Merge branch '3.0' into 3.0test/jcy

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()
@@ -289,7 +284,6 @@ 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 +292,8 @@ 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 +324,9 @@ 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 +336,6 @@ 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 +392,9 @@ 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)
-                            }
-                        }
                     }
                 }
             }

examples/c/stream_demo.c

@@ -99,8 +99,8 @@ int32_t create_stream() {
   /*const char* sql = "select sum(k) from tu1 interval(10m)";*/
   /*pRes = tmq_create_stream(pConn, "stream1", "out1", sql);*/
   pRes = taos_query(pConn,
-                    "create stream stream1 trigger at_once into outstb as select _wstartts, sum(k) from st1 "
+                    "create stream stream1 trigger at_once into outstb as select _wstartts, sum(k) from st1 partition "
-                    "partition by tbname interval(10s) ");
+                    "by tbname interval(10s) ");
   if (taos_errno(pRes) != 0) {
     printf("failed to create stream stream1, reason:%s\n", taos_errstr(pRes));
     return -1;

include/common/tcommon.h

@@ -47,11 +47,14 @@ typedef enum EStreamType {
   STREAM_GET_ALL,
   STREAM_DELETE,
   STREAM_RETRIEVE,
+  STREAM_PUSH_DATA,
 } EStreamType;
 
 typedef struct {
+  SArray*   pGroupList;
   SArray*   pTableList;
   SHashObj* map;  // speedup acquire the tableQueryInfo by table uid
+  bool      needSortTableByGroupId;
   void*     pTagCond;
   void*     pTagIndexCond;
   uint64_t  suid;

include/common/tmsg.h

@@ -851,7 +851,6 @@ typedef struct {
 int32_t tSerializeSServerVerRsp(void* buf, int32_t bufLen, SServerVerRsp* pRsp);
 int32_t tDeserializeSServerVerRsp(void* buf, int32_t bufLen, SServerVerRsp* pRsp);
 
-
 typedef struct SQueryNodeAddr {
   int32_t nodeId;  // vgId or qnodeId
   SEpSet  epSet;
@@ -878,7 +877,6 @@ int32_t tSerializeSDnodeListRsp(void* buf, int32_t bufLen, SDnodeListRsp* pRsp);
 int32_t tDeserializeSDnodeListRsp(void* buf, int32_t bufLen, SDnodeListRsp* pRsp);
 void    tFreeSDnodeListRsp(SDnodeListRsp* pRsp);
 
-
 typedef struct {
   SArray* pArray;  // Array of SUseDbRsp
 } SUseDbBatchRsp;
@@ -1258,7 +1256,6 @@ int32_t tDeserializeSShowVariablesRsp(void* buf, int32_t bufLen, SShowVariablesR
 
 void tFreeSShowVariablesRsp(SShowVariablesRsp* pRsp);
 
-
 /*
  * sql: show tables like '%a_%'
  * payload is the query condition, e.g., '%a_%'
@@ -1287,6 +1284,7 @@ void    tFreeSShowRsp(SShowRsp* pRsp);
 typedef struct {
   char    db[TSDB_DB_FNAME_LEN];
   char    tb[TSDB_TABLE_NAME_LEN];
+  char    user[TSDB_USER_LEN];
   int64_t showId;
 } SRetrieveTableReq;
 
@@ -1308,6 +1306,8 @@ typedef struct {
   int32_t compLen;
   int32_t numOfRows;
   int32_t numOfCols;
+  int64_t skey;
+  int64_t ekey;
   char    data[];
 } SRetrieveTableRsp;
 

include/libs/executor/executor.h

@@ -36,7 +36,7 @@ typedef struct SReadHandle {
   void*   vnode;
   void*   mnd;
   SMsgCb* pMsgCb;
-  int8_t  initTsdbReader;
+//  int8_t  initTsdbReader;
 } SReadHandle;
 
 enum {

include/libs/nodes/plannodes.h

@@ -75,6 +75,7 @@ typedef struct SScanLogicNode {
   double        filesFactor;
   SArray*       pSmaIndexes;
   SNodeList*    pPartTags;
+  bool          partSort;
 } SScanLogicNode;
 
 typedef struct SJoinLogicNode {

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);
 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,

include/os/os.h

@@ -52,6 +52,7 @@ extern "C" {
 #endif
 #else
 
+#include <malloc.h>
 #include <time.h>
 #ifndef TD_USE_WINSOCK
 #include <winsock2.h>

include/os/osMath.h

@@ -25,11 +25,10 @@ extern "C" {
 
 #define TSWAP(a, b)                              \
   do {                                           \
-    char *__tmp = taosMemoryMalloc(sizeof(a));   \
+    char *__tmp = alloca(sizeof(a));             \
     memcpy(__tmp, &(a), sizeof(a));              \
     memcpy(&(a), &(b), sizeof(a));               \
     memcpy(&(b), __tmp, sizeof(a));              \
-    taosMemoryFree(__tmp);                       \
   } while (0)
 
 #ifdef WINDOWS

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"
@@ -85,7 +85,8 @@ void closeTransporter(STscObj *pTscObj) {
 }
 
 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;
@@ -137,7 +138,8 @@ void destroyTscObj(void *pObj) {
     // TODO
     // 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);
 }

source/common/src/tglobal.c

@@ -963,6 +963,10 @@ int32_t taosSetCfg(SConfig *pCfg, char* name) {
         tsTelemPort = (uint16_t)cfgGetItem(pCfg, "telemetryPort")->i32;
       } else if (strcasecmp("transPullupInterval", name) == 0) {
         tsTransPullupInterval = cfgGetItem(pCfg, "transPullupInterval")->i32;
+      } else if (strcasecmp("ttlUnit", name) == 0) {
+        tsTtlUnit = cfgGetItem(pCfg, "ttlUnit")->i32;
+      } else if (strcasecmp("ttlPushInterval", name) == 0) {
+        tsTtlPushInterval = cfgGetItem(pCfg, "ttlPushInterval")->i32;
       } else if (strcasecmp("tmrDebugFlag", name) == 0) {
         tmrDebugFlag = cfgGetItem(pCfg, "tmrDebugFlag")->i32;
       } else if (strcasecmp("tsdbDebugFlag", name) == 0) {

source/common/src/tmsg.c

@@ -3017,6 +3017,7 @@ int32_t tSerializeSRetrieveTableReq(void *buf, int32_t bufLen, SRetrieveTableReq
   if (tEncodeI64(&encoder, pReq->showId) < 0) return -1;
   if (tEncodeCStr(&encoder, pReq->db) < 0) return -1;
   if (tEncodeCStr(&encoder, pReq->tb) < 0) return -1;
+  if (tEncodeCStr(&encoder, pReq->user) < 0) return -1;
   tEndEncode(&encoder);
 
   int32_t tlen = encoder.pos;
@@ -3032,6 +3033,8 @@ int32_t tDeserializeSRetrieveTableReq(void *buf, int32_t bufLen, SRetrieveTableR
   if (tDecodeI64(&decoder, &pReq->showId) < 0) return -1;
   if (tDecodeCStrTo(&decoder, pReq->db) < 0) return -1;
   if (tDecodeCStrTo(&decoder, pReq->tb) < 0) return -1;
+  if (tDecodeCStrTo(&decoder, pReq->user) < 0) return -1;
+
   tEndDecode(&decoder);
   tDecoderClear(&decoder);
   return 0;

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

@@ -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/mgmt/node_util/src/dmFile.c

@@ -133,10 +133,10 @@ TdFilePtr dmCheckRunning(const char *dataDir) {
     ret = taosLockFile(pFile);
     if (ret == 0) break;
     terrno = TAOS_SYSTEM_ERROR(errno);
-    taosMsleep(100);
+    taosMsleep(1000);
     retryTimes++;
     dError("failed to lock file:%s since %s, retryTimes:%d", filepath, terrstr(), retryTimes);
-  } while (retryTimes < 120);
+  } while (retryTimes < 12);
 
   if (ret < 0) {
     terrno = TAOS_SYSTEM_ERROR(errno);

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

@@ -13,8 +13,8 @@
  * along with this program. If not, see <http://www.gnu.org/licenses/>.
  */
 
-#ifndef _TD_MND_AUTH_H_
+#ifndef _TD_MND_PRIVILEGE_H
-#define _TD_MND_AUTH_H_
+#define _TD_MND_PRIVILEGE_H
 
 #include "mndInt.h"
 
@@ -23,13 +23,18 @@ extern "C" {
 #endif
 
 typedef enum {
-  MND_OPER_CREATE_USER = 1,
+  MND_OPER_CONNECT = 1,
+  MND_OPER_CREATE_ACCT,
+  MND_OPER_DROP_ACCT,
+  MND_OPER_ALTER_ACCT,
+  MND_OPER_CREATE_USER,
   MND_OPER_DROP_USER,
   MND_OPER_ALTER_USER,
   MND_OPER_CREATE_BNODE,
   MND_OPER_DROP_BNODE,
   MND_OPER_CREATE_DNODE,
   MND_OPER_DROP_DNODE,
+  MND_OPER_CONFIG_DNODE,
   MND_OPER_CREATE_MNODE,
   MND_OPER_DROP_MNODE,
   MND_OPER_CREATE_QNODE,
@@ -37,11 +42,14 @@ typedef enum {
   MND_OPER_CREATE_SNODE,
   MND_OPER_DROP_SNODE,
   MND_OPER_REDISTRIBUTE_VGROUP,
+  MND_OPER_MERGE_VGROUP,
   MND_OPER_SPLIT_VGROUP,
   MND_OPER_BALANCE_VGROUP,
   MND_OPER_CREATE_FUNC,
   MND_OPER_DROP_FUNC,
   MND_OPER_KILL_TRANS,
+  MND_OPER_KILL_CONN,
+  MND_OPER_KILL_QUERY,
   MND_OPER_CREATE_DB,
   MND_OPER_ALTER_DB,
   MND_OPER_DROP_DB,
@@ -51,16 +59,17 @@ typedef enum {
   MND_OPER_READ_DB,
 } EOperType;
 
-int32_t mndInitAuth(SMnode *pMnode);
+int32_t mndInitPrivilege(SMnode *pMnode);
-void    mndCleanupAuth(SMnode *pMnode);
+void    mndCleanupPrivilege(SMnode *pMnode);
 
-int32_t mndCheckOperAuth(SMnode *pMnode, const char *user, EOperType operType);
+int32_t mndCheckOperPrivilege(SMnode *pMnode, const char *user, EOperType operType);
-int32_t mndCheckDbAuth(SMnode *pMnode, const char *user, EOperType operType, SDbObj *pDb);
+int32_t mndCheckDbPrivilege(SMnode *pMnode, const char *user, EOperType operType, SDbObj *pDb);
-int32_t mndCheckShowAuth(SMnode *pMnode, const char *user, int32_t showType);
+int32_t mndCheckDbPrivilegeByName(SMnode *pMnode, const char *user, EOperType operType, const char *name);
-int32_t mndCheckAlterUserAuth(SUserObj *pOperUser, SUserObj *pUser, SAlterUserReq *pAlter);
+int32_t mndCheckShowPrivilege(SMnode *pMnode, const char *user, int32_t showType);
+int32_t mndCheckAlterUserPrivilege(SUserObj *pOperUser, SUserObj *pUser, SAlterUserReq *pAlter);
 
 #ifdef __cplusplus
 }
 #endif
 
-#endif /*_TD_MND_AUTH_H_*/
+#endif /*_TD_MND_PRIVILEGE_H*/

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

@@ -15,6 +15,7 @@
 
 #define _DEFAULT_SOURCE
 #include "mndAcct.h"
+#include "mndPrivilege.h"
 #include "mndShow.h"
 #include "mndTrans.h"
 
@@ -212,18 +213,30 @@ static int32_t mndAcctActionUpdate(SSdb *pSdb, SAcctObj *pOld, SAcctObj *pNew) {
 }
 
 static int32_t mndProcessCreateAcctReq(SRpcMsg *pReq) {
+  if (mndCheckOperPrivilege(pReq->info.node, pReq->info.conn.user, MND_OPER_CREATE_ACCT) != 0) {
+    return -1;
+  }
+
   terrno = TSDB_CODE_MSG_NOT_PROCESSED;
   mError("failed to process create acct request since %s", terrstr());
   return -1;
 }
 
 static int32_t mndProcessAlterAcctReq(SRpcMsg *pReq) {
+  if (mndCheckOperPrivilege(pReq->info.node, pReq->info.conn.user, MND_OPER_ALTER_ACCT) != 0) {
+    return -1;
+  }
+
   terrno = TSDB_CODE_MSG_NOT_PROCESSED;
   mError("failed to process create acct request since %s", terrstr());
   return -1;
 }
 
 static int32_t mndProcessDropAcctReq(SRpcMsg *pReq) {
+  if (mndCheckOperPrivilege(pReq->info.node, pReq->info.conn.user, MND_OPER_DROP_ACCT) != 0) {
+    return -1;
+  }
+
   terrno = TSDB_CODE_MSG_NOT_PROCESSED;
   mError("failed to process create acct request since %s", terrstr());
   return -1;

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

@@ -15,7 +15,7 @@
 
 #define _DEFAULT_SOURCE
 #include "mndBnode.h"
-#include "mndAuth.h"
+#include "mndPrivilege.h"
 #include "mndDnode.h"
 #include "mndShow.h"
 #include "mndTrans.h"
@@ -277,6 +277,9 @@ static int32_t mndProcessCreateBnodeReq(SRpcMsg *pReq) {
   }
 
   mDebug("bnode:%d, start to create", createReq.dnodeId);
+  if (mndCheckOperPrivilege(pMnode, pReq->info.conn.user, MND_OPER_CREATE_BNODE) != 0) {
+    goto _OVER;
+  }
 
   pObj = mndAcquireBnode(pMnode, createReq.dnodeId);
   if (pObj != NULL) {
@@ -292,10 +295,6 @@ static int32_t mndProcessCreateBnodeReq(SRpcMsg *pReq) {
     goto _OVER;
   }
 
-  if (mndCheckOperAuth(pMnode, pReq->info.conn.user, MND_OPER_CREATE_BNODE) != 0) {
-    goto _OVER;
-  }
-
   code = mndCreateBnode(pMnode, pReq, pDnode, &createReq);
   if (code == 0) code = TSDB_CODE_ACTION_IN_PROGRESS;
 
@@ -383,6 +382,9 @@ static int32_t mndProcessDropBnodeReq(SRpcMsg *pReq) {
   }
 
   mDebug("bnode:%d, start to drop", dropReq.dnodeId);
+  if (mndCheckOperPrivilege(pMnode, pReq->info.conn.user, MND_OPER_DROP_BNODE) != 0) {
+    goto _OVER;
+  }
 
   if (dropReq.dnodeId <= 0) {
     terrno = TSDB_CODE_INVALID_MSG;
@@ -394,10 +396,6 @@ static int32_t mndProcessDropBnodeReq(SRpcMsg *pReq) {
     goto _OVER;
   }
 
-  if (mndCheckOperAuth(pMnode, pReq->info.conn.user, MND_OPER_DROP_BNODE) != 0) {
-    goto _OVER;
-  }
-
   code = mndDropBnode(pMnode, pReq, pObj);
   if (code == 0) code = TSDB_CODE_ACTION_IN_PROGRESS;
 

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

@@ -15,7 +15,7 @@
 
 #define _DEFAULT_SOURCE
 #include "mndConsumer.h"
-#include "mndAuth.h"
+#include "mndPrivilege.h"
 #include "mndDb.h"
 #include "mndDnode.h"
 #include "mndMnode.h"
@@ -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

@@ -15,7 +15,7 @@
 
 #define _DEFAULT_SOURCE
 #include "mndDb.h"
-#include "mndAuth.h"
+#include "mndPrivilege.h"
 #include "mndDnode.h"
 #include "mndOffset.h"
 #include "mndShow.h"
@@ -506,6 +506,9 @@ static int32_t mndProcessCreateDbReq(SRpcMsg *pReq) {
   }
 
   mDebug("db:%s, start to create, vgroups:%d", createReq.db, createReq.numOfVgroups);
+  if (mndCheckDbPrivilege(pMnode, pReq->info.conn.user, MND_OPER_CREATE_DB, NULL) != 0) {
+    goto _OVER;
+  }
 
   pDb = mndAcquireDb(pMnode, createReq.db);
   if (pDb != NULL) {
@@ -526,10 +529,6 @@ static int32_t mndProcessCreateDbReq(SRpcMsg *pReq) {
     goto _OVER;
   }
 
-  if (mndCheckDbAuth(pMnode, pReq->info.conn.user, MND_OPER_CREATE_DB, NULL) != 0) {
-    goto _OVER;
-  }
-
   code = mndCreateDb(pMnode, pReq, &createReq, pUser);
   if (code == 0) code = TSDB_CODE_ACTION_IN_PROGRESS;
 
@@ -700,7 +699,7 @@ static int32_t mndProcessAlterDbReq(SRpcMsg *pReq) {
     goto _OVER;
   }
 
-  if (mndCheckDbAuth(pMnode, pReq->info.conn.user, MND_OPER_ALTER_DB, pDb) != 0) {
+  if (mndCheckDbPrivilege(pMnode, pReq->info.conn.user, MND_OPER_ALTER_DB, pDb) != 0) {
     goto _OVER;
   }
 
@@ -980,7 +979,7 @@ static int32_t mndProcessDropDbReq(SRpcMsg *pReq) {
     }
   }
 
-  if (mndCheckDbAuth(pMnode, pReq->info.conn.user, MND_OPER_DROP_DB, pDb) != 0) {
+  if (mndCheckDbPrivilege(pMnode, pReq->info.conn.user, MND_OPER_DROP_DB, pDb) != 0) {
     goto _OVER;
   }
 
@@ -1127,7 +1126,7 @@ static int32_t mndProcessUseDbReq(SRpcMsg *pReq) {
 
       mError("db:%s, failed to process use db req since %s", usedbReq.db, terrstr());
     } else {
-      if (mndCheckDbAuth(pMnode, pReq->info.conn.user, MND_OPER_USE_DB, pDb) != 0) {
+      if (mndCheckDbPrivilege(pMnode, pReq->info.conn.user, MND_OPER_USE_DB, pDb) != 0) {
         goto _OVER;
       }
 
@@ -1252,7 +1251,7 @@ static int32_t mndProcessCompactDbReq(SRpcMsg *pReq) {
     goto _OVER;
   }
 
-  if (mndCheckDbAuth(pMnode, pReq->info.conn.user, MND_OPER_COMPACT_DB, pDb) != 0) {
+  if (mndCheckDbPrivilege(pMnode, pReq->info.conn.user, MND_OPER_COMPACT_DB, pDb) != 0) {
     goto _OVER;
   }
 

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

@@ -15,7 +15,7 @@
 
 #define _DEFAULT_SOURCE
 #include "mndDnode.h"
-#include "mndAuth.h"
+#include "mndPrivilege.h"
 #include "mndMnode.h"
 #include "mndQnode.h"
 #include "mndShow.h"
@@ -609,7 +609,6 @@ _OVER:
   return code;
 }
 
-
 static int32_t mndProcessCreateDnodeReq(SRpcMsg *pReq) {
   SMnode         *pMnode = pReq->info.node;
   int32_t         code = -1;
@@ -622,6 +621,9 @@ static int32_t mndProcessCreateDnodeReq(SRpcMsg *pReq) {
   }
 
   mInfo("dnode:%s:%d, start to create", createReq.fqdn, createReq.port);
+  if (mndCheckOperPrivilege(pMnode, pReq->info.conn.user, MND_OPER_CREATE_DNODE) != 0) {
+    goto _OVER;
+  }
 
   if (createReq.fqdn[0] == 0 || createReq.port <= 0 || createReq.port > UINT16_MAX) {
     terrno = TSDB_CODE_MND_INVALID_DNODE_EP;
@@ -635,10 +637,6 @@ static int32_t mndProcessCreateDnodeReq(SRpcMsg *pReq) {
     goto _OVER;
   }
 
-  if (mndCheckOperAuth(pMnode, pReq->info.conn.user, MND_OPER_CREATE_DNODE) != 0) {
-    goto _OVER;
-  }
-
   code = mndCreateDnode(pMnode, pReq, &createReq);
   if (code == 0) code = TSDB_CODE_ACTION_IN_PROGRESS;
 
@@ -717,6 +715,9 @@ static int32_t mndProcessDropDnodeReq(SRpcMsg *pReq) {
   }
 
   mInfo("dnode:%d, start to drop, ep:%s:%d", dropReq.dnodeId, dropReq.fqdn, dropReq.port);
+  if (mndCheckOperPrivilege(pMnode, pReq->info.conn.user, MND_OPER_DROP_MNODE) != 0) {
+    goto _OVER;
+  }
 
   pDnode = mndAcquireDnode(pMnode, dropReq.dnodeId);
   if (pDnode == NULL) {
@@ -753,10 +754,6 @@ static int32_t mndProcessDropDnodeReq(SRpcMsg *pReq) {
     }
   }
 
-  if (mndCheckOperAuth(pMnode, pReq->info.conn.user, MND_OPER_DROP_MNODE) != 0) {
-    goto _OVER;
-  }
-
   code = mndDropDnode(pMnode, pReq, pDnode, pMObj, pQObj, pSObj, numOfVnodes);
   if (code == 0) code = TSDB_CODE_ACTION_IN_PROGRESS;
 
@@ -782,6 +779,10 @@ static int32_t mndProcessConfigDnodeReq(SRpcMsg *pReq) {
   }
 
   mInfo("dnode:%d, start to config, option:%s, value:%s", cfgReq.dnodeId, cfgReq.config, cfgReq.value);
+  if (mndCheckOperPrivilege(pMnode, pReq->info.conn.user, MND_OPER_CONFIG_DNODE) != 0) {
+    return -1;
+  }
+
   SDnodeObj *pDnode = mndAcquireDnode(pMnode, cfgReq.dnodeId);
   if (pDnode == NULL) {
     mError("dnode:%d, failed to config since %s ", cfgReq.dnodeId, terrstr());

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

@@ -15,7 +15,7 @@
 
 #define _DEFAULT_SOURCE
 #include "mndFunc.h"
-#include "mndAuth.h"
+#include "mndPrivilege.h"
 #include "mndShow.h"
 #include "mndSync.h"
 #include "mndTrans.h"
@@ -283,6 +283,9 @@ static int32_t mndProcessCreateFuncReq(SRpcMsg *pReq) {
   }
 
   mDebug("func:%s, start to create", createReq.name);
+  if (mndCheckOperPrivilege(pMnode, pReq->info.conn.user, MND_OPER_CREATE_FUNC) != 0) {
+    goto _OVER;
+  }
 
   pFunc = mndAcquireFunc(pMnode, createReq.name);
   if (pFunc != NULL) {
@@ -318,10 +321,6 @@ static int32_t mndProcessCreateFuncReq(SRpcMsg *pReq) {
     goto _OVER;
   }
 
-  if (mndCheckOperAuth(pMnode, pReq->info.conn.user, MND_OPER_CREATE_FUNC) != 0) {
-    goto _OVER;
-  }
-
   code = mndCreateFunc(pMnode, pReq, &createReq);
   if (code == 0) code = TSDB_CODE_ACTION_IN_PROGRESS;
 
@@ -347,6 +346,9 @@ static int32_t mndProcessDropFuncReq(SRpcMsg *pReq) {
   }
 
   mDebug("func:%s, start to drop", dropReq.name);
+  if (mndCheckOperPrivilege(pMnode, pReq->info.conn.user, MND_OPER_DROP_FUNC) != 0) {
+    goto _OVER;
+  }
 
   if (dropReq.name[0] == 0) {
     terrno = TSDB_CODE_MND_INVALID_FUNC_NAME;
@@ -365,10 +367,6 @@ static int32_t mndProcessDropFuncReq(SRpcMsg *pReq) {
     }
   }
 
-  if (mndCheckOperAuth(pMnode, pReq->info.conn.user, MND_OPER_DROP_FUNC) != 0) {
-    goto _OVER;
-  }
-
   code = mndDropFunc(pMnode, pReq, pFunc);
   if (code == 0) code = TSDB_CODE_ACTION_IN_PROGRESS;
 

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

@@ -15,7 +15,7 @@
 
 #define _DEFAULT_SOURCE
 #include "mndAcct.h"
-#include "mndAuth.h"
+#include "mndPrivilege.h"
 #include "mndBnode.h"
 #include "mndCluster.h"
 #include "mndConsumer.h"
@@ -100,7 +100,7 @@ static void *mndThreadFp(void *param) {
     taosMsleep(100);
     if (mndGetStop(pMnode)) break;
 
-    if (lastTime % (tsTransPullupInterval * 10) == 1) {
+    if (lastTime % (tsTtlPushInterval * 10) == 1) {
       mndTtlTimer(pMnode);
     }
 
@@ -239,7 +239,7 @@ static int32_t mndInitSteps(SMnode *pMnode) {
   if (mndAllocStep(pMnode, "mnode-dnode", mndInitDnode, mndCleanupDnode) != 0) return -1;
   if (mndAllocStep(pMnode, "mnode-user", mndInitUser, mndCleanupUser) != 0) return -1;
   if (mndAllocStep(pMnode, "mnode-grant", mndInitGrant, mndCleanupGrant) != 0) return -1;
-  if (mndAllocStep(pMnode, "mnode-auth", mndInitAuth, mndCleanupAuth) != 0) return -1;
+  if (mndAllocStep(pMnode, "mnode-privilege", mndInitPrivilege, mndCleanupPrivilege) != 0) return -1;
   if (mndAllocStep(pMnode, "mnode-acct", mndInitAcct, mndCleanupAcct) != 0) return -1;
   if (mndAllocStep(pMnode, "mnode-stream", mndInitStream, mndCleanupStream) != 0) return -1;
   if (mndAllocStep(pMnode, "mnode-topic", mndInitTopic, mndCleanupTopic) != 0) return -1;

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

@@ -15,7 +15,7 @@
 
 #define _DEFAULT_SOURCE
 #include "mndMnode.h"
-#include "mndAuth.h"
+#include "mndPrivilege.h"
 #include "mndDnode.h"
 #include "mndShow.h"
 #include "mndSync.h"
@@ -389,6 +389,9 @@ static int32_t mndProcessCreateMnodeReq(SRpcMsg *pReq) {
   }
 
   mDebug("mnode:%d, start to create", createReq.dnodeId);
+  if (mndCheckOperPrivilege(pMnode, pReq->info.conn.user, MND_OPER_CREATE_MNODE) != 0) {
+    goto _OVER;
+  }
 
   pObj = mndAcquireMnode(pMnode, createReq.dnodeId);
   if (pObj != NULL) {
@@ -414,10 +417,6 @@ static int32_t mndProcessCreateMnodeReq(SRpcMsg *pReq) {
     goto _OVER;
   }
 
-  if (mndCheckOperAuth(pMnode, pReq->info.conn.user, MND_OPER_CREATE_MNODE) != 0) {
-    goto _OVER;
-  }
-
   code = mndCreateMnode(pMnode, pReq, pDnode, &createReq);
   if (code == 0) code = TSDB_CODE_ACTION_IN_PROGRESS;
 
@@ -595,6 +594,9 @@ static int32_t mndProcessDropMnodeReq(SRpcMsg *pReq) {
   }
 
   mDebug("mnode:%d, start to drop", dropReq.dnodeId);
+  if (mndCheckOperPrivilege(pMnode, pReq->info.conn.user, MND_OPER_DROP_MNODE) != 0) {
+    goto _OVER;
+  }
 
   if (dropReq.dnodeId <= 0) {
     terrno = TSDB_CODE_INVALID_MSG;
@@ -621,10 +623,6 @@ static int32_t mndProcessDropMnodeReq(SRpcMsg *pReq) {
     goto _OVER;
   }
 
-  if (mndCheckOperAuth(pMnode, pReq->info.conn.user, MND_OPER_DROP_MNODE) != 0) {
-    goto _OVER;
-  }
-
   code = mndDropMnode(pMnode, pReq, pObj);
   if (code == 0) code = TSDB_CODE_ACTION_IN_PROGRESS;
 

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

@@ -15,7 +15,7 @@
 
 #define _DEFAULT_SOURCE
 #include "mndOffset.h"
-#include "mndAuth.h"
+#include "mndPrivilege.h"
 #include "mndDb.h"
 #include "mndDnode.h"
 #include "mndMnode.h"
@@ -36,13 +36,15 @@ static int32_t mndOffsetActionUpdate(SSdb *pSdb, SMqOffsetObj *pOffset, SMqOffse
 static int32_t mndProcessCommitOffsetReq(SRpcMsg *pReq);
 
 int32_t mndInitOffset(SMnode *pMnode) {
-  SSdbTable table = {.sdbType = SDB_OFFSET,
+  SSdbTable table = {
+      .sdbType = SDB_OFFSET,
       .keyType = SDB_KEY_BINARY,
       .encodeFp = (SdbEncodeFp)mndOffsetActionEncode,
       .decodeFp = (SdbDecodeFp)mndOffsetActionDecode,
       .insertFp = (SdbInsertFp)mndOffsetActionInsert,
       .updateFp = (SdbUpdateFp)mndOffsetActionUpdate,
-                     .deleteFp = (SdbDeleteFp)mndOffsetActionDelete};
+      .deleteFp = (SdbDeleteFp)mndOffsetActionDelete,
+  };
 
   mndSetMsgHandle(pMnode, TDMT_MND_MQ_COMMIT_OFFSET, mndProcessCommitOffsetReq);
 

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

@@ -14,66 +14,15 @@
  */
 
 #define _DEFAULT_SOURCE
-#include "mndAuth.h"
+#include "mndPrivilege.h"
 #include "mndUser.h"
+#include "mndDb.h"
 
-static int32_t mndProcessAuthReq(SRpcMsg *pReq);
+int32_t mndInitPrivilege(SMnode *pMnode) { return 0; }
 
-int32_t mndInitAuth(SMnode *pMnode) {
+void mndCleanupPrivilege(SMnode *pMnode) {}
-  mndSetMsgHandle(pMnode, TDMT_MND_AUTH, mndProcessAuthReq);
-  return 0;
-}
 
-void mndCleanupAuth(SMnode *pMnode) {}
+int32_t mndCheckOperPrivilege(SMnode *pMnode, const char *user, EOperType operType) {
-
-static int32_t mndRetriveAuth(SMnode *pMnode, SAuthRsp *pRsp) {
-  SUserObj *pUser = mndAcquireUser(pMnode, pRsp->user);
-  if (pUser == NULL) {
-    *pRsp->secret = 0;
-    mError("user:%s, failed to auth user since %s", pRsp->user, terrstr());
-    return -1;
-  }
-
-  pRsp->spi = 1;
-  pRsp->encrypt = 0;
-  *pRsp->ckey = 0;
-
-  memcpy(pRsp->secret, pUser->pass, TSDB_PASSWORD_LEN);
-  mndReleaseUser(pMnode, pUser);
-
-  mDebug("user:%s, auth info is returned", pRsp->user);
-  return 0;
-}
-
-static int32_t mndProcessAuthReq(SRpcMsg *pReq) {
-  SAuthReq authReq = {0};
-  if (tDeserializeSAuthReq(pReq->pCont, pReq->contLen, &authReq) != 0) {
-    terrno = TSDB_CODE_INVALID_MSG;
-    return -1;
-  }
-
-  SAuthReq authRsp = {0};
-  memcpy(authRsp.user, authReq.user, TSDB_USER_LEN);
-
-  int32_t code = mndRetriveAuth(pReq->info.node, &authRsp);
-  mTrace("user:%s, auth req received, spi:%d encrypt:%d ruser:%s", pReq->info.conn.user, authRsp.spi, authRsp.encrypt,
-         authRsp.user);
-
-  int32_t contLen = tSerializeSAuthReq(NULL, 0, &authRsp);
-  void   *pRsp = rpcMallocCont(contLen);
-  if (pRsp == NULL) {
-    terrno = TSDB_CODE_OUT_OF_MEMORY;
-    return -1;
-  }
-
-  tSerializeSAuthReq(pRsp, contLen, &authRsp);
-
-  pReq->info.rsp = pRsp;
-  pReq->info.rspLen = contLen;
-  return code;
-}
-
-int32_t mndCheckOperAuth(SMnode *pMnode, const char *user, EOperType operType) {
   int32_t   code = 0;
   SUserObj *pUser = mndAcquireUser(pMnode, user);
 
@@ -93,16 +42,29 @@ int32_t mndCheckOperAuth(SMnode *pMnode, const char *user, EOperType operType) {
     goto _OVER;
   }
 
+  switch (operType) {
+    case MND_OPER_CONNECT:
+    case MND_OPER_CREATE_FUNC:
+    case MND_OPER_DROP_FUNC:
+      break;
+    default:
       terrno = TSDB_CODE_MND_NO_RIGHTS;
       code = -1;
+  }
 
 _OVER:
   mndReleaseUser(pMnode, pUser);
   return code;
 }
 
-int32_t mndCheckAlterUserAuth(SUserObj *pOperUser, SUserObj *pUser, SAlterUserReq *pAlter) {
+int32_t mndCheckAlterUserPrivilege(SUserObj *pOperUser, SUserObj *pUser, SAlterUserReq *pAlter) {
+  if (pUser->superUser && pAlter->alterType != TSDB_ALTER_USER_PASSWD) {
+    terrno = TSDB_CODE_MND_NO_RIGHTS;
+    return -1;
+  }
+
   if (pOperUser->superUser) return 0;
+
   if (!pOperUser->enable) {
     terrno = TSDB_CODE_MND_USER_DISABLED;
     return -1;
@@ -118,7 +80,7 @@ int32_t mndCheckAlterUserAuth(SUserObj *pOperUser, SUserObj *pUser, SAlterUserRe
   return -1;
 }
 
-int32_t mndCheckShowAuth(SMnode *pMnode, const char *user, int32_t showType) {
+int32_t mndCheckShowPrivilege(SMnode *pMnode, const char *user, int32_t showType) {
   int32_t   code = 0;
   SUserObj *pUser = mndAcquireUser(pMnode, user);
 
@@ -151,7 +113,7 @@ _OVER:
   return code;
 }
 
-int32_t mndCheckDbAuth(SMnode *pMnode, const char *user, EOperType operType, SDbObj *pDb) {
+int32_t mndCheckDbPrivilege(SMnode *pMnode, const char *user, EOperType operType, SDbObj *pDb) {
   int32_t   code = 0;
   SUserObj *pUser = mndAcquireUser(pMnode, user);
 
@@ -172,15 +134,7 @@ int32_t mndCheckDbAuth(SMnode *pMnode, const char *user, EOperType operType, SDb
     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 (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;
   }
 
@@ -207,3 +161,12 @@ _OVER:
   mndReleaseUser(pMnode, pUser);
   return code;
 }
+
+int32_t mndCheckDbPrivilegeByName(SMnode *pMnode, const char *user, EOperType operType, const char *name) {
+  SDbObj *pDb = mndAcquireDb(pMnode, name);
+  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

@@ -15,6 +15,7 @@
 
 #define _DEFAULT_SOURCE
 #include "mndProfile.h"
+#include "mndPrivilege.h"
 #include "mndDb.h"
 #include "mndDnode.h"
 #include "mndMnode.h"
@@ -217,36 +218,41 @@ static int32_t mndProcessConnectReq(SRpcMsg *pReq) {
   SConnObj       *pConn = NULL;
   int32_t         code = -1;
   SConnectReq     connReq = {0};
-  char            ip[30] = {0};
+  char            ip[24] = {0};
   const STraceId *trace = &pReq->info.traceId;
 
   if (tDeserializeSConnectReq(pReq->pCont, pReq->contLen, &connReq) != 0) {
     terrno = TSDB_CODE_INVALID_MSG;
-    goto CONN_OVER;
+    goto _OVER;
   }
 
   taosIp2String(pReq->info.conn.clientIp, ip);
+  if (mndCheckOperPrivilege(pMnode, pReq->info.conn.user, MND_OPER_CONNECT) != 0) {
+    mGError("user:%s, failed to login from %s since %s", pReq->info.conn.user, ip, terrstr());
+    goto _OVER;
+  }
 
   pUser = mndAcquireUser(pMnode, pReq->info.conn.user);
   if (pUser == NULL) {
-    mGError("user:%s, failed to login while acquire user since %s", pReq->info.conn.user, terrstr());
+    mGError("user:%s, failed to login from %s while acquire user since %s", pReq->info.conn.user, ip, terrstr());
-    goto CONN_OVER;
+    goto _OVER;
   }
-  if (0 != strncmp(connReq.passwd, pUser->pass, TSDB_PASSWORD_LEN - 1)) {
+
-    mGError("user:%s, failed to auth while acquire user, input:%s", pReq->info.conn.user, connReq.passwd);
+  if (strncmp(connReq.passwd, pUser->pass, TSDB_PASSWORD_LEN - 1) != 0) {
+    mGError("user:%s, failed to login from %s since invalid pass, input:%s", pReq->info.conn.user, ip, connReq.passwd);
     code = TSDB_CODE_RPC_AUTH_FAILURE;
-    goto CONN_OVER;
+    goto _OVER;
   }
 
   if (connReq.db[0]) {
-    char db[TSDB_DB_FNAME_LEN];
+    char db[TSDB_DB_FNAME_LEN] = {0};
     snprintf(db, TSDB_DB_FNAME_LEN, "%d%s%s", pUser->acctId, TS_PATH_DELIMITER, connReq.db);
     pDb = mndAcquireDb(pMnode, db);
     if (pDb == NULL) {
       terrno = TSDB_CODE_MND_INVALID_DB;
       mGError("user:%s, failed to login from %s while use db:%s since %s", pReq->info.conn.user, ip, connReq.db,
               terrstr());
-      goto CONN_OVER;
+      goto _OVER;
     }
   }
 
@@ -254,7 +260,7 @@ static int32_t mndProcessConnectReq(SRpcMsg *pReq) {
                         pReq->info.conn.clientPort, connReq.pid, connReq.app, connReq.startTime);
   if (pConn == NULL) {
     mGError("user:%s, failed to login from %s while create connection since %s", pReq->info.conn.user, ip, terrstr());
-    goto CONN_OVER;
+    goto _OVER;
   }
 
   SConnectRsp connectRsp = {0};
@@ -271,9 +277,9 @@ static int32_t mndProcessConnectReq(SRpcMsg *pReq) {
   mndGetMnodeEpSet(pMnode, &connectRsp.epSet);
 
   int32_t contLen = tSerializeSConnectRsp(NULL, 0, &connectRsp);
-  if (contLen < 0) goto CONN_OVER;
+  if (contLen < 0) goto _OVER;
   void *pRsp = rpcMallocCont(contLen);
-  if (pRsp == NULL) goto CONN_OVER;
+  if (pRsp == NULL) goto _OVER;
   tSerializeSConnectRsp(pRsp, contLen, &connectRsp);
 
   pReq->info.rspLen = contLen;
@@ -283,7 +289,7 @@ static int32_t mndProcessConnectReq(SRpcMsg *pReq) {
 
   code = 0;
 
-CONN_OVER:
+_OVER:
 
   mndReleaseUser(pMnode, pUser);
   mndReleaseDb(pMnode, pDb);
@@ -645,15 +651,6 @@ static int32_t mndProcessKillQueryReq(SRpcMsg *pReq) {
   SMnode       *pMnode = pReq->info.node;
   SProfileMgmt *pMgmt = &pMnode->profileMgmt;
 
-  SUserObj *pUser = mndAcquireUser(pMnode, pReq->info.conn.user);
-  if (pUser == NULL) return 0;
-  if (!pUser->superUser) {
-    mndReleaseUser(pMnode, pUser);
-    terrno = TSDB_CODE_MND_NO_RIGHTS;
-    return -1;
-  }
-  mndReleaseUser(pMnode, pUser);
-
   SKillQueryReq killReq = {0};
   if (tDeserializeSKillQueryReq(pReq->pCont, pReq->contLen, &killReq) != 0) {
     terrno = TSDB_CODE_INVALID_MSG;
@@ -661,6 +658,10 @@ static int32_t mndProcessKillQueryReq(SRpcMsg *pReq) {
   }
 
   mInfo("kill query msg is received, queryId:%s", killReq.queryStrId);
+  if (mndCheckOperPrivilege(pMnode, pReq->info.conn.user, MND_OPER_KILL_QUERY) != 0) {
+    return -1;
+  }
+
   int32_t  connId = 0;
   uint64_t queryId = 0;
   char    *p = strchr(killReq.queryStrId, ':');
@@ -690,21 +691,16 @@ static int32_t mndProcessKillConnReq(SRpcMsg *pReq) {
   SMnode       *pMnode = pReq->info.node;
   SProfileMgmt *pMgmt = &pMnode->profileMgmt;
 
-  SUserObj *pUser = mndAcquireUser(pMnode, pReq->info.conn.user);
-  if (pUser == NULL) return 0;
-  if (!pUser->superUser) {
-    mndReleaseUser(pMnode, pUser);
-    terrno = TSDB_CODE_MND_NO_RIGHTS;
-    return -1;
-  }
-  mndReleaseUser(pMnode, pUser);
-
   SKillConnReq killReq = {0};
   if (tDeserializeSKillConnReq(pReq->pCont, pReq->contLen, &killReq) != 0) {
     terrno = TSDB_CODE_INVALID_MSG;
     return -1;
   }
 
+  if (mndCheckOperPrivilege(pMnode, pReq->info.conn.user, MND_OPER_KILL_CONN) != 0) {
+    return -1;
+  }
+
   SConnObj *pConn = taosCacheAcquireByKey(pMgmt->connCache, &killReq.connId, sizeof(uint32_t));
   if (pConn == NULL) {
     mError("connId:%u, failed to kill connection, conn not exist", killReq.connId);
@@ -739,7 +735,6 @@ _over:
   return code;
 }
 
-
 static int32_t mndRetrieveConns(SRpcMsg *pReq, SShowObj *pShow, SSDataBlock *pBlock, int32_t rows) {
   SMnode   *pMnode = pReq->info.node;
   SSdb     *pSdb = pMnode->pSdb;

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

@@ -15,7 +15,7 @@
 
 #define _DEFAULT_SOURCE
 #include "mndQnode.h"
-#include "mndAuth.h"
+#include "mndPrivilege.h"
 #include "mndDnode.h"
 #include "mndShow.h"
 #include "mndTrans.h"
@@ -279,6 +279,9 @@ static int32_t mndProcessCreateQnodeReq(SRpcMsg *pReq) {
   }
 
   mDebug("qnode:%d, start to create", createReq.dnodeId);
+  if (mndCheckOperPrivilege(pMnode, pReq->info.conn.user, MND_OPER_CREATE_QNODE) != 0) {
+    goto _OVER;
+  }
 
   pObj = mndAcquireQnode(pMnode, createReq.dnodeId);
   if (pObj != NULL) {
@@ -294,10 +297,6 @@ static int32_t mndProcessCreateQnodeReq(SRpcMsg *pReq) {
     goto _OVER;
   }
 
-  if (mndCheckOperAuth(pMnode, pReq->info.conn.user, MND_OPER_CREATE_QNODE) != 0) {
-    goto _OVER;
-  }
-
   code = mndCreateQnode(pMnode, pReq, pDnode, &createReq);
   if (code == 0) code = TSDB_CODE_ACTION_IN_PROGRESS;
 
@@ -391,6 +390,9 @@ static int32_t mndProcessDropQnodeReq(SRpcMsg *pReq) {
   }
 
   mDebug("qnode:%d, start to drop", dropReq.dnodeId);
+  if (mndCheckOperPrivilege(pMnode, pReq->info.conn.user, MND_OPER_DROP_QNODE) != 0) {
+    goto _OVER;
+  }
 
   if (dropReq.dnodeId <= 0) {
     terrno = TSDB_CODE_INVALID_MSG;
@@ -402,10 +404,6 @@ static int32_t mndProcessDropQnodeReq(SRpcMsg *pReq) {
     goto _OVER;
   }
 
-  if (mndCheckOperAuth(pMnode, pReq->info.conn.user, MND_OPER_DROP_QNODE) != 0) {
-    goto _OVER;
-  }
-
   code = mndDropQnode(pMnode, pReq, pObj);
   if (code == 0) code = TSDB_CODE_ACTION_IN_PROGRESS;
 
@@ -457,7 +455,6 @@ int32_t mndCreateQnodeList(SMnode       *pMnode, SArray** pList, int32_t limit)
   return TSDB_CODE_SUCCESS;
 }
 
-
 static int32_t mndProcessQnodeListReq(SRpcMsg *pReq) {
   int32_t       code = -1;
   SMnode       *pMnode = pReq->info.node;

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

@@ -16,7 +16,7 @@
 #define _DEFAULT_SOURCE
 #include "mndShow.h"
 #include "systable.h"
-#include "mndAuth.h"
+#include "mndPrivilege.h"
 
 #define SHOW_STEP_SIZE 100
 
@@ -122,6 +122,7 @@ static SShowObj *mndCreateShowObj(SMnode *pMnode, SRetrieveTableReq *pReq) {
   int32_t size = sizeof(SShowObj);
 
   SShowObj showObj = {0};
+
   showObj.id = showId;
   showObj.pMnode = pMnode;
   showObj.type = convertToRetrieveType(pReq->tb, tListLen(pReq->tb));
@@ -231,7 +232,7 @@ static int32_t mndProcessRetrieveSysTableReq(SRpcMsg *pReq) {
 
   mDebug("show:0x%" PRIx64 ", start retrieve data, type:%d", pShow->id, pShow->type);
 
-  // if (mndCheckShowAuth(pMnode, pReq->info.conn.user, pShow->type) != 0) return -1;
+  // if (mndCheckShowPrivilege(pMnode, pReq->info.conn.user, pShow->type) != 0) return -1;
 
   int32_t      numOfCols = pShow->pMeta->numOfColumns;
   SSDataBlock *pBlock = taosMemoryCalloc(1, sizeof(SSDataBlock));

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

@@ -15,7 +15,7 @@
 
 #define _DEFAULT_SOURCE
 #include "mndSma.h"
-#include "mndAuth.h"
+#include "mndPrivilege.h"
 #include "mndDb.h"
 #include "mndDnode.h"
 #include "mndInfoSchema.h"
@@ -713,7 +713,7 @@ static int32_t mndProcessCreateSmaReq(SRpcMsg *pReq) {
     goto _OVER;
   }
 
-  if (mndCheckDbAuth(pMnode, pReq->info.conn.user, MND_OPER_WRITE_DB, pDb) != 0) {
+  if (mndCheckDbPrivilege(pMnode, pReq->info.conn.user, MND_OPER_WRITE_DB, pDb) != 0) {
     goto _OVER;
   }
 
@@ -974,7 +974,7 @@ static int32_t mndProcessDropSmaReq(SRpcMsg *pReq) {
     goto _OVER;
   }
 
-  if (mndCheckDbAuth(pMnode, pReq->info.conn.user, MND_OPER_WRITE_DB, pDb) != 0) {
+  if (mndCheckDbPrivilege(pMnode, pReq->info.conn.user, MND_OPER_WRITE_DB, pDb) != 0) {
     goto _OVER;
   }
 

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

@@ -15,7 +15,7 @@
 
 #define _DEFAULT_SOURCE
 #include "mndSnode.h"
-#include "mndAuth.h"
+#include "mndPrivilege.h"
 #include "mndDnode.h"
 #include "mndShow.h"
 #include "mndTrans.h"
@@ -285,6 +285,9 @@ static int32_t mndProcessCreateSnodeReq(SRpcMsg *pReq) {
   }
 
   mDebug("snode:%d, start to create", createReq.dnodeId);
+  if (mndCheckOperPrivilege(pMnode, pReq->info.conn.user, MND_OPER_CREATE_SNODE) != 0) {
+    goto _OVER;
+  }
 
   pObj = mndAcquireSnode(pMnode, createReq.dnodeId);
   if (pObj != NULL) {
@@ -300,10 +303,6 @@ static int32_t mndProcessCreateSnodeReq(SRpcMsg *pReq) {
     goto _OVER;
   }
 
-  if (mndCheckOperAuth(pMnode, pReq->info.conn.user, MND_OPER_CREATE_SNODE) != 0) {
-    goto _OVER;
-  }
-
   code = mndCreateSnode(pMnode, pReq, pDnode, &createReq);
   if (code == 0) code = TSDB_CODE_ACTION_IN_PROGRESS;
 
@@ -398,6 +397,9 @@ static int32_t mndProcessDropSnodeReq(SRpcMsg *pReq) {
   }
 
   mDebug("snode:%d, start to drop", dropReq.dnodeId);
+  if (mndCheckOperPrivilege(pMnode, pReq->info.conn.user, MND_OPER_DROP_SNODE) != 0) {
+    goto _OVER;
+  }
 
   if (dropReq.dnodeId <= 0) {
     terrno = TSDB_CODE_INVALID_MSG;
@@ -409,10 +411,6 @@ static int32_t mndProcessDropSnodeReq(SRpcMsg *pReq) {
     goto _OVER;
   }
 
-  if (mndCheckOperAuth(pMnode, pReq->info.conn.user, MND_OPER_DROP_SNODE) != 0) {
-    goto _OVER;
-  }
-
   // check deletable
   code = mndDropSnode(pMnode, pReq, pObj);
   if (code == 0) code = TSDB_CODE_ACTION_IN_PROGRESS;

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

@@ -15,7 +15,7 @@
 
 #define _DEFAULT_SOURCE
 #include "mndStb.h"
-#include "mndAuth.h"
+#include "mndPrivilege.h"
 #include "mndDb.h"
 #include "mndDnode.h"
 #include "mndInfoSchema.h"
@@ -876,7 +876,7 @@ static int32_t mndProcessCreateStbReq(SRpcMsg *pReq) {
     goto _OVER;
   }
 
-  if (mndCheckDbAuth(pMnode, pReq->info.conn.user, MND_OPER_WRITE_DB, pDb) != 0) {
+  if (mndCheckDbPrivilege(pMnode, pReq->info.conn.user, MND_OPER_WRITE_DB, pDb) != 0) {
     goto _OVER;
   }
 
@@ -1607,7 +1607,7 @@ static int32_t mndProcessAlterStbReq(SRpcMsg *pReq) {
     goto _OVER;
   }
 
-  if (mndCheckDbAuth(pMnode, pReq->info.conn.user, MND_OPER_WRITE_DB, pDb) != 0) {
+  if (mndCheckDbPrivilege(pMnode, pReq->info.conn.user, MND_OPER_WRITE_DB, pDb) != 0) {
     goto _OVER;
   }
 
@@ -1737,7 +1737,7 @@ static int32_t mndProcessDropStbReq(SRpcMsg *pReq) {
     goto _OVER;
   }
 
-  if (mndCheckDbAuth(pMnode, pReq->info.conn.user, MND_OPER_WRITE_DB, pDb) != 0) {
+  if (mndCheckDbPrivilege(pMnode, pReq->info.conn.user, MND_OPER_WRITE_DB, pDb) != 0) {
     goto _OVER;
   }
 

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

@@ -14,7 +14,7 @@
  */
 
 #include "mndStream.h"
-#include "mndAuth.h"
+#include "mndPrivilege.h"
 #include "mndDb.h"
 #include "mndDnode.h"
 #include "mndMnode.h"
@@ -437,10 +437,6 @@ static int32_t mndCreateStbForStream(SMnode *pMnode, STrans *pTrans, const SStre
     goto _OVER;
   }
 
-  if (mndCheckDbAuth(pMnode, user, MND_OPER_WRITE_DB, pDb) != 0) {
-    goto _OVER;
-  }
-
   int32_t numOfStbs = -1;
   if (mndGetNumOfStbs(pMnode, pDb->name, &numOfStbs) != 0) {
     goto _OVER;
@@ -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 (mndCheckDbAuth(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());
@@ -634,20 +627,16 @@ static int32_t mndProcessDropStreamReq(SRpcMsg *pReq) {
     if (dropReq.igNotExists) {
       mDebug("stream:%s, not exist, ignore not exist is set", dropReq.name);
       sdbRelease(pMnode->pSdb, pStream);
-      return -1;
+      return 0;
     } else {
       terrno = TSDB_CODE_MND_STREAM_NOT_EXIST;
       return -1;
     }
   }
 
-#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 "mndAuth.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"
@@ -480,7 +480,7 @@ static int32_t mndProcessCreateTopicReq(SRpcMsg *pReq) {
     goto _OVER;
   }
 
-  if (mndCheckDbAuth(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 +571,10 @@ 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/mnode/impl/src/mndTrans.c

@@ -15,7 +15,7 @@
 
 #define _DEFAULT_SOURCE
 #include "mndTrans.h"
-#include "mndAuth.h"
+#include "mndPrivilege.h"
 #include "mndConsumer.h"
 #include "mndDb.h"
 #include "mndShow.h"
@@ -1384,8 +1384,7 @@ static int32_t mndProcessKillTransReq(SRpcMsg *pReq) {
   }
 
   mInfo("trans:%d, start to kill", killReq.transId);
-
+  if (mndCheckOperPrivilege(pMnode, pReq->info.conn.user, MND_OPER_KILL_TRANS) != 0) {
-  if (mndCheckOperAuth(pMnode, pReq->info.conn.user, MND_OPER_KILL_TRANS) != 0) {
     goto _OVER;
   }
 

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

@@ -15,7 +15,7 @@
 
 #define _DEFAULT_SOURCE
 #include "mndUser.h"
-#include "mndAuth.h"
+#include "mndPrivilege.h"
 #include "mndDb.h"
 #include "mndShow.h"
 #include "mndTrans.h"
@@ -295,7 +295,7 @@ static int32_t mndCreateUser(SMnode *pMnode, char *acct, SCreateUserReq *pCreate
   tstrncpy(userObj.acct, acct, TSDB_USER_LEN);
   userObj.createdTime = taosGetTimestampMs();
   userObj.updateTime = userObj.createdTime;
-  userObj.superUser = pCreate->superUser;
+  userObj.superUser = 0;  // pCreate->superUser;
   userObj.sysInfo = pCreate->sysInfo;
   userObj.enable = pCreate->enable;
 
@@ -337,6 +337,9 @@ static int32_t mndProcessCreateUserReq(SRpcMsg *pReq) {
   }
 
   mDebug("user:%s, start to create", createReq.user);
+  if (mndCheckOperPrivilege(pMnode, pReq->info.conn.user, MND_OPER_CREATE_USER) != 0) {
+    goto _OVER;
+  }
 
   if (createReq.user[0] == 0) {
     terrno = TSDB_CODE_MND_INVALID_USER_FORMAT;
@@ -360,10 +363,6 @@ static int32_t mndProcessCreateUserReq(SRpcMsg *pReq) {
     goto _OVER;
   }
 
-  if (mndCheckOperAuth(pMnode, pReq->info.conn.user, MND_OPER_CREATE_USER) != 0) {
-    goto _OVER;
-  }
-
   code = mndCreateUser(pMnode, pOperUser->acct, &createReq, pReq);
   if (code == 0) code = TSDB_CODE_ACTION_IN_PROGRESS;
 
@@ -466,7 +465,7 @@ static int32_t mndProcessAlterUserReq(SRpcMsg *pReq) {
     goto _OVER;
   }
 
-  if (mndCheckAlterUserAuth(pOperUser, pUser, &alterReq) != 0) {
+  if (mndCheckAlterUserPrivilege(pOperUser, pUser, &alterReq) != 0) {
     goto _OVER;
   }
 
@@ -631,6 +630,9 @@ static int32_t mndProcessDropUserReq(SRpcMsg *pReq) {
   }
 
   mDebug("user:%s, start to drop", dropReq.user);
+  if (mndCheckOperPrivilege(pMnode, pReq->info.conn.user, MND_OPER_DROP_USER) != 0) {
+    goto _OVER;
+  }
 
   if (dropReq.user[0] == 0) {
     terrno = TSDB_CODE_MND_INVALID_USER_FORMAT;
@@ -643,10 +645,6 @@ static int32_t mndProcessDropUserReq(SRpcMsg *pReq) {
     goto _OVER;
   }
 
-  if (mndCheckOperAuth(pMnode, pReq->info.conn.user, MND_OPER_DROP_USER) != 0) {
-    goto _OVER;
-  }
-
   code = mndDropUser(pMnode, pReq, pUser);
   if (code == 0) code = TSDB_CODE_ACTION_IN_PROGRESS;
 

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

@@ -15,7 +15,7 @@
 
 #define _DEFAULT_SOURCE
 #include "mndVgroup.h"
-#include "mndAuth.h"
+#include "mndPrivilege.h"
 #include "mndDb.h"
 #include "mndDnode.h"
 #include "mndMnode.h"
@@ -1212,8 +1212,9 @@ static int32_t mndProcessRedistributeVgroupMsg(SRpcMsg *pReq) {
   }
 
   mInfo("vgId:%d, start to redistribute vgroup to dnode %d:%d:%d", req.vgId, req.dnodeId1, req.dnodeId2, req.dnodeId3);
-
+  if (mndCheckOperPrivilege(pMnode, pReq->info.conn.user, MND_OPER_REDISTRIBUTE_VGROUP) != 0) {
-  if (mndCheckOperAuth(pMnode, pReq->info.conn.user, MND_OPER_REDISTRIBUTE_VGROUP) != 0) goto _OVER;
+    goto _OVER;
+  }
 
   pVgroup = mndAcquireVgroup(pMnode, req.vgId);
   if (pVgroup == NULL) goto _OVER;
@@ -1506,6 +1507,9 @@ static int32_t mndProcessSplitVgroupMsg(SRpcMsg *pReq) {
   SDbObj *pDb = NULL;
 
   mDebug("vgId:%d, start to split", vgId);
+  if (mndCheckOperPrivilege(pMnode, pReq->info.conn.user, MND_OPER_SPLIT_VGROUP) != 0) {
+    goto _OVER;
+  }
 
   pVgroup = mndAcquireVgroup(pMnode, vgId);
   if (pVgroup == NULL) goto _OVER;
@@ -1513,8 +1517,6 @@ static int32_t mndProcessSplitVgroupMsg(SRpcMsg *pReq) {
   pDb = mndAcquireDb(pMnode, pVgroup->dbName);
   if (pDb == NULL) goto _OVER;
 
-  if (mndCheckOperAuth(pMnode, pReq->info.conn.user, MND_OPER_SPLIT_VGROUP) != 0) goto _OVER;
-
   code = mndSplitVgroup(pMnode, pReq, pDb, pVgroup);
   if (code == 0) code = TSDB_CODE_ACTION_IN_PROGRESS;
 
@@ -1655,8 +1657,9 @@ static int32_t mndProcessBalanceVgroupMsg(SRpcMsg *pReq) {
   }
 
   mInfo("start to balance vgroup");
-
+  if (mndCheckOperPrivilege(pMnode, pReq->info.conn.user, MND_OPER_BALANCE_VGROUP) != 0) {
-  if (mndCheckOperAuth(pMnode, pReq->info.conn.user, MND_OPER_BALANCE_VGROUP) != 0) goto _OVER;
+    goto _OVER;
+  }
 
   while (1) {
     SDnodeObj *pDnode = NULL;

source/dnode/mnode/impl/test/user/user.cpp

@@ -33,6 +33,8 @@ TEST_F(MndTestUser, 01_Show_User) {
 TEST_F(MndTestUser, 02_Create_User) {
   {
     SCreateUserReq createReq = {0};
+    createReq.enable = 1;
+    createReq.sysInfo = 1;
     strcpy(createReq.user, "");
     strcpy(createReq.pass, "p1");
 
@@ -47,6 +49,8 @@ TEST_F(MndTestUser, 02_Create_User) {
 
   {
     SCreateUserReq createReq = {0};
+    createReq.enable = 1;
+    createReq.sysInfo = 1;
     strcpy(createReq.user, "u1");
     strcpy(createReq.pass, "");
 
@@ -61,6 +65,8 @@ TEST_F(MndTestUser, 02_Create_User) {
 
   {
     SCreateUserReq createReq = {0};
+    createReq.enable = 1;
+    createReq.sysInfo = 1;
     strcpy(createReq.user, "root");
     strcpy(createReq.pass, "1");
 
@@ -75,6 +81,8 @@ TEST_F(MndTestUser, 02_Create_User) {
 
   {
     SCreateUserReq createReq = {0};
+    createReq.enable = 1;
+    createReq.sysInfo = 1;
     strcpy(createReq.user, "u1");
     strcpy(createReq.pass, "p1");
 
@@ -108,9 +116,11 @@ TEST_F(MndTestUser, 02_Create_User) {
 
   {
     SCreateUserReq createReq = {0};
+    createReq.enable = 1;
+    createReq.sysInfo = 1;
     strcpy(createReq.user, "u2");
     strcpy(createReq.pass, "p1");
-    createReq.superUser = 1;
+    createReq.superUser = 0;
 
     int32_t contLen = tSerializeSCreateUserReq(NULL, 0, &createReq);
     void*   pReq = rpcMallocCont(contLen);
@@ -144,9 +154,11 @@ TEST_F(MndTestUser, 02_Create_User) {
 TEST_F(MndTestUser, 03_Alter_User) {
   {
     SCreateUserReq createReq = {0};
+    createReq.enable = 1;
+    createReq.sysInfo = 1;
     strcpy(createReq.user, "u3");
     strcpy(createReq.pass, "p1");
-    createReq.superUser = 1;
+    createReq.superUser = 0;
 
     int32_t contLen = tSerializeSCreateUserReq(NULL, 0, &createReq);
     void*   pReq = rpcMallocCont(contLen);
@@ -225,7 +237,7 @@ TEST_F(MndTestUser, 03_Alter_User) {
     alterReq.alterType = TSDB_ALTER_USER_SUPERUSER;
     strcpy(alterReq.user, "u3");
     strcpy(alterReq.pass, "1");
-    alterReq.superUser = 1;
+    alterReq.superUser = 0;
 
     int32_t contLen = tSerializeSAlterUserReq(NULL, 0, &alterReq);
     void*   pReq = rpcMallocCont(contLen);
@@ -361,7 +373,7 @@ TEST_F(MndTestUser, 03_Alter_User) {
     SGetUserAuthRsp authRsp = {0};
     tDeserializeSGetUserAuthRsp(pRsp->pCont, pRsp->contLen, &authRsp);
     EXPECT_STREQ(authRsp.user, "u3");
-    EXPECT_EQ(authRsp.superAuth, 1);
+    EXPECT_EQ(authRsp.superAuth, 0);
     int32_t numOfReadDbs = taosHashGetSize(authRsp.readDbs);
     int32_t numOfWriteDbs = taosHashGetSize(authRsp.writeDbs);
     EXPECT_EQ(numOfReadDbs, 1);
@@ -436,6 +448,8 @@ TEST_F(MndTestUser, 05_Drop_User) {
 
   {
     SCreateUserReq createReq = {0};
+    createReq.enable = 1;
+    createReq.sysInfo = 1;
     strcpy(createReq.user, "u1");
     strcpy(createReq.pass, "p1");
 
@@ -468,6 +482,8 @@ TEST_F(MndTestUser, 05_Drop_User) {
 TEST_F(MndTestUser, 06_Create_Drop_Alter_User) {
   {
     SCreateUserReq createReq = {0};
+    createReq.enable = 1;
+    createReq.sysInfo = 1;
     strcpy(createReq.user, "u1");
     strcpy(createReq.pass, "p1");
 
@@ -482,6 +498,8 @@ TEST_F(MndTestUser, 06_Create_Drop_Alter_User) {
 
   {
     SCreateUserReq createReq = {0};
+    createReq.enable = 1;
+    createReq.sysInfo = 1;
     strcpy(createReq.user, "u2");
     strcpy(createReq.pass, "p2");
 

source/dnode/vnode/inc/vnode.h

@@ -116,7 +116,8 @@ typedef void *tsdbReaderT;
 #define BLOCK_LOAD_TABLE_SEQ_ORDER  2
 #define BLOCK_LOAD_TABLE_RR_ORDER   3
 
-tsdbReaderT  tsdbReaderOpen(SVnode *pVnode, SQueryTableDataCond *pCond, STableListInfo *tableInfoGroup, uint64_t qId,
+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,
                                 void *pMemRef);
@@ -195,7 +196,6 @@ struct SVnodeCfg {
 typedef struct {
   TSKEY    lastKey;
   uint64_t uid;
-  uint64_t groupId;
 } STableKeyInfo;
 
 struct SMetaEntry {

source/dnode/vnode/src/inc/vnodeInt.h

@@ -121,7 +121,7 @@ int         tsdbInsertData(STsdb* pTsdb, int64_t version, SSubmitReq* pMsg, SSub
 int32_t     tsdbInsertTableData(STsdb* pTsdb, int64_t version, SSubmitMsgIter* pMsgIter, SSubmitBlk* pBlock,
                                 SSubmitBlkRsp* pRsp);
 int32_t     tsdbDeleteTableData(STsdb* pTsdb, int64_t version, tb_uid_t suid, tb_uid_t uid, TSKEY sKey, TSKEY eKey);
-tsdbReaderT tsdbReaderOpen(SVnode* pVnode, SQueryTableDataCond* pCond, STableListInfo* tableList, uint64_t qId,
+tsdbReaderT tsdbReaderOpen(SVnode* pVnode, SQueryTableDataCond* pCond, SArray* tableList, uint64_t qId,
                            uint64_t taskId);
 tsdbReaderT tsdbQueryCacheLastT(STsdb* tsdb, SQueryTableDataCond* pCond, STableListInfo* tableList, uint64_t qId,
                                 void* pMemRef);

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

@@ -381,13 +381,14 @@ int metaTtlDropTable(SMeta *pMeta, int64_t ttl, SArray *tbUids) {
   for (int i = 0; i < taosArrayGetSize(tbUids); ++i) {
     tb_uid_t *uid = (tb_uid_t *)taosArrayGet(tbUids, i);
     metaDropTableByUid(pMeta, *uid, NULL);
+    metaDebug("ttl drop table:%"PRId64, *uid);
   }
   metaULock(pMeta);
   return 0;
 }
 
 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;
@@ -443,7 +444,6 @@ static int metaDropTableByUid(SMeta *pMeta, tb_uid_t uid, int *type) {
     // drop schema.db (todo)
   }
 
-  metaError("ttl drop table:%s", e.name);
   tDecoderClear(&dc);
   tdbFree(pData);
 
@@ -976,7 +976,9 @@ static int metaUpdateTagIdx(SMeta *pMeta, const SMetaEntry *pCtbEntry) {
   SDecoder       dc = {0};
 
   // get super table
-  tdbTbGet(pMeta->pUidIdx, &pCtbEntry->ctbEntry.suid, sizeof(tb_uid_t), &pData, &nData);
+  if(tdbTbGet(pMeta->pUidIdx, &pCtbEntry->ctbEntry.suid, sizeof(tb_uid_t), &pData, &nData) != 0){
+    return -1;
+  }
   tbDbKey.uid = pCtbEntry->ctbEntry.suid;
   tbDbKey.version = *(int64_t *)pData;
   tdbTbGet(pMeta->pTbDb, &tbDbKey, sizeof(tbDbKey), &pData, &nData);

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

@@ -403,7 +403,7 @@ int32_t tqProcessVgChangeReq(STQ* pTq, char* msg, int32_t msgLen) {
             .reader = pHandle->execHandle.pExecReader[i],
             .meta = pTq->pVnode->pMeta,
             .vnode = pTq->pVnode,
-            .initTsdbReader = 1,
+//            .initTsdbReader = 1,
         };
         pHandle->execHandle.execCol.task[i] = qCreateStreamExecTaskInfo(pHandle->execHandle.execCol.qmsg, &handle);
         ASSERT(pHandle->execHandle.execCol.task[i]);
@@ -479,7 +479,7 @@ int32_t tqProcessTaskDeployReq(STQ* pTq, char* msg, int32_t msgLen) {
              .reader = pStreamReader,
              .meta = pTq->pVnode->pMeta,
              .vnode = pTq->pVnode,
-             .initTsdbReader = 1,
+//             .initTsdbReader = 1,
       };
       /*pTask->exec.inputHandle = pStreamReader;*/
       pTask->exec.executor = qCreateStreamExecTaskInfo(pTask->exec.qmsg, &handle);

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

@@ -223,9 +223,8 @@ int64_t tsdbGetNumOfRowsInMemTable(tsdbReaderT* pHandle) {
   return rows;
 }
 
-static SArray* createCheckInfoFromTableGroup(STsdbReadHandle* pTsdbReadHandle, STableListInfo* pTableList) {
+static SArray* createCheckInfoFromTableGroup(STsdbReadHandle* pTsdbReadHandle, SArray* pTableList) {
-  size_t tableSize = taosArrayGetSize(pTableList->pTableList);
+  size_t tableSize = taosArrayGetSize(pTableList);
-  assert(tableSize >= 1);
 
   // allocate buffer in order to load data blocks from file
   SArray* pTableCheckInfo = taosArrayInit(tableSize, sizeof(STableCheckInfo));
@@ -235,7 +234,7 @@ static SArray* createCheckInfoFromTableGroup(STsdbReadHandle* pTsdbReadHandle, S
 
   // todo apply the lastkey of table check to avoid to load header file
   for (int32_t j = 0; j < tableSize; ++j) {
-    STableKeyInfo* pKeyInfo = (STableKeyInfo*)taosArrayGet(pTableList->pTableList, j);
+    STableKeyInfo* pKeyInfo = (STableKeyInfo*)taosArrayGet(pTableList, j);
 
     STableCheckInfo info = {.lastKey = pKeyInfo->lastKey, .tableId = pKeyInfo->uid};
     info.suid = pTsdbReadHandle->suid;
@@ -254,8 +253,6 @@ static SArray* createCheckInfoFromTableGroup(STsdbReadHandle* pTsdbReadHandle, S
               pTsdbReadHandle->idStr);
   }
 
-  // TODO  group table according to the tag value.
-  taosArraySort(pTableCheckInfo, tsdbCheckInfoCompar);
   return pTableCheckInfo;
 }
 
@@ -497,8 +494,21 @@ static int32_t setCurrentSchema(SVnode* pVnode, STsdbReadHandle* pTsdbReadHandle
   return TSDB_CODE_SUCCESS;
 }
 
-tsdbReaderT tsdbReaderOpen(SVnode* pVnode, SQueryTableDataCond* pCond, STableListInfo* tableList, uint64_t qId,
+int32_t tsdbSetTableList(tsdbReaderT reader, SArray* tableList){
+  STsdbReadHandle* pTsdbReadHandle = reader;
+  if(pTsdbReadHandle->pTableCheckInfo) taosArrayDestroy(pTsdbReadHandle->pTableCheckInfo);
+  pTsdbReadHandle->pTableCheckInfo = createCheckInfoFromTableGroup(pTsdbReadHandle, tableList);
+  if (pTsdbReadHandle->pTableCheckInfo == NULL) {
+    return TSDB_CODE_TDB_OUT_OF_MEMORY;
+  }
+  return TDB_CODE_SUCCESS;
+}
+
+tsdbReaderT tsdbReaderOpen(SVnode* pVnode, SQueryTableDataCond* pCond, SArray* tableList, uint64_t qId,
                             uint64_t taskId) {
+  if(taosArrayGetSize(tableList) == 0){
+    return NULL;
+  }
   STsdbReadHandle* pTsdbReadHandle = tsdbQueryTablesImpl(pVnode, pCond, qId, taskId);
   if (pTsdbReadHandle == NULL) {
     return NULL;
@@ -543,7 +553,7 @@ tsdbReaderT tsdbReaderOpen(SVnode* pVnode, SQueryTableDataCond* pCond, STableLis
   }
 
   tsdbDebug("%p total numOfTable:%" PRIzu " in this query, table %" PRIzu " %s", pTsdbReadHandle,
-            taosArrayGetSize(pTsdbReadHandle->pTableCheckInfo), taosArrayGetSize(tableList->pTableList),
+            taosArrayGetSize(pTsdbReadHandle->pTableCheckInfo), taosArrayGetSize(tableList),
             pTsdbReadHandle->idStr);
 
   return (tsdbReaderT)pTsdbReadHandle;
@@ -639,7 +649,7 @@ tsdbReaderT tsdbQueryLastRow(SVnode* pVnode, SQueryTableDataCond* pCond, STableL
     return NULL;
   }
 
-  STsdbReadHandle* pTsdbReadHandle = (STsdbReadHandle*)tsdbReaderOpen(pVnode, pCond, pList, qId, taskId);
+  STsdbReadHandle* pTsdbReadHandle = (STsdbReadHandle*)tsdbReaderOpen(pVnode, pCond, pList->pTableList, qId, taskId);
   if (pTsdbReadHandle == NULL) {
     return NULL;
   }
@@ -2842,7 +2852,7 @@ int32_t tsdbGetAllTableList(SMeta* pMeta, uint64_t uid, SArray* list) {
       break;
     }
 
-    STableKeyInfo info = {.lastKey = TSKEY_INITIAL_VAL, uid = id, .groupId = 0};
+    STableKeyInfo info = {.lastKey = TSKEY_INITIAL_VAL, uid = id};
     taosArrayPush(list, &info);
   }
 
@@ -3644,17 +3654,6 @@ SArray* tsdbRetrieveDataBlock(tsdbReaderT* pTsdbReadHandle, SArray* pIdList) {
   }
 }
 
-static int tsdbCheckInfoCompar(const void* key1, const void* key2) {
-  if (((STableCheckInfo*)key1)->tableId < ((STableCheckInfo*)key2)->tableId) {
-    return -1;
-  } else if (((STableCheckInfo*)key1)->tableId > ((STableCheckInfo*)key2)->tableId) {
-    return 1;
-  } else {
-    ASSERT(false);
-    return 0;
-  }
-}
-
 static void* doFreeColumnInfoData(SArray* pColumnInfoData) {
   if (pColumnInfoData == NULL) {
     return NULL;

source/libs/executor/inc/executorimpl.h

@@ -273,6 +273,10 @@ typedef struct STableScanInfo {
 
   SSampleExecInfo sample;       // sample execution info
   int32_t         curTWinIdx;
+
+  int32_t         currentGroupId;
+  uint64_t        queryId;
+  uint64_t        taskId;
 } STableScanInfo;
 
 typedef struct STagScanInfo {
@@ -354,6 +358,7 @@ typedef struct SSysTableScanInfo {
   tsem_t                 ready;
   SReadHandle            readHandle;
   int32_t                accountId;
+  const char*            pUser;
   bool                   showRewrite;
   SNode*                 pCondition;  // db_name filter condition, to discard data that are not in current database
   SMTbCursor*            pCur;        // cursor for iterate the local table meta store.
@@ -706,10 +711,10 @@ SResultRow* doSetResultOutBufByKey(SDiskbasedBuf* pResultBuf, SResultRowInfo* pR
 
 SOperatorInfo* createExchangeOperatorInfo(void* pTransporter, SExchangePhysiNode* pExNode, SExecTaskInfo* pTaskInfo);
 
-SOperatorInfo* createTableScanOperatorInfo(STableScanPhysiNode* pTableScanNode, tsdbReaderT pDataReader, SReadHandle* pHandle, SExecTaskInfo* pTaskInfo);
+SOperatorInfo* createTableScanOperatorInfo(STableScanPhysiNode* pTableScanNode, SReadHandle* pHandle, SExecTaskInfo* pTaskInfo, uint64_t queryId, uint64_t taskId);
 SOperatorInfo* createTagScanOperatorInfo(SReadHandle* pReadHandle, STagScanPhysiNode* pPhyNode,
                                          STableListInfo* pTableListInfo, SExecTaskInfo* pTaskInfo);
-SOperatorInfo* createSysTableScanOperatorInfo(void* readHandle, SSystemTableScanPhysiNode *pScanPhyNode, SExecTaskInfo* pTaskInfo);
+SOperatorInfo* createSysTableScanOperatorInfo(void* readHandle, SSystemTableScanPhysiNode *pScanPhyNode, const char* pUser, SExecTaskInfo* pTaskInfo);
 
 SOperatorInfo* createAggregateOperatorInfo(SOperatorInfo* downstream, SExprInfo* pExprInfo, int32_t numOfCols, SSDataBlock* pResultBlock, SExprInfo* pScalarExprInfo,
                                            int32_t numOfScalarExpr, SExecTaskInfo* pTaskInfo);
@@ -749,8 +754,8 @@ SOperatorInfo* createGroupOperatorInfo(SOperatorInfo* downstream, SExprInfo* pEx
 SOperatorInfo* createDataBlockInfoScanOperator(void* dataReader, SReadHandle* readHandle, uint64_t uid, SBlockDistScanPhysiNode* pBlockScanNode,
                                                SExecTaskInfo* pTaskInfo);
 
-SOperatorInfo* createStreamScanOperatorInfo(void* pDataReader, SReadHandle* pHandle,
+SOperatorInfo* createStreamScanOperatorInfo(SReadHandle* pHandle,
-    STableScanPhysiNode* pTableScanNode, SExecTaskInfo* pTaskInfo, STimeWindowAggSupp* pTwSup);
+    STableScanPhysiNode* pTableScanNode, SExecTaskInfo* pTaskInfo, STimeWindowAggSupp* pTwSup, uint64_t queryId, uint64_t taskId);
 
 SOperatorInfo* createFillOperatorInfo(SOperatorInfo* downstream, SFillPhysiNode* pPhyFillNode, bool multigroupResult,
                                       SExecTaskInfo* pTaskInfo);
@@ -845,7 +850,7 @@ SOperatorInfo* createTableMergeScanOperatorInfo(STableScanPhysiNode* pTableScanN
 
 void copyUpdateDataBlock(SSDataBlock* pDest, SSDataBlock* pSource, int32_t tsColIndex);
 
-int32_t generateGroupIdMap(STableListInfo* pTableListInfo, SReadHandle* pHandle, SArray* groupKey);
+int32_t generateGroupIdMap(STableListInfo* pTableListInfo, SReadHandle* pHandle, SNodeList* groupKey);
 
 #ifdef __cplusplus
 }

source/libs/executor/src/executil.c

@@ -287,6 +287,7 @@ static bool isTableOk(STableKeyInfo* info, SNode *pTagCond, SMeta *metaHandle){
 int32_t getTableList(void* metaHandle, SScanPhysiNode* pScanNode, STableListInfo* pListInfo) {
   int32_t code = TSDB_CODE_SUCCESS;
   pListInfo->pTableList = taosArrayInit(8, sizeof(STableKeyInfo));
+  if(pListInfo->pTableList == NULL) return TSDB_CODE_OUT_OF_MEMORY;
 
   uint64_t tableUid = pScanNode->uid;
 
@@ -314,7 +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), .groupId = 0};
+        STableKeyInfo info = {.lastKey = TSKEY_INITIAL_VAL, .uid = *(uint64_t*)taosArrayGet(res, i)};
         taosArrayPush(pListInfo->pTableList, &info);
       }
       taosArrayDestroy(res);
@@ -335,9 +336,14 @@ int32_t getTableList(void* metaHandle, SScanPhysiNode* pScanNode, STableListInfo
       }
     }
   }else {  // Create one table group.
-    STableKeyInfo info = {.lastKey = 0, .uid = tableUid, .groupId = 0};
+    STableKeyInfo info = {.lastKey = 0, .uid = tableUid};
     taosArrayPush(pListInfo->pTableList, &info);
   }
+  pListInfo->pGroupList = taosArrayInit(4, POINTER_BYTES);
+  if(pListInfo->pGroupList == NULL) return TSDB_CODE_OUT_OF_MEMORY;
+
+  //put into list as default group, remove it if grouping sorting is required later
+  taosArrayPush(pListInfo->pGroupList, &pListInfo->pTableList);
 
   return code;
 }

source/libs/executor/src/executorimpl.c

@@ -3862,9 +3862,6 @@ static SExecTaskInfo* createExecTaskInfo(uint64_t queryId, uint64_t taskId, EOPT
   return pTaskInfo;
 }
 
-static tsdbReaderT doCreateDataReader(STableScanPhysiNode* pTableScanNode, SReadHandle* pHandle,
-                                      STableListInfo* pTableListInfo, uint64_t queryId, uint64_t taskId);
-
 static SArray* extractColumnInfo(SNodeList* pNodeList);
 
 int32_t extractTableSchemaVersion(SReadHandle* pHandle, uint64_t uid, SExecTaskInfo* pTaskInfo) {
@@ -3895,8 +3892,67 @@ int32_t extractTableSchemaVersion(SReadHandle* pHandle, uint64_t uid, SExecTaskI
   return TSDB_CODE_SUCCESS;
 }
 
-int32_t generateGroupIdMap(STableListInfo* pTableListInfo, SReadHandle* pHandle, SArray* groupKey) {
+static int32_t sortTableGroup(STableListInfo* pTableListInfo, int32_t groupNum){
-  if (groupKey == NULL) {
+  taosArrayClear(pTableListInfo->pGroupList);
+  SArray *sortSupport = taosArrayInit(groupNum, sizeof(uint64_t));
+  if(sortSupport == NULL) return TSDB_CODE_OUT_OF_MEMORY;
+  for (int32_t i = 0; i < taosArrayGetSize(pTableListInfo->pTableList); i++) {
+    STableKeyInfo* info = taosArrayGet(pTableListInfo->pTableList, i);
+    uint64_t* groupId = taosHashGet(pTableListInfo->map, &info->uid, sizeof(uint64_t));
+
+    int32_t index = taosArraySearchIdx(sortSupport, groupId, compareUint64Val, TD_EQ);
+    if (index == -1){
+      void *p = taosArraySearch(sortSupport, groupId, compareUint64Val, TD_GT);
+      SArray *tGroup = taosArrayInit(8, sizeof(STableKeyInfo));
+      if(tGroup == NULL) {
+        taosArrayDestroy(sortSupport);
+        return TSDB_CODE_OUT_OF_MEMORY;
+      }
+      if(taosArrayPush(tGroup, info) == NULL){
+        qError("taos push info array error");
+        taosArrayDestroy(sortSupport);
+        return TSDB_CODE_QRY_APP_ERROR;
+      }
+      if(p == NULL){
+        if(taosArrayPush(sortSupport, groupId) != NULL){
+          qError("taos push support array error");
+          taosArrayDestroy(sortSupport);
+          return TSDB_CODE_QRY_APP_ERROR;
+        }
+        if(taosArrayPush(pTableListInfo->pGroupList, &tGroup) != NULL){
+          qError("taos push group array error");
+          taosArrayDestroy(sortSupport);
+          return TSDB_CODE_QRY_APP_ERROR;
+        }
+      }else{
+        int32_t pos = TARRAY_ELEM_IDX(sortSupport, p);
+        if(taosArrayInsert(sortSupport, pos, groupId) == NULL){
+          qError("taos insert support array error");
+          taosArrayDestroy(sortSupport);
+          return TSDB_CODE_QRY_APP_ERROR;
+        }
+        if(taosArrayInsert(pTableListInfo->pGroupList, pos, &tGroup) == NULL){
+          qError("taos insert group array error");
+          taosArrayDestroy(sortSupport);
+          return TSDB_CODE_QRY_APP_ERROR;
+        }
+      }
+    }else{
+      SArray* tGroup = (SArray*)taosArrayGetP(pTableListInfo->pGroupList, index);
+      if(taosArrayPush(tGroup, info) == NULL){
+        qError("taos push uid array error");
+        taosArrayDestroy(sortSupport);
+        return TSDB_CODE_QRY_APP_ERROR;
+      }
+    }
+
+  }
+  taosArrayDestroy(sortSupport);
+  return TDB_CODE_SUCCESS;
+}
+
+int32_t generateGroupIdMap(STableListInfo* pTableListInfo, SReadHandle* pHandle, SNodeList* group) {
+  if (group == NULL) {
     return TDB_CODE_SUCCESS;
   }
 
@@ -3906,13 +3962,14 @@ int32_t generateGroupIdMap(STableListInfo* pTableListInfo, SReadHandle* pHandle,
   }
   int32_t keyLen = 0;
   void*   keyBuf = NULL;
-  int32_t numOfGroupCols = taosArrayGetSize(groupKey);
+
-  for (int32_t j = 0; j < numOfGroupCols; ++j) {
+  SNode*           node;
-    SColumn* pCol = taosArrayGet(groupKey, j);
+  FOREACH(node, group) {
-    keyLen += pCol->bytes;  // actual data + null_flag
+    SExprNode *pExpr =  (SExprNode *)node;
+    keyLen += pExpr->resType.bytes;
   }
 
-  int32_t nullFlagSize = sizeof(int8_t) * numOfGroupCols;
+  int32_t nullFlagSize = sizeof(int8_t) * LIST_LENGTH(group);
   keyLen += nullFlagSize;
 
   keyBuf = taosMemoryCalloc(1, keyLen);
@@ -3920,103 +3977,109 @@ int32_t generateGroupIdMap(STableListInfo* pTableListInfo, SReadHandle* pHandle,
     return TSDB_CODE_OUT_OF_MEMORY;
   }
 
+  int32_t groupNum = 0;
   for (int32_t i = 0; i < taosArrayGetSize(pTableListInfo->pTableList); i++) {
     STableKeyInfo* info = taosArrayGet(pTableListInfo->pTableList, i);
     SMetaReader    mr = {0};
     metaReaderInit(&mr, pHandle->meta, 0);
     metaGetTableEntryByUid(&mr, info->uid);
 
-    char* isNull = (char*)keyBuf;
+    SNodeList *groupNew = nodesCloneList(group);
-    char* pStart = (char*)keyBuf + sizeof(int8_t) * numOfGroupCols;
-    for (int32_t j = 0; j < numOfGroupCols; ++j) {
-      SColumn* pCol = taosArrayGet(groupKey, j);
 
-      if (strcmp(pCol->name, "tbname") == 0) {
+    nodesRewriteExprsPostOrder(groupNew, doTranslateTagExpr, &mr);
-        isNull[i] = 0;
+    char* isNull = (char*)keyBuf;
-        memcpy(pStart, mr.me.name, strlen(mr.me.name));
+    char* pStart = (char*)keyBuf + nullFlagSize;
-        pStart += strlen(mr.me.name);
+
+    SNode* pNode;
+    int32_t index = 0;
+    FOREACH(pNode, groupNew){
+      SNode*  pNew = NULL;
+      int32_t code = scalarCalculateConstants(pNode, &pNew);
+      if (TSDB_CODE_SUCCESS == code) {
+        REPLACE_NODE(pNew);
       } else {
-        STagVal tagVal = {0};
+        taosMemoryFree(keyBuf);
-        tagVal.cid = pCol->colId;
+        nodesClearList(groupNew);
-        const char* p = metaGetTableTagVal(&mr.me, pCol->type, &tagVal);
+        return code;
-        if (p == NULL) {
-          isNull[j] = 1;
-          continue;
       }
-        isNull[i] = 0;
+
-        if (pCol->type == TSDB_DATA_TYPE_JSON) {
+      ASSERT(nodeType(pNew) == QUERY_NODE_VALUE);
-          //          int32_t dataLen = getJsonValueLen(pkey->pData);
+      SValueNode *pValue = (SValueNode *)pNew;
-          //          memcpy(pStart, (pkey->pData), dataLen);
+
-          //          pStart += dataLen;
+      if (pValue->node.resType.type == TSDB_DATA_TYPE_NULL) {
-        } else if (IS_VAR_DATA_TYPE(pCol->type)) {
+        isNull[index++] = 1;
-          memcpy(pStart, tagVal.pData, tagVal.nData);
+        continue;
-          pStart += tagVal.nData;
-          ASSERT(tagVal.nData <= pCol->bytes);
       } else {
-          memcpy(pStart, &(tagVal.i64), pCol->bytes);
+        isNull[index++] = 0;
-          pStart += pCol->bytes;
+        char*       data = nodesGetValueFromNode(pValue);
+        if (pValue->node.resType.type == TSDB_DATA_TYPE_JSON){
+          int32_t len = getJsonValueLen(data);
+          memcpy(pStart, data, len);
+          pStart += len;
+        } else if (IS_VAR_DATA_TYPE(pValue->node.resType.type)) {
+          memcpy(pStart, data, varDataTLen(data));
+          pStart += varDataTLen(data);
+        } else {
+          memcpy(pStart, data, pValue->node.resType.bytes);
+          pStart += pValue->node.resType.bytes;
         }
       }
     }
     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));
+    groupNum++;
 
-    uint64_t* pGroupId = taosHashGet(pTableListInfo->map, keyBuf, len);
+    nodesClearList(groupNew);
-
-    if (!pGroupId) {
-      uint64_t tmpId = calcGroupId(keyBuf, len);
-      info->groupId = tmpId;
-      taosHashPut(pTableListInfo->map, &(info->uid), sizeof(uint64_t), &tmpId, sizeof(uint64_t));
-    } else {
-      info->groupId = *pGroupId;
-    }
-
     metaReaderClear(&mr);
   }
   taosMemoryFree(keyBuf);
+
+  if(pTableListInfo->needSortTableByGroupId){
+    return sortTableGroup(pTableListInfo, groupNum);
+  }
+
   return TDB_CODE_SUCCESS;
 }
 
 SOperatorInfo* createOperatorTree(SPhysiNode* pPhyNode, SExecTaskInfo* pTaskInfo, SReadHandle* pHandle,
-                                  uint64_t queryId, uint64_t taskId, STableListInfo* pTableListInfo) {
+                                  uint64_t queryId, uint64_t taskId, STableListInfo* pTableListInfo, const char* pUser) {
   int32_t type = nodeType(pPhyNode);
 
   if (pPhyNode->pChildren == NULL || LIST_LENGTH(pPhyNode->pChildren) == 0) {
     if (QUERY_NODE_PHYSICAL_PLAN_TABLE_SCAN == type) {
       STableScanPhysiNode* pTableScanNode = (STableScanPhysiNode*)pPhyNode;
 
-      tsdbReaderT pDataReader = doCreateDataReader(pTableScanNode, pHandle, pTableListInfo, (uint64_t)queryId, taskId);
+      int32_t code = createScanTableListInfo(pTableScanNode, pHandle, pTableListInfo, queryId, taskId);
-      if (pDataReader == NULL && terrno != 0) {
-        pTaskInfo->code = terrno;
-        return NULL;
-      }
-
-      int32_t code = extractTableSchemaVersion(pHandle, pTableScanNode->scan.uid, pTaskInfo);
       if(code){
-        tsdbCleanupReadHandle(pDataReader);
+        pTaskInfo->code = code;
-        pTaskInfo->code = terrno;
         return NULL;
       }
-
+      code = extractTableSchemaVersion(pHandle, pTableScanNode->scan.uid, pTaskInfo);
-      SArray* groupKeys = extractPartitionColInfo(pTableScanNode->pPartitionTags);
-      code = generateGroupIdMap(pTableListInfo, pHandle, groupKeys);  // todo for json
-      taosArrayDestroy(groupKeys);
       if (code) {
-        tsdbCleanupReadHandle(pDataReader);
         pTaskInfo->code = terrno;
         return NULL;
       }
 
-      SOperatorInfo*  pOperator = createTableScanOperatorInfo(pTableScanNode, pDataReader, pHandle, pTaskInfo);
+      SOperatorInfo*  pOperator = createTableScanOperatorInfo(pTableScanNode, pHandle, pTaskInfo, queryId, taskId);
       STableScanInfo* pScanInfo = pOperator->info;
       pTaskInfo->cost.pRecoder = &pScanInfo->readRecorder;
       return pOperator;
 
     } else if (QUERY_NODE_PHYSICAL_PLAN_TABLE_MERGE_SCAN == type) {
       STableMergeScanPhysiNode* pTableScanNode = (STableMergeScanPhysiNode*)pPhyNode;
-      createScanTableListInfo(pTableScanNode, pHandle, pTableListInfo, queryId, taskId);
+      int32_t code = createScanTableListInfo(pTableScanNode, pHandle, pTableListInfo, queryId, taskId);
-      extractTableSchemaVersion(pHandle, pTableScanNode->scan.uid, pTaskInfo);
+      if(code){
-      SOperatorInfo* pOperator =
+        return NULL;
-          createTableMergeScanOperatorInfo(pTableScanNode, pTableListInfo, pHandle, pTaskInfo, queryId, taskId);
+      }
+      code = extractTableSchemaVersion(pHandle, pTableScanNode->scan.uid, pTaskInfo);
+      if (code) {
+        pTaskInfo->code = terrno;
+        return NULL;
+      }
+
+      SOperatorInfo*  pOperator = createTableMergeScanOperatorInfo(pTableScanNode, pTableListInfo, pHandle, pTaskInfo, queryId, taskId);
+
       STableScanInfo* pScanInfo = pOperator->info;
       pTaskInfo->cost.pRecoder = &pScanInfo->readRecorder;
       return pOperator;
@@ -4025,52 +4088,22 @@ SOperatorInfo* createOperatorTree(SPhysiNode* pPhyNode, SExecTaskInfo* pTaskInfo
       return createExchangeOperatorInfo(pHandle->pMsgCb->clientRpc, (SExchangePhysiNode*)pPhyNode, pTaskInfo);
 
     } else if (QUERY_NODE_PHYSICAL_PLAN_STREAM_SCAN == type) {
-      SScanPhysiNode*      pScanPhyNode = (SScanPhysiNode*)pPhyNode;  // simple child table.
       STableScanPhysiNode* pTableScanNode = (STableScanPhysiNode*)pPhyNode;
       STimeWindowAggSupp   twSup = {
             .waterMark = pTableScanNode->watermark,
             .calTrigger = pTableScanNode->triggerType,
             .maxTs = INT64_MIN,
       };
-      tsdbReaderT pDataReader = NULL;
-
       if (pHandle) {
-        if (pHandle->initTsdbReader) {
+        createScanTableListInfo(pTableScanNode, pHandle, pTableListInfo, queryId, taskId);
-          // for stream
-          ASSERT(pHandle->vnode);
-          pDataReader = doCreateDataReader(pTableScanNode, pHandle, pTableListInfo, (uint64_t)queryId, taskId);
-        } else {
-          // for tq
-          ASSERT(pHandle->meta);
-          getTableList(pHandle->meta, pScanPhyNode, pTableListInfo);
-        }
       }
 
-#if 0
+      SOperatorInfo* pOperator = createStreamScanOperatorInfo(pHandle, pTableScanNode, pTaskInfo, &twSup, queryId, taskId);
-      if (pDataReader == NULL && terrno != 0) {
-        qDebug("%s pDataReader is NULL", GET_TASKID(pTaskInfo));
-        // return NULL;
-      } else {
-        qDebug("%s pDataReader is not NULL", GET_TASKID(pTaskInfo));
-      }
-#endif
-
-      SArray* groupKeys = extractPartitionColInfo(pTableScanNode->pPartitionTags);
-      int32_t code = generateGroupIdMap(pTableListInfo, pHandle, groupKeys);  // todo for json
-      taosArrayDestroy(groupKeys);
-      if (code) {
-        tsdbCleanupReadHandle(pDataReader);
-        return NULL;
-      }
-
-      SOperatorInfo* pOperator = createStreamScanOperatorInfo(pDataReader, pHandle, pTableScanNode, pTaskInfo, &twSup);
-
       return pOperator;
 
     } else if (QUERY_NODE_PHYSICAL_PLAN_SYSTABLE_SCAN == type) {
       SSystemTableScanPhysiNode* pSysScanPhyNode = (SSystemTableScanPhysiNode*)pPhyNode;
-      return createSysTableScanOperatorInfo(pHandle, pSysScanPhyNode, pTaskInfo);
+      return createSysTableScanOperatorInfo(pHandle, pSysScanPhyNode, pUser, pTaskInfo);
-
     } else if (QUERY_NODE_PHYSICAL_PLAN_TAG_SCAN == type) {
       STagScanPhysiNode* pScanPhyNode = (STagScanPhysiNode*)pPhyNode;
 
@@ -4093,7 +4126,7 @@ SOperatorInfo* createOperatorTree(SPhysiNode* pPhyNode, SExecTaskInfo* pTaskInfo
           return NULL;
         }
       } else {  // Create one table group.
-        STableKeyInfo info = {.lastKey = 0, .uid = pBlockNode->uid, .groupId = 0};
+        STableKeyInfo info = {.lastKey = 0, .uid = pBlockNode->uid};
         taosArrayPush(pTableListInfo->pTableList, &info);
       }
 
@@ -4118,7 +4151,7 @@ SOperatorInfo* createOperatorTree(SPhysiNode* pPhyNode, SExecTaskInfo* pTaskInfo
         cond.suid = pBlockNode->suid;
         cond.type = BLOCK_LOAD_OFFSET_SEQ_ORDER;
       }
-      tsdbReaderT* pReader = tsdbReaderOpen(pHandle->vnode, &cond, pTableListInfo, queryId, taskId);
+      tsdbReaderT* pReader = tsdbReaderOpen(pHandle->vnode, &cond, pTableListInfo->pTableList, queryId, taskId);
       cleanupQueryTableDataCond(&cond);
 
       return createDataBlockInfoScanOperator(pReader, pHandle, cond.suid, pBlockNode, pTaskInfo);
@@ -4133,7 +4166,7 @@ SOperatorInfo* createOperatorTree(SPhysiNode* pPhyNode, SExecTaskInfo* pTaskInfo
   SOperatorInfo** ops = taosMemoryCalloc(size, POINTER_BYTES);
   for (int32_t i = 0; i < size; ++i) {
     SPhysiNode* pChildNode = (SPhysiNode*)nodesListGetNode(pPhyNode->pChildren, i);
-    ops[i] = createOperatorTree(pChildNode, pTaskInfo, pHandle, queryId, taskId, pTableListInfo);
+    ops[i] = createOperatorTree(pChildNode, pTaskInfo, pHandle, queryId, taskId, pTableListInfo, pUser);
     if (ops[i] == NULL) {
       return NULL;
     }
@@ -4355,7 +4388,7 @@ tsdbReaderT doCreateDataReader(STableScanPhysiNode* pTableScanNode, SReadHandle*
     goto _error;
   }
 
-  tsdbReaderT pReader = tsdbReaderOpen(pHandle->vnode, &cond, pTableListInfo, queryId, taskId);
+  tsdbReaderT pReader = tsdbReaderOpen(pHandle->vnode, &cond, pTableListInfo->pTableList, queryId, taskId);
   cleanupQueryTableDataCond(&cond);
 
   return pReader;
@@ -4566,7 +4599,7 @@ int32_t createExecTaskInfoImpl(SSubplan* pPlan, SExecTaskInfo** pTaskInfo, SRead
   (*pTaskInfo)->tableqinfoList.pTagCond = pPlan->pTagCond;
   (*pTaskInfo)->tableqinfoList.pTagIndexCond = pPlan->pTagIndexCond;
   (*pTaskInfo)->pRoot =
-      createOperatorTree(pPlan->pNode, *pTaskInfo, pHandle, queryId, taskId, &(*pTaskInfo)->tableqinfoList);
+      createOperatorTree(pPlan->pNode, *pTaskInfo, pHandle, queryId, taskId, &(*pTaskInfo)->tableqinfoList, pPlan->user);
 
   if (NULL == (*pTaskInfo)->pRoot) {
     code = (*pTaskInfo)->code;
@@ -4584,6 +4617,13 @@ _complete:
 static void doDestroyTableList(STableListInfo* pTableqinfoList) {
   taosArrayDestroy(pTableqinfoList->pTableList);
   taosHashCleanup(pTableqinfoList->map);
+  if(pTableqinfoList->needSortTableByGroupId){
+    for(int32_t i = 0; i < taosArrayGetSize(pTableqinfoList->pGroupList); i++){
+      SArray* tmp = taosArrayGetP(pTableqinfoList->pGroupList, i);
+      taosArrayDestroy(tmp);
+    }
+  }
+  taosArrayDestroy(pTableqinfoList->pGroupList);
 
   pTableqinfoList->pTableList = NULL;
   pTableqinfoList->map = NULL;

source/libs/executor/src/scanoperator.c

@@ -418,7 +418,7 @@ static SSDataBlock* doTableScanImpl(SOperatorInfo* pOperator) {
   return NULL;
 }
 
-static SSDataBlock* doTableScan(SOperatorInfo* pOperator) {
+static SSDataBlock* doTableScanGroup(SOperatorInfo* pOperator) {
   STableScanInfo* pTableScanInfo = pOperator->info;
   SExecTaskInfo*  pTaskInfo = pOperator->pTaskInfo;
 
@@ -500,6 +500,48 @@ static SSDataBlock* doTableScan(SOperatorInfo* pOperator) {
     }
   }
 
+  return NULL;
+}
+
+static SSDataBlock* doTableScan(SOperatorInfo* pOperator) {
+  STableScanInfo* pInfo = pOperator->info;
+  SExecTaskInfo*  pTaskInfo = pOperator->pTaskInfo;
+
+  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);
+    tsdbCleanupReadHandle(pInfo->dataReader);
+    tsdbReaderT* pReader = tsdbReaderOpen(pInfo->readHandle.vnode, &pInfo->cond, tableList, pInfo->queryId, pInfo->taskId);
+    pInfo->dataReader = pReader;
+  }
+
+  SSDataBlock* result = doTableScanGroup(pOperator);
+  if(result){
+    return result;
+  }
+
+  pInfo->currentGroupId++;
+  if (pInfo->currentGroupId >= taosArrayGetSize(pTaskInfo->tableqinfoList.pGroupList)) {
+    setTaskStatus(pTaskInfo, TASK_COMPLETED);
+    return NULL;
+  }
+
+  SArray *tableList = taosArrayGetP(pTaskInfo->tableqinfoList.pGroupList, pInfo->currentGroupId);
+  tsdbSetTableList(pInfo->dataReader, tableList);
+
+  tsdbResetReadHandle(pInfo->dataReader, &pInfo->cond, 0);
+  pInfo->curTWinIdx = 0;
+  pInfo->scanTimes = 0;
+
+  result = doTableScanGroup(pOperator);
+  if(result){
+    return result;
+  }
+
   setTaskStatus(pTaskInfo, TASK_COMPLETED);
   return NULL;
 }
@@ -525,8 +567,8 @@ static void destroyTableScanOperatorInfo(void* param, int32_t numOfOutput) {
   }
 }
 
-SOperatorInfo* createTableScanOperatorInfo(STableScanPhysiNode* pTableScanNode, tsdbReaderT pDataReader,
+SOperatorInfo* createTableScanOperatorInfo(STableScanPhysiNode* pTableScanNode, SReadHandle* readHandle,
-                                           SReadHandle* readHandle, SExecTaskInfo* pTaskInfo) {
+                                           SExecTaskInfo* pTaskInfo, uint64_t queryId, uint64_t taskId) {
   STableScanInfo* pInfo = taosMemoryCalloc(1, sizeof(STableScanInfo));
   SOperatorInfo*  pOperator = taosMemoryCalloc(1, sizeof(SOperatorInfo));
   if (pInfo == NULL || pOperator == NULL) {
@@ -561,10 +603,12 @@ SOperatorInfo* createTableScanOperatorInfo(STableScanPhysiNode* pTableScanNode,
   pInfo->dataBlockLoadFlag = pTableScanNode->dataRequired;
   pInfo->pResBlock = createResDataBlock(pDescNode);
   pInfo->pFilterNode = pTableScanNode->scan.node.pConditions;
-  pInfo->dataReader = pDataReader;
   pInfo->scanFlag = MAIN_SCAN;
   pInfo->pColMatchInfo = pColList;
   pInfo->curTWinIdx = 0;
+  pInfo->queryId = queryId;
+  pInfo->taskId = taskId;
+  pInfo->currentGroupId = -1;
 
   pOperator->name = "TableScanOperator";  // for debug purpose
   pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_TABLE_SCAN;
@@ -778,8 +822,9 @@ 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);
   pTableScanInfo->scanTimes = 0;
+  pTableScanInfo->currentGroupId = -1;
   return true;
 }
 
@@ -1022,9 +1067,7 @@ static SSDataBlock* doStreamBlockScan(SOperatorInfo* pOperator) {
         }
       }
 
-      // TODO refactor @liao
       taosArrayDestroy(block.pDataBlock);
-
       if (pInfo->pRes->pDataBlock == NULL) {
         // TODO add log
         pOperator->status = OP_EXEC_DONE;
@@ -1087,9 +1130,9 @@ static SArray* extractTableIdList(const STableListInfo* pTableGroupInfo) {
   return tableIdList;
 }
 
-SOperatorInfo* createStreamScanOperatorInfo(void* pDataReader, SReadHandle* pHandle,
+SOperatorInfo* createStreamScanOperatorInfo(SReadHandle* pHandle,
                                             STableScanPhysiNode* pTableScanNode, SExecTaskInfo* pTaskInfo,
-                                            STimeWindowAggSupp* pTwSup) {
+                                            STimeWindowAggSupp* pTwSup, uint64_t queryId, uint64_t taskId) {
   SStreamBlockScanInfo* pInfo = taosMemoryCalloc(1, sizeof(SStreamBlockScanInfo));
   SOperatorInfo*        pOperator = taosMemoryCalloc(1, sizeof(SOperatorInfo));
 
@@ -1129,7 +1172,7 @@ SOperatorInfo* createStreamScanOperatorInfo(void* pDataReader, SReadHandle* pHan
   }
 
   if (pHandle) {
-    SOperatorInfo*  pTableScanDummy = createTableScanOperatorInfo(pTableScanNode, pDataReader, pHandle, pTaskInfo);
+    SOperatorInfo*  pTableScanDummy = createTableScanOperatorInfo(pTableScanNode, pHandle, pTaskInfo, queryId, taskId);
     STableScanInfo* pSTInfo = (STableScanInfo*)pTableScanDummy->info;
     if (pSTInfo->interval.interval > 0) {
       pInfo->pUpdateInfo = updateInfoInitP(&pSTInfo->interval, pTwSup->waterMark);
@@ -1529,6 +1572,7 @@ static SSDataBlock* doSysTableScan(SOperatorInfo* pOperator) {
     while (1) {
       int64_t startTs = taosGetTimestampUs();
       strncpy(pInfo->req.tb, tNameGetTableName(&pInfo->name), tListLen(pInfo->req.tb));
+      strcpy(pInfo->req.user, pInfo->pUser);
 
       if (pInfo->showRewrite) {
         char dbName[TSDB_DB_NAME_LEN] = {0};
@@ -1661,7 +1705,7 @@ int32_t buildDbTableInfoBlock(const SSDataBlock* p, const SSysTableMeta* pSysDbT
 }
 
 SOperatorInfo* createSysTableScanOperatorInfo(void* readHandle, SSystemTableScanPhysiNode* pScanPhyNode,
-                                              SExecTaskInfo* pTaskInfo) {
+                                              const char* pUser, SExecTaskInfo* pTaskInfo) {
   SSysTableScanInfo* pInfo = taosMemoryCalloc(1, sizeof(SSysTableScanInfo));
   SOperatorInfo*     pOperator = taosMemoryCalloc(1, sizeof(SOperatorInfo));
   if (pInfo == NULL || pOperator == NULL) {
@@ -1677,6 +1721,7 @@ 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->showRewrite = pScanPhyNode->showRewrite;
   pInfo->pRes        = pResBlock;
   pInfo->pCondition  = pScanNode->node.pConditions;
@@ -1894,6 +1939,7 @@ SOperatorInfo* createTagScanOperatorInfo(SReadHandle* pReadHandle, STagScanPhysi
   pInfo->pRes             = createResDataBlock(pDescNode);
   pInfo->readHandle       = *pReadHandle;
   pInfo->curPos           = 0;
+
   pOperator->name = "TagScanOperator";
   pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_TAG_SCAN;
 
@@ -1919,10 +1965,7 @@ _error:
 
 typedef struct STableMergeScanInfo {
   STableListInfo* tableListInfo;
-  int32_t         tableStartIndex;
+  int32_t         currentGroupId;
-  int32_t         tableEndIndex;
-  bool            hasGroupId;
-  uint64_t        groupId;
 
   SArray*     dataReaders;  // array of tsdbReaderT*
   SReadHandle readHandle;
@@ -1968,12 +2011,6 @@ typedef struct STableMergeScanInfo {
   SSampleExecInfo sample;  // sample execution info
 } STableMergeScanInfo;
 
-int32_t compareTableKeyInfoByGid(const void* p1, const void* p2) {
-  const STableKeyInfo* info1 = p1;
-  const STableKeyInfo* info2 = p2;
-  return info1->groupId - info2->groupId;
-}
-
 int32_t createScanTableListInfo(STableScanPhysiNode* pTableScanNode, SReadHandle* pHandle,
                                 STableListInfo* pTableListInfo, uint64_t queryId, uint64_t taskId) {
   int32_t code = getTableList(pHandle->meta, &pTableScanNode->scan, pTableListInfo);
@@ -1985,57 +2022,11 @@ int32_t createScanTableListInfo(STableScanPhysiNode* pTableScanNode, SReadHandle
     qDebug("no table qualified for query, TID:0x%" PRIx64 ", QID:0x%" PRIx64, taskId, queryId);
     return TSDB_CODE_SUCCESS;
   }
-  SArray* groupKeys = extractPartitionColInfo(pTableScanNode->pPartitionTags);
+  code = generateGroupIdMap(pTableListInfo, pHandle, pTableScanNode->pPartitionTags);
-  generateGroupIdMap(pTableListInfo, pHandle, groupKeys);  // todo for json
-  if (groupKeys) {
-    taosArraySort(pTableListInfo->pTableList, compareTableKeyInfoByGid);
-  }
-  taosArrayDestroy(groupKeys);
-  return TSDB_CODE_SUCCESS;
-}
-
-int32_t doCreateMultipleDataReaders(STableScanPhysiNode* pTableScanNode, SReadHandle* pHandle,
-                                    STableListInfo* pTableListInfo, SArray* arrayReader, uint64_t queryId,
-                                    uint64_t taskId) {
-  SQueryTableDataCond cond = {0};
-  int32_t             code = initQueryTableDataCond(&cond, pTableScanNode);
   if (code != TSDB_CODE_SUCCESS) {
-    goto _error;
-  }
-  for (int32_t i = 0; i < taosArrayGetSize(pTableListInfo->pTableList); ++i) {
-    STableListInfo* subListInfo = taosMemoryCalloc(1, sizeof(subListInfo));
-    subListInfo->pTableList = taosArrayInit(1, sizeof(STableKeyInfo));
-    taosArrayPush(subListInfo->pTableList, taosArrayGet(pTableListInfo->pTableList, i));
-
-    tsdbReaderT* pReader = tsdbReaderOpen(pHandle->vnode, &cond, subListInfo, queryId, taskId);
-    taosArrayPush(arrayReader, &pReader);
-
-    taosArrayDestroy(subListInfo->pTableList);
-    taosMemoryFree(subListInfo);
-  }
-  cleanupQueryTableDataCond(&cond);
-
-  return TSDB_CODE_SUCCESS;
-
-_error:
     return code;
   }
 
-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) {
-    STableListInfo* subListInfo = taosMemoryCalloc(1, sizeof(subListInfo));
-    subListInfo->pTableList = taosArrayInit(1, sizeof(STableKeyInfo));
-    taosArrayPush(subListInfo->pTableList, taosArrayGet(pTableListInfo->pTableList, i));
-
-    tsdbReaderT* pReader = tsdbReaderOpen(pHandle->vnode, pQueryCond, subListInfo, queryId, taskId);
-    taosArrayPush(arrayReader, &pReader);
-
-    taosArrayDestroy(subListInfo->pTableList);
-    taosMemoryFree(subListInfo);
-  }
-
   return TSDB_CODE_SUCCESS;
 }
 
@@ -2225,32 +2216,34 @@ 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;
-  }
 
-  int32_t tableStartIdx = pInfo->tableStartIndex;
+  createMultipleDataReaders(&pInfo->cond, &pInfo->readHandle, tableList,
-  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
-  pInfo->sortBufSize = pInfo->bufPageSize * (tableEndIdx - tableStartIdx + 1 + 1);
+  int32_t tableLen = taosArrayGetSize(tableList);
+  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);
@@ -2337,37 +2330,42 @@ 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 (tableListSize == 0) {
+  if (pInfo->currentGroupId == -1) {
+    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 = NULL;
+  SSDataBlock* pBlock = getSortedTableMergeScanBlockData(pInfo->pSortHandle, pOperator->resultInfo.capacity, pOperator);
-  while (pInfo->tableStartIndex < tableListSize) {
-    pBlock = getSortedTableMergeScanBlockData(pInfo->pSortHandle, pOperator->resultInfo.capacity, pOperator);
   if (pBlock != NULL) {
-      pBlock->info.groupId = pInfo->groupId;
+    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;
-    } else {
+  }
+
   stopGroupTableMergeScan(pOperator);
-      if (pInfo->tableEndIndex >= tableListSize - 1) {
+  pInfo->currentGroupId++;
+  if (pInfo->currentGroupId >= taosArrayGetSize(pInfo->tableListInfo->pGroupList)) {
     doSetOperatorCompleted(pOperator);
-        break;
+    return NULL;
   }
-      pInfo->tableStartIndex = pInfo->tableEndIndex + 1;
-      pInfo->groupId =
-          ((STableKeyInfo*)taosArrayGet(pInfo->tableListInfo->pTableList, pInfo->tableStartIndex))->groupId;
   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));
+    if(groupId) pBlock->info.groupId = *groupId;
+
+    pOperator->resultInfo.totalRows += pBlock->info.rows;
+    return pBlock;
   }
-  }
+
+  doSetOperatorCompleted(pOperator);
 
   return pBlock;
 }
@@ -2445,6 +2443,7 @@ 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/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)) {
@@ -1279,8 +1280,7 @@ static bool validateTimestampDigits(const SValueNode* pVal) {
   int32_t tsDigits = (int32_t)strlen(fraction);
 
   if (tsDigits > TSDB_TIME_PRECISION_SEC_DIGITS) {
-    if (tsDigits == TSDB_TIME_PRECISION_MILLI_DIGITS ||
+    if (tsDigits == TSDB_TIME_PRECISION_MILLI_DIGITS || tsDigits == TSDB_TIME_PRECISION_MICRO_DIGITS ||
-        tsDigits == TSDB_TIME_PRECISION_MICRO_DIGITS ||
         tsDigits == TSDB_TIME_PRECISION_NANO_DIGITS) {
       return true;
     } else {
@@ -1510,6 +1510,11 @@ static int32_t translateBlockDistInfoFunc(SFunctionNode* pFunc, char* pErrBuf, i
   return TSDB_CODE_SUCCESS;
 }
 
+static int32_t translateGroupKeyFunc(SFunctionNode* pFunc, char* pErrBuf, int32_t len) {
+  pFunc->node.resType = ((SExprNode*)nodesListGetNode(pFunc->pParameterList, 0))->resType;
+  return TSDB_CODE_SUCCESS;
+}
+
 static bool getBlockDistFuncEnv(SFunctionNode* UNUSED_PARAM(pFunc), SFuncExecEnv* pEnv) {
   pEnv->calcMemSize = sizeof(STableBlockDistInfo);
   return true;
@@ -2519,6 +2524,8 @@ const SBuiltinFuncDefinition funcMgtBuiltins[] = {
     .initFunc     = functionSetup,
     .processFunc  = groupKeyFunction,
     .finalizeFunc = groupKeyFinalize,
+    .pPartialFunc = "_group_key",
+    .pMergeFunc   = "_group_key"
   },
 };
 // clang-format on

source/libs/function/src/builtinsimpl.c

@@ -264,6 +264,7 @@ typedef struct SRateInfo {
 
 typedef struct SGroupKeyInfo{
   bool  hasResult;
+  bool  isNull;
   char  data[];
 } SGroupKeyInfo;
 
@@ -5371,14 +5372,21 @@ int32_t groupKeyFunction(SqlFunctionCtx* pCtx) {
   int32_t bytes = pInputCol->info.bytes;
 
   int32_t startIndex = pInput->startRowIndex;
-  if (colDataIsNull_s(pInputCol, startIndex)) {
+
-    pInfo->hasResult = false;
+  //escape rest of data blocks to avoid first entry be overwritten.
+  if (pInfo->hasResult) {
     goto _group_key_over;
   }
 
+  if (colDataIsNull_s(pInputCol, startIndex)) {
+    pInfo->isNull = true;
     pInfo->hasResult = true;
+    goto _group_key_over;
+  }
+
   char* data = colDataGetData(pInputCol, startIndex);
   memcpy(pInfo->data, data, bytes);
+  pInfo->hasResult = true;
 
 _group_key_over:
 
@@ -5393,7 +5401,12 @@ int32_t groupKeyFinalize(SqlFunctionCtx* pCtx, SSDataBlock* pBlock) {
   SResultRowEntryInfo* pResInfo = GET_RES_INFO(pCtx);
 
   SGroupKeyInfo* pInfo = GET_ROWCELL_INTERBUF(pResInfo);
-  colDataAppend(pCol, pBlock->info.rows, pInfo->data, pInfo->hasResult ? false : true);
+
+  if (pInfo->hasResult) {
+    colDataAppend(pCol, pBlock->info.rows, pInfo->data, pInfo->isNull ? true : false);
+  } else {
+    pResInfo->numOfRes = 0;
+  }
 
   return pResInfo->numOfRes;
 }

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;
@@ -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;
     }
   }

source/libs/nodes/src/nodesUtilFuncs.c

@@ -346,9 +346,11 @@ static void destroyVgDataBlockArray(SArray* pArray) {
 }
 
 static void destroyLogicNode(SLogicNode* pNode) {
-  nodesDestroyList(pNode->pChildren);
-  nodesDestroyNode(pNode->pConditions);
   nodesDestroyList(pNode->pTargets);
+  nodesDestroyNode(pNode->pConditions);
+  nodesDestroyList(pNode->pChildren);
+  nodesDestroyNode(pNode->pLimit);
+  nodesDestroyNode(pNode->pSlimit);
 }
 
 static void destroyPhysiNode(SPhysiNode* pNode) {
@@ -368,6 +370,7 @@ static void destroyWinodwPhysiNode(SWinodwPhysiNode* pNode) {
 static void destroyScanPhysiNode(SScanPhysiNode* pNode) {
   destroyPhysiNode((SPhysiNode*)pNode);
   nodesDestroyList(pNode->pScanCols);
+  nodesDestroyList(pNode->pScanPseudoCols);
 }
 
 static void destroyDataSinkNode(SDataSinkNode* pNode) { nodesDestroyNode((SNode*)pNode->pInputDataBlockDesc); }
@@ -516,6 +519,9 @@ void nodesDestroyNode(SNode* pNode) {
       nodesDestroyNode(pStmt->pWindow);
       nodesDestroyList(pStmt->pGroupByList);
       nodesDestroyNode(pStmt->pHaving);
+      nodesDestroyNode(pStmt->pRange);
+      nodesDestroyNode(pStmt->pEvery);
+      nodesDestroyNode(pStmt->pFill);
       nodesDestroyList(pStmt->pOrderByList);
       nodesDestroyNode((SNode*)pStmt->pLimit);
       nodesDestroyNode((SNode*)pStmt->pSlimit);
@@ -779,6 +785,8 @@ void nodesDestroyNode(SNode* pNode) {
       SInterpFuncLogicNode* pLogicNode = (SInterpFuncLogicNode*)pNode;
       destroyLogicNode((SLogicNode*)pLogicNode);
       nodesDestroyList(pLogicNode->pFuncs);
+      nodesDestroyNode(pLogicNode->pFillValues);
+      nodesDestroyNode(pLogicNode->pTimeSeries);
       break;
     }
     case QUERY_NODE_LOGIC_SUBPLAN: {
@@ -793,14 +801,21 @@ void nodesDestroyNode(SNode* pNode) {
       nodesDestroyList(((SQueryLogicPlan*)pNode)->pTopSubplans);
       break;
     case QUERY_NODE_PHYSICAL_PLAN_TAG_SCAN:
-    case QUERY_NODE_PHYSICAL_PLAN_TABLE_SCAN:
-    case QUERY_NODE_PHYSICAL_PLAN_TABLE_SEQ_SCAN:
-    case QUERY_NODE_PHYSICAL_PLAN_STREAM_SCAN:
     case QUERY_NODE_PHYSICAL_PLAN_SYSTABLE_SCAN:
     case QUERY_NODE_PHYSICAL_PLAN_BLOCK_DIST_SCAN:
     case QUERY_NODE_PHYSICAL_PLAN_LAST_ROW_SCAN:
       destroyScanPhysiNode((SScanPhysiNode*)pNode);
       break;
+    case QUERY_NODE_PHYSICAL_PLAN_TABLE_SCAN:
+    case QUERY_NODE_PHYSICAL_PLAN_TABLE_SEQ_SCAN:
+    case QUERY_NODE_PHYSICAL_PLAN_TABLE_MERGE_SCAN:
+    case QUERY_NODE_PHYSICAL_PLAN_STREAM_SCAN: {
+      STableScanPhysiNode* pPhyNode = (STableScanPhysiNode*)pNode;
+      destroyScanPhysiNode((SScanPhysiNode*)pNode);
+      nodesDestroyList(pPhyNode->pDynamicScanFuncs);
+      nodesDestroyList(pPhyNode->pPartitionTags);
+      break;
+    }
     case QUERY_NODE_PHYSICAL_PLAN_PROJECT: {
       SProjectPhysiNode* pPhyNode = (SProjectPhysiNode*)pNode;
       destroyPhysiNode((SPhysiNode*)pPhyNode);
@@ -891,6 +906,8 @@ void nodesDestroyNode(SNode* pNode) {
       destroyPhysiNode((SPhysiNode*)pPhyNode);
       nodesDestroyList(pPhyNode->pExprs);
       nodesDestroyList(pPhyNode->pFuncs);
+      nodesDestroyNode(pPhyNode->pFillValues);
+      nodesDestroyNode(pPhyNode->pTimeSeries);
       break;
     }
     case QUERY_NODE_PHYSICAL_PLAN_DISPATCH:
@@ -1220,6 +1237,7 @@ int32_t nodesSetValueNodeValue(SValueNode* pNode, void* value) {
     case TSDB_DATA_TYPE_NCHAR:
     case TSDB_DATA_TYPE_VARCHAR:
     case TSDB_DATA_TYPE_VARBINARY:
+    case TSDB_DATA_TYPE_JSON:
       pNode->datum.p = (char*)value;
       break;
     default:

source/libs/parser/src/parTranslater.c

@@ -1355,6 +1355,25 @@ static EDealRes rewriteColToSelectValFunc(STranslateContext* pCxt, SNode** pNode
   return TSDB_CODE_SUCCESS == pCxt->errCode ? DEAL_RES_IGNORE_CHILD : DEAL_RES_ERROR;
 }
 
+static EDealRes rewriteExprToGroupKeyFunc(STranslateContext* pCxt, SNode** pNode) {
+  SFunctionNode* pFunc = (SFunctionNode*)nodesMakeNode(QUERY_NODE_FUNCTION);
+  if (NULL == pFunc) {
+    pCxt->errCode = TSDB_CODE_OUT_OF_MEMORY;
+    return DEAL_RES_ERROR;
+  }
+
+  strcpy(pFunc->functionName, "_group_key");
+  strcpy(pFunc->node.aliasName, ((SExprNode*)*pNode)->aliasName);
+  pCxt->errCode = nodesListMakeAppend(&pFunc->pParameterList, *pNode);
+  if (TSDB_CODE_SUCCESS == pCxt->errCode) {
+    *pNode = (SNode*)pFunc;
+    pCxt->errCode = fmGetFuncInfo(pFunc, pCxt->msgBuf.buf, pCxt->msgBuf.len);
+  }
+  pCxt->pCurrSelectStmt->hasAggFuncs = true;
+
+  return (TSDB_CODE_SUCCESS == pCxt->errCode ? DEAL_RES_IGNORE_CHILD : DEAL_RES_ERROR);
+}
+
 static EDealRes doCheckExprForGroupBy(SNode** pNode, void* pContext) {
   SCheckExprForGroupByCxt* pCxt = (SCheckExprForGroupByCxt*)pContext;
   if (!nodesIsExprNode(*pNode) || isAliasColumn(*pNode)) {
@@ -1371,10 +1390,10 @@ static EDealRes doCheckExprForGroupBy(SNode** pNode, void* pContext) {
   if (isAggFunc(*pNode) && !isDistinctOrderBy(pCxt->pTranslateCxt)) {
     return DEAL_RES_IGNORE_CHILD;
   }
-  SNode* pGroupNode;
+  SNode* pGroupNode = NULL;
   FOREACH(pGroupNode, getGroupByList(pCxt->pTranslateCxt)) {
     if (nodesEqualNode(getGroupByNode(pGroupNode), *pNode)) {
-      return DEAL_RES_IGNORE_CHILD;
+      return rewriteExprToGroupKeyFunc(pCxt->pTranslateCxt, pNode);
     }
   }
   if (isScanPseudoColumnFunc(*pNode) || QUERY_NODE_COLUMN == nodeType(*pNode)) {
@@ -1432,6 +1451,25 @@ static int32_t rewriteColsToSelectValFunc(STranslateContext* pCxt, SSelectStmt*
   return pCxt->errCode;
 }
 
+static EDealRes rewriteExprsToGroupKeyFuncImpl(SNode** pNode, void* pContext) {
+  STranslateContext* pCxt = pContext;
+  SNode*             pPartKey = NULL;
+  FOREACH(pPartKey, pCxt->pCurrSelectStmt->pPartitionByList) {
+    if (nodesEqualNode(pPartKey, *pNode)) {
+      return rewriteExprToGroupKeyFunc(pCxt, pNode);
+    }
+  }
+  return DEAL_RES_CONTINUE;
+}
+
+static int32_t rewriteExprsToGroupKeyFunc(STranslateContext* pCxt, SSelectStmt* pSelect) {
+  nodesRewriteExprs(pSelect->pProjectionList, rewriteExprsToGroupKeyFuncImpl, pCxt);
+  if (TSDB_CODE_SUCCESS == pCxt->errCode && !pSelect->isDistinct) {
+    nodesRewriteExprs(pSelect->pOrderByList, rewriteExprsToGroupKeyFuncImpl, pCxt);
+  }
+  return pCxt->errCode;
+}
+
 typedef struct CheckAggColCoexistCxt {
   STranslateContext* pTranslateCxt;
   bool               existAggFunc;
@@ -1456,6 +1494,12 @@ static EDealRes doCheckAggColCoexist(SNode* pNode, void* pContext) {
     pCxt->existIndefiniteRowsFunc = true;
     return DEAL_RES_IGNORE_CHILD;
   }
+  SNode* pPartKey = NULL;
+  FOREACH(pPartKey, pCxt->pTranslateCxt->pCurrSelectStmt->pPartitionByList) {
+    if (nodesEqualNode(pPartKey, pNode)) {
+      return DEAL_RES_IGNORE_CHILD;
+    }
+  }
   if (isScanPseudoColumnFunc(pNode) || QUERY_NODE_COLUMN == nodeType(pNode)) {
     pCxt->existCol = true;
   }
@@ -1485,6 +1529,9 @@ static int32_t checkAggColCoexist(STranslateContext* pCxt, SSelectStmt* pSelect)
   if (cxt.existIndefiniteRowsFunc && cxt.existCol) {
     return generateSyntaxErrMsg(&pCxt->msgBuf, TSDB_CODE_PAR_NOT_ALLOWED_FUNC);
   }
+  if (cxt.existAggFunc && NULL != pSelect->pPartitionByList) {
+    return rewriteExprsToGroupKeyFunc(pCxt, pSelect);
+  }
   return TSDB_CODE_SUCCESS;
 }
 
@@ -1840,10 +1887,7 @@ static int32_t createMultiResFuncsFromStar(STranslateContext* pCxt, SFunctionNod
     code = createMultiResFuncs(pSrcFunc, pExprs, pOutput);
   }
 
-  if (TSDB_CODE_SUCCESS != code) {
   nodesDestroyList(pExprs);
-  }
-
   return code;
 }
 
@@ -2463,8 +2507,24 @@ static SNode* createOrderByExpr(STranslateContext* pCxt) {
   return (SNode*)pOrder;
 }
 
-// from: select tail(expr, k, f) from t where_clause partition_by_clause order_by_clause ...
+/* case 1:
-// to: select expr from t where_clause order by _rowts desc limit k offset f
+ * in:  select tail(expr, k, f) from t where_clause
+ * out: select expr from t where_clause order by _rowts desc limit k offset f
+ *
+ * case 2:
+ * in:  select tail(expr, k, f) from t where_clause partition_by_clause
+ * out: select expr from t where_clause partition_by_clause sort by _rowts desc limit k offset f
+ *
+ * case 3:
+ * in:  select tail(expr, k, f) from t where_clause order_by_clause limit_clause
+ * out: select expr from (
+ *        select expr from t where_clause order by _rowts desc limit k offset f
+ *      ) order_by_clause limit_clause
+ *
+ * case 4:
+ * in:  select tail(expr, k, f) from t where_clause partition_by_clause limit_clause
+ * out:
+ */
 static int32_t rewriteTailStmt(STranslateContext* pCxt, SSelectStmt* pSelect) {
   if (!pSelect->hasTailFunc) {
     return TSDB_CODE_SUCCESS;
@@ -4963,6 +5023,7 @@ typedef struct SVgroupCreateTableBatch {
 
 static void destroyCreateTbReq(SVCreateTbReq* pReq) {
   taosMemoryFreeClear(pReq->name);
+  taosMemoryFreeClear(pReq->comment);
   taosMemoryFreeClear(pReq->ntb.schemaRow.pSchema);
 }
 
@@ -4980,6 +5041,7 @@ 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);
@@ -5051,6 +5113,7 @@ 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/parser/test/parSelectTest.cpp

@@ -199,6 +199,20 @@ TEST_F(ParserSelectTest, tailFuncSemanticCheck) {
   run("SELECT TAIL(c1, 10) FROM t1 GROUP BY c2", TSDB_CODE_PAR_GROUP_BY_NOT_ALLOWED_FUNC);
 }
 
+TEST_F(ParserSelectTest, partitionBy) {
+  useDb("root", "test");
+
+  run("SELECT c1, c2 FROM t1 PARTITION BY c2");
+
+  run("SELECT SUM(c1), c2 FROM t1 PARTITION BY c2");
+}
+
+TEST_F(ParserSelectTest, partitionBySemanticCheck) {
+  useDb("root", "test");
+
+  run("SELECT SUM(c1), c2, c3 FROM t1 PARTITION BY c2", TSDB_CODE_PAR_NOT_SINGLE_GROUP);
+}
+
 TEST_F(ParserSelectTest, groupBy) {
   useDb("root", "test");
 
@@ -213,6 +227,15 @@ TEST_F(ParserSelectTest, groupBy) {
   run("SELECT COUNT(*), c1 + 10, c2 cnt FROM t1 WHERE c1 > 0 GROUP BY c1 + 10, c2");
 }
 
+TEST_F(ParserSelectTest, groupBySemanticCheck) {
+  useDb("root", "test");
+
+  run("SELECT COUNT(*) cnt, c1 FROM t1 WHERE c1 > 0", TSDB_CODE_PAR_NOT_SINGLE_GROUP);
+  run("SELECT COUNT(*) cnt, c2 FROM t1 WHERE c1 > 0 GROUP BY c1", TSDB_CODE_PAR_GROUPBY_LACK_EXPRESSION);
+  run("SELECT COUNT(*) cnt, c2 FROM t1 WHERE c1 > 0 PARTITION BY c2 GROUP BY c1",
+      TSDB_CODE_PAR_GROUPBY_LACK_EXPRESSION);
+}
+
 TEST_F(ParserSelectTest, orderBy) {
   useDb("root", "test");
 

source/libs/planner/src/planLogicCreater.c

@@ -450,6 +450,15 @@ static int32_t createAggLogicNode(SLogicPlanContext* pCxt, SSelectStmt* pSelect,
   int32_t code = TSDB_CODE_SUCCESS;
 
   // set grouyp keys, agg funcs and having conditions
+  if (TSDB_CODE_SUCCESS == code && pSelect->hasAggFuncs) {
+    code = nodesCollectFuncs(pSelect, SQL_CLAUSE_GROUP_BY, fmIsAggFunc, &pAgg->pAggFuncs);
+  }
+
+  // rewrite the expression in subsequent clauses
+  if (TSDB_CODE_SUCCESS == code) {
+    code = rewriteExprsForSelect(pAgg->pAggFuncs, pSelect, SQL_CLAUSE_GROUP_BY);
+  }
+
   if (NULL != pSelect->pGroupByList) {
     pAgg->pGroupKeys = nodesCloneList(pSelect->pGroupByList);
     if (NULL == pAgg->pGroupKeys) {
@@ -462,15 +471,6 @@ static int32_t createAggLogicNode(SLogicPlanContext* pCxt, SSelectStmt* pSelect,
     code = rewriteExprsForSelect(pAgg->pGroupKeys, pSelect, SQL_CLAUSE_GROUP_BY);
   }
 
-  if (TSDB_CODE_SUCCESS == code && pSelect->hasAggFuncs) {
-    code = nodesCollectFuncs(pSelect, SQL_CLAUSE_GROUP_BY, fmIsAggFunc, &pAgg->pAggFuncs);
-  }
-
-  // rewrite the expression in subsequent clauses
-  if (TSDB_CODE_SUCCESS == code) {
-    code = rewriteExprsForSelect(pAgg->pAggFuncs, pSelect, SQL_CLAUSE_GROUP_BY);
-  }
-
   if (TSDB_CODE_SUCCESS == code && NULL != pSelect->pHaving) {
     pAgg->node.pConditions = nodesCloneNode(pSelect->pHaving);
     if (NULL == pAgg->node.pConditions) {
@@ -780,8 +780,8 @@ static int32_t createProjectLogicNode(SLogicPlanContext* pCxt, SSelectStmt* pSel
     return TSDB_CODE_OUT_OF_MEMORY;
   }
 
-  pProject->node.pLimit = (SNode*)pSelect->pLimit;
+  TSWAP(pProject->node.pLimit, pSelect->pLimit);
-  pProject->node.pSlimit = (SNode*)pSelect->pSlimit;
+  TSWAP(pProject->node.pSlimit, pSelect->pSlimit);
 
   int32_t code = TSDB_CODE_SUCCESS;
 
@@ -940,7 +940,7 @@ static int32_t createSetOpSortLogicNode(SLogicPlanContext* pCxt, SSetOperator* p
     return TSDB_CODE_OUT_OF_MEMORY;
   }
 
-  pSort->node.pLimit = pSetOperator->pLimit;
+  TSWAP(pSort->node.pLimit, pSetOperator->pLimit);
 
   int32_t code = TSDB_CODE_SUCCESS;
 
@@ -973,7 +973,7 @@ static int32_t createSetOpProjectLogicNode(SLogicPlanContext* pCxt, SSetOperator
   }
 
   if (NULL == pSetOperator->pOrderByList) {
-    pProject->node.pLimit = pSetOperator->pLimit;
+    TSWAP(pProject->node.pLimit, pSetOperator->pLimit);
   }
 
   int32_t code = TSDB_CODE_SUCCESS;
@@ -1004,7 +1004,7 @@ static int32_t createSetOpAggLogicNode(SLogicPlanContext* pCxt, SSetOperator* pS
   }
 
   if (NULL == pSetOperator->pOrderByList) {
-    pAgg->node.pLimit = pSetOperator->pLimit;
+    TSWAP(pAgg->node.pSlimit, pSetOperator->pLimit);
   }
 
   int32_t code = TSDB_CODE_SUCCESS;

source/libs/planner/src/planOptimizer.c

@@ -682,7 +682,7 @@ static EOrder opkGetPrimaryKeyOrder(SSortLogicNode* pSort) {
 static SNode* opkRewriteDownNode(SSortLogicNode* pSort) {
   SNode* pDownNode = nodesListGetNode(pSort->node.pChildren, 0);
   // todo
-  pSort->node.pChildren = NULL;
+  NODES_CLEAR_LIST(pSort->node.pChildren);
   return pDownNode;
 }
 
@@ -1061,6 +1061,7 @@ static EDealRes partTagsOptRebuildTbanmeImpl(SNode** pNode, void* pContext) {
     }
     strcpy(pFunc->functionName, "tbname");
     pFunc->funcType = FUNCTION_TYPE_TBNAME;
+    pFunc->node.resType = ((SColumnNode*)*pNode)->node.resType;
     nodesDestroyNode(*pNode);
     *pNode = (SNode*)pFunc;
     return DEAL_RES_IGNORE_CHILD;
@@ -1188,7 +1189,7 @@ static const SOptimizeRule optimizeRuleSet[] = {
   {.pName = "ConditionPushDown", .optimizeFunc = cpdOptimize},
   {.pName = "OrderByPrimaryKey", .optimizeFunc = opkOptimize},
   {.pName = "SmaIndex",          .optimizeFunc = smaOptimize},
-  // {.pName = "PartitionTags",     .optimizeFunc = partTagsOptimize},
+  {.pName = "PartitionTags",     .optimizeFunc = partTagsOptimize},
   {.pName = "EliminateProject",  .optimizeFunc = eliminateProjOptimize}
 };
 // clang-format on

source/libs/planner/src/planPhysiCreater.c

@@ -348,8 +348,8 @@ static SPhysiNode* makePhysiNode(SPhysiPlanContext* pCxt, SLogicNode* pLogicNode
     return NULL;
   }
 
-  pPhysiNode->pLimit = pLogicNode->pLimit;
+  TSWAP(pPhysiNode->pLimit, pLogicNode->pLimit);
-  pPhysiNode->pSlimit = pLogicNode->pSlimit;
+  TSWAP(pPhysiNode->pSlimit, pLogicNode->pSlimit);
 
   int32_t code = createDataBlockDesc(pCxt, pLogicNode->pTargets, &pPhysiNode->pOutputDataBlockDesc);
   if (TSDB_CODE_SUCCESS != code) {
@@ -862,6 +862,9 @@ static int32_t createIndefRowsFuncPhysiNode(SPhysiPlanContext* pCxt, SNodeList*
     nodesDestroyNode((SNode*)pIdfRowsFunc);
   }
 
+  nodesDestroyList(pPrecalcExprs);
+  nodesDestroyList(pFuncs);
+
   return code;
 }
 
@@ -913,6 +916,9 @@ static int32_t createInterpFuncPhysiNode(SPhysiPlanContext* pCxt, SNodeList* pCh
     nodesDestroyNode((SNode*)pInterpFunc);
   }
 
+  nodesDestroyList(pPrecalcExprs);
+  nodesDestroyList(pFuncs);
+
   return code;
 }
 
@@ -1048,6 +1054,9 @@ static int32_t createWindowPhysiNodeFinalize(SPhysiPlanContext* pCxt, SNodeList*
     nodesDestroyNode((SNode*)pWindow);
   }
 
+  nodesDestroyList(pPrecalcExprs);
+  nodesDestroyList(pFuncs);
+
   return code;
 }
 
@@ -1241,6 +1250,9 @@ static int32_t createPartitionPhysiNode(SPhysiPlanContext* pCxt, SNodeList* pChi
     nodesDestroyNode((SNode*)pPart);
   }
 
+  nodesDestroyList(pPrecalcExprs);
+  nodesDestroyList(pPartitionKeys);
+
   return code;
 }
 

source/libs/planner/src/planSpliter.c

@@ -374,7 +374,7 @@ static int32_t stbSplCreateMergeNode(SSplitContext* pCxt, SLogicSubplan* pSubpla
 
   int32_t code = TSDB_CODE_SUCCESS;
   pMerge->pInputs = nodesCloneList(pPartChild->pTargets);
-  // NULL == pSubplan means 'merge node' replaces 'split node'.
+  // NULL != pSubplan means 'merge node' replaces 'split node'.
   if (NULL == pSubplan) {
     pMerge->node.pTargets = nodesCloneList(pPartChild->pTargets);
   } else {
@@ -561,6 +561,8 @@ static int32_t stbSplSplitState(SSplitContext* pCxt, SStableSplitInfo* pInfo) {
 static SNodeList* stbSplGetPartKeys(SLogicNode* pNode) {
   if (QUERY_NODE_LOGIC_PLAN_SCAN == nodeType(pNode)) {
     return ((SScanLogicNode*)pNode)->pPartTags;
+  } else if (QUERY_NODE_LOGIC_PLAN_PARTITION == nodeType(pNode)) {
+    return ((SPartitionLogicNode*)pNode)->pPartitionKeys;
   } else {
     return NULL;
   }
@@ -571,14 +573,15 @@ static bool stbSplIsPartTbanme(SNodeList* pPartKeys) {
     return false;
   }
   SNode* pPartKey = nodesListGetNode(pPartKeys, 0);
-  return QUERY_NODE_FUNCTION == nodeType(pPartKey) && FUNCTION_TYPE_TBNAME == ((SFunctionNode*)pPartKey)->funcType;
+  return (QUERY_NODE_FUNCTION == nodeType(pPartKey) && FUNCTION_TYPE_TBNAME == ((SFunctionNode*)pPartKey)->funcType) ||
+         (QUERY_NODE_COLUMN == nodeType(pPartKey) && COLUMN_TYPE_TBNAME == ((SColumnNode*)pPartKey)->colType);
 }
 
-static bool stbSplIsMultiTableWinodw(SWindowLogicNode* pWindow) {
+static bool stbSplIsPartTableWinodw(SWindowLogicNode* pWindow) {
   return stbSplIsPartTbanme(stbSplGetPartKeys((SLogicNode*)nodesListGetNode(pWindow->node.pChildren, 0)));
 }
 
-static int32_t stbSplSplitWindowForMergeTable(SSplitContext* pCxt, SStableSplitInfo* pInfo) {
+static int32_t stbSplSplitWindowForCrossTable(SSplitContext* pCxt, SStableSplitInfo* pInfo) {
   switch (((SWindowLogicNode*)pInfo->pSplitNode)->winType) {
     case WINDOW_TYPE_INTERVAL:
       return stbSplSplitInterval(pCxt, pInfo);
@@ -592,7 +595,7 @@ static int32_t stbSplSplitWindowForMergeTable(SSplitContext* pCxt, SStableSplitI
   return TSDB_CODE_PLAN_INTERNAL_ERROR;
 }
 
-static int32_t stbSplSplitWindowForMultiTable(SSplitContext* pCxt, SStableSplitInfo* pInfo) {
+static int32_t stbSplSplitWindowForPartTable(SSplitContext* pCxt, SStableSplitInfo* pInfo) {
   if (pCxt->pPlanCxt->streamQuery) {
     SPLIT_FLAG_SET_MASK(pInfo->pSubplan->splitFlag, SPLIT_FLAG_STABLE_SPLIT);
     return TSDB_CODE_SUCCESS;
@@ -613,10 +616,10 @@ static int32_t stbSplSplitWindowForMultiTable(SSplitContext* pCxt, SStableSplitI
 }
 
 static int32_t stbSplSplitWindowNode(SSplitContext* pCxt, SStableSplitInfo* pInfo) {
-  if (stbSplIsMultiTableWinodw((SWindowLogicNode*)pInfo->pSplitNode)) {
+  if (stbSplIsPartTableWinodw((SWindowLogicNode*)pInfo->pSplitNode)) {
-    return stbSplSplitWindowForMultiTable(pCxt, pInfo);
+    return stbSplSplitWindowForPartTable(pCxt, pInfo);
   } else {
-    return stbSplSplitWindowForMergeTable(pCxt, pInfo);
+    return stbSplSplitWindowForCrossTable(pCxt, pInfo);
   }
 }
 

source/libs/planner/test/planDistinctTest.cpp

@@ -40,3 +40,9 @@ TEST_F(PlanDistinctTest, withOrderBy) {
 
   run("select distinct c1 + 10 a from t1 order by a");
 }
+
+TEST_F(PlanDistinctTest, withLimit) {
+  useDb("root", "test");
+
+  run("SELECT DISTINCT c1 FROM t1 LIMIT 3");
+}

source/libs/planner/test/planGroupByTest.cpp

@@ -53,14 +53,6 @@ TEST_F(PlanGroupByTest, aggFunc) {
   run("SELECT SUM(10), COUNT(c1) FROM t1 GROUP BY c2");
 }
 
-TEST_F(PlanGroupByTest, rewriteFunc) {
-  useDb("root", "test");
-
-  run("SELECT AVG(c1) FROM t1");
-
-  run("SELECT AVG(c1) FROM t1 GROUP BY c2");
-}
-
 TEST_F(PlanGroupByTest, selectFunc) {
   useDb("root", "test");
 
@@ -81,6 +73,8 @@ TEST_F(PlanGroupByTest, stable) {
 
   run("SELECT COUNT(*) FROM st1");
 
+  run("SELECT c1 FROM st1 GROUP BY c1");
+
   run("SELECT COUNT(*) FROM st1 GROUP BY c1");
 
   run("SELECT COUNT(*) FROM st1 PARTITION BY c2 GROUP BY c1");

source/libs/planner/test/planOptimizeTest.cpp

@@ -57,9 +57,15 @@ TEST_F(PlanOptimizeTest, orderByPrimaryKey) {
 TEST_F(PlanOptimizeTest, PartitionTags) {
   useDb("root", "test");
 
-  run("SELECT c1 FROM st1 PARTITION BY tag1");
+  run("SELECT c1, tag1 FROM st1 PARTITION BY tag1");
 
-  run("SELECT SUM(c1) FROM st1 GROUP BY tag1");
+  run("SELECT SUM(c1), tag1 FROM st1 PARTITION BY tag1");
+
+  run("SELECT SUM(c1), tag1 + 10 FROM st1 PARTITION BY tag1 + 10");
+
+  run("SELECT SUM(c1), tag1 FROM st1 GROUP BY tag1");
+
+  run("SELECT SUM(c1), tag1 + 10 FROM st1 GROUP BY tag1 + 10");
 }
 
 TEST_F(PlanOptimizeTest, eliminateProjection) {

source/libs/planner/test/planPartByTest.cpp

@@ -34,6 +34,8 @@ TEST_F(PlanPartitionByTest, withAggFunc) {
   useDb("root", "test");
 
   run("select count(*) from t1 partition by c1");
+
+  run("select count(*), c1 from t1 partition by c1");
 }
 
 TEST_F(PlanPartitionByTest, withInterval) {
@@ -41,8 +43,12 @@ TEST_F(PlanPartitionByTest, withInterval) {
 
   // normal/child table
   run("select count(*) from t1 partition by c1 interval(10s)");
+
+  run("select count(*), c1 from t1 partition by c1 interval(10s)");
   // super table
   run("select count(*) from st1 partition by tag1, tag2 interval(10s)");
+
+  run("select count(*), tag1 from st1 partition by tag1, tag2 interval(10s)");
 }
 
 TEST_F(PlanPartitionByTest, withGroupBy) {

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);
@@ -221,6 +224,7 @@ 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 n = 0;
@@ -372,7 +376,6 @@ end:
   return string;
 }
 
-
 int32_t cloneTableMeta(STableMeta* pSrc, STableMeta** pDst) {
   if (NULL == pSrc) {
     *pDst = NULL;
@@ -400,7 +403,8 @@ 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;
     }
@@ -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/streamData.c

@@ -27,11 +27,14 @@ int32_t streamDispatchReqToData(const SStreamDispatchReq* pReq, SStreamDataBlock
   ASSERT(pReq->blockNum == taosArrayGetSize(pReq->dataLen));
 
   for (int32_t i = 0; i < blockNum; i++) {
-    int32_t            len = *(int32_t*)taosArrayGet(pReq->dataLen, i);
+    /*int32_t            len = *(int32_t*)taosArrayGet(pReq->dataLen, i);*/
     SRetrieveTableRsp* pRetrieve = taosArrayGetP(pReq->data, i);
     SSDataBlock*       pDataBlock = taosArrayGet(pArray, i);
     blockCompressDecode(pDataBlock, htonl(pRetrieve->numOfCols), htonl(pRetrieve->numOfRows), pRetrieve->data);
     // TODO: refactor
+    pDataBlock->info.window.skey = be64toh(pRetrieve->skey);
+    pDataBlock->info.window.ekey = be64toh(pRetrieve->ekey);
+
     pDataBlock->info.type = pRetrieve->streamBlockType;
     pDataBlock->info.childId = pReq->upstreamChildId;
   }
@@ -46,8 +49,14 @@ int32_t streamRetrieveReqToData(const SStreamRetrieveReq* pReq, SStreamDataBlock
   }
   taosArraySetSize(pArray, 1);
   SRetrieveTableRsp* pRetrieve = pReq->pRetrieve;
-  SSDataBlock*       pBlock = taosArrayGet(pArray, 0);
+  SSDataBlock*       pDataBlock = taosArrayGet(pArray, 0);
-  blockCompressDecode(pBlock, htonl(pRetrieve->numOfCols), htonl(pRetrieve->numOfRows), pRetrieve->data);
+  blockCompressDecode(pDataBlock, htonl(pRetrieve->numOfCols), htonl(pRetrieve->numOfRows), pRetrieve->data);
+  // TODO: refactor
+  pDataBlock->info.window.skey = be64toh(pRetrieve->skey);
+  pDataBlock->info.window.ekey = be64toh(pRetrieve->ekey);
+
+  pDataBlock->info.type = pRetrieve->streamBlockType;
+
   pData->blocks = pArray;
   return 0;
 }

source/libs/stream/src/streamDispatch.c

@@ -104,6 +104,8 @@ int32_t streamBroadcastToChildren(SStreamTask* pTask, const SSDataBlock* pBlock)
   pRetrieve->streamBlockType = pBlock->info.type;
   pRetrieve->numOfRows = htonl(pBlock->info.rows);
   pRetrieve->numOfCols = htonl(numOfCols);
+  pRetrieve->skey = htobe64(pBlock->info.window.skey);
+  pRetrieve->ekey = htobe64(pBlock->info.window.ekey);
 
   int32_t actualLen = 0;
   blockCompressEncode(pBlock, pRetrieve->data, &actualLen, numOfCols, false);
@@ -171,6 +173,8 @@ static int32_t streamAddBlockToDispatchMsg(const SSDataBlock* pBlock, SStreamDis
   pRetrieve->completed = 1;
   pRetrieve->streamBlockType = pBlock->info.type;
   pRetrieve->numOfRows = htonl(pBlock->info.rows);
+  pRetrieve->skey = htobe64(pBlock->info.window.skey);
+  pRetrieve->ekey = htobe64(pBlock->info.window.ekey);
 
   int32_t numOfCols = (int32_t)taosArrayGetSize(pBlock->pDataBlock);
   pRetrieve->numOfCols = htonl(numOfCols);
@@ -296,7 +300,7 @@ 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 +44,16 @@ 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 && !hasData) {
+        SSDataBlock block = {0};
+        block.info.type = STREAM_PUSH_DATA;
+        block.info.childId = pTask->selfChildId;
+        taosArrayPush(pRes, &block);
+      }
+      break;
+    }
+    hasData = true;
 
     if (output->info.type == STREAM_RETRIEVE) {
       if (streamBroadcastToChildren(pTask, output) < 0) {

source/libs/sync/src/syncAppendEntries.c

@@ -89,6 +89,280 @@
 //       /\ UNCHANGED <<candidateVars, leaderVars>>
 //
 
+int32_t syncNodeOnAppendEntriesCb(SSyncNode* ths, SyncAppendEntries* pMsg) {
+  int32_t ret = 0;
+
+  // print log
+  syncAppendEntriesLog2("==syncNodeOnAppendEntriesCb==", pMsg);
+
+  // if already drop replica, do not process
+  if (!syncNodeInRaftGroup(ths, &(pMsg->srcId)) && !ths->pRaftCfg->isStandBy) {
+    syncNodeEventLog(ths, "recv sync-append-entries, maybe replica already dropped");
+    return ret;
+  }
+
+  // maybe update term
+  if (pMsg->term > ths->pRaftStore->currentTerm) {
+    syncNodeUpdateTerm(ths, pMsg->term);
+  }
+  ASSERT(pMsg->term <= ths->pRaftStore->currentTerm);
+
+  // reset elect timer
+  if (pMsg->term == ths->pRaftStore->currentTerm) {
+    ths->leaderCache = pMsg->srcId;
+    syncNodeResetElectTimer(ths);
+  }
+  ASSERT(pMsg->dataLen >= 0);
+
+  do {
+    // return to follower state
+    if (pMsg->term == ths->pRaftStore->currentTerm && ths->state == TAOS_SYNC_STATE_CANDIDATE) {
+      syncNodeEventLog(ths, "recv sync-append-entries, candidate to follower");
+
+      syncNodeBecomeFollower(ths, "from candidate by append entries");
+
+      // ret or reply?
+      return ret;
+    }
+  } while (0);
+
+  SyncTerm localPreLogTerm = 0;
+  if (pMsg->prevLogIndex >= SYNC_INDEX_BEGIN && pMsg->prevLogIndex <= ths->pLogStore->getLastIndex(ths->pLogStore)) {
+    SSyncRaftEntry* pEntry = ths->pLogStore->getEntry(ths->pLogStore, pMsg->prevLogIndex);
+    if (pEntry == NULL) {
+      char logBuf[128];
+      snprintf(logBuf, sizeof(logBuf), "getEntry error, index:%ld, since %s", pMsg->prevLogIndex, terrstr());
+      syncNodeErrorLog(ths, logBuf);
+      return -1;
+    }
+
+    localPreLogTerm = pEntry->term;
+    syncEntryDestory(pEntry);
+  }
+
+  bool logOK =
+      (pMsg->prevLogIndex == SYNC_INDEX_INVALID) ||
+      ((pMsg->prevLogIndex >= SYNC_INDEX_BEGIN) &&
+       (pMsg->prevLogIndex <= ths->pLogStore->getLastIndex(ths->pLogStore)) && (pMsg->prevLogTerm == localPreLogTerm));
+
+  // reject request
+  if ((pMsg->term < ths->pRaftStore->currentTerm) ||
+      ((pMsg->term == ths->pRaftStore->currentTerm) && (ths->state == TAOS_SYNC_STATE_FOLLOWER) && !logOK)) {
+    do {
+      char logBuf[128];
+      snprintf(logBuf, sizeof(logBuf), "recv sync-append-entries, reject, pre-index:%ld, pre-term:%lu, datalen:%d",
+               pMsg->prevLogIndex, pMsg->prevLogTerm, pMsg->dataLen);
+      syncNodeEventLog(ths, logBuf);
+    } while (0);
+
+    SyncAppendEntriesReply* pReply = syncAppendEntriesReplyBuild(ths->vgId);
+    pReply->srcId = ths->myRaftId;
+    pReply->destId = pMsg->srcId;
+    pReply->term = ths->pRaftStore->currentTerm;
+    pReply->success = false;
+    pReply->matchIndex = SYNC_INDEX_INVALID;
+
+    SRpcMsg rpcMsg;
+    syncAppendEntriesReply2RpcMsg(pReply, &rpcMsg);
+    syncNodeSendMsgById(&pReply->destId, ths, &rpcMsg);
+    syncAppendEntriesReplyDestroy(pReply);
+
+    return ret;
+  }
+
+  // accept request
+  if (pMsg->term == ths->pRaftStore->currentTerm && ths->state == TAOS_SYNC_STATE_FOLLOWER && logOK) {
+    // preIndex = -1, or has preIndex entry in local log
+    ASSERT(pMsg->prevLogIndex <= ths->pLogStore->getLastIndex(ths->pLogStore));
+
+    // has extra entries (> preIndex) in local log
+    bool hasExtraEntries = pMsg->prevLogIndex < ths->pLogStore->getLastIndex(ths->pLogStore);
+
+    // has entries in SyncAppendEntries msg
+    bool hasAppendEntries = pMsg->dataLen > 0;
+
+    do {
+      char logBuf[128];
+      snprintf(logBuf, sizeof(logBuf), "recv sync-append-entries, accept, pre-index:%ld, pre-term:%lu, datalen:%d",
+               pMsg->prevLogIndex, pMsg->prevLogTerm, pMsg->dataLen);
+      syncNodeEventLog(ths, logBuf);
+    } while (0);
+
+    if (hasExtraEntries && hasAppendEntries) {
+      // not conflict by default
+      bool conflict = false;
+
+      SyncIndex       extraIndex = pMsg->prevLogIndex + 1;
+      SSyncRaftEntry* pExtraEntry = ths->pLogStore->getEntry(ths->pLogStore, extraIndex);
+      if (pExtraEntry == NULL) {
+        char logBuf[128];
+        snprintf(logBuf, sizeof(logBuf), "getEntry error2, index:%ld, since %s", extraIndex, terrstr());
+        syncNodeErrorLog(ths, logBuf);
+        return -1;
+      }
+
+      SSyncRaftEntry* pAppendEntry = syncEntryDeserialize(pMsg->data, pMsg->dataLen);
+      if (pAppendEntry == NULL) {
+        syncNodeErrorLog(ths, "syncEntryDeserialize pAppendEntry error");
+        return -1;
+      }
+
+      // log not match, conflict
+      ASSERT(extraIndex == pAppendEntry->index);
+      if (pExtraEntry->term != pAppendEntry->term) {
+        conflict = true;
+      }
+
+      if (conflict) {
+        // roll back
+        SyncIndex delBegin = ths->pLogStore->getLastIndex(ths->pLogStore);
+        SyncIndex delEnd = extraIndex;
+
+        sTrace("syncNodeOnAppendEntriesCb --> conflict:%d, delBegin:%ld, delEnd:%ld", conflict, delBegin, delEnd);
+
+        // notice! reverse roll back!
+        for (SyncIndex index = delEnd; index >= delBegin; --index) {
+          if (ths->pFsm->FpRollBackCb != NULL) {
+            SSyncRaftEntry* pRollBackEntry = ths->pLogStore->getEntry(ths->pLogStore, index);
+            if (pRollBackEntry == NULL) {
+              char logBuf[128];
+              snprintf(logBuf, sizeof(logBuf), "getEntry error3, index:%ld, since %s", index, terrstr());
+              syncNodeErrorLog(ths, logBuf);
+              return -1;
+            }
+
+            // if (pRollBackEntry->msgType != TDMT_SYNC_NOOP) {
+            if (syncUtilUserRollback(pRollBackEntry->msgType)) {
+              SRpcMsg rpcMsg;
+              syncEntry2OriginalRpc(pRollBackEntry, &rpcMsg);
+
+              SFsmCbMeta cbMeta = {0};
+              cbMeta.index = pRollBackEntry->index;
+              cbMeta.lastConfigIndex = syncNodeGetSnapshotConfigIndex(ths, cbMeta.index);
+              cbMeta.isWeak = pRollBackEntry->isWeak;
+              cbMeta.code = 0;
+              cbMeta.state = ths->state;
+              cbMeta.seqNum = pRollBackEntry->seqNum;
+              ths->pFsm->FpRollBackCb(ths->pFsm, &rpcMsg, cbMeta);
+              rpcFreeCont(rpcMsg.pCont);
+            }
+
+            syncEntryDestory(pRollBackEntry);
+          }
+        }
+
+        // delete confict entries
+        ths->pLogStore->truncate(ths->pLogStore, extraIndex);
+
+        // append new entries
+        ths->pLogStore->appendEntry(ths->pLogStore, pAppendEntry);
+
+        // pre commit
+        SRpcMsg rpcMsg;
+        syncEntry2OriginalRpc(pAppendEntry, &rpcMsg);
+        if (ths->pFsm != NULL) {
+          // if (ths->pFsm->FpPreCommitCb != NULL && pAppendEntry->originalRpcType != TDMT_SYNC_NOOP) {
+          if (ths->pFsm->FpPreCommitCb != NULL && syncUtilUserPreCommit(pAppendEntry->originalRpcType)) {
+            SFsmCbMeta cbMeta = {0};
+            cbMeta.index = pAppendEntry->index;
+            cbMeta.lastConfigIndex = syncNodeGetSnapshotConfigIndex(ths, cbMeta.index);
+            cbMeta.isWeak = pAppendEntry->isWeak;
+            cbMeta.code = 2;
+            cbMeta.state = ths->state;
+            cbMeta.seqNum = pAppendEntry->seqNum;
+            ths->pFsm->FpPreCommitCb(ths->pFsm, &rpcMsg, cbMeta);
+          }
+        }
+        rpcFreeCont(rpcMsg.pCont);
+      }
+
+      // free memory
+      syncEntryDestory(pExtraEntry);
+      syncEntryDestory(pAppendEntry);
+
+    } else if (hasExtraEntries && !hasAppendEntries) {
+      // do nothing
+
+    } else if (!hasExtraEntries && hasAppendEntries) {
+      SSyncRaftEntry* pAppendEntry = syncEntryDeserialize(pMsg->data, pMsg->dataLen);
+      if (pAppendEntry == NULL) {
+        syncNodeErrorLog(ths, "syncEntryDeserialize pAppendEntry2 error");
+        return -1;
+      }
+
+      // append new entries
+      ths->pLogStore->appendEntry(ths->pLogStore, pAppendEntry);
+
+      // pre commit
+      SRpcMsg rpcMsg;
+      syncEntry2OriginalRpc(pAppendEntry, &rpcMsg);
+      if (ths->pFsm != NULL) {
+        // if (ths->pFsm->FpPreCommitCb != NULL && pAppendEntry->originalRpcType != TDMT_SYNC_NOOP) {
+        if (ths->pFsm->FpPreCommitCb != NULL && syncUtilUserPreCommit(pAppendEntry->originalRpcType)) {
+          SFsmCbMeta cbMeta = {0};
+          cbMeta.index = pAppendEntry->index;
+          cbMeta.lastConfigIndex = syncNodeGetSnapshotConfigIndex(ths, cbMeta.index);
+          cbMeta.isWeak = pAppendEntry->isWeak;
+          cbMeta.code = 3;
+          cbMeta.state = ths->state;
+          cbMeta.seqNum = pAppendEntry->seqNum;
+          ths->pFsm->FpPreCommitCb(ths->pFsm, &rpcMsg, cbMeta);
+        }
+      }
+      rpcFreeCont(rpcMsg.pCont);
+
+      // free memory
+      syncEntryDestory(pAppendEntry);
+
+    } else if (!hasExtraEntries && !hasAppendEntries) {
+      // do nothing
+
+    } else {
+      syncNodeLog3("", ths);
+      ASSERT(0);
+    }
+
+    SyncAppendEntriesReply* pReply = syncAppendEntriesReplyBuild(ths->vgId);
+    pReply->srcId = ths->myRaftId;
+    pReply->destId = pMsg->srcId;
+    pReply->term = ths->pRaftStore->currentTerm;
+    pReply->success = true;
+
+    if (hasAppendEntries) {
+      pReply->matchIndex = pMsg->prevLogIndex + 1;
+    } else {
+      pReply->matchIndex = pMsg->prevLogIndex;
+    }
+
+    SRpcMsg rpcMsg;
+    syncAppendEntriesReply2RpcMsg(pReply, &rpcMsg);
+    syncNodeSendMsgById(&pReply->destId, ths, &rpcMsg);
+    syncAppendEntriesReplyDestroy(pReply);
+
+    // maybe update commit index from leader
+    if (pMsg->commitIndex > ths->commitIndex) {
+      // has commit entry in local
+      if (pMsg->commitIndex <= ths->pLogStore->getLastIndex(ths->pLogStore)) {
+        SyncIndex beginIndex = ths->commitIndex + 1;
+        SyncIndex endIndex = pMsg->commitIndex;
+
+        // update commit index
+        ths->commitIndex = pMsg->commitIndex;
+
+        // call back Wal
+        ths->pLogStore->updateCommitIndex(ths->pLogStore, ths->commitIndex);
+
+        int32_t code = syncNodeCommit(ths, beginIndex, endIndex, ths->state);
+        ASSERT(code == 0);
+      }
+    }
+  }
+
+  return ret;
+}
+
+#if 0
+
 int32_t syncNodeOnAppendEntriesCb(SSyncNode* ths, SyncAppendEntries* pMsg) {
   int32_t ret = 0;
 
@@ -352,6 +626,8 @@ int32_t syncNodeOnAppendEntriesCb(SSyncNode* ths, SyncAppendEntries* pMsg) {
   return ret;
 }
 
+#endif
+
 static int32_t syncNodeMakeLogSame(SSyncNode* ths, SyncAppendEntries* pMsg) {
   int32_t code;
 

source/libs/sync/src/syncAppendEntriesReply.c

@@ -40,6 +40,78 @@
 int32_t syncNodeOnAppendEntriesReplyCb(SSyncNode* ths, SyncAppendEntriesReply* pMsg) {
   int32_t ret = 0;
 
+  // print log
+  syncAppendEntriesReplyLog2("==syncNodeOnAppendEntriesReplyCb==", pMsg);
+
+  // if already drop replica, do not process
+  if (!syncNodeInRaftGroup(ths, &(pMsg->srcId)) && !ths->pRaftCfg->isStandBy) {
+    syncNodeEventLog(ths, "recv sync-append-entries-reply,  maybe replica already dropped");
+    return 0;
+  }
+
+  // drop stale response
+  if (pMsg->term < ths->pRaftStore->currentTerm) {
+    char logBuf[128];
+    snprintf(logBuf, sizeof(logBuf), "recv sync-append-entries-reply, recv-term:%lu, drop stale response", pMsg->term);
+    syncNodeEventLog(ths, logBuf);
+    return 0;
+  }
+
+  if (gRaftDetailLog) {
+    syncNodeEventLog(ths, "recv sync-append-entries-reply, before");
+  }
+  syncIndexMgrLog2("==syncNodeOnAppendEntriesReplyCb== before pNextIndex", ths->pNextIndex);
+  syncIndexMgrLog2("==syncNodeOnAppendEntriesReplyCb== before pMatchIndex", ths->pMatchIndex);
+
+  // no need this code, because if I receive reply.term, then I must have sent for that term.
+  //  if (pMsg->term > ths->pRaftStore->currentTerm) {
+  //    syncNodeUpdateTerm(ths, pMsg->term);
+  //  }
+
+  if (pMsg->term > ths->pRaftStore->currentTerm) {
+    char logBuf[128];
+    snprintf(logBuf, sizeof(logBuf), "recv sync-append-entries-reply, error term, recv-term:%lu", pMsg->term);
+    syncNodeErrorLog(ths, logBuf);
+    return -1;
+  }
+
+  ASSERT(pMsg->term == ths->pRaftStore->currentTerm);
+
+  if (pMsg->success) {
+    // nextIndex'  = [nextIndex  EXCEPT ![i][j] = m.mmatchIndex + 1]
+    syncIndexMgrSetIndex(ths->pNextIndex, &(pMsg->srcId), pMsg->matchIndex + 1);
+
+    // matchIndex' = [matchIndex EXCEPT ![i][j] = m.mmatchIndex]
+    syncIndexMgrSetIndex(ths->pMatchIndex, &(pMsg->srcId), pMsg->matchIndex);
+
+    // maybe commit
+    syncMaybeAdvanceCommitIndex(ths);
+
+  } else {
+    SyncIndex nextIndex = syncIndexMgrGetIndex(ths->pNextIndex, &(pMsg->srcId));
+
+    // notice! int64, uint64
+    if (nextIndex > SYNC_INDEX_BEGIN) {
+      --nextIndex;
+    } else {
+      nextIndex = SYNC_INDEX_BEGIN;
+    }
+    syncIndexMgrSetIndex(ths->pNextIndex, &(pMsg->srcId), nextIndex);
+  }
+
+  if (gRaftDetailLog) {
+    syncNodeEventLog(ths, "recv sync-append-entries-reply, after");
+  }
+  syncIndexMgrLog2("==syncNodeOnAppendEntriesReplyCb== after pNextIndex", ths->pNextIndex);
+  syncIndexMgrLog2("==syncNodeOnAppendEntriesReplyCb== after pMatchIndex", ths->pMatchIndex);
+
+  return ret;
+}
+
+#if 0
+int32_t syncNodeOnAppendEntriesReplyCb(SSyncNode* ths, SyncAppendEntriesReply* pMsg) {
+  int32_t ret = 0;
+
   char logBuf[128] = {0};
   snprintf(logBuf, sizeof(logBuf), "==syncNodeOnAppendEntriesReplyCb== term:%lu", ths->pRaftStore->currentTerm);
   syncAppendEntriesReplyLog2(logBuf, pMsg);
@@ -96,6 +168,7 @@ int32_t syncNodeOnAppendEntriesReplyCb(SSyncNode* ths, SyncAppendEntriesReply* p
 
   return ret;
 }
+#endif
 
 int32_t syncNodeOnAppendEntriesReplySnapshotCb(SSyncNode* ths, SyncAppendEntriesReply* pMsg) {
   int32_t ret = 0;

source/libs/sync/src/syncRequestVote.c

@@ -45,6 +45,75 @@
 int32_t syncNodeOnRequestVoteCb(SSyncNode* ths, SyncRequestVote* pMsg) {
   int32_t ret = 0;
 
+  syncRequestVoteLog2("==syncNodeOnRequestVoteCb==", pMsg);
+
+  // if already drop replica, do not process
+  if (!syncNodeInRaftGroup(ths, &(pMsg->srcId)) && !ths->pRaftCfg->isStandBy) {
+    do {
+      char     logBuf[256];
+      char     host[64];
+      uint16_t port;
+      syncUtilU642Addr(pMsg->srcId.addr, host, sizeof(host), &port);
+      snprintf(logBuf, sizeof(logBuf),
+               "recv sync-request-vote from %s:%d, term:%lu, lindex:%ld, lterm:%lu, maybe replica already dropped",
+               host, port, pMsg->term, pMsg->lastLogIndex, pMsg->lastLogTerm);
+      syncNodeEventLog(ths, logBuf);
+    } while (0);
+
+    return -1;
+  }
+
+  // maybe update term
+  if (pMsg->term > ths->pRaftStore->currentTerm) {
+    syncNodeUpdateTerm(ths, pMsg->term);
+  }
+  ASSERT(pMsg->term <= ths->pRaftStore->currentTerm);
+
+  bool logOK = (pMsg->lastLogTerm > ths->pLogStore->getLastTerm(ths->pLogStore)) ||
+               ((pMsg->lastLogTerm == ths->pLogStore->getLastTerm(ths->pLogStore)) &&
+                (pMsg->lastLogIndex >= ths->pLogStore->getLastIndex(ths->pLogStore)));
+  bool grant = (pMsg->term == ths->pRaftStore->currentTerm) && logOK &&
+               ((!raftStoreHasVoted(ths->pRaftStore)) || (syncUtilSameId(&(ths->pRaftStore->voteFor), &(pMsg->srcId))));
+  if (grant) {
+    // maybe has already voted for pMsg->srcId
+    // vote again, no harm
+    raftStoreVote(ths->pRaftStore, &(pMsg->srcId));
+
+    // forbid elect for this round
+    syncNodeResetElectTimer(ths);
+  }
+
+  // send msg
+  SyncRequestVoteReply* pReply = syncRequestVoteReplyBuild(ths->vgId);
+  pReply->srcId = ths->myRaftId;
+  pReply->destId = pMsg->srcId;
+  pReply->term = ths->pRaftStore->currentTerm;
+  pReply->voteGranted = grant;
+
+  // trace log
+  do {
+    char     logBuf[256];
+    char     host[64];
+    uint16_t port;
+    syncUtilU642Addr(pMsg->srcId.addr, host, sizeof(host), &port);
+    snprintf(logBuf, sizeof(logBuf),
+             "recv sync-request-vote from %s:%d, term:%lu, lindex:%ld, lterm:%lu, reply-grant:%d", host, port,
+             pMsg->term, pMsg->lastLogIndex, pMsg->lastLogTerm, pReply->voteGranted);
+    syncNodeEventLog(ths, logBuf);
+  } while (0);
+
+  SRpcMsg rpcMsg;
+  syncRequestVoteReply2RpcMsg(pReply, &rpcMsg);
+  syncNodeSendMsgById(&pReply->destId, ths, &rpcMsg);
+  syncRequestVoteReplyDestroy(pReply);
+
+  return ret;
+}
+
+#if 0
+int32_t syncNodeOnRequestVoteCb(SSyncNode* ths, SyncRequestVote* pMsg) {
+  int32_t ret = 0;
+
   char logBuf[128] = {0};
   snprintf(logBuf, sizeof(logBuf), "==syncNodeOnRequestVoteCb== term:%lu", ths->pRaftStore->currentTerm);
   syncRequestVoteLog2(logBuf, pMsg);
@@ -81,6 +150,7 @@ int32_t syncNodeOnRequestVoteCb(SSyncNode* ths, SyncRequestVote* pMsg) {
 
   return ret;
 }
+#endif
 
 static bool syncNodeOnRequestVoteLogOK(SSyncNode* pSyncNode, SyncRequestVote* pMsg) {
   SyncTerm  myLastTerm = syncNodeGetLastTerm(pSyncNode);

source/libs/sync/src/syncRequestVoteReply.c

@@ -40,6 +40,72 @@
 int32_t syncNodeOnRequestVoteReplyCb(SSyncNode* ths, SyncRequestVoteReply* pMsg) {
   int32_t ret = 0;
 
+  // print log
+  char logBuf[128] = {0};
+  snprintf(logBuf, sizeof(logBuf), "==syncNodeOnRequestVoteReplyCb== term:%lu", ths->pRaftStore->currentTerm);
+  syncRequestVoteReplyLog2(logBuf, pMsg);
+
+  // if already drop replica, do not process
+  if (!syncNodeInRaftGroup(ths, &(pMsg->srcId)) && !ths->pRaftCfg->isStandBy) {
+    sInfo("recv SyncRequestVoteReply, maybe replica already dropped");
+    return ret;
+  }
+
+  // drop stale response
+  if (pMsg->term < ths->pRaftStore->currentTerm) {
+    sTrace("recv SyncRequestVoteReply, drop stale response, receive_term:%lu current_term:%lu", pMsg->term,
+           ths->pRaftStore->currentTerm);
+    return ret;
+  }
+
+  // ASSERT(!(pMsg->term > ths->pRaftStore->currentTerm));
+  //  no need this code, because if I receive reply.term, then I must have sent for that term.
+  //   if (pMsg->term > ths->pRaftStore->currentTerm) {
+  //     syncNodeUpdateTerm(ths, pMsg->term);
+  //   }
+
+  if (pMsg->term > ths->pRaftStore->currentTerm) {
+    char logBuf[128] = {0};
+    snprintf(logBuf, sizeof(logBuf), "syncNodeOnRequestVoteReplyCb error term, receive:%lu current:%lu", pMsg->term,
+             ths->pRaftStore->currentTerm);
+    syncNodePrint2(logBuf, ths);
+    sError("%s", logBuf);
+    return ret;
+  }
+
+  ASSERT(pMsg->term == ths->pRaftStore->currentTerm);
+
+  // This tallies votes even when the current state is not Candidate,
+  // but they won't be looked at, so it doesn't matter.
+  if (ths->state == TAOS_SYNC_STATE_CANDIDATE) {
+    votesRespondAdd(ths->pVotesRespond, pMsg);
+    if (pMsg->voteGranted) {
+      // add vote
+      voteGrantedVote(ths->pVotesGranted, pMsg);
+
+      // maybe to leader
+      if (voteGrantedMajority(ths->pVotesGranted)) {
+        if (!ths->pVotesGranted->toLeader) {
+          syncNodeCandidate2Leader(ths);
+
+          // prevent to leader again!
+          ths->pVotesGranted->toLeader = true;
+        }
+      }
+    } else {
+      ;
+      // do nothing
+      // UNCHANGED <<votesGranted, voterLog>>
+    }
+  }
+
+  return ret;
+}
+
+#if 0
+int32_t syncNodeOnRequestVoteReplyCb(SSyncNode* ths, SyncRequestVoteReply* pMsg) {
+  int32_t ret = 0;
+
   char logBuf[128] = {0};
   snprintf(logBuf, sizeof(logBuf), "==syncNodeOnRequestVoteReplyCb== term:%lu", ths->pRaftStore->currentTerm);
   syncRequestVoteReplyLog2(logBuf, pMsg);
@@ -93,6 +159,7 @@ int32_t syncNodeOnRequestVoteReplyCb(SSyncNode* ths, SyncRequestVoteReply* pMsg)
 
   return ret;
 }
+#endif
 
 int32_t syncNodeOnRequestVoteReplySnapshotCb(SSyncNode* ths, SyncRequestVoteReply* pMsg) {
   int32_t ret = 0;

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

@@ -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);
@@ -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

@@ -455,7 +455,7 @@ 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);

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/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/osSocket.c

@@ -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/script/jenkins/basic.txt

@@ -2,12 +2,10 @@
 #======================b1-start===============
 
 # ---- user
-./test.sh -f tsim/user/basic1.sim
+./test.sh -f tsim/user/basic.sim
-./test.sh -f tsim/user/pass_alter.sim
+./test.sh -f tsim/user/password.sim
-./test.sh -f tsim/user/pass_len.sim
+./test.sh -f tsim/user/privilege_db.sim
-./test.sh -f tsim/user/user_len.sim
+./test.sh -f tsim/user/privilege_sysinfo.sim
-./test.sh -f tsim/user/privilege1.sim
-./test.sh -f tsim/user/privilege2.sim
 
 ## ---- db
 ./test.sh -f tsim/db/create_all_options.sim
@@ -70,7 +68,7 @@
 ./test.sh -f tsim/mnode/basic2.sim
 ./test.sh -f tsim/mnode/basic3.sim
 ./test.sh -f tsim/mnode/basic4.sim
-./test.sh -f tsim/mnode/basic5.sim
+#./test.sh -f tsim/mnode/basic5.sim
 
 # ---- show
 ./test.sh -f tsim/show/basic.sim
@@ -135,7 +133,7 @@
 ./test.sh -f tsim/stable/tag_filter.sim
 
 # --- for multi process mode
-./test.sh -f tsim/user/basic1.sim -m
+./test.sh -f tsim/user/basic.sim -m
 ./test.sh -f tsim/db/basic3.sim -m
 ./test.sh -f tsim/db/error1.sim -m
 ./test.sh -f tsim/insert/backquote.sim -m

tests/script/tsim/user/basic.sim

@@ -0,0 +1,157 @@
+system sh/stop_dnodes.sh
+system sh/deploy.sh -n dnode1 -i 1
+system sh/exec.sh -n dnode1 -s start
+sql connect
+
+print =============== step0
+sql show users
+if $data(root)[1] != 1 then
+  return -1
+endi
+if $data(root)[2] != 1 then
+  return -1
+endi
+if $data(root)[3] != 1 then
+  return -1
+endi
+
+sql alter user root pass 'taosdata'
+
+sql_error ALTER USER root SYSINFO 0
+sql_error ALTER USER root SYSINFO 1
+sql_error ALTER USER root enable 0
+sql_error ALTER USER root enable 1
+
+sql_error create database db vgroups 1;
+sql_error GRANT read ON db.* to root;
+sql_error GRANT read ON *.* to root;
+sql_error REVOKE read ON db.* from root;
+sql_error REVOKE read ON *.* from root;
+sql_error GRANT write ON db.* to root;
+sql_error GRANT write ON *.* to root;
+sql_error REVOKE write ON db.* from root;
+sql_error REVOKE write ON *.* from root;
+sql_error REVOKE write ON *.* from root;
+
+sql_error GRANT all ON *.* to root;
+sql_error REVOKE all ON *.* from root;
+sql_error GRANT read,write ON *.* to root;
+sql_error REVOKE read,write ON *.* from root;
+
+print =============== step1: sysinfo create
+sql CREATE USER u1 PASS 'taosdata' SYSINFO 0;
+sql show users
+if $rows != 2 then 
+  return -1
+endi
+if $data(u1)[1] != 0 then
+  return -1
+endi
+if $data(u1)[2] != 1 then
+  return -1
+endi
+if $data(u1)[3] != 0 then
+  return -1
+endi
+
+sql CREATE USER u2 PASS 'taosdata' SYSINFO 1;
+sql show users
+if $rows != 3 then 
+  return -1
+endi
+if $data(u2)[1] != 0 then
+  return -1
+endi
+if $data(u2)[2] != 1 then
+  return -1
+endi
+if $data(u2)[3] != 1 then
+  return -1
+endi
+
+print =============== step2: sysinfo alter
+sql ALTER USER u1 SYSINFO 1
+sql show users
+if $data(u1)[1] != 0 then
+  return -1
+endi
+if $data(u1)[2] != 1 then
+  return -1
+endi
+if $data(u1)[3] != 1 then
+  return -1
+endi
+
+sql ALTER USER u1 SYSINFO 0
+sql show users
+if $data(u1)[1] != 0 then
+  return -1
+endi
+if $data(u1)[2] != 1 then
+  return -1
+endi
+if $data(u1)[3] != 0 then
+  return -1
+endi
+
+sql ALTER USER u1 SYSINFO 0
+sql ALTER USER u1 SYSINFO 0
+
+sql drop user u1
+sql show users
+if $rows != 2 then 
+  return -1
+endi
+
+print =============== step3: enable alter
+sql ALTER USER u2 enable 0
+sql show users
+if $rows != 2 then 
+  return -1
+endi
+if $data(u2)[1] != 0 then
+  return -1
+endi
+if $data(u2)[2] != 0 then
+  return -1
+endi
+if $data(u2)[3] != 1 then
+  return -1
+endi
+
+sql ALTER USER u2 enable 1
+sql show users
+if $data(u2)[1] != 0 then
+  return -1
+endi
+if $data(u2)[2] != 1 then
+  return -1
+endi
+if $data(u2)[3] != 1 then
+  return -1
+endi
+
+sql ALTER USER u2 enable 1
+sql ALTER USER u2 enable 1
+
+print =============== restart taosd
+system sh/exec.sh -n dnode1 -s stop
+system sh/exec.sh -n dnode1 -s start
+
+print =============== step4: enable privilege
+sql show users
+if $rows != 2 then 
+  return -1
+endi
+if $data(u2)[1] != 0 then
+  return -1
+endi
+if $data(u2)[2] != 1 then
+  return -1
+endi
+if $data(u2)[3] != 1 then
+  return -1
+endi
+
+
+system sh/exec.sh -n dnode1 -s stop -x SIGINT

tests/script/tsim/user/basic1.sim

@@ -1,74 +0,0 @@
-system sh/stop_dnodes.sh
-system sh/deploy.sh -n dnode1 -i 1
-system sh/exec.sh -n dnode1 -s start
-sql connect
-
-print =============== show users
-sql show users
-if $rows != 1 then 
-  return -1
-endi
-
-print $data[0][0] $data[0][1] $data[0][2]
-print $data[1][0] $data[1][1] $data[1][2]
-print $data[2][0] $data[1][2] $data[2][2]
-
-sql_error show accounts;
-sql_error create account a pass "a"
-sql_error drop account a
-sql_error drop account root
-
-print =============== create user1
-sql create user user1 PASS 'user1'
-sql show users
-if $rows != 2 then 
-  return -1
-endi
-
-print $data[0][0] $data[0][1] $data[0][2]
-print $data[1][0] $data[1][1] $data[1][2]
-print $data[2][0] $data[1][2] $data[2][2]
-print $data[3][0] $data[3][1] $data[3][2]
-
-print =============== create user2
-sql create user user2 PASS 'user2'
-sql show users
-if $rows != 3 then 
-  return -1
-endi
-
-print $data[0][0] $data[0][1] $data[0][2]
-print $data[1][0] $data[1][1] $data[1][2]
-print $data[2][0] $data[1][2] $data[2][2]
-print $data[3][0] $data[3][1] $data[3][2]
-print $data40 $data41 $data42
-
-print =============== drop user1
-sql drop user user1
-sql show users
-if $rows != 2 then 
-  return -1
-endi
-
-print $data[0][0] $data[0][1] $data[0][2]
-print $data[1][0] $data[1][1] $data[1][2]
-print $data[2][0] $data[1][2] $data[2][2]
-print $data[3][0] $data[3][1] $data[3][2]
-
-print =============== restart taosd
-system sh/exec.sh -n dnode1 -s stop
-sleep 1000
-system sh/exec.sh -n dnode1 -s start
-
-print =============== show users
-sql show users
-if $rows != 2 then 
-  return -1
-endi
-
-print $data[0][0] $data[0][1] $data[0][2]
-print $data[1][0] $data[1][1] $data[1][2]
-print $data[2][0] $data[1][2] $data[2][2]
-print $data[3][0] $data[3][1] $data[3][2]
-
-system sh/exec.sh -n dnode1 -s stop -x SIGINT

tests/script/tsim/user/pass_alter.sim

@@ -1,66 +0,0 @@
-system sh/stop_dnodes.sh
-system sh/deploy.sh -n dnode1 -i 1
-system sh/exec.sh -n dnode1 -s start
-sql connect
-
-print ============= step1
-sql create user u_read pass 'taosdata1'
-sql create user u_write pass 'taosdata1'
-
-sql alter user u_read pass 'taosdata'
-sql alter user u_write pass 'taosdata'
-
-sql show users
-if $rows != 3 then 
-  return -1
-endi
-
-print ============= step2
-sql close
-sleep 2500
-print user u_read login
-sql connect u_read
-sql alter user u_read pass 'taosdata'
-sql alter user u_write pass 'taosdata1' -x step2
-  return -1
-step2:
-
-sql_error create user read1 pass 'taosdata1'
-sql_error create user write1 pass 'taosdata1'
-
-sql show users
-if $rows != 3 then 
-  return -1
-endi
-
-print ============= step3
-sql close
-sleep 2500
-print user u_write login
-sql connect u_write
-
-sql_error create user read2 pass 'taosdata1'
-sql_error create user write2 pass 'taosdata1'
-sql alter user u_write pass 'taosdata'
-sql alter user u_read pass 'taosdata' -x step3
-  return -1
-step3:
-
-sql show users
-if $rows != 3 then 
-  return -1
-endi
-
-print ============= step4
-sql close 
-sleep 2500
-print user root login
-sql connect
-sql create user oroot pass 'taosdata'
-
-sql show users
-if $rows != 4 then 
-  return -1
-endi
-
-system sh/exec.sh -n dnode1 -s stop -x SIGINT

tests/script/tsim/user/pass_len.sim

@@ -1,79 +0,0 @@
-system sh/stop_dnodes.sh
-system sh/deploy.sh -n dnode1 -i 1
-system sh/exec.sh -n dnode1 -s start
-sql connect
-
-$i = 0
-$dbPrefix = apdb
-$tbPrefix = aptb
-$db = $dbPrefix . $i
-$tb = $tbPrefix . $i
-$userPrefix = apusr
-
-print =============== step1
-$i = 0
-$user = $userPrefix . $i
-
-sql drop user $user -x step11
-  return -1
-step11:
-
-sql create user $user PASS -x step12
-  return -1
-step12:
-
-sql create user $user PASS 'taosdata'
-
-sql show users
-if $rows != 2 then 
-  return -1
-endi
-
-print =============== step2
-$i = 1
-$user = $userPrefix . $i
-sql drop user $user -x step2
-step2:
-sql create user $user PASS '1'
-sql show users
-if $rows != 3 then 
-  return -1
-endi
-
-print =============== step3
-$i = 2
-$user = $userPrefix . $i
-sql drop user $user -x step3
-step3:
-
-sql create user $user PASS 'abc0123456789'
-sql show users
-if $rows != 4 then 
-  return -1
-endi
-
-print =============== step4
-$i = 3
-$user = $userPrefix . $i
-sql create user $user PASS 'abcd012345678901234567891234567890abcd012345678901234567891234567890abcd012345678901234567891234567890abcd012345678901234567891234567890123' -x step4
-	return -1
-
-step4:
-sql show users
-if $rows != 4 then 
-  return -1
-endi
-
-$i = 0
-while $i < 3
-  $user = $userPrefix . $i
-  sql drop user $user 
-  $i = $i + 1
-endw
-
-sql show users
-if $rows != 1 then 
-  return -1
-endi
-
-system sh/exec.sh -n dnode1 -s stop -x SIGINT

tests/script/tsim/user/password.sim

@@ -0,0 +1,87 @@
+system sh/stop_dnodes.sh
+system sh/deploy.sh -n dnode1 -i 1
+system sh/exec.sh -n dnode1 -s start
+sql connect
+
+print ============= step1
+sql create user u_read pass 'taosdata1'
+sql create user u_write pass 'taosdata1'
+
+sql alter user u_read pass 'taosdata'
+sql alter user u_write pass 'taosdata'
+
+sql show users
+if $rows != 3 then 
+  return -1
+endi
+
+print ============= step2
+print user u_read login
+sql close
+sql connect u_read
+
+sql alter user u_read pass 'taosdata'
+sql_error alter user u_write pass 'taosdata1' 
+
+sql_error create user read1 pass 'taosdata1'
+sql_error create user write1 pass 'taosdata1'
+
+sql show users
+if $rows != 3 then 
+  return -1
+endi
+
+print ============= step3
+print user u_write login
+sql close
+sql connect u_write
+
+sql_error create user read2 pass 'taosdata1'
+sql_error create user write2 pass 'taosdata1'
+sql alter user u_write pass 'taosdata'
+sql_error alter user u_read pass 'taosdata' 
+
+sql show users
+if $rows != 3 then 
+  return -1
+endi
+
+print ============= step4
+print user root login
+sql close 
+sql connect
+sql create user oroot pass 'taosdata'
+sql_error create user $user PASS 'abcd012345678901234567891234567890abcd012345678901234567891234567890abcd012345678901234567891234567890abcd012345678901234567891234567890123'
+sql_error create userabcd012345678901234567891234567890abcd01234567890123456789123456789  PASS 'taosdata'
+sql_error create user abcd0123456789012345678901234567890111 PASS '123'
+sql create user abc01234567890123456789 PASS '123'
+
+sql show users
+if $rows != 5 then 
+  return -1
+endi
+
+print ============= step5
+sql create database db vgroups 1
+sql_error ALTER USER o_root SYSINFO 0
+sql_error ALTER USER o_root SYSINFO 1
+sql_error ALTER USER o_root enable 0
+sql_error ALTER USER o_root enable 1
+
+sql_error create database db vgroups 1;
+sql_error GRANT read ON db.* to o_root;
+sql_error GRANT read ON *.* to o_root;
+sql_error REVOKE read ON db.* from o_root;
+sql_error REVOKE read ON *.* from o_root;
+sql_error GRANT write ON db.* to o_root;
+sql_error GRANT write ON *.* to o_root;
+sql_error REVOKE write ON db.* from o_root;
+sql_error REVOKE write ON *.* from o_root;
+sql_error REVOKE write ON *.* from o_root;
+
+sql_error GRANT all ON *.* to o_root;
+sql_error REVOKE all ON *.* from o_root;
+sql_error GRANT read,write ON *.* to o_root;
+sql_error REVOKE read,write ON *.* from o_root;
+
+system sh/exec.sh -n dnode1 -s stop -x SIGINT

tests/script/tsim/user/privilege2.sim

@@ -1,38 +0,0 @@
-system sh/stop_dnodes.sh
-system sh/deploy.sh -n dnode1 -i 1
-system sh/exec.sh -n dnode1 -s start
-sql connect
-
-print =============== show users
-sql create database d1 vgroups 1;
-sql create database d2 vgroups 1;
-sql create database d3 vgroups 1;
-sql show databases
-if $rows != 5 then 
-  return -1
-endi
-
-print =============== create users
-sql create user user1 PASS 'taosdata'
-sql create user user2 PASS 'taosdata'
-sql show users
-if $rows != 3 then 
-  return -1
-endi
-
-sql GRANT read ON d1.* to user1;
-sql GRANT write ON d2.* to user1;
-
-print =============== re connect
-sql close
-sleep 2500
-print user user1 login
-sql connect user1
-
-sql_error drop database d1;
-sql_error drop database d2;
-
-sql_error create stable d1.st (ts timestamp, i int) tags (j int)
-sql create stable d2.st (ts timestamp, i int) tags (j int)
-
-system sh/exec.sh -n dnode1 -s stop -x SIGINT

tests/script/tsim/user/privilege_db.sim

@@ -3,7 +3,7 @@ system sh/deploy.sh -n dnode1 -i 1
 system sh/exec.sh -n dnode1 -s start
 sql connect
 
-print =============== show users
+print =============== create db
 sql create database d1 vgroups 1;
 sql create database d2 vgroups 1;
 sql create database d3 vgroups 1;
@@ -68,4 +68,26 @@ sql REVOKE read,write ON d1.* from user1;
 sql REVOKE read,write ON d2.* from user1;
 sql REVOKE read,write ON *.* from user1;
 
+
+print =============== create users
+sql create user u1 PASS 'taosdata'
+sql show users
+if $rows != 4 then 
+  return -1
+endi
+
+sql GRANT read ON d1.* to u1;
+sql GRANT write ON d2.* to u1;
+
+print =============== re connect
+print user u1 login
+sql close
+sql connect u1
+
+sql_error drop database d1;
+sql_error drop database d2;
+
+sql_error create stable d1.st (ts timestamp, i int) tags (j int)
+sql create stable d2.st (ts timestamp, i int) tags (j int)
+
 system sh/exec.sh -n dnode1 -s stop -x SIGINT

tests/script/tsim/user/privilege_sysinfo.sim

@@ -0,0 +1,44 @@
+system sh/stop_dnodes.sh
+system sh/deploy.sh -n dnode1 -i 1
+system sh/exec.sh -n dnode1 -s start
+sql connect
+
+print =============== create user and login
+sql create user sysinfo0 pass 'taosdata'
+sql create user sysinfo1 pass 'taosdata'
+sql alter user sysinfo0 sysinfo 0
+sql alter user sysinfo1 sysinfo 1
+
+print user sysinfo0 login
+sql close
+sql connect sysinfo0
+
+print =============== check oper
+sql_error create user u1 pass 'u1'
+sql_error drop user sysinfo1
+sql_error alter user sysinfo1 pass '1'
+sql_error alter user sysinfo0 pass '1'
+
+sql_error create dnode $hostname port 7200
+sql_error drop dnode 1
+
+sql_error create qnode on dnode 1
+sql_error drop qnode on dnode 1
+
+sql_error create mnode on dnode 1
+sql_error drop mnode on dnode 1
+
+sql_error create snode on dnode 1
+sql_error drop snode on dnode 1
+
+sql_error redistribute vgroup 2 dnode 1 dnode 2
+sql_error balance vgroup
+
+sql_error kill transaction 1
+sql_error kill connection 1
+sql_error kill query 1
+
+print =============== check db
+sql_error create database db
+
+system sh/exec.sh -n dnode1 -s stop -x SIGINT

tests/script/tsim/user/user_len.sim

@@ -1,85 +0,0 @@
-system sh/stop_dnodes.sh
-system sh/deploy.sh -n dnode1 -i 1
-system sh/exec.sh -n dnode1 -s start
-sql connect
-
-$i = 0
-$dbPrefix = lm_us_db
-$tbPrefix = lm_us_tb
-$db = $dbPrefix . $i
-$tb = $tbPrefix . $i
-
-print =============== step1
-sql drop user ac -x step0
-	return -1
-step0:
-
-sql create user PASS '123' -x step1
-	return -1
-step1:
-
-sql show users
-if $rows != 1 then 
-  return -1
-endi
-
-print =============== step2
-sql drop user a -x step2
-step2:
-sql create user a PASS '123'
-sql show users
-if $rows != 2 then 
-  return -1
-endi
-
-sql drop user a
-sql show users
-if $rows != 1 then 
-  return -1
-endi
-
-print =============== step3
-sql drop user abc01234567890123456789 -x step3
-step3:
-
-sql create user abc01234567890123456789 PASS '123'
-sql show users
-if $rows != 2 then 
-  return -1
-endi
-
-sql drop user abc01234567890123456789
-sql show users
-if $rows != 1 then 
-  return -1
-endi
-
-print =============== step4
-sql create user abcd0123456789012345678901234567890111 PASS '123' -x step4
-	return -1
-step4:
-sql show users
-if $rows != 1 then 
-  return -1
-endi
-
-print =============== step5
-sql drop user 123 -x step5
-step5:
-sql create user 123 PASS '123' -x step61
-  return -1
-step61:
-
-sql create user a123 PASS '123'
-sql show users
-if $rows != 2 then 
-  return -1
-endi
-
-sql drop user a123
-sql show users
-if $rows != 1 then 
-  return -1
-endi
-
-system sh/exec.sh -n dnode1 -s stop -x SIGINT

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

@@ -412,52 +412,59 @@ class TDTestCase:
         tdSql.checkColNameList(res, cname_list)
         #
         # test group by & order by  json tag
+        tdSql.query("select ts,jtag->'tag1' from jsons1 partition by jtag->'tag1' order by jtag->'tag1' desc")
+        tdSql.checkRows(11)
+        tdSql.checkData(0, 1, '"femail"')
+        tdSql.checkData(2, 1, '"收到货"')
+        tdSql.checkData(7, 1, "false")
+
+
         # tdSql.error("select count(*) from jsons1 group by jtag")
         # tdSql.error("select count(*) from jsons1 partition by jtag")
         # tdSql.error("select count(*) from jsons1 group by jtag order by jtag")
         tdSql.error("select count(*) from jsons1 group by jtag->'tag1' order by jtag->'tag2'")
         tdSql.error("select count(*) from jsons1 group by jtag->'tag1' order by jtag")
-        tdSql.query("select count(*),jtag->'tag1' from jsons1 group by jtag->'tag1' order by jtag->'tag1' desc")
+        # tdSql.query("select count(*),jtag->'tag1' from jsons1 group by jtag->'tag1' order by jtag->'tag1' desc")
-        tdSql.checkRows(8)
+        # tdSql.checkRows(8)
-        tdSql.checkData(0, 0, 2)
+        # tdSql.checkData(0, 0, 2)
-        tdSql.checkData(0, 1, '"femail"')
+        # tdSql.checkData(0, 1, '"femail"')
-        tdSql.checkData(1, 0, 2)
+        # tdSql.checkData(1, 0, 2)
-        tdSql.checkData(1, 1, '"收到货"')
+        # tdSql.checkData(1, 1, '"收到货"')
-        tdSql.checkData(2, 0, 1)
+        # tdSql.checkData(2, 0, 1)
-        tdSql.checkData(2, 1, "11.000000000")
+        # tdSql.checkData(2, 1, "11.000000000")
-        tdSql.checkData(5, 0, 1)
+        # tdSql.checkData(5, 0, 1)
-        tdSql.checkData(5, 1, "false")
+        # tdSql.checkData(5, 1, "false")
-        tdSql.checkData(6, 0, 1)
+        # tdSql.checkData(6, 0, 1)
-        tdSql.checkData(6, 1, "null")
+        # tdSql.checkData(6, 1, "null")
-        tdSql.checkData(7, 0, 2)
+        # tdSql.checkData(7, 0, 2)
-        tdSql.checkData(7, 1, None)
+        # tdSql.checkData(7, 1, None)
 
-        tdSql.query("select count(*),jtag->'tag1' from jsons1 group by jtag->'tag1' order by jtag->'tag1' asc")
+        # tdSql.query("select count(*),jtag->'tag1' from jsons1 group by jtag->'tag1' order by jtag->'tag1' asc")
-        tdSql.checkRows(8)
+        # tdSql.checkRows(8)
-        tdSql.checkData(0, 0, 2)
+        # tdSql.checkData(0, 0, 2)
-        tdSql.checkData(0, 1, None)
+        # tdSql.checkData(0, 1, None)
-        tdSql.checkData(2, 0, 1)
+        # tdSql.checkData(2, 0, 1)
-        tdSql.checkData(2, 1, "false")
+        # tdSql.checkData(2, 1, "false")
-        tdSql.checkData(5, 0, 1)
+        # tdSql.checkData(5, 0, 1)
-        tdSql.checkData(5, 1, "11.000000000")
+        # tdSql.checkData(5, 1, "11.000000000")
-        tdSql.checkData(7, 0, 2)
+        # tdSql.checkData(7, 0, 2)
-        tdSql.checkData(7, 1, '"femail"')
+        # tdSql.checkData(7, 1, '"femail"')
         #
         # test stddev with group by json tag
-        tdSql.query("select stddev(dataint),jtag->'tag1' from jsons1 group by jtag->'tag1' order by jtag->'tag1'")
+        # tdSql.query("select stddev(dataint),jtag->'tag1' from jsons1 group by jtag->'tag1' order by jtag->'tag1'")
-        tdSql.checkRows(8)
+        # tdSql.checkRows(8)
-        tdSql.checkData(0, 0, 10)
+        # tdSql.checkData(0, 0, 10)
-        tdSql.checkData(0, 1, None)
+        # tdSql.checkData(0, 1, None)
-        tdSql.checkData(4, 0, 0)
+        # tdSql.checkData(4, 0, 0)
-        tdSql.checkData(4, 1, "5.000000000")
+        # tdSql.checkData(4, 1, "5.000000000")
-        tdSql.checkData(7, 0, 11)
+        # tdSql.checkData(7, 0, 11)
-        tdSql.checkData(7, 1, '"femail"')
+        # tdSql.checkData(7, 1, '"femail"')
-
+        #
-        res = tdSql.getColNameList("select stddev(dataint),jsons1.jtag->'tag1' from jsons1 group by jsons1.jtag->'tag1' order by jtag->'tag1'")
+        # res = tdSql.getColNameList("select stddev(dataint),jsons1.jtag->'tag1' from jsons1 group by jsons1.jtag->'tag1' order by jtag->'tag1'")
-        cname_list = []
+        # cname_list = []
-        cname_list.append("stddev(dataint)")
+        # cname_list.append("stddev(dataint)")
-        cname_list.append("jsons1.jtag->'tag1'")
+        # cname_list.append("jsons1.jtag->'tag1'")
-        tdSql.checkColNameList(res, cname_list)
+        # tdSql.checkColNameList(res, cname_list)
 
         # test top/bottom with group by json tag
         # tdSql.query("select top(dataint,2),jtag->'tag1' from jsons1 group by jtag->'tag1' order by jtag->'tag1'")
@@ -470,8 +477,8 @@ class TDTestCase:
         # tdSql.checkData(10, 1, '"femail"')
 
         # test having
-        tdSql.query("select count(*),jtag->'tag1' from jsons1 group by jtag->'tag1' having count(*) > 1")
+        # tdSql.query("select count(*),jtag->'tag1' from jsons1 group by jtag->'tag1' having count(*) > 1")
-        tdSql.checkRows(3)
+        # tdSql.checkRows(3)
 
         # subquery with json tag
         tdSql.query("select * from (select jtag, dataint from jsons1) order by dataint")

tools/taos-tools

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