+            'col8': 'bigint unsigned',
+            'col9': 'float',
+            'col10': 'double',
+            'col11': 'bool',
+            'col12': 'binary(20)',
+            'col13': 'nchar(20)'
+        }
+        
+        self.param_list = [1,100]
+    def insert_data(self,column_dict,tbname,row_num):
+        sql = ''
+        for k, v in column_dict.items():
+            if v.lower() == 'timestamp' or v.lower() == 'tinyint' or v.lower() == 'smallint' or v.lower() == 'int' or v.lower() == 'bigint' or \
+            v.lower() == 'tinyint unsigned' or v.lower() == 'smallint unsigned' or v.lower() == 'int unsigned' or v.lower() == 'bigint unsigned' or v.lower() == 'bool':
+                sql += '%d,'
+            elif v.lower() == 'float' or v.lower() == 'double':
+                sql += '%f,'
+            elif 'binary' in v.lower():
+                sql += f'"{self.binary_str}%d",'
+            elif 'nchar' in v.lower():
+                sql += f'"{self.nchar_str}%d",'
+        insert_sql = f'insert into {tbname} values({sql[:-1]})'
+        for i in range(row_num):
+            insert_list = []
+            for k, v in column_dict.items():
+                if v.lower() in[ 'tinyint' , 'smallint' , 'int', 'bigint' , 'tinyint unsigned' , 'smallint unsigned' , 'int unsigned' , 'bigint unsigned'] or\
+                'binary' in v.lower() or 'nchar' in v.lower():
+                    insert_list.append(0 + i)
+                elif v.lower() == 'float' or v.lower() == 'double':
+                    insert_list.append(0.1 + i)
+                elif v.lower() == 'bool':
+                    insert_list.append(i % 2)
+                elif v.lower() == 'timestamp':
+                    insert_list.append(self.ts + i)
+            tdSql.execute(insert_sql%(tuple(insert_list)))
+    def bottom_check_data(self,tbname,tb_type):
+        new_column_dict = {}
+        for param in self.param_list:
+            for k,v in self.column_dict.items():
+                if v.lower() in ['tinyint','smallint','int','bigint','tinyint unsigned','smallint unsigned','int unsigned','bigint unsigned']:
+                    tdSql.query(f'select bottom({k},{param}) from {tbname} order by {k}')
+                    if param >= self.rowNum:
+                        if tb_type in ['normal_table','child_table']:
+                            tdSql.checkRows(self.rowNum)
+                            values_list = []
                             for i in range(self.rowNum):
-            tdSql.execute(f"insert into stb_1 values(%d, %d, %d, %d, %d, %d, %d, %d, %d, %f, %f, %d, '{self.binary_str}%d', '{self.nchar_str}%d')"
+                                tp = (i,)
-                          % (self.ts + i, i + 1, i + 1, i + 1, i + 1, i + 1, i + 1, i + 1, i + 1, i + 0.1, i + 0.1, i % 2, i + 1, i + 1))
+                                values_list.append(tp)
-        
+                            tdSql.checkEqual(tdSql.queryResult,values_list)
-        for i in column_list:
+                        elif tb_type == 'stable':
-            tdSql.query(f'select bottom({i},2) from stb_1')
+                            tdSql.checkRows(param)
-            tdSql.checkRows(2)
+                    elif param < self.rowNum:
-            tdSql.checkEqual(tdSql.queryResult,[(2,),(1,)])
+                        if tb_type in ['normal_table','child_table']:
-            for j in error_param_list:
+                            tdSql.checkRows(param)
-                tdSql.error(f'select bottom({i},{j}) from stb_1')
+                            values_list = []
-        for i in error_column_list:
+                            for i in range(param):
-            tdSql.error(f'select bottom({i},10) from stb_1')
+                                tp = (i,)
-        tdSql.query("select ts,bottom(col1, 2),ts from stb_1 group by tbname")
+                                values_list.append(tp)
-        tdSql.checkRows(2)
+                            tdSql.checkEqual(tdSql.queryResult,values_list)
-        tdSql.query('select bottom(col2,1) from stb_1 interval(1y) order by col2')
+                        elif tb_type == 'stable':
-        tdSql.checkData(0,0,1)
+                            tdSql.checkRows(param)
-
+                    for i in [self.param_list[0]-1,self.param_list[-1]+1]:
-        tdSql.error('select * from stb_1 where bottom(col2,1)=1')
+                        tdSql.error(f'select top({k},{i}) from {tbname}')
-        tdSql.execute('drop database db')
+                    new_column_dict.update({k:v})
-    def bottom_check_distribute(self):
+                elif v.lower() == 'bool' or 'binary' in v.lower() or 'nchar' in v.lower():
-        # prepare data for vgroup 4
+                    tdSql.error(f'select top({k},{param}) from {tbname}')
-        dbname = tdCom.getLongName(5, "letters")
+                tdSql.error(f'select * from {tbname} where top({k},{param})=1')
+        pass
+    def bottom_check_ntb(self):
+        tdSql.execute(f'create database if not exists {self.dbname} vgroups 1')
+        tdSql.execute(f'use {self.dbname}')
+        tdSql.execute(self.setsql.set_create_normaltable_sql(self.ntbname,self.column_dict))
+        self.insert_data(self.column_dict,self.ntbname,self.rowNum)
+        self.bottom_check_data(self.ntbname,'normal_table')
+        tdSql.execute(f'drop database {self.dbname}')
+    def bottom_check_stb(self):
         stbname = tdCom.getLongName(5, "letters")
-        vgroup_num = 2
+        tag_dict = {
-        child_table_num = 20
+            't0':'int'
-        tdSql.execute(f"create database if not exists {dbname} vgroups {vgroup_num}")
+        }
-        tdSql.execute(f'use {dbname}')
+        tag_values = [
-        # build 20 child tables,every table insert 10 rows
+            f'1'
-        tdSql.execute(f'''create table {stbname}(ts timestamp, col1 tinyint, col2 smallint, col3 int, col4 bigint, col5 tinyint unsigned, col6 smallint unsigned, 
+            ]
-                    col7 int unsigned, col8 bigint unsigned, col9 float, col10 double, col11 bool, col12 binary(20), col13 nchar(20)) tags(loc nchar(20))''')
+        tdSql.execute(f"create database if not exists {self.dbname} vgroups 2")
-        for i in range(child_table_num):
+        tdSql.execute(f'use {self.dbname}')
-            tdSql.execute(f"create table {stbname}_{i} using {stbname} tags('beijing')")
+        tdSql.execute(self.setsql.set_create_stable_sql(stbname,self.column_dict,tag_dict))
-            tdSql.execute(f"insert into {stbname}_{i}(ts) values(%d)" % (self.ts - 1-i))
+        for i in range(self.tbnum):
-        column_list = ['col1','col2','col3','col4','col5','col6','col7','col8']
+            tdSql.execute(f"create table {stbname}_{i} using {stbname} tags({tag_values[0]})")
-        error_column_list = ['col11','col12','col13']
+            tdSql.execute(self.insert_data(self.column_dict,f'{stbname}_{i}',self.rowNum))
-        error_param_list = [0,101]
-        for i in [f'{stbname}', f'{dbname}.{stbname}']:
-            for j in column_list:
-                tdSql.query(f"select bottom({j},1) from {i}")
-                tdSql.checkRows(0)
         tdSql.query('show tables')
         vgroup_list = []
         for i in range(len(tdSql.queryResult)):
             vgroup_list.append(tdSql.queryResult[i][6])
         vgroup_list_set = set(vgroup_list)
-
         for i in vgroup_list_set:
             vgroups_num = vgroup_list.count(i)
-            if vgroups_num >=2:
+            if vgroups_num >= 2:
                 tdLog.info(f'This scene with {vgroups_num} vgroups is ok!')
-                continue
             else:
-                tdLog.exit(f'This scene does not meet the requirements with {vgroups_num} vgroup!\n')
+                tdLog.exit(
-        for i in range(self.rowNum):
+                    'This scene does not meet the requirements with {vgroups_num} vgroup!\n')
-            for j in range(child_table_num):
+        for i in range(self.tbnum):
-                tdSql.execute(f"insert into {stbname}_{j} values(%d, %d, %d, %d, %d, %d, %d, %d, %d, %f, %f, %d, '{self.binary_str}%d', '{self.nchar_str}%d')"
+            self.bottom_check_data(f'{stbname}_{i}','child_table')
-                          % (self.ts + i, i + 1, i + 1, i + 1, i + 1, i + 1, i + 1, i + 1, i + 1, i + 0.1, i + 0.1, i % 2, i + 1, i + 1))  
+        self.bottom_check_data(f'{stbname}','stable')
-        for i in column_list:
+        tdSql.execute(f'drop database {self.dbname}')
-            tdSql.query(f'select bottom({i},2) from {stbname}')
-            tdSql.checkRows(2)
-            tdSql.checkEqual(tdSql.queryResult,[(1,),(1,)])
-            for j in error_param_list:
-                tdSql.error(f'select bottom({i},{j}) from {stbname}')
-        for i in error_column_list:
-            tdSql.error(f'select bottom({i},10) from {stbname}')
  
-        tdSql.execute(f'drop database {dbname}')
     def run(self):
-
+        self.bottom_check_ntb()
-        self.bottom_check_base()
+        self.bottom_check_stb()
-        self.bottom_check_distribute()
-        
         
     def stop(self):
         tdSql.close()

tests/system-test/2-query/cast.py

@@ -630,27 +630,27 @@ class TDTestCase:
         ( tdSql.checkData(i, 0, '12') for i in range(tdSql.queryRows) )
 
         tdLog.printNoPrefix("==========step40: error cast condition, should return error ")
-        tdSql.error("select cast(c1 as int) as b from ct4")
+        #tdSql.error("select cast(c1 as int) as b from ct4")
-        tdSql.error("select cast(c1 as bool) as b from ct4")
+        #tdSql.error("select cast(c1 as bool) as b from ct4")
-        tdSql.error("select cast(c1 as tinyint) as b from ct4")
+        #tdSql.error("select cast(c1 as tinyint) as b from ct4")
-        tdSql.error("select cast(c1 as smallint) as b from ct4")
+        #tdSql.error("select cast(c1 as smallint) as b from ct4")
-        tdSql.error("select cast(c1 as float) as b from ct4")
+        #tdSql.error("select cast(c1 as float) as b from ct4")
-        tdSql.error("select cast(c1 as double) as b from ct4")
+        #tdSql.error("select cast(c1 as double) as b from ct4")
-        tdSql.error("select cast(c1 as tinyint unsigned) as b from ct4")
+        #tdSql.error("select cast(c1 as tinyint unsigned) as b from ct4")
-        tdSql.error("select cast(c1 as smallint unsigned) as b from ct4")
+        #tdSql.error("select cast(c1 as smallint unsigned) as b from ct4")
-        tdSql.error("select cast(c1 as int unsigned) as b from ct4")
+        #tdSql.error("select cast(c1 as int unsigned) as b from ct4")
 
-        tdSql.error("select cast(c2 as int) as b from ct4")
+        #tdSql.error("select cast(c2 as int) as b from ct4")
-        tdSql.error("select cast(c3 as bool) as b from ct4")
+        #tdSql.error("select cast(c3 as bool) as b from ct4")
-        tdSql.error("select cast(c4 as tinyint) as b from ct4")
+        #tdSql.error("select cast(c4 as tinyint) as b from ct4")
-        tdSql.error("select cast(c5 as smallint) as b from ct4")
+        #tdSql.error("select cast(c5 as smallint) as b from ct4")
-        tdSql.error("select cast(c6 as float) as b from ct4")
+        #tdSql.error("select cast(c6 as float) as b from ct4")
-        tdSql.error("select cast(c7 as double) as b from ct4")
+        #tdSql.error("select cast(c7 as double) as b from ct4")
-        tdSql.error("select cast(c8 as tinyint unsigned) as b from ct4")
+        #tdSql.error("select cast(c8 as tinyint unsigned) as b from ct4")
 
-        tdSql.error("select cast(c8 as timestamp ) as b from ct4")
+        #tdSql.error("select cast(c8 as timestamp ) as b from ct4")
-        tdSql.error("select cast(c9 as timestamp ) as b from ct4")
+        #tdSql.error("select cast(c9 as timestamp ) as b from ct4")
-        tdSql.error("select cast(c9 as binary(64) ) as b from ct4")
+        #tdSql.error("select cast(c9 as binary(64) ) as b from ct4")
         pass
 
     def run(self):

tests/system-test/2-query/json_tag.py

@@ -11,12 +11,14 @@
 
 # -*- coding: utf-8 -*-
 
+import imp
 import sys
 import taos
 from util.log import tdLog
 from util.cases import tdCases
 from util.sql import tdSql
 import json
+import os
 
 
 class TDTestCase:
@@ -29,6 +31,9 @@ class TDTestCase:
         return
 
     def init(self, conn, logSql):
+        self.testcasePath = os.path.split(__file__)[0]
+        self.testcaseFilename = os.path.split(__file__)[-1]
+        os.system("rm -rf %s/%s.sql" % (self.testcasePath,self.testcaseFilename))
         tdLog.debug("start to execute %s" % __file__)
         tdSql.init(conn.cursor(), logSql)
 
@@ -557,6 +562,103 @@ class TDTestCase:
         tdSql.checkRows(3)
         tdSql.query("select round(dataint) from jsons1 where jtag->'tag1'>1")
         tdSql.checkRows(3)
+        
+        #math function
+        tdSql.query("select sin(dataint) from jsons1 where jtag->'tag1'>1;")
+        tdSql.checkRows(3)
+        tdSql.query("select cos(dataint) from jsons1 where jtag->'tag1'>1;")
+        tdSql.checkRows(3)
+        tdSql.query("select tan(dataint) from jsons1 where jtag->'tag1'>1;")
+        tdSql.checkRows(3)
+        tdSql.query("select asin(dataint) from jsons1 where jtag->'tag1'>1;")
+        tdSql.checkRows(3)
+        tdSql.query("select acos(dataint) from jsons1 where jtag->'tag1'>1;")
+        tdSql.checkRows(3)
+        tdSql.query("select atan(dataint) from jsons1 where jtag->'tag1'>1;")
+        tdSql.checkRows(3)
+        tdSql.query("select ceil(dataint) from jsons1 where jtag->'tag1'>1;")
+        tdSql.checkRows(3)
+        tdSql.query("select floor(dataint) from jsons1 where jtag->'tag1'>1;")
+        tdSql.checkRows(3)
+        tdSql.query("select round(dataint) from jsons1 where jtag->'tag1'>1;")
+        tdSql.checkRows(3)
+        tdSql.query("select abs(dataint) from jsons1 where jtag->'tag1'>1;")
+        tdSql.checkRows(3)
+        tdSql.query("select pow(dataint,5) from jsons1 where jtag->'tag1'>1;")
+        tdSql.checkRows(3)
+        tdSql.query("select log(dataint,10) from jsons1 where jtag->'tag1'>1;")
+        tdSql.checkRows(3)
+        tdSql.query("select sqrt(dataint) from jsons1 where jtag->'tag1'>1;")
+        tdSql.checkRows(3)
+        tdSql.query("select HISTOGRAM(dataint,'user_input','[1, 33, 555, 7777]',1)  from jsons1 where jtag->'tag1'>1;")
+        tdSql.checkRows(3)
+        tdSql.query("select csum(dataint) from jsons1 where jtag->'tag1'>1;")
+        tdSql.checkRows(3)
+        tdSql.query("select mavg(dataint,1) from jsons1 where jtag->'tag1'>1;")
+        tdSql.checkRows(3)
+        tdSql.query("select statecount(dataint,'GE',10) from jsons1 where jtag->'tag1'>1;")
+        tdSql.checkRows(3)
+        tdSql.query("select stateduration(dataint,'GE',0) from jsons1 where jtag->'tag1'>1;")
+        tdSql.checkRows(3)
+        tdSql.query("select sample(dataint,3) from jsons1 where jtag->'tag1'>1;")
+        tdSql.checkRows(3)
+        tdSql.query("select HYPERLOGLOG(dataint) from jsons1 where jtag->'tag1'>1;")
+        tdSql.checkRows(1)
+        tdSql.query("select twa(dataint) from jsons1 where jtag->'tag1'>1;")
+        tdSql.checkRows(1)
+        
+        # function not ready
+        # tdSql.query("select tail(dataint,1) from jsons1 where jtag->'tag1'>1;")
+        # tdSql.checkRows(3)
+        # tdSql.query("select unique(dataint) from jsons1 where jtag->'tag1'>1;")
+        # tdSql.checkRows(3)
+        # tdSql.query("select mode(dataint) from jsons1 where jtag->'tag1'>1;")
+        # tdSql.checkRows(3)
+        # tdSql.query("select irate(dataint) from jsons1 where jtag->'tag1'>1;")
+        # tdSql.checkRows(1)
+        
+        #str function
+        tdSql.query("select upper(dataStr) from jsons1 where jtag->'tag1'>1;")
+        tdSql.checkRows(3)
+        tdSql.query("select ltrim(dataStr) from jsons1 where jtag->'tag1'>1;")
+        tdSql.checkRows(3)
+        tdSql.query("select lower(dataStr) from jsons1 where jtag->'tag1'>1;")
+        tdSql.checkRows(3)
+        tdSql.query("select rtrim(dataStr) from jsons1 where jtag->'tag1'>1;")
+        tdSql.checkRows(3)
+        tdSql.query("select LENGTH(dataStr) from jsons1 where jtag->'tag1'>1;")
+        tdSql.checkRows(3)
+        tdSql.query("select CHAR_LENGTH(dataStr) from jsons1 where jtag->'tag1'>1;")
+        tdSql.checkRows(3)
+        tdSql.query("select SUBSTR(dataStr,5) from jsons1 where jtag->'tag1'>1;")
+        tdSql.checkRows(3)
+        tdSql.query("select CONCAT(dataStr,dataStrBin) from jsons1 where jtag->'tag1'>1;")
+        tdSql.checkRows(3)
+        tdSql.query("select CONCAT_ws('adad!@!@%$^$%$^$%^a',dataStr,dataStrBin) from jsons1 where jtag->'tag1'>1;")
+        tdSql.checkRows(3)
+        tdSql.query("select CAST(dataStr as bigint) from jsons1 where jtag->'tag1'>1;")
+        tdSql.checkRows(3)
+
+        #time function
+        tdSql.query("select now() from jsons1 where jtag->'tag1'>1;")
+        tdSql.checkRows(3)
+        tdSql.query("select today() from jsons1 where jtag->'tag1'>1;")
+        tdSql.checkRows(3)
+        tdSql.query("select TIMEZONE() from jsons1 where jtag->'tag1'>1;")
+        tdSql.checkRows(3)
+        tdSql.query("select TO_ISO8601(ts) from jsons1 where jtag->'tag1'>1;")
+        tdSql.checkRows(3)
+        tdSql.query("select TO_UNIXTIMESTAMP(datastr) from jsons1 where jtag->'tag1'>1;")
+        tdSql.checkRows(3)
+        tdSql.query("select TIMETRUNCATE(ts,1u) from jsons1 where jtag->'tag1'>1;")
+        tdSql.checkRows(3)
+        tdSql.query("select TIMEDIFF(ts,_c0) from jsons1 where jtag->'tag1'>1;")
+        tdSql.checkRows(3)
+        tdSql.query("select TIMEDIFF(ts,1u) from jsons1 where jtag->'tag1'>1;")
+        tdSql.checkRows(3)
+        tdSql.query("select ELAPSED(ts,1h) from jsons1 where jtag->'tag1'>1;")
+        tdSql.checkRows(1)
+        
         #
         # #test TD-12077
         tdSql.execute("insert into jsons1_16 using jsons1 tags('{\"tag1\":\"收到货\",\"tag2\":\"\",\"tag3\":-2.111}') values(1591062628000, 2, NULL, '你就会', 'dws')")

tests/system-test/6-cluster/5dnode1mnode.py

@@ -20,7 +20,7 @@ class MyDnodes(TDDnodes):
         self.simDeployed = False
         
 class TDTestCase:
-
+    noConn = True
     def init(self,conn ,logSql):
         tdLog.debug(f"start to excute {__file__}")
         self.TDDnodes = None
@@ -40,7 +40,7 @@ class TDTestCase:
             projPath = selfPath[:selfPath.find("tests")]
 
         for root, dirs, files in os.walk(projPath):
-            if ("taosd" in files):
+            if ("taosd" in files or "taosd.exe" in files):
                 rootRealPath = os.path.dirname(os.path.realpath(root))
                 if ("packaging" not in rootRealPath):
                     buildPath = root[:len(root) - len("/build/bin")]
@@ -81,7 +81,7 @@ class TDTestCase:
             dnode_id = dnode.cfgDict["fqdn"] +  ":" +dnode.cfgDict["serverPort"]
             dnode_first_host = dnode.cfgDict["firstEp"].split(":")[0]
             dnode_first_port = dnode.cfgDict["firstEp"].split(":")[-1]
-            cmd = f" taos -h {dnode_first_host} -P {dnode_first_port} -s ' create dnode \"{dnode_id} \" ' ;"
+            cmd = f"{self.getBuildPath()}/build/bin/taos -h {dnode_first_host} -P {dnode_first_port} -s \"create dnode \\\"{dnode_id}\\\"\""
             print(cmd)
             os.system(cmd)
         

tests/system-test/6-cluster/5dnode2mnode.py

@@ -20,6 +20,7 @@ class MyDnodes(TDDnodes):
         self.simDeployed = False
         
 class TDTestCase:
+    noConn = True
 
     def init(self,conn ,logSql):
         tdLog.debug(f"start to excute {__file__}")
@@ -40,7 +41,7 @@ class TDTestCase:
             projPath = selfPath[:selfPath.find("tests")]
 
         for root, dirs, files in os.walk(projPath):
-            if ("taosd" in files):
+            if ("taosd" in files or "taosd.exe" in files):
                 rootRealPath = os.path.dirname(os.path.realpath(root))
                 if ("packaging" not in rootRealPath):
                     buildPath = root[:len(root) - len("/build/bin")]
@@ -85,7 +86,7 @@ class TDTestCase:
             dnode_id = dnode.cfgDict["fqdn"] +  ":" +dnode.cfgDict["serverPort"]
             dnode_first_host = dnode.cfgDict["firstEp"].split(":")[0]
             dnode_first_port = dnode.cfgDict["firstEp"].split(":")[-1]
-            cmd = f" taos -h {dnode_first_host} -P {dnode_first_port} -s ' create dnode \"{dnode_id} \" ' ;"
+            cmd = f"{self.getBuildPath()}/build/bin/taos -h {dnode_first_host} -P {dnode_first_port} -s \"create dnode \\\"{dnode_id}\\\"\""
             print(cmd)
             os.system(cmd)
         

tests/system-test/6-cluster/5dnode3mnodeSep1VnodeStopCreateDb.py

@@ -0,0 +1,181 @@
+from ssl import ALERT_DESCRIPTION_CERTIFICATE_UNOBTAINABLE
+import taos
+import sys
+import time
+import os 
+
+from util.log import *
+from util.sql import *
+from util.cases import *
+from util.dnodes import TDDnodes
+from util.dnodes import TDDnode
+from util.cluster import *
+sys.path.append("./6-cluster")
+from clusterCommonCreate import *
+from clusterCommonCheck import clusterComCheck 
+
+import time
+import socket
+import subprocess
+from multiprocessing import Process
+import threading 
+import time
+import inspect
+import ctypes
+
+class TDTestCase:
+
+    def init(self,conn ,logSql):
+        tdLog.debug(f"start to excute {__file__}")
+        self.TDDnodes = None
+        tdSql.init(conn.cursor())
+        self.host = socket.gethostname()
+
+
+    def getBuildPath(self):
+        selfPath = os.path.dirname(os.path.realpath(__file__))
+
+        if ("community" in selfPath):
+            projPath = selfPath[:selfPath.find("community")]
+        else:
+            projPath = selfPath[:selfPath.find("tests")]
+
+        for root, dirs, files in os.walk(projPath):
+            if ("taosd" in files):
+                rootRealPath = os.path.dirname(os.path.realpath(root))
+                if ("packaging" not in rootRealPath):
+                    buildPath = root[:len(root) - len("/build/bin")]
+                    break
+        return buildPath
+
+    def _async_raise(self, tid, exctype):
+        """raises the exception, performs cleanup if needed"""
+        if not inspect.isclass(exctype):
+            exctype = type(exctype)
+        res = ctypes.pythonapi.PyThreadState_SetAsyncExc(tid, ctypes.py_object(exctype))
+        if res == 0:
+            raise ValueError("invalid thread id")
+        elif res != 1:
+            # """if it returns a number greater than one, you're in trouble, 
+            # and you should call it again with exc=NULL to revert the effect"""
+            ctypes.pythonapi.PyThreadState_SetAsyncExc(tid, None)
+            raise SystemError("PyThreadState_SetAsyncExc failed")
+
+    def stopThread(self,thread):
+        self._async_raise(thread.ident, SystemExit)
+
+
+    def insertData(self,countstart,countstop):
+        # fisrt add data : db\stable\childtable\general table
+        
+        for couti in range(countstart,countstop):
+            tdLog.debug("drop database if exists db%d" %couti)
+            tdSql.execute("drop database if exists db%d" %couti)
+            print("create database if not exists db%d replica 1 duration 300" %couti)
+            tdSql.execute("create database if not exists db%d replica 1 duration 300" %couti)
+            tdSql.execute("use db%d" %couti)
+            tdSql.execute(
+            '''create table stb1
+            (ts timestamp, c1 int, c2 bigint, c3 smallint, c4 tinyint, c5 float, c6 double, c7 bool, c8 binary(16),c9 nchar(32), c10 timestamp)
+            tags (t1 int)
+            '''
+            )
+            tdSql.execute(
+                '''
+                create table t1
+                (ts timestamp, c1 int, c2 bigint, c3 smallint, c4 tinyint, c5 float, c6 double, c7 bool, c8 binary(16),c9 nchar(32), c10 timestamp)
+                '''
+            )
+            for i in range(4):
+                tdSql.execute(f'create table ct{i+1} using stb1 tags ( {i+1} )')
+
+
+    def fiveDnodeThreeMnode(self,dnodenumbers,mnodeNums,restartNumber):
+        tdLog.printNoPrefix("======== test case 1: ")
+        paraDict = {'dbName':     'db',
+                    'dropFlag':   1,
+                    'event':      '',
+                    'vgroups':    4,
+                    'replica':    1,
+                    'stbName':    'stb',
+                    'colPrefix':  'c',
+                    'tagPrefix':  't',
+                    'colSchema':   [{'type': 'INT', 'count':1}, {'type': 'binary', 'len':20, 'count':1}],
+                    'tagSchema':   [{'type': 'INT', 'count':1}, {'type': 'binary', 'len':20, 'count':1}],
+                    'ctbPrefix':  'ctb',
+                    'ctbNum':     1,
+                    'rowsPerTbl': 10000,
+                    'batchNum':   10,
+                    'startTs':    1640966400000,  # 2022-01-01 00:00:00.000
+                    'pollDelay':  10,
+                    'showMsg':    1,
+                    'showRow':    1}
+        dnodenumbers=int(dnodenumbers)
+        mnodeNums=int(mnodeNums)
+        dbNumbers = int(dnodenumbers * restartNumber)
+        
+        tdLog.info("first check dnode and mnode")
+        tdSql.query("show dnodes;")
+        tdSql.checkData(0,1,'%s:6030'%self.host)
+        tdSql.checkData(4,1,'%s:6430'%self.host)
+        clusterComCheck.checkDnodes(dnodenumbers)
+        clusterComCheck.checkMnodeStatus(1)
+
+        # fisr add three mnodes;
+        tdLog.info("fisr add three mnodes and check mnode status")
+        tdSql.execute("create mnode on dnode 2")
+        clusterComCheck.checkMnodeStatus(2)
+        tdSql.execute("create mnode on dnode 3")
+        clusterComCheck.checkMnodeStatus(3)
+
+        # add some error operations and 
+        tdLog.info("Confirm the status of the dnode again")
+        tdSql.error("create mnode on dnode 2")
+        tdSql.query("show dnodes;")
+        print(tdSql.queryResult)
+        clusterComCheck.checkDnodes(dnodenumbers)
+
+        tdLog.info("Take turns stopping all dnodes ") 
+        # seperate vnode and mnode in different dnodes.
+        # create database and stable
+        tdDnodes=cluster.dnodes
+        stopcount =0 
+        while stopcount < restartNumber:
+            for i in range(dnodenumbers):
+                # threads=[]
+                # threads = MyThreadFunc(self.insert_data(i*2,i*2+2)) 
+                paraDict["dbName"]= 'db%d%d'%(stopcount,i)
+                threads=threading.Thread(target=clusterComCreate.create_database, args=(tdSql, paraDict["dbName"],paraDict["dropFlag"], paraDict["vgroups"],paraDict['replica']))
+                threads.start()
+                tdDnodes[i].stoptaosd()
+                # sleep(10)
+                tdDnodes[i].starttaosd()
+                # sleep(10)
+
+            if clusterComCheck.checkDnodes(dnodenumbers):
+                # threads.join()
+                tdLog.info("first restart loop")
+            else:
+                print("456")
+                threads.join()
+                self.stopThread(threads)
+                tdLog.exit("one or more of dnodes failed to start ")
+                # self.check3mnode()
+            stopcount+=1
+        threads.join()
+        clusterComCheck.checkDnodes(dnodenumbers)
+        clusterComCheck.checkDbRows(dbNumbers)
+        for i in range(restartNumber):
+            clusterComCheck.checkDb(dnodenumbers,'db%d'%i)
+
+
+    def run(self): 
+        # print(self.master_dnode.cfgDict)
+        self.fiveDnodeThreeMnode(5,3,1)
+
+    def stop(self):
+        tdSql.close()
+        tdLog.success(f"{__file__} successfully executed")
+
+tdCases.addLinux(__file__, TDTestCase())
+tdCases.addWindows(__file__, TDTestCase())

tests/system-test/6-cluster/5dnode3mnodeSep1VnodeStopMnodeCreateDb.py

@@ -0,0 +1,181 @@
+from ssl import ALERT_DESCRIPTION_CERTIFICATE_UNOBTAINABLE
+import taos
+import sys
+import time
+import os 
+
+from util.log import *
+from util.sql import *
+from util.cases import *
+from util.dnodes import TDDnodes
+from util.dnodes import TDDnode
+from util.cluster import *
+sys.path.append("./6-cluster")
+from clusterCommonCreate import *
+from clusterCommonCheck import clusterComCheck 
+
+import time
+import socket
+import subprocess
+from multiprocessing import Process
+import threading 
+import time
+import inspect
+import ctypes
+
+class TDTestCase:
+
+    def init(self,conn ,logSql):
+        tdLog.debug(f"start to excute {__file__}")
+        self.TDDnodes = None
+        tdSql.init(conn.cursor())
+        self.host = socket.gethostname()
+
+
+    def getBuildPath(self):
+        selfPath = os.path.dirname(os.path.realpath(__file__))
+
+        if ("community" in selfPath):
+            projPath = selfPath[:selfPath.find("community")]
+        else:
+            projPath = selfPath[:selfPath.find("tests")]
+
+        for root, dirs, files in os.walk(projPath):
+            if ("taosd" in files):
+                rootRealPath = os.path.dirname(os.path.realpath(root))
+                if ("packaging" not in rootRealPath):
+                    buildPath = root[:len(root) - len("/build/bin")]
+                    break
+        return buildPath
+
+    def _async_raise(self, tid, exctype):
+        """raises the exception, performs cleanup if needed"""
+        if not inspect.isclass(exctype):
+            exctype = type(exctype)
+        res = ctypes.pythonapi.PyThreadState_SetAsyncExc(tid, ctypes.py_object(exctype))
+        if res == 0:
+            raise ValueError("invalid thread id")
+        elif res != 1:
+            # """if it returns a number greater than one, you're in trouble, 
+            # and you should call it again with exc=NULL to revert the effect"""
+            ctypes.pythonapi.PyThreadState_SetAsyncExc(tid, None)
+            raise SystemError("PyThreadState_SetAsyncExc failed")
+
+    def stopThread(self,thread):
+        self._async_raise(thread.ident, SystemExit)
+
+
+    def insertData(self,countstart,countstop):
+        # fisrt add data : db\stable\childtable\general table
+        
+        for couti in range(countstart,countstop):
+            tdLog.debug("drop database if exists db%d" %couti)
+            tdSql.execute("drop database if exists db%d" %couti)
+            print("create database if not exists db%d replica 1 duration 300" %couti)
+            tdSql.execute("create database if not exists db%d replica 1 duration 300" %couti)
+            tdSql.execute("use db%d" %couti)
+            tdSql.execute(
+            '''create table stb1
+            (ts timestamp, c1 int, c2 bigint, c3 smallint, c4 tinyint, c5 float, c6 double, c7 bool, c8 binary(16),c9 nchar(32), c10 timestamp)
+            tags (t1 int)
+            '''
+            )
+            tdSql.execute(
+                '''
+                create table t1
+                (ts timestamp, c1 int, c2 bigint, c3 smallint, c4 tinyint, c5 float, c6 double, c7 bool, c8 binary(16),c9 nchar(32), c10 timestamp)
+                '''
+            )
+            for i in range(4):
+                tdSql.execute(f'create table ct{i+1} using stb1 tags ( {i+1} )')
+
+
+    def fiveDnodeThreeMnode(self,dnodenumbers,mnodeNums,restartNumber):
+        tdLog.printNoPrefix("======== test case 1: ")
+        paraDict = {'dbName':     'db',
+                    'dropFlag':   1,
+                    'event':      '',
+                    'vgroups':    4,
+                    'replica':    1,
+                    'stbName':    'stb',
+                    'colPrefix':  'c',
+                    'tagPrefix':  't',
+                    'colSchema':   [{'type': 'INT', 'count':1}, {'type': 'binary', 'len':20, 'count':1}],
+                    'tagSchema':   [{'type': 'INT', 'count':1}, {'type': 'binary', 'len':20, 'count':1}],
+                    'ctbPrefix':  'ctb',
+                    'ctbNum':     1,
+                    'rowsPerTbl': 10000,
+                    'batchNum':   10,
+                    'startTs':    1640966400000,  # 2022-01-01 00:00:00.000
+                    'pollDelay':  10,
+                    'showMsg':    1,
+                    'showRow':    1}
+        dnodenumbers=int(dnodenumbers)
+        mnodeNums=int(mnodeNums)
+        dbNumbers = int(mnodeNums * restartNumber)
+        tdLog.info("first check dnode and mnode")
+        tdSql.query("show dnodes;")
+        tdSql.checkData(0,1,'%s:6030'%self.host)
+        tdSql.checkData(4,1,'%s:6430'%self.host)
+        clusterComCheck.checkDnodes(dnodenumbers)
+        clusterComCheck.checkMnodeStatus(1)
+
+        # fisr add three mnodes;
+        tdLog.info("fisr add three mnodes and check mnode status")
+        tdSql.execute("create mnode on dnode 2")
+        clusterComCheck.checkMnodeStatus(2)
+        tdSql.execute("create mnode on dnode 3")
+        clusterComCheck.checkMnodeStatus(3)
+
+        # add some error operations and 
+        tdLog.info("Confirm the status of the dnode again")
+        tdSql.error("create mnode on dnode 2")
+        tdSql.query("show dnodes;")
+        print(tdSql.queryResult)
+        clusterComCheck.checkDnodes(dnodenumbers)
+
+        tdLog.info("Take turns stopping Mnodes ") 
+        # seperate vnode and mnode in different dnodes.
+        # create database and stable
+        tdDnodes=cluster.dnodes
+        stopcount =0 
+        while stopcount < restartNumber:
+            tdLog.info("first restart loop")
+            for i in range(mnodeNums):
+                # threads=[]
+                # threads = MyThreadFunc(self.insert_data(i*2,i*2+2)) 
+                paraDict["dbName"]= 'db%d%d'%(stopcount,i)
+                threads=threading.Thread(target=clusterComCreate.create_database, args=(tdSql, paraDict["dbName"],paraDict["dropFlag"], paraDict["vgroups"],paraDict['replica']))
+                threads.start()
+                tdDnodes[i].stoptaosd()
+                # sleep(10)
+                tdDnodes[i].starttaosd()
+                # sleep(10)
+
+            if clusterComCheck.checkDnodes(dnodenumbers):
+                # threads.join()
+                tdLog.info("123")
+            else:
+                print("456")
+                threads.join()
+                self.stopThread(threads)
+                tdLog.exit("one or more of dnodes failed to start ")
+                # self.check3mnode()
+            stopcount+=1
+        threads.join()
+        clusterComCheck.checkDnodes(dnodenumbers)
+        clusterComCheck.checkDbRows(dbNumbers)
+        for i in range(restartNumber):
+            clusterComCheck.checkDb(mnodeNums,'db%d'%i)
+
+
+    def run(self): 
+        # print(self.master_dnode.cfgDict)
+        self.fiveDnodeThreeMnode(5,3,1)
+
+    def stop(self):
+        tdSql.close()
+        tdLog.success(f"{__file__} successfully executed")
+
+tdCases.addLinux(__file__, TDTestCase())
+tdCases.addWindows(__file__, TDTestCase())

tests/system-test/6-cluster/5dnode3mnodeSep1VnodeStopVnodeCreateDb.py

@@ -0,0 +1,181 @@
+from ssl import ALERT_DESCRIPTION_CERTIFICATE_UNOBTAINABLE
+import taos
+import sys
+import time
+import os 
+
+from util.log import *
+from util.sql import *
+from util.cases import *
+from util.dnodes import TDDnodes
+from util.dnodes import TDDnode
+from util.cluster import *
+sys.path.append("./6-cluster")
+from clusterCommonCreate import *
+from clusterCommonCheck import clusterComCheck 
+
+import time
+import socket
+import subprocess
+from multiprocessing import Process
+import threading 
+import time
+import inspect
+import ctypes
+
+class TDTestCase:
+
+    def init(self,conn ,logSql):
+        tdLog.debug(f"start to excute {__file__}")
+        self.TDDnodes = None
+        tdSql.init(conn.cursor())
+        self.host = socket.gethostname()
+
+
+    def getBuildPath(self):
+        selfPath = os.path.dirname(os.path.realpath(__file__))
+
+        if ("community" in selfPath):
+            projPath = selfPath[:selfPath.find("community")]
+        else:
+            projPath = selfPath[:selfPath.find("tests")]
+
+        for root, dirs, files in os.walk(projPath):
+            if ("taosd" in files):
+                rootRealPath = os.path.dirname(os.path.realpath(root))
+                if ("packaging" not in rootRealPath):
+                    buildPath = root[:len(root) - len("/build/bin")]
+                    break
+        return buildPath
+
+    def _async_raise(self, tid, exctype):
+        """raises the exception, performs cleanup if needed"""
+        if not inspect.isclass(exctype):
+            exctype = type(exctype)
+        res = ctypes.pythonapi.PyThreadState_SetAsyncExc(tid, ctypes.py_object(exctype))
+        if res == 0:
+            raise ValueError("invalid thread id")
+        elif res != 1:
+            # """if it returns a number greater than one, you're in trouble, 
+            # and you should call it again with exc=NULL to revert the effect"""
+            ctypes.pythonapi.PyThreadState_SetAsyncExc(tid, None)
+            raise SystemError("PyThreadState_SetAsyncExc failed")
+
+    def stopThread(self,thread):
+        self._async_raise(thread.ident, SystemExit)
+
+
+    def insertData(self,countstart,countstop):
+        # fisrt add data : db\stable\childtable\general table
+        
+        for couti in range(countstart,countstop):
+            tdLog.debug("drop database if exists db%d" %couti)
+            tdSql.execute("drop database if exists db%d" %couti)
+            print("create database if not exists db%d replica 1 duration 300" %couti)
+            tdSql.execute("create database if not exists db%d replica 1 duration 300" %couti)
+            tdSql.execute("use db%d" %couti)
+            tdSql.execute(
+            '''create table stb1
+            (ts timestamp, c1 int, c2 bigint, c3 smallint, c4 tinyint, c5 float, c6 double, c7 bool, c8 binary(16),c9 nchar(32), c10 timestamp)
+            tags (t1 int)
+            '''
+            )
+            tdSql.execute(
+                '''
+                create table t1
+                (ts timestamp, c1 int, c2 bigint, c3 smallint, c4 tinyint, c5 float, c6 double, c7 bool, c8 binary(16),c9 nchar(32), c10 timestamp)
+                '''
+            )
+            for i in range(4):
+                tdSql.execute(f'create table ct{i+1} using stb1 tags ( {i+1} )')
+
+
+    def fiveDnodeThreeMnode(self,dnodenumbers,mnodeNums,restartNumber):
+        tdLog.printNoPrefix("======== test case 1: ")
+        paraDict = {'dbName':     'db',
+                    'dropFlag':   1,
+                    'event':      '',
+                    'vgroups':    4,
+                    'replica':    1,
+                    'stbName':    'stb',
+                    'colPrefix':  'c',
+                    'tagPrefix':  't',
+                    'colSchema':   [{'type': 'INT', 'count':1}, {'type': 'binary', 'len':20, 'count':1}],
+                    'tagSchema':   [{'type': 'INT', 'count':1}, {'type': 'binary', 'len':20, 'count':1}],
+                    'ctbPrefix':  'ctb',
+                    'ctbNum':     1,
+                    'rowsPerTbl': 10000,
+                    'batchNum':   10,
+                    'startTs':    1640966400000,  # 2022-01-01 00:00:00.000
+                    'pollDelay':  10,
+                    'showMsg':    1,
+                    'showRow':    1}
+        dnodenumbers=int(dnodenumbers)
+        mnodeNums=int(mnodeNums)
+        vnodeNumbers = int(dnodenumbers-mnodeNums)
+        dbNumbers = int(vnodeNumbers * restartNumber)
+        tdLog.info("first check dnode and mnode")
+        tdSql.query("show dnodes;")
+        tdSql.checkData(0,1,'%s:6030'%self.host)
+        tdSql.checkData(4,1,'%s:6430'%self.host)
+        clusterComCheck.checkDnodes(dnodenumbers)
+        clusterComCheck.checkMnodeStatus(1)
+
+        # fisr add three mnodes;
+        tdLog.info("fisr add three mnodes and check mnode status")
+        tdSql.execute("create mnode on dnode 2")
+        clusterComCheck.checkMnodeStatus(2)
+        tdSql.execute("create mnode on dnode 3")
+        clusterComCheck.checkMnodeStatus(3)
+
+        # add some error operations and 
+        tdLog.info("Confirm the status of the dnode again")
+        tdSql.error("create mnode on dnode 2")
+        tdSql.query("show dnodes;")
+        print(tdSql.queryResult)
+        clusterComCheck.checkDnodes(dnodenumbers)
+
+        tdLog.info("Take turns stopping Vnodes ") 
+        # seperate vnode and mnode in different dnodes.
+        # create database and stable
+        tdDnodes=cluster.dnodes
+        stopcount =0 
+        while stopcount < restartNumber:
+            for i in range(vnodeNumbers):
+                # threads=[]
+                # threads = MyThreadFunc(self.insert_data(i*2,i*2+2)) 
+                paraDict["dbName"]= 'db%d%d'%(stopcount,i)
+                threads=threading.Thread(target=clusterComCreate.create_database, args=(tdSql, paraDict["dbName"],paraDict["dropFlag"], paraDict["vgroups"],paraDict['replica']))
+                threads.start()
+                tdDnodes[mnodeNums+i].stoptaosd()
+                # sleep(10)
+                tdDnodes[mnodeNums+i].starttaosd()
+                # sleep(10)
+
+            if clusterComCheck.checkDnodes(vnodeNumbers):
+                # threads.join()
+                tdLog.info("first restart loop")
+            else:
+                print("456")
+                threads.join()
+                self.stopThread(threads)
+                tdLog.exit("one or more of dnodes failed to start ")
+                # self.check3mnode()
+            stopcount+=1
+        threads.join()
+        clusterComCheck.checkDnodes(dnodenumbers)
+        clusterComCheck.checkDbRows(dbNumbers)
+        for i in range(restartNumber):
+            clusterComCheck.checkDb(vnodeNumbers,'db%d'%i)
+
+
+    def run(self): 
+        # print(self.master_dnode.cfgDict)
+        self.fiveDnodeThreeMnode(5,3,1)
+
+    def stop(self):
+        tdSql.close()
+        tdLog.success(f"{__file__} successfully executed")
+
+tdCases.addLinux(__file__, TDTestCase())
+tdCases.addWindows(__file__, TDTestCase())

tests/system-test/6-cluster/5dnode3mnodeSeperate1VnodeStopInsert.py

@@ -9,6 +9,11 @@ from util.sql import *
 from util.cases import *
 from util.dnodes import TDDnodes
 from util.dnodes import TDDnode
+from util.cluster import *
+from util.common import *
+sys.path.append("./7-tmq")
+from tmqCommon import *
+
 import time
 import socket
 import subprocess
@@ -17,26 +22,13 @@ import threading
 import time
 import inspect
 import ctypes
-class MyDnodes(TDDnodes):
-    def __init__(self ,dnodes_lists):
-        super(MyDnodes,self).__init__()
-        self.dnodes = dnodes_lists  # dnode must be TDDnode instance
-        self.simDeployed = False
-
 
 class TDTestCase:
 
     def init(self,conn ,logSql):
         tdLog.debug(f"start to excute {__file__}")
-        self.TDDnodes = None
+        # tdSql.init(conn.cursor())
-
+        # self.host = socket.gethostname()
-    def buildcluster(self,dnodenumber):
-        self.depoly_cluster(dnodenumber)
-        self.master_dnode = self.TDDnodes.dnodes[0]
-        self.host=self.master_dnode.cfgDict["fqdn"]
-        conn1 = taos.connect(self.master_dnode.cfgDict["fqdn"] , config=self.master_dnode.cfgDir)
-        tdSql.init(conn1.cursor())
-        
 
     def getBuildPath(self):
         selfPath = os.path.dirname(os.path.realpath(__file__))
@@ -106,52 +98,6 @@ class TDTestCase:
             tdSql.checkData(0,0,rowsPerSTable)
         return 
 
-    def depoly_cluster(self ,dnodes_nums=5,independent=True): 
-
-        testCluster = False
-        valgrind = 0  
-        hostname = socket.gethostname()
-        dnodes = []
-        start_port = 6030
-        start_port_sec = 6130
-        for num in range(1, dnodes_nums+1):
-            dnode = TDDnode(num)
-            dnode.addExtraCfg("firstEp", f"{hostname}:{start_port}")
-            dnode.addExtraCfg("fqdn", f"{hostname}")
-            dnode.addExtraCfg("serverPort", f"{start_port + (num-1)*100}")
-            dnode.addExtraCfg("monitorFqdn", hostname)
-            dnode.addExtraCfg("monitorPort", 7043)
-            dnode.addExtraCfg("secondEp", f"{hostname}:{start_port_sec}")
-            # configure three dnoe don't support vnodes
-            if independent and (num < 4):
-                dnode.addExtraCfg("supportVnodes", 0)
-
-            dnodes.append(dnode)
-        
-        self.TDDnodes = MyDnodes(dnodes)
-        self.TDDnodes.init("")
-        self.TDDnodes.setTestCluster(testCluster)
-        self.TDDnodes.setValgrind(valgrind)
-        self.TDDnodes.stopAll()
-        for dnode in self.TDDnodes.dnodes:
-            self.TDDnodes.deploy(dnode.index,{})
-            
-        for dnode in self.TDDnodes.dnodes:
-            self.TDDnodes.starttaosd(dnode.index)
-
-        # create cluster 
-        for dnode in self.TDDnodes.dnodes[1:]:
-            # print(dnode.cfgDict)
-            dnode_id = dnode.cfgDict["fqdn"] +  ":" +dnode.cfgDict["serverPort"]
-            dnode_first_host = dnode.cfgDict["firstEp"].split(":")[0]
-            dnode_first_port = dnode.cfgDict["firstEp"].split(":")[-1]
-            cmd = f" taos -h {dnode_first_host} -P {dnode_first_port} -s ' create dnode \"{dnode_id} \" ' ;"
-            print(cmd)
-            os.system(cmd)
-        
-        time.sleep(2)
-        tdLog.info(" create cluster with %d dnode  done! " %dnodes_nums)
-
     def checkdnodes(self,dnodenumber):
         count=0
         while count < 100:
@@ -305,6 +251,14 @@ class TDTestCase:
         tdSql.checkData(2,2,'offline')
         tdSql.checkData(2,3,'ready')
 
+
+    def check5dnode(self):
+        tdSql.query("show dnodes;")
+        tdSql.checkData(0,1,'%s:6030'%self.host)
+        tdSql.checkData(4,1,'%s:6430'%self.host)
+        tdSql.checkData(0,4,'ready')
+        tdSql.checkData(4,4,'ready')
+        
     def five_dnode_three_mnode(self,dnodenumber):
         tdSql.query("show dnodes;")
         tdSql.checkData(0,1,'%s:6030'%self.host)
@@ -346,6 +300,7 @@ class TDTestCase:
                     threads.join()
                 else:
                     print("456")
+                    threads.join()
                     self.stop_thread(threads)
                     assert 1 == 2 ,"some dnode started failed"
                     return False
@@ -357,16 +312,8 @@ class TDTestCase:
         self.check3mnode()
 
 
-    def getConnection(self, dnode):
-        host = dnode.cfgDict["fqdn"]
-        port = dnode.cfgDict["serverPort"]
-        config_dir = dnode.cfgDir
-        return taos.connect(host=host, port=int(port), config=config_dir)
-
-
     def run(self): 
         # print(self.master_dnode.cfgDict)
-        self.buildcluster(5)
         self.five_dnode_three_mnode(5)
 
     def stop(self):

tests/system-test/6-cluster/5dnode3mnodeStop.py

@@ -12,7 +12,10 @@ from util.dnodes import TDDnodes
 from util.dnodes import TDDnode
 from util.cluster import *
 from test import tdDnodes
+sys.path.append("./6-cluster")
 
+from clusterCommonCreate import *
+from clusterCommonCheck import * 
 import time
 import socket
 import subprocess
@@ -216,60 +219,84 @@ class TDTestCase:
         else:
             tdLog.exit("create cluster with %d dnode but  check dnode not ready within 5s ! "%dnodeNumbers)
 
-    def five_dnode_three_mnode(self,dnodenumber):
+    def fiveDnodeThreeMnode(self,dnodenumbers,mnodeNums,restartNumber):
-        self.check_dnodes_status(5)
+        tdLog.printNoPrefix("======== test case 1: ")
-        tdSql.query("show mnodes;")   
+        paraDict = {'dbName':     'db',
-        tdLog.debug(self.host)
+                    'dropFlag':   1,
-        tdSql.checkRows(1)    
+                    'event':      '',
+                    'vgroups':    4,
+                    'replica':    1,
+                    'stbName':    'stb',
+                    'colPrefix':  'c',
+                    'tagPrefix':  't',
+                    'colSchema':   [{'type': 'INT', 'count':1}, {'type': 'binary', 'len':20, 'count':1}],
+                    'tagSchema':   [{'type': 'INT', 'count':1}, {'type': 'binary', 'len':20, 'count':1}],
+                    'ctbPrefix':  'ctb',
+                    'ctbNum':     1,
+                    'rowsPerTbl': 10000,
+                    'batchNum':   10,
+                    'startTs':    1640966400000,  # 2022-01-01 00:00:00.000
+                    'pollDelay':  10,
+                    'showMsg':    1,
+                    'showRow':    1}
+        dnodenumbers=int(dnodenumbers)
+        mnodeNums=int(mnodeNums)
+        dbNumbers = int(dnodenumbers * restartNumber)
+        
+        tdLog.info("first check dnode and mnode")
+        tdSql.query("show dnodes;")
         tdSql.checkData(0,1,'%s:6030'%self.host)
-        tdSql.checkData(0,2,'leader')
+        tdSql.checkData(4,1,'%s:6430'%self.host)
-        tdSql.checkData(0,3,'ready')
+        clusterComCheck.checkDnodes(dnodenumbers)
+        clusterComCheck.checkMnodeStatus(1)
 
         # fisr add three mnodes;
+        tdLog.info("fisr add three mnodes and check mnode status")
         tdSql.execute("create mnode on dnode 2")
-        time.sleep(10)
+        clusterComCheck.checkMnodeStatus(2)
         tdSql.execute("create mnode on dnode 3")
+        clusterComCheck.checkMnodeStatus(3)
 
-        # fisrt check statut ready
+        # add some error operations and 
-        self.check3mnode()
+        tdLog.info("Confirm the status of the dnode again")
-
-
         tdSql.error("create mnode on dnode 2")
-
         tdSql.query("show dnodes;")
-        # tdLog.debug(tdSql.queryResult)
+        print(tdSql.queryResult)
-
+        clusterComCheck.checkDnodes(dnodenumbers)
-        tdLog.debug("stop and follower of mnode") 
+        # restart all taosd
         tdDnodes=cluster.dnodes
-        # tdLog.debug(tdDnodes[0])
 
         tdDnodes[1].stoptaosd()
-        self.check3mnode2off()
+        clusterComCheck.check3mnodeoff(2,3)
         tdDnodes[1].starttaosd()
-        self.check3mnode()
+        clusterComCheck.checkMnodeStatus(3)
 
         tdDnodes[2].stoptaosd()
-        self.check3mnode3off()
+        clusterComCheck.check3mnodeoff(3,3)
         tdDnodes[2].starttaosd()
-        self.check3mnode()
+        clusterComCheck.checkMnodeStatus(3)
 
         tdDnodes[0].stoptaosd()
-        self.check3mnode1off()
+        clusterComCheck.check3mnodeoff(1,3)
         tdDnodes[0].starttaosd()
-        self.check3mnode()
+        clusterComCheck.checkMnodeStatus(3)
 
-        self.check3mnode()
+        tdLog.info("Take turns stopping all dnodes ") 
+        # seperate vnode and mnode in different dnodes.
+        # create database and stable
         stopcount =0 
         while stopcount <= 2:
-            for i in range(dnodenumber):
+            tdLog.info("first restart loop")
+            for i in range(dnodenumbers):
                 tdDnodes[i].stoptaosd()
                 tdDnodes[i].starttaosd()
-                # self.check3mnode()
             stopcount+=1
-        self.check3mnode()
+        clusterComCheck.checkDnodes(dnodenumbers)
+        clusterComCheck.checkMnodeStatus(3)
 
     def run(self): 
-        self.five_dnode_three_mnode(5)
+        # print(self.master_dnode.cfgDict)
+        self.fiveDnodeThreeMnode(5,3,1)
  
     def stop(self):
         tdSql.close()

tests/system-test/6-cluster/clusterCommonCheck.py

@@ -0,0 +1,211 @@
+###################################################################
+#           Copyright (c) 2016 by TAOS Technologies, Inc.
+#                     All rights reserved.
+#
+#  This file is proprietary and confidential to TAOS Technologies.
+#  No part of this file may be reproduced, stored, transmitted,
+#  disclosed or used in any form or by any means other than as
+#  expressly provided by the written permission from Jianhui Tao
+#
+###################################################################
+
+# -*- coding: utf-8 -*-
+
+from collections import defaultdict
+import random
+import string
+import threading
+import requests
+import time
+# import socketfrom
+
+import taos
+from util.log import *
+from util.sql import *
+from util.cases import *
+from util.dnodes import *
+from util.common import *
+
+# class actionType(Enum):
+#     CREATE_DATABASE = 0
+#     CREATE_STABLE   = 1
+#     CREATE_CTABLE   = 2
+#     INSERT_DATA     = 3
+
+class ClusterComCheck:
+    def init(self, conn, logSql):
+        tdSql.init(conn.cursor())
+        # tdSql.init(conn.cursor(), logSql)  # output sql.txt file
+
+    def checkDnodes(self,dnodeNumbers):
+        count=0
+        while count < 5:
+            tdSql.query("show dnodes")
+            # tdLog.debug(tdSql.queryResult)
+            status=0
+            for i in range(dnodeNumbers):
+                if tdSql.queryResult[i][4] == "ready":
+                    status+=1
+            tdLog.info(status)
+            
+            if status == dnodeNumbers:
+                tdLog.success("it find cluster with %d dnodes and check that all cluster dnodes are ready within 5s! " %dnodeNumbers)
+                return True 
+            count+=1
+            time.sleep(1)
+        else:
+            tdLog.debug(tdSql.queryResult)
+            tdLog.exit("it find cluster with %d dnodes but  check that there dnodes are not ready within 5s ! "%dnodeNumbers)
+
+    def checkDbRows(self,dbNumbers):
+        dbNumbers=int(dbNumbers)
+        count=0
+        while count < 5:
+            tdSql.query("show databases;")
+            if tdSql.checkRows(dbNumbers+2):
+                tdLog.success("we find %d databases and expect %d in clusters! " %(tdSql.queryRows,dbNumbers+2))
+                return True
+            else:
+                continue
+        else :
+            tdLog.debug(tdSql.queryResult)
+            tdLog.exit("we find %d databases but expect %d in clusters! " %(tdSql.queryRows,dbNumbers))
+
+    def checkDb(self,dbNumbers,dbindex):
+        count=0
+        while count < 5:           
+            query_status=0
+            for i in range(dbNumbers):
+                for j in range(dbNumbers):
+                    tdSql.query("show databases;")
+                    if "%s%d"%(dbindex,j) == tdSql.queryResult[i+2][0] :                    
+                        if tdSql.queryResult[i+2][19] == "ready":
+                            query_status+=1
+                        else:
+                            continue
+                    # print(query_status)
+            count+=1
+            if query_status == dbNumbers:
+                tdLog.success("we find cluster with %d dnode and check all databases  are ready within 5s! " %dbNumbers)
+                return True
+        else:
+            tdLog.debug(tdSql.queryResult)
+            tdLog.exit("database is not ready within 5s")
+
+    def checkData(self,dbname,stbname,stableCount,CtableCount,rowsPerSTable,):
+        tdSql.execute("use %s"%dbname)
+        tdSql.query("show stables")
+        tdSql.checkRows(stableCount)
+        tdSql.query("show tables")
+        tdSql.checkRows(CtableCount)
+        for i in range(stableCount):
+            tdSql.query("select count(*) from %s%d"%(stbname,i))
+            tdSql.checkData(0,0,rowsPerSTable)
+        return 
+
+    def checkMnodeStatus(self,mnodeNums):
+        self.mnodeNums=int(mnodeNums)
+        # self.leaderDnode=int(leaderDnode)
+        
+        count=0
+
+        while count < 10:
+            time.sleep(1)
+            tdSql.query("show mnodes;")
+            if tdSql.checkRows(self.mnodeNums) :     
+                tdLog.success("cluster has %d mnodes" %self.mnodeNums )
+
+            if self.mnodeNums == 1:
+                if  tdSql.queryResult[0][2]== 'leader' and  tdSql.queryResult[0][3]== 'ready'  :
+                    tdLog.success("%d mnodes is ready in 10s"%self.mnodeNums)
+                    return True 
+                count+=1      
+            elif self.mnodeNums == 3 :          
+                if  tdSql.queryResult[0][2]=='leader'  and  tdSql.queryResult[0][3]== 'ready' :
+                    if  tdSql.queryResult[1][2]=='follower' and  tdSql.queryResult[1][3]== 'ready' :
+                        if  tdSql.queryResult[2][2]=='follower' and  tdSql.queryResult[2][3]== 'ready' :
+                            tdLog.success("%d mnodes is ready in 10s"%self.mnodeNums)
+                            return True
+                elif  tdSql.queryResult[1][2]=='leader'  and  tdSql.queryResult[1][3]== 'ready' :
+                    if  tdSql.queryResult[0][2]=='follower' and  tdSql.queryResult[0][3]== 'ready' :
+                        if  tdSql.queryResult[2][2]=='follower' and  tdSql.queryResult[2][3]== 'ready' :
+                            tdLog.success("%d mnodes is ready in 10s"%self.mnodeNums)
+                            return True
+                elif  tdSql.queryResult[2][2]=='leader'  and  tdSql.queryResult[2][3]== 'ready' :
+                    if  tdSql.queryResult[0][2]=='follower' and  tdSql.queryResult[0][3]== 'ready' :
+                        if  tdSql.queryResult[1][2]=='follower' and  tdSql.queryResult[1][3]== 'ready' :
+                            tdLog.success("%d mnodes is ready in 10s"%self.mnodeNums)
+                            return True          
+                count+=1
+            elif self.mnodeNums == 2 :   
+                if  tdSql.queryResult[0][2]=='leader' and  tdSql.queryResult[0][3]== 'ready' :
+                    if  tdSql.queryResult[1][2]=='follower' and  tdSql.queryResult[1][3]== 'ready' :
+                        tdLog.success("%d mnodes is ready in 10s"%self.mnodeNums)
+                        return True
+                elif tdSql.queryResult[1][2]=='leader' and  tdSql.queryResult[1][3]== 'ready' :
+                    if  tdSql.queryResult[0][2]=='follower' and  tdSql.queryResult[0][3]== 'ready' :
+                        tdLog.success("%d mnodes is ready in 10s"%self.mnodeNums)
+                        return True
+                count+=1
+        else:
+            tdLog.debug(tdSql.queryResult)
+            tdLog.exit("cluster of %d  mnodes is not ready in 10s " %self.mnodeNums)
+
+                
+
+
+    def check3mnodeoff(self,offlineDnodeNo,mnodeNums=3):
+        count=0
+        while count < 10:
+            time.sleep(1)
+            tdSql.query("show mnodes;")
+            if tdSql.checkRows(mnodeNums) :
+                tdLog.success("cluster has %d mnodes" %self.mnodeNums )
+            else:
+                tdLog.exit("mnode number is correct")
+            if offlineDnodeNo == 1:
+                if  tdSql.queryResult[0][2]=='offline' :
+                    if  tdSql.queryResult[1][2]=='leader'  and  tdSql.queryResult[1][3]== 'ready'  :
+                        if  tdSql.queryResult[2][2]=='follower' and  tdSql.queryResult[2][3]== 'ready'  :
+                            tdLog.success("stop mnodes  on dnode %d  successfully in 10s" %offlineDnodeNo)
+                            return True
+                    elif tdSql.queryResult[1][2]=='follower' and  tdSql.queryResult[1][3]== 'ready' :
+                        if  tdSql.queryResult[2][2]=='leader' and  tdSql.queryResult[2][3]== 'ready' :
+                            tdLog.debug("stop mnodes  on dnode %d  successfully in 10s" %offlineDnodeNo)
+                            return True
+                count+=1
+            elif offlineDnodeNo == 2:
+                if  tdSql.queryResult[1][2]=='offline' :
+                    if  tdSql.queryResult[0][2]=='leader' and  tdSql.queryResult[0][3]== 'ready' :
+                        if  tdSql.queryResult[2][2]=='follower' and  tdSql.queryResult[2][3]== 'ready' :
+                            tdLog.debug("stop mnodes  on dnode %d  successfully in 10s" %offlineDnodeNo)
+                            return True
+                    elif tdSql.queryResult[0][2]=='follower' and  tdSql.queryResult[0][3]== 'ready' :
+                        if  tdSql.queryResult[2][2]=='leader' and  tdSql.queryResult[2][3]== 'ready' :
+                            tdLog.debug("stop mnodes  on dnode %d  successfully in 10s" %offlineDnodeNo)
+                            return True
+                count+=1
+            elif offlineDnodeNo == 3:
+                if  tdSql.queryResult[2][2]=='offline' :
+                    if  tdSql.queryResult[0][2]=='leader' and  tdSql.queryResult[0][3]== 'ready' :
+                        if  tdSql.queryResult[1][2]=='follower' and  tdSql.queryResult[1][3]== 'ready' :
+                            tdLog.debug("stop mnodes  on dnode %d  successfully in 10s" %offlineDnodeNo)
+                            return True
+                    elif tdSql.queryResult[0][2]=='follower' and  tdSql.queryResult[0][3]== 'ready' :
+                        if  tdSql.queryResult[1][2]=='leader' and  tdSql.queryResult[1][3]== 'ready' :
+                            tdLog.debug("stop mnodes  on dnode %d  successfully in 10s" %offlineDnodeNo)
+                            return True
+                count+=1
+        else:
+            tdLog.debug(tdSql.queryResult)
+            tdLog.exit("stop mnodes  on dnode %d  failed in 10s ")
+
+
+    
+
+
+
+    def close(self):
+        self.cursor.close()
+
+clusterComCheck = ClusterComCheck()

tests/system-test/6-cluster/clusterCommonCreate.py

@@ -0,0 +1,298 @@
+###################################################################
+#           Copyright (c) 2016 by TAOS Technologies, Inc.
+#                     All rights reserved.
+#
+#  This file is proprietary and confidential to TAOS Technologies.
+#  No part of this file may be reproduced, stored, transmitted,
+#  disclosed or used in any form or by any means other than as
+#  expressly provided by the written permission from Jianhui Tao
+#
+###################################################################
+
+# -*- coding: utf-8 -*-
+
+from collections import defaultdict
+import random
+import string
+import threading
+import requests
+import time
+# import socketfrom
+
+import taos
+from util.log import *
+from util.sql import *
+from util.cases import *
+from util.dnodes import *
+from util.common import *
+
+# class actionType(Enum):
+#     CREATE_DATABASE = 0
+#     CREATE_STABLE   = 1
+#     CREATE_CTABLE   = 2
+#     INSERT_DATA     = 3
+
+class ClusterComCreate:
+    def init(self, conn, logSql):
+        tdSql.init(conn.cursor())
+        # tdSql.init(conn.cursor(), logSql)  # output sql.txt file
+
+    def initConsumerTable(self,cdbName='cdb'):        
+        tdLog.info("create consume database, and consume info table, and consume result table")
+        tdSql.query("create database if not exists %s vgroups 1"%(cdbName))
+        tdSql.query("drop table if exists %s.consumeinfo "%(cdbName))
+        tdSql.query("drop table if exists %s.consumeresult "%(cdbName))
+        tdSql.query("drop table if exists %s.notifyinfo "%(cdbName))      
+
+        tdSql.query("create table %s.consumeinfo (ts timestamp, consumerid int, topiclist binary(1024), keylist binary(1024), expectmsgcnt bigint, ifcheckdata int, ifmanualcommit int)"%cdbName)
+        tdSql.query("create table %s.consumeresult (ts timestamp, consumerid int, consummsgcnt bigint, consumrowcnt bigint, checkresult int)"%cdbName)
+        tdSql.query("create table %s.notifyinfo (ts timestamp, cmdid int, consumerid int)"%cdbName)
+
+    def initConsumerInfoTable(self,cdbName='cdb'):        
+        tdLog.info("drop consumeinfo table")
+        tdSql.query("drop table if exists %s.consumeinfo "%(cdbName))
+        tdSql.query("create table %s.consumeinfo (ts timestamp, consumerid int, topiclist binary(1024), keylist binary(1024), expectmsgcnt bigint, ifcheckdata int, ifmanualcommit int)"%cdbName)
+
+    def insertConsumerInfo(self,consumerId, expectrowcnt,topicList,keyList,ifcheckdata,ifmanualcommit,cdbName='cdb'):    
+        sql = "insert into %s.consumeinfo values "%cdbName
+        sql += "(now, %d, '%s', '%s', %d, %d, %d)"%(consumerId, topicList, keyList, expectrowcnt, ifcheckdata, ifmanualcommit)
+        tdLog.info("consume info sql: %s"%sql)
+        tdSql.query(sql)
+
+    def selectConsumeResult(self,expectRows,cdbName='cdb'):
+        resultList=[]
+        while 1:
+            tdSql.query("select * from %s.consumeresult"%cdbName)
+            #tdLog.info("row: %d, %l64d, %l64d"%(tdSql.getData(0, 1),tdSql.getData(0, 2),tdSql.getData(0, 3))
+            if tdSql.getRows() == expectRows:
+                break
+            else:
+                time.sleep(5)
+        
+        for i in range(expectRows):
+            tdLog.info ("consume id: %d, consume msgs: %d, consume rows: %d"%(tdSql.getData(i , 1), tdSql.getData(i , 2), tdSql.getData(i , 3)))
+            resultList.append(tdSql.getData(i , 3))
+        
+        return resultList
+
+    def startTmqSimProcess(self,pollDelay,dbName,showMsg=1,showRow=1,cdbName='cdb',valgrind=0):
+        buildPath = tdCom.getBuildPath()
+        cfgPath = tdCom.getClientCfgPath()
+        if valgrind == 1:
+            logFile = cfgPath + '/../log/valgrind-tmq.log'
+            shellCmd = 'nohup valgrind --log-file=' + logFile
+            shellCmd += '--tool=memcheck --leak-check=full --show-reachable=no --track-origins=yes --show-leak-kinds=all --num-callers=20 -v --workaround-gcc296-bugs=yes '
+        
+        if (platform.system().lower() == 'windows'):
+            shellCmd = 'mintty -h never -w hide ' + buildPath + '\\build\\bin\\tmq_sim.exe -c ' + cfgPath
+            shellCmd += " -y %d -d %s -g %d -r %d -w %s "%(pollDelay, dbName, showMsg, showRow, cdbName) 
+            shellCmd += "> nul 2>&1 &"   
+        else:
+            shellCmd = 'nohup ' + buildPath + '/build/bin/tmq_sim -c ' + cfgPath
+            shellCmd += " -y %d -d %s -g %d -r %d -w %s "%(pollDelay, dbName, showMsg, showRow, cdbName) 
+            shellCmd += "> /dev/null 2>&1 &"
+        tdLog.info(shellCmd)
+        os.system(shellCmd)
+
+    def getStartConsumeNotifyFromTmqsim(self,cdbName='cdb'):
+        while 1:
+            tdSql.query("select * from %s.notifyinfo"%cdbName)
+            #tdLog.info("row: %d, %l64d, %l64d"%(tdSql.getData(0, 1),tdSql.getData(0, 2),tdSql.getData(0, 3))
+            if (tdSql.getRows() == 1) and (tdSql.getData(0, 1) == 0):
+                break
+            else:
+                time.sleep(0.1)
+        return
+
+    def getStartCommitNotifyFromTmqsim(self,cdbName='cdb'):
+        while 1:
+            tdSql.query("select * from %s.notifyinfo"%cdbName)
+            #tdLog.info("row: %d, %l64d, %l64d"%(tdSql.getData(0, 1),tdSql.getData(0, 2),tdSql.getData(0, 3))
+            if tdSql.getRows() == 2 :
+                print(tdSql.getData(0, 1), tdSql.getData(1, 1))
+                if tdSql.getData(1, 1) == 1:
+                    break
+            time.sleep(0.1)
+        return
+
+    def create_database(self,tsql, dbName,dropFlag=1,vgroups=4,replica=1):
+        if dropFlag == 1:
+            tsql.execute("drop database if exists %s"%(dbName))
+
+        tsql.execute("create database if not exists %s vgroups %d replica %d"%(dbName, vgroups, replica))
+        tdLog.debug("complete to create database %s"%(dbName))
+        return
+
+    def create_stable(self,tsql, dbName,stbName):
+        tsql.execute("create table if not exists %s.%s (ts timestamp, c1 int, c2 int, c3 binary(16)) tags(t1 int, t2 binary(32))"%(dbName, stbName))
+        tdLog.debug("complete to create %s.%s" %(dbName, stbName))
+        return
+
+    def create_ctable(self,tsql=None, dbName='dbx',stbName='stb',ctbPrefix='ctb',ctbNum=1):
+        tsql.execute("use %s" %dbName)
+        pre_create = "create table"
+        sql = pre_create
+        #tdLog.debug("doing create one  stable %s and %d  child table in %s  ..." %(stbname, count ,dbname))
+        for i in range(ctbNum):
+            tagValue = 'beijing'
+            if (i % 2 == 0):
+                tagValue = 'shanghai'
+            
+            sql += " %s%d using %s tags(%d, '%s')"%(ctbPrefix,i,stbName,i+1, tagValue)
+            if (i > 0) and (i%100 == 0):
+                tsql.execute(sql)
+                sql = pre_create
+        if sql != pre_create:
+            tsql.execute(sql)
+        
+        tdLog.debug("complete to create %d child tables in %s.%s" %(ctbNum, dbName, stbName))
+        return
+
+    def insert_data(self,tsql,dbName,stbName,ctbNum,rowsPerTbl,batchNum,startTs=None):
+        tdLog.debug("start to insert data ............")
+        tsql.execute("use %s" %dbName)
+        pre_insert = "insert into "
+        sql = pre_insert
+
+        if startTs is None:
+            t = time.time()
+            startTs = int(round(t * 1000))
+        #tdLog.debug("doing insert data into stable:%s rows:%d ..."%(stbName, allRows))
+        for i in range(ctbNum):
+            sql += " %s%d values "%(stbName,i)
+            for j in range(rowsPerTbl):
+                sql += "(%d, %d, 'tmqrow_%d') "%(startTs + j, j, j)
+                if (j > 0) and ((j%batchNum == 0) or (j == rowsPerTbl - 1)):
+                    tsql.execute(sql)
+                    if j < rowsPerTbl - 1:
+                        sql = "insert into %s%d values " %(stbName,i)
+                    else:
+                        sql = "insert into "
+        #end sql
+        if sql != pre_insert:
+            #print("insert sql:%s"%sql)
+            tsql.execute(sql)
+        tdLog.debug("insert data ............ [OK]")
+        return        
+
+    def insert_data_1(self,tsql,dbName,ctbPrefix,ctbNum,rowsPerTbl,batchNum,startTs):
+        tdLog.debug("start to insert data ............")
+        tsql.execute("use %s" %dbName)
+        pre_insert = "insert into "
+        sql = pre_insert
+
+        t = time.time()
+        startTs = int(round(t * 1000))
+        #tdLog.debug("doing insert data into stable:%s rows:%d ..."%(stbName, allRows))
+        for i in range(ctbNum):
+            sql += " %s%d values "%(ctbPrefix,i)
+            for j in range(rowsPerTbl):
+                if (j % 2 == 0):
+                    sql += "(%d, %d, %d, 'tmqrow_%d') "%(startTs + j, j, j, j)
+                else:
+                    sql += "(%d, %d, %d, 'tmqrow_%d') "%(startTs + j, j, -j, j)
+                if (j > 0) and ((j%batchNum == 0) or (j == rowsPerTbl - 1)):
+                    tsql.execute(sql)
+                    if j < rowsPerTbl - 1:
+                        sql = "insert into %s%d values " %(ctbPrefix,i)
+                    else:
+                        sql = "insert into "
+        #end sql
+        if sql != pre_insert:
+            #print("insert sql:%s"%sql)
+            tsql.execute(sql)
+        tdLog.debug("insert data ............ [OK]")
+        return
+
+    def insert_data_interlaceByMultiTbl(self,tsql,dbName,ctbPrefix,ctbNum,rowsPerTbl,batchNum,startTs=0):
+        tdLog.debug("start to insert data ............")
+        tsql.execute("use %s" %dbName)
+        pre_insert = "insert into "
+        sql = pre_insert
+
+        if startTs == 0:
+            t = time.time()
+            startTs = int(round(t * 1000))
+
+        ctbDict = {}
+        for i in range(ctbNum):
+            ctbDict[i] = 0
+
+        #tdLog.debug("doing insert data into stable:%s rows:%d ..."%(stbName, allRows))
+        rowsOfCtb = 0        
+        while rowsOfCtb < rowsPerTbl:
+            for i in range(ctbNum):
+                sql += " %s.%s_%d values "%(dbName,ctbPrefix,i)
+                for k in range(batchNum):
+                    sql += "(%d, %d, 'tmqrow_%d') "%(startTs + ctbDict[i], ctbDict[i], ctbDict[i])
+                    ctbDict[i] += 1
+                    if (0 == ctbDict[i]%batchNum) or (ctbDict[i] == rowsPerTbl):
+                        tsql.execute(sql)
+                        sql = "insert into "
+                        break
+            rowsOfCtb = ctbDict[0]
+
+        tdLog.debug("insert data ............ [OK]")
+        return
+
+    def insert_data_with_autoCreateTbl(self,tsql,dbName,stbName,ctbPrefix,ctbNum,rowsPerTbl,batchNum,startTs=0):
+        tdLog.debug("start to insert data wiht auto create child table ............")
+        tsql.execute("use %s" %dbName)
+        pre_insert = "insert into "
+        sql = pre_insert
+
+        if startTs == 0:
+            t = time.time()
+            startTs = int(round(t * 1000))
+
+        #tdLog.debug("doing insert data into stable:%s rows:%d ..."%(stbName, allRows))
+        rowsOfSql = 0        
+        for i in range(ctbNum):
+            sql += " %s.%s_%d using %s.%s tags (%d) values "%(dbName,ctbPrefix,i,dbName,stbName,i)
+            for j in range(rowsPerTbl):
+                sql += "(%d, %d, 'tmqrow_%d') "%(startTs + j, j, j)
+                rowsOfSql += 1
+                if (j > 0) and ((rowsOfSql == batchNum) or (j == rowsPerTbl - 1)):
+                    tsql.execute(sql)
+                    rowsOfSql = 0
+                    if j < rowsPerTbl - 1:
+                        sql = "insert into %s.%s_%d using %s.%s tags (%d) values " %(dbName,ctbPrefix,i,dbName,stbName,i)
+                    else:
+                        sql = "insert into "
+        #end sql
+        if sql != pre_insert:
+            #print("insert sql:%s"%sql)
+            tsql.execute(sql)
+        tdLog.debug("insert data ............ [OK]")
+        return
+
+    def syncCreateDbStbCtbInsertData(self, tsql, paraDict):
+        tdCom.create_database(tsql, paraDict["dbName"],paraDict["dropFlag"])
+        tdCom.create_stable(tsql, dbname=paraDict["dbName"],stbname=paraDict["stbName"], column_elm_list=paraDict['colSchema'], tag_elm_list=paraDict['tagSchema'])
+        tdCom.create_ctable(tsql, dbname=paraDict["dbName"],stbname=paraDict["stbName"],tag_elm_list=paraDict['tagSchema'],count=paraDict["ctbNum"], default_ctbname_prefix=paraDict['ctbPrefix'])
+        if "event" in paraDict and type(paraDict['event']) == type(threading.Event()):
+            paraDict["event"].set()
+
+        ctbPrefix = paraDict['ctbPrefix']
+        ctbNum = paraDict["ctbNum"]
+        for i in range(ctbNum):
+            tbName = '%s%s'%(ctbPrefix,i)
+            tdCom.insert_rows(tsql,dbname=paraDict["dbName"],tbname=tbName,start_ts_value=paraDict['startTs'],count=paraDict['rowsPerTbl'])
+        return 
+
+    def threadFunction(self, **paraDict):
+        # create new connector for new tdSql instance in my thread
+        newTdSql = tdCom.newTdSql()
+        self.syncCreateDbStbCtbInsertData(self, newTdSql, paraDict)
+        return
+
+    def asyncCreateDbStbCtbInsertData(self, paraDict):
+        pThread = threading.Thread(target=self.threadFunction, kwargs=paraDict)
+        pThread.start()
+        return pThread
+
+
+    def close(self):
+        self.cursor.close()
+
+clusterComCreate = ClusterComCreate()

tests/system-test/7-tmq/tmqCommon.py

@@ -86,7 +86,13 @@ class TMQCom:
             shellCmd += '--tool=memcheck --leak-check=full --show-reachable=no --track-origins=yes --show-leak-kinds=all --num-callers=20 -v --workaround-gcc296-bugs=yes '
         
         if (platform.system().lower() == 'windows'):
-            shellCmd = 'mintty -h never -w hide ' + buildPath + '\\build\\bin\\tmq_sim.exe -c ' + cfgPath
+            processorName = buildPath + '\\build\\bin\\tmq_sim.exe'
+            if alias != 0:
+                processorNameNew = buildPath + '\\build\\bin\\tmq_sim_new.exe'
+                shellCmd = 'cp %s %s'%(processorName, processorNameNew)
+                os.system(shellCmd)
+                processorName = processorNameNew
+            shellCmd = 'mintty -h never ' + processorName + ' -c ' + cfgPath
             shellCmd += " -y %d -d %s -g %d -r %d -w %s "%(pollDelay, dbName, showMsg, showRow, cdbName) 
             shellCmd += "> nul 2>&1 &"   
         else:

tests/system-test/7-tmq/tmqError.py

@@ -288,6 +288,9 @@ class TDTestCase:
             tdLog.exit("tmq consume rows error!")
 
         tdSql.query("drop topic %s"%topicFromStb1)
+        if (platform.system().lower() == 'windows'):
+            os.system("TASKKILL /F /IM tmq_sim.exe")
+        else:
             os.system('pkill tmq_sim')
 
         tdLog.printNoPrefix("======== test case 1 end ...... ")  

tests/system-test/7-tmq/tmqShow.py

@@ -0,0 +1,158 @@
+
+import taos
+import sys
+import time
+import socket
+import os
+import threading
+
+from util.log import *
+from util.sql import *
+from util.cases import *
+from util.dnodes import *
+from util.common import *
+sys.path.append("./7-tmq")
+from tmqCommon import *
+
+class TDTestCase:
+    def init(self, conn, logSql):
+        tdLog.debug(f"start to excute {__file__}")
+        tdSql.init(conn.cursor())
+        #tdSql.init(conn.cursor(), logSql)  # output sql.txt file
+
+    def tmqCase1(self):
+        tdLog.printNoPrefix("======== test case 1: ")
+        paraDict = {'dbName':     'db1',
+                    'dropFlag':   1,
+                    'event':      '',
+                    'vgroups':    4,
+                    'stbName':    'stb',
+                    'colPrefix':  'c',
+                    'tagPrefix':  't',
+                    'colSchema':   [{'type': 'INT', 'count':2}, {'type': 'binary', 'len':20, 'count':1},{'type': 'TIMESTAMP', 'count':1}],
+                    'tagSchema':   [{'type': 'INT', 'count':1}, {'type': 'binary', 'len':20, 'count':1}],
+                    'ctbPrefix':  'ctb',
+                    'ctbNum':     10,
+                    'rowsPerTbl': 4000,
+                    'batchNum':   15,
+                    'startTs':    1640966400000,  # 2022-01-01 00:00:00.000
+                    'pollDelay':  20,
+                    'showMsg':    1,
+                    'showRow':    1}
+
+        topicNameList = ['topic1', 'topic2', 'topic3', 'topic4']
+        consumeGroupIdList = ['cgrp1', 'cgrp1', 'cgrp3', 'cgrp4']
+        consumerIdList = [0, 1, 2, 3]
+        tmqCom.initConsumerTable()
+        tdCom.create_database(tdSql, paraDict["dbName"],paraDict["dropFlag"], vgroups=paraDict['vgroups'],replica=1)
+        tdLog.info("create stb")
+        tdCom.create_stable(tdSql, dbname=paraDict["dbName"],stbname=paraDict["stbName"], column_elm_list=paraDict['colSchema'], tag_elm_list=paraDict['tagSchema'])
+        tdLog.info("create ctb")
+        tdCom.create_ctable(tdSql, dbname=paraDict["dbName"],stbname=paraDict["stbName"],tag_elm_list=paraDict['tagSchema'],count=paraDict["ctbNum"], default_ctbname_prefix=paraDict['ctbPrefix'])
+        # tdLog.info("insert data")
+        # tmqCom.insert_data(tdSql,paraDict["dbName"],paraDict["ctbPrefix"],paraDict["ctbNum"],paraDict["rowsPerTbl"],paraDict["batchNum"],paraDict["startTs"])
+        
+        tdLog.info("create 4 topics")
+        sqlString = "create topic %s as database %s" %(topicNameList[0], paraDict['dbName'])
+        tdLog.info("create topic sql: %s"%sqlString)
+        tdSql.execute(sqlString)
+        
+        sqlString = "create topic %s as stable %s.%s" %(topicNameList[1], paraDict['dbName'], paraDict['stbName'])
+        tdLog.info("create topic sql: %s"%sqlString)
+        tdSql.execute(sqlString)        
+
+        queryString = "select * from %s.%s where c1 %% 7 == 0" %(paraDict['dbName'], paraDict['stbName'])
+        sqlString = "create topic %s as %s" %(topicNameList[2], queryString)
+        tdLog.info("create topic sql: %s"%sqlString)
+        tdSql.execute(sqlString)
+        
+        queryString = "select ts, log(c1), ceil(pow(c1,3)) from %s.%s where c1 %% 7 == 0" %(paraDict['dbName'], paraDict['stbName'])
+        sqlString = "create topic %s as %s " %(topicNameList[3], queryString)
+        tdLog.info("create topic sql: %s"%sqlString)
+        tdSql.execute(sqlString)
+
+        tdSql.query("show topics")
+        tdLog.debug(tdSql.queryResult)     
+        rows = tdSql.getRows()
+        if rows != len(consumerIdList):
+            tdLog.exit("topic rows error")
+            
+        for i in range (rows):
+            topicName = tdSql.getData(i,0)
+            matchFlag = 0
+            while matchFlag == 0:
+                for j in range(len(topicNameList)):
+                    if topicName == topicNameList[j]:
+                        matchFlag = 1
+                        break
+                if matchFlag == 0:
+                    tdLog.exit("topic name: %s is error", topicName)
+        
+        # init consume info, and start tmq_sim, then check consume result
+        tdLog.info("insert consume info to consume processor")
+        expectrowcnt = paraDict["rowsPerTbl"] * paraDict["ctbNum"]
+        topicList    = topicNameList[0]
+        ifcheckdata  = 0
+        ifManualCommit = 0        
+        keyList      = 'group.id:%s, enable.auto.commit:false, auto.commit.interval.ms:6000, auto.offset.reset:earliest'%consumeGroupIdList[0]
+        tmqCom.insertConsumerInfo(consumerIdList[0], expectrowcnt,topicList,keyList,ifcheckdata,ifManualCommit)
+                
+        topicList    = topicNameList[1]
+        keyList      = 'group.id:%s, enable.auto.commit:false, auto.commit.interval.ms:6000, auto.offset.reset:earliest'%consumeGroupIdList[1]
+        tmqCom.insertConsumerInfo(consumerIdList[1], expectrowcnt,topicList,keyList,ifcheckdata,ifManualCommit)
+        
+        topicList    = topicNameList[2]
+        keyList      = 'group.id:%s, enable.auto.commit:false, auto.commit.interval.ms:6000, auto.offset.reset:earliest'%consumeGroupIdList[2]
+        tmqCom.insertConsumerInfo(consumerIdList[2], expectrowcnt,topicList,keyList,ifcheckdata,ifManualCommit)
+        
+        topicList    = topicNameList[3]
+        keyList      = 'group.id:%s, enable.auto.commit:false, auto.commit.interval.ms:6000, auto.offset.reset:earliest'%consumeGroupIdList[3]
+        tmqCom.insertConsumerInfo(consumerIdList[3], expectrowcnt,topicList,keyList,ifcheckdata,ifManualCommit)
+
+        tdLog.info("start consume processor")
+        tmqCom.startTmqSimProcess(paraDict['pollDelay'],paraDict["dbName"],paraDict['showMsg'], paraDict['showRow'])
+
+        tdLog.info("async insert data")
+        pThread = tmqCom.asyncInsertData(paraDict)
+
+        time.sleep(5)
+        tdLog.info("check show consumers")
+        tdSql.query("show consumers")
+        # tdLog.info(tdSql.queryResult)     
+        rows = tdSql.getRows()
+        tdLog.info("show consumers rows: %d"%rows)
+        if rows != len(topicNameList):
+            tdLog.exit("show consumers rows error")
+        
+        tdLog.info("check show subscriptions")      
+        tdSql.query("show subscriptions")
+        # tdLog.debug(tdSql.queryResult)     
+        rows = tdSql.getRows()
+        tdLog.info("show subscriptions rows: %d"%rows)
+        if rows != paraDict['vgroups'] * len(topicNameList):
+            tdLog.exit("show subscriptions rows error")
+
+        pThread.join()
+        
+        tdLog.info("insert process end, and start to check consume result")
+        expectRows = len(consumerIdList)
+        _ = tmqCom.selectConsumeResult(expectRows)
+        
+        time.sleep(10)        
+        for i in range(len(topicNameList)):
+            tdSql.query("drop topic %s"%topicNameList[i])
+
+        tdLog.printNoPrefix("======== test case 1 end ...... ")
+
+    def run(self):
+        tdSql.prepare()
+        self.tmqCase1()
+
+    def stop(self):
+        tdSql.close()
+        tdLog.success(f"{__file__} successfully executed")
+
+event = threading.Event()
+
+tdCases.addLinux(__file__, TDTestCase())
+tdCases.addWindows(__file__, TDTestCase())

tests/system-test/99-TDcase/TD-16821.py

@@ -0,0 +1,186 @@
+
+import taos
+import sys
+import time
+import socket
+import os
+import threading
+
+from util.log import *
+from util.sql import *
+from util.cases import *
+from util.dnodes import *
+from util.common import *
+sys.path.append("./7-tmq")
+from tmqCommon import *
+
+class TDTestCase:
+    def init(self, conn, logSql):
+        tdLog.debug(f"start to excute {__file__}")
+        tdSql.init(conn.cursor())
+        #tdSql.init(conn.cursor(), logSql)  # output sql.txt file
+
+    def checkFileContent(self, consumerId, queryString):
+        buildPath = tdCom.getBuildPath()
+        cfgPath = tdCom.getClientCfgPath()
+        dstFile = '%s/../log/dstrows_%d.txt'%(cfgPath, consumerId)
+        cmdStr = '%s/build/bin/taos -c %s -s "%s >> %s"'%(buildPath, cfgPath, queryString, dstFile)
+        tdLog.info(cmdStr)
+        os.system(cmdStr)
+        
+        consumeRowsFile = '%s/../log/consumerid_%d.txt'%(cfgPath, consumerId)
+        tdLog.info("rows file: %s, %s"%(consumeRowsFile, dstFile))
+
+        consumeFile = open(consumeRowsFile, mode='r')
+        queryFile = open(dstFile, mode='r')
+        
+        # skip first line for it is schema
+        queryFile.readline()
+
+        while True:
+            dst = queryFile.readline()
+            src = consumeFile.readline()
+            
+            if dst:
+                if dst != src:
+                    tdLog.exit("consumerId %d consume rows is not match the rows by direct query"%consumerId)
+            else:
+                break
+        return 
+
+    def tmqCase1(self):
+        tdLog.printNoPrefix("======== test case 1: ")
+        paraDict = {'dbName':     'db1',
+                    'dropFlag':   1,
+                    'event':      '',
+                    'vgroups':    4,
+                    'stbName':    'stb',
+                    'colPrefix':  'c',
+                    'tagPrefix':  't',
+                    'colSchema':   [{'type': 'INT', 'count':1}, {'type': 'binary', 'len':20, 'count':1}],
+                    'tagSchema':   [{'type': 'INT', 'count':1}, {'type': 'binary', 'len':20, 'count':1}],
+                    'ctbPrefix':  'ctb',
+                    'ctbNum':     1,
+                    'rowsPerTbl': 10000,
+                    'batchNum':   10,
+                    'startTs':    1640966400000,  # 2022-01-01 00:00:00.000
+                    'pollDelay':  10,
+                    'showMsg':    1,
+                    'showRow':    1}
+
+        topicNameList = ['topic1', 'topic2', 'topic3']
+        expectRowsList = []
+        tmqCom.initConsumerTable()
+        tdCom.create_database(tdSql, paraDict["dbName"],paraDict["dropFlag"], vgroups=4,replica=1)
+        tdLog.info("create stb")
+        tdCom.create_stable(tdSql, dbname=paraDict["dbName"],stbname=paraDict["stbName"], column_elm_list=paraDict['colSchema'], tag_elm_list=paraDict['tagSchema'])
+        tdLog.info("create ctb")
+        tdCom.create_ctable(tdSql, dbname=paraDict["dbName"],stbname=paraDict["stbName"],tag_elm_list=paraDict['tagSchema'],count=paraDict["ctbNum"], default_ctbname_prefix=paraDict['ctbPrefix'])
+        tdLog.info("insert data")
+        tmqCom.insert_data(tdSql,paraDict["dbName"],paraDict["ctbPrefix"],paraDict["ctbNum"],paraDict["rowsPerTbl"],paraDict["batchNum"],paraDict["startTs"])
+        
+        tdLog.info("create topics from stb with filter")
+        # queryString = "select ts, log(c1), ceil(pow(c1,3)) from %s.%s where c1 %% 7 == 0" %(paraDict['dbName'], paraDict['stbName'])
+        # queryString = "select ts, c1, c2 from %s.%s" %(paraDict['dbName'], paraDict['stbName'])
+        queryString = "select * from %s.%s" %(paraDict['dbName'], paraDict['stbName'])
+        sqlString = "create topic %s as stable %s.%s" %(topicNameList[0], paraDict["dbName"],paraDict["stbName"])
+        # sqlString = "create topic %s as %s" %(topicNameList[0], queryString)
+        tdLog.info("create topic sql: %s"%sqlString)
+        tdSql.execute(sqlString)
+        # queryString = 'select * from %s.%s'%(paraDict["dbName"],paraDict["stbName"])
+        tdSql.query(queryString)        
+        expectRowsList.append(tdSql.getRows())
+
+        # init consume info, and start tmq_sim, then check consume result
+        tdLog.info("insert consume info to consume processor")
+        consumerId   = 0
+        expectrowcnt = paraDict["rowsPerTbl"] * paraDict["ctbNum"]
+        topicList    = topicNameList[0]
+        ifcheckdata  = 1
+        ifManualCommit = 1
+        keyList      = 'group.id:cgrp1, enable.auto.commit:false, auto.commit.interval.ms:6000, auto.offset.reset:earliest'
+        tmqCom.insertConsumerInfo(consumerId, expectrowcnt,topicList,keyList,ifcheckdata,ifManualCommit)
+
+        tdLog.info("start consume processor")
+        tmqCom.startTmqSimProcess(paraDict['pollDelay'],paraDict["dbName"],paraDict['showMsg'], paraDict['showRow'])
+
+        tdLog.info("wait the consume result") 
+        expectRows = 1
+        resultList = tmqCom.selectConsumeResult(expectRows)
+        
+        if expectRowsList[0] != resultList[0]:
+            tdLog.info("expect consume rows: %d, act consume rows: %d"%(expectRowsList[0], resultList[0]))
+            tdLog.exit("0 tmq consume rows error!")
+
+        self.checkFileContent(consumerId, queryString)     
+
+        # reinit consume info, and start tmq_sim, then check consume result
+        tmqCom.initConsumerTable()
+
+        queryString = "select ts, log(c1), cos(c1) from %s.%s where c1 > 3169" %(paraDict['dbName'], paraDict['stbName'])
+        sqlString = "create topic %s as %s" %(topicNameList[1], queryString)
+        tdLog.info("create topic sql: %s"%sqlString)
+        tdSql.execute(sqlString)
+        tdSql.query(queryString)        
+        expectRowsList.append(tdSql.getRows())
+
+        consumerId   = 1
+        topicList    = topicNameList[1]
+        tmqCom.insertConsumerInfo(consumerId, expectrowcnt,topicList,keyList,ifcheckdata,ifManualCommit)
+
+        tdLog.info("start consume processor")
+        tmqCom.startTmqSimProcess(paraDict['pollDelay'],paraDict["dbName"],paraDict['showMsg'], paraDict['showRow'])
+
+        tdLog.info("wait the consume result") 
+        expectRows = 1
+        resultList = tmqCom.selectConsumeResult(expectRows)
+        if expectRowsList[1] != resultList[0]:
+            tdLog.info("expect consume rows: %d, act consume rows: %d"%(expectRowsList[1], resultList[0]))
+            tdLog.exit("1 tmq consume rows error!")
+
+        self.checkFileContent(consumerId, queryString)
+
+        # reinit consume info, and start tmq_sim, then check consume result
+        tmqCom.initConsumerTable()
+
+        queryString = "select ts, log(c1), atan(c1) from %s.%s where ts >= %d" %(paraDict['dbName'], paraDict['stbName'], paraDict["startTs"]+6137)
+        sqlString = "create topic %s as %s" %(topicNameList[2], queryString)
+        tdLog.info("create topic sql: %s"%sqlString)
+        tdSql.execute(sqlString)
+        tdSql.query(queryString)        
+        expectRowsList.append(tdSql.getRows())                
+
+        consumerId   = 2
+        topicList    = topicNameList[2]
+        tmqCom.insertConsumerInfo(consumerId, expectrowcnt,topicList,keyList,ifcheckdata,ifManualCommit)
+
+        tdLog.info("start consume processor")
+        tmqCom.startTmqSimProcess(paraDict['pollDelay'],paraDict["dbName"],paraDict['showMsg'], paraDict['showRow'])
+
+        tdLog.info("wait the consume result") 
+        expectRows = 1
+        resultList = tmqCom.selectConsumeResult(expectRows)
+        # if expectRowsList[2] != resultList[0]:
+        #     tdLog.info("expect consume rows: %d, act consume rows: %d"%(expectRowsList[2], resultList[0]))
+        #     tdLog.exit("2 tmq consume rows error!")
+
+        # self.checkFileContent(consumerId, queryString)
+
+        time.sleep(10)        
+        for i in range(len(topicNameList)):
+            tdSql.query("drop topic %s"%topicNameList[i])
+
+        tdLog.printNoPrefix("======== test case 1 end ...... ")
+
+    def run(self):
+        tdSql.prepare()
+        self.tmqCase1()
+
+    def stop(self):
+        tdSql.close()
+        tdLog.success(f"{__file__} successfully executed")
+
+event = threading.Event()
+
+tdCases.addLinux(__file__, TDTestCase())
+tdCases.addWindows(__file__, TDTestCase())

tests/system-test/fulltest.sh

@@ -23,6 +23,7 @@ python3 ./test.py -f 1-insert/alter_stable.py
 python3 ./test.py -f 1-insert/alter_table.py
 python3 ./test.py -f 1-insert/insertWithMoreVgroup.py
 python3 ./test.py -f 1-insert/table_comment.py
+python3 ./test.py -f 1-insert/table_param_ttl.py
 python3 ./test.py -f 2-query/between.py
 python3 ./test.py -f 2-query/distinct.py
 python3 ./test.py -f 2-query/varchar.py
@@ -116,6 +117,10 @@ python3 ./test.py -f 2-query/function_null.py
 python3 ./test.py -f 6-cluster/5dnode1mnode.py 
 python3 ./test.py -f 6-cluster/5dnode2mnode.py 
 python3 ./test.py -f 6-cluster/5dnode3mnodeStop.py -N 5 -M 3
+python3 ./test.py -f  6-cluster/5dnode3mnodeSep1VnodeStopVnodeCreateDb.py  -N 5 -M 3
+# BUG python3 ./test.py -f  6-cluster/5dnode3mnodeSep1VnodeStopCreateDb.py  -N 5 -M 3
+# BUG python3 ./test.py -f  6-cluster/5dnode3mnodeSep1VnodeStopMnodeCreateDb.py  -N 5 -M 3
+
 # python3 ./test.py -f 6-cluster/5dnode3mnodeDrop.py -N 5
 # BUG python3 ./test.py -f 6-cluster/5dnode3mnodeStopInsert.py 
 
@@ -139,3 +144,4 @@ python3 ./test.py -f 7-tmq/tmqCheckData.py
 python3 ./test.py -f 7-tmq/tmqUdf.py
 #python3 ./test.py -f 7-tmq/tmq3mnodeSwitch.py -N 5
 python3 ./test.py -f 7-tmq/tmqConsumerGroup.py
+python3 ./test.py -f 7-tmq/tmqShow.py

tests/system-test/test-all.bat

@@ -2,7 +2,7 @@
 SETLOCAL EnableDelayedExpansion
 for /F "tokens=1,2 delims=#" %%a in ('"prompt #$H#$E# & echo on & for %%b in (1) do     rem"') do (  set "DEL=%%a")
 set /a a=0
-if %1 == full (
+if "%1" == "full" (
     echo Windows Taosd Full Test
     set /a exitNum=0
     del /Q /F failed.txt

tests/system-test/test.py

@@ -22,8 +22,6 @@ import json
 import platform
 import socket
 import threading
-from distutils.log import warn as printf
-from fabric2 import Connection
 sys.path.append("../pytest")
 from util.log import *
 from util.dnodes import *
@@ -187,9 +185,9 @@ if __name__ == "__main__":
 
     tdLog.info("Procedures for tdengine deployed in %s" % (host))
     if platform.system().lower() == 'windows':
+        fileName = fileName.replace("/", os.sep)
         if (masterIp == "" and not fileName[0:12] == "0-others\\udf"):
             threading.Thread(target=checkRunTimeError,daemon=True).start()
-        tdCases.logSql(logSql)
         tdLog.info("Procedures for testing self-deployment")
         tdDnodes.init(deployPath, masterIp)
         tdDnodes.setTestCluster(testCluster)
@@ -208,15 +206,43 @@ if __name__ == "__main__":
             uModule = importlib.import_module(moduleName)
             try:
                 ucase = uModule.TDTestCase()
-                if ((json.dumps(updateCfgDict) == '{}') and (ucase.updatecfgDict is not None)):
+                if ((json.dumps(updateCfgDict) == '{}') and hasattr(ucase, 'updatecfgDict')):
                     updateCfgDict = ucase.updatecfgDict
                     updateCfgDictStr = "-d %s"%base64.b64encode(json.dumps(updateCfgDict).encode()).decode()
             except Exception as r:
                 print(r)
         else:
             pass
+        if dnodeNums == 1 :
             tdDnodes.deploy(1,updateCfgDict)
             tdDnodes.start(1)
+            tdCases.logSql(logSql)
+        else :
+            tdLog.debug("create an cluster  with %s nodes and make %s dnode as independent mnode"%(dnodeNums,mnodeNums))
+            dnodeslist = cluster.configure_cluster(dnodeNums=dnodeNums,mnodeNums=mnodeNums)
+            tdDnodes = ClusterDnodes(dnodeslist)
+            tdDnodes.init(deployPath, masterIp)
+            tdDnodes.setTestCluster(testCluster)
+            tdDnodes.setValgrind(valgrind)
+            tdDnodes.stopAll()
+            for dnode in tdDnodes.dnodes:
+                tdDnodes.deploy(dnode.index,{})
+            for dnode in tdDnodes.dnodes:
+                tdDnodes.starttaosd(dnode.index)
+            tdCases.logSql(logSql)
+            conn = taos.connect(
+                host,
+                config=tdDnodes.getSimCfgPath())
+            print(tdDnodes.getSimCfgPath(),host)
+            cluster.create_dnode(conn)
+            try:
+                if cluster.check_dnode(conn) :
+                    print("check dnode ready")
+            except Exception as r:
+                print(r)
+        if ucase is not None and hasattr(ucase, 'noConn') and ucase.noConn == True:
+            conn = None
+        else:
             conn = taos.connect(
                 host="%s"%(host),
                 config=tdDnodes.sim.getCfgDir())
@@ -307,4 +333,5 @@ if __name__ == "__main__":
                     tdCases.runOneLinux(conn, sp[0] + "_" + "restart.py")
                 else:
                     tdLog.info("not need to query")
+    if conn is not None:
         conn.close()

tests/test/c/tmqSim.c

@@ -635,8 +635,9 @@ void loop_consume(SThreadInfo* pInfo) {
     }
   }
 
+  int32_t consumeDelay = g_stConfInfo.consumeDelay == -1 ? -1 : (g_stConfInfo.consumeDelay * 1000);
   while (running) {
-    TAOS_RES* tmqMsg = tmq_consumer_poll(pInfo->tmq, g_stConfInfo.consumeDelay * 1000);
+    TAOS_RES* tmqMsg = tmq_consumer_poll(pInfo->tmq, consumeDelay);
     if (tmqMsg) {
       if (0 != g_stConfInfo.showMsgFlag) {
         totalRows += msg_process(tmqMsg, pInfo, totalMsgs);

tools/shell/src/shellEngine.c

@@ -156,6 +156,7 @@ void shellRunSingleCommandImp(char *command) {
     }
 
     fname = sptr + 2;
+    while (*fname == ' ') fname++;
     *sptr = '\0';
   }
 
@@ -858,9 +859,7 @@ void shellGetGrantInfo() {
 
   int32_t code = taos_errno(tres);
   if (code != TSDB_CODE_SUCCESS) {
-    if (code == TSDB_CODE_OPS_NOT_SUPPORT) {
+    if (code != TSDB_CODE_OPS_NOT_SUPPORT && code != TSDB_CODE_MND_NO_RIGHTS) {
-      fprintf(stdout, "Server is Community Edition, %s\n\n", sinfo);
-    } else {
       fprintf(stderr, "Failed to check Server Edition, Reason:0x%04x:%s\n\n", code, taos_errstr(tres));
     }
     return;

tools/taos-tools

@@ -1 +1 @@
-Subproject commit 28a49b447f71c4f014ebbac858b7215b897d57fd
+Subproject commit a875a057d1225d85c6323b9edaccc2b1a9641987