merge main

cmake/cmake.version

@@ -2,7 +2,7 @@
 IF (DEFINED VERNUMBER)
   SET(TD_VER_NUMBER ${VERNUMBER})
 ELSE ()
-  SET(TD_VER_NUMBER "3.0.3.1")
+  SET(TD_VER_NUMBER "3.0.3.2")
 ENDIF ()
 
 IF (DEFINED VERCOMPATIBLE)

cmake/taosadapter_CMakeLists.txt.in

@@ -2,7 +2,7 @@
 # taosadapter
 ExternalProject_Add(taosadapter
         GIT_REPOSITORY https://github.com/taosdata/taosadapter.git
-        GIT_TAG d8059ff
+        GIT_TAG cb1e89c
         SOURCE_DIR "${TD_SOURCE_DIR}/tools/taosadapter"
         BINARY_DIR ""
         #BUILD_IN_SOURCE TRUE

cmake/taostools_CMakeLists.txt.in

@@ -2,7 +2,7 @@
 # taos-tools
 ExternalProject_Add(taos-tools
         GIT_REPOSITORY https://github.com/taosdata/taos-tools.git
-        GIT_TAG 2864326
+        GIT_TAG 149ac34
         SOURCE_DIR "${TD_SOURCE_DIR}/tools/taos-tools"
         BINARY_DIR ""
         #BUILD_IN_SOURCE TRUE

docs/en/07-develop/07-tmq.mdx

@@ -293,7 +293,6 @@ You configure the following parameters when creating a consumer:
 |      `auto.offset.reset`       |  enum   | Initial offset for the consumer group                                     | Specify `earliest`, `latest`, or `none`(default) |
 |      `enable.auto.commit`      | boolean | Commit automatically                                             | Specify `true` or `false`.                   |
 |   `auto.commit.interval.ms`    | integer | Interval for automatic commits, in milliseconds                           |
-| `enable.heartbeat.background`  | boolean | Backend heartbeat; if enabled, the consumer does not go offline even if it has not polled for a long time |                                             |
 | `experimental.snapshot.enable` | boolean | Specify whether to consume messages from the WAL or from TSBS                    |                                             |
 |     `msg.with.table.name`      | boolean | Specify whether to deserialize table names from messages                                 |
 
@@ -368,7 +367,6 @@ conf := &tmq.ConfigMap{
  "td.connect.port":              "6030",
  "client.id":                    "test_tmq_c",
  "enable.auto.commit":           "false",
- "enable.heartbeat.background":  "true",
  "experimental.snapshot.enable": "true",
  "msg.with.table.name":          "true",
 }
@@ -418,7 +416,6 @@ Python programs use the following parameters:
 | `auto.commit.interval.ms` | string | Interval for automatic commits, in milliseconds | |
 | `auto.offset.reset` | string | Initial offset for the consumer group | Specify `earliest`, `latest`, or `none`(default) |
 | `experimental.snapshot.enable` | string | Specify whether to consume messages from the WAL or from TSDB | Specify `true` or `false` |
-| `enable.heartbeat.background` | string | Backend heartbeat; if enabled, the consumer does not go offline even if it has not polled for a long time | Specify `true` or `false` |
 
 </TabItem>
 

docs/examples/go/sub/main.go

@@ -35,7 +35,6 @@ func main() {
 		"td.connect.port":              "6030",
 		"client.id":                    "test_tmq_client",
 		"enable.auto.commit":           "false",
-		"enable.heartbeat.background":  "true",
 		"experimental.snapshot.enable": "true",
 		"msg.with.table.name":          "true",
 	})

docs/zh/07-develop/07-tmq.mdx

@@ -291,7 +291,6 @@ CREATE TOPIC topic_name AS DATABASE db_name;
 |      `auto.offset.reset`       |  enum   | 消费组订阅的初始位置                                     | 可选：`earliest`(default), `latest`, `none` |
 |      `enable.auto.commit`      | boolean | 是否启用消费位点自动提交                                 | 合法值：`true`, `false`。                   |
 |   `auto.commit.interval.ms`    | integer | 以毫秒为单位的消费记录自动提交消费位点时间间             | 默认 5000 m                                 |
-| `enable.heartbeat.background`  | boolean | 启用后台心跳，启用后即使长时间不 poll 消息也不会造成离线 | 默认开启                                    |
 | `experimental.snapshot.enable` | boolean | 是否允许从 TSDB 消费数据                                 | 实验功能，默认关闭                          |
 |     `msg.with.table.name`      | boolean | 是否允许从消息中解析表名, 不适用于列订阅（列订阅时可将 tbname 作为列写入 subquery 语句）               | |
 
@@ -366,7 +365,6 @@ conf := &tmq.ConfigMap{
  "td.connect.port":              "6030",
  "client.id":                    "test_tmq_c",
  "enable.auto.commit":           "false",
- "enable.heartbeat.background":  "true",
  "experimental.snapshot.enable": "true",
  "msg.with.table.name":          "true",
 }
@@ -418,7 +416,6 @@ consumer = Consumer({"group.id": "local", "td.connect.ip": "127.0.0.1"})
 | `auto.commit.interval.ms` | string | 以毫秒为单位的自动提交时间间隔 | 默认值：5000 ms |
 | `auto.offset.reset` | string | 消费组订阅的初始位置 | 可选：`earliest`(default), `latest`, `none` |
 | `experimental.snapshot.enable` | string | 是否允许从 TSDB 消费数据 | 合法值：`true`, `false` |
-| `enable.heartbeat.background` | string | 启用后台心跳，启用后即使长时间不 poll 消息也不会造成离线 | 合法值：`true`, `false` |
 
 </TabItem>
 

include/common/tcommon.h

@@ -185,7 +185,7 @@ typedef struct SBlockID {
 typedef struct SDataBlockInfo {
   STimeWindow window;
   int32_t     rowSize;
-  int32_t     rows;  // todo hide this attribute
+  int64_t     rows;  // todo hide this attribute
   uint32_t    capacity;
   SBlockID    id;
   int16_t     hasVarCol;
@@ -208,15 +208,12 @@ typedef struct SSDataBlock {
 } SSDataBlock;
 
 enum {
-  FETCH_TYPE__DATA = 1,
-  FETCH_TYPE__META,
-  FETCH_TYPE__SEP,
+  FETCH_TYPE__DATA = 0,
   FETCH_TYPE__NONE,
 };
 
 typedef struct {
   int8_t       fetchType;
-  STqOffsetVal offset;
   union {
     SSDataBlock data;
     void*       meta;

include/common/tdatablock.h

@@ -178,7 +178,7 @@ int32_t getJsonValueLen(const char* data);
 
 int32_t colDataSetVal(SColumnInfoData* pColumnInfoData, uint32_t rowIndex, const char* pData, bool isNull);
 int32_t colDataAppend(SColumnInfoData* pColumnInfoData, uint32_t rowIndex, const char* pData, bool isNull);
-int32_t colDataSetNItems(SColumnInfoData* pColumnInfoData, uint32_t rowIndex, const char* pData, uint32_t numOfRows);
+int32_t colDataSetNItems(SColumnInfoData* pColumnInfoData, uint32_t rowIndex, const char* pData, uint32_t numOfRows, bool trimValue);
 int32_t colDataMergeCol(SColumnInfoData* pColumnInfoData, int32_t numOfRow1, int32_t* capacity,
                         const SColumnInfoData* pSource, int32_t numOfRow2);
 int32_t colDataAssign(SColumnInfoData* pColumnInfoData, const SColumnInfoData* pSource, int32_t numOfRows,

include/common/tmsgcb.h

@@ -34,7 +34,7 @@ typedef enum {
   WRITE_QUEUE,
   APPLY_QUEUE,
   SYNC_QUEUE,
-  SYNC_CTRL_QUEUE,
+  SYNC_RD_QUEUE,
   STREAM_QUEUE,
   QUEUE_MAX,
 } EQueueType;

include/common/tmsgdef.h

@@ -259,7 +259,7 @@ enum {
 
   TD_NEW_MSG_SEG(TDMT_SYNC_MSG)
   TD_DEF_MSG_TYPE(TDMT_SYNC_TIMEOUT, "sync-timer", NULL, NULL)
-  TD_DEF_MSG_TYPE(TDMT_SYNC_PING, "sync-ping", NULL, NULL)             // no longer used
+  TD_DEF_MSG_TYPE(TDMT_SYNC_TIMEOUT_ELECTION, "sync-elect", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_SYNC_PING_REPLY, "sync-ping-reply", NULL, NULL) // no longer used
   TD_DEF_MSG_TYPE(TDMT_SYNC_CLIENT_REQUEST, "sync-client-request", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_SYNC_CLIENT_REQUEST_BATCH, "sync-client-request-batch", NULL, NULL)

include/common/ttokendef.h

@@ -127,24 +127,24 @@
 #define TK_NK_EQ                109
 #define TK_USING                110
 #define TK_TAGS                 111
-#define TK_COMMENT              112
-#define TK_BOOL                 113
-#define TK_TINYINT              114
-#define TK_SMALLINT             115
-#define TK_INT                  116
-#define TK_INTEGER              117
-#define TK_BIGINT               118
-#define TK_FLOAT                119
-#define TK_DOUBLE               120
-#define TK_BINARY               121
-#define TK_NCHAR                122
-#define TK_UNSIGNED             123
-#define TK_JSON                 124
-#define TK_VARCHAR              125
-#define TK_MEDIUMBLOB           126
-#define TK_BLOB                 127
-#define TK_VARBINARY            128
-#define TK_DECIMAL              129
+#define TK_BOOL                 112
+#define TK_TINYINT              113
+#define TK_SMALLINT             114
+#define TK_INT                  115
+#define TK_INTEGER              116
+#define TK_BIGINT               117
+#define TK_FLOAT                118
+#define TK_DOUBLE               119
+#define TK_BINARY               120
+#define TK_NCHAR                121
+#define TK_UNSIGNED             122
+#define TK_JSON                 123
+#define TK_VARCHAR              124
+#define TK_MEDIUMBLOB           125
+#define TK_BLOB                 126
+#define TK_VARBINARY            127
+#define TK_DECIMAL              128
+#define TK_COMMENT              129
 #define TK_MAX_DELAY            130
 #define TK_WATERMARK            131
 #define TK_ROLLUP               132

include/libs/executor/executor.h

@@ -78,7 +78,7 @@ qTaskInfo_t qCreateStreamExecTaskInfo(void* msg, SReadHandle* readers, int32_t v
  * @param SReadHandle
  * @return
  */
-qTaskInfo_t qCreateQueueExecTaskInfo(void* msg, SReadHandle* pReaderHandle, int32_t vgId, int32_t* numOfCols, SSchemaWrapper** pSchema);
+qTaskInfo_t qCreateQueueExecTaskInfo(void* msg, SReadHandle* pReaderHandle, int32_t vgId, int32_t* numOfCols, uint64_t id);
 
 /**
  * set the task Id, usually used by message queue process
@@ -89,6 +89,7 @@ qTaskInfo_t qCreateQueueExecTaskInfo(void* msg, SReadHandle* pReaderHandle, int3
 void qSetTaskId(qTaskInfo_t tinfo, uint64_t taskId, uint64_t queryId);
 
 int32_t qSetStreamOpOpen(qTaskInfo_t tinfo);
+
 /**
  * Set multiple input data blocks for the stream scan.
  * @param tinfo
@@ -197,11 +198,11 @@ int32_t qStreamPrepareScan(qTaskInfo_t tinfo, STqOffsetVal* pOffset, int8_t subT
 //
 int32_t qStreamSetScanMemData(qTaskInfo_t tinfo, SPackedData submit);
 
-int32_t qStreamExtractOffset(qTaskInfo_t tinfo, STqOffsetVal* pOffset);
+void qStreamSetOpen(qTaskInfo_t tinfo);
 
-SMqMetaRsp* qStreamExtractMetaMsg(qTaskInfo_t tinfo);
+void qStreamExtractOffset(qTaskInfo_t tinfo, STqOffsetVal* pOffset);
 
-int64_t qStreamExtractPrepareUid(qTaskInfo_t tinfo);
+SMqMetaRsp* qStreamExtractMetaMsg(qTaskInfo_t tinfo);
 
 const SSchemaWrapper* qExtractSchemaFromTask(qTaskInfo_t tinfo);
 

include/libs/function/function.h

@@ -112,7 +112,7 @@ typedef struct SResultDataInfo {
 typedef struct SInputColumnInfoData {
   int32_t           totalRows;        // total rows in current columnar data
   int32_t           startRowIndex;    // handle started row index
-  int32_t           numOfRows;        // the number of rows needs to be handled
+  int64_t           numOfRows;        // the number of rows needs to be handled
   int32_t           numOfInputCols;   // PTS is not included
   bool              colDataSMAIsSet;  // if agg is set or not
   SColumnInfoData  *pPTS;             // primary timestamp column

include/libs/nodes/querynodes.h

@@ -298,6 +298,7 @@ typedef struct SSelectStmt {
   bool        hasUniqueFunc;
   bool        hasTailFunc;
   bool        hasInterpFunc;
+  bool        hasInterpPseudoColFunc;
   bool        hasLastRowFunc;
   bool        hasLastFunc;
   bool        hasTimeLineFunc;

include/libs/qcom/query.h

@@ -33,6 +33,7 @@ typedef enum {
   JOB_TASK_STATUS_INIT,
   JOB_TASK_STATUS_EXEC,
   JOB_TASK_STATUS_PART_SUCC,
+  JOB_TASK_STATUS_FETCH,
   JOB_TASK_STATUS_SUCC,
   JOB_TASK_STATUS_FAIL,
   JOB_TASK_STATUS_DROP,

include/libs/stream/streamState.h

@@ -23,20 +23,19 @@ extern "C" {
 #ifndef _STREAM_STATE_H_
 #define _STREAM_STATE_H_
 
-typedef struct SStreamTask SStreamTask;
-
 typedef bool (*state_key_cmpr_fn)(void* pKey1, void* pKey2);
 
 typedef struct STdbState {
-  SStreamTask* pOwner;
-  TDB*         db;
-  TTB*         pStateDb;
-  TTB*         pFuncStateDb;
-  TTB*         pFillStateDb;  // todo refactor
-  TTB*         pSessionStateDb;
-  TTB*         pParNameDb;
-  TTB*         pParTagDb;
-  TXN*         txn;
+  struct SStreamTask* pOwner;
+
+  TDB* db;
+  TTB* pStateDb;
+  TTB* pFuncStateDb;
+  TTB* pFillStateDb;  // todo refactor
+  TTB* pSessionStateDb;
+  TTB* pParNameDb;
+  TTB* pParTagDb;
+  TXN* txn;
 } STdbState;
 
 // incremental state storage
@@ -45,7 +44,7 @@ typedef struct {
   int32_t    number;
 } SStreamState;
 
-SStreamState* streamStateOpen(char* path, SStreamTask* pTask, bool specPath, int32_t szPage, int32_t pages);
+SStreamState* streamStateOpen(char* path, struct SStreamTask* pTask, bool specPath, int32_t szPage, int32_t pages);
 void          streamStateClose(SStreamState* pState);
 int32_t       streamStateBegin(SStreamState* pState);
 int32_t       streamStateCommit(SStreamState* pState);

include/libs/stream/tstream.h

@@ -223,27 +223,12 @@ static FORCE_INLINE void* streamQueueCurItem(SStreamQueue* queue) {
   return queue->qItem;
 }
 
-static FORCE_INLINE void* streamQueueNextItem(SStreamQueue* queue) {
-  int8_t dequeueFlag = atomic_exchange_8(&queue->status, STREAM_QUEUE__PROCESSING);
-  if (dequeueFlag == STREAM_QUEUE__FAILED) {
-    ASSERT(queue->qItem != NULL);
-    return streamQueueCurItem(queue);
-  } else {
-    queue->qItem = NULL;
-    taosGetQitem(queue->qall, &queue->qItem);
-    if (queue->qItem == NULL) {
-      taosReadAllQitems(queue->queue, queue->qall);
-      taosGetQitem(queue->qall, &queue->qItem);
-    }
-    return streamQueueCurItem(queue);
-  }
-}
+void* streamQueueNextItem(SStreamQueue* queue);
 
 SStreamDataSubmit2* streamDataSubmitNew(SPackedData submit);
+void streamDataSubmitDestroy(SStreamDataSubmit2* pDataSubmit);
 
-void streamDataSubmitRefDec(SStreamDataSubmit2* pDataSubmit);
-
-SStreamDataSubmit2* streamSubmitRefClone(SStreamDataSubmit2* pSubmit);
+SStreamDataSubmit2* streamSubmitBlockClone(SStreamDataSubmit2* pSubmit);
 
 typedef struct {
   char* qmsg;
@@ -295,7 +280,7 @@ typedef struct {
   SEpSet  epSet;
 } SStreamChildEpInfo;
 
-typedef struct SStreamTask {
+struct SStreamTask {
   int64_t streamId;
   int32_t taskId;
   int32_t totalLevel;
@@ -362,8 +347,7 @@ typedef struct SStreamTask {
 
   int64_t checkpointingId;
   int32_t checkpointAlignCnt;
-
-} SStreamTask;
+};
 
 int32_t tEncodeStreamEpInfo(SEncoder* pEncoder, const SStreamChildEpInfo* pInfo);
 int32_t tDecodeStreamEpInfo(SDecoder* pDecoder, SStreamChildEpInfo* pInfo);
@@ -372,43 +356,7 @@ SStreamTask* tNewSStreamTask(int64_t streamId);
 int32_t      tEncodeSStreamTask(SEncoder* pEncoder, const SStreamTask* pTask);
 int32_t      tDecodeSStreamTask(SDecoder* pDecoder, SStreamTask* pTask);
 void         tFreeSStreamTask(SStreamTask* pTask);
-
-static FORCE_INLINE int32_t streamTaskInput(SStreamTask* pTask, SStreamQueueItem* pItem) {
-  int8_t type = pItem->type;
-  if (type == STREAM_INPUT__DATA_SUBMIT) {
-    SStreamDataSubmit2* pSubmitClone = streamSubmitRefClone((SStreamDataSubmit2*)pItem);
-    if (pSubmitClone == NULL) {
-      qDebug("task %d %p submit enqueue failed since out of memory", pTask->taskId, pTask);
-      terrno = TSDB_CODE_OUT_OF_MEMORY;
-      atomic_store_8(&pTask->inputStatus, TASK_INPUT_STATUS__FAILED);
-      return -1;
-    }
-    qDebug("task %d %p submit enqueue %p %p %p %d %" PRId64, pTask->taskId, pTask, pItem, pSubmitClone,
-           pSubmitClone->submit.msgStr, pSubmitClone->submit.msgLen, pSubmitClone->submit.ver);
-    taosWriteQitem(pTask->inputQueue->queue, pSubmitClone);
-    // qStreamInput(pTask->exec.executor, pSubmitClone);
-  } else if (type == STREAM_INPUT__DATA_BLOCK || type == STREAM_INPUT__DATA_RETRIEVE ||
-             type == STREAM_INPUT__REF_DATA_BLOCK) {
-    taosWriteQitem(pTask->inputQueue->queue, pItem);
-    // qStreamInput(pTask->exec.executor, pItem);
-  } else if (type == STREAM_INPUT__CHECKPOINT) {
-    taosWriteQitem(pTask->inputQueue->queue, pItem);
-    // qStreamInput(pTask->exec.executor, pItem);
-  } else if (type == STREAM_INPUT__GET_RES) {
-    taosWriteQitem(pTask->inputQueue->queue, pItem);
-    // qStreamInput(pTask->exec.executor, pItem);
-  }
-
-  if (type != STREAM_INPUT__GET_RES && type != STREAM_INPUT__CHECKPOINT && pTask->triggerParam != 0) {
-    atomic_val_compare_exchange_8(&pTask->triggerStatus, TASK_TRIGGER_STATUS__INACTIVE, TASK_TRIGGER_STATUS__ACTIVE);
-  }
-
-#if 0
-  // TODO: back pressure
-  atomic_store_8(&pTask->inputStatus, TASK_INPUT_STATUS__NORMAL);
-#endif
-  return 0;
-}
+int32_t      tAppendDataForStream(SStreamTask* pTask, SStreamQueueItem* pItem);
 
 static FORCE_INLINE void streamTaskInputFail(SStreamTask* pTask) {
   atomic_store_8(&pTask->inputStatus, TASK_INPUT_STATUS__FAILED);
@@ -587,7 +535,7 @@ int32_t streamProcessDispatchReq(SStreamTask* pTask, SStreamDispatchReq* pReq, S
 int32_t streamProcessDispatchRsp(SStreamTask* pTask, SStreamDispatchRsp* pRsp, int32_t code);
 
 int32_t streamProcessRetrieveReq(SStreamTask* pTask, SStreamRetrieveReq* pReq, SRpcMsg* pMsg);
-int32_t streamProcessRetrieveRsp(SStreamTask* pTask, SStreamRetrieveRsp* pRsp);
+// int32_t streamProcessRetrieveRsp(SStreamTask* pTask, SStreamRetrieveRsp* pRsp);
 
 int32_t streamTryExec(SStreamTask* pTask);
 int32_t streamSchedExec(SStreamTask* pTask);

include/libs/wal/wal.h

@@ -146,8 +146,8 @@ typedef struct {
   int64_t        curFileFirstVer;
   int64_t        curVersion;
   int64_t        capacity;
-  int8_t         curInvalid;
-  int8_t         curStopped;
+//  int8_t         curInvalid;
+//  int8_t         curStopped;
   TdThreadMutex  mutex;
   SWalFilterCond cond;
   // TODO remove it

include/os/osDir.h

@@ -31,21 +31,49 @@
 extern "C" {
 #endif
 
+#if defined(CUS_NAME) || defined(CUS_PROMPT) || defined(CUS_EMAIL)
+#include "cus_name.h"
+#endif
+
 #ifdef WINDOWS
+
 #define TD_TMP_DIR_PATH  "C:\\Windows\\Temp\\"
+#ifdef CUS_NAME
+#define TD_CFG_DIR_PATH  "C:\\"CUS_NAME"\\cfg\\"
+#define TD_DATA_DIR_PATH "C:\\"CUS_NAME"\\data\\"
+#define TD_LOG_DIR_PATH  "C:\\"CUS_NAME"\\log\\"
+#else
 #define TD_CFG_DIR_PATH  "C:\\TDengine\\cfg\\"
 #define TD_DATA_DIR_PATH "C:\\TDengine\\data\\"
 #define TD_LOG_DIR_PATH  "C:\\TDengine\\log\\"
+#endif  // CUS_NAME
+
 #elif defined(_TD_DARWIN_64)
+
+#ifdef CUS_PROMPT
+#define TD_TMP_DIR_PATH  "/tmp/"CUS_PROMPT"d/"
+#define TD_CFG_DIR_PATH  "/etc/"CUS_PROMPT"/"
+#define TD_DATA_DIR_PATH "/var/lib/"CUS_PROMPT"/"
+#define TD_LOG_DIR_PATH  "/var/log/"CUS_PROMPT"/"
+#else
 #define TD_TMP_DIR_PATH  "/tmp/taosd/"
 #define TD_CFG_DIR_PATH  "/etc/taos/"
 #define TD_DATA_DIR_PATH "/var/lib/taos/"
 #define TD_LOG_DIR_PATH  "/var/log/taos/"
+#endif  // CUS_PROMPT
+
 #else
+
 #define TD_TMP_DIR_PATH  "/tmp/"
+#ifdef CUS_PROMPT
+#define TD_CFG_DIR_PATH  "/etc/"CUS_PROMPT"/"
+#define TD_DATA_DIR_PATH "/var/lib/"CUS_PROMPT"/"
+#define TD_LOG_DIR_PATH  "/var/log/"CUS_PROMPT"/"
+#else
 #define TD_CFG_DIR_PATH  "/etc/taos/"
 #define TD_DATA_DIR_PATH "/var/lib/taos/"
 #define TD_LOG_DIR_PATH  "/var/log/taos/"
+#endif   // CUS_PROMPT
 #endif
 
 typedef struct TdDir      *TdDirPtr;

include/util/cus_name.h

+/*
+ * Copyright (c) 2019 TAOS Data, Inc. <jhtao@taosdata.com>
+ *
+ * This program is free software: you can use, redistribute, and/or modify
+ * it under the terms of the GNU Affero General Public License, version 3
+ * or later ("AGPL"), as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.
+ *
+ * You should have received a copy of the GNU Affero General Public License
+ * along with this program. If not, see <http://www.gnu.org/licenses/>.
+ */
+
+#ifndef _CUS_NAME_H_
+#define _CUS_NAME_H_
+
+#ifndef CUS_NAME
+#define CUS_NAME    "TDengine"
+#endif
+
+#ifndef CUS_PROMPT
+#define CUS_PROMPT  "taos"
+#endif
+
+#ifndef CUS_EMAIL
+#define CUS_EMAIL   "<support@taosdata.com>"
+#endif
+
+#endif  // _CUS_NAME_H_

include/util/taoserror.h

@@ -759,6 +759,7 @@ int32_t* taosGetErrno();
 #define TSDB_CODE_TMQ_INVALID_MSG                TAOS_DEF_ERROR_CODE(0, 0x4000)
 #define TSDB_CODE_TMQ_CONSUMER_MISMATCH          TAOS_DEF_ERROR_CODE(0, 0x4001)
 #define TSDB_CODE_TMQ_CONSUMER_CLOSED            TAOS_DEF_ERROR_CODE(0, 0x4002)
+#define TSDB_CODE_TMQ_CONSUMER_ERROR             TAOS_DEF_ERROR_CODE(0, 0x4003)
 
 // stream
 #define TSDB_CODE_STREAM_TASK_NOT_EXIST          TAOS_DEF_ERROR_CODE(0, 0x4100)

include/util/tqueue.h

@@ -61,7 +61,7 @@ typedef void (*FItems)(SQueueInfo *pInfo, STaosQall *qall, int32_t numOfItems);
 
 typedef struct STaosQnode STaosQnode;
 
-typedef struct STaosQnode {
+struct STaosQnode {
   STaosQnode *next;
   STaosQueue *queue;
   int64_t     timestamp;
@@ -70,9 +70,9 @@ typedef struct STaosQnode {
   int8_t      itype;
   int8_t      reserved[3];
   char        item[];
-} STaosQnode;
+};
 
-typedef struct STaosQueue {
+struct STaosQueue {
   STaosQnode   *head;
   STaosQnode   *tail;
   STaosQueue   *next;     // for queue set
@@ -84,22 +84,22 @@ typedef struct STaosQueue {
   int64_t       memOfItems;
   int32_t       numOfItems;
   int64_t       threadId;
-} STaosQueue;
+};
 
-typedef struct STaosQset {
+struct STaosQset {
   STaosQueue   *head;
   STaosQueue   *current;
   TdThreadMutex mutex;
   tsem_t        sem;
   int32_t       numOfQueues;
   int32_t       numOfItems;
-} STaosQset;
+};
 
-typedef struct STaosQall {
+struct STaosQall {
   STaosQnode *current;
   STaosQnode *start;
   int32_t     numOfItems;
-} STaosQall;
+};
 
 STaosQueue *taosOpenQueue();
 void        taosCloseQueue(STaosQueue *queue);

include/util/tsimplehash.h

@@ -48,6 +48,13 @@ SSHashObj *tSimpleHashInit(size_t capacity, _hash_fn_t fn);
  */
 int32_t tSimpleHashGetSize(const SSHashObj *pHashObj);
 
+/**
+ * set the free function pointer
+ * @param pHashObj
+ * @param freeFp
+ */
+void tSimpleHashSetFreeFp(SSHashObj* pHashObj, _hash_free_fn_t freeFp);
+
 int32_t tSimpleHashPrint(const SSHashObj *pHashObj);
 
 /**

packaging/cfg/taos.cfg

 ########################################################
 #                                                      #
 #                  Configuration                       #
-#   Any questions, please email support@taosdata.com   #
 #                                                      #
 ########################################################
 
@@ -13,7 +12,7 @@
 
 ############### 1. Cluster End point ############################
 
-# The end point of the first dnode in the cluster to be connected to when this dnode or a CLI `taos` is started
+# The end point of the first dnode in the cluster to be connected to when this dnode or the CLI utility is started
 # firstEp                   hostname:6030
 
 # The end point of the second dnode to be connected to if the firstEp is not available
@@ -25,7 +24,7 @@
 # The FQDN of the host on which this dnode will be started. It can be IP address
 # fqdn                      hostname
 
-# The port for external access after this dnode is started     
+# The port for external access after this dnode is started
 # serverPort                6030
 
 # The maximum number of connections a dnode can accept
@@ -96,7 +95,7 @@
 # if free disk space is less than this value, this dnode will fail to start
 # minimalDataDirGB          2.0
 
-# enable/disable system monitor 
+# enable/disable system monitor
 # monitor                   1
 
 # The following parameter is used to limit the maximum number of lines in log files.
@@ -114,8 +113,8 @@
 
 # The following parameters are used for debug purpose only by this dnode.
 # debugFlag is a 8 bits mask: FILE-SCREEN-UNUSED-HeartBeat-DUMP-TRACE_WARN-ERROR
-# Available debug levels are: 
-# 131: output warning and error 
+# Available debug levels are:
+# 131: output warning and error
 # 135: output debug, warning and error
 # 143: output trace, debug, warning and error to log
 # 199: output debug, warning and error to both screen and file
@@ -130,7 +129,7 @@
 # debug flag for util
 # uDebugFlag                131
 
-# debug flag for rpc 
+# debug flag for rpc
 # rpcDebugFlag              131
 
 # debug flag for jni
@@ -139,7 +138,7 @@
 # debug flag for query
 # qDebugFlag                131
 
-# debug flag for taosc driver 
+# debug flag for client driver
 # cDebugFlag                131
 
 # debug flag for dnode messages

packaging/cfg/taosd.service

 [Unit]
-Description=TDengine server service
+Description=server service
 After=network-online.target
 Wants=network-online.target
 

packaging/rpm/tdengine.spec

@@ -90,45 +90,33 @@ cp %{_compiledir}/../include/libs/function/taosudf.h       %{buildroot}%{homepat
 cp -r %{_compiledir}/../examples/*                  %{buildroot}%{homepath}/examples
 
 if [ -f %{_compiledir}/build/bin/jemalloc-config ]; then
-    mkdir -p %{buildroot}%{userlocalpath}/bin
-    mkdir -p %{buildroot}%{userlocalpath}/lib
-    mkdir -p %{buildroot}%{userlocalpath}/lib/pkgconfig
-    mkdir -p %{buildroot}%{userlocalpath}/include
-    mkdir -p %{buildroot}%{userlocalpath}/include/jemalloc
-    mkdir -p %{buildroot}%{userlocalpath}/share
-    mkdir -p %{buildroot}%{userlocalpath}/share/doc
-    mkdir -p %{buildroot}%{userlocalpath}/share/doc/jemalloc
-    mkdir -p %{buildroot}%{userlocalpath}/share/man
-    mkdir -p %{buildroot}%{userlocalpath}/share/man/man3
-
-    cp %{_compiledir}/build/bin/jemalloc-config %{buildroot}%{userlocalpath}/bin/
+    mkdir -p %{buildroot}%{homepath}/jemalloc/ ||:
+    mkdir -p %{buildroot}%{homepath}/jemalloc/include/jemalloc/ ||:
+    mkdir -p %{buildroot}%{homepath}/jemalloc/lib/ ||:
+    mkdir -p %{buildroot}%{homepath}/jemalloc/lib/pkgconfig ||:
+
+    cp %{_compiledir}/build/bin/jemalloc-config %{buildroot}%{homepath}/jemalloc/bin
     if [ -f %{_compiledir}/build/bin/jemalloc.sh ]; then
-        cp %{_compiledir}/build/bin/jemalloc.sh %{buildroot}%{userlocalpath}/bin/
+        cp %{_compiledir}/build/bin/jemalloc.sh %{buildroot}%{homepath}/jemalloc/bin
     fi
     if [ -f %{_compiledir}/build/bin/jeprof ]; then
-        cp %{_compiledir}/build/bin/jeprof %{buildroot}%{userlocalpath}/bin/
+        cp %{_compiledir}/build/bin/jeprof %{buildroot}%{homepath}/jemalloc/bin
     fi
     if [ -f %{_compiledir}/build/include/jemalloc/jemalloc.h ]; then
-        cp %{_compiledir}/build/include/jemalloc/jemalloc.h %{buildroot}%{userlocalpath}/include/jemalloc/
+        cp %{_compiledir}/build/include/jemalloc/jemalloc.h %{buildroot}%{homepath}/jemalloc/include/jemalloc/
     fi
     if [ -f %{_compiledir}/build/lib/libjemalloc.so.2 ]; then
-        cp %{_compiledir}/build/lib/libjemalloc.so.2 %{buildroot}%{userlocalpath}/lib/
-        ln -sf libjemalloc.so.2 %{buildroot}%{userlocalpath}/lib/libjemalloc.so
+        cp %{_compiledir}/build/lib/libjemalloc.so.2 %{buildroot}%{homepath}/jemalloc/lib
+        ln -sf libjemalloc.so.2 %{buildroot}%{homepath}/jemalloc/lib/libjemalloc.so
     fi
     if [ -f %{_compiledir}/build/lib/libjemalloc.a ]; then
-        cp %{_compiledir}/build/lib/libjemalloc.a %{buildroot}%{userlocalpath}/lib/
+        cp %{_compiledir}/build/lib/libjemalloc.a %{buildroot}%{homepath}/jemalloc/lib
     fi
     if [ -f %{_compiledir}/build/lib/libjemalloc_pic.a ]; then
-        cp %{_compiledir}/build/lib/libjemalloc_pic.a %{buildroot}%{userlocalpath}/lib/
+        cp %{_compiledir}/build/lib/libjemalloc_pic.a %{buildroot}%{homepath}/jemalloc/lib
     fi
     if [ -f %{_compiledir}/build/lib/pkgconfig/jemalloc.pc ]; then
-        cp %{_compiledir}/build/lib/pkgconfig/jemalloc.pc %{buildroot}%{userlocalpath}/lib/pkgconfig/
-    fi
-    if [ -f %{_compiledir}/build/share/doc/jemalloc/jemalloc.html ]; then
-        cp %{_compiledir}/build/share/doc/jemalloc/jemalloc.html %{buildroot}%{userlocalpath}/share/doc/jemalloc/
-    fi
-    if [ -f %{_compiledir}/build/share/man/man3/jemalloc.3 ]; then
-        cp %{_compiledir}/build/share/man/man3/jemalloc.3 %{buildroot}%{userlocalpath}/share/man/man3/
+        cp %{_compiledir}/build/lib/pkgconfig/jemalloc.pc %{buildroot}%{homepath}/jemalloc/lib/pkgconfig
     fi
 fi
 

packaging/tools/makeclient.sh

@@ -23,9 +23,12 @@ clientName2="${12}"
 
 productName="TDengine"
 clientName="taos"
+benchmarkName="taosBenchmark"
 configFile="taos.cfg"
 tarName="package.tar.gz"
 
+benchmarkName2="${clientName2}Benchmark"
+
 if [ "$osType" != "Darwin" ]; then
   script_dir="$(dirname $(readlink -f $0))"
   top_dir="$(readlink -f ${script_dir}/../..)"
@@ -53,11 +56,12 @@ fi
 
 # Directories and files.
 
-if [ "$verMode" == "cluster" ]; then
-  sed -i 's/verMode=edge/verMode=cluster/g' ${script_dir}/remove_client.sh
-  sed -i "s/clientName2=\"taos\"/clientName2=\"${clientName2}\"/g" ${script_dir}/remove_client.sh
-  sed -i "s/productName2=\"TDengine\"/productName2=\"${productName2}\"/g" ${script_dir}/remove_client.sh
-fi
+#if [ "$verMode" == "cluster" ]; then
+#  sed -i 's/verMode=edge/verMode=cluster/g' ${script_dir}/remove_client.sh
+#  sed -i "s/clientName2=\"taos\"/clientName2=\"${clientName2}\"/g" ${script_dir}/remove_client.sh
+#  sed -i "s/configFile2=\"taos\"/configFile2=\"${clientName2}\"/g" ${script_dir}/remove_client.sh
+#  sed -i "s/productName2=\"TDengine\"/productName2=\"${productName2}\"/g" ${script_dir}/remove_client.sh
+#fi
 
 if [ "$osType" != "Darwin" ]; then
   if [ "$pagMode" == "lite" ]; then
@@ -66,6 +70,7 @@ if [ "$osType" != "Darwin" ]; then
         ${script_dir}/remove_client.sh"
   else
     bin_files="${build_dir}/bin/${clientName} \
+        ${build_dir}/bin/${benchmarkName} \
         ${script_dir}/remove_client.sh \
         ${script_dir}/set_core.sh \
         ${script_dir}/get_client.sh"
@@ -153,6 +158,7 @@ if [ "$verMode" == "cluster" ]; then
   sed -i 's/verMode=edge/verMode=cluster/g' install_client_temp.sh
   sed -i "s/serverName2=\"taosd\"/serverName2=\"${serverName2}\"/g" install_client_temp.sh
   sed -i "s/clientName2=\"taos\"/clientName2=\"${clientName2}\"/g" install_client_temp.sh
+  sed -i "s/configFile2=\"taos.cfg\"/configFile2=\"${clientName2}.cfg\"/g" install_client_temp.sh
   sed -i "s/productName2=\"TDengine\"/productName2=\"${productName2}\"/g" install_client_temp.sh
   sed -i "s/emailName2=\"taosdata.com\"/emailName2=\"${cusEmail2}\"/g" install_client_temp.sh
 

packaging/tools/makepkg.sh

@@ -51,9 +51,9 @@ fi
 
 if [ -d ${top_dir}/tools/taos-tools/packaging/deb ]; then
   cd ${top_dir}/tools/taos-tools/packaging/deb
-  [ -z "$taos_tools_ver" ] && taos_tools_ver="0.1.0"
 
   taostools_ver=$(git for-each-ref --sort=taggerdate --format '%(tag)' refs/tags|grep -v taos | tail -1)
+  [ -z "$taostools_ver" ] && taostools_ver="0.1.0"
   taostools_install_dir="${release_dir}/${clientName2}Tools-${taostools_ver}"
 
   cd ${curr_dir}
@@ -96,7 +96,7 @@ else
       ${taostools_bin_files} \
       ${taosx_bin} \
       ${explorer_bin_files} \
-      ${build_dir}/bin/taosadapter \
+      ${build_dir}/bin/${clientName}adapter \
       ${build_dir}/bin/udfd \
       ${script_dir}/remove.sh \
       ${script_dir}/set_core.sh \
@@ -135,12 +135,12 @@ mkdir -p ${install_dir}/inc && cp ${header_files} ${install_dir}/inc
 
 mkdir -p ${install_dir}/cfg && cp ${cfg_dir}/${configFile} ${install_dir}/cfg/${configFile}
 
-if [ -f "${compile_dir}/test/cfg/taosadapter.toml" ]; then
-  cp ${compile_dir}/test/cfg/taosadapter.toml ${install_dir}/cfg || :
+if [ -f "${compile_dir}/test/cfg/${clientName}adapter.toml" ]; then
+  cp ${compile_dir}/test/cfg/${clientName}adapter.toml ${install_dir}/cfg || :
 fi
 
-if [ -f "${compile_dir}/test/cfg/taosadapter.service" ]; then
-  cp ${compile_dir}/test/cfg/taosadapter.service ${install_dir}/cfg || :
+if [ -f "${compile_dir}/test/cfg/${clientName}adapter.service" ]; then
+  cp ${compile_dir}/test/cfg/${clientName}adapter.service ${install_dir}/cfg || :
 fi
 
 if [ -f "${cfg_dir}/${serverName}.service" ]; then
@@ -150,18 +150,19 @@ fi
 mkdir -p ${install_dir}/bin && cp ${bin_files} ${install_dir}/bin && chmod a+x ${install_dir}/bin/* || :
 mkdir -p ${install_dir}/init.d && cp ${init_file_deb} ${install_dir}/init.d/${serverName}.deb
 mkdir -p ${install_dir}/init.d && cp ${init_file_rpm} ${install_dir}/init.d/${serverName}.rpm
+mkdir -p ${install_dir}/share && cp -rf ${build_dir}/share/{etc,srv} ${install_dir}/share
 
 if [ $adapterName != "taosadapter" ]; then
-  mv ${install_dir}/cfg/taosadapter.toml ${install_dir}/cfg/$adapterName.toml
+  mv ${install_dir}/cfg/${clientName2}adapter.toml ${install_dir}/cfg/$adapterName.toml
   sed -i "s/path = \"\/var\/log\/taos\"/path = \"\/var\/log\/${productName}\"/g" ${install_dir}/cfg/$adapterName.toml
   sed -i "s/password = \"taosdata\"/password = \"${defaultPasswd}\"/g" ${install_dir}/cfg/$adapterName.toml
 
-  mv ${install_dir}/cfg/taosadapter.service ${install_dir}/cfg/$adapterName.service
+  mv ${install_dir}/cfg/${clientName2}adapter.service ${install_dir}/cfg/$adapterName.service
   sed -i "s/TDengine/${productName}/g" ${install_dir}/cfg/$adapterName.service
   sed -i "s/taosAdapter/${adapterName}/g" ${install_dir}/cfg/$adapterName.service
   sed -i "s/taosadapter/${adapterName}/g" ${install_dir}/cfg/$adapterName.service
 
-  mv ${install_dir}/bin/taosadapter ${install_dir}/bin/${adapterName}
+  mv ${install_dir}/bin/${clientName2}adapter ${install_dir}/bin/${adapterName}
   mv ${install_dir}/bin/taosd-dump-cfg.gdb ${install_dir}/bin/${serverName}-dump-cfg.gdb
 fi
 
@@ -233,8 +234,10 @@ if [ "$verMode" == "cluster" ]; then
   sed 's/verMode=edge/verMode=cluster/g' ${install_dir}/bin/remove.sh >>remove_temp.sh
   sed -i "s/serverName2=\"taosd\"/serverName2=\"${serverName2}\"/g" remove_temp.sh
   sed -i "s/clientName2=\"taos\"/clientName2=\"${clientName2}\"/g" remove_temp.sh
+  sed -i "s/configFile2=\"taos.cfg\"/configFile2=\"${clientName2}.cfg\"/g" remove_temp.sh
   sed -i "s/productName2=\"TDengine\"/productName2=\"${productName2}\"/g" remove_temp.sh
-  sed -i "s/emailName2=\"taosdata.com\"/emailName2=\"${cusEmail2}\"/g" remove_temp.sh
+  cusDomain=`echo "${cusEmail2}" | sed 's/^[^@]*@//'`
+  sed -i "s/emailName2=\"taosdata.com\"/emailName2=\"${cusDomain}\"/g" remove_temp.sh
   mv remove_temp.sh ${install_dir}/bin/remove.sh
 fi
 if [ "$verMode" == "cloud" ]; then
@@ -262,8 +265,10 @@ if [ "$verMode" == "cluster" ]; then
   sed -i 's/verMode=edge/verMode=cluster/g' install_temp.sh
   sed -i "s/serverName2=\"taosd\"/serverName2=\"${serverName2}\"/g" install_temp.sh
   sed -i "s/clientName2=\"taos\"/clientName2=\"${clientName2}\"/g" install_temp.sh
+  sed -i "s/configFile2=\"taos.cfg\"/configFile2=\"${clientName2}.cfg\"/g" install_temp.sh
   sed -i "s/productName2=\"TDengine\"/productName2=\"${productName2}\"/g" install_temp.sh
-  sed -i "s/emailName2=\"taosdata.com\"/emailName2=\"${cusEmail2}\"/g" install_temp.sh
+  cusDomain=`echo "${cusEmail2}" | sed 's/^[^@]*@//'`
+  sed -i "s/emailName2=\"taosdata.com\"/emailName2=\"${cusDomain}\"/g" install_temp.sh
   mv install_temp.sh ${install_dir}/install.sh
 fi
 if [ "$verMode" == "cloud" ]; then

packaging/tools/post.sh

@@ -145,6 +145,59 @@ function install_include() {
     log_print "install include success"
 }
 
+function install_jemalloc() {
+  jemalloc_dir=${script_dir}/../jemalloc
+
+  if [ -d ${jemalloc_dir} ]; then
+    ${csudo}/usr/bin/install -c -d /usr/local/bin
+
+    if [ -f ${jemalloc_dir}/bin/jemalloc-config ]; then
+      ${csudo}/usr/bin/install -c -m 755 ${jemalloc_dir}/bin/jemalloc-config /usr/local/bin
+    fi
+    if [ -f ${jemalloc_dir}/bin/jemalloc.sh ]; then
+      ${csudo}/usr/bin/install -c -m 755 ${jemalloc_dir}/bin/jemalloc.sh /usr/local/bin
+    fi
+    if [ -f ${jemalloc_dir}/bin/jeprof ]; then
+      ${csudo}/usr/bin/install -c -m 755 ${jemalloc_dir}/bin/jeprof /usr/local/bin
+    fi
+    if [ -f ${jemalloc_dir}/include/jemalloc/jemalloc.h ]; then
+      ${csudo}/usr/bin/install -c -d /usr/local/include/jemalloc
+      ${csudo}/usr/bin/install -c -m 644 ${jemalloc_dir}/include/jemalloc/jemalloc.h /usr/local/include/jemalloc
+    fi
+    if [ -f ${jemalloc_dir}/lib/libjemalloc.so.2 ]; then
+      ${csudo}/usr/bin/install -c -d /usr/local/lib
+      ${csudo}/usr/bin/install -c -m 755 ${jemalloc_dir}/lib/libjemalloc.so.2 /usr/local/lib
+      ${csudo}ln -sf libjemalloc.so.2 /usr/local/lib/libjemalloc.so
+      ${csudo}/usr/bin/install -c -d /usr/local/lib
+      if [ -f ${jemalloc_dir}/lib/libjemalloc.a ]; then
+        ${csudo}/usr/bin/install -c -m 755 ${jemalloc_dir}/lib/libjemalloc.a /usr/local/lib
+      fi
+      if [ -f ${jemalloc_dir}/lib/libjemalloc_pic.a ]; then
+        ${csudo}/usr/bin/install -c -m 755 ${jemalloc_dir}/lib/libjemalloc_pic.a /usr/local/lib
+      fi
+      if [ -f ${jemalloc_dir}/lib/libjemalloc_pic.a ]; then
+        ${csudo}/usr/bin/install -c -d /usr/local/lib/pkgconfig
+        ${csudo}/usr/bin/install -c -m 644 ${jemalloc_dir}/lib/pkgconfig/jemalloc.pc /usr/local/lib/pkgconfig
+      fi
+    fi
+    if [ -f ${jemalloc_dir}/share/doc/jemalloc/jemalloc.html ]; then
+      ${csudo}/usr/bin/install -c -d /usr/local/share/doc/jemalloc
+      ${csudo}/usr/bin/install -c -m 644 ${jemalloc_dir}/share/doc/jemalloc/jemalloc.html /usr/local/share/doc/jemalloc
+    fi
+    if [ -f ${jemalloc_dir}/share/man/man3/jemalloc.3 ]; then
+      ${csudo}/usr/bin/install -c -d /usr/local/share/man/man3
+      ${csudo}/usr/bin/install -c -m 644 ${jemalloc_dir}/share/man/man3/jemalloc.3 /usr/local/share/man/man3
+    fi
+
+    if [ -d /etc/ld.so.conf.d ]; then
+      echo "/usr/local/lib" | ${csudo}tee /etc/ld.so.conf.d/jemalloc.conf >/dev/null || echo -e "failed to write /etc/ld.so.conf.d/jemalloc.conf"
+      ${csudo}ldconfig
+    else
+      echo "/etc/ld.so.conf.d not found!"
+    fi
+  fi
+}
+
 function install_lib() {
     log_print "start install lib from ${lib_dir} to ${lib_link_dir}"
     ${csudo}rm -f ${lib_link_dir}/libtaos* || :
@@ -663,6 +716,7 @@ function install_TDengine() {
     # Install include, lib, binary and service
     install_include &&
     install_lib &&
+    install_jemalloc
     install_bin
 
     if [[ "$?" != 0 ]];then

packaging/tools/remove.sh

@@ -40,11 +40,16 @@ serverName2="taosd"
 clientName2="taos"
 productName2="TDengine"
 
+adapterName2="${clientName2}adapter"
+demoName2="${clientName2}demo"
 benchmarkName2="${clientName2}Benchmark"
 dumpName2="${clientName2}dump"
+keeperName2="${clientName2}keeper"
+xName2="${clientName2}x"
+explorerName2="${clientName2}-explorer"
 uninstallScript2="rm${clientName2}"
 
-installDir="/usr/local/${clientName}"
+installDir="/usr/local/${clientName2}"
 
 #install main path
 install_main_dir=${installDir}
@@ -55,8 +60,8 @@ local_bin_link_dir="/usr/local/bin"
 
 
 service_config_dir="/etc/systemd/system"
-taos_service_name=${serverName}
-taosadapter_service_name="taosadapter"
+taos_service_name=${serverName2}
+taosadapter_service_name="${clientName2}adapter"
 tarbitrator_service_name="tarbitratord"
 csudo=""
 if command -v sudo >/dev/null; then
@@ -84,14 +89,14 @@ else
 fi
 
 function kill_taosadapter() {
-  pid=$(ps -ef | grep "taosadapter" | grep -v "grep" | awk '{print $2}')
+  pid=$(ps -ef | grep "${adapterName2}" | grep -v "grep" | awk '{print $2}')
   if [ -n "$pid" ]; then
     ${csudo}kill -9 $pid || :
   fi
 }
 
 function kill_taosd() {
-  pid=$(ps -ef | grep ${serverName} | grep -v "grep" | awk '{print $2}')
+  pid=$(ps -ef | grep ${serverName2} | grep -v "grep" | awk '{print $2}')
   if [ -n "$pid" ]; then
     ${csudo}kill -9 $pid || :
   fi
@@ -109,17 +114,17 @@ function clean_bin() {
   ${csudo}rm -f ${bin_link_dir}/${clientName} || :
   ${csudo}rm -f ${bin_link_dir}/${serverName} || :
   ${csudo}rm -f ${bin_link_dir}/udfd || :
-  ${csudo}rm -f ${bin_link_dir}/taosadapter || :
-  ${csudo}rm -f ${bin_link_dir}/taosBenchmark || :
-  ${csudo}rm -f ${bin_link_dir}/taosdemo || :
-  ${csudo}rm -f ${bin_link_dir}/taosdump || :
-  ${csudo}rm -f ${bin_link_dir}/${uninstallScript} || :
+  ${csudo}rm -f ${bin_link_dir}/${adapterName2}     || :
+  ${csudo}rm -f ${bin_link_dir}/${benchmarkName2}   || :
+  ${csudo}rm -f ${bin_link_dir}/${demoName2}        || :
+  ${csudo}rm -f ${bin_link_dir}/${dumpName2}        || :
+  ${csudo}rm -f ${bin_link_dir}/${uninstallScript}  || :
   ${csudo}rm -f ${bin_link_dir}/tarbitrator || :
   ${csudo}rm -f ${bin_link_dir}/set_core || :
   ${csudo}rm -f ${bin_link_dir}/TDinsight.sh || :
-  ${csudo}rm -f ${bin_link_dir}/taoskeeper || :
-  ${csudo}rm -f ${bin_link_dir}/taosx || :
-  ${csudo}rm -f ${bin_link_dir}/taos-explorer || :
+  ${csudo}rm -f ${bin_link_dir}/${keeperName2}      || :
+  ${csudo}rm -f ${bin_link_dir}/${xName2}           || :
+  ${csudo}rm -f ${bin_link_dir}/${explorerName2}    || :
 
   if [ "$verMode" == "cluster" ] && [ "$clientName" != "$clientName2" ]; then
     ${csudo}rm -f ${bin_link_dir}/${clientName2} || :
@@ -130,8 +135,8 @@ function clean_bin() {
 }
 
 function clean_local_bin() {
-  ${csudo}rm -f ${local_bin_link_dir}/taosBenchmark || :
-  ${csudo}rm -f ${local_bin_link_dir}/taosdemo || :
+  ${csudo}rm -f ${local_bin_link_dir}/${benchmarkName2} || :
+  ${csudo}rm -f ${local_bin_link_dir}/${demoName2}      || :
 }
 
 function clean_lib() {
@@ -173,7 +178,7 @@ function clean_service_on_systemd() {
   ${csudo}systemctl disable ${taos_service_name} &>/dev/null || echo &>/dev/null
   ${csudo}rm -f ${taosd_service_config}
 
-  taosadapter_service_config="${service_config_dir}/taosadapter.service"
+  taosadapter_service_config="${service_config_dir}/${clientName2}adapter.service"
   if systemctl is-active --quiet ${taosadapter_service_name}; then
     echo "${productName2}  ${clientName2}Adapter is running, stopping it..."
     ${csudo}systemctl stop ${taosadapter_service_name} &>/dev/null || echo &>/dev/null
@@ -187,7 +192,27 @@ function clean_service_on_systemd() {
     ${csudo}systemctl stop ${tarbitrator_service_name} &>/dev/null || echo &>/dev/null
   fi
   ${csudo}systemctl disable ${tarbitrator_service_name} &>/dev/null || echo &>/dev/null
-  ${csudo}rm -f ${tarbitratord_service_config}
+
+  x_service_config="${service_config_dir}/${xName2}.service"
+  if [ -e "$x_service_config" ]; then
+    if systemctl is-active --quiet ${xName2}; then
+      echo "${productName2} ${xName2} is running, stopping it..."
+      ${csudo}systemctl stop ${xName2} &>/dev/null || echo &>/dev/null
+    fi
+    ${csudo}systemctl disable ${xName2} &>/dev/null || echo &>/dev/null
+    ${csudo}rm -f ${x_service_config}
+  fi
+
+  explorer_service_config="${service_config_dir}/${explorerName2}.service"
+  if [ -e "$explorer_service_config" ]; then
+    if systemctl is-active --quiet ${explorerName2}; then
+      echo "${productName2} ${explorerName2} is running, stopping it..."
+      ${csudo}systemctl stop ${explorerName2} &>/dev/null || echo &>/dev/null
+    fi
+    ${csudo}systemctl disable ${explorerName2} &>/dev/null || echo &>/dev/null
+    ${csudo}rm -f ${explorer_service_config}
+    ${csudo}rm -f /etc/${clientName2}/explorer.toml
+  fi
 }
 
 function clean_service_on_sysvinit() {
@@ -235,8 +260,8 @@ function clean_service_on_sysvinit() {
 function clean_service_on_launchctl() {
   ${csudouser}launchctl unload -w /Library/LaunchDaemons/com.taosdata.taosd.plist > /dev/null 2>&1 || :
   ${csudo}rm /Library/LaunchDaemons/com.taosdata.taosd.plist > /dev/null 2>&1 || :
-  ${csudouser}launchctl unload -w /Library/LaunchDaemons/com.taosdata.taosadapter.plist > /dev/null 2>&1 || :
-  ${csudo}rm /Library/LaunchDaemons/com.taosdata.taosadapter.plist > /dev/null 2>&1 || :
+  ${csudouser}launchctl unload -w /Library/LaunchDaemons/com.taosdata.${clientName2}adapter.plist > /dev/null 2>&1 || :
+  ${csudo}rm /Library/LaunchDaemons/com.taosdata.${clientName2}adapter.plist > /dev/null 2>&1 || :
 }
 
 function clean_service() {

packaging/tools/remove_client.sh

@@ -15,11 +15,12 @@ uninstallScript="rmtaos"
 clientName2="taos"
 productName2="TDengine"
 
-benchmarkName2="${clientName}Benchmark"
-dumpName2="${clientName}dump"
-uninstallScript2="rm${clientName}"
+benchmarkName2="${clientName2}Benchmark"
+demoName2="${clientName2}demo"
+dumpName2="${clientName2}dump"
+uninstallScript2="rm${clientName2}"
 
-installDir="/usr/local/${clientName}"
+installDir="/usr/local/${clientName2}"
 
 #install main path
 install_main_dir=${installDir}
@@ -44,14 +45,17 @@ function kill_client() {
 
 function clean_bin() {
     # Remove link
-    ${csudo}rm -f ${bin_link_dir}/${clientName}      || :
-    ${csudo}rm -f ${bin_link_dir}/taosdemo  || :
-    ${csudo}rm -f ${bin_link_dir}/taosdump  || :
+    ${csudo}rm -f ${bin_link_dir}/${clientName2}      || :
+    ${csudo}rm -f ${bin_link_dir}/${demoName2}        || :
+    ${csudo}rm -f ${bin_link_dir}/${benchmarkName2}   || :
+    ${csudo}rm -f ${bin_link_dir}/${dumpName2}        || :
     ${csudo}rm -f ${bin_link_dir}/${uninstallScript}    || :
     ${csudo}rm -f ${bin_link_dir}/set_core  || :
 
     if [ "$verMode" == "cluster" ] && [ "$clientName" != "$clientName2" ]; then
         ${csudo}rm -f ${bin_link_dir}/${clientName2} || :
+        ${csudo}rm -f ${bin_link_dir}/${demoName2}        || :
+        ${csudo}rm -f ${bin_link_dir}/${benchmarkName2}   || :
         ${csudo}rm -f ${bin_link_dir}/${dumpName2} || :
         ${csudo}rm -f ${bin_link_dir}/${uninstallScript2} || :
     fi

source/client/inc/clientLog.h

@@ -26,6 +26,7 @@ extern "C" {
 #define tscFatal(...)  do { if (cDebugFlag & DEBUG_FATAL) { taosPrintLog("TSC FATAL ", DEBUG_FATAL, cDebugFlag, __VA_ARGS__); }} while(0)
 #define tscError(...)  do { if (cDebugFlag & DEBUG_ERROR) { taosPrintLog("TSC ERROR ", DEBUG_ERROR, cDebugFlag, __VA_ARGS__); }} while(0)
 #define tscWarn(...)   do { if (cDebugFlag & DEBUG_WARN)  { taosPrintLog("TSC WARN ", DEBUG_WARN, cDebugFlag, __VA_ARGS__); }}  while(0)
+#define tscWarnL(...)  do { if (cDebugFlag & DEBUG_WARN)  { taosPrintLongString("TSC WARN ", DEBUG_WARN, cDebugFlag, __VA_ARGS__); }}  while(0)
 #define tscInfo(...)   do { if (cDebugFlag & DEBUG_INFO)  { taosPrintLog("TSC ", DEBUG_INFO, cDebugFlag, __VA_ARGS__); }} while(0)
 #define tscDebug(...)  do { if (cDebugFlag & DEBUG_DEBUG) { taosPrintLog("TSC ", DEBUG_DEBUG, cDebugFlag, __VA_ARGS__); }} while(0)
 #define tscTrace(...)  do { if (cDebugFlag & DEBUG_TRACE) { taosPrintLog("TSC ", DEBUG_TRACE, cDebugFlag, __VA_ARGS__); }} while(0)

source/client/inc/clientStmt.h

@@ -102,6 +102,7 @@ typedef struct STscStmt {
   SStmtBindInfo bInfo;
 
   int64_t reqid;
+  int32_t errCode;
 } STscStmt;
 
 extern char *gStmtStatusStr[];
@@ -121,6 +122,7 @@ extern char *gStmtStatusStr[];
     int32_t _code = c;                \
     if (_code != TSDB_CODE_SUCCESS) { \
       terrno = _code;                 \
+      pStmt->errCode = _code;         \
       return _code;                   \
     }                                 \
   } while (0)
@@ -129,6 +131,7 @@ extern char *gStmtStatusStr[];
     int32_t _code = c;                \
     if (_code != TSDB_CODE_SUCCESS) { \
       terrno = _code;                 \
+      pStmt->errCode = _code;         \
     }                                 \
     return _code;                     \
   } while (0)
@@ -137,6 +140,7 @@ extern char *gStmtStatusStr[];
     code = c;                        \
     if (code != TSDB_CODE_SUCCESS) { \
       terrno = code;                 \
+      pStmt->errCode = code;         \
       goto _return;                  \
     }                                \
   } while (0)

source/client/src/clientEnv.c

@@ -30,6 +30,10 @@
 #include "tsched.h"
 #include "ttime.h"
 
+#if defined(CUS_NAME) || defined(CUS_PROMPT) || defined(CUS_EMAIL)
+#include "cus_name.h"
+#endif
+
 #define TSC_VAR_NOT_RELEASE 1
 #define TSC_VAR_RELEASED    0
 
@@ -103,6 +107,7 @@ static void deregisterRequest(SRequestObj *pRequest) {
 
   if (duration >= SLOW_QUERY_INTERVAL) {
     atomic_add_fetch_64((int64_t *)&pActivity->numOfSlowQueries, 1);
+    tscWarnL("slow query: %s, duration:%" PRId64, pRequest->sqlstr, duration);
   }
 
   releaseTscObj(pTscObj->id);
@@ -541,9 +546,15 @@ void taos_init_imp(void) {
 
   deltaToUtcInitOnce();
 
-  if (taosCreateLog("taoslog", 10, configDir, NULL, NULL, NULL, NULL, 1) != 0) {
+  char logDirName[64] = {0};
+#ifdef CUS_PROMPT
+  snprintf(logDirName, 64, "%slog", CUS_PROMPT);
+#else
+  snprintf(logDirName, 64, "taoslog");
+#endif
+  if (taosCreateLog(logDirName, 10, configDir, NULL, NULL, NULL, NULL, 1) != 0) {
     // ignore create log failed, only print
-    printf(" WARING: Create taoslog failed:%s. configDir=%s\n", strerror(errno), configDir);
+    printf(" WARING: Create %s failed:%s. configDir=%s\n", logDirName, strerror(errno), configDir);
   }
 
   if (taosInitCfg(configDir, NULL, NULL, NULL, NULL, 1) != 0) {

source/client/src/clientMsgHandler.c

@@ -76,7 +76,7 @@ int32_t processConnectRsp(void* param, SDataBuf* pMsg, int32_t code) {
     goto End;
   }
 
-  if ((code = taosCheckVersionCompatibleFromStr(version, connectRsp.sVer, 2)) != 0) {
+  if ((code = taosCheckVersionCompatibleFromStr(version, connectRsp.sVer, 3)) != 0) {
     setErrno(pRequest, code);
     tsem_post(&pRequest->body.rspSem);
     goto End;
@@ -506,7 +506,7 @@ int32_t processShowVariablesRsp(void* param, SDataBuf* pMsg, int32_t code) {
       code = setQueryResultFromRsp(&pRequest->body.resInfo, pRes, false, true);
     }
 
-    if(code != 0){
+    if (code != 0) {
       taosMemoryFree(pRes);
     }
     tFreeSShowVariablesRsp(&rsp);

source/client/src/clientSml.c

@@ -174,6 +174,7 @@ static int32_t smlParseTableName(SArray *tags, char *childTableName) {
     if (childTableNameLen == tag->keyLen && strncmp(tag->key, tsSmlChildTableName, tag->keyLen) == 0) {
       memset(childTableName, 0, TSDB_TABLE_NAME_LEN);
       strncpy(childTableName, tag->value, (tag->length < TSDB_TABLE_NAME_LEN ? tag->length : TSDB_TABLE_NAME_LEN));
+      taosArrayRemove(tags, i);
       break;
     }
   }
@@ -257,19 +258,14 @@ cleanup:
   kvVal->f = (float)result;
 
 #define SET_BIGINT                                                                                         \
-  if (smlDoubleToInt64OverFlow(result)) {                                                                  \
-    errno = 0;                                                                                             \
-    int64_t tmp = taosStr2Int64(pVal, &endptr, 10);                                                        \
-    if (errno == ERANGE) {                                                                                 \
-      smlBuildInvalidDataMsg(msg, "big int out of range[-9223372036854775808,9223372036854775807]", pVal); \
-      return false;                                                                                        \
-    }                                                                                                      \
-    kvVal->type = TSDB_DATA_TYPE_BIGINT;                                                                   \
-    kvVal->i = tmp;                                                                                        \
-    return true;                                                                                           \
-  }                                                                                                        \
-  kvVal->type = TSDB_DATA_TYPE_BIGINT;                                                                     \
-  kvVal->i = (int64_t)result;
+  errno = 0;                                                                                             \
+  int64_t tmp = taosStr2Int64(pVal, &endptr, 10);                                                        \
+  if (errno == ERANGE) {                                                                                 \
+    smlBuildInvalidDataMsg(msg, "big int out of range[-9223372036854775808,9223372036854775807]", pVal); \
+    return false;                                                                                        \
+  }                                                                                                      \
+  kvVal->type = TSDB_DATA_TYPE_BIGINT;                                                                   \
+  kvVal->i = tmp;
 
 #define SET_INT                                                                    \
   if (!IS_VALID_INT(result)) {                                                     \
@@ -288,19 +284,14 @@ cleanup:
   kvVal->i = result;
 
 #define SET_UBIGINT                                                                               \
-  if (result >= (double)UINT64_MAX || result < 0) {                                               \
-    errno = 0;                                                                                    \
-    uint64_t tmp = taosStr2UInt64(pVal, &endptr, 10);                                             \
-    if (errno == ERANGE || result < 0) {                                                          \
-      smlBuildInvalidDataMsg(msg, "unsigned big int out of range[0,18446744073709551615]", pVal); \
-      return false;                                                                               \
-    }                                                                                             \
-    kvVal->type = TSDB_DATA_TYPE_UBIGINT;                                                         \
-    kvVal->u = tmp;                                                                               \
-    return true;                                                                                  \
-  }                                                                                               \
-  kvVal->type = TSDB_DATA_TYPE_UBIGINT;                                                           \
-  kvVal->u = result;
+  errno = 0;                                                                                    \
+  uint64_t tmp = taosStr2UInt64(pVal, &endptr, 10);                                             \
+  if (errno == ERANGE || result < 0) {                                                          \
+    smlBuildInvalidDataMsg(msg, "unsigned big int out of range[0,18446744073709551615]", pVal); \
+    return false;                                                                               \
+  }                                                                                             \
+  kvVal->type = TSDB_DATA_TYPE_UBIGINT;                                                         \
+  kvVal->u = tmp;
 
 #define SET_UINT                                                                  \
   if (!IS_VALID_UINT(result)) {                                                   \
@@ -543,8 +534,7 @@ static int32_t smlGenerateSchemaAction(SSchema *colField, SHashObj *colHash, SSm
   uint16_t *index = colHash ? (uint16_t *)taosHashGet(colHash, kv->key, kv->keyLen) : NULL;
   if (index) {
     if (colField[*index].type != kv->type) {
-      uError("SML:0x%" PRIx64 " point type and db type mismatch. key: %s. point type: %d, db type: %d", info->id,
-             kv->key, colField[*index].type, kv->type);
+      uError("SML:0x%" PRIx64 " point type and db type mismatch. point type: %d, db type: %d, key: %s", info->id, colField[*index].type, kv->type, kv->key);
       return TSDB_CODE_TSC_INVALID_VALUE;
     }
 

source/client/src/clientStmt.c

@@ -32,8 +32,14 @@ int32_t stmtSwitchStatus(STscStmt* pStmt, STMT_STATUS newStatus) {
     STMT_LOG_SEQ(newStatus);
   }
 
+  if (pStmt->errCode && newStatus != STMT_PREPARE) {
+    STMT_DLOG("stmt already failed with err: %s", tstrerror(pStmt->errCode));
+    return pStmt->errCode;
+  }
+
   switch (newStatus) {
     case STMT_PREPARE:
+      pStmt->errCode = 0;    
       break;
     case STMT_SETTBNAME:
       if (STMT_STATUS_EQ(INIT) || STMT_STATUS_EQ(BIND) || STMT_STATUS_EQ(BIND_COL)) {
@@ -197,7 +203,10 @@ int32_t stmtGetExecInfo(TAOS_STMT* stmt, SHashObj** pVgHash, SHashObj** pBlockHa
   STscStmt* pStmt = (STscStmt*)stmt;
 
   *pVgHash = pStmt->sql.pVgHash;
+  pStmt->sql.pVgHash = NULL;
+  
   *pBlockHash = pStmt->exec.pBlockHash;
+  pStmt->exec.pBlockHash = NULL;
 
   return TSDB_CODE_SUCCESS;
 }
@@ -325,6 +334,8 @@ int32_t stmtCleanExecInfo(STscStmt* pStmt, bool keepTable, bool deepClean) {
 }
 
 int32_t stmtCleanSQLInfo(STscStmt* pStmt) {
+  STMT_DLOG_E("start to free SQL info");
+  
   taosMemoryFree(pStmt->sql.queryRes.fields);
   taosMemoryFree(pStmt->sql.queryRes.userFields);
   taosMemoryFree(pStmt->sql.sqlStr);
@@ -351,6 +362,8 @@ int32_t stmtCleanSQLInfo(STscStmt* pStmt) {
 
   memset(&pStmt->sql, 0, sizeof(pStmt->sql));
 
+  STMT_DLOG_E("end to free SQL info");
+
   return TSDB_CODE_SUCCESS;
 }
 
@@ -441,11 +454,10 @@ int32_t stmtGetFromCache(STscStmt* pStmt) {
                            .mgmtEps = getEpSet_s(&pStmt->taos->pAppInfo->mgmtEp)};
   int32_t          code = catalogGetTableMeta(pStmt->pCatalog, &conn, &pStmt->bInfo.sname, &pTableMeta);
   if (TSDB_CODE_PAR_TABLE_NOT_EXIST == code) {
-    STMT_ERR_RET(stmtCleanBindInfo(pStmt));
-
     tscDebug("tb %s not exist", pStmt->bInfo.tbFName);
+    stmtCleanBindInfo(pStmt);
 
-    return TSDB_CODE_SUCCESS;
+    STMT_ERR_RET(code);
   }
 
   STMT_ERR_RET(code);
@@ -922,9 +934,13 @@ _return:
 int stmtClose(TAOS_STMT* stmt) {
   STscStmt* pStmt = (STscStmt*)stmt;
 
+  STMT_DLOG_E("start to free stmt");
+
   stmtCleanSQLInfo(pStmt);
   taosMemoryFree(stmt);
 
+  STMT_DLOG_E("stmt freed");
+
   return TSDB_CODE_SUCCESS;
 }
 

source/client/test/clientTests.cpp

@@ -161,10 +161,10 @@ void *queryThread(void *arg) {
   return NULL;
 }
 
-static int32_t numOfThreads = 1;
+int32_t numOfThreads = 1;
 
 void tmq_commit_cb_print(tmq_t *pTmq, int32_t code, void *param) {
-  printf("success, code:%d\n", code);
+  printf("auto commit success, code:%d\n\n\n\n", code);
 }
 
 void* doConsumeData(void* param) {
@@ -173,7 +173,7 @@ void* doConsumeData(void* param) {
   tmq_conf_t* conf = tmq_conf_new();
   tmq_conf_set(conf, "enable.auto.commit", "true");
   tmq_conf_set(conf, "auto.commit.interval.ms", "1000");
-  tmq_conf_set(conf, "group.id", "cgrpName12");
+  tmq_conf_set(conf, "group.id", "cgrpName41");
   tmq_conf_set(conf, "td.connect.user", "root");
   tmq_conf_set(conf, "td.connect.pass", "taosdata");
   tmq_conf_set(conf, "auto.offset.reset", "earliest");
@@ -998,7 +998,8 @@ TEST(clientCase, sub_db_test) {
 
   // 创建订阅 topics 列表
   tmq_list_t* topicList = tmq_list_new();
-  tmq_list_append(topicList, "topic_db1");
+  tmq_list_append(topicList, "topic_t1");
+  tmq_list_append(topicList, "topic_s2");
 
   // 启动订阅
   tmq_subscribe(tmq, topicList);
@@ -1059,7 +1060,7 @@ TEST(clientCase, sub_tb_test) {
   tmq_conf_t* conf = tmq_conf_new();
   tmq_conf_set(conf, "enable.auto.commit", "true");
   tmq_conf_set(conf, "auto.commit.interval.ms", "1000");
-  tmq_conf_set(conf, "group.id", "cgrpName27");
+  tmq_conf_set(conf, "group.id", "cgrpName45");
   tmq_conf_set(conf, "td.connect.user", "root");
   tmq_conf_set(conf, "td.connect.pass", "taosdata");
   tmq_conf_set(conf, "auto.offset.reset", "earliest");

source/common/src/tdatablock.c

@@ -147,9 +147,17 @@ int32_t colDataReserve(SColumnInfoData* pColumnInfoData, size_t newSize) {
   return TSDB_CODE_SUCCESS;
 }
 
-static void doCopyNItems(struct SColumnInfoData* pColumnInfoData, int32_t currentRow, const char* pData,
-                         int32_t itemLen, int32_t numOfRows) {
-  ASSERT(pColumnInfoData->info.bytes >= itemLen);
+static int32_t doCopyNItems(struct SColumnInfoData* pColumnInfoData, int32_t currentRow, const char* pData,
+                         int32_t itemLen, int32_t numOfRows, bool trimValue) {
+  if (pColumnInfoData->info.bytes < itemLen) {
+    uWarn("column/tag actual data len %d is bigger than schema len %d, trim it:%d", itemLen, pColumnInfoData->info.bytes, trimValue);
+    if (trimValue) {
+      itemLen = pColumnInfoData->info.bytes;
+    } else {
+      return TSDB_CODE_TDB_INVALID_TABLE_SCHEMA_VER;
+    }
+  }
+  
   size_t start = 1;
 
   // the first item
@@ -178,10 +186,12 @@ static void doCopyNItems(struct SColumnInfoData* pColumnInfoData, int32_t curren
 
     pColumnInfoData->varmeta.length += numOfRows * itemLen;
   }
+
+  return TSDB_CODE_SUCCESS;
 }
 
 int32_t colDataSetNItems(SColumnInfoData* pColumnInfoData, uint32_t currentRow, const char* pData,
-                            uint32_t numOfRows) {
+                            uint32_t numOfRows, bool trimValue) {
   int32_t len = pColumnInfoData->info.bytes;
   if (IS_VAR_DATA_TYPE(pColumnInfoData->info.type)) {
     len = varDataTLen(pData);
@@ -193,8 +203,7 @@ int32_t colDataSetNItems(SColumnInfoData* pColumnInfoData, uint32_t currentRow,
     }
   }
 
-  doCopyNItems(pColumnInfoData, currentRow, pData, len, numOfRows);
-  return TSDB_CODE_SUCCESS;
+  return doCopyNItems(pColumnInfoData, currentRow, pData, len, numOfRows, trimValue);
 }
 
 static void doBitmapMerge(SColumnInfoData* pColumnInfoData, int32_t numOfRow1, const SColumnInfoData* pSource,
@@ -973,7 +982,7 @@ int32_t blockDataSort(SSDataBlock* pDataBlock, SArray* pOrderInfo) {
         taosSort(pColInfoData->pData, pDataBlock->info.rows, pColInfoData->info.bytes, fn);
 
         int64_t p1 = taosGetTimestampUs();
-        uDebug("blockDataSort easy cost:%" PRId64 ", rows:%d\n", p1 - p0, pDataBlock->info.rows);
+        uDebug("blockDataSort easy cost:%" PRId64 ", rows:%" PRId64 "\n", p1 - p0, pDataBlock->info.rows);
 
         return TSDB_CODE_SUCCESS;
       } else {  // var data type
@@ -1026,6 +1035,7 @@ int32_t blockDataSort(SSDataBlock* pDataBlock, SArray* pOrderInfo) {
   return TSDB_CODE_SUCCESS;
 }
 
+#if 0
 typedef struct SHelper {
   int32_t index;
   union {
@@ -1074,59 +1084,20 @@ SHelper* createTupleIndex_rv(int32_t numOfRows, SArray* pOrderInfo, SSDataBlock*
 int32_t dataBlockCompar_rv(const void* p1, const void* p2, const void* param) {
   const SSDataBlockSortHelper* pHelper = (const SSDataBlockSortHelper*)param;
 
-  //  SSDataBlock* pDataBlock = pHelper->pDataBlock;
-
   SHelper* left = (SHelper*)p1;
   SHelper* right = (SHelper*)p2;
 
   SArray* pInfo = pHelper->orderInfo;
 
   int32_t offset = 0;
-  //  for(int32_t i = 0; i < pInfo->size; ++i) {
-  //    SBlockOrderInfo* pOrder = TARRAY_GET_ELEM(pInfo, 0);
-  //    SColumnInfoData* pColInfoData = pOrder->pColData;//TARRAY_GET_ELEM(pDataBlock->pDataBlock, pOrder->colIndex);
-
-  //    if (pColInfoData->hasNull) {
-  //      bool leftNull  = colDataIsNull(pColInfoData, pDataBlock->info.rows, left, pDataBlock->pBlockAgg);
-  //      bool rightNull = colDataIsNull(pColInfoData, pDataBlock->info.rows, right, pDataBlock->pBlockAgg);
-  //      if (leftNull && rightNull) {
-  //        continue; // continue to next slot
-  //      }
-  //
-  //      if (rightNull) {
-  //        return pHelper->nullFirst? 1:-1;
-  //      }
-  //
-  //      if (leftNull) {
-  //        return pHelper->nullFirst? -1:1;
-  //      }
-  //    }
-
-  //    void* left1  = colDataGetData(pColInfoData, left);
-  //    void* right1 = colDataGetData(pColInfoData, right);
-
-  //    switch(pColInfoData->info.type) {
-  //      case TSDB_DATA_TYPE_INT: {
   int32_t leftx = *(int32_t*)left->pData;    //*(int32_t*)(left->pData + offset);
   int32_t rightx = *(int32_t*)right->pData;  //*(int32_t*)(right->pData + offset);
 
-  //        offset += pColInfoData->info.bytes;
   if (leftx == rightx) {
-    //          break;
     return 0;
   } else {
-    //          if (pOrder->order == TSDB_ORDER_ASC) {
     return (leftx < rightx) ? -1 : 1;
-    //          } else {
-    //            return (leftx < rightx)? 1:-1;
-    //          }
-  }
-  //      }
-  //      default:
-  //        assert(0);
-  //    }
-  //  }
-
+  }
   return 0;
 }
 
@@ -1170,6 +1141,7 @@ int32_t blockDataSort_rv(SSDataBlock* pDataBlock, SArray* pOrderInfo, bool nullF
   //  destroyTupleIndex(index);
   return 0;
 }
+#endif
 
 void blockDataCleanup(SSDataBlock* pDataBlock) {
   blockDataEmpty(pDataBlock);
@@ -1180,7 +1152,7 @@ void blockDataCleanup(SSDataBlock* pDataBlock) {
 
 void blockDataEmpty(SSDataBlock* pDataBlock) {
   SDataBlockInfo* pInfo = &pDataBlock->info;
-  if (pInfo->capacity == 0 || pInfo->rows > pDataBlock->info.capacity) {
+  if (pInfo->capacity == 0) {
     return;
   }
 
@@ -1618,12 +1590,13 @@ static void doShiftBitmap(char* nullBitmap, size_t n, size_t total) {
         i += 1;
       }
     } else if (n > 8) {
-      int32_t gap = len - newLen;
+      int32_t remain = (total % 8 != 0 && total % 8 <= tail) ? 1 : 0;
+      int32_t gap = len - newLen - remain;
       while (i < newLen) {
         uint8_t v = p[i + gap];
         p[i] = (v << tail);
 
-        if (i < newLen - 1) {
+        if (i < newLen - 1 + remain) {
           uint8_t next = p[i + gap + 1];
           p[i] |= (next >> (8 - tail));
         }
@@ -1739,14 +1712,14 @@ int32_t tEncodeDataBlock(void** buf, const SSDataBlock* pBlock) {
   int64_t tbUid = pBlock->info.id.uid;
   int16_t numOfCols = taosArrayGetSize(pBlock->pDataBlock);
   int16_t hasVarCol = pBlock->info.hasVarCol;
-  int32_t rows = pBlock->info.rows;
+  int64_t rows = pBlock->info.rows;
   int32_t sz = taosArrayGetSize(pBlock->pDataBlock);
 
   int32_t tlen = 0;
   tlen += taosEncodeFixedI64(buf, tbUid);
   tlen += taosEncodeFixedI16(buf, numOfCols);
   tlen += taosEncodeFixedI16(buf, hasVarCol);
-  tlen += taosEncodeFixedI32(buf, rows);
+  tlen += taosEncodeFixedI64(buf, rows);
   tlen += taosEncodeFixedI32(buf, sz);
   for (int32_t i = 0; i < sz; i++) {
     SColumnInfoData* pColData = (SColumnInfoData*)taosArrayGet(pBlock->pDataBlock, i);
@@ -1777,7 +1750,7 @@ void* tDecodeDataBlock(const void* buf, SSDataBlock* pBlock) {
   buf = taosDecodeFixedU64(buf, &pBlock->info.id.uid);
   buf = taosDecodeFixedI16(buf, &numOfCols);
   buf = taosDecodeFixedI16(buf, &pBlock->info.hasVarCol);
-  buf = taosDecodeFixedI32(buf, &pBlock->info.rows);
+  buf = taosDecodeFixedI64(buf, &pBlock->info.rows);
   buf = taosDecodeFixedI32(buf, &sz);
   pBlock->pDataBlock = taosArrayInit(sz, sizeof(SColumnInfoData));
   for (int32_t i = 0; i < sz; i++) {
@@ -1878,6 +1851,7 @@ static char* formatTimestamp(char* buf, int64_t val, int precision) {
   return buf;
 }
 
+#if 0
 void blockDebugShowDataBlock(SSDataBlock* pBlock, const char* flag) {
   SArray* dataBlocks = taosArrayInit(1, sizeof(SSDataBlock*));
   taosArrayPush(dataBlocks, &pBlock);
@@ -1970,6 +1944,8 @@ void blockDebugShowDataBlocks(const SArray* dataBlocks, const char* flag) {
   }
 }
 
+#endif
+
 // for debug
 char* dumpBlockData(SSDataBlock* pDataBlock, const char* flag, char** pDataBuf) {
   int32_t size = 2048;
@@ -1981,7 +1957,7 @@ char* dumpBlockData(SSDataBlock* pDataBlock, const char* flag, char** pDataBuf)
   int32_t len = 0;
   len += snprintf(dumpBuf + len, size - len,
                   "===stream===%s|block type %d|child id %d|group id:%" PRIu64 "|uid:%" PRId64
-                  "|rows:%d|version:%" PRIu64 "|cal start:%" PRIu64 "|cal end:%" PRIu64 "|tbl:%s\n",
+                  "|rows:%" PRId64 "|version:%" PRIu64 "|cal start:%" PRIu64 "|cal end:%" PRIu64 "|tbl:%s\n",
                   flag, (int32_t)pDataBlock->info.type, pDataBlock->info.childId, pDataBlock->info.id.groupId,
                   pDataBlock->info.id.uid, pDataBlock->info.rows, pDataBlock->info.version,
                   pDataBlock->info.calWin.skey, pDataBlock->info.calWin.ekey, pDataBlock->info.parTbName);

source/common/src/tdataformat.c

@@ -2453,6 +2453,11 @@ int32_t tColDataAddValueByDataBlock(SColData *pColData, int8_t type, int32_t byt
         code = tColDataAppendValueImpl[pColData->flag][CV_FLAG_NULL](pColData, NULL, 0);
         if (code) goto _exit;
       } else {
+        if(ASSERT(varDataTLen(data + offset) <= bytes)){
+          uError("var data length invalid, varDataTLen(data + offset):%d <= bytes:%d", (int)varDataTLen(data + offset), bytes);
+          code = TSDB_CODE_INVALID_PARA;
+          goto _exit;
+        }
         code = tColDataAppendValueImpl[pColData->flag][CV_FLAG_VALUE](pColData, (uint8_t *)varDataVal(data + offset),
                                                                       varDataLen(data + offset));
       }

source/common/src/tglobal.c

@@ -20,6 +20,10 @@
 #include "tlog.h"
 #include "tmisce.h"
 
+#if defined(CUS_NAME) || defined(CUS_PROMPT) || defined(CUS_EMAIL)
+#include "cus_name.h"
+#endif
+
 GRANT_CFG_DECLARE;
 
 SConfig *tsCfg = NULL;
@@ -228,7 +232,11 @@ static int32_t taosLoadCfg(SConfig *pCfg, const char **envCmd, const char *input
 
   taosExpandDir(inputCfgDir, cfgDir, PATH_MAX);
   if (taosIsDir(cfgDir)) {
+#ifdef CUS_PROMPT
+    snprintf(cfgFile, sizeof(cfgFile), "%s" TD_DIRSEP "%s.cfg", cfgDir, CUS_PROMPT);
+#else
     snprintf(cfgFile, sizeof(cfgFile), "%s" TD_DIRSEP "taos.cfg", cfgDir);
+#endif
   } else {
     tstrncpy(cfgFile, cfgDir, sizeof(cfgDir));
   }

source/common/src/tmsg.c

@@ -7305,6 +7305,7 @@ void tDestroySSubmitReq2(SSubmitReq2 *pReq, int32_t flag) {
     tDestroySSubmitTbData(&aSubmitTbData[i], flag);
   }
   taosArrayDestroy(pReq->aSubmitTbData);
+  pReq->aSubmitTbData = NULL;
 }
 
 int32_t tEncodeSSubmitRsp2(SEncoder *pCoder, const SSubmitRsp2 *pRsp) {

source/dnode/mgmt/exe/dmMain.c

@@ -20,6 +20,21 @@
 #include "tglobal.h"
 #include "version.h"
 
+#if defined(CUS_NAME) || defined(CUS_PROMPT) || defined(CUS_EMAIL)
+#include "cus_name.h"
+#else
+#ifndef CUS_NAME
+    #define CUS_NAME      "TDengine"
+#endif
+
+#ifndef CUS_PROMPT
+    #define CUS_PROMPT    "taos"
+#endif
+
+#ifndef CUS_EMAIL
+    #define CUS_EMAIL     "<support@taosdata.com>"
+#endif
+#endif
 // clang-format off
 #define DM_APOLLO_URL    "The apollo string to use when configuring the server, such as: -a 'jsonFile:./tests/cfg.json', cfg.json text can be '{\"fqdn\":\"td1\"}'."
 #define DM_CFG_DIR       "Configuration directory."
@@ -232,7 +247,7 @@ static void dmDumpCfg() {
 }
 
 static int32_t dmInitLog() {
-  return taosCreateLog("taosdlog", 1, configDir, global.envCmd, global.envFile, global.apolloUrl, global.pArgs, 0);
+  return taosCreateLog(CUS_PROMPT"dlog", 1, configDir, global.envCmd, global.envFile, global.apolloUrl, global.pArgs, 0);
 }
 
 static void taosCleanupArgs() {

source/dnode/mgmt/mgmt_mnode/inc/mmInt.h

@@ -34,7 +34,7 @@ typedef struct SMnodeMgmt {
   SSingleWorker  readWorker;
   SSingleWorker  writeWorker;
   SSingleWorker  syncWorker;
-  SSingleWorker  syncCtrlWorker;
+  SSingleWorker  syncRdWorker;
   bool           stopped;
   int32_t        refCount;
   TdThreadRwlock lock;
@@ -54,7 +54,7 @@ int32_t mmStartWorker(SMnodeMgmt *pMgmt);
 void    mmStopWorker(SMnodeMgmt *pMgmt);
 int32_t mmPutMsgToWriteQueue(SMnodeMgmt *pMgmt, SRpcMsg *pMsg);
 int32_t mmPutMsgToSyncQueue(SMnodeMgmt *pMgmt, SRpcMsg *pMsg);
-int32_t mmPutMsgToSyncCtrlQueue(SMnodeMgmt *pMgmt, SRpcMsg *pMsg);
+int32_t mmPutMsgToSyncRdQueue(SMnodeMgmt *pMgmt, SRpcMsg *pMsg);
 int32_t mmPutMsgToReadQueue(SMnodeMgmt *pMgmt, SRpcMsg *pMsg);
 int32_t mmPutMsgToQueryQueue(SMnodeMgmt *pMgmt, SRpcMsg *pMsg);
 int32_t mmPutMsgToFetchQueue(SMnodeMgmt *pMgmt, SRpcMsg *pMsg);

source/dnode/mgmt/mgmt_mnode/src/mmHandle.c

@@ -188,7 +188,8 @@ SArray *mmGetMsgHandles() {
   if (dmSetMgmtHandle(pArray, TDMT_VND_DROP_INDEX_RSP, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_VND_DISABLE_WRITE_RSP, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
 
-  if (dmSetMgmtHandle(pArray, TDMT_SYNC_TIMEOUT, mmPutMsgToSyncQueue, 1) == NULL) goto _OVER;
+  if (dmSetMgmtHandle(pArray, TDMT_SYNC_TIMEOUT_ELECTION, mmPutMsgToSyncQueue, 1) == NULL) goto _OVER;
+  if (dmSetMgmtHandle(pArray, TDMT_SYNC_HEARTBEAT, mmPutMsgToSyncRdQueue, 1) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_SYNC_CLIENT_REQUEST, mmPutMsgToSyncQueue, 1) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_SYNC_CLIENT_REQUEST_BATCH, mmPutMsgToSyncQueue, 1) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_SYNC_CLIENT_REQUEST_REPLY, mmPutMsgToSyncQueue, 1) == NULL) goto _OVER;
@@ -198,11 +199,12 @@ SArray *mmGetMsgHandles() {
   if (dmSetMgmtHandle(pArray, TDMT_SYNC_APPEND_ENTRIES_BATCH, mmPutMsgToSyncQueue, 1) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_SYNC_APPEND_ENTRIES_REPLY, mmPutMsgToSyncQueue, 1) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_SYNC_SNAPSHOT_SEND, mmPutMsgToSyncQueue, 1) == NULL) goto _OVER;
-  if (dmSetMgmtHandle(pArray, TDMT_SYNC_SNAPSHOT_RSP, mmPutMsgToSyncQueue, 1) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_SYNC_PRE_SNAPSHOT, mmPutMsgToSyncQueue, 1) == NULL) goto _OVER;
-  if (dmSetMgmtHandle(pArray, TDMT_SYNC_PRE_SNAPSHOT_REPLY, mmPutMsgToSyncQueue, 1) == NULL) goto _OVER;
-  if (dmSetMgmtHandle(pArray, TDMT_SYNC_HEARTBEAT, mmPutMsgToSyncCtrlQueue, 1) == NULL) goto _OVER;
-  if (dmSetMgmtHandle(pArray, TDMT_SYNC_HEARTBEAT_REPLY, mmPutMsgToSyncCtrlQueue, 1) == NULL) goto _OVER;
+
+  if (dmSetMgmtHandle(pArray, TDMT_SYNC_TIMEOUT, mmPutMsgToSyncRdQueue, 1) == NULL) goto _OVER;
+  if (dmSetMgmtHandle(pArray, TDMT_SYNC_HEARTBEAT_REPLY, mmPutMsgToSyncRdQueue, 1) == NULL) goto _OVER;
+  if (dmSetMgmtHandle(pArray, TDMT_SYNC_SNAPSHOT_RSP, mmPutMsgToSyncRdQueue, 1) == NULL) goto _OVER;
+  if (dmSetMgmtHandle(pArray, TDMT_SYNC_PRE_SNAPSHOT_REPLY, mmPutMsgToSyncRdQueue, 1) == NULL) goto _OVER;
 
   code = 0;
 

source/dnode/mgmt/mgmt_mnode/src/mmWorker.c

@@ -111,8 +111,8 @@ int32_t mmPutMsgToSyncQueue(SMnodeMgmt *pMgmt, SRpcMsg *pMsg) {
   return mmPutMsgToWorker(pMgmt, &pMgmt->syncWorker, pMsg);
 }
 
-int32_t mmPutMsgToSyncCtrlQueue(SMnodeMgmt *pMgmt, SRpcMsg *pMsg) {
-  return mmPutMsgToWorker(pMgmt, &pMgmt->syncCtrlWorker, pMsg);
+int32_t mmPutMsgToSyncRdQueue(SMnodeMgmt *pMgmt, SRpcMsg *pMsg) {
+  return mmPutMsgToWorker(pMgmt, &pMgmt->syncRdWorker, pMsg);
 }
 
 int32_t mmPutMsgToReadQueue(SMnodeMgmt *pMgmt, SRpcMsg *pMsg) {
@@ -151,8 +151,8 @@ int32_t mmPutMsgToQueue(SMnodeMgmt *pMgmt, EQueueType qtype, SRpcMsg *pRpc) {
     case SYNC_QUEUE:
       pWorker = &pMgmt->syncWorker;
       break;
-    case SYNC_CTRL_QUEUE:
-      pWorker = &pMgmt->syncCtrlWorker;
+    case SYNC_RD_QUEUE:
+      pWorker = &pMgmt->syncRdWorker;
       break;
     default:
       terrno = TSDB_CODE_INVALID_PARA;
@@ -238,12 +238,12 @@ int32_t mmStartWorker(SMnodeMgmt *pMgmt) {
   SSingleWorkerCfg scCfg = {
       .min = 1,
       .max = 1,
-      .name = "mnode-sync-ctrl",
+      .name = "mnode-sync-rd",
       .fp = (FItem)mmProcessSyncMsg,
       .param = pMgmt,
   };
-  if (tSingleWorkerInit(&pMgmt->syncCtrlWorker, &scCfg) != 0) {
-    dError("failed to start mnode mnode-sync-ctrl worker since %s", terrstr());
+  if (tSingleWorkerInit(&pMgmt->syncRdWorker, &scCfg) != 0) {
+    dError("failed to start mnode mnode-sync-rd worker since %s", terrstr());
     return -1;
   }
 
@@ -259,6 +259,6 @@ void mmStopWorker(SMnodeMgmt *pMgmt) {
   tSingleWorkerCleanup(&pMgmt->readWorker);
   tSingleWorkerCleanup(&pMgmt->writeWorker);
   tSingleWorkerCleanup(&pMgmt->syncWorker);
-  tSingleWorkerCleanup(&pMgmt->syncCtrlWorker);
+  tSingleWorkerCleanup(&pMgmt->syncRdWorker);
   dDebug("mnode workers are closed");
 }

source/dnode/mgmt/mgmt_vnode/inc/vmInt.h

@@ -59,7 +59,7 @@ typedef struct {
   SVnode       *pImpl;
   SMultiWorker  pWriteW;
   SMultiWorker  pSyncW;
-  SMultiWorker  pSyncCtrlW;
+  SMultiWorker  pSyncRdW;
   SMultiWorker  pApplyW;
   STaosQueue   *pQueryQ;
   STaosQueue   *pStreamQ;
@@ -107,7 +107,7 @@ int32_t vmPutRpcMsgToQueue(SVnodeMgmt *pMgmt, EQueueType qtype, SRpcMsg *pRpc);
 
 int32_t vmPutMsgToWriteQueue(SVnodeMgmt *pMgmt, SRpcMsg *pMsg);
 int32_t vmPutMsgToSyncQueue(SVnodeMgmt *pMgmt, SRpcMsg *pMsg);
-int32_t vmPutMsgToSyncCtrlQueue(SVnodeMgmt *pMgmt, SRpcMsg *pMsg);
+int32_t vmPutMsgToSyncRdQueue(SVnodeMgmt *pMgmt, SRpcMsg *pMsg);
 int32_t vmPutMsgToQueryQueue(SVnodeMgmt *pMgmt, SRpcMsg *pMsg);
 int32_t vmPutMsgToFetchQueue(SVnodeMgmt *pMgmt, SRpcMsg *pMsg);
 int32_t vmPutMsgToStreamQueue(SVnodeMgmt *pMgmt, SRpcMsg *pMsg);

source/dnode/mgmt/mgmt_vnode/src/vmHandle.c

@@ -549,7 +549,8 @@ SArray *vmGetMsgHandles() {
   if (dmSetMgmtHandle(pArray, TDMT_DND_CREATE_VNODE, vmPutMsgToMgmtQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_DND_DROP_VNODE, vmPutMsgToMgmtQueue, 0) == NULL) goto _OVER;
 
-  if (dmSetMgmtHandle(pArray, TDMT_SYNC_TIMEOUT, vmPutMsgToSyncQueue, 0) == NULL) goto _OVER;
+  if (dmSetMgmtHandle(pArray, TDMT_SYNC_TIMEOUT_ELECTION, vmPutMsgToSyncQueue, 0) == NULL) goto _OVER;
+  if (dmSetMgmtHandle(pArray, TDMT_SYNC_HEARTBEAT, vmPutMsgToSyncRdQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_SYNC_CLIENT_REQUEST, vmPutMsgToSyncQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_SYNC_CLIENT_REQUEST_BATCH, vmPutMsgToSyncQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_SYNC_CLIENT_REQUEST_REPLY, vmPutMsgToSyncQueue, 0) == NULL) goto _OVER;
@@ -559,12 +560,12 @@ SArray *vmGetMsgHandles() {
   if (dmSetMgmtHandle(pArray, TDMT_SYNC_APPEND_ENTRIES_BATCH, vmPutMsgToSyncQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_SYNC_APPEND_ENTRIES_REPLY, vmPutMsgToSyncQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_SYNC_SNAPSHOT_SEND, vmPutMsgToSyncQueue, 0) == NULL) goto _OVER;
-  if (dmSetMgmtHandle(pArray, TDMT_SYNC_SNAPSHOT_RSP, vmPutMsgToSyncQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_SYNC_PRE_SNAPSHOT, vmPutMsgToSyncQueue, 0) == NULL) goto _OVER;
-  if (dmSetMgmtHandle(pArray, TDMT_SYNC_PRE_SNAPSHOT_REPLY, vmPutMsgToSyncQueue, 0) == NULL) goto _OVER;
 
-  if (dmSetMgmtHandle(pArray, TDMT_SYNC_HEARTBEAT, vmPutMsgToSyncCtrlQueue, 0) == NULL) goto _OVER;
-  if (dmSetMgmtHandle(pArray, TDMT_SYNC_HEARTBEAT_REPLY, vmPutMsgToSyncCtrlQueue, 0) == NULL) goto _OVER;
+  if (dmSetMgmtHandle(pArray, TDMT_SYNC_TIMEOUT, vmPutMsgToSyncRdQueue, 0) == NULL) goto _OVER;
+  if (dmSetMgmtHandle(pArray, TDMT_SYNC_HEARTBEAT_REPLY, vmPutMsgToSyncRdQueue, 0) == NULL) goto _OVER;
+  if (dmSetMgmtHandle(pArray, TDMT_SYNC_SNAPSHOT_RSP, vmPutMsgToSyncRdQueue, 0) == NULL) goto _OVER;
+  if (dmSetMgmtHandle(pArray, TDMT_SYNC_PRE_SNAPSHOT_REPLY, vmPutMsgToSyncRdQueue, 0) == NULL) goto _OVER;
 
   code = 0;
 

source/dnode/mgmt/mgmt_vnode/src/vmInt.c

@@ -98,9 +98,9 @@ void vmCloseVnode(SVnodeMgmt *pMgmt, SVnodeObj *pVnode, bool commitAndRemoveWal)
         pVnode->pSyncW.queue->threadId);
   tMultiWorkerCleanup(&pVnode->pSyncW);
 
-  dInfo("vgId:%d, wait for vnode sync ctrl queue:%p is empty, thread:%08" PRId64, pVnode->vgId,
-        pVnode->pSyncCtrlW.queue, pVnode->pSyncCtrlW.queue->threadId);
-  tMultiWorkerCleanup(&pVnode->pSyncCtrlW);
+  dInfo("vgId:%d, wait for vnode sync rd queue:%p is empty, thread:%08" PRId64, pVnode->vgId, pVnode->pSyncRdW.queue,
+        pVnode->pSyncRdW.queue->threadId);
+  tMultiWorkerCleanup(&pVnode->pSyncRdW);
 
   dInfo("vgId:%d, wait for vnode apply queue:%p is empty, thread:%08" PRId64, pVnode->vgId, pVnode->pApplyW.queue,
         pVnode->pApplyW.queue->threadId);

source/dnode/mgmt/mgmt_vnode/src/vmWorker.c

@@ -216,9 +216,9 @@ static int32_t vmPutMsgToQueue(SVnodeMgmt *pMgmt, SRpcMsg *pMsg, EQueueType qtyp
       dGTrace("vgId:%d, msg:%p put into vnode-sync queue", pVnode->vgId, pMsg);
       taosWriteQitem(pVnode->pSyncW.queue, pMsg);
       break;
-    case SYNC_CTRL_QUEUE:
-      dGTrace("vgId:%d, msg:%p put into vnode-sync-ctrl queue", pVnode->vgId, pMsg);
-      taosWriteQitem(pVnode->pSyncCtrlW.queue, pMsg);
+    case SYNC_RD_QUEUE:
+      dGTrace("vgId:%d, msg:%p put into vnode-sync-rd queue", pVnode->vgId, pMsg);
+      taosWriteQitem(pVnode->pSyncRdW.queue, pMsg);
       break;
     case APPLY_QUEUE:
       dGTrace("vgId:%d, msg:%p put into vnode-apply queue", pVnode->vgId, pMsg);
@@ -234,9 +234,7 @@ static int32_t vmPutMsgToQueue(SVnodeMgmt *pMgmt, SRpcMsg *pMsg, EQueueType qtyp
   return code;
 }
 
-int32_t vmPutMsgToSyncCtrlQueue(SVnodeMgmt *pMgmt, SRpcMsg *pMsg) {
-  return vmPutMsgToQueue(pMgmt, pMsg, SYNC_CTRL_QUEUE);
-}
+int32_t vmPutMsgToSyncRdQueue(SVnodeMgmt *pMgmt, SRpcMsg *pMsg) { return vmPutMsgToQueue(pMgmt, pMsg, SYNC_RD_QUEUE); }
 
 int32_t vmPutMsgToSyncQueue(SVnodeMgmt *pMgmt, SRpcMsg *pMsg) { return vmPutMsgToQueue(pMgmt, pMsg, SYNC_QUEUE); }
 
@@ -327,18 +325,18 @@ int32_t vmGetQueueSize(SVnodeMgmt *pMgmt, int32_t vgId, EQueueType qtype) {
 int32_t vmAllocQueue(SVnodeMgmt *pMgmt, SVnodeObj *pVnode) {
   SMultiWorkerCfg wcfg = {.max = 1, .name = "vnode-write", .fp = (FItems)vnodeProposeWriteMsg, .param = pVnode->pImpl};
   SMultiWorkerCfg scfg = {.max = 1, .name = "vnode-sync", .fp = (FItems)vmProcessSyncQueue, .param = pVnode};
-  SMultiWorkerCfg sccfg = {.max = 1, .name = "vnode-sync-ctrl", .fp = (FItems)vmProcessSyncQueue, .param = pVnode};
+  SMultiWorkerCfg sccfg = {.max = 1, .name = "vnode-sync-rd", .fp = (FItems)vmProcessSyncQueue, .param = pVnode};
   SMultiWorkerCfg acfg = {.max = 1, .name = "vnode-apply", .fp = (FItems)vnodeApplyWriteMsg, .param = pVnode->pImpl};
   (void)tMultiWorkerInit(&pVnode->pWriteW, &wcfg);
   (void)tMultiWorkerInit(&pVnode->pSyncW, &scfg);
-  (void)tMultiWorkerInit(&pVnode->pSyncCtrlW, &sccfg);
+  (void)tMultiWorkerInit(&pVnode->pSyncRdW, &sccfg);
   (void)tMultiWorkerInit(&pVnode->pApplyW, &acfg);
 
   pVnode->pQueryQ = tQWorkerAllocQueue(&pMgmt->queryPool, pVnode, (FItem)vmProcessQueryQueue);
   pVnode->pStreamQ = tAutoQWorkerAllocQueue(&pMgmt->streamPool, pVnode, (FItem)vmProcessStreamQueue);
   pVnode->pFetchQ = tWWorkerAllocQueue(&pMgmt->fetchPool, pVnode, (FItems)vmProcessFetchQueue);
 
-  if (pVnode->pWriteW.queue == NULL || pVnode->pSyncW.queue == NULL || pVnode->pSyncCtrlW.queue == NULL ||
+  if (pVnode->pWriteW.queue == NULL || pVnode->pSyncW.queue == NULL || pVnode->pSyncRdW.queue == NULL ||
       pVnode->pApplyW.queue == NULL || pVnode->pQueryQ == NULL || pVnode->pStreamQ == NULL || pVnode->pFetchQ == NULL) {
     terrno = TSDB_CODE_OUT_OF_MEMORY;
     return -1;
@@ -348,8 +346,8 @@ int32_t vmAllocQueue(SVnodeMgmt *pMgmt, SVnodeObj *pVnode) {
         pVnode->pWriteW.queue->threadId);
   dInfo("vgId:%d, sync-queue:%p is alloced, thread:%08" PRId64, pVnode->vgId, pVnode->pSyncW.queue,
         pVnode->pSyncW.queue->threadId);
-  dInfo("vgId:%d, sync-ctrl-queue:%p is alloced, thread:%08" PRId64, pVnode->vgId, pVnode->pSyncCtrlW.queue,
-        pVnode->pSyncCtrlW.queue->threadId);
+  dInfo("vgId:%d, sync-rd-queue:%p is alloced, thread:%08" PRId64, pVnode->vgId, pVnode->pSyncRdW.queue,
+        pVnode->pSyncRdW.queue->threadId);
   dInfo("vgId:%d, apply-queue:%p is alloced, thread:%08" PRId64, pVnode->vgId, pVnode->pApplyW.queue,
         pVnode->pApplyW.queue->threadId);
   dInfo("vgId:%d, query-queue:%p is alloced", pVnode->vgId, pVnode->pQueryQ);

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

@@ -24,10 +24,10 @@ extern "C" {
 
 enum {
   MQ_CONSUMER_STATUS__MODIFY = 1,
-  MQ_CONSUMER_STATUS__MODIFY_IN_REB,     // this value is not used anymore
+//  MQ_CONSUMER_STATUS__MODIFY_IN_REB,     // this value is not used anymore
   MQ_CONSUMER_STATUS__READY,
   MQ_CONSUMER_STATUS__LOST,
-  MQ_CONSUMER_STATUS__LOST_IN_REB,       // this value is not used anymore
+//  MQ_CONSUMER_STATUS__LOST_IN_REB,       // this value is not used anymore
   MQ_CONSUMER_STATUS__LOST_REBD,
   MQ_CONSUMER_STATUS__REMOVED,
 };

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

@@ -394,7 +394,7 @@ typedef struct {
 } SSmaObj;
 
 typedef struct {
-  char    name[TSDB_TABLE_FNAME_LEN];
+  char    name[TSDB_INDEX_FNAME_LEN];
   char    stb[TSDB_TABLE_FNAME_LEN];
   char    db[TSDB_DB_FNAME_LEN];
   char    dstTbName[TSDB_TABLE_FNAME_LEN];

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

@@ -247,6 +247,13 @@ static SMqRebInfo *mndGetOrCreateRebSub(SHashObj *pHash, const char *key) {
   return pRebInfo;
 }
 
+static void freeRebalanceItem(void* param) {
+  SMqRebInfo* pInfo = param;
+  taosArrayDestroy(pInfo->lostConsumers);
+  taosArrayDestroy(pInfo->newConsumers);
+  taosArrayDestroy(pInfo->removedConsumers);
+}
+
 static int32_t mndProcessMqTimerMsg(SRpcMsg *pMsg) {
   SMnode         *pMnode = pMsg->info.node;
   SSdb           *pSdb = pMnode->pSdb;
@@ -262,8 +269,21 @@ static int32_t mndProcessMqTimerMsg(SRpcMsg *pMsg) {
   }
 
   SMqDoRebalanceMsg *pRebMsg = rpcMallocCont(sizeof(SMqDoRebalanceMsg));
+  if (pRebMsg == NULL) {
+    mError("failed to create the rebalance msg, size:%d, quit mq timer", (int32_t) sizeof(SMqDoRebalanceMsg));
+    mndRebEnd();
+    return TSDB_CODE_OUT_OF_MEMORY;
+  }
+
   pRebMsg->rebSubHash = taosHashInit(64, MurmurHash3_32, true, HASH_NO_LOCK);
-  // TODO set cleanfp
+  if (pRebMsg->rebSubHash == NULL) {
+    mError("failed to create rebalance hashmap");
+    rpcFreeCont(pRebMsg);
+    mndRebEnd();
+    return TSDB_CODE_OUT_OF_MEMORY;
+  }
+
+  taosHashSetFreeFp(pRebMsg->rebSubHash, freeRebalanceItem);
 
   // iterate all consumers, find all modification
   while (1) {
@@ -317,7 +337,7 @@ static int32_t mndProcessMqTimerMsg(SRpcMsg *pMsg) {
         taosArrayPush(pRebSub->removedConsumers, &pConsumer->consumerId);
       }
       taosRUnLockLatch(&pConsumer->lock);
-    } else if (status == MQ_CONSUMER_STATUS__MODIFY || status == MQ_CONSUMER_STATUS__MODIFY_IN_REB) {
+    } else if (status == MQ_CONSUMER_STATUS__MODIFY) {
       taosRLockLatch(&pConsumer->lock);
 
       int32_t newTopicNum = taosArrayGetSize(pConsumer->rebNewTopics);
@@ -356,7 +376,7 @@ static int32_t mndProcessMqTimerMsg(SRpcMsg *pMsg) {
   } else {
     taosHashCleanup(pRebMsg->rebSubHash);
     rpcFreeCont(pRebMsg);
-    mDebug("mq rebalance finished, no modification");
+    mDebug("mq timer finished, no need to re-balance");
     mndRebEnd();
   }
   return 0;
@@ -852,6 +872,47 @@ static int32_t mndConsumerActionDelete(SSdb *pSdb, SMqConsumerObj *pConsumer) {
   return 0;
 }
 
+static void updateConsumerStatus(SMqConsumerObj* pConsumer) {
+  int32_t status = pConsumer->status;
+
+  if (taosArrayGetSize(pConsumer->rebNewTopics) == 0 && taosArrayGetSize(pConsumer->rebRemovedTopics) == 0) {
+    if (status == MQ_CONSUMER_STATUS__MODIFY) {
+      pConsumer->status = MQ_CONSUMER_STATUS__READY;
+    } else if (status == MQ_CONSUMER_STATUS__LOST) {
+      pConsumer->status = MQ_CONSUMER_STATUS__LOST_REBD;
+    }
+  }
+}
+
+// remove from new topic
+static void removeFromNewTopicList(SMqConsumerObj* pConsumer, const char* pTopic) {
+  int32_t size = taosArrayGetSize(pConsumer->rebNewTopics);
+  for (int32_t i = 0; i < taosArrayGetSize(pConsumer->rebNewTopics); i++) {
+    char *p = taosArrayGetP(pConsumer->rebNewTopics, i);
+    if (strcmp(pTopic, p) == 0) {
+      taosArrayRemove(pConsumer->rebNewTopics, i);
+      taosMemoryFree(p);
+
+      mDebug("consumer:0x%" PRIx64 " remove new topic:%s in the topic list, remain newTopics:%d", pConsumer->consumerId,
+             pTopic, (int) taosArrayGetSize(pConsumer->rebNewTopics));
+      break;
+    }
+  }
+}
+
+// remove from removed topic
+static void removeFromRemoveTopicList(SMqConsumerObj* pConsumer, const char* pTopic) {
+  int32_t size = taosArrayGetSize(pConsumer->rebRemovedTopics);
+  for (int32_t i = 0; i < size; i++) {
+    char *p = taosArrayGetP(pConsumer->rebRemovedTopics, i);
+    if (strcmp(pTopic, p) == 0) {
+      taosArrayRemove(pConsumer->rebRemovedTopics, i);
+      taosMemoryFree(p);
+      break;
+    }
+  }
+}
+
 static int32_t mndConsumerActionUpdate(SSdb *pSdb, SMqConsumerObj *pOldConsumer, SMqConsumerObj *pNewConsumer) {
   mDebug("consumer:0x%" PRIx64 " perform update action, update type:%d, subscribe-time:%" PRId64 ", uptime:%" PRId64,
          pOldConsumer->consumerId, pNewConsumer->updateType, pOldConsumer->subscribeTime, pOldConsumer->upTime);
@@ -862,6 +923,7 @@ static int32_t mndConsumerActionUpdate(SSdb *pSdb, SMqConsumerObj *pOldConsumer,
     /*A(taosArrayGetSize(pOldConsumer->rebNewTopics) == 0);*/
     /*A(taosArrayGetSize(pOldConsumer->rebRemovedTopics) == 0);*/
 
+    // this new consumer has identical topics with one existed consumers.
     if (taosArrayGetSize(pNewConsumer->rebNewTopics) == 0 && taosArrayGetSize(pNewConsumer->rebRemovedTopics) == 0) {
       pOldConsumer->status = MQ_CONSUMER_STATUS__READY;
     } else {
@@ -878,7 +940,6 @@ static int32_t mndConsumerActionUpdate(SSdb *pSdb, SMqConsumerObj *pOldConsumer,
       pNewConsumer->assignedTopics = tmp;
 
       pOldConsumer->subscribeTime = pNewConsumer->upTime;
-
       pOldConsumer->status = MQ_CONSUMER_STATUS__MODIFY;
     }
   } else if (pNewConsumer->updateType == CONSUMER_UPDATE__LOST) {
@@ -918,71 +979,48 @@ static int32_t mndConsumerActionUpdate(SSdb *pSdb, SMqConsumerObj *pOldConsumer,
     pOldConsumer->rebalanceTime = pNewConsumer->upTime;
 
   } else if (pNewConsumer->updateType == CONSUMER_UPDATE__ADD) {
-    /*A(taosArrayGetSize(pNewConsumer->rebNewTopics) == 1);*/
-    /*A(taosArrayGetSize(pNewConsumer->rebRemovedTopics) == 0);*/
+    ASSERT(taosArrayGetSize(pNewConsumer->rebNewTopics) == 1 && taosArrayGetSize(pNewConsumer->rebRemovedTopics) == 0);
+    char *pNewTopic = taosStrdup(taosArrayGetP(pNewConsumer->rebNewTopics, 0));
 
-    char *addedTopic = taosStrdup(taosArrayGetP(pNewConsumer->rebNewTopics, 0));
     // not exist in current topic
-
-    bool existing = false;
-#if 1
+    bool    existing = false;
     int32_t numOfExistedTopics = taosArrayGetSize(pOldConsumer->currentTopics);
     for (int32_t i = 0; i < numOfExistedTopics; i++) {
       char *topic = taosArrayGetP(pOldConsumer->currentTopics, i);
-      if (strcmp(topic, addedTopic) == 0) {
+      if (strcmp(topic, pNewTopic) == 0) {
         existing = true;
       }
     }
-#endif
 
-    // remove from new topic
-    for (int32_t i = 0; i < taosArrayGetSize(pOldConsumer->rebNewTopics); i++) {
-      char *topic = taosArrayGetP(pOldConsumer->rebNewTopics, i);
-      if (strcmp(addedTopic, topic) == 0) {
-        taosArrayRemove(pOldConsumer->rebNewTopics, i);
-        taosMemoryFree(topic);
-        break;
-      }
-    }
+    removeFromNewTopicList(pOldConsumer, pNewTopic);
 
     // add to current topic
     if (!existing) {
-      taosArrayPush(pOldConsumer->currentTopics, &addedTopic);
+      taosArrayPush(pOldConsumer->currentTopics, &pNewTopic);
       taosArraySort(pOldConsumer->currentTopics, taosArrayCompareString);
     } else {
-      taosMemoryFree(addedTopic);
+      taosMemoryFree(pNewTopic);
     }
 
     // set status
     int32_t status = pOldConsumer->status;
-    if (taosArrayGetSize(pOldConsumer->rebNewTopics) == 0 && taosArrayGetSize(pOldConsumer->rebRemovedTopics) == 0) {
-      if (status == MQ_CONSUMER_STATUS__MODIFY || status == MQ_CONSUMER_STATUS__MODIFY_IN_REB) {
-        pOldConsumer->status = MQ_CONSUMER_STATUS__READY;
-      } else if (status == MQ_CONSUMER_STATUS__LOST_IN_REB || status == MQ_CONSUMER_STATUS__LOST) {
-        pOldConsumer->status = MQ_CONSUMER_STATUS__LOST_REBD;
-      }
-    } else {
-      if (status == MQ_CONSUMER_STATUS__MODIFY || status == MQ_CONSUMER_STATUS__MODIFY_IN_REB) {
-        pOldConsumer->status = MQ_CONSUMER_STATUS__MODIFY_IN_REB;
-      } else if (status == MQ_CONSUMER_STATUS__LOST || status == MQ_CONSUMER_STATUS__LOST_IN_REB) {
-        pOldConsumer->status = MQ_CONSUMER_STATUS__LOST_IN_REB;
-      }
-    }
+    updateConsumerStatus(pOldConsumer);
 
     // the re-balance is triggered when the new consumer is launched.
     pOldConsumer->rebalanceTime = pNewConsumer->upTime;
 
     atomic_add_fetch_32(&pOldConsumer->epoch, 1);
-    mDebug("consumer:0x%" PRIx64 " state (%d)%s -> (%d)%s, new epoch:%d, reb-time:%" PRId64 ", current topics:%d",
+    mDebug("consumer:0x%" PRIx64 " state (%d)%s -> (%d)%s, new epoch:%d, reb-time:%" PRId64
+           ", current topics:%d, newTopics:%d, removeTopics:%d",
            pOldConsumer->consumerId, status, mndConsumerStatusName(status), pOldConsumer->status,
-           mndConsumerStatusName(pOldConsumer->status),
-           pOldConsumer->epoch, pOldConsumer->rebalanceTime, (int)taosArrayGetSize(pOldConsumer->currentTopics));
+           mndConsumerStatusName(pOldConsumer->status), pOldConsumer->epoch, pOldConsumer->rebalanceTime,
+           (int)taosArrayGetSize(pOldConsumer->currentTopics), (int)taosArrayGetSize(pOldConsumer->rebNewTopics),
+           (int)taosArrayGetSize(pOldConsumer->rebRemovedTopics));
+
   } else if (pNewConsumer->updateType == CONSUMER_UPDATE__REMOVE) {
     /*A(taosArrayGetSize(pNewConsumer->rebNewTopics) == 0);*/
     /*A(taosArrayGetSize(pNewConsumer->rebRemovedTopics) == 1);*/
     char *removedTopic = taosArrayGetP(pNewConsumer->rebRemovedTopics, 0);
-
-    // not exist in new topic
 #if 0
     for (int32_t i = 0; i < taosArrayGetSize(pOldConsumer->rebNewTopics); i++) {
       char *topic = taosArrayGetP(pOldConsumer->rebNewTopics, i);
@@ -991,14 +1029,7 @@ static int32_t mndConsumerActionUpdate(SSdb *pSdb, SMqConsumerObj *pOldConsumer,
 #endif
 
     // remove from removed topic
-    for (int32_t i = 0; i < taosArrayGetSize(pOldConsumer->rebRemovedTopics); i++) {
-      char *topic = taosArrayGetP(pOldConsumer->rebRemovedTopics, i);
-      if (strcmp(removedTopic, topic) == 0) {
-        taosArrayRemove(pOldConsumer->rebRemovedTopics, i);
-        taosMemoryFree(topic);
-        break;
-      }
-    }
+    removeFromRemoveTopicList(pOldConsumer, removedTopic);
 
     // remove from current topic
     int32_t i = 0;
@@ -1011,32 +1042,20 @@ static int32_t mndConsumerActionUpdate(SSdb *pSdb, SMqConsumerObj *pOldConsumer,
         break;
       }
     }
-    // must find the topic
-    /*A(i < sz);*/
 
     // set status
     int32_t status = pOldConsumer->status;
-    if (taosArrayGetSize(pOldConsumer->rebNewTopics) == 0 && taosArrayGetSize(pOldConsumer->rebRemovedTopics) == 0) {
-      if (status == MQ_CONSUMER_STATUS__MODIFY || status == MQ_CONSUMER_STATUS__MODIFY_IN_REB) {
-        pOldConsumer->status = MQ_CONSUMER_STATUS__READY;
-      } else if (status == MQ_CONSUMER_STATUS__LOST_IN_REB || status == MQ_CONSUMER_STATUS__LOST) {
-        pOldConsumer->status = MQ_CONSUMER_STATUS__LOST_REBD;
-      }
-    } else {
-      if (status == MQ_CONSUMER_STATUS__MODIFY || status == MQ_CONSUMER_STATUS__MODIFY_IN_REB) {
-        pOldConsumer->status = MQ_CONSUMER_STATUS__MODIFY_IN_REB;
-      } else if (status == MQ_CONSUMER_STATUS__LOST || status == MQ_CONSUMER_STATUS__LOST_IN_REB) {
-        pOldConsumer->status = MQ_CONSUMER_STATUS__LOST_IN_REB;
-      }
-    }
+    updateConsumerStatus(pOldConsumer);
 
     pOldConsumer->rebalanceTime = pNewConsumer->upTime;
     atomic_add_fetch_32(&pOldConsumer->epoch, 1);
 
-    mDebug("consumer:0x%" PRIx64 " state %d(%s) -> %d(%s), new epoch:%d, reb-time:%" PRId64 ", current topics:%d",
+    mDebug("consumer:0x%" PRIx64 " state (%d)%s -> (%d)%s, new epoch:%d, reb-time:%" PRId64
+               ", current topics:%d, newTopics:%d, removeTopics:%d",
            pOldConsumer->consumerId, status, mndConsumerStatusName(status), pOldConsumer->status,
-           mndConsumerStatusName(pOldConsumer->status),
-           pOldConsumer->epoch, pOldConsumer->rebalanceTime, (int)taosArrayGetSize(pOldConsumer->currentTopics));
+           mndConsumerStatusName(pOldConsumer->status), pOldConsumer->epoch, pOldConsumer->rebalanceTime,
+           (int)taosArrayGetSize(pOldConsumer->currentTopics), (int)taosArrayGetSize(pOldConsumer->rebNewTopics),
+           (int)taosArrayGetSize(pOldConsumer->rebRemovedTopics));
   }
 
   taosWUnLockLatch(&pOldConsumer->lock);
@@ -1170,10 +1189,8 @@ static const char *mndConsumerStatusName(int status) {
       return "ready";
     case MQ_CONSUMER_STATUS__LOST:
     case MQ_CONSUMER_STATUS__LOST_REBD:
-    case MQ_CONSUMER_STATUS__LOST_IN_REB:
       return "lost";
     case MQ_CONSUMER_STATUS__MODIFY:
-    case MQ_CONSUMER_STATUS__MODIFY_IN_REB:
       return "rebalancing";
     default:
       return "unknown";

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

@@ -138,7 +138,7 @@ static void *mndBuildDropIdxReq(SMnode *pMnode, SVgObj *pVgroup, SStbObj *pStbOb
   mInfo("idx: %s start to build drop index req", pIdx->name);
 
   len = tSerializeSDropIdxReq(NULL, 0, &req);
-  if (ret < 0) {
+  if (len < 0) {
     goto _err;
   }
 
@@ -672,7 +672,7 @@ _OVER:
 static int32_t mndAddIndex(SMnode *pMnode, SRpcMsg *pReq, SCreateTagIndexReq *req, SDbObj *pDb, SStbObj *pStb) {
   int32_t code = -1;
   SIdxObj idxObj = {0};
-  memcpy(idxObj.name, req->idxName, TSDB_TABLE_FNAME_LEN);
+  memcpy(idxObj.name, req->idxName, TSDB_INDEX_FNAME_LEN);
   memcpy(idxObj.stb, pStb->name, TSDB_TABLE_FNAME_LEN);
   memcpy(idxObj.db, pDb->name, TSDB_DB_FNAME_LEN);
   memcpy(idxObj.colName, req->colName, TSDB_COL_NAME_LEN);

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

@@ -226,7 +226,7 @@ static int32_t mndProcessConnectReq(SRpcMsg *pReq) {
     goto _OVER;
   }
 
-  if ((code = taosCheckVersionCompatibleFromStr(connReq.sVer, version, 2)) != 0) {
+  if ((code = taosCheckVersionCompatibleFromStr(connReq.sVer, version, 3)) != 0) {
     terrno = code;
     goto _OVER;
   }

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

@@ -197,24 +197,20 @@ static SMqRebInfo *mndGetOrCreateRebSub(SHashObj *pHash, const char *key) {
   return pRebSub;
 }
 
-static int32_t mndDoRebalance(SMnode *pMnode, const SMqRebInputObj *pInput, SMqRebOutputObj *pOutput) {
-  int32_t     totalVgNum = pOutput->pSub->vgNum;
-  const char *sub = pOutput->pSub->key;
-  mInfo("sub:%s mq re-balance %d vgroups", sub, pOutput->pSub->vgNum);
-
-  // 1. build temporary hash(vgId -> SMqRebOutputVg) to store modified vg
-  SHashObj *pHash = taosHashInit(64, taosGetDefaultHashFunction(TSDB_DATA_TYPE_INT), false, HASH_NO_LOCK);
+static void doRemoveExistedConsumers(SMqRebOutputObj *pOutput, SHashObj *pHash, const SMqRebInputObj *pInput) {
+  int32_t     numOfRemoved = taosArrayGetSize(pInput->pRebInfo->removedConsumers);
+  const char *pSubKey = pOutput->pSub->key;
 
-  // 2. check and get actual removed consumers, put their vg into hash
-  int32_t removedNum = taosArrayGetSize(pInput->pRebInfo->removedConsumers);
   int32_t actualRemoved = 0;
-  for (int32_t i = 0; i < removedNum; i++) {
+  for (int32_t i = 0; i < numOfRemoved; i++) {
     uint64_t consumerId = *(uint64_t *)taosArrayGet(pInput->pRebInfo->removedConsumers, i);
 
     SMqConsumerEp *pConsumerEp = taosHashGet(pOutput->pSub->consumerHash, &consumerId, sizeof(int64_t));
 
+    // consumer exists till now
     if (pConsumerEp) {
       actualRemoved++;
+
       int32_t consumerVgNum = taosArrayGetSize(pConsumerEp->vgs);
       for (int32_t j = 0; j < consumerVgNum; j++) {
         SMqVgEp       *pVgEp = taosArrayGetP(pConsumerEp->vgs, j);
@@ -223,52 +219,66 @@ static int32_t mndDoRebalance(SMnode *pMnode, const SMqRebInputObj *pInput, SMqR
             .newConsumerId = -1,
             .pVgEp = pVgEp,
         };
+
         taosHashPut(pHash, &pVgEp->vgId, sizeof(int32_t), &outputVg, sizeof(SMqRebOutputVg));
-        mInfo("sub:%s mq re-balance remove vgId:%d from consumer:0x%" PRIx64, sub, pVgEp->vgId, consumerId);
+        mInfo("sub:%s mq re-balance remove vgId:%d from consumer:%" PRIx64, pSubKey, pVgEp->vgId, consumerId);
       }
+
       taosArrayDestroy(pConsumerEp->vgs);
       taosHashRemove(pOutput->pSub->consumerHash, &consumerId, sizeof(int64_t));
+
       // put into removed
       taosArrayPush(pOutput->removedConsumers, &consumerId);
     }
   }
 
-  if (removedNum != actualRemoved) {
-    mError("sub:%s mq re-balance removedNum:%d not matched with actual:%d", sub, removedNum, actualRemoved);
+  if (numOfRemoved != actualRemoved) {
+    mError("sub:%s mq re-balance removedNum:%d not matched with actual:%d", pSubKey, numOfRemoved, actualRemoved);
+  } else {
+    mInfo("sub:%s removed %d consumers", pSubKey, numOfRemoved);
   }
+}
 
-  // if previously no consumer, there are vgs not assigned
-  {
-    int32_t consumerVgNum = taosArrayGetSize(pOutput->pSub->unassignedVgs);
-    for (int32_t i = 0; i < consumerVgNum; i++) {
-      SMqVgEp       *pVgEp = *(SMqVgEp **)taosArrayPop(pOutput->pSub->unassignedVgs);
-      SMqRebOutputVg rebOutput = {
-          .oldConsumerId = -1,
-          .newConsumerId = -1,
-          .pVgEp = pVgEp,
-      };
-      taosHashPut(pHash, &pVgEp->vgId, sizeof(int32_t), &rebOutput, sizeof(SMqRebOutputVg));
-      mInfo("sub:%s mq re-balance remove vgId:%d from unassigned", sub, pVgEp->vgId);
-    }
+static void doAddNewConsumers(SMqRebOutputObj *pOutput, const SMqRebInputObj *pInput) {
+  int32_t numOfNewConsumers = taosArrayGetSize(pInput->pRebInfo->newConsumers);
+  const char *pSubKey = pOutput->pSub->key;
+
+  for (int32_t i = 0; i < numOfNewConsumers; i++) {
+    int64_t consumerId = *(int64_t *)taosArrayGet(pInput->pRebInfo->newConsumers, i);
+
+    SMqConsumerEp newConsumerEp;
+    newConsumerEp.consumerId = consumerId;
+    newConsumerEp.vgs = taosArrayInit(0, sizeof(void *));
+
+    taosHashPut(pOutput->pSub->consumerHash, &consumerId, sizeof(int64_t), &newConsumerEp, sizeof(SMqConsumerEp));
+    taosArrayPush(pOutput->newConsumers, &consumerId);
+    mInfo("sub:%s mq rebalance add new consumer:%" PRIx64, pSubKey, consumerId);
   }
+}
 
-  // 3. calc vg number of each consumer
-  int32_t afterRebConsumerNum = pInput->oldConsumerNum + taosArrayGetSize(pInput->pRebInfo->newConsumers) -
-                                taosArrayGetSize(pInput->pRebInfo->removedConsumers);
-  int32_t minVgCnt = 0;
-  int32_t imbConsumerNum = 0;
-  // calc num
-  if (afterRebConsumerNum) {
-    minVgCnt = totalVgNum / afterRebConsumerNum;
-    imbConsumerNum = totalVgNum % afterRebConsumerNum;
+static void addUnassignedVgroups(SMqRebOutputObj *pOutput, SHashObj* pHash) {
+  const char *pSubKey = pOutput->pSub->key;
+  int32_t numOfVgroups = taosArrayGetSize(pOutput->pSub->unassignedVgs);
+
+  for (int32_t i = 0; i < numOfVgroups; i++) {
+    SMqVgEp       *pVgEp = *(SMqVgEp **)taosArrayPop(pOutput->pSub->unassignedVgs);
+    SMqRebOutputVg rebOutput = {
+        .oldConsumerId = -1,
+        .newConsumerId = -1,
+        .pVgEp = pVgEp,
+    };
+
+    taosHashPut(pHash, &pVgEp->vgId, sizeof(int32_t), &rebOutput, sizeof(SMqRebOutputVg));
+    mInfo("sub:%s mq re-balance remove vgId:%d from unassigned", pSubKey, pVgEp->vgId);
   }
+}
 
-  mInfo("sub:%s mq re-balance %d consumers: at least %d vgs each, %d consumers has more vgs", sub,
-        afterRebConsumerNum, minVgCnt, imbConsumerNum);
+static void transferVgroupsForConsumers(SMqRebOutputObj *pOutput, SHashObj* pHash, int32_t minVgCnt, int32_t imbConsumerNum) {
+  const char *pSubKey = pOutput->pSub->key;
 
-  // 4. first scan: remove consumer more than wanted, put to remove hash
   int32_t imbCnt = 0;
   void   *pIter = NULL;
+
   while (1) {
     pIter = taosHashIterate(pOutput->pSub->consumerHash, pIter);
     if (pIter == NULL) {
@@ -276,8 +286,8 @@ static int32_t mndDoRebalance(SMnode *pMnode, const SMqRebInputObj *pInput, SMqR
     }
 
     SMqConsumerEp *pConsumerEp = (SMqConsumerEp *)pIter;
+    int32_t        consumerVgNum = taosArrayGetSize(pConsumerEp->vgs);
 
-    int32_t consumerVgNum = taosArrayGetSize(pConsumerEp->vgs);
     // all old consumers still existing are touched
     // TODO optimize: touch only consumer whose vgs changed
     taosArrayPush(pOutput->touchedConsumers, &pConsumerEp->consumerId);
@@ -296,13 +306,13 @@ static int32_t mndDoRebalance(SMnode *pMnode, const SMqRebInputObj *pInput, SMqR
                 .pVgEp = pVgEp,
             };
             taosHashPut(pHash, &pVgEp->vgId, sizeof(int32_t), &outputVg, sizeof(SMqRebOutputVg));
-            mInfo("sub:%s mq rebalance remove vgId:%d from consumer:0x%" PRIx64 ",(first scan)", sub, pVgEp->vgId,
+            mInfo("sub:%s mq rebalance remove vgId:%d from consumer:0x%" PRIx64 ",(first scan)", pSubKey, pVgEp->vgId,
                   pConsumerEp->consumerId);
           }
           imbCnt++;
         }
       } else {
-        // pop until equal minVg
+        // all the remain consumers should only have the number of vgroups, which is equalled to the value of minVg
         while (taosArrayGetSize(pConsumerEp->vgs) > minVgCnt) {
           SMqVgEp       *pVgEp = *(SMqVgEp **)taosArrayPop(pConsumerEp->vgs);
           SMqRebOutputVg outputVg = {
@@ -311,36 +321,67 @@ static int32_t mndDoRebalance(SMnode *pMnode, const SMqRebInputObj *pInput, SMqR
               .pVgEp = pVgEp,
           };
           taosHashPut(pHash, &pVgEp->vgId, sizeof(int32_t), &outputVg, sizeof(SMqRebOutputVg));
-          mInfo("sub:%s mq rebalance remove vgId:%d from consumer:0x%" PRIx64 ",(first scan)", sub, pVgEp->vgId,
+          mInfo("sub:%s mq rebalance remove vgId:%d from consumer:0x%" PRIx64 ",(first scan)", pSubKey, pVgEp->vgId,
                 pConsumerEp->consumerId);
         }
       }
     }
   }
+}
 
-  // 5. add new consumer into sub
-  {
-    int32_t consumerNum = taosArrayGetSize(pInput->pRebInfo->newConsumers);
-    for (int32_t i = 0; i < consumerNum; i++) {
-      int64_t consumerId = *(int64_t *)taosArrayGet(pInput->pRebInfo->newConsumers, i);
-
-      SMqConsumerEp newConsumerEp;
-      newConsumerEp.consumerId = consumerId;
-      newConsumerEp.vgs = taosArrayInit(0, sizeof(void *));
-      taosHashPut(pOutput->pSub->consumerHash, &consumerId, sizeof(int64_t), &newConsumerEp, sizeof(SMqConsumerEp));
-      taosArrayPush(pOutput->newConsumers, &consumerId);
-      mInfo("sub:%s mq rebalance add new consumer:0x%" PRIx64, sub, consumerId);
-    }
+static int32_t mndDoRebalance(SMnode *pMnode, const SMqRebInputObj *pInput, SMqRebOutputObj *pOutput) {
+  int32_t     totalVgNum = pOutput->pSub->vgNum;
+  const char *pSubKey = pOutput->pSub->key;
+
+  int32_t numOfRemoved = taosArrayGetSize(pInput->pRebInfo->removedConsumers);
+  int32_t numOfAdded = taosArrayGetSize(pInput->pRebInfo->newConsumers);
+  mInfo("sub:%s mq re-balance %d vgroups, existed consumers:%d, added:%d, removed:%d", pSubKey, totalVgNum,
+        pInput->oldConsumerNum, numOfAdded, numOfRemoved);
+
+  // 1. build temporary hash(vgId -> SMqRebOutputVg) to store modified vg
+  SHashObj *pHash = taosHashInit(64, taosGetDefaultHashFunction(TSDB_DATA_TYPE_INT), false, HASH_NO_LOCK);
+
+  // 2. check and get actual removed consumers, put their vg into hash
+  doRemoveExistedConsumers(pOutput, pHash, pInput);
+
+  // 3. if previously no consumer, there are vgs not assigned
+  addUnassignedVgroups(pOutput, pHash);
+
+  // 4. calc vg number of each consumer
+  int32_t numOfFinal = pInput->oldConsumerNum + numOfAdded - numOfRemoved;
+
+  int32_t minVgCnt = 0;
+  int32_t imbConsumerNum = 0;
+
+  // calc num
+  if (numOfFinal) {
+    minVgCnt = totalVgNum / numOfFinal;
+    imbConsumerNum = totalVgNum % numOfFinal;
+    mInfo("sub:%s mq re-balance %d consumers: at least %d vgs each, %d consumers has 1 more vgroups than avg value",
+          pSubKey, numOfFinal, minVgCnt, imbConsumerNum);
+  } else {
+    mInfo("sub:%s no consumer subscribe this topic", pSubKey);
   }
 
-  // 6. second scan: find consumer do not have enough vg, extract from temporary hash and assign to new consumer.
+  // 5. first scan: remove vgroups from te consumers, who have more vgroups than the threashold value that is
+  // minVgCnt, and then put them into the recycled hash list
+  transferVgroupsForConsumers(pOutput, pHash, minVgCnt, imbConsumerNum);
+
+  // 6. add new consumer into sub
+  doAddNewConsumers(pOutput, pInput);
+
+  // 7. second scan: find consumer do not have enough vgroups, extract from temporary hash and assign to them
   // All related vg should be put into rebVgs
   SMqRebOutputVg *pRebVg = NULL;
   void           *pRemovedIter = NULL;
-  pIter = NULL;
+  void           *pIter = NULL;
+
   while (1) {
     pIter = taosHashIterate(pOutput->pSub->consumerHash, pIter);
-    if (pIter == NULL) break;
+    if (pIter == NULL) {
+      break;
+    }
+
     SMqConsumerEp *pConsumerEp = (SMqConsumerEp *)pIter;
 
     // push until equal minVg
@@ -348,8 +389,8 @@ static int32_t mndDoRebalance(SMnode *pMnode, const SMqRebInputObj *pInput, SMqR
       // iter hash and find one vg
       pRemovedIter = taosHashIterate(pHash, pRemovedIter);
       if (pRemovedIter == NULL) {
-        mError("sub:%s removed iter is null", sub);
-        continue;
+        mError("sub:%s removed iter is null", pSubKey);
+        break;
       }
 
       pRebVg = (SMqRebOutputVg *)pRemovedIter;
@@ -409,15 +450,15 @@ static int32_t mndDoRebalance(SMnode *pMnode, const SMqRebInputObj *pInput, SMqR
 
       taosArrayPush(pOutput->pSub->unassignedVgs, &pRebOutput->pVgEp);
       taosArrayPush(pOutput->rebVgs, pRebOutput);
-      mInfo("sub:%s mq re-balance unassign vgId:%d (second scan)", sub, pRebOutput->pVgEp->vgId);
+      mInfo("sub:%s mq re-balance unassign vgId:%d (second scan)", pSubKey, pRebOutput->pVgEp->vgId);
     }
   }
 
   // 8. generate logs
-  mInfo("sub:%s mq re-balance calculation completed, re-balanced vg", sub);
+  mInfo("sub:%s mq re-balance calculation completed, re-balanced vg", pSubKey);
   for (int32_t i = 0; i < taosArrayGetSize(pOutput->rebVgs); i++) {
     SMqRebOutputVg *pOutputRebVg = taosArrayGet(pOutput->rebVgs, i);
-    mInfo("sub:%s mq re-balance vgId:%d, moved from consumer:0x%" PRIx64 ", to consumer:0x%" PRIx64, sub,
+    mInfo("sub:%s mq re-balance vgId:%d, moved from consumer:0x%" PRIx64 ", to consumer:0x%" PRIx64, pSubKey,
           pOutputRebVg->pVgEp->vgId, pOutputRebVg->oldConsumerId, pOutputRebVg->newConsumerId);
   }
   {
@@ -427,10 +468,10 @@ static int32_t mndDoRebalance(SMnode *pMnode, const SMqRebInputObj *pInput, SMqR
       if (pIter == NULL) break;
       SMqConsumerEp *pConsumerEp = (SMqConsumerEp *)pIter;
       int32_t        sz = taosArrayGetSize(pConsumerEp->vgs);
-      mInfo("sub:%s mq re-balance final cfg: consumer:0x%" PRIx64 " has %d vg", sub, pConsumerEp->consumerId, sz);
+      mInfo("sub:%s mq re-balance final cfg: consumer:0x%" PRIx64 " has %d vg", pSubKey, pConsumerEp->consumerId, sz);
       for (int32_t i = 0; i < sz; i++) {
         SMqVgEp *pVgEp = taosArrayGetP(pConsumerEp->vgs, i);
-        mInfo("sub:%s mq re-balance final cfg: vg %d to consumer:0x%" PRIx64, sub, pVgEp->vgId,
+        mInfo("sub:%s mq re-balance final cfg: vg %d to consumer:0x%" PRIx64, pSubKey, pVgEp->vgId,
               pConsumerEp->consumerId);
       }
     }
@@ -555,17 +596,23 @@ static int32_t mndProcessRebalanceReq(SRpcMsg *pMsg) {
   SMnode            *pMnode = pMsg->info.node;
   SMqDoRebalanceMsg *pReq = pMsg->pCont;
   void              *pIter = NULL;
+  bool               rebalanceOnce = false;  // to ensure only once.
 
-  mInfo("mq re-balance start");
+  mInfo("mq re-balance start, total required re-balanced trans:%d", taosHashGetSize(pReq->rebSubHash));
 
+  // here we only handle one topic rebalance requirement to ensure the atomic execution of this transaction.
   while (1) {
+    if (rebalanceOnce) {
+      break;
+    }
+
     pIter = taosHashIterate(pReq->rebSubHash, pIter);
     if (pIter == NULL) {
       break;
     }
 
+    // todo handle the malloc failure
     SMqRebInputObj rebInput = {0};
-
     SMqRebOutputObj rebOutput = {0};
     rebOutput.newConsumers = taosArrayInit(0, sizeof(int64_t));
     rebOutput.removedConsumers = taosArrayInit(0, sizeof(int64_t));
@@ -582,9 +629,10 @@ static int32_t mndProcessRebalanceReq(SRpcMsg *pMsg) {
       char topic[TSDB_TOPIC_FNAME_LEN];
       char cgroup[TSDB_CGROUP_LEN];
       mndSplitSubscribeKey(pRebInfo->key, topic, cgroup, true);
+
       SMqTopicObj *pTopic = mndAcquireTopic(pMnode, topic);
       if (pTopic == NULL) {
-        mError("mq re-balance %s ignored since topic %s not exist", pRebInfo->key, topic);
+        mError("mq re-balance %s ignored since topic %s doesn't exist", pRebInfo->key, topic);
         continue;
       }
 
@@ -604,11 +652,13 @@ static int32_t mndProcessRebalanceReq(SRpcMsg *pMsg) {
       mndReleaseTopic(pMnode, pTopic);
 
       rebInput.oldConsumerNum = 0;
+      mInfo("topic:%s has no consumers sub yet", topic);
     } else {
       taosRLockLatch(&pSub->lock);
       rebInput.oldConsumerNum = taosHashGetSize(pSub->consumerHash);
       rebOutput.pSub = tCloneSubscribeObj(pSub);
       taosRUnLockLatch(&pSub->lock);
+      mInfo("topic:%s has %d consumers sub till now", pRebInfo->key, rebInput.oldConsumerNum);
       mndReleaseSubscribe(pMnode, pSub);
     }
 
@@ -623,16 +673,14 @@ static int32_t mndProcessRebalanceReq(SRpcMsg *pMsg) {
       mError("mq re-balance persist output error, possibly vnode splitted or dropped");
     }
 
-    taosArrayDestroy(pRebInfo->lostConsumers);
-    taosArrayDestroy(pRebInfo->newConsumers);
-    taosArrayDestroy(pRebInfo->removedConsumers);
-
     taosArrayDestroy(rebOutput.newConsumers);
     taosArrayDestroy(rebOutput.touchedConsumers);
     taosArrayDestroy(rebOutput.removedConsumers);
     taosArrayDestroy(rebOutput.rebVgs);
     tDeleteSubscribeObj(rebOutput.pSub);
     taosMemoryFree(rebOutput.pSub);
+
+    rebalanceOnce = true;
   }
 
   // reset flag

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

@@ -33,7 +33,7 @@ static int32_t mndSyncEqCtrlMsg(const SMsgCb *msgcb, SRpcMsg *pMsg) {
     return -1;
   }
 
-  int32_t code = tmsgPutToQueue(msgcb, SYNC_CTRL_QUEUE, pMsg);
+  int32_t code = tmsgPutToQueue(msgcb, SYNC_RD_QUEUE, pMsg);
   if (code != 0) {
     rpcFreeCont(pMsg->pCont);
     pMsg->pCont = NULL;

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

@@ -1091,10 +1091,13 @@ int32_t mndTransProcessRsp(SRpcMsg *pRsp) {
     pAction->msgReceived = 1;
     pAction->errCode = pRsp->code;
     pTrans->lastErrorNo = pRsp->code;
+
+    mInfo("trans:%d, %s:%d response is received, code:0x%x, accept:0x%x retry:0x%x", transId,
+          mndTransStr(pAction->stage), action, pRsp->code, pAction->acceptableCode, pAction->retryCode);
+  } else {
+    mInfo("trans:%d, invalid action, index:%d, code:0x%x", transId, action, pRsp->code);
   }
 
-  mInfo("trans:%d, %s:%d response is received, code:0x%x, accept:0x%x retry:0x%x", transId, mndTransStr(pAction->stage),
-        action, pRsp->code, pAction->acceptableCode, pAction->retryCode);
   mndTransExecute(pMnode, pTrans, true);
 
 _OVER:

source/dnode/vnode/CMakeLists.txt

@@ -57,7 +57,7 @@ target_sources(
 
     # tq
     "src/tq/tq.c"
-    "src/tq/tqExec.c"
+    "src/tq/tqScan.c"
     "src/tq/tqMeta.c"
     "src/tq/tqRead.c"
     "src/tq/tqOffset.c"

source/dnode/vnode/inc/vnode.h

@@ -178,11 +178,11 @@ typedef struct STsdbReader STsdbReader;
 
 int32_t tsdbSetTableList(STsdbReader *pReader, const void *pTableList, int32_t num);
 int32_t tsdbReaderOpen(SVnode *pVnode, SQueryTableDataCond *pCond, void *pTableList, int32_t numOfTables,
-                       SSDataBlock *pResBlock, STsdbReader **ppReader, const char *idstr);
+                       SSDataBlock *pResBlock, STsdbReader **ppReader, const char *idstr, bool countOnly);
 
 void         tsdbReaderSetId(STsdbReader* pReader, const char* idstr);
 void         tsdbReaderClose(STsdbReader *pReader);
-bool         tsdbNextDataBlock(STsdbReader *pReader);
+int32_t      tsdbNextDataBlock(STsdbReader *pReader, bool *hasNext);
 int32_t      tsdbRetrieveDatablockSMA(STsdbReader *pReader, SSDataBlock *pDataBlock, bool *allHave);
 void         tsdbReleaseDataBlock(STsdbReader *pReader);
 SSDataBlock *tsdbRetrieveDataBlock(STsdbReader *pTsdbReadHandle, SArray *pColumnIdList);
@@ -228,19 +228,12 @@ typedef struct SSnapContext {
   SArray   *idList;
   int32_t   index;
   bool      withMeta;
-  bool      queryMetaOrData;  // true-get meta, false-get data
+  bool      queryMeta;  // true-get meta, false-get data
 } SSnapContext;
 
 typedef struct STqReader {
-  // const SSubmitReq *pMsg;
-  //   SSubmitBlk       *pBlock;
-  //   SSubmitMsgIter    msgIter;
-  //   SSubmitBlkIter    blkIter;
-
-  int64_t     ver;
   SPackedData msg2;
 
-  int8_t      setMsg;
   SSubmitReq2 submit;
   int32_t     nextBlk;
 
@@ -267,7 +260,7 @@ int32_t tqReaderAddTbUidList(STqReader *pReader, const SArray *tbUidList);
 int32_t tqReaderRemoveTbUidList(STqReader *pReader, const SArray *tbUidList);
 
 int32_t tqSeekVer(STqReader *pReader, int64_t ver, const char *id);
-int32_t tqNextBlock(STqReader *pReader, SFetchRet *ret);
+void tqNextBlock(STqReader *pReader, SFetchRet *ret);
 
 int32_t tqReaderSetSubmitReq2(STqReader *pReader, void *msgStr, int32_t msgLen, int64_t ver);
 // int32_t tqReaderSetDataMsg(STqReader *pReader, const SSubmitReq *pMsg, int64_t ver);

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

@@ -123,12 +123,12 @@ int32_t  tsdbRowIterOpen(STSDBRowIter *pIter, TSDBROW *pRow, STSchema *pTSchema)
 void     tsdbRowClose(STSDBRowIter *pIter);
 SColVal *tsdbRowIterNext(STSDBRowIter *pIter);
 // SRowMerger
-int32_t tsdbRowMergerInit2(SRowMerger *pMerger, STSchema *pResTSchema, TSDBROW *pRow, STSchema *pTSchema);
+int32_t tsdbRowMergerInit(SRowMerger *pMerger, STSchema *pResTSchema, TSDBROW *pRow, STSchema *pTSchema);
 int32_t tsdbRowMergerAdd(SRowMerger *pMerger, TSDBROW *pRow, STSchema *pTSchema);
 
-int32_t tsdbRowMergerInit(SRowMerger *pMerger, TSDBROW *pRow, STSchema *pTSchema);
-void    tsdbRowMergerClear(SRowMerger *pMerger);
-int32_t tsdbRowMerge(SRowMerger *pMerger, TSDBROW *pRow);
+// int32_t tsdbRowMergerInit(SRowMerger *pMerger, TSDBROW *pRow, STSchema *pTSchema);
+void tsdbRowMergerClear(SRowMerger *pMerger);
+// int32_t tsdbRowMerge(SRowMerger *pMerger, TSDBROW *pRow);
 int32_t tsdbRowMergerGetRow(SRowMerger *pMerger, SRow **ppRow);
 // TABLEID
 int32_t tTABLEIDCmprFn(const void *p1, const void *p2);
@@ -224,6 +224,8 @@ int32_t tsdbTbDataIterCreate(STbData *pTbData, TSDBKEY *pFrom, int8_t backward,
 void   *tsdbTbDataIterDestroy(STbDataIter *pIter);
 void    tsdbTbDataIterOpen(STbData *pTbData, TSDBKEY *pFrom, int8_t backward, STbDataIter *pIter);
 bool    tsdbTbDataIterNext(STbDataIter *pIter);
+void    tsdbMemTableCountRows(SMemTable *pMemTable, SHashObj *pTableMap, int64_t *rowsNum);
+
 // STbData
 int32_t tsdbGetNRowsInTbData(STbData *pTbData);
 // tsdbFile.c ==============================================================================================

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

@@ -193,7 +193,7 @@ STQ*    tqOpen(const char* path, SVnode* pVnode);
 void    tqClose(STQ*);
 int     tqPushMsg(STQ*, void* msg, int32_t msgLen, tmsg_t msgType, int64_t ver);
 int     tqRegisterPushEntry(STQ* pTq, void* pHandle, const SMqPollReq* pRequest, SRpcMsg* pRpcMsg, SMqDataRsp* pDataRsp, int32_t type);
-int     tqRemovePushEntry(STQ* pTq, const char* pKey, int32_t keyLen, uint64_t consumerId, bool rspConsumer);
+int     tqUnregisterPushEntry(STQ* pTq, const char* pKey, int32_t keyLen, uint64_t consumerId, bool rspConsumer);
 
 int     tqCommit(STQ*);
 int32_t tqUpdateTbUidList(STQ* pTq, const SArray* tbUidList, bool isAdd);

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

@@ -268,7 +268,7 @@ int32_t buildSnapContext(SMeta* pMeta, int64_t snapVersion, int64_t suid, int8_t
   ctx->snapVersion = snapVersion;
   ctx->suid = suid;
   ctx->subType = subType;
-  ctx->queryMetaOrData = withMeta;
+  ctx->queryMeta = withMeta;
   ctx->withMeta = withMeta;
   ctx->idVersion = taosHashInit(100, taosGetDefaultHashFunction(TSDB_DATA_TYPE_BIGINT), true, HASH_NO_LOCK);
   if (ctx->idVersion == NULL) {
@@ -475,7 +475,7 @@ int32_t getMetafromSnapShot(SSnapContext* ctx, void** pBuf, int32_t* contLen, in
     if (ctx->index >= taosArrayGetSize(ctx->idList)) {
       metaDebug("tmqsnap get meta end");
       ctx->index = 0;
-      ctx->queryMetaOrData = false;  // change to get data
+      ctx->queryMeta = false;  // change to get data
       return 0;
     }
 

source/dnode/vnode/src/sma/smaRollup.c

@@ -669,7 +669,7 @@ static int32_t tdRSmaExecAndSubmitResult(SSma *pSma, qTaskInfo_t taskInfo, SRSma
 #endif
     for (int32_t i = 0; i < taosArrayGetSize(pResList); ++i) {
       SSDataBlock *output = taosArrayGetP(pResList, i);
-      smaDebug("result block, uid:%" PRIu64 ", groupid:%" PRIu64 ", rows:%d", output->info.id.uid,
+      smaDebug("result block, uid:%" PRIu64 ", groupid:%" PRIu64 ", rows:%" PRId64, output->info.id.uid,
                output->info.id.groupId, output->info.rows);
 
       STsdb       *sinkTsdb = (pItem->level == TSDB_RETENTION_L1 ? pSma->pRSmaTsdb[0] : pSma->pRSmaTsdb[1]);

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

@@ -223,19 +223,6 @@ static int32_t doSendDataRsp(const SRpcHandleInfo* pRpcHandleInfo, const SMqData
 
 int32_t tqPushDataRsp(STQ* pTq, STqPushEntry* pPushEntry) {
   SMqDataRsp* pRsp = pPushEntry->pDataRsp;
-
-#if 0
-  A(taosArrayGetSize(pRsp->blockData) == pRsp->blockNum);
-  A(taosArrayGetSize(pRsp->blockDataLen) == pRsp->blockNum);
-
-  A(!pRsp->withSchema);
-  A(taosArrayGetSize(pRsp->blockSchema) == 0);
-
-  if (pRsp->reqOffset.type == TMQ_OFFSET__LOG) {
-    A(pRsp->rspOffset.version > pRsp->reqOffset.version);
-  }
-#endif
-
   SMqRspHead* pHeader = &pPushEntry->pDataRsp->head;
   doSendDataRsp(&pPushEntry->info, pRsp, pHeader->epoch, pHeader->consumerId, pHeader->mqMsgType);
 
@@ -249,21 +236,6 @@ int32_t tqPushDataRsp(STQ* pTq, STqPushEntry* pPushEntry) {
 }
 
 int32_t tqSendDataRsp(STQ* pTq, const SRpcMsg* pMsg, const SMqPollReq* pReq, const SMqDataRsp* pRsp, int32_t type) {
-#if 0
-  A(taosArrayGetSize(pRsp->blockData) == pRsp->blockNum);
-  A(taosArrayGetSize(pRsp->blockDataLen) == pRsp->blockNum);
-
-  A(!pRsp->withSchema);
-  A(taosArrayGetSize(pRsp->blockSchema) == 0);
-
-  if (pRsp->reqOffset.type == TMQ_OFFSET__LOG) {
-    if (pRsp->blockNum > 0) {
-      A(pRsp->rspOffset.version > pRsp->reqOffset.version);
-    } else {
-      A(pRsp->rspOffset.version >= pRsp->reqOffset.version);
-    }
-  }
-#endif
   doSendDataRsp(&pMsg->info, pRsp, pReq->epoch, pReq->consumerId, type);
 
   char buf1[80] = {0};
@@ -365,17 +337,6 @@ static int32_t tqInitDataRsp(SMqDataRsp* pRsp, const SMqPollReq* pReq, int8_t su
   }
 
   pRsp->withTbName = 0;
-#if 0
-  pRsp->withTbName = pReq->withTbName;
-  if (pRsp->withTbName) {
-    pRsp->blockTbName = taosArrayInit(0, sizeof(void*));
-    if (pRsp->blockTbName == NULL) {
-      // TODO free
-      return -1;
-    }
-  }
-#endif
-
   pRsp->withSchema = false;
   return 0;
 }
@@ -412,8 +373,8 @@ static int32_t extractResetOffsetVal(STqOffsetVal* pOffsetVal, STQ* pTq, STqHand
 
     char formatBuf[80];
     tFormatOffset(formatBuf, 80, pOffsetVal);
-    tqDebug("tmq poll: consumer:0x%" PRIx64 ", subkey %s, vgId:%d, existed offset found, offset reset to %s and continue.",
-            consumerId, pHandle->subKey, vgId, formatBuf);
+    tqDebug("tmq poll: consumer:0x%" PRIx64 ", subkey %s, vgId:%d, existed offset found, offset reset to %s and continue. reqId:0x%"PRIx64,
+            consumerId, pHandle->subKey, vgId, formatBuf, pRequest->reqId);
     return 0;
   } else {
     // no poll occurs in this vnode for this topic, let's seek to the right offset value.
@@ -434,6 +395,7 @@ static int32_t extractResetOffsetVal(STqOffsetVal* pOffsetVal, STQ* pTq, STqHand
           return -1;
         }
 
+        // offset set to previous version when init
         tqOffsetResetToLog(pOffsetVal, pHandle->pRef->refVer - 1);
       }
     } else if (reqOffset.type == TMQ_OFFSET__RESET_LATEST) {
@@ -474,7 +436,6 @@ static int32_t extractResetOffsetVal(STqOffsetVal* pOffsetVal, STQ* pTq, STqHand
 
 static int32_t extractDataAndRspForNormalSubscribe(STQ* pTq, STqHandle* pHandle, const SMqPollReq* pRequest,
                                                    SRpcMsg* pMsg, STqOffsetVal* pOffset) {
-  int32_t  code = 0;
   uint64_t consumerId = pRequest->consumerId;
   int32_t  vgId = TD_VID(pTq->pVnode);
 
@@ -485,7 +446,10 @@ static int32_t extractDataAndRspForNormalSubscribe(STQ* pTq, STqHandle* pHandle,
   taosWLockLatch(&pTq->lock);
 
   qSetTaskId(pHandle->execHandle.task, consumerId, pRequest->reqId);
-  code = tqScanData(pTq, pHandle, &dataRsp, pOffset);
+  int code = tqScanData(pTq, pHandle, &dataRsp, pOffset);
+  if(code != 0) {
+    goto end;
+  }
 
   // till now, all data has been transferred to consumer, new data needs to push client once arrived.
   if (dataRsp.blockNum == 0 && dataRsp.reqOffset.type == TMQ_OFFSET__LOG &&
@@ -495,89 +459,60 @@ static int32_t extractDataAndRspForNormalSubscribe(STQ* pTq, STqHandle* pHandle,
     return code;
   }
 
-  taosWUnLockLatch(&pTq->lock);
+
   code = tqSendDataRsp(pTq, pMsg, pRequest, (SMqDataRsp*)&dataRsp, TMQ_MSG_TYPE__POLL_RSP);
 
   // NOTE: this pHandle->consumerId may have been changed already.
-  tqDebug("tmq poll: consumer:0x%" PRIx64 ", subkey %s, vgId:%d, rsp block:%d, offset type:%d, uid/version:%" PRId64
-          ", ts:%" PRId64 ", reqId:0x%" PRIx64,
-          consumerId, pHandle->subKey, vgId, dataRsp.blockNum, dataRsp.rspOffset.type, dataRsp.rspOffset.uid,
-          dataRsp.rspOffset.ts, pRequest->reqId);
 
-  tDeleteSMqDataRsp(&dataRsp);
+end:
+  {
+    char buf[80] = {0};
+    tFormatOffset(buf, 80, &dataRsp.rspOffset);
+    tqDebug("tmq poll: consumer:0x%" PRIx64 ", subkey %s, vgId:%d, rsp block:%d, rsp offset type:%s, reqId:0x%" PRIx64 " code:%d",
+            consumerId, pHandle->subKey, vgId, dataRsp.blockNum, buf, pRequest->reqId, code);
+    taosWUnLockLatch(&pTq->lock);
+    tDeleteSMqDataRsp(&dataRsp);
+  }
   return code;
 }
 
-static int32_t doPollDataForMq(STQ* pTq, STqHandle* pHandle, const SMqPollReq* pRequest, SRpcMsg* pMsg) {
-  int32_t      code = -1;
-  STqOffsetVal offset = {0};
-  SWalCkHead*  pCkHead = NULL;
-  int32_t      vgId = TD_VID(pTq->pVnode);
-
-  STqOffsetVal reqOffset = pRequest->reqOffset;
-  uint64_t     consumerId = pRequest->consumerId;
 
-  // 1. reset the offset if needed
-  if (IS_OFFSET_RESET_TYPE(reqOffset.type)) {
-    // handle the reset offset cases, according to the consumer's choice.
-    bool blockReturned = false;
-    code = extractResetOffsetVal(&offset, pTq, pHandle, pRequest, pMsg, &blockReturned);
-    if (code != 0) {
-      return code;
-    }
-
-    // empty block returned, quit
-    if (blockReturned) {
-      return 0;
-    }
-  } else { // use the consumer specified offset
-    // the offset value can not be monotonious increase??
-    offset = reqOffset;
-  }
-
-  // this is a normal subscribe requirement
-  if (pHandle->execHandle.subType == TOPIC_SUB_TYPE__COLUMN) {
-    return extractDataAndRspForNormalSubscribe(pTq, pHandle, pRequest, pMsg, &offset);
-  }
-
-  // todo handle the case where re-balance occurs.
-  // for taosx
+static int32_t extractDataAndRspForDbStbSubscribe(STQ* pTq, STqHandle* pHandle, const SMqPollReq* pRequest, SRpcMsg* pMsg, STqOffsetVal *offset) {
+  int code = 0;
+  int32_t      vgId = TD_VID(pTq->pVnode);
+  SWalCkHead*  pCkHead = NULL;
   SMqMetaRsp metaRsp = {0};
   STaosxRsp taosxRsp = {0};
   tqInitTaosxRsp(&taosxRsp, pRequest);
 
-  if (offset.type != TMQ_OFFSET__LOG) {
-    if (tqScanTaosx(pTq, pHandle, &taosxRsp, &metaRsp, &offset) < 0) {
+  if (offset->type != TMQ_OFFSET__LOG) {
+    if (tqScanTaosx(pTq, pHandle, &taosxRsp, &metaRsp, offset) < 0) {
       return -1;
     }
 
     if (metaRsp.metaRspLen > 0) {
       code = tqSendMetaPollRsp(pTq, pMsg, pRequest, &metaRsp);
-      tqDebug("tmq poll: consumer:0x%" PRIx64 " subkey:%s vgId:%d, send meta offset type:%d,uid:%" PRId64
-                  ",ts:%" PRId64,
-              consumerId, pHandle->subKey, vgId, metaRsp.rspOffset.type, metaRsp.rspOffset.uid,
-              metaRsp.rspOffset.ts);
+      tqDebug("tmq poll: consumer:0x%" PRIx64 " subkey:%s vgId:%d, send meta offset type:%d,uid:%" PRId64 ",ts:%" PRId64,
+              pRequest->consumerId, pHandle->subKey, vgId, metaRsp.rspOffset.type, metaRsp.rspOffset.uid, metaRsp.rspOffset.ts);
       taosMemoryFree(metaRsp.metaRsp);
       tDeleteSTaosxRsp(&taosxRsp);
       return code;
     }
 
+    tqDebug("taosx poll: consumer:0x%" PRIx64 " subkey:%s vgId:%d, send data blockNum:%d, offset type:%d,uid:%" PRId64
+    ",ts:%" PRId64,pRequest->consumerId, pHandle->subKey, vgId, taosxRsp.blockNum, taosxRsp.rspOffset.type, taosxRsp.rspOffset.uid,taosxRsp.rspOffset.ts);
     if (taosxRsp.blockNum > 0) {
       code = tqSendDataRsp(pTq, pMsg, pRequest, (SMqDataRsp*)&taosxRsp, TMQ_MSG_TYPE__TAOSX_RSP);
       tDeleteSTaosxRsp(&taosxRsp);
       return code;
-    } else {
-      offset = taosxRsp.rspOffset;
+    }else {
+      *offset = taosxRsp.rspOffset;
     }
-
-    tqDebug("taosx poll: consumer:0x%" PRIx64 " subkey:%s vgId:%d, send data blockNum:%d, offset type:%d,uid:%" PRId64
-                ",version:%" PRId64,
-            consumerId, pHandle->subKey, vgId, taosxRsp.blockNum, taosxRsp.rspOffset.type, taosxRsp.rspOffset.uid,
-            taosxRsp.rspOffset.version);
   }
 
-  if (offset.type == TMQ_OFFSET__LOG) {
-    int64_t fetchVer = offset.version + 1;
+
+  if (offset->type == TMQ_OFFSET__LOG) {
+    int64_t fetchVer = offset->version + 1;
     pCkHead = taosMemoryMalloc(sizeof(SWalCkHead) + 2048);
     if (pCkHead == NULL) {
       tDeleteSTaosxRsp(&taosxRsp);
@@ -587,12 +522,10 @@ static int32_t doPollDataForMq(STQ* pTq, STqHandle* pHandle, const SMqPollReq* p
     walSetReaderCapacity(pHandle->pWalReader, 2048);
     int totalRows = 0;
     while (1) {
-      // todo refactor: this is not correct.
       int32_t savedEpoch = atomic_load_32(&pHandle->epoch);
       if (savedEpoch > pRequest->epoch) {
         tqWarn("tmq poll: consumer:0x%" PRIx64 " (epoch %d), subkey:%s vgId:%d offset %" PRId64
-                   ", found new consumer epoch %d, discard req epoch %d",
-               consumerId, pRequest->epoch, pHandle->subKey, vgId, fetchVer, savedEpoch, pRequest->epoch);
+          ", found new consumer epoch %d, discard req epoch %d", pRequest->consumerId, pRequest->epoch, pHandle->subKey, vgId, fetchVer, savedEpoch, pRequest->epoch);
         break;
       }
 
@@ -605,7 +538,7 @@ static int32_t doPollDataForMq(STQ* pTq, STqHandle* pHandle, const SMqPollReq* p
       }
 
       SWalCont* pHead = &pCkHead->head;
-      tqDebug("tmq poll: consumer:0x%" PRIx64 " (epoch %d) iter log, vgId:%d offset %" PRId64 " msgType %d", consumerId,
+      tqDebug("tmq poll: consumer:0x%" PRIx64 " (epoch %d) iter log, vgId:%d offset %" PRId64 " msgType %d", pRequest->consumerId,
               pRequest->epoch, vgId, fetchVer, pHead->msgType);
 
       // process meta
@@ -643,7 +576,7 @@ static int32_t doPollDataForMq(STQ* pTq, STqHandle* pHandle, const SMqPollReq* p
       };
 
       if (tqTaosxScanLog(pTq, pHandle, submit, &taosxRsp, &totalRows) < 0) {
-        tqError("tmq poll: tqTaosxScanLog error %" PRId64 ", in vgId:%d, subkey %s", consumerId, vgId,
+        tqError("tmq poll: tqTaosxScanLog error %" PRId64 ", in vgId:%d, subkey %s", pRequest->consumerId, vgId,
                 pRequest->subKey);
         taosMemoryFreeClear(pCkHead);
         tDeleteSTaosxRsp(&taosxRsp);
@@ -667,6 +600,39 @@ static int32_t doPollDataForMq(STQ* pTq, STqHandle* pHandle, const SMqPollReq* p
   return 0;
 }
 
+static int32_t doPollDataForMq(STQ* pTq, STqHandle* pHandle, const SMqPollReq* pRequest, SRpcMsg* pMsg) {
+  int32_t      code = -1;
+  STqOffsetVal offset = {0};
+  STqOffsetVal reqOffset = pRequest->reqOffset;
+
+  // 1. reset the offset if needed
+  if (IS_OFFSET_RESET_TYPE(reqOffset.type)) {
+    // handle the reset offset cases, according to the consumer's choice.
+    bool blockReturned = false;
+    code = extractResetOffsetVal(&offset, pTq, pHandle, pRequest, pMsg, &blockReturned);
+    if (code != 0) {
+      return code;
+    }
+
+    // empty block returned, quit
+    if (blockReturned) {
+      return 0;
+    }
+  } else { // use the consumer specified offset
+    // the offset value can not be monotonious increase??
+    offset = reqOffset;
+  }
+
+  // this is a normal subscribe requirement
+  if (pHandle->execHandle.subType == TOPIC_SUB_TYPE__COLUMN) {
+    return extractDataAndRspForNormalSubscribe(pTq, pHandle, pRequest, pMsg, &offset);
+  }
+
+  // todo handle the case where re-balance occurs.
+  // for taosx
+  return extractDataAndRspForDbStbSubscribe(pTq, pHandle, pRequest, pMsg, &offset);
+}
+
 int32_t tqProcessPollReq(STQ* pTq, SRpcMsg* pMsg) {
   SMqPollReq   req = {0};
   if (tDeserializeSMqPollReq(pMsg->pCont, pMsg->contLen, &req) < 0) {
@@ -808,7 +774,6 @@ int32_t tqProcessSubscribeReq(STQ* pTq, int64_t sversion, char* msg, int32_t msg
 
     STqHandle tqHandle = {0};
     pHandle = &tqHandle;
-    /*taosInitRWLatch(&pExec->lock);*/
 
     uint64_t oldConsumerId = pHandle->consumerId;
     memcpy(pHandle->subKey, req.subKey, TSDB_SUBSCRIBE_KEY_LEN);
@@ -829,13 +794,7 @@ int32_t tqProcessSubscribeReq(STQ* pTq, int64_t sversion, char* msg, int32_t msg
     pHandle->pRef = pRef;
 
     SReadHandle handle = {
-        .meta = pVnode->pMeta,
-        .vnode = pVnode,
-        .initTableReader = true,
-        .initTqReader = true,
-        .version = ver,
-    };
-
+        .meta = pVnode->pMeta, .vnode = pVnode, .initTableReader = true, .initTqReader = true, .version = ver};
     pHandle->snapshotVer = ver;
 
     if (pHandle->execHandle.subType == TOPIC_SUB_TYPE__COLUMN) {
@@ -843,7 +802,7 @@ int32_t tqProcessSubscribeReq(STQ* pTq, int64_t sversion, char* msg, int32_t msg
       req.qmsg = NULL;
 
       pHandle->execHandle.task =
-          qCreateQueueExecTaskInfo(pHandle->execHandle.execCol.qmsg, &handle, vgId, &pHandle->execHandle.numOfCols, NULL);
+          qCreateQueueExecTaskInfo(pHandle->execHandle.execCol.qmsg, &handle, vgId, &pHandle->execHandle.numOfCols, req.newConsumerId);
       void* scanner = NULL;
       qExtractStreamScanner(pHandle->execHandle.task, &scanner);
       pHandle->execHandle.pExecReader = qExtractReaderFromStreamScanner(scanner);
@@ -856,7 +815,7 @@ int32_t tqProcessSubscribeReq(STQ* pTq, int64_t sversion, char* msg, int32_t msg
       buildSnapContext(handle.meta, handle.version, 0, pHandle->execHandle.subType, pHandle->fetchMeta,
                        (SSnapContext**)(&handle.sContext));
 
-      pHandle->execHandle.task = qCreateQueueExecTaskInfo(NULL, &handle, vgId, NULL, NULL);
+      pHandle->execHandle.task = qCreateQueueExecTaskInfo(NULL, &handle, vgId, NULL, req.newConsumerId);
     } else if (pHandle->execHandle.subType == TOPIC_SUB_TYPE__TABLE) {
       pHandle->pWalReader = walOpenReader(pVnode->pWal, NULL);
       pHandle->execHandle.execTb.suid = req.suid;
@@ -874,7 +833,7 @@ int32_t tqProcessSubscribeReq(STQ* pTq, int64_t sversion, char* msg, int32_t msg
 
       buildSnapContext(handle.meta, handle.version, req.suid, pHandle->execHandle.subType, pHandle->fetchMeta,
                        (SSnapContext**)(&handle.sContext));
-      pHandle->execHandle.task = qCreateQueueExecTaskInfo(NULL, &handle, vgId, NULL, NULL);
+      pHandle->execHandle.task = qCreateQueueExecTaskInfo(NULL, &handle, vgId, NULL, req.newConsumerId);
     }
 
     taosHashPut(pTq->pHandle, req.subKey, strlen(req.subKey), pHandle, sizeof(STqHandle));
@@ -905,7 +864,7 @@ int32_t tqProcessSubscribeReq(STQ* pTq, int64_t sversion, char* msg, int32_t msg
       atomic_store_32(&pHandle->epoch, -1);
 
       // remove if it has been register in the push manager, and return one empty block to consumer
-      tqRemovePushEntry(pTq, req.subKey, (int32_t)strlen(req.subKey), pHandle->consumerId, true);
+      tqUnregisterPushEntry(pTq, req.subKey, (int32_t)strlen(req.subKey), pHandle->consumerId, true);
 
       atomic_store_64(&pHandle->consumerId, req.newConsumerId);
       atomic_add_fetch_32(&pHandle->epoch, 1);
@@ -1348,7 +1307,7 @@ int32_t tqProcessDelReq(STQ* pTq, void* pReq, int32_t len, int64_t ver) {
       pRefBlock->dataRef = pRef;
       atomic_add_fetch_32(pRefBlock->dataRef, 1);
 
-      if (streamTaskInput(pTask, (SStreamQueueItem*)pRefBlock) < 0) {
+      if (tAppendDataForStream(pTask, (SStreamQueueItem*)pRefBlock) < 0) {
         qError("stream task input del failed, task id %d", pTask->taskId);
 
         atomic_sub_fetch_32(pRef, 1);
@@ -1383,7 +1342,7 @@ int32_t tqProcessDelReq(STQ* pTq, void* pReq, int32_t len, int64_t ver) {
     taosArrayPush(pStreamBlock->blocks, &block);
 
     if (!failed) {
-      if (streamTaskInput(pTask, (SStreamQueueItem*)pStreamBlock) < 0) {
+      if (tAppendDataForStream(pTask, (SStreamQueueItem*)pStreamBlock) < 0) {
         qError("stream task input del failed, task id %d", pTask->taskId);
         continue;
       }
@@ -1403,15 +1362,14 @@ int32_t tqProcessDelReq(STQ* pTq, void* pReq, int32_t len, int64_t ver) {
 }
 
 int32_t tqProcessSubmitReq(STQ* pTq, SPackedData submit) {
-  void*               pIter = NULL;
-  bool                failed = false;
-  SStreamDataSubmit2* pSubmit = NULL;
+  void* pIter = NULL;
+  bool  succ = true;
 
-  pSubmit = streamDataSubmitNew(submit);
+  SStreamDataSubmit2* pSubmit = streamDataSubmitNew(submit);
   if (pSubmit == NULL) {
     terrno = TSDB_CODE_OUT_OF_MEMORY;
     tqError("failed to create data submit for stream since out of memory");
-    failed = true;
+    succ = false;
   }
 
   while (1) {
@@ -1421,22 +1379,27 @@ int32_t tqProcessSubmitReq(STQ* pTq, SPackedData submit) {
     }
 
     SStreamTask* pTask = *(SStreamTask**)pIter;
-    if (pTask->taskLevel != TASK_LEVEL__SOURCE) continue;
+    if (pTask->taskLevel != TASK_LEVEL__SOURCE) {
+      continue;
+    }
+
     if (pTask->taskStatus == TASK_STATUS__RECOVER_PREPARE || pTask->taskStatus == TASK_STATUS__WAIT_DOWNSTREAM) {
-      tqDebug("skip push task %d, task status %d", pTask->taskId, pTask->taskStatus);
+      tqDebug("stream task:%d skip push data, not ready for processing, status %d", pTask->taskId, pTask->taskStatus);
       continue;
     }
 
-    tqDebug("data submit enqueue stream task: %d, ver: %" PRId64, pTask->taskId, submit.ver);
+    tqDebug("data submit enqueue stream task:%d, ver: %" PRId64, pTask->taskId, submit.ver);
+    if (succ) {
+      int32_t code = tAppendDataForStream(pTask, (SStreamQueueItem*)pSubmit);
+      if (code < 0) {
+        // let's handle the back pressure
 
-    if (!failed) {
-      if (streamTaskInput(pTask, (SStreamQueueItem*)pSubmit) < 0) {
-        tqError("stream task input failed, task id %d", pTask->taskId);
+        tqError("stream task:%d failed to put into queue for, too many", pTask->taskId);
         continue;
       }
 
       if (streamSchedExec(pTask) < 0) {
-        tqError("stream task launch failed, task id %d", pTask->taskId);
+        tqError("stream task:%d launch failed, code:%s", pTask->taskId, tstrerror(terrno));
         continue;
       }
     } else {
@@ -1444,12 +1407,12 @@ int32_t tqProcessSubmitReq(STQ* pTq, SPackedData submit) {
     }
   }
 
-  if (pSubmit) {
-    streamDataSubmitRefDec(pSubmit);
+  if (pSubmit != NULL) {
+    streamDataSubmitDestroy(pSubmit);
     taosFreeQitem(pSubmit);
   }
 
-  return failed ? -1 : 0;
+  return succ ? 0 : -1;
 }
 
 int32_t tqProcessTaskRunReq(STQ* pTq, SRpcMsg* pMsg) {
@@ -1569,6 +1532,8 @@ int32_t vnodeEnqueueStreamMsg(SVnode* pVnode, SRpcMsg* pMsg) {
     rpcFreeCont(pMsg->pCont);
     taosFreeQitem(pMsg);
     return 0;
+  } else {
+    tDeleteStreamDispatchReq(&req);
   }
 
   code = TSDB_CODE_STREAM_TASK_NOT_EXIST;

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

@@ -307,7 +307,7 @@ int32_t tqMetaRestoreHandle(STQ* pTq) {
 
     if (handle.execHandle.subType == TOPIC_SUB_TYPE__COLUMN) {
       handle.execHandle.task =
-          qCreateQueueExecTaskInfo(handle.execHandle.execCol.qmsg, &reader, vgId, &handle.execHandle.numOfCols, NULL);
+          qCreateQueueExecTaskInfo(handle.execHandle.execCol.qmsg, &reader, vgId, &handle.execHandle.numOfCols, 0);
       if (handle.execHandle.task == NULL) {
         tqError("cannot create exec task for %s", handle.subKey);
         code = -1;
@@ -332,7 +332,7 @@ int32_t tqMetaRestoreHandle(STQ* pTq) {
 
       buildSnapContext(reader.meta, reader.version, 0, handle.execHandle.subType, handle.fetchMeta,
                        (SSnapContext**)(&reader.sContext));
-      handle.execHandle.task = qCreateQueueExecTaskInfo(NULL, &reader, vgId, NULL, NULL);
+      handle.execHandle.task = qCreateQueueExecTaskInfo(NULL, &reader, vgId, NULL, 0);
     } else if (handle.execHandle.subType == TOPIC_SUB_TYPE__TABLE) {
       handle.pWalReader = walOpenReader(pTq->pVnode->pWal, NULL);
 
@@ -341,7 +341,7 @@ int32_t tqMetaRestoreHandle(STQ* pTq) {
       tqDebug("vgId:%d, tq try to get all ctb, suid:%" PRId64, pTq->pVnode->config.vgId, handle.execHandle.execTb.suid);
       for (int32_t i = 0; i < taosArrayGetSize(tbUidList); i++) {
         int64_t tbUid = *(int64_t*)taosArrayGet(tbUidList, i);
-        tqDebug("vgId:%d, idx %d, uid:%" PRId64, TD_VID(pTq->pVnode), i, tbUid);
+        tqDebug("vgId:%d, idx %d, uid:%" PRId64, vgId, i, tbUid);
       }
       handle.execHandle.pExecReader = tqOpenReader(pTq->pVnode);
       tqReaderSetTbUidList(handle.execHandle.pExecReader, tbUidList);
@@ -349,9 +349,9 @@ int32_t tqMetaRestoreHandle(STQ* pTq) {
 
       buildSnapContext(reader.meta, reader.version, handle.execHandle.execTb.suid, handle.execHandle.subType,
                        handle.fetchMeta, (SSnapContext**)(&reader.sContext));
-      handle.execHandle.task = qCreateQueueExecTaskInfo(NULL, &reader, vgId, NULL, NULL);
+      handle.execHandle.task = qCreateQueueExecTaskInfo(NULL, &reader, vgId, NULL, 0);
     }
-    tqDebug("tq restore %s consumer %" PRId64 " vgId:%d", handle.subKey, handle.consumerId, TD_VID(pTq->pVnode));
+    tqDebug("tq restore %s consumer %" PRId64 " vgId:%d", handle.subKey, handle.consumerId, vgId);
     taosHashPut(pTq->pHandle, pKey, kLen, &handle, sizeof(STqHandle));
   }
 

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

@@ -30,7 +30,7 @@ static int32_t tqLoopExecFromQueue(STQ* pTq, STqHandle* pHandle, SStreamDataSubm
     // update processed
     atomic_store_64(&pHandle->pushHandle.processedVer, pSubmit->ver);
     streamQueueProcessSuccess(&pHandle->pushHandle.inputQ);
-    streamDataSubmitRefDec(pSubmit);
+    streamDataSubmitDestroy(pSubmit);
     if (pRsp->blockNum > 0) {
       *ppSubmit = pSubmit;
       return 0;
@@ -58,7 +58,7 @@ int32_t tqExecFromInputQ(STQ* pTq, STqHandle* pHandle) {
     }
     while (pHandle->pushHandle.processedVer > pSubmit->ver + 1) {
       streamQueueProcessSuccess(&pHandle->pushHandle.inputQ);
-      streamDataSubmitRefDec(pSubmit);
+      streamDataSubmitDestroy(pSubmit);
       pSubmit = streamQueueNextItem(&pHandle->pushHandle.inputQ);
       if (pSubmit == NULL) break;
     }
@@ -120,7 +120,7 @@ int32_t tqPreparePush(STQ* pTq, STqHandle* pHandle, int64_t reqId, const SRpcHan
 int32_t tqEnqueue(STqHandle* pHandle, SStreamDataSubmit* pSubmit) {
   int8_t inputStatus = atomic_load_8(&pHandle->pushHandle.inputStatus);
   if (inputStatus == TASK_INPUT_STATUS__NORMAL) {
-    SStreamDataSubmit* pSubmitClone = streamSubmitRefClone(pSubmit);
+    SStreamDataSubmit* pSubmitClone = streamSubmitBlockClone(pSubmit);
     if (pSubmitClone == NULL) {
       return -1;
     }
@@ -206,7 +206,69 @@ int32_t tqPushMsgNew(STQ* pTq, void* msg, int32_t msgLen, tmsg_t msgType, int64_
 }
 #endif
 
-int tqPushMsg(STQ* pTq, void* msg, int32_t msgLen, tmsg_t msgType, int64_t ver) {
+typedef struct {
+  void*   pKey;
+  int64_t keyLen;
+} SItem;
+
+static void recordPushedEntry(SArray* cachedKey, void* pIter);
+static void doRemovePushedEntry(SArray* pCachedKeys, STQ* pTq);
+
+static void freeItem(void* param) {
+  SItem* p = (SItem*) param;
+  taosMemoryFree(p->pKey);
+}
+
+static void doPushDataForEntry(void* pIter, STqExecHandle* pExec, STQ* pTq, int64_t ver, int32_t vgId, char* pData,
+                               int32_t dataLen, SArray* pCachedKey) {
+  STqPushEntry* pPushEntry = *(STqPushEntry**)pIter;
+
+  SMqDataRsp* pRsp = pPushEntry->pDataRsp;
+  if (pRsp->reqOffset.version >= ver) {
+    tqDebug("vgId:%d, push entry req version %" PRId64 ", while push version %" PRId64 ", skip", vgId,
+            pRsp->reqOffset.version, ver);
+    return;
+  }
+
+  qTaskInfo_t pTaskInfo = pExec->task;
+
+  // prepare scan mem data
+  SPackedData submit = {.msgStr = pData, .msgLen = dataLen, .ver = ver};
+
+  if (qStreamSetScanMemData(pTaskInfo, submit) != 0) {
+    return;
+  }
+  qStreamSetOpen(pTaskInfo);
+  // here start to scan submit block to extract the subscribed data
+  int32_t totalRows = 0;
+
+  while (1) {
+    SSDataBlock* pDataBlock = NULL;
+    uint64_t     ts = 0;
+    if (qExecTask(pTaskInfo, &pDataBlock, &ts) < 0) {
+      tqDebug("vgId:%d, tq exec error since %s", vgId, terrstr());
+    }
+
+    if (pDataBlock == NULL) {
+      break;
+    }
+
+    tqAddBlockDataToRsp(pDataBlock, pRsp, pExec->numOfCols, pTq->pVnode->config.tsdbCfg.precision);
+    pRsp->blockNum++;
+    totalRows += pDataBlock->info.rows;
+  }
+
+  tqDebug("vgId:%d, tq handle push, subkey:%s, block num:%d, rows:%d", vgId, pPushEntry->subKey, pRsp->blockNum,
+      totalRows);
+
+  if (pRsp->blockNum > 0) {
+    tqOffsetResetToLog(&pRsp->rspOffset, ver);
+    tqPushDataRsp(pTq, pPushEntry);
+    recordPushedEntry(pCachedKey, pIter);
+  }
+}
+
+int32_t tqPushMsg(STQ* pTq, void* msg, int32_t msgLen, tmsg_t msgType, int64_t ver) {
   void*   pReq = POINTER_SHIFT(msg, sizeof(SSubmitReq2Msg));
   int32_t len = msgLen - sizeof(SSubmitReq2Msg);
   int32_t vgId = TD_VID(pTq->pVnode);
@@ -220,24 +282,19 @@ int tqPushMsg(STQ* pTq, void* msg, int32_t msgLen, tmsg_t msgType, int64_t ver)
       tqDebug("vgId:%d tq push msg version:%" PRId64 " type:%s, head:%p, body:%p len:%d, numOfPushed consumers:%d",
           vgId, ver, TMSG_INFO(msgType), msg, pReq, len, numOfRegisteredPush);
 
-      SArray* cachedKeys = taosArrayInit(0, sizeof(void*));
-      SArray* cachedKeyLens = taosArrayInit(0, sizeof(size_t));
-
-      void*   data = taosMemoryMalloc(len);
+      void* data = taosMemoryMalloc(len);
       if (data == NULL) {
         terrno = TSDB_CODE_OUT_OF_MEMORY;
-        tqError("failed to copy data for stream since out of memory");
-        taosArrayDestroyP(cachedKeys, (FDelete)taosMemoryFree);
-        taosArrayDestroy(cachedKeyLens);
-
-        // unlock
+        tqError("failed to copy data for stream since out of memory, vgId:%d", vgId);
         taosWUnLockLatch(&pTq->lock);
         return -1;
       }
 
       memcpy(data, pReq, len);
 
-      void* pIter = NULL;
+      SArray* cachedKey = taosArrayInit(0, sizeof(SItem));
+      void*   pIter = NULL;
+
       while (1) {
         pIter = taosHashIterate(pTq->pPushMgr, pIter);
         if (pIter == NULL) {
@@ -248,96 +305,39 @@ int tqPushMsg(STQ* pTq, void* msg, int32_t msgLen, tmsg_t msgType, int64_t ver)
 
         STqHandle* pHandle = taosHashGet(pTq->pHandle, pPushEntry->subKey, strlen(pPushEntry->subKey));
         if (pHandle == NULL) {
-          tqDebug("vgId:%d, cannot find handle %s", pTq->pVnode->config.vgId, pPushEntry->subKey);
-          continue;
-        }
-
-        SMqDataRsp* pRsp = pPushEntry->pDataRsp;
-        if (pRsp->reqOffset.version >= ver) {
-          tqDebug("vgId:%d, push entry req version %" PRId64 ", while push version %" PRId64 ", skip", vgId,
-                  pRsp->reqOffset.version, ver);
+          tqDebug("vgId:%d, failed to find handle %s in pushing data to consumer, ignore", pTq->pVnode->config.vgId, pPushEntry->subKey);
           continue;
         }
 
         STqExecHandle* pExec = &pHandle->execHandle;
-        qTaskInfo_t    task = pExec->task;
-
-        // prepare scan mem data
-        SPackedData submit = {
-            .msgStr = data,
-            .msgLen = len,
-            .ver = ver,
-        };
-        if(qStreamSetScanMemData(task, submit) != 0){
-          continue;
-        }
-
-        // here start to scan submit block to extract the subscribed data
-        while (1) {
-          SSDataBlock* pDataBlock = NULL;
-          uint64_t     ts = 0;
-          if (qExecTask(task, &pDataBlock, &ts) < 0) {
-            tqDebug("vgId:%d, tq exec error since %s", vgId, terrstr());
-          }
-
-          if (pDataBlock == NULL) {
-            break;
-          }
-
-          tqAddBlockDataToRsp(pDataBlock, pRsp, pExec->numOfCols, pTq->pVnode->config.tsdbCfg.precision);
-          pRsp->blockNum++;
-        }
-
-        tqDebug("vgId:%d, tq handle push, subkey:%s, block num:%d", vgId, pPushEntry->subKey, pRsp->blockNum);
-        if (pRsp->blockNum > 0) {
-          // set offset
-          tqOffsetResetToLog(&pRsp->rspOffset, ver);
-
-          // remove from hash
-          size_t kLen;
-          void*  key = taosHashGetKey(pIter, &kLen);
-          void*  keyCopy = taosMemoryCalloc(1, kLen + 1);
-          memcpy(keyCopy, key, kLen);
-
-          taosArrayPush(cachedKeys, &keyCopy);
-          taosArrayPush(cachedKeyLens, &kLen);
-
-          tqPushDataRsp(pTq, pPushEntry);
-        }
+        doPushDataForEntry(pIter, pExec, pTq, ver, vgId, data, len, cachedKey);
       }
 
-      // delete entry
-      for (int32_t i = 0; i < taosArrayGetSize(cachedKeys); i++) {
-        void*  key = taosArrayGetP(cachedKeys, i);
-        size_t kLen = *(size_t*)taosArrayGet(cachedKeyLens, i);
-        if (taosHashRemove(pTq->pPushMgr, key, kLen) != 0) {
-          tqError("vgId:%d, tq push hash remove key error, key: %s", pTq->pVnode->config.vgId, (char*)key);
-        }
-      }
-
-      taosArrayDestroyP(cachedKeys, (FDelete)taosMemoryFree);
-      taosArrayDestroy(cachedKeyLens);
+      doRemovePushedEntry(cachedKey, pTq);
+      taosArrayDestroyEx(cachedKey, freeItem);
       taosMemoryFree(data);
     }
+
     // unlock
     taosWUnLockLatch(&pTq->lock);
   }
 
+  // push data for stream processing
   if (!tsDisableStream && vnodeIsRoleLeader(pTq->pVnode)) {
-    if (taosHashGetSize(pTq->pStreamMeta->pTasks) == 0) return 0;
+    if (taosHashGetSize(pTq->pStreamMeta->pTasks) == 0) {
+      return 0;
+    }
+
     if (msgType == TDMT_VND_SUBMIT) {
       void* data = taosMemoryMalloc(len);
       if (data == NULL) {
         terrno = TSDB_CODE_OUT_OF_MEMORY;
-        tqError("failed to copy data for stream since out of memory");
+        tqError("vgId:%d, failed to copy submit data for stream processing, since out of memory", vgId);
         return -1;
       }
+
       memcpy(data, pReq, len);
-      SPackedData submit = {
-          .msgStr = data,
-          .msgLen = len,
-          .ver = ver,
-      };
+      SPackedData submit = {.msgStr = data, .msgLen = len, .ver = ver};
 
       tqDebug("tq copy write msg %p %d %" PRId64 " from %p", data, len, ver, pReq);
       tqProcessSubmitReq(pTq, submit);
@@ -388,8 +388,8 @@ int32_t tqRegisterPushEntry(STQ* pTq, void* pHandle, const SMqPollReq* pRequest,
   return 0;
 }
 
-int32_t tqRemovePushEntry(STQ* pTq, const char* pKey, int32_t keyLen, uint64_t consumerId, bool rspConsumer) {
-  int32_t       vgId = TD_VID(pTq->pVnode);
+int32_t tqUnregisterPushEntry(STQ* pTq, const char* pKey, int32_t keyLen, uint64_t consumerId, bool rspConsumer) {
+  int32_t        vgId = TD_VID(pTq->pVnode);
   STqPushEntry** pEntry = taosHashGet(pTq->pPushMgr, pKey, keyLen);
 
   if (pEntry != NULL) {
@@ -408,3 +408,26 @@ int32_t tqRemovePushEntry(STQ* pTq, const char* pKey, int32_t keyLen, uint64_t c
 
   return 0;
 }
+
+void recordPushedEntry(SArray* cachedKey, void* pIter) {
+  size_t kLen = 0;
+  void*  key = taosHashGetKey(pIter, &kLen);
+  SItem item = {.pKey = strndup(key, kLen), .keyLen = kLen};
+  taosArrayPush(cachedKey, &item);
+}
+
+void doRemovePushedEntry(SArray* pCachedKeys, STQ* pTq) {
+  int32_t vgId = TD_VID(pTq->pVnode);
+  int32_t  numOfKeys = (int32_t) taosArrayGetSize(pCachedKeys);
+
+  for (int32_t i = 0; i < numOfKeys; i++) {
+    SItem* pItem = taosArrayGet(pCachedKeys, i);
+    if (taosHashRemove(pTq->pPushMgr, pItem->pKey, pItem->keyLen) != 0) {
+      tqError("vgId:%d, tq push hash remove key error, key: %s", vgId, (char*) pItem->pKey);
+    }
+  }
+
+  if (numOfKeys > 0) {
+    tqDebug("vgId:%d, pushed %d items and remain:%d", vgId, numOfKeys, (int32_t)taosHashGetSize(pTq->pPushMgr));
+  }
+}

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

@@ -18,7 +18,9 @@
 int32_t tqAddBlockDataToRsp(const SSDataBlock* pBlock, SMqDataRsp* pRsp, int32_t numOfCols, int8_t precision) {
   int32_t dataStrLen = sizeof(SRetrieveTableRsp) + blockGetEncodeSize(pBlock);
   void*   buf = taosMemoryCalloc(1, dataStrLen);
-  if (buf == NULL) return -1;
+  if (buf == NULL) {
+    return TSDB_CODE_OUT_OF_MEMORY;
+  }
 
   SRetrieveTableRsp* pRetrieve = (SRetrieveTableRsp*)buf;
   pRetrieve->useconds = 0;
@@ -31,7 +33,8 @@ int32_t tqAddBlockDataToRsp(const SSDataBlock* pBlock, SMqDataRsp* pRsp, int32_t
   actualLen += sizeof(SRetrieveTableRsp);
   taosArrayPush(pRsp->blockDataLen, &actualLen);
   taosArrayPush(pRsp->blockData, &buf);
-  return 0;
+
+  return TSDB_CODE_SUCCESS;
 }
 
 static int32_t tqAddBlockSchemaToRsp(const STqExecHandle* pExec, STaosxRsp* pRsp) {
@@ -62,71 +65,50 @@ static int32_t tqAddTbNameToRsp(const STQ* pTq, int64_t uid, STaosxRsp* pRsp, in
 }
 
 int32_t tqScanData(STQ* pTq, const STqHandle* pHandle, SMqDataRsp* pRsp, STqOffsetVal* pOffset) {
-  const STqExecHandle* pExec = &pHandle->execHandle;
+  const int32_t MAX_ROWS_TO_RETURN = 4096;
+  int32_t       vgId = TD_VID(pTq->pVnode);
+  int32_t       code = 0;
+  int32_t       totalRows = 0;
 
+  const STqExecHandle* pExec = &pHandle->execHandle;
   qTaskInfo_t task = pExec->task;
-  int32_t vgId = TD_VID(pTq->pVnode);
 
   if (qStreamPrepareScan(task, pOffset, pHandle->execHandle.subType) < 0) {
-    tqDebug("prepare scan failed, return, consumer:0x%"PRIx64, pHandle->consumerId);
-    if (pOffset->type == TMQ_OFFSET__LOG) {
-      pRsp->rspOffset = *pOffset;
-      return 0;
-    } else {
-      tqOffsetResetToLog(pOffset, pHandle->snapshotVer);
-      if (qStreamPrepareScan(task, pOffset, pHandle->execHandle.subType) < 0) {
-        tqDebug("prepare scan failed, return, consumer:0x%"PRIx64, pHandle->consumerId);
-        pRsp->rspOffset = *pOffset;
-        return 0;
-      }
-    }
+    tqError("prepare scan failed, return");
+    return -1;
   }
 
-  int32_t rowCnt = 0;
   while (1) {
     SSDataBlock* pDataBlock = NULL;
     uint64_t     ts = 0;
-
-    tqDebug("vgId:%d, tmq task start to execute, consumer:0x%"PRIx64, vgId, pHandle->consumerId);
-    if (qExecTask(task, &pDataBlock, &ts) < 0) {
-      tqError("vgId:%d, task exec error since %s, consumer:0x%" PRIx64, vgId, terrstr(),
-              pHandle->consumerId);
+    qStreamSetOpen(task);
+    tqDebug("consumer:0x%"PRIx64" vgId:%d, tmq one task start execute", pHandle->consumerId, vgId);
+    if (qExecTask(task, &pDataBlock, &ts) != TSDB_CODE_SUCCESS) {
+      tqError("consumer:0x%"PRIx64" vgId:%d, task exec error since %s", pHandle->consumerId, vgId, terrstr());
       return -1;
     }
 
-    tqDebug("consumer:0x%"PRIx64" vgId:%d, tmq task executed, get %p", pHandle->consumerId, vgId, pDataBlock);
-
+    tqDebug("consumer:0x%"PRIx64" vgId:%d tmq one task end executed, pDataBlock:%p", pHandle->consumerId, vgId, pDataBlock);
     // current scan should be stopped asap, since the rebalance occurs.
     if (pDataBlock == NULL) {
       break;
     }
 
-    tqAddBlockDataToRsp(pDataBlock, pRsp, pExec->numOfCols, pTq->pVnode->config.tsdbCfg.precision);
-    pRsp->blockNum++;
-
-    if (pOffset->type == TMQ_OFFSET__SNAPSHOT_DATA) {
-      rowCnt += pDataBlock->info.rows;
-      if (rowCnt >= 4096) {
-        break;
-      }
+    code = tqAddBlockDataToRsp(pDataBlock, pRsp, pExec->numOfCols, pTq->pVnode->config.tsdbCfg.precision);
+    if (code != TSDB_CODE_SUCCESS) {
+      tqError("vgId:%d, failed to add block to rsp msg", vgId);
+      return code;
     }
-  }
 
-  if (qStreamExtractOffset(task, &pRsp->rspOffset) < 0) {
-    return -1;
-  }
-
-  if (pRsp->rspOffset.type == 0) {
-    tqError("vgId:%d, expected rsp offset: type %d %" PRId64 " %" PRId64 " %" PRId64, vgId, pRsp->rspOffset.type,
-            pRsp->rspOffset.ts, pRsp->rspOffset.uid, pRsp->rspOffset.version);
-    return -1;
-  }
-
-  if (pRsp->withTbName || pRsp->withSchema) {
-    tqError("vgId:%d, get column should not with meta:%d,%d", vgId, pRsp->withTbName, pRsp->withSchema);
-    return -1;
+    pRsp->blockNum++;
+    totalRows += pDataBlock->info.rows;
+    if (totalRows >= MAX_ROWS_TO_RETURN) {
+      break;
+    }
   }
 
+  tqDebug("consumer:0x%"PRIx64" vgId:%d tmq task executed finished, total blocks:%d, totalRows:%d", pHandle->consumerId, vgId, pRsp->blockNum, totalRows);
+  qStreamExtractOffset(task, &pRsp->rspOffset);
   return 0;
 }
 
@@ -135,18 +117,8 @@ int32_t tqScanTaosx(STQ* pTq, const STqHandle* pHandle, STaosxRsp* pRsp, SMqMeta
   qTaskInfo_t          task = pExec->task;
 
   if (qStreamPrepareScan(task, pOffset, pHandle->execHandle.subType) < 0) {
-    tqDebug("prepare scan failed, return");
-    if (pOffset->type == TMQ_OFFSET__LOG) {
-      pRsp->rspOffset = *pOffset;
-      return 0;
-    } else {
-      tqOffsetResetToLog(pOffset, pHandle->snapshotVer);
-      if (qStreamPrepareScan(task, pOffset, pHandle->execHandle.subType) < 0) {
-        tqDebug("prepare scan failed, return");
-        pRsp->rspOffset = *pOffset;
-        return 0;
-      }
-    }
+    tqDebug("tqScanTaosx prepare scan failed, return");
+    return -1;
   }
 
   int32_t rowCnt = 0;
@@ -192,42 +164,31 @@ int32_t tqScanTaosx(STQ* pTq, const STqHandle* pHandle, STaosxRsp* pRsp, SMqMeta
       }
     }
 
-    if (pDataBlock == NULL && pOffset->type == TMQ_OFFSET__SNAPSHOT_DATA) {
-      if (qStreamExtractPrepareUid(task) != 0) {
+    // get meta
+    SMqMetaRsp* tmp = qStreamExtractMetaMsg(task);
+    if (tmp->metaRspLen > 0) {
+      qStreamExtractOffset(task, &tmp->rspOffset);
+      *pMetaRsp = *tmp;
+
+      tqDebug("tmqsnap task get meta");
+      break;
+    }
+
+    if (pDataBlock == NULL) {
+      qStreamExtractOffset(task, pOffset);
+      if (pOffset->type == TMQ_OFFSET__SNAPSHOT_DATA) {
         continue;
       }
-      tqDebug("tmqsnap vgId: %d, tsdb consume over, switch to wal, ver %" PRId64, TD_VID(pTq->pVnode),
-              pHandle->snapshotVer + 1);
+      tqDebug("tmqsnap vgId: %d, tsdb consume over, switch to wal, ver %" PRId64, TD_VID(pTq->pVnode), pHandle->snapshotVer + 1);
+      qStreamExtractOffset(task, &pRsp->rspOffset);
       break;
     }
 
     if (pRsp->blockNum > 0) {
       tqDebug("tmqsnap task exec exited, get data");
+      qStreamExtractOffset(task, &pRsp->rspOffset);
       break;
     }
-
-    SMqMetaRsp* tmp = qStreamExtractMetaMsg(task);
-    if (tmp->rspOffset.type == TMQ_OFFSET__SNAPSHOT_DATA) {
-      tqOffsetResetToData(pOffset, tmp->rspOffset.uid, tmp->rspOffset.ts);
-      qStreamPrepareScan(task, pOffset, pHandle->execHandle.subType);
-      tmp->rspOffset.type = TMQ_OFFSET__SNAPSHOT_META;
-      tqDebug("tmqsnap task exec change to get data");
-      continue;
-    }
-
-    *pMetaRsp = *tmp;
-    tqDebug("tmqsnap task exec exited, get meta");
-
-    tqDebug("task exec exited");
-    break;
-  }
-
-  qStreamExtractOffset(task, &pRsp->rspOffset);
-
-  if (pRsp->rspOffset.type == 0) {
-    tqError("expected rsp offset: type %d %" PRId64 " %" PRId64 " %" PRId64, pRsp->rspOffset.type, pRsp->rspOffset.ts,
-            pRsp->rspOffset.uid, pRsp->rspOffset.version);
-    return -1;
   }
 
   return 0;

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

@@ -282,6 +282,40 @@ bool tsdbTbDataIterNext(STbDataIter *pIter) {
   return true;
 }
 
+int64_t tsdbCountTbDataRows(STbData *pTbData) {
+  SMemSkipListNode *pNode = pTbData->sl.pHead;
+  int64_t rowsNum = 0;
+  
+  while (NULL != pNode) {
+    pNode = SL_GET_NODE_FORWARD(pNode, 0);
+    if (pNode == pTbData->sl.pTail) {
+      return rowsNum;
+    }
+
+    rowsNum++;
+  }
+
+  return rowsNum;
+}
+
+void tsdbMemTableCountRows(SMemTable *pMemTable, SHashObj*        pTableMap, int64_t *rowsNum) {
+  taosRLockLatch(&pMemTable->latch);
+  for (int32_t i = 0; i < pMemTable->nBucket; ++i) {
+    STbData *pTbData = pMemTable->aBucket[i];
+    while (pTbData) {
+      void* p = taosHashGet(pTableMap, &pTbData->uid, sizeof(pTbData->uid));
+      if (p == NULL) {
+        pTbData = pTbData->next;
+        continue;
+      }
+      
+      *rowsNum += tsdbCountTbDataRows(pTbData);
+      pTbData = pTbData->next;
+    }
+  }
+  taosRUnLockLatch(&pMemTable->latch);
+}
+
 static int32_t tsdbMemTableRehash(SMemTable *pMemTable) {
   int32_t code = 0;
 
@@ -787,4 +821,4 @@ SArray *tsdbMemTableGetTbDataArray(SMemTable *pMemTable) {
 
 _exit:
   return aTbDataP;
-}
+}
\ No newline at end of file

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

@@ -638,13 +638,17 @@ SColVal *tsdbRowIterNext(STSDBRowIter *pIter) {
 
 // SRowMerger ======================================================
 
-int32_t tsdbRowMergerInit2(SRowMerger *pMerger, STSchema *pResTSchema, TSDBROW *pRow, STSchema *pTSchema) {
+int32_t tsdbRowMergerInit(SRowMerger *pMerger, STSchema *pResTSchema, TSDBROW *pRow, STSchema *pTSchema) {
   int32_t   code = 0;
   TSDBKEY   key = TSDBROW_KEY(pRow);
   SColVal  *pColVal = &(SColVal){0};
   STColumn *pTColumn;
   int32_t   iCol, jCol = 0;
 
+  if (NULL == pResTSchema) {
+    pResTSchema = pTSchema;
+  }
+
   pMerger->pTSchema = pResTSchema;
   pMerger->version = key.version;
 
@@ -712,6 +716,9 @@ int32_t tsdbRowMergerAdd(SRowMerger *pMerger, TSDBROW *pRow, STSchema *pTSchema)
   STColumn *pTColumn;
   int32_t   iCol, jCol = 1;
 
+  if (NULL == pTSchema) {
+    pTSchema = pMerger->pTSchema;
+  }
   ASSERT(((SColVal *)pMerger->pArray->pData)->value.val == key.ts);
 
   for (iCol = 1; iCol < pMerger->pTSchema->numOfCols && jCol < pTSchema->numOfCols; ++iCol) {
@@ -727,23 +734,6 @@ int32_t tsdbRowMergerAdd(SRowMerger *pMerger, TSDBROW *pRow, STSchema *pTSchema)
     tsdbRowGetColVal(pRow, pTSchema, jCol++, pColVal);
 
     if (key.version > pMerger->version) {
-#if 0
-      if (!COL_VAL_IS_NONE(pColVal)) {
-        if ((!COL_VAL_IS_NULL(pColVal)) && IS_VAR_DATA_TYPE(pColVal->type)) {
-          SColVal *tColVal = taosArrayGet(pMerger->pArray, iCol);
-          code = tRealloc(&tColVal->value.pData, pColVal->value.nData);
-          if (code) return code;
-
-          tColVal->value.nData = pColVal->value.nData;
-          if (pColVal->value.nData) {
-            memcpy(tColVal->value.pData, pColVal->value.pData, pColVal->value.nData);
-          }
-          tColVal->flag = 0;
-        } else {
-          taosArraySet(pMerger->pArray, iCol, pColVal);
-        }
-      }
-#endif
       if (!COL_VAL_IS_NONE(pColVal)) {
         if (IS_VAR_DATA_TYPE(pColVal->type)) {
           SColVal *pTColVal = taosArrayGet(pMerger->pArray, iCol);
@@ -758,7 +748,6 @@ int32_t tsdbRowMergerAdd(SRowMerger *pMerger, TSDBROW *pRow, STSchema *pTSchema)
             pTColVal->flag = 0;
           } else {
             tFree(pTColVal->value.pData);
-            pTColVal->value.pData = NULL;
             taosArraySet(pMerger->pArray, iCol, pColVal);
           }
         } else {
@@ -789,7 +778,7 @@ int32_t tsdbRowMergerAdd(SRowMerger *pMerger, TSDBROW *pRow, STSchema *pTSchema)
   pMerger->version = key.version;
   return code;
 }
-
+/*
 int32_t tsdbRowMergerInit(SRowMerger *pMerger, TSDBROW *pRow, STSchema *pTSchema) {
   int32_t   code = 0;
   TSDBKEY   key = TSDBROW_KEY(pRow);
@@ -840,7 +829,7 @@ int32_t tsdbRowMergerInit(SRowMerger *pMerger, TSDBROW *pRow, STSchema *pTSchema
 _exit:
   return code;
 }
-
+*/
 void tsdbRowMergerClear(SRowMerger *pMerger) {
   for (int32_t iCol = 1; iCol < pMerger->pTSchema->numOfCols; iCol++) {
     SColVal *pTColVal = taosArrayGet(pMerger->pArray, iCol);
@@ -851,7 +840,7 @@ void tsdbRowMergerClear(SRowMerger *pMerger) {
 
   taosArrayDestroy(pMerger->pArray);
 }
-
+/*
 int32_t tsdbRowMerge(SRowMerger *pMerger, TSDBROW *pRow) {
   int32_t  code = 0;
   TSDBKEY  key = TSDBROW_KEY(pRow);
@@ -916,7 +905,7 @@ int32_t tsdbRowMerge(SRowMerger *pMerger, TSDBROW *pRow) {
 _exit:
   return code;
 }
-
+*/
 int32_t tsdbRowMergerGetRow(SRowMerger *pMerger, SRow **ppRow) {
   return tRowBuild(pMerger->pArray, pMerger->pTSchema, ppRow);
 }

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

@@ -541,6 +541,7 @@ int32_t vnodeProcessFetchMsg(SVnode *pVnode, SRpcMsg *pMsg, SQueueInfo *pInfo) {
       return vnodeGetBatchMeta(pVnode, pMsg);
     case TDMT_VND_TMQ_CONSUME:
       return tqProcessPollReq(pVnode->pTq, pMsg);
+
     case TDMT_STREAM_TASK_RUN:
       return tqProcessTaskRunReq(pVnode->pTq, pMsg);
 #if 1

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

@@ -378,7 +378,7 @@ static int32_t vnodeSyncEqCtrlMsg(const SMsgCb *msgcb, SRpcMsg *pMsg) {
     return -1;
   }
 
-  int32_t code = tmsgPutToQueue(msgcb, SYNC_CTRL_QUEUE, pMsg);
+  int32_t code = tmsgPutToQueue(msgcb, SYNC_RD_QUEUE, pMsg);
   if (code != 0) {
     rpcFreeCont(pMsg->pCont);
     pMsg->pCont = NULL;

source/libs/executor/inc/executil.h

@@ -107,6 +107,7 @@ int32_t         tableListGetGroupList(const STableListInfo* pTableList, int32_t
 uint64_t        tableListGetSize(const STableListInfo* pTableList);
 uint64_t        tableListGetSuid(const STableListInfo* pTableList);
 STableKeyInfo*  tableListGetInfo(const STableListInfo* pTableList, int32_t index);
+int32_t         tableListFind(const STableListInfo* pTableList, uint64_t uid, int32_t startIndex);
 
 size_t getResultRowSize(struct SqlFunctionCtx* pCtx, int32_t numOfOutput);
 void   initResultRowInfo(SResultRowInfo* pResultRowInfo);

source/libs/executor/src/cachescanoperator.c

@@ -37,6 +37,7 @@ typedef struct SCacheRowsScanInfo {
   SSDataBlock*    pBufferredRes;
   SArray*         pUidList;
   int32_t         indexOfBufferedRes;
+  STableListInfo* pTableList;
 } SCacheRowsScanInfo;
 
 static SSDataBlock* doScanCache(SOperatorInfo* pOperator);
@@ -47,15 +48,17 @@ static int32_t      removeRedundantTsCol(SLastRowScanPhysiNode* pScanNode, SColM
 #define SCAN_ROW_TYPE(_t)  ((_t)? CACHESCAN_RETRIEVE_LAST : CACHESCAN_RETRIEVE_LAST_ROW)
 
 SOperatorInfo* createCacherowsScanOperator(SLastRowScanPhysiNode* pScanNode, SReadHandle* readHandle,
-                                           SExecTaskInfo* pTaskInfo) {
+                                           STableListInfo* pTableListInfo, SExecTaskInfo* pTaskInfo) {
   int32_t           code = TSDB_CODE_SUCCESS;
   SCacheRowsScanInfo* pInfo = taosMemoryCalloc(1, sizeof(SCacheRowsScanInfo));
   SOperatorInfo*    pOperator = taosMemoryCalloc(1, sizeof(SOperatorInfo));
   if (pInfo == NULL || pOperator == NULL) {
     code = TSDB_CODE_OUT_OF_MEMORY;
+    tableListDestroy(pTableListInfo);
     goto _error;
   }
 
+  pInfo->pTableList = pTableListInfo;
   pInfo->readHandle = *readHandle;
 
   SDataBlockDescNode* pDescNode = pScanNode->scan.node.pOutputDataBlockDesc;
@@ -75,20 +78,18 @@ SOperatorInfo* createCacherowsScanOperator(SLastRowScanPhysiNode* pScanNode, SRe
     goto _error;
   }
 
-  STableListInfo* pTableList = pTaskInfo->pTableInfoList;
-
-  int32_t totalTables = tableListGetSize(pTableList);
+  int32_t totalTables = tableListGetSize(pTableListInfo);
   int32_t capacity = 0;
 
   pInfo->pUidList = taosArrayInit(4, sizeof(int64_t));
 
   // partition by tbname
-  if (oneTableForEachGroup(pTableList) || (totalTables == 1)) {
+  if (oneTableForEachGroup(pTableListInfo) || (totalTables == 1)) {
     pInfo->retrieveType = CACHESCAN_RETRIEVE_TYPE_ALL | SCAN_ROW_TYPE(pScanNode->ignoreNull);
 
-    STableKeyInfo* pList = tableListGetInfo(pTableList, 0);
+    STableKeyInfo* pList = tableListGetInfo(pTableListInfo, 0);
 
-    uint64_t suid = tableListGetSuid(pTableList);
+    uint64_t suid = tableListGetSuid(pTableListInfo);
     code = tsdbCacherowsReaderOpen(pInfo->readHandle.vnode, pInfo->retrieveType, pList, totalTables,
                                    taosArrayGetSize(pInfo->matchInfo.pList), suid, &pInfo->pLastrowReader, pTaskInfo->id.str);
     if (code != TSDB_CODE_SUCCESS) {
@@ -136,7 +137,7 @@ SSDataBlock* doScanCache(SOperatorInfo* pOperator) {
 
   SCacheRowsScanInfo* pInfo = pOperator->info;
   SExecTaskInfo*    pTaskInfo = pOperator->pTaskInfo;
-  STableListInfo*   pTableList = pTaskInfo->pTableInfoList;
+  STableListInfo*   pTableList = pInfo->pTableList;
 
   uint64_t suid = tableListGetSuid(pTableList);
   int32_t  size = tableListGetSize(pTableList);
@@ -281,6 +282,7 @@ void destroyCacheScanOperator(void* param) {
   taosMemoryFree(pInfo->pSlotIds);
   taosArrayDestroy(pInfo->pUidList);
   taosArrayDestroy(pInfo->matchInfo.pList);
+  tableListDestroy(pInfo->pTableList);
 
   if (pInfo->pLastrowReader != NULL) {
     pInfo->pLastrowReader = tsdbCacherowsReaderClose(pInfo->pLastrowReader);

source/libs/executor/src/executil.c

@@ -34,7 +34,7 @@ struct STableListInfo {
   int32_t   numOfOuputGroups;  // the data block will be generated one by one
   int32_t*  groupOffset;       // keep the offset value for each group in the tableList
   SArray*   pTableList;
-  SHashObj* map;  // speedup acquire the tableQueryInfo by table uid
+  SHashObj* map;               // speedup acquire the tableQueryInfo by table uid
   uint64_t  suid;
 };
 
@@ -571,6 +571,10 @@ int32_t getColInfoResultForGroupby(void* metaHandle, SNodeList* group, STableLis
           memcpy(pStart, data, len);
           pStart += len;
         } else if (IS_VAR_DATA_TYPE(pValue->info.type)) {
+          if (varDataTLen(data) > pValue->info.bytes) {
+            code = TSDB_CODE_TDB_INVALID_TABLE_SCHEMA_VER;
+            goto end;
+          }
           memcpy(pStart, data, varDataTLen(data));
           pStart += varDataTLen(data);
         } else {
@@ -1800,6 +1804,21 @@ STableKeyInfo* tableListGetInfo(const STableListInfo* pTableList, int32_t index)
   return taosArrayGet(pTableList->pTableList, index);
 }
 
+int32_t tableListFind(const STableListInfo* pTableList, uint64_t uid, int32_t startIndex) {
+  int32_t numOfTables = taosArrayGetSize(pTableList->pTableList);
+  if (startIndex >= numOfTables) {
+    return -1;
+  }
+
+  for (int32_t i = startIndex; i < numOfTables; ++i) {
+    STableKeyInfo* p = taosArrayGet(pTableList->pTableList, i);
+    if (p->uid == uid) {
+      return i;
+    }
+  }
+  return -1;
+}
+
 uint64_t getTableGroupId(const STableListInfo* pTableList, uint64_t tableUid) {
   int32_t* slot = taosHashGet(pTableList->map, &tableUid, sizeof(tableUid));
   ASSERT(pTableList->map != NULL && slot != NULL);