Merge branch '3.0' into 3.0test/jcy

Jenkinsfile2

@@ -130,6 +130,9 @@ def pre_test(){
 def pre_test_win(){
     bat '''
         hostname
+        taskkill /f /t /im python.exe
+        taskkill /f /t /im bash.exe
+        taskkill /f /t /im taosd.exe
         ipconfig
         set
         date /t

cmake/cmake.options

@@ -52,12 +52,22 @@ IF(${TD_WINDOWS})
         ON
     )
 ELSE ()
-
+    include(CheckCXXCompilerFlag)
-    option(
+    CHECK_CXX_COMPILER_FLAG("-std=c++13" COMPILER_SUPPORTS_CXX13)
-        BUILD_TEST 
+    IF(${COMPILER_SUPPORTS_CXX13})
-        "If build unit tests using googletest" 
+        add_definitions(-DCOMPILER_SUPPORTS_CXX13)
-        ON
+        option(
-    )
+            BUILD_TEST
+            "If build unit tests using googletest"
+            ON
+        )
+    ELSE ()
+        option(
+            BUILD_TEST
+            "If build unit tests using googletest"
+            OFF
+        )
+    ENDIF ()
 ENDIF ()
 
 option(

docs/en/14-reference/04-taosadapter.md

@@ -206,8 +206,8 @@ Note: InfluxDB token authorization is not supported at present. Only Basic autho
 You can use any client that supports the http protocol to access the RESTful interface address `http://<fqdn>:6041/<APIEndPoint>` to write data in OpenTSDB compatible format to TDengine.
 
 ```text
-/opentsdb/v1/put/json/:db
+/opentsdb/v1/put/json/<db>
-/opentsdb/v1/put/telnet/:db
+/opentsdb/v1/put/telnet/<db>
 ```
 
 ### collectd

docs/zh/14-reference/04-taosadapter.md

@@ -207,8 +207,8 @@ AllowWebSockets
 您可以使用任何支持 http 协议的客户端访问 Restful 接口地址 `http://<fqdn>:6041/<APIEndPoint>` 来写入 OpenTSDB 兼容格式的数据到 TDengine。EndPoint 如下：
 
 ```text
-/opentsdb/v1/put/json/:db
+/opentsdb/v1/put/json/<db>
-/opentsdb/v1/put/telnet/:db
+/opentsdb/v1/put/telnet/<db>
 ```
 
 ### collectd

examples/c/tmq.c

@@ -199,7 +199,7 @@ tmq_t* build_consumer() {
   tmq_conf_set(conf, "msg.with.table.name", "true");
   tmq_conf_set(conf, "enable.auto.commit", "true");
 
-  tmq_conf_set(conf, "experiment.use.snapshot", "false");
+  tmq_conf_set(conf, "experimental.snapshot.enable", "true");
 
   tmq_conf_set_auto_commit_cb(conf, tmq_commit_cb_print, NULL);
   tmq_t* tmq = tmq_consumer_new(conf, NULL, 0);

include/common/tdatablock.h

@@ -222,6 +222,7 @@ void blockDataCleanup(SSDataBlock* pDataBlock);
 size_t blockDataGetCapacityInRow(const SSDataBlock* pBlock, size_t pageSize);
 
 int32_t blockDataTrimFirstNRows(SSDataBlock* pBlock, size_t n);
+int32_t blockDataKeepFirstNRows(SSDataBlock* pBlock, size_t n);
 
 int32_t      assignOneDataBlock(SSDataBlock* dst, const SSDataBlock* src);
 int32_t      copyDataBlock(SSDataBlock* dst, const SSDataBlock* src);
@@ -235,6 +236,7 @@ SColumnInfoData* bdGetColumnInfoData(SSDataBlock* pBlock, int32_t index);
 void blockEncode(const SSDataBlock* pBlock, char* data, int32_t* dataLen, int32_t numOfCols, int8_t needCompress);
 const char* blockDecode(SSDataBlock* pBlock, int32_t numOfCols, int32_t numOfRows, const char* pData);
 
+void blockDebugShowDataBlock(SSDataBlock* pBlock, const char* flag);
 void blockDebugShowDataBlocks(const SArray* dataBlocks, const char* flag);
 // for debug
 char* dumpBlockData(SSDataBlock* pDataBlock, const char* flag, char** dumpBuf);

include/common/tdataformat.h

@@ -78,6 +78,7 @@ int32_t tEncodeTag(SEncoder *pEncoder, const STag *pTag);
 int32_t tDecodeTag(SDecoder *pDecoder, STag **ppTag);
 int32_t tTagToValArray(const STag *pTag, SArray **ppArray);
 void    debugPrintSTag(STag *pTag, const char *tag, int32_t ln);  // TODO: remove
+int32_t parseJsontoTagData(const char* json, SArray* pTagVals, STag** ppTag, void* pMsgBuf);
 
 // STRUCT =================
 struct STColumn {

include/common/tmsg.h

@@ -436,15 +436,16 @@ typedef struct {
   int32_t ttl;
   int32_t numOfColumns;
   int32_t numOfTags;
+  int32_t numOfFuncs;
   int32_t commentLen;
   int32_t ast1Len;
   int32_t ast2Len;
   SArray* pColumns;  // array of SField
   SArray* pTags;     // array of SField
-  char*   comment;
+  SArray* pFuncs;
+  char*   pComment;
   char*   pAst1;
   char*   pAst2;
-  SArray* pFuncs;
 } SMCreateStbReq;
 
 int32_t tSerializeSMCreateStbReq(void* buf, int32_t bufLen, SMCreateStbReq* pReq);
@@ -1495,6 +1496,7 @@ typedef struct SSubQueryMsg {
   uint64_t queryId;
   uint64_t taskId;
   int64_t  refId;
+  int32_t  execId;
   int8_t   taskType;
   int8_t   explain;
   uint32_t sqlLen;  // the query sql,
@@ -1514,6 +1516,7 @@ typedef struct {
   uint64_t sId;
   uint64_t queryId;
   uint64_t taskId;
+  int32_t  execId;
 } SQueryContinueReq;
 
 typedef struct {
@@ -1535,6 +1538,7 @@ typedef struct {
   uint64_t sId;
   uint64_t queryId;
   uint64_t taskId;
+  int32_t  execId;
 } SResFetchReq;
 
 typedef struct {
@@ -1546,6 +1550,7 @@ typedef struct {
   uint64_t queryId;
   uint64_t taskId;
   int64_t  refId;
+  int32_t  execId;
   int8_t   status;
 } STaskStatus;
 
@@ -1591,6 +1596,7 @@ typedef struct {
   uint64_t queryId;
   uint64_t taskId;
   int64_t  refId;
+  int32_t  execId;
 } STaskCancelReq;
 
 typedef struct {
@@ -1603,6 +1609,7 @@ typedef struct {
   uint64_t queryId;
   uint64_t taskId;
   int64_t  refId;
+  int32_t  execId;
 } STaskDropReq;
 
 typedef struct {
@@ -1623,6 +1630,7 @@ typedef struct {
   int8_t  triggerType;
   int64_t maxDelay;
   int64_t watermark;
+  int8_t  igExpired;
 } SCMCreateStreamReq;
 
 typedef struct {

include/common/tmsgdef.h

@@ -167,10 +167,6 @@ enum {
 
   TD_NEW_MSG_SEG(TDMT_VND_MSG)
   TD_DEF_MSG_TYPE(TDMT_VND_SUBMIT, "submit", SSubmitReq, SSubmitRsp)
-  TD_DEF_MSG_TYPE(TDMT_VND_QUERY, "query", NULL, NULL)
-  TD_DEF_MSG_TYPE(TDMT_VND_QUERY_CONTINUE, "query-continue", NULL, NULL)
-  TD_DEF_MSG_TYPE(TDMT_VND_QUERY_HEARTBEAT, "query-heartbeat", NULL, NULL)
-  TD_DEF_MSG_TYPE(TDMT_VND_FETCH, "fetch", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_VND_CREATE_TABLE, "create-table", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_VND_ALTER_TABLE, "alter-table", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_VND_DROP_TABLE, "drop-table", NULL, NULL)
@@ -184,12 +180,9 @@ enum {
   TD_DEF_MSG_TYPE(TDMT_VND_MQ_VG_CHANGE, "vnode-mq-vg-change", SMqRebVgReq, SMqRebVgRsp)
   TD_DEF_MSG_TYPE(TDMT_VND_MQ_VG_DELETE, "vnode-mq-vg-delete", SMqVDeleteReq, SMqVDeleteRsp)
   TD_DEF_MSG_TYPE(TDMT_VND_MQ_COMMIT_OFFSET, "vnode-commit-offset", STqOffset, STqOffset)
-  TD_DEF_MSG_TYPE(TDMT_VND_CANCEL_TASK, "vnode-cancel-task", NULL, NULL)
-  TD_DEF_MSG_TYPE(TDMT_VND_DROP_TASK, "vnode-drop-task", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_VND_CREATE_TOPIC, "vnode-create-topic", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_VND_ALTER_TOPIC, "vnode-alter-topic", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_VND_DROP_TOPIC, "vnode-drop-topic", NULL, NULL)
-  TD_DEF_MSG_TYPE(TDMT_VND_EXPLAIN, "vnode-explain", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_VND_SUBSCRIBE, "vnode-subscribe", SMVSubscribeReq, SMVSubscribeRsp)
   TD_DEF_MSG_TYPE(TDMT_VND_CONSUME, "vnode-consume", SMqPollReq, SMqDataBlkRsp)
   TD_DEF_MSG_TYPE(TDMT_VND_STREAM_TRIGGER, "vnode-stream-trigger", NULL, NULL)
@@ -206,6 +199,17 @@ enum {
   TD_DEF_MSG_TYPE(TDMT_VND_COMPACT, "compact", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_VND_DROP_TTL_TABLE, "drop-ttl-stb", NULL, NULL)
 
+  TD_NEW_MSG_SEG(TDMT_SCH_MSG)
+  TD_DEF_MSG_TYPE(TDMT_SCH_QUERY, "query", NULL, NULL)
+  TD_DEF_MSG_TYPE(TDMT_SCH_MERGE_QUERY, "merge-query", NULL, NULL)
+  TD_DEF_MSG_TYPE(TDMT_SCH_QUERY_CONTINUE, "query-continue", NULL, NULL)
+  TD_DEF_MSG_TYPE(TDMT_SCH_QUERY_HEARTBEAT, "query-heartbeat", NULL, NULL)
+  TD_DEF_MSG_TYPE(TDMT_SCH_FETCH, "fetch", NULL, NULL)
+  TD_DEF_MSG_TYPE(TDMT_SCH_CANCEL_TASK, "cancel-task", NULL, NULL)
+  TD_DEF_MSG_TYPE(TDMT_SCH_DROP_TASK, "drop-task", NULL, NULL)
+  TD_DEF_MSG_TYPE(TDMT_SCH_EXPLAIN, "explain", NULL, NULL)
+  TD_DEF_MSG_TYPE(TDMT_SCH_LINK_BROKEN, "link-broken", NULL, NULL)
+
   TD_NEW_MSG_SEG(TDMT_STREAM_MSG)
   TD_DEF_MSG_TYPE(TDMT_STREAM_TASK_DEPLOY, "stream-task-deploy", SStreamTaskDeployReq, SStreamTaskDeployRsp)
   TD_DEF_MSG_TYPE(TDMT_STREAM_TASK_DROP, "stream-task-drop", NULL, NULL)
@@ -214,9 +218,6 @@ enum {
   TD_DEF_MSG_TYPE(TDMT_STREAM_TASK_RECOVER, "stream-task-recover", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_STREAM_RETRIEVE, "stream-retrieve", NULL, NULL)
 
-  TD_NEW_MSG_SEG(TDMT_SCH_MSG)
-  TD_DEF_MSG_TYPE(TDMT_SCH_LINK_BROKEN, "link-broken", NULL, NULL)
-
   TD_NEW_MSG_SEG(TDMT_MON_MSG)
   TD_DEF_MSG_TYPE(TDMT_MON_MM_INFO, "monitor-minfo", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_MON_VM_INFO, "monitor-vinfo", NULL, NULL)
@@ -232,10 +233,12 @@ enum {
   TD_DEF_MSG_TYPE(TDMT_SYNC_PING, "sync-ping", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_SYNC_PING_REPLY, "sync-ping-reply", NULL, NULL)
   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)
   TD_DEF_MSG_TYPE(TDMT_SYNC_CLIENT_REQUEST_REPLY, "sync-client-request-reply", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_SYNC_REQUEST_VOTE, "sync-request-vote", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_SYNC_REQUEST_VOTE_REPLY, "sync-request-vote-reply", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_SYNC_APPEND_ENTRIES, "sync-append-entries", NULL, NULL)
+  TD_DEF_MSG_TYPE(TDMT_SYNC_APPEND_ENTRIES_BATCH, "sync-append-entries-batch", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_SYNC_APPEND_ENTRIES_REPLY, "sync-append-entries-reply", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_SYNC_NOOP, "sync-noop", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_SYNC_UNKNOWN, "sync-unknown", NULL, NULL)

include/common/ttokendef.h

@@ -192,79 +192,81 @@
 #define TK_TRIGGER        174
 #define TK_AT_ONCE        175
 #define TK_WINDOW_CLOSE   176
-#define TK_KILL           177
+#define TK_IGNORE         177
-#define TK_CONNECTION     178
+#define TK_EXPIRED        178
-#define TK_TRANSACTION    179
+#define TK_KILL           179
-#define TK_BALANCE        180
+#define TK_CONNECTION     180
-#define TK_VGROUP         181
+#define TK_TRANSACTION    181
-#define TK_MERGE          182
+#define TK_BALANCE        182
-#define TK_REDISTRIBUTE   183
+#define TK_VGROUP         183
-#define TK_SPLIT          184
+#define TK_MERGE          184
-#define TK_SYNCDB         185
+#define TK_REDISTRIBUTE   185
-#define TK_DELETE         186
+#define TK_SPLIT          186
-#define TK_NULL           187
+#define TK_SYNCDB         187
-#define TK_NK_QUESTION    188
+#define TK_DELETE         188
-#define TK_NK_ARROW       189
+#define TK_NULL           189
-#define TK_ROWTS          190
+#define TK_NK_QUESTION    190
-#define TK_TBNAME         191
+#define TK_NK_ARROW       191
-#define TK_QSTARTTS       192
+#define TK_ROWTS          192
-#define TK_QENDTS         193
+#define TK_TBNAME         193
-#define TK_WSTARTTS       194
+#define TK_QSTARTTS       194
-#define TK_WENDTS         195
+#define TK_QENDTS         195
-#define TK_WDURATION      196
+#define TK_WSTARTTS       196
-#define TK_CAST           197
+#define TK_WENDTS         197
-#define TK_NOW            198
+#define TK_WDURATION      198
-#define TK_TODAY          199
+#define TK_CAST           199
-#define TK_TIMEZONE       200
+#define TK_NOW            200
-#define TK_CLIENT_VERSION 201
+#define TK_TODAY          201
-#define TK_SERVER_VERSION 202
+#define TK_TIMEZONE       202
-#define TK_SERVER_STATUS  203
+#define TK_CLIENT_VERSION 203
-#define TK_CURRENT_USER   204
+#define TK_SERVER_VERSION 204
-#define TK_COUNT          205
+#define TK_SERVER_STATUS  205
-#define TK_LAST_ROW       206
+#define TK_CURRENT_USER   206
-#define TK_BETWEEN        207
+#define TK_COUNT          207
-#define TK_IS             208
+#define TK_LAST_ROW       208
-#define TK_NK_LT          209
+#define TK_BETWEEN        209
-#define TK_NK_GT          210
+#define TK_IS             210
-#define TK_NK_LE          211
+#define TK_NK_LT          211
-#define TK_NK_GE          212
+#define TK_NK_GT          212
-#define TK_NK_NE          213
+#define TK_NK_LE          213
-#define TK_MATCH          214
+#define TK_NK_GE          214
-#define TK_NMATCH         215
+#define TK_NK_NE          215
-#define TK_CONTAINS       216
+#define TK_MATCH          216
-#define TK_JOIN           217
+#define TK_NMATCH         217
-#define TK_INNER          218
+#define TK_CONTAINS       218
-#define TK_SELECT         219
+#define TK_JOIN           219
-#define TK_DISTINCT       220
+#define TK_INNER          220
-#define TK_WHERE          221
+#define TK_SELECT         221
-#define TK_PARTITION      222
+#define TK_DISTINCT       222
-#define TK_BY             223
+#define TK_WHERE          223
-#define TK_SESSION        224
+#define TK_PARTITION      224
-#define TK_STATE_WINDOW   225
+#define TK_BY             225
-#define TK_SLIDING        226
+#define TK_SESSION        226
-#define TK_FILL           227
+#define TK_STATE_WINDOW   227
-#define TK_VALUE          228
+#define TK_SLIDING        228
-#define TK_NONE           229
+#define TK_FILL           229
-#define TK_PREV           230
+#define TK_VALUE          230
-#define TK_LINEAR         231
+#define TK_NONE           231
-#define TK_NEXT           232
+#define TK_PREV           232
-#define TK_HAVING         233
+#define TK_LINEAR         233
-#define TK_RANGE          234
+#define TK_NEXT           234
-#define TK_EVERY          235
+#define TK_HAVING         235
-#define TK_ORDER          236
+#define TK_RANGE          236
-#define TK_SLIMIT         237
+#define TK_EVERY          237
-#define TK_SOFFSET        238
+#define TK_ORDER          238
-#define TK_LIMIT          239
+#define TK_SLIMIT         239
-#define TK_OFFSET         240
+#define TK_SOFFSET        240
-#define TK_ASC            241
+#define TK_LIMIT          241
-#define TK_NULLS          242
+#define TK_OFFSET         242
-#define TK_ID             243
+#define TK_ASC            243
-#define TK_NK_BITNOT      244
+#define TK_NULLS          244
-#define TK_INSERT         245
+#define TK_ID             245
-#define TK_VALUES         246
+#define TK_NK_BITNOT      246
-#define TK_IMPORT         247
+#define TK_INSERT         247
-#define TK_NK_SEMI        248
+#define TK_VALUES         248
-#define TK_FILE           249
+#define TK_IMPORT         249
+#define TK_NK_SEMI        250
+#define TK_FILE           251
 
 #define TK_NK_SPACE   300
 #define TK_NK_COMMENT 301

include/libs/executor/executor.h

@@ -36,6 +36,7 @@ typedef struct SReadHandle {
   void*   vnode;
   void*   mnd;
   SMsgCb* pMsgCb;
+  bool    tqReader;
 } SReadHandle;
 
 typedef enum {
@@ -133,7 +134,6 @@ int32_t qKillTask(qTaskInfo_t tinfo);
  */
 int32_t qAsyncKillTask(qTaskInfo_t tinfo);
 
-
 /**
  * destroy query info structure
  * @param qHandle
@@ -172,6 +172,8 @@ int32_t qDeserializeTaskStatus(qTaskInfo_t tinfo, const char* pInput, int32_t le
  */
 int32_t qGetStreamScanStatus(qTaskInfo_t tinfo, uint64_t* uid, int64_t* ts);
 
+int32_t qStreamPrepareScan(qTaskInfo_t tinfo, uint64_t uid, int64_t ts);
+
 #ifdef __cplusplus
 }
 #endif

include/libs/function/functionMgt.h

@@ -34,6 +34,7 @@ typedef enum EFunctionType {
   FUNCTION_TYPE_ELAPSED,
   FUNCTION_TYPE_IRATE,
   FUNCTION_TYPE_LAST_ROW,
+  FUNCTION_TYPE_LAST_ROWT, //TODO: removed
   FUNCTION_TYPE_MAX,
   FUNCTION_TYPE_MIN,
   FUNCTION_TYPE_MODE,
@@ -195,6 +196,7 @@ bool fmIsIntervalInterpoFunc(int32_t funcId);
 bool fmIsInterpFunc(int32_t funcId);
 bool fmIsLastRowFunc(int32_t funcId);
 bool fmIsSystemInfoFunc(int32_t funcId);
+bool fmIsImplicitTsFunc(int32_t funcId);
 
 int32_t fmGetDistMethod(const SFunctionNode* pFunc, SFunctionNode** pPartialFunc, SFunctionNode** pMergeFunc);
 

include/libs/nodes/cmdnodes.h

@@ -338,6 +338,7 @@ typedef struct SStreamOptions {
   int8_t    triggerType;
   SNode*    pDelay;
   SNode*    pWatermark;
+  bool      ignoreExpired;
 } SStreamOptions;
 
 typedef struct SCreateStreamStmt {

include/libs/nodes/plannodes.h

@@ -73,8 +73,7 @@ typedef struct SScanLogicNode {
   SNode*        pTagIndexCond;
   int8_t        triggerType;
   int64_t       watermark;
-  int16_t       tsColId;
+  int8_t        igExpired;
-  double        filesFactor;
   SArray*       pSmaIndexes;
   SNodeList*    pGroupTags;
   bool          groupSort;
@@ -105,6 +104,7 @@ typedef struct SIndefRowsFuncLogicNode {
   SNodeList* pFuncs;
   bool       isTailFunc;
   bool       isUniqueFunc;
+  bool       isTimeLineFunc;
 } SIndefRowsFuncLogicNode;
 
 typedef struct SInterpFuncLogicNode {
@@ -175,7 +175,7 @@ typedef struct SWindowLogicNode {
   SNode*           pStateExpr;
   int8_t           triggerType;
   int64_t          watermark;
-  double           filesFactor;
+  int8_t           igExpired;
   EWindowAlgorithm windowAlgo;
 } SWindowLogicNode;
 
@@ -296,8 +296,7 @@ typedef struct STableScanPhysiNode {
   int8_t         slidingUnit;
   int8_t         triggerType;
   int64_t        watermark;
-  int16_t        tsColId;
+  int8_t         igExpired;
-  double         filesFactor;
 } STableScanPhysiNode;
 
 typedef STableScanPhysiNode STableSeqScanPhysiNode;
@@ -347,6 +346,7 @@ typedef struct SDownstreamSourceNode {
   SQueryNodeAddr addr;
   uint64_t       taskId;
   uint64_t       schedId;
+  int32_t        execId;
 } SDownstreamSourceNode;
 
 typedef struct SExchangePhysiNode {
@@ -373,7 +373,7 @@ typedef struct SWinodwPhysiNode {
   SNode*     pTsEnd;  // window end timestamp
   int8_t     triggerType;
   int64_t    watermark;
-  double     filesFactor;
+  int8_t     igExpired;
 } SWinodwPhysiNode;
 
 typedef struct SIntervalPhysiNode {

include/libs/nodes/querynodes.h

@@ -251,7 +251,7 @@ typedef struct SSelectStmt {
   char        stmtName[TSDB_TABLE_NAME_LEN];
   uint8_t     precision;
   bool        isEmptyResult;
-  bool        isTimeOrderQuery;
+  bool        isTimeLineResult;
   bool        hasAggFuncs;
   bool        hasRepeatScanFuncs;
   bool        hasIndefiniteRowsFunc;
@@ -261,6 +261,7 @@ typedef struct SSelectStmt {
   bool        hasTailFunc;
   bool        hasInterpFunc;
   bool        hasLastRowFunc;
+  bool        hasTimeLineFunc;
   bool        groupSort;
 } SSelectStmt;
 

include/libs/planner/planner.h

@@ -34,6 +34,7 @@ typedef struct SPlanContext {
   bool        showRewrite;
   int8_t      triggerType;
   int64_t     watermark;
+  int8_t      igExpired;
   char*       pMsg;
   int32_t     msgLen;
   const char* pUser;

include/libs/qcom/query.h

@@ -135,9 +135,11 @@ typedef struct STableMetaOutput {
 } STableMetaOutput;
 
 typedef struct SDataBuf {
+  int32_t  msgType;
   void*    pData;
   uint32_t len;
   void*    handle;
+  SEpSet*  pEpSet;
 } SDataBuf;
 
 typedef struct STargetInfo {
@@ -146,7 +148,7 @@ typedef struct STargetInfo {
   int32_t     vgId;
 } STargetInfo;
 
-typedef int32_t (*__async_send_cb_fn_t)(void* param, const SDataBuf* pMsg, int32_t code);
+typedef int32_t (*__async_send_cb_fn_t)(void* param, SDataBuf* pMsg, int32_t code);
 typedef int32_t (*__async_exec_fn_t)(void* param);
 
 typedef struct SRequestConnInfo {
@@ -234,13 +236,18 @@ extern int32_t (*queryProcessMsgRsp[TDMT_MAX])(void* output, char* msg, int32_t
 #define NEED_CLIENT_HANDLE_ERROR(_code)                                          \
   (NEED_CLIENT_RM_TBLMETA_ERROR(_code) || NEED_CLIENT_REFRESH_VG_ERROR(_code) || \
    NEED_CLIENT_REFRESH_TBLMETA_ERROR(_code))
+#define NEED_REDIRECT_ERROR(_code)                                                      \
+  ((_code) == TSDB_CODE_RPC_REDIRECT || (_code) == TSDB_CODE_RPC_NETWORK_UNAVAIL ||     \
+   (_code) == TSDB_CODE_NODE_NOT_DEPLOYED || (_code) == TSDB_CODE_SYN_NOT_LEADER ||     \
+   (_code) == TSDB_CODE_APP_NOT_READY || (_code) == TSDB_CODE_RPC_BROKEN_LINK)
+  
 #define NEED_CLIENT_RM_TBLMETA_REQ(_type)                                                                  \
   ((_type) == TDMT_VND_CREATE_TABLE || (_type) == TDMT_VND_CREATE_STB || (_type) == TDMT_VND_DROP_TABLE || \
    (_type) == TDMT_VND_DROP_STB)
 
-#define NEED_SCHEDULER_RETRY_ERROR(_code)                                           \
+#define NEED_SCHEDULER_REDIRECT_ERROR(_code)                                                      \
-  ((_code) == TSDB_CODE_RPC_REDIRECT || (_code) == TSDB_CODE_RPC_NETWORK_UNAVAIL || \
+  ((_code) == TSDB_CODE_RPC_REDIRECT || (_code) == TSDB_CODE_NODE_NOT_DEPLOYED ||                 \
-   (_code) == TSDB_CODE_SCH_TIMEOUT_ERROR)
+   (_code) == TSDB_CODE_SYN_NOT_LEADER || (_code) == TSDB_CODE_APP_NOT_READY)
 
 #define REQUEST_TOTAL_EXEC_TIMES 2
 

include/libs/sync/sync.h

@@ -26,6 +26,7 @@ extern "C" {
 
 extern bool gRaftDetailLog;
 
+#define SYNC_MAX_BATCH_SIZE 100
 #define SYNC_INDEX_BEGIN 0
 #define SYNC_INDEX_INVALID -1
 #define SYNC_TERM_INVALID 0xFFFFFFFFFFFFFFFF
@@ -120,7 +121,7 @@ typedef struct SSyncFSM {
   int32_t (*FpGetSnapshot)(struct SSyncFSM* pFsm, SSnapshot* pSnapshot, void* pReaderParam, void** ppReader);
   int32_t (*FpGetSnapshotInfo)(struct SSyncFSM* pFsm, SSnapshot* pSnapshot);
 
-  int32_t (*FpSnapshotStartRead)(struct SSyncFSM* pFsm, void** ppReader);
+  int32_t (*FpSnapshotStartRead)(struct SSyncFSM* pFsm, void* pReaderParam, void** ppReader);
   int32_t (*FpSnapshotStopRead)(struct SSyncFSM* pFsm, void* pReader);
   int32_t (*FpSnapshotDoRead)(struct SSyncFSM* pFsm, void* pReader, void** ppBuf, int32_t* len);
 
@@ -164,6 +165,7 @@ typedef struct SSyncLogStore {
   bool (*syncLogIsEmpty)(struct SSyncLogStore* pLogStore);
   int32_t (*syncLogEntryCount)(struct SSyncLogStore* pLogStore);
   int32_t (*syncLogRestoreFromSnapshot)(struct SSyncLogStore* pLogStore, SyncIndex index);
+  bool (*syncLogExist)(struct SSyncLogStore* pLogStore, SyncIndex index);
 
   SyncIndex (*syncLogWriteIndex)(struct SSyncLogStore* pLogStore);
   SyncIndex (*syncLogLastIndex)(struct SSyncLogStore* pLogStore);
@@ -179,6 +181,7 @@ typedef struct SSyncInfo {
   bool        isStandBy;
   bool        snapshotEnable;
   SyncGroupId vgId;
+  int32_t     batchSize;
   SSyncCfg    syncCfg;
   char        path[TSDB_FILENAME_LEN];
   SWal*       pWal;
@@ -202,6 +205,7 @@ SyncGroupId syncGetVgId(int64_t rid);
 void        syncGetEpSet(int64_t rid, SEpSet* pEpSet);
 void        syncGetRetryEpSet(int64_t rid, SEpSet* pEpSet);
 int32_t     syncPropose(int64_t rid, SRpcMsg* pMsg, bool isWeak);
+// int32_t     syncProposeBatch(int64_t rid, SRpcMsg* pMsgArr, bool* pIsWeakArr, int32_t arrSize);
 bool        syncEnvIsStart();
 const char* syncStr(ESyncState state);
 bool        syncIsRestoreFinish(int64_t rid);

include/libs/sync/syncTools.h

@@ -219,6 +219,34 @@ void syncClientRequestPrint2(char* s, const SyncClientRequest* pMsg);
 void syncClientRequestLog(const SyncClientRequest* pMsg);
 void syncClientRequestLog2(char* s, const SyncClientRequest* pMsg);
 
+// ---------------------------------------------
+typedef struct SOffsetAndContLen {
+  int32_t offset;
+  int32_t contLen;
+} SOffsetAndContLen;
+
+typedef struct SRaftMeta {
+  uint64_t seqNum;
+  bool     isWeak;
+} SRaftMeta;
+
+// block1:
+// block2: SRaftMeta array
+// block3: rpc msg array (with pCont)
+
+typedef struct SyncClientRequestBatch {
+  uint32_t bytes;
+  int32_t  vgId;
+  uint32_t msgType;  // SyncClientRequestBatch msgType
+  uint32_t dataCount;
+  uint32_t dataLen;  // user RpcMsg.contLen
+  char     data[];   // user RpcMsg.pCont
+} SyncClientRequestBatch;
+
+SyncClientRequestBatch* syncClientRequestBatchBuild(SRpcMsg* rpcMsgArr, SRaftMeta* raftArr, int32_t arrSize,
+                                                    int32_t vgId);
+void                    syncClientRequestBatch2RpcMsg(const SyncClientRequestBatch* pSyncMsg, SRpcMsg* pRpcMsg);
+
 // ---------------------------------------------
 typedef struct SyncClientRequestReply {
   uint32_t bytes;
@@ -325,22 +353,58 @@ void syncAppendEntriesLog(const SyncAppendEntries* pMsg);
 void syncAppendEntriesLog2(char* s, const SyncAppendEntries* pMsg);
 
 // ---------------------------------------------
+
+// define ahead
+/*
+typedef struct SOffsetAndContLen {
+  int32_t offset;
+  int32_t contLen;
+} SOffsetAndContLen;
+*/
+
+// block1: SOffsetAndContLen
+// block2: SOffsetAndContLen Array
+// block3: SRpcMsg Array
+// block4: SRpcMsg pCont Array
+
 typedef struct SyncAppendEntriesBatch {
   uint32_t bytes;
   int32_t  vgId;
   uint32_t msgType;
   SRaftId  srcId;
   SRaftId  destId;
+
   // private data
   SyncTerm  term;
   SyncIndex prevLogIndex;
   SyncTerm  prevLogTerm;
   SyncIndex commitIndex;
   SyncTerm  privateTerm;
+  int32_t   dataCount;
   uint32_t  dataLen;
   char      data[];
 } SyncAppendEntriesBatch;
 
+SyncAppendEntriesBatch* syncAppendEntriesBatchBuild(SRpcMsg* rpcMsgArr, int32_t arrSize, int32_t vgId);
+void                    syncAppendEntriesBatchDestroy(SyncAppendEntriesBatch* pMsg);
+void                    syncAppendEntriesBatchSerialize(const SyncAppendEntriesBatch* pMsg, char* buf, uint32_t bufLen);
+void                    syncAppendEntriesBatchDeserialize(const char* buf, uint32_t len, SyncAppendEntriesBatch* pMsg);
+char*                   syncAppendEntriesBatchSerialize2(const SyncAppendEntriesBatch* pMsg, uint32_t* len);
+SyncAppendEntriesBatch* syncAppendEntriesBatchDeserialize2(const char* buf, uint32_t len);
+void                    syncAppendEntriesBatch2RpcMsg(const SyncAppendEntriesBatch* pMsg, SRpcMsg* pRpcMsg);
+void                    syncAppendEntriesBatchFromRpcMsg(const SRpcMsg* pRpcMsg, SyncAppendEntriesBatch* pMsg);
+SyncAppendEntriesBatch* syncAppendEntriesBatchFromRpcMsg2(const SRpcMsg* pRpcMsg);
+cJSON*                  syncAppendEntriesBatch2Json(const SyncAppendEntriesBatch* pMsg);
+char*                   syncAppendEntriesBatch2Str(const SyncAppendEntriesBatch* pMsg);
+void syncAppendEntriesBatch2RpcMsgArray(SyncAppendEntriesBatch* pSyncMsg, SRpcMsg* rpcMsgArr, int32_t maxArrSize,
+                                        int32_t* pRetArrSize);
+
+// for debug ----------------------
+void syncAppendEntriesBatchPrint(const SyncAppendEntriesBatch* pMsg);
+void syncAppendEntriesBatchPrint2(char* s, const SyncAppendEntriesBatch* pMsg);
+void syncAppendEntriesBatchLog(const SyncAppendEntriesBatch* pMsg);
+void syncAppendEntriesBatchLog2(char* s, const SyncAppendEntriesBatch* pMsg);
+
 // ---------------------------------------------
 typedef struct SyncAppendEntriesReply {
   uint32_t bytes;
@@ -542,6 +606,7 @@ int32_t syncNodeOnPingCb(SSyncNode* ths, SyncPing* pMsg);
 int32_t syncNodeOnPingReplyCb(SSyncNode* ths, SyncPingReply* pMsg);
 int32_t syncNodeOnTimeoutCb(SSyncNode* ths, SyncTimeout* pMsg);
 int32_t syncNodeOnClientRequestCb(SSyncNode* ths, SyncClientRequest* pMsg, SyncIndex* pRetIndex);
+int32_t syncNodeOnClientRequestBatchCb(SSyncNode* ths, SyncClientRequestBatch* pMsg);
 int32_t syncNodeOnRequestVoteCb(SSyncNode* ths, SyncRequestVote* pMsg);
 int32_t syncNodeOnRequestVoteReplyCb(SSyncNode* ths, SyncRequestVoteReply* pMsg);
 int32_t syncNodeOnAppendEntriesCb(SSyncNode* ths, SyncAppendEntries* pMsg);

include/libs/wal/wal.h

@@ -61,45 +61,23 @@ extern "C" {
     }                                                             \
   }
 
-#define WAL_HEAD_VER     0
+#define WAL_PROTO_VER    0
 #define WAL_NOSUFFIX_LEN 20
 #define WAL_SUFFIX_AT    (WAL_NOSUFFIX_LEN + 1)
 #define WAL_LOG_SUFFIX   "log"
 #define WAL_INDEX_SUFFIX "idx"
 #define WAL_REFRESH_MS   1000
-#define WAL_MAX_SIZE     (TSDB_MAX_WAL_SIZE + sizeof(SWalHead))
+#define WAL_MAX_SIZE     (TSDB_MAX_WAL_SIZE + sizeof(SWalCkHead))
 #define WAL_PATH_LEN     (TSDB_FILENAME_LEN + 12)
 #define WAL_FILE_LEN     (WAL_PATH_LEN + 32)
 #define WAL_MAGIC        0xFAFBFCFDULL
 
-#pragma pack(push, 1)
 typedef enum {
   TAOS_WAL_NOLOG = 0,
   TAOS_WAL_WRITE = 1,
   TAOS_WAL_FSYNC = 2,
 } EWalType;
 
-// used by sync module
-typedef struct {
-  int8_t   isWeek;
-  uint64_t seqNum;
-  uint64_t term;
-} SSyncLogMeta;
-
-typedef struct SWalReadHead {
-  int8_t  headVer;
-  int8_t  reserved;
-  int16_t msgType;
-  int32_t bodyLen;
-  int64_t ingestTs;  // not implemented
-  int64_t version;
-
-  // sync meta
-  SSyncLogMeta syncMeta;
-
-  char body[];
-} SWalReadHead;
-
 typedef struct {
   int32_t  vgId;
   int32_t  fsyncPeriod;      // millisecond
@@ -110,13 +88,6 @@ typedef struct {
   EWalType level;  // wal level
 } SWalCfg;
 
-typedef struct {
-  uint64_t     magic;
-  uint32_t     cksumHead;
-  uint32_t     cksumBody;
-  SWalReadHead head;
-} SWalHead;
-
 typedef struct SWalVer {
   int64_t firstVer;
   int64_t verInSnapshotting;
@@ -125,6 +96,35 @@ typedef struct SWalVer {
   int64_t lastVer;
 } SWalVer;
 
+#pragma pack(push, 1)
+// used by sync module
+typedef struct {
+  int8_t   isWeek;
+  uint64_t seqNum;
+  uint64_t term;
+} SSyncLogMeta;
+
+typedef struct {
+  int8_t  protoVer;
+  int64_t version;
+  int16_t msgType;
+  int32_t bodyLen;
+  int64_t ingestTs;  // not implemented
+
+  // sync meta
+  SSyncLogMeta syncMeta;
+
+  char body[];
+} SWalCont;
+
+typedef struct {
+  uint64_t magic;
+  uint32_t cksumHead;
+  uint32_t cksumBody;
+  SWalCont head;
+} SWalCkHead;
+#pragma pack(pop)
+
 typedef struct SWal {
   // cfg
   SWalCfg cfg;
@@ -134,7 +134,7 @@ typedef struct SWal {
   TdFilePtr pWriteLogTFile;
   TdFilePtr pWriteIdxTFile;
   int32_t   writeCur;
-  SArray   *fileInfoSet;
+  SArray   *fileInfoSet;  // SArray<SWalFileInfo>
   // status
   int64_t totSize;
   int64_t lastRollSeq;
@@ -146,7 +146,7 @@ typedef struct SWal {
   // path
   char path[WAL_PATH_LEN];
   // reusable write head
-  SWalHead writeHead;
+  SWalCkHead writeHead;
 } SWal;  // WAL HANDLE
 
 typedef struct SWalReadHandle {
@@ -158,11 +158,8 @@ typedef struct SWalReadHandle {
   int64_t       capacity;
   int64_t       status;  // if cursor valid
   TdThreadMutex mutex;
-  SWalHead     *pHead;
+  SWalCkHead   *pHead;
 } SWalReadHandle;
-#pragma pack(pop)
-
-// typedef int32_t (*FWalWrite)(void *ahandle, void *pHead);
 
 // module initialization
 int32_t walInit();
@@ -174,9 +171,9 @@ int32_t walAlter(SWal *, SWalCfg *pCfg);
 void    walClose(SWal *);
 
 // write
-int64_t walWriteWithSyncInfo(SWal *, int64_t index, tmsg_t msgType, SSyncLogMeta syncMeta, const void *body,
+int32_t walWriteWithSyncInfo(SWal *, int64_t index, tmsg_t msgType, SSyncLogMeta syncMeta, const void *body,
                              int32_t bodyLen);
-int64_t walWrite(SWal *, int64_t index, tmsg_t msgType, const void *body, int32_t bodyLen);
+int32_t walWrite(SWal *, int64_t index, tmsg_t msgType, const void *body, int32_t bodyLen);
 void    walFsync(SWal *, bool force);
 
 // apis for lifecycle management
@@ -196,9 +193,9 @@ int32_t         walReadWithHandle(SWalReadHandle *pRead, int64_t ver);
 
 // only for tq usage
 void    walSetReaderCapacity(SWalReadHandle *pRead, int32_t capacity);
-int32_t walFetchHead(SWalReadHandle *pRead, int64_t ver, SWalHead *pHead);
+int32_t walFetchHead(SWalReadHandle *pRead, int64_t ver, SWalCkHead *pHead);
-int32_t walFetchBody(SWalReadHandle *pRead, SWalHead **ppHead);
+int32_t walFetchBody(SWalReadHandle *pRead, SWalCkHead **ppHead);
-int32_t walSkipFetchBody(SWalReadHandle *pRead, const SWalHead *pHead);
+int32_t walSkipFetchBody(SWalReadHandle *pRead, const SWalCkHead *pHead);
 
 typedef struct {
   int64_t refId;
@@ -210,6 +207,8 @@ void     walCloseRef(SWalRef *);
 int32_t  walRefVer(SWalRef *, int64_t ver);
 int32_t  walUnrefVer(SWal *);
 
+bool walLogExist(SWal *, int64_t ver);
+
 // lifecycle check
 bool    walIsEmpty(SWal *);
 int64_t walGetFirstVer(SWal *);

include/os/osFile.h

@@ -25,17 +25,17 @@ extern "C" {
 // If the error is in a third-party library, place this header file under the third-party library header file.
 // When you want to use this feature, you should find or add the same function in the following sectio
 #ifndef ALLOW_FORBID_FUNC
-    #define open OPEN_FUNC_TAOS_FORBID
+#define open    OPEN_FUNC_TAOS_FORBID
-    #define fopen FOPEN_FUNC_TAOS_FORBID
+#define fopen   FOPEN_FUNC_TAOS_FORBID
-    #define access ACCESS_FUNC_TAOS_FORBID
+#define access  ACCESS_FUNC_TAOS_FORBID
-    #define stat STAT_FUNC_TAOS_FORBID
+#define stat    STAT_FUNC_TAOS_FORBID
-    #define lstat LSTAT_FUNC_TAOS_FORBID
+#define lstat   LSTAT_FUNC_TAOS_FORBID
-    #define fstat FSTAT_FUNC_TAOS_FORBID
+#define fstat   FSTAT_FUNC_TAOS_FORBID
-    #define close CLOSE_FUNC_TAOS_FORBID
+#define close   CLOSE_FUNC_TAOS_FORBID
-    #define fclose FCLOSE_FUNC_TAOS_FORBID
+#define fclose  FCLOSE_FUNC_TAOS_FORBID
-    #define fsync FSYNC_FUNC_TAOS_FORBID
+#define fsync   FSYNC_FUNC_TAOS_FORBID
-    #define getline GETLINE_FUNC_TAOS_FORBID
+#define getline GETLINE_FUNC_TAOS_FORBID
-    // #define fflush FFLUSH_FUNC_TAOS_FORBID
+// #define fflush FFLUSH_FUNC_TAOS_FORBID
 #endif
 
 #ifndef PATH_MAX
@@ -43,54 +43,54 @@ extern "C" {
 #endif
 
 typedef struct TdFile *TdFilePtr;
- 
+
-#define TD_FILE_CREATE    0x0001
+#define TD_FILE_CREATE   0x0001
-#define TD_FILE_WRITE     0x0002
+#define TD_FILE_WRITE    0x0002
-#define TD_FILE_READ      0x0004
+#define TD_FILE_READ     0x0004
-#define TD_FILE_TRUNC     0x0008
+#define TD_FILE_TRUNC    0x0008
-#define TD_FILE_APPEND    0x0010
+#define TD_FILE_APPEND   0x0010
-#define TD_FILE_TEXT      0x0020
+#define TD_FILE_TEXT     0x0020
-#define TD_FILE_AUTO_DEL  0x0040
+#define TD_FILE_AUTO_DEL 0x0040
-#define TD_FILE_EXCL      0x0080
+#define TD_FILE_EXCL     0x0080
-#define TD_FILE_STREAM    0x0100   // Only support taosFprintfFile, taosGetLineFile, taosEOFFile
+#define TD_FILE_STREAM   0x0100  // Only support taosFprintfFile, taosGetLineFile, taosEOFFile
-TdFilePtr taosOpenFile(const char *path,int32_t tdFileOptions);
+TdFilePtr taosOpenFile(const char *path, int32_t tdFileOptions);
 
 #define TD_FILE_ACCESS_EXIST_OK 0x1
 #define TD_FILE_ACCESS_READ_OK  0x2
 #define TD_FILE_ACCESS_WRITE_OK 0x4
-bool    taosCheckAccessFile(const char *pathname, int mode);
+bool taosCheckAccessFile(const char *pathname, int mode);
- 
+
 int32_t taosLockFile(TdFilePtr pFile);
 int32_t taosUnLockFile(TdFilePtr pFile);
- 
+
 int32_t taosUmaskFile(int32_t maskVal);
- 
+
 int32_t taosStatFile(const char *path, int64_t *size, int32_t *mtime);
 int32_t taosDevInoFile(TdFilePtr pFile, int64_t *stDev, int64_t *stIno);
 int32_t taosFStatFile(TdFilePtr pFile, int64_t *size, int32_t *mtime);
 bool    taosCheckExistFile(const char *pathname);
- 
+
 int64_t taosLSeekFile(TdFilePtr pFile, int64_t offset, int32_t whence);
 int32_t taosFtruncateFile(TdFilePtr pFile, int64_t length);
 int32_t taosFsyncFile(TdFilePtr pFile);
- 
+
 int64_t taosReadFile(TdFilePtr pFile, void *buf, int64_t count);
 int64_t taosPReadFile(TdFilePtr pFile, void *buf, int64_t count, int64_t offset);
 int64_t taosWriteFile(TdFilePtr pFile, const void *buf, int64_t count);
 void    taosFprintfFile(TdFilePtr pFile, const char *format, ...);
 
-int64_t taosGetLineFile(TdFilePtr pFile, char ** __restrict ptrBuf);
+int64_t taosGetLineFile(TdFilePtr pFile, char **__restrict ptrBuf);
 int64_t taosGetsFile(TdFilePtr pFile, int32_t maxSize, char *__restrict buf);
 
 int32_t taosEOFFile(TdFilePtr pFile);
- 
+
-int64_t taosCloseFile(TdFilePtr *ppFile);
+int32_t taosCloseFile(TdFilePtr *ppFile);
- 
+
 int32_t taosRenameFile(const char *oldName, const char *newName);
 int64_t taosCopyFile(const char *from, const char *to);
 int32_t taosRemoveFile(const char *path);
- 
+
-void    taosGetTmpfilePath(const char *inputTmpDir, const char *fileNamePrefix, char *dstPath);
+void taosGetTmpfilePath(const char *inputTmpDir, const char *fileNamePrefix, char *dstPath);
 
 int64_t taosFSendFile(TdFilePtr pFileOut, TdFilePtr pFileIn, int64_t *offset, int64_t size);
 

include/os/osMath.h

@@ -60,6 +60,13 @@ extern "C" {
   })
 #endif
 
+#ifndef __COMPAR_FN_T
+#define __COMPAR_FN_T
+typedef int32_t (*__compar_fn_t)(const void *, const void *);
+#endif
+
+void taosSort(void* arr, int64_t sz, int64_t width, __compar_fn_t compar);
+
 #ifdef __cplusplus
 }
 #endif

include/util/taoserror.h

@@ -86,6 +86,7 @@ int32_t* taosGetErrno();
 #define TSDB_CODE_RPC_NETWORK_UNAVAIL           TAOS_DEF_ERROR_CODE(0, 0x0102)
 #define TSDB_CODE_RPC_FQDN_ERROR                TAOS_DEF_ERROR_CODE(0, 0x0103)
 #define TSDB_CODE_RPC_PORT_EADDRINUSE           TAOS_DEF_ERROR_CODE(0, 0x0104)
+#define TSDB_CODE_RPC_BROKEN_LINK               TAOS_DEF_ERROR_CODE(0, 0x0105)
 
 //client
 #define TSDB_CODE_TSC_INVALID_OPERATION         TAOS_DEF_ERROR_CODE(0, 0x0200)
@@ -387,6 +388,10 @@ int32_t* taosGetErrno();
 #define TSDB_CODE_QRY_TASK_MSG_ERROR            TAOS_DEF_ERROR_CODE(0, 0x0719)
 #define TSDB_CODE_QRY_JOB_FREED                 TAOS_DEF_ERROR_CODE(0, 0x071A)
 #define TSDB_CODE_QRY_TASK_STATUS_ERROR         TAOS_DEF_ERROR_CODE(0, 0x071B)
+//json
+#define TSDB_CODE_QRY_JSON_IN_ERROR             TAOS_DEF_ERROR_CODE(0, 0x071C)
+#define TSDB_CODE_QRY_JSON_NOT_SUPPORT_ERROR    TAOS_DEF_ERROR_CODE(0, 0x071D)
+#define TSDB_CODE_QRY_JSON_IN_GROUP_ERROR       TAOS_DEF_ERROR_CODE(0, 0x071E)
 
 // grant
 #define TSDB_CODE_GRANT_EXPIRED                 TAOS_DEF_ERROR_CODE(0, 0x0800)
@@ -422,6 +427,7 @@ int32_t* taosGetErrno();
 #define TSDB_CODE_SYN_RECONFIG_NOT_READY        TAOS_DEF_ERROR_CODE(0, 0x0910)
 #define TSDB_CODE_SYN_PROPOSE_NOT_READY         TAOS_DEF_ERROR_CODE(0, 0x0911)
 #define TSDB_CODE_SYN_STANDBY_NOT_READY         TAOS_DEF_ERROR_CODE(0, 0x0912)
+#define TSDB_CODE_SYN_BATCH_ERROR               TAOS_DEF_ERROR_CODE(0, 0x0913)
 #define TSDB_CODE_SYN_INTERNAL_ERROR            TAOS_DEF_ERROR_CODE(0, 0x09FF)
 
 // tq
@@ -571,6 +577,7 @@ int32_t* taosGetErrno();
 #define TSDB_CODE_PAR_GROUP_BY_NOT_ALLOWED_FUNC TAOS_DEF_ERROR_CODE(0, 0x265B)
 #define TSDB_CODE_PAR_INVALID_TABLE_OPTION      TAOS_DEF_ERROR_CODE(0, 0x265C)
 #define TSDB_CODE_PAR_INVALID_INTERP_CLAUSE     TAOS_DEF_ERROR_CODE(0, 0x265D)
+#define TSDB_CODE_PAR_NO_VALID_FUNC_IN_WIN      TAOS_DEF_ERROR_CODE(0, 0x265E)
 
 //planner
 #define TSDB_CODE_PLAN_INTERNAL_ERROR           TAOS_DEF_ERROR_CODE(0, 0x2700)
@@ -616,6 +623,7 @@ int32_t* taosGetErrno();
 #define TSDB_CODE_RSMA_INVALID_ENV               TAOS_DEF_ERROR_CODE(0, 0x3150)
 #define TSDB_CODE_RSMA_INVALID_STAT              TAOS_DEF_ERROR_CODE(0, 0x3151)
 #define TSDB_CODE_RSMA_QTASKINFO_CREATE          TAOS_DEF_ERROR_CODE(0, 0x3152)
+#define TSDB_CODE_RSMA_FILE_CORRUPTED            TAOS_DEF_ERROR_CODE(0, 0x3153)
 
 //index
 #define TSDB_CODE_INDEX_REBUILDING               TAOS_DEF_ERROR_CODE(0, 0x3200)

include/util/tutil.h

@@ -44,6 +44,8 @@ uint32_t ip2uint(const char *const ip_addr);
 void     taosIp2String(uint32_t ip, char *str);
 void     taosIpPort2String(uint32_t ip, uint16_t port, char *str);
 
+void *tmemmem(const char *haystack, int hlen, const char *needle, int nlen);
+
 static FORCE_INLINE void taosEncryptPass(uint8_t *inBuf, size_t inLen, char *target) {
   T_MD5_CTX context;
   tMD5Init(&context);
@@ -59,10 +61,10 @@ static FORCE_INLINE void taosEncryptPass_c(uint8_t *inBuf, size_t len, char *tar
   tMD5Final(&context);
   char buf[TSDB_PASSWORD_LEN + 1];
 
-  sprintf(buf, "%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x", context.digest[0],
+  sprintf(buf, "%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x", context.digest[0], context.digest[1],
-          context.digest[1], context.digest[2], context.digest[3], context.digest[4], context.digest[5],
+          context.digest[2], context.digest[3], context.digest[4], context.digest[5], context.digest[6],
-          context.digest[6], context.digest[7], context.digest[8], context.digest[9], context.digest[10],
+          context.digest[7], context.digest[8], context.digest[9], context.digest[10], context.digest[11],
-          context.digest[11], context.digest[12], context.digest[13], context.digest[14], context.digest[15]);
+          context.digest[12], context.digest[13], context.digest[14], context.digest[15]);
   memcpy(target, buf, TSDB_PASSWORD_LEN);
 }
 

packaging/cfg/tarbitratord.service

@@ -1,20 +0,0 @@
-[Unit]
-Description=TDengine arbitrator service
-After=network-online.target
-Wants=network-online.target
-
-[Service]
-Type=simple
-ExecStart=/usr/bin/tarbitrator
-TimeoutStopSec=1000000s
-LimitNOFILE=infinity
-LimitNPROC=infinity
-LimitCORE=infinity
-TimeoutStartSec=0
-StandardOutput=null
-Restart=always
-StartLimitBurst=3
-StartLimitInterval=60s
-
-[Install]
-WantedBy=multi-user.target

packaging/deb/tarbitratord

@@ -1,88 +0,0 @@
-#!/bin/bash 
-#
-# Modified from original source: Elastic Search
-# https://github.com/elasticsearch/elasticsearch
-# Thank you to the Elastic Search authors
-#
-# chkconfig: 2345 99 01
-#
-### BEGIN INIT INFO
-# Provides:          taoscluster
-# Required-Start:    $local_fs $network $syslog
-# Required-Stop:     $local_fs $network $syslog
-# Default-Start:     2 3 4 5
-# Default-Stop:      0 1 6
-# Short-Description: Starts taoscluster tarbitrator
-# Description:       Starts taoscluster tarbitrator, a arbitrator
-### END INIT INFO
-
-set -e
-
-PATH="/bin:/usr/bin:/sbin:/usr/sbin"
-NAME="taoscluster"
-USER="root"
-GROUP="root"
-DAEMON="/usr/local/taos/bin/tarbitrator"
-DAEMON_OPTS=""
-PID_FILE="/var/run/$NAME.pid"
-APPARGS=""
-
-# Maximum number of open files
-MAX_OPEN_FILES=65535
-
-. /lib/lsb/init-functions
-
-case "$1" in
-    start)
-
-        log_action_begin_msg "Starting tarbitrator..."
-        if start-stop-daemon --test --start --chuid "$USER:$GROUP" --background --make-pidfile --pidfile "$PID_FILE" --exec "$DAEMON" -- $APPARGS &> /dev/null; then
-
-            touch "$PID_FILE" && chown "$USER":"$GROUP" "$PID_FILE"
-
-            if [ -n "$MAX_OPEN_FILES" ]; then
-                ulimit -n $MAX_OPEN_FILES
-            fi
-
-            start-stop-daemon --start --chuid "$USER:$GROUP" --background --make-pidfile --pidfile "$PID_FILE" --exec "$DAEMON" -- $APPARGS
-
-            log_end_msg $?
-        fi
-        ;;
-
-    stop)
-        log_action_begin_msg "Stopping tarbitrator..."
-        set +e
-        if [ -f "$PID_FILE" ]; then
-            start-stop-daemon --stop --pidfile "$PID_FILE" --user "$USER" --retry=TERM/120/KILL/5 > /dev/null
-            if [ $? -eq 1 ]; then
-                log_action_cont_msg "TSD is not running but pid file exists, cleaning up"
-            elif [ $? -eq 3 ]; then
-                PID="`cat $PID_FILE`"
-                log_failure_msg "Failed to stop tarbitrator (pid $PID)"
-                exit 1
-            fi
-            rm -f "$PID_FILE"
-        else
-            log_action_cont_msg "tarbitrator was not running"
-        fi
-        log_action_end_msg 0
-        set -e
-        ;;
-
-    restart|force-reload)
-        if [ -f "$PID_FILE" ]; then
-            $0 stop
-            sleep 1
-        fi
-        $0 start
-        ;;
-    status)
-        status_of_proc -p "$PID_FILE" "$DAEMON" "$NAME"
-        ;;
-    *)
-        exit 1
-        ;;
-esac
-
-exit 0

packaging/release.sh

@@ -302,7 +302,6 @@ if [ "$osType" != "Darwin" ]; then
 
   ${csudo}./makepkg.sh ${compile_dir} ${verNumber} "${build_time}" ${cpuType} ${osType} ${verMode} ${verType} ${pagMode} ${verNumberComp} ${dbName}
   ${csudo}./makeclient.sh ${compile_dir} ${verNumber} "${build_time}" ${cpuType} ${osType} ${verMode} ${verType} ${pagMode} ${dbName}
-  ${csudo}./makearbi.sh ${compile_dir} ${verNumber} "${build_time}" ${cpuType} ${osType} ${verMode} ${verType} ${pagMode}
 
 else
   # only make client for Darwin

packaging/rpm/tarbitratord

@@ -1,141 +0,0 @@
-#!/bin/bash
-#
-# tarbitratord      This shell script takes care of starting and stopping tarbitrator.
-#
-# chkconfig: 2345 99 01
-# description: tarbitrator is a arbitrator used in TDengine cluster.
-#
-#
-### BEGIN INIT INFO
-# Provides: taoscluster
-# Required-Start: $network $local_fs $remote_fs
-# Required-Stop: $network $local_fs $remote_fs
-# Short-Description: start and stop tarbitrator
-# Description: tarbitrator is a arbitrator used in TDengine cluster.
-### END INIT INFO
-
-# Source init functions
-. /etc/init.d/functions
-
-# Maximum number of open files
-MAX_OPEN_FILES=65535
-
-# Default program options
-NAME=tarbitrator
-PROG=/usr/local/taos/bin/tarbitrator
-USER=root
-GROUP=root
-
-# Default directories
-LOCK_DIR=/var/lock/subsys
-PID_DIR=/var/run/$NAME
-
-# Set file names
-LOCK_FILE=$LOCK_DIR/$NAME
-PID_FILE=$PID_DIR/$NAME.pid
-
-[ -e $PID_DIR ] || mkdir -p $PID_DIR
-
-PROG_OPTS=""
-
-start() {
-    echo -n "Starting ${NAME}: "
-    # check identity
-    curid="`id -u -n`"
-    if [ "$curid" != root ] && [ "$curid" != "$USER" ] ; then
-        echo "Must be run as root or $USER, but was run as $curid"
-        return 1
-    fi
-    # Sets the maximum number of open file descriptors allowed.
-    ulimit -n $MAX_OPEN_FILES
-    curulimit="`ulimit -n`"
-    if [ "$curulimit" -lt $MAX_OPEN_FILES ] ; then
-        echo "'ulimit -n' must be greater than or equal to $MAX_OPEN_FILES, is $curulimit"
-        return 1
-    fi
-
-    if [ "`id -u -n`" == root ] ; then
-        # Changes the owner of the lock, and the pid files to allow
-        # non-root OpenTSDB daemons to run /usr/share/opentsdb/bin/opentsdb_restart.py.
-        touch $LOCK_FILE && chown $USER:$GROUP $LOCK_FILE
-        touch $PID_FILE && chown $USER:$GROUP $PID_FILE
-        daemon --user $USER --pidfile $PID_FILE "$PROG $PROG_OPTS &> /dev/null &"
-    else
-        # Don't have to change user.
-        daemon --pidfile $PID_FILE "$PROG $PROG_OPTS &> /dev/null &"
-    fi
-    retval=$?
-    sleep 2
-    echo
-    [ $retval -eq 0 ] && (findproc > $PID_FILE && touch $LOCK_FILE)
-    return $retval
-}
-
-stop() {
-    echo -n "Stopping ${NAME}: "
-    killproc -p $PID_FILE $NAME
-    retval=$?
-    echo
-    # Non-root users don't have enough permission to remove pid and lock files.
-    # So, the opentsdb_restart.py cannot get rid of the files, and the command
-    # "service opentsdb status" will complain about the existing pid file.
-    # Makes the pid file empty.
-    echo > $PID_FILE
-    [ $retval -eq 0 ] && (rm -f $PID_FILE && rm -f $LOCK_FILE)
-    return $retval
-}
-
-restart() {
-    stop
-    start
-}
-
-reload() {
-    restart
-}
-
-force_reload() {
-    restart
-}
-
-rh_status() {
-    # run checks to determine if the service is running or use generic status
-    status -p $PID_FILE -l $LOCK_FILE $NAME
-}
-
-rh_status_q() {
-    rh_status >/dev/null 2>&1
-}
-
-case "$1" in
-    start)
-        rh_status_q && exit 0
-        $1
-        ;;
-    stop)
-        rh_status_q || exit 0
-        $1
-        ;;
-    restart)
-        $1
-        ;;
-    reload)
-        rh_status_q || exit 7
-        $1
-        ;;
-    force-reload)
-        force_reload
-        ;;
-    status)
-        rh_status
-        ;;
-    condrestart|try-restart)
-        rh_status_q || exit 0
-        restart
-        ;;
-    *)
-        echo "Usage: $0 {start|stop|status|restart|condrestart|try-restart|reload|force-reload}"
-        exit 2
-esac
-
-exit $?

packaging/tools/install.sh

@@ -194,7 +194,6 @@ function install_bin() {
   ${csudo}rm -f ${bin_link_dir}/${serverName} || :
   ${csudo}rm -f ${bin_link_dir}/${adapterName} || :
   ${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 || :
 
@@ -210,7 +209,6 @@ function install_bin() {
   [ -x ${install_main_dir}/bin/TDinsight.sh ] && ${csudo}ln -s ${install_main_dir}/bin/TDinsight.sh ${bin_link_dir}/TDinsight.sh || :
   [ -x ${install_main_dir}/bin/remove.sh ] && ${csudo}ln -s ${install_main_dir}/bin/remove.sh ${bin_link_dir}/${uninstallScript} || :
   [ -x ${install_main_dir}/bin/set_core.sh ] && ${csudo}ln -s ${install_main_dir}/bin/set_core.sh ${bin_link_dir}/set_core || :
-  [ -x ${install_main_dir}/bin/tarbitrator ] && ${csudo}ln -s ${install_main_dir}/bin/tarbitrator ${bin_link_dir}/tarbitrator || :
 
   if [ "$verMode" == "cluster" ]; then
     ${csudo}cp -r ${script_dir}/nginxd/* ${nginx_dir} && ${csudo}chmod 0555 ${nginx_dir}/*
@@ -606,28 +604,19 @@ function install_service_on_sysvinit() {
   if ((${os_type} == 1)); then
     #    ${csudo}cp -f ${script_dir}/init.d/${serverName}.deb ${install_main_dir}/init.d/${serverName}
     ${csudo}cp ${script_dir}/init.d/${serverName}.deb ${service_config_dir}/${serverName} && ${csudo}chmod a+x ${service_config_dir}/${serverName}
-    #    ${csudo}cp -f ${script_dir}/init.d/tarbitratord.deb ${install_main_dir}/init.d/tarbitratord
-    ${csudo}cp ${script_dir}/init.d/tarbitratord.deb ${service_config_dir}/tarbitratord && ${csudo}chmod a+x ${service_config_dir}/tarbitratord
   elif ((${os_type} == 2)); then
     #    ${csudo}cp -f ${script_dir}/init.d/${serverName}.rpm ${install_main_dir}/init.d/${serverName}
     ${csudo}cp ${script_dir}/init.d/${serverName}.rpm ${service_config_dir}/${serverName} && ${csudo}chmod a+x ${service_config_dir}/${serverName}
-    #    ${csudo}cp -f ${script_dir}/init.d/tarbitratord.rpm ${install_main_dir}/init.d/tarbitratord
-    ${csudo}cp ${script_dir}/init.d/tarbitratord.rpm ${service_config_dir}/tarbitratord && ${csudo}chmod a+x ${service_config_dir}/tarbitratord
   fi
 
   if ((${initd_mod} == 1)); then
     ${csudo}chkconfig --add ${serverName} || :
     ${csudo}chkconfig --level 2345 ${serverName} on || :
-    ${csudo}chkconfig --add tarbitratord || :
-    ${csudo}chkconfig --level 2345 tarbitratord on || :
   elif ((${initd_mod} == 2)); then
     ${csudo}insserv ${serverName} || :
     ${csudo}insserv -d ${serverName} || :
-    ${csudo}insserv tarbitratord || :
-    ${csudo}insserv -d tarbitratord || :
   elif ((${initd_mod} == 3)); then
     ${csudo}update-rc.d ${serverName} defaults || :
-    ${csudo}update-rc.d tarbitratord defaults || :
   fi
 }
 
@@ -669,9 +658,6 @@ function install_service_on_systemd() {
 
   ${csudo}systemctl enable ${serverName}
 
-  [ -f ${script_dir}/cfg/tarbitratord.service ] &&
-    ${csudo}cp ${script_dir}/cfg/tarbitratord.service \
-      ${service_config_dir}/ || :
   ${csudo}systemctl daemon-reload
 
   if [ "$verMode" == "cluster" ]; then

packaging/tools/install_arbi.sh

@@ -1,340 +0,0 @@
-#!/bin/bash
-#
-# This file is used to install database on linux systems. The operating system
-# is required to use systemd to manage services at boot
-
-set -e
-#set -x
-
-# -----------------------Variables definition---------------------
-script_dir=$(dirname $(readlink -f "$0"))
-
-bin_link_dir="/usr/bin"
-#inc_link_dir="/usr/include"
-
-#install main path
-install_main_dir="/usr/local/tarbitrator"
-
-# old bin dir
-bin_dir="/usr/local/tarbitrator/bin"
-
-service_config_dir="/etc/systemd/system"
-
-# Color setting
-RED='\033[0;31m'
-GREEN='\033[1;32m'
-GREEN_DARK='\033[0;32m'
-GREEN_UNDERLINE='\033[4;32m'
-NC='\033[0m'
-
-csudo=""
-if command -v sudo >/dev/null; then
-  csudo="sudo "
-fi
-
-update_flag=0
-
-initd_mod=0
-service_mod=2
-if pidof systemd &>/dev/null; then
-  service_mod=0
-elif $(which service &>/dev/null); then
-  service_mod=1
-  service_config_dir="/etc/init.d"
-  if $(which chkconfig &>/dev/null); then
-    initd_mod=1
-  elif $(which insserv &>/dev/null); then
-    initd_mod=2
-  elif $(which update-rc.d &>/dev/null); then
-    initd_mod=3
-  else
-    service_mod=2
-  fi
-else
-  service_mod=2
-fi
-
-# get the operating system type for using the corresponding init file
-# ubuntu/debian(deb), centos/fedora(rpm), others: opensuse, redhat, ..., no verification
-#osinfo=$(awk -F= '/^NAME/{print $2}' /etc/os-release)
-if [[ -e /etc/os-release ]]; then
-  osinfo=$(cat /etc/os-release | grep "NAME" | cut -d '"' -f2) || :
-else
-  osinfo=""
-fi
-#echo "osinfo: ${osinfo}"
-os_type=0
-if echo $osinfo | grep -qwi "ubuntu"; then
-  #  echo "This is ubuntu system"
-  os_type=1
-elif echo $osinfo | grep -qwi "debian"; then
-  #  echo "This is debian system"
-  os_type=1
-elif echo $osinfo | grep -qwi "Kylin"; then
-  #  echo "This is Kylin system"
-  os_type=1
-elif echo $osinfo | grep -qwi "centos"; then
-  #  echo "This is centos system"
-  os_type=2
-elif echo $osinfo | grep -qwi "fedora"; then
-  #  echo "This is fedora system"
-  os_type=2
-else
-  echo " osinfo: ${osinfo}"
-  echo " This is an officially unverified linux system,"
-  echo " if there are any problems with the installation and operation, "
-  echo " please feel free to contact taosdata.com for support."
-  os_type=1
-fi
-
-function kill_tarbitrator() {
-  pid=$(ps -ef | grep "tarbitrator" | grep -v "grep" | awk '{print $2}')
-  if [ -n "$pid" ]; then
-    ${csudo}kill -9 $pid || :
-  fi
-}
-
-function install_main_path() {
-  #create install main dir and all sub dir
-  ${csudo}rm -rf ${install_main_dir} || :
-  ${csudo}mkdir -p ${install_main_dir}
-  ${csudo}mkdir -p ${install_main_dir}/bin
-  #${csudo}mkdir -p ${install_main_dir}/include
-  ${csudo}mkdir -p ${install_main_dir}/init.d
-}
-
-function install_bin() {
-  # Remove links
-  ${csudo}rm -f ${bin_link_dir}/rmtarbitrator || :
-  ${csudo}rm -f ${bin_link_dir}/tarbitrator || :
-  ${csudo}cp -r ${script_dir}/bin/* ${install_main_dir}/bin && ${csudo}chmod 0555 ${install_main_dir}/bin/*
-
-  #Make link
-  [ -x ${install_main_dir}/bin/remove_arbi.sh ] && ${csudo}ln -s ${install_main_dir}/bin/remove_arbi.sh ${bin_link_dir}/rmtarbitrator || :
-  [ -x ${install_main_dir}/bin/tarbitrator ] && ${csudo}ln -s ${install_main_dir}/bin/tarbitrator ${bin_link_dir}/tarbitrator || :
-}
-
-function install_header() {
-  ${csudo}rm -f ${inc_link_dir}/taos.h ${inc_link_dir}/taosdef.h ${inc_link_dir}/taoserror.h ${inc_link_dir}/taosudf.h || :
-  ${csudo}cp -f ${script_dir}/inc/* ${install_main_dir}/include && ${csudo}chmod 644 ${install_main_dir}/include/*
-  ${csudo}ln -s ${install_main_dir}/include/taos.h ${inc_link_dir}/taos.h
-  ${csudo}ln -s ${install_main_dir}/include/taosdef.h ${inc_link_dir}/taosdef.h
-  ${csudo}ln -s ${install_main_dir}/include/taoserror.h ${inc_link_dir}/taoserror.h
-  ${csudo}ln -s ${install_main_dir}/include/taosudf.h ${inc_link_dir}/taosudf.h
-}
-
-function install_jemalloc() {
-  jemalloc_dir=${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 clean_service_on_sysvinit() {
-  if pidof tarbitrator &>/dev/null; then
-    ${csudo}service tarbitratord stop || :
-  fi
-
-  if ((${initd_mod} == 1)); then
-    if [ -e ${service_config_dir}/tarbitratord ]; then
-      ${csudo}chkconfig --del tarbitratord || :
-    fi
-  elif ((${initd_mod} == 2)); then
-    if [ -e ${service_config_dir}/tarbitratord ]; then
-      ${csudo}insserv -r tarbitratord || :
-    fi
-  elif ((${initd_mod} == 3)); then
-    if [ -e ${service_config_dir}/tarbitratord ]; then
-      ${csudo}update-rc.d -f tarbitratord remove || :
-    fi
-  fi
-
-  ${csudo}rm -f ${service_config_dir}/tarbitratord || :
-
-  if $(which init &>/dev/null); then
-    ${csudo}init q || :
-  fi
-}
-
-function install_service_on_sysvinit() {
-  clean_service_on_sysvinit
-  sleep 1
-
-  if ((${os_type} == 1)); then
-    ${csudo}cp -f ${script_dir}/init.d/tarbitratord.deb ${install_main_dir}/init.d/tarbitratord
-    ${csudo}cp ${script_dir}/init.d/tarbitratord.deb ${service_config_dir}/tarbitratord && ${csudo}chmod a+x ${service_config_dir}/tarbitratord
-  elif ((${os_type} == 2)); then
-    ${csudo}cp -f ${script_dir}/init.d/tarbitratord.rpm ${install_main_dir}/init.d/tarbitratord
-    ${csudo}cp ${script_dir}/init.d/tarbitratord.rpm ${service_config_dir}/tarbitratord && ${csudo}chmod a+x ${service_config_dir}/tarbitratord
-  fi
-
-  if ((${initd_mod} == 1)); then
-    ${csudo}chkconfig --add tarbitratord || :
-    ${csudo}chkconfig --level 2345 tarbitratord on || :
-  elif ((${initd_mod} == 2)); then
-    ${csudo}insserv tarbitratord || :
-    ${csudo}insserv -d tarbitratord || :
-  elif ((${initd_mod} == 3)); then
-    ${csudo}update-rc.d tarbitratord defaults || :
-  fi
-}
-
-function clean_service_on_systemd() {
-  tarbitratord_service_config="${service_config_dir}/tarbitratord.service"
-  if systemctl is-active --quiet tarbitratord; then
-    echo "tarbitrator is running, stopping it..."
-    ${csudo}systemctl stop tarbitratord &>/dev/null || echo &>/dev/null
-  fi
-  ${csudo}systemctl disable tarbitratord &>/dev/null || echo &>/dev/null
-
-  ${csudo}rm -f ${tarbitratord_service_config}
-}
-
-function install_service_on_systemd() {
-  clean_service_on_systemd
-
-  tarbitratord_service_config="${service_config_dir}/tarbitratord.service"
-
-  ${csudo}bash -c "echo '[Unit]'                                  >> ${tarbitratord_service_config}"
-  ${csudo}bash -c "echo 'Description=TDengine arbitrator service' >> ${tarbitratord_service_config}"
-  ${csudo}bash -c "echo 'After=network-online.target'             >> ${tarbitratord_service_config}"
-  ${csudo}bash -c "echo 'Wants=network-online.target'             >> ${tarbitratord_service_config}"
-  ${csudo}bash -c "echo                                           >> ${tarbitratord_service_config}"
-  ${csudo}bash -c "echo '[Service]'                               >> ${tarbitratord_service_config}"
-  ${csudo}bash -c "echo 'Type=simple'                             >> ${tarbitratord_service_config}"
-  ${csudo}bash -c "echo 'ExecStart=/usr/bin/tarbitrator'          >> ${tarbitratord_service_config}"
-  ${csudo}bash -c "echo 'TimeoutStopSec=1000000s'                 >> ${tarbitratord_service_config}"
-  ${csudo}bash -c "echo 'LimitNOFILE=infinity'                    >> ${tarbitratord_service_config}"
-  ${csudo}bash -c "echo 'LimitNPROC=infinity'                     >> ${tarbitratord_service_config}"
-  ${csudo}bash -c "echo 'LimitCORE=infinity'                      >> ${tarbitratord_service_config}"
-  ${csudo}bash -c "echo 'TimeoutStartSec=0'                       >> ${tarbitratord_service_config}"
-  ${csudo}bash -c "echo 'StandardOutput=null'                     >> ${tarbitratord_service_config}"
-  ${csudo}bash -c "echo 'Restart=always'                          >> ${tarbitratord_service_config}"
-  ${csudo}bash -c "echo 'StartLimitBurst=3'                       >> ${tarbitratord_service_config}"
-  ${csudo}bash -c "echo 'StartLimitInterval=60s'                  >> ${tarbitratord_service_config}"
-  ${csudo}bash -c "echo                                           >> ${tarbitratord_service_config}"
-  ${csudo}bash -c "echo '[Install]'                               >> ${tarbitratord_service_config}"
-  ${csudo}bash -c "echo 'WantedBy=multi-user.target'              >> ${tarbitratord_service_config}"
-  ${csudo}systemctl enable tarbitratord
-}
-
-function install_service() {
-  if ((${service_mod} == 0)); then
-    install_service_on_systemd
-  elif ((${service_mod} == 1)); then
-    install_service_on_sysvinit
-  else
-    kill_tarbitrator
-  fi
-}
-
-function update_TDengine() {
-  # Start to update
-  echo -e "${GREEN}Start to update TDengine's arbitrator ...${NC}"
-  # Stop the service if running
-  if pidof tarbitrator &>/dev/null; then
-    if ((${service_mod} == 0)); then
-      ${csudo}systemctl stop tarbitratord || :
-    elif ((${service_mod} == 1)); then
-      ${csudo}service tarbitratord stop || :
-    else
-      kill_tarbitrator
-    fi
-    sleep 1
-  fi
-
-  install_main_path
-  #install_header
-  install_bin
-  install_service
-  install_jemalloc
-
-  echo
-  if ((${service_mod} == 0)); then
-    echo -e "${GREEN_DARK}To start arbitrator     ${NC}: ${csudo}systemctl start tarbitratord${NC}"
-  elif ((${service_mod} == 1)); then
-    echo -e "${GREEN_DARK}To start arbitrator     ${NC}: ${csudo}service tarbitratord start${NC}"
-  else
-    echo -e "${GREEN_DARK}To start arbitrator     ${NC}: ./tarbitrator${NC}"
-  fi
-  echo
-  echo -e "\033[44;32;1mTDengine's arbitrator is updated successfully!${NC}"
-}
-
-function install_TDengine() {
-  # Start to install
-  echo -e "${GREEN}Start to install TDengine's arbitrator ...${NC}"
-
-  install_main_path
-  #install_header
-  install_bin
-  install_service
-  install_jemalloc
-
-  echo
-  if ((${service_mod} == 0)); then
-    echo -e "${GREEN_DARK}To start arbitrator     ${NC}: ${csudo}systemctl start tarbitratord${NC}"
-  elif ((${service_mod} == 1)); then
-    echo -e "${GREEN_DARK}To start arbitrator     ${NC}: ${csudo}service tarbitratord start${NC}"
-  else
-    echo -e "${GREEN_DARK}To start arbitrator     ${NC}: tarbitrator${NC}"
-  fi
-
-  echo -e "\033[44;32;1mTDengine's arbitrator is installed successfully!${NC}"
-  echo
-}
-
-## ==============================Main program starts from here============================
-# Install server and client
-if [ -x ${bin_dir}/tarbitrator ]; then
-  update_flag=1
-  update_TDengine
-else
-  install_TDengine
-fi

packaging/tools/make_install.sh

@@ -185,7 +185,6 @@ function install_bin() {
     [ -f ${binary_dir}/build/bin/taosadapter ] && ${csudo}cp -r ${binary_dir}/build/bin/taosadapter ${install_main_dir}/bin || :
     [ -f ${binary_dir}/build/bin/udfd ] && ${csudo}cp -r ${binary_dir}/build/bin/udfd ${install_main_dir}/bin || :
     ${csudo}cp -r ${binary_dir}/build/bin/${serverName} ${install_main_dir}/bin || :
-    # ${csudo}cp -r ${binary_dir}/build/bin/tarbitrator ${install_main_dir}/bin || :
 
     ${csudo}cp -r ${script_dir}/taosd-dump-cfg.gdb ${install_main_dir}/bin || :
     ${csudo}cp -r ${script_dir}/remove.sh ${install_main_dir}/bin || :

packaging/tools/makearbi.sh

@@ -1,71 +0,0 @@
-#!/bin/bash
-#
-# Generate arbitrator's tar.gz setup package for all os system
-
-set -e
-#set -x
-
-curr_dir=$(pwd)
-compile_dir=$1
-version=$2
-build_time=$3
-cpuType=$4
-osType=$5
-verMode=$6
-verType=$7
-pagMode=$8
-
-script_dir="$(dirname $(readlink -f $0))"
-top_dir="$(readlink -f ${script_dir}/../..)"
-
-productName="TDengine"
-
-# create compressed install file.
-build_dir="${compile_dir}/build"
-code_dir="${top_dir}"
-release_dir="${top_dir}/release"
-
-#package_name='linux'
-if [ "$verMode" == "cluster" ]; then
-  install_dir="${release_dir}/${productName}-enterprise-arbitrator-${version}"
-else
-  install_dir="${release_dir}/${productName}-arbitrator-${version}"
-fi
-
-# Directories and files.
-bin_files="${build_dir}/bin/tarbitrator ${script_dir}/remove_arbi.sh"
-install_files="${script_dir}/install_arbi.sh"
-
-#header_files="${code_dir}/include/client/taos.h ${code_dir}/include/common/taosdef.h ${code_dir}/include/util/taoserror.h ${code_dir}/include/libs/function/taosudf.h"
-init_file_tarbitrator_deb=${script_dir}/../deb/tarbitratord
-init_file_tarbitrator_rpm=${script_dir}/../rpm/tarbitratord
-
-# make directories.
-mkdir -p ${install_dir} && cp ${install_files} ${install_dir} && chmod a+x ${install_dir}/install_arbi.sh || :
-#mkdir -p ${install_dir}/inc && cp ${header_files} ${install_dir}/inc || :
-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_tarbitrator_deb} ${install_dir}/init.d/tarbitratord.deb || :
-mkdir -p ${install_dir}/init.d && cp ${init_file_tarbitrator_rpm} ${install_dir}/init.d/tarbitratord.rpm || :
-
-cd ${release_dir}
-
-#  install_dir has been distinguishes  cluster from  edege, so comments this code
-pkg_name=${install_dir}-${osType}-${cpuType}
-
-if [[ "$verType" == "beta" ]] || [[ "$verType" == "preRelease" ]]; then
-  pkg_name=${install_dir}-${verType}-${osType}-${cpuType}
-elif [ "$verType" == "stable" ]; then
-  pkg_name=${pkg_name}
-else
-  echo "unknow verType, nor stabel or beta"
-  exit 1
-fi
-
-tar -zcv -f "$(basename ${pkg_name}).tar.gz" $(basename ${install_dir}) --remove-files || :
-exitcode=$?
-if [ "$exitcode" != "0" ]; then
-  echo "tar ${pkg_name}.tar.gz error !!!"
-  exit $exitcode
-fi
-
-cd ${curr_dir}

packaging/tools/makepkg.sh

@@ -85,7 +85,6 @@ else
       ${build_dir}/bin/${clientName} \
       ${taostools_bin_files} \
       ${build_dir}/bin/taosadapter \
-      ${build_dir}/bin/tarbitrator\
       ${script_dir}/remove.sh \
       ${script_dir}/set_core.sh \
       ${script_dir}/startPre.sh \
@@ -106,8 +105,6 @@ nginx_dir="${top_dir}/../enterprise/src/plugins/web"
 
 init_file_deb=${script_dir}/../deb/taosd
 init_file_rpm=${script_dir}/../rpm/taosd
-init_file_tarbitrator_deb=${script_dir}/../deb/tarbitratord
-init_file_tarbitrator_rpm=${script_dir}/../rpm/tarbitratord
 
 # make directories.
 mkdir -p ${install_dir}
@@ -126,10 +123,6 @@ if [ -f "${cfg_dir}/${serverName}.service" ]; then
   cp ${cfg_dir}/${serverName}.service ${install_dir}/cfg || :
 fi
 
-if [ -f "${top_dir}/packaging/cfg/tarbitratord.service" ]; then
-  cp ${top_dir}/packaging/cfg/tarbitratord.service ${install_dir}/cfg || :
-fi
-
 if [ -f "${top_dir}/packaging/cfg/nginxd.service" ]; then
   cp ${top_dir}/packaging/cfg/nginxd.service ${install_dir}/cfg || :
 fi
@@ -137,8 +130,6 @@ 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}/init.d && cp ${init_file_tarbitrator_deb} ${install_dir}/init.d/tarbitratord.deb || :
-mkdir -p ${install_dir}/init.d && cp ${init_file_tarbitrator_rpm} ${install_dir}/init.d/tarbitratord.rpm || :
 
 if [ $adapterName != "taosadapter" ]; then
   mv ${install_dir}/cfg/taosadapter.toml ${install_dir}/cfg/$adapterName.toml

packaging/tools/remove.sh

@@ -1,6 +1,6 @@
 #!/bin/bash
 #
-# Script to stop the service and uninstall TDengine, but retain the config, data and log files.
+# Script to stop and uninstall the service, but retain the config, data and log files.
 
 set -e
 #set -x
@@ -11,121 +11,216 @@ RED='\033[0;31m'
 GREEN='\033[1;32m'
 NC='\033[0m'
 
+installDir="/usr/local/taos"
+serverName="taosd"
+clientName="taos"
+uninstallScript="rmtaos"
+productName="TDengine"
+
 #install main path
-install_main_dir="/usr/local/taos"
+install_main_dir=${installDir}
-data_link_dir="/usr/local/taos/data"
+data_link_dir=${installDir}/data
-log_link_dir="/usr/local/taos/log"
+log_link_dir=${installDir}/log
-cfg_link_dir="/usr/local/taos/cfg"
+cfg_link_dir=${installDir}/cfg
 bin_link_dir="/usr/bin"
+local_bin_link_dir="/usr/local/bin"
 lib_link_dir="/usr/lib"
 lib64_link_dir="/usr/lib64"
 inc_link_dir="/usr/include"
+install_nginxd_dir="/usr/local/nginxd"
 
 service_config_dir="/etc/systemd/system"
-taos_service_name="taosd"
+taos_service_name=${serverName}
-
+taosadapter_service_name="taosadapter"
+tarbitrator_service_name="tarbitratord"
+nginx_service_name="nginxd"
 csudo=""
-if command -v sudo > /dev/null; then
+if command -v sudo >/dev/null; then
-    csudo="sudo"
+  csudo="sudo "
 fi
 
 initd_mod=0
 service_mod=2
-if pidof systemd &> /dev/null; then
+if pidof systemd &>/dev/null; then
-    service_mod=0
+  service_mod=0
-elif $(which service &> /dev/null); then
+elif $(which service &>/dev/null); then
-    service_mod=1
+  service_mod=1
-    service_config_dir="/etc/init.d"
+  service_config_dir="/etc/init.d"
-    if $(which chkconfig &> /dev/null); then
+  if $(which chkconfig &>/dev/null); then
-         initd_mod=1
+    initd_mod=1
-    elif $(which insserv &> /dev/null); then
+  elif $(which insserv &>/dev/null); then
-        initd_mod=2
+    initd_mod=2
-    elif $(which update-rc.d &> /dev/null); then
+  elif $(which update-rc.d &>/dev/null); then
-        initd_mod=3
+    initd_mod=3
-    else
+  else
-        service_mod=2
-    fi
-else
     service_mod=2
+  fi
+else
+  service_mod=2
 fi
 
-
+function kill_taosadapter() {
-function kill_taosd() {
+  pid=$(ps -ef | grep "taosadapter" | grep -v "grep" | awk '{print $2}')
-  pid=$(ps -ef | grep "taosd" | grep -v "grep" | awk '{print $2}')
   if [ -n "$pid" ]; then
-    ${csudo} kill -9 $pid   || :
+    ${csudo}kill -9 $pid || :
   fi
 }
 
+function kill_taosd() {
+  pid=$(ps -ef | grep ${serverName} | grep -v "grep" | awk '{print $2}')
+  if [ -n "$pid" ]; then
+    ${csudo}kill -9 $pid || :
+  fi
+}
+
+function kill_tarbitrator() {
+  pid=$(ps -ef | grep "tarbitrator" | grep -v "grep" | awk '{print $2}')
+  if [ -n "$pid" ]; then
+    ${csudo}kill -9 $pid || :
+  fi
+}
 
 function clean_bin() {
-    # Remove link
+  # Remove link
-    ${csudo} rm -f ${bin_link_dir}/taos        || :
+  ${csudo}rm -f ${bin_link_dir}/${clientName} || :
-    ${csudo} rm -f ${bin_link_dir}/taosd       || :
+  ${csudo}rm -f ${bin_link_dir}/${serverName} || :
-    ${csudo} rm -f ${bin_link_dir}/create_table  || :
+  ${csudo}rm -f ${bin_link_dir}/taosadapter || :
-    ${csudo} rm -f ${bin_link_dir}/tmq_sim || :
+  ${csudo}rm -f ${bin_link_dir}/taosBenchmark || :
-    ${csudo} rm -f ${bin_link_dir}/taosdemo    || :
+  ${csudo}rm -f ${bin_link_dir}/taosdemo || :
-    ${csudo} rm -f ${bin_link_dir}/taosdump    || :
+  ${csudo}rm -f ${bin_link_dir}/taosdump || :
-    ${csudo} rm -f ${bin_link_dir}/rmtaos      || :
+  ${csudo}rm -f ${bin_link_dir}/${uninstallScript} || :
-    #${csudo} rm -f ${bin_link_dir}/set_core    || :
+  ${csudo}rm -f ${bin_link_dir}/tarbitrator || :
+  ${csudo}rm -f ${bin_link_dir}/set_core || :
+  ${csudo}rm -f ${bin_link_dir}/TDinsight.sh || :
+}
+
+function clean_local_bin() {
+  ${csudo}rm -f ${local_bin_link_dir}/taosBenchmark || :
+  ${csudo}rm -f ${local_bin_link_dir}/taosdemo || :
 }
 
 function clean_lib() {
-    # Remove link
+  # Remove link
-    ${csudo} rm -f ${lib_link_dir}/libtaos.*      || :
+  ${csudo}rm -f ${lib_link_dir}/libtaos.* || :
-    ${csudo} rm -f ${lib64_link_dir}/libtaos.*    || :    
+  ${csudo}rm -f ${lib64_link_dir}/libtaos.* || :
-    
+  #${csudo}rm -rf ${v15_java_app_dir}           || :
-    ${csudo} rm -f ${lib_link_dir}/libtdb.*      || :
-    ${csudo} rm -f ${lib64_link_dir}/libtdb.*    || :
 }
 
 function clean_header() {
-    # Remove link
+  # Remove link
-    ${csudo} rm -f ${inc_link_dir}/taos.h       || :
+  ${csudo}rm -f ${inc_link_dir}/taos.h || :
-    ${csudo} rm -f ${inc_link_dir}/taosdef.h    || :
+  ${csudo}rm -f ${inc_link_dir}/taosdef.h || :
-    ${csudo} rm -f ${inc_link_dir}/taoserror.h  || :
+  ${csudo}rm -f ${inc_link_dir}/taoserror.h || :
-    ${csudo} rm -f ${inc_link_dir}/taosudf.h  || :
 }
 
 function clean_config() {
-    # Remove link
+  # Remove link
-    ${csudo} rm -f ${cfg_link_dir}/*            || :
+  ${csudo}rm -f ${cfg_link_dir}/* || :
 }
 
 function clean_log() {
-    # Remove link
+  # Remove link
-    ${csudo} rm -rf ${log_link_dir}    || :
+  ${csudo}rm -rf ${log_link_dir} || :
 }
 
 function clean_service_on_systemd() {
-    taosd_service_config="${service_config_dir}/${taos_service_name}.service"
+  taosd_service_config="${service_config_dir}/${taos_service_name}.service"
-    if systemctl is-active --quiet ${taos_service_name}; then
+  if systemctl is-active --quiet ${taos_service_name}; then
-        echo "TDengine taosd is running, stopping it..."
+    echo "${productName} ${serverName} is running, stopping it..."
-        ${csudo} systemctl stop ${taos_service_name} &> /dev/null || echo &> /dev/null
+    ${csudo}systemctl stop ${taos_service_name} &>/dev/null || echo &>/dev/null
+  fi
+  ${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"
+  if systemctl is-active --quiet ${taosadapter_service_name}; then
+    echo "${productName} taosAdapter is running, stopping it..."
+    ${csudo}systemctl stop ${taosadapter_service_name} &>/dev/null || echo &>/dev/null
+  fi
+  ${csudo}systemctl disable ${taosadapter_service_name} &>/dev/null || echo &>/dev/null
+  [ -f ${taosadapter_service_config} ] && ${csudo}rm -f ${taosadapter_service_config}
+
+  tarbitratord_service_config="${service_config_dir}/${tarbitrator_service_name}.service"
+  if systemctl is-active --quiet ${tarbitrator_service_name}; then
+    echo "${productName} tarbitrator is running, stopping it..."
+    ${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 ${tarbitratord_service_config}
+
+  if [ "$verMode" == "cluster" ]; then
+    nginx_service_config="${service_config_dir}/${nginx_service_name}.service"
+    if [ -d ${install_nginxd_dir} ]; then
+      if systemctl is-active --quiet ${nginx_service_name}; then
+        echo "Nginx for ${productName} is running, stopping it..."
+        ${csudo}systemctl stop ${nginx_service_name} &>/dev/null || echo &>/dev/null
+      fi
+      ${csudo}systemctl disable ${nginx_service_name} &>/dev/null || echo &>/dev/null
+      ${csudo}rm -f ${nginx_service_config}
     fi
-    ${csudo} systemctl disable ${taos_service_name} &> /dev/null || echo &> /dev/null
+  fi
-    ${csudo} rm -f ${taosd_service_config}
 }
 
 function clean_service_on_sysvinit() {
-    echo " "
+  if pidof ${serverName} &>/dev/null; then
+    echo "${productName} ${serverName} is running, stopping it..."
+    ${csudo}service ${serverName} stop || :
+  fi
+
+  if pidof tarbitrator &>/dev/null; then
+    echo "${productName} tarbitrator is running, stopping it..."
+    ${csudo}service tarbitratord stop || :
+  fi
+
+  if ((${initd_mod} == 1)); then
+    if [ -e ${service_config_dir}/${serverName} ]; then
+      ${csudo}chkconfig --del ${serverName} || :
+    fi
+    if [ -e ${service_config_dir}/tarbitratord ]; then
+      ${csudo}chkconfig --del tarbitratord || :
+    fi
+  elif ((${initd_mod} == 2)); then
+    if [ -e ${service_config_dir}/${serverName} ]; then
+      ${csudo}insserv -r ${serverName} || :
+    fi
+    if [ -e ${service_config_dir}/tarbitratord ]; then
+      ${csudo}insserv -r tarbitratord || :
+    fi
+  elif ((${initd_mod} == 3)); then
+    if [ -e ${service_config_dir}/${serverName} ]; then
+      ${csudo}update-rc.d -f ${serverName} remove || :
+    fi
+    if [ -e ${service_config_dir}/tarbitratord ]; then
+      ${csudo}update-rc.d -f tarbitratord remove || :
+    fi
+  fi
+
+  ${csudo}rm -f ${service_config_dir}/${serverName} || :
+  ${csudo}rm -f ${service_config_dir}/tarbitratord || :
+
+  if $(which init &>/dev/null); then
+    ${csudo}init q || :
+  fi
 }
 
 function clean_service() {
-    if ((${service_mod}==0)); then
+  if ((${service_mod} == 0)); then
-        clean_service_on_systemd
+    clean_service_on_systemd
-    elif ((${service_mod}==1)); then
+  elif ((${service_mod} == 1)); then
-        clean_service_on_sysvinit
+    clean_service_on_sysvinit
-    else
+  else
-        # must manual stop taosd
+    kill_taosadapter
-        kill_taosd
+    kill_taosd
-    fi
+    kill_tarbitrator
+  fi
 }
 
 # Stop service and disable booting start.
 clean_service
 # Remove binary file and links
 clean_bin
+# Remove links of local bin
+clean_local_bin
 # Remove header file.
 clean_header
 # Remove lib file
@@ -135,15 +230,26 @@ clean_log
 # Remove link configuration file
 clean_config
 # Remove data link directory
-${csudo} rm -rf ${data_link_dir}    || :
+${csudo}rm -rf ${data_link_dir} || :
-
-${csudo} rm -rf ${install_main_dir}
 
+${csudo}rm -rf ${install_main_dir}
+${csudo}rm -rf ${install_nginxd_dir}
 if [[ -e /etc/os-release ]]; then
   osinfo=$(awk -F= '/^NAME/{print $2}' /etc/os-release)
 else
   osinfo=""
 fi
 
-echo -e "${GREEN}TDengine is removed successfully!${NC}"
+if echo $osinfo | grep -qwi "ubuntu"; then
+  #  echo "this is ubuntu system"
+  ${csudo}dpkg --force-all -P tdengine >/dev/null 2>&1 || :
+elif echo $osinfo | grep -qwi "debian"; then
+  #  echo "this is debian system"
+  ${csudo}dpkg --force-all -P tdengine >/dev/null 2>&1 || :
+elif echo $osinfo | grep -qwi "centos"; then
+  #  echo "this is centos system"
+  ${csudo}rpm -e --noscripts tdengine >/dev/null 2>&1 || :
+fi
+
+echo -e "${GREEN}${productName} is removed successfully!${NC}"
 echo

packaging/tools/remove_arbi.sh

@@ -1,132 +0,0 @@
-#!/bin/bash
-#
-# Script to stop the service and uninstall TDengine's arbitrator
-
-set -e
-#set -x
-
-verMode=edge
-
-RED='\033[0;31m'
-GREEN='\033[1;32m'
-NC='\033[0m'
-
-#install main path
-install_main_dir="/usr/local/tarbitrator"
-bin_link_dir="/usr/bin"
-#inc_link_dir="/usr/include"
-
-service_config_dir="/etc/systemd/system"
-tarbitrator_service_name="tarbitratord"
-csudo=""
-if command -v sudo > /dev/null; then
-    csudo="sudo "
-fi
-
-initd_mod=0
-service_mod=2
-if pidof systemd &> /dev/null; then
-    service_mod=0
-elif $(which service &> /dev/null); then    
-    service_mod=1
-    service_config_dir="/etc/init.d" 
-    if $(which chkconfig &> /dev/null); then
-         initd_mod=1 
-    elif $(which insserv &> /dev/null); then
-        initd_mod=2
-    elif $(which update-rc.d &> /dev/null); then
-        initd_mod=3
-    else
-        service_mod=2
-    fi
-else 
-    service_mod=2
-fi
-
-function kill_tarbitrator() {
-  pid=$(ps -ef | grep "tarbitrator" | grep -v "grep" | awk '{print $2}')
-  if [ -n "$pid" ]; then
-    ${csudo}kill -9 $pid   || :
-  fi
-}
-function clean_bin() {
-    # Remove link
-    ${csudo}rm -f ${bin_link_dir}/tarbitrator      || :
-}
-
-function clean_header() {
-    # Remove link
-    ${csudo}rm -f ${inc_link_dir}/taos.h       || :
-    ${csudo}rm -f ${inc_link_dir}/taosdef.h    || :
-    ${csudo}rm -f ${inc_link_dir}/taoserror.h  || :
-    ${csudo}rm -f ${inc_link_dir}/taosudf.h  || :
-
-}
-
-function clean_log() {
-    # Remove link
-    ${csudo}rm -rf /arbitrator.log    || :
-}
-
-function clean_service_on_systemd() {
-  tarbitratord_service_config="${service_config_dir}/${tarbitrator_service_name}.service"
-
-  if systemctl is-active --quiet ${tarbitrator_service_name}; then
-      echo "TDengine tarbitrator is running, stopping it..."
-      ${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}
-}
-
-function clean_service_on_sysvinit() {
-    if pidof tarbitrator &> /dev/null; then
-        echo "TDengine's tarbitrator is running, stopping it..."
-        ${csudo}service tarbitratord stop || :
-    fi
-    
-    if ((${initd_mod}==1)); then    
-      if [ -e ${service_config_dir}/tarbitratord ]; then 
-        ${csudo}chkconfig --del tarbitratord || :
-      fi
-    elif ((${initd_mod}==2)); then   
-      if [ -e ${service_config_dir}/tarbitratord ]; then 
-        ${csudo}insserv -r tarbitratord || :
-      fi
-    elif ((${initd_mod}==3)); then  
-      if [ -e ${service_config_dir}/tarbitratord ]; then 
-        ${csudo}update-rc.d -f tarbitratord remove || :
-      fi
-    fi
-
-    ${csudo}rm -f ${service_config_dir}/tarbitratord || :
-   
-    if $(which init &> /dev/null); then
-        ${csudo}init q || :
-    fi
-}
-
-function clean_service() {
-    if ((${service_mod}==0)); then
-        clean_service_on_systemd
-    elif ((${service_mod}==1)); then
-        clean_service_on_sysvinit
-    else
-        # must manual stop  
-        kill_tarbitrator
-    fi
-}
-
-# Stop service and disable booting start.
-clean_service
-# Remove binary file and links
-clean_bin
-# Remove header file.
-##clean_header
-# Remove log file
-clean_log
-
-${csudo}rm -rf ${install_main_dir}
-
-echo -e "${GREEN}TDengine's arbitrator is removed successfully!${NC}"

source/client/src/clientEnv.c

@@ -89,8 +89,10 @@ void closeTransporter(SAppInstInfo *pAppInfo) {
 }
 
 static bool clientRpcRfp(int32_t code, tmsg_t msgType) {
-  if (code == TSDB_CODE_RPC_REDIRECT || code == TSDB_CODE_RPC_NETWORK_UNAVAIL || code == TSDB_CODE_NODE_NOT_DEPLOYED ||
+  if (NEED_REDIRECT_ERROR(code)) {
-      code == TSDB_CODE_SYN_NOT_LEADER || code == TSDB_CODE_APP_NOT_READY) {
+    if (msgType == TDMT_SCH_QUERY || msgType == TDMT_SCH_MERGE_QUERY || msgType == TDMT_SCH_FETCH) {
+      return false;
+    }
     return true;
   } else {
     return false;

source/client/src/clientHb.c

@@ -262,7 +262,7 @@ static int32_t hbQueryHbRspHandle(SAppHbMgr *pAppHbMgr, SClientHbRsp *pRsp) {
   return TSDB_CODE_SUCCESS;
 }
 
-static int32_t hbAsyncCallBack(void *param, const SDataBuf *pMsg, int32_t code) {
+static int32_t hbAsyncCallBack(void *param, SDataBuf *pMsg, int32_t code) {
   static int32_t emptyRspNum = 0;
   if (code != 0) {
     taosMemoryFreeClear(param);

source/client/src/clientImpl.c

@@ -769,7 +769,8 @@ int32_t handleQueryExecRsp(SRequestObj* pRequest) {
       code = handleSubmitExecRes(pRequest, pRes->res, pCatalog, &epset);
       break;
     }
-    case TDMT_VND_QUERY: {
+    case TDMT_SCH_QUERY: 
+    case TDMT_SCH_MERGE_QUERY: {
       code = handleQueryExecRes(pRequest, pRes->res, pCatalog, &epset);
       break;
     }
@@ -785,6 +786,7 @@ int32_t handleQueryExecRsp(SRequestObj* pRequest) {
 void schedulerExecCb(SQueryResult* pResult, void* param, int32_t code) {
   SRequestObj* pRequest = (SRequestObj*)param;
   pRequest->code = code;
+  pRequest->body.resInfo.execRes = pResult->res;
 
   if (TDMT_VND_SUBMIT == pRequest->type || TDMT_VND_DELETE == pRequest->type ||
       TDMT_VND_CREATE_TABLE == pRequest->type) {
@@ -796,6 +798,8 @@ void schedulerExecCb(SQueryResult* pResult, void* param, int32_t code) {
     }
   }
 
+  taosMemoryFree(pResult);
+
   tscDebug("0x%" PRIx64 " enter scheduler exec cb, code:%d - %s, reqId:0x%" PRIx64, pRequest->self, code,
            tstrerror(code), pRequest->requestId);
 
@@ -1255,10 +1259,10 @@ void processMsgFromServer(void* parent, SRpcMsg* pMsg, SEpSet* pEpSet) {
        */
       int32_t elapsed = pRequest->metric.rsp - pRequest->metric.start;
       if (pMsg->code == TSDB_CODE_SUCCESS) {
-        tscDebug("0x%" PRIx64 " message:%s, code:%s rspLen:%d, elapsed:%d ms, reqId:0x%" PRIx64, pRequest->self,
+        tscDebug("0x%" PRIx64 " rsp msg:%s, code:%s rspLen:%d, elapsed:%d ms, reqId:0x%" PRIx64, pRequest->self,
                  TMSG_INFO(pMsg->msgType), tstrerror(pMsg->code), pMsg->contLen, elapsed / 1000, pRequest->requestId);
       } else {
-        tscError("0x%" PRIx64 " SQL cmd:%s, code:%s rspLen:%d, elapsed time:%d ms, reqId:0x%" PRIx64, pRequest->self,
+        tscError("0x%" PRIx64 " rsp msg:%s, code:%s rspLen:%d, elapsed time:%d ms, reqId:0x%" PRIx64, pRequest->self,
                  TMSG_INFO(pMsg->msgType), tstrerror(pMsg->code), pMsg->contLen, elapsed / 1000, pRequest->requestId);
       }
 
@@ -1268,7 +1272,7 @@ void processMsgFromServer(void* parent, SRpcMsg* pMsg, SEpSet* pEpSet) {
 
   updateTargetEpSet(pSendInfo, pTscObj, pMsg, pEpSet);
 
-  SDataBuf buf = {.len = pMsg->contLen, .pData = NULL, .handle = pMsg->info.handle};
+  SDataBuf buf = {.msgType = pMsg->msgType, .len = pMsg->contLen, .pData = NULL, .handle = pMsg->info.handle, .pEpSet = pEpSet};
 
   if (pMsg->contLen > 0) {
     buf.pData = taosMemoryCalloc(1, pMsg->contLen);
@@ -1407,7 +1411,11 @@ void* doAsyncFetchRows(SRequestObj* pRequest, bool setupOneRowPtr, bool convertU
     }
 
     SSyncQueryParam* pParam = pRequest->body.param;
-
+    if (NULL == pParam) {
+      pParam = taosMemoryCalloc(1, sizeof(SSyncQueryParam));
+      tsem_init(&pParam->sem, 0, 0);
+    }
+    
     // convert ucs4 to native multi-bytes string
     pResultInfo->convertUcs4 = convertUcs4;
 

source/client/src/clientMsgHandler.c

@@ -26,7 +26,7 @@ static void setErrno(SRequestObj* pRequest, int32_t code) {
   terrno = code;
 }
 
-int32_t genericRspCallback(void* param, const SDataBuf* pMsg, int32_t code) {
+int32_t genericRspCallback(void* param, SDataBuf* pMsg, int32_t code) {
   SRequestObj* pRequest = param;
   setErrno(pRequest, code);
 
@@ -39,7 +39,7 @@ int32_t genericRspCallback(void* param, const SDataBuf* pMsg, int32_t code) {
   return code;
 }
 
-int32_t processConnectRsp(void* param, const SDataBuf* pMsg, int32_t code) {
+int32_t processConnectRsp(void* param, SDataBuf* pMsg, int32_t code) {
   SRequestObj* pRequest = param;
   if (code != TSDB_CODE_SUCCESS) {
     taosMemoryFree(pMsg->pData);
@@ -116,7 +116,7 @@ SMsgSendInfo* buildMsgInfoImpl(SRequestObj* pRequest) {
   return pMsgSendInfo;
 }
 
-int32_t processCreateDbRsp(void* param, const SDataBuf* pMsg, int32_t code) {
+int32_t processCreateDbRsp(void* param, SDataBuf* pMsg, int32_t code) {
   // todo rsp with the vnode id list
   SRequestObj* pRequest = param;
   taosMemoryFree(pMsg->pData);
@@ -132,7 +132,7 @@ int32_t processCreateDbRsp(void* param, const SDataBuf* pMsg, int32_t code) {
   return code;
 }
 
-int32_t processUseDbRsp(void* param, const SDataBuf* pMsg, int32_t code) {
+int32_t processUseDbRsp(void* param, SDataBuf* pMsg, int32_t code) {
   SRequestObj* pRequest = param;
 
   if (TSDB_CODE_MND_DB_NOT_EXIST == code) {
@@ -211,7 +211,7 @@ int32_t processUseDbRsp(void* param, const SDataBuf* pMsg, int32_t code) {
   return 0;
 }
 
-int32_t processCreateSTableRsp(void* param, const SDataBuf* pMsg, int32_t code) {
+int32_t processCreateSTableRsp(void* param, SDataBuf* pMsg, int32_t code) {
   assert(pMsg != NULL && param != NULL);
   SRequestObj* pRequest = param;
 
@@ -229,7 +229,7 @@ int32_t processCreateSTableRsp(void* param, const SDataBuf* pMsg, int32_t code)
   return code;
 }
 
-int32_t processDropDbRsp(void* param, const SDataBuf* pMsg, int32_t code) {
+int32_t processDropDbRsp(void* param, SDataBuf* pMsg, int32_t code) {
   SRequestObj* pRequest = param;
   if (code != TSDB_CODE_SUCCESS) {
     setErrno(pRequest, code);
@@ -250,7 +250,7 @@ int32_t processDropDbRsp(void* param, const SDataBuf* pMsg, int32_t code) {
   return code;
 }
 
-int32_t processAlterStbRsp(void* param, const SDataBuf* pMsg, int32_t code) {
+int32_t processAlterStbRsp(void* param, SDataBuf* pMsg, int32_t code) {
   SRequestObj* pRequest = param;
   if (code != TSDB_CODE_SUCCESS) {
     setErrno(pRequest, code);
@@ -356,7 +356,7 @@ static int32_t buildShowVariablesRsp(SArray* pVars, SRetrieveTableRsp** pRsp) {
   return TSDB_CODE_SUCCESS;
 }
 
-int32_t processShowVariablesRsp(void* param, const SDataBuf* pMsg, int32_t code) {
+int32_t processShowVariablesRsp(void* param, SDataBuf* pMsg, int32_t code) {
   SRequestObj* pRequest = param;
   if (code != TSDB_CODE_SUCCESS) {
     setErrno(pRequest, code);

source/client/src/clientSml.c

@@ -2456,10 +2456,10 @@ TAOS_RES* taos_schemaless_insert(TAOS* taos, char* lines[], int numLines, int pr
     return NULL;
   }
 
+  int    batchs = 0;
   pTscObj->schemalessType = 1;
   SSmlMsgBuf msg = {ERROR_MSG_BUF_DEFAULT_SIZE, request->msgBuf};
 
-  int    cnt = ceil(((double)numLines) / LINE_BATCH);
   Params params;
   params.request = request;
   tsem_init(&params.sem, 0, 0);
@@ -2496,7 +2496,16 @@ TAOS_RES* taos_schemaless_insert(TAOS* taos, char* lines[], int numLines, int pr
     goto end;
   }
 
-  for (int i = 0; i < cnt; ++i) {
+  if(protocol == TSDB_SML_JSON_PROTOCOL){
+    numLines = 1;
+  }else if(numLines <= 0){
+    request->code = TSDB_CODE_SML_INVALID_DATA;
+    smlBuildInvalidDataMsg(&msg, "line num is invalid", NULL);
+    goto end;
+  }
+
+  batchs = ceil(((double)numLines) / LINE_BATCH);
+  for (int i = 0; i < batchs; ++i) {
     SRequestObj* req = (SRequestObj*)createRequest(pTscObj, TSDB_SQL_INSERT);
     if(!req){
       request->code = TSDB_CODE_OUT_OF_MEMORY;

source/client/src/tmq.c

@@ -54,7 +54,8 @@ struct tmq_conf_t {
   int8_t   autoCommit;
   int8_t   resetOffset;
   int8_t   withTbName;
-  int8_t   useSnapshot;
+  int8_t   spEnable;
+  int32_t  spBatchSize;
   uint16_t port;
   int32_t  autoCommitInterval;
   char*    ip;
@@ -288,18 +289,23 @@ tmq_conf_res_t tmq_conf_set(tmq_conf_t* conf, const char* key, const char* value
     }
   }
 
-  if (strcmp(key, "experiment.use.snapshot") == 0) {
+  if (strcmp(key, "experimental.snapshot.enable") == 0) {
     if (strcmp(value, "true") == 0) {
-      conf->useSnapshot = true;
+      conf->spEnable = true;
       return TMQ_CONF_OK;
     } else if (strcmp(value, "false") == 0) {
-      conf->useSnapshot = false;
+      conf->spEnable = false;
       return TMQ_CONF_OK;
     } else {
       return TMQ_CONF_INVALID;
     }
   }
 
+  if (strcmp(key, "experimental.snapshot.batch.size") == 0) {
+    conf->spBatchSize = atoi(value);
+    return TMQ_CONF_OK;
+  }
+
   if (strcmp(key, "td.connect.ip") == 0) {
     conf->ip = strdup(value);
     return TMQ_CONF_OK;
@@ -378,7 +384,7 @@ int32_t tmqCommitCb(void* param, const SDataBuf* pMsg, int32_t code) {
 }
 #endif
 
-int32_t tmqCommitCb2(void* param, const SDataBuf* pBuf, int32_t code) {
+int32_t tmqCommitCb2(void* param, SDataBuf* pBuf, int32_t code) {
   SMqCommitCbParam2*   pParam = (SMqCommitCbParam2*)param;
   SMqCommitCbParamSet* pParamSet = (SMqCommitCbParamSet*)pParam->params;
   // push into array
@@ -813,7 +819,7 @@ void tmqClearUnhandleMsg(tmq_t* tmq) {
   }
 }
 
-int32_t tmqSubscribeCb(void* param, const SDataBuf* pMsg, int32_t code) {
+int32_t tmqSubscribeCb(void* param, SDataBuf* pMsg, int32_t code) {
   SMqSubscribeCbParam* pParam = (SMqSubscribeCbParam*)param;
   pParam->rspErr = code;
   /*tmq_t* tmq = pParam->tmq;*/
@@ -918,7 +924,7 @@ tmq_t* tmq_consumer_new(tmq_conf_t* conf, char* errstr, int32_t errstrLen) {
   strcpy(pTmq->clientId, conf->clientId);
   strcpy(pTmq->groupId, conf->groupId);
   pTmq->withTbName = conf->withTbName;
-  pTmq->useSnapshot = conf->useSnapshot;
+  pTmq->useSnapshot = conf->spEnable;
   pTmq->autoCommit = conf->autoCommit;
   pTmq->autoCommitInterval = conf->autoCommitInterval;
   pTmq->commitCb = conf->commitCb;
@@ -1067,7 +1073,7 @@ int32_t tmqGetSkipLogNum(tmq_message_t* tmq_message) {
 }
 #endif
 
-int32_t tmqPollCb(void* param, const SDataBuf* pMsg, int32_t code) {
+int32_t tmqPollCb(void* param, SDataBuf* pMsg, int32_t code) {
   SMqPollCbParam* pParam = (SMqPollCbParam*)param;
   SMqClientVg*    pVg = pParam->pVg;
   SMqClientTopic* pTopic = pParam->pTopic;
@@ -1324,7 +1330,7 @@ bool tmqUpdateEp(tmq_t* tmq, int32_t epoch, SMqAskEpRsp* pRsp) {
 }
 #endif
 
-int32_t tmqAskEpCb(void* param, const SDataBuf* pMsg, int32_t code) {
+int32_t tmqAskEpCb(void* param, SDataBuf* pMsg, int32_t code) {
   SMqAskEpCbParam* pParam = (SMqAskEpCbParam*)param;
   tmq_t*           tmq = pParam->tmq;
   int8_t           async = pParam->async;
@@ -1541,7 +1547,7 @@ SMqRspObj* tmqBuildRspFromWrapper(SMqPollRspWrapper* pWrapper) {
 }
 
 int32_t tmqPollImpl(tmq_t* tmq, int64_t timeout) {
-  /*printf("call poll\n");*/
+  /*tscDebug("call poll");*/
   for (int i = 0; i < taosArrayGetSize(tmq->clientTopics); i++) {
     SMqClientTopic* pTopic = taosArrayGet(tmq->clientTopics, i);
     for (int j = 0; j < taosArrayGetSize(pTopic->vgs); j++) {
@@ -1603,8 +1609,8 @@ int32_t tmqPollImpl(tmq_t* tmq, int64_t timeout) {
       int64_t transporterId = 0;
       /*printf("send poll\n");*/
 
-      char offsetFormatBuf[50];
+      char offsetFormatBuf[80];
-      tFormatOffset(offsetFormatBuf, 50, &pVg->currentOffsetNew);
+      tFormatOffset(offsetFormatBuf, 80, &pVg->currentOffsetNew);
       tscDebug("consumer %ld send poll to %s : vg %d, epoch %d, req offset %s, reqId %lu", tmq->consumerId,
                pTopic->topicName, pVg->vgId, tmq->epoch, offsetFormatBuf, pReq->reqId);
       /*printf("send vg %d %ld\n", pVg->vgId, pVg->currentOffset);*/
@@ -1702,6 +1708,7 @@ void* tmqHandleAllRsp(tmq_t* tmq, int64_t timeout, bool pollIfReset) {
 }
 
 TAOS_RES* tmq_consumer_poll(tmq_t* tmq, int64_t timeout) {
+  /*tscDebug("call poll1");*/
   void*   rspObj;
   int64_t startTime = taosGetTimestampMs();
 

source/common/src/tdatablock.c

@@ -716,7 +716,12 @@ int32_t dataBlockCompar(const void* p1, const void* p2, const void* param) {
 
     void* left1 = colDataGetData(pColInfoData, left);
     void* right1 = colDataGetData(pColInfoData, right);
-
+    if (pColInfoData->info.type == TSDB_DATA_TYPE_JSON) {
+      if (tTagIsJson(left1) || tTagIsJson(right1)) {
+        terrno = TSDB_CODE_QRY_JSON_NOT_SUPPORT_ERROR;
+        return 0;
+      }
+    }
     __compar_fn_t fn = getKeyComparFunc(pColInfoData->info.type, pOrder->order);
 
     int ret = fn(left1, right1);
@@ -890,9 +895,9 @@ int32_t blockDataSort(SSDataBlock* pDataBlock, SArray* pOrderInfo) {
         SBlockOrderInfo* pOrder = taosArrayGet(pOrderInfo, 0);
 
         int64_t p0 = taosGetTimestampUs();
-
+        
         __compar_fn_t fn = getKeyComparFunc(pColInfoData->info.type, pOrder->order);
-        qsort(pColInfoData->pData, pDataBlock->info.rows, pColInfoData->info.bytes, fn);
+        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);
@@ -919,6 +924,7 @@ int32_t blockDataSort(SSDataBlock* pDataBlock, SArray* pOrderInfo) {
   }
 
   taosqsort(index, rows, sizeof(int32_t), &helper, dataBlockCompar);
+  if(terrno) return terrno;
 
   int64_t p1 = taosGetTimestampUs();
 
@@ -1218,7 +1224,7 @@ int32_t assignOneDataBlock(SSDataBlock* dst, const SSDataBlock* src) {
   for (int32_t i = 0; i < numOfCols; ++i) {
     SColumnInfoData* pDst = taosArrayGet(dst->pDataBlock, i);
     SColumnInfoData* pSrc = taosArrayGet(src->pDataBlock, i);
-    if (pSrc->pData == NULL) {
+    if (pSrc->pData == NULL && (!IS_VAR_DATA_TYPE(pSrc->info.type))) {
       continue;
     }
 
@@ -1431,9 +1437,39 @@ static void doShiftBitmap(char* nullBitmap, size_t n, size_t total) {
   }
 }
 
+static int32_t colDataMoveVarData(SColumnInfoData* pColInfoData, size_t start, size_t end){
+  int32_t dataOffset = -1;
+  int32_t dataLen = 0;
+  int32_t beigin = start;
+  while(beigin < end){
+    int32_t offset = pColInfoData->varmeta.offset[beigin];
+    if(offset == -1) {
+      beigin++;
+      continue;
+    }
+    if(start != 0) {
+      pColInfoData->varmeta.offset[beigin] = dataLen;
+    }
+    char *data = pColInfoData->pData + offset;
+    if(dataOffset == -1) dataOffset = offset;   // mark the begin of data
+    int32_t type = pColInfoData->info.type;
+    if (type == TSDB_DATA_TYPE_JSON) {
+      dataLen += getJsonValueLen(data);
+    } else {
+      dataLen += varDataTLen(data);
+    }
+    beigin++;
+  }
+  if(dataOffset > 0){
+    memmove(pColInfoData->pData, pColInfoData->pData + dataOffset, dataLen);
+    memmove(pColInfoData->varmeta.offset, &pColInfoData->varmeta.offset[start], (end - start) * sizeof(int32_t));
+  }
+  return dataLen;
+}
+
 static void colDataTrimFirstNRows(SColumnInfoData* pColInfoData, size_t n, size_t total) {
   if (IS_VAR_DATA_TYPE(pColInfoData->info.type)) {
-    memmove(pColInfoData->varmeta.offset, &pColInfoData->varmeta.offset[n], (total - n) * sizeof(int32_t));
+    pColInfoData->varmeta.length = colDataMoveVarData(pColInfoData, n, total);
     memset(&pColInfoData->varmeta.offset[total - n], 0, n);
   } else {
     int32_t bytes = pColInfoData->info.bytes;
@@ -1461,6 +1497,33 @@ int32_t blockDataTrimFirstNRows(SSDataBlock* pBlock, size_t n) {
   return TSDB_CODE_SUCCESS;
 }
 
+static void colDataKeepFirstNRows(SColumnInfoData* pColInfoData, size_t n, size_t total) {
+  if (IS_VAR_DATA_TYPE(pColInfoData->info.type)) {
+    pColInfoData->varmeta.length = colDataMoveVarData(pColInfoData, 0, n);
+    memset(&pColInfoData->varmeta.offset[n], 0, total - n);
+  }
+}
+
+int32_t blockDataKeepFirstNRows(SSDataBlock* pBlock, size_t n) {
+  if (n == 0) {
+    blockDataCleanup(pBlock);
+    return TSDB_CODE_SUCCESS;
+  }
+
+  if (pBlock->info.rows <= n) {
+    return TSDB_CODE_SUCCESS;
+  } else {
+    size_t numOfCols = taosArrayGetSize(pBlock->pDataBlock);
+    for (int32_t i = 0; i < numOfCols; ++i) {
+      SColumnInfoData* pColInfoData = taosArrayGet(pBlock->pDataBlock, i);
+      colDataKeepFirstNRows(pColInfoData, n, pBlock->info.rows);
+    }
+
+    pBlock->info.rows = n;
+  }
+  return TSDB_CODE_SUCCESS;
+}
+
 int32_t tEncodeDataBlock(void** buf, const SSDataBlock* pBlock) {
   int64_t tbUid = pBlock->info.uid;
   int16_t numOfCols = taosArrayGetSize(pBlock->pDataBlock);
@@ -1605,6 +1668,13 @@ static char* formatTimestamp(char* buf, int64_t val, int precision) {
   return buf;
 }
 
+void blockDebugShowDataBlock(SSDataBlock* pBlock, const char* flag) {
+  SArray* dataBlocks = taosArrayInit(1, sizeof(SSDataBlock));
+  taosArrayPush(dataBlocks, pBlock);
+  blockDebugShowDataBlocks(dataBlocks, flag);
+  taosArrayDestroy(dataBlocks);
+}
+
 void blockDebugShowDataBlocks(const SArray* dataBlocks, const char* flag) {
   char    pBuf[128] = {0};
   int32_t sz = taosArrayGetSize(dataBlocks);

source/common/src/tdataformat.c

@@ -931,9 +931,9 @@ int32_t tTagNew(SArray *pArray, int32_t version, int8_t isJson, STag **ppTag) {
 
   // sort
   if (isJson) {
-    qsort(pArray->pData, nTag, sizeof(STagVal), tTagValJsonCmprFn);
+    taosSort(pArray->pData, nTag, sizeof(STagVal), tTagValJsonCmprFn);
   } else {
-    qsort(pArray->pData, nTag, sizeof(STagVal), tTagValCmprFn);
+    taosSort(pArray->pData, nTag, sizeof(STagVal), tTagValCmprFn);
   }
 
   // get size

source/common/src/tmsg.c

@@ -503,6 +503,7 @@ int32_t tSerializeSMCreateStbReq(void *buf, int32_t bufLen, SMCreateStbReq *pReq
   if (tEncodeI32(&encoder, pReq->ttl) < 0) return -1;
   if (tEncodeI32(&encoder, pReq->numOfColumns) < 0) return -1;
   if (tEncodeI32(&encoder, pReq->numOfTags) < 0) return -1;
+  if (tEncodeI32(&encoder, pReq->numOfFuncs) < 0) return -1;
   if (tEncodeI32(&encoder, pReq->commentLen) < 0) return -1;
   if (tEncodeI32(&encoder, pReq->ast1Len) < 0) return -1;
   if (tEncodeI32(&encoder, pReq->ast2Len) < 0) return -1;
@@ -510,21 +511,26 @@ int32_t tSerializeSMCreateStbReq(void *buf, int32_t bufLen, SMCreateStbReq *pReq
   for (int32_t i = 0; i < pReq->numOfColumns; ++i) {
     SField *pField = taosArrayGet(pReq->pColumns, i);
     if (tEncodeI8(&encoder, pField->type) < 0) return -1;
+    if (tEncodeI8(&encoder, pField->flags) < 0) return -1;
     if (tEncodeI32(&encoder, pField->bytes) < 0) return -1;
     if (tEncodeCStr(&encoder, pField->name) < 0) return -1;
-    if (tEncodeI8(&encoder, pField->flags) < 0) return -1;
   }
 
   for (int32_t i = 0; i < pReq->numOfTags; ++i) {
     SField *pField = taosArrayGet(pReq->pTags, i);
     if (tEncodeI8(&encoder, pField->type) < 0) return -1;
+    if (tEncodeI8(&encoder, pField->flags) < 0) return -1;
     if (tEncodeI32(&encoder, pField->bytes) < 0) return -1;
     if (tEncodeCStr(&encoder, pField->name) < 0) return -1;
-    if (tEncodeI8(&encoder, pField->flags) < 0) return -1;
+  }
+
+  for (int32_t i = 0; i < pReq->numOfFuncs; ++i) {
+    const char *pFunc = taosArrayGet(pReq->pFuncs, i);
+    if (tEncodeCStr(&encoder, pFunc) < 0) return -1;
   }
 
   if (pReq->commentLen > 0) {
-    if (tEncodeCStr(&encoder, pReq->comment) < 0) return -1;
+    if (tEncodeCStr(&encoder, pReq->pComment) < 0) return -1;
   }
   if (pReq->ast1Len > 0) {
     if (tEncodeBinary(&encoder, pReq->pAst1, pReq->ast1Len) < 0) return -1;
@@ -533,13 +539,6 @@ int32_t tSerializeSMCreateStbReq(void *buf, int32_t bufLen, SMCreateStbReq *pReq
     if (tEncodeBinary(&encoder, pReq->pAst2, pReq->ast2Len) < 0) return -1;
   }
 
-  int32_t numOfFuncs = taosArrayGetSize(pReq->pFuncs);
-  if (tEncodeI32(&encoder, numOfFuncs) < 0) return -1;
-  for (int32_t i = 0; i < numOfFuncs; ++i) {
-    const char *pFunc = taosArrayGet(pReq->pFuncs, i);
-    if (tEncodeCStr(&encoder, pFunc) < 0) return -1;
-  }
-
   tEndEncode(&encoder);
 
   int32_t tlen = encoder.pos;
@@ -561,13 +560,15 @@ int32_t tDeserializeSMCreateStbReq(void *buf, int32_t bufLen, SMCreateStbReq *pR
   if (tDecodeI32(&decoder, &pReq->ttl) < 0) return -1;
   if (tDecodeI32(&decoder, &pReq->numOfColumns) < 0) return -1;
   if (tDecodeI32(&decoder, &pReq->numOfTags) < 0) return -1;
+  if (tDecodeI32(&decoder, &pReq->numOfFuncs) < 0) return -1;
   if (tDecodeI32(&decoder, &pReq->commentLen) < 0) return -1;
   if (tDecodeI32(&decoder, &pReq->ast1Len) < 0) return -1;
   if (tDecodeI32(&decoder, &pReq->ast2Len) < 0) return -1;
 
   pReq->pColumns = taosArrayInit(pReq->numOfColumns, sizeof(SField));
   pReq->pTags = taosArrayInit(pReq->numOfTags, sizeof(SField));
-  if (pReq->pColumns == NULL || pReq->pTags == NULL) {
+  pReq->pFuncs = taosArrayInit(pReq->numOfFuncs, TSDB_FUNC_NAME_LEN);
+  if (pReq->pColumns == NULL || pReq->pTags == NULL || pReq->pFuncs == NULL) {
     terrno = TSDB_CODE_OUT_OF_MEMORY;
     return -1;
   }
@@ -575,9 +576,9 @@ int32_t tDeserializeSMCreateStbReq(void *buf, int32_t bufLen, SMCreateStbReq *pR
   for (int32_t i = 0; i < pReq->numOfColumns; ++i) {
     SField field = {0};
     if (tDecodeI8(&decoder, &field.type) < 0) return -1;
+    if (tDecodeI8(&decoder, &field.flags) < 0) return -1;
     if (tDecodeI32(&decoder, &field.bytes) < 0) return -1;
     if (tDecodeCStrTo(&decoder, field.name) < 0) return -1;
-    if (tDecodeI8(&decoder, &field.flags) < 0) return -1;
     if (taosArrayPush(pReq->pColumns, &field) == NULL) {
       terrno = TSDB_CODE_OUT_OF_MEMORY;
       return -1;
@@ -587,19 +588,28 @@ int32_t tDeserializeSMCreateStbReq(void *buf, int32_t bufLen, SMCreateStbReq *pR
   for (int32_t i = 0; i < pReq->numOfTags; ++i) {
     SField field = {0};
     if (tDecodeI8(&decoder, &field.type) < 0) return -1;
+    if (tDecodeI8(&decoder, &field.flags) < 0) return -1;
     if (tDecodeI32(&decoder, &field.bytes) < 0) return -1;
     if (tDecodeCStrTo(&decoder, field.name) < 0) return -1;
-    if (tDecodeI8(&decoder, &field.flags) < 0) return -1;
     if (taosArrayPush(pReq->pTags, &field) == NULL) {
       terrno = TSDB_CODE_OUT_OF_MEMORY;
       return -1;
     }
   }
 
+  for (int32_t i = 0; i < pReq->numOfFuncs; ++i) {
+    char pFunc[TSDB_FUNC_NAME_LEN] = {0};
+    if (tDecodeCStrTo(&decoder, pFunc) < 0) return -1;
+    if (taosArrayPush(pReq->pFuncs, pFunc) == NULL) {
+      terrno = TSDB_CODE_OUT_OF_MEMORY;
+      return -1;
+    }
+  }
+
   if (pReq->commentLen > 0) {
-    pReq->comment = taosMemoryMalloc(pReq->commentLen + 1);
+    pReq->pComment = taosMemoryMalloc(pReq->commentLen + 1);
-    if (pReq->comment == NULL) return -1;
+    if (pReq->pComment == NULL) return -1;
-    if (tDecodeCStrTo(&decoder, pReq->comment) < 0) return -1;
+    if (tDecodeCStrTo(&decoder, pReq->pComment) < 0) return -1;
   }
 
   if (pReq->ast1Len > 0) {
@@ -614,23 +624,7 @@ int32_t tDeserializeSMCreateStbReq(void *buf, int32_t bufLen, SMCreateStbReq *pR
     if (tDecodeCStrTo(&decoder, pReq->pAst2) < 0) return -1;
   }
 
-  int32_t numOfFuncs = 0;
-  if (tDecodeI32(&decoder, &numOfFuncs) < 0) return -1;
-  if (numOfFuncs > 0) {
-    pReq->pFuncs = taosArrayInit(numOfFuncs, TSDB_FUNC_NAME_LEN);
-    if (NULL == pReq->pFuncs) return -1;
-  }
-  for (int32_t i = 0; i < numOfFuncs; ++i) {
-    char pFunc[TSDB_FUNC_NAME_LEN] = {0};
-    if (tDecodeCStrTo(&decoder, pFunc) < 0) return -1;
-    if (taosArrayPush(pReq->pFuncs, pFunc) == NULL) {
-      terrno = TSDB_CODE_OUT_OF_MEMORY;
-      return -1;
-    }
-  }
-
   tEndDecode(&decoder);
-
   tDecoderClear(&decoder);
   return 0;
 }
@@ -638,10 +632,10 @@ int32_t tDeserializeSMCreateStbReq(void *buf, int32_t bufLen, SMCreateStbReq *pR
 void tFreeSMCreateStbReq(SMCreateStbReq *pReq) {
   taosArrayDestroy(pReq->pColumns);
   taosArrayDestroy(pReq->pTags);
-  taosMemoryFreeClear(pReq->comment);
+  taosArrayDestroy(pReq->pFuncs);
+  taosMemoryFreeClear(pReq->pComment);
   taosMemoryFreeClear(pReq->pAst1);
   taosMemoryFreeClear(pReq->pAst2);
-  taosArrayDestroy(pReq->pFuncs);
 }
 
 int32_t tSerializeSMDropStbReq(void *buf, int32_t bufLen, SMDropStbReq *pReq) {
@@ -4309,6 +4303,7 @@ int32_t tSerializeSSchedulerHbRsp(void *buf, int32_t bufLen, SSchedulerHbRsp *pR
       if (tEncodeU64(&encoder, status->queryId) < 0) return -1;
       if (tEncodeU64(&encoder, status->taskId) < 0) return -1;
       if (tEncodeI64(&encoder, status->refId) < 0) return -1;
+      if (tEncodeI32(&encoder, status->execId) < 0) return -1;
       if (tEncodeI8(&encoder, status->status) < 0) return -1;
     }
   } else {
@@ -4339,6 +4334,7 @@ int32_t tDeserializeSSchedulerHbRsp(void *buf, int32_t bufLen, SSchedulerHbRsp *
       if (tDecodeU64(&decoder, &status.queryId) < 0) return -1;
       if (tDecodeU64(&decoder, &status.taskId) < 0) return -1;
       if (tDecodeI64(&decoder, &status.refId) < 0) return -1;
+      if (tDecodeI32(&decoder, &status.execId) < 0) return -1;
       if (tDecodeI8(&decoder, &status.status) < 0) return -1;
       taosArrayPush(pRsp->taskStatus, &status);
     }
@@ -4641,6 +4637,7 @@ int32_t tSerializeSCMCreateStreamReq(void *buf, int32_t bufLen, const SCMCreateS
   if (tEncodeI8(&encoder, pReq->triggerType) < 0) return -1;
   if (tEncodeI64(&encoder, pReq->maxDelay) < 0) return -1;
   if (tEncodeI64(&encoder, pReq->watermark) < 0) return -1;
+  if (tEncodeI8(&encoder, pReq->igExpired) < 0) return -1;
   if (sqlLen > 0 && tEncodeCStr(&encoder, pReq->sql) < 0) return -1;
   if (astLen > 0 && tEncodeCStr(&encoder, pReq->ast) < 0) return -1;
 
@@ -4668,6 +4665,7 @@ int32_t tDeserializeSCMCreateStreamReq(void *buf, int32_t bufLen, SCMCreateStrea
   if (tDecodeI8(&decoder, &pReq->triggerType) < 0) return -1;
   if (tDecodeI64(&decoder, &pReq->maxDelay) < 0) return -1;
   if (tDecodeI64(&decoder, &pReq->watermark) < 0) return -1;
+  if (tDecodeI8(&decoder, &pReq->igExpired) < 0) return -1;
 
   if (sqlLen > 0) {
     pReq->sql = taosMemoryCalloc(1, sqlLen + 1);
@@ -5398,9 +5396,9 @@ int32_t tFormatOffset(char *buf, int32_t maxLen, const STqOffsetVal *pVal) {
   } else if (pVal->type == TMQ_OFFSET__LOG) {
     snprintf(buf, maxLen, "offset(log) ver:%ld", pVal->version);
   } else if (pVal->type == TMQ_OFFSET__SNAPSHOT_DATA) {
-    snprintf(buf, maxLen, "offset(snapshot data) uid:%ld, ts:%ld", pVal->uid, pVal->ts);
+    snprintf(buf, maxLen, "offset(ss data) uid:%ld, ts:%ld", pVal->uid, pVal->ts);
   } else if (pVal->type == TMQ_OFFSET__SNAPSHOT_META) {
-    snprintf(buf, maxLen, "offset(snapshot meta) uid:%ld, ts:%ld", pVal->uid, pVal->ts);
+    snprintf(buf, maxLen, "offset(ss meta) uid:%ld, ts:%ld", pVal->uid, pVal->ts);
   } else {
     ASSERT(0);
   }
@@ -5502,4 +5500,3 @@ int32_t tDecodeSMqDataRsp(SDecoder *pDecoder, SMqDataRsp *pRsp) {
   }
   return 0;
 }
-

source/common/src/tvariant.c

@@ -155,7 +155,8 @@ void taosVariantCreateFromBinary(SVariant *pVar, const char *pz, size_t len, uin
 void taosVariantDestroy(SVariant *pVar) {
   if (pVar == NULL) return;
 
-  if (pVar->nType == TSDB_DATA_TYPE_BINARY || pVar->nType == TSDB_DATA_TYPE_NCHAR) {
+  if (pVar->nType == TSDB_DATA_TYPE_BINARY || pVar->nType == TSDB_DATA_TYPE_NCHAR
+     || pVar->nType == TSDB_DATA_TYPE_JSON) {
     taosMemoryFreeClear(pVar->pz);
     pVar->nLen = 0;
   }
@@ -184,7 +185,8 @@ void taosVariantAssign(SVariant *pDst, const SVariant *pSrc) {
   if (pSrc == NULL || pDst == NULL) return;
 
   pDst->nType = pSrc->nType;
-  if (pSrc->nType == TSDB_DATA_TYPE_BINARY || pSrc->nType == TSDB_DATA_TYPE_NCHAR) {
+  if (pSrc->nType == TSDB_DATA_TYPE_BINARY || pSrc->nType == TSDB_DATA_TYPE_NCHAR
+      || pSrc->nType == TSDB_DATA_TYPE_JSON) {
     int32_t len = pSrc->nLen + TSDB_NCHAR_SIZE;
     char   *p = taosMemoryRealloc(pDst->pz, len);
     assert(p);
@@ -976,6 +978,7 @@ char *taosVariantGet(SVariant *pVar, int32_t type) {
     case TSDB_DATA_TYPE_FLOAT:
       return (char *)&pVar->d;
     case TSDB_DATA_TYPE_BINARY:
+    case TSDB_DATA_TYPE_JSON:
       return (char *)pVar->pz;
     case TSDB_DATA_TYPE_NCHAR:
       return (char *)pVar->ucs4;

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

@@ -209,10 +209,11 @@ SArray *mmGetMsgHandles() {
   if (dmSetMgmtHandle(pArray, TDMT_MND_SHOW_VARIABLES, mmPutMsgToReadQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_MND_SERVER_VERSION, mmPutMsgToReadQueue, 0) == NULL) goto _OVER;
 
-  if (dmSetMgmtHandle(pArray, TDMT_VND_QUERY, mmPutMsgToQueryQueue, 1) == NULL) goto _OVER;
+  if (dmSetMgmtHandle(pArray, TDMT_SCH_QUERY, mmPutMsgToQueryQueue, 1) == NULL) goto _OVER;
-  if (dmSetMgmtHandle(pArray, TDMT_VND_QUERY_CONTINUE, mmPutMsgToQueryQueue, 1) == NULL) goto _OVER;
+  if (dmSetMgmtHandle(pArray, TDMT_SCH_MERGE_QUERY, mmPutMsgToQueryQueue, 1) == NULL) goto _OVER;
-  if (dmSetMgmtHandle(pArray, TDMT_VND_QUERY_HEARTBEAT, mmPutMsgToFetchQueue, 1) == NULL) goto _OVER;
+  if (dmSetMgmtHandle(pArray, TDMT_SCH_QUERY_CONTINUE, mmPutMsgToQueryQueue, 1) == NULL) goto _OVER;
-  if (dmSetMgmtHandle(pArray, TDMT_VND_FETCH, mmPutMsgToFetchQueue, 1) == NULL) goto _OVER;
+  if (dmSetMgmtHandle(pArray, TDMT_SCH_QUERY_HEARTBEAT, mmPutMsgToFetchQueue, 1) == NULL) goto _OVER;
+  if (dmSetMgmtHandle(pArray, TDMT_SCH_FETCH, mmPutMsgToFetchQueue, 1) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_VND_CREATE_STB_RSP, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_VND_ALTER_STB_RSP, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_VND_DROP_STB_RSP, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
@@ -220,7 +221,7 @@ SArray *mmGetMsgHandles() {
   if (dmSetMgmtHandle(pArray, TDMT_VND_DROP_SMA_RSP, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_VND_MQ_VG_CHANGE_RSP, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_VND_MQ_VG_DELETE_RSP, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
-  if (dmSetMgmtHandle(pArray, TDMT_VND_DROP_TASK, mmPutMsgToFetchQueue, 1) == NULL) goto _OVER;
+  if (dmSetMgmtHandle(pArray, TDMT_SCH_DROP_TASK, mmPutMsgToFetchQueue, 1) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_STREAM_TASK_DEPLOY_RSP, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_STREAM_TASK_DROP_RSP, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_VND_ALTER_CONFIG_RSP, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;

source/dnode/mgmt/mgmt_qnode/src/qmHandle.c

@@ -107,14 +107,15 @@ SArray *qmGetMsgHandles() {
   if (dmSetMgmtHandle(pArray, TDMT_MON_QM_INFO, qmPutNodeMsgToMonitorQueue, 0) == NULL) goto _OVER;
 
   // Requests handled by VNODE
-  if (dmSetMgmtHandle(pArray, TDMT_VND_QUERY, qmPutNodeMsgToQueryQueue, 1) == NULL) goto _OVER;
+  if (dmSetMgmtHandle(pArray, TDMT_SCH_QUERY, qmPutNodeMsgToQueryQueue, 1) == NULL) goto _OVER;
-  if (dmSetMgmtHandle(pArray, TDMT_VND_QUERY_CONTINUE, qmPutNodeMsgToQueryQueue, 1) == NULL) goto _OVER;
+  if (dmSetMgmtHandle(pArray, TDMT_SCH_MERGE_QUERY, qmPutNodeMsgToQueryQueue, 1) == NULL) goto _OVER;
-  if (dmSetMgmtHandle(pArray, TDMT_VND_FETCH, qmPutNodeMsgToFetchQueue, 1) == NULL) goto _OVER;
+  if (dmSetMgmtHandle(pArray, TDMT_SCH_QUERY_CONTINUE, qmPutNodeMsgToQueryQueue, 1) == NULL) goto _OVER;
-  if (dmSetMgmtHandle(pArray, TDMT_VND_FETCH_RSP, qmPutNodeMsgToFetchQueue, 1) == NULL) goto _OVER;
+  if (dmSetMgmtHandle(pArray, TDMT_SCH_FETCH, qmPutNodeMsgToFetchQueue, 1) == NULL) goto _OVER;
-  if (dmSetMgmtHandle(pArray, TDMT_VND_QUERY_HEARTBEAT, qmPutNodeMsgToFetchQueue, 1) == NULL) goto _OVER;
+  if (dmSetMgmtHandle(pArray, TDMT_SCH_FETCH_RSP, qmPutNodeMsgToFetchQueue, 1) == NULL) goto _OVER;
+  if (dmSetMgmtHandle(pArray, TDMT_SCH_QUERY_HEARTBEAT, qmPutNodeMsgToFetchQueue, 1) == NULL) goto _OVER;
 
-  if (dmSetMgmtHandle(pArray, TDMT_VND_CANCEL_TASK, qmPutNodeMsgToFetchQueue, 1) == NULL) goto _OVER;
+  if (dmSetMgmtHandle(pArray, TDMT_SCH_CANCEL_TASK, qmPutNodeMsgToFetchQueue, 1) == NULL) goto _OVER;
-  if (dmSetMgmtHandle(pArray, TDMT_VND_DROP_TASK, qmPutNodeMsgToFetchQueue, 1) == NULL) goto _OVER;
+  if (dmSetMgmtHandle(pArray, TDMT_SCH_DROP_TASK, qmPutNodeMsgToFetchQueue, 1) == NULL) goto _OVER;
 
   code = 0;
 _OVER:

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

@@ -324,16 +324,17 @@ SArray *vmGetMsgHandles() {
   if (dmSetMgmtHandle(pArray, TDMT_MON_VM_LOAD, vmPutMsgToMonitorQueue, 0) == NULL) goto _OVER;
 
   if (dmSetMgmtHandle(pArray, TDMT_VND_SUBMIT, vmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
-  if (dmSetMgmtHandle(pArray, TDMT_VND_QUERY, vmPutMsgToQueryQueue, 0) == NULL) goto _OVER;
+  if (dmSetMgmtHandle(pArray, TDMT_SCH_QUERY, vmPutMsgToQueryQueue, 0) == NULL) goto _OVER;
-  if (dmSetMgmtHandle(pArray, TDMT_VND_QUERY_CONTINUE, vmPutMsgToQueryQueue, 0) == NULL) goto _OVER;
+  if (dmSetMgmtHandle(pArray, TDMT_SCH_MERGE_QUERY, vmPutMsgToQueryQueue, 0) == NULL) goto _OVER;
-  if (dmSetMgmtHandle(pArray, TDMT_VND_FETCH, vmPutMsgToFetchQueue, 0) == NULL) goto _OVER;
+  if (dmSetMgmtHandle(pArray, TDMT_SCH_QUERY_CONTINUE, vmPutMsgToQueryQueue, 0) == NULL) goto _OVER;
+  if (dmSetMgmtHandle(pArray, TDMT_SCH_FETCH, vmPutMsgToFetchQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_VND_ALTER_TABLE, vmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_VND_UPDATE_TAG_VAL, vmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_VND_TABLE_META, vmPutMsgToFetchQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_VND_TABLE_CFG, vmPutMsgToFetchQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_VND_TABLES_META, vmPutMsgToFetchQueue, 0) == NULL) goto _OVER;
-  if (dmSetMgmtHandle(pArray, TDMT_VND_CANCEL_TASK, vmPutMsgToFetchQueue, 0) == NULL) goto _OVER;
+  if (dmSetMgmtHandle(pArray, TDMT_SCH_CANCEL_TASK, vmPutMsgToFetchQueue, 0) == NULL) goto _OVER;
-  if (dmSetMgmtHandle(pArray, TDMT_VND_DROP_TASK, vmPutMsgToFetchQueue, 0) == NULL) goto _OVER;
+  if (dmSetMgmtHandle(pArray, TDMT_SCH_DROP_TASK, vmPutMsgToFetchQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_VND_CREATE_STB, vmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_VND_DROP_TTL_TABLE, vmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_VND_ALTER_STB, vmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
@@ -349,7 +350,7 @@ SArray *vmGetMsgHandles() {
   if (dmSetMgmtHandle(pArray, TDMT_VND_MQ_COMMIT_OFFSET, vmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_VND_CONSUME, vmPutMsgToFetchQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_VND_DELETE, vmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
-  if (dmSetMgmtHandle(pArray, TDMT_VND_QUERY_HEARTBEAT, vmPutMsgToFetchQueue, 0) == NULL) goto _OVER;
+  if (dmSetMgmtHandle(pArray, TDMT_SCH_QUERY_HEARTBEAT, vmPutMsgToFetchQueue, 0) == NULL) goto _OVER;
 
   if (dmSetMgmtHandle(pArray, TDMT_VND_STREAM_TRIGGER, vmPutMsgToFetchQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_STREAM_TASK_DROP, vmPutMsgToWriteQueue, 0) == NULL) goto _OVER;

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

@@ -41,6 +41,7 @@ static inline void dmBuildMnodeRedirectRsp(SDnode *pDnode, SRpcMsg *pMsg) {
 }
 
 static inline void dmSendRedirectRsp(SRpcMsg *pMsg, const SEpSet *pNewEpSet) {
+  pMsg->info.hasEpSet = 1;
   SRpcMsg rsp = {.code = TSDB_CODE_RPC_REDIRECT, .info = pMsg->info};
   int32_t contLen = tSerializeSEpSet(NULL, 0, pNewEpSet);
 
@@ -86,7 +87,7 @@ static void dmProcessRpcMsg(SDnode *pDnode, SRpcMsg *pRpc, SEpSet *pEpSet) {
       return;
     case TDMT_MND_SYSTABLE_RETRIEVE_RSP:
     case TDMT_DND_SYSTABLE_RETRIEVE_RSP:
-    case TDMT_VND_FETCH_RSP:
+    case TDMT_SCH_FETCH_RSP:
       qWorkerProcessFetchRsp(NULL, NULL, pRpc, 0);
       return;
     case TDMT_MND_STATUS_RSP:
@@ -250,7 +251,10 @@ static inline void dmReleaseHandle(SRpcHandleInfo *pHandle, int8_t type) {
 
 static bool rpcRfp(int32_t code, tmsg_t msgType) {
   if (code == TSDB_CODE_RPC_REDIRECT || code == TSDB_CODE_RPC_NETWORK_UNAVAIL || code == TSDB_CODE_NODE_NOT_DEPLOYED ||
-      code == TSDB_CODE_SYN_NOT_LEADER || code == TSDB_CODE_APP_NOT_READY) {
+      code == TSDB_CODE_SYN_NOT_LEADER || code == TSDB_CODE_APP_NOT_READY || code == TSDB_CODE_RPC_BROKEN_LINK) {
+    if (msgType == TDMT_SCH_QUERY || msgType == TDMT_SCH_MERGE_QUERY || msgType == TDMT_SCH_FETCH) {
+      return false;
+    }
     return true;
   } else {
     return false;

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

@@ -341,11 +341,12 @@ typedef struct {
   int32_t  colVer;
   int32_t  smaVer;
   int32_t  nextColId;
-  int64_t  watermark[2];
   int64_t  maxdelay[2];
+  int64_t  watermark[2];
   int32_t  ttl;
   int32_t  numOfColumns;
   int32_t  numOfTags;
+  int32_t  numOfFuncs;
   int32_t  commentLen;
   int32_t  ast1Len;
   int32_t  ast2Len;
@@ -555,7 +556,7 @@ typedef struct {
   int64_t uid;
   int8_t  status;
   // config
-  int8_t  dropPolicy;
+  int8_t  igExpired;
   int8_t  trigger;
   int64_t triggerParam;
   int64_t watermark;

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

@@ -47,6 +47,7 @@ int32_t mndBuildAlterVgroupAction(SMnode *pMnode, STrans *pTrans, SDbObj *pDb, S
 void *mndBuildCreateVnodeReq(SMnode *, SDnodeObj *pDnode, SDbObj *pDb, SVgObj *pVgroup, int32_t *cntlen, bool standby);
 void *mndBuildDropVnodeReq(SMnode *, SDnodeObj *pDnode, SDbObj *pDb, SVgObj *pVgroup, int32_t *pContLen);
 void *mndBuildAlterVnodeReq(SMnode *, SDbObj *pDb, SVgObj *pVgroup, int32_t *pContLen);
+bool  mndVgroupInDb(SVgObj *pVgroup, int64_t dbUid);
 
 #ifdef __cplusplus
 }

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

@@ -645,7 +645,7 @@ static int32_t mndSetAlterDbRedoActions(SMnode *pMnode, STrans *pTrans, SDbObj *
     pIter = sdbFetch(pSdb, SDB_VGROUP, pIter, (void **)&pVgroup);
     if (pIter == NULL) break;
 
-    if (pVgroup->dbUid == pNew->uid) {
+    if (mndVgroupInDb(pVgroup, pNew->uid)) {
       if (mndBuildAlterVgroupAction(pMnode, pTrans, pNew, pVgroup, pArray) != 0) {
         sdbCancelFetch(pSdb, pIter);
         sdbRelease(pSdb, pVgroup);
@@ -1006,7 +1006,7 @@ static int32_t mndGetDBTableNum(SDbObj *pDb, SMnode *pMnode) {
     pIter = sdbFetch(pSdb, SDB_VGROUP, pIter, (void **)&pVgroup);
     if (pIter == NULL) break;
 
-    if (pVgroup->dbUid == pDb->uid) {
+    if (mndVgroupInDb(pVgroup, pDb->uid)) {
       numOfTables += pVgroup->numOfTables / TSDB_TABLE_NUM_UNIT;
       vindex++;
     }

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

@@ -28,7 +28,7 @@ int32_t tEncodeSStreamObj(SEncoder *pEncoder, const SStreamObj *pObj) {
   if (tEncodeI64(pEncoder, pObj->uid) < 0) return -1;
   if (tEncodeI8(pEncoder, pObj->status) < 0) return -1;
 
-  if (tEncodeI8(pEncoder, pObj->dropPolicy) < 0) return -1;
+  if (tEncodeI8(pEncoder, pObj->igExpired) < 0) return -1;
   if (tEncodeI8(pEncoder, pObj->trigger) < 0) return -1;
   if (tEncodeI64(pEncoder, pObj->triggerParam) < 0) return -1;
   if (tEncodeI64(pEncoder, pObj->watermark) < 0) return -1;
@@ -73,7 +73,7 @@ int32_t tDecodeSStreamObj(SDecoder *pDecoder, SStreamObj *pObj) {
   if (tDecodeI64(pDecoder, &pObj->uid) < 0) return -1;
   if (tDecodeI8(pDecoder, &pObj->status) < 0) return -1;
 
-  if (tDecodeI8(pDecoder, &pObj->dropPolicy) < 0) return -1;
+  if (tDecodeI8(pDecoder, &pObj->igExpired) < 0) return -1;
   if (tDecodeI8(pDecoder, &pObj->trigger) < 0) return -1;
   if (tDecodeI64(pDecoder, &pObj->triggerParam) < 0) return -1;
   if (tDecodeI64(pDecoder, &pObj->watermark) < 0) return -1;

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

@@ -58,7 +58,7 @@ static void *mndBuildTimerMsg(int32_t *pContLen) {
 
 static void mndPullupTrans(SMnode *pMnode) {
   int32_t contLen = 0;
-  void   *pReq = mndBuildTimerMsg(&contLen);
+  void *  pReq = mndBuildTimerMsg(&contLen);
   if (pReq != NULL) {
     SRpcMsg rpcMsg = {.msgType = TDMT_MND_TRANS_TIMER, .pCont = pReq, .contLen = contLen};
     tmsgPutToQueue(&pMnode->msgCb, WRITE_QUEUE, &rpcMsg);
@@ -67,14 +67,14 @@ static void mndPullupTrans(SMnode *pMnode) {
 
 static void mndTtlTimer(SMnode *pMnode) {
   int32_t contLen = 0;
-  void   *pReq = mndBuildTimerMsg(&contLen);
+  void *  pReq = mndBuildTimerMsg(&contLen);
   SRpcMsg rpcMsg = {.msgType = TDMT_MND_TTL_TIMER, .pCont = pReq, .contLen = contLen};
   tmsgPutToQueue(&pMnode->msgCb, WRITE_QUEUE, &rpcMsg);
 }
 
 static void mndCalMqRebalance(SMnode *pMnode) {
   int32_t contLen = 0;
-  void   *pReq = mndBuildTimerMsg(&contLen);
+  void *  pReq = mndBuildTimerMsg(&contLen);
   if (pReq != NULL) {
     SRpcMsg rpcMsg = {.msgType = TDMT_MND_MQ_TIMER, .pCont = pReq, .contLen = contLen};
     tmsgPutToQueue(&pMnode->msgCb, READ_QUEUE, &rpcMsg);
@@ -83,7 +83,7 @@ static void mndCalMqRebalance(SMnode *pMnode) {
 
 static void mndPullupTelem(SMnode *pMnode) {
   int32_t contLen = 0;
-  void   *pReq = mndBuildTimerMsg(&contLen);
+  void *  pReq = mndBuildTimerMsg(&contLen);
   if (pReq != NULL) {
     SRpcMsg rpcMsg = {.msgType = TDMT_MND_TELEM_TIMER, .pCont = pReq, .contLen = contLen};
     tmsgPutToQueue(&pMnode->msgCb, READ_QUEUE, &rpcMsg);
@@ -395,7 +395,7 @@ void mndStop(SMnode *pMnode) {
 }
 
 int32_t mndProcessSyncMsg(SRpcMsg *pMsg) {
-  SMnode    *pMnode = pMsg->info.node;
+  SMnode *   pMnode = pMsg->info.node;
   SSyncMgmt *pMgmt = &pMnode->syncMgmt;
   int32_t    code = 0;
 
@@ -413,7 +413,7 @@ int32_t mndProcessSyncMsg(SRpcMsg *pMsg) {
   }
 
   do {
-    char          *syncNodeStr = sync2SimpleStr(pMgmt->sync);
+    char *         syncNodeStr = sync2SimpleStr(pMgmt->sync);
     static int64_t mndTick = 0;
     if (++mndTick % 10 == 1) {
       mTrace("vgId:%d, sync trace msg:%s, %s", syncGetVgId(pMgmt->sync), TMSG_INFO(pMsg->msgType), syncNodeStr);
@@ -527,9 +527,9 @@ int32_t mndProcessSyncMsg(SRpcMsg *pMsg) {
 
 static int32_t mndCheckMnodeState(SRpcMsg *pMsg) {
   if (!IsReq(pMsg)) return 0;
-  if (pMsg->msgType == TDMT_VND_QUERY || pMsg->msgType == TDMT_VND_QUERY_CONTINUE ||
+  if (pMsg->msgType == TDMT_SCH_QUERY || pMsg->msgType == TDMT_SCH_MERGE_QUERY ||
-      pMsg->msgType == TDMT_VND_QUERY_HEARTBEAT || pMsg->msgType == TDMT_VND_FETCH ||
+      pMsg->msgType == TDMT_SCH_QUERY_CONTINUE || pMsg->msgType == TDMT_SCH_QUERY_HEARTBEAT ||
-      pMsg->msgType == TDMT_VND_DROP_TASK) {
+      pMsg->msgType == TDMT_SCH_FETCH || pMsg->msgType == TDMT_SCH_DROP_TASK) {
     return 0;
   }
   if (mndAcquireRpcRef(pMsg->info.node) == 0) return 0;
@@ -552,6 +552,7 @@ static int32_t mndCheckMnodeState(SRpcMsg *pMsg) {
 
     int32_t contLen = tSerializeSEpSet(NULL, 0, &epSet);
     pMsg->info.rsp = rpcMallocCont(contLen);
+    pMsg->info.hasEpSet = 1;
     if (pMsg->info.rsp != NULL) {
       tSerializeSEpSet(pMsg->info.rsp, contLen, &epSet);
       pMsg->info.rspLen = contLen;
@@ -578,7 +579,7 @@ static int32_t mndCheckMsgContent(SRpcMsg *pMsg) {
 }
 
 int32_t mndProcessRpcMsg(SRpcMsg *pMsg) {
-  SMnode         *pMnode = pMsg->info.node;
+  SMnode *        pMnode = pMsg->info.node;
   const STraceId *trace = &pMsg->info.traceId;
 
   MndMsgFp fp = pMnode->msgFp[TMSG_INDEX(pMsg->msgType)];
@@ -631,7 +632,7 @@ int32_t mndGetMonitorInfo(SMnode *pMnode, SMonClusterInfo *pClusterInfo, SMonVgr
                           SMonStbInfo *pStbInfo, SMonGrantInfo *pGrantInfo) {
   if (mndAcquireRpcRef(pMnode) != 0) return -1;
 
-  SSdb   *pSdb = pMnode->pSdb;
+  SSdb *  pSdb = pMnode->pSdb;
   int64_t ms = taosGetTimestampMs();
 
   pClusterInfo->dnodes = taosArrayInit(sdbGetSize(pSdb, SDB_DNODE), sizeof(SMonDnodeDesc));
@@ -712,7 +713,7 @@ int32_t mndGetMonitorInfo(SMnode *pMnode, SMonClusterInfo *pClusterInfo, SMonVgr
     pGrantInfo->timeseries_used += pVgroup->numOfTimeSeries;
     tstrncpy(desc.status, "unsynced", sizeof(desc.status));
     for (int32_t i = 0; i < pVgroup->replica; ++i) {
-      SVnodeGid     *pVgid = &pVgroup->vnodeGid[i];
+      SVnodeGid *    pVgid = &pVgroup->vnodeGid[i];
       SMonVnodeDesc *pVnDesc = &desc.vnodes[i];
       pVnDesc->dnode_id = pVgid->dnodeId;
       tstrncpy(pVnDesc->vnode_role, syncStr(pVgid->role), sizeof(pVnDesc->vnode_role));

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

@@ -19,13 +19,13 @@
 #include "qworker.h"
 
 int32_t mndPreProcessQueryMsg(SRpcMsg *pMsg) {
-  if (TDMT_VND_QUERY != pMsg->msgType) return 0;
+  if (TDMT_SCH_QUERY != pMsg->msgType && TDMT_SCH_MERGE_QUERY != pMsg->msgType) return 0;
   SMnode *pMnode = pMsg->info.node;
   return qWorkerPreprocessQueryMsg(pMnode->pQuery, pMsg);
 }
 
 void mndPostProcessQueryMsg(SRpcMsg *pMsg) {
-  if (TDMT_VND_QUERY != pMsg->msgType) return;
+  if (TDMT_SCH_QUERY != pMsg->msgType && TDMT_SCH_MERGE_QUERY != pMsg->msgType) return;
   SMnode *pMnode = pMsg->info.node;
   qWorkerAbortPreprocessQueryMsg(pMnode->pQuery, pMsg);
 }
@@ -37,19 +37,20 @@ int32_t mndProcessQueryMsg(SRpcMsg *pMsg) {
 
   mTrace("msg:%p, in query queue is processing", pMsg);
   switch (pMsg->msgType) {
-    case TDMT_VND_QUERY:
+    case TDMT_SCH_QUERY:
+    case TDMT_SCH_MERGE_QUERY:
       code = qWorkerProcessQueryMsg(&handle, pMnode->pQuery, pMsg, 0);
       break;
-    case TDMT_VND_QUERY_CONTINUE:
+    case TDMT_SCH_QUERY_CONTINUE:
       code = qWorkerProcessCQueryMsg(&handle, pMnode->pQuery, pMsg, 0);
       break;
-    case TDMT_VND_FETCH:
+    case TDMT_SCH_FETCH:
       code = qWorkerProcessFetchMsg(pMnode, pMnode->pQuery, pMsg, 0);
       break;
-    case TDMT_VND_DROP_TASK:
+    case TDMT_SCH_DROP_TASK:
       code = qWorkerProcessDropMsg(pMnode, pMnode->pQuery, pMsg, 0);
       break;
-    case TDMT_VND_QUERY_HEARTBEAT:
+    case TDMT_SCH_QUERY_HEARTBEAT:
       code = qWorkerProcessHbMsg(pMnode, pMnode->pQuery, pMsg, 0);
       break;
     default:
@@ -67,11 +68,12 @@ int32_t mndInitQuery(SMnode *pMnode) {
     return -1;
   }
 
-  mndSetMsgHandle(pMnode, TDMT_VND_QUERY, mndProcessQueryMsg);
+  mndSetMsgHandle(pMnode, TDMT_SCH_QUERY, mndProcessQueryMsg);
-  mndSetMsgHandle(pMnode, TDMT_VND_QUERY_CONTINUE, mndProcessQueryMsg);
+  mndSetMsgHandle(pMnode, TDMT_SCH_MERGE_QUERY, mndProcessQueryMsg);
-  mndSetMsgHandle(pMnode, TDMT_VND_FETCH, mndProcessQueryMsg);
+  mndSetMsgHandle(pMnode, TDMT_SCH_QUERY_CONTINUE, mndProcessQueryMsg);
-  mndSetMsgHandle(pMnode, TDMT_VND_DROP_TASK, mndProcessQueryMsg);
+  mndSetMsgHandle(pMnode, TDMT_SCH_FETCH, mndProcessQueryMsg);
-  mndSetMsgHandle(pMnode, TDMT_VND_QUERY_HEARTBEAT, mndProcessQueryMsg);
+  mndSetMsgHandle(pMnode, TDMT_SCH_DROP_TASK, mndProcessQueryMsg);
+  mndSetMsgHandle(pMnode, TDMT_SCH_QUERY_HEARTBEAT, mndProcessQueryMsg);
 
   return 0;
 }

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

@@ -225,10 +225,11 @@ int32_t mndAddShuffleSinkTasksToStream(SMnode* pMnode, SStreamObj* pStream) {
     SVgObj* pVgroup;
     pIter = sdbFetch(pSdb, SDB_VGROUP, pIter, (void**)&pVgroup);
     if (pIter == NULL) break;
-    if (strcmp(pVgroup->dbName, pStream->targetDb) != 0) {
+    if (!mndVgroupInDb(pVgroup, pStream->targetDbUid)) {
       sdbRelease(pSdb, pVgroup);
       continue;
     }
+
     SStreamTask* pTask = tNewSStreamTask(pStream->uid);
     if (pTask == NULL) {
       sdbRelease(pSdb, pVgroup);
@@ -420,10 +421,11 @@ int32_t mndScheduleStream(SMnode* pMnode, SStreamObj* pStream) {
       SVgObj* pVgroup;
       pIter = sdbFetch(pSdb, SDB_VGROUP, pIter, (void**)&pVgroup);
       if (pIter == NULL) break;
-      if (pVgroup->dbUid != pStream->sourceDbUid) {
+      if (!mndVgroupInDb(pVgroup, pStream->sourceDbUid)) {
         sdbRelease(pSdb, pVgroup);
         continue;
       }
+
       SStreamTask* pTask = tNewSStreamTask(pStream->uid);
       if (pInnerTask == NULL) {
         terrno = TSDB_CODE_OUT_OF_MEMORY;
@@ -483,10 +485,11 @@ int32_t mndScheduleStream(SMnode* pMnode, SStreamObj* pStream) {
       SVgObj* pVgroup;
       pIter = sdbFetch(pSdb, SDB_VGROUP, pIter, (void**)&pVgroup);
       if (pIter == NULL) break;
-      if (pVgroup->dbUid != pStream->sourceDbUid) {
+      if (!mndVgroupInDb(pVgroup, pStream->sourceDbUid)) {
         sdbRelease(pSdb, pVgroup);
         continue;
       }
+
       SStreamTask* pTask = tNewSStreamTask(pStream->uid);
       if (pTask == NULL) {
         sdbRelease(pSdb, pVgroup);
@@ -559,7 +562,7 @@ int32_t mndSchedInitSubEp(SMnode* pMnode, const SMqTopicObj* pTopic, SMqSubscrib
   while (1) {
     pIter = sdbFetch(pSdb, SDB_VGROUP, pIter, (void**)&pVgroup);
     if (pIter == NULL) break;
-    if (pVgroup->dbUid != pTopic->dbUid) {
+    if (!mndVgroupInDb(pVgroup, pTopic->dbUid)) {
       sdbRelease(pSdb, pVgroup);
       continue;
     }

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

@@ -396,6 +396,8 @@ static int32_t mndSetUpdateSmaStbCommitLogs(SMnode *pMnode, STrans *pTrans, SStb
   stbObj.pColumns = NULL;
   stbObj.numOfTags = 0;
   stbObj.pTags = NULL;
+  stbObj.numOfFuncs = 0;
+  stbObj.pFuncs = NULL;
   stbObj.updateTime = taosGetTimestampMs();
   stbObj.lock = 0;
   stbObj.smaVer++;
@@ -408,47 +410,6 @@ static int32_t mndSetUpdateSmaStbCommitLogs(SMnode *pMnode, STrans *pTrans, SStb
   return 0;
 }
 
-#if 0
-static int32_t mndSetCreateSmaRedoActions(SMnode *pMnode, STrans *pTrans, SDbObj *pDb, SSmaObj *pSma) {
-  SSdb   *pSdb = pMnode->pSdb;
-  SVgObj *pVgroup = NULL;
-  void   *pIter = NULL;
-  int32_t contLen;
-
-  while (1) {
-    pIter = sdbFetch(pSdb, SDB_VGROUP, pIter, (void **)&pVgroup);
-    if (pIter == NULL) break;
-    if (pVgroup->dbUid != pDb->uid) {
-      sdbRelease(pSdb, pVgroup);
-      continue;
-    }
-
-    void *pReq = mndBuildVCreateSmaReq(pMnode, pVgroup, pSma, &contLen);
-    if (pReq == NULL) {
-      sdbCancelFetch(pSdb, pIter);
-      sdbRelease(pSdb, pVgroup);
-      terrno = TSDB_CODE_OUT_OF_MEMORY;
-      return -1;
-    }
-
-    STransAction action = {0};
-    action.epSet = mndGetVgroupEpset(pMnode, pVgroup);
-    action.pCont = pReq;
-    action.contLen = contLen;
-    action.msgType = TDMT_VND_CREATE_SMA;
-    if (mndTransAppendRedoAction(pTrans, &action) != 0) {
-      taosMemoryFree(pReq);
-      sdbCancelFetch(pSdb, pIter);
-      sdbRelease(pSdb, pVgroup);
-      return -1;
-    }
-    sdbRelease(pSdb, pVgroup);
-  }
-
-  return 0;
-}
-#endif
-
 static int32_t mndSetCreateSmaVgroupRedoActions(SMnode *pMnode, STrans *pTrans, SDbObj *pDb, SVgObj *pVgroup,
                                                 SSmaObj *pSma) {
   SVnodeGid *pVgid = pVgroup->vnodeGid + 0;
@@ -621,7 +582,6 @@ static int32_t mndCreateSma(SMnode *pMnode, SRpcMsg *pReq, SMCreateSmaReq *pCrea
   if (mndSetCreateSmaCommitLogs(pMnode, pTrans, &smaObj) != 0) goto _OVER;
   if (mndSetCreateSmaVgroupCommitLogs(pMnode, pTrans, &streamObj.fixedSinkVg) != 0) goto _OVER;
   if (mndSetUpdateSmaStbCommitLogs(pMnode, pTrans, pStb) != 0) goto _OVER;
-  // if (mndSetCreateSmaRedoActions(pMnode, pTrans, pDb, &smaObj) != 0) goto _OVER;
   if (mndSetCreateSmaVgroupRedoActions(pMnode, pTrans, pDb, &streamObj.fixedSinkVg, &smaObj) != 0) goto _OVER;
   if (mndScheduleStream(pMnode, &streamObj) != 0) goto _OVER;
   if (mndPersistStream(pMnode, pTrans, &streamObj) != 0) goto _OVER;
@@ -770,49 +730,6 @@ static int32_t mndSetDropSmaVgroupCommitLogs(SMnode *pMnode, STrans *pTrans, SVg
   return 0;
 }
 
-#if 0
-static int32_t mndSetDropSmaRedoActions(SMnode *pMnode, STrans *pTrans, SDbObj *pDb, SSmaObj *pSma) {
-  SSdb   *pSdb = pMnode->pSdb;
-  SVgObj *pVgroup = NULL;
-  void   *pIter = NULL;
-  int32_t contLen;
-
-  while (1) {
-    pIter = sdbFetch(pSdb, SDB_VGROUP, pIter, (void **)&pVgroup);
-    if (pIter == NULL) break;
-    if (pVgroup->dbUid != pDb->uid) {
-      sdbRelease(pSdb, pVgroup);
-      continue;
-    }
-
-    int32_t contLen = 0;
-    void   *pReq = mndBuildVDropSmaReq(pMnode, pVgroup, pSma, &contLen);
-    if (pReq == NULL) {
-      sdbCancelFetch(pSdb, pIter);
-      sdbRelease(pSdb, pVgroup);
-      terrno = TSDB_CODE_OUT_OF_MEMORY;
-      return -1;
-    }
-
-    STransAction action = {0};
-    action.epSet = mndGetVgroupEpset(pMnode, pVgroup);
-    action.pCont = pReq;
-    action.contLen = contLen;
-    action.msgType = TDMT_VND_DROP_SMA;
-    action.acceptableCode = TSDB_CODE_VND_SMA_NOT_EXIST;
-    if (mndTransAppendRedoAction(pTrans, &action) != 0) {
-      taosMemoryFree(pReq);
-      sdbCancelFetch(pSdb, pIter);
-      sdbRelease(pSdb, pVgroup);
-      return -1;
-    }
-    sdbRelease(pSdb, pVgroup);
-  }
-
-  return 0;
-}
-#endif
-
 static int32_t mndSetDropSmaVgroupRedoActions(SMnode *pMnode, STrans *pTrans, SDbObj *pDb, SVgObj *pVgroup) {
   SVnodeGid *pVgid = pVgroup->vnodeGid + 0;
   SDnodeObj *pDnode = mndAcquireDnode(pMnode, pVgid->dnodeId);
@@ -879,7 +796,6 @@ static int32_t mndDropSma(SMnode *pMnode, SRpcMsg *pReq, SDbObj *pDb, SSmaObj *p
   if (mndSetDropSmaCommitLogs(pMnode, pTrans, pSma) != 0) goto _OVER;
   if (mndSetDropSmaVgroupCommitLogs(pMnode, pTrans, pVgroup) != 0) goto _OVER;
   if (mndSetUpdateSmaStbCommitLogs(pMnode, pTrans, pStb) != 0) goto _OVER;
-  // if (mndSetDropSmaRedoActions(pMnode, pTrans, pDb, pSma) != 0) goto _OVER;
   if (mndSetDropSmaVgroupRedoActions(pMnode, pTrans, pDb, pVgroup) != 0) goto _OVER;
   if (mndTransPrepare(pMnode, pTrans) != 0) goto _OVER;
 
@@ -906,10 +822,20 @@ int32_t mndDropSmasByStb(SMnode *pMnode, STrans *pTrans, SDbObj *pDb, SStbObj *p
     if (pSma->stbUid == pStb->uid) {
       pVgroup = mndAcquireVgroup(pMnode, pSma->dstVgId);
       if (pVgroup == NULL) goto _OVER;
+
+      SStreamObj *pStream = mndAcquireStream(pMnode, pSma->name);
+      if (pStream != NULL && pStream->smaId == pSma->uid) {
+        if (mndDropStreamTasks(pMnode, pTrans, pStream) < 0) {
+          mError("stream:%s, failed to drop task since %s", pStream->name, terrstr());
+          goto _OVER;
+        }
+        if (mndPersistDropStreamLog(pMnode, pTrans, pStream) < 0) {
+          goto _OVER;
+        }
+      }
       if (mndSetDropSmaVgroupCommitLogs(pMnode, pTrans, pVgroup) != 0) goto _OVER;
       if (mndSetDropSmaVgroupRedoActions(pMnode, pTrans, pDb, pVgroup) != 0) goto _OVER;
       if (mndSetDropSmaCommitLogs(pMnode, pTrans, pSma) != 0) goto _OVER;
-      // if (mndSetDropSmaRedoActions(pMnode, pTrans, pDb, pSma) != 0) goto _OVER;
       mndReleaseVgroup(pMnode, pVgroup);
       pVgroup = NULL;
     }

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

@@ -78,7 +78,7 @@ void mndCleanupStb(SMnode *pMnode) {}
 SSdbRaw *mndStbActionEncode(SStbObj *pStb) {
   terrno = TSDB_CODE_OUT_OF_MEMORY;
 
-  int32_t size = sizeof(SStbObj) + (pStb->numOfColumns + pStb->numOfTags) * sizeof(SSchema) + +pStb->commentLen +
+  int32_t size = sizeof(SStbObj) + (pStb->numOfColumns + pStb->numOfTags) * sizeof(SSchema) + pStb->commentLen +
                  pStb->ast1Len + pStb->ast2Len + STB_RESERVE_SIZE + taosArrayGetSize(pStb->pFuncs) * TSDB_FUNC_NAME_LEN;
   SSdbRaw *pRaw = sdbAllocRaw(SDB_STB, STB_VER_NUMBER, size);
   if (pRaw == NULL) goto _OVER;
@@ -92,6 +92,7 @@ SSdbRaw *mndStbActionEncode(SStbObj *pStb) {
   SDB_SET_INT64(pRaw, dataPos, pStb->dbUid, _OVER)
   SDB_SET_INT32(pRaw, dataPos, pStb->tagVer, _OVER)
   SDB_SET_INT32(pRaw, dataPos, pStb->colVer, _OVER)
+  SDB_SET_INT32(pRaw, dataPos, pStb->smaVer, _OVER)
   SDB_SET_INT32(pRaw, dataPos, pStb->nextColId, _OVER)
   SDB_SET_INT64(pRaw, dataPos, pStb->maxdelay[0], _OVER)
   SDB_SET_INT64(pRaw, dataPos, pStb->maxdelay[1], _OVER)
@@ -100,17 +101,11 @@ SSdbRaw *mndStbActionEncode(SStbObj *pStb) {
   SDB_SET_INT32(pRaw, dataPos, pStb->ttl, _OVER)
   SDB_SET_INT32(pRaw, dataPos, pStb->numOfColumns, _OVER)
   SDB_SET_INT32(pRaw, dataPos, pStb->numOfTags, _OVER)
+  SDB_SET_INT32(pRaw, dataPos, pStb->numOfFuncs, _OVER)
   SDB_SET_INT32(pRaw, dataPos, pStb->commentLen, _OVER)
   SDB_SET_INT32(pRaw, dataPos, pStb->ast1Len, _OVER)
   SDB_SET_INT32(pRaw, dataPos, pStb->ast2Len, _OVER)
 
-  int32_t funcNum = taosArrayGetSize(pStb->pFuncs);
-  SDB_SET_INT32(pRaw, dataPos, funcNum, _OVER)
-  for (int32_t i = 0; i < funcNum; ++i) {
-    char *func = taosArrayGet(pStb->pFuncs, i);
-    SDB_SET_BINARY(pRaw, dataPos, func, TSDB_FUNC_NAME_LEN, _OVER)
-  }
-
   for (int32_t i = 0; i < pStb->numOfColumns; ++i) {
     SSchema *pSchema = &pStb->pColumns[i];
     SDB_SET_INT8(pRaw, dataPos, pSchema->type, _OVER)
@@ -129,15 +124,23 @@ SSdbRaw *mndStbActionEncode(SStbObj *pStb) {
     SDB_SET_BINARY(pRaw, dataPos, pSchema->name, TSDB_COL_NAME_LEN, _OVER)
   }
 
+  for (int32_t i = 0; i < pStb->numOfFuncs; ++i) {
+    char *func = taosArrayGet(pStb->pFuncs, i);
+    SDB_SET_BINARY(pRaw, dataPos, func, TSDB_FUNC_NAME_LEN, _OVER)
+  }
+
   if (pStb->commentLen > 0) {
     SDB_SET_BINARY(pRaw, dataPos, pStb->comment, pStb->commentLen + 1, _OVER)
   }
+
   if (pStb->ast1Len > 0) {
     SDB_SET_BINARY(pRaw, dataPos, pStb->pAst1, pStb->ast1Len, _OVER)
   }
+
   if (pStb->ast2Len > 0) {
     SDB_SET_BINARY(pRaw, dataPos, pStb->pAst2, pStb->ast2Len, _OVER)
   }
+
   SDB_SET_RESERVE(pRaw, dataPos, STB_RESERVE_SIZE, _OVER)
   SDB_SET_DATALEN(pRaw, dataPos, _OVER)
 
@@ -180,6 +183,7 @@ static SSdbRow *mndStbActionDecode(SSdbRaw *pRaw) {
   SDB_GET_INT64(pRaw, dataPos, &pStb->dbUid, _OVER)
   SDB_GET_INT32(pRaw, dataPos, &pStb->tagVer, _OVER)
   SDB_GET_INT32(pRaw, dataPos, &pStb->colVer, _OVER)
+  SDB_GET_INT32(pRaw, dataPos, &pStb->smaVer, _OVER)
   SDB_GET_INT32(pRaw, dataPos, &pStb->nextColId, _OVER)
   SDB_GET_INT64(pRaw, dataPos, &pStb->maxdelay[0], _OVER)
   SDB_GET_INT64(pRaw, dataPos, &pStb->maxdelay[1], _OVER)
@@ -188,27 +192,15 @@ static SSdbRow *mndStbActionDecode(SSdbRaw *pRaw) {
   SDB_GET_INT32(pRaw, dataPos, &pStb->ttl, _OVER)
   SDB_GET_INT32(pRaw, dataPos, &pStb->numOfColumns, _OVER)
   SDB_GET_INT32(pRaw, dataPos, &pStb->numOfTags, _OVER)
+  SDB_GET_INT32(pRaw, dataPos, &pStb->numOfFuncs, _OVER)
   SDB_GET_INT32(pRaw, dataPos, &pStb->commentLen, _OVER)
   SDB_GET_INT32(pRaw, dataPos, &pStb->ast1Len, _OVER)
   SDB_GET_INT32(pRaw, dataPos, &pStb->ast2Len, _OVER)
 
-  int32_t funcNum = 0;
-  SDB_GET_INT32(pRaw, dataPos, &funcNum, _OVER)
-  if (funcNum > 0) {
-    pStb->pFuncs = taosArrayInit(funcNum, TSDB_FUNC_NAME_LEN);
-    if (NULL == pStb->pFuncs) {
-      goto _OVER;
-    }
-    char funcName[TSDB_FUNC_NAME_LEN];
-    for (int32_t i = 0; i < funcNum; ++i) {
-      SDB_GET_BINARY(pRaw, dataPos, funcName, TSDB_FUNC_NAME_LEN, _OVER)
-      taosArrayPush(pStb->pFuncs, funcName);
-    }
-  }
-
   pStb->pColumns = taosMemoryCalloc(pStb->numOfColumns, sizeof(SSchema));
   pStb->pTags = taosMemoryCalloc(pStb->numOfTags, sizeof(SSchema));
-  if (pStb->pColumns == NULL || pStb->pTags == NULL) {
+  pStb->pFuncs = taosArrayInit(pStb->numOfFuncs, TSDB_FUNC_NAME_LEN);
+  if (pStb->pColumns == NULL || pStb->pTags == NULL || pStb->pFuncs == NULL) {
     goto _OVER;
   }
 
@@ -230,16 +222,24 @@ static SSdbRow *mndStbActionDecode(SSdbRaw *pRaw) {
     SDB_GET_BINARY(pRaw, dataPos, pSchema->name, TSDB_COL_NAME_LEN, _OVER)
   }
 
+  for (int32_t i = 0; i < pStb->numOfFuncs; ++i) {
+    char funcName[TSDB_FUNC_NAME_LEN] = {0};
+    SDB_GET_BINARY(pRaw, dataPos, funcName, TSDB_FUNC_NAME_LEN, _OVER)
+    taosArrayPush(pStb->pFuncs, funcName);
+  }
+
   if (pStb->commentLen > 0) {
     pStb->comment = taosMemoryCalloc(pStb->commentLen + 1, 1);
     if (pStb->comment == NULL) goto _OVER;
     SDB_GET_BINARY(pRaw, dataPos, pStb->comment, pStb->commentLen + 1, _OVER)
   }
+
   if (pStb->ast1Len > 0) {
     pStb->pAst1 = taosMemoryCalloc(pStb->ast1Len, 1);
     if (pStb->pAst1 == NULL) goto _OVER;
     SDB_GET_BINARY(pRaw, dataPos, pStb->pAst1, pStb->ast1Len, _OVER)
   }
+
   if (pStb->ast2Len > 0) {
     pStb->pAst2 = taosMemoryCalloc(pStb->ast2Len, 1);
     if (pStb->pAst2 == NULL) goto _OVER;
@@ -273,6 +273,7 @@ static int32_t mndStbActionDelete(SSdb *pSdb, SStbObj *pStb) {
   taosMemoryFreeClear(pStb->pColumns);
   taosMemoryFreeClear(pStb->pTags);
   taosMemoryFreeClear(pStb->comment);
+  taosMemoryFreeClear(pStb->pFuncs);
   taosMemoryFreeClear(pStb->pAst1);
   taosMemoryFreeClear(pStb->pAst2);
   taosArrayDestroy(pStb->pFuncs);
@@ -322,7 +323,7 @@ static int32_t mndStbActionUpdate(SSdb *pSdb, SStbObj *pOld, SStbObj *pNew) {
   pOld->commentLen = pNew->commentLen;
 
   if (pOld->ast1Len < pNew->ast1Len) {
-    void *pAst1 = taosMemoryMalloc(pNew->ast1Len);
+    void *pAst1 = taosMemoryMalloc(pNew->ast1Len + 1);
     if (pAst1 != NULL) {
       taosMemoryFree(pOld->pAst1);
       pOld->pAst1 = pAst1;
@@ -334,7 +335,7 @@ static int32_t mndStbActionUpdate(SSdb *pSdb, SStbObj *pOld, SStbObj *pNew) {
   }
 
   if (pOld->ast2Len < pNew->ast2Len) {
-    void *pAst2 = taosMemoryMalloc(pNew->ast2Len);
+    void *pAst2 = taosMemoryMalloc(pNew->ast2Len + 1);
     if (pAst2 != NULL) {
       taosMemoryFree(pOld->pAst2);
       pOld->pAst2 = pAst2;
@@ -361,12 +362,15 @@ static int32_t mndStbActionUpdate(SSdb *pSdb, SStbObj *pOld, SStbObj *pNew) {
   }
   if (pNew->commentLen > 0) {
     memcpy(pOld->comment, pNew->comment, pNew->commentLen + 1);
+    pOld->commentLen = pNew->commentLen;
   }
   if (pNew->ast1Len != 0) {
     memcpy(pOld->pAst1, pNew->pAst1, pNew->ast1Len);
+    pOld->ast1Len = pNew->ast1Len;
   }
   if (pNew->ast2Len != 0) {
     memcpy(pOld->pAst2, pNew->pAst2, pNew->ast2Len);
+    pOld->ast2Len = pNew->ast2Len;
   }
   taosWUnLockLatch(&pOld->lock);
   return 0;
@@ -575,7 +579,10 @@ int32_t mndCheckCreateStbReq(SMCreateStbReq *pCreate) {
 static int32_t mndSetCreateStbRedoLogs(SMnode *pMnode, STrans *pTrans, SDbObj *pDb, SStbObj *pStb) {
   SSdbRaw *pRedoRaw = mndStbActionEncode(pStb);
   if (pRedoRaw == NULL) return -1;
-  if (mndTransAppendRedolog(pTrans, pRedoRaw) != 0) return -1;
+  if (mndTransAppendRedolog(pTrans, pRedoRaw) != 0) {
+    sdbFreeRaw(pRedoRaw);
+    return -1;
+  }
   if (sdbSetRawStatus(pRedoRaw, SDB_STATUS_CREATING) != 0) return -1;
 
   return 0;
@@ -584,7 +591,10 @@ static int32_t mndSetCreateStbRedoLogs(SMnode *pMnode, STrans *pTrans, SDbObj *p
 static int32_t mndSetCreateStbUndoLogs(SMnode *pMnode, STrans *pTrans, SDbObj *pDb, SStbObj *pStb) {
   SSdbRaw *pUndoRaw = mndStbActionEncode(pStb);
   if (pUndoRaw == NULL) return -1;
-  if (mndTransAppendUndolog(pTrans, pUndoRaw) != 0) return -1;
+  if (mndTransAppendUndolog(pTrans, pUndoRaw) != 0) {
+    sdbFreeRaw(pUndoRaw);
+    return -1;
+  }
   if (sdbSetRawStatus(pUndoRaw, SDB_STATUS_DROPPED) != 0) return -1;
 
   return 0;
@@ -593,7 +603,10 @@ static int32_t mndSetCreateStbUndoLogs(SMnode *pMnode, STrans *pTrans, SDbObj *p
 static int32_t mndSetCreateStbCommitLogs(SMnode *pMnode, STrans *pTrans, SDbObj *pDb, SStbObj *pStb) {
   SSdbRaw *pCommitRaw = mndStbActionEncode(pStb);
   if (pCommitRaw == NULL) return -1;
-  if (mndTransAppendCommitlog(pTrans, pCommitRaw) != 0) return -1;
+  if (mndTransAppendCommitlog(pTrans, pCommitRaw) != 0) {
+    sdbFreeRaw(pCommitRaw);
+    return -1;
+  }
   if (sdbSetRawStatus(pCommitRaw, SDB_STATUS_READY) != 0) return -1;
 
   return 0;
@@ -608,7 +621,7 @@ static int32_t mndSetCreateStbRedoActions(SMnode *pMnode, STrans *pTrans, SDbObj
   while (1) {
     pIter = sdbFetch(pSdb, SDB_VGROUP, pIter, (void **)&pVgroup);
     if (pIter == NULL) break;
-    if (pVgroup->dbUid != pDb->uid) {
+    if (!mndVgroupInDb(pVgroup, pDb->uid)) {
       sdbRelease(pSdb, pVgroup);
       continue;
     }
@@ -646,7 +659,7 @@ static int32_t mndSetCreateStbUndoActions(SMnode *pMnode, STrans *pTrans, SDbObj
   while (1) {
     pIter = sdbFetch(pSdb, SDB_VGROUP, pIter, (void **)&pVgroup);
     if (pIter == NULL) break;
-    if (pVgroup->dbUid != pDb->uid) {
+    if (!mndVgroupInDb(pVgroup, pDb->uid)) {
       sdbRelease(pSdb, pVgroup);
       continue;
     }
@@ -697,6 +710,7 @@ int32_t mndBuildStbFromReq(SMnode *pMnode, SStbObj *pDst, SMCreateStbReq *pCreat
   pDst->dbUid = pDb->uid;
   pDst->tagVer = 1;
   pDst->colVer = 1;
+  pDst->smaVer = 1;
   pDst->nextColId = 1;
   pDst->maxdelay[0] = pCreate->delay1;
   pDst->maxdelay[1] = pCreate->delay2;
@@ -705,6 +719,7 @@ int32_t mndBuildStbFromReq(SMnode *pMnode, SStbObj *pDst, SMCreateStbReq *pCreat
   pDst->ttl = pCreate->ttl;
   pDst->numOfColumns = pCreate->numOfColumns;
   pDst->numOfTags = pCreate->numOfTags;
+  pDst->numOfFuncs = pCreate->numOfFuncs;
   pDst->commentLen = pCreate->commentLen;
   pDst->pFuncs = pCreate->pFuncs;
   pCreate->pFuncs = NULL;
@@ -715,7 +730,7 @@ int32_t mndBuildStbFromReq(SMnode *pMnode, SStbObj *pDst, SMCreateStbReq *pCreat
       terrno = TSDB_CODE_OUT_OF_MEMORY;
       return -1;
     }
-    memcpy(pDst->comment, pCreate->comment, pDst->commentLen + 1);
+    memcpy(pDst->comment, pCreate->pComment, pDst->commentLen + 1);
   }
 
   pDst->ast1Len = pCreate->ast1Len;
@@ -770,20 +785,15 @@ int32_t mndBuildStbFromReq(SMnode *pMnode, SStbObj *pDst, SMCreateStbReq *pCreat
 
 static int32_t mndCreateStb(SMnode *pMnode, SRpcMsg *pReq, SMCreateStbReq *pCreate, SDbObj *pDb) {
   SStbObj stbObj = {0};
-
   int32_t code = -1;
 
   STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_ROLLBACK, TRN_CONFLICT_DB_INSIDE, pReq);
   if (pTrans == NULL) goto _OVER;
 
   mDebug("trans:%d, used to create stb:%s", pTrans->id, pCreate->name);
-
   if (mndBuildStbFromReq(pMnode, &stbObj, pCreate, pDb) != 0) goto _OVER;
-
   if (mndAddStbToTrans(pMnode, pTrans, pDb, &stbObj) < 0) goto _OVER;
-
   if (mndTransPrepare(pMnode, pTrans) != 0) goto _OVER;
-
   code = 0;
 
 _OVER:
@@ -906,7 +916,8 @@ _OVER:
 }
 
 static int32_t mndCheckAlterStbReq(SMAlterStbReq *pAlter) {
-  if (pAlter->commentLen >= 0 || pAlter->ttl != 0) return 0;
+  if (pAlter->commentLen >= 0) return 0;
+  if (pAlter->ttl != 0) return 0;
 
   if (pAlter->numOfFields < 1 || pAlter->numOfFields != (int32_t)taosArrayGetSize(pAlter->pFields)) {
     terrno = TSDB_CODE_MND_INVALID_STB_OPTION;
@@ -969,6 +980,7 @@ static int32_t mndUpdateStbCommentAndTTL(const SStbObj *pOld, SStbObj *pNew, cha
     memcpy(pNew->comment, pComment, commentLen + 1);
   } else if (commentLen == 0) {
     pNew->commentLen = 0;
+  } else {
   }
 
   if (ttl >= 0) {
@@ -1245,7 +1257,10 @@ static int32_t mndAlterStbColumnBytes(SMnode *pMnode, const SStbObj *pOld, SStbO
 static int32_t mndSetAlterStbRedoLogs(SMnode *pMnode, STrans *pTrans, SDbObj *pDb, SStbObj *pStb) {
   SSdbRaw *pRedoRaw = mndStbActionEncode(pStb);
   if (pRedoRaw == NULL) return -1;
-  if (mndTransAppendRedolog(pTrans, pRedoRaw) != 0) return -1;
+  if (mndTransAppendRedolog(pTrans, pRedoRaw) != 0) {
+    sdbFreeRaw(pRedoRaw);
+    return -1;
+  }
   if (sdbSetRawStatus(pRedoRaw, SDB_STATUS_READY) != 0) return -1;
 
   return 0;
@@ -1254,7 +1269,10 @@ static int32_t mndSetAlterStbRedoLogs(SMnode *pMnode, STrans *pTrans, SDbObj *pD
 static int32_t mndSetAlterStbCommitLogs(SMnode *pMnode, STrans *pTrans, SDbObj *pDb, SStbObj *pStb) {
   SSdbRaw *pCommitRaw = mndStbActionEncode(pStb);
   if (pCommitRaw == NULL) return -1;
-  if (mndTransAppendCommitlog(pTrans, pCommitRaw) != 0) return -1;
+  if (mndTransAppendCommitlog(pTrans, pCommitRaw) != 0) {
+    sdbFreeRaw(pCommitRaw);
+    return -1;
+  }
   if (sdbSetRawStatus(pCommitRaw, SDB_STATUS_READY) != 0) return -1;
 
   return 0;
@@ -1269,7 +1287,7 @@ static int32_t mndSetAlterStbRedoActions(SMnode *pMnode, STrans *pTrans, SDbObj
   while (1) {
     pIter = sdbFetch(pSdb, SDB_VGROUP, pIter, (void **)&pVgroup);
     if (pIter == NULL) break;
-    if (pVgroup->dbUid != pDb->uid) {
+    if (!mndVgroupInDb(pVgroup, pDb->uid)) {
       sdbRelease(pSdb, pVgroup);
       continue;
     }
@@ -1388,7 +1406,7 @@ static int32_t mndBuildStbCfgImp(SDbObj *pDb, SStbObj *pStb, const char *tbName,
     pSchema->bytes = pSrcSchema->bytes;
   }
 
-  if (pStb->pFuncs) {
+  if (pStb->numOfFuncs > 0) {
     pRsp->pFuncs = taosArrayDup(pStb->pFuncs);
   }
 
@@ -1626,7 +1644,10 @@ _OVER:
 static int32_t mndSetDropStbRedoLogs(SMnode *pMnode, STrans *pTrans, SStbObj *pStb) {
   SSdbRaw *pRedoRaw = mndStbActionEncode(pStb);
   if (pRedoRaw == NULL) return -1;
-  if (mndTransAppendRedolog(pTrans, pRedoRaw) != 0) return -1;
+  if (mndTransAppendRedolog(pTrans, pRedoRaw) != 0) {
+    sdbFreeRaw(pRedoRaw);
+    return -1;
+  }
   if (sdbSetRawStatus(pRedoRaw, SDB_STATUS_DROPPING) != 0) return -1;
 
   return 0;
@@ -1635,7 +1656,10 @@ static int32_t mndSetDropStbRedoLogs(SMnode *pMnode, STrans *pTrans, SStbObj *pS
 static int32_t mndSetDropStbCommitLogs(SMnode *pMnode, STrans *pTrans, SStbObj *pStb) {
   SSdbRaw *pCommitRaw = mndStbActionEncode(pStb);
   if (pCommitRaw == NULL) return -1;
-  if (mndTransAppendCommitlog(pTrans, pCommitRaw) != 0) return -1;
+  if (mndTransAppendCommitlog(pTrans, pCommitRaw) != 0) {
+    sdbFreeRaw(pCommitRaw);
+    return -1;
+  }
   if (sdbSetRawStatus(pCommitRaw, SDB_STATUS_DROPPED) != 0) return -1;
 
   return 0;
@@ -1649,7 +1673,7 @@ static int32_t mndSetDropStbRedoActions(SMnode *pMnode, STrans *pTrans, SDbObj *
   while (1) {
     pIter = sdbFetch(pSdb, SDB_VGROUP, pIter, (void **)&pVgroup);
     if (pIter == NULL) break;
-    if (pVgroup->dbUid != pDb->uid) {
+    if (!mndVgroupInDb(pVgroup, pDb->uid)) {
       sdbRelease(pSdb, pVgroup);
       continue;
     }
@@ -1683,7 +1707,7 @@ static int32_t mndSetDropStbRedoActions(SMnode *pMnode, STrans *pTrans, SDbObj *
 
 static int32_t mndDropStb(SMnode *pMnode, SRpcMsg *pReq, SDbObj *pDb, SStbObj *pStb) {
   int32_t code = -1;
-  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_ROLLBACK, TRN_CONFLICT_DB_INSIDE, pReq);
+  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_RETRY, TRN_CONFLICT_DB_INSIDE, pReq);
   if (pTrans == NULL) goto _OVER;
 
   mDebug("trans:%d, used to drop stb:%s", pTrans->id, pStb->name);

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

@@ -248,7 +248,7 @@ static int32_t mndBuildStreamObjFromCreateReq(SMnode *pMnode, SStreamObj *pObj,
   pObj->status = 0;
 
   // TODO
-  pObj->dropPolicy = 0;
+  pObj->igExpired = pCreate->igExpired;
   pObj->trigger = pCreate->triggerType;
   pObj->triggerParam = pCreate->maxDelay;
   pObj->watermark = pCreate->watermark;
@@ -301,6 +301,7 @@ static int32_t mndBuildStreamObjFromCreateReq(SMnode *pMnode, SStreamObj *pObj,
       .streamQuery = true,
       .triggerType = pObj->trigger == STREAM_TRIGGER_MAX_DELAY ? STREAM_TRIGGER_WINDOW_CLOSE : pObj->trigger,
       .watermark = pObj->watermark,
+      .igExpired = pObj->igExpired,
   };
 
   // using ast and param to build physical plan

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

@@ -117,7 +117,7 @@ void mndReConfig(struct SSyncFSM *pFsm, const SRpcMsg *pMsg, SReConfigCbMeta cbM
   }
 }
 
-int32_t mndSnapshotStartRead(struct SSyncFSM *pFsm, void **ppReader) {
+int32_t mndSnapshotStartRead(struct SSyncFSM *pFsm, void *pParam, void **ppReader) {
   mDebug("start to read snapshot from sdb");
   SMnode *pMnode = pFsm->data;
   return sdbStartRead(pMnode->pSdb, (SSdbIter **)ppReader, NULL, NULL, NULL);

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

@@ -1146,13 +1146,13 @@ static int32_t mndTransExecuteRedoActionsSerial(SMnode *pMnode, STrans *pTrans)
         } else {
           code = TSDB_CODE_ACTION_IN_PROGRESS;
         }
-      }
+      } else if (pAction->rawWritten) {
-      if (pAction->rawWritten) {
         if (pAction->errCode != 0 && pAction->errCode != pAction->acceptableCode) {
           code = pAction->errCode;
         } else {
           mDebug("trans:%d, %s:%d write successfully", pTrans->id, mndTransStr(pAction->stage), action);
         }
+      } else {
       }
     }
 

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

@@ -426,7 +426,15 @@ static int32_t mndCompareDnodeId(int32_t *dnode1Id, int32_t *dnode2Id) { return
 static int32_t mndCompareDnodeVnodes(SDnodeObj *pDnode1, SDnodeObj *pDnode2) {
   float d1Score = (float)pDnode1->numOfVnodes / pDnode1->numOfSupportVnodes;
   float d2Score = (float)pDnode2->numOfVnodes / pDnode2->numOfSupportVnodes;
+#if 0  
+  if (d1Score == d2Score) {
+    return pDnode2->id - pDnode1->id;
+  } else {
+    return d1Score >= d2Score ? 1 : 0;
+  }
+#else
   return d1Score >= d2Score ? 1 : 0;
+#endif
 }
 
 void mndSortVnodeGid(SVgObj *pVgroup) {
@@ -1751,4 +1759,6 @@ _OVER:
 
   taosArrayDestroy(pArray);
   return code;
-}
+}
+
+bool mndVgroupInDb(SVgObj *pVgroup, int64_t dbUid) { return !pVgroup->isTsma && pVgroup->dbUid == dbUid; }

source/dnode/qnode/src/qnode.c

@@ -65,7 +65,7 @@ int32_t qndGetLoad(SQnode *pQnode, SQnodeLoad *pLoad) {
 }
 
 int32_t qndPreprocessQueryMsg(SQnode *pQnode, SRpcMsg * pMsg) {
-  if (TDMT_VND_QUERY != pMsg->msgType) {
+  if (TDMT_SCH_QUERY != pMsg->msgType && TDMT_SCH_MERGE_QUERY != pMsg->msgType) {
     return 0;
   }
 
@@ -78,28 +78,29 @@ int32_t qndProcessQueryMsg(SQnode *pQnode, int64_t ts, SRpcMsg *pMsg) {
   qTrace("message in qnode queue is processing");
 
   switch (pMsg->msgType) {
-    case TDMT_VND_QUERY:
+    case TDMT_SCH_QUERY:
+    case TDMT_SCH_MERGE_QUERY:
       code = qWorkerProcessQueryMsg(&handle, pQnode->pQuery, pMsg, ts);
       break;
-    case TDMT_VND_QUERY_CONTINUE:
+    case TDMT_SCH_QUERY_CONTINUE:
       code = qWorkerProcessCQueryMsg(&handle, pQnode->pQuery, pMsg, ts);
       break;
-    case TDMT_VND_FETCH:
+    case TDMT_SCH_FETCH:
       code = qWorkerProcessFetchMsg(pQnode, pQnode->pQuery, pMsg, ts);
       break;
-    case TDMT_VND_FETCH_RSP:
+    case TDMT_SCH_FETCH_RSP:
       code = qWorkerProcessFetchRsp(pQnode, pQnode->pQuery, pMsg, ts);
       break;
-    case TDMT_VND_CANCEL_TASK:
+    case TDMT_SCH_CANCEL_TASK:
       code = qWorkerProcessCancelMsg(pQnode, pQnode->pQuery, pMsg, ts);
       break;
-    case TDMT_VND_DROP_TASK:
+    case TDMT_SCH_DROP_TASK:
       code = qWorkerProcessDropMsg(pQnode, pQnode->pQuery, pMsg, ts);
       break;
     case TDMT_VND_CONSUME:
       // code =  tqProcessConsumeReq(pQnode->pTq, pMsg);
       // break;
-    case TDMT_VND_QUERY_HEARTBEAT:
+    case TDMT_SCH_QUERY_HEARTBEAT:
       code = qWorkerProcessHbMsg(pQnode, pQnode->pQuery, pMsg, ts);
       break;
     default:

source/dnode/vnode/CMakeLists.txt

@@ -31,6 +31,8 @@ target_sources(
     "src/sma/smaEnv.c"
     "src/sma/smaUtil.c"
     "src/sma/smaOpen.c"
+    "src/sma/smaCommit.c"
+    "src/sma/smaSnapshot.c"
     "src/sma/smaRollup.c"
     "src/sma/smaTimeRange.c"
 

source/dnode/vnode/inc/vnode.h

@@ -116,6 +116,7 @@ typedef void *tsdbReaderT;
 #define BLOCK_LOAD_TABLE_SEQ_ORDER  2
 #define BLOCK_LOAD_TABLE_RR_ORDER   3
 
+int32_t     tsdbSetTableId(tsdbReaderT reader, int64_t uid);
 int32_t     tsdbSetTableList(tsdbReaderT reader, SArray *tableList);
 tsdbReaderT tsdbReaderOpen(SVnode *pVnode, SQueryTableDataCond *pCond, SArray *tableList, uint64_t qId,
                            uint64_t taskId);

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

@@ -92,8 +92,9 @@ enum {
   TASK_TRIGGER_STAT_INIT = 0,
   TASK_TRIGGER_STAT_ACTIVE = 1,
   TASK_TRIGGER_STAT_INACTIVE = 2,
-  TASK_TRIGGER_STAT_CANCELLED = 3,
+  TASK_TRIGGER_STAT_PAUSED = 3,
-  TASK_TRIGGER_STAT_FINISHED = 4,
+  TASK_TRIGGER_STAT_CANCELLED = 4,
+  TASK_TRIGGER_STAT_FINISHED = 5,
 };
 void  tdDestroySmaEnv(SSmaEnv *pSmaEnv);
 void *tdFreeSmaEnv(SSmaEnv *pSmaEnv);
@@ -214,25 +215,22 @@ struct STFInfo {
 };
 
 struct STFile {
-  STFInfo   info;
-  STfsFile  f;
-  TdFilePtr pFile;
   uint8_t   state;
+  STFInfo   info;
+  char     *fname;
+  TdFilePtr pFile;
 };
 
-#define TD_TFILE_F(tf)            (&((tf)->f))
 #define TD_TFILE_PFILE(tf)        ((tf)->pFile)
 #define TD_TFILE_OPENED(tf)       (TD_TFILE_PFILE(tf) != NULL)
-#define TD_TFILE_FULL_NAME(tf)    (TD_TFILE_F(tf)->aname)
+#define TD_TFILE_FULL_NAME(tf)    ((tf)->fname)
-#define TD_TFILE_REL_NAME(tf)     (TD_TFILE_F(tf)->rname)
 #define TD_TFILE_OPENED(tf)       (TD_TFILE_PFILE(tf) != NULL)
 #define TD_TFILE_CLOSED(tf)       (!TD_TFILE_OPENED(tf))
 #define TD_TFILE_SET_CLOSED(f)    (TD_TFILE_PFILE(f) = NULL)
 #define TD_TFILE_SET_STATE(tf, s) ((tf)->state = (s))
-#define TD_TFILE_DID(tf)          (TD_TFILE_F(tf)->did)
 
-int32_t tdInitTFile(STFile *pTFile, STfs *pTfs, const char *fname);
+int32_t tdInitTFile(STFile *pTFile, const char *dname, const char *fname);
-int32_t tdCreateTFile(STFile *pTFile, STfs *pTfs, bool updateHeader, int8_t fType);
+int32_t tdCreateTFile(STFile *pTFile, bool updateHeader, int8_t fType);
 int32_t tdOpenTFile(STFile *pTFile, int flags);
 int64_t tdReadTFile(STFile *pTFile, void *buf, int64_t nbyte);
 int64_t tdSeekTFile(STFile *pTFile, int64_t offset, int whence);
@@ -244,8 +242,10 @@ int32_t tdLoadTFileHeader(STFile *pTFile, STFInfo *pInfo);
 int32_t tdUpdateTFileHeader(STFile *pTFile);
 void    tdUpdateTFileMagic(STFile *pTFile, void *pCksm);
 void    tdCloseTFile(STFile *pTFile);
+void    tdDestroyTFile(STFile *pTFile);
 
-void tdGetVndFileName(int32_t vgId, const char *dname, const char *fname, char *outputName);
+void tdGetVndFileName(int32_t vgId, const char *dname, const char *fname, int64_t version, char *outputName);
+void tdGetVndDirName(int32_t vgId, const char *dname, char *outputName);
 
 #ifdef __cplusplus
 }

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

@@ -157,11 +157,11 @@ typedef struct {
 static STqMgmt tqMgmt = {0};
 
 // tqRead
-int64_t tqFetchLog(STQ* pTq, STqHandle* pHandle, int64_t* fetchOffset, SWalHead** pHeadWithCkSum);
+int64_t tqFetchLog(STQ* pTq, STqHandle* pHandle, int64_t* fetchOffset, SWalCkHead** pHeadWithCkSum);
 
 // tqExec
 int32_t tqLogScanExec(STQ* pTq, STqExecHandle* pExec, SSubmitReq* pReq, SMqDataRsp* pRsp, int32_t workerId);
-int32_t tqScanSnapshot(STQ* pTq, const STqExecHandle* pExec, SMqDataRsp* pRsp, int32_t workerId);
+int32_t tqScanSnapshot(STQ* pTq, const STqExecHandle* pExec, SMqDataRsp* pRsp, STqOffsetVal offset, int32_t workerId);
 int32_t tqSendDataRsp(STQ* pTq, const SRpcMsg* pMsg, const SMqPollReq* pReq, const SMqDataRsp* pRsp);
 
 // tqMeta
@@ -178,11 +178,23 @@ STqOffsetStore* tqOffsetOpen();
 void            tqOffsetClose(STqOffsetStore*);
 STqOffset*      tqOffsetRead(STqOffsetStore* pStore, const char* subscribeKey);
 int32_t         tqOffsetWrite(STqOffsetStore* pStore, const STqOffset* pOffset);
+int32_t         tqOffsetDelete(STqOffsetStore* pStore, const char* subscribeKey);
 int32_t         tqOffsetSnapshot(STqOffsetStore* pStore);
 
 // tqSink
 void tqTableSink(SStreamTask* pTask, void* vnode, int64_t ver, void* data);
 
+static FORCE_INLINE void tqOffsetResetToData(STqOffsetVal* pOffsetVal, int64_t uid, int64_t ts) {
+  pOffsetVal->type = TMQ_OFFSET__SNAPSHOT_DATA;
+  pOffsetVal->uid = uid;
+  pOffsetVal->ts = ts;
+}
+
+static FORCE_INLINE void tqOffsetResetToLog(STqOffsetVal* pOffsetVal, int64_t ver) {
+  pOffsetVal->type = TMQ_OFFSET__LOG;
+  pOffsetVal->version = ver;
+}
+
 #ifdef __cplusplus
 }
 #endif

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

@@ -165,6 +165,9 @@ SSubmitReq* tdBlockToSubmit(const SArray* pBlocks, const STSchema* pSchema, bool
 int32_t smaOpen(SVnode* pVnode);
 int32_t smaCloseEnv(SSma* pSma);
 int32_t smaCloseEx(SSma* pSma);
+int32_t smaPreCommit(SSma* pSma);
+int32_t smaCommit(SSma* pSma);
+int32_t smaPostCommit(SSma* pSma);
 
 int32_t tdProcessTSmaCreate(SSma* pSma, int64_t version, const char* msg);
 int32_t tdProcessTSmaInsert(SSma* pSma, int64_t indexUid, const char* msg);

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

@@ -0,0 +1,150 @@
+/*
+ * Copyright (c) 2019 TAOS Data, Inc. <jhtao@taosdata.com>
+ *
+ * This program is free software: you can use, redistribute, and/or modify
+ * it under the terms of the GNU Affero General Public License, version 3
+ * or later ("AGPL"), as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.
+ *
+ * You should have received a copy of the GNU Affero General Public License
+ * along with this program. If not, see <http://www.gnu.org/licenses/>.
+ */
+
+#include "sma.h"
+
+static int32_t tdProcessRSmaPreCommitImpl(SSma *pSma);
+static int32_t tdProcessRSmaCommitImpl(SSma *pSma);
+static int32_t tdProcessRSmaPostCommitImpl(SSma *pSma);
+
+/**
+ * @brief Only applicable to Rollup SMA
+ *
+ * @param pSma
+ * @return int32_t
+ */
+int32_t smaPreCommit(SSma *pSma) { return tdProcessRSmaPreCommitImpl(pSma); }
+
+/**
+ * @brief Only applicable to Rollup SMA
+ *
+ * @param pSma
+ * @return int32_t
+ */
+int32_t smaCommit(SSma *pSma) { return tdProcessRSmaCommitImpl(pSma); }
+
+/**
+ * @brief Only applicable to Rollup SMA
+ *
+ * @param pSma
+ * @return int32_t
+ */
+int32_t smaPostCommit(SSma *pSma) { return tdProcessRSmaPostCommitImpl(pSma); }
+
+/**
+ * @brief pre-commit for rollup sma.
+ *  1) set trigger stat of rsma timer TASK_TRIGGER_STAT_PAUSED.
+ *  2) perform persist task for qTaskInfo
+ *  3) wait all triggered fetch tasks finished
+ *  4) set trigger stat of rsma timer TASK_TRIGGER_STAT_ACTIVE.
+ *  5) finish
+ *
+ * @param pSma
+ * @return int32_t
+ */
+static int32_t tdProcessRSmaPreCommitImpl(SSma *pSma) {
+  SSmaEnv *pSmaEnv = SMA_RSMA_ENV(pSma);
+  if (!pSmaEnv) {
+    return TSDB_CODE_SUCCESS;
+  }
+
+  SSmaStat  *pStat = SMA_ENV_STAT(pSmaEnv);
+  SRSmaStat *pRSmaStat = SMA_RSMA_STAT(pStat);
+
+  // step 1
+  atomic_store_8(RSMA_TRIGGER_STAT(pRSmaStat), TASK_TRIGGER_STAT_PAUSED);
+
+  // step 2
+
+  return TSDB_CODE_SUCCESS;
+}
+
+/**
+ * @brief commit for rollup sma
+ *
+ * @param pSma
+ * @return int32_t
+ */
+static int32_t tdProcessRSmaCommitImpl(SSma *pSma) {
+  SSmaEnv *pSmaEnv = SMA_RSMA_ENV(pSma);
+  if (!pSmaEnv) {
+    return TSDB_CODE_SUCCESS;
+  }
+  return TSDB_CODE_SUCCESS;
+}
+
+/**
+ * @brief post-commit for rollup sma
+ *  1) clean up the outdated qtaskinfo files
+ *
+ * @param pSma
+ * @return int32_t
+ */
+static int32_t tdProcessRSmaPostCommitImpl(SSma *pSma) {
+  SVnode *pVnode = pSma->pVnode;
+
+  if (!VND_IS_RSMA(pVnode)) {
+    return TSDB_CODE_SUCCESS;
+  }
+
+  int64_t       committed = pVnode->state.committed;
+  TdDirPtr      pDir = NULL;
+  TdDirEntryPtr pDirEntry = NULL;
+  char          dir[TSDB_FILENAME_LEN];
+  char          bname[TSDB_FILENAME_LEN];
+  const char   *pattern = "^v[0-9]+qtaskinfo\\.ver([0-9]+)?$";
+  regex_t       regex;
+
+  tdGetVndDirName(TD_VID(pVnode), VNODE_RSMA_DIR, dir);
+
+  // Resource allocation and init
+  regcomp(&regex, pattern, REG_EXTENDED);
+
+  if ((pDir = taosOpenDir(dir)) == NULL) {
+    terrno = TAOS_SYSTEM_ERROR(errno);
+    smaWarn("rsma post-commit open dir %s failed since %s", dir, terrstr());
+    return TSDB_CODE_FAILED;
+  }
+
+  regmatch_t regMatch[2];
+  while ((pDirEntry = taosReadDir(pDir)) != NULL) {
+    char *entryName = taosGetDirEntryName(pDirEntry);
+    if (!entryName) {
+      continue;
+    }
+    char *fileName = taosDirEntryBaseName(entryName);
+    int   code = regexec(&regex, bname, 2, regMatch, 0);
+
+    if (code == 0) {
+      // match
+      printf("match 0 = %s\n", (char *)POINTER_SHIFT(fileName, regMatch[0].rm_so));
+      printf("match 1 = %s\n", (char *)POINTER_SHIFT(fileName, regMatch[1].rm_so));
+    } else if (code == REG_NOMATCH) {
+      // not match
+      smaInfo("rsma post-commit not match %s", fileName);
+      continue;
+    } else {
+      // has other error
+      terrno = TAOS_SYSTEM_ERROR(code);
+      smaWarn("rsma post-commit regexec failed since %s", terrstr());
+
+      taosCloseDir(&pDir);
+      regfree(&regex);
+      return TSDB_CODE_FAILED;
+    }
+  }
+  taosCloseDir(&pDir);
+  return TSDB_CODE_SUCCESS;
+}

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

@@ -23,8 +23,7 @@ SSmaMgmt smaMgmt = {
     .smaRef = -1,
 };
 
-typedef enum { TD_QTASK_TMP_F = 0, TD_QTASK_CUR_F } TD_QTASK_FILE_T;
+#define TD_QTASKINFO_FNAME_PREFIX "qtaskinfo.ver"
-static const char                *tdQTaskInfoFname[] = {"qtaskinfo.t", "qtaskinfo"};
 typedef struct SRSmaQTaskInfoItem SRSmaQTaskInfoItem;
 typedef struct SRSmaQTaskInfoIter SRSmaQTaskInfoIter;
 
@@ -37,7 +36,7 @@ static int32_t tdExecuteRSmaImpl(SSma *pSma, const void *pMsg, int32_t inputType
 static void    tdRSmaFetchTrigger(void *param, void *tmrId);
 static void    tdRSmaPersistTrigger(void *param, void *tmrId);
 static void   *tdRSmaPersistExec(void *param);
-static void    tdRSmaQTaskInfoGetFName(int32_t vid, int8_t ftype, char *outputName);
+static void    tdRSmaQTaskInfoGetFName(int32_t vid, int64_t version, char *outputName);
 
 static int32_t tdRSmaQTaskInfoIterInit(SRSmaQTaskInfoIter *pIter, STFile *pTFile);
 static int32_t tdRSmaQTaskInfoIterNextBlock(SRSmaQTaskInfoIter *pIter, bool *isFinish);
@@ -88,8 +87,8 @@ struct SRSmaQTaskInfoIter {
   int32_t nBufPos;
 };
 
-static void tdRSmaQTaskInfoGetFName(int32_t vgId, int8_t ftype, char *outputName) {
+static void tdRSmaQTaskInfoGetFName(int32_t vgId, int64_t version, char *outputName) {
-  tdGetVndFileName(vgId, VNODE_RSMA_DIR, tdQTaskInfoFname[ftype], outputName);
+  tdGetVndFileName(vgId, VNODE_RSMA_DIR, TD_QTASKINFO_FNAME_PREFIX, version, outputName);
 }
 
 static FORCE_INLINE int32_t tdRSmaQTaskInfoContLen(int32_t lenWithHead) {
@@ -493,7 +492,6 @@ static int32_t tdProcessSubmitReq(STsdb *pTsdb, int64_t version, void *pReq) {
 }
 
 static int32_t tdFetchSubmitReqSuids(SSubmitReq *pMsg, STbUidStore *pStore) {
-  ASSERT(pMsg != NULL);
   SSubmitMsgIter msgIter = {0};
   SSubmitBlk    *pBlock = NULL;
   SSubmitBlkIter blkIter = {0};
@@ -501,19 +499,26 @@ static int32_t tdFetchSubmitReqSuids(SSubmitReq *pMsg, STbUidStore *pStore) {
 
   terrno = TSDB_CODE_SUCCESS;
 
-  if (tInitSubmitMsgIter(pMsg, &msgIter) < 0) return -1;
+  if (tInitSubmitMsgIter(pMsg, &msgIter) < 0) {
+    return -1;
+  }
   while (true) {
-    if (tGetSubmitMsgNext(&msgIter, &pBlock) < 0) return -1;
+    if (tGetSubmitMsgNext(&msgIter, &pBlock) < 0) {
+      return -1;
+    }
 
     if (!pBlock) break;
     tdUidStorePut(pStore, msgIter.suid, NULL);
   }
 
-  if (terrno != TSDB_CODE_SUCCESS) return -1;
+  if (terrno != TSDB_CODE_SUCCESS) {
+    return -1;
+  }
   return 0;
 }
 
 static void tdDestroySDataBlockArray(SArray *pArray) {
+  // TODO
 #if 0
   for (int32_t i = 0; i < taosArrayGetSize(pArray); ++i) {
     SSDataBlock *pDataBlock = taosArrayGet(pArray, i);
@@ -598,33 +603,54 @@ static void tdRSmaFetchTrigger(void *param, void *tmrId) {
 
   pSma = RSMA_INFO_SMA(pItem->pRsmaInfo);
 
-  // if rsma trigger stat in cancelled or finished, not start fetch task anymore
+  // if rsma trigger stat in paused, cancelled or finished, not start fetch task
   int8_t rsmaTriggerStat = atomic_load_8(RSMA_TRIGGER_STAT(pStat));
-  if (rsmaTriggerStat == TASK_TRIGGER_STAT_CANCELLED || rsmaTriggerStat == TASK_TRIGGER_STAT_FINISHED) {
+  switch (rsmaTriggerStat) {
-    taosReleaseRef(smaMgmt.smaRef, pItem->refId);
+    case TASK_TRIGGER_STAT_PAUSED:
-    smaDebug("vgId:%d, not fetch rsma level %" PRIi8 " data for table:%" PRIi64 " since stat is cancelled",
+    case TASK_TRIGGER_STAT_CANCELLED:
-             SMA_VID(pSma), pItem->level, pItem->pRsmaInfo->suid);
+    case TASK_TRIGGER_STAT_FINISHED: {
-    return;
+      taosReleaseRef(smaMgmt.smaRef, pItem->refId);
+      smaDebug("vgId:%d, not fetch rsma level %" PRIi8 " data for table:%" PRIi64 " since stat is cancelled",
+               SMA_VID(pSma), pItem->level, pItem->pRsmaInfo->suid);
+      return;
+    }
+    default:
+      break;
   }
 
   int8_t fetchTriggerStat =
       atomic_val_compare_exchange_8(&pItem->triggerStat, TASK_TRIGGER_STAT_ACTIVE, TASK_TRIGGER_STAT_INACTIVE);
-  if (fetchTriggerStat == TASK_TRIGGER_STAT_ACTIVE) {
+  switch (fetchTriggerStat) {
-    smaDebug("vgId:%d, fetch rsma level %" PRIi8 " data for table:%" PRIi64 " since stat is active", SMA_VID(pSma),
+    case TASK_TRIGGER_STAT_ACTIVE: {
-             pItem->level, pItem->pRsmaInfo->suid);
+      smaDebug("vgId:%d, fetch rsma level %" PRIi8 " data for table:%" PRIi64 " since stat is active", SMA_VID(pSma),
+               pItem->level, pItem->pRsmaInfo->suid);
 
-    tdRefSmaStat(pSma, (SSmaStat *)pStat);
+      tdRefSmaStat(pSma, (SSmaStat *)pStat);
 
-    SSDataBlock dataBlock = {.info.type = STREAM_GET_ALL};
+      SSDataBlock dataBlock = {.info.type = STREAM_GET_ALL};
-    qSetStreamInput(pItem->taskInfo, &dataBlock, STREAM_INPUT__DATA_BLOCK, false);
+      qSetStreamInput(pItem->taskInfo, &dataBlock, STREAM_INPUT__DATA_BLOCK, false);
-    tdFetchAndSubmitRSmaResult(pItem, STREAM_INPUT__DATA_BLOCK);
+      tdFetchAndSubmitRSmaResult(pItem, STREAM_INPUT__DATA_BLOCK);
 
-    tdUnRefSmaStat(pSma, (SSmaStat *)pStat);
+      tdUnRefSmaStat(pSma, (SSmaStat *)pStat);
-
+    } break;
-  } else {
+    case TASK_TRIGGER_STAT_PAUSED: {
-    smaDebug("vgId:%d, not fetch rsma level %" PRIi8 " data for table:%" PRIi64 " since stat is inactive",
+      smaDebug("vgId:%d, not fetch rsma level %" PRIi8 " data for table:%" PRIi64 " since stat is paused",
-             SMA_VID(pSma), pItem->level, pItem->pRsmaInfo->suid);
+               SMA_VID(pSma), pItem->level, pItem->pRsmaInfo->suid);
+    } break;
+    case TASK_TRIGGER_STAT_INACTIVE: {
+      smaDebug("vgId:%d, not fetch rsma level %" PRIi8 " data for table:%" PRIi64 " since stat is inactive",
+               SMA_VID(pSma), pItem->level, pItem->pRsmaInfo->suid);
+    } break;
+    case TASK_TRIGGER_STAT_INIT: {
+      smaDebug("vgId:%d, not fetch rsma level %" PRIi8 " data for table:%" PRIi64 " since stat is init", SMA_VID(pSma),
+               pItem->level, pItem->pRsmaInfo->suid);
+    } break;
+    default: {
+      smaWarn("vgId:%d, not fetch rsma level %" PRIi8 " data for table:%" PRIi64 " since stat is unknown",
+              SMA_VID(pSma), pItem->level, pItem->pRsmaInfo->suid);
+    } break;
   }
+
 _end:
   taosReleaseRef(smaMgmt.smaRef, pItem->refId);
 }
@@ -780,10 +806,10 @@ _err:
 static int32_t tdRSmaRestoreQTaskInfoReload(SSma *pSma) {
   SVnode *pVnode = pSma->pVnode;
   STFile  tFile = {0};
-  char    qTaskInfoFName[TSDB_FILENAME_LEN];
+  char    qTaskInfoFName[TSDB_FILENAME_LEN] = {0};
 
-  tdRSmaQTaskInfoGetFName(TD_VID(pVnode), TD_QTASK_TMP_F, qTaskInfoFName);
+  tdRSmaQTaskInfoGetFName(TD_VID(pVnode), pVnode->state.committed, qTaskInfoFName);
-  if (tdInitTFile(&tFile, pVnode->pTfs, qTaskInfoFName) < 0) {
+  if (tdInitTFile(&tFile, tfsGetPrimaryPath(pVnode->pTfs), qTaskInfoFName) < 0) {
     goto _err;
   }
 
@@ -799,17 +825,20 @@ static int32_t tdRSmaRestoreQTaskInfoReload(SSma *pSma) {
   if (tdRSmaQTaskInfoIterInit(&fIter, &tFile) < 0) {
     tdRSmaQTaskInfoIterDestroy(&fIter);
     tdCloseTFile(&tFile);
+    tdDestroyTFile(&tFile);
     goto _err;
   }
 
   if (tdRSmaQTaskInfoRestore(pSma, &fIter) < 0) {
     tdRSmaQTaskInfoIterDestroy(&fIter);
     tdCloseTFile(&tFile);
+    tdDestroyTFile(&tFile);
     goto _err;
   }
 
   tdRSmaQTaskInfoIterDestroy(&fIter);
   tdCloseTFile(&tFile);
+  tdDestroyTFile(&tFile);
   return TSDB_CODE_SUCCESS;
 _err:
   smaError("rsma restore, qtaskinfo reload failed since %s", terrstr());
@@ -931,19 +960,21 @@ static int32_t tdRSmaQTaskInfoIterNextBlock(SRSmaQTaskInfoIter *pIter, bool *isF
   }
 
   if (tdSeekTFile(pTFile, pIter->offset, SEEK_SET) < 0) {
-    ASSERT(0);
     return TSDB_CODE_FAILED;
   }
 
   if (tdReadTFile(pTFile, pIter->qBuf, nBytes) != nBytes) {
-    ASSERT(0);
     return TSDB_CODE_FAILED;
   }
 
   int32_t infoLen = 0;
   taosDecodeFixedI32(pIter->qBuf, &infoLen);
   if (infoLen > nBytes) {
-    ASSERT(infoLen > RSMA_QTASKINFO_BUFSIZE);
+    if (infoLen <= RSMA_QTASKINFO_BUFSIZE) {
+      terrno = TSDB_CODE_RSMA_FILE_CORRUPTED;
+      smaError("iterate rsma qtaskinfo file %s failed since %s", TD_TFILE_FULL_NAME(pIter->pTFile), terrstr());
+      return TSDB_CODE_FAILED;
+    }
     pIter->nAlloc = infoLen;
     void *pBuf = taosMemoryRealloc(pIter->pBuf, infoLen);
     if (!pBuf) {
@@ -955,12 +986,10 @@ static int32_t tdRSmaQTaskInfoIterNextBlock(SRSmaQTaskInfoIter *pIter, bool *isF
     nBytes = infoLen;
 
     if (tdSeekTFile(pTFile, pIter->offset, SEEK_SET)) {
-      ASSERT(0);
       return TSDB_CODE_FAILED;
     }
 
     if (tdReadTFile(pTFile, pIter->pBuf, nBytes) != nBytes) {
-      ASSERT(0);
       return TSDB_CODE_FAILED;
     }
   }
@@ -977,7 +1006,6 @@ static int32_t tdRSmaQTaskInfoRestore(SSma *pSma, SRSmaQTaskInfoIter *pIter) {
     // block iter
     bool isFinish = false;
     if (tdRSmaQTaskInfoIterNextBlock(pIter, &isFinish) < 0) {
-      ASSERT(0);
       return TSDB_CODE_FAILED;
     }
     if (isFinish) {
@@ -989,6 +1017,7 @@ static int32_t tdRSmaQTaskInfoRestore(SSma *pSma, SRSmaQTaskInfoIter *pIter) {
     pIter->qBuf = taosDecodeFixedI32(pIter->qBuf, &qTaskInfoLenWithHead);
     if (qTaskInfoLenWithHead < RSMA_QTASKINFO_HEAD_LEN) {
       terrno = TSDB_CODE_TDB_FILE_CORRUPTED;
+      smaError("restore rsma qtaskinfo file %s failed since %s", TD_TFILE_FULL_NAME(pIter->pTFile), terrstr());
       return TSDB_CODE_FAILED;
     }
 
@@ -1025,22 +1054,16 @@ static int32_t tdRSmaQTaskInfoRestore(SSma *pSma, SRSmaQTaskInfoIter *pIter) {
   return TSDB_CODE_SUCCESS;
 }
 
-static void *tdRSmaPersistExec(void *param) {
+static int32_t tdRSmaPersistExecImpl(SRSmaStat *pRSmaStat) {
-  setThreadName("rsma-task-persist");
+  SSma   *pSma = pRSmaStat->pSma;
-  SRSmaStat *pRSmaStat = param;
+  SVnode *pVnode = pSma->pVnode;
-  SSma      *pSma = pRSmaStat->pSma;
+  int32_t vid = SMA_VID(pSma);
-  STfs      *pTfs = pSma->pVnode->pTfs;
+  int64_t toffset = 0;
-  int32_t    vid = SMA_VID(pSma);
+  bool    isFileCreated = false;
-  int64_t    toffset = 0;
-  bool       isFileCreated = false;
-
-  if (TASK_TRIGGER_STAT_CANCELLED == atomic_load_8(RSMA_TRIGGER_STAT(pRSmaStat))) {
-    goto _end;
-  }
 
   void *infoHash = taosHashIterate(RSMA_INFO_HASH(pRSmaStat), NULL);
   if (!infoHash) {
-    goto _end;
+    return TSDB_CODE_SUCCESS;
   }
 
   STFile tFile = {0};
@@ -1074,9 +1097,13 @@ static void *tdRSmaPersistExec(void *param) {
 
       if (!isFileCreated) {
         char qTaskInfoFName[TSDB_FILENAME_LEN];
-        tdRSmaQTaskInfoGetFName(vid, TD_QTASK_TMP_F, qTaskInfoFName);
+        tdRSmaQTaskInfoGetFName(vid, pSma->pVnode->state.applied, qTaskInfoFName);
-        tdInitTFile(&tFile, pTfs, qTaskInfoFName);
+        if (tdInitTFile(&tFile, tfsGetPrimaryPath(pVnode->pTfs), qTaskInfoFName) < 0) {
-        tdCreateTFile(&tFile, pTfs, true, -1);
+          goto _err;
+        }
+        if (tdCreateTFile(&tFile, true, -1) < 0) {
+          goto _err;
+        }
 
         isFileCreated = true;
       }
@@ -1101,49 +1128,55 @@ static void *tdRSmaPersistExec(void *param) {
 
     infoHash = taosHashIterate(RSMA_INFO_HASH(pRSmaStat), infoHash);
   }
-_normal:
+
   if (isFileCreated) {
     if (tdUpdateTFileHeader(&tFile) < 0) {
       smaError("vgId:%d, rsma, failed to update tfile %s header since %s", vid, TD_TFILE_FULL_NAME(&tFile),
                tstrerror(terrno));
-      tdCloseTFile(&tFile);
-      tdRemoveTFile(&tFile);
       goto _err;
     } else {
       smaDebug("vgId:%d, rsma, succeed to update tfile %s header", vid, TD_TFILE_FULL_NAME(&tFile));
     }
 
     tdCloseTFile(&tFile);
-
+    tdDestroyTFile(&tFile);
-    char newFName[TSDB_FILENAME_LEN];
-    strncpy(newFName, TD_TFILE_FULL_NAME(&tFile), TSDB_FILENAME_LEN);
-    char *pos = strstr(newFName, tdQTaskInfoFname[TD_QTASK_TMP_F]);
-    strncpy(pos, tdQTaskInfoFname[TD_QTASK_TMP_F], TSDB_FILENAME_LEN - POINTER_DISTANCE(pos, newFName));
-    if (taosRenameFile(TD_TFILE_FULL_NAME(&tFile), newFName) != 0) {
-      smaError("vgId:%d, rsma, failed to rename %s to %s", vid, TD_TFILE_FULL_NAME(&tFile), newFName);
-      goto _err;
-    } else {
-      smaDebug("vgId:%d, rsma, succeed to rename %s to %s", vid, TD_TFILE_FULL_NAME(&tFile), newFName);
-    }
   }
-  goto _end;
+  return TSDB_CODE_SUCCESS;
 _err:
   if (isFileCreated) {
     tdRemoveTFile(&tFile);
+    tdDestroyTFile(&tFile);
   }
+  return TSDB_CODE_FAILED;
+}
+
+static void *tdRSmaPersistExec(void *param) {
+  setThreadName("rsma-task-persist");
+  SRSmaStat *pRSmaStat = param;
+  SSma      *pSma = pRSmaStat->pSma;
+
+  int8_t triggerStat = atomic_load_8(RSMA_TRIGGER_STAT(pRSmaStat));
+
+  if (TASK_TRIGGER_STAT_CANCELLED == triggerStat || TASK_TRIGGER_STAT_PAUSED == triggerStat) {
+    goto _end;
+  }
+
+  // execution
+  tdRSmaPersistExecImpl(pRSmaStat);
+
 _end:
   if (TASK_TRIGGER_STAT_INACTIVE == atomic_val_compare_exchange_8(RSMA_TRIGGER_STAT(pRSmaStat),
                                                                   TASK_TRIGGER_STAT_INACTIVE,
                                                                   TASK_TRIGGER_STAT_ACTIVE)) {
-    smaDebug("vgId:%d, rsma persist task is active again", vid);
+    smaDebug("vgId:%d, rsma persist task is active again", SMA_VID(pSma));
   } else if (TASK_TRIGGER_STAT_CANCELLED == atomic_val_compare_exchange_8(RSMA_TRIGGER_STAT(pRSmaStat),
                                                                           TASK_TRIGGER_STAT_CANCELLED,
                                                                           TASK_TRIGGER_STAT_FINISHED)) {
-    smaDebug("vgId:%d, rsma persist task is cancelled", vid);
+    smaDebug("vgId:%d, rsma persist task is cancelled", SMA_VID(pSma));
   } else {
-    smaWarn("vgId:%d, rsma persist task in abnormal stat %" PRIi8, vid, atomic_load_8(RSMA_TRIGGER_STAT(pRSmaStat)));
+    smaWarn("vgId:%d, rsma persist task in stat %" PRIi8, SMA_VID(pSma), atomic_load_8(RSMA_TRIGGER_STAT(pRSmaStat)));
-    ASSERT(0);
   }
+
   atomic_store_8(RSMA_RUNNING_STAT(pRSmaStat), 0);
   taosReleaseRef(smaMgmt.smaRef, pRSmaStat->refId);
   taosThreadExit(NULL);
@@ -1166,8 +1199,8 @@ static void tdRSmaPersistTask(SRSmaStat *pRSmaStat) {
                                                                             TASK_TRIGGER_STAT_FINISHED)) {
       smaDebug("vgId:%d, persist task is cancelled and set finished", SMA_VID(pRSmaStat->pSma));
     } else {
-      smaWarn("vgId:%d, persist task in abnormal stat %" PRIi8, atomic_load_8(RSMA_TRIGGER_STAT(pRSmaStat)),
+      smaWarn("vgId:%d, persist task in abnormal stat %" PRIi8, SMA_VID(pRSmaStat->pSma),
-              SMA_VID(pRSmaStat->pSma));
+              atomic_load_8(RSMA_TRIGGER_STAT(pRSmaStat)));
       ASSERT(0);
     }
     atomic_store_8(RSMA_RUNNING_STAT(pRSmaStat), 0);
@@ -1216,6 +1249,9 @@ static void tdRSmaPersistTrigger(void *param, void *tmrId) {
       atomic_store_8(RSMA_TRIGGER_STAT(pRSmaStat), TASK_TRIGGER_STAT_FINISHED);
       smaDebug("rsma persistence not start since cancelled and finished");
     } break;
+    case TASK_TRIGGER_STAT_PAUSED: {
+      smaDebug("rsma persistence not start since paused");
+    } break;
     case TASK_TRIGGER_STAT_INACTIVE: {
       smaDebug("rsma persistence not start since inactive");
     } break;

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

@@ -0,0 +1,16 @@
+/*
+ * Copyright (c) 2019 TAOS Data, Inc. <jhtao@taosdata.com>
+ *
+ * This program is free software: you can use, redistribute, and/or modify
+ * it under the terms of the GNU Affero General Public License, version 3
+ * or later ("AGPL"), as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.
+ *
+ * You should have received a copy of the GNU Affero General Public License
+ * along with this program. If not, see <http://www.gnu.org/licenses/>.
+ */
+
+#include "sma.h"

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

@@ -113,7 +113,7 @@ int32_t tdProcessTSmaInsertImpl(SSma *pSma, int64_t indexUid, const char *msg) {
   if (!pDataBlocks) {
     terrno = TSDB_CODE_TSMA_INVALID_PTR;
     smaWarn("vgId:%d, insert tsma data failed since pDataBlocks is NULL", SMA_VID(pSma));
-    return terrno;
+    return TSDB_CODE_FAILED;
   }
 
   if (taosArrayGetSize(pDataBlocks) <= 0) {
@@ -127,9 +127,9 @@ int32_t tdProcessTSmaInsertImpl(SSma *pSma, int64_t indexUid, const char *msg) {
     return TSDB_CODE_FAILED;
   }
 
-  SSmaEnv      *pEnv = SMA_TSMA_ENV(pSma);
+  SSmaEnv   *pEnv = SMA_TSMA_ENV(pSma);
-  SSmaStat     *pStat = NULL;
+  SSmaStat  *pStat = NULL;
-  STSmaStat    *pItem = NULL;
+  STSmaStat *pTsmaStat = NULL;
 
   if (!pEnv || !(pStat = SMA_ENV_STAT(pEnv))) {
     terrno = TSDB_CODE_TSMA_INVALID_STAT;
@@ -137,32 +137,43 @@ int32_t tdProcessTSmaInsertImpl(SSma *pSma, int64_t indexUid, const char *msg) {
   }
 
   tdRefSmaStat(pSma, pStat);
-  pItem = &pStat->tsmaStat;
+  pTsmaStat = SMA_TSMA_STAT(pStat);
 
-  ASSERT(pItem);
+  if (!pTsmaStat->pTSma) {
-
-  if (!pItem->pTSma) {
     STSma *pTSma = metaGetSmaInfoByIndex(SMA_META(pSma), indexUid);
     if (!pTSma) {
-      terrno = TSDB_CODE_TSMA_NO_INDEX_IN_META;
+      smaError("vgId:%d, failed to get STSma while tsma insert for smaIndex %" PRIi64 " since %s", SMA_VID(pSma),
-      smaWarn("vgId:%d, tsma insert for smaIndex %" PRIi64 " since %s", SMA_VID(pSma), indexUid, tstrerror(terrno));
+               indexUid, tstrerror(terrno));
-      return TSDB_CODE_FAILED;
+      goto _err;
+    }
+    pTsmaStat->pTSma = pTSma;
+    pTsmaStat->pTSchema = metaGetTbTSchema(SMA_META(pSma), pTSma->dstTbUid, -1);
+    if (!pTsmaStat->pTSchema) {
+      smaError("vgId:%d, failed to get STSchema while tsma insert for smaIndex %" PRIi64 " since %s", SMA_VID(pSma),
+               indexUid, tstrerror(terrno));
+      goto _err;
     }
-    pItem->pTSma = pTSma;
-    pItem->pTSchema = metaGetTbTSchema(SMA_META(pSma), pTSma->dstTbUid, -1);
-    ASSERT(pItem->pTSchema);  // TODO
   }
 
-  ASSERT(pItem->pTSma->indexUid == indexUid);
+  if (pTsmaStat->pTSma->indexUid != indexUid) {
+    terrno = TSDB_CODE_VND_APP_ERROR;
+    smaError("vgId:%d, tsma insert for smaIndex %" PRIi64 "(!=%" PRIi64 ") failed since %s", SMA_VID(pSma), indexUid,
+             pTsmaStat->pTSma->indexUid, tstrerror(terrno));
+    goto _err;
+  }
 
-  SSubmitReq *pSubmitReq = NULL;
+  SSubmitReq *pSubmitReq = tdBlockToSubmit((const SArray *)msg, pTsmaStat->pTSchema, true, pTsmaStat->pTSma->dstTbUid,
+                                           pTsmaStat->pTSma->dstTbName, pTsmaStat->pTSma->dstVgId);
 
-  pSubmitReq = tdBlockToSubmit((const SArray *)msg, pItem->pTSchema, true, pItem->pTSma->dstTbUid,
+  if (!pSubmitReq) {
-                               pItem->pTSma->dstTbName, pItem->pTSma->dstVgId);
+    smaError("vgId:%d, failed to gen submit blk while tsma insert for smaIndex %" PRIi64 " since %s", SMA_VID(pSma),
+             indexUid, tstrerror(terrno));
+    goto _err;
+  }
 
-  ASSERT(pSubmitReq);  // TODO
+#if 0
-
+   ASSERT(!strncasecmp("td.tsma.rst.tb", pTsmaStat->pTSma->dstTbName, 14));
-  ASSERT(!strncasecmp("td.tsma.rst.tb", pItem->pTSma->dstTbName, 14));
+#endif
 
   SRpcMsg submitReqMsg = {
       .msgType = TDMT_VND_SUBMIT,
@@ -170,9 +181,15 @@ int32_t tdProcessTSmaInsertImpl(SSma *pSma, int64_t indexUid, const char *msg) {
       .contLen = ntohl(pSubmitReq->length),
   };
 
-  ASSERT(tmsgPutToQueue(&pSma->pVnode->msgCb, WRITE_QUEUE, &submitReqMsg) == 0);
+  if (tmsgPutToQueue(&pSma->pVnode->msgCb, WRITE_QUEUE, &submitReqMsg) < 0) {
+    smaError("vgId:%d, failed to put SubmitReq msg while tsma insert for smaIndex %" PRIi64 " since %s", SMA_VID(pSma),
+             indexUid, tstrerror(terrno));
+    goto _err;
+  }
 
   tdUnRefSmaStat(pSma, pStat);
-
   return TSDB_CODE_SUCCESS;
+_err:
+  tdUnRefSmaStat(pSma, pStat);
+  return TSDB_CODE_FAILED;
 }

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

@@ -179,72 +179,83 @@ void tdCloseTFile(STFile *pTFile) {
   }
 }
 
-void tdGetVndFileName(int32_t vgId, const char *dname, const char *fname, char *outputName) {
+void tdDestroyTFile(STFile *pTFile) { taosMemoryFreeClear(TD_TFILE_FULL_NAME(pTFile)); }
-  snprintf(outputName, TSDB_FILENAME_LEN, "vnode/vnode%d/%s/%s", vgId, dname, fname);
+
+void tdGetVndFileName(int32_t vgId, const char *dname, const char *fname, int64_t version, char *outputName) {
+  if (version < 0) {
+    snprintf(outputName, TSDB_FILENAME_LEN, "vnode/vnode%d/%s/v%d%s", vgId, dname, vgId, fname);
+  } else {
+    snprintf(outputName, TSDB_FILENAME_LEN, "vnode/vnode%d/%s/v%d%s%" PRIi64, vgId, dname, vgId, fname, version);
+  }
 }
 
-int32_t tdInitTFile(STFile *pTFile, STfs *pTfs, const char *fname) {
+void tdGetVndDirName(int32_t vgId, const char *dname, char *outputName) {
-  char    fullname[TSDB_FILENAME_LEN];
+  snprintf(outputName, TSDB_FILENAME_LEN, "vnode/vnode%d/%s", vgId, dname);
-  SDiskID did = {0};
+}
 
+int32_t tdInitTFile(STFile *pTFile, const char *dname, const char *fname) {
   TD_TFILE_SET_STATE(pTFile, TD_FILE_STATE_OK);
   TD_TFILE_SET_CLOSED(pTFile);
 
   memset(&(pTFile->info), 0, sizeof(pTFile->info));
   pTFile->info.magic = TD_FILE_INIT_MAGIC;
 
-  if (tfsAllocDisk(pTfs, 0, &did) < 0) {
+  char tmpName[TSDB_FILENAME_LEN * 2 + 32] = {0};
-    terrno = TSDB_CODE_NO_AVAIL_DISK;
+  snprintf(tmpName, TSDB_FILENAME_LEN * 2 + 32, "%s%s%s", dname, TD_DIRSEP, fname);
+  int32_t tmpNameLen = strlen(tmpName) + 1;
+  pTFile->fname = taosMemoryMalloc(tmpNameLen);
+  if (!pTFile->fname) {
+    terrno = TSDB_CODE_OUT_OF_MEMORY;
     return -1;
   }
-
+  tstrncpy(pTFile->fname, tmpName, tmpNameLen);
-  tfsInitFile(pTfs, &(pTFile->f), did, fname);
 
   return 0;
 }
 
-int32_t tdCreateTFile(STFile *pTFile, STfs *pTfs, bool updateHeader, int8_t fType) {
+int32_t tdCreateTFile(STFile *pTFile, bool updateHeader, int8_t fType) {
   ASSERT(pTFile->info.fsize == 0 && pTFile->info.magic == TD_FILE_INIT_MAGIC);
 
   pTFile->pFile = taosOpenFile(TD_TFILE_FULL_NAME(pTFile), TD_FILE_CREATE | TD_FILE_WRITE | TD_FILE_TRUNC);
   if (pTFile->pFile == NULL) {
     if (errno == ENOENT) {
       // Try to create directory recursively
-      char *s = strdup(TD_TFILE_REL_NAME(pTFile));
+      if (taosMulMkDir(taosDirName(TD_TFILE_FULL_NAME(pTFile))) != 0) {
-      if (tfsMkdirRecurAt(pTfs, taosDirName(s), TD_TFILE_DID(pTFile)) < 0) {
-        taosMemoryFreeClear(s);
-        return -1;
-      }
-      taosMemoryFreeClear(s);
-
-      pTFile->pFile = taosOpenFile(TD_TFILE_FULL_NAME(pTFile), TD_FILE_CREATE | TD_FILE_WRITE | TD_FILE_TRUNC);
-      if (pTFile->pFile == NULL) {
         terrno = TAOS_SYSTEM_ERROR(errno);
         return -1;
+      } else {
+        pTFile->pFile = taosOpenFile(TD_TFILE_FULL_NAME(pTFile), TD_FILE_CREATE | TD_FILE_WRITE | TD_FILE_TRUNC);
+        if (pTFile->pFile == NULL) {
+          terrno = TAOS_SYSTEM_ERROR(errno);
+          return -1;
+        }
       }
-    } else {
+    }
-      terrno = TAOS_SYSTEM_ERROR(errno);
+
+    if (!updateHeader) {
+      return 0;
+    }
+
+    pTFile->info.fsize += TD_FILE_HEAD_SIZE;
+    pTFile->info.fver = 0;
+
+    if (tdUpdateTFileHeader(pTFile) < 0) {
+      tdCloseTFile(pTFile);
+      tdRemoveTFile(pTFile);
       return -1;
     }
   }
 
-  if (!updateHeader) {
-    return 0;
-  }
-
-  pTFile->info.fsize += TD_FILE_HEAD_SIZE;
-  pTFile->info.fver = 0;
-
-  if (tdUpdateTFileHeader(pTFile) < 0) {
-    tdCloseTFile(pTFile);
-    tdRemoveTFile(pTFile);
-    return -1;
-  }
-
   return 0;
 }
 
-int32_t tdRemoveTFile(STFile *pTFile) { return tfsRemoveFile(TD_TFILE_F(pTFile)); }
+int32_t tdRemoveTFile(STFile *pTFile) {
+  if (taosRemoveFile(TD_TFILE_FULL_NAME(pTFile)) != 0) {
+    terrno = TAOS_SYSTEM_ERROR(errno);
+    return -1;
+  };
+  return 0;
+}
 
 // smaXXXUtil ================
 // ...

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

@@ -154,10 +154,10 @@ int32_t tqSendDataRsp(STQ* pTq, const SRpcMsg* pMsg, const SMqPollReq* pReq, con
   };
   tmsgSendRsp(&rsp);
 
-  char buf1[50];
+  char buf1[80];
-  char buf2[50];
+  char buf2[80];
-  tFormatOffset(buf1, 50, &pRsp->reqOffset);
+  tFormatOffset(buf1, 80, &pRsp->reqOffset);
-  tFormatOffset(buf2, 50, &pRsp->rspOffset);
+  tFormatOffset(buf2, 80, &pRsp->rspOffset);
   tqDebug("vg %d from consumer %ld (epoch %d) send rsp, block num: %d, reqOffset: %s, rspOffset: %s",
           TD_VID(pTq->pVnode), pReq->consumerId, pReq->epoch, pRsp->blockNum, buf1, buf2);
 
@@ -183,13 +183,15 @@ int32_t tqProcessOffsetCommitReq(STQ* pTq, char* msg, int32_t msgLen) {
   } else {
     ASSERT(0);
   }
-  STqOffset* pOffset = tqOffsetRead(pTq->pOffsetStore, offset.subKey);
+  /*STqOffset* pOffset = tqOffsetRead(pTq->pOffsetStore, offset.subKey);*/
-  if (pOffset == NULL || pOffset->val.version < offset.val.version) {
+  /*if (pOffset != NULL) {*/
-    if (tqOffsetWrite(pTq->pOffsetStore, &offset) < 0) {
+  /*if (pOffset->val.type == TMQ_OFFSET__LOG && pOffset->val.version < offset.val.version) {*/
-      ASSERT(0);
+  if (tqOffsetWrite(pTq->pOffsetStore, &offset) < 0) {
-      return -1;
+    ASSERT(0);
-    }
+    return -1;
   }
+  /*}*/
+  /*}*/
 
   return 0;
 }
@@ -228,17 +230,6 @@ static int32_t tqInitDataRsp(SMqDataRsp* pRsp, const SMqPollReq* pReq, int8_t su
 
 static int32_t tqInitMetaRsp(SMqMetaRsp* pRsp, const SMqPollReq* pReq) { return 0; }
 
-static FORCE_INLINE void tqOffsetResetToData(STqOffsetVal* pOffsetVal, int64_t uid, int64_t ts) {
-  pOffsetVal->type = TMQ_OFFSET__SNAPSHOT_DATA;
-  pOffsetVal->uid = uid;
-  pOffsetVal->ts = ts;
-}
-
-static FORCE_INLINE void tqOffsetResetToLog(STqOffsetVal* pOffsetVal, int64_t ver) {
-  pOffsetVal->type = TMQ_OFFSET__LOG;
-  pOffsetVal->version = ver;
-}
-
 int32_t tqProcessPollReq(STQ* pTq, SRpcMsg* pMsg, int32_t workerId) {
   SMqPollReq*  pReq = pMsg->pCont;
   int64_t      consumerId = pReq->consumerId;
@@ -249,8 +240,8 @@ int32_t tqProcessPollReq(STQ* pTq, SRpcMsg* pMsg, int32_t workerId) {
   STqOffsetVal fetchOffsetNew;
 
   // 1.find handle
-  char buf[50];
+  char buf[80];
-  tFormatOffset(buf, 50, &reqOffset);
+  tFormatOffset(buf, 80, &reqOffset);
   tqDebug("tmq poll: consumer %ld (epoch %d) recv poll req in vg %d, req offset %s", consumerId, pReq->epoch,
           TD_VID(pTq->pVnode), buf);
 
@@ -282,12 +273,12 @@ int32_t tqProcessPollReq(STQ* pTq, SRpcMsg* pMsg, int32_t workerId) {
     STqOffset* pOffset = tqOffsetRead(pTq->pOffsetStore, pReq->subKey);
     if (pOffset != NULL) {
       fetchOffsetNew = pOffset->val;
-      char formatBuf[50];
+      char formatBuf[80];
-      tFormatOffset(formatBuf, 50, &fetchOffsetNew);
+      tFormatOffset(formatBuf, 80, &fetchOffsetNew);
       tqDebug("tmq poll: consumer %ld, offset reset to %s", consumerId, formatBuf);
     } else {
       if (reqOffset.type == TMQ_OFFSET__RESET_EARLIEAST) {
-        if (pReq->useSnapshot) {
+        if (pReq->useSnapshot && pHandle->execHandle.subType == TOPIC_SUB_TYPE__COLUMN) {
           if (!pHandle->fetchMeta) {
             tqOffsetResetToData(&fetchOffsetNew, 0, 0);
           } else {
@@ -313,9 +304,9 @@ int32_t tqProcessPollReq(STQ* pTq, SRpcMsg* pMsg, int32_t workerId) {
   tqInitDataRsp(&dataRsp, pReq, pHandle->execHandle.subType);
 
   if (fetchOffsetNew.type == TMQ_OFFSET__LOG) {
-    int64_t   fetchVer = fetchOffsetNew.version + 1;
+    int64_t     fetchVer = fetchOffsetNew.version + 1;
-    SWalHead* pHeadWithCkSum = taosMemoryMalloc(sizeof(SWalHead) + 2048);
+    SWalCkHead* pCkHead = taosMemoryMalloc(sizeof(SWalCkHead) + 2048);
-    if (pHeadWithCkSum == NULL) {
+    if (pCkHead == NULL) {
       return -1;
     }
 
@@ -329,7 +320,7 @@ int32_t tqProcessPollReq(STQ* pTq, SRpcMsg* pMsg, int32_t workerId) {
         break;
       }
 
-      if (tqFetchLog(pTq, pHandle, &fetchVer, &pHeadWithCkSum) < 0) {
+      if (tqFetchLog(pTq, pHandle, &fetchVer, &pCkHead) < 0) {
         // TODO add push mgr
 
         tqOffsetResetToLog(&dataRsp.rspOffset, fetchVer);
@@ -340,7 +331,7 @@ int32_t tqProcessPollReq(STQ* pTq, SRpcMsg* pMsg, int32_t workerId) {
         goto OVER;
       }
 
-      SWalReadHead* pHead = &pHeadWithCkSum->head;
+      SWalCont* pHead = &pCkHead->head;
 
       tqDebug("tmq poll: consumer %ld (epoch %d) iter log, vg %d offset %ld msgType %d", consumerId, pReq->epoch,
               TD_VID(pTq->pVnode), fetchVer, pHead->msgType);
@@ -371,7 +362,6 @@ int32_t tqProcessPollReq(STQ* pTq, SRpcMsg* pMsg, int32_t workerId) {
         tqInfo("fetch meta msg, ver: %ld, type: %d", pHead->version, pHead->msgType);
         SMqMetaRsp metaRsp = {0};
         metaRsp.reqOffset = pReq->reqOffset.version;
-        /*tqOffsetResetToLog(&metaR)*/
         metaRsp.rspOffset = fetchVer;
         metaRsp.resMsgType = pHead->msgType;
         metaRsp.metaRspLen = pHead->bodyLen;
@@ -385,25 +375,16 @@ int32_t tqProcessPollReq(STQ* pTq, SRpcMsg* pMsg, int32_t workerId) {
       }
     }
 
-    taosMemoryFree(pHeadWithCkSum);
+    taosMemoryFree(pCkHead);
   } else if (fetchOffsetNew.type == TMQ_OFFSET__SNAPSHOT_DATA) {
-    // 1. set uid and ts
+    tqInfo("retrieve using snapshot actual offset: uid %ld ts %ld", fetchOffsetNew.uid, fetchOffsetNew.ts);
-    // 2. get data (rebuild reader if needed)
+    if (tqScanSnapshot(pTq, &pHandle->execHandle, &dataRsp, fetchOffsetNew, workerId) < 0) {
-    // 3. get new uid and ts
-
-    char formatBuf[50];
-    tFormatOffset(formatBuf, 50, &dataRsp.reqOffset);
-    tqInfo("retrieve using snapshot req offset %s", formatBuf);
-    if (tqScanSnapshot(pTq, &pHandle->execHandle, &dataRsp, workerId) < 0) {
       ASSERT(0);
     }
 
     // 4. send rsp
-    if (dataRsp.blockNum != 0) {
+    if (tqSendDataRsp(pTq, pMsg, pReq, &dataRsp) < 0) {
-      tqOffsetResetToData(&dataRsp.rspOffset, 0, 0);
+      code = -1;
-      if (tqSendDataRsp(pTq, pMsg, pReq, &dataRsp) < 0) {
-        code = -1;
-      }
     }
   } else if (fetchOffsetNew.type == TMQ_OFFSET__SNAPSHOT_META) {
     ASSERT(0);
@@ -543,7 +524,7 @@ int32_t tqProcessPollReq(STQ* pTq, SRpcMsg* pMsg, int32_t workerId) {
       break;
     }
 
-    SWalReadHead* pHead = &pHeadWithCkSum->head;
+    SWalCont* pHead = &pHeadWithCkSum->head;
 
     tqDebug("tmq poll: consumer %ld (epoch %d) iter log, vg %d offset %ld msgType %d", consumerId, pReq->epoch,
             TD_VID(pTq->pVnode), fetchOffset, pHead->msgType);
@@ -618,6 +599,8 @@ int32_t tqProcessVgDeleteReq(STQ* pTq, char* msg, int32_t msgLen) {
   int32_t code = taosHashRemove(pTq->handles, pReq->subKey, strlen(pReq->subKey));
   ASSERT(code == 0);
 
+  tqOffsetDelete(pTq->pOffsetStore, pReq->subKey);
+
   if (tqMetaDeleteHandle(pTq, pReq->subKey) < 0) {
     ASSERT(0);
   }
@@ -655,6 +638,7 @@ int32_t tqProcessVgChangeReq(STQ* pTq, char* msg, int32_t msgLen) {
             .reader = pHandle->execHandle.pExecReader[i],
             .meta = pTq->pVnode->pMeta,
             .vnode = pTq->pVnode,
+            .tqReader = true,
         };
         pHandle->execHandle.execCol.task[i] = qCreateStreamExecTaskInfo(pHandle->execHandle.execCol.qmsg, &handle);
         ASSERT(pHandle->execHandle.execCol.task[i]);

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

@@ -59,13 +59,19 @@ static int32_t tqAddTbNameToRsp(const STQ* pTq, int64_t uid, SMqDataRsp* pRsp, i
   return 0;
 }
 
-int32_t tqScanSnapshot(STQ* pTq, const STqExecHandle* pExec, SMqDataRsp* pRsp, int32_t workerId) {
+int32_t tqScanSnapshot(STQ* pTq, const STqExecHandle* pExec, SMqDataRsp* pRsp, STqOffsetVal offset, int32_t workerId) {
   ASSERT(pExec->subType == TOPIC_SUB_TYPE__COLUMN);
   qTaskInfo_t task = pExec->execCol.task[workerId];
-  // TODO set uid and ts
+
-  if (qStreamScanSnapshot(task) < 0) {
+  /*if (qStreamScanSnapshot(task) < 0) {*/
+  /*ASSERT(0);*/
+  /*}*/
+
+  if (qStreamPrepareScan(task, offset.uid, offset.ts) < 0) {
     ASSERT(0);
   }
+
+  int32_t rowCnt = 0;
   while (1) {
     SSDataBlock* pDataBlock = NULL;
     uint64_t     ts = 0;
@@ -80,13 +86,27 @@ int32_t tqScanSnapshot(STQ* pTq, const STqExecHandle* pExec, SMqDataRsp* pRsp, i
     tqAddBlockDataToRsp(pDataBlock, pRsp);
 
     if (pRsp->withTbName) {
-      // TODO
       pRsp->withTbName = 0;
-      /*int64_t uid = 0;*/
+#if 0
-      /*tqAddTbNameToRsp(pTq, uid, pRsp, workerId);*/
+      int64_t uid;
+      int64_t ts;
+      if (qGetStreamScanStatus(task, &uid, &ts) < 0) {
+        ASSERT(0);
+      }
+      tqAddTbNameToRsp(pTq, uid, pRsp, workerId);
+#endif
     }
     pRsp->blockNum++;
+
+    rowCnt += pDataBlock->info.rows;
+    if (rowCnt >= 4096) break;
   }
+  int64_t uid;
+  int64_t ts;
+  if (qGetStreamScanStatus(task, &uid, &ts) < 0) {
+    ASSERT(0);
+  }
+  tqOffsetResetToData(&pRsp->rspOffset, uid, ts);
 
   return 0;
 }

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

@@ -97,6 +97,10 @@ int32_t tqOffsetWrite(STqOffsetStore* pStore, const STqOffset* pOffset) {
   return taosHashPut(pStore->pHash, pOffset->subKey, strlen(pOffset->subKey), pOffset, sizeof(STqOffset));
 }
 
+int32_t tqOffsetDelete(STqOffsetStore* pStore, const char* subscribeKey) {
+  return taosHashRemove(pStore->pHash, subscribeKey, strlen(subscribeKey));
+}
+
 int32_t tqOffsetSnapshot(STqOffsetStore* pStore) {
   // open file
   // TODO file name should be with a version

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

@@ -15,13 +15,13 @@
 
 #include "tq.h"
 
-int64_t tqFetchLog(STQ* pTq, STqHandle* pHandle, int64_t* fetchOffset, SWalHead** ppHeadWithCkSum) {
+int64_t tqFetchLog(STQ* pTq, STqHandle* pHandle, int64_t* fetchOffset, SWalCkHead** ppCkHead) {
   int32_t code = 0;
   taosThreadMutexLock(&pHandle->pWalReader->mutex);
   int64_t offset = *fetchOffset;
 
   while (1) {
-    if (walFetchHead(pHandle->pWalReader, offset, *ppHeadWithCkSum) < 0) {
+    if (walFetchHead(pHandle->pWalReader, offset, *ppCkHead) < 0) {
       tqDebug("tmq poll: consumer %ld (epoch %d) vg %d offset %ld, no more log to return", pHandle->consumerId,
               pHandle->epoch, TD_VID(pTq->pVnode), offset);
       *fetchOffset = offset - 1;
@@ -29,8 +29,8 @@ int64_t tqFetchLog(STQ* pTq, STqHandle* pHandle, int64_t* fetchOffset, SWalHead*
       goto END;
     }
 
-    if ((*ppHeadWithCkSum)->head.msgType == TDMT_VND_SUBMIT) {
+    if ((*ppCkHead)->head.msgType == TDMT_VND_SUBMIT) {
-      code = walFetchBody(pHandle->pWalReader, ppHeadWithCkSum);
+      code = walFetchBody(pHandle->pWalReader, ppCkHead);
 
       if (code < 0) {
         ASSERT(0);
@@ -43,9 +43,9 @@ int64_t tqFetchLog(STQ* pTq, STqHandle* pHandle, int64_t* fetchOffset, SWalHead*
       goto END;
     } else {
       if (pHandle->fetchMeta) {
-        SWalReadHead* pHead = &((*ppHeadWithCkSum)->head);
+        SWalCont* pHead = &((*ppCkHead)->head);
         if (IS_META_MSG(pHead->msgType)) {
-          code = walFetchBody(pHandle->pWalReader, ppHeadWithCkSum);
+          code = walFetchBody(pHandle->pWalReader, ppCkHead);
 
           if (code < 0) {
             ASSERT(0);
@@ -58,7 +58,7 @@ int64_t tqFetchLog(STQ* pTq, STqHandle* pHandle, int64_t* fetchOffset, SWalHead*
           goto END;
         }
       }
-      code = walSkipFetchBody(pHandle->pWalReader, *ppHeadWithCkSum);
+      code = walSkipFetchBody(pHandle->pWalReader, *ppCkHead);
       if (code < 0) {
         ASSERT(0);
         *fetchOffset = offset;
@@ -120,7 +120,9 @@ bool tqNextDataBlock(SStreamReader* pHandle) {
       return true;
     }
     void* ret = taosHashGet(pHandle->tbIdHash, &pHandle->msgIter.uid, sizeof(int64_t));
+    /*tqDebug("search uid %ld", pHandle->msgIter.uid);*/
     if (ret != NULL) {
+      /*tqDebug("find   uid %ld", pHandle->msgIter.uid);*/
       return true;
     }
   }

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

@@ -112,9 +112,9 @@ typedef struct STsdbReadHandle {
   STimeWindow   window;  // the primary query time window that applies to all queries
   //  SColumnDataAgg* statis;  // query level statistics, only one table block statistics info exists at any time
   //  SColumnDataAgg** pstatis;// the ptr array list to return to caller
-  int32_t numOfBlocks;
+  int32_t      numOfBlocks;
   SSDataBlock* pResBlock;
-//  SArray* pColumns;  // column list, SColumnInfoData array list
+  //  SArray* pColumns;  // column list, SColumnInfoData array list
   bool    locateStart;
   int32_t outputCapacity;
   int32_t realNumOfRows;
@@ -223,6 +223,22 @@ int64_t tsdbGetNumOfRowsInMemTable(tsdbReaderT* pHandle) {
   return rows;
 }
 
+static SArray* createCheckInfoFromUid(STsdbReadHandle* pTsdbReadHandle, int64_t uid) {
+  SArray* pTableCheckInfo = taosArrayInit(1, sizeof(STableCheckInfo));
+  if (pTableCheckInfo == NULL) {
+    return NULL;
+  }
+  STableCheckInfo info = {
+      .tableId = uid,
+  };
+  info.suid = pTsdbReadHandle->suid;
+
+  taosArrayPush(pTableCheckInfo, &info);
+  tsdbDebug("%p check table uid:%" PRId64 " from lastKey:%" PRId64 " %s", pTsdbReadHandle, info.tableId, info.lastKey,
+            pTsdbReadHandle->idStr);
+  return pTableCheckInfo;
+}
+
 static SArray* createCheckInfoFromTableGroup(STsdbReadHandle* pTsdbReadHandle, SArray* pTableList) {
   size_t tableSize = taosArrayGetSize(pTableList);
 
@@ -428,8 +444,8 @@ static STsdbReadHandle* tsdbQueryTablesImpl(SVnode* pVnode, SQueryTableDataCond*
 
     for (int32_t i = 0; i < pCond->numOfCols; ++i) {
       SColumnInfoData colInfo = {.info = pCond->colList[i], 0};
-      int32_t code = blockDataAppendColInfo(pReadHandle->pResBlock, &colInfo);
+      int32_t         code = blockDataAppendColInfo(pReadHandle->pResBlock, &colInfo);
-      if (code != TSDB_CODE_SUCCESS){
+      if (code != TSDB_CODE_SUCCESS) {
         goto _end;
       }
     }
@@ -494,9 +510,19 @@ static int32_t setCurrentSchema(SVnode* pVnode, STsdbReadHandle* pTsdbReadHandle
   return TSDB_CODE_SUCCESS;
 }
 
-int32_t tsdbSetTableList(tsdbReaderT reader, SArray* tableList){
+int32_t tsdbSetTableId(tsdbReaderT reader, int64_t uid) {
   STsdbReadHandle* pTsdbReadHandle = reader;
-  if(pTsdbReadHandle->pTableCheckInfo) taosArrayDestroy(pTsdbReadHandle->pTableCheckInfo);
+  if (pTsdbReadHandle->pTableCheckInfo) taosArrayDestroy(pTsdbReadHandle->pTableCheckInfo);
+  pTsdbReadHandle->pTableCheckInfo = createCheckInfoFromUid(pTsdbReadHandle, uid);
+  if (pTsdbReadHandle->pTableCheckInfo == NULL) {
+    return TSDB_CODE_TDB_OUT_OF_MEMORY;
+  }
+  return TDB_CODE_SUCCESS;
+}
+
+int32_t tsdbSetTableList(tsdbReaderT reader, SArray* tableList) {
+  STsdbReadHandle* pTsdbReadHandle = reader;
+  if (pTsdbReadHandle->pTableCheckInfo) taosArrayDestroy(pTsdbReadHandle->pTableCheckInfo);
   pTsdbReadHandle->pTableCheckInfo = createCheckInfoFromTableGroup(pTsdbReadHandle, tableList);
   if (pTsdbReadHandle->pTableCheckInfo == NULL) {
     return TSDB_CODE_TDB_OUT_OF_MEMORY;
@@ -505,8 +531,8 @@ int32_t tsdbSetTableList(tsdbReaderT reader, SArray* tableList){
 }
 
 tsdbReaderT tsdbReaderOpen(SVnode* pVnode, SQueryTableDataCond* pCond, SArray* tableList, uint64_t qId,
-                            uint64_t taskId) {
+                           uint64_t taskId) {
-  if(taosArrayGetSize(tableList) == 0){
+  if (taosArrayGetSize(tableList) == 0) {
     return NULL;
   }
   STsdbReadHandle* pTsdbReadHandle = tsdbQueryTablesImpl(pVnode, pCond, qId, taskId);
@@ -553,8 +579,7 @@ tsdbReaderT tsdbReaderOpen(SVnode* pVnode, SQueryTableDataCond* pCond, SArray* t
   }
 
   tsdbDebug("%p total numOfTable:%" PRIzu " in this query, table %" PRIzu " %s", pTsdbReadHandle,
-            taosArrayGetSize(pTsdbReadHandle->pTableCheckInfo), taosArrayGetSize(tableList),
+            taosArrayGetSize(pTsdbReadHandle->pTableCheckInfo), taosArrayGetSize(tableList), pTsdbReadHandle->idStr);
-            pTsdbReadHandle->idStr);
 
   return (tsdbReaderT)pTsdbReadHandle;
 }
@@ -1073,7 +1098,7 @@ static int32_t getFileIdFromKey(TSKEY key, int32_t daysPerFile, int32_t precisio
   }
 
   int64_t fid = (int64_t)(key / (daysPerFile * tsTickPerMin[precision]));  // set the starting fileId
-  if (fid < 0LL && llabs(fid) > INT32_MAX) {                                // data value overflow for INT32
+  if (fid < 0LL && llabs(fid) > INT32_MAX) {                               // data value overflow for INT32
     fid = INT32_MIN;
   }
 
@@ -2612,7 +2637,7 @@ int32_t tsdbGetFileBlocksDistInfo(tsdbReaderT* queryHandle, STableBlockDistInfo*
     tsdbGetFidKeyRange(pCfg->days, pCfg->precision, pTsdbReadHandle->pFileGroup->fid, &win.skey, &win.ekey);
 
     // current file are not overlapped with query time window, ignore remain files
-    if ((win.skey > pTsdbReadHandle->window.ekey)/* || (!ascTraverse && win.ekey < pTsdbReadHandle->window.ekey)*/) {
+    if ((win.skey > pTsdbReadHandle->window.ekey) /* || (!ascTraverse && win.ekey < pTsdbReadHandle->window.ekey)*/) {
       tsdbUnLockFS(REPO_FS(pTsdbReadHandle->pTsdb));
       tsdbDebug("%p remain files are not qualified for qrange:%" PRId64 "-%" PRId64 ", ignore, %s", pTsdbReadHandle,
                 pTsdbReadHandle->window.skey, pTsdbReadHandle->window.ekey, pTsdbReadHandle->idStr);
@@ -2886,7 +2911,7 @@ int32_t tsdbGetCtbIdList(SMeta* pMeta, int64_t suid, SArray* list) {
  */
 int32_t tsdbGetStbIdList(SMeta* pMeta, int64_t suid, SArray* list) {
   SMStbCursor* pCur = metaOpenStbCursor(pMeta, suid);
-  if(!pCur) {
+  if (!pCur) {
     return TSDB_CODE_FAILED;
   }
 

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

@@ -28,12 +28,12 @@ const SVnodeCfg vnodeCfgDefault = {
                 .update = 1,
                 .compression = 2,
                 .slLevel = 5,
-                .days = 10,
+                .days = 14400,
                 .minRows = 100,
                 .maxRows = 4096,
-                .keep2 = 3650,
+                .keep2 = 5256000,
-                .keep0 = 3650,
+                .keep0 = 5256000,
-                .keep1 = 3650},
+                .keep1 = 5256000},
     .walCfg =
         {.vgId = -1, .fsyncPeriod = 0, .retentionPeriod = 0, .rollPeriod = 0, .segSize = 0, .level = TAOS_WAL_WRITE},
     .hashBegin = 0,

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

@@ -229,6 +229,9 @@ int vnodeCommit(SVnode *pVnode) {
     return -1;
   }
 
+  // preCommit
+  // TODO
+  
   // commit each sub-system
   if (metaCommit(pVnode->pMeta) < 0) {
     ASSERT(0);
@@ -266,6 +269,11 @@ int vnodeCommit(SVnode *pVnode) {
     ASSERT(0);
     return -1;
   }
+  
+  pVnode->state.committed = info.state.committed;
+
+  // postCommit
+  smaPostCommit(pVnode->pSma);
 
   // apply the commit (TODO)
   vnodeBufPoolReset(pVnode->onCommit);

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

@@ -216,7 +216,7 @@ _err:
 }
 
 int32_t vnodePreprocessQueryMsg(SVnode *pVnode, SRpcMsg *pMsg) {
-  if (TDMT_VND_QUERY != pMsg->msgType) {
+  if (TDMT_SCH_QUERY != pMsg->msgType && TDMT_SCH_MERGE_QUERY != pMsg->msgType) {
     return 0;
   }
 
@@ -227,9 +227,10 @@ int32_t vnodeProcessQueryMsg(SVnode *pVnode, SRpcMsg *pMsg) {
   vTrace("message in vnode query queue is processing");
   SReadHandle handle = {.meta = pVnode->pMeta, .config = &pVnode->config, .vnode = pVnode, .pMsgCb = &pVnode->msgCb};
   switch (pMsg->msgType) {
-    case TDMT_VND_QUERY:
+    case TDMT_SCH_QUERY:
+    case TDMT_SCH_MERGE_QUERY:
       return qWorkerProcessQueryMsg(&handle, pVnode->pQuery, pMsg, 0);
-    case TDMT_VND_QUERY_CONTINUE:
+    case TDMT_SCH_QUERY_CONTINUE:
       return qWorkerProcessCQueryMsg(&handle, pVnode->pQuery, pMsg, 0);
     default:
       vError("unknown msg type:%d in query queue", pMsg->msgType);
@@ -243,15 +244,15 @@ int32_t vnodeProcessFetchMsg(SVnode *pVnode, SRpcMsg *pMsg, SQueueInfo *pInfo) {
   int32_t msgLen = pMsg->contLen - sizeof(SMsgHead);
 
   switch (pMsg->msgType) {
-    case TDMT_VND_FETCH:
+    case TDMT_SCH_FETCH:
       return qWorkerProcessFetchMsg(pVnode, pVnode->pQuery, pMsg, 0);
-    case TDMT_VND_FETCH_RSP:
+    case TDMT_SCH_FETCH_RSP:
       return qWorkerProcessFetchRsp(pVnode, pVnode->pQuery, pMsg, 0);
-    case TDMT_VND_CANCEL_TASK:
+    case TDMT_SCH_CANCEL_TASK:
       return qWorkerProcessCancelMsg(pVnode, pVnode->pQuery, pMsg, 0);
-    case TDMT_VND_DROP_TASK:
+    case TDMT_SCH_DROP_TASK:
       return qWorkerProcessDropMsg(pVnode, pVnode->pQuery, pMsg, 0);
-    case TDMT_VND_QUERY_HEARTBEAT:
+    case TDMT_SCH_QUERY_HEARTBEAT:
       return qWorkerProcessHbMsg(pVnode, pVnode->pQuery, pMsg, 0);
     case TDMT_VND_TABLE_META:
       return vnodeGetTableMeta(pVnode, pMsg);
@@ -778,6 +779,7 @@ _exit:
   taosArrayDestroy(submitRsp.pArray);
 
   // TODO: the partial success scenario and the error case
+  // => If partial success, extract the success submitted rows and reconstruct a new submit msg, and push to level 1/level 2.
   // TODO: refactor
   if ((terrno == TSDB_CODE_SUCCESS) && (pRsp->code == TSDB_CODE_SUCCESS)) {
     tdProcessRSmaSubmit(pVnode->pSma, pReq, STREAM_INPUT__DATA_SUBMIT);

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

@@ -178,7 +178,7 @@ void vnodeProposeMsg(SQueueInfo *pInfo, STaosQall *qall, int32_t numOfMsgs) {
       for (int32_t i = 0; i < newEpSet.numOfEps; ++i) {
         vGTrace("vgId:%d, msg:%p redirect:%d ep:%s:%u", vgId, pMsg, i, newEpSet.eps[i].fqdn, newEpSet.eps[i].port);
       }
-
+      pMsg->info.hasEpSet = 1;
       SRpcMsg rsp = {.code = TSDB_CODE_RPC_REDIRECT, .info = pMsg->info};
       tmsgSendRedirectRsp(&rsp, &newEpSet);
     } else {
@@ -409,7 +409,7 @@ static void vnodeSyncRollBackMsg(SSyncFSM *pFsm, const SRpcMsg *pMsg, SFsmCbMeta
   syncRpcMsgLog2(logBuf, (SRpcMsg *)pMsg);
 }
 
-static int32_t vnodeSnapshotStartRead(struct SSyncFSM *pFsm, void **ppReader) { return 0; }
+static int32_t vnodeSnapshotStartRead(struct SSyncFSM *pFsm, void *pParam, void **ppReader) { return 0; }
 
 static int32_t vnodeSnapshotStopRead(struct SSyncFSM *pFsm, void *pReader) { return 0; }
 

source/libs/catalog/src/ctgRemote.c

@@ -241,7 +241,7 @@ int32_t ctgProcessRspMsg(void* out, int32_t reqType, char* msg, int32_t msgSize,
 }
 
 
-int32_t ctgHandleMsgCallback(void *param, const SDataBuf *pMsg, int32_t rspCode) {
+int32_t ctgHandleMsgCallback(void *param, SDataBuf *pMsg, int32_t rspCode) {
   SCtgTaskCallbackParam* cbParam = (SCtgTaskCallbackParam*)param;
   int32_t code = 0;
   

source/libs/command/src/command.c

@@ -297,7 +297,7 @@ int32_t appendTagValues(char* buf, int32_t* len, STableCfg* pCfg) {
   SArray* pTagVals = NULL;
   STag*   pTag = (STag*)pCfg->pTags;
 
-  if (pCfg->pTags && pTag->flags & TD_TAG_JSON) {
+  if (pCfg->pTags && tTagIsJson(pTag)) {
     char* pJson = parseTagDatatoJson(pTag);
     if (pJson) {
       *len += sprintf(buf + VARSTR_HEADER_SIZE + *len, "%s", pJson);

source/libs/executor/inc/executorimpl.h

@@ -248,6 +248,11 @@ typedef struct SSampleExecInfo {
   uint32_t        seed;         // random seed value
 } SSampleExecInfo;
 
+enum {
+  TABLE_SCAN__TABLE_ORDER = 1,
+  TABLE_SCAN__BLOCK_ORDER = 2,
+};
+
 typedef struct STableScanInfo {
   void*           dataReader;
   SReadHandle     readHandle;
@@ -272,13 +277,17 @@ typedef struct STableScanInfo {
   int32_t         curTWinIdx;
 
   int32_t         currentGroupId;
+  int32_t         currentTable;
   uint64_t        queryId;   // todo remove it
   uint64_t        taskId;    // todo remove it
 
   struct {
     uint64_t uid;
-    int64_t t;
+    int64_t ts;
-  } scanStatus;
+  } lastStatus;
+
+  int8_t scanMode;
+  int8_t noTable;
 } STableScanInfo;
 
 typedef struct STagScanInfo {
@@ -347,6 +356,7 @@ typedef struct SStreamBlockScanInfo {
   SUpdateInfo*    pUpdateInfo;
 
   EStreamScanMode scanMode;
+  SOperatorInfo* pStreamScanOp;
   SOperatorInfo* pSnapshotReadOp;
   SArray*        childIds;
   SessionWindowSupporter sessionSup;
@@ -713,6 +723,7 @@ void    destroyBasicOperatorInfo(void* param, int32_t numOfOutput);
 void    appendOneRowToDataBlock(SSDataBlock* pBlock, STupleHandle* pTupleHandle);
 void    setTbNameColData(void* pMeta, const SSDataBlock* pBlock, SColumnInfoData* pColInfoData, int32_t functionId);
 
+int32_t doPrepareScan(SOperatorInfo* pOperator, uint64_t uid, int64_t ts);
 int32_t doGetScanStatus(SOperatorInfo* pOperator, uint64_t* uid, int64_t* ts);
 
 SSDataBlock* loadNextDataBlock(void* param);

source/libs/executor/src/executil.c

@@ -63,6 +63,7 @@ size_t getResultRowSize(SqlFunctionCtx* pCtx, int32_t numOfOutput) {
     rowSize += pCtx[i].resDataInfo.interBufSize;
   }
 
+  rowSize += (numOfOutput * sizeof(bool));    // expand rowSize to mark if col is null for top/bottom result(saveTupleData)
   return rowSize;
 }
 
@@ -112,13 +113,15 @@ void initGroupedResultInfo(SGroupResInfo* pGroupResInfo, SHashObj* pHashmap, int
     p->groupId = *(uint64_t*)key;
     p->pos = *(SResultRowPosition*)pData;
     memcpy(p->key, (char*)key + sizeof(uint64_t), keyLen - sizeof(uint64_t));
-
+#ifdef BUF_PAGE_DEBUG
+    qDebug("page_groupRes, groupId:%"PRIu64",pageId:%d,offset:%d\n", p->groupId, p->pos.pageId, p->pos.offset);
+#endif
     taosArrayPush(pGroupResInfo->pRows, &p);
   }
 
   if (order == TSDB_ORDER_ASC || order == TSDB_ORDER_DESC) {
     __compar_fn_t fn = (order == TSDB_ORDER_ASC) ? resultrowComparAsc : resultrowComparDesc;
-    qsort(pGroupResInfo->pRows->pData, taosArrayGetSize(pGroupResInfo->pRows), POINTER_BYTES, fn);
+    taosSort(pGroupResInfo->pRows->pData, taosArrayGetSize(pGroupResInfo->pRows), POINTER_BYTES, fn);
   }
 
   pGroupResInfo->index = 0;
@@ -271,6 +274,7 @@ static bool isTableOk(STableKeyInfo* info, SNode* pTagCond, SMeta* metaHandle) {
   SNode*  pNew = NULL;
   int32_t code = scalarCalculateConstants(pTagCondTmp, &pNew);
   if (TSDB_CODE_SUCCESS != code) {
+    terrno = code;
     nodesDestroyNode(pTagCondTmp);
     return false;
   }
@@ -323,12 +327,19 @@ int32_t getTableList(void* metaHandle, SScanPhysiNode* pScanNode, STableListInfo
       code = tsdbGetAllTableList(metaHandle, tableUid, pListInfo->pTableList);
     }
 
-    if (pTagCond) {
+    if (code != TSDB_CODE_SUCCESS) {
+      qError("failed  to  get tableIds, reason: %s, suid: %" PRIu64 "", tstrerror(code), tableUid);
+      terrno = code;
+      return code;
+    }
+
+    if(pTagCond){
       int32_t i = 0;
       while (i < taosArrayGetSize(pListInfo->pTableList)) {
         STableKeyInfo* info = taosArrayGet(pListInfo->pTableList, i);
-        bool           isOk = isTableOk(info, pTagCond, metaHandle);
+        bool isOk = isTableOk(info, pTagCond, metaHandle);
-        if (!isOk) {
+        if(terrno) return terrno;
+        if(!isOk){
           taosArrayRemove(pListInfo->pTableList, i);
           continue;
         }
@@ -586,13 +597,16 @@ static int32_t setSelectValueColumnInfo(SqlFunctionCtx* pCtx, int32_t numOfOutpu
   }
 
   for (int32_t i = 0; i < numOfOutput; ++i) {
-    if (strcmp(pCtx[i].pExpr->pExpr->_function.functionName, "_select_value") == 0) {
+    if (strcmp(pCtx[i].pExpr->pExpr->_function.functionName, "_select_value") == 0 ||
+        strcmp(pCtx[i].pExpr->pExpr->_function.functionName, "_group_key") == 0) {
       pValCtx[num++] = &pCtx[i];
     } else if (fmIsSelectFunc(pCtx[i].functionId)) {
       p = &pCtx[i];
     }
   }
-
+#ifdef BUF_PAGE_DEBUG
+  qDebug("page_setSelect num:%d", num);
+#endif
   if (p != NULL) {
     p->subsidiaries.pCtx = pValCtx;
     p->subsidiaries.num = num;

source/libs/executor/src/executor.c

@@ -145,10 +145,12 @@ static SArray* filterQualifiedChildTables(const SStreamBlockScanInfo* pScanInfo,
       continue;
     }
 
+    // TODO handle ntb case
     if (mr.me.type != TSDB_CHILD_TABLE || mr.me.ctbEntry.suid != pScanInfo->tableUid) {
       continue;
     }
-    // TODO handle ntb case
+    /*pScanInfo->pStreamScanOp->pTaskInfo->tableqinfoList.*/
+    // handle multiple partition
 
     taosArrayPush(qa, id);
   }

source/libs/executor/src/executorMain.c

@@ -222,7 +222,7 @@ int32_t qSerializeTaskStatus(qTaskInfo_t tinfo, char** pOutput, int32_t* len) {
 }
 
 int32_t qDeserializeTaskStatus(qTaskInfo_t tinfo, const char* pInput, int32_t len) {
-  SExecTaskInfo* pTaskInfo = (struct SExecTaskInfo*) tinfo;
+  SExecTaskInfo* pTaskInfo = (struct SExecTaskInfo*)tinfo;
 
   if (pTaskInfo == NULL || pInput == NULL || len == 0) {
     return TSDB_CODE_INVALID_PARA;
@@ -231,11 +231,22 @@ int32_t qDeserializeTaskStatus(qTaskInfo_t tinfo, const char* pInput, int32_t le
   return decodeOperator(pTaskInfo->pRoot, pInput, len);
 }
 
+int32_t qStreamPrepareScan(qTaskInfo_t tinfo, uint64_t uid, int64_t ts) {
+  SExecTaskInfo* pTaskInfo = (SExecTaskInfo*)tinfo;
 
-int32_t qGetStreamScanStatus(qTaskInfo_t tinfo, uint64_t* uid, int64_t* ts) {
+  if (uid == 0) {
-  SExecTaskInfo* pTaskInfo = (SExecTaskInfo*) tinfo;
+    if (taosArrayGetSize(pTaskInfo->tableqinfoList.pTableList) != 0) {
+      STableKeyInfo* pTableInfo = taosArrayGet(pTaskInfo->tableqinfoList.pTableList, 0);
+      uid = pTableInfo->uid;
+      ts = INT64_MIN;
+    }
+  }
 
-  return TSDB_CODE_SUCCESS;
+  return doPrepareScan(pTaskInfo->pRoot, uid, ts);
 }
 
+int32_t qGetStreamScanStatus(qTaskInfo_t tinfo, uint64_t* uid, int64_t* ts) {
+  SExecTaskInfo* pTaskInfo = (SExecTaskInfo*)tinfo;
 
+  return doGetScanStatus(pTaskInfo->pRoot, uid, ts);
+}

source/libs/executor/src/executorimpl.c

@@ -274,6 +274,9 @@ SResultRow* doSetResultOutBufByKey(SDiskbasedBuf* pResultBuf, SResultRowInfo* pR
 
   // 1. close current opened time window
   if (pResultRowInfo->cur.pageId != -1 && ((pResult == NULL) || (pResult->pageId != pResultRowInfo->cur.pageId))) {
+#ifdef BUF_PAGE_DEBUG
+    qDebug("page_1");
+#endif
     SResultRowPosition pos = pResultRowInfo->cur;
     SFilePage*         pPage = getBufPage(pResultBuf, pos.pageId);
     releaseBufPage(pResultBuf, pPage);
@@ -281,6 +284,9 @@ SResultRow* doSetResultOutBufByKey(SDiskbasedBuf* pResultBuf, SResultRowInfo* pR
 
   // allocate a new buffer page
   if (pResult == NULL) {
+#ifdef BUF_PAGE_DEBUG
+    qDebug("page_2");
+#endif
     ASSERT(pSup->resultRowSize > 0);
     pResult = getNewResultRow(pResultBuf, groupId, pSup->resultRowSize);
 
@@ -538,7 +544,9 @@ static int32_t doAggregateImpl(SOperatorInfo* pOperator, TSKEY startTs, SqlFunct
       if (pCtx[k].fpSet.process == NULL) {
         continue;
       }
-
+#ifdef BUF_PAGE_DEBUG
+      qDebug("page_process");
+#endif
       int32_t code = pCtx[k].fpSet.process(&pCtx[k]);
       if (code != TSDB_CODE_SUCCESS) {
         qError("%s aggregate function error happens, code: %s", GET_TASKID(pOperator->pTaskInfo), tstrerror(code));
@@ -1033,7 +1041,7 @@ static uint32_t doFilterByBlockTimeWindow(STableScanInfo* pTableScanInfo, SSData
   SqlFunctionCtx* pCtx = pTableScanInfo->pCtx;
   uint32_t        status = BLK_DATA_NOT_LOAD;
 
-  int32_t numOfOutput = 0;//pTableScanInfo->numOfOutput;
+  int32_t numOfOutput = 0;  // pTableScanInfo->numOfOutput;
   for (int32_t i = 0; i < numOfOutput; ++i) {
     int32_t functionId = pCtx[i].functionId;
     int32_t colId = pTableScanInfo->pExpr[i].base.pParam[0].pCol->colId;
@@ -1348,7 +1356,7 @@ void extractQualifiedTupleByFilterResult(SSDataBlock* pBlock, const int8_t* rowR
       SColumnInfoData* pSrc = taosArrayGet(px->pDataBlock, i);
       SColumnInfoData* pDst = taosArrayGet(pBlock->pDataBlock, i);
       // it is a reserved column for scalar function, and no data in this column yet.
-      if (pDst->pData == NULL) {
+      if (pDst->pData == NULL || pSrc->pData == NULL) {
         continue;
       }
 
@@ -1413,7 +1421,9 @@ void setExecutionContext(SOperatorInfo* pOperator, int32_t numOfOutput, uint64_t
   if (pAggInfo->groupId != INT32_MIN && pAggInfo->groupId == groupId) {
     return;
   }
-
+#ifdef BUF_PAGE_DEBUG
+  qDebug("page_setbuf, groupId:%"PRIu64, groupId);
+#endif
   doSetTableGroupOutputBuf(pOperator, pAggInfo, numOfOutput, groupId);
 
   // record the current active group id
@@ -1489,11 +1499,15 @@ int32_t doCopyToSDataBlock(SExecTaskInfo* pTaskInfo, SSDataBlock* pBlock, SExprI
                            int32_t numOfExprs) {
   int32_t numOfRows = getNumOfTotalRes(pGroupResInfo);
   int32_t start = pGroupResInfo->index;
-
+#ifdef BUF_PAGE_DEBUG
+  qDebug("\npage_copytoblock rows:%d", numOfRows);
+#endif
   for (int32_t i = start; i < numOfRows; i += 1) {
     SResKeyPos* pPos = taosArrayGetP(pGroupResInfo->pRows, i);
     SFilePage*  page = getBufPage(pBuf, pPos->pos.pageId);
-
+#ifdef BUF_PAGE_DEBUG
+    qDebug("page_copytoblock pos pageId:%d, offset:%d", pPos->pos.pageId, pPos->pos.offset);
+#endif
     SResultRow* pRow = (SResultRow*)((char*)page + pPos->pos.offset);
 
     doUpdateNumOfRows(pRow, numOfExprs, rowCellOffset);
@@ -1525,6 +1539,9 @@ int32_t doCopyToSDataBlock(SExecTaskInfo* pTaskInfo, SSDataBlock* pBlock, SExprI
 
       pCtx[j].resultInfo = getResultEntryInfo(pRow, j, rowCellOffset);
       if (pCtx[j].fpSet.finalize) {
+#ifdef BUF_PAGE_DEBUG
+        qDebug("\npage_finalize %d", numOfExprs);
+#endif
         int32_t code = pCtx[j].fpSet.finalize(&pCtx[j], pBlock);
         if (TAOS_FAILED(code)) {
           qError("%s build result data block error, code %s", GET_TASKID(pTaskInfo), tstrerror(code));
@@ -1553,9 +1570,9 @@ int32_t doCopyToSDataBlock(SExecTaskInfo* pTaskInfo, SSDataBlock* pBlock, SExprI
 
     releaseBufPage(pBuf, page);
     pBlock->info.rows += pRow->numOfRows;
-    if (pBlock->info.rows >= pBlock->info.capacity) {  // output buffer is full
+//    if (pBlock->info.rows >= pBlock->info.capacity) {  // output buffer is full
-      break;
+//      break;
-    }
+//    }
   }
 
   qDebug("%s result generated, rows:%d, groupId:%" PRIu64, GET_TASKID(pTaskInfo), pBlock->info.rows,
@@ -1933,7 +1950,7 @@ typedef struct SFetchRspHandleWrapper {
   int32_t  sourceIndex;
 } SFetchRspHandleWrapper;
 
-int32_t loadRemoteDataCallback(void* param, const SDataBuf* pMsg, int32_t code) {
+int32_t loadRemoteDataCallback(void* param, SDataBuf* pMsg, int32_t code) {
   SFetchRspHandleWrapper* pWrapper = (SFetchRspHandleWrapper*)param;
 
   SExchangeInfo* pExchangeInfo = taosAcquireRef(exchangeObjRefPool, pWrapper->exchangeId);
@@ -2010,13 +2027,15 @@ static int32_t doSendFetchDataRequest(SExchangeInfo* pExchangeInfo, SExecTaskInf
 
   ASSERT(pDataInfo->status == EX_SOURCE_DATA_NOT_READY);
 
-  qDebug("%s build fetch msg and send to vgId:%d, ep:%s, taskId:0x%" PRIx64 ", %d/%" PRIzu, GET_TASKID(pTaskInfo),
+  qDebug("%s build fetch msg and send to vgId:%d, ep:%s, taskId:0x%" PRIx64 ", execId:%d, %d/%" PRIzu,
-         pSource->addr.nodeId, pSource->addr.epSet.eps[0].fqdn, pSource->taskId, sourceIndex, totalSources);
+         GET_TASKID(pTaskInfo), pSource->addr.nodeId, pSource->addr.epSet.eps[0].fqdn, pSource->taskId, pSource->execId,
+         sourceIndex, totalSources);
 
   pMsg->header.vgId = htonl(pSource->addr.nodeId);
   pMsg->sId = htobe64(pSource->schedId);
   pMsg->taskId = htobe64(pSource->taskId);
   pMsg->queryId = htobe64(pTaskInfo->id.queryId);
+  pMsg->execId = htonl(pSource->execId);
 
   // send the fetch remote task result reques
   SMsgSendInfo* pMsgSendInfo = taosMemoryCalloc(1, sizeof(SMsgSendInfo));
@@ -2034,7 +2053,7 @@ static int32_t doSendFetchDataRequest(SExchangeInfo* pExchangeInfo, SExecTaskInf
   pMsgSendInfo->param = pWrapper;
   pMsgSendInfo->msgInfo.pData = pMsg;
   pMsgSendInfo->msgInfo.len = sizeof(SResFetchReq);
-  pMsgSendInfo->msgType = TDMT_VND_FETCH;
+  pMsgSendInfo->msgType = TDMT_SCH_FETCH;
   pMsgSendInfo->fp = loadRemoteDataCallback;
 
   int64_t transporterId = 0;
@@ -2145,9 +2164,9 @@ static SSDataBlock* concurrentlyLoadRemoteDataImpl(SOperatorInfo* pOperator, SEx
       SSDataBlock*         pRes = pExchangeInfo->pResult;
       SLoadRemoteDataInfo* pLoadInfo = &pExchangeInfo->loadInfo;
       if (pRsp->numOfRows == 0) {
-        qDebug("%s vgId:%d, taskId:0x%" PRIx64 " index:%d completed, rowsOfSource:%" PRIu64 ", totalRows:%" PRIu64
+        qDebug("%s vgId:%d, taskId:0x%" PRIx64 " execId:%d index:%d completed, rowsOfSource:%" PRIu64
-               ", completed:%d try next %d/%" PRIzu,
+               ", totalRows:%" PRIu64 ", completed:%d try next %d/%" PRIzu,
-               GET_TASKID(pTaskInfo), pSource->addr.nodeId, pSource->taskId, i, pDataInfo->totalRows,
+               GET_TASKID(pTaskInfo), pSource->addr.nodeId, pSource->taskId, pSource->execId, i, pDataInfo->totalRows,
                pExchangeInfo->loadInfo.totalRows, completed + 1, i + 1, totalSources);
         pDataInfo->status = EX_SOURCE_DATA_EXHAUSTED;
         completed += 1;
@@ -2166,17 +2185,18 @@ static SSDataBlock* concurrentlyLoadRemoteDataImpl(SOperatorInfo* pOperator, SEx
 
       if (pRsp->completed == 1) {
         qDebug("%s fetch msg rsp from vgId:%d, taskId:0x%" PRIx64
+               " execId:%d"
                " index:%d completed, numOfRows:%d, rowsOfSource:%" PRIu64 ", totalRows:%" PRIu64 ", totalBytes:%" PRIu64
                ", completed:%d try next %d/%" PRIzu,
-               GET_TASKID(pTaskInfo), pSource->addr.nodeId, pSource->taskId, i, pRes->info.rows, pDataInfo->totalRows,
+               GET_TASKID(pTaskInfo), pSource->addr.nodeId, pSource->taskId, pSource->execId, i, pRes->info.rows,
-               pLoadInfo->totalRows, pLoadInfo->totalSize, completed + 1, i + 1, totalSources);
+               pDataInfo->totalRows, pLoadInfo->totalRows, pLoadInfo->totalSize, completed + 1, i + 1, totalSources);
         completed += 1;
         pDataInfo->status = EX_SOURCE_DATA_EXHAUSTED;
       } else {
-        qDebug("%s fetch msg rsp from vgId:%d, taskId:0x%" PRIx64 " numOfRows:%d, totalRows:%" PRIu64
+        qDebug("%s fetch msg rsp from vgId:%d, taskId:0x%" PRIx64 " execId:%d numOfRows:%d, totalRows:%" PRIu64
                ", totalBytes:%" PRIu64,
-               GET_TASKID(pTaskInfo), pSource->addr.nodeId, pSource->taskId, pRes->info.rows, pLoadInfo->totalRows,
+               GET_TASKID(pTaskInfo), pSource->addr.nodeId, pSource->taskId, pSource->execId, pRes->info.rows,
-               pLoadInfo->totalSize);
+               pLoadInfo->totalRows, pLoadInfo->totalSize);
       }
 
       taosMemoryFreeClear(pDataInfo->pRsp);
@@ -2249,8 +2269,8 @@ static SSDataBlock* seqLoadRemoteData(SOperatorInfo* pOperator) {
     SDownstreamSourceNode* pSource = taosArrayGet(pExchangeInfo->pSources, pExchangeInfo->current);
 
     if (pDataInfo->code != TSDB_CODE_SUCCESS) {
-      qError("%s vgId:%d, taskID:0x%" PRIx64 " error happens, code:%s", GET_TASKID(pTaskInfo), pSource->addr.nodeId,
+      qError("%s vgId:%d, taskID:0x%" PRIx64 " execId:%d error happens, code:%s", GET_TASKID(pTaskInfo),
-             pSource->taskId, tstrerror(pDataInfo->code));
+             pSource->addr.nodeId, pSource->taskId, pSource->execId, tstrerror(pDataInfo->code));
       pOperator->pTaskInfo->code = pDataInfo->code;
       return NULL;
     }
@@ -2258,9 +2278,9 @@ static SSDataBlock* seqLoadRemoteData(SOperatorInfo* pOperator) {
     SRetrieveTableRsp*   pRsp = pDataInfo->pRsp;
     SLoadRemoteDataInfo* pLoadInfo = &pExchangeInfo->loadInfo;
     if (pRsp->numOfRows == 0) {
-      qDebug("%s vgId:%d, taskID:0x%" PRIx64 " %d of total completed, rowsOfSource:%" PRIu64 ", totalRows:%" PRIu64
+      qDebug("%s vgId:%d, taskID:0x%" PRIx64 " execId:%d %d of total completed, rowsOfSource:%" PRIu64
-             " try next",
+             ", totalRows:%" PRIu64 " try next",
-             GET_TASKID(pTaskInfo), pSource->addr.nodeId, pSource->taskId, pExchangeInfo->current + 1,
+             GET_TASKID(pTaskInfo), pSource->addr.nodeId, pSource->taskId, pSource->execId, pExchangeInfo->current + 1,
              pDataInfo->totalRows, pLoadInfo->totalRows);
 
       pDataInfo->status = EX_SOURCE_DATA_EXHAUSTED;
@@ -2276,18 +2296,19 @@ static SSDataBlock* seqLoadRemoteData(SOperatorInfo* pOperator) {
                                      pTableRsp->compLen, pTableRsp->numOfCols, startTs, &pDataInfo->totalRows, NULL);
 
     if (pRsp->completed == 1) {
-      qDebug("%s fetch msg rsp from vgId:%d, taskId:0x%" PRIx64 " numOfRows:%d, rowsOfSource:%" PRIu64
+      qDebug("%s fetch msg rsp from vgId:%d, taskId:0x%" PRIx64 " execId:%d numOfRows:%d, rowsOfSource:%" PRIu64
              ", totalRows:%" PRIu64 ", totalBytes:%" PRIu64 " try next %d/%" PRIzu,
-             GET_TASKID(pTaskInfo), pSource->addr.nodeId, pSource->taskId, pRes->info.rows, pDataInfo->totalRows,
+             GET_TASKID(pTaskInfo), pSource->addr.nodeId, pSource->taskId, pSource->execId, pRes->info.rows,
-             pLoadInfo->totalRows, pLoadInfo->totalSize, pExchangeInfo->current + 1, totalSources);
+             pDataInfo->totalRows, pLoadInfo->totalRows, pLoadInfo->totalSize, pExchangeInfo->current + 1,
+             totalSources);
 
       pDataInfo->status = EX_SOURCE_DATA_EXHAUSTED;
       pExchangeInfo->current += 1;
     } else {
-      qDebug("%s fetch msg rsp from vgId:%d, taskId:0x%" PRIx64 " numOfRows:%d, totalRows:%" PRIu64
+      qDebug("%s fetch msg rsp from vgId:%d, taskId:0x%" PRIx64 " execId:%d numOfRows:%d, totalRows:%" PRIu64
              ", totalBytes:%" PRIu64,
-             GET_TASKID(pTaskInfo), pSource->addr.nodeId, pSource->taskId, pRes->info.rows, pLoadInfo->totalRows,
+             GET_TASKID(pTaskInfo), pSource->addr.nodeId, pSource->taskId, pSource->execId, pRes->info.rows,
-             pLoadInfo->totalSize);
+             pLoadInfo->totalRows, pLoadInfo->totalSize);
     }
 
     pOperator->resultInfo.totalRows += pRes->info.rows;
@@ -2372,13 +2393,12 @@ static int32_t initExchangeOperator(SExchangePhysiNode* pExNode, SExchangeInfo*
   }
 
   pInfo->pSources = taosArrayInit(numOfSources, sizeof(SDownstreamSourceNode));
-  pInfo->pSourceDataInfo = taosArrayInit(numOfSources, sizeof(SSourceDataInfo));
+  if (pInfo->pSources == NULL) {
-  if (pInfo->pSourceDataInfo == NULL || pInfo->pSources == NULL) {
     return TSDB_CODE_OUT_OF_MEMORY;
   }
 
   for (int32_t i = 0; i < numOfSources; ++i) {
-    SNodeListNode* pNode = (SNodeListNode*)nodesListGetNode((SNodeList*)pExNode->pSrcEndPoints, i);
+    SDownstreamSourceNode* pNode = (SDownstreamSourceNode*)nodesListGetNode((SNodeList*)pExNode->pSrcEndPoints, i);
     taosArrayPush(pInfo->pSources, pNode);
   }
 
@@ -2821,13 +2841,56 @@ int32_t getTableScanInfo(SOperatorInfo* pOperator, int32_t* order, int32_t* scan
   }
 }
 
+int32_t doPrepareScan(SOperatorInfo* pOperator, uint64_t uid, int64_t ts) {
+  int32_t type = pOperator->operatorType;
+  if (type == QUERY_NODE_PHYSICAL_PLAN_STREAM_SCAN) {
+    SStreamBlockScanInfo* pScanInfo = pOperator->info;
+    pScanInfo->blockType = STREAM_INPUT__DATA_SCAN;
+
+    STableScanInfo* pInfo = pScanInfo->pSnapshotReadOp->info;
+    if (uid == 0) {
+      pInfo->noTable = 1;
+      return TSDB_CODE_SUCCESS;
+    }
+
+    /*if (pSnapShotScanInfo->dataReader == NULL) {*/
+    /*pSnapShotScanInfo->dataReader = tsdbReaderOpen(pHandle->vnode, &pSTInfo->cond, tableList, 0, 0);*/
+    /*pSnapShotScanInfo->scanMode = TABLE_SCAN__TABLE_ORDER;*/
+    /*}*/
+
+    pInfo->noTable = 0;
+
+    if (pInfo->lastStatus.uid != uid || pInfo->lastStatus.ts != ts) {
+      tsdbSetTableId(pInfo->dataReader, uid);
+      int64_t oldSkey = pInfo->cond.twindows[0].skey;
+      pInfo->cond.twindows[0].skey = ts + 1;
+      tsdbResetReadHandle(pInfo->dataReader, &pInfo->cond, 0);
+      pInfo->cond.twindows[0].skey = oldSkey;
+      pInfo->scanTimes = 0;
+      pInfo->curTWinIdx = 0;
+    }
+    return TSDB_CODE_SUCCESS;
+
+  } else {
+    if (pOperator->numOfDownstream == 1) {
+      return doPrepareScan(pOperator->pDownstream[0], uid, ts);
+    } else if (pOperator->numOfDownstream == 0) {
+      qError("failed to find stream scan operator to set the input data block");
+      return TSDB_CODE_QRY_APP_ERROR;
+    } else {
+      qError("join not supported for stream block scan");
+      return TSDB_CODE_QRY_APP_ERROR;
+    }
+  }
+}
+
 int32_t doGetScanStatus(SOperatorInfo* pOperator, uint64_t* uid, int64_t* ts) {
   int32_t type = pOperator->operatorType;
   if (type == QUERY_NODE_PHYSICAL_PLAN_STREAM_SCAN) {
     SStreamBlockScanInfo* pScanInfo = pOperator->info;
-    STableScanInfo* pSnapShotScanInfo = pScanInfo->pSnapshotReadOp->info;
+    STableScanInfo*       pSnapShotScanInfo = pScanInfo->pSnapshotReadOp->info;
-    *uid = pSnapShotScanInfo->scanStatus.uid;
+    *uid = pSnapShotScanInfo->lastStatus.uid;
-    *ts = pSnapShotScanInfo->scanStatus.t;
+    *ts = pSnapShotScanInfo->lastStatus.ts;
   } else {
     if (pOperator->pDownstream[0] == NULL) {
       return TSDB_CODE_INVALID_PARA;
@@ -3121,8 +3184,9 @@ static int32_t handleLimitOffset(SOperatorInfo* pOperator, SSDataBlock* pBlock)
   }
 
   // check for the limitation in each group
-  if (pProjectInfo->limit.limit > 0 && pProjectInfo->curOutput + pRes->info.rows >= pProjectInfo->limit.limit) {
+  if (pProjectInfo->limit.limit >= 0 && pProjectInfo->curOutput + pRes->info.rows >= pProjectInfo->limit.limit) {
-    pRes->info.rows = (int32_t)(pProjectInfo->limit.limit - pProjectInfo->curOutput);
+    int32_t keepRows = (int32_t)(pProjectInfo->limit.limit - pProjectInfo->curOutput);
+    blockDataKeepFirstNRows(pRes, keepRows);
     if (pProjectInfo->slimit.limit > 0 && pProjectInfo->slimit.limit <= pProjectInfo->curGroupOutput) {
       pOperator->status = OP_EXEC_DONE;
     }
@@ -3196,8 +3260,6 @@ static SSDataBlock* doProjectOperation(SOperatorInfo* pOperator) {
       longjmp(pTaskInfo->env, code);
     }
 
-    doFilter(pProjectInfo->pFilterNode, pBlock);
-
     setInputDataBlock(pOperator, pSup->pCtx, pBlock, order, scanFlag, false);
     blockDataEnsureCapacity(pInfo->pRes, pInfo->pRes->info.rows + pBlock->info.rows);
 
@@ -3208,6 +3270,10 @@ static SSDataBlock* doProjectOperation(SOperatorInfo* pOperator) {
     }
 
     int32_t status = handleLimitOffset(pOperator, pBlock);
+
+    // filter shall be applied after apply functions and limit/offset on the result
+    doFilter(pProjectInfo->pFilterNode, pInfo->pRes);
+
     if (status == PROJECT_RETRIEVE_CONTINUE) {
       continue;
     } else if (status == PROJECT_RETRIEVE_DONE) {
@@ -3434,11 +3500,12 @@ int32_t initAggInfo(SExprSupp* pSup, SAggSupporter* pAggSup, SExprInfo* pExprInf
 }
 
 void initResultSizeInfo(SOperatorInfo* pOperator, int32_t numOfRows) {
+  ASSERT(numOfRows != 0);
   pOperator->resultInfo.capacity = numOfRows;
   pOperator->resultInfo.threshold = numOfRows * 0.75;
 
   if (pOperator->resultInfo.threshold == 0) {
-    pOperator->resultInfo.capacity = numOfRows;
+    pOperator->resultInfo.threshold = numOfRows;
   }
 }
 
@@ -3938,12 +4005,12 @@ static int32_t sortTableGroup(STableListInfo* pTableListInfo, int32_t groupNum)
         return TSDB_CODE_QRY_APP_ERROR;
       }
       if (p == NULL) {
-        if (taosArrayPush(sortSupport, groupId) != NULL) {
+        if (taosArrayPush(sortSupport, groupId) == NULL) {
           qError("taos push support array error");
           taosArrayDestroy(sortSupport);
           return TSDB_CODE_QRY_APP_ERROR;
         }
-        if (taosArrayPush(pTableListInfo->pGroupList, &tGroup) != NULL) {
+        if (taosArrayPush(pTableListInfo->pGroupList, &tGroup) == NULL) {
           qError("taos push group array error");
           taosArrayDestroy(sortSupport);
           return TSDB_CODE_QRY_APP_ERROR;
@@ -4023,6 +4090,7 @@ int32_t generateGroupIdMap(STableListInfo* pTableListInfo, SReadHandle* pHandle,
       } else {
         taosMemoryFree(keyBuf);
         nodesClearList(groupNew);
+        metaReaderClear(&mr);
         return code;
       }
 
@@ -4035,7 +4103,14 @@ int32_t generateGroupIdMap(STableListInfo* pTableListInfo, SReadHandle* pHandle,
       } else {
         isNull[index++] = 0;
         char* data = nodesGetValueFromNode(pValue);
-        if (pValue->node.resType.type == TSDB_DATA_TYPE_JSON) {
+        if (pValue->node.resType.type == TSDB_DATA_TYPE_JSON){
+          if(tTagIsJson(data)){
+            terrno = TSDB_CODE_QRY_JSON_IN_GROUP_ERROR;
+            taosMemoryFree(keyBuf);
+            nodesClearList(groupNew);
+            metaReaderClear(&mr);
+            return terrno;
+          }
           int32_t len = getJsonValueLen(data);
           memcpy(pStart, data, len);
           pStart += len;
@@ -4094,7 +4169,8 @@ SOperatorInfo* createOperatorTree(SPhysiNode* pPhyNode, SExecTaskInfo* pTaskInfo
     } else if (QUERY_NODE_PHYSICAL_PLAN_TABLE_MERGE_SCAN == type) {
       STableMergeScanPhysiNode* pTableScanNode = (STableMergeScanPhysiNode*)pPhyNode;
       int32_t code = createScanTableListInfo(pTableScanNode, pHandle, pTableListInfo, queryId, taskId);
-      if (code) {
+      if(code){
+        pTaskInfo->code = code;
         return NULL;
       }
       code = extractTableSchemaVersion(pHandle, pTableScanNode->scan.uid, pTaskInfo);
@@ -4121,7 +4197,11 @@ SOperatorInfo* createOperatorTree(SPhysiNode* pPhyNode, SExecTaskInfo* pTaskInfo
             .maxTs = INT64_MIN,
       };
       if (pHandle) {
-        createScanTableListInfo(pTableScanNode, pHandle, pTableListInfo, queryId, taskId);
+        int32_t code = createScanTableListInfo(pTableScanNode, pHandle, pTableListInfo, queryId, taskId);
+        if(code){
+          pTaskInfo->code = code;
+          return NULL;
+        }
       }
 
       SOperatorInfo* pOperator =

source/libs/executor/src/groupoperator.c

@@ -141,6 +141,10 @@ static void recordNewGroupKeys(SArray* pGroupCols, SArray* pGroupColVals, SSData
       pkey->isNull = false;
       char* val = colDataGetData(pColInfoData, rowIndex);
       if (pkey->type == TSDB_DATA_TYPE_JSON) {
+        if(tTagIsJson(val)){
+          terrno = TSDB_CODE_QRY_JSON_IN_GROUP_ERROR;
+          return;
+        }
         int32_t dataLen = getJsonValueLen(val);
         memcpy(pkey->pData, val, dataLen);
       } else if (IS_VAR_DATA_TYPE(pkey->type)) {
@@ -227,11 +231,15 @@ static void doHashGroupbyAgg(SOperatorInfo* pOperator, SSDataBlock* pBlock) {
   int32_t     len = 0;
   STimeWindow w = TSWINDOW_INITIALIZER;
 
+  terrno = TSDB_CODE_SUCCESS;
   int32_t num = 0;
   for (int32_t j = 0; j < pBlock->info.rows; ++j) {
     // Compare with the previous row of this column, and do not set the output buffer again if they are identical.
     if (!pInfo->isInit) {
       recordNewGroupKeys(pInfo->pGroupCols, pInfo->pGroupColVals, pBlock, j);
+      if (terrno != TSDB_CODE_SUCCESS) {  // group by json error
+        longjmp(pTaskInfo->env, terrno);
+      }
       pInfo->isInit = true;
       num++;
       continue;
@@ -247,6 +255,9 @@ static void doHashGroupbyAgg(SOperatorInfo* pOperator, SSDataBlock* pBlock) {
     if (j == 0) {
       num++;
       recordNewGroupKeys(pInfo->pGroupCols, pInfo->pGroupColVals, pBlock, j);
+      if (terrno != TSDB_CODE_SUCCESS) {  // group by json error
+        longjmp(pTaskInfo->env, terrno);
+      }
       continue;
     }
 
@@ -661,7 +672,11 @@ static SSDataBlock* hashPartition(SOperatorInfo* pOperator) {
       }
     }
 
+    terrno = TSDB_CODE_SUCCESS;
     doHashPartition(pOperator, pBlock);
+    if (terrno != TSDB_CODE_SUCCESS) {  // group by json error
+      longjmp(pTaskInfo->env, terrno);
+    }
   }
 
   SArray* groupArray = taosArrayInit(taosHashGetSize(pInfo->pGroupSet), sizeof(SDataGroupInfo));

source/libs/executor/src/scanoperator.c

@@ -392,6 +392,7 @@ static SSDataBlock* doTableScanImpl(SOperatorInfo* pOperator) {
     binfo.capacity = binfo.rows;
     blockDataEnsureCapacity(pBlock, binfo.rows);
     pBlock->info = binfo;
+    ASSERT(binfo.uid != 0);
 
     uint32_t status = 0;
     int32_t  code = loadDataBlock(pOperator, pTableScanInfo, pBlock, &status);
@@ -416,9 +417,10 @@ static SSDataBlock* doTableScanImpl(SOperatorInfo* pOperator) {
     pOperator->cost.totalCost = pTableScanInfo->readRecorder.elapsedTime;
 
     // todo refactor
-    pTableScanInfo->scanStatus.uid = pBlock->info.uid;
+    pTableScanInfo->lastStatus.uid = pBlock->info.uid;
-    pTableScanInfo->scanStatus.t = pBlock->info.window.ekey;
+    pTableScanInfo->lastStatus.ts = pBlock->info.window.ekey;
 
+    ASSERT(pBlock->info.uid != 0);
     return pBlock;
   }
   return NULL;
@@ -438,6 +440,7 @@ static SSDataBlock* doTableScanGroup(SOperatorInfo* pOperator) {
     while (pTableScanInfo->curTWinIdx < pTableScanInfo->cond.numOfTWindows) {
       SSDataBlock* p = doTableScanImpl(pOperator);
       if (p != NULL) {
+        ASSERT(p->info.uid != 0);
         return p;
       }
       pTableScanInfo->curTWinIdx += 1;
@@ -513,6 +516,27 @@ static SSDataBlock* doTableScan(SOperatorInfo* pOperator) {
   STableScanInfo* pInfo = pOperator->info;
   SExecTaskInfo*  pTaskInfo = pOperator->pTaskInfo;
 
+  // if scan table by table
+  if (pInfo->scanMode == TABLE_SCAN__TABLE_ORDER) {
+    if (pInfo->noTable) return NULL;
+    while (1) {
+      SSDataBlock* result = doTableScanGroup(pOperator);
+      if (result) {
+        return result;
+      }
+      // if no data, switch to next table and continue scan
+      pInfo->currentTable++;
+      if (pInfo->currentTable >= taosArrayGetSize(pTaskInfo->tableqinfoList.pTableList)) {
+        return NULL;
+      }
+      STableKeyInfo* pTableInfo = taosArrayGet(pTaskInfo->tableqinfoList.pTableList, pInfo->currentTable);
+      tsdbSetTableId(pInfo->dataReader, pTableInfo->uid);
+      tsdbResetReadHandle(pInfo->dataReader, &pInfo->cond, 0);
+      pInfo->scanTimes = 0;
+      pInfo->curTWinIdx = 0;
+    }
+  }
+
   if (pInfo->currentGroupId == -1) {
     pInfo->currentGroupId++;
     if (pInfo->currentGroupId >= taosArrayGetSize(pTaskInfo->tableqinfoList.pGroupList)) {
@@ -1188,7 +1212,7 @@ SOperatorInfo* createStreamScanOperatorInfo(SReadHandle* pHandle, STableScanPhys
 
     int16_t colId = id->colId;
     taosArrayPush(pColIds, &colId);
-    if (id->colId == pTableScanNode->tsColId) {
+    if (id->colId == PRIMARYKEY_TIMESTAMP_COL_ID) {
       pInfo->primaryTsIndex = id->targetSlotId;
     }
   }
@@ -1207,6 +1231,13 @@ SOperatorInfo* createStreamScanOperatorInfo(SReadHandle* pHandle, STableScanPhys
   if (pHandle) {
     SOperatorInfo*  pTableScanDummy = createTableScanOperatorInfo(pTableScanNode, pHandle, pTaskInfo, queryId, taskId);
     STableScanInfo* pSTInfo = (STableScanInfo*)pTableScanDummy->info;
+
+    SArray* tableList = taosArrayGetP(pTaskInfo->tableqinfoList.pGroupList, 0);
+    if (pHandle->tqReader) {
+      pSTInfo->scanMode = TABLE_SCAN__TABLE_ORDER;
+      pSTInfo->dataReader = tsdbReaderOpen(pHandle->vnode, &pSTInfo->cond, tableList, 0, 0);
+    }
+
     if (pSTInfo->interval.interval > 0) {
       pInfo->pUpdateInfo = updateInfoInitP(&pSTInfo->interval, pTwSup->waterMark);
     } else {
@@ -1243,6 +1274,7 @@ SOperatorInfo* createStreamScanOperatorInfo(SReadHandle* pHandle, STableScanPhys
   pInfo->sessionSup = (SessionWindowSupporter){.pStreamAggSup = NULL, .gap = -1};
   pInfo->groupId = 0;
   pInfo->pPullDataRes = createPullDataBlock();
+  pInfo->pStreamScanOp = pOperator;
 
   pOperator->name = "StreamBlockScanOperator";
   pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_STREAM_SCAN;
@@ -1330,7 +1362,7 @@ static void getDBNameFromCondition(SNode* pCondition, const char* dbName) {
   nodesWalkExpr(pCondition, getDBNameFromConditionWalker, (char*)dbName);
 }
 
-static int32_t loadSysTableCallback(void* param, const SDataBuf* pMsg, int32_t code) {
+static int32_t loadSysTableCallback(void* param, SDataBuf* pMsg, int32_t code) {
   SOperatorInfo*     operator=(SOperatorInfo*) param;
   SSysTableScanInfo* pScanResInfo = (SSysTableScanInfo*)operator->info;
   if (TSDB_CODE_SUCCESS == code) {
@@ -2059,6 +2091,7 @@ int32_t createScanTableListInfo(STableScanPhysiNode* pTableScanNode, SReadHandle
     qDebug("no table qualified for query, TID:0x%" PRIx64 ", QID:0x%" PRIx64, taskId, queryId);
     return TSDB_CODE_SUCCESS;
   }
+  pTableListInfo->needSortTableByGroupId = pTableScanNode->groupSort;
   code = generateGroupIdMap(pTableListInfo, pHandle, pTableScanNode->pGroupTags);
   if (code != TSDB_CODE_SUCCESS) {
     return code;

source/libs/executor/src/timewindowoperator.c

@@ -1246,6 +1246,9 @@ void doClearWindowImpl(SResultRowPosition* p1, SDiskbasedBuf* pResultBuf, SExprS
       pCtx[i].fpSet.init(&pCtx[i], pResInfo);
     }
   }
+  SFilePage* bufPage = getBufPage(pResultBuf, p1->pageId);
+  setBufPageDirty(bufPage, true);
+  releaseBufPage(pResultBuf, bufPage);
 }
 
 bool doClearWindow(SAggSupporter* pAggSup, SExprSupp* pSup, char* pData, int16_t bytes, uint64_t groupId,
@@ -3171,6 +3174,10 @@ static void doClearSessionWindows(SStreamAggSupporter* pAggSup, SExprSupp* pSup,
         getSessionTimeWindow(pAggSup, tsCols[i], INT64_MIN, pBlock->info.groupId, gap, &winIndex);
     step = updateSessionWindowInfo(pCurWin, tsCols, NULL, pBlock->info.rows, i, gap, NULL);
     ASSERT(isInWindow(pCurWin, tsCols[i], gap));
+    if (pCurWin->pos.pageId == -1) {
+      // window has been closed.
+      continue;
+    }
     doClearWindowImpl(&pCurWin->pos, pAggSup->pResultBuf, pSup, numOfOutput);
     if (result) {
       taosArrayPush(result, pCurWin);
@@ -3246,12 +3253,12 @@ static void rebuildTimeWindow(SStreamSessionAggOperatorInfo* pInfo, SArray* pWin
           setWindowOutputBuf(pChWin, &pChResult, pChild->exprSupp.pCtx, groupId, numOfOutput,
                              pChild->exprSupp.rowEntryInfoOffset, &pChInfo->streamAggSup, pTaskInfo);
           compactFunctions(pSup->pCtx, pChild->exprSupp.pCtx, numOfOutput, pTaskInfo);
-          SFilePage* bufPage = getBufPage(pInfo->streamAggSup.pResultBuf, pChWin->pos.pageId);
+          SFilePage* bufPage = getBufPage(pChInfo->streamAggSup.pResultBuf, pChWin->pos.pageId);
-          setBufPageDirty(bufPage, true);
+          releaseBufPage(pChInfo->streamAggSup.pResultBuf, bufPage);
-          releaseBufPage(pInfo->streamAggSup.pResultBuf, bufPage);
           continue;
+        } else if (!pChWin->isClosed) {
+          break;
         }
-        break;
       }
     }
     SFilePage* bufPage = getBufPage(pInfo->streamAggSup.pResultBuf, pParentWin->pos.pageId);
@@ -3265,7 +3272,8 @@ typedef SResultWindowInfo* (*__get_win_info_)(void*);
 SResultWindowInfo* getResWinForSession(void* pData) { return (SResultWindowInfo*)pData; }
 SResultWindowInfo* getResWinForState(void* pData) { return &((SStateWindowInfo*)pData)->winInfo; }
 
-int32_t closeSessionWindow(SHashObj* pHashMap, STimeWindowAggSupp* pTwSup, SArray* pClosed, __get_win_info_ fn) {
+int32_t closeSessionWindow(SHashObj* pHashMap, STimeWindowAggSupp* pTwSup,
+    SArray* pClosed, __get_win_info_ fn, bool delete) {
   // Todo(liuyao) save window to tdb
   void** pIte = NULL;
   size_t keyLen = 0;
@@ -3279,10 +3287,18 @@ int32_t closeSessionWindow(SHashObj* pHashMap, STimeWindowAggSupp* pTwSup, SArra
       if (isCloseWindow(&pSeWin->win, pTwSup)) {
         if (!pSeWin->isClosed) {
           pSeWin->isClosed = true;
-          if (pTwSup->calTrigger == STREAM_TRIGGER_WINDOW_CLOSE) {
+          if (pTwSup->calTrigger == STREAM_TRIGGER_WINDOW_CLOSE && pClosed) {
             int32_t code = saveResult(pSeWin->win.skey, pSeWin->pos.pageId, pSeWin->pos.offset, *pGroupId, pClosed);
+            if (code != TSDB_CODE_SUCCESS) {
+              return code;
+            }
             pSeWin->isOutput = true;
           }
+          if (delete) {
+            taosArrayRemove(pWins, i);
+            i--;
+            size = taosArrayGetSize(pWins);
+          }
         }
         continue;
       }
@@ -3292,6 +3308,16 @@ int32_t closeSessionWindow(SHashObj* pHashMap, STimeWindowAggSupp* pTwSup, SArra
   return TSDB_CODE_SUCCESS;
 }
 
+static void closeChildSessionWindow(SArray* pChildren, TSKEY maxTs, bool delete) {
+  int32_t size = taosArrayGetSize(pChildren);
+  for (int32_t i = 0; i < size; i++) {
+    SOperatorInfo*                 pChildOp = taosArrayGetP(pChildren, i);
+    SStreamSessionAggOperatorInfo* pChInfo = pChildOp->info;
+    pChInfo->twAggSup.maxTs = TMAX(pChInfo->twAggSup.maxTs, maxTs);
+    closeSessionWindow(pChInfo->streamAggSup.pResultRows, &pChInfo->twAggSup, NULL, getResWinForSession, delete);
+  }
+}
+
 int32_t getAllSessionWindow(SHashObj* pHashMap, SArray* pClosed, __get_win_info_ fn) {
   void** pIte = NULL;
   while ((pIte = taosHashIterate(pHashMap, pIte)) != NULL) {
@@ -3339,6 +3365,7 @@ static SSDataBlock* doStreamSessionAgg(SOperatorInfo* pOperator) {
     if (pBlock == NULL) {
       break;
     }
+    printDataBlock(pBlock, IS_FINAL_OP(pInfo) ? "Final  Session Recv" : "Single Session Recv");
 
     if (pBlock->info.type == STREAM_CLEAR) {
       SArray* pWins = taosArrayInit(16, sizeof(SResultWindowInfo));
@@ -3385,7 +3412,9 @@ static SSDataBlock* doStreamSessionAgg(SOperatorInfo* pOperator) {
   // restore the value
   pOperator->status = OP_RES_TO_RETURN;
 
-  closeSessionWindow(pInfo->streamAggSup.pResultRows, &pInfo->twAggSup, pUpdated, getResWinForSession);
+  closeSessionWindow(pInfo->streamAggSup.pResultRows, &pInfo->twAggSup, pUpdated,
+      getResWinForSession, pInfo->ignoreCloseWindow);
+  closeChildSessionWindow(pInfo->pChildren, pInfo->twAggSup.maxTs, pInfo->ignoreCloseWindow);
   copyUpdateResult(pStUpdated, pUpdated);
   taosHashCleanup(pStUpdated);
 
@@ -3437,10 +3466,11 @@ static SSDataBlock* doStreamSessionSemiAgg(SOperatorInfo* pOperator) {
   } else if (pOperator->status == OP_RES_TO_RETURN) {
     doBuildDeleteDataBlock(pInfo->pStDeleted, pInfo->pDelRes, &pInfo->pDelIterator);
     if (pInfo->pDelRes->info.rows > 0) {
+      printDataBlock(pInfo->pDelRes, "Semi  Session");
       return pInfo->pDelRes;
     }
     doBuildResultDatablock(pOperator, pBInfo, &pInfo->groupResInfo, pInfo->streamAggSup.pResultBuf);
-    if (pInfo->binfo.pRes->info.rows == 0) {
+    if (pBInfo->pRes->info.rows == 0) {
       pOperator->status = OP_EXEC_DONE;
       if (pInfo->pUpdateRes->info.rows == 0) {
         // semi interval operator clear disk buffer
@@ -3449,9 +3479,11 @@ static SSDataBlock* doStreamSessionSemiAgg(SOperatorInfo* pOperator) {
       }
       // process the rest of the data
       pOperator->status = OP_OPENED;
+      printDataBlock(pInfo->pUpdateRes, "Semi  Session");
       return pInfo->pUpdateRes;
     }
-    return pInfo->binfo.pRes;
+    printDataBlock(pBInfo->pRes, "Semi  Session");
+    return pBInfo->pRes;
   }
 
   _hash_fn_t     hashFn = taosGetDefaultHashFunction(TSDB_DATA_TYPE_BINARY);
@@ -3495,21 +3527,24 @@ static SSDataBlock* doStreamSessionSemiAgg(SOperatorInfo* pOperator) {
   finalizeUpdatedResult(pOperator->exprSupp.numOfExprs, pInfo->streamAggSup.pResultBuf, pUpdated,
                         pSup->rowEntryInfoOffset);
   initMultiResInfoFromArrayList(&pInfo->groupResInfo, pUpdated);
-  blockDataEnsureCapacity(pInfo->binfo.pRes, pOperator->resultInfo.capacity);
+  blockDataEnsureCapacity(pBInfo->pRes, pOperator->resultInfo.capacity);
   doBuildDeleteDataBlock(pInfo->pStDeleted, pInfo->pDelRes, &pInfo->pDelIterator);
   if (pInfo->pDelRes->info.rows > 0) {
+    printDataBlock(pInfo->pDelRes, "Semi  Session");
     return pInfo->pDelRes;
   }
-  doBuildResultDatablock(pOperator, &pInfo->binfo, &pInfo->groupResInfo, pInfo->streamAggSup.pResultBuf);
+  doBuildResultDatablock(pOperator, pBInfo, &pInfo->groupResInfo, pInfo->streamAggSup.pResultBuf);
-  if (pInfo->binfo.pRes->info.rows == 0) {
+  if (pBInfo->pRes->info.rows == 0) {
     pOperator->status = OP_EXEC_DONE;
     if (pInfo->pUpdateRes->info.rows == 0) {
       return NULL;
     }
     // process the rest of the data
     pOperator->status = OP_OPENED;
+    printDataBlock(pInfo->pUpdateRes, "Semi  Session");
     return pInfo->pUpdateRes;
   }
+  printDataBlock(pBInfo->pRes, "Semi  Session");
   return pBInfo->pRes->info.rows == 0 ? NULL : pBInfo->pRes;
 }
 
@@ -3867,7 +3902,9 @@ static SSDataBlock* doStreamStateAgg(SOperatorInfo* pOperator) {
   // restore the value
   pOperator->status = OP_RES_TO_RETURN;
 
-  closeSessionWindow(pInfo->streamAggSup.pResultRows, &pInfo->twAggSup, pUpdated, getResWinForState);
+  closeSessionWindow(pInfo->streamAggSup.pResultRows, &pInfo->twAggSup, pUpdated,
+      getResWinForState, pInfo->ignoreCloseWindow);
+  closeChildSessionWindow(pInfo->pChildren, pInfo->twAggSup.maxTs, pInfo->ignoreCloseWindow);
   copyUpdateResult(pSeUpdated, pUpdated);
   taosHashCleanup(pSeUpdated);
 

source/libs/executor/src/tsort.c

@@ -593,7 +593,10 @@ static int32_t createInitialSources(SSortHandle* pHandle) {
       if (size > sortBufSize) {
         // Perform the in-memory sort and then flush data in the buffer into disk.
         int64_t p = taosGetTimestampUs();
-        blockDataSort(pHandle->pDataBlock, pHandle->pSortInfo);
+        code = blockDataSort(pHandle->pDataBlock, pHandle->pSortInfo);
+        if (code != 0) {
+          return code;
+        }
 
         int64_t el = taosGetTimestampUs() - p;
         pHandle->sortElapsed += el;
@@ -608,7 +611,10 @@ static int32_t createInitialSources(SSortHandle* pHandle) {
       // Perform the in-memory sort and then flush data in the buffer into disk.
       int64_t p = taosGetTimestampUs();
 
-      blockDataSort(pHandle->pDataBlock, pHandle->pSortInfo);
+      int32_t code = blockDataSort(pHandle->pDataBlock, pHandle->pSortInfo);
+      if (code != 0) {
+        return code;
+      }
 
       int64_t el = taosGetTimestampUs() - p;
       pHandle->sortElapsed += el;

source/libs/function/inc/builtinsimpl.h

@@ -117,6 +117,9 @@ int32_t firstCombine(SqlFunctionCtx* pDestCtx, SqlFunctionCtx* pSourceCtx);
 int32_t lastCombine(SqlFunctionCtx* pDestCtx, SqlFunctionCtx* pSourceCtx);
 int32_t getFirstLastInfoSize(int32_t resBytes);
 
+int32_t lastRowFunction(SqlFunctionCtx *pCtx);
+int32_t lastRowFinalize(SqlFunctionCtx* pCtx, SSDataBlock* pBlock);
+
 bool getTopBotFuncEnv(SFunctionNode* UNUSED_PARAM(pFunc), SFuncExecEnv* pEnv);
 bool getTopBotMergeFuncEnv(SFunctionNode* UNUSED_PARAM(pFunc), SFuncExecEnv* pEnv);
 bool topBotFunctionSetup(SqlFunctionCtx *pCtx, SResultRowEntryInfo* pResultInfo);

source/libs/function/inc/functionMgtInt.h

@@ -32,7 +32,7 @@ extern "C" {
 #define FUNC_MGT_STRING_FUNC            FUNC_MGT_FUNC_CLASSIFICATION_MASK(3)
 #define FUNC_MGT_DATETIME_FUNC          FUNC_MGT_FUNC_CLASSIFICATION_MASK(4)
 #define FUNC_MGT_TIMELINE_FUNC          FUNC_MGT_FUNC_CLASSIFICATION_MASK(5)
-#define FUNC_MGT_TIMEORDER_FUNC         FUNC_MGT_FUNC_CLASSIFICATION_MASK(6)
+#define FUNC_MGT_IMPLICIT_TS_FUNC       FUNC_MGT_FUNC_CLASSIFICATION_MASK(6)
 #define FUNC_MGT_PSEUDO_COLUMN_FUNC     FUNC_MGT_FUNC_CLASSIFICATION_MASK(7)
 #define FUNC_MGT_WINDOW_PC_FUNC         FUNC_MGT_FUNC_CLASSIFICATION_MASK(8)
 #define FUNC_MGT_SPECIAL_DATA_REQUIRED  FUNC_MGT_FUNC_CLASSIFICATION_MASK(9)

source/libs/function/src/builtins.c

@@ -44,6 +44,7 @@ void static addDbPrecisonParam(SNodeList** pList, uint8_t precision) {
   pVal->literal = NULL;
   pVal->isDuration = false;
   pVal->translate = true;
+  pVal->notReserved = true;
   pVal->node.resType.type = TSDB_DATA_TYPE_TINYINT;
   pVal->node.resType.bytes = tDataTypes[TSDB_DATA_TYPE_TINYINT].bytes;
   pVal->node.resType.precision = precision;
@@ -237,7 +238,7 @@ static int32_t translateWduration(SFunctionNode* pFunc, char* pErrBuf, int32_t l
 static int32_t translateNowToday(SFunctionNode* pFunc, char* pErrBuf, int32_t len) {
   // pseudo column do not need to check parameters
 
-  //add database precision as param
+  // add database precision as param
   uint8_t dbPrec = pFunc->node.resType.precision;
   addDbPrecisonParam(&pFunc->pParameterList, dbPrec);
 
@@ -633,6 +634,12 @@ static int32_t translateHistogram(SFunctionNode* pFunc, char* pErrBuf, int32_t l
   }
 
   // param1 ~ param3
+  if (((SExprNode*)nodesListGetNode(pFunc->pParameterList, 1))->resType.type != TSDB_DATA_TYPE_BINARY ||
+      ((SExprNode*)nodesListGetNode(pFunc->pParameterList, 2))->resType.type != TSDB_DATA_TYPE_BINARY ||
+      ((SExprNode*)nodesListGetNode(pFunc->pParameterList, 3))->resType.type != TSDB_DATA_TYPE_BIGINT) {
+    return invaildFuncParaTypeErrMsg(pErrBuf, len, pFunc->functionName);
+  }
+
   for (int32_t i = 1; i < numOfParams; ++i) {
     SNode* pParamNode = nodesListGetNode(pFunc->pParameterList, i);
     if (QUERY_NODE_VALUE != nodeType(pParamNode)) {
@@ -642,12 +649,11 @@ static int32_t translateHistogram(SFunctionNode* pFunc, char* pErrBuf, int32_t l
     SValueNode* pValue = (SValueNode*)pParamNode;
 
     pValue->notReserved = true;
-  }
 
-  if (((SExprNode*)nodesListGetNode(pFunc->pParameterList, 1))->resType.type != TSDB_DATA_TYPE_BINARY ||
+    if (i == 3 && pValue->datum.i != 1 && pValue->datum.i != 0) {
-      ((SExprNode*)nodesListGetNode(pFunc->pParameterList, 2))->resType.type != TSDB_DATA_TYPE_BINARY ||
+        return buildFuncErrMsg(pErrBuf, len, TSDB_CODE_FUNC_FUNTION_ERROR,
-      ((SExprNode*)nodesListGetNode(pFunc->pParameterList, 3))->resType.type != TSDB_DATA_TYPE_BIGINT) {
+                               "HISTOGRAM function normalized parameter should be 0/1");
-    return invaildFuncParaTypeErrMsg(pErrBuf, len, pFunc->functionName);
+    }
   }
 
   pFunc->node.resType = (SDataType){.bytes = 512, .type = TSDB_DATA_TYPE_BINARY};
@@ -667,6 +673,12 @@ static int32_t translateHistogramImpl(SFunctionNode* pFunc, char* pErrBuf, int32
     }
 
     // param1 ~ param3
+    if (((SExprNode*)nodesListGetNode(pFunc->pParameterList, 1))->resType.type != TSDB_DATA_TYPE_BINARY ||
+        ((SExprNode*)nodesListGetNode(pFunc->pParameterList, 2))->resType.type != TSDB_DATA_TYPE_BINARY ||
+        ((SExprNode*)nodesListGetNode(pFunc->pParameterList, 3))->resType.type != TSDB_DATA_TYPE_BIGINT) {
+      return invaildFuncParaTypeErrMsg(pErrBuf, len, pFunc->functionName);
+    }
+
     for (int32_t i = 1; i < numOfParams; ++i) {
       SNode* pParamNode = nodesListGetNode(pFunc->pParameterList, i);
       if (QUERY_NODE_VALUE != nodeType(pParamNode)) {
@@ -676,12 +688,11 @@ static int32_t translateHistogramImpl(SFunctionNode* pFunc, char* pErrBuf, int32
       SValueNode* pValue = (SValueNode*)pParamNode;
 
       pValue->notReserved = true;
-    }
 
-    if (((SExprNode*)nodesListGetNode(pFunc->pParameterList, 1))->resType.type != TSDB_DATA_TYPE_BINARY ||
+      if (i == 3 && pValue->datum.i != 1 && pValue->datum.i != 0) {
-        ((SExprNode*)nodesListGetNode(pFunc->pParameterList, 2))->resType.type != TSDB_DATA_TYPE_BINARY ||
+          return buildFuncErrMsg(pErrBuf, len, TSDB_CODE_FUNC_FUNTION_ERROR,
-        ((SExprNode*)nodesListGetNode(pFunc->pParameterList, 3))->resType.type != TSDB_DATA_TYPE_BIGINT) {
+                                 "HISTOGRAM function normalized parameter should be 0/1");
-      return invaildFuncParaTypeErrMsg(pErrBuf, len, pFunc->functionName);
+      }
     }
 
     pFunc->node.resType =
@@ -1016,6 +1027,10 @@ static int32_t translateIrate(SFunctionNode* pFunc, char* pErrBuf, int32_t len)
     return invaildFuncParaTypeErrMsg(pErrBuf, len, pFunc->functionName);
   }
 
+  // add database precision as param
+  uint8_t dbPrec = pFunc->node.resType.precision;
+  addDbPrecisonParam(&pFunc->pParameterList, dbPrec);
+
   pFunc->node.resType = (SDataType){.bytes = tDataTypes[TSDB_DATA_TYPE_DOUBLE].bytes, .type = TSDB_DATA_TYPE_DOUBLE};
   return TSDB_CODE_SUCCESS;
 }
@@ -1026,13 +1041,7 @@ static int32_t translateFirstLast(SFunctionNode* pFunc, char* pErrBuf, int32_t l
     return TSDB_CODE_SUCCESS;
   }
 
-  SNode* pPara = nodesListGetNode(pFunc->pParameterList, 0);
+  pFunc->node.resType = ((SExprNode*)nodesListGetNode(pFunc->pParameterList, 0))->resType;
-  if (QUERY_NODE_COLUMN != nodeType(pPara)) {
-    return buildFuncErrMsg(pErrBuf, len, TSDB_CODE_FUNC_FUNTION_ERROR,
-                           "The parameters of first/last can only be columns");
-  }
-
-  pFunc->node.resType = ((SExprNode*)pPara)->resType;
   return TSDB_CODE_SUCCESS;
 }
 
@@ -1046,11 +1055,6 @@ static int32_t translateFirstLastImpl(SFunctionNode* pFunc, char* pErrBuf, int32
   uint8_t paraType = ((SExprNode*)pPara)->resType.type;
   int32_t paraBytes = ((SExprNode*)pPara)->resType.bytes;
   if (isPartial) {
-    if (QUERY_NODE_COLUMN != nodeType(pPara)) {
-      return buildFuncErrMsg(pErrBuf, len, TSDB_CODE_FUNC_FUNTION_ERROR,
-                             "The parameters of first/last can only be columns");
-    }
-
     pFunc->node.resType =
         (SDataType){.bytes = getFirstLastInfoSize(paraBytes) + VARSTR_HEADER_SIZE, .type = TSDB_DATA_TYPE_BINARY};
   } else {
@@ -1472,7 +1476,7 @@ static int32_t translateToUnixtimestamp(SFunctionNode* pFunc, char* pErrBuf, int
     return invaildFuncParaTypeErrMsg(pErrBuf, len, pFunc->functionName);
   }
 
-  //add database precision as param
+  // add database precision as param
   uint8_t dbPrec = pFunc->node.resType.precision;
   addDbPrecisonParam(&pFunc->pParameterList, dbPrec);
 
@@ -1492,7 +1496,7 @@ static int32_t translateTimeTruncate(SFunctionNode* pFunc, char* pErrBuf, int32_
     return invaildFuncParaTypeErrMsg(pErrBuf, len, pFunc->functionName);
   }
 
-  //add database precision as param
+  // add database precision as param
   uint8_t dbPrec = pFunc->node.resType.precision;
   addDbPrecisonParam(&pFunc->pParameterList, dbPrec);
 
@@ -1520,7 +1524,7 @@ static int32_t translateTimeDiff(SFunctionNode* pFunc, char* pErrBuf, int32_t le
     }
   }
 
-  //add database precision as param
+  // add database precision as param
   uint8_t dbPrec = pFunc->node.resType.precision;
   addDbPrecisonParam(&pFunc->pParameterList, dbPrec);
 
@@ -1538,7 +1542,7 @@ static int32_t translateToJson(SFunctionNode* pFunc, char* pErrBuf, int32_t len)
     return invaildFuncParaTypeErrMsg(pErrBuf, len, pFunc->functionName);
   }
 
-  pFunc->node.resType = (SDataType){.bytes = tDataTypes[TSDB_DATA_TYPE_BINARY].bytes, .type = TSDB_DATA_TYPE_BINARY};
+  pFunc->node.resType = (SDataType){.bytes = tDataTypes[TSDB_DATA_TYPE_JSON].bytes, .type = TSDB_DATA_TYPE_JSON};
   return TSDB_CODE_SUCCESS;
 }
 
@@ -1919,7 +1923,7 @@ const SBuiltinFuncDefinition funcMgtBuiltins[] = {
   {
     .name = "interp",
     .type = FUNCTION_TYPE_INTERP,
-    .classification = FUNC_MGT_TIMELINE_FUNC | FUNC_MGT_INTERVAL_INTERPO_FUNC,
+    .classification = FUNC_MGT_TIMELINE_FUNC | FUNC_MGT_INTERVAL_INTERPO_FUNC | FUNC_MGT_IMPLICIT_TS_FUNC,
     .translateFunc = translateFirstLast,
     .getEnvFunc    = getSelectivityFuncEnv,
     .initFunc      = functionSetup,
@@ -1929,7 +1933,7 @@ const SBuiltinFuncDefinition funcMgtBuiltins[] = {
   {
     .name = "derivative",
     .type = FUNCTION_TYPE_DERIVATIVE,
-    .classification = FUNC_MGT_INDEFINITE_ROWS_FUNC | FUNC_MGT_TIMELINE_FUNC,
+    .classification = FUNC_MGT_INDEFINITE_ROWS_FUNC | FUNC_MGT_TIMELINE_FUNC | FUNC_MGT_IMPLICIT_TS_FUNC,
     .translateFunc = translateDerivative,
     .getEnvFunc   = getDerivativeFuncEnv,
     .initFunc     = derivativeFuncSetup,
@@ -1939,7 +1943,7 @@ const SBuiltinFuncDefinition funcMgtBuiltins[] = {
   {
     .name = "irate",
     .type = FUNCTION_TYPE_IRATE,
-    .classification = FUNC_MGT_AGG_FUNC | FUNC_MGT_TIMELINE_FUNC,
+    .classification = FUNC_MGT_AGG_FUNC | FUNC_MGT_TIMELINE_FUNC | FUNC_MGT_IMPLICIT_TS_FUNC,
     .translateFunc = translateIrate,
     .getEnvFunc   = getIrateFuncEnv,
     .initFunc     = irateFuncSetup,
@@ -1949,20 +1953,17 @@ const SBuiltinFuncDefinition funcMgtBuiltins[] = {
   {
     .name = "last_row",
     .type = FUNCTION_TYPE_LAST_ROW,
-    .classification = FUNC_MGT_AGG_FUNC | FUNC_MGT_MULTI_RES_FUNC | FUNC_MGT_TIMELINE_FUNC,
+    .classification = FUNC_MGT_AGG_FUNC | FUNC_MGT_SELECT_FUNC | FUNC_MGT_MULTI_RES_FUNC | FUNC_MGT_TIMELINE_FUNC | FUNC_MGT_IMPLICIT_TS_FUNC,
     .translateFunc = translateFirstLast,
     .getEnvFunc   = getFirstLastFuncEnv,
     .initFunc     = functionSetup,
-    .processFunc  = lastFunction,
+    .processFunc  = lastRowFunction,
-    .finalizeFunc = firstLastFinalize,
+    .finalizeFunc = lastRowFinalize,
-    .pPartialFunc = "_last_partial",
-    .pMergeFunc   = "_last_merge",
-    .combineFunc  = lastCombine,
   },
   {
     .name = "_cache_last_row",
     .type = FUNCTION_TYPE_CACHE_LAST_ROW,
-    .classification = FUNC_MGT_AGG_FUNC | FUNC_MGT_MULTI_RES_FUNC | FUNC_MGT_TIMELINE_FUNC,
+    .classification = FUNC_MGT_AGG_FUNC | FUNC_MGT_MULTI_RES_FUNC | FUNC_MGT_TIMELINE_FUNC | FUNC_MGT_IMPLICIT_TS_FUNC,
     .translateFunc = translateLastRow,
     .getEnvFunc   = getMinmaxFuncEnv,
     .initFunc     = minmaxFunctionSetup,
@@ -1972,7 +1973,7 @@ const SBuiltinFuncDefinition funcMgtBuiltins[] = {
   {
     .name = "first",
     .type = FUNCTION_TYPE_FIRST,
-    .classification = FUNC_MGT_AGG_FUNC | FUNC_MGT_SELECT_FUNC | FUNC_MGT_MULTI_RES_FUNC | FUNC_MGT_TIMELINE_FUNC,
+    .classification = FUNC_MGT_AGG_FUNC | FUNC_MGT_SELECT_FUNC | FUNC_MGT_MULTI_RES_FUNC | FUNC_MGT_TIMELINE_FUNC | FUNC_MGT_IMPLICIT_TS_FUNC,
     .translateFunc = translateFirstLast,
     .getEnvFunc   = getFirstLastFuncEnv,
     .initFunc     = functionSetup,
@@ -1985,7 +1986,7 @@ const SBuiltinFuncDefinition funcMgtBuiltins[] = {
   {
     .name = "_first_partial",
     .type = FUNCTION_TYPE_FIRST_PARTIAL,
-    .classification = FUNC_MGT_AGG_FUNC | FUNC_MGT_SELECT_FUNC | FUNC_MGT_MULTI_RES_FUNC | FUNC_MGT_TIMELINE_FUNC,
+    .classification = FUNC_MGT_AGG_FUNC | FUNC_MGT_SELECT_FUNC | FUNC_MGT_MULTI_RES_FUNC | FUNC_MGT_TIMELINE_FUNC | FUNC_MGT_IMPLICIT_TS_FUNC,
     .translateFunc = translateFirstLastPartial,
     .getEnvFunc   = getFirstLastFuncEnv,
     .initFunc     = functionSetup,
@@ -1996,7 +1997,7 @@ const SBuiltinFuncDefinition funcMgtBuiltins[] = {
   {
     .name = "_first_merge",
     .type = FUNCTION_TYPE_FIRST_MERGE,
-    .classification = FUNC_MGT_AGG_FUNC | FUNC_MGT_SELECT_FUNC | FUNC_MGT_MULTI_RES_FUNC | FUNC_MGT_TIMELINE_FUNC,
+    .classification = FUNC_MGT_AGG_FUNC | FUNC_MGT_SELECT_FUNC | FUNC_MGT_MULTI_RES_FUNC | FUNC_MGT_TIMELINE_FUNC | FUNC_MGT_IMPLICIT_TS_FUNC,
     .translateFunc = translateFirstLastMerge,
     .getEnvFunc   = getFirstLastFuncEnv,
     .initFunc     = functionSetup,
@@ -2007,7 +2008,7 @@ const SBuiltinFuncDefinition funcMgtBuiltins[] = {
   {
     .name = "last",
     .type = FUNCTION_TYPE_LAST,
-    .classification = FUNC_MGT_AGG_FUNC | FUNC_MGT_SELECT_FUNC | FUNC_MGT_MULTI_RES_FUNC | FUNC_MGT_TIMELINE_FUNC,
+    .classification = FUNC_MGT_AGG_FUNC | FUNC_MGT_SELECT_FUNC | FUNC_MGT_MULTI_RES_FUNC | FUNC_MGT_TIMELINE_FUNC | FUNC_MGT_IMPLICIT_TS_FUNC,
     .translateFunc = translateFirstLast,
     .getEnvFunc   = getFirstLastFuncEnv,
     .initFunc     = functionSetup,
@@ -2020,7 +2021,7 @@ const SBuiltinFuncDefinition funcMgtBuiltins[] = {
   {
     .name = "_last_partial",
     .type = FUNCTION_TYPE_LAST_PARTIAL,
-    .classification = FUNC_MGT_AGG_FUNC | FUNC_MGT_SELECT_FUNC | FUNC_MGT_MULTI_RES_FUNC | FUNC_MGT_TIMELINE_FUNC,
+    .classification = FUNC_MGT_AGG_FUNC | FUNC_MGT_SELECT_FUNC | FUNC_MGT_MULTI_RES_FUNC | FUNC_MGT_TIMELINE_FUNC | FUNC_MGT_IMPLICIT_TS_FUNC,
     .translateFunc = translateFirstLastPartial,
     .getEnvFunc   = getFirstLastFuncEnv,
     .initFunc     = functionSetup,
@@ -2031,7 +2032,7 @@ const SBuiltinFuncDefinition funcMgtBuiltins[] = {
   {
     .name = "_last_merge",
     .type = FUNCTION_TYPE_LAST_MERGE,
-    .classification = FUNC_MGT_AGG_FUNC | FUNC_MGT_SELECT_FUNC | FUNC_MGT_MULTI_RES_FUNC | FUNC_MGT_TIMELINE_FUNC,
+    .classification = FUNC_MGT_AGG_FUNC | FUNC_MGT_SELECT_FUNC | FUNC_MGT_MULTI_RES_FUNC | FUNC_MGT_TIMELINE_FUNC | FUNC_MGT_IMPLICIT_TS_FUNC,
     .translateFunc = translateFirstLastMerge,
     .getEnvFunc   = getFirstLastFuncEnv,
     .initFunc     = functionSetup,
@@ -2042,7 +2043,7 @@ const SBuiltinFuncDefinition funcMgtBuiltins[] = {
   {
     .name = "twa",
     .type = FUNCTION_TYPE_TWA,
-    .classification = FUNC_MGT_AGG_FUNC | FUNC_MGT_TIMELINE_FUNC | FUNC_MGT_INTERVAL_INTERPO_FUNC | FUNC_MGT_FORBID_STREAM_FUNC,
+    .classification = FUNC_MGT_AGG_FUNC | FUNC_MGT_TIMELINE_FUNC | FUNC_MGT_INTERVAL_INTERPO_FUNC | FUNC_MGT_FORBID_STREAM_FUNC | FUNC_MGT_IMPLICIT_TS_FUNC,
     .translateFunc = translateInNumOutDou,
     .dataRequiredFunc = statisDataRequired,
     .getEnvFunc    = getTwaFuncEnv,
@@ -2139,7 +2140,7 @@ const SBuiltinFuncDefinition funcMgtBuiltins[] = {
   {
     .name = "statecount",
     .type = FUNCTION_TYPE_STATE_COUNT,
-    .classification = FUNC_MGT_INDEFINITE_ROWS_FUNC | FUNC_MGT_FORBID_STREAM_FUNC | FUNC_MGT_FORBID_WINDOW_FUNC,
+    .classification = FUNC_MGT_INDEFINITE_ROWS_FUNC | FUNC_MGT_TIMELINE_FUNC | FUNC_MGT_FORBID_STREAM_FUNC | FUNC_MGT_FORBID_WINDOW_FUNC,
     .translateFunc = translateStateCount,
     .getEnvFunc   = getStateFuncEnv,
     .initFunc     = functionSetup,
@@ -2149,7 +2150,7 @@ const SBuiltinFuncDefinition funcMgtBuiltins[] = {
   {
     .name = "stateduration",
     .type = FUNCTION_TYPE_STATE_DURATION,
-    .classification = FUNC_MGT_INDEFINITE_ROWS_FUNC | FUNC_MGT_TIMELINE_FUNC | FUNC_MGT_FORBID_STREAM_FUNC | FUNC_MGT_FORBID_WINDOW_FUNC,
+    .classification = FUNC_MGT_INDEFINITE_ROWS_FUNC | FUNC_MGT_TIMELINE_FUNC | FUNC_MGT_IMPLICIT_TS_FUNC | FUNC_MGT_FORBID_STREAM_FUNC | FUNC_MGT_FORBID_WINDOW_FUNC,
     .translateFunc = translateStateDuration,
     .getEnvFunc   = getStateFuncEnv,
     .initFunc     = functionSetup,
@@ -2189,7 +2190,8 @@ const SBuiltinFuncDefinition funcMgtBuiltins[] = {
   {
     .name = "tail",
     .type = FUNCTION_TYPE_TAIL,
-    .classification = FUNC_MGT_INDEFINITE_ROWS_FUNC | FUNC_MGT_TIMELINE_FUNC | FUNC_MGT_FORBID_STREAM_FUNC | FUNC_MGT_FORBID_WINDOW_FUNC | FUNC_MGT_FORBID_GROUP_BY_FUNC,
+    .classification = FUNC_MGT_SELECT_FUNC | FUNC_MGT_INDEFINITE_ROWS_FUNC | FUNC_MGT_TIMELINE_FUNC | FUNC_MGT_FORBID_STREAM_FUNC | 
+                      FUNC_MGT_FORBID_WINDOW_FUNC | FUNC_MGT_FORBID_GROUP_BY_FUNC | FUNC_MGT_IMPLICIT_TS_FUNC,
     .translateFunc = translateTail,
     .getEnvFunc   = getTailFuncEnv,
     .initFunc     = tailFunctionSetup,
@@ -2200,7 +2202,7 @@ const SBuiltinFuncDefinition funcMgtBuiltins[] = {
     .name = "unique",
     .type = FUNCTION_TYPE_UNIQUE,
     .classification = FUNC_MGT_SELECT_FUNC | FUNC_MGT_INDEFINITE_ROWS_FUNC | FUNC_MGT_TIMELINE_FUNC | 
-                      FUNC_MGT_FORBID_STREAM_FUNC | FUNC_MGT_FORBID_WINDOW_FUNC | FUNC_MGT_FORBID_GROUP_BY_FUNC,
+                      FUNC_MGT_FORBID_STREAM_FUNC | FUNC_MGT_FORBID_WINDOW_FUNC | FUNC_MGT_FORBID_GROUP_BY_FUNC | FUNC_MGT_IMPLICIT_TS_FUNC,
     .translateFunc = translateUnique,
     .getEnvFunc   = getUniqueFuncEnv,
     .initFunc     = uniqueFunctionSetup,
@@ -2618,7 +2620,7 @@ const SBuiltinFuncDefinition funcMgtBuiltins[] = {
   {
     .name = "_group_key",
     .type = FUNCTION_TYPE_GROUP_KEY,
-    .classification = FUNC_MGT_AGG_FUNC,
+    .classification = FUNC_MGT_AGG_FUNC | FUNC_MGT_SELECT_FUNC,
     .translateFunc = translateGroupKey,
     .getEnvFunc   = getGroupKeyFuncEnv,
     .initFunc     = functionSetup,

source/libs/function/src/builtinsimpl.c

@@ -16,6 +16,7 @@
 #include "builtinsimpl.h"
 #include "cJSON.h"
 #include "function.h"
+#include "query.h"
 #include "querynodes.h"
 #include "taggfunction.h"
 #include "tcompare.h"
@@ -80,6 +81,7 @@ typedef struct STopBotRes {
 
 typedef struct SFirstLastRes {
   bool    hasResult;
+  bool    isNull;  // used for last_row function only
   int32_t bytes;
   char    buf[];
 } SFirstLastRes;
@@ -208,13 +210,13 @@ typedef struct SMavgInfo {
 } SMavgInfo;
 
 typedef struct SSampleInfo {
-  int32_t     samples;
+  int32_t    samples;
-  int32_t     totalPoints;
+  int32_t    totalPoints;
-  int32_t     numSampled;
+  int32_t    numSampled;
-  uint8_t     colType;
+  uint8_t    colType;
-  int16_t     colBytes;
+  int16_t    colBytes;
-  char*       data;
+  char*      data;
-  STuplePos*  tuplePos;
+  STuplePos* tuplePos;
 } SSampleInfo;
 
 typedef struct STailItem {
@@ -269,20 +271,19 @@ typedef struct SDerivInfo {
 } SDerivInfo;
 
 typedef struct SRateInfo {
-  double  firstValue;
+  double firstValue;
-  TSKEY   firstKey;
+  TSKEY  firstKey;
-  double  lastValue;
+  double lastValue;
-  TSKEY   lastKey;
+  TSKEY  lastKey;
-  int8_t  hasResult;  // flag to denote has value
+  int8_t hasResult;  // flag to denote has value
 } SRateInfo;
 
-typedef struct SGroupKeyInfo{
+typedef struct SGroupKeyInfo {
-  bool  hasResult;
+  bool hasResult;
-  bool  isNull;
+  bool isNull;
-  char  data[];
+  char data[];
 } SGroupKeyInfo;
 
-
 #define SET_VAL(_info, numOfElem, res) \
   do {                                 \
     if ((numOfElem) <= 0) {            \
@@ -1471,8 +1472,8 @@ void setSelectivityValue(SqlFunctionCtx* pCtx, SSDataBlock* pBlock, const STuple
   int32_t pageId = pTuplePos->pageId;
   int32_t offset = pTuplePos->offset;
 
-  if (pTuplePos->pageId != -1) {
+  if (pTuplePos->pageId != -1 && pCtx->subsidiaries.num > 0) {
-    int32_t numOfCols = taosArrayGetSize(pCtx->pSrcBlock->pDataBlock);
+    int32_t    numOfCols = pCtx->subsidiaries.num;
     SFilePage* pPage = getBufPage(pCtx->pBuf, pageId);
 
     bool* nullList = (bool*)((char*)pPage + offset);
@@ -1483,22 +1484,21 @@ void setSelectivityValue(SqlFunctionCtx* pCtx, SSDataBlock* pBlock, const STuple
       SqlFunctionCtx* pc = pCtx->subsidiaries.pCtx[j];
 
       SFunctParam* pFuncParam = &pc->pExpr->base.pParam[0];
-      int32_t      srcSlotId = pFuncParam->pCol->slotId;
       int32_t      dstSlotId = pc->pExpr->base.resSchema.slotId;
 
       int32_t ps = 0;
-      for (int32_t k = 0; k < srcSlotId; ++k) {
-        SColumnInfoData* pSrcCol = taosArrayGet(pCtx->pSrcBlock->pDataBlock, k);
-        ps += pSrcCol->info.bytes;
-      }
 
       SColumnInfoData* pDstCol = taosArrayGet(pBlock->pDataBlock, dstSlotId);
-      if (nullList[srcSlotId]) {
+      ASSERT(pc->pExpr->base.resSchema.bytes == pDstCol->info.bytes);
+      if (nullList[j]) {
         colDataAppendNULL(pDstCol, rowIndex);
       } else {
-        colDataAppend(pDstCol, rowIndex, (pStart + ps), false);
+        colDataAppend(pDstCol, rowIndex, pStart, false);
       }
+      pStart += pDstCol->info.bytes;
     }
+
+    releaseBufPage(pCtx->pBuf, pPage);
   }
 }
 
@@ -2409,7 +2409,9 @@ int32_t apercentileCombine(SqlFunctionCtx* pDestCtx, SqlFunctionCtx* pSourceCtx)
   return TSDB_CODE_SUCCESS;
 }
 
-int32_t getFirstLastInfoSize(int32_t resBytes) { return sizeof(SFirstLastRes) + resBytes + sizeof(int64_t) + sizeof(STuplePos); }
+int32_t getFirstLastInfoSize(int32_t resBytes) {
+  return sizeof(SFirstLastRes) + resBytes + sizeof(int64_t) + sizeof(STuplePos);
+}
 
 bool getFirstLastFuncEnv(SFunctionNode* pFunc, SFuncExecEnv* pEnv) {
   SColumnNode* pNode = (SColumnNode*)nodesListGetNode(pFunc->pParameterList, 0);
@@ -2491,7 +2493,7 @@ int32_t firstFunction(SqlFunctionCtx* pCtx) {
         }
         memcpy(pInfo->buf, data, bytes);
         *(TSKEY*)(pInfo->buf + bytes) = cts;
-        //handle selectivity
+        // handle selectivity
         if (pCtx->subsidiaries.num > 0) {
           STuplePos* pTuplePos = (STuplePos*)(pInfo->buf + bytes + sizeof(TSKEY));
           if (!pInfo->hasResult) {
@@ -2501,7 +2503,7 @@ int32_t firstFunction(SqlFunctionCtx* pCtx) {
           }
         }
         pInfo->hasResult = true;
-        //DO_UPDATE_TAG_COLUMNS(pCtx, ts);
+        // DO_UPDATE_TAG_COLUMNS(pCtx, ts);
         pResInfo->numOfRes = 1;
         break;
       }
@@ -2533,7 +2535,7 @@ int32_t firstFunction(SqlFunctionCtx* pCtx) {
         }
         memcpy(pInfo->buf, data, bytes);
         *(TSKEY*)(pInfo->buf + bytes) = cts;
-        //handle selectivity
+        // handle selectivity
         if (pCtx->subsidiaries.num > 0) {
           STuplePos* pTuplePos = (STuplePos*)(pInfo->buf + bytes + sizeof(TSKEY));
           if (!pInfo->hasResult) {
@@ -2543,7 +2545,7 @@ int32_t firstFunction(SqlFunctionCtx* pCtx) {
           }
         }
         pInfo->hasResult = true;
-        //DO_UPDATE_TAG_COLUMNS(pCtx, ts);
+        // DO_UPDATE_TAG_COLUMNS(pCtx, ts);
         pResInfo->numOfRes = 1;
         break;
       }
@@ -2597,7 +2599,7 @@ int32_t lastFunction(SqlFunctionCtx* pCtx) {
         }
         memcpy(pInfo->buf, data, bytes);
         *(TSKEY*)(pInfo->buf + bytes) = cts;
-        //handle selectivity
+        // handle selectivity
         if (pCtx->subsidiaries.num > 0) {
           STuplePos* pTuplePos = (STuplePos*)(pInfo->buf + bytes + sizeof(TSKEY));
           if (!pInfo->hasResult) {
@@ -2607,7 +2609,7 @@ int32_t lastFunction(SqlFunctionCtx* pCtx) {
           }
         }
         pInfo->hasResult = true;
-        //DO_UPDATE_TAG_COLUMNS(pCtx, ts);
+        // DO_UPDATE_TAG_COLUMNS(pCtx, ts);
         pResInfo->numOfRes = 1;
       }
       break;
@@ -2629,7 +2631,7 @@ int32_t lastFunction(SqlFunctionCtx* pCtx) {
         }
         memcpy(pInfo->buf, data, bytes);
         *(TSKEY*)(pInfo->buf + bytes) = cts;
-        //handle selectivity
+        // handle selectivity
         if (pCtx->subsidiaries.num > 0) {
           STuplePos* pTuplePos = (STuplePos*)(pInfo->buf + bytes + sizeof(TSKEY));
           if (!pInfo->hasResult) {
@@ -2640,7 +2642,7 @@ int32_t lastFunction(SqlFunctionCtx* pCtx) {
         }
         pInfo->hasResult = true;
         pResInfo->numOfRes = 1;
-        //DO_UPDATE_TAG_COLUMNS(pCtx, ts);
+        // DO_UPDATE_TAG_COLUMNS(pCtx, ts);
       }
       break;
     }
@@ -2652,7 +2654,7 @@ int32_t lastFunction(SqlFunctionCtx* pCtx) {
 
 static void firstLastTransferInfo(SqlFunctionCtx* pCtx, SFirstLastRes* pInput, SFirstLastRes* pOutput, bool isFirst) {
   SInputColumnInfoData* pColInfo = &pCtx->input;
-  int32_t start = pColInfo->startRowIndex;
+  int32_t               start = pColInfo->startRowIndex;
 
   pOutput->bytes = pInput->bytes;
   TSKEY* tsIn = (TSKEY*)(pInput->buf + pInput->bytes);
@@ -2670,7 +2672,7 @@ static void firstLastTransferInfo(SqlFunctionCtx* pCtx, SFirstLastRes* pInput, S
   }
   *tsOut = *tsIn;
   memcpy(pOutput->buf, pInput->buf, pOutput->bytes);
-  //handle selectivity
+  // handle selectivity
   STuplePos* pTuplePos = (STuplePos*)(pOutput->buf + pOutput->bytes + sizeof(TSKEY));
   if (pCtx->subsidiaries.num > 0) {
     if (!pOutput->hasResult) {
@@ -2717,7 +2719,7 @@ int32_t firstLastFinalize(SqlFunctionCtx* pCtx, SSDataBlock* pBlock) {
 
   SFirstLastRes* pRes = GET_ROWCELL_INTERBUF(pResInfo);
   colDataAppend(pCol, pBlock->info.rows, pRes->buf, pResInfo->isNullRes);
-  //handle selectivity
+  // handle selectivity
   STuplePos* pTuplePos = (STuplePos*)(pRes->buf + pRes->bytes + sizeof(TSKEY));
   setSelectivityValue(pCtx, pBlock, pTuplePos, pBlock->info.rows);
 
@@ -2728,8 +2730,8 @@ int32_t firstLastPartialFinalize(SqlFunctionCtx* pCtx, SSDataBlock* pBlock) {
   SResultRowEntryInfo* pEntryInfo = GET_RES_INFO(pCtx);
   SFirstLastRes*       pRes = GET_ROWCELL_INTERBUF(GET_RES_INFO(pCtx));
 
-  int32_t              resultBytes = getFirstLastInfoSize(pRes->bytes);
+  int32_t resultBytes = getFirstLastInfoSize(pRes->bytes);
-  char*                res = taosMemoryCalloc(resultBytes + VARSTR_HEADER_SIZE, sizeof(char));
+  char*   res = taosMemoryCalloc(resultBytes + VARSTR_HEADER_SIZE, sizeof(char));
 
   memcpy(varDataVal(res), pRes, resultBytes);
   varDataSetLen(res, resultBytes);
@@ -2738,7 +2740,7 @@ int32_t firstLastPartialFinalize(SqlFunctionCtx* pCtx, SSDataBlock* pBlock) {
   SColumnInfoData* pCol = taosArrayGet(pBlock->pDataBlock, slotId);
 
   colDataAppend(pCol, pBlock->info.rows, res, false);
-  //handle selectivity
+  // handle selectivity
   STuplePos* pTuplePos = (STuplePos*)(pRes->buf + pRes->bytes + sizeof(TSKEY));
   setSelectivityValue(pCtx, pBlock, pTuplePos, pBlock->info.rows);
 
@@ -2763,6 +2765,111 @@ int32_t lastCombine(SqlFunctionCtx* pDestCtx, SqlFunctionCtx* pSourceCtx) {
   return TSDB_CODE_SUCCESS;
 }
 
+int32_t lastRowFunction(SqlFunctionCtx* pCtx) {
+  int32_t numOfElems = 0;
+
+  SResultRowEntryInfo* pResInfo = GET_RES_INFO(pCtx);
+  SFirstLastRes*       pInfo = GET_ROWCELL_INTERBUF(pResInfo);
+
+  SInputColumnInfoData* pInput = &pCtx->input;
+  SColumnInfoData*      pInputCol = pInput->pData[0];
+
+  int32_t type = pInputCol->info.type;
+  int32_t bytes = pInputCol->info.bytes;
+  pInfo->bytes = bytes;
+
+  SColumnDataAgg* pColAgg = (pInput->colDataAggIsSet) ? pInput->pColumnDataAgg[0] : NULL;
+
+  TSKEY startKey = getRowPTs(pInput->pPTS, 0);
+  TSKEY endKey = getRowPTs(pInput->pPTS, pInput->totalRows - 1);
+
+  int32_t blockDataOrder = (startKey <= endKey) ? TSDB_ORDER_ASC : TSDB_ORDER_DESC;
+
+  if (blockDataOrder == TSDB_ORDER_ASC) {
+    for (int32_t i = pInput->numOfRows + pInput->startRowIndex - 1; i >= pInput->startRowIndex; --i) {
+      char* data = colDataGetData(pInputCol, i);
+      TSKEY cts = getRowPTs(pInput->pPTS, i);
+      if (pResInfo->numOfRes == 0 || *(TSKEY*)(pInfo->buf) < cts) {
+        if (pInputCol->hasNull && colDataIsNull(pInputCol, pInput->totalRows, i, pColAgg)) {
+          pInfo->isNull = true;
+        } else {
+          pInfo->isNull = false;
+          if (IS_VAR_DATA_TYPE(type)) {
+            bytes = varDataTLen(data);
+            pInfo->bytes = bytes;
+          }
+          memcpy(pInfo->buf + sizeof(TSKEY), data, bytes);
+        }
+        *(TSKEY*)(pInfo->buf) = cts;
+        numOfElems++;
+        // handle selectivity
+        if (pCtx->subsidiaries.num > 0) {
+          STuplePos* pTuplePos = (STuplePos*)(pInfo->buf + bytes + sizeof(TSKEY));
+          if (!pInfo->hasResult) {
+            saveTupleData(pCtx, i, pCtx->pSrcBlock, pTuplePos);
+          } else {
+            copyTupleData(pCtx, i, pCtx->pSrcBlock, pTuplePos);
+          }
+        }
+        pInfo->hasResult = true;
+        // DO_UPDATE_TAG_COLUMNS(pCtx, ts);
+        pResInfo->numOfRes = 1;
+      }
+      break;
+    }
+  } else {  // descending order
+    for (int32_t i = pInput->startRowIndex; i < pInput->numOfRows + pInput->startRowIndex; ++i) {
+      char* data = colDataGetData(pInputCol, i);
+      TSKEY cts = getRowPTs(pInput->pPTS, i);
+      if (pResInfo->numOfRes == 0 || *(TSKEY*)(pInfo->buf) < cts) {
+        if (pInputCol->hasNull && colDataIsNull(pInputCol, pInput->totalRows, i, pColAgg)) {
+          pInfo->isNull = true;
+        } else {
+          pInfo->isNull = false;
+          if (IS_VAR_DATA_TYPE(type)) {
+            bytes = varDataTLen(data);
+            pInfo->bytes = bytes;
+          }
+          memcpy(pInfo->buf + sizeof(TSKEY), data, bytes);
+        }
+        *(TSKEY*)(pInfo->buf) = cts;
+        numOfElems++;
+        // handle selectivity
+        if (pCtx->subsidiaries.num > 0) {
+          STuplePos* pTuplePos = (STuplePos*)(pInfo->buf + bytes + sizeof(TSKEY));
+          if (!pInfo->hasResult) {
+            saveTupleData(pCtx, i, pCtx->pSrcBlock, pTuplePos);
+          } else {
+            copyTupleData(pCtx, i, pCtx->pSrcBlock, pTuplePos);
+          }
+        }
+        pInfo->hasResult = true;
+        pResInfo->numOfRes = 1;
+        // DO_UPDATE_TAG_COLUMNS(pCtx, ts);
+      }
+      break;
+    }
+  }
+
+  SET_VAL(pResInfo, numOfElems, 1);
+  return TSDB_CODE_SUCCESS;
+}
+
+int32_t lastRowFinalize(SqlFunctionCtx* pCtx, SSDataBlock* pBlock) {
+  int32_t          slotId = pCtx->pExpr->base.resSchema.slotId;
+  SColumnInfoData* pCol = taosArrayGet(pBlock->pDataBlock, slotId);
+
+  SResultRowEntryInfo* pResInfo = GET_RES_INFO(pCtx);
+
+  SFirstLastRes* pRes = GET_ROWCELL_INTERBUF(pResInfo);
+  colDataAppend(pCol, pBlock->info.rows, pRes->buf + sizeof(TSKEY), pRes->isNull);
+  // handle selectivity
+  STuplePos* pTuplePos = (STuplePos*)(pRes->buf + pRes->bytes + sizeof(TSKEY));
+  setSelectivityValue(pCtx, pBlock, pTuplePos, pBlock->info.rows);
+
+  return pResInfo->numOfRes;
+}
+
 bool getDiffFuncEnv(SFunctionNode* UNUSED_PARAM(pFunc), SFuncExecEnv* pEnv) {
   pEnv->calcMemSize = sizeof(SDiffInfo);
   return true;
@@ -2776,7 +2883,11 @@ bool diffFunctionSetup(SqlFunctionCtx* pCtx, SResultRowEntryInfo* pResInfo) {
   SDiffInfo* pDiffInfo = GET_ROWCELL_INTERBUF(pResInfo);
   pDiffInfo->hasPrev = false;
   pDiffInfo->prev.i64 = 0;
-  pDiffInfo->ignoreNegative = pCtx->param[1].param.i;  // TODO set correct param
+  if (pCtx->numOfParams > 1) {
+    pDiffInfo->ignoreNegative = pCtx->param[1].param.i;  // TODO set correct param
+  } else {
+    pDiffInfo->ignoreNegative = false;
+  }
   pDiffInfo->includeNull = false;
   pDiffInfo->firstOutput = false;
   return true;
@@ -2891,10 +3002,8 @@ int32_t diffFunction(SqlFunctionCtx* pCtx) {
   SInputColumnInfoData* pInput = &pCtx->input;
 
   SColumnInfoData* pInputCol = pInput->pData[0];
-  SColumnInfoData* pTsOutput = pCtx->pTsOutput;
 
   int32_t numOfElems = 0;
-  TSKEY*  tsList = (int64_t*)pInput->pPTS->pData;
   int32_t startOffset = pCtx->offset;
 
   SColumnInfoData* pOutput = (SColumnInfoData*)pCtx->pOutput;
@@ -2906,9 +3015,6 @@ int32_t diffFunction(SqlFunctionCtx* pCtx) {
       if (colDataIsNull_f(pInputCol->nullbitmap, i)) {
         if (pDiffInfo->includeNull) {
           colDataSetNull_f(pOutput->nullbitmap, pos);
-          if (tsList != NULL) {
-            colDataAppendInt64(pTsOutput, pos, &tsList[i]);
-          }
 
           numOfElems += 1;
         }
@@ -2919,9 +3025,6 @@ int32_t diffFunction(SqlFunctionCtx* pCtx) {
 
       if (pDiffInfo->hasPrev) {
         doHandleDiff(pDiffInfo, pInputCol->info.type, pv, pOutput, pos, pCtx->order);
-        if (pTsOutput != NULL) {
-          colDataAppendInt64(pTsOutput, pos, &tsList[i]);
-        }
 
         numOfElems++;
       } else {
@@ -2937,9 +3040,6 @@ int32_t diffFunction(SqlFunctionCtx* pCtx) {
       if (colDataIsNull_f(pInputCol->nullbitmap, i)) {
         if (pDiffInfo->includeNull) {
           colDataSetNull_f(pOutput->nullbitmap, pos);
-          if (tsList != NULL) {
-            colDataAppendInt64(pTsOutput, pos, &tsList[i]);
-          }
 
           numOfElems += 1;
         }
@@ -2951,9 +3051,6 @@ int32_t diffFunction(SqlFunctionCtx* pCtx) {
       // there is a row of previous data block to be handled in the first place.
       if (pDiffInfo->hasPrev) {
         doHandleDiff(pDiffInfo, pInputCol->info.type, pv, pOutput, pos, pCtx->order);
-        if (pTsOutput != NULL) {
-          colDataAppendInt64(pTsOutput, pos, &pDiffInfo->prevTs);
-        }
 
         numOfElems++;
       } else {
@@ -2961,9 +3058,6 @@ int32_t diffFunction(SqlFunctionCtx* pCtx) {
       }
 
       pDiffInfo->hasPrev = true;
-      if (pTsOutput != NULL) {
-        pDiffInfo->prevTs = tsList[i];
-      }
     }
   }
 
@@ -3100,7 +3194,10 @@ void doAddIntoResult(SqlFunctionCtx* pCtx, void* pData, int32_t rowIndex, SSData
     if (pCtx->subsidiaries.num > 0) {
       saveTupleData(pCtx, rowIndex, pSrcBlock, &pItem->tuplePos);
     }
-
+#ifdef BUF_PAGE_DEBUG
+    qDebug("page_saveTuple i:%d, item:%p,pageId:%d, offset:%d\n", pEntryInfo->numOfRes, pItem, pItem->tuplePos.pageId,
+           pItem->tuplePos.offset);
+#endif
     // allocate the buffer and keep the data of this row into the new allocated buffer
     pEntryInfo->numOfRes++;
     taosheapsort((void*)pItems, sizeof(STopBotResItem), pEntryInfo->numOfRes, (const void*)&type, topBotResComparFn,
@@ -3121,7 +3218,9 @@ void doAddIntoResult(SqlFunctionCtx* pCtx, void* pData, int32_t rowIndex, SSData
       if (pCtx->subsidiaries.num > 0) {
         copyTupleData(pCtx, rowIndex, pSrcBlock, &pItem->tuplePos);
       }
-
+#ifdef BUF_PAGE_DEBUG
+      qDebug("page_copyTuple pageId:%d, offset:%d", pItem->tuplePos.pageId, pItem->tuplePos.offset);
+#endif
       taosheapadjust((void*)pItems, sizeof(STopBotResItem), 0, pEntryInfo->numOfRes - 1, (const void*)&type,
                      topBotResComparFn, NULL, !isTopQuery);
     }
@@ -3131,7 +3230,11 @@ void doAddIntoResult(SqlFunctionCtx* pCtx, void* pData, int32_t rowIndex, SSData
 void saveTupleData(SqlFunctionCtx* pCtx, int32_t rowIndex, const SSDataBlock* pSrcBlock, STuplePos* pPos) {
   SFilePage* pPage = NULL;
 
-  int32_t completeRowSize = pSrcBlock->info.rowSize + (int32_t) taosArrayGetSize(pSrcBlock->pDataBlock) * sizeof(bool);
+  int32_t completeRowSize = pCtx->subsidiaries.num * sizeof(bool);
+  for (int32_t j = 0; j < pCtx->subsidiaries.num; ++j) {
+    SqlFunctionCtx* pc = pCtx->subsidiaries.pCtx[j];
+    completeRowSize += pc->pExpr->base.resSchema.bytes;
+  }
 
   if (pCtx->curBufPage == -1) {
     pPage = getNewBufPage(pCtx->pBuf, 0, &pCtx->curBufPage);
@@ -3149,19 +3252,22 @@ void saveTupleData(SqlFunctionCtx* pCtx, int32_t rowIndex, const SSDataBlock* pS
   // keep the current row data, extract method
   int32_t offset = 0;
   bool*   nullList = (bool*)((char*)pPage + pPage->num);
-  char*   pStart = (char*)(nullList + sizeof(bool) * (int32_t) taosArrayGetSize(pSrcBlock->pDataBlock));
+  char*   pStart = (char*)(nullList + sizeof(bool) * pCtx->subsidiaries.num);
-  for (int32_t i = 0; i < (int32_t) taosArrayGetSize(pSrcBlock->pDataBlock); ++i) {
+  for (int32_t i = 0; i < pCtx->subsidiaries.num; ++i) {
-    SColumnInfoData* pCol = taosArrayGet(pSrcBlock->pDataBlock, i);
+    SqlFunctionCtx* pc = pCtx->subsidiaries.pCtx[i];
-    bool             isNull = colDataIsNull_s(pCol, rowIndex);
+
-    if (isNull) {
+    SFunctParam* pFuncParam = &pc->pExpr->base.pParam[0];
-      nullList[i] = true;
+    int32_t      srcSlotId = pFuncParam->pCol->slotId;
+
+    SColumnInfoData* pCol = taosArrayGet(pSrcBlock->pDataBlock, srcSlotId);
+    if ((nullList[i] = colDataIsNull_s(pCol, rowIndex)) == true) {
       offset += pCol->info.bytes;
       continue;
     }
 
     char* p = colDataGetData(pCol, rowIndex);
     if (IS_VAR_DATA_TYPE(pCol->info.type)) {
-      memcpy(pStart + offset, p, varDataTLen(p));
+      memcpy(pStart + offset, p, (pCol->info.type == TSDB_DATA_TYPE_JSON) ? getJsonValueLen(p) : varDataTLen(p));
     } else {
       memcpy(pStart + offset, p, pCol->info.bytes);
     }
@@ -3179,14 +3285,18 @@ void saveTupleData(SqlFunctionCtx* pCtx, int32_t rowIndex, const SSDataBlock* pS
 void copyTupleData(SqlFunctionCtx* pCtx, int32_t rowIndex, const SSDataBlock* pSrcBlock, STuplePos* pPos) {
   SFilePage* pPage = getBufPage(pCtx->pBuf, pPos->pageId);
 
-  int32_t numOfCols = taosArrayGetSize(pSrcBlock->pDataBlock);
+  int32_t numOfCols = pCtx->subsidiaries.num;
 
   bool* nullList = (bool*)((char*)pPage + pPos->offset);
   char* pStart = (char*)(nullList + numOfCols * sizeof(bool));
 
   int32_t offset = 0;
   for (int32_t i = 0; i < numOfCols; ++i) {
-    SColumnInfoData* pCol = taosArrayGet(pSrcBlock->pDataBlock, i);
+    SqlFunctionCtx* pc = pCtx->subsidiaries.pCtx[i];
+    SFunctParam*    pFuncParam = &pc->pExpr->base.pParam[0];
+    int32_t         srcSlotId = pFuncParam->pCol->slotId;
+
+    SColumnInfoData* pCol = taosArrayGet(pSrcBlock->pDataBlock, srcSlotId);
     if ((nullList[i] = colDataIsNull_s(pCol, rowIndex)) == true) {
       offset += pCol->info.bytes;
       continue;
@@ -3194,7 +3304,7 @@ void copyTupleData(SqlFunctionCtx* pCtx, int32_t rowIndex, const SSDataBlock* pS
 
     char* p = colDataGetData(pCol, rowIndex);
     if (IS_VAR_DATA_TYPE(pCol->info.type)) {
-      memcpy(pStart + offset, p, varDataTLen(p));
+      memcpy(pStart + offset, p, (pCol->info.type == TSDB_DATA_TYPE_JSON) ? getJsonValueLen(p) : varDataTLen(p));
     } else {
       memcpy(pStart + offset, p, pCol->info.bytes);
     }
@@ -3208,7 +3318,7 @@ void copyTupleData(SqlFunctionCtx* pCtx, int32_t rowIndex, const SSDataBlock* pS
 
 int32_t topBotFinalize(SqlFunctionCtx* pCtx, SSDataBlock* pBlock) {
   SResultRowEntryInfo* pEntryInfo = GET_RES_INFO(pCtx);
-  STopBotRes*          pRes = GET_ROWCELL_INTERBUF(pEntryInfo);
+  STopBotRes*          pRes = getTopBotOutputInfo(pCtx);
 
   int16_t type = pCtx->input.pData[0]->info.type;
   int32_t slotId = pCtx->pExpr->base.resSchema.slotId;
@@ -3225,7 +3335,10 @@ int32_t topBotFinalize(SqlFunctionCtx* pCtx, SSDataBlock* pBlock) {
     } else {
       colDataAppend(pCol, currentRow, (const char*)&pItem->v.i, false);
     }
-
+#ifdef BUF_PAGE_DEBUG
+    qDebug("page_finalize i:%d,item:%p,pageId:%d, offset:%d\n", i, pItem, pItem->tuplePos.pageId,
+           pItem->tuplePos.offset);
+#endif
     setSelectivityValue(pCtx, pBlock, &pRes->pItems[i].tuplePos, currentRow);
     currentRow += 1;
   }
@@ -3464,7 +3577,7 @@ bool elapsedFunctionSetup(SqlFunctionCtx* pCtx, SResultRowEntryInfo* pResultInfo
   pInfo->min = MAX_TS_KEY;
   pInfo->max = 0;
 
-  if (pCtx->numOfParams > 2) {
+  if (pCtx->numOfParams > 1) {
     pInfo->timeUnit = pCtx->param[1].param.i;
   } else {
     pInfo->timeUnit = 1;
@@ -4405,7 +4518,6 @@ int32_t mavgFunction(SqlFunctionCtx* pCtx) {
   SMavgInfo*           pInfo = GET_ROWCELL_INTERBUF(pResInfo);
 
   SInputColumnInfoData* pInput = &pCtx->input;
-  TSKEY*                tsList = (int64_t*)pInput->pPTS->pData;
 
   SColumnInfoData* pInputCol = pInput->pData[0];
   SColumnInfoData* pTsOutput = pCtx->pTsOutput;
@@ -4445,10 +4557,6 @@ int32_t mavgFunction(SqlFunctionCtx* pCtx) {
         colDataAppend(pOutput, pos, (char*)&result, false);
       }
 
-      // TODO: remove this after pTsOutput is handled
-      if (pTsOutput != NULL) {
-        colDataAppendInt64(pTsOutput, pos, &tsList[i]);
-      }
       numOfElems++;
     }
 
@@ -4795,11 +4903,11 @@ static void doModeAdd(SModeInfo* pInfo, char* data, bool isNull) {
     return;
   }
 
-  int32_t      hashKeyBytes = IS_VAR_DATA_TYPE(pInfo->colType) ? varDataTLen(data) : pInfo->colBytes;
+  int32_t     hashKeyBytes = IS_VAR_DATA_TYPE(pInfo->colType) ? varDataTLen(data) : pInfo->colBytes;
-  SModeItem**     pHashItem = taosHashGet(pInfo->pHash, data, hashKeyBytes);
+  SModeItem** pHashItem = taosHashGet(pInfo->pHash, data, hashKeyBytes);
   if (pHashItem == NULL) {
-    int32_t      size = sizeof(SModeItem) + pInfo->colBytes;
+    int32_t    size = sizeof(SModeItem) + pInfo->colBytes;
-    SModeItem* pItem  = (SModeItem*)(pInfo->pItems + pInfo->numOfPoints * size);
+    SModeItem* pItem = (SModeItem*)(pInfo->pItems + pInfo->numOfPoints * size);
     memcpy(pItem->data, data, pInfo->colBytes);
     pItem->count += 1;
 
@@ -4812,7 +4920,7 @@ static void doModeAdd(SModeInfo* pInfo, char* data, bool isNull) {
 
 int32_t modeFunction(SqlFunctionCtx* pCtx) {
   SResultRowEntryInfo* pResInfo = GET_RES_INFO(pCtx);
-  SModeInfo*              pInfo = GET_ROWCELL_INTERBUF(pResInfo);
+  SModeInfo*           pInfo = GET_ROWCELL_INTERBUF(pResInfo);
 
   SInputColumnInfoData* pInput = &pCtx->input;
 
@@ -4860,7 +4968,6 @@ int32_t modeFinalize(SqlFunctionCtx* pCtx, SSDataBlock* pBlock) {
   return pResInfo->numOfRes;
 }
 
-
 bool getTwaFuncEnv(struct SFunctionNode* pFunc, SFuncExecEnv* pEnv) {
   pEnv->calcMemSize = sizeof(STwaInfo);
   return true;
@@ -5112,7 +5219,7 @@ int32_t twaFinalize(struct SqlFunctionCtx* pCtx, SSDataBlock* pBlock) {
   return functionFinalize(pCtx, pBlock);
 }
 
-bool blockDistSetup(SqlFunctionCtx *pCtx, SResultRowEntryInfo* pResultInfo) {
+bool blockDistSetup(SqlFunctionCtx* pCtx, SResultRowEntryInfo* pResultInfo) {
   if (!functionSetup(pCtx, pResultInfo)) {
     return false;
   }
@@ -5144,7 +5251,7 @@ int32_t blockDistFunction(SqlFunctionCtx* pCtx) {
 
   pDistInfo->defMinRows = p1.defMinRows;
   pDistInfo->defMaxRows = p1.defMaxRows;
-  pDistInfo->rowSize    = p1.rowSize;
+  pDistInfo->rowSize = p1.rowSize;
   pDistInfo->numOfSmallBlocks = p1.numOfSmallBlocks;
 
   if (pDistInfo->minRows > p1.minRows) {
@@ -5230,16 +5337,18 @@ int32_t blockDistFinalize(SqlFunctionCtx* pCtx, SSDataBlock* pBlock) {
   int32_t row = 0;
   char    st[256] = {0};
   double  totalRawSize = pData->totalRows * pData->rowSize;
-  int32_t len =
+  int32_t len = sprintf(st + VARSTR_HEADER_SIZE,
-      sprintf(st + VARSTR_HEADER_SIZE, "Total_Blocks=[%d] Total_Size=[%.2f Kb] Average_size=[%.2f Kb] Compression_Ratio=[%.2f %c]",
+                        "Total_Blocks=[%d] Total_Size=[%.2f Kb] Average_size=[%.2f Kb] Compression_Ratio=[%.2f %c]",
-              pData->numOfBlocks, pData->totalSize / 1024.0, ((double)pData->totalSize) / pData->numOfBlocks,
+                        pData->numOfBlocks, pData->totalSize / 1024.0, ((double)pData->totalSize) / pData->numOfBlocks,
-              pData->totalSize * 100 / totalRawSize, '%');
+                        pData->totalSize * 100 / totalRawSize, '%');
 
   varDataSetLen(st, len);
   colDataAppend(pColInfo, row++, st, false);
 
-  len = sprintf(st + VARSTR_HEADER_SIZE, "Total_Rows=[%"PRId64"] Inmem_Rows=[%d] MinRows=[%d] MaxRows=[%d] Average_Rows=[%"PRId64"]",
+  len = sprintf(st + VARSTR_HEADER_SIZE,
-                pData->totalRows, pData->numOfInmemRows, pData->minRows, pData->maxRows, pData->totalRows / pData->numOfBlocks);
+                "Total_Rows=[%" PRId64 "] Inmem_Rows=[%d] MinRows=[%d] MaxRows=[%d] Average_Rows=[%" PRId64 "]",
+                pData->totalRows, pData->numOfInmemRows, pData->minRows, pData->maxRows,
+                pData->totalRows / pData->numOfBlocks);
 
   varDataSetLen(st, len);
   colDataAppend(pColInfo, row++, st, false);
@@ -5415,25 +5524,25 @@ bool irateFuncSetup(SqlFunctionCtx* pCtx, SResultRowEntryInfo* pResInfo) {
 
   SRateInfo* pInfo = GET_ROWCELL_INTERBUF(pResInfo);
 
-  pInfo->firstKey   = INT64_MIN;
+  pInfo->firstKey = INT64_MIN;
-  pInfo->lastKey    = INT64_MIN;
+  pInfo->lastKey = INT64_MIN;
   pInfo->firstValue = (double)INT64_MIN;
-  pInfo->lastValue  = (double)INT64_MIN;
+  pInfo->lastValue = (double)INT64_MIN;
 
-  pInfo->hasResult  = 0;
+  pInfo->hasResult = 0;
   return true;
 }
 
 int32_t irateFunction(SqlFunctionCtx* pCtx) {
   SResultRowEntryInfo* pResInfo = GET_RES_INFO(pCtx);
-  SRateInfo*          pRateInfo = GET_ROWCELL_INTERBUF(pResInfo);
+  SRateInfo*           pRateInfo = GET_ROWCELL_INTERBUF(pResInfo);
 
   SInputColumnInfoData* pInput = &pCtx->input;
   SColumnInfoData*      pInputCol = pInput->pData[0];
 
   SColumnInfoData* pOutput = (SColumnInfoData*)pCtx->pOutput;
 
-  TSKEY*  tsList = (int64_t*)pInput->pPTS->pData;
+  TSKEY* tsList = (int64_t*)pInput->pPTS->pData;
 
   int32_t numOfElems = 0;
   int32_t type = pInputCol->info.type;
@@ -5445,13 +5554,13 @@ int32_t irateFunction(SqlFunctionCtx* pCtx) {
 
     numOfElems++;
 
-    char* data = colDataGetData(pInputCol, i);
+    char*  data = colDataGetData(pInputCol, i);
-    double  v = 0;
+    double v = 0;
     GET_TYPED_DATA(v, double, type, data);
 
     if (INT64_MIN == pRateInfo->lastKey) {
       pRateInfo->lastValue = v;
-      pRateInfo->lastKey   = tsList[i];
+      pRateInfo->lastKey = tsList[i];
       continue;
     }
 
@@ -5462,7 +5571,7 @@ int32_t irateFunction(SqlFunctionCtx* pCtx) {
       }
 
       pRateInfo->lastValue = v;
-      pRateInfo->lastKey   = tsList[i];
+      pRateInfo->lastKey = tsList[i];
 
       continue;
     }
@@ -5471,7 +5580,6 @@ int32_t irateFunction(SqlFunctionCtx* pCtx) {
       pRateInfo->firstValue = v;
       pRateInfo->firstKey = tsList[i];
     }
-
   }
 
   SET_VAL(pResInfo, numOfElems, 1);
@@ -5497,7 +5605,7 @@ static double doCalcRate(const SRateInfo* pRateInfo, double tickPerSec) {
     return 0;
   }
 
-  return (duration > 0)? ((double)diff) / (duration/tickPerSec):0.0;
+  return (duration > 0) ? ((double)diff) / (duration / tickPerSec) : 0.0;
 }
 
 int32_t irateFinalize(SqlFunctionCtx* pCtx, SSDataBlock* pBlock) {
@@ -5508,7 +5616,7 @@ int32_t irateFinalize(SqlFunctionCtx* pCtx, SSDataBlock* pBlock) {
   pResInfo->isNullRes = (pResInfo->numOfRes == 0) ? 1 : 0;
 
   SRateInfo* pInfo = GET_ROWCELL_INTERBUF(pResInfo);
-  double result = doCalcRate(pInfo, 1000);
+  double     result = doCalcRate(pInfo, (double)TSDB_TICK_PER_SECOND(pCtx->param[1].param.i));
   colDataAppend(pCol, pBlock->info.rows, (const char*)&result, pResInfo->isNullRes);
 
   return pResInfo->numOfRes;
@@ -5516,16 +5624,14 @@ int32_t irateFinalize(SqlFunctionCtx* pCtx, SSDataBlock* pBlock) {
 
 int32_t groupKeyFunction(SqlFunctionCtx* pCtx) {
   SResultRowEntryInfo* pResInfo = GET_RES_INFO(pCtx);
-  SGroupKeyInfo* pInfo = GET_ROWCELL_INTERBUF(pResInfo);
+  SGroupKeyInfo*       pInfo = GET_ROWCELL_INTERBUF(pResInfo);
 
   SInputColumnInfoData* pInput = &pCtx->input;
-  SColumnInfoData*   pInputCol = pInput->pData[0];
+  SColumnInfoData*      pInputCol = pInput->pData[0];
-
-  int32_t bytes = pInputCol->info.bytes;
 
   int32_t startIndex = pInput->startRowIndex;
 
-  //escape rest of data blocks to avoid first entry to be overwritten.
+  // escape rest of data blocks to avoid first entry to be overwritten.
   if (pInfo->hasResult) {
     goto _group_key_over;
   }
@@ -5537,7 +5643,12 @@ int32_t groupKeyFunction(SqlFunctionCtx* pCtx) {
   }
 
   char* data = colDataGetData(pInputCol, startIndex);
-  memcpy(pInfo->data, data, bytes);
+  if (IS_VAR_DATA_TYPE(pInputCol->info.type)) {
+    memcpy(pInfo->data, data,
+           (pInputCol->info.type == TSDB_DATA_TYPE_JSON) ? getJsonValueLen(data) : varDataTLen(data));
+  } else {
+    memcpy(pInfo->data, data, pInputCol->info.bytes);
+  }
   pInfo->hasResult = true;
 
 _group_key_over:

source/libs/function/src/functionMgt.c

@@ -181,6 +181,8 @@ bool fmIsForbidGroupByFunc(int32_t funcId) { return isSpecificClassifyFunc(funcI
 
 bool fmIsSystemInfoFunc(int32_t funcId) { return isSpecificClassifyFunc(funcId, FUNC_MGT_SYSTEM_INFO_FUNC); }
 
+bool fmIsImplicitTsFunc(int32_t funcId) { return isSpecificClassifyFunc(funcId, FUNC_MGT_IMPLICIT_TS_FUNC); }
+
 bool fmIsInterpFunc(int32_t funcId) {
   if (funcId < 0 || funcId >= funcMgtBuiltinsNum) {
     return false;

source/libs/function/src/taggfunction.c

@@ -1887,10 +1887,10 @@ static void top_bottom_func_finalizer(SqlFunctionCtx *pCtx) {
   // user specify the order of output by sort the result according to timestamp
   if (pCtx->param[1].param.i == PRIMARYKEY_TIMESTAMP_COL_ID) {
     __compar_fn_t comparator = (pCtx->param[2].param.i == TSDB_ORDER_ASC) ? resAscComparFn : resDescComparFn;
-    qsort(tvp, (size_t)pResInfo->numOfRes, POINTER_BYTES, comparator);
+    taosSort(tvp, (size_t)pResInfo->numOfRes, POINTER_BYTES, comparator);
   } else /*if (pCtx->param[1].param.i > PRIMARYKEY_TIMESTAMP_COL_ID)*/ {
     __compar_fn_t comparator = (pCtx->param[2].param.i == TSDB_ORDER_ASC) ? resDataAscComparFn : resDataDescComparFn;
-    qsort(tvp, (size_t)pResInfo->numOfRes, POINTER_BYTES, comparator);
+    taosSort(tvp, (size_t)pResInfo->numOfRes, POINTER_BYTES, comparator);
   }
   
   GET_TRUE_DATA_TYPE();

source/libs/function/src/tpercentile.c

@@ -44,7 +44,7 @@ static SFilePage *loadDataFromFilePage(tMemBucket *pMemBucket, int32_t slotIdx)
     offset += (int32_t)(pg->num * pMemBucket->bytes);
   }
 
-  qsort(buffer->data, pMemBucket->pSlots[slotIdx].info.size, pMemBucket->bytes, pMemBucket->comparFn);
+  taosSort(buffer->data, pMemBucket->pSlots[slotIdx].info.size, pMemBucket->bytes, pMemBucket->comparFn);
   return buffer;
 }