Merge branch '3.0' of https://github.com/taosdata/TDengine into feature/3.0_mhli

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.define

@@ -98,12 +98,12 @@ ELSE ()
     ENDIF ()
 
     IF (${SANITIZER} MATCHES "true")
-        SET(CMAKE_C_FLAGS "${CMAKE_C_FLAGS}     -Werror -Werror=return-type -fPIC -gdwarf-2 -fsanitize=address -fsanitize=undefined -fsanitize-recover=all -fsanitize=float-divide-by-zero -fsanitize=float-cast-overflow -fno-sanitize=shift-base -fno-sanitize=alignment -g3")
+        SET(CMAKE_C_FLAGS "${CMAKE_C_FLAGS}     -Werror -Werror=return-type -fPIC -gdwarf-2 -fsanitize=address -fsanitize=undefined -fsanitize-recover=all -fsanitize=float-divide-by-zero -fsanitize=float-cast-overflow -fno-sanitize=shift-base -fno-sanitize=alignment -g3 -Wformat=0")
-        SET(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Werror -Wno-literal-suffix -Werror=return-type -fPIC -gdwarf-2 -fsanitize=address -fsanitize=undefined -fsanitize-recover=all -fsanitize=float-divide-by-zero -fsanitize=float-cast-overflow -fno-sanitize=shift-base -fno-sanitize=alignment -g3")
+        SET(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Werror -Wno-literal-suffix -Werror=return-type -fPIC -gdwarf-2 -fsanitize=address -fsanitize=undefined -fsanitize-recover=all -fsanitize=float-divide-by-zero -fsanitize=float-cast-overflow -fno-sanitize=shift-base -fno-sanitize=alignment -g3 -Wformat=0")
 	MESSAGE(STATUS "Will compile with Address Sanitizer!")
     ELSE ()
-        SET(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -Werror -Werror=return-type -fPIC -gdwarf-2 -g3")
+        SET(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -Werror -Werror=return-type -fPIC -gdwarf-2 -g3 -Wformat=0")
-        SET(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Werror -Wno-literal-suffix -Werror=return-type -fPIC -gdwarf-2 -g3")
+        SET(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Werror -Wno-literal-suffix -Werror=return-type -fPIC -gdwarf-2 -g3 -Wformat=0")
     ENDIF ()
 
     MESSAGE("System processor ID: ${CMAKE_SYSTEM_PROCESSOR}")

include/common/tcommon.h

@@ -56,7 +56,8 @@ typedef enum EStreamType {
   STREAM_CLEAR,
   STREAM_INVALID,
   STREAM_GET_ALL,
-  STREAM_DELETE,
+  STREAM_DELETE_RESULT,
+  STREAM_DELETE_DATA,
   STREAM_RETRIEVE,
   STREAM_PULL_DATA,
   STREAM_PULL_OVER,

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);

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/plannodes.h

@@ -104,6 +104,7 @@ typedef struct SIndefRowsFuncLogicNode {
   SNodeList* pFuncs;
   bool       isTailFunc;
   bool       isUniqueFunc;
+  bool       isTimeLineFunc;
 } SIndefRowsFuncLogicNode;
 
 typedef struct SInterpFuncLogicNode {

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/qcom/query.h

@@ -65,7 +65,7 @@ typedef struct SQueryExecRes {
 } SQueryExecRes;
 
 typedef struct SIndexMeta {
-#ifdef WINDOWS
+#if defined(WINDOWS) || defined(_TD_DARWIN_64)
   size_t avoidCompilationErrors;
 #endif
 
@@ -249,13 +249,6 @@ extern int32_t (*queryProcessMsgRsp[TDMT_MAX])(void* output, char* msg, int32_t
   ((_code) == TSDB_CODE_RPC_REDIRECT || (_code) == TSDB_CODE_NODE_NOT_DEPLOYED ||                 \
    (_code) == TSDB_CODE_SYN_NOT_LEADER || (_code) == TSDB_CODE_APP_NOT_READY)
 
-#define NEED_SCHEDULER_RETRY_ERROR(_code)                                                         \
-  (NEED_SCHEDULER_REDIRECT_ERROR(_code) || (_code) == TSDB_CODE_RPC_NETWORK_UNAVAIL ||            \
-   (_code) == TSDB_CODE_SCH_TIMEOUT_ERROR || (_code) == TSDB_CODE_RPC_BROKEN_LINK)
-  
-
-
-
 #define REQUEST_TOTAL_EXEC_TIMES 2
 
 #define qFatal(...)                                                                           \

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/os.h

@@ -41,7 +41,6 @@ extern "C" {
 #include <sys/types.h>
 #include <termios.h>
 #include <sys/statvfs.h>
-#include <sys/prctl.h>
 #include <sys/shm.h>
 #include <sys/wait.h>
 

include/os/osAtomic.h

@@ -63,7 +63,7 @@ int8_t  atomic_add_fetch_8(int8_t volatile *ptr, int8_t val);
 int16_t atomic_add_fetch_16(int16_t volatile *ptr, int16_t val);
 int32_t atomic_add_fetch_32(int32_t volatile *ptr, int32_t val);
 int64_t atomic_add_fetch_64(int64_t volatile *ptr, int64_t val);
-void   *atomic_add_fetch_ptr(void *ptr, void *val);
+void   *atomic_add_fetch_ptr(void *ptr, int64_t val);
 int8_t  atomic_fetch_add_8(int8_t volatile *ptr, int8_t val);
 int16_t atomic_fetch_add_16(int16_t volatile *ptr, int16_t val);
 int32_t atomic_fetch_add_32(int32_t volatile *ptr, int32_t val);
@@ -73,7 +73,7 @@ int8_t  atomic_sub_fetch_8(int8_t volatile *ptr, int8_t val);
 int16_t atomic_sub_fetch_16(int16_t volatile *ptr, int16_t val);
 int32_t atomic_sub_fetch_32(int32_t volatile *ptr, int32_t val);
 int64_t atomic_sub_fetch_64(int64_t volatile *ptr, int64_t val);
-void   *atomic_sub_fetch_ptr(void *ptr, void *val);
+void   *atomic_sub_fetch_ptr(void *ptr, int64_t val);
 int8_t  atomic_fetch_sub_8(int8_t volatile *ptr, int8_t val);
 int16_t atomic_fetch_sub_16(int16_t volatile *ptr, int16_t val);
 int32_t atomic_fetch_sub_32(int32_t volatile *ptr, int32_t val);

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

@@ -20,6 +20,12 @@
 extern "C" {
 #endif
 
+// 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 qsort  QSORT_FUNC_TAOS_FORBID
+#endif
+
 #define TPOW2(x) ((x) * (x))
 #define TABS(x) ((x) > 0 ? (x) : -(x))
 
@@ -60,6 +66,13 @@ 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/os/osSemaphore.h

@@ -24,7 +24,9 @@ extern "C" {
 
 #if defined(_TD_DARWIN_64)
 
-typedef struct tsem_s *tsem_t;
+// typedef struct tsem_s *tsem_t;
+typedef struct bosal_sem_t *tsem_t;
+
 
 int tsem_init(tsem_t *sem, int pshared, unsigned int value);
 int tsem_wait(tsem_t *sem);
@@ -51,11 +53,11 @@ int tsem_timewait(tsem_t *sim, int64_t nanosecs);
 //  #define taosThreadRwlockRdlock(lock) taosThreadMutexLock(lock)
 //  #define taosThreadRwlockUnlock(lock) taosThreadMutexUnlock(lock)
 
-#define TdThreadSpinlock               TdThreadMutex
+// #define TdThreadSpinlock               TdThreadMutex
-#define taosThreadSpinInit(lock, NULL) taosThreadMutexInit(lock, NULL)
+// #define taosThreadSpinInit(lock, NULL) taosThreadMutexInit(lock, NULL)
-#define taosThreadSpinDestroy(lock)    taosThreadMutexDestroy(lock)
+// #define taosThreadSpinDestroy(lock)    taosThreadMutexDestroy(lock)
-#define taosThreadSpinLock(lock)       taosThreadMutexLock(lock)
+// #define taosThreadSpinLock(lock)       taosThreadMutexLock(lock)
-#define taosThreadSpinUnlock(lock)     taosThreadMutexUnlock(lock)
+// #define taosThreadSpinUnlock(lock)     taosThreadMutexUnlock(lock)
 #endif
 
 bool    taosCheckPthreadValid(TdThread thread);

include/os/osSocket.h

@@ -64,7 +64,6 @@
 #include <osEok.h>
 #else
 #include <netinet/in.h>
-#include <sys/epoll.h>
 #endif
 #endif
 
@@ -77,15 +76,12 @@ typedef int socklen_t;
 #define TAOS_EPOLL_WAIT_TIME 100
 typedef SOCKET eventfd_t;
 #define eventfd(a, b)      -1
-#define EpollClose(pollFd) epoll_close(pollFd)
 #ifndef EPOLLWAKEUP
 #define EPOLLWAKEUP (1u << 29)
 #endif
 #elif defined(_TD_DARWIN_64)
 #define TAOS_EPOLL_WAIT_TIME 500
 typedef int32_t SOCKET;
-typedef SOCKET  EpollFd;
-#define EpollClose(pollFd)   epoll_close(pollFd)
 #else
 #define TAOS_EPOLL_WAIT_TIME 500
 typedef int32_t SOCKET;
@@ -122,14 +118,6 @@ typedef SOCKET  EpollFd;
 typedef int32_t  SocketFd;
 typedef SocketFd EpollFd;
 
-typedef struct TdSocket {
-#if SOCKET_WITH_LOCK
-  TdThreadRwlock rwlock;
-#endif
-  int      refId;
-  SocketFd fd;
-} * TdSocketPtr, TdSocket;
-
 typedef struct TdSocketServer *TdSocketServerPtr;
 typedef struct TdSocket *      TdSocketPtr;
 typedef struct TdEpoll *       TdEpollPtr;
@@ -181,11 +169,6 @@ void        taosSetMaskSIGPIPE();
 uint32_t    taosInetAddr(const char *ipAddr);
 const char *taosInetNtoa(struct in_addr ipInt);
 
-TdEpollPtr taosCreateEpoll(int32_t size);
-int32_t    taosCtlEpoll(TdEpollPtr pEpoll, int32_t epollOperate, TdSocketPtr pSocket, struct epoll_event *event);
-int32_t    taosWaitEpoll(TdEpollPtr pEpoll, struct epoll_event *event, int32_t maxEvents, int32_t timeout);
-int32_t    taosCloseEpoll(TdEpollPtr *ppEpoll);
-
 #ifdef __cplusplus
 }
 #endif

include/os/osString.h

@@ -67,7 +67,7 @@ bool    taosMbsToUcs4(const char *mbs, size_t mbs_len, TdUcs4 *ucs4, int32_t ucs
 int32_t tasoUcs4Compare(TdUcs4 *f1_ucs4, TdUcs4 *f2_ucs4, int32_t bytes);
 TdUcs4* tasoUcs4Copy(TdUcs4 *target_ucs4, TdUcs4 *source_ucs4, int32_t len_ucs4);
 bool    taosValidateEncodec(const char *encodec);
-int32_t taosHexEncode(const char *src, char *dst, int32_t len);
+int32_t taosHexEncode(const unsigned char *src, char *dst, int32_t len);
 int32_t taosHexDecode(const char *src, char *dst, int32_t len);
 
 int32_t taosWcharWidth(TdWchar wchar);

include/os/osThread.h

@@ -188,27 +188,27 @@ int32_t taosThreadJoin(TdThread thread, void **valuePtr);
 int32_t taosThreadKeyCreate(TdThreadKey * key, void(*destructor)(void *));
 int32_t taosThreadKeyDelete(TdThreadKey key);
 int32_t taosThreadKill(TdThread thread, int32_t sig);
-int32_t taosThreadMutexConsistent(TdThreadMutex* mutex);
+// int32_t taosThreadMutexConsistent(TdThreadMutex* mutex);
 int32_t taosThreadMutexDestroy(TdThreadMutex * mutex);
 int32_t taosThreadMutexInit(TdThreadMutex * mutex, const TdThreadMutexAttr * attr);
 int32_t taosThreadMutexLock(TdThreadMutex * mutex);
-int32_t taosThreadMutexTimedLock(TdThreadMutex * mutex, const struct timespec *abstime);
+// int32_t taosThreadMutexTimedLock(TdThreadMutex * mutex, const struct timespec *abstime);
 int32_t taosThreadMutexTryLock(TdThreadMutex * mutex);
 int32_t taosThreadMutexUnlock(TdThreadMutex * mutex);
 int32_t taosThreadMutexAttrDestroy(TdThreadMutexAttr * attr);
 int32_t taosThreadMutexAttrGetPshared(const TdThreadMutexAttr * attr, int32_t *pshared);
-int32_t taosThreadMutexAttrGetRobust(const TdThreadMutexAttr * attr, int32_t * robust);
+// int32_t taosThreadMutexAttrGetRobust(const TdThreadMutexAttr * attr, int32_t * robust);
 int32_t taosThreadMutexAttrGetType(const TdThreadMutexAttr * attr, int32_t *kind);
 int32_t taosThreadMutexAttrInit(TdThreadMutexAttr * attr);
 int32_t taosThreadMutexAttrSetPshared(TdThreadMutexAttr * attr, int32_t pshared);
-int32_t taosThreadMutexAttrSetRobust(TdThreadMutexAttr * attr, int32_t robust);
+// int32_t taosThreadMutexAttrSetRobust(TdThreadMutexAttr * attr, int32_t robust);
 int32_t taosThreadMutexAttrSetType(TdThreadMutexAttr * attr, int32_t kind);
 int32_t taosThreadOnce(TdThreadOnce * onceControl, void(*initRoutine)(void));
 int32_t taosThreadRwlockDestroy(TdThreadRwlock * rwlock);
 int32_t taosThreadRwlockInit(TdThreadRwlock * rwlock, const TdThreadRwlockAttr * attr);
 int32_t taosThreadRwlockRdlock(TdThreadRwlock * rwlock);
-int32_t taosThreadRwlockTimedRdlock(TdThreadRwlock * rwlock, const struct timespec *abstime);
+// int32_t taosThreadRwlockTimedRdlock(TdThreadRwlock * rwlock, const struct timespec *abstime);
-int32_t taosThreadRwlockTimedWrlock(TdThreadRwlock * rwlock, const struct timespec *abstime);
+// int32_t taosThreadRwlockTimedWrlock(TdThreadRwlock * rwlock, const struct timespec *abstime);
 int32_t taosThreadRwlockTryRdlock(TdThreadRwlock * rwlock);
 int32_t taosThreadRwlockTryWrlock(TdThreadRwlock * rwlock);
 int32_t taosThreadRwlockUnlock(TdThreadRwlock * rwlock);

include/util/taoserror.h

@@ -388,6 +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)
@@ -573,6 +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)
@@ -618,9 +623,12 @@ 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)
+#define TSDB_CODE_INDEX_INVALID_FILE             TAOS_DEF_ERROR_CODE(0, 0x3201)
+
 
 //tmq
 #define TSDB_CODE_TMQ_INVALID_MSG                TAOS_DEF_ERROR_CODE(0, 0x4000)

include/util/tdef.h

@@ -287,7 +287,7 @@ typedef enum ELogicConditionType {
 #define TSDB_MULTI_TABLEMETA_MAX_NUM 100000  // maximum batch size allowed to load table meta
 
 #define TSDB_MIN_VNODES_PER_DB          1
-#define TSDB_MAX_VNODES_PER_DB          4096
+#define TSDB_MAX_VNODES_PER_DB          1024
 #define TSDB_DEFAULT_VN_PER_DB          2
 #define TSDB_MIN_BUFFER_PER_VNODE       3      // unit MB
 #define TSDB_MAX_BUFFER_PER_VNODE       16384  // unit MB

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);
 }
 

include/util/types.h

@@ -83,6 +83,7 @@ typedef uint16_t VarDataLenT;  // maxVarDataLen: 32767
 
 #define varDataLen(v) ((VarDataLenT *)(v))[0]
 #define varDataVal(v) ((char *)(v) + VARSTR_HEADER_SIZE)
+#define varDataTLen(v) (sizeof(VarDataLenT) + varDataLen(v))
 
 #define NCHAR_WIDTH_TO_BYTES(n)  ((n) * TSDB_NCHAR_SIZE + VARSTR_HEADER_SIZE)
 

source/client/src/clientImpl.c

@@ -786,6 +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) {
@@ -797,6 +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);
 

source/client/src/tmq.c

@@ -1609,8 +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);*/

source/common/src/systable.c

@@ -108,13 +108,11 @@ static const SSysDbTableSchema userFuncSchema[] = {
 };
 
 static const SSysDbTableSchema userIdxSchema[] = {
+    {.name = "index_name", .bytes = SYSTABLE_SCH_TABLE_NAME_LEN, .type = TSDB_DATA_TYPE_VARCHAR},
     {.name = "db_name", .bytes = SYSTABLE_SCH_DB_NAME_LEN, .type = TSDB_DATA_TYPE_VARCHAR},
     {.name = "table_name", .bytes = SYSTABLE_SCH_TABLE_NAME_LEN, .type = TSDB_DATA_TYPE_VARCHAR},
-    {.name = "index_database", .bytes = SYSTABLE_SCH_DB_NAME_LEN, .type = TSDB_DATA_TYPE_VARCHAR},
+    {.name = "vgroup_id", .bytes = 4, .type = TSDB_DATA_TYPE_INT},
-    {.name = "index_name", .bytes = SYSTABLE_SCH_TABLE_NAME_LEN, .type = TSDB_DATA_TYPE_VARCHAR},
+    {.name = "create_time", .bytes = 8, .type = TSDB_DATA_TYPE_TIMESTAMP},
-    {.name = "column_name", .bytes = SYSTABLE_SCH_COL_NAME_LEN, .type = TSDB_DATA_TYPE_VARCHAR},
-    {.name = "index_type", .bytes = 10, .type = TSDB_DATA_TYPE_VARCHAR},
-    {.name = "index_extensions", .bytes = 256, .type = TSDB_DATA_TYPE_VARCHAR},
 };
 
 static const SSysDbTableSchema userStbsSchema[] = {
@@ -313,7 +311,7 @@ static const SSysDbTableSchema querySchema[] = {
     {.name = "query_id", .bytes = TSDB_QUERY_ID_LEN + VARSTR_HEADER_SIZE, .type = TSDB_DATA_TYPE_VARCHAR},
     {.name = "req_id", .bytes = 8, .type = TSDB_DATA_TYPE_UBIGINT},
     {.name = "connId", .bytes = 4, .type = TSDB_DATA_TYPE_UINT},
-    {.name = "app", .bytes = TSDB_APP_NAME_LEN + VARSTR_HEADER_SIZE, .type = TSDB_DATA_TYPE_VARCHAR},    
+    {.name = "app", .bytes = TSDB_APP_NAME_LEN + VARSTR_HEADER_SIZE, .type = TSDB_DATA_TYPE_VARCHAR},
     {.name = "pid", .bytes = 4, .type = TSDB_DATA_TYPE_INT},
     {.name = "user", .bytes = TSDB_USER_LEN + VARSTR_HEADER_SIZE, .type = TSDB_DATA_TYPE_VARCHAR},
     {.name = "end_point", .bytes = TSDB_IPv4ADDR_LEN + 6 + VARSTR_HEADER_SIZE, .type = TSDB_DATA_TYPE_VARCHAR},
@@ -329,17 +327,17 @@ static const SSysDbTableSchema appSchema[] = {
     {.name = "app_id", .bytes = 8, .type = TSDB_DATA_TYPE_UBIGINT},
     {.name = "ip", .bytes = TSDB_IPv4ADDR_LEN + VARSTR_HEADER_SIZE, .type = TSDB_DATA_TYPE_VARCHAR},
     {.name = "pid", .bytes = 4, .type = TSDB_DATA_TYPE_INT},
-    {.name = "name", .bytes = TSDB_APP_NAME_LEN + VARSTR_HEADER_SIZE, .type = TSDB_DATA_TYPE_VARCHAR},    
+    {.name = "name", .bytes = TSDB_APP_NAME_LEN + VARSTR_HEADER_SIZE, .type = TSDB_DATA_TYPE_VARCHAR},
-    {.name = "start_time", .bytes = 8 , .type = TSDB_DATA_TYPE_TIMESTAMP},
+    {.name = "start_time", .bytes = 8, .type = TSDB_DATA_TYPE_TIMESTAMP},
-    {.name = "insert_req", .bytes = 8 , .type = TSDB_DATA_TYPE_UBIGINT},
+    {.name = "insert_req", .bytes = 8, .type = TSDB_DATA_TYPE_UBIGINT},
-    {.name = "insert_row", .bytes = 8 , .type = TSDB_DATA_TYPE_UBIGINT},
+    {.name = "insert_row", .bytes = 8, .type = TSDB_DATA_TYPE_UBIGINT},
-    {.name = "insert_time", .bytes = 8 , .type = TSDB_DATA_TYPE_UBIGINT},
+    {.name = "insert_time", .bytes = 8, .type = TSDB_DATA_TYPE_UBIGINT},
-    {.name = "insert_bytes", .bytes = 8 , .type = TSDB_DATA_TYPE_UBIGINT},
+    {.name = "insert_bytes", .bytes = 8, .type = TSDB_DATA_TYPE_UBIGINT},
-    {.name = "fetch_bytes", .bytes = 8 , .type = TSDB_DATA_TYPE_UBIGINT},
+    {.name = "fetch_bytes", .bytes = 8, .type = TSDB_DATA_TYPE_UBIGINT},
-    {.name = "query_time", .bytes = 8 , .type = TSDB_DATA_TYPE_UBIGINT},
+    {.name = "query_time", .bytes = 8, .type = TSDB_DATA_TYPE_UBIGINT},
-    {.name = "show_query", .bytes = 8 , .type = TSDB_DATA_TYPE_UBIGINT},
+    {.name = "show_query", .bytes = 8, .type = TSDB_DATA_TYPE_UBIGINT},
-    {.name = "total_req", .bytes = 8 , .type = TSDB_DATA_TYPE_UBIGINT},
+    {.name = "total_req", .bytes = 8, .type = TSDB_DATA_TYPE_UBIGINT},
-    {.name = "current_req", .bytes = 8 , .type = TSDB_DATA_TYPE_UBIGINT},
+    {.name = "current_req", .bytes = 8, .type = TSDB_DATA_TYPE_UBIGINT},
     {.name = "last_access", .bytes = 8, .type = TSDB_DATA_TYPE_TIMESTAMP},
 };
 
@@ -353,8 +351,7 @@ static const SSysTableMeta perfsMeta[] = {
     {TSDB_PERFS_TABLE_TRANS, transSchema, tListLen(transSchema)},
     {TSDB_PERFS_TABLE_SMAS, smaSchema, tListLen(smaSchema)},
     {TSDB_PERFS_TABLE_STREAMS, streamSchema, tListLen(streamSchema)},
-    {TSDB_PERFS_TABLE_APPS, appSchema, tListLen(appSchema)}
+    {TSDB_PERFS_TABLE_APPS, appSchema, tListLen(appSchema)}};
-};
 
 void getInfosDbMeta(const SSysTableMeta** pInfosTableMeta, size_t* size) {
   *pInfosTableMeta = infosMeta;

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;
     }
 
@@ -1253,6 +1259,7 @@ int32_t copyDataBlock(SSDataBlock* dst, const SSDataBlock* src) {
 
   dst->info.rows = src->info.rows;
   dst->info.window = src->info.window;
+  dst->info.type = src->info.type;
   return TSDB_CODE_SUCCESS;
 }
 
@@ -1431,9 +1438,39 @@ 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 +1498,33 @@ 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 +1669,13 @@ 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) {

source/common/src/ttime.c

@@ -589,7 +589,7 @@ int32_t convertStringToTimestamp(int16_t type, char *inputData, int64_t timePrec
       return TSDB_CODE_FAILED;
     }
     newColData[len] = 0;
-    int32_t ret = taosParseTime(newColData, timeVal, len + 1, (int32_t)timePrec, tsDaylight);
+    int32_t ret = taosParseTime(newColData, timeVal, len, (int32_t)timePrec, tsDaylight);
     if (ret != TSDB_CODE_SUCCESS) {
       taosMemoryFree(newColData);
       return ret;

source/common/src/ttszip.c

@@ -845,11 +845,7 @@ int32_t tsBufMerge(STSBuf* pDestBuf, const STSBuf* pSrcBuf) {
 
   int64_t offset = getDataStartOffset();
   int32_t size = (int32_t)pSrcBuf->fileSize - (int32_t)offset;
-#if defined(_TD_DARWIN_64)
-  int64_t written = taosFSendFile(pDestBuf->pFile->fp, pSrcBuf->pFile->fp, &offset, size);
-#else
   int64_t written = taosFSendFile(pDestBuf->pFile, pSrcBuf->pFile, &offset, size);
-#endif
 
   if (written == -1 || written != size) {
     return -1;

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_vnode/src/vmInt.c

@@ -21,7 +21,7 @@ SVnodeObj *vmAcquireVnode(SVnodeMgmt *pMgmt, int32_t vgId) {
 
   taosThreadRwlockRdlock(&pMgmt->lock);
   taosHashGetDup(pMgmt->hash, &vgId, sizeof(int32_t), (void *)&pVnode);
-  if (pVnode == NULL) {
+  if (pVnode == NULL || pVnode->dropped) {
     terrno = TSDB_CODE_VND_INVALID_VGROUP_ID;
   } else {
     int32_t refCount = atomic_add_fetch_32(&pVnode->refCount, 1);
@@ -81,16 +81,18 @@ void vmCloseVnode(SVnodeMgmt *pMgmt, SVnodeObj *pVnode) {
 
   vmReleaseVnode(pMgmt, pVnode);
   while (pVnode->refCount > 0) taosMsleep(10);
+  dTrace("vgId:%d, wait for vnode queue is empty", pVnode->vgId);
+
   while (!taosQueueEmpty(pVnode->pWriteQ)) taosMsleep(10);
   while (!taosQueueEmpty(pVnode->pSyncQ)) taosMsleep(10);
   while (!taosQueueEmpty(pVnode->pApplyQ)) taosMsleep(10);
   while (!taosQueueEmpty(pVnode->pQueryQ)) taosMsleep(10);
   while (!taosQueueEmpty(pVnode->pFetchQ)) taosMsleep(10);
+  dTrace("vgId:%d, vnode-fetch queue is empty", pVnode->vgId);
 
   vmFreeQueue(pMgmt, pVnode);
   vnodeClose(pVnode->pImpl);
   pVnode->pImpl = NULL;
-
   dDebug("vgId:%d, vnode is closed", pVnode->vgId);
 
   if (pVnode->dropped) {

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

@@ -107,7 +107,7 @@ static void vmProcessSyncQueue(SQueueInfo *pInfo, STaosQall *qall, int32_t numOf
     const STraceId *trace = &pMsg->info.traceId;
     dGTrace("vgId:%d, msg:%p get from vnode-sync queue", pVnode->vgId, pMsg);
 
-    int32_t code = vnodeProcessSyncReq(pVnode->pImpl, pMsg, NULL); // no response here
+    int32_t code = vnodeProcessSyncReq(pVnode->pImpl, pMsg, NULL);  // no response here
     dGTrace("vgId:%d, msg:%p is freed, code:0x%x", pVnode->vgId, pMsg, code);
     rpcFreeCont(pMsg->pCont);
     taosFreeQitem(pMsg);
@@ -146,8 +146,8 @@ static int32_t vmPutMsgToQueue(SVnodeMgmt *pMgmt, SRpcMsg *pMsg, EQueueType qtyp
 
   SVnodeObj *pVnode = vmAcquireVnode(pMgmt, pHead->vgId);
   if (pVnode == NULL) {
-    dGError("vgId:%d, msg:%p failed to put into vnode queue since %s, type:%s", pHead->vgId, pMsg, terrstr(),
+    dGError("vgId:%d, msg:%p failed to put into vnode queue since %s, msgtype:%s qtype:%d", pHead->vgId, pMsg,
-            TMSG_INFO(pMsg->msgType));
+            terrstr(), TMSG_INFO(pMsg->msgType), qtype);
     return terrno != 0 ? terrno : -1;
   }
 

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/mndInt.h

@@ -87,7 +87,7 @@ typedef struct {
 } STelemMgmt;
 
 typedef struct {
-  sem_t    syncSem;
+  tsem_t    syncSem;
   int64_t  sync;
   bool     standby;
   SReplica replica;

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

@@ -38,6 +38,8 @@ int32_t mndPersistStream(SMnode *pMnode, STrans *pTrans, SStreamObj *pStream);
 int32_t mndDropStreamTasks(SMnode *pMnode, STrans *pTrans, SStreamObj *pStream);
 int32_t mndPersistDropStreamLog(SMnode *pMnode, STrans *pTrans, SStreamObj *pStream);
 
+int32_t mndDropStreamByDb(SMnode *pMnode, STrans *pTrans, SDbObj *pDb);
+
 #ifdef __cplusplus
 }
 #endif

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

@@ -21,6 +21,7 @@
 #include "mndShow.h"
 #include "mndSma.h"
 #include "mndStb.h"
+#include "mndStream.h"
 #include "mndSubscribe.h"
 #include "mndTopic.h"
 #include "mndTrans.h"
@@ -548,18 +549,33 @@ static int32_t mndSetDbCfgFromAlterDbReq(SDbObj *pDb, SAlterDbReq *pAlter) {
   terrno = TSDB_CODE_MND_DB_OPTION_UNCHANGED;
 
   if (pAlter->buffer > 0 && pAlter->buffer != pDb->cfg.buffer) {
+#if 1
+    terrno = TSDB_CODE_OPS_NOT_SUPPORT;
+    return terrno;
+#else
     pDb->cfg.buffer = pAlter->buffer;
     terrno = 0;
+#endif
   }
 
   if (pAlter->pages > 0 && pAlter->pages != pDb->cfg.pages) {
+#if 1
+    terrno = TSDB_CODE_OPS_NOT_SUPPORT;
+    return terrno;
+#else
     pDb->cfg.pages = pAlter->pages;
     terrno = 0;
+#endif
   }
 
   if (pAlter->pageSize > 0 && pAlter->pageSize != pDb->cfg.pageSize) {
+#if 1
+    terrno = TSDB_CODE_OPS_NOT_SUPPORT;
+    return terrno;
+#else
     pDb->cfg.pageSize = pAlter->pageSize;
     terrno = 0;
+#endif
   }
 
   if (pAlter->daysPerFile > 0 && pAlter->daysPerFile != pDb->cfg.daysPerFile) {
@@ -593,8 +609,12 @@ static int32_t mndSetDbCfgFromAlterDbReq(SDbObj *pDb, SAlterDbReq *pAlter) {
   }
 
   if (pAlter->strict >= 0 && pAlter->strict != pDb->cfg.strict) {
+#if 1
+    terrno = TSDB_CODE_OPS_NOT_SUPPORT;
+#else
     pDb->cfg.strict = pAlter->strict;
     terrno = 0;
+#endif
   }
 
   if (pAlter->cacheLastRow >= 0 && pAlter->cacheLastRow != pDb->cfg.cacheLastRow) {
@@ -603,9 +623,13 @@ static int32_t mndSetDbCfgFromAlterDbReq(SDbObj *pDb, SAlterDbReq *pAlter) {
   }
 
   if (pAlter->replications > 0 && pAlter->replications != pDb->cfg.replications) {
+#if 1
+    terrno = TSDB_CODE_OPS_NOT_SUPPORT;
+#else
     pDb->cfg.replications = pAlter->replications;
     pDb->vgVersion++;
     terrno = 0;
+#endif
   }
 
   return terrno;
@@ -645,7 +669,7 @@ 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);
@@ -927,6 +951,7 @@ static int32_t mndDropDb(SMnode *pMnode, SRpcMsg *pReq, SDbObj *pDb) {
   if (mndDropOffsetByDB(pMnode, pTrans, pDb) != 0) goto _OVER;
   if (mndDropSubByDB(pMnode, pTrans, pDb) != 0) goto _OVER;
   if (mndDropTopicByDB(pMnode, pTrans, pDb) != 0) goto _OVER;
+  if (mndDropStreamByDb(pMnode, pTrans, pDb) != 0) goto _OVER;
   if (mndDropSmasByDb(pMnode, pTrans, pDb) != 0) goto _OVER;
   if (mndSetDropDbRedoActions(pMnode, pTrans, pDb) != 0) goto _OVER;
 
@@ -947,7 +972,6 @@ static int32_t mndDropDb(SMnode *pMnode, SRpcMsg *pReq, SDbObj *pDb) {
   mndTransSetRpcRsp(pTrans, pRsp, rspLen);
 
   if (mndTransPrepare(pMnode, pTrans) != 0) goto _OVER;
-
   code = 0;
 
 _OVER:
@@ -1006,7 +1030,7 @@ 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/mndDnode.c

@@ -758,6 +758,11 @@ static int32_t mndProcessDropDnodeReq(SRpcMsg *pReq) {
     }
   }
 
+  if (numOfVnodes > 0) {
+    terrno = TSDB_CODE_OPS_NOT_SUPPORT;
+    goto _OVER;
+  }
+
   code = mndDropDnode(pMnode, pReq, pDnode, pMObj, pQObj, pSObj, numOfVnodes);
   if (code == 0) code = TSDB_CODE_ACTION_IN_PROGRESS;
 

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/mndShow.c

@@ -71,7 +71,7 @@ static int32_t convertToRetrieveType(char *name, int32_t len) {
   } else if (strncasecmp(name, TSDB_INS_TABLE_USER_FUNCTIONS, len) == 0) {
     type = TSDB_MGMT_TABLE_FUNC;
   } else if (strncasecmp(name, TSDB_INS_TABLE_USER_INDEXES, len) == 0) {
-    //    type = TSDB_MGMT_TABLE_INDEX;
+    type = TSDB_MGMT_TABLE_INDEX;
   } else if (strncasecmp(name, TSDB_INS_TABLE_USER_STABLES, len) == 0) {
     type = TSDB_MGMT_TABLE_STB;
   } else if (strncasecmp(name, TSDB_INS_TABLE_USER_TABLES, len) == 0) {

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;
@@ -562,6 +523,7 @@ static int32_t mndCreateSma(SMnode *pMnode, SRpcMsg *pReq, SMCreateSmaReq *pCrea
   streamObj.updateTime = streamObj.createTime;
   streamObj.uid = mndGenerateUid(pCreate->name, strlen(pCreate->name));
   streamObj.sourceDbUid = pDb->uid;
+  streamObj.targetDbUid = pDb->uid;
   streamObj.version = 1;
   streamObj.sql = pCreate->sql;
   streamObj.smaId = smaObj.uid;
@@ -621,7 +583,6 @@ 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 +731,6 @@ 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 +797,6 @@ 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 +823,20 @@ 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;
     }
@@ -927,36 +854,26 @@ _OVER:
 }
 
 int32_t mndDropSmasByDb(SMnode *pMnode, STrans *pTrans, SDbObj *pDb) {
-  SSdb    *pSdb = pMnode->pSdb;
+  SSdb *pSdb = pMnode->pSdb;
-  SSmaObj *pSma = NULL;
+  void *pIter = NULL;
-  void    *pIter = NULL;
-  SVgObj  *pVgroup = NULL;
-  int32_t  code = -1;
 
   while (1) {
+    SSmaObj *pSma = NULL;
     pIter = sdbFetch(pSdb, SDB_SMA, pIter, (void **)&pSma);
     if (pIter == NULL) break;
 
     if (pSma->dbUid == pDb->uid) {
-      pVgroup = mndAcquireVgroup(pMnode, pSma->dstVgId);
+      if (mndSetDropSmaCommitLogs(pMnode, pTrans, pSma) != 0) {
-      if (pVgroup == NULL) goto _OVER;
+        sdbRelease(pSdb, pSma);
-      if (mndSetDropSmaVgroupCommitLogs(pMnode, pTrans, pVgroup) != 0) goto _OVER;
+        sdbCancelFetch(pSdb, pSma);
-      if (mndSetDropSmaVgroupRedoActions(pMnode, pTrans, pDb, pVgroup) != 0) goto _OVER;
+        return -1;
-      if (mndSetDropSmaCommitLogs(pMnode, pTrans, pSma) != 0) goto _OVER;
+      }
-      mndReleaseVgroup(pMnode, pVgroup);
-      pVgroup = NULL;
     }
 
     sdbRelease(pSdb, pSma);
   }
 
-  code = 0;
+  return 0;
-
-_OVER:
-  sdbCancelFetch(pSdb, pIter);
-  sdbRelease(pSdb, pSma);
-  mndReleaseVgroup(pMnode, pVgroup);
-  return code;
 }
 
 static int32_t mndProcessDropSmaReq(SRpcMsg *pReq) {
@@ -1230,29 +1147,32 @@ static int32_t mndRetrieveSma(SRpcMsg *pReq, SShowObj *pShow, SSDataBlock *pBloc
 
     SName smaName = {0};
     tNameFromString(&smaName, pSma->name, T_NAME_ACCT | T_NAME_DB | T_NAME_TABLE);
+    char n1[TSDB_TABLE_FNAME_LEN + VARSTR_HEADER_SIZE] = {0};
+    STR_TO_VARSTR(n1, (char *)tNameGetTableName(&smaName));
 
-    char n[TSDB_TABLE_FNAME_LEN + VARSTR_HEADER_SIZE] = {0};
+    char n2[TSDB_DB_FNAME_LEN + VARSTR_HEADER_SIZE] = {0};
-    STR_TO_VARSTR(n, (char *)tNameGetTableName(&smaName));
+    STR_TO_VARSTR(n2, (char *)mndGetDbStr(pDb->name));
-    cols++;
 
     SName stbName = {0};
     tNameFromString(&stbName, pSma->stb, T_NAME_ACCT | T_NAME_DB | T_NAME_TABLE);
-
+    char n3[TSDB_TABLE_FNAME_LEN + VARSTR_HEADER_SIZE] = {0};
-    char n1[TSDB_TABLE_FNAME_LEN + VARSTR_HEADER_SIZE] = {0};
+    STR_TO_VARSTR(n3, (char *)tNameGetTableName(&stbName));
-    STR_TO_VARSTR(n1, (char *)tNameGetTableName(&stbName));
 
     SColumnInfoData *pColInfo = taosArrayGet(pBlock->pDataBlock, cols++);
-    colDataAppend(pColInfo, numOfRows, (const char *)n, false);
-
-    pColInfo = taosArrayGet(pBlock->pDataBlock, cols++);
-    colDataAppend(pColInfo, numOfRows, (const char *)&pSma->createdTime, false);
-
-    pColInfo = taosArrayGet(pBlock->pDataBlock, cols++);
     colDataAppend(pColInfo, numOfRows, (const char *)n1, false);
 
+    pColInfo = taosArrayGet(pBlock->pDataBlock, cols++);
+    colDataAppend(pColInfo, numOfRows, (const char *)n2, false);
+
+    pColInfo = taosArrayGet(pBlock->pDataBlock, cols++);
+    colDataAppend(pColInfo, numOfRows, (const char *)n3, false);
+
     pColInfo = taosArrayGet(pBlock->pDataBlock, cols++);
     colDataAppend(pColInfo, numOfRows, (const char *)&pSma->dstVgId, false);
 
+    pColInfo = taosArrayGet(pBlock->pDataBlock, cols++);
+    colDataAppend(pColInfo, numOfRows, (const char *)&pSma->createdTime, false);
+
     numOfRows++;
     sdbRelease(pSdb, pSma);
   }

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

@@ -273,6 +273,9 @@ _OVER:
 }
 
 static int32_t mndProcessCreateSnodeReq(SRpcMsg *pReq) {
+#if 1
+  return TSDB_CODE_OPS_NOT_SUPPORT;
+#else
   SMnode          *pMnode = pReq->info.node;
   int32_t          code = -1;
   SSnodeObj       *pObj = NULL;
@@ -315,6 +318,7 @@ _OVER:
   mndReleaseSnode(pMnode, pObj);
   mndReleaseDnode(pMnode, pDnode);
   return code;
+#endif
 }
 
 static int32_t mndSetDropSnodeRedoLogs(STrans *pTrans, SSnodeObj *pObj) {
@@ -386,9 +390,12 @@ _OVER:
 }
 
 static int32_t mndProcessDropSnodeReq(SRpcMsg *pReq) {
-  SMnode        *pMnode = pReq->info.node;
+#if 1
-  int32_t        code = -1;
+  return TSDB_CODE_OPS_NOT_SUPPORT;
-  SSnodeObj     *pObj = NULL;
+#else
+  SMnode *pMnode = pReq->info.node;
+  int32_t code = -1;
+  SSnodeObj *pObj = NULL;
   SMDropSnodeReq dropReq = {0};
 
   if (tDeserializeSCreateDropMQSBNodeReq(pReq->pCont, pReq->contLen, &dropReq) != 0) {
@@ -422,6 +429,7 @@ _OVER:
 
   mndReleaseSnode(pMnode, pObj);
   return code;
+#endif
 }
 
 static int32_t mndRetrieveSnodes(SRpcMsg *pReq, SShowObj *pShow, SSDataBlock *pBlock, int32_t rows) {

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

@@ -247,8 +247,7 @@ static int32_t mndBuildStreamObjFromCreateReq(SMnode *pMnode, SStreamObj *pObj,
   pObj->uid = mndGenerateUid(pObj->name, strlen(pObj->name));
   pObj->status = 0;
 
-  // TODO
+  pObj->igExpired = pCreate->igExpired;
-  pObj->dropPolicy = 0;
   pObj->trigger = pCreate->triggerType;
   pObj->triggerParam = pCreate->maxDelay;
   pObj->watermark = pCreate->watermark;
@@ -301,6 +300,7 @@ 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
@@ -673,27 +673,29 @@ static int32_t mndProcessDropStreamReq(SRpcMsg *pReq) {
 
 int32_t mndDropStreamByDb(SMnode *pMnode, STrans *pTrans, SDbObj *pDb) {
   SSdb *pSdb = pMnode->pSdb;
+  void *pIter = NULL;
 
-  void       *pIter = NULL;
-  SStreamObj *pStream = NULL;
   while (1) {
+    SStreamObj *pStream = NULL;
     pIter = sdbFetch(pSdb, SDB_STREAM, pIter, (void **)&pStream);
     if (pIter == NULL) break;
 
     if (pStream->sourceDbUid == pDb->uid || pStream->targetDbUid == pDb->uid) {
       if (pStream->sourceDbUid != pStream->targetDbUid) {
         sdbRelease(pSdb, pStream);
+        sdbCancelFetch(pSdb, pIter);
+        mError("db:%s, failed to drop stream:%s since sourceDbUid:%" PRId64 " not match with targetDbUid:%" PRId64,
+               pDb->name, pStream->name, pStream->sourceDbUid, pStream->targetDbUid);
+        terrno = TSDB_CODE_MND_STREAM_ALREADY_EXIST;
         return -1;
       } else {
         // TODO drop all task on snode
         if (mndPersistDropStreamLog(pMnode, pTrans, pStream) < 0) {
           sdbRelease(pSdb, pStream);
+          sdbCancelFetch(pSdb, pIter);
           return -1;
         }
       }
-    } else {
-      sdbRelease(pSdb, pStream);
-      continue;
     }
 
 #if 0

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

@@ -90,7 +90,7 @@ static int32_t mndTransGetActionsSize(SArray *pArray) {
   for (int32_t i = 0; i < actionNum; ++i) {
     STransAction *pAction = taosArrayGet(pArray, i);
     if (pAction->actionType == TRANS_ACTION_RAW) {
-      rawDataLen += (sdbGetRawTotalSize(pAction->pRaw) + sizeof(int32_t));
+      rawDataLen += (sizeof(STransAction) + sdbGetRawTotalSize(pAction->pRaw));
     } else if (pAction->actionType == TRANS_ACTION_MSG) {
       rawDataLen += (sizeof(STransAction) + pAction->contLen);
     } else {
@@ -105,7 +105,7 @@ static int32_t mndTransGetActionsSize(SArray *pArray) {
 static SSdbRaw *mndTransActionEncode(STrans *pTrans) {
   terrno = TSDB_CODE_OUT_OF_MEMORY;
 
-  int32_t rawDataLen = sizeof(STrans) + TRANS_RESERVE_SIZE;
+  int32_t rawDataLen = sizeof(STrans) + TRANS_RESERVE_SIZE + pTrans->paramLen;
   rawDataLen += mndTransGetActionsSize(pTrans->redoActions);
   rawDataLen += mndTransGetActionsSize(pTrans->undoActions);
   rawDataLen += mndTransGetActionsSize(pTrans->commitActions);
@@ -226,7 +226,8 @@ static SSdbRaw *mndTransActionEncode(STrans *pTrans) {
 
 _OVER:
   if (terrno != 0) {
-    mError("trans:%d, failed to encode to raw:%p len:%d since %s", pTrans->id, pRaw, dataPos, terrstr());
+    mError("trans:%d, failed to encode to raw:%p maxlen:%d len:%d since %s", pTrans->id, pRaw, sdbGetRawTotalSize(pRaw),
+           dataPos, terrstr());
     sdbFreeRaw(pRaw);
     return NULL;
   }
@@ -1025,7 +1026,7 @@ static int32_t mndTransExecNullMsg(SMnode *pMnode, STrans *pTrans, STransAction
   pTrans->lastAction = pAction->id;
   pTrans->lastMsgType = pAction->msgType;
   pTrans->lastEpset = pAction->epSet;
-  pTrans->lastErrorNo == 0;
+  pTrans->lastErrorNo = 0;
   return 0;
 }
 
@@ -1146,13 +1147,13 @@ 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) {
@@ -1211,6 +1219,9 @@ _OVER:
 }
 
 static int32_t mndProcessRedistributeVgroupMsg(SRpcMsg *pReq) {
+#if 1
+  return TSDB_CODE_OPS_NOT_SUPPORT;
+#else
   SMnode    *pMnode = pReq->info.node;
   SDnodeObj *pNew1 = NULL;
   SDnodeObj *pNew2 = NULL;
@@ -1404,6 +1415,7 @@ _OVER:
   mndReleaseDb(pMnode, pDb);
 
   return code;
+#endif  
 }
 
 int32_t mndBuildAlterVgroupAction(SMnode *pMnode, STrans *pTrans, SDbObj *pDb, SVgObj *pVgroup, SArray *pArray) {
@@ -1703,6 +1715,9 @@ _OVER:
 }
 
 static int32_t mndProcessBalanceVgroupMsg(SRpcMsg *pReq) {
+#if 1
+  return TSDB_CODE_OPS_NOT_SUPPORT;
+#else
   SMnode *pMnode = pReq->info.node;
   int32_t code = -1;
   SArray *pArray = NULL;
@@ -1751,4 +1766,7 @@ _OVER:
 
   taosArrayDestroy(pArray);
   return code;
-}
+#endif 
+}
+
+bool mndVgroupInDb(SVgObj *pVgroup, int64_t dbUid) { return !pVgroup->isTsma && pVgroup->dbUid == dbUid; }

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

@@ -93,7 +93,7 @@ TEST_F(MndTestDb, 02_Create_Alter_Drop_Db) {
 
     SRpcMsg* pRsp = test.SendReq(TDMT_MND_ALTER_DB, pReq, contLen);
     ASSERT_NE(pRsp, nullptr);
-    ASSERT_EQ(pRsp->code, 0);
+    ASSERT_EQ(pRsp->code, TSDB_CODE_OPS_NOT_SUPPORT);
   }
 
   test.SendShowReq(TSDB_MGMT_TABLE_DB, "user_databases", "");

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/src/inc/sma.h

@@ -62,12 +62,10 @@ struct STSmaStat {
 
 struct SRSmaStat {
   SSma     *pSma;
-  int64_t   refId;         // shared by persistence/fetch tasks
+  int64_t   refId;         // shared by fetch tasks
-  void     *tmrHandle;     // for persistence task
+  void     *tmrHandle;     // shared by fetch tasks
-  tmr_h     tmrId;         // for persistence task
+  int8_t    triggerStat;   // shared by fetch tasks
-  int32_t   tmrSeconds;    // for persistence task
+  int8_t    runningStat;   // for persistence task 
-  int8_t    triggerStat;   // for persistence task
-  int8_t    runningStat;   // for persistence task
   SHashObj *rsmaInfoHash;  // key: stbUid, value: SRSmaInfo;
 };
 
@@ -82,7 +80,6 @@ struct SSmaStat {
 #define SMA_TSMA_STAT(s)     (&(s)->tsmaStat)
 #define SMA_RSMA_STAT(s)     (&(s)->rsmaStat)
 #define RSMA_INFO_HASH(r)    ((r)->rsmaInfoHash)
-#define RSMA_TMR_ID(r)       ((r)->tmrId)
 #define RSMA_TMR_HANDLE(r)   ((r)->tmrHandle)
 #define RSMA_TRIGGER_STAT(r) (&(r)->triggerStat)
 #define RSMA_RUNNING_STAT(r) (&(r)->runningStat)
@@ -92,8 +89,9 @@ 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);
@@ -184,9 +182,11 @@ static FORCE_INLINE void tdSmaStatSetDropped(STSmaStat *pTStat) {
 static int32_t tdDestroySmaState(SSmaStat *pSmaStat, int8_t smaType);
 void          *tdFreeSmaState(SSmaStat *pSmaStat, int8_t smaType);
 void          *tdFreeRSmaInfo(SRSmaInfo *pInfo);
+int32_t        tdRSmaPersistExecImpl(SRSmaStat *pRSmaStat);
 
 int32_t tdProcessRSmaCreateImpl(SSma *pSma, SRSmaParam *param, int64_t suid, const char *tbName);
 int32_t tdProcessRSmaRestoreImpl(SSma *pSma);
+
 int32_t tdProcessTSmaCreateImpl(SSma *pSma, int64_t version, const char *pMsg);
 int32_t tdProcessTSmaInsertImpl(SSma *pSma, int64_t indexUid, const char *msg);
 int32_t tdProcessTSmaGetDaysImpl(SVnodeCfg *pCfg, void *pCont, uint32_t contLen, int32_t *days);
@@ -214,25 +214,22 @@ 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 +241,10 @@ 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 *pdname, const char *dname, const char *fname, int64_t version, char *outputName);
+void tdGetVndDirName(int32_t vgId,const char *pdname,  const char *dname, bool endWithSep, char *outputName);
 
 #ifdef __cplusplus
 }

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

@@ -157,7 +157,7 @@ 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);
@@ -178,6 +178,7 @@ 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

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

@@ -163,8 +163,11 @@ SSubmitReq* tdBlockToSubmit(const SArray* pBlocks, const STSchema* pSchema, bool
 
 // sma
 int32_t smaOpen(SVnode* pVnode);
-int32_t smaCloseEnv(SSma* pSma);
+int32_t smaClose(SSma* pSma);
-int32_t smaCloseEx(SSma* pSma);
+int32_t smaBegin(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);
@@ -238,7 +241,7 @@ struct SVnode {
   tsem_t     canCommit;
   int64_t    sync;
   int32_t    syncCount;
-  sem_t      syncSem;
+  tsem_t     syncSem;
   SQHandle*  pQuery;
 };
 

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

@@ -0,0 +1,225 @@
+/*
+ * 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 set rsma trigger stat active
+ *
+ * @param pSma
+ * @return int32_t
+ */
+int32_t smaBegin(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);
+
+  int8_t rsmaTriggerStat =
+      atomic_val_compare_exchange_8(RSMA_TRIGGER_STAT(pRSmaStat), TASK_TRIGGER_STAT_PAUSED, TASK_TRIGGER_STAT_ACTIVE);
+  switch (rsmaTriggerStat) {
+    case TASK_TRIGGER_STAT_PAUSED: {
+      smaDebug("vgId:%d rsma trigger stat from paused to active", SMA_VID(pSma));
+      break;
+    }
+    case TASK_TRIGGER_STAT_INIT: {
+      atomic_store_8(RSMA_TRIGGER_STAT(pRSmaStat), TASK_TRIGGER_STAT_ACTIVE);
+      smaDebug("vgId:%d rsma trigger stat from init to active", SMA_VID(pSma));
+      break;
+    }
+    default: {
+      atomic_store_8(RSMA_TRIGGER_STAT(pRSmaStat), TASK_TRIGGER_STAT_ACTIVE);
+      smaWarn("vgId:%d rsma trigger stat %" PRIi8 " is unexpected", SMA_VID(pSma), rsmaTriggerStat);
+      ASSERT(0);
+      break;
+    }
+  }
+  return TSDB_CODE_SUCCESS;
+}
+
+/**
+ * @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
+ *
+ * @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: set persistence task paused
+  atomic_store_8(RSMA_TRIGGER_STAT(pRSmaStat), TASK_TRIGGER_STAT_PAUSED);
+
+  // step 2: perform persist task for qTaskInfo
+  tdRSmaPersistExecImpl(pRSmaStat);
+
+  // step 3: wait all triggered fetch tasks finished
+  int32_t nLoops = 0;
+  while (1) {
+    if (T_REF_VAL_GET(pStat) == 0) {
+      smaDebug("vgId:%d, rsma fetch tasks all finished", SMA_VID(pSma));
+      break;
+    } else {
+      smaDebug("vgId:%d, rsma fetch tasks not all finished yet", SMA_VID(pSma));
+    }
+    ++nLoops;
+    if (nLoops > 1000) {
+      sched_yield();
+      nLoops = 0;
+    }
+  }
+
+  smaDebug("vgId:%d, rsma pre commit succeess", SMA_VID(pSma));
+
+  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];
+  const char   *pattern = "v[0-9]+qtaskinfo\\.ver([0-9]+)?$";
+  regex_t       regex;
+  int           code = 0;
+
+  tdGetVndDirName(TD_VID(pVnode), tfsGetPrimaryPath(pVnode->pTfs), VNODE_RSMA_DIR, true, dir);
+
+  // Resource allocation and init
+  if ((code = regcomp(&regex, pattern, REG_EXTENDED)) != 0) {
+    char errbuf[128];
+    regerror(code, &regex, errbuf, sizeof(errbuf));
+    smaWarn("vgId:%d, rsma post commit, regcomp for %s failed since %s", TD_VID(pVnode), dir, errbuf);
+    return TSDB_CODE_FAILED;
+  }
+
+  if ((pDir = taosOpenDir(dir)) == NULL) {
+    terrno = TAOS_SYSTEM_ERROR(errno);
+    smaWarn("vgId:%d, rsma post commit, open dir %s failed since %s", TD_VID(pVnode), dir, terrstr());
+    return TSDB_CODE_FAILED;
+  }
+
+  int32_t    dirLen = strlen(dir);
+  char      *dirEnd = POINTER_SHIFT(dir, dirLen);
+  regmatch_t regMatch[2];
+  while ((pDirEntry = taosReadDir(pDir)) != NULL) {
+    char *entryName = taosGetDirEntryName(pDirEntry);
+    if (!entryName) {
+      continue;
+    }
+
+    code = regexec(&regex, entryName, 2, regMatch, 0);
+
+    if (code == 0) {
+      // match
+      int64_t version = -1;
+      sscanf((const char *)POINTER_SHIFT(entryName, regMatch[1].rm_so), "%" PRIi64, &version);
+      if ((version < committed) && (version > -1)) {
+        strncpy(dirEnd, entryName, TSDB_FILENAME_LEN - dirLen);
+        if (taosRemoveFile(dir) != 0) {
+          terrno = TAOS_SYSTEM_ERROR(errno);
+          smaWarn("vgId:%d, committed version:%" PRIi64 ", failed to remove %s since %s", TD_VID(pVnode), committed,
+                  dir, terrstr());
+        } else {
+          smaDebug("vgId:%d, committed version:%" PRIi64 ", success to remove %s", TD_VID(pVnode), committed, dir);
+        }
+      }
+    } else if (code == REG_NOMATCH) {
+      // not match
+      smaTrace("vgId:%d, rsma post commit, not match %s", TD_VID(pVnode), entryName);
+      continue;
+    } else {
+      // has other error
+      char errbuf[128];
+      regerror(code, &regex, errbuf, sizeof(errbuf));
+      smaWarn("vgId:%d, rsma post commit, regexec failed since %s", TD_VID(pVnode), errbuf);
+
+      taosCloseDir(&pDir);
+      regfree(&regex);
+      return TSDB_CODE_FAILED;
+    }
+  }
+
+  taosCloseDir(&pDir);
+  regfree(&regex);
+  return TSDB_CODE_SUCCESS;
+}

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

@@ -132,6 +132,7 @@ static int32_t tdInitSmaStat(SSmaStat **pSmaStat, int8_t smaType, const SSma *pS
     if (smaType == TSDB_SMA_TYPE_ROLLUP) {
       SRSmaStat *pRSmaStat = (SRSmaStat *)(*pSmaStat);
       pRSmaStat->pSma = (SSma *)pSma;
+      atomic_store_8(RSMA_TRIGGER_STAT(pRSmaStat), TASK_TRIGGER_STAT_INIT);
 
       // init smaMgmt
       smaMgmt.smaRef = taosOpenRef(SMA_MGMT_REF_NUM, tdDestroyRSmaStat);
@@ -192,22 +193,20 @@ static void *tdFreeTSmaStat(STSmaStat *pStat) {
 static void tdDestroyRSmaStat(void *pRSmaStat) {
   if (pRSmaStat) {
     SRSmaStat *pStat = (SRSmaStat *)pRSmaStat;
-    smaDebug("vgId:%d %s:%d destroy rsma stat %p", SMA_VID(pStat->pSma), __func__, __LINE__, pRSmaStat);
+    SSma      *pSma = pStat->pSma;
-    // step 1: set persistence task cancelled
+    smaDebug("vgId:%d, destroy rsma stat %p", SMA_VID(pSma), pRSmaStat);
+    // step 1: set rsma trigger stat cancelled
     atomic_store_8(RSMA_TRIGGER_STAT(pStat), TASK_TRIGGER_STAT_CANCELLED);
 
-    // step 2: stop the persistence timer
+    // step 2: wait the persistence thread to finish
-    taosTmrStopA(&RSMA_TMR_ID(pStat));
-
-    // step 3: wait the persistence thread to finish
     int32_t nLoops = 0;
     if (atomic_load_8(RSMA_RUNNING_STAT(pStat)) == 1) {
       while (1) {
         if (atomic_load_8(RSMA_TRIGGER_STAT(pStat)) == TASK_TRIGGER_STAT_FINISHED) {
-          smaDebug("rsma, persist task finished already");
+          smaDebug("vgId:%d, rsma persist task finished already", SMA_VID(pSma));
           break;
         } else {
-          smaDebug("rsma, persist task not finished yet since rsma stat in %" PRIi8,
+          smaDebug("vgId:%d, rsma persist task not finished yet since rsma stat in %" PRIi8, SMA_VID(pSma),
                    atomic_load_8(RSMA_TRIGGER_STAT(pStat)));
         }
         ++nLoops;
@@ -218,13 +217,15 @@ static void tdDestroyRSmaStat(void *pRSmaStat) {
       }
     }
 
-    // step 4: destroy the rsma info and associated fetch tasks
+    // step 3: destroy the rsma info and associated fetch tasks
     // TODO: use taosHashSetFreeFp when taosHashSetFreeFp is ready.
-    void *infoHash = taosHashIterate(RSMA_INFO_HASH(pStat), NULL);
+    if (taosHashGetSize(RSMA_INFO_HASH(pStat)) > 0) {
-    while (infoHash) {
+      void *infoHash = taosHashIterate(RSMA_INFO_HASH(pStat), NULL);
-      SRSmaInfo *pSmaInfo = *(SRSmaInfo **)infoHash;
+      while (infoHash) {
-      tdFreeRSmaInfo(pSmaInfo);
+        SRSmaInfo *pSmaInfo = *(SRSmaInfo **)infoHash;
-      infoHash = taosHashIterate(RSMA_INFO_HASH(pStat), infoHash);
+        tdFreeRSmaInfo(pSmaInfo);
+        infoHash = taosHashIterate(RSMA_INFO_HASH(pStat), infoHash);
+      }
     }
     taosHashCleanup(RSMA_INFO_HASH(pStat));
 
@@ -232,10 +233,10 @@ static void tdDestroyRSmaStat(void *pRSmaStat) {
     nLoops = 0;
     while (1) {
       if (T_REF_VAL_GET((SSmaStat *)pStat) == 0) {
-        smaDebug("rsma, all fetch task finished already");
+        smaDebug("vgId:%d, rsma fetch tasks all finished", SMA_VID(pSma));
         break;
       } else {
-        smaDebug("rsma, fetch tasks not all finished yet");
+        smaDebug("vgId:%d, rsma fetch tasks not all finished yet", SMA_VID(pSma));
       }
       ++nLoops;
       if (nLoops > 1000) {
@@ -275,7 +276,7 @@ int32_t tdDestroySmaState(SSmaStat *pSmaStat, int8_t smaType) {
     } else if (smaType == TSDB_SMA_TYPE_ROLLUP) {
       SRSmaStat *pRSmaStat = SMA_RSMA_STAT(pSmaStat);
       if (taosRemoveRef(smaMgmt.smaRef, RSMA_REF_ID(pRSmaStat)) < 0) {
-        smaError("remove refId from smaRef failed, refId:0x%" PRIx64, RSMA_REF_ID(pRSmaStat));
+        smaError("remove refId from rsmaRef:0x%" PRIx64 " failed since %s", RSMA_REF_ID(pRSmaStat), terrstr());
       }
     } else {
       ASSERT(0);

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

@@ -135,17 +135,11 @@ _err:
   return -1;
 }
 
-int32_t smaCloseEnv(SSma *pSma) {
+int32_t smaClose(SSma *pSma) {
-  if (pSma) {
-    SMA_TSMA_ENV(pSma) = tdFreeSmaEnv(SMA_TSMA_ENV(pSma));
-    SMA_RSMA_ENV(pSma) = tdFreeSmaEnv(SMA_RSMA_ENV(pSma));
-  }
-  return 0;
-}
-
-int32_t smaCloseEx(SSma *pSma) {
   if (pSma) {
     taosThreadMutexDestroy(&pSma->mutex);
+    SMA_TSMA_ENV(pSma) = tdFreeSmaEnv(SMA_TSMA_ENV(pSma));
+    SMA_RSMA_ENV(pSma) = tdFreeSmaEnv(SMA_RSMA_ENV(pSma));
     if SMA_RSMA_TSDB0 (pSma) tsdbClose(&SMA_RSMA_TSDB0(pSma));
     if SMA_RSMA_TSDB1 (pSma) tsdbClose(&SMA_RSMA_TSDB1(pSma));
     if SMA_RSMA_TSDB2 (pSma) tsdbClose(&SMA_RSMA_TSDB2(pSma));

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

@@ -15,16 +15,14 @@
 
 #include "sma.h"
 
-#define RSMA_QTASKINFO_PERSIST_MS 7200000
+#define RSMA_QTASKINFO_BUFSIZE  32768
-#define RSMA_QTASKINFO_BUFSIZE    32768
+#define RSMA_QTASKINFO_HEAD_LEN (sizeof(int32_t) + sizeof(int8_t) + sizeof(int64_t))  // len + type + suid
-#define RSMA_QTASKINFO_HEAD_LEN   (sizeof(int32_t) + sizeof(int8_t) + sizeof(int64_t))  // len + type + suid
 
 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;
 
@@ -37,16 +35,16 @@ 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);
 static int32_t tdRSmaQTaskInfoRestore(SSma *pSma, SRSmaQTaskInfoIter *pIter);
 static int32_t tdRSmaQTaskInfoItemRestore(SSma *pSma, const SRSmaQTaskInfoItem *infoItem);
 
-static int32_t tdRSmaRestoreQTaskInfoInit(SSma *pSma);
+static int32_t tdRSmaRestoreQTaskInfoInit(SSma *pSma, int64_t *nTables);
-static int32_t tdRSmaRestoreQTaskInfoReload(SSma *pSma);
+static int32_t tdRSmaRestoreQTaskInfoReload(SSma *pSma, int64_t *committed);
-static int32_t tdRSmaRestoreTSDataReload(SSma *pSma);
+static int32_t tdRSmaRestoreTSDataReload(SSma *pSma, int64_t committed);
 
 struct SRSmaInfoItem {
   SRSmaInfo *pRsmaInfo;
@@ -88,8 +86,8 @@ 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, NULL, VNODE_RSMA_DIR, TD_QTASKINFO_FNAME_PREFIX, version, outputName);
 }
 
 static FORCE_INLINE int32_t tdRSmaQTaskInfoContLen(int32_t lenWithHead) {
@@ -115,12 +113,14 @@ void *tdFreeRSmaInfo(SRSmaInfo *pInfo) {
     for (int32_t i = 0; i < TSDB_RETENTION_L2; ++i) {
       SRSmaInfoItem *pItem = &pInfo->items[i];
       if (pItem->taskInfo) {
-        smaDebug("vgId:%d, stb %" PRIi64 " stop fetch-timer %p level %d", SMA_VID(pSma), pInfo->suid, pItem->tmrId,
+        if (pItem->tmrId) {
-                 i + 1);
+          smaDebug("vgId:%d, table %" PRIi64 " stop fetch timer %p level %d", SMA_VID(pSma), pInfo->suid, pItem->tmrId,
-        taosTmrStopA(&pItem->tmrId);
+                   i + 1);
+          taosTmrStopA(&pItem->tmrId);
+        }
         tdFreeTaskHandle(&pItem->taskInfo, SMA_VID(pSma), i + 1);
       } else {
-        smaDebug("vgId:%d, stb %" PRIi64 " no need to destroy rsma info level %d since empty taskInfo", SMA_VID(pSma),
+        smaDebug("vgId:%d, table %" PRIi64 " no need to destroy rsma info level %d since empty taskInfo", SMA_VID(pSma),
                  pInfo->suid, i + 1);
       }
     }
@@ -359,13 +359,7 @@ int32_t tdProcessRSmaCreateImpl(SSma *pSma, SRSmaParam *param, int64_t suid, con
     goto _err;
   }
 
-  smaDebug("vgId:%d, register rsma info succeed for suid:%" PRIi64, SMA_VID(pSma), suid);
+  smaDebug("vgId:%d, register rsma info succeed for table %" PRIi64, SMA_VID(pSma), suid);
-
-  // start the persist timer
-  if (TASK_TRIGGER_STAT_INIT ==
-      atomic_val_compare_exchange_8(RSMA_TRIGGER_STAT(pStat), TASK_TRIGGER_STAT_INIT, TASK_TRIGGER_STAT_ACTIVE)) {
-    taosTmrStart(tdRSmaPersistTrigger, RSMA_QTASKINFO_PERSIST_MS, pStat, RSMA_TMR_HANDLE(pStat));
-  }
 
   return TSDB_CODE_SUCCESS;
 _err:
@@ -493,7 +487,6 @@ 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 +494,26 @@ 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 +598,54 @@ 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);
 }
@@ -722,7 +743,7 @@ int32_t tdProcessRSmaSubmit(SSma *pSma, void *pMsg, int32_t inputType) {
   return TSDB_CODE_SUCCESS;
 }
 
-static int32_t tdRSmaRestoreQTaskInfoInit(SSma *pSma) {
+static int32_t tdRSmaRestoreQTaskInfoInit(SSma *pSma, int64_t *nTables) {
   SVnode *pVnode = pSma->pVnode;
 
   SArray *suidList = taosArrayInit(1, sizeof(tb_uid_t));
@@ -732,7 +753,12 @@ static int32_t tdRSmaRestoreQTaskInfoInit(SSma *pSma) {
     return TSDB_CODE_FAILED;
   }
 
-  int32_t arrSize = taosArrayGetSize(suidList);
+  int64_t arrSize = taosArrayGetSize(suidList);
+
+  if (nTables) {
+    *nTables = arrSize;
+  }
+
   if (arrSize == 0) {
     taosArrayDestroy(suidList);
     smaDebug("vgId:%d, no need to restore rsma env since empty stb id list", TD_VID(pVnode));
@@ -741,9 +767,9 @@ static int32_t tdRSmaRestoreQTaskInfoInit(SSma *pSma) {
 
   SMetaReader mr = {0};
   metaReaderInit(&mr, SMA_META(pSma), 0);
-  for (int32_t i = 0; i < arrSize; ++i) {
+  for (int64_t i = 0; i < arrSize; ++i) {
     tb_uid_t suid = *(tb_uid_t *)taosArrayGet(suidList, i);
-    smaDebug("vgId:%d, rsma restore, suid[%d] is %" PRIi64, TD_VID(pVnode), i, suid);
+    smaDebug("vgId:%d, rsma restore, suid is %" PRIi64, TD_VID(pVnode), suid);
     if (metaGetTableEntryByUid(&mr, suid) < 0) {
       smaError("vgId:%d, rsma restore, failed to get table meta for %" PRIi64 " since %s", TD_VID(pVnode), suid,
                terrstr());
@@ -777,17 +803,24 @@ _err:
   return TSDB_CODE_FAILED;
 }
 
-static int32_t tdRSmaRestoreQTaskInfoReload(SSma *pSma) {
+static int32_t tdRSmaRestoreQTaskInfoReload(SSma *pSma, int64_t *committed) {
   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;
   }
 
   if (!taosCheckExistFile(TD_TFILE_FULL_NAME(&tFile))) {
+    if (pVnode->state.committed > 0) {
+      smaWarn("vgId:%d, rsma restore for version %" PRIi64 ", not start as %s not exist", TD_VID(pVnode),
+              pVnode->state.committed, TD_TFILE_FULL_NAME(&tFile));
+    } else {
+      smaDebug("vgId:%d, rsma restore for version %" PRIi64 ", no need as %s not exist", TD_VID(pVnode),
+               pVnode->state.committed, TD_TFILE_FULL_NAME(&tFile));
+    }
     return TSDB_CODE_SUCCESS;
   }
 
@@ -799,20 +832,28 @@ 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);
+
+  // restored successfully from committed
+  *committed = pVnode->state.committed;
+
   return TSDB_CODE_SUCCESS;
 _err:
-  smaError("rsma restore, qtaskinfo reload failed since %s", terrstr());
+  smaError("vgId:%d, rsma restore for version %" PRIi64 ", qtaskinfo reload failed since %s", TD_VID(pVnode),
+           pVnode->state.committed, terrstr());
   return TSDB_CODE_FAILED;
 }
 
@@ -820,35 +861,45 @@ _err:
  * @brief reload ts data from checkpoint
  *
  * @param pSma
+ * @param committed restore from committed version
  * @return int32_t
  */
-static int32_t tdRSmaRestoreTSDataReload(SSma *pSma) {
+static int32_t tdRSmaRestoreTSDataReload(SSma *pSma, int64_t committed) {
   // TODO
+  smaDebug("vgId:%d, rsma restore from %" PRIi64 ", ts data reload success", SMA_VID(pSma), committed);
   return TSDB_CODE_SUCCESS;
 _err:
-  smaError("rsma restore, ts data reload failed since %s", terrstr());
+  smaError("vgId:%d, rsma restore from %" PRIi64 ", ts data reload failed since %s", SMA_VID(pSma), committed,
+           terrstr());
   return TSDB_CODE_FAILED;
 }
 
 int32_t tdProcessRSmaRestoreImpl(SSma *pSma) {
   // step 1: iterate all stables to restore the rsma env
-  if (tdRSmaRestoreQTaskInfoInit(pSma) < 0) {
+  int64_t nTables = 0;
+  if (tdRSmaRestoreQTaskInfoInit(pSma, &nTables) < 0) {
     goto _err;
   }
 
+  if (nTables <= 0) {
+    smaDebug("vgId:%d, no need to restore rsma task since no tables", SMA_VID(pSma));
+    return TSDB_CODE_SUCCESS;
+  }
+
   // step 2: retrieve qtaskinfo items from the persistence file(rsma/qtaskinfo) and restore
-  if (tdRSmaRestoreQTaskInfoReload(pSma) < 0) {
+  int64_t committed = -1;
+  if (tdRSmaRestoreQTaskInfoReload(pSma, &committed) < 0) {
     goto _err;
   }
 
   // step 3: reload ts data from checkpoint
-  if (tdRSmaRestoreTSDataReload(pSma) < 0) {
+  if (tdRSmaRestoreTSDataReload(pSma, committed) < 0) {
     goto _err;
   }
 
   return TSDB_CODE_SUCCESS;
 _err:
-  smaError("failed to restore rsma task since %s", terrstr());
+  smaError("vgId:%d failed to restore rsma task since %s", SMA_VID(pSma), terrstr());
   return TSDB_CODE_FAILED;
 }
 
@@ -931,19 +982,21 @@ 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 +1008,10 @@ 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 +1028,6 @@ 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 +1039,8 @@ 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("vgId:%d, restore rsma qtaskinfo file %s failed since %s", SMA_VID(pSma),
+               TD_TFILE_FULL_NAME(pIter->pTFile), terrstr());
       return TSDB_CODE_FAILED;
     }
 
@@ -1025,22 +1077,20 @@ static int32_t tdRSmaQTaskInfoRestore(SSma *pSma, SRSmaQTaskInfoIter *pIter) {
   return TSDB_CODE_SUCCESS;
 }
 
-static void *tdRSmaPersistExec(void *param) {
+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))) {
+  if (taosHashGetSize(RSMA_INFO_HASH(pRSmaStat)) <= 0) {
-    goto _end;
+    return TSDB_CODE_SUCCESS;
   }
 
   void *infoHash = taosHashIterate(RSMA_INFO_HASH(pRSmaStat), NULL);
   if (!infoHash) {
-    goto _end;
+    return TSDB_CODE_SUCCESS;
   }
 
   STFile tFile = {0};
@@ -1074,9 +1124,17 @@ 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);
+          smaError("vgId:%d, rsma persit, init %s failed since %s", vid, qTaskInfoFName, terrstr());
+          goto _err;
+        }
+        if (tdCreateTFile(&tFile, true, -1) < 0) {
+          smaError("vgId:%d, rsma persit, create %s failed since %s", vid, TD_TFILE_FULL_NAME(&tFile), terrstr());
+          goto _err;
+        }
+        smaDebug("vgId:%d, rsma, table %" PRIi64 " level %d serialize qTaskInfo, file %s created", vid, pRSmaInfo->suid,
+                 i + 1, TD_TFILE_FULL_NAME(&tFile));
 
         isFileCreated = true;
       }
@@ -1101,49 +1159,56 @@ 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:
+  smaError("vgId:%d, rsma persit failed since %s", vid, terrstr());
   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 +1231,8 @@ 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);
@@ -1205,8 +1270,8 @@ static void tdRSmaPersistTrigger(void *param, void *tmrId) {
         // start persist task
         tdRSmaPersistTask(pRSmaStat);
 
-        taosTmrReset(tdRSmaPersistTrigger, RSMA_QTASKINFO_PERSIST_MS, pRSmaStat, pRSmaStat->tmrHandle,
+        // taosTmrReset(tdRSmaPersistTrigger, 5000, pRSmaStat, pRSmaStat->tmrHandle,
-                     &pRSmaStat->tmrId);
+        //              RSMA_TMR_ID(pRSmaStat));
       } else {
         atomic_store_8(RSMA_RUNNING_STAT(pRSmaStat), 0);
       }
@@ -1216,6 +1281,9 @@ 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/smaUtil.c

@@ -140,7 +140,7 @@ int64_t tdAppendTFile(STFile *pTFile, void *buf, int64_t nbyte, int64_t *offset)
     return -1;
   }
 
-#if 1
+#if 0
   smaDebug("append to file %s, offset:%" PRIi64 " nbyte:%" PRIi64 " fsize:%" PRIi64, TD_TFILE_FULL_NAME(pTFile),
            toffset, nbyte, toffset + nbyte);
 #endif
@@ -179,52 +179,85 @@ 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 *pdname, const char *dname, const char *fname, int64_t version,
+                      char *outputName) {
+  if (version < 0) {
+    if (pdname) {
+      snprintf(outputName, TSDB_FILENAME_LEN, "%s%svnode%svnode%d%s%s%sv%d%s", pdname, TD_DIRSEP, TD_DIRSEP, vgId,
+               TD_DIRSEP, dname, TD_DIRSEP, vgId, fname);
+    } else {
+      snprintf(outputName, TSDB_FILENAME_LEN, "vnode%svnode%d%s%s%sv%d%s", TD_DIRSEP, vgId, TD_DIRSEP, dname, TD_DIRSEP,
+               vgId, fname);
+    }
+  } else {
+    if (pdname) {
+      snprintf(outputName, TSDB_FILENAME_LEN, "%s%svnode%svnode%d%s%s%sv%d%s%" PRIi64, pdname, TD_DIRSEP, TD_DIRSEP,
+               vgId, TD_DIRSEP, dname, TD_DIRSEP, vgId, fname, version);
+    } else {
+      snprintf(outputName, TSDB_FILENAME_LEN, "vnode%svnode%d%s%s%sv%d%s%" PRIi64, TD_DIRSEP, vgId, TD_DIRSEP, dname,
+               TD_DIRSEP, vgId, fname, version);
+    }
+  }
 }
 
-int32_t tdInitTFile(STFile *pTFile, STfs *pTfs, const char *fname) {
+void tdGetVndDirName(int32_t vgId, const char *pdname, const char *dname, bool endWithSep, char *outputName) {
-  char    fullname[TSDB_FILENAME_LEN];
+  if (pdname) {
-  SDiskID did = {0};
+    if (endWithSep) {
+      snprintf(outputName, TSDB_FILENAME_LEN, "%s%svnode%svnode%d%s%s%s", pdname, TD_DIRSEP, TD_DIRSEP, vgId, TD_DIRSEP,
+               dname, TD_DIRSEP);
+    } else {
+      snprintf(outputName, TSDB_FILENAME_LEN, "%s%svnode%svnode%d%s%s", pdname, TD_DIRSEP, TD_DIRSEP, vgId, TD_DIRSEP,
+               dname);
+    }
+  } else {
+    if (endWithSep) {
+      snprintf(outputName, TSDB_FILENAME_LEN, "vnode%svnode%d%s%s%s", TD_DIRSEP, vgId, TD_DIRSEP, dname, TD_DIRSEP);
+    } else {
+      snprintf(outputName, TSDB_FILENAME_LEN, "vnode%svnode%d%s%s", TD_DIRSEP, vgId, TD_DIRSEP, dname);
+    }
+  }
+}
 
+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));
+      char *s = strdup(TD_TFILE_FULL_NAME(pTFile));
-      if (tfsMkdirRecurAt(pTfs, taosDirName(s), TD_TFILE_DID(pTFile)) < 0) {
+      if (taosMulMkDir(taosDirName(s)) != 0) {
-        taosMemoryFreeClear(s);
+        terrno = TAOS_SYSTEM_ERROR(errno);
+        taosMemoryFree(s);
         return -1;
       }
-      taosMemoryFreeClear(s);
+      taosMemoryFree(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 {
-      terrno = TAOS_SYSTEM_ERROR(errno);
-      return -1;
     }
   }
 
@@ -244,7 +277,13 @@ int32_t tdCreateTFile(STFile *pTFile, STfs *pTfs, bool updateHeader, int8_t fTyp
   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

@@ -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;
 }
@@ -271,12 +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 {
@@ -302,9 +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;
     }
 
@@ -318,7 +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);
@@ -329,7 +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);
@@ -373,13 +375,9 @@ 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)
-    // 3. get new uid and ts
-
-    tqInfo("retrieve using snapshot req offset: uid %ld ts %ld", dataRsp.reqOffset.uid, dataRsp.reqOffset.ts);
     if (tqScanSnapshot(pTq, &pHandle->execHandle, &dataRsp, fetchOffsetNew, workerId) < 0) {
       ASSERT(0);
     }
@@ -408,199 +406,14 @@ OVER:
   return code;
 }
 
-#if 0
-int32_t tqProcessPollReq(STQ* pTq, SRpcMsg* pMsg, int32_t workerId) {
-  SMqPollReq* pReq = pMsg->pCont;
-  int64_t     consumerId = pReq->consumerId;
-  int64_t     timeout = pReq->timeout;
-  int32_t     reqEpoch = pReq->epoch;
-  int64_t     fetchOffset;
-  int32_t     code = 0;
-
-  // get offset to fetch message
-  if (pReq->currentOffset >= 0) {
-    fetchOffset = pReq->currentOffset + 1;
-  } else {
-    STqOffset* pOffset = tqOffsetRead(pTq->pOffsetStore, pReq->subKey);
-    if (pOffset != NULL) {
-      ASSERT(pOffset->val.type == TMQ_OFFSET__LOG);
-      tqDebug("consumer %ld, restore offset of %s on vg %d, offset(type:log) version: %ld", consumerId, pReq->subKey,
-              TD_VID(pTq->pVnode), pOffset->val.version);
-      fetchOffset = pOffset->val.version + 1;
-    } else {
-      if (pReq->currentOffset == TMQ_CONF__RESET_OFFSET__EARLIEAST) {
-        fetchOffset = walGetFirstVer(pTq->pWal);
-      } else if (pReq->currentOffset == TMQ_CONF__RESET_OFFSET__LATEST) {
-        fetchOffset = walGetCommittedVer(pTq->pWal);
-      } else if (pReq->currentOffset == TMQ_CONF__RESET_OFFSET__NONE) {
-        tqError("tmq poll: no offset committed for consumer %ld in vg %d, subkey %s", consumerId, TD_VID(pTq->pVnode),
-                pReq->subKey);
-        terrno = TSDB_CODE_TQ_NO_COMMITTED_OFFSET;
-        return -1;
-      }
-      tqDebug("consumer %ld, restore offset of %s on vg %d failed, config is %ld, set to %ld", consumerId, pReq->subKey,
-              TD_VID(pTq->pVnode), pReq->currentOffset, fetchOffset);
-    }
-  }
-
-  tqDebug("tmq poll: consumer %ld (epoch %d) recv poll req in vg %d, req offset %ld fetch offset %ld", consumerId,
-          pReq->epoch, TD_VID(pTq->pVnode), pReq->currentOffset, fetchOffset);
-
-  STqHandle* pHandle = taosHashGet(pTq->handles, pReq->subKey, strlen(pReq->subKey));
-  /*ASSERT(pHandle);*/
-  if (pHandle == NULL) {
-    tqError("tmq poll: no consumer handle for consumer %ld in vg %d, subkey %s", consumerId, TD_VID(pTq->pVnode),
-            pReq->subKey);
-    return -1;
-  }
-
-  if (pHandle->consumerId != consumerId) {
-    tqError("tmq poll: consumer handle mismatch for consumer %ld in vg %d, subkey %s, handle consumer id %ld",
-            consumerId, TD_VID(pTq->pVnode), pReq->subKey, pHandle->consumerId);
-    return -1;
-  }
-
-  int32_t consumerEpoch = atomic_load_32(&pHandle->epoch);
-  while (consumerEpoch < reqEpoch) {
-    consumerEpoch = atomic_val_compare_exchange_32(&pHandle->epoch, consumerEpoch, reqEpoch);
-  }
-
-  SMqDataBlkRsp rsp = {0};
-  rsp.reqOffset = pReq->currentOffset;
-
-  rsp.blockData = taosArrayInit(0, sizeof(void*));
-  rsp.blockDataLen = taosArrayInit(0, sizeof(int32_t));
-
-  if (rsp.blockData == NULL || rsp.blockDataLen == NULL) {
-    return -1;
-  }
-
-  rsp.withTbName = pReq->withTbName;
-  if (rsp.withTbName) {
-    rsp.blockTbName = taosArrayInit(0, sizeof(void*));
-  }
-
-  if (pHandle->execHandle.subType == TOPIC_SUB_TYPE__COLUMN) {
-    rsp.withSchema = false;
-  } else {
-    rsp.withSchema = true;
-    rsp.blockSchema = taosArrayInit(0, sizeof(void*));
-  }
-
-#if 1
-  if (pReq->useSnapshot) {
-    // TODO set ver into snapshot
-    int64_t lastVer = walGetCommittedVer(pTq->pWal);
-    if (rsp.reqOffset < lastVer) {
-      tqInfo("retrieve using snapshot req offset %ld last ver %ld", rsp.reqOffset, lastVer);
-      tqScanSnapshot(pTq, &pHandle->execHandle, &rsp, workerId);
-
-      if (rsp.blockNum != 0) {
-        rsp.withTbName = false;
-        rsp.rspOffset = lastVer;
-        tqInfo("direct send by snapshot req offset %ld rsp offset %ld", rsp.reqOffset, rsp.rspOffset);
-        fetchOffset = lastVer;
-        goto SEND_RSP;
-      }
-    }
-  }
-#endif
-
-  SWalHead* pHeadWithCkSum = taosMemoryMalloc(sizeof(SWalHead) + 2048);
-  if (pHeadWithCkSum == NULL) {
-    return -1;
-  }
-
-  walSetReaderCapacity(pHandle->pWalReader, 2048);
-
-  while (1) {
-    consumerEpoch = atomic_load_32(&pHandle->epoch);
-    if (consumerEpoch > reqEpoch) {
-      tqWarn("tmq poll: consumer %ld (epoch %d) vg %d offset %ld, found new consumer epoch %d, discard req epoch %d",
-             consumerId, pReq->epoch, TD_VID(pTq->pVnode), fetchOffset, consumerEpoch, reqEpoch);
-      break;
-    }
-
-    if (tqFetchLog(pTq, pHandle, &fetchOffset, &pHeadWithCkSum) < 0) {
-      // TODO add push mgr
-      break;
-    }
-
-    SWalReadHead* 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);
-
-    if (pHead->msgType == TDMT_VND_SUBMIT) {
-      SSubmitReq* pCont = (SSubmitReq*)&pHead->body;
-
-      if (tqDataExec(pTq, &pHandle->execHandle, pCont, &rsp, workerId) < 0) {
-        /*ASSERT(0);*/
-      }
-    } else {
-      ASSERT(pHandle->fetchMeta);
-      ASSERT(IS_META_MSG(pHead->msgType));
-      tqInfo("fetch meta msg, ver: %ld, type: %d", pHead->version, pHead->msgType);
-      SMqMetaRsp metaRsp = {0};
-      metaRsp.reqOffset = pReq->currentOffset;
-      metaRsp.rspOffset = fetchOffset;
-      metaRsp.resMsgType = pHead->msgType;
-      metaRsp.metaRspLen = pHead->bodyLen;
-      metaRsp.metaRsp = pHead->body;
-      if (tqSendMetaPollRsp(pTq, pMsg, pReq, &metaRsp) < 0) {
-        code = -1;
-        goto OVER;
-      }
-      code = 0;
-      goto OVER;
-    }
-
-    // TODO batch optimization:
-    // TODO continue scan until meeting batch requirement
-    if (rsp.blockNum > 0 /* threshold */) {
-      break;
-    } else {
-      fetchOffset++;
-    }
-  }
-
-  taosMemoryFree(pHeadWithCkSum);
-
-SEND_RSP:
-  ASSERT(taosArrayGetSize(rsp.blockData) == rsp.blockNum);
-  ASSERT(taosArrayGetSize(rsp.blockDataLen) == rsp.blockNum);
-  if (rsp.withSchema) {
-    ASSERT(taosArrayGetSize(rsp.blockSchema) == rsp.blockNum);
-  }
-
-  rsp.rspOffset = fetchOffset;
-
-  if (tqSendDataRsp(pTq, pMsg, pReq, &rsp) < 0) {
-    code = -1;
-  }
-OVER:
-  // TODO wrap in destroy func
-  taosArrayDestroy(rsp.blockDataLen);
-  taosArrayDestroyP(rsp.blockData, (FDelete)taosMemoryFree);
-
-  if (rsp.withSchema) {
-    taosArrayDestroyP(rsp.blockSchema, (FDelete)tDeleteSSchemaWrapper);
-  }
-
-  if (rsp.withTbName) {
-    taosArrayDestroyP(rsp.blockTbName, (FDelete)taosMemoryFree);
-  }
-
-  return code;
-}
-#endif
-
 int32_t tqProcessVgDeleteReq(STQ* pTq, char* msg, int32_t msgLen) {
   SMqVDeleteReq* pReq = (SMqVDeleteReq*)msg;
 
   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);
   }

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/tq/tqSink.c

@@ -185,5 +185,7 @@ void tqTableSink(SStreamTask* pTask, void* vnode, int64_t ver, void* data) {
       .contLen = ntohl(pReq->length),
   };
 
-  ASSERT(tmsgPutToQueue(&pVnode->msgCb, WRITE_QUEUE, &msg) == 0);
+  if (tmsgPutToQueue(&pVnode->msgCb, WRITE_QUEUE, &msg) != 0) {
+    tqDebug("failed to put into write-queue since %s", terrstr());
+  }
 }

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

@@ -69,6 +69,9 @@ int vnodeBegin(SVnode *pVnode) {
     }
   }
 
+  // begin sma
+  smaBegin(pVnode->pSma);  // TODO: refactor to include the rsma1/rsma2 tsdbBegin() after tsdb_refact branch merged
+
   return 0;
 }
 
@@ -229,6 +232,9 @@ int vnodeCommit(SVnode *pVnode) {
     return -1;
   }
 
+  // preCommit
+  smaPreCommit(pVnode->pSma);
+
   // commit each sub-system
   if (metaCommit(pVnode->pMeta) < 0) {
     ASSERT(0);
@@ -269,6 +275,9 @@ int vnodeCommit(SVnode *pVnode) {
   
   pVnode->state.committed = info.state.committed;
 
+  // postCommit
+  smaPostCommit(pVnode->pSma);
+
   // apply the commit (TODO)
   vnodeBufPoolReset(pVnode->onCommit);
   pVnode->onCommit->next = pVnode->pPool;

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

@@ -152,12 +152,11 @@ SVnode *vnodeOpen(const char *path, STfs *pTfs, SMsgCb msgCb) {
   return pVnode;
 
 _err:
-  if (pVnode->pSma) smaCloseEnv(pVnode->pSma);
   if (pVnode->pQuery) vnodeQueryClose(pVnode);
   if (pVnode->pTq) tqClose(pVnode->pTq);
   if (pVnode->pWal) walClose(pVnode->pWal);
   if (pVnode->pTsdb) tsdbClose(&pVnode->pTsdb);
-  if (pVnode->pSma) smaCloseEx(pVnode->pSma);
+  if (pVnode->pSma) smaClose(pVnode->pSma);
   if (pVnode->pMeta) metaClose(pVnode->pMeta);
 
   tsem_destroy(&(pVnode->canCommit));
@@ -167,14 +166,13 @@ _err:
 
 void vnodeClose(SVnode *pVnode) {
   if (pVnode) {
-    smaCloseEnv(pVnode->pSma);
     vnodeCommit(pVnode);
     vnodeSyncClose(pVnode);
     vnodeQueryClose(pVnode);
     walClose(pVnode->pWal);
     tqClose(pVnode->pTq);
     if (pVnode->pTsdb) tsdbClose(&pVnode->pTsdb);
-    smaCloseEx(pVnode->pSma);
+    smaClose(pVnode->pSma);
     metaClose(pVnode->pMeta);
     vnodeCloseBufPool(pVnode);
     // destroy handle

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

@@ -779,6 +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/libs/catalog/inc/catalogInt.h

@@ -278,7 +278,7 @@ typedef struct SCtgAsyncFps {
 
 typedef struct SCtgApiStat {
 
-#ifdef WINDOWS
+#if defined(WINDOWS) || defined(_TD_DARWIN_64)
   size_t avoidCompilationErrors;
 #endif
 

source/libs/catalog/test/CMakeLists.txt

@@ -1,23 +1,25 @@
 
 MESSAGE(STATUS "build catalog unit test")
 
-# GoogleTest requires at least C++11
+IF(NOT TD_DARWIN)
-SET(CMAKE_CXX_STANDARD 11)
+        # GoogleTest requires at least C++11
-AUX_SOURCE_DIRECTORY(${CMAKE_CURRENT_SOURCE_DIR} SOURCE_LIST)
+        SET(CMAKE_CXX_STANDARD 11)
+        AUX_SOURCE_DIRECTORY(${CMAKE_CURRENT_SOURCE_DIR} SOURCE_LIST)
 
-ADD_EXECUTABLE(catalogTest ${SOURCE_LIST})
+        ADD_EXECUTABLE(catalogTest ${SOURCE_LIST})
-TARGET_LINK_LIBRARIES(
+        TARGET_LINK_LIBRARIES(
-        catalogTest
+                catalogTest
-        PUBLIC os util common catalog transport gtest qcom taos_static
+                PUBLIC os util common catalog transport gtest qcom taos_static
-)
+        )
 
-TARGET_INCLUDE_DIRECTORIES(
+        TARGET_INCLUDE_DIRECTORIES(
-        catalogTest
+                catalogTest
-        PUBLIC "${TD_SOURCE_DIR}/include/libs/catalog/"
+                PUBLIC "${TD_SOURCE_DIR}/include/libs/catalog/"
-        PRIVATE "${TD_SOURCE_DIR}/source/libs/catalog/inc"
+                PRIVATE "${TD_SOURCE_DIR}/source/libs/catalog/inc"
-)
+        )
 
-# add_test(
+        # add_test(
-#     NAME catalogTest
+        #     NAME catalogTest
-#     COMMAND catalogTest
+        #     COMMAND catalogTest
-# )
+        # )
+ENDIF()

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

@@ -287,6 +287,7 @@ typedef struct STableScanInfo {
   } lastStatus;
 
   int8_t scanMode;
+  int8_t noTable;
 } STableScanInfo;
 
 typedef struct STagScanInfo {
@@ -355,6 +356,7 @@ typedef struct SStreamBlockScanInfo {
   SUpdateInfo*    pUpdateInfo;
 
   EStreamScanMode scanMode;
+  SOperatorInfo* pStreamScanOp;
   SOperatorInfo* pSnapshotReadOp;
   SArray*        childIds;
   SessionWindowSupporter sessionSup;
@@ -426,7 +428,7 @@ typedef struct SIntervalAggOperatorInfo {
   STimeWindowAggSupp twAggSup;
   bool               invertible;
   SArray*            pPrevValues;        //  SArray<SGroupKeys> used to keep the previous not null value for interpolation.
-  bool               ignoreCloseWindow;
+  bool               ignoreExpiredData;
 } SIntervalAggOperatorInfo;
 
 typedef struct SStreamFinalIntervalOperatorInfo {
@@ -448,7 +450,7 @@ typedef struct SStreamFinalIntervalOperatorInfo {
   SArray*            pPullWins;          // SPullWindowInfo
   int32_t            pullIndex;
   SSDataBlock*       pPullDataRes;
-  bool               ignoreCloseWindow;
+  bool               ignoreExpiredData;
 } SStreamFinalIntervalOperatorInfo;
 
 typedef struct SAggOperatorInfo {
@@ -586,7 +588,7 @@ typedef struct SStreamSessionAggOperatorInfo {
   SArray*              pChildren;       // cache for children's result; final stream operator
   SPhysiNode*          pPhyNode;           // create new child
   bool                 isFinal;
-  bool                 ignoreCloseWindow;
+  bool                 ignoreExpiredData;
 } SStreamSessionAggOperatorInfo;
 
 typedef struct STimeSliceOperatorInfo {
@@ -630,7 +632,7 @@ typedef struct SStreamStateAggOperatorInfo {
   void*                pDelIterator;
   SArray*              pScanWindow;
   SArray*              pChildren;       // cache for children's result;
-  bool                 ignoreCloseWindow;
+  bool                 ignoreExpiredData;
 } SStreamStateAggOperatorInfo;
 
 typedef struct SSortedMergeOperatorInfo {

source/libs/executor/src/dataSinkMgt.c

@@ -35,7 +35,7 @@ int32_t dsDataSinkGetCacheSize(SDataSinkStat *pStat) {
 
 
 int32_t dsCreateDataSinker(const SDataSinkNode *pDataSink, DataSinkHandle* pHandle, void* pParam) {
-  switch (nodeType(pDataSink)) {
+  switch ((int)nodeType(pDataSink)) {
     case QUERY_NODE_PHYSICAL_PLAN_DISPATCH:
       return createDataDispatcher(&gDataSinkManager, pDataSink, pHandle);
     case QUERY_NODE_PHYSICAL_PLAN_DELETE:

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

@@ -235,9 +235,11 @@ int32_t qStreamPrepareScan(qTaskInfo_t tinfo, uint64_t uid, int64_t ts) {
   SExecTaskInfo* pTaskInfo = (SExecTaskInfo*)tinfo;
 
   if (uid == 0) {
-    STableKeyInfo* pTableInfo = taosArrayGet(pTaskInfo->tableqinfoList.pTableList, 0);
+    if (taosArrayGetSize(pTaskInfo->tableqinfoList.pTableList) != 0) {
-    uid = pTableInfo->uid;
+      STableKeyInfo* pTableInfo = taosArrayGet(pTaskInfo->tableqinfoList.pTableList, 0);
-    ts = INT64_MIN;
+      uid = pTableInfo->uid;
+      ts = INT64_MIN;
+    }
   }
 
   return doPrepareScan(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));
@@ -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,
@@ -2010,8 +2027,9 @@ 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 ", execId:%d, %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, pSource->execId, 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);
@@ -2146,8 +2164,8 @@ 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 " execId:%d 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, pSource->execId, i, pDataInfo->totalRows,
                pExchangeInfo->loadInfo.totalRows, completed + 1, i + 1, totalSources);
         pDataInfo->status = EX_SOURCE_DATA_EXHAUSTED;
@@ -2166,18 +2184,19 @@ static SSDataBlock* concurrentlyLoadRemoteDataImpl(SOperatorInfo* pOperator, SEx
       }
 
       if (pRsp->completed == 1) {
-        qDebug("%s fetch msg rsp from vgId:%d, taskId:0x%" PRIx64 " execId:%d"
+        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, pSource->execId, 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 " execId:%d numOfRows:%d, totalRows:%" PRIu64
                ", totalBytes:%" PRIu64,
-               GET_TASKID(pTaskInfo), pSource->addr.nodeId, pSource->taskId, pSource->execId, 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);
@@ -2250,8 +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 " execId:%d 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, pSource->execId, tstrerror(pDataInfo->code));
+             pSource->addr.nodeId, pSource->taskId, pSource->execId, tstrerror(pDataInfo->code));
       pOperator->pTaskInfo->code = pDataInfo->code;
       return NULL;
     }
@@ -2259,8 +2278,8 @@ static SSDataBlock* seqLoadRemoteData(SOperatorInfo* pOperator) {
     SRetrieveTableRsp*   pRsp = pDataInfo->pRsp;
     SLoadRemoteDataInfo* pLoadInfo = &pExchangeInfo->loadInfo;
     if (pRsp->numOfRows == 0) {
-      qDebug("%s vgId:%d, taskID:0x%" PRIx64 " execId:%d %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, pSource->execId, pExchangeInfo->current + 1,
              pDataInfo->totalRows, pLoadInfo->totalRows);
 
@@ -2279,16 +2298,17 @@ static SSDataBlock* seqLoadRemoteData(SOperatorInfo* pOperator) {
     if (pRsp->completed == 1) {
       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, pSource->execId, 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 " execId:%d numOfRows:%d, totalRows:%" PRIu64
              ", totalBytes:%" PRIu64,
-             GET_TASKID(pTaskInfo), pSource->addr.nodeId, pSource->taskId, pSource->execId, 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;
@@ -2373,8 +2393,7 @@ 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;
   }
 
@@ -2829,22 +2848,46 @@ int32_t doPrepareScan(SOperatorInfo* pOperator, uint64_t uid, int64_t ts) {
     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;
+      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;
+
+      SExecTaskInfo* pTaskInfo = pOperator->pTaskInfo;
+
+      int32_t tableSz = taosArrayGetSize(pTaskInfo->tableqinfoList.pTableList);
+      bool    found = false;
+      for (int32_t i = 0; i < tableSz; i++) {
+        STableKeyInfo* pTableInfo = taosArrayGet(pTaskInfo->tableqinfoList.pTableList, i);
+        if (pTableInfo->uid == uid) {
+          found = true;
+          pInfo->currentTable = i;
+        }
+      }
+      // TODO after processing drop,
+      ASSERT(found);
+      qDebug("tsdb reader offset seek to uid %ld ts %ld, table cur set to %d , all table num %d", uid, ts,
+             pInfo->currentTable, tableSz);
     }
 
+    return TSDB_CODE_SUCCESS;
+
   } else {
     if (pOperator->numOfDownstream == 1) {
       return doPrepareScan(pOperator->pDownstream[0], uid, ts);
@@ -2856,8 +2899,6 @@ int32_t doPrepareScan(SOperatorInfo* pOperator, uint64_t uid, int64_t ts) {
       return TSDB_CODE_QRY_APP_ERROR;
     }
   }
-
-  return TSDB_CODE_SUCCESS;
 }
 
 int32_t doGetScanStatus(SOperatorInfo* pOperator, uint64_t* uid, int64_t* ts) {
@@ -3160,8 +3201,9 @@ 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;
     }
@@ -3228,6 +3270,10 @@ static SSDataBlock* doProjectOperation(SOperatorInfo* pOperator) {
       doSetOperatorCompleted(pOperator);
       break;
     }
+    if (pBlock->info.type == STREAM_RETRIEVE) {
+      // for stream interval
+      return pBlock;
+    }
 
     // the pDataBlock are always the same one, no need to call this again
     int32_t code = getTableScanInfo(pOperator->pDownstream[0], &order, &scanFlag);
@@ -3235,8 +3281,6 @@ 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);
 
@@ -3247,6 +3291,10 @@ 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) {
@@ -3473,11 +3521,12 @@ 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;
   }
 }
 
@@ -3977,12 +4026,12 @@ 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;
@@ -4062,19 +4111,27 @@ int32_t generateGroupIdMap(STableListInfo* pTableListInfo, SReadHandle* pHandle,
       } else {
         taosMemoryFree(keyBuf);
         nodesClearList(groupNew);
+        metaReaderClear(&mr);
         return code;
       }
 
       ASSERT(nodeType(pNew) == QUERY_NODE_VALUE);
       SValueNode* pValue = (SValueNode*)pNew;
 
-      if (pValue->node.resType.type == TSDB_DATA_TYPE_NULL) {
+      if (pValue->node.resType.type == TSDB_DATA_TYPE_NULL || pValue->isNull) {
         isNull[index++] = 1;
         continue;
       } else {
         isNull[index++] = 0;
         char* data = nodesGetValueFromNode(pValue);
         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;
@@ -4134,6 +4191,7 @@ SOperatorInfo* createOperatorTree(SPhysiNode* pPhyNode, SExecTaskInfo* pTaskInfo
       STableMergeScanPhysiNode* pTableScanNode = (STableMergeScanPhysiNode*)pPhyNode;
       int32_t code = createScanTableListInfo(pTableScanNode, pHandle, pTableListInfo, queryId, taskId);
       if (code) {
+        pTaskInfo->code = code;
         return NULL;
       }
       code = extractTableSchemaVersion(pHandle, pTableScanNode->scan.uid, pTaskInfo);
@@ -4160,7 +4218,11 @@ 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

@@ -518,6 +518,7 @@ static SSDataBlock* doTableScan(SOperatorInfo* pOperator) {
 
   // 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) {
@@ -806,6 +807,23 @@ static bool isStateWindow(SStreamBlockScanInfo* pInfo) {
   return pInfo->sessionSup.parentType == QUERY_NODE_PHYSICAL_PLAN_STREAM_STATE;
 }
 
+static void setGroupId(SStreamBlockScanInfo* pInfo, SSDataBlock* pBlock, int32_t groupColIndex, int32_t rowIndex) {
+  ASSERT(rowIndex < pBlock->info.rows);
+  switch (pBlock->info.type)
+  {
+  case STREAM_RETRIEVE: {
+    SColumnInfoData* pColInfo = taosArrayGet(pBlock->pDataBlock, groupColIndex);
+    uint64_t* groupCol = (uint64_t*)pColInfo->pData;
+    pInfo->groupId = groupCol[rowIndex];
+  }
+    break;
+  case STREAM_DELETE_DATA:
+    break;
+  default:
+    break;
+  }
+}
+
 static bool prepareDataScan(SStreamBlockScanInfo* pInfo, SSDataBlock* pSDB, int32_t tsColIndex, int32_t* pRowIndex) {
   STimeWindow win = {
       .skey = INT64_MIN,
@@ -828,6 +846,7 @@ static bool prepareDataScan(SStreamBlockScanInfo* pInfo, SSDataBlock* pSDB, int3
     } else {
       win =
           getActiveTimeWindow(NULL, &dumyInfo, tsCols[(*pRowIndex)], &pInfo->interval, pInfo->interval.precision, NULL);
+      setGroupId(pInfo, pSDB, 2, *pRowIndex);
       (*pRowIndex) += getNumOfRowsInTimeWindow(&pSDB->info, tsCols, (*pRowIndex), win.ekey, binarySearchForKey, NULL,
                                                TSDB_ORDER_ASC);
     }
@@ -1030,10 +1049,12 @@ static SSDataBlock* doStreamBlockScan(SOperatorInfo* pOperator) {
       }
       pInfo->scanMode = STREAM_SCAN_FROM_DATAREADER;
     } else {
-      if (isStateWindow(pInfo) && taosArrayGetSize(pInfo->sessionSup.pStreamAggSup->pScanWindow) > 0) {
+      if (isStateWindow(pInfo)) {
         pInfo->scanMode = STREAM_SCAN_FROM_DATAREADER;
         pInfo->updateResIndex = pInfo->pUpdateRes->info.rows;
-        prepareDataScan(pInfo, pInfo->pUpdateRes, pInfo->primaryTsIndex, &pInfo->updateResIndex);
+        if (!prepareDataScan(pInfo, pInfo->pUpdateRes, pInfo->primaryTsIndex, &pInfo->updateResIndex)) {
+          pInfo->scanMode = STREAM_SCAN_FROM_READERHANDLE;
+        }
       }
       if (pInfo->scanMode == STREAM_SCAN_FROM_DATAREADER) {
         SSDataBlock* pSDB = doDataScan(pInfo, pInfo->pUpdateRes, pInfo->primaryTsIndex, &pInfo->updateResIndex);
@@ -1273,6 +1294,7 @@ 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;
@@ -1307,6 +1329,13 @@ static void destroySysScanOperator(void* param, int32_t numOfOutput) {
   taosArrayDestroy(pInfo->scanCols);
 }
 
+static int32_t getSysTableDbNameColId(const char* pTable) {
+  // if (0 == strcmp(TSDB_INS_TABLE_USER_INDEXES, pTable)) {
+  //   return 1;
+  // }
+  return TSDB_INS_USER_STABLES_DBNAME_COLID;
+}
+
 EDealRes getDBNameFromConditionWalker(SNode* pNode, void* pContext) {
   int32_t   code = TSDB_CODE_SUCCESS;
   ENodeType nType = nodeType(pNode);
@@ -1328,7 +1357,7 @@ EDealRes getDBNameFromConditionWalker(SNode* pNode, void* pContext) {
       }
 
       SColumnNode* node = (SColumnNode*)pNode;
-      if (TSDB_INS_USER_STABLES_DBNAME_COLID == node->colId) {
+      if (getSysTableDbNameColId(node->tableName) == node->colId) {
         *(int32_t*)pContext = 2;
         return DEAL_RES_CONTINUE;
       }
@@ -2089,6 +2118,7 @@ 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

@@ -838,7 +838,7 @@ static void hashIntervalAgg(SOperatorInfo* pOperatorInfo, SResultRowInfo* pResul
   STimeWindow win = getActiveTimeWindow(pInfo->aggSup.pResultBuf, pResultRowInfo, ts, &pInfo->interval,
                                         pInfo->interval.precision, &pInfo->win);
   int32_t     ret = TSDB_CODE_SUCCESS;
-  if (!pInfo->ignoreCloseWindow || !isCloseWindow(&win, &pInfo->twAggSup)) {
+  if (!pInfo->ignoreExpiredData || !isCloseWindow(&win, &pInfo->twAggSup)) {
     ret = setTimeWindowOutputBuf(pResultRowInfo, &win, (scanFlag == MAIN_SCAN), &pResult, tableGroupId, pSup->pCtx,
                                  numOfOutput, pSup->rowEntryInfoOffset, &pInfo->aggSup, pTaskInfo);
     if (ret != TSDB_CODE_SUCCESS || pResult == NULL) {
@@ -871,7 +871,7 @@ static void hashIntervalAgg(SOperatorInfo* pOperatorInfo, SResultRowInfo* pResul
     doWindowBorderInterpolation(pInfo, pBlock, pResult, &win, startPos, forwardRows, pSup);
   }
 
-  if (!pInfo->ignoreCloseWindow || !isCloseWindow(&win, &pInfo->twAggSup)) {
+  if (!pInfo->ignoreExpiredData || !isCloseWindow(&win, &pInfo->twAggSup)) {
     updateTimeWindowInfo(&pInfo->twAggSup.timeWindowData, &win, true);
     doApplyFunctions(pTaskInfo, pSup->pCtx, &win, &pInfo->twAggSup.timeWindowData, startPos, forwardRows, tsCols,
                      pBlock->info.rows, numOfOutput, pInfo->order);
@@ -886,7 +886,7 @@ static void hashIntervalAgg(SOperatorInfo* pOperatorInfo, SResultRowInfo* pResul
     if (startPos < 0) {
       break;
     }
-    if (pInfo->ignoreCloseWindow && isCloseWindow(&nextWin, &pInfo->twAggSup)) {
+    if (pInfo->ignoreExpiredData && isCloseWindow(&nextWin, &pInfo->twAggSup)) {
       ekey = ascScan ? nextWin.ekey : nextWin.skey;
       forwardRows =
           getNumOfRowsInTimeWindow(&pBlock->info, tsCols, startPos, ekey, binarySearchForKey, NULL, pInfo->order);
@@ -1535,7 +1535,7 @@ SOperatorInfo* createIntervalOperatorInfo(SOperatorInfo* downstream, SExprInfo*
   pInfo->interval = *pInterval;
   pInfo->execModel = pTaskInfo->execModel;
   pInfo->twAggSup = *pTwAggSupp;
-  pInfo->ignoreCloseWindow = false;
+  pInfo->ignoreExpiredData = pPhyNode->window.igExpired;
 
   if (pPhyNode->window.pExprs != NULL) {
     int32_t    numOfScalar = 0;
@@ -2292,7 +2292,7 @@ static void doHashInterval(SOperatorInfo* pOperatorInfo, SSDataBlock* pSDataBloc
                                             pInfo->interval.precision, NULL);
   while (1) {
     bool isClosed = isCloseWindow(&nextWin, &pInfo->twAggSup);
-    if (pInfo->ignoreCloseWindow && isClosed) {
+    if (pInfo->ignoreExpiredData && isClosed) {
       startPos = getNexWindowPos(&pInfo->interval, &pSDataBlock->info, tsCols, startPos, nextWin.ekey, &nextWin);
       if (startPos < 0) {
         break;
@@ -2710,7 +2710,7 @@ SOperatorInfo* createStreamFinalIntervalOperatorInfo(SOperatorInfo* downstream,
   _hash_fn_t hashFn = taosGetDefaultHashFunction(TSDB_DATA_TYPE_BINARY);
   pInfo->pPullDataMap = taosHashInit(64, hashFn, false, HASH_NO_LOCK);
   pInfo->pPullDataRes = createPullDataBlock();
-  pInfo->ignoreCloseWindow = false;
+  pInfo->ignoreExpiredData = pIntervalPhyNode->window.igExpired;
 
   pOperator->operatorType = pPhyNode->type;
   pOperator->blocking = true;
@@ -2852,12 +2852,12 @@ SOperatorInfo* createStreamSessionAggOperatorInfo(SOperatorInfo* downstream, SPh
   pInfo->pStDeleted = taosHashInit(64, hashFn, true, HASH_NO_LOCK);
   pInfo->pDelIterator = NULL;
   pInfo->pDelRes = createOneDataBlock(pResBlock, false);
-  pInfo->pDelRes->info.type = STREAM_DELETE;
+  pInfo->pDelRes->info.type = STREAM_DELETE_RESULT;
   blockDataEnsureCapacity(pInfo->pDelRes, 64);
   pInfo->pChildren = NULL;
   pInfo->isFinal = false;
   pInfo->pPhyNode = pPhyNode;
-  pInfo->ignoreCloseWindow = false;
+  pInfo->ignoreExpiredData = pSessionNode->window.igExpired;
 
   pOperator->name = "StreamSessionWindowAggOperator";
   pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_STREAM_SESSION;
@@ -3133,7 +3133,7 @@ static void doStreamSessionAggImpl(SOperatorInfo* pOperator, SSDataBlock* pSData
 
   SStreamAggSupporter* pAggSup = &pInfo->streamAggSup;
   for (int32_t i = 0; i < pSDataBlock->info.rows;) {
-    if (pInfo->ignoreCloseWindow && isOverdue(endTsCols[i], &pInfo->twAggSup)) {
+    if (pInfo->ignoreExpiredData && isOverdue(endTsCols[i], &pInfo->twAggSup)) {
       i++;
       continue;
     }
@@ -3413,8 +3413,8 @@ static SSDataBlock* doStreamSessionAgg(SOperatorInfo* pOperator) {
   pOperator->status = OP_RES_TO_RETURN;
 
   closeSessionWindow(pInfo->streamAggSup.pResultRows, &pInfo->twAggSup, pUpdated,
-      getResWinForSession, pInfo->ignoreCloseWindow);
+      getResWinForSession, pInfo->ignoreExpiredData);
-  closeChildSessionWindow(pInfo->pChildren, pInfo->twAggSup.maxTs, pInfo->ignoreCloseWindow);
+  closeChildSessionWindow(pInfo->pChildren, pInfo->twAggSup.maxTs, pInfo->ignoreExpiredData);
   copyUpdateResult(pStUpdated, pUpdated);
   taosHashCleanup(pStUpdated);
 
@@ -3822,7 +3822,7 @@ static void doStreamStateAggImpl(SOperatorInfo* pOperator, SSDataBlock* pSDataBl
   SStreamAggSupporter* pAggSup = &pInfo->streamAggSup;
   SColumnInfoData*     pKeyColInfo = taosArrayGet(pSDataBlock->pDataBlock, pInfo->stateCol.slotId);
   for (int32_t i = 0; i < pSDataBlock->info.rows; i += winRows) {
-    if (pInfo->ignoreCloseWindow && isOverdue(tsCols[i], &pInfo->twAggSup)) {
+    if (pInfo->ignoreExpiredData && isOverdue(tsCols[i], &pInfo->twAggSup)) {
       i++;
       continue;
     }
@@ -3866,12 +3866,14 @@ static SSDataBlock* doStreamStateAgg(SOperatorInfo* pOperator) {
   if (pOperator->status == OP_RES_TO_RETURN) {
     doBuildDeleteDataBlock(pInfo->pSeDeleted, pInfo->pDelRes, &pInfo->pDelIterator);
     if (pInfo->pDelRes->info.rows > 0) {
+      printDataBlock(pInfo->pDelRes, "single state");
       return pInfo->pDelRes;
     }
     doBuildResultDatablock(pOperator, pBInfo, &pInfo->groupResInfo, pInfo->streamAggSup.pResultBuf);
     if (pBInfo->pRes->info.rows == 0 || !hasDataInGroupInfo(&pInfo->groupResInfo)) {
       doSetOperatorCompleted(pOperator);
     }
+    printDataBlock(pBInfo->pRes, "single state");
     return pBInfo->pRes->info.rows == 0 ? NULL : pBInfo->pRes;
   }
 
@@ -3884,6 +3886,7 @@ static SSDataBlock* doStreamStateAgg(SOperatorInfo* pOperator) {
     if (pBlock == NULL) {
       break;
     }
+    printDataBlock(pBlock, "single state recv");
 
     if (pBlock->info.type == STREAM_CLEAR) {
       doClearStateWindows(&pInfo->streamAggSup, pBlock, pInfo->primaryTsIndex, &pInfo->stateCol, pInfo->stateCol.slotId,
@@ -3903,8 +3906,8 @@ static SSDataBlock* doStreamStateAgg(SOperatorInfo* pOperator) {
   pOperator->status = OP_RES_TO_RETURN;
 
   closeSessionWindow(pInfo->streamAggSup.pResultRows, &pInfo->twAggSup, pUpdated,
-      getResWinForState, pInfo->ignoreCloseWindow);
+      getResWinForState, pInfo->ignoreExpiredData);
-  closeChildSessionWindow(pInfo->pChildren, pInfo->twAggSup.maxTs, pInfo->ignoreCloseWindow);
+  closeChildSessionWindow(pInfo->pChildren, pInfo->twAggSup.maxTs, pInfo->ignoreExpiredData);
   copyUpdateResult(pSeUpdated, pUpdated);
   taosHashCleanup(pSeUpdated);
 
@@ -3914,9 +3917,11 @@ static SSDataBlock* doStreamStateAgg(SOperatorInfo* pOperator) {
   blockDataEnsureCapacity(pInfo->binfo.pRes, pOperator->resultInfo.capacity);
   doBuildDeleteDataBlock(pInfo->pSeDeleted, pInfo->pDelRes, &pInfo->pDelIterator);
   if (pInfo->pDelRes->info.rows > 0) {
+    printDataBlock(pInfo->pDelRes, "single state");
     return pInfo->pDelRes;
   }
   doBuildResultDatablock(pOperator, &pInfo->binfo, &pInfo->groupResInfo, pInfo->streamAggSup.pResultBuf);
+  printDataBlock(pBInfo->pRes, "single state");
   return pBInfo->pRes->info.rows == 0 ? NULL : pBInfo->pRes;
 }
 
@@ -3975,10 +3980,10 @@ SOperatorInfo* createStreamStateAggOperatorInfo(SOperatorInfo* downstream, SPhys
   pInfo->pSeDeleted = taosHashInit(64, hashFn, true, HASH_NO_LOCK);
   pInfo->pDelIterator = NULL;
   pInfo->pDelRes = createOneDataBlock(pResBlock, false);
-  pInfo->pDelRes->info.type = STREAM_DELETE;
+  pInfo->pDelRes->info.type = STREAM_DELETE_RESULT;
   blockDataEnsureCapacity(pInfo->pDelRes, 64);
   pInfo->pChildren = NULL;
-  pInfo->ignoreCloseWindow = false;
+  pInfo->ignoreExpiredData = pStateNode->window.igExpired;
 
   pOperator->name = "StreamStateAggOperator";
   pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_STREAM_STATE;

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

@@ -39,6 +39,174 @@ static int32_t invaildFuncParaValueErrMsg(char* pErrBuf, int32_t len, const char
   return buildFuncErrMsg(pErrBuf, len, TSDB_CODE_FUNC_FUNTION_PARA_VALUE, "Invalid parameter value : %s", pFuncName);
 }
 
+#define TIME_UNIT_INVALID 1
+#define TIME_UNIT_TOO_SMALL 2
+
+static int32_t validateTimeUnitParam(uint8_t dbPrec, const SValueNode* pVal) {
+  if (!pVal->isDuration) {
+    return TIME_UNIT_INVALID;
+  }
+
+  if (TSDB_TIME_PRECISION_MILLI == dbPrec && 0 == strcasecmp(pVal->literal, "1u")) {
+    return TIME_UNIT_TOO_SMALL;
+  }
+
+  if (pVal->literal[0] != '1' || (pVal->literal[1] != 'u' && pVal->literal[1] != 'a' &&
+                                  pVal->literal[1] != 's' && pVal->literal[1] != 'm' &&
+                                  pVal->literal[1] != 'h' && pVal->literal[1] != 'd' &&
+                                  pVal->literal[1] != 'w')) {
+    return TIME_UNIT_INVALID;
+  }
+
+  return TSDB_CODE_SUCCESS;
+}
+
+/* Following are valid ISO-8601 timezone format:
+ * 1 z/Z
+ * 2 ±hh:mm
+ * 3 ±hhmm
+ * 4 ±hh
+ *
+ */
+
+static bool validateHourRange(int8_t hour) {
+  if (hour < 0 || hour > 12) {
+    return false;
+  }
+
+  return true;
+}
+
+static bool validateMinuteRange(int8_t hour, int8_t minute, char sign) {
+  if (minute == 0 || (minute == 30 && (hour == 3 || hour == 5) && sign == '+')) {
+    return true;
+  }
+
+  return false;
+}
+
+static bool validateTimestampDigits(const SValueNode* pVal) {
+  if (!IS_INTEGER_TYPE(pVal->node.resType.type)) {
+    return false;
+  }
+
+  int64_t tsVal = pVal->datum.i;
+  char    fraction[20] = {0};
+  NUM_TO_STRING(pVal->node.resType.type, &tsVal, sizeof(fraction), fraction);
+  int32_t tsDigits = (int32_t)strlen(fraction);
+
+  if (tsDigits > TSDB_TIME_PRECISION_SEC_DIGITS) {
+    if (tsDigits == TSDB_TIME_PRECISION_MILLI_DIGITS || tsDigits == TSDB_TIME_PRECISION_MICRO_DIGITS ||
+        tsDigits == TSDB_TIME_PRECISION_NANO_DIGITS) {
+      return true;
+    } else {
+      return false;
+    }
+  }
+
+  return true;
+}
+
+static bool validateTimezoneFormat(const SValueNode* pVal) {
+  if (TSDB_DATA_TYPE_BINARY != pVal->node.resType.type) {
+    return false;
+  }
+
+  char*   tz = varDataVal(pVal->datum.p);
+  int32_t len = varDataLen(pVal->datum.p);
+
+  char   buf[3] = {0};
+  int8_t hour = -1, minute = -1;
+  if (len == 0) {
+    return false;
+  } else if (len == 1 && (tz[0] == 'z' || tz[0] == 'Z')) {
+    return true;
+  } else if ((tz[0] == '+' || tz[0] == '-')) {
+    switch (len) {
+      case 3:
+      case 5: {
+        for (int32_t i = 1; i < len; ++i) {
+          if (!isdigit(tz[i])) {
+            return false;
+          }
+
+          if (i == 2) {
+            memcpy(buf, &tz[i - 1], 2);
+            hour = taosStr2Int8(buf, NULL, 10);
+            if (!validateHourRange(hour)) {
+              return false;
+            }
+          } else if (i == 4) {
+            memcpy(buf, &tz[i - 1], 2);
+            minute = taosStr2Int8(buf, NULL, 10);
+            if (!validateMinuteRange(hour, minute, tz[0])) {
+              return false;
+            }
+          }
+        }
+        break;
+      }
+      case 6: {
+        for (int32_t i = 1; i < len; ++i) {
+          if (i == 3) {
+            if (tz[i] != ':') {
+              return false;
+            }
+            continue;
+          }
+
+          if (!isdigit(tz[i])) {
+            return false;
+          }
+
+          if (i == 2) {
+            memcpy(buf, &tz[i - 1], 2);
+            hour = taosStr2Int8(buf, NULL, 10);
+            if (!validateHourRange(hour)) {
+              return false;
+            }
+          } else if (i == 5) {
+            memcpy(buf, &tz[i - 1], 2);
+            minute = taosStr2Int8(buf, NULL, 10);
+            if (!validateMinuteRange(hour, minute, tz[0])) {
+              return false;
+            }
+          }
+        }
+        break;
+      }
+      default: {
+        return false;
+      }
+    }
+  } else {
+    return false;
+  }
+
+  return true;
+}
+
+void static addTimezoneParam(SNodeList* pList) {
+  char       buf[6] = {0};
+  time_t     t = taosTime(NULL);
+  struct tm* tmInfo = taosLocalTime(&t, NULL);
+  strftime(buf, sizeof(buf), "%z", tmInfo);
+  int32_t len = (int32_t)strlen(buf);
+
+  SValueNode* pVal = (SValueNode*)nodesMakeNode(QUERY_NODE_VALUE);
+  pVal->literal = strndup(buf, len);
+  pVal->isDuration = false;
+  pVal->translate = true;
+  pVal->node.resType.type = TSDB_DATA_TYPE_BINARY;
+  pVal->node.resType.bytes = len + VARSTR_HEADER_SIZE;
+  pVal->node.resType.precision = TSDB_TIME_PRECISION_MILLI;
+  pVal->datum.p = taosMemoryCalloc(1, len + VARSTR_HEADER_SIZE + 1);
+  varDataSetLen(pVal->datum.p, len);
+  strncpy(varDataVal(pVal->datum.p), pVal->literal, len);
+
+  nodesListAppend(pList, (SNode*)pVal);
+}
+
 void static addDbPrecisonParam(SNodeList** pList, uint8_t precision) {
   SValueNode* pVal = (SValueNode*)nodesMakeNode(QUERY_NODE_VALUE);
   pVal->literal = NULL;
@@ -238,7 +406,7 @@ 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);
 
@@ -525,9 +693,15 @@ static int32_t translateElapsed(SFunctionNode* pFunc, char* pErrBuf, int32_t len
       return invaildFuncParaTypeErrMsg(pErrBuf, len, pFunc->functionName);
     }
 
-    if (pValue->datum.i == 0) {
+    uint8_t dbPrec = pFunc->node.resType.precision;
+
+    int32_t ret = validateTimeUnitParam(dbPrec, (SValueNode *)nodesListGetNode(pFunc->pParameterList, 1));
+    if (ret == TIME_UNIT_TOO_SMALL) {
       return buildFuncErrMsg(pErrBuf, len, TSDB_CODE_FUNC_FUNTION_ERROR,
                              "ELAPSED function time unit parameter should be greater than db precision");
+    } else if (ret == TIME_UNIT_INVALID) {
+      return buildFuncErrMsg(pErrBuf, len, TSDB_CODE_FUNC_FUNTION_ERROR,
+                             "ELAPSED function time unit parameter should be one of the following: [1u, 1a, 1s, 1m, 1h, 1d, 1w]");
     }
   }
 
@@ -634,6 +808,12 @@ 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)) {
@@ -643,12 +823,11 @@ 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};
@@ -668,6 +847,12 @@ 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)) {
@@ -677,12 +862,11 @@ 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 =
@@ -833,6 +1017,19 @@ static int32_t translateStateDuration(SFunctionNode* pFunc, char* pErrBuf, int32
     return invaildFuncParaTypeErrMsg(pErrBuf, len, pFunc->functionName);
   }
 
+  if (numOfParams == 4) {
+    uint8_t dbPrec = pFunc->node.resType.precision;
+
+    int32_t ret = validateTimeUnitParam(dbPrec, (SValueNode *)nodesListGetNode(pFunc->pParameterList, 3));
+    if (ret == TIME_UNIT_TOO_SMALL) {
+      return buildFuncErrMsg(pErrBuf, len, TSDB_CODE_FUNC_FUNTION_ERROR,
+                             "STATEDURATION function time unit parameter should be greater than db precision");
+    } else if (ret == TIME_UNIT_INVALID) {
+      return buildFuncErrMsg(pErrBuf, len, TSDB_CODE_FUNC_FUNTION_ERROR,
+                             "STATEDURATION function time unit parameter should be one of the following: [1u, 1a, 1s, 1m, 1h, 1d, 1w]");
+    }
+  }
+
   // set result type
   pFunc->node.resType = (SDataType){.bytes = tDataTypes[TSDB_DATA_TYPE_BIGINT].bytes, .type = TSDB_DATA_TYPE_BIGINT};
   return TSDB_CODE_SUCCESS;
@@ -1017,7 +1214,7 @@ static int32_t translateIrate(SFunctionNode* pFunc, char* pErrBuf, int32_t len)
     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);
 
@@ -1031,13 +1228,7 @@ 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;
 }
 
@@ -1051,11 +1242,6 @@ 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 {
@@ -1260,19 +1446,9 @@ static int32_t translateSubstr(SFunctionNode* pFunc, char* pErrBuf, int32_t len)
 
 static int32_t translateCast(SFunctionNode* pFunc, char* pErrBuf, int32_t len) {
   // The number of parameters has been limited by the syntax definition
-  // uint8_t para1Type = ((SExprNode*)nodesListGetNode(pFunc->pParameterList, 0))->resType.type;
 
   // The function return type has been set during syntax parsing
   uint8_t para2Type = pFunc->node.resType.type;
-  // if (para2Type != TSDB_DATA_TYPE_BIGINT && para2Type != TSDB_DATA_TYPE_UBIGINT &&
-  //     para2Type != TSDB_DATA_TYPE_VARCHAR && para2Type != TSDB_DATA_TYPE_NCHAR &&
-  //     para2Type != TSDB_DATA_TYPE_TIMESTAMP) {
-  //   return invaildFuncParaTypeErrMsg(pErrBuf, len, pFunc->functionName);
-  // }
-  // if ((para2Type == TSDB_DATA_TYPE_TIMESTAMP && IS_VAR_DATA_TYPE(para1Type)) ||
-  //     (para2Type == TSDB_DATA_TYPE_BINARY && para1Type == TSDB_DATA_TYPE_NCHAR)) {
-  //   return invaildFuncParaTypeErrMsg(pErrBuf, len, pFunc->functionName);
-  // }
 
   int32_t para2Bytes = pFunc->node.resType.bytes;
   if (IS_VAR_DATA_TYPE(para2Type)) {
@@ -1282,155 +1458,14 @@ static int32_t translateCast(SFunctionNode* pFunc, char* pErrBuf, int32_t len) {
     return buildFuncErrMsg(pErrBuf, len, TSDB_CODE_FUNC_FUNTION_ERROR,
                            "CAST function converted length should be in range [0, 1000]");
   }
+
+  // add database precision as param
+  uint8_t dbPrec = pFunc->node.resType.precision;
+  addDbPrecisonParam(&pFunc->pParameterList, dbPrec);
+
   return TSDB_CODE_SUCCESS;
 }
 
-/* Following are valid ISO-8601 timezone format:
- * 1 z/Z
- * 2 ±hh:mm
- * 3 ±hhmm
- * 4 ±hh
- *
- */
-
-static bool validateHourRange(int8_t hour) {
-  if (hour < 0 || hour > 12) {
-    return false;
-  }
-
-  return true;
-}
-
-static bool validateMinuteRange(int8_t hour, int8_t minute, char sign) {
-  if (minute == 0 || (minute == 30 && (hour == 3 || hour == 5) && sign == '+')) {
-    return true;
-  }
-
-  return false;
-}
-
-static bool validateTimestampDigits(const SValueNode* pVal) {
-  if (!IS_INTEGER_TYPE(pVal->node.resType.type)) {
-    return false;
-  }
-
-  int64_t tsVal = pVal->datum.i;
-  char    fraction[20] = {0};
-  NUM_TO_STRING(pVal->node.resType.type, &tsVal, sizeof(fraction), fraction);
-  int32_t tsDigits = (int32_t)strlen(fraction);
-
-  if (tsDigits > TSDB_TIME_PRECISION_SEC_DIGITS) {
-    if (tsDigits == TSDB_TIME_PRECISION_MILLI_DIGITS || tsDigits == TSDB_TIME_PRECISION_MICRO_DIGITS ||
-        tsDigits == TSDB_TIME_PRECISION_NANO_DIGITS) {
-      return true;
-    } else {
-      return false;
-    }
-  }
-
-  return true;
-}
-
-static bool validateTimezoneFormat(const SValueNode* pVal) {
-  if (TSDB_DATA_TYPE_BINARY != pVal->node.resType.type) {
-    return false;
-  }
-
-  char*   tz = varDataVal(pVal->datum.p);
-  int32_t len = varDataLen(pVal->datum.p);
-
-  char   buf[3] = {0};
-  int8_t hour = -1, minute = -1;
-  if (len == 0) {
-    return false;
-  } else if (len == 1 && (tz[0] == 'z' || tz[0] == 'Z')) {
-    return true;
-  } else if ((tz[0] == '+' || tz[0] == '-')) {
-    switch (len) {
-      case 3:
-      case 5: {
-        for (int32_t i = 1; i < len; ++i) {
-          if (!isdigit(tz[i])) {
-            return false;
-          }
-
-          if (i == 2) {
-            memcpy(buf, &tz[i - 1], 2);
-            hour = taosStr2Int8(buf, NULL, 10);
-            if (!validateHourRange(hour)) {
-              return false;
-            }
-          } else if (i == 4) {
-            memcpy(buf, &tz[i - 1], 2);
-            minute = taosStr2Int8(buf, NULL, 10);
-            if (!validateMinuteRange(hour, minute, tz[0])) {
-              return false;
-            }
-          }
-        }
-        break;
-      }
-      case 6: {
-        for (int32_t i = 1; i < len; ++i) {
-          if (i == 3) {
-            if (tz[i] != ':') {
-              return false;
-            }
-            continue;
-          }
-
-          if (!isdigit(tz[i])) {
-            return false;
-          }
-
-          if (i == 2) {
-            memcpy(buf, &tz[i - 1], 2);
-            hour = taosStr2Int8(buf, NULL, 10);
-            if (!validateHourRange(hour)) {
-              return false;
-            }
-          } else if (i == 5) {
-            memcpy(buf, &tz[i - 1], 2);
-            minute = taosStr2Int8(buf, NULL, 10);
-            if (!validateMinuteRange(hour, minute, tz[0])) {
-              return false;
-            }
-          }
-        }
-        break;
-      }
-      default: {
-        return false;
-      }
-    }
-  } else {
-    return false;
-  }
-
-  return true;
-}
-
-void static addTimezoneParam(SNodeList* pList) {
-  char       buf[6] = {0};
-  time_t     t = taosTime(NULL);
-  struct tm* tmInfo = taosLocalTime(&t, NULL);
-  strftime(buf, sizeof(buf), "%z", tmInfo);
-  int32_t len = (int32_t)strlen(buf);
-
-  SValueNode* pVal = (SValueNode*)nodesMakeNode(QUERY_NODE_VALUE);
-  pVal->literal = strndup(buf, len);
-  pVal->isDuration = false;
-  pVal->translate = true;
-  pVal->node.resType.type = TSDB_DATA_TYPE_BINARY;
-  pVal->node.resType.bytes = len + VARSTR_HEADER_SIZE;
-  pVal->node.resType.precision = TSDB_TIME_PRECISION_MILLI;
-  pVal->datum.p = taosMemoryCalloc(1, len + VARSTR_HEADER_SIZE + 1);
-  varDataSetLen(pVal->datum.p, len);
-  strncpy(varDataVal(pVal->datum.p), pVal->literal, len);
-
-  nodesListAppend(pList, (SNode*)pVal);
-}
-
 static int32_t translateToIso8601(SFunctionNode* pFunc, char* pErrBuf, int32_t len) {
   int32_t numOfParams = LIST_LENGTH(pFunc->pParameterList);
   if (1 != numOfParams && 2 != numOfParams) {
@@ -1477,7 +1512,7 @@ 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);
 
@@ -1497,8 +1532,18 @@ 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;
+
+  int32_t ret = validateTimeUnitParam(dbPrec, (SValueNode *)nodesListGetNode(pFunc->pParameterList, 1));
+  if (ret == TIME_UNIT_TOO_SMALL) {
+    return buildFuncErrMsg(pErrBuf, len, TSDB_CODE_FUNC_FUNTION_ERROR,
+                           "TIMETRUNCATE function time unit parameter should be greater than db precision");
+  } else if (ret == TIME_UNIT_INVALID) {
+    return buildFuncErrMsg(pErrBuf, len, TSDB_CODE_FUNC_FUNTION_ERROR,
+                           "TIMETRUNCATE function time unit parameter should be one of the following: [1u, 1a, 1s, 1m, 1h, 1d, 1w]");
+  }
+
   addDbPrecisonParam(&pFunc->pParameterList, dbPrec);
 
   pFunc->node.resType =
@@ -1525,8 +1570,20 @@ 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;
+
+  if (3 == numOfParams) {
+    int32_t ret = validateTimeUnitParam(dbPrec, (SValueNode *)nodesListGetNode(pFunc->pParameterList, 2));
+    if (ret == TIME_UNIT_TOO_SMALL) {
+      return buildFuncErrMsg(pErrBuf, len, TSDB_CODE_FUNC_FUNTION_ERROR,
+                             "TIMEDIFF function time unit parameter should be greater than db precision");
+    } else if (ret == TIME_UNIT_INVALID) {
+      return buildFuncErrMsg(pErrBuf, len, TSDB_CODE_FUNC_FUNTION_ERROR,
+                             "TIMEDIFF function time unit parameter should be one of the following: [1u, 1a, 1s, 1m, 1h, 1d, 1w]");
+    }
+  }
+
   addDbPrecisonParam(&pFunc->pParameterList, dbPrec);
 
   pFunc->node.resType = (SDataType){.bytes = tDataTypes[TSDB_DATA_TYPE_BIGINT].bytes, .type = TSDB_DATA_TYPE_BIGINT};
@@ -1543,7 +1600,7 @@ 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;
 }
 
@@ -1924,7 +1981,7 @@ 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,
@@ -1934,7 +1991,7 @@ 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,
@@ -1944,7 +2001,7 @@ 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,
@@ -1954,20 +2011,17 @@ 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,
@@ -1977,7 +2031,7 @@ 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,
@@ -1990,7 +2044,7 @@ 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,
@@ -2001,7 +2055,7 @@ 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,
@@ -2012,7 +2066,7 @@ 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,
@@ -2025,7 +2079,7 @@ 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,
@@ -2036,7 +2090,7 @@ 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,
@@ -2047,7 +2101,7 @@ 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,
@@ -2144,7 +2198,7 @@ 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,
@@ -2154,7 +2208,7 @@ 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,
@@ -2194,7 +2248,8 @@ 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,
@@ -2205,7 +2260,7 @@ 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,
@@ -2623,7 +2678,7 @@ 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, (double)TSDB_TICK_PER_SECOND(pCtx->param[1].param.i));
+  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;
 }
 

source/libs/index/CMakeLists.txt

@@ -35,7 +35,7 @@ if (${BUILD_WITH_INVERTEDINDEX})
 endif(${BUILD_WITH_INVERTEDINDEX})
 
 
-if (${BUILD_TEST})
+# if (${BUILD_TEST})
-  add_subdirectory(test)
+#   add_subdirectory(test)
-endif(${BUILD_TEST})
+# endif(${BUILD_TEST})
 

source/libs/index/inc/indexFst.h

@@ -21,7 +21,7 @@ extern "C" {
 #endif
 
 #include "indexFstAutomation.h"
-#include "indexFstCountingWriter.h"
+#include "indexFstFile.h"
 #include "indexFstNode.h"
 #include "indexFstRegistry.h"
 #include "indexFstUtil.h"
@@ -90,8 +90,8 @@ FstBuilderNode*     fstUnFinishedNodesPopEmpty(FstUnFinishedNodes* nodes);
 uint64_t fstUnFinishedNodesFindCommPrefixAndSetOutput(FstUnFinishedNodes* node, FstSlice bs, Output in, Output* out);
 
 typedef struct FstBuilder {
-  FstCountingWriter*  wrt;         // The FST raw data is written directly to `wtr`.
+  IdxFstFile*         wrt;         // The FST raw data is written directly to `wtr`.
-  FstUnFinishedNodes* unfinished;  // The stack of unfinished nodes
+  FstUnFinishedNodes* unfinished;  // The stack of unfinished  nodes
   FstRegistry*        registry;    // A map of finished nodes.
   FstSlice            last;        // The last word added
   CompiledAddr        lastAddr;    // The address of the last compiled node
@@ -125,9 +125,9 @@ FstState fstStateCreateFrom(FstSlice* data, CompiledAddr addr);
 FstState fstStateCreate(State state);
 
 // compile
-void fstStateCompileForOneTransNext(FstCountingWriter* w, CompiledAddr addr, uint8_t inp);
+void fstStateCompileForOneTransNext(IdxFstFile* w, CompiledAddr addr, uint8_t inp);
-void fstStateCompileForOneTrans(FstCountingWriter* w, CompiledAddr addr, FstTransition* trn);
+void fstStateCompileForOneTrans(IdxFstFile* w, CompiledAddr addr, FstTransition* trn);
-void fstStateCompileForAnyTrans(FstCountingWriter* w, CompiledAddr addr, FstBuilderNode* node);
+void fstStateCompileForAnyTrans(IdxFstFile* w, CompiledAddr addr, FstBuilderNode* node);
 
 // set_comm_input
 void fstStateSetCommInput(FstState* state, uint8_t inp);
@@ -282,7 +282,7 @@ FStmSt* stmBuilderIntoStm(FStmBuilder* sb);
 bool fstVerify(Fst* fst);
 
 // refactor this function
-bool fstBuilderNodeCompileTo(FstBuilderNode* b, FstCountingWriter* wrt, CompiledAddr lastAddr, CompiledAddr startAddr);
+bool fstBuilderNodeCompileTo(FstBuilderNode* b, IdxFstFile* wrt, CompiledAddr lastAddr, CompiledAddr startAddr);
 
 typedef struct StreamState {
   FstNode*  node;

source/libs/index/inc/indexFstCountingWriter.h

@@ -1,96 +0,0 @@
-/*
- * Copyright (c) 2019 TAOS Data, Inc. <jhtao@taosdata.com>
- *
- * This program is free software: you can use, redistribute, and/or modify
- * it under the terms of the GNU Affero General Public License, version 3
- * or later ("AGPL"), as published by the Free Software Foundation.
- *
- * This program is distributed in the hope that it will be useful, but WITHOUT
- * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
- * FITNESS FOR A PARTICULAR PURPOSE.
- *
- * You should have received a copy of the GNU Affero General Public License
- * along with this program. If not, see <http://www.gnu.org/licenses/>.
- */
-
-#ifndef __INDEX_FST_COUNTING_WRITER_H__
-#define __INDEX_FST_COUNTING_WRITER_H__
-
-#include "indexInt.h"
-
-#ifdef __cplusplus
-extern "C" {
-#endif
-
-//#define USE_MMAP 1
-
-#define DefaultMem 1024 * 1024
-
-static char tmpFile[] = "./index";
-typedef enum WriterType { TMemory, TFile } WriterType;
-
-typedef struct WriterCtx {
-  int (*write)(struct WriterCtx* ctx, uint8_t* buf, int len);
-  int (*read)(struct WriterCtx* ctx, uint8_t* buf, int len);
-  int (*flush)(struct WriterCtx* ctx);
-  int (*readFrom)(struct WriterCtx* ctx, uint8_t* buf, int len, int32_t offset);
-  int (*size)(struct WriterCtx* ctx);
-  WriterType type;
-  union {
-    struct {
-      TdFilePtr pFile;
-      bool readOnly;
-      char buf[256];
-      int  size;
-#ifdef USE_MMAP
-      char* ptr;
-#endif
-    } file;
-    struct {
-      int32_t capa;
-      char*   buf;
-    } mem;
-  };
-  int32_t offset;
-  int32_t limit;
-} WriterCtx;
-
-static int writeCtxDoWrite(WriterCtx* ctx, uint8_t* buf, int len);
-static int writeCtxDoRead(WriterCtx* ctx, uint8_t* buf, int len);
-static int writeCtxDoReadFrom(WriterCtx* ctx, uint8_t* buf, int len, int32_t offset);
-static int writeCtxDoFlush(WriterCtx* ctx);
-
-WriterCtx* writerCtxCreate(WriterType type, const char* path, bool readOnly, int32_t capacity);
-void       writerCtxDestroy(WriterCtx* w, bool remove);
-
-typedef uint32_t CheckSummer;
-
-typedef struct FstCountingWriter {
-  void*       wrt;  // wrap any writer that counts and checksum bytes written
-  uint64_t    count;
-  CheckSummer summer;
-} FstCountingWriter;
-
-int fstCountingWriterWrite(FstCountingWriter* write, uint8_t* buf, uint32_t len);
-
-int fstCountingWriterRead(FstCountingWriter* write, uint8_t* buf, uint32_t len);
-
-int fstCountingWriterFlush(FstCountingWriter* write);
-
-uint32_t fstCountingWriterMaskedCheckSum(FstCountingWriter* write);
-
-FstCountingWriter* fstCountingWriterCreate(void* wtr);
-void               fstCountingWriterDestroy(FstCountingWriter* w);
-
-void    fstCountingWriterPackUintIn(FstCountingWriter* writer, uint64_t n, uint8_t nBytes);
-uint8_t fstCountingWriterPackUint(FstCountingWriter* writer, uint64_t n);
-
-#define FST_WRITER_COUNT(writer) (writer->count)
-#define FST_WRITER_INTER_WRITER(writer) (writer->wtr)
-#define FST_WRITE_CHECK_SUMMER(writer) (writer->summer)
-
-#ifdef __cplusplus
-}
-#endif
-
-#endif

source/libs/index/inc/indexFstFile.h

@@ -0,0 +1,96 @@
+/*
+ * Copyright (c) 2019 TAOS Data, Inc. <jhtao@taosdata.com>
+ *
+ * This program is free software: you can use, redistribute, and/or modify
+ * it under the terms of the GNU Affero General Public License, version 3
+ * or later ("AGPL"), as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.
+ *
+ * You should have received a copy of the GNU Affero General Public License
+ * along with this program. If not, see <http://www.gnu.org/licenses/>.
+ */
+
+#ifndef __INDEX_FST_FILE_H__
+#define __INDEX_FST_FILE_H__
+
+#include "indexInt.h"
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+//#define USE_MMAP 1
+
+#define DefaultMem 1024 * 1024
+
+static char tmpFile[] = "./index";
+typedef enum WriterType { TMemory, TFile } WriterType;
+
+typedef struct IFileCtx {
+  int (*write)(struct IFileCtx* ctx, uint8_t* buf, int len);
+  int (*read)(struct IFileCtx* ctx, uint8_t* buf, int len);
+  int (*flush)(struct IFileCtx* ctx);
+  int (*readFrom)(struct IFileCtx* ctx, uint8_t* buf, int len, int32_t offset);
+  int (*size)(struct IFileCtx* ctx);
+  WriterType type;
+  union {
+    struct {
+      TdFilePtr pFile;
+      bool      readOnly;
+      char      buf[256];
+      int64_t   size;
+#ifdef USE_MMAP
+      char* ptr;
+#endif
+    } file;
+    struct {
+      int32_t cap;
+      char*   buf;
+    } mem;
+  };
+  int32_t offset;
+  int32_t limit;
+} IFileCtx;
+
+static int idxFileCtxDoWrite(IFileCtx* ctx, uint8_t* buf, int len);
+static int idxFileCtxDoRead(IFileCtx* ctx, uint8_t* buf, int len);
+static int idxFileCtxDoReadFrom(IFileCtx* ctx, uint8_t* buf, int len, int32_t offset);
+static int idxFileCtxDoFlush(IFileCtx* ctx);
+
+IFileCtx* idxFileCtxCreate(WriterType type, const char* path, bool readOnly, int32_t capacity);
+void      idxFileCtxDestroy(IFileCtx* w, bool remove);
+
+typedef uint32_t CheckSummer;
+
+typedef struct IdxFstFile {
+  void*       wrt;  // wrap any writer that counts and checksum bytes written
+  uint64_t    count;
+  CheckSummer summer;
+} IdxFstFile;
+
+int idxFileWrite(IdxFstFile* write, uint8_t* buf, uint32_t len);
+
+int idxFileRead(IdxFstFile* write, uint8_t* buf, uint32_t len);
+
+int idxFileFlush(IdxFstFile* write);
+
+uint32_t idxFileMaskedCheckSum(IdxFstFile* write);
+
+IdxFstFile* idxFileCreate(void* wtr);
+void        idxFileDestroy(IdxFstFile* w);
+
+void    idxFilePackUintIn(IdxFstFile* writer, uint64_t n, uint8_t nBytes);
+uint8_t idxFilePackUint(IdxFstFile* writer, uint64_t n);
+
+#define FST_WRITER_COUNT(writer)        (writer->count)
+#define FST_WRITER_INTER_WRITER(writer) (writer->wtr)
+#define FST_WRITE_CHECK_SUMMER(writer)  (writer->summer)
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif

source/libs/index/inc/indexFstNode.h

@@ -20,12 +20,12 @@
 extern "C" {
 #endif
 
-#include "indexFstCountingWriter.h"
+#include "indexFstFile.h"
 #include "indexFstUtil.h"
 #include "indexInt.h"
 
-#define FST_BUILDER_NODE_IS_FINAL(bn) (bn->isFinal)
+#define FST_BUILDER_NODE_IS_FINAL(bn)           (bn->isFinal)
-#define FST_BUILDER_NODE_TRANS_ISEMPTY(bn) (taosArrayGetSize(bn->trans) == 0)
+#define FST_BUILDER_NODE_TRANS_ISEMPTY(bn)      (taosArrayGetSize(bn->trans) == 0)
 #define FST_BUILDER_NODE_FINALOUTPUT_ISZERO(bn) (bn->finalOutput == 0)
 
 typedef struct FstTransition {
@@ -46,7 +46,7 @@ FstBuilderNode* fstBuilderNodeClone(FstBuilderNode* src);
 
 void fstBuilderNodeCloneFrom(FstBuilderNode* dst, FstBuilderNode* src);
 
-// bool fstBuilderNodeCompileTo(FstBuilderNode *b, FstCountingWriter *wrt,
+// bool fstBuilderNodeCompileTo(FstBuilderNode *b, IdxFile' *wrt,
 // CompiledAddr lastAddr, CompiledAddr startAddr);
 bool fstBuilderNodeEqual(FstBuilderNode* n1, FstBuilderNode* n2);
 

source/libs/index/inc/indexTfile.h

@@ -16,7 +16,7 @@
 #define __INDEX_TFILE_H__
 
 #include "indexFst.h"
-#include "indexFstCountingWriter.h"
+#include "indexFstFile.h"
 #include "indexInt.h"
 #include "indexTfile.h"
 #include "indexUtil.h"
@@ -59,7 +59,7 @@ typedef struct TFileCache {
 
 typedef struct TFileWriter {
   FstBuilder* fb;
-  WriterCtx*  ctx;
+  IFileCtx*   ctx;
   TFileHeader header;
   uint32_t    offset;
 } TFileWriter;
@@ -68,7 +68,7 @@ typedef struct TFileWriter {
 typedef struct TFileReader {
   T_REF_DECLARE()
   Fst*        fst;
-  WriterCtx*  ctx;
+  IFileCtx*   ctx;
   TFileHeader header;
   bool        remove;
 } TFileReader;
@@ -103,7 +103,7 @@ void         tfileCachePut(TFileCache* tcache, ICacheKey* key, TFileReader* read
 TFileReader* tfileGetReaderByCol(IndexTFile* tf, uint64_t suid, char* colName);
 
 TFileReader* tfileReaderOpen(char* path, uint64_t suid, int64_t version, const char* colName);
-TFileReader* tfileReaderCreate(WriterCtx* ctx);
+TFileReader* tfileReaderCreate(IFileCtx* ctx);
 void         tfileReaderDestroy(TFileReader* reader);
 int          tfileReaderSearch(TFileReader* reader, SIndexTermQuery* query, SIdxTRslt* tr);
 void         tfileReaderRef(TFileReader* reader);
@@ -111,7 +111,7 @@ void         tfileReaderUnRef(TFileReader* reader);
 
 TFileWriter* tfileWriterOpen(char* path, uint64_t suid, int64_t version, const char* colName, uint8_t type);
 void         tfileWriterClose(TFileWriter* tw);
-TFileWriter* tfileWriterCreate(WriterCtx* ctx, TFileHeader* header);
+TFileWriter* tfileWriterCreate(IFileCtx* ctx, TFileHeader* header);
 void         tfileWriterDestroy(TFileWriter* tw);
 int          tfileWriterPut(TFileWriter* tw, void* data, bool order);
 int          tfileWriterFinish(TFileWriter* tw);

source/libs/index/src/index.c

@@ -39,7 +39,7 @@
 #define INDEX_DATA_BIGINT_NULL    0x8000000000000000LL
 #define INDEX_DATA_TIMESTAMP_NULL TSDB_DATA_BIGINT_NULL
 
-#define INDEX_DATA_FLOAT_NULL    0x7FF00000           // it is an NAN
+#define INDEX_DATA_FLOAT_NULL    0x7FF00000            // it is an NAN
 #define INDEX_DATA_DOUBLE_NULL   0x7FFFFF0000000000LL  // an NAN
 #define INDEX_DATA_NCHAR_NULL    0xFFFFFFFF
 #define INDEX_DATA_BINARY_NULL   0xFF
@@ -614,7 +614,7 @@ static int idxGenTFile(SIndex* sIdx, IndexCache* cache, SArray* batch) {
   return ret;
 END:
   if (tw != NULL) {
-    writerCtxDestroy(tw->ctx, true);
+    idxFileCtxDestroy(tw->ctx, true);
     taosMemoryFree(tw);
   }
   return -1;

source/libs/index/src/indexFst.c

@@ -19,11 +19,11 @@
 #include "tchecksum.h"
 #include "tcoding.h"
 
-static void fstPackDeltaIn(FstCountingWriter* wrt, CompiledAddr nodeAddr, CompiledAddr transAddr, uint8_t nBytes) {
+static void fstPackDeltaIn(IdxFstFile* wrt, CompiledAddr nodeAddr, CompiledAddr transAddr, uint8_t nBytes) {
   CompiledAddr deltaAddr = (transAddr == EMPTY_ADDRESS) ? EMPTY_ADDRESS : nodeAddr - transAddr;
-  fstCountingWriterPackUintIn(wrt, deltaAddr, nBytes);
+  idxFilePackUintIn(wrt, deltaAddr, nBytes);
 }
-static uint8_t fstPackDetla(FstCountingWriter* wrt, CompiledAddr nodeAddr, CompiledAddr transAddr) {
+static uint8_t fstPackDetla(IdxFstFile* wrt, CompiledAddr nodeAddr, CompiledAddr transAddr) {
   uint8_t nBytes = packDeltaSize(nodeAddr, transAddr);
   fstPackDeltaIn(wrt, nodeAddr, transAddr, nBytes);
   return nBytes;
@@ -208,7 +208,7 @@ FstState fstStateCreate(State state) {
   return fstStateDict[idx];
 }
 // compile
-void fstStateCompileForOneTransNext(FstCountingWriter* w, CompiledAddr addr, uint8_t inp) {
+void fstStateCompileForOneTransNext(IdxFstFile* w, CompiledAddr addr, uint8_t inp) {
   FstState s = fstStateCreate(OneTransNext);
   fstStateSetCommInput(&s, inp);
 
@@ -216,21 +216,21 @@ void fstStateCompileForOneTransNext(FstCountingWriter* w, CompiledAddr addr, uin
   uint8_t v = fstStateCommInput(&s, &null);
   if (null) {
     // w->write_all(&[inp])
-    fstCountingWriterWrite(w, &inp, 1);
+    idxFileWrite(w, &inp, 1);
   }
-  fstCountingWriterWrite(w, &(s.val), 1);
+  idxFileWrite(w, &(s.val), 1);
   // w->write_all(&[s.val])
   return;
 }
-void fstStateCompileForOneTrans(FstCountingWriter* w, CompiledAddr addr, FstTransition* trn) {
+void fstStateCompileForOneTrans(IdxFstFile* w, CompiledAddr addr, FstTransition* trn) {
   Output    out = trn->out;
-  uint8_t   outPackSize = (out == 0 ? 0 : fstCountingWriterPackUint(w, out));
+  uint8_t   outPackSize = (out == 0 ? 0 : idxFilePackUint(w, out));
   uint8_t   transPackSize = fstPackDetla(w, addr, trn->addr);
   PackSizes packSizes = 0;
 
   FST_SET_OUTPUT_PACK_SIZE(packSizes, outPackSize);
   FST_SET_TRANSITION_PACK_SIZE(packSizes, transPackSize);
-  fstCountingWriterWrite(w, (char*)&packSizes, sizeof(packSizes));
+  idxFileWrite(w, (char*)&packSizes, sizeof(packSizes));
 
   FstState st = fstStateCreate(OneTrans);
 
@@ -239,12 +239,12 @@ void fstStateCompileForOneTrans(FstCountingWriter* w, CompiledAddr addr, FstTran
   bool    null = false;
   uint8_t inp = fstStateCommInput(&st, &null);
   if (null == true) {
-    fstCountingWriterWrite(w, (char*)&trn->inp, sizeof(trn->inp));
+    idxFileWrite(w, (char*)&trn->inp, sizeof(trn->inp));
   }
-  fstCountingWriterWrite(w, (char*)(&(st.val)), sizeof(st.val));
+  idxFileWrite(w, (char*)(&(st.val)), sizeof(st.val));
   return;
 }
-void fstStateCompileForAnyTrans(FstCountingWriter* w, CompiledAddr addr, FstBuilderNode* node) {
+void fstStateCompileForAnyTrans(IdxFstFile* w, CompiledAddr addr, FstBuilderNode* node) {
   int32_t sz = taosArrayGetSize(node->trans);
   assert(sz <= 256);
 
@@ -275,11 +275,11 @@ void fstStateCompileForAnyTrans(FstCountingWriter* w, CompiledAddr addr, FstBuil
 
   if (anyOuts) {
     if (FST_BUILDER_NODE_IS_FINAL(node)) {
-      fstCountingWriterPackUintIn(w, node->finalOutput, oSize);
+      idxFilePackUintIn(w, node->finalOutput, oSize);
     }
     for (int32_t i = sz - 1; i >= 0; i--) {
       FstTransition* t = taosArrayGet(node->trans, i);
-      fstCountingWriterPackUintIn(w, t->out, oSize);
+      idxFilePackUintIn(w, t->out, oSize);
     }
   }
   for (int32_t i = sz - 1; i >= 0; i--) {
@@ -288,7 +288,7 @@ void fstStateCompileForAnyTrans(FstCountingWriter* w, CompiledAddr addr, FstBuil
   }
   for (int32_t i = sz - 1; i >= 0; i--) {
     FstTransition* t = taosArrayGet(node->trans, i);
-    fstCountingWriterWrite(w, (char*)&t->inp, 1);
+    idxFileWrite(w, (char*)&t->inp, 1);
     // fstPackDeltaIn(w, addr, t->addr, tSize);
   }
   if (sz > TRANS_INDEX_THRESHOLD) {
@@ -306,10 +306,10 @@ void fstStateCompileForAnyTrans(FstCountingWriter* w, CompiledAddr addr, FstBuil
       index[t->inp] = i;
       // fstPackDeltaIn(w, addr, t->addr, tSize);
     }
-    fstCountingWriterWrite(w, (char*)index, 256);
+    idxFileWrite(w, (char*)index, 256);
     taosMemoryFree(index);
   }
-  fstCountingWriterWrite(w, (char*)&packSizes, 1);
+  idxFileWrite(w, (char*)&packSizes, 1);
   bool null = false;
   fstStateStateNtrans(&st, &null);
   if (null == true) {
@@ -318,12 +318,12 @@ void fstStateCompileForAnyTrans(FstCountingWriter* w, CompiledAddr addr, FstBuil
     // encoded in the state byte.
     uint8_t v = 1;
     if (sz == 256) {
-      fstCountingWriterWrite(w, (char*)&v, 1);
+      idxFileWrite(w, (char*)&v, 1);
     } else {
-      fstCountingWriterWrite(w, (char*)&sz, 1);
+      idxFileWrite(w, (char*)&sz, 1);
     }
   }
-  fstCountingWriterWrite(w, (char*)(&(st.val)), 1);
+  idxFileWrite(w, (char*)(&(st.val)), 1);
   return;
 }
 
@@ -753,7 +753,7 @@ bool fstNodeCompile(FstNode* node, void* w, CompiledAddr lastAddr, CompiledAddr
   return true;
 }
 
-bool fstBuilderNodeCompileTo(FstBuilderNode* b, FstCountingWriter* wrt, CompiledAddr lastAddr, CompiledAddr startAddr) {
+bool fstBuilderNodeCompileTo(FstBuilderNode* b, IdxFstFile* wrt, CompiledAddr lastAddr, CompiledAddr startAddr) {
   return fstNodeCompile(NULL, wrt, lastAddr, startAddr, b);
 }
 
@@ -763,7 +763,7 @@ FstBuilder* fstBuilderCreate(void* w, FstType ty) {
     return b;
   }
 
-  b->wrt = fstCountingWriterCreate(w);
+  b->wrt = idxFileCreate(w);
   b->unfinished = fstUnFinishedNodesCreate();
   b->registry = fstRegistryCreate(10000, 2);
   b->last = fstSliceCreate(NULL, 0);
@@ -773,12 +773,12 @@ FstBuilder* fstBuilderCreate(void* w, FstType ty) {
   char  buf64[8] = {0};
   void* pBuf64 = buf64;
   taosEncodeFixedU64(&pBuf64, VERSION);
-  fstCountingWriterWrite(b->wrt, buf64, sizeof(buf64));
+  idxFileWrite(b->wrt, buf64, sizeof(buf64));
 
   pBuf64 = buf64;
   memset(buf64, 0, sizeof(buf64));
   taosEncodeFixedU64(&pBuf64, ty);
-  fstCountingWriterWrite(b->wrt, buf64, sizeof(buf64));
+  idxFileWrite(b->wrt, buf64, sizeof(buf64));
 
   return b;
 }
@@ -787,7 +787,7 @@ void fstBuilderDestroy(FstBuilder* b) {
     return;
   }
 
-  fstCountingWriterDestroy(b->wrt);
+  idxFileDestroy(b->wrt);
   fstUnFinishedNodesDestroy(b->unfinished);
   fstRegistryDestroy(b->registry);
   fstSliceDestroy(&b->last);
@@ -905,21 +905,19 @@ void* fstBuilderInsertInner(FstBuilder* b) {
 
   void* pBuf64 = buf64;
   taosEncodeFixedU64(&pBuf64, b->len);
-  fstCountingWriterWrite(b->wrt, buf64, sizeof(buf64));
+  idxFileWrite(b->wrt, buf64, sizeof(buf64));
 
   pBuf64 = buf64;
   taosEncodeFixedU64(&pBuf64, rootAddr);
-  fstCountingWriterWrite(b->wrt, buf64, sizeof(buf64));
+  idxFileWrite(b->wrt, buf64, sizeof(buf64));
 
   char     buf32[4] = {0};
   void*    pBuf32 = buf32;
-  uint32_t sum = fstCountingWriterMaskedCheckSum(b->wrt);
+  uint32_t sum = idxFileMaskedCheckSum(b->wrt);
   taosEncodeFixedU32(&pBuf32, sum);
-  fstCountingWriterWrite(b->wrt, buf32, sizeof(buf32));
+  idxFileWrite(b->wrt, buf32, sizeof(buf32));
 
-  fstCountingWriterFlush(b->wrt);
+  idxFileFlush(b->wrt);
-  // fstCountingWriterDestroy(b->wrt);
-  // b->wrt = NULL;
   return b->wrt;
 }
 void fstBuilderFinish(FstBuilder* b) { fstBuilderInsertInner(b); }

source/libs/index/src/indexFstDfa.c

@@ -61,9 +61,10 @@ void dfaBuilderDestroy(FstDfaBuilder *builder) {
     pIter = taosHashIterate(builder->cache, pIter);
   }
   taosHashCleanup(builder->cache);
+  taosMemoryFree(builder);
 }
 
-FstDfa *dfaBuilderBuild(FstDfaBuilder *builder) {
+FstDfa *dfaBuilder(FstDfaBuilder *builder) {
   uint32_t      sz = taosArrayGetSize(builder->dfa->insts);
   FstSparseSet *cur = sparSetCreate(sz);
   FstSparseSet *nxt = sparSetCreate(sz);

source/libs/index/src/indexFstFile.c

@@ -13,13 +13,13 @@
  * along with this program. If not, see <http://www.gnu.org/licenses/>.
  */
 
-#include "indexFstCountingWriter.h"
+#include "indexFstFile.h"
 #include "indexFstUtil.h"
 #include "indexInt.h"
 #include "os.h"
 #include "tutil.h"
 
-static int writeCtxDoWrite(WriterCtx* ctx, uint8_t* buf, int len) {
+static int idxFileCtxDoWrite(IFileCtx* ctx, uint8_t* buf, int len) {
   if (ctx->type == TFile) {
     assert(len == taosWriteFile(ctx->file.pFile, buf, len));
   } else {
@@ -28,7 +28,7 @@ static int writeCtxDoWrite(WriterCtx* ctx, uint8_t* buf, int len) {
   ctx->offset += len;
   return len;
 }
-static int writeCtxDoRead(WriterCtx* ctx, uint8_t* buf, int len) {
+static int idxFileCtxDoRead(IFileCtx* ctx, uint8_t* buf, int len) {
   int nRead = 0;
   if (ctx->type == TFile) {
 #ifdef USE_MMAP
@@ -44,7 +44,7 @@ static int writeCtxDoRead(WriterCtx* ctx, uint8_t* buf, int len) {
 
   return nRead;
 }
-static int writeCtxDoReadFrom(WriterCtx* ctx, uint8_t* buf, int len, int32_t offset) {
+static int idxFileCtxDoReadFrom(IFileCtx* ctx, uint8_t* buf, int len, int32_t offset) {
   int nRead = 0;
   if (ctx->type == TFile) {
     // tfLseek(ctx->file.pFile, offset, 0);
@@ -61,7 +61,7 @@ static int writeCtxDoReadFrom(WriterCtx* ctx, uint8_t* buf, int len, int32_t off
   }
   return nRead;
 }
-static int writeCtxGetSize(WriterCtx* ctx) {
+static int idxFileCtxGetSize(IFileCtx* ctx) {
   if (ctx->type == TFile) {
     int64_t file_size = 0;
     taosStatFile(ctx->file.buf, &file_size, NULL);
@@ -69,7 +69,7 @@ static int writeCtxGetSize(WriterCtx* ctx) {
   }
   return 0;
 }
-static int writeCtxDoFlush(WriterCtx* ctx) {
+static int idxFileCtxDoFlush(IFileCtx* ctx) {
   if (ctx->type == TFile) {
     // taosFsyncFile(ctx->file.pFile);
     taosFsyncFile(ctx->file.pFile);
@@ -80,8 +80,8 @@ static int writeCtxDoFlush(WriterCtx* ctx) {
   return 1;
 }
 
-WriterCtx* writerCtxCreate(WriterType type, const char* path, bool readOnly, int32_t capacity) {
+IFileCtx* idxFileCtxCreate(WriterType type, const char* path, bool readOnly, int32_t capacity) {
-  WriterCtx* ctx = taosMemoryCalloc(1, sizeof(WriterCtx));
+  IFileCtx* ctx = taosMemoryCalloc(1, sizeof(IFileCtx));
   if (ctx == NULL) {
     return NULL;
   }
@@ -90,39 +90,36 @@ WriterCtx* writerCtxCreate(WriterType type, const char* path, bool readOnly, int
   if (ctx->type == TFile) {
     // ugly code, refactor later
     ctx->file.readOnly = readOnly;
+    memcpy(ctx->file.buf, path, strlen(path));
     if (readOnly == false) {
-      // ctx->file.pFile = open(path, O_WRONLY | O_CREAT | O_APPEND, S_IRWXU | S_IRWXG | S_IRWXO);
       ctx->file.pFile = taosOpenFile(path, TD_FILE_CREATE | TD_FILE_WRITE | TD_FILE_APPEND);
       taosFtruncateFile(ctx->file.pFile, 0);
-      int64_t file_size;
+      taosStatFile(path, &ctx->file.size, NULL);
-      taosStatFile(path, &file_size, NULL);
+      // ctx->file.size = (int)size;
-      ctx->file.size = (int)file_size;
 
     } else {
-      // ctx->file.pFile = open(path, O_RDONLY, S_IRWXU | S_IRWXG | S_IRWXO);
       ctx->file.pFile = taosOpenFile(path, TD_FILE_READ);
 
-      int64_t file_size = 0;
+      int64_t size = 0;
-      taosFStatFile(ctx->file.pFile, &file_size, NULL);
+      taosFStatFile(ctx->file.pFile, &ctx->file.size, NULL);
-      ctx->file.size = (int)file_size;
+      ctx->file.size = (int)size;
 #ifdef USE_MMAP
       ctx->file.ptr = (char*)tfMmapReadOnly(ctx->file.pFile, ctx->file.size);
 #endif
     }
-    memcpy(ctx->file.buf, path, strlen(path));
     if (ctx->file.pFile == NULL) {
       indexError("failed to open file, error %d", errno);
       goto END;
     }
   } else if (ctx->type == TMemory) {
     ctx->mem.buf = taosMemoryCalloc(1, sizeof(char) * capacity);
-    ctx->mem.capa = capacity;
+    ctx->mem.cap = capacity;
   }
-  ctx->write = writeCtxDoWrite;
+  ctx->write = idxFileCtxDoWrite;
-  ctx->read = writeCtxDoRead;
+  ctx->read = idxFileCtxDoRead;
-  ctx->flush = writeCtxDoFlush;
+  ctx->flush = idxFileCtxDoFlush;
-  ctx->readFrom = writeCtxDoReadFrom;
+  ctx->readFrom = idxFileCtxDoReadFrom;
-  ctx->size = writeCtxGetSize;
+  ctx->size = idxFileCtxGetSize;
 
   ctx->offset = 0;
   ctx->limit = capacity;
@@ -135,7 +132,7 @@ END:
   taosMemoryFree(ctx);
   return NULL;
 }
-void writerCtxDestroy(WriterCtx* ctx, bool remove) {
+void idxFileCtxDestroy(IFileCtx* ctx, bool remove) {
   if (ctx->type == TMemory) {
     taosMemoryFree(ctx->mem.buf);
   } else {
@@ -149,9 +146,6 @@ void writerCtxDestroy(WriterCtx* ctx, bool remove) {
     if (ctx->file.readOnly == false) {
       int64_t file_size = 0;
       taosStatFile(ctx->file.buf, &file_size, NULL);
-      // struct stat fstat;
-      // stat(ctx->file.buf, &fstat);
-      // indexError("write file size: %d", (int)(fstat.st_size));
     }
     if (remove) {
       unlink(ctx->file.buf);
@@ -160,30 +154,29 @@ void writerCtxDestroy(WriterCtx* ctx, bool remove) {
   taosMemoryFree(ctx);
 }
 
-FstCountingWriter* fstCountingWriterCreate(void* wrt) {
+IdxFstFile* idxFileCreate(void* wrt) {
-  FstCountingWriter* cw = taosMemoryCalloc(1, sizeof(FstCountingWriter));
+  IdxFstFile* cw = taosMemoryCalloc(1, sizeof(IdxFstFile));
   if (cw == NULL) {
     return NULL;
   }
 
   cw->wrt = wrt;
-  //(void *)(writerCtxCreate(TFile, readOnly));
   return cw;
 }
-void fstCountingWriterDestroy(FstCountingWriter* cw) {
+void idxFileDestroy(IdxFstFile* cw) {
   // free wrt object: close fd or free mem
-  fstCountingWriterFlush(cw);
+  idxFileFlush(cw);
-  // writerCtxDestroy((WriterCtx *)(cw->wrt));
+  // idxFileCtxDestroy((IFileCtx *)(cw->wrt));
   taosMemoryFree(cw);
 }
 
-int fstCountingWriterWrite(FstCountingWriter* write, uint8_t* buf, uint32_t len) {
+int idxFileWrite(IdxFstFile* write, uint8_t* buf, uint32_t len) {
   if (write == NULL) {
     return 0;
   }
   // update checksum
   // write data to file/socket or mem
-  WriterCtx* ctx = write->wrt;
+  IFileCtx* ctx = write->wrt;
 
   int nWrite = ctx->write(ctx, buf, len);
   assert(nWrite == len);
@@ -192,42 +185,41 @@ int fstCountingWriterWrite(FstCountingWriter* write, uint8_t* buf, uint32_t len)
   write->summer = taosCalcChecksum(write->summer, buf, len);
   return len;
 }
-int fstCountingWriterRead(FstCountingWriter* write, uint8_t* buf, uint32_t len) {
+int idxFileRead(IdxFstFile* write, uint8_t* buf, uint32_t len) {
   if (write == NULL) {
     return 0;
   }
-  WriterCtx* ctx = write->wrt;
+  IFileCtx* ctx = write->wrt;
-  int        nRead = ctx->read(ctx, buf, len);
+  int       nRead = ctx->read(ctx, buf, len);
   // assert(nRead == len);
   return nRead;
 }
 
-uint32_t fstCountingWriterMaskedCheckSum(FstCountingWriter* write) {
+uint32_t idxFileMaskedCheckSum(IdxFstFile* write) {
   // opt
   return write->summer;
 }
 
-int fstCountingWriterFlush(FstCountingWriter* write) {
+int idxFileFlush(IdxFstFile* write) {
-  WriterCtx* ctx = write->wrt;
+  IFileCtx* ctx = write->wrt;
   ctx->flush(ctx);
-  // write->wtr->flush
   return 1;
 }
 
-void fstCountingWriterPackUintIn(FstCountingWriter* writer, uint64_t n, uint8_t nBytes) {
+void idxFilePackUintIn(IdxFstFile* writer, uint64_t n, uint8_t nBytes) {
   assert(1 <= nBytes && nBytes <= 8);
   uint8_t* buf = taosMemoryCalloc(8, sizeof(uint8_t));
   for (uint8_t i = 0; i < nBytes; i++) {
     buf[i] = (uint8_t)n;
     n = n >> 8;
   }
-  fstCountingWriterWrite(writer, buf, nBytes);
+  idxFileWrite(writer, buf, nBytes);
   taosMemoryFree(buf);
   return;
 }
 
-uint8_t fstCountingWriterPackUint(FstCountingWriter* writer, uint64_t n) {
+uint8_t idxFilePackUint(IdxFstFile* writer, uint64_t n) {
   uint8_t nBytes = packSize(n);
-  fstCountingWriterPackUintIn(writer, n, nBytes);
+  idxFilePackUintIn(writer, n, nBytes);
   return nBytes;
 }

source/libs/index/src/indexFstNode.c

@@ -95,7 +95,7 @@ void fstBuilderNodeCloneFrom(FstBuilderNode* dst, FstBuilderNode* src) {
   }
 }
 
-// bool fstBuilderNodeCompileTo(FstBuilderNode *b, FstCountingWriter *wrt, CompiledAddr lastAddr, CompiledAddr
+// bool fstBuilderNodeCompileTo(FstBuilderNode *b, IdxFile *wrt, CompiledAddr lastAddr, CompiledAddr
 // startAddr) {
 
 // size_t sz = taosArrayGetSize(b->trans);

source/libs/index/src/indexFstSparse.c

@@ -17,7 +17,7 @@
 
 FstSparseSet *sparSetCreate(int32_t sz) {
   FstSparseSet *ss = taosMemoryCalloc(1, sizeof(FstSparseSet));
-  if (ss = NULL) {
+  if (ss == NULL) {
     return NULL;
   }