Merge branch '3.0' of https://github.com/taosdata/TDengine into test/td-14999

Jenkinsfile2

@@ -168,13 +168,6 @@ pipeline {
     stages {
         stage('run test') {
             parallel {
-                stage('windows test') {
-                    agent {label " windows11 "}
-                    steps {
-                        pre_test_win()
-                        pre_test_build_win()
-                    }
-                }
                 stage('linux test') {
                     agent{label " slave3_0 || slave15 || slave16 || slave17 "}
                     options { skipDefaultCheckout() } 

cmake/cmake.define

@@ -46,7 +46,7 @@ ENDIF ()
 
 IF (TD_WINDOWS)
     MESSAGE("${Yellow} set compiler flag for Windows! ${ColourReset}")
-    SET(COMMON_FLAGS "/W3 /D_WIN32")
+    SET(COMMON_FLAGS "/W3 /D_WIN32 /vmg")
 
     # IF (MSVC AND (MSVC_VERSION GREATER_EQUAL 1900))
     #  SET(COMMON_FLAGS "${COMMON_FLAGS} /Wv:18")
@@ -62,7 +62,7 @@ ELSE ()
 	MESSAGE(STATUS "Will compile with Address Sanitizer!")
     ELSE ()
         SET(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -Werror -Werror=return-type -fPIC -gdwarf-2 -g3")
-        SET(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Werror -Werror=return-type -fPIC -gdwarf-2 -g3")
+        SET(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wno-literal-suffix -Werror=return-type -fpermissive -fPIC -gdwarf-2 -g3")
     ENDIF ()
 
     MESSAGE("System processor ID: ${CMAKE_SYSTEM_PROCESSOR}")

example/src/tstream.c

@@ -81,10 +81,10 @@ int32_t create_stream() {
   /*const char* sql = "select min(k), max(k), sum(k) as sum_of_k from st1";*/
   /*const char* sql = "select sum(k) from tu1 interval(10m)";*/
   /*pRes = tmq_create_stream(pConn, "stream1", "out1", sql);*/
-  pRes =
+  pRes = taos_query(
-      taos_query(pConn,
+      pConn,
-                 "create stream stream1 trigger window_close as select _wstartts, min(k), max(k), sum(k) as sum_of_k "
+      "create stream stream1 trigger window_close into outstb as select _wstartts, min(k), max(k), sum(k) as sum_of_k "
-                 "from tu1 interval(10m)");
+      "from tu1 interval(10m)");
   if (taos_errno(pRes) != 0) {
     printf("failed to create stream stream1, reason:%s\n", taos_errstr(pRes));
     return -1;

include/client/taos.h

@@ -195,7 +195,7 @@ DLL_EXPORT void      taos_unsubscribe(TAOS_SUB *tsub, int keepProgress);
 #endif
 
 DLL_EXPORT int       taos_load_table_info(TAOS *taos, const char *tableNameList);
-DLL_EXPORT TAOS_RES *taos_schemaless_insert(TAOS *taos, char *lines[], int numLines, int protocol, int precision);
+DLL_EXPORT TAOS_RES *taos_schemaless_insert(TAOS *taos, char *lines[], int numLines, int protocol, int precision, bool dataFormat);
 
 /* --------------------------TMQ INTERFACE------------------------------- */
 

include/common/tcommon.h

@@ -194,6 +194,21 @@ typedef struct SExprInfo {
   struct tExprNode*     pExpr;
 } SExprInfo;
 
+typedef struct {
+  const char* key;
+  int32_t keyLen;
+  uint8_t type;
+  int16_t length;
+  union{
+    const char* value;
+    int64_t  i;
+    uint64_t u;
+    double   d;
+    float    f;
+  };
+  int32_t valueLen;
+} SSmlKv;
+
 #define QUERY_ASC_FORWARD_STEP  1
 #define QUERY_DESC_FORWARD_STEP -1
 

include/common/tdataformat.h

@@ -372,10 +372,9 @@ static FORCE_INLINE void tdCopyColOfRowBySchema(SDataRow dst, STSchema *pDstSche
 // ----------------- Data column structure
 // SDataCol arrangement: data => bitmap => dataOffset
 typedef struct SDataCol {
-  int8_t          type;            // column type
+  int8_t          type;        // column type
-  uint8_t         bitmap : 1;      // 0: no bitmap if all rows are NORM, 1: has bitmap if has NULL/NORM rows
+  uint8_t         bitmap : 1;  // 0: no bitmap if all rows are NORM, 1: has bitmap if has NULL/NORM rows
-  uint8_t         bitmapMode : 1;  // default is 0(2 bits), otherwise 1(1 bit)
+  uint8_t         reserve : 7;
-  uint8_t         reserve : 6;
   int16_t         colId;      // column ID
   int32_t         bytes;      // column data bytes defined
   int32_t         offset;     // data offset in a SDataRow (including the header size)
@@ -387,8 +386,6 @@ typedef struct SDataCol {
   TSKEY           ts;         // only used in last NULL column
 } SDataCol;
 
-
-
 #define isAllRowsNull(pCol) ((pCol)->len == 0)
 #define isAllRowsNone(pCol) ((pCol)->len == 0)
 static FORCE_INLINE void dataColReset(SDataCol *pDataCol) { pDataCol->len = 0; }
@@ -482,7 +479,7 @@ void       tdResetDataCols(SDataCols *pCols);
 int32_t    tdInitDataCols(SDataCols *pCols, STSchema *pSchema);
 SDataCols *tdDupDataCols(SDataCols *pCols, bool keepData);
 SDataCols *tdFreeDataCols(SDataCols *pCols);
-int32_t tdMergeDataCols(SDataCols *target, SDataCols *source, int32_t rowsToMerge, int32_t *pOffset, bool forceSetNull, TDRowVerT maxVer);
+int32_t tdMergeDataCols(SDataCols *target, SDataCols *source, int32_t rowsToMerge, int32_t *pOffset, bool update, TDRowVerT maxVer);
 
 // ----------------- K-V data row structure
 /* |<-------------------------------------- len -------------------------------------------->|

include/common/tmsg.h

@@ -672,7 +672,6 @@ typedef struct {
   SArray* pArray;  // Array of SUseDbRsp
 } SUseDbBatchRsp;
 
-
 int32_t tSerializeSUseDbBatchRsp(void* buf, int32_t bufLen, SUseDbBatchRsp* pRsp);
 int32_t tDeserializeSUseDbBatchRsp(void* buf, int32_t bufLen, SUseDbBatchRsp* pRsp);
 void    tFreeSUseDbBatchRsp(SUseDbBatchRsp* pRsp);
@@ -685,7 +684,6 @@ int32_t tSerializeSUserAuthBatchRsp(void* buf, int32_t bufLen, SUserAuthBatchRsp
 int32_t tDeserializeSUserAuthBatchRsp(void* buf, int32_t bufLen, SUserAuthBatchRsp* pRsp);
 void    tFreeSUserAuthBatchRsp(SUserAuthBatchRsp* pRsp);
 
-
 typedef struct {
   char db[TSDB_DB_FNAME_LEN];
 } SCompactDbReq;
@@ -1554,7 +1552,9 @@ typedef struct SVDropStbReq {
 int32_t tEncodeSVDropStbReq(SCoder* pCoder, const SVDropStbReq* pReq);
 int32_t tDecodeSVDropStbReq(SCoder* pCoder, SVDropStbReq* pReq);
 
+#define TD_CREATE_IF_NOT_EXISTS 0x1
 typedef struct SVCreateTbReq {
+  int32_t     flags;
   tb_uid_t    uid;
   int64_t     ctime;
   const char* name;

include/common/trow.h

@@ -42,7 +42,7 @@ extern "C" {
  * @brief value type
  *  - for data from client input and STSRow in memory, 3 types of value none/null/norm available
  */
-#define TD_VTYPE_NORM 0x00U  // normal val: not none, not null(no need assign value)
+#define TD_VTYPE_NORM 0x00U  // normal val: not none, not null
 #define TD_VTYPE_NULL 0x01U  // null val
 #define TD_VTYPE_NONE 0x02U  // none or unknown/undefined
 #define TD_VTYPE_MAX  0x03U  //
@@ -140,8 +140,6 @@ typedef struct {
   };
   /// row total length
   uint32_t len;
-  /// row version
-  // uint64_t ver;
   /// the inline data, maybe a tuple or a k-v tuple
   char data[];
 } STSRow;
@@ -241,13 +239,13 @@ static FORCE_INLINE int32_t tdGetBitmapValType(const void *pBitmap, int16_t colI
 static FORCE_INLINE bool    tdIsBitmapValTypeNorm(const void *pBitmap, int16_t idx, int8_t bitmapMode);
 bool                        tdIsBitmapBlkNorm(const void *pBitmap, int32_t numOfBits, int8_t bitmapMode);
 int32_t tdAppendValToDataCol(SDataCol *pCol, TDRowValT valType, const void *val, int32_t numOfRows, int32_t maxPoints,
-                             int8_t bitmapMode);
+                             int8_t bitmapMode, bool isMerge);
 static FORCE_INLINE int32_t tdAppendColValToTpRow(SRowBuilder *pBuilder, TDRowValT valType, const void *val,
                                                   bool isCopyVarData, int8_t colType, int16_t colIdx, int32_t offset);
 static FORCE_INLINE int32_t tdAppendColValToKvRow(SRowBuilder *pBuilder, TDRowValT valType, const void *val,
                                                   bool isCopyVarData, int8_t colType, int16_t colIdx, int32_t offset,
                                                   col_id_t colId);
-int32_t                     tdAppendSTSRowToDataCol(STSRow *pRow, STSchema *pSchema, SDataCols *pCols);
+int32_t                     tdAppendSTSRowToDataCol(STSRow *pRow, STSchema *pSchema, SDataCols *pCols, bool isMerge);
 
 /**
  * @brief
@@ -718,6 +716,7 @@ static int32_t tdSRowResetBuf(SRowBuilder *pBuilder, void *pBuf) {
       terrno = TSDB_CODE_INVALID_PARA;
       return terrno;
   }
+
   return TSDB_CODE_SUCCESS;
 }
 

include/common/ttypes.h

@@ -186,14 +186,14 @@ typedef struct {
 #define IS_NUMERIC_TYPE(_t) ((IS_SIGNED_NUMERIC_TYPE(_t)) || (IS_UNSIGNED_NUMERIC_TYPE(_t)) || (IS_FLOAT_TYPE(_t)))
 #define IS_MATHABLE_TYPE(_t) (IS_NUMERIC_TYPE(_t) || (_t) == (TSDB_DATA_TYPE_BOOL) || (_t) == (TSDB_DATA_TYPE_TIMESTAMP))
 
-#define IS_VALID_TINYINT(_t)   ((_t) > INT8_MIN && (_t) <= INT8_MAX)
+#define IS_VALID_TINYINT(_t)   ((_t) >= INT8_MIN && (_t) <= INT8_MAX)
-#define IS_VALID_SMALLINT(_t)  ((_t) > INT16_MIN && (_t) <= INT16_MAX)
+#define IS_VALID_SMALLINT(_t)  ((_t) >= INT16_MIN && (_t) <= INT16_MAX)
-#define IS_VALID_INT(_t)       ((_t) > INT32_MIN && (_t) <= INT32_MAX)
+#define IS_VALID_INT(_t)       ((_t) >= INT32_MIN && (_t) <= INT32_MAX)
-#define IS_VALID_BIGINT(_t)    ((_t) > INT64_MIN && (_t) <= INT64_MAX)
+#define IS_VALID_BIGINT(_t)    ((_t) >= INT64_MIN && (_t) <= INT64_MAX)
-#define IS_VALID_UTINYINT(_t)  ((_t) >= 0 && (_t) < UINT8_MAX)
+#define IS_VALID_UTINYINT(_t)  ((_t) >= 0 && (_t) <= UINT8_MAX)
-#define IS_VALID_USMALLINT(_t) ((_t) >= 0 && (_t) < UINT16_MAX)
+#define IS_VALID_USMALLINT(_t) ((_t) >= 0 && (_t) <= UINT16_MAX)
-#define IS_VALID_UINT(_t)      ((_t) >= 0 && (_t) < UINT32_MAX)
+#define IS_VALID_UINT(_t)      ((_t) >= 0 && (_t) <= UINT32_MAX)
-#define IS_VALID_UBIGINT(_t)   ((_t) >= 0 && (_t) < UINT64_MAX)
+#define IS_VALID_UBIGINT(_t)   ((_t) >= 0 && (_t) <= UINT64_MAX)
 #define IS_VALID_FLOAT(_t)     ((_t) >= -FLT_MAX && (_t) <= FLT_MAX)
 #define IS_VALID_DOUBLE(_t)    ((_t) >= -DBL_MAX && (_t) <= DBL_MAX)
 

include/libs/index/index.h

@@ -56,7 +56,8 @@ typedef enum {
   QUERY_LESS_EQUAL,
   QUERY_GREATER_THAN,
   QUERY_GREATER_EQUAL,
-  QUERY_RANGE
+  QUERY_RANGE,
+  QUERY_MAX
 } EIndexQueryType;
 
 /*
@@ -178,8 +179,8 @@ void        indexOptsDestroy(SIndexOpts* opts);
  * @param:
  */
 
-SIndexTerm* indexTermCreate(int64_t suid, SIndexOperOnColumn operType, int8_t qType, uint8_t colType,
+SIndexTerm* indexTermCreate(int64_t suid, SIndexOperOnColumn operType, uint8_t colType, const char* colName,
-                            const char* colName, int32_t nColName, const char* colVal, int32_t nColVal);
+                            int32_t nColName, const char* colVal, int32_t nColVal);
 void        indexTermDestroy(SIndexTerm* p);
 
 /*

include/libs/parser/parser.h

@@ -91,17 +91,23 @@ int32_t qCloneStmtDataBlock(void** pDst, void* pSrc);
 void    qFreeStmtDataBlock(void* pDataBlock);
 int32_t qRebuildStmtDataBlock(void** pDst, void* pSrc);
 void    qDestroyStmtDataBlock(void* pBlock);
+
 int32_t qBindStmtColsValue(void* pBlock, TAOS_MULTI_BIND* bind, char* msgBuf, int32_t msgBufLen);
 int32_t qBindStmtSingleColValue(void* pBlock, TAOS_MULTI_BIND* bind, char* msgBuf, int32_t msgBufLen, int32_t colIdx,
                                 int32_t rowNum);
 int32_t qBuildStmtColFields(void* pDataBlock, int32_t* fieldNum, TAOS_FIELD** fields);
 int32_t qBuildStmtTagFields(void* pBlock, void* boundTags, int32_t* fieldNum, TAOS_FIELD** fields);
-int32_t qBindStmtTagsValue(void* pBlock, void* boundTags, int64_t suid, SName* pName, TAOS_MULTI_BIND* bind,
+int32_t qBindStmtTagsValue(void* pBlock, void* boundTags, int64_t suid, char* tName, TAOS_MULTI_BIND* bind,
                            char* msgBuf, int32_t msgBufLen);
 void    destroyBoundColumnInfo(void* pBoundInfo);
 int32_t qCreateSName(SName* pName, const char* pTableName, int32_t acctId, char* dbName, char* msgBuf,
                      int32_t msgBufLen);
 
+void*   smlInitHandle(SQuery *pQuery);
+void    smlDestroyHandle(void *pHandle);
+int32_t smlBindData(void *handle, SArray *tags, SArray *colsFormat, SHashObj *colsHash, SArray *cols, bool format, STableMeta *pTableMeta, char *tableName, char *msgBuf, int16_t msgBufLen);
+int32_t smlBuildOutput(void* handle, SHashObj* pVgHash);
+
 #ifdef __cplusplus
 }
 #endif

include/util/taoserror.h

@@ -642,6 +642,10 @@ int32_t* taosGetErrno();
 #define TSDB_CODE_FUNC_FUNTION_PARA_VALUE       TAOS_DEF_ERROR_CODE(0, 0x2803)
 #define TSDB_CODE_FUNC_INVALID_FUNTION          TAOS_DEF_ERROR_CODE(0, 0x2604)
 
+#define TSDB_CODE_SML_INVALID_PROTOCOL_TYPE     TAOS_DEF_ERROR_CODE(0, 0x3000)
+#define TSDB_CODE_SML_INVALID_PRECISION_TYPE    TAOS_DEF_ERROR_CODE(0, 0x3001)
+#define TSDB_CODE_SML_INVALID_DATA              TAOS_DEF_ERROR_CODE(0, 0x3002)
+
 #ifdef __cplusplus
 }
 #endif

include/util/tdef.h

@@ -29,6 +29,8 @@ extern "C" {
 #define TSKEY_MAX         (INT64_MAX - 1)
 #define TSKEY_INITIAL_VAL TSKEY_MIN
 
+#define TD_VER_MAX UINT64_MAX  // TODO: use the real max version from query handle
+
 // Bytes for each type.
 extern const int32_t TYPE_BYTES[15];
 
@@ -84,9 +86,13 @@ extern const int32_t TYPE_BYTES[15];
 #define TS_PATH_DELIMITER "."
 #define TS_ESCAPE_CHAR    '`'
 
+
 #define TSDB_TIME_PRECISION_MILLI 0
 #define TSDB_TIME_PRECISION_MICRO 1
 #define TSDB_TIME_PRECISION_NANO  2
+#define TSDB_TIME_PRECISION_HOURS   3
+#define TSDB_TIME_PRECISION_MINUTES 4
+#define TSDB_TIME_PRECISION_SECONDS 5
 
 #define TSDB_TIME_PRECISION_MILLI_STR "ms"
 #define TSDB_TIME_PRECISION_MICRO_STR "us"

source/client/src/clientMain.c

@@ -787,8 +787,3 @@ int taos_stmt_close(TAOS_STMT *stmt) {
   return stmtClose(stmt);
 }
 
-TAOS_RES *taos_schemaless_insert(TAOS *taos, char *lines[], int numLines, int protocol, int precision) {
-  // TODO
-  return NULL;
-}
-

source/client/src/clientStmt.c

@@ -458,7 +458,7 @@ int stmtSetTbTags(TAOS_STMT *stmt, TAOS_MULTI_BIND *tags) {
     STMT_ERR_RET(TSDB_CODE_QRY_APP_ERROR);
   }
   
-  STMT_ERR_RET(qBindStmtTagsValue(*pDataBlock, pStmt->bInfo.boundTags, pStmt->bInfo.tbSuid, &pStmt->bInfo.sname, tags, pStmt->exec.pRequest->msgBuf, pStmt->exec.pRequest->msgBufLen));
+  STMT_ERR_RET(qBindStmtTagsValue(*pDataBlock, pStmt->bInfo.boundTags, pStmt->bInfo.tbSuid, pStmt->bInfo.sname.tname, tags, pStmt->exec.pRequest->msgBuf, pStmt->exec.pRequest->msgBufLen));
 
   return TSDB_CODE_SUCCESS;
 }

source/client/test/CMakeLists.txt

@@ -17,14 +17,27 @@ TARGET_LINK_LIBRARIES(
         PUBLIC os util common transport parser catalog scheduler function gtest taos_static qcom
 )
 
+ADD_EXECUTABLE(smlTest smlTest.cpp)
+TARGET_LINK_LIBRARIES(
+        smlTest
+        PUBLIC os util common transport parser catalog scheduler function gtest taos_static qcom
+)
+
 TARGET_INCLUDE_DIRECTORIES(
         clientTest
-        PUBLIC "${TD_SOURCE_DIR}/include/libs/client/"
+        PUBLIC "${TD_SOURCE_DIR}/include/client/"
-        PRIVATE "${TD_SOURCE_DIR}/source/libs/client/inc"
+        PRIVATE "${TD_SOURCE_DIR}/source/client/inc"
 )
 
 TARGET_INCLUDE_DIRECTORIES(
         tmqTest
-        PUBLIC "${TD_SOURCE_DIR}/include/libs/client/"
+        PUBLIC "${TD_SOURCE_DIR}/include/client/"
-        PRIVATE "${TD_SOURCE_DIR}/source/libs/client/inc"
+        PRIVATE "${TD_SOURCE_DIR}/source/client/inc"
 )
+
+TARGET_INCLUDE_DIRECTORIES(
+        smlTest
+        PUBLIC "${TD_SOURCE_DIR}/include/client/"
+        PRIVATE "${TD_SOURCE_DIR}/source/client/inc"
+)
+

source/client/test/smlTest.cpp

@@ -0,0 +1,498 @@
+/*
+ * 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 <gtest/gtest.h>
+
+#include <taoserror.h>
+#include <tglobal.h>
+#include <iostream>
+
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wwrite-strings"
+#pragma GCC diagnostic ignored "-Wunused-function"
+#pragma GCC diagnostic ignored "-Wunused-variable"
+#pragma GCC diagnostic ignored "-Wsign-compare"
+
+#include "../src/clientSml.c"
+#include "taos.h"
+
+int main(int argc, char **argv) {
+  testing::InitGoogleTest(&argc, argv);
+  return RUN_ALL_TESTS();
+}
+
+TEST(testCase, smlParseString_Test) {
+  char       msg[256] = {0};
+  SSmlMsgBuf msgBuf;
+  msgBuf.buf = msg;
+  msgBuf.len = 256;
+  SSmlLineInfo elements = {0};
+
+  // case 1
+  char *sql = "st,t1=3,t2=4,t3=t3 c1=3i64,c3=\"passit hello,c1=2\",c2=false,c4=4f64 1626006833639000000    ,32,c=3";
+  int   ret = smlParseString(sql, &elements, &msgBuf);
+  ASSERT_EQ(ret, 0);
+  ASSERT_EQ(elements.measure, sql);
+  ASSERT_EQ(elements.measureLen, strlen("st"));
+  ASSERT_EQ(elements.measureTagsLen, strlen("st,t1=3,t2=4,t3=t3"));
+
+  ASSERT_EQ(elements.tags, sql + elements.measureLen + 1);
+  ASSERT_EQ(elements.tagsLen, strlen("t1=3,t2=4,t3=t3"));
+
+  ASSERT_EQ(elements.cols, sql + elements.measureTagsLen + 1);
+  ASSERT_EQ(elements.colsLen, strlen("c1=3i64,c3=\"passit hello,c1=2\",c2=false,c4=4f64"));
+
+  ASSERT_EQ(elements.timestamp, sql + elements.measureTagsLen + 1 + elements.colsLen + 1);
+  ASSERT_EQ(elements.timestampLen, strlen("1626006833639000000"));
+
+  // case 2  false
+  sql = "st,t1=3,t2=4,t3=t3 c1=3i64,c3=\"passit hello,c1=2,c2=false,c4=4f64 1626006833639000000";
+  memset(&elements, 0, sizeof(SSmlLineInfo));
+  ret = smlParseString(sql, &elements, &msgBuf);
+  ASSERT_NE(ret, 0);
+
+  // case 3  false
+  sql = "st, t1=3,t2=4,t3=t3 c1=3i64,c3=\"passit hello,c1=2,c2=false,c4=4f64 1626006833639000000";
+  memset(&elements, 0, sizeof(SSmlLineInfo));
+  ret = smlParseString(sql, &elements, &msgBuf);
+  ASSERT_EQ(ret, 0);
+  ASSERT_EQ(elements.cols, sql + elements.measureTagsLen + 2);
+  ASSERT_EQ(elements.colsLen, strlen("t1=3,t2=4,t3=t3"));
+
+  // case 4  tag is null
+  sql = "st, c1=3i64,c3=\"passit hello,c1=2\",c2=false,c4=4f64 1626006833639000000";
+  memset(&elements, 0, sizeof(SSmlLineInfo));
+  ret = smlParseString(sql, &elements, &msgBuf);
+  ASSERT_EQ(ret, 0);
+  ASSERT_EQ(elements.measure, sql);
+  ASSERT_EQ(elements.measureLen, strlen("st"));
+  ASSERT_EQ(elements.measureTagsLen, strlen("st"));
+
+  ASSERT_EQ(elements.tags, sql + elements.measureLen + 1);
+  ASSERT_EQ(elements.tagsLen, 0);
+
+  ASSERT_EQ(elements.cols, sql + elements.measureTagsLen + 2);
+  ASSERT_EQ(elements.colsLen, strlen("c1=3i64,c3=\"passit hello,c1=2\",c2=false,c4=4f64"));
+
+  ASSERT_EQ(elements.timestamp, sql + elements.measureTagsLen + 2 + elements.colsLen + 1);
+  ASSERT_EQ(elements.timestampLen, strlen("1626006833639000000"));
+
+  // case 5 tag is null
+  sql = " st   c1=3i64,c3=\"passit hello,c1=2\",c2=false,c4=4f64  1626006833639000000 ";
+  memset(&elements, 0, sizeof(SSmlLineInfo));
+  ret = smlParseString(sql, &elements, &msgBuf);
+  sql++;
+  ASSERT_EQ(ret, 0);
+  ASSERT_EQ(elements.measure, sql);
+  ASSERT_EQ(elements.measureLen, strlen("st"));
+  ASSERT_EQ(elements.measureTagsLen, strlen("st"));
+
+  ASSERT_EQ(elements.tags, sql + elements.measureLen);
+  ASSERT_EQ(elements.tagsLen, 0);
+
+  ASSERT_EQ(elements.cols, sql + elements.measureTagsLen + 3);
+  ASSERT_EQ(elements.colsLen, strlen("c1=3i64,c3=\"passit hello,c1=2\",c2=false,c4=4f64"));
+
+  ASSERT_EQ(elements.timestamp, sql + elements.measureTagsLen + 3 + elements.colsLen + 2);
+  ASSERT_EQ(elements.timestampLen, strlen("1626006833639000000"));
+
+  // case 6
+  sql = " st   c1=3i64,c3=\"passit hello,c1=2\",c2=false,c4=4f64   ";
+  memset(&elements, 0, sizeof(SSmlLineInfo));
+  ret = smlParseString(sql, &elements, &msgBuf);
+  ASSERT_EQ(ret, 0);
+
+  // case 7
+  sql = " st   ,   ";
+  memset(&elements, 0, sizeof(SSmlLineInfo));
+  ret = smlParseString(sql, &elements, &msgBuf);
+  sql++;
+  ASSERT_EQ(ret, 0);
+  ASSERT_EQ(elements.cols, sql + elements.measureTagsLen + 3);
+  ASSERT_EQ(elements.colsLen, strlen(","));
+
+  // case 8 false
+  sql = ", st   ,   ";
+  memset(&elements, 0, sizeof(SSmlLineInfo));
+  ret = smlParseString(sql, &elements, &msgBuf);
+  ASSERT_NE(ret, 0);
+}
+
+TEST(testCase, smlParseCols_Error_Test) {
+  const char *data[] = {
+    "c=\"89sd",           // binary, nchar
+    "c=j\"89sd\"",
+    "c=\"89sd\"k",
+    "c=u",                // bool
+    "c=truet",
+    "c=f64",              // double
+    "c=8f64f",
+    "c=8ef64",
+    "c=1.7976931348623158e+390f64",
+    "c=f32",              // float
+    "c=8f32f",
+    "c=8wef32",
+    "c=-3.402823466e+39f32",
+    "c=",                 // float
+    "c=8f",
+    "c=8we",
+    "c=3.402823466e+39",
+    "c=i8",              // tiny int
+    "c=-8i8f",
+    "c=8wei8",
+    "c=-999i8",
+    "c=u8",              // u tiny int
+    "c=8fu8",
+    "c=8weu8",
+    "c=999u8",
+    "c=-8u8",
+    "c=i16",              // small int
+    "c=8fi16u",
+    "c=8wei16",
+    "c=-67787i16",
+    "c=u16",              // u small int
+    "c=8u16f",
+    "c=8weu16",
+    "c=-9u16",
+    "c=67787u16",
+    "c=i32",                // int
+    "c=8i32f",
+    "c=8wei32",
+    "c=2147483649i32",
+    "c=u32",                // u int
+    "c=8u32f",
+    "c=8weu32",
+    "c=-4u32",
+    "c=42949672958u32",
+    "c=i64",                // big int
+    "c=8i64i",
+    "c=8wei64",
+    "c=-9223372036854775809i64",
+    "c=i",                  // big int
+    "c=8fi",
+    "c=8wei",
+    "c=9223372036854775808i",
+    "c=u64",                // u big int
+    "c=8u64f",
+    "c=8weu64",
+    "c=-3.402823466e+39u64",
+    "c=-339u64",
+    "c=18446744073709551616u64",
+  };
+
+  for(int i = 0; i < sizeof(data)/sizeof(data[0]); i++){
+    char       msg[256] = {0};
+    SSmlMsgBuf msgBuf;
+    msgBuf.buf = msg;
+    msgBuf.len = 256;
+    int32_t len = strlen(data[i]);
+    int32_t ret = smlParseCols(data[i], len, NULL, false, &msgBuf);
+    ASSERT_NE(ret, TSDB_CODE_SUCCESS);
+  }
+}
+
+TEST(testCase, smlParseCols_tag_Test) {
+  char       msg[256] = {0};
+  SSmlMsgBuf msgBuf;
+  msgBuf.buf = msg;
+  msgBuf.len = 256;
+
+  SArray *cols = taosArrayInit(16, POINTER_BYTES);
+  ASSERT_NE(cols, NULL);
+
+  const char *data =
+      "cbin=\"passit hello,c=2\",cnch=L\"iisdfsf\",cbool=false,cf64=4.31f64,cf32_=8.32,cf32=8.23f32,ci8=-34i8,cu8=89u8,ci16=233i16,cu16=898u16,ci32=98289i32,cu32=12323u32,ci64=-89238i64,ci=989i,cu64=8989323u64,cbooltrue=true,cboolt=t,cboolf=f,cnch_=l\"iuwq\"";
+  int32_t len = strlen(data);
+  int32_t ret = smlParseCols(data, len, cols, true, &msgBuf);
+  ASSERT_EQ(ret, TSDB_CODE_SUCCESS);
+  int32_t size = taosArrayGetSize(cols);
+  ASSERT_EQ(size, 19);
+
+  // nchar
+  SSmlKv *kv = taosArrayGetP(cols, 0);
+  ASSERT_EQ(strncasecmp(kv->key, "cbin", 4), 0);
+  ASSERT_EQ(kv->keyLen, 4);
+  ASSERT_EQ(kv->type, TSDB_DATA_TYPE_NCHAR);
+  ASSERT_EQ(kv->valueLen, 18);
+  ASSERT_EQ(strncasecmp(kv->value, "\"passit", 7), 0);
+  taosMemoryFree(kv);
+
+  // nchar
+  kv = taosArrayGetP(cols, 3);
+  ASSERT_EQ(strncasecmp(kv->key, "cf64", 4), 0);
+  ASSERT_EQ(kv->keyLen, 4);
+  ASSERT_EQ(kv->type, TSDB_DATA_TYPE_NCHAR);
+  ASSERT_EQ(kv->valueLen, 7);
+  ASSERT_EQ(strncasecmp(kv->value, "4.31f64", 7), 0);
+  taosMemoryFree(kv);
+
+  taosArrayClear(cols);
+  data = "t=3e";
+  len = 0;
+  memset(msgBuf.buf, 0, msgBuf.len);
+  ret = smlParseCols(data, len, cols, true, &msgBuf);
+  ASSERT_EQ(ret, TSDB_CODE_SUCCESS);
+  size = taosArrayGetSize(cols);
+  ASSERT_EQ(size, 1);
+
+  // nchar
+  kv = taosArrayGetP(cols, 0);
+  ASSERT_EQ(strncasecmp(kv->key, TAG, strlen(TAG)), 0);
+  ASSERT_EQ(kv->keyLen, strlen(TAG));
+  ASSERT_EQ(kv->type, TSDB_DATA_TYPE_NCHAR);
+  ASSERT_EQ(kv->valueLen, strlen(TAG));
+  ASSERT_EQ(strncasecmp(kv->value, TAG, strlen(TAG)), 0);
+  taosMemoryFree(kv);
+}
+
+TEST(testCase, smlParseCols_Test) {
+  char       msg[256] = {0};
+  SSmlMsgBuf msgBuf;
+  msgBuf.buf = msg;
+  msgBuf.len = 256;
+
+  SArray *cols = taosArrayInit(16, POINTER_BYTES);
+  ASSERT_NE(cols, NULL);
+
+  const char *data = "cbin=\"passit hello,c=2\",cnch=L\"iisdfsf\",cbool=false,cf64=4.31f64,cf32_=8.32,cf32=8.23f32,ci8=-34i8,cu8=89u8,ci16=233i16,cu16=898u16,ci32=98289i32,cu32=12323u32,ci64=-89238i64,ci=989i,cu64=8989323u64,cbooltrue=true,cboolt=t,cboolf=f,cnch_=l\"iuwq\"";
+  int32_t len = strlen(data);
+  int32_t ret = smlParseCols(data, len, cols, false, &msgBuf);
+  ASSERT_EQ(ret, TSDB_CODE_SUCCESS);
+  int32_t size = taosArrayGetSize(cols);
+  ASSERT_EQ(size, 19);
+
+  // binary
+  SSmlKv *kv = taosArrayGetP(cols, 0);
+  ASSERT_EQ(strncasecmp(kv->key, "cbin", 4), 0);
+  ASSERT_EQ(kv->keyLen, 4);
+  ASSERT_EQ(kv->type, TSDB_DATA_TYPE_BINARY);
+  ASSERT_EQ(kv->length, 16);
+  ASSERT_EQ(strncasecmp(kv->value, "passit", 6), 0);
+  taosMemoryFree(kv);
+
+  // nchar
+  kv = taosArrayGetP(cols, 1);
+  ASSERT_EQ(strncasecmp(kv->key, "cnch", 4), 0);
+  ASSERT_EQ(kv->keyLen, 4);
+  ASSERT_EQ(kv->type, TSDB_DATA_TYPE_NCHAR);
+  ASSERT_EQ(kv->length, 7);
+  ASSERT_EQ(strncasecmp(kv->value, "iisd", 4), 0);
+  taosMemoryFree(kv);
+
+  // bool
+  kv = taosArrayGetP(cols, 2);
+  ASSERT_EQ(strncasecmp(kv->key, "cbool", 5), 0);
+  ASSERT_EQ(kv->keyLen, 5);
+  ASSERT_EQ(kv->type, TSDB_DATA_TYPE_BOOL);
+  ASSERT_EQ(kv->length, 1);
+  ASSERT_EQ(kv->i, false);
+  taosMemoryFree(kv);
+
+  // double
+  kv = taosArrayGetP(cols, 3);
+  ASSERT_EQ(strncasecmp(kv->key, "cf64", 4), 0);
+  ASSERT_EQ(kv->keyLen, 4);
+  ASSERT_EQ(kv->type, TSDB_DATA_TYPE_DOUBLE);
+  ASSERT_EQ(kv->length, 8);
+  //ASSERT_EQ(kv->d, 4.31);
+  printf("4.31 = kv->f:%f\n", kv->d);
+  taosMemoryFree(kv);
+
+  // float
+  kv = taosArrayGetP(cols, 4);
+  ASSERT_EQ(strncasecmp(kv->key, "cf32_", 5), 0);
+  ASSERT_EQ(kv->keyLen, 5);
+  ASSERT_EQ(kv->type, TSDB_DATA_TYPE_FLOAT);
+  ASSERT_EQ(kv->length, 4);
+  //ASSERT_EQ(kv->f, 8.32);
+  printf("8.32 = kv->f:%f\n", kv->f);
+  taosMemoryFree(kv);
+
+  // float
+  kv = taosArrayGetP(cols, 5);
+  ASSERT_EQ(strncasecmp(kv->key, "cf32", 4), 0);
+  ASSERT_EQ(kv->keyLen, 4);
+  ASSERT_EQ(kv->type, TSDB_DATA_TYPE_FLOAT);
+  ASSERT_EQ(kv->length, 4);
+  //ASSERT_EQ(kv->f, 8.23);
+  printf("8.23 = kv->f:%f\n", kv->f);
+  taosMemoryFree(kv);
+
+  // tiny int
+  kv = taosArrayGetP(cols, 6);
+  ASSERT_EQ(strncasecmp(kv->key, "ci8", 3), 0);
+  ASSERT_EQ(kv->keyLen, 3);
+  ASSERT_EQ(kv->type, TSDB_DATA_TYPE_TINYINT);
+  ASSERT_EQ(kv->length, 1);
+  ASSERT_EQ(kv->i, -34);
+  taosMemoryFree(kv);
+
+  // unsigned tiny int
+  kv = taosArrayGetP(cols, 7);
+  ASSERT_EQ(strncasecmp(kv->key, "cu8", 3), 0);
+  ASSERT_EQ(kv->keyLen, 3);
+  ASSERT_EQ(kv->type, TSDB_DATA_TYPE_UTINYINT);
+  ASSERT_EQ(kv->length, 1);
+  ASSERT_EQ(kv->u, 89);
+  taosMemoryFree(kv);
+
+  // small int
+  kv = taosArrayGetP(cols, 8);
+  ASSERT_EQ(strncasecmp(kv->key, "ci16", 4), 0);
+  ASSERT_EQ(kv->keyLen, 4);
+  ASSERT_EQ(kv->type, TSDB_DATA_TYPE_SMALLINT);
+  ASSERT_EQ(kv->length, 2);
+  ASSERT_EQ(kv->u, 233);
+  taosMemoryFree(kv);
+
+  // unsigned smallint
+  kv = taosArrayGetP(cols, 9);
+  ASSERT_EQ(strncasecmp(kv->key, "cu16", 4), 0);
+  ASSERT_EQ(kv->keyLen, 4);
+  ASSERT_EQ(kv->type, TSDB_DATA_TYPE_USMALLINT);
+  ASSERT_EQ(kv->length, 2);
+  ASSERT_EQ(kv->u, 898);
+  taosMemoryFree(kv);
+
+  // int
+  kv = taosArrayGetP(cols, 10);
+  ASSERT_EQ(strncasecmp(kv->key, "ci32", 4), 0);
+  ASSERT_EQ(kv->keyLen, 4);
+  ASSERT_EQ(kv->type, TSDB_DATA_TYPE_INT);
+  ASSERT_EQ(kv->length, 4);
+  ASSERT_EQ(kv->u, 98289);
+  taosMemoryFree(kv);
+
+  // unsigned int
+  kv = taosArrayGetP(cols, 11);
+  ASSERT_EQ(strncasecmp(kv->key, "cu32", 4), 0);
+  ASSERT_EQ(kv->keyLen, 4);
+  ASSERT_EQ(kv->type, TSDB_DATA_TYPE_UINT);
+  ASSERT_EQ(kv->length, 4);
+  ASSERT_EQ(kv->u, 12323);
+  taosMemoryFree(kv);
+
+
+  // bigint
+  kv = taosArrayGetP(cols, 12);
+  ASSERT_EQ(strncasecmp(kv->key, "ci64", 4), 0);
+  ASSERT_EQ(kv->keyLen, 4);
+  ASSERT_EQ(kv->type, TSDB_DATA_TYPE_BIGINT);
+  ASSERT_EQ(kv->length, 8);
+  ASSERT_EQ(kv->i, -89238);
+  taosMemoryFree(kv);
+
+  // bigint
+  kv = taosArrayGetP(cols, 13);
+  ASSERT_EQ(strncasecmp(kv->key, "ci", 2), 0);
+  ASSERT_EQ(kv->keyLen, 2);
+  ASSERT_EQ(kv->type, TSDB_DATA_TYPE_BIGINT);
+  ASSERT_EQ(kv->length, 8);
+  ASSERT_EQ(kv->i, 989);
+  taosMemoryFree(kv);
+
+  // unsigned bigint
+  kv = taosArrayGetP(cols, 14);
+  ASSERT_EQ(strncasecmp(kv->key, "cu64", 4), 0);
+  ASSERT_EQ(kv->keyLen, 4);
+  ASSERT_EQ(kv->type, TSDB_DATA_TYPE_UBIGINT);
+  ASSERT_EQ(kv->length, 8);
+  ASSERT_EQ(kv->u, 8989323);
+  taosMemoryFree(kv);
+
+  // bool
+  kv = taosArrayGetP(cols, 15);
+  ASSERT_EQ(strncasecmp(kv->key, "cbooltrue", 9), 0);
+  ASSERT_EQ(kv->keyLen, 9);
+  ASSERT_EQ(kv->type, TSDB_DATA_TYPE_BOOL);
+  ASSERT_EQ(kv->length, 1);
+  ASSERT_EQ(kv->i, true);
+  taosMemoryFree(kv);
+
+
+  // bool
+  kv = taosArrayGetP(cols, 16);
+  ASSERT_EQ(strncasecmp(kv->key, "cboolt", 6), 0);
+  ASSERT_EQ(kv->keyLen, 6);
+  ASSERT_EQ(kv->type, TSDB_DATA_TYPE_BOOL);
+  ASSERT_EQ(kv->length, 1);
+  ASSERT_EQ(kv->i, true);
+  taosMemoryFree(kv);
+
+  // bool
+  kv = taosArrayGetP(cols, 17);
+  ASSERT_EQ(strncasecmp(kv->key, "cboolf", 6), 0);
+  ASSERT_EQ(kv->keyLen, 6);
+  ASSERT_EQ(kv->type, TSDB_DATA_TYPE_BOOL);
+  ASSERT_EQ(kv->length, 1);
+  ASSERT_EQ(kv->i, false);
+  taosMemoryFree(kv);
+
+  // nchar
+  kv = taosArrayGetP(cols, 18);
+  ASSERT_EQ(strncasecmp(kv->key, "cnch_", 5), 0);
+  ASSERT_EQ(kv->keyLen, 5);
+  ASSERT_EQ(kv->type, TSDB_DATA_TYPE_NCHAR);
+  ASSERT_EQ(kv->length, 4);
+  ASSERT_EQ(strncasecmp(kv->value, "iuwq", 4), 0);
+  taosMemoryFree(kv);
+
+  taosArrayDestroy(cols);
+}
+
+TEST(testCase, smlParseLine_Test) {
+  TAOS *taos = taos_connect("localhost", "root", "taosdata", NULL, 0);
+  ASSERT_NE(taos, NULL);
+
+  TAOS_RES* pRes = taos_query(taos, "create database if not exists sml_db");
+  taos_free_result(pRes);
+
+  pRes = taos_query(taos, "use sml_db");
+  taos_free_result(pRes);
+
+  SRequestObj *request = createRequest(taos, NULL, NULL, TSDB_SQL_INSERT);
+  ASSERT_NE(request, NULL);
+
+  SSmlHandle *info = smlBuildSmlInfo(taos, request, TSDB_SML_LINE_PROTOCOL, TSDB_SML_TIMESTAMP_NANO_SECONDS, true);
+  ASSERT_NE(info, NULL);
+
+  const char *sql[3] = {
+    "readings,name=truck_0,fleet=South,driver=Trish,model=H-2,device_version=v2.3 load_capacity=1500,fuel_capacity=150,nominal_fuel_consumption=12,latitude=52.31854,longitude=4.72037,elevation=124,velocity=0,heading=221,grade=0,fuel_consumption=25 1451606400000000000",
+    "readings,name=truck_1,fleet=South,driver=Albert,model=F-150,device_version=v1.5 load_capacity=2000,fuel_capacity=200,nominal_fuel_consumption=15,latitude=72.45258,longitude=68.83761,elevation=255,velocity=0,heading=181,grade=0,fuel_consumption=25 1451606400000000000",
+    "readings,name=truck_2,fleet=North,driver=Derek,model=F-150,device_version=v1.5 load_capacity=2000,fuel_capacity=200,nominal_fuel_consumption=15,latitude=24.5208,longitude=28.09377,elevation=428,velocity=0,heading=304,grade=0,fuel_consumption=25 1451606400000000000"
+  };
+  smlInsertLines(info, sql, 3);
+//  for (int i = 0; i < 3; i++) {
+//    smlParseLine(info, sql[i]);
+//  }
+}
+
+// TEST(testCase, smlParseTS_Test) {
+//   char       msg[256] = {0};
+//   SSmlMsgBuf msgBuf;
+//   msgBuf.buf = msg;
+//   msgBuf.len = 256;
+//   SSmlLineInfo elements = {0};
+//
+//   SSmlHandle* info = smlBuildSmlInfo(taos, request, protocol, precision, dataFormat);
+//   if(!info){
+//     return (TAOS_RES*)request;
+//   }
+//   ret = smlParseTS(info, elements.timestamp, elements.timestampLen, cols);
+//   if(ret != TSDB_CODE_SUCCESS){
+//     uError("SML:0x%"PRIx64" smlParseTS failed", info->id);
+//     return ret;
+//   }
+// }

source/common/src/tdatablock.c

@@ -1478,11 +1478,11 @@ void blockDebugShowData(const SArray* dataBlocks) {
  * @param uid set as parameter temporarily // TODO: remove this parameter, and the executor should set uid in
  * SDataBlock->info.uid
  * @param suid  // TODO: check with Liao whether suid response is reasonable
- * 
+ *
  * TODO: colId should be set
  */
-int32_t buildSubmitReqFromDataBlock(SSubmitReq** pReq, const SArray* pDataBlocks, STSchema *pTSchema, int32_t vgId, tb_uid_t uid,
+int32_t buildSubmitReqFromDataBlock(SSubmitReq** pReq, const SArray* pDataBlocks, STSchema* pTSchema, int32_t vgId,
-                                        tb_uid_t suid) {
+                                    tb_uid_t uid, tb_uid_t suid) {
   int32_t sz = taosArrayGetSize(pDataBlocks);
   int32_t bufSize = sizeof(SSubmitReq);
   for (int32_t i = 0; i < sz; ++i) {
@@ -1494,16 +1494,16 @@ int32_t buildSubmitReqFromDataBlock(SSubmitReq** pReq, const SArray* pDataBlocks
   ASSERT(bufSize < 3 * 1024 * 1024);
 
   *pReq = taosMemoryCalloc(1, bufSize);
-  if(!(*pReq)) {
+  if (!(*pReq)) {
     terrno = TSDB_CODE_OUT_OF_MEMORY;
     return TSDB_CODE_FAILED;
   }
   void* pDataBuf = *pReq;
 
-  int32_t msgLen = sizeof(SSubmitReq);
+  int32_t     msgLen = sizeof(SSubmitReq);
   int32_t     numOfBlks = 0;
   SRowBuilder rb = {0};
-  tdSRowInit(&rb, 0); // TODO: use the latest version
+  tdSRowInit(&rb, 0);  // TODO: use the latest version
 
   for (int32_t i = 0; i < sz; ++i) {
     SSDataBlock* pDataBlock = taosArrayGet(pDataBlocks, i);
@@ -1511,8 +1511,8 @@ int32_t buildSubmitReqFromDataBlock(SSubmitReq** pReq, const SArray* pDataBlocks
     int32_t      rows = pDataBlock->info.rows;
     int32_t      rowSize = pDataBlock->info.rowSize;
     int64_t      groupId = pDataBlock->info.groupId;
-    
+
-    if(rb.nCols != colNum) {
+    if (rb.nCols != colNum) {
       tdSRowSetTpInfo(&rb, colNum, pTSchema->flen);
     }
 
@@ -1525,10 +1525,10 @@ int32_t buildSubmitReqFromDataBlock(SSubmitReq** pReq, const SArray* pDataBlocks
 
     msgLen += sizeof(SSubmitBlk);
     int32_t dataLen = 0;
-    for (int32_t j = 0; j < rows; ++j) {     // iterate by row
+    for (int32_t j = 0; j < rows; ++j) {                     // iterate by row
-      tdSRowResetBuf(&rb, POINTER_SHIFT(pDataBuf, msgLen)); // set row buf
+      tdSRowResetBuf(&rb, POINTER_SHIFT(pDataBuf, msgLen));  // set row buf
       printf("|");
-      bool  isStartKey = false;
+      bool isStartKey = false;
       for (int32_t k = 0; k < colNum; ++k) {  // iterate by column
         SColumnInfoData* pColInfoData = taosArrayGet(pDataBlock->pDataBlock, k);
         void*            var = POINTER_SHIFT(pColInfoData->pData, j * pColInfoData->info.bytes);
@@ -1536,7 +1536,8 @@ int32_t buildSubmitReqFromDataBlock(SSubmitReq** pReq, const SArray* pDataBlocks
           case TSDB_DATA_TYPE_TIMESTAMP:
             if (!isStartKey) {
               isStartKey = true;
-              tdAppendColValToRow(&rb, PRIMARYKEY_TIMESTAMP_COL_ID, TSDB_DATA_TYPE_TIMESTAMP, TD_VTYPE_NORM, var, true, 0, 0);
+              tdAppendColValToRow(&rb, PRIMARYKEY_TIMESTAMP_COL_ID, TSDB_DATA_TYPE_TIMESTAMP, TD_VTYPE_NORM, var, true,
+                                  0, 0);
             } else {
               tdAppendColValToRow(&rb, 2, TSDB_DATA_TYPE_TIMESTAMP, TD_VTYPE_NORM, var, true, 8, k);
               break;
@@ -1629,14 +1630,14 @@ SSubmitReq* tdBlockToSubmit(const SArray* pBlocks, const STSchema* pTSchema) {
     blkHead->uid = htobe64(pDataBlock->info.uid);
 
     int32_t rows = pDataBlock->info.rows;
-    int32_t maxLen = TD_ROW_MAX_BYTES_FROM_SCHEMA(pTSchema);
+    /*int32_t maxLen = TD_ROW_MAX_BYTES_FROM_SCHEMA(pTSchema);*/
     /*blkHead->dataLen = htonl(rows * maxLen);*/
     blkHead->dataLen = 0;
 
     void*   blockData = POINTER_SHIFT(submitBlk, sizeof(SSubmitBlk));
     STSRow* rowData = blockData;
 
-    for (int32_t j = 0; j < pDataBlock->info.rows; j++) {
+    for (int32_t j = 0; j < rows; j++) {
       SRowBuilder rb = {0};
       tdSRowInit(&rb, pTSchema->version);
       tdSRowSetTpInfo(&rb, pTSchema->numOfCols, pTSchema->flen);

source/common/src/tmsg.c

@@ -1255,7 +1255,6 @@ int32_t tDeserializeSGetUserAuthRspImpl(SCoder *pDecoder, SGetUserAuthRsp *pRsp)
   return 0;
 }
 
-
 int32_t tDeserializeSGetUserAuthRsp(void *buf, int32_t bufLen, SGetUserAuthRsp *pRsp) {
   SCoder decoder = {0};
   tCoderInit(&decoder, TD_LITTLE_ENDIAN, buf, bufLen, TD_DECODER);
@@ -2091,7 +2090,7 @@ void tFreeSUseDbBatchRsp(SUseDbBatchRsp *pRsp) {
   taosArrayDestroy(pRsp->pArray);
 }
 
-int32_t tSerializeSUserAuthBatchRsp(void* buf, int32_t bufLen, SUserAuthBatchRsp* pRsp){
+int32_t tSerializeSUserAuthBatchRsp(void *buf, int32_t bufLen, SUserAuthBatchRsp *pRsp) {
   SCoder encoder = {0};
   tCoderInit(&encoder, TD_LITTLE_ENDIAN, buf, bufLen, TD_ENCODER);
 
@@ -2110,7 +2109,7 @@ int32_t tSerializeSUserAuthBatchRsp(void* buf, int32_t bufLen, SUserAuthBatchRsp
   return tlen;
 }
 
-int32_t tDeserializeSUserAuthBatchRsp(void* buf, int32_t bufLen, SUserAuthBatchRsp* pRsp){
+int32_t tDeserializeSUserAuthBatchRsp(void *buf, int32_t bufLen, SUserAuthBatchRsp *pRsp) {
   SCoder decoder = {0};
   tCoderInit(&decoder, TD_LITTLE_ENDIAN, buf, bufLen, TD_DECODER);
 
@@ -2136,7 +2135,7 @@ int32_t tDeserializeSUserAuthBatchRsp(void* buf, int32_t bufLen, SUserAuthBatchR
   return 0;
 }
 
-void    tFreeSUserAuthBatchRsp(SUserAuthBatchRsp* pRsp){
+void tFreeSUserAuthBatchRsp(SUserAuthBatchRsp *pRsp) {
   int32_t numOfBatch = taosArrayGetSize(pRsp->pArray);
   for (int32_t i = 0; i < numOfBatch; ++i) {
     SGetUserAuthRsp *pUserAuthRsp = taosArrayGet(pRsp->pArray, i);
@@ -2146,7 +2145,6 @@ void    tFreeSUserAuthBatchRsp(SUserAuthBatchRsp* pRsp){
   taosArrayDestroy(pRsp->pArray);
 }
 
-
 int32_t tSerializeSDbCfgReq(void *buf, int32_t bufLen, SDbCfgReq *pReq) {
   SCoder encoder = {0};
   tCoderInit(&encoder, TD_LITTLE_ENDIAN, buf, bufLen, TD_ENCODER);
@@ -3746,6 +3744,7 @@ STSchema *tdGetSTSChemaFromSSChema(SSchema **pSchema, int32_t nCols) {
 int tEncodeSVCreateTbReq(SCoder *pCoder, const SVCreateTbReq *pReq) {
   if (tStartEncode(pCoder) < 0) return -1;
 
+  if (tEncodeI32v(pCoder, pReq->flags) < 0) return -1;
   if (tEncodeI64(pCoder, pReq->uid) < 0) return -1;
   if (tEncodeI64(pCoder, pReq->ctime) < 0) return -1;
 
@@ -3771,6 +3770,7 @@ int tDecodeSVCreateTbReq(SCoder *pCoder, SVCreateTbReq *pReq) {
 
   if (tStartDecode(pCoder) < 0) return -1;
 
+  if (tDecodeI32v(pCoder, &pReq->flags) < 0) return -1;
   if (tDecodeI64(pCoder, &pReq->uid) < 0) return -1;
   if (tDecodeI64(pCoder, &pReq->ctime) < 0) return -1;
 

source/common/src/trow.c

@@ -24,7 +24,8 @@ const uint8_t tdVTypeByte[2][3] = {{
                                    },
                                    {
                                        // 1 bit
-                                       TD_VTYPE_NORM_BYTE_I, TD_VTYPE_NULL_BYTE_I,
+                                       TD_VTYPE_NORM_BYTE_I,  // normal
+                                       TD_VTYPE_NULL_BYTE_I,
                                        TD_VTYPE_NULL_BYTE_I,  // padding
                                    }
 
@@ -33,6 +34,24 @@ const uint8_t tdVTypeByte[2][3] = {{
 // declaration
 static uint8_t tdGetBitmapByte(uint8_t byte);
 
+// static void dataColSetNEleNull(SDataCol *pCol, int nEle);
+
+/**
+ * @brief src2 data has more priority than src1
+ *
+ * @param target
+ * @param src1
+ * @param iter1
+ * @param limit1
+ * @param src2
+ * @param iter2
+ * @param limit2
+ * @param tRows
+ * @param update
+ */
+static void tdMergeTwoDataCols(SDataCols *target, SDataCols *src1, int *iter1, int limit1, SDataCols *src2, int *iter2,
+                               int limit2, int tRows, bool update);
+
 // implementation
 /**
  * @brief Compress bitmap bytes comprised of 2-bits to counterpart of 1-bit.
@@ -229,23 +248,23 @@ static uint8_t tdGetMergedBitmapByte(uint8_t byte) {
 void tdMergeBitmap(uint8_t *srcBitmap, int32_t nBits, uint8_t *dstBitmap) {
   int32_t i = 0, j = 0;
   int32_t nBytes = TD_BITMAP_BYTES(nBits);
-  int32_t nStrictBytes = nBits / 4;
+  int32_t nRoundBytes = nBits / 4;
-  int32_t nPartialBits = nBits - nStrictBytes * 4;
+  int32_t nRemainderBits = nBits - nRoundBytes * 4;
 
-  switch (nPartialBits) {
+  switch (nRemainderBits) {
     case 0:
       // NOTHING TODO
       break;
     case 1: {
-      void *lastByte = POINTER_SHIFT(srcBitmap, nStrictBytes);
+      void *lastByte = POINTER_SHIFT(srcBitmap, nRoundBytes);
       *(uint8_t *)lastByte &= 0xC0;
     } break;
     case 2: {
-      void *lastByte = POINTER_SHIFT(srcBitmap, nStrictBytes);
+      void *lastByte = POINTER_SHIFT(srcBitmap, nRoundBytes);
       *(uint8_t *)lastByte &= 0xF0;
     } break;
     case 3: {
-      void *lastByte = POINTER_SHIFT(srcBitmap, nStrictBytes);
+      void *lastByte = POINTER_SHIFT(srcBitmap, nRoundBytes);
       *(uint8_t *)lastByte &= 0xFC;
     } break;
     default:
@@ -266,10 +285,6 @@ void tdMergeBitmap(uint8_t *srcBitmap, int32_t nBits, uint8_t *dstBitmap) {
   }
 }
 
-// static void dataColSetNEleNull(SDataCol *pCol, int nEle);
-static void tdMergeTwoDataCols(SDataCols *target, SDataCols *src1, int *iter1, int limit1, SDataCols *src2, int *iter2,
-                               int limit2, int tRows, bool forceSetNull);
-
 static FORCE_INLINE void dataColSetNullAt(SDataCol *pCol, int index, bool setBitmap, int8_t bitmapMode) {
   if (IS_VAR_DATA_TYPE(pCol->type)) {
     pCol->dataOff[index] = pCol->len;
@@ -329,7 +344,7 @@ bool tdIsBitmapBlkNorm(const void *pBitmap, int32_t numOfBits, int8_t bitmapMode
     if (*((uint8_t *)pBitmap) != vTypeByte) {
       return false;
     }
-    pBitmap = POINTER_SHIFT(pBitmap, 1);
+    pBitmap = POINTER_SHIFT(pBitmap, i);
   }
 
   int32_t nLeft = numOfBits - nBytes * (bitmapMode == 0 ? TD_VTYPE_BITS : TD_VTYPE_BITS_I);
@@ -410,11 +425,12 @@ STSRow *tdRowDup(STSRow *row) {
  * @param val
  * @param numOfRows
  * @param maxPoints
- * @param bitmapMode  default is 0(2 bits), otherwise 1(1 bit)
+ * @param bitmapMode default is 0(2 bits), otherwise 1(1 bit)
+ * @param isMerge merge to current row
  * @return int
  */
 int tdAppendValToDataCol(SDataCol *pCol, TDRowValT valType, const void *val, int numOfRows, int maxPoints,
-                         int8_t bitmapMode) {
+                         int8_t bitmapMode, bool isMerge) {
   TASSERT(pCol != NULL);
 
   // Assume that the columns not specified during insert/upsert mean None.
@@ -430,33 +446,58 @@ int tdAppendValToDataCol(SDataCol *pCol, TDRowValT valType, const void *val, int
       dataColSetNEleNone(pCol, numOfRows, bitmapMode);
     }
   }
-  if (!tdValTypeIsNorm(valType)) {
+  const void *value = val;
+  if (!tdValTypeIsNorm(valType) || !val) {
     // TODO:
     // 1. back compatibility and easy to debug with codes of 2.0 to save NULL values.
     // 2. later on, considering further optimization, don't save Null/None for VarType.
-    val = getNullValue(pCol->type);
+    value = getNullValue(pCol->type);
   }
-  if (IS_VAR_DATA_TYPE(pCol->type)) {
+  if (!isMerge) {
-    // set offset
+    if (IS_VAR_DATA_TYPE(pCol->type)) {
-    pCol->dataOff[numOfRows] = pCol->len;
+      // set offset
-    // Copy data
+      pCol->dataOff[numOfRows] = pCol->len;
-    memcpy(POINTER_SHIFT(pCol->pData, pCol->len), val, varDataTLen(val));
+      // Copy data
-    // Update the length
+      memcpy(POINTER_SHIFT(pCol->pData, pCol->len), value, varDataTLen(value));
-    pCol->len += varDataTLen(val);
+      // Update the length
-  } else {
+      pCol->len += varDataTLen(value);
-    ASSERT(pCol->len == TYPE_BYTES[pCol->type] * numOfRows);
+    } else {
-    memcpy(POINTER_SHIFT(pCol->pData, pCol->len), val, pCol->bytes);
+      ASSERT(pCol->len == TYPE_BYTES[pCol->type] * numOfRows);
-    pCol->len += pCol->bytes;
+      memcpy(POINTER_SHIFT(pCol->pData, pCol->len), value, pCol->bytes);
+      pCol->len += pCol->bytes;
+    }
+  } else if (!tdValTypeIsNone(valType)) {
+    if (IS_VAR_DATA_TYPE(pCol->type)) {
+      // keep the last offset
+      // discard the last var data
+      int32_t lastVarLen = varDataTLen(POINTER_SHIFT(pCol->pData, pCol->dataOff[numOfRows]));
+      pCol->len -= lastVarLen;
+      // Copy data
+      memcpy(POINTER_SHIFT(pCol->pData, pCol->len), value, varDataTLen(value));
+      // Update the length
+      pCol->len += varDataTLen(value);
+    } else {
+      ASSERT(pCol->len - TYPE_BYTES[pCol->type] == TYPE_BYTES[pCol->type] * numOfRows);
+      memcpy(POINTER_SHIFT(pCol->pData, pCol->len - TYPE_BYTES[pCol->type]), value, pCol->bytes);
+    }
   }
+
 #ifdef TD_SUPPORT_BITMAP
-  tdSetBitmapValType(pCol->pBitmap, numOfRows, valType, bitmapMode);
+  if (!isMerge || !tdValTypeIsNone(valType)) {
+    tdSetBitmapValType(pCol->pBitmap, numOfRows, valType, bitmapMode);
+  }
 #endif
   return 0;
 }
 
 // internal
-static int32_t tdAppendTpRowToDataCol(STSRow *pRow, STSchema *pSchema, SDataCols *pCols) {
+static int32_t tdAppendTpRowToDataCol(STSRow *pRow, STSchema *pSchema, SDataCols *pCols, bool isMerge) {
+#if 0
   ASSERT(pCols->numOfRows == 0 || dataColsKeyLast(pCols) < TD_ROW_KEY(pRow));
+#endif
+
+  // Multi-Version rows with the same key and different versions supported
+  ASSERT(pCols->numOfRows == 0 || dataColsKeyLast(pCols) <= TD_ROW_KEY(pRow));
 
   int   rcol = 1;
   int   dcol = 1;
@@ -464,12 +505,14 @@ static int32_t tdAppendTpRowToDataCol(STSRow *pRow, STSchema *pSchema, SDataCols
 
   SDataCol *pDataCol = &(pCols->cols[0]);
   ASSERT(pDataCol->colId == PRIMARYKEY_TIMESTAMP_COL_ID);
-  tdAppendValToDataCol(pDataCol, TD_VTYPE_NORM, &pRow->ts, pCols->numOfRows, pCols->maxPoints, pCols->bitmapMode);
+  tdAppendValToDataCol(pDataCol, TD_VTYPE_NORM, &pRow->ts, pCols->numOfRows, pCols->maxPoints, pCols->bitmapMode,
+                       isMerge);
 
   while (dcol < pCols->numOfCols) {
     pDataCol = &(pCols->cols[dcol]);
     if (rcol >= schemaNCols(pSchema)) {
-      tdAppendValToDataCol(pDataCol, TD_VTYPE_NULL, NULL, pCols->numOfRows, pCols->maxPoints, pCols->bitmapMode);
+      tdAppendValToDataCol(pDataCol, TD_VTYPE_NULL, NULL, pCols->numOfRows, pCols->maxPoints, pCols->bitmapMode,
+                           isMerge);
       ++dcol;
       continue;
     }
@@ -480,22 +523,26 @@ static int32_t tdAppendTpRowToDataCol(STSRow *pRow, STSchema *pSchema, SDataCols
       if (tdGetTpRowValOfCol(&sVal, pRow, pBitmap, pRowCol->type, pRowCol->offset - sizeof(TSKEY), rcol - 1) < 0) {
         return terrno;
       }
-      tdAppendValToDataCol(pDataCol, sVal.valType, sVal.val, pCols->numOfRows, pCols->maxPoints, pCols->bitmapMode);
+      tdAppendValToDataCol(pDataCol, sVal.valType, sVal.val, pCols->numOfRows, pCols->maxPoints, pCols->bitmapMode,
+                           isMerge);
       ++dcol;
       ++rcol;
     } else if (pRowCol->colId < pDataCol->colId) {
       ++rcol;
     } else {
-      tdAppendValToDataCol(pDataCol, TD_VTYPE_NULL, NULL, pCols->numOfRows, pCols->maxPoints, pCols->bitmapMode);
+      tdAppendValToDataCol(pDataCol, TD_VTYPE_NULL, NULL, pCols->numOfRows, pCols->maxPoints, pCols->bitmapMode,
+                           isMerge);
       ++dcol;
     }
   }
+#if 0  
   ++pCols->numOfRows;
+#endif
 
   return TSDB_CODE_SUCCESS;
 }
 // internal
-static int32_t tdAppendKvRowToDataCol(STSRow *pRow, STSchema *pSchema, SDataCols *pCols) {
+static int32_t tdAppendKvRowToDataCol(STSRow *pRow, STSchema *pSchema, SDataCols *pCols, bool isMerge) {
   ASSERT(pCols->numOfRows == 0 || dataColsKeyLast(pCols) < TD_ROW_KEY(pRow));
 
   int   rcol = 0;
@@ -506,12 +553,14 @@ static int32_t tdAppendKvRowToDataCol(STSRow *pRow, STSchema *pSchema, SDataCols
 
   SDataCol *pDataCol = &(pCols->cols[0]);
   ASSERT(pDataCol->colId == PRIMARYKEY_TIMESTAMP_COL_ID);
-  tdAppendValToDataCol(pDataCol, TD_VTYPE_NORM, &pRow->ts, pCols->numOfRows, pCols->maxPoints, pCols->bitmapMode);
+  tdAppendValToDataCol(pDataCol, TD_VTYPE_NORM, &pRow->ts, pCols->numOfRows, pCols->maxPoints, pCols->bitmapMode,
+                       isMerge);
 
   while (dcol < pCols->numOfCols) {
     pDataCol = &(pCols->cols[dcol]);
     if (rcol >= tRowCols || rcol >= tSchemaCols) {
-      tdAppendValToDataCol(pDataCol, TD_VTYPE_NULL, NULL, pCols->numOfRows, pCols->maxPoints, pCols->bitmapMode);
+      tdAppendValToDataCol(pDataCol, TD_VTYPE_NULL, NULL, pCols->numOfRows, pCols->maxPoints, pCols->bitmapMode,
+                           isMerge);
       ++dcol;
       continue;
     }
@@ -527,17 +576,21 @@ static int32_t tdAppendKvRowToDataCol(STSRow *pRow, STSchema *pSchema, SDataCols
       if (tdGetKvRowValOfCol(&sVal, pRow, pBitmap, pIdx->offset, colIdx) < 0) {
         return terrno;
       }
-      tdAppendValToDataCol(pDataCol, sVal.valType, sVal.val, pCols->numOfRows, pCols->maxPoints, pCols->bitmapMode);
+      tdAppendValToDataCol(pDataCol, sVal.valType, sVal.val, pCols->numOfRows, pCols->maxPoints, pCols->bitmapMode,
+                           isMerge);
       ++dcol;
       ++rcol;
     } else if (pIdx->colId < pDataCol->colId) {
       ++rcol;
     } else {
-      tdAppendValToDataCol(pDataCol, TD_VTYPE_NULL, NULL, pCols->numOfRows, pCols->maxPoints, pCols->bitmapMode);
+      tdAppendValToDataCol(pDataCol, TD_VTYPE_NULL, NULL, pCols->numOfRows, pCols->maxPoints, pCols->bitmapMode,
+                           isMerge);
       ++dcol;
     }
   }
+#if 0
   ++pCols->numOfRows;
+#endif
 
   return TSDB_CODE_SUCCESS;
 }
@@ -548,20 +601,30 @@ static int32_t tdAppendKvRowToDataCol(STSRow *pRow, STSchema *pSchema, SDataCols
  * @param pRow
  * @param pSchema
  * @param pCols
- * @param forceSetNull
  */
-int32_t tdAppendSTSRowToDataCol(STSRow *pRow, STSchema *pSchema, SDataCols *pCols) {
+int32_t tdAppendSTSRowToDataCol(STSRow *pRow, STSchema *pSchema, SDataCols *pCols, bool isMerge) {
   if (TD_IS_TP_ROW(pRow)) {
-    return tdAppendTpRowToDataCol(pRow, pSchema, pCols);
+    return tdAppendTpRowToDataCol(pRow, pSchema, pCols, isMerge);
   } else if (TD_IS_KV_ROW(pRow)) {
-    return tdAppendKvRowToDataCol(pRow, pSchema, pCols);
+    return tdAppendKvRowToDataCol(pRow, pSchema, pCols, isMerge);
   } else {
     ASSERT(0);
   }
   return TSDB_CODE_SUCCESS;
 }
 
-int tdMergeDataCols(SDataCols *target, SDataCols *source, int rowsToMerge, int *pOffset, bool forceSetNull,
+/**
+ * @brief source data has more priority than target
+ *
+ * @param target
+ * @param source
+ * @param rowsToMerge
+ * @param pOffset
+ * @param update
+ * @param maxVer
+ * @return int
+ */
+int tdMergeDataCols(SDataCols *target, SDataCols *source, int rowsToMerge, int *pOffset, bool update,
                     TDRowVerT maxVer) {
   ASSERT(rowsToMerge > 0 && rowsToMerge <= source->numOfRows);
   ASSERT(target->numOfCols == source->numOfCols);
@@ -576,17 +639,38 @@ int tdMergeDataCols(SDataCols *target, SDataCols *source, int rowsToMerge, int *
   if ((target->numOfRows == 0) || (dataColsKeyLast(target) < dataColsKeyAtRow(source, *pOffset))) {  // No overlap
     ASSERT(target->numOfRows + rowsToMerge <= target->maxPoints);
     // TODO: filter the maxVer
-    for (int i = 0; i < rowsToMerge; i++) {
+    TSKEY lastKey = TSKEY_INITIAL_VAL;
+    for (int i = 0; i < rowsToMerge; ++i) {
+      bool merge = false;
       for (int j = 0; j < source->numOfCols; j++) {
         if (source->cols[j].len > 0 || target->cols[j].len > 0) {
           SCellVal sVal = {0};
           if (tdGetColDataOfRow(&sVal, source->cols + j, i + (*pOffset), source->bitmapMode) < 0) {
             TASSERT(0);
           }
+
+          if (j == 0) {
+            if (lastKey == *(TSKEY *)sVal.val) {
+              if (!update) {
+                break;
+              }
+              merge = true;
+            } else if (lastKey != TSKEY_INITIAL_VAL) {
+              ++target->numOfRows;
+            }
+
+            lastKey = *(TSKEY *)sVal.val;
+          }
+          if (i == 0) {
+            (target->cols + j)->bitmap = (source->cols + j)->bitmap;
+          }
+
           tdAppendValToDataCol(target->cols + j, sVal.valType, sVal.val, target->numOfRows, target->maxPoints,
-                               target->bitmapMode);
+                               target->bitmapMode, merge);
         }
       }
+    }
+    if (lastKey != TSKEY_INITIAL_VAL) {
       ++target->numOfRows;
     }
     (*pOffset) += rowsToMerge;
@@ -596,7 +680,7 @@ int tdMergeDataCols(SDataCols *target, SDataCols *source, int rowsToMerge, int *
 
     int iter1 = 0;
     tdMergeTwoDataCols(target, pTarget, &iter1, pTarget->numOfRows, source, pOffset, source->numOfRows,
-                       pTarget->numOfRows + rowsToMerge, forceSetNull);
+                       pTarget->numOfRows + rowsToMerge, update);
   }
 
   tdFreeDataCols(pTarget);
@@ -607,67 +691,95 @@ _err:
   return -1;
 }
 
-// src2 data has more priority than src1
+static void tdAppendValToDataCols(SDataCols *target, SDataCols *src, int iter, bool isMerge) {
+  for (int i = 0; i < src->numOfCols; ++i) {
+    ASSERT(target->cols[i].type == src->cols[i].type);
+    if (src->cols[i].len > 0 || target->cols[i].len > 0) {
+      SCellVal sVal = {0};
+      if (tdGetColDataOfRow(&sVal, src->cols + i, iter, src->bitmapMode) < 0) {
+        TASSERT(0);
+      }
+      if (isMerge) {
+        if (!tdValTypeIsNone(sVal.valType)) {
+          tdAppendValToDataCol(&(target->cols[i]), sVal.valType, sVal.val, target->numOfRows, target->maxPoints,
+                               target->bitmapMode, isMerge);
+        } else {
+          // Keep the origin value for None
+        }
+      } else {
+        tdAppendValToDataCol(&(target->cols[i]), sVal.valType, sVal.val, target->numOfRows, target->maxPoints,
+                             target->bitmapMode, isMerge);
+      }
+    }
+  }
+}
+/**
+ * @brief src2 data has more priority than src1
+ *
+ * @param target
+ * @param src1
+ * @param iter1
+ * @param limit1
+ * @param src2
+ * @param iter2
+ * @param limit2
+ * @param tRows
+ * @param update
+ */
 static void tdMergeTwoDataCols(SDataCols *target, SDataCols *src1, int *iter1, int limit1, SDataCols *src2, int *iter2,
-                               int limit2, int tRows, bool forceSetNull) {
+                               int limit2, int tRows, bool update) {
   tdResetDataCols(target);
+  target->bitmapMode = src1->bitmapMode;
   ASSERT(limit1 <= src1->numOfRows && limit2 <= src2->numOfRows);
+  int32_t nRows = 0;
 
-  while (target->numOfRows < tRows) {
+  // TODO: filter the maxVer
+  // TODO: handle the delete function
+  TSKEY lastKey = TSKEY_INITIAL_VAL;
+  while (nRows < tRows) {
     if (*iter1 >= limit1 && *iter2 >= limit2) break;
 
     TSKEY key1 = (*iter1 >= limit1) ? INT64_MAX : dataColsKeyAt(src1, *iter1);
-    TKEY  tkey1 = (*iter1 >= limit1) ? TKEY_NULL : dataColsTKeyAt(src1, *iter1);
+    // TKEY  tkey1 = (*iter1 >= limit1) ? TKEY_NULL : dataColsTKeyAt(src1, *iter1);
     TSKEY key2 = (*iter2 >= limit2) ? INT64_MAX : dataColsKeyAt(src2, *iter2);
     // TKEY  tkey2 = (*iter2 >= limit2) ? TKEY_NULL : dataColsTKeyAt(src2, *iter2);
 
-    ASSERT(tkey1 == TKEY_NULL || (!TKEY_IS_DELETED(tkey1)));
+    // ASSERT(tkey1 == TKEY_NULL || (!TKEY_IS_DELETED(tkey1)));
-    // TODO: filter the maxVer
-    if (key1 < key2) {
-      for (int i = 0; i < src1->numOfCols; ++i) {
-        ASSERT(target->cols[i].type == src1->cols[i].type);
-        if (src1->cols[i].len > 0 || target->cols[i].len > 0) {
-          SCellVal sVal = {0};
-          if (tdGetColDataOfRow(&sVal, src1->cols + i, *iter1, src1->bitmapMode) < 0) {
-            TASSERT(0);
-          }
-          tdAppendValToDataCol(&(target->cols[i]), sVal.valType, sVal.val, target->numOfRows, target->maxPoints,
-                               target->bitmapMode);
-        }
-      }
 
-      ++target->numOfRows;
+    if (key1 <= key2) {
+      // select key1 if not delete
+      if (update && (lastKey == key1)) {
+        tdAppendValToDataCols(target, src1, *iter1, true);
+      } else if (lastKey != key1) {
+        if (lastKey != TSKEY_INITIAL_VAL) {
+          ++target->numOfRows;
+        }
+        tdAppendValToDataCols(target, src1, *iter1, false);
+      }
+      ++nRows;
       ++(*iter1);
-    } else if (key1 >= key2) {
+      lastKey = key1;
-      // TODO: filter the maxVer
+    } else {
-      if ((key1 > key2) || ((key1 == key2) && !TKEY_IS_DELETED(key2))) {
+      // use key2 if not deleted
-        for (int i = 0; i < src2->numOfCols; ++i) {
+      // TODO: handle the delete function
-          SCellVal sVal = {0};
+      if (update && (lastKey == key2)) {
-          ASSERT(target->cols[i].type == src2->cols[i].type);
+        tdAppendValToDataCols(target, src2, *iter2, true);
-          if (tdGetColDataOfRow(&sVal, src2->cols + i, *iter2, src2->bitmapMode) < 0) {
+      } else if (lastKey != key2) {
-            TASSERT(0);
+        if (lastKey != TSKEY_INITIAL_VAL) {
-          }
+          ++target->numOfRows;
-          if (src2->cols[i].len > 0 && !tdValTypeIsNull(sVal.valType)) {
-            tdAppendValToDataCol(&(target->cols[i]), sVal.valType, sVal.val, target->numOfRows, target->maxPoints,
-                                 target->bitmapMode);
-          } else if (!forceSetNull && key1 == key2 && src1->cols[i].len > 0) {
-            if (tdGetColDataOfRow(&sVal, src1->cols + i, *iter1, src1->bitmapMode) < 0) {
-              TASSERT(0);
-            }
-            tdAppendValToDataCol(&(target->cols[i]), sVal.valType, sVal.val, target->numOfRows, target->maxPoints,
-                                 target->bitmapMode);
-          } else if (target->cols[i].len > 0) {
-            dataColSetNullAt(&target->cols[i], target->numOfRows, true, target->bitmapMode);
-          }
         }
-        ++target->numOfRows;
+        tdAppendValToDataCols(target, src2, *iter2, false);
       }
 
+      ++nRows;
       ++(*iter2);
-      if (key1 == key2) ++(*iter1);
+      lastKey = key2;
     }
 
-    ASSERT(target->numOfRows <= target->maxPoints);
+    ASSERT(target->numOfRows <= target->maxPoints - 1);
+  }
+  if (lastKey != TSKEY_INITIAL_VAL) {
+    ++target->numOfRows;
   }
 }
 
@@ -777,7 +889,7 @@ SDataCols *tdDupDataCols(SDataCols *pDataCols, bool keepData) {
   pRet->sversion = pDataCols->sversion;
   if (keepData) pRet->numOfRows = pDataCols->numOfRows;
 
-  for (int i = 0; i < pDataCols->numOfCols; i++) {
+  for (int i = 0; i < pDataCols->numOfCols; ++i) {
     pRet->cols[i].type = pDataCols->cols[i].type;
     pRet->cols[i].bitmap = pDataCols->cols[i].bitmap;
     pRet->cols[i].colId = pDataCols->cols[i].colId;
@@ -797,8 +909,7 @@ SDataCols *tdDupDataCols(SDataCols *pDataCols, bool keepData) {
           memcpy(pRet->cols[i].dataOff, pDataCols->cols[i].dataOff, dataOffSize);
         }
         if (!TD_COL_ROWS_NORM(pRet->cols + i)) {
-          int32_t nBitmapBytes = (int32_t)TD_BITMAP_BYTES(pDataCols->numOfRows);
+          memcpy(pRet->cols[i].pBitmap, pDataCols->cols[i].pBitmap, TD_BITMAP_BYTES(pDataCols->numOfRows));
-          memcpy(pRet->cols[i].pBitmap, pDataCols->cols[i].pBitmap, nBitmapBytes);
         }
       }
     }

source/common/src/ttypes.c

@@ -557,7 +557,7 @@ static const void *nullValues[] = {
 };
 
 const void *getNullValue(int32_t type) {
-  assert(type >= TSDB_DATA_TYPE_BOOL && type <= TSDB_DATA_TYPE_UBIGINT);
+  assert(type >= TSDB_DATA_TYPE_BOOL && type <= TSDB_DATA_TYPE_UBIGINT);  // TODO: extend the types
   return nullValues[type - 1];
 }
 

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

@@ -293,14 +293,10 @@ void vmInitMsgHandle(SMgmtWrapper *pWrapper) {
   dmSetMsgHandle(pWrapper, TDMT_VND_TASK_MERGE_EXEC, vmProcessMergeMsg, DEFAULT_HANDLE);
   dmSetMsgHandle(pWrapper, TDMT_VND_TASK_WRITE_EXEC, vmProcessWriteMsg, DEFAULT_HANDLE);
   dmSetMsgHandle(pWrapper, TDMT_VND_STREAM_TRIGGER, vmProcessFetchMsg, DEFAULT_HANDLE);
-
   dmSetMsgHandle(pWrapper, TDMT_VND_ALTER_VNODE, vmProcessWriteMsg, DEFAULT_HANDLE);
   dmSetMsgHandle(pWrapper, TDMT_VND_COMPACT_VNODE, vmProcessWriteMsg, DEFAULT_HANDLE);
-
   dmSetMsgHandle(pWrapper, TDMT_DND_CREATE_VNODE, vmProcessMgmtMsg, DEFAULT_HANDLE);
   dmSetMsgHandle(pWrapper, TDMT_DND_DROP_VNODE, vmProcessMgmtMsg, DEFAULT_HANDLE);
-  // dmSetMsgHandle(pWrapper, TDMT_DND_SYNC_VNODE, vmProcessMgmtMsg, DEFAULT_HANDLE);
-  // dmSetMsgHandle(pWrapper, TDMT_DND_COMPACT_VNODE, vmProcessMgmtMsg, DEFAULT_HANDLE);
 
   // sync integration
   dmSetMsgHandle(pWrapper, TDMT_VND_SYNC_TIMEOUT, (NodeMsgFp)vmProcessSyncMsg, DEFAULT_HANDLE);

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

@@ -31,6 +31,7 @@ int32_t mndCheckDropUserAuth(SUserObj *pOperUser);
 
 int32_t mndCheckNodeAuth(SUserObj *pOperUser);
 int32_t mndCheckFuncAuth(SUserObj *pOperUser);
+int32_t mndCheckTransAuth(SUserObj *pOperUser);
 
 int32_t mndCheckCreateDbAuth(SUserObj *pOperUser);
 int32_t mndCheckAlterDropCompactDbAuth(SUserObj *pOperUser, SDbObj *pDb);

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

@@ -72,7 +72,7 @@ typedef enum {
   TRN_TYPE_DROP_USER = 1003,
   TRN_TYPE_CREATE_FUNC = 1004,
   TRN_TYPE_DROP_FUNC = 1005,
-  
+
   TRN_TYPE_CREATE_SNODE = 1006,
   TRN_TYPE_DROP_SNODE = 1007,
   TRN_TYPE_CREATE_QNODE = 1008,
@@ -601,7 +601,6 @@ typedef struct {
   int32_t        triggerParam;
   int64_t        waterMark;
   char*          sql;
-  char*          logicalPlan;
   char*          physicalPlan;
   SArray*        tasks;  // SArray<SArray<SStreamTask>>
   SSchemaWrapper outputSchema;

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

@@ -31,6 +31,11 @@ int32_t  mndValidateStbInfo(SMnode *pMnode, SSTableMetaVersion *pStbs, int32_t n
                             int32_t *pRspLen);
 int32_t  mndGetNumOfStbs(SMnode *pMnode, char *dbName, int32_t *pNumOfStbs);
 
+int32_t mndCheckCreateStbReq(SMCreateStbReq *pCreate);
+SDbObj *mndAcquireDbByStb(SMnode *pMnode, const char *stbName);
+int32_t mndBuildStbFromReq(SMnode *pMnode, SStbObj *pDst, SMCreateStbReq *pCreate, SDbObj *pDb);
+int32_t mndAddStbToTrans(SMnode *pMnode, STrans *pTrans, SDbObj *pDb, SStbObj *pStb);
+
 #ifdef __cplusplus
 }
 #endif

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

@@ -27,10 +27,14 @@ void     mndCleanupVgroup(SMnode *pMnode);
 SVgObj  *mndAcquireVgroup(SMnode *pMnode, int32_t vgId);
 void     mndReleaseVgroup(SMnode *pMnode, SVgObj *pVgroup);
 SSdbRaw *mndVgroupActionEncode(SVgObj *pVgroup);
-int32_t  mndAllocVgroup(SMnode *pMnode, SDbObj *pDb, SVgObj **ppVgroups);
 SEpSet   mndGetVgroupEpset(SMnode *pMnode, const SVgObj *pVgroup);
 int32_t  mndGetVnodesNum(SMnode *pMnode, int32_t dnodeId);
 
+int32_t mndAllocVgroup(SMnode *pMnode, SDbObj *pDb, SVgObj **ppVgroups);
+SArray *mndBuildDnodesArray(SMnode *pMnode);
+int32_t mndAddVnodeToVgroup(SMnode *pMnode, SVgObj *pVgroup, SArray *pArray);
+int32_t mndRemoveVnodeFromVgroup(SMnode *pMnode, SVgObj *pVgroup, SArray *pArray, SVnodeGid *del1, SVnodeGid *del2);
+
 void *mndBuildCreateVnodeReq(SMnode *pMnode, SDnodeObj *pDnode, SDbObj *pDb, SVgObj *pVgroup, int32_t *pContLen);
 void *mndBuildDropVnodeReq(SMnode *pMnode, SDnodeObj *pDnode, SDbObj *pDb, SVgObj *pVgroup, int32_t *pContLen);
 void *mndBuildAlterVnodeReq(SMnode *pMnode, SDnodeObj *pDnode, SDbObj *pDb, SVgObj *pVgroup, int32_t *pContLen);

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

@@ -74,10 +74,7 @@ static int32_t mndProcessAuthReq(SNodeMsg *pReq) {
 }
 
 int32_t mndCheckCreateUserAuth(SUserObj *pOperUser) {
-  if (pOperUser->superUser) {
+  if (pOperUser->superUser) return 0;
-    return 0;
-  }
-
   terrno = TSDB_CODE_MND_NO_RIGHTS;
   return -1;
 }
@@ -118,28 +115,25 @@ int32_t mndCheckAlterUserAuth(SUserObj *pOperUser, SUserObj *pUser, SDbObj *pDb,
 }
 
 int32_t mndCheckDropUserAuth(SUserObj *pOperUser) {
-  if (pOperUser->superUser) {
+  if (pOperUser->superUser) return 0;
-    return 0;
-  }
-
   terrno = TSDB_CODE_MND_NO_RIGHTS;
   return -1;
 }
 
 int32_t mndCheckNodeAuth(SUserObj *pOperUser) {
-  if (pOperUser->superUser) {
+  if (pOperUser->superUser) return 0;
-    return 0;
-  }
-
   terrno = TSDB_CODE_MND_NO_RIGHTS;
   return -1;
 }
 
 int32_t mndCheckFuncAuth(SUserObj *pOperUser) {
-  if (pOperUser->superUser) {
+  if (pOperUser->superUser) return 0;
-    return 0;
+  terrno = TSDB_CODE_MND_NO_RIGHTS;
-  }
+  return -1;
+}
 
+int32_t mndCheckTransAuth(SUserObj *pOperUser) {
+  if (pOperUser->superUser) return 0;
   terrno = TSDB_CODE_MND_NO_RIGHTS;
   return -1;
 }

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

@@ -261,6 +261,104 @@ void mndReleaseDb(SMnode *pMnode, SDbObj *pDb) {
   sdbRelease(pSdb, pDb);
 }
 
+static int32_t mndAddCreateVnodeAction(SMnode *pMnode, STrans *pTrans, SDbObj *pDb, SVgObj *pVgroup, SVnodeGid *pVgid,
+                                       bool isRedo) {
+  STransAction action = {0};
+
+  SDnodeObj *pDnode = mndAcquireDnode(pMnode, pVgid->dnodeId);
+  if (pDnode == NULL) return -1;
+  action.epSet = mndGetDnodeEpset(pDnode);
+  mndReleaseDnode(pMnode, pDnode);
+
+  int32_t contLen = 0;
+  void   *pReq = mndBuildCreateVnodeReq(pMnode, pDnode, pDb, pVgroup, &contLen);
+  if (pReq == NULL) return -1;
+
+  action.pCont = pReq;
+  action.contLen = contLen;
+  action.msgType = TDMT_DND_CREATE_VNODE;
+  action.acceptableCode = TSDB_CODE_NODE_ALREADY_DEPLOYED;
+
+  if (isRedo) {
+    if (mndTransAppendRedoAction(pTrans, &action) != 0) {
+      taosMemoryFree(pReq);
+      return -1;
+    }
+  } else {
+    if (mndTransAppendUndoAction(pTrans, &action) != 0) {
+      taosMemoryFree(pReq);
+      return -1;
+    }
+  }
+
+  return 0;
+}
+
+static int32_t mndAddAlterVnodeAction(SMnode *pMnode, STrans *pTrans, SDbObj *pDb, SVgObj *pVgroup, SVnodeGid *pVgid,
+                                      bool isRedo) {
+  STransAction action = {0};
+
+  SDnodeObj *pDnode = mndAcquireDnode(pMnode, pVgid->dnodeId);
+  if (pDnode == NULL) return -1;
+  action.epSet = mndGetDnodeEpset(pDnode);
+  mndReleaseDnode(pMnode, pDnode);
+
+  int32_t contLen = 0;
+  void   *pReq = mndBuildAlterVnodeReq(pMnode, pDnode, pDb, pVgroup, &contLen);
+  if (pReq == NULL) return -1;
+
+  action.pCont = pReq;
+  action.contLen = contLen;
+  action.msgType = TDMT_VND_ALTER_VNODE;
+
+  if (isRedo) {
+    if (mndTransAppendRedoAction(pTrans, &action) != 0) {
+      taosMemoryFree(pReq);
+      return -1;
+    }
+  } else {
+    if (mndTransAppendUndoAction(pTrans, &action) != 0) {
+      taosMemoryFree(pReq);
+      return -1;
+    }
+  }
+
+  return 0;
+}
+
+static int32_t mndAddDropVnodeAction(SMnode *pMnode, STrans *pTrans, SDbObj *pDb, SVgObj *pVgroup, SVnodeGid *pVgid,
+                                     bool isRedo) {
+  STransAction action = {0};
+
+  SDnodeObj *pDnode = mndAcquireDnode(pMnode, pVgid->dnodeId);
+  if (pDnode == NULL) return -1;
+  action.epSet = mndGetDnodeEpset(pDnode);
+  mndReleaseDnode(pMnode, pDnode);
+
+  int32_t contLen = 0;
+  void   *pReq = mndBuildDropVnodeReq(pMnode, pDnode, pDb, pVgroup, &contLen);
+  if (pReq == NULL) return -1;
+
+  action.pCont = pReq;
+  action.contLen = contLen;
+  action.msgType = TDMT_DND_DROP_VNODE;
+  action.acceptableCode = TSDB_CODE_NODE_NOT_DEPLOYED;
+
+  if (isRedo) {
+    if (mndTransAppendRedoAction(pTrans, &action) != 0) {
+      taosMemoryFree(pReq);
+      return -1;
+    }
+  } else {
+    if (mndTransAppendUndoAction(pTrans, &action) != 0) {
+      taosMemoryFree(pReq);
+      return -1;
+    }
+  }
+
+  return 0;
+}
+
 static int32_t mndCheckDbName(const char *dbName, SUserObj *pUser) {
   char *pos = strstr(dbName, TS_PATH_DELIMITER);
   if (pos == NULL) {
@@ -278,6 +376,8 @@ static int32_t mndCheckDbName(const char *dbName, SUserObj *pUser) {
 }
 
 static int32_t mndCheckDbCfg(SMnode *pMnode, SDbCfg *pCfg) {
+  terrno = TSDB_CODE_MND_INVALID_DB_OPTION;
+
   if (pCfg->numOfVgroups < TSDB_MIN_VNODES_PER_DB || pCfg->numOfVgroups > TSDB_MAX_VNODES_PER_DB) return -1;
   if (pCfg->numOfStables < TSDB_DB_STREAM_MODE_OFF || pCfg->numOfStables > TSDB_DB_STREAM_MODE_ON) return -1;
   if (pCfg->buffer < TSDB_MIN_BUFFER_PER_VNODE || pCfg->buffer > TSDB_MAX_BUFFER_PER_VNODE) return -1;
@@ -299,10 +399,12 @@ static int32_t mndCheckDbCfg(SMnode *pMnode, SDbCfg *pCfg) {
   if (pCfg->compression < TSDB_MIN_COMP_LEVEL || pCfg->compression > TSDB_MAX_COMP_LEVEL) return -1;
   if (pCfg->replications < TSDB_MIN_DB_REPLICA || pCfg->replications > TSDB_MAX_DB_REPLICA) return -1;
   if (pCfg->replications > mndGetDnodeSize(pMnode)) return -1;
+  if (pCfg->replications != 1 && pCfg->replications != 3) return -1;
   if (pCfg->strict < TSDB_DB_STRICT_OFF || pCfg->strict > TSDB_DB_STRICT_ON) return -1;
-  if (pCfg->strict > pCfg->replications) return -1;
   if (pCfg->cacheLastRow < TSDB_MIN_DB_CACHE_LAST_ROW || pCfg->cacheLastRow > TSDB_MAX_DB_CACHE_LAST_ROW) return -1;
   if (pCfg->hashMethod != 1) return -1;
+
+  terrno = 0;
   return TSDB_CODE_SUCCESS;
 }
 
@@ -381,24 +483,8 @@ static int32_t mndSetCreateDbRedoActions(SMnode *pMnode, STrans *pTrans, SDbObj
     SVgObj *pVgroup = pVgroups + vg;
 
     for (int32_t vn = 0; vn < pVgroup->replica; ++vn) {
-      STransAction action = {0};
+      SVnodeGid *pVgid = pVgroup->vnodeGid + vn;
-      SVnodeGid   *pVgid = pVgroup->vnodeGid + vn;
+      if (mndAddCreateVnodeAction(pMnode, pTrans, pDb, pVgroup, pVgid, true) != 0) {
-
-      SDnodeObj *pDnode = mndAcquireDnode(pMnode, pVgid->dnodeId);
-      if (pDnode == NULL) return -1;
-      action.epSet = mndGetDnodeEpset(pDnode);
-      mndReleaseDnode(pMnode, pDnode);
-
-      int32_t contLen = 0;
-      void   *pReq = mndBuildCreateVnodeReq(pMnode, pDnode, pDb, pVgroup, &contLen);
-      if (pReq == NULL) return -1;
-
-      action.pCont = pReq;
-      action.contLen = contLen;
-      action.msgType = TDMT_DND_CREATE_VNODE;
-      action.acceptableCode = TSDB_CODE_NODE_ALREADY_DEPLOYED;
-      if (mndTransAppendRedoAction(pTrans, &action) != 0) {
-        taosMemoryFree(pReq);
         return -1;
       }
     }
@@ -412,24 +498,8 @@ static int32_t mndSetCreateDbUndoActions(SMnode *pMnode, STrans *pTrans, SDbObj
     SVgObj *pVgroup = pVgroups + vg;
 
     for (int32_t vn = 0; vn < pVgroup->replica; ++vn) {
-      STransAction action = {0};
+      SVnodeGid *pVgid = pVgroup->vnodeGid + vn;
-      SVnodeGid   *pVgid = pVgroup->vnodeGid + vn;
+      if (mndAddDropVnodeAction(pMnode, pTrans, pDb, pVgroup, pVgid, false) != 0) {
-
-      SDnodeObj *pDnode = mndAcquireDnode(pMnode, pVgid->dnodeId);
-      if (pDnode == NULL) return -1;
-      action.epSet = mndGetDnodeEpset(pDnode);
-      mndReleaseDnode(pMnode, pDnode);
-
-      int32_t contLen = 0;
-      void   *pReq = mndBuildDropVnodeReq(pMnode, pDnode, pDb, pVgroup, &contLen);
-      if (pReq == NULL) return -1;
-
-      action.pCont = pReq;
-      action.contLen = contLen;
-      action.msgType = TDMT_DND_DROP_VNODE;
-      action.acceptableCode = TSDB_CODE_NODE_NOT_DEPLOYED;
-      if (mndTransAppendUndoAction(pTrans, &action) != 0) {
-        taosMemoryFree(pReq);
         return -1;
       }
     }
@@ -482,7 +552,6 @@ static int32_t mndCreateDb(SMnode *pMnode, SNodeMsg *pReq, SCreateDbReq *pCreate
   }
 
   if (mndCheckDbCfg(pMnode, &dbObj.cfg) != 0) {
-    terrno = TSDB_CODE_MND_INVALID_DB_OPTION;
     mError("db:%s, failed to create since %s", pCreate->db, terrstr());
     return -1;
   }
@@ -570,37 +639,37 @@ _OVER:
 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 (pAlter->buffer > 0 && pAlter->buffer != pDb->cfg.buffer) {
     pDb->cfg.buffer = pAlter->buffer;
     terrno = 0;
   }
 
-  if (pAlter->pages >= 0 && pAlter->pages != pDb->cfg.pages) {
+  if (pAlter->pages > 0 && pAlter->pages != pDb->cfg.pages) {
     pDb->cfg.pages = pAlter->pages;
     terrno = 0;
   }
 
-  if (pAlter->pageSize >= 0 && pAlter->pageSize != pDb->cfg.pageSize) {
+  if (pAlter->pageSize > 0 && pAlter->pageSize != pDb->cfg.pageSize) {
     pDb->cfg.pageSize = pAlter->pageSize;
     terrno = 0;
   }
 
-  if (pAlter->daysPerFile >= 0 && pAlter->daysPerFile != pDb->cfg.daysPerFile) {
+  if (pAlter->daysPerFile > 0 && pAlter->daysPerFile != pDb->cfg.daysPerFile) {
     pDb->cfg.daysPerFile = pAlter->daysPerFile;
     terrno = 0;
   }
 
-  if (pAlter->daysToKeep0 >= 0 && pAlter->daysToKeep0 != pDb->cfg.daysToKeep0) {
+  if (pAlter->daysToKeep0 > 0 && pAlter->daysToKeep0 != pDb->cfg.daysToKeep0) {
     pDb->cfg.daysToKeep0 = pAlter->daysToKeep0;
     terrno = 0;
   }
 
-  if (pAlter->daysToKeep1 >= 0 && pAlter->daysToKeep1 != pDb->cfg.daysToKeep1) {
+  if (pAlter->daysToKeep1 > 0 && pAlter->daysToKeep1 != pDb->cfg.daysToKeep1) {
     pDb->cfg.daysToKeep1 = pAlter->daysToKeep1;
     terrno = 0;
   }
 
-  if (pAlter->daysToKeep2 >= 0 && pAlter->daysToKeep2 != pDb->cfg.daysToKeep2) {
+  if (pAlter->daysToKeep2 > 0 && pAlter->daysToKeep2 != pDb->cfg.daysToKeep2) {
     pDb->cfg.daysToKeep2 = pAlter->daysToKeep2;
     terrno = 0;
   }
@@ -625,8 +694,9 @@ static int32_t mndSetDbCfgFromAlterDbReq(SDbObj *pDb, SAlterDbReq *pAlter) {
     terrno = 0;
   }
 
-  if (pAlter->replications >= 0 && pAlter->replications != pDb->cfg.replications) {
+  if (pAlter->replications > 0 && pAlter->replications != pDb->cfg.replications) {
     pDb->cfg.replications = pAlter->replications;
+    pDb->vgVersion++;
     terrno = 0;
   }
 
@@ -651,35 +721,57 @@ static int32_t mndSetAlterDbCommitLogs(SMnode *pMnode, STrans *pTrans, SDbObj *p
   return 0;
 }
 
-static int32_t mndBuildAlterVgroupAction(SMnode *pMnode, STrans *pTrans, SDbObj *pDb, SVgObj *pVgroup) {
+static int32_t mndBuildAlterVgroupAction(SMnode *pMnode, STrans *pTrans, SDbObj *pDb, SVgObj *pVgroup, SArray *pArray) {
-  for (int32_t vn = 0; vn < pVgroup->replica; ++vn) {
+  if (pVgroup->replica <= 0 || pVgroup->replica == pDb->cfg.replications) {
-    STransAction action = {0};
+    for (int32_t vn = 0; vn < pVgroup->replica; ++vn) {
-    SVnodeGid   *pVgid = pVgroup->vnodeGid + vn;
+      SVnodeGid *pVgid = pVgroup->vnodeGid + vn;
-
+      if (mndAddAlterVnodeAction(pMnode, pTrans, pDb, pVgroup, pVgid, true) != 0) {
-    SDnodeObj *pDnode = mndAcquireDnode(pMnode, pVgid->dnodeId);
+        return -1;
-    if (pDnode == NULL) return -1;
+      }
-    action.epSet = mndGetDnodeEpset(pDnode);
-    mndReleaseDnode(pMnode, pDnode);
-
-    int32_t contLen = 0;
-    void   *pReq = mndBuildAlterVnodeReq(pMnode, pDnode, pDb, pVgroup, &contLen);
-    if (pReq == NULL) return -1;
-
-    action.pCont = pReq;
-    action.contLen = contLen;
-    action.msgType = TDMT_VND_ALTER_VNODE;
-    if (mndTransAppendRedoAction(pTrans, &action) != 0) {
-      taosMemoryFree(pReq);
-      return -1;
     }
+  } else {
+    SVgObj newVgroup = {0};
+    memcpy(&newVgroup, pVgroup, sizeof(SVgObj));
+    if (newVgroup.replica < pDb->cfg.replications) {
+      mInfo("db:%s, vgId:%d, will add 2 vnodes, vn:0 dnode:%d", pVgroup->dbName, pVgroup->vgId,
+            pVgroup->vnodeGid[0].dnodeId);
+
+      if (mndAddVnodeToVgroup(pMnode, &newVgroup, pArray) != 0) {
+        mError("db:%s, failed to add vnode to vgId:%d since %s", pDb->name, newVgroup.vgId, terrstr());
+        return -1;
+      }
+      newVgroup.replica = pDb->cfg.replications;
+      if (mndAddAlterVnodeAction(pMnode, pTrans, pDb, &newVgroup, &newVgroup.vnodeGid[0], true) != 0) return -1;
+      if (mndAddCreateVnodeAction(pMnode, pTrans, pDb, &newVgroup, &newVgroup.vnodeGid[1], true) != 0) return -1;
+      if (mndAddCreateVnodeAction(pMnode, pTrans, pDb, &newVgroup, &newVgroup.vnodeGid[2], true) != 0) return -1;
+    } else {
+      mInfo("db:%s, vgId:%d, will remove 2 vnodes", pVgroup->dbName, pVgroup->vgId);
+
+      SVnodeGid del1 = {0};
+      SVnodeGid del2 = {0};
+      if (mndRemoveVnodeFromVgroup(pMnode, &newVgroup, pArray, &del1, &del2) != 0) {
+        mError("db:%s, failed to remove vnode from vgId:%d since %s", pDb->name, newVgroup.vgId, terrstr());
+        return -1;
+      }
+      newVgroup.replica = pDb->cfg.replications;
+      if (mndAddAlterVnodeAction(pMnode, pTrans, pDb, &newVgroup, &newVgroup.vnodeGid[0], true) != 0) return -1;
+      if (mndAddDropVnodeAction(pMnode, pTrans, pDb, &newVgroup, &del1, true) != 0) return -1;
+      if (mndAddDropVnodeAction(pMnode, pTrans, pDb, &newVgroup, &del2, true) != 0) return -1;
+    }
+
+    SSdbRaw *pVgRaw = mndVgroupActionEncode(&newVgroup);
+    if (pVgRaw == NULL) return -1;
+    if (mndTransAppendCommitlog(pTrans, pVgRaw) != 0) return -1;
+    if (sdbSetRawStatus(pVgRaw, SDB_STATUS_READY) != 0) return -1;
   }
 
   return 0;
 }
 
 static int32_t mndSetAlterDbRedoActions(SMnode *pMnode, STrans *pTrans, SDbObj *pOld, SDbObj *pNew) {
-  SSdb *pSdb = pMnode->pSdb;
+  SSdb   *pSdb = pMnode->pSdb;
-  void *pIter = NULL;
+  void   *pIter = NULL;
+  SArray *pArray = mndBuildDnodesArray(pMnode);
 
   while (1) {
     SVgObj *pVgroup = NULL;
@@ -687,9 +779,10 @@ static int32_t mndSetAlterDbRedoActions(SMnode *pMnode, STrans *pTrans, SDbObj *
     if (pIter == NULL) break;
 
     if (pVgroup->dbUid == pNew->uid) {
-      if (mndBuildAlterVgroupAction(pMnode, pTrans, pNew, pVgroup) != 0) {
+      if (mndBuildAlterVgroupAction(pMnode, pTrans, pNew, pVgroup, pArray) != 0) {
         sdbCancelFetch(pSdb, pIter);
         sdbRelease(pSdb, pVgroup);
+        taosArrayDestroy(pArray);
         return -1;
       }
     }
@@ -697,6 +790,7 @@ static int32_t mndSetAlterDbRedoActions(SMnode *pMnode, STrans *pTrans, SDbObj *
     sdbRelease(pSdb, pVgroup);
   }
 
+  taosArrayDestroy(pArray);
   return 0;
 }
 
@@ -726,6 +820,7 @@ static int32_t mndProcessAlterDbReq(SNodeMsg *pReq) {
   SDbObj     *pDb = NULL;
   SUserObj   *pUser = NULL;
   SAlterDbReq alterReq = {0};
+  SDbObj      dbObj = {0};
 
   if (tDeserializeSAlterDbReq(pReq->rpcMsg.pCont, pReq->rpcMsg.contLen, &alterReq) != 0) {
     terrno = TSDB_CODE_INVALID_MSG;
@@ -749,15 +844,17 @@ static int32_t mndProcessAlterDbReq(SNodeMsg *pReq) {
     goto _OVER;
   }
 
-  SDbObj dbObj = {0};
   memcpy(&dbObj, pDb, sizeof(SDbObj));
-  dbObj.cfg.numOfRetensions = 0;
+  if (dbObj.cfg.pRetensions != NULL) {
-  dbObj.cfg.pRetensions = NULL;
+    dbObj.cfg.pRetensions = taosArrayDup(pDb->cfg.pRetensions);
+    if (dbObj.cfg.pRetensions == NULL) goto _OVER;
+  }
 
   code = mndSetDbCfgFromAlterDbReq(&dbObj, &alterReq);
-  if (code != 0) {
+  if (code != 0) goto _OVER;
-    goto _OVER;
+
-  }
+  code = mndCheckDbCfg(pMnode, &dbObj.cfg);
+  if (code != 0) goto _OVER;
 
   dbObj.cfgVersion++;
   dbObj.updateTime = taosGetTimestampMs();
@@ -771,6 +868,7 @@ _OVER:
 
   mndReleaseDb(pMnode, pDb);
   mndReleaseUser(pMnode, pUser);
+  taosArrayDestroy(dbObj.cfg.pRetensions);
 
   return code;
 }
@@ -899,24 +997,8 @@ static int32_t mndSetDropDbCommitLogs(SMnode *pMnode, STrans *pTrans, SDbObj *pD
 
 static int32_t mndBuildDropVgroupAction(SMnode *pMnode, STrans *pTrans, SDbObj *pDb, SVgObj *pVgroup) {
   for (int32_t vn = 0; vn < pVgroup->replica; ++vn) {
-    STransAction action = {0};
+    SVnodeGid *pVgid = pVgroup->vnodeGid + vn;
-    SVnodeGid   *pVgid = pVgroup->vnodeGid + vn;
+    if (mndAddDropVnodeAction(pMnode, pTrans, pDb, pVgroup, pVgid, true) != 0) {
-
-    SDnodeObj *pDnode = mndAcquireDnode(pMnode, pVgid->dnodeId);
-    if (pDnode == NULL) return -1;
-    action.epSet = mndGetDnodeEpset(pDnode);
-    mndReleaseDnode(pMnode, pDnode);
-
-    int32_t contLen = 0;
-    void   *pReq = mndBuildDropVnodeReq(pMnode, pDnode, pDb, pVgroup, &contLen);
-    if (pReq == NULL) return -1;
-
-    action.pCont = pReq;
-    action.contLen = contLen;
-    action.msgType = TDMT_DND_DROP_VNODE;
-    action.acceptableCode = TSDB_CODE_NODE_NOT_DEPLOYED;
-    if (mndTransAppendRedoAction(pTrans, &action) != 0) {
-      taosMemoryFree(pReq);
       return -1;
     }
   }
@@ -988,6 +1070,17 @@ static int32_t mndDropDb(SMnode *pMnode, SNodeMsg *pReq, SDbObj *pDb) {
   /*if (mndDropTopicByDB(pMnode, pTrans, pDb) != 0) goto _OVER;*/
   if (mndSetDropDbRedoActions(pMnode, pTrans, pDb) != 0) goto _OVER;
 
+  SUserObj *pUser = mndAcquireUser(pMnode, pDb->createUser);
+  if (pUser != NULL) {
+    pUser->authVersion++;
+    SSdbRaw *pCommitRaw = mndUserActionEncode(pUser);
+    if (pCommitRaw == NULL || mndTransAppendCommitlog(pTrans, pCommitRaw) != 0) {
+      mError("trans:%d, failed to append redo log since %s", pTrans->id, terrstr());
+      goto _OVER;
+    }
+    sdbSetRawStatus(pCommitRaw, SDB_STATUS_READY);
+  }
+
   int32_t rspLen = 0;
   void   *pRsp = NULL;
   if (mndBuildDropDbRsp(pDb, &rspLen, &pRsp, false) < 0) goto _OVER;

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

@@ -21,7 +21,7 @@
 #include "mndInfoSchema.h"
 #include "mndMnode.h"
 #include "mndShow.h"
-#include "mndStb.c"
+#include "mndStb.h"
 #include "mndStream.h"
 #include "mndTrans.h"
 #include "mndUser.h"
@@ -419,7 +419,6 @@ static int32_t mndCreateSma(SMnode *pMnode, SNodeMsg *pReq, SMCreateSmaReq *pCre
   streamObj.fixedSinkVgId = smaObj.dstVgId;
   streamObj.smaId = smaObj.uid;
   /*streamObj.physicalPlan = "";*/
-  streamObj.logicalPlan = "not implemented";
 
   int32_t code = -1;
   STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_RETRY, TRN_TYPE_CREATE_SMA, &pReq->rpcMsg);

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

@@ -13,6 +13,7 @@
  * along with this program. If not, see <http://www.gnu.org/licenses/>.
  */
 
+#define _DEFAULT_SOURCE
 #include "mndStb.h"
 #include "mndAuth.h"
 #include "mndDb.h"
@@ -342,7 +343,7 @@ void mndReleaseStb(SMnode *pMnode, SStbObj *pStb) {
   sdbRelease(pSdb, pStb);
 }
 
-static SDbObj *mndAcquireDbByStb(SMnode *pMnode, const char *stbName) {
+SDbObj *mndAcquireDbByStb(SMnode *pMnode, const char *stbName) {
   SName name = {0};
   tNameFromString(&name, stbName, T_NAME_ACCT | T_NAME_DB | T_NAME_TABLE);
 
@@ -463,7 +464,7 @@ static void *mndBuildVDropStbReq(SMnode *pMnode, SVgObj *pVgroup, SStbObj *pStb,
   return pHead;
 }
 
-static int32_t mndCheckCreateStbReq(SMCreateStbReq *pCreate) {
+int32_t mndCheckCreateStbReq(SMCreateStbReq *pCreate) {
   if (pCreate->igExists < 0 || pCreate->igExists > 1) {
     terrno = TSDB_CODE_MND_INVALID_STB_OPTION;
     return -1;
@@ -634,91 +635,96 @@ static SSchema *mndFindStbColumns(const SStbObj *pStb, const char *colName) {
   return NULL;
 }
 
-static int32_t mndCreateStb(SMnode *pMnode, SNodeMsg *pReq, SMCreateStbReq *pCreate, SDbObj *pDb) {
+int32_t mndBuildStbFromReq(SMnode *pMnode, SStbObj *pDst, SMCreateStbReq *pCreate, SDbObj *pDb) {
-  SStbObj stbObj = {0};
+  memcpy(pDst->name, pCreate->name, TSDB_TABLE_FNAME_LEN);
-  memcpy(stbObj.name, pCreate->name, TSDB_TABLE_FNAME_LEN);
+  memcpy(pDst->db, pDb->name, TSDB_DB_FNAME_LEN);
-  memcpy(stbObj.db, pDb->name, TSDB_DB_FNAME_LEN);
+  pDst->createdTime = taosGetTimestampMs();
-  stbObj.createdTime = taosGetTimestampMs();
+  pDst->updateTime = pDst->createdTime;
-  stbObj.updateTime = stbObj.createdTime;
+  pDst->uid = mndGenerateUid(pCreate->name, TSDB_TABLE_FNAME_LEN);
-  stbObj.uid = mndGenerateUid(pCreate->name, TSDB_TABLE_FNAME_LEN);
+  pDst->dbUid = pDb->uid;
-  stbObj.dbUid = pDb->uid;
+  pDst->version = 1;
-  stbObj.version = 1;
+  pDst->nextColId = 1;
-  stbObj.nextColId = 1;
+  pDst->xFilesFactor = pCreate->xFilesFactor;
-  stbObj.xFilesFactor = pCreate->xFilesFactor;
+  pDst->delay = pCreate->delay;
-  stbObj.delay = pCreate->delay;
+  pDst->ttl = pCreate->ttl;
-  stbObj.ttl = pCreate->ttl;
+  pDst->numOfColumns = pCreate->numOfColumns;
-  stbObj.numOfColumns = pCreate->numOfColumns;
+  pDst->numOfTags = pCreate->numOfTags;
-  stbObj.numOfTags = pCreate->numOfTags;
+  pDst->commentLen = pCreate->commentLen;
-  stbObj.commentLen = pCreate->commentLen;
+  if (pDst->commentLen > 0) {
-  if (stbObj.commentLen > 0) {
+    pDst->comment = taosMemoryCalloc(pDst->commentLen, 1);
-    stbObj.comment = taosMemoryCalloc(stbObj.commentLen, 1);
+    if (pDst->comment == NULL) {
-    if (stbObj.comment == NULL) {
       terrno = TSDB_CODE_OUT_OF_MEMORY;
       return -1;
     }
-    memcpy(stbObj.comment, pCreate->comment, stbObj.commentLen);
+    memcpy(pDst->comment, pCreate->comment, pDst->commentLen);
   }
 
-  stbObj.ast1Len = pCreate->ast1Len;
+  pDst->ast1Len = pCreate->ast1Len;
-  if (stbObj.ast1Len > 0) {
+  if (pDst->ast1Len > 0) {
-    stbObj.pAst1 = taosMemoryCalloc(stbObj.ast1Len, 1);
+    pDst->pAst1 = taosMemoryCalloc(pDst->ast1Len, 1);
-    if (stbObj.pAst1 == NULL) {
+    if (pDst->pAst1 == NULL) {
       terrno = TSDB_CODE_OUT_OF_MEMORY;
       return -1;
     }
-    memcpy(stbObj.pAst1, pCreate->pAst1, stbObj.ast1Len);
+    memcpy(pDst->pAst1, pCreate->pAst1, pDst->ast1Len);
   }
 
-  stbObj.ast2Len = pCreate->ast2Len;
+  pDst->ast2Len = pCreate->ast2Len;
-  if (stbObj.ast2Len > 0) {
+  if (pDst->ast2Len > 0) {
-    stbObj.pAst2 = taosMemoryCalloc(stbObj.ast2Len, 1);
+    pDst->pAst2 = taosMemoryCalloc(pDst->ast2Len, 1);
-    if (stbObj.pAst2 == NULL) {
+    if (pDst->pAst2 == NULL) {
       terrno = TSDB_CODE_OUT_OF_MEMORY;
       return -1;
     }
-    memcpy(stbObj.pAst2, pCreate->pAst2, stbObj.ast2Len);
+    memcpy(pDst->pAst2, pCreate->pAst2, pDst->ast2Len);
   }
 
-  stbObj.pColumns = taosMemoryCalloc(1, stbObj.numOfColumns * sizeof(SSchema));
+  pDst->pColumns = taosMemoryCalloc(1, pDst->numOfColumns * sizeof(SSchema));
-  stbObj.pTags = taosMemoryCalloc(1, stbObj.numOfTags * sizeof(SSchema));
+  pDst->pTags = taosMemoryCalloc(1, pDst->numOfTags * sizeof(SSchema));
-  if (stbObj.pColumns == NULL || stbObj.pTags == NULL) {
+  if (pDst->pColumns == NULL || pDst->pTags == NULL) {
     terrno = TSDB_CODE_OUT_OF_MEMORY;
     return -1;
   }
 
-  for (int32_t i = 0; i < stbObj.numOfColumns; ++i) {
+  for (int32_t i = 0; i < pDst->numOfColumns; ++i) {
     SField  *pField = taosArrayGet(pCreate->pColumns, i);
-    SSchema *pSchema = &stbObj.pColumns[i];
+    SSchema *pSchema = &pDst->pColumns[i];
     pSchema->type = pField->type;
     pSchema->bytes = pField->bytes;
     pSchema->flags = pField->flags;
     memcpy(pSchema->name, pField->name, TSDB_COL_NAME_LEN);
-    pSchema->colId = stbObj.nextColId;
+    pSchema->colId = pDst->nextColId;
-    stbObj.nextColId++;
+    pDst->nextColId++;
   }
 
-  for (int32_t i = 0; i < stbObj.numOfTags; ++i) {
+  for (int32_t i = 0; i < pDst->numOfTags; ++i) {
     SField  *pField = taosArrayGet(pCreate->pTags, i);
-    SSchema *pSchema = &stbObj.pTags[i];
+    SSchema *pSchema = &pDst->pTags[i];
     pSchema->type = pField->type;
     pSchema->bytes = pField->bytes;
     memcpy(pSchema->name, pField->name, TSDB_COL_NAME_LEN);
-    pSchema->colId = stbObj.nextColId;
+    pSchema->colId = pDst->nextColId;
-    stbObj.nextColId++;
+    pDst->nextColId++;
   }
+  return 0;
+}
+
+static int32_t mndCreateStb(SMnode *pMnode, SNodeMsg *pReq, SMCreateStbReq *pCreate, SDbObj *pDb) {
+  SStbObj stbObj = {0};
 
   int32_t code = -1;
+
   STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_ROLLBACK, TRN_TYPE_CREATE_STB, &pReq->rpcMsg);
   if (pTrans == NULL) goto _OVER;
 
   mDebug("trans:%d, used to create stb:%s", pTrans->id, pCreate->name);
-  mndTransSetDbInfo(pTrans, pDb);
 
-  if (mndSetCreateStbRedoLogs(pMnode, pTrans, pDb, &stbObj) != 0) goto _OVER;
+  if (mndBuildStbFromReq(pMnode, &stbObj, pCreate, pDb) != 0) {
-  if (mndSetCreateStbUndoLogs(pMnode, pTrans, pDb, &stbObj) != 0) goto _OVER;
+    goto _OVER;
-  if (mndSetCreateStbCommitLogs(pMnode, pTrans, pDb, &stbObj) != 0) goto _OVER;
+  }
-  if (mndSetCreateStbRedoActions(pMnode, pTrans, pDb, &stbObj) != 0) goto _OVER;
+
-  if (mndSetCreateStbUndoActions(pMnode, pTrans, pDb, &stbObj) != 0) goto _OVER;
+  if (mndAddStbToTrans(pMnode, pTrans, pDb, &stbObj) < 0) goto _OVER;
+
   if (mndTransPrepare(pMnode, pTrans) != 0) goto _OVER;
 
   code = 0;
@@ -727,6 +733,15 @@ _OVER:
   mndTransDrop(pTrans);
   return code;
 }
+int32_t mndAddStbToTrans(SMnode *pMnode, STrans *pTrans, SDbObj *pDb, SStbObj *pStb) {
+  mndTransSetDbInfo(pTrans, pDb);
+  if (mndSetCreateStbRedoLogs(pMnode, pTrans, pDb, pStb) != 0) return -1;
+  if (mndSetCreateStbUndoLogs(pMnode, pTrans, pDb, pStb) != 0) return -1;
+  if (mndSetCreateStbCommitLogs(pMnode, pTrans, pDb, pStb) != 0) return -1;
+  if (mndSetCreateStbRedoActions(pMnode, pTrans, pDb, pStb) != 0) return -1;
+  if (mndSetCreateStbUndoActions(pMnode, pTrans, pDb, pStb) != 0) return -1;
+  return 0;
+}
 
 static int32_t mndProcessMCreateStbReq(SNodeMsg *pReq) {
   SMnode        *pMnode = pReq->pNode;

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

@@ -290,6 +290,83 @@ int32_t mndAddStreamToTrans(SMnode *pMnode, SStreamObj *pStream, const char *ast
   return 0;
 }
 
+static int32_t mndCreateStbForStream(SMnode *pMnode, STrans *pTrans, const SStreamObj *pStream, const char *user) {
+  SStbObj  *pStb = NULL;
+  SDbObj   *pDb = NULL;
+  SUserObj *pUser = NULL;
+
+  SMCreateStbReq createReq = {0};
+  tstrncpy(createReq.name, pStream->targetSTbName, TSDB_TABLE_FNAME_LEN);
+  createReq.numOfColumns = pStream->outputSchema.nCols;
+  createReq.numOfTags = 1;  // group id
+  createReq.pColumns = taosArrayInit(createReq.numOfColumns, sizeof(SField));
+  // build fields
+  taosArraySetSize(createReq.pColumns, createReq.numOfColumns);
+  for (int32_t i = 0; i < createReq.numOfColumns; i++) {
+    SField *pField = taosArrayGet(createReq.pColumns, i);
+    tstrncpy(pField->name, pStream->outputSchema.pSchema[i].name, TSDB_COL_NAME_LEN);
+    pField->flags = pStream->outputSchema.pSchema[i].flags;
+    pField->type = pStream->outputSchema.pSchema[i].type;
+    pField->bytes = pStream->outputSchema.pSchema[i].bytes;
+  }
+  createReq.pTags = taosArrayInit(createReq.numOfTags, sizeof(SField));
+  taosArraySetSize(createReq.pTags, 1);
+  // build tags
+  SField *pField = taosArrayGet(createReq.pTags, 0);
+  strcpy(pField->name, "group_id");
+  pField->type = TSDB_DATA_TYPE_UBIGINT;
+  pField->flags = 0;
+  pField->bytes = 8;
+
+  if (mndCheckCreateStbReq(&createReq) != 0) {
+    terrno = TSDB_CODE_INVALID_MSG;
+    goto _OVER;
+  }
+
+  pStb = mndAcquireStb(pMnode, createReq.name);
+  if (pStb != NULL) {
+    terrno = TSDB_CODE_MND_STB_ALREADY_EXIST;
+    goto _OVER;
+  }
+
+  pDb = mndAcquireDbByStb(pMnode, createReq.name);
+  if (pDb == NULL) {
+    terrno = TSDB_CODE_MND_DB_NOT_SELECTED;
+    goto _OVER;
+  }
+
+  pUser = mndAcquireUser(pMnode, user);
+  if (pUser == NULL) {
+    goto _OVER;
+  }
+
+  if (mndCheckWriteAuth(pUser, pDb) != 0) {
+    goto _OVER;
+  }
+
+  int32_t numOfStbs = -1;
+  mndGetNumOfStbs(pMnode, pDb->name, &numOfStbs);
+  if (pDb->cfg.numOfStables == 1 && numOfStbs != 0) {
+    terrno = TSDB_CODE_MND_SINGLE_STB_MODE_DB;
+    goto _OVER;
+  }
+
+  SStbObj stbObj = {0};
+
+  if (mndBuildStbFromReq(pMnode, &stbObj, &createReq, pDb) != 0) {
+    goto _OVER;
+  }
+
+  if (mndAddStbToTrans(pMnode, pTrans, pDb, &stbObj) < 0) goto _OVER;
+
+  return 0;
+_OVER:
+  mndReleaseStb(pMnode, pStb);
+  mndReleaseDb(pMnode, pDb);
+  mndReleaseUser(pMnode, pUser);
+  return -1;
+}
+
 static int32_t mndCreateStream(SMnode *pMnode, SNodeMsg *pReq, SCMCreateStreamReq *pCreate, SDbObj *pDb) {
   mDebug("stream:%s to create", pCreate->name);
   SStreamObj streamObj = {0};
@@ -307,11 +384,10 @@ static int32_t mndCreateStream(SMnode *pMnode, SNodeMsg *pReq, SCMCreateStreamRe
   streamObj.fixedSinkVgId = 0;
   streamObj.smaId = 0;
   /*streamObj.physicalPlan = "";*/
-  streamObj.logicalPlan = "not implemented";
   streamObj.trigger = pCreate->triggerType;
   streamObj.waterMark = pCreate->watermark;
 
-  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_RETRY, TRN_TYPE_CREATE_STREAM, &pReq->rpcMsg);
+  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_ROLLBACK, TRN_TYPE_CREATE_STREAM, &pReq->rpcMsg);
   if (pTrans == NULL) {
     mError("stream:%s, failed to create since %s", pCreate->name, terrstr());
     return -1;
@@ -324,6 +400,12 @@ static int32_t mndCreateStream(SMnode *pMnode, SNodeMsg *pReq, SCMCreateStreamRe
     return -1;
   }
 
+  if (streamObj.targetSTbName[0] && mndCreateStbForStream(pMnode, pTrans, &streamObj, pReq->user) < 0) {
+    mError("trans:%d, failed to create stb for stream since %s", pTrans->id, terrstr());
+    mndTransDrop(pTrans);
+    return -1;
+  }
+
   if (mndTransPrepare(pMnode, pTrans) != 0) {
     mError("trans:%d, failed to prepare since %s", pTrans->id, terrstr());
     mndTransDrop(pTrans);

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

@@ -1336,8 +1336,7 @@ static int32_t mndProcessKillTransReq(SNodeMsg *pReq) {
     goto _OVER;
   }
 
-  if (!pUser->superUser) {
+  if (mndCheckTransAuth(pUser) != 0) {
-    terrno = TSDB_CODE_MND_NO_RIGHTS;
     goto _OVER;
   }
 

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

@@ -106,6 +106,7 @@ SSdbRaw *mndUserActionEncode(SUserObj *pUser) {
   SDB_SET_INT64(pRaw, dataPos, pUser->createdTime, _OVER)
   SDB_SET_INT64(pRaw, dataPos, pUser->updateTime, _OVER)
   SDB_SET_INT8(pRaw, dataPos, pUser->superUser, _OVER)
+  SDB_SET_INT32(pRaw, dataPos, pUser->authVersion, _OVER)
   SDB_SET_INT32(pRaw, dataPos, numOfReadDbs, _OVER)
   SDB_SET_INT32(pRaw, dataPos, numOfWriteDbs, _OVER)
 
@@ -161,6 +162,7 @@ static SSdbRow *mndUserActionDecode(SSdbRaw *pRaw) {
   SDB_GET_INT64(pRaw, dataPos, &pUser->createdTime, _OVER)
   SDB_GET_INT64(pRaw, dataPos, &pUser->updateTime, _OVER)
   SDB_GET_INT8(pRaw, dataPos, &pUser->superUser, _OVER)
+  SDB_GET_INT32(pRaw, dataPos, &pUser->authVersion, _OVER)
 
   int32_t numOfReadDbs = 0;
   int32_t numOfWriteDbs = 0;
@@ -588,7 +590,7 @@ _OVER:
   return code;
 }
 
-static int32_t mndSetUserAuthRsp(SMnode       *pMnode, SUserObj *pUser, SGetUserAuthRsp *pRsp) {
+static int32_t mndSetUserAuthRsp(SMnode *pMnode, SUserObj *pUser, SGetUserAuthRsp *pRsp) {
   memcpy(pRsp->user, pUser->user, TSDB_USER_LEN);
   pRsp->superAuth = pUser->superUser;
   pRsp->version = pUser->authVersion;
@@ -601,7 +603,7 @@ static int32_t mndSetUserAuthRsp(SMnode       *pMnode, SUserObj *pUser, SGetUser
     terrno = TSDB_CODE_OUT_OF_MEMORY;
     return -1;
   }
-  
+
   SSdb *pSdb = pMnode->pSdb;
   void *pIter = NULL;
   while (1) {
@@ -659,7 +661,7 @@ static int32_t mndProcessGetUserAuthReq(SNodeMsg *pReq) {
   code = 0;
 
 _OVER:
-  
+
   mndReleaseUser(pMnode, pUser);
   tFreeSGetUserAuthRsp(&authRsp);
 
@@ -711,7 +713,8 @@ static void mndCancelGetNextUser(SMnode *pMnode, void *pIter) {
   sdbCancelFetch(pSdb, pIter);
 }
 
-int32_t mndValidateUserAuthInfo(SMnode *pMnode, SUserAuthVersion *pUsers, int32_t numOfUses, void **ppRsp, int32_t *pRspLen) {
+int32_t mndValidateUserAuthInfo(SMnode *pMnode, SUserAuthVersion *pUsers, int32_t numOfUses, void **ppRsp,
+                                int32_t *pRspLen) {
   SUserAuthBatchRsp batchRsp = {0};
   batchRsp.pArray = taosArrayInit(numOfUses, sizeof(SGetUserAuthRsp));
   if (batchRsp.pArray == NULL) {
@@ -731,7 +734,7 @@ int32_t mndValidateUserAuthInfo(SMnode *pMnode, SUserAuthVersion *pUsers, int32_
       mndReleaseUser(pMnode, pUser);
       continue;
     }
-    
+
     SGetUserAuthRsp rsp = {0};
     code = mndSetUserAuthRsp(pMnode, pUser, &rsp);
     if (code) {
@@ -740,7 +743,6 @@ int32_t mndValidateUserAuthInfo(SMnode *pMnode, SUserAuthVersion *pUsers, int32_
       goto _OVER;
     }
 
-
     taosArrayPush(batchRsp.pArray, &rsp);
     mndReleaseUser(pMnode, pUser);
   }
@@ -748,7 +750,7 @@ int32_t mndValidateUserAuthInfo(SMnode *pMnode, SUserAuthVersion *pUsers, int32_
   if (taosArrayGetSize(batchRsp.pArray) <= 0) {
     *ppRsp = NULL;
     *pRspLen = 0;
-    
+
     tFreeSUserAuthBatchRsp(&batchRsp);
     return 0;
   }
@@ -772,10 +774,7 @@ _OVER:
 
   *ppRsp = NULL;
   *pRspLen = 0;
-  
+
   tFreeSUserAuthBatchRsp(&batchRsp);
   return code;
 }
-
-
-

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

@@ -370,7 +370,7 @@ static bool mndBuildDnodesArrayFp(SMnode *pMnode, void *pObj, void *p1, void *p2
   return true;
 }
 
-static SArray *mndBuildDnodesArray(SMnode *pMnode) {
+SArray *mndBuildDnodesArray(SMnode *pMnode) {
   SSdb   *pSdb = pMnode->pSdb;
   int32_t numOfDnodes = mndGetDnodeSize(pMnode);
 
@@ -421,7 +421,7 @@ static int32_t mndGetAvailableDnode(SMnode *pMnode, SVgObj *pVgroup, SArray *pAr
       pVgid->role = TAOS_SYNC_STATE_FOLLOWER;
     }
 
-    mDebug("db:%s, vgId:%d, vn:%d dnode:%d is alloced", pVgroup->dbName, pVgroup->vgId, v, pVgid->dnodeId);
+    mInfo("db:%s, vgId:%d, vn:%d dnode:%d is alloced", pVgroup->dbName, pVgroup->vgId, v, pVgid->dnodeId);
     pDnode->numOfVnodes++;
   }
 
@@ -440,12 +440,10 @@ int32_t mndAllocVgroup(SMnode *pMnode, SDbObj *pDb, SVgObj **ppVgroups) {
   }
 
   pArray = mndBuildDnodesArray(pMnode);
-  if (pArray == NULL) {
+  if (pArray == NULL) goto _OVER;
-    goto _OVER;
-  }
 
-  mDebug("db:%s, total %d dnodes used to create %d vgroups (%d vnodes)", pDb->name, (int32_t)taosArrayGetSize(pArray),
+  mInfo("db:%s, total %d dnodes used to create %d vgroups (%d vnodes)", pDb->name, (int32_t)taosArrayGetSize(pArray),
-         pDb->cfg.numOfVgroups, pDb->cfg.numOfVgroups * pDb->cfg.replications);
+        pDb->cfg.numOfVgroups, pDb->cfg.numOfVgroups * pDb->cfg.replications);
 
   int32_t  allocedVgroups = 0;
   int32_t  maxVgId = sdbGetMaxId(pMnode->pSdb, SDB_VGROUP);
@@ -483,7 +481,7 @@ int32_t mndAllocVgroup(SMnode *pMnode, SDbObj *pDb, SVgObj **ppVgroups) {
   *ppVgroups = pVgroups;
   code = 0;
 
-  mDebug("db:%s, %d vgroups is alloced, replica:%d", pDb->name, pDb->cfg.numOfVgroups, pDb->cfg.replications);
+  mInfo("db:%s, %d vgroups is alloced, replica:%d", pDb->name, pDb->cfg.numOfVgroups, pDb->cfg.replications);
 
 _OVER:
   if (code != 0) taosMemoryFree(pVgroups);
@@ -491,6 +489,88 @@ _OVER:
   return code;
 }
 
+int32_t mndAddVnodeToVgroup(SMnode *pMnode, SVgObj *pVgroup, SArray *pArray) {
+  taosArraySort(pArray, (__compar_fn_t)mndCompareDnodeVnodes);
+  for (int32_t i = 0; i < taosArrayGetSize(pArray); ++i) {
+    SDnodeObj *pDnode = taosArrayGet(pArray, i);
+    mDebug("dnode:%d, equivalent vnodes:%d", pDnode->id, pDnode->numOfVnodes);
+  }
+
+  int32_t maxPos = 1;
+  for (int32_t d = 0; d < taosArrayGetSize(pArray); ++d) {
+    SDnodeObj *pDnode = taosArrayGet(pArray, d);
+
+    bool used = false;
+    for (int32_t vn = 0; vn < maxPos; ++vn) {
+      if (pDnode->id == pVgroup->vnodeGid[vn].dnodeId) {
+        used = true;
+        break;
+      }
+    }
+    if (used) continue;
+
+    if (pDnode == NULL || pDnode->numOfVnodes > pDnode->numOfSupportVnodes) {
+      terrno = TSDB_CODE_MND_NO_ENOUGH_DNODES;
+      return -1;
+    }
+
+    SVnodeGid *pVgid = &pVgroup->vnodeGid[maxPos];
+    pVgid->dnodeId = pDnode->id;
+    pVgid->role = TAOS_SYNC_STATE_FOLLOWER;
+    pDnode->numOfVnodes++;
+
+    mInfo("db:%s, vgId:%d, vn:%d dnode:%d is added", pVgroup->dbName, pVgroup->vgId, maxPos, pVgid->dnodeId);
+    maxPos++;
+    if (maxPos == 3) return 0;
+  }
+
+  terrno = TSDB_CODE_MND_NO_ENOUGH_DNODES;
+  return -1;
+}
+
+int32_t mndRemoveVnodeFromVgroup(SMnode *pMnode, SVgObj *pVgroup, SArray *pArray, SVnodeGid *del1, SVnodeGid *del2) {
+  int32_t removedNum = 0;
+
+  taosArraySort(pArray, (__compar_fn_t)mndCompareDnodeVnodes);
+  for (int32_t i = 0; i < taosArrayGetSize(pArray); ++i) {
+    SDnodeObj *pDnode = taosArrayGet(pArray, i);
+    mDebug("dnode:%d, equivalent vnodes:%d", pDnode->id, pDnode->numOfVnodes);
+  }
+
+  for (int32_t d = taosArrayGetSize(pArray) - 1; d >= 0; --d) {
+    SDnodeObj *pDnode = taosArrayGet(pArray, d);
+
+    for (int32_t vn = 0; vn < TSDB_MAX_REPLICA; ++vn) {
+      SVnodeGid *pVgid = &pVgroup->vnodeGid[vn];
+      if (pVgid->dnodeId == pDnode->id) {
+        if (removedNum == 0) *del1 = *pVgid;
+        if (removedNum == 1) *del2 = *pVgid;
+
+        mInfo("db:%s, vgId:%d, vn:%d dnode:%d is removed", pVgroup->dbName, pVgroup->vgId, vn, pVgid->dnodeId);
+        memset(pVgid, 0, sizeof(SVnodeGid));
+        removedNum++;
+        pDnode->numOfVnodes--;
+
+        if (removedNum == 2) goto _OVER;
+      }
+    }
+  }
+
+_OVER:
+  if (removedNum != 2) return -1;
+
+  for (int32_t vn = 1; vn < TSDB_MAX_REPLICA; ++vn) {
+    SVnodeGid *pVgid = &pVgroup->vnodeGid[vn];
+    if (pVgid->dnodeId != 0) {
+      memcpy(&pVgroup->vnodeGid[0], pVgid, sizeof(SVnodeGid));
+      memset(pVgid, 0, sizeof(SVnodeGid));
+    }
+  }
+
+  mInfo("db:%s, vgId:%d, dnode:%d is keeped", pVgroup->dbName, pVgroup->vgId, pVgroup->vnodeGid[0].dnodeId);
+  return 0;
+}
+
 SEpSet mndGetVgroupEpset(SMnode *pMnode, const SVgObj *pVgroup) {
   SEpSet epset = {0};
 

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

@@ -73,14 +73,19 @@ TEST_F(MndTestDb, 02_Create_Alter_Drop_Db) {
   {
     SAlterDbReq alterdbReq = {0};
     strcpy(alterdbReq.db, "1.d1");
+
     alterdbReq.buffer = 12;
-    alterdbReq.daysToKeep0 = 300;
+    alterdbReq.pageSize = -1;
-    alterdbReq.daysToKeep1 = 400;
+    alterdbReq.pages = -1;
-    alterdbReq.daysToKeep2 = 500;
+    alterdbReq.daysPerFile = -1;
+    alterdbReq.daysToKeep0 = -1;
+    alterdbReq.daysToKeep1 = -1;
+    alterdbReq.daysToKeep2 = -1;
     alterdbReq.fsyncPeriod = 4000;
     alterdbReq.walLevel = 2;
-    alterdbReq.strict = 2;
+    alterdbReq.strict = 1;
     alterdbReq.cacheLastRow = 1;
+    alterdbReq.replications = 1;
 
     int32_t contLen = tSerializeSAlterDbReq(NULL, 0, &alterdbReq);
     void*   pReq = rpcMallocCont(contLen);

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

@@ -266,4 +266,96 @@ TEST_F(MndTestDnode, 05_Create_Drop_Restart_Dnode) {
   taosMsleep(1300);
   test.SendShowReq(TSDB_MGMT_TABLE_DNODE, "dnodes", "");
   EXPECT_EQ(test.GetShowRows(), 4);
+
+  // alter replica
+#if 0
+  {
+    SCreateDbReq createReq = {0};
+    strcpy(createReq.db, "1.d2");
+    createReq.numOfVgroups = 2;
+    createReq.buffer = -1;
+    createReq.pageSize = -1;
+    createReq.pages = -1;
+    createReq.daysPerFile = 1000;
+    createReq.daysToKeep0 = 3650;
+    createReq.daysToKeep1 = 3650;
+    createReq.daysToKeep2 = 3650;
+    createReq.minRows = 100;
+    createReq.maxRows = 4096;
+    createReq.fsyncPeriod = 3000;
+    createReq.walLevel = 1;
+    createReq.precision = 0;
+    createReq.compression = 2;
+    createReq.replications = 1;
+    createReq.strict = 1;
+    createReq.cacheLastRow = 0;
+    createReq.ignoreExist = 1;
+    createReq.numOfStables = 0;
+    createReq.numOfRetensions = 0;
+
+    int32_t contLen = tSerializeSCreateDbReq(NULL, 0, &createReq);
+    void*   pReq = rpcMallocCont(contLen);
+    tSerializeSCreateDbReq(pReq, contLen, &createReq);
+
+    SRpcMsg* pRsp = test.SendReq(TDMT_MND_CREATE_DB, pReq, contLen);
+    ASSERT_NE(pRsp, nullptr);
+    ASSERT_EQ(pRsp->code, 0);
+
+    test.SendShowReq(TSDB_MGMT_TABLE_DB, "user_databases", "");
+    EXPECT_EQ(test.GetShowRows(), 3);
+  }
+
+  {
+    SAlterDbReq alterdbReq = {0};
+    strcpy(alterdbReq.db, "1.d2");
+
+    alterdbReq.buffer = 12;
+    alterdbReq.pageSize = -1;
+    alterdbReq.pages = -1;
+    alterdbReq.daysPerFile = -1;
+    alterdbReq.daysToKeep0 = -1;
+    alterdbReq.daysToKeep1 = -1;
+    alterdbReq.daysToKeep2 = -1;
+    alterdbReq.fsyncPeriod = 4000;
+    alterdbReq.walLevel = 2;
+    alterdbReq.strict = 1;
+    alterdbReq.cacheLastRow = 1;
+    alterdbReq.replications = 3;
+
+    int32_t contLen = tSerializeSAlterDbReq(NULL, 0, &alterdbReq);
+    void*   pReq = rpcMallocCont(contLen);
+    tSerializeSAlterDbReq(pReq, contLen, &alterdbReq);
+
+    SRpcMsg* pRsp = test.SendReq(TDMT_MND_ALTER_DB, pReq, contLen);
+    ASSERT_NE(pRsp, nullptr);
+    ASSERT_EQ(pRsp->code, 0);
+  }
+
+  {
+    SAlterDbReq alterdbReq = {0};
+    strcpy(alterdbReq.db, "1.d2");
+
+    alterdbReq.buffer = 12;
+    alterdbReq.pageSize = -1;
+    alterdbReq.pages = -1;
+    alterdbReq.daysPerFile = -1;
+    alterdbReq.daysToKeep0 = -1;
+    alterdbReq.daysToKeep1 = -1;
+    alterdbReq.daysToKeep2 = -1;
+    alterdbReq.fsyncPeriod = 4000;
+    alterdbReq.walLevel = 2;
+    alterdbReq.strict = 1;
+    alterdbReq.cacheLastRow = 1;
+    alterdbReq.replications = 1;
+
+    int32_t contLen = tSerializeSAlterDbReq(NULL, 0, &alterdbReq);
+    void*   pReq = rpcMallocCont(contLen);
+    tSerializeSAlterDbReq(pReq, contLen, &alterdbReq);
+
+    SRpcMsg* pRsp = test.SendReq(TDMT_MND_ALTER_DB, pReq, contLen);
+    ASSERT_NE(pRsp, nullptr);
+    ASSERT_EQ(pRsp->code, 0);
+  }
+
+#endif
 }

source/dnode/vnode/inc/vnode.h

@@ -126,7 +126,7 @@ int     tqReadHandleSetTbUidList(STqReadHandle *pHandle, const SArray *tbUidList
 int     tqReadHandleAddTbUidList(STqReadHandle *pHandle, const SArray *tbUidList);
 int32_t tqReadHandleSetMsg(STqReadHandle *pHandle, SSubmitReq *pMsg, int64_t ver);
 bool    tqNextDataBlock(STqReadHandle *pHandle);
-int32_t tqRetrieveDataBlock(SArray **ppCols, STqReadHandle *pHandle, uint64_t *pGroupId, int32_t *pNumOfRows,
+int32_t tqRetrieveDataBlock(SArray **ppCols, STqReadHandle *pHandle, uint64_t *pGroupId, uint64_t* pUid, int32_t *pNumOfRows,
                             int16_t *pNumOfCols);
 
 // need to reposition

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

@@ -273,7 +273,8 @@ typedef enum {
 
 typedef struct {
   uint8_t  last : 1;
-  uint8_t  blkVer : 7;
+  uint8_t  hasDupKey : 1;  // 0: no dup TS key, 1: has dup TS key(since supporting Multi-Version)
+  uint8_t  blkVer : 6;
   uint8_t  numOfSubBlocks;
   col_id_t numOfCols;    // not including timestamp column
   uint32_t len;          // data block length
@@ -324,9 +325,8 @@ typedef struct {
 typedef struct {
   int16_t  colId;
   uint16_t type : 6;
-  uint16_t blen : 10;   // bitmap length(TODO: full UT for the bitmap compress of various data input)
+  uint16_t blen : 10;  // 0 no bitmap if all rows are NORM, > 0 bitmap length
-  uint32_t bitmap : 1;  // 0: no bitmap if all rows are NORM, 1: has bitmap if has NULL/NORM rows
+  uint32_t len;        // data length + bitmap length
-  uint32_t len : 31;    // data length + bitmap length
   uint32_t offset;
 } SBlockColV0;
 

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

@@ -196,7 +196,7 @@ int metaDropTable(SMeta *pMeta, int64_t version, SVDropTbReq *pReq) {
   int8_t      type;
   int64_t     ctime;
   tb_uid_t    suid;
-  int         c, ret;
+  int         c = 0, ret;
 
   // search & delete the name idx
   tdbDbcOpen(pMeta->pNameIdx, &pNameIdxc, &pMeta->txn);

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

@@ -161,7 +161,7 @@ int32_t tqPushMsgNew(STQ* pTq, void* msg, int32_t msgLen, tmsg_t msgType, int64_
       tqReadHandleSetMsg(pReader, pReq, 0);
       while (tqNextDataBlock(pReader)) {
         SSDataBlock block = {0};
-        if (tqRetrieveDataBlock(&block.pDataBlock, pReader, &block.info.groupId, &block.info.rows,
+        if (tqRetrieveDataBlock(&block.pDataBlock, pReader, &block.info.groupId, &block.info.uid, &block.info.rows,
                                 &block.info.numOfCols) < 0) {
           ASSERT(0);
         }
@@ -540,7 +540,7 @@ int32_t tqProcessPollReq(STQ* pTq, SRpcMsg* pMsg, int32_t workerId) {
         tqReadHandleSetMsg(pReader, pCont, 0);
         while (tqNextDataBlock(pReader)) {
           SSDataBlock block = {0};
-          if (tqRetrieveDataBlock(&block.pDataBlock, pReader, &block.info.groupId, &block.info.rows,
+          if (tqRetrieveDataBlock(&block.pDataBlock, pReader, &block.info.groupId, &block.info.uid, &block.info.rows,
                                   &block.info.numOfCols) < 0) {
             ASSERT(0);
           }

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

@@ -84,10 +84,12 @@ bool tqNextDataBlock(STqReadHandle* pHandle) {
   return false;
 }
 
-int32_t tqRetrieveDataBlock(SArray** ppCols, STqReadHandle* pHandle, uint64_t* pGroupId, int32_t* pNumOfRows,
+int32_t tqRetrieveDataBlock(SArray** ppCols, STqReadHandle* pHandle, uint64_t* pGroupId, uint64_t* pUid, int32_t* pNumOfRows,
                             int16_t* pNumOfCols) {
   /*int32_t         sversion = pHandle->pBlock->sversion;*/
   // TODO set to real sversion
+  *pUid = 0;
+
   int32_t sversion = 0;
   if (pHandle->sver != sversion) {
     pHandle->pSchema = metaGetTbTSchema(pHandle->pVnodeMeta, pHandle->msgIter.uid, sversion);
@@ -169,7 +171,10 @@ int32_t tqRetrieveDataBlock(SArray** ppCols, STqReadHandle* pHandle, uint64_t* p
   tdSTSRowIterInit(&iter, pTschema);
   STSRow* row;
   int32_t curRow = 0;
+
   tInitSubmitBlkIter(&pHandle->msgIter, pHandle->pBlock, &pHandle->blkIter);
+  *pUid = pHandle->msgIter.uid;  // set the uid of table for submit block
+
   while ((row = tGetSubmitBlkNext(&pHandle->blkIter)) != NULL) {
     tdSTSRowIterReset(&iter, row);
     // get all wanted col of that block

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

@@ -943,16 +943,16 @@ int tsdbWriteBlockImpl(STsdb *pRepo, STable *pTable, SDFile *pDFile, SDFile *pDF
                                                &(pAggrBlkCol->numOfNull));
 
       if (pAggrBlkCol->numOfNull == 0) {
-        TD_SET_COL_ROWS_NORM(pBlockCol);
+        pBlockCol->blen = 0;
       } else {
-        TD_SET_COL_ROWS_MISC(pBlockCol);
+        pBlockCol->blen = 1;
       }
       ++nColsOfBlockSma;
     } else if (tdIsBitmapBlkNorm(pDataCol->pBitmap, rowsToWrite, pDataCols->bitmapMode)) {
       // check if all rows normal
-      TD_SET_COL_ROWS_NORM(pBlockCol);
+      pBlockCol->blen = 0;
     } else {
-      TD_SET_COL_ROWS_MISC(pBlockCol);
+      pBlockCol->blen = 1;
     }
 
     ++nColsNotAllNull;
@@ -985,7 +985,7 @@ int tsdbWriteBlockImpl(STsdb *pRepo, STable *pTable, SDFile *pDFile, SDFile *pDF
 #ifdef TD_SUPPORT_BITMAP
     int32_t tBitmaps = 0;
     int32_t tBitmapsLen = 0;
-    if ((ncol != 0) && !TD_COL_ROWS_NORM(pBlockCol)) {
+    if ((ncol != 0) && (pBlockCol->blen > 0)) {
       tBitmaps = isSuper ? sBitmaps : nBitmaps;
     }
 #endif
@@ -1330,13 +1330,15 @@ static void tsdbLoadAndMergeFromCache(SDataCols *pDataCols, int *iter, SCommitIt
                                       TSKEY maxKey, int maxRows, int8_t update) {
   TSKEY     key1 = INT64_MAX;
   TSKEY     key2 = INT64_MAX;
+  TSKEY     lastKey = TSKEY_INITIAL_VAL;
   STSchema *pSchema = NULL;
 
   ASSERT(maxRows > 0 && dataColsKeyLast(pDataCols) <= maxKey);
   tdResetDataCols(pTarget);
 
   pTarget->bitmapMode = pDataCols->bitmapMode;
-
+  // TODO: filter Multi-Version
+  // TODO: support delete function
   while (true) {
     key1 = (*iter >= pDataCols->numOfRows) ? INT64_MAX : dataColsKeyAt(pDataCols, *iter);
     STSRow *row = tsdbNextIterRow(pCommitIter->pIter);
@@ -1349,6 +1351,9 @@ static void tsdbLoadAndMergeFromCache(SDataCols *pDataCols, int *iter, SCommitIt
     if (key1 == INT64_MAX && key2 == INT64_MAX) break;
 
     if (key1 < key2) {
+      if (lastKey != TSKEY_INITIAL_VAL) {
+        ++pTarget->numOfRows;
+      }
       for (int i = 0; i < pDataCols->numOfCols; ++i) {
         // TODO: dataColAppendVal may fail
         SCellVal sVal = {0};
@@ -1356,10 +1361,10 @@ static void tsdbLoadAndMergeFromCache(SDataCols *pDataCols, int *iter, SCommitIt
           TASSERT(0);
         }
         tdAppendValToDataCol(pTarget->cols + i, sVal.valType, sVal.val, pTarget->numOfRows, pTarget->maxPoints,
-                             pTarget->bitmapMode);
+                             pTarget->bitmapMode, false);
       }
 
-      ++pTarget->numOfRows;
+      lastKey = key1;
       ++(*iter);
     } else if (key1 > key2) {
       if (pSchema == NULL || schemaVersion(pSchema) != TD_ROW_SVER(row)) {
@@ -1367,7 +1372,17 @@ static void tsdbLoadAndMergeFromCache(SDataCols *pDataCols, int *iter, SCommitIt
         ASSERT(pSchema != NULL);
       }
 
-      tdAppendSTSRowToDataCol(row, pSchema, pTarget);
+      if (key2 == lastKey) {
+        if (TD_SUPPORT_UPDATE(update)) {
+          tdAppendSTSRowToDataCol(row, pSchema, pTarget, true);
+        }
+      } else {
+        if (lastKey != TSKEY_INITIAL_VAL) {
+          ++pTarget->numOfRows;
+        }
+        tdAppendSTSRowToDataCol(row, pSchema, pTarget, false);
+        lastKey = key2;
+      }
 
       tSkipListIterNext(pCommitIter->pIter);
     } else {
@@ -1397,6 +1412,12 @@ static void tsdbLoadAndMergeFromCache(SDataCols *pDataCols, int *iter, SCommitIt
       ++(*iter);
       tSkipListIterNext(pCommitIter->pIter);
 #endif
+
+      if(lastKey != key1) {
+        lastKey = key1;
+        ++pTarget->numOfRows;
+      }
+
       // copy disk data
       for (int i = 0; i < pDataCols->numOfCols; ++i) {
         SCellVal sVal = {0};
@@ -1405,7 +1426,7 @@ static void tsdbLoadAndMergeFromCache(SDataCols *pDataCols, int *iter, SCommitIt
         }
         // TODO: tdAppendValToDataCol may fail
         tdAppendValToDataCol(pTarget->cols + i, sVal.valType, sVal.val, pTarget->numOfRows, pTarget->maxPoints,
-                             pTarget->bitmapMode);
+                             pTarget->bitmapMode, false);
       }
 
       if (TD_SUPPORT_UPDATE(update)) {
@@ -1416,26 +1437,17 @@ static void tsdbLoadAndMergeFromCache(SDataCols *pDataCols, int *iter, SCommitIt
         }
 
         // TODO: merge with Multi-Version
-        STSRow *curRow = row;
+        tdAppendSTSRowToDataCol(row, pSchema, pTarget, true);
-
-        ++(*iter);
-        tSkipListIterNext(pCommitIter->pIter);
-        STSRow *nextRow = tsdbNextIterRow(pCommitIter->pIter);
-
-        if (key2 < TD_ROW_KEY(nextRow)) {
-          tdAppendSTSRowToDataCol(row, pSchema, pTarget);
-        } else {
-          tdAppendSTSRowToDataCol(row, pSchema, pTarget);
-        }
-        // TODO: merge with Multi-Version
-      } else {
-        ++pTarget->numOfRows;
-        ++(*iter);
-        tSkipListIterNext(pCommitIter->pIter);
       }
+      ++(*iter);
+      tSkipListIterNext(pCommitIter->pIter);
     }
 
-    if (pTarget->numOfRows >= maxRows) break;
+    if (pTarget->numOfRows >= (maxRows - 1)) break;
+  }
+
+  if (lastKey != TSKEY_INITIAL_VAL) {
+    ++pTarget->numOfRows;
   }
 }
 

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

@@ -20,7 +20,7 @@ static void     tsdbFreeTbData(STbData *pTbData);
 static char    *tsdbGetTsTupleKey(const void *data);
 static int      tsdbTbDataComp(const void *arg1, const void *arg2);
 static char    *tsdbTbDataGetUid(const void *arg);
-static int      tsdbAppendTableRowToCols(STable *pTable, SDataCols *pCols, STSchema **ppSchema, STSRow *row);
+static int tsdbAppendTableRowToCols(STable *pTable, SDataCols *pCols, STSchema **ppSchema, STSRow *row, bool merge);
 
 int tsdbMemTableCreate(STsdb *pTsdb, STsdbMemTable **ppMemTable) {
   STsdbMemTable *pMemTable;
@@ -82,14 +82,19 @@ int tsdbLoadDataFromCache(STable *pTable, SSkipListIterator *pIter, TSKEY maxKey
                           TKEY *filterKeys, int nFilterKeys, bool keepDup, SMergeInfo *pMergeInfo) {
   ASSERT(maxRowsToRead > 0 && nFilterKeys >= 0);
   if (pIter == NULL) return 0;
-  STSchema  *pSchema = NULL;
+  STSchema *pSchema = NULL;
-  TSKEY      rowKey = 0;
+  TSKEY     rowKey = 0;
-  TSKEY      fKey = 0;
+  TSKEY     fKey = 0;
+  // only fetch lastKey from mem data as file data not used in this function actually
+  TSKEY      lastKey = TSKEY_INITIAL_VAL;
   bool       isRowDel = false;
   int        filterIter = 0;
   STSRow    *row = NULL;
   SMergeInfo mInfo;
 
+  // TODO: support Multi-Version(the rows with the same TS keys in memory can't be merged if its version refered by
+  // query handle)
+
   if (pMergeInfo == NULL) pMergeInfo = &mInfo;
 
   memset(pMergeInfo, 0, sizeof(*pMergeInfo));
@@ -111,7 +116,8 @@ int tsdbLoadDataFromCache(STable *pTable, SSkipListIterator *pIter, TSKEY maxKey
   } else {
     fKey = tdGetKey(filterKeys[filterIter]);
   }
-
+  // 1. fkey - no dup since merged up to maxVersion of each query handle by tsdbLoadBlockDataCols
+  // 2. rowKey - would dup since Multi-Version supported
   while (true) {
     if (fKey == INT64_MAX && rowKey == INT64_MAX) break;
 
@@ -125,12 +131,14 @@ int tsdbLoadDataFromCache(STable *pTable, SSkipListIterator *pIter, TSKEY maxKey
       } else {
         fKey = tdGetKey(filterKeys[filterIter]);
       }
+#if 0
     } else if (fKey > rowKey) {
       if (isRowDel) {
         pMergeInfo->rowsDeleteFailed++;
       } else {
         if (pMergeInfo->rowsInserted - pMergeInfo->rowsDeleteSucceed >= maxRowsToRead) break;
         if (pCols && pMergeInfo->nOperations >= pCols->maxPoints) break;
+
         pMergeInfo->rowsInserted++;
         pMergeInfo->nOperations++;
         pMergeInfo->keyFirst = TMIN(pMergeInfo->keyFirst, rowKey);
@@ -185,6 +193,94 @@ int tsdbLoadDataFromCache(STable *pTable, SSkipListIterator *pIter, TSKEY maxKey
         fKey = tdGetKey(filterKeys[filterIter]);
       }
     }
+#endif
+#if 1
+    } else if (fKey > rowKey) {
+      if (isRowDel) {
+        // TODO: support delete function
+        pMergeInfo->rowsDeleteFailed++;
+      } else {
+        if (pMergeInfo->rowsInserted - pMergeInfo->rowsDeleteSucceed >= maxRowsToRead) break;
+        if (pCols && pMergeInfo->nOperations >= pCols->maxPoints) break;
+
+        if (lastKey != rowKey) {
+          pMergeInfo->rowsInserted++;
+          pMergeInfo->nOperations++;
+          pMergeInfo->keyFirst = TMIN(pMergeInfo->keyFirst, rowKey);
+          pMergeInfo->keyLast = TMAX(pMergeInfo->keyLast, rowKey);
+          if (pCols) {
+            if (lastKey != TSKEY_INITIAL_VAL) {
+              ++pCols->numOfRows;
+            }
+            tsdbAppendTableRowToCols(pTable, pCols, &pSchema, row, false);
+          }
+          lastKey = rowKey;
+        } else {
+          if (keepDup) {
+            tsdbAppendTableRowToCols(pTable, pCols, &pSchema, row, true);
+          } else {
+            // discard
+          }
+        }
+      }
+
+      tSkipListIterNext(pIter);
+      row = tsdbNextIterRow(pIter);
+      if (row == NULL || TD_ROW_KEY(row) > maxKey) {
+        rowKey = INT64_MAX;
+        isRowDel = false;
+      } else {
+        rowKey = TD_ROW_KEY(row);
+        isRowDel = TD_ROW_IS_DELETED(row);
+      }
+    } else {           // fkey == rowKey
+      if (isRowDel) {  // TODO: support delete function(How to stands for delete in file? rowVersion = -1?)
+        ASSERT(!keepDup);
+        if (pCols && pMergeInfo->nOperations >= pCols->maxPoints) break;
+        pMergeInfo->rowsDeleteSucceed++;
+        pMergeInfo->nOperations++;
+        tsdbAppendTableRowToCols(pTable, pCols, &pSchema, row, false);
+      } else {
+        if (keepDup) {
+          if (pCols && pMergeInfo->nOperations >= pCols->maxPoints) break;
+          if (lastKey != rowKey) {
+            pMergeInfo->rowsUpdated++;
+            pMergeInfo->nOperations++;
+            pMergeInfo->keyFirst = TMIN(pMergeInfo->keyFirst, rowKey);
+            pMergeInfo->keyLast = TMAX(pMergeInfo->keyLast, rowKey);
+            lastKey = rowKey;
+            ++pCols->numOfRows;
+            tsdbAppendTableRowToCols(pTable, pCols, &pSchema, row, false);
+          } else {
+            tsdbAppendTableRowToCols(pTable, pCols, &pSchema, row, true);
+          }
+        } else {
+          pMergeInfo->keyFirst = TMIN(pMergeInfo->keyFirst, fKey);
+          pMergeInfo->keyLast = TMAX(pMergeInfo->keyLast, fKey);
+        }
+      }
+
+      tSkipListIterNext(pIter);
+      row = tsdbNextIterRow(pIter);
+      if (row == NULL || TD_ROW_KEY(row) > maxKey) {
+        rowKey = INT64_MAX;
+        isRowDel = false;
+      } else {
+        rowKey = TD_ROW_KEY(row);
+        isRowDel = TD_ROW_IS_DELETED(row);
+      }
+
+      filterIter++;
+      if (filterIter >= nFilterKeys) {
+        fKey = INT64_MAX;
+      } else {
+        fKey = tdGetKey(filterKeys[filterIter]);
+      }
+    }
+#endif
+  }
+  if (lastKey != TSKEY_INITIAL_VAL) {
+    ++pCols->numOfRows;
   }
 
   return 0;
@@ -254,9 +350,12 @@ static STbData *tsdbNewTbData(tb_uid_t uid) {
   pTbData->keyMin = TSKEY_MAX;
   pTbData->keyMax = TSKEY_MIN;
   pTbData->nrows = 0;
-
+#if 0
   pTbData->pData = tSkipListCreate(5, TSDB_DATA_TYPE_TIMESTAMP, sizeof(int64_t), tkeyComparFn, SL_DISCARD_DUP_KEY,
                                    tsdbGetTsTupleKey);
+#endif
+  pTbData->pData =
+      tSkipListCreate(5, TSDB_DATA_TYPE_TIMESTAMP, sizeof(int64_t), tkeyComparFn, SL_ALLOW_DUP_KEY, tsdbGetTsTupleKey);
   if (pTbData->pData == NULL) {
     taosMemoryFree(pTbData);
     return NULL;
@@ -291,7 +390,7 @@ static char *tsdbTbDataGetUid(const void *arg) {
   STbData *pTbData = (STbData *)arg;
   return (char *)(&(pTbData->uid));
 }
-static int tsdbAppendTableRowToCols(STable *pTable, SDataCols *pCols, STSchema **ppSchema, STSRow *row) {
+static int tsdbAppendTableRowToCols(STable *pTable, SDataCols *pCols, STSchema **ppSchema, STSRow *row, bool merge) {
   if (pCols) {
     if (*ppSchema == NULL || schemaVersion(*ppSchema) != TD_ROW_SVER(row)) {
       *ppSchema = tsdbGetTableSchemaImpl(pTable, false, false, TD_ROW_SVER(row));
@@ -301,7 +400,7 @@ static int tsdbAppendTableRowToCols(STable *pTable, SDataCols *pCols, STSchema *
       }
     }
 
-    tdAppendSTSRowToDataCol(row, *ppSchema, pCols);
+    tdAppendSTSRowToDataCol(row, *ppSchema, pCols, merge);
   }
 
   return 0;

source/dnode/vnode/src/tsdb/tsdbMemTable2.c

@@ -15,10 +15,11 @@
 
 #include "tsdb.h"
 
-typedef struct SMemTable       SMemTable;
+typedef struct SMemTable          SMemTable;
-typedef struct SMemData        SMemData;
+typedef struct SMemData           SMemData;
-typedef struct SMemSkipList    SMemSkipList;
+typedef struct SMemSkipList       SMemSkipList;
-typedef struct SMemSkipListCfg SMemSkipListCfg;
+typedef struct SMemSkipListNode   SMemSkipListNode;
+typedef struct SMemSkipListCurosr SMemSkipListCurosr;
 
 struct SMemTable {
   STsdb     *pTsdb;
@@ -32,15 +33,17 @@ struct SMemTable {
   SMemData **pBuckets;
 };
 
-struct SMemSkipListCfg {
+struct SMemSkipListNode {
-  int8_t  maxLevel;
+  int8_t            level;
-  int32_t nKey;
+  SMemSkipListNode *forwards[];
-  int32_t nData;
 };
 
 struct SMemSkipList {
-  int8_t   level;
+  uint32_t         seed;
-  uint32_t seed;
+  int8_t           maxLevel;
+  int8_t           level;
+  int32_t          size;
+  SMemSkipListNode pHead[];
 };
 
 struct SMemData {
@@ -55,6 +58,20 @@ struct SMemData {
   SMemSkipList sl;
 };
 
+struct SMemSkipListCurosr {
+  SMemSkipList     *pSl;
+  SMemSkipListNode *pNodeC;
+};
+
+#define SL_NODE_SIZE(l)        (sizeof(SMemSkipListNode) + sizeof(SMemSkipListNode *) * (l)*2)
+#define SL_NODE_HALF_SIZE(l)   (sizeof(SMemSkipListNode) + sizeof(SMemSkipListNode *) * (l))
+#define SL_NODE_FORWARD(n, l)  ((n)->forwards[l])
+#define SL_NODE_BACKWARD(n, l) ((n)->forwards[(n)->level + (l)])
+#define SL_NODE_DATA(n)        (&SL_NODE_BACKWARD(n, (n)->level))
+
+#define SL_HEAD_NODE(sl) ((sl)->pHead)
+#define SL_TAIL_NODE(sl) ((SMemSkipListNode *)&SL_NODE_FORWARD(SL_HEAD_NODE(sl), (sl)->maxLevel))
+
 // SMemTable
 int32_t tsdbMemTableCreate2(STsdb *pTsdb, SMemTable **ppMemTb) {
   SMemTable *pMemTb = NULL;
@@ -76,6 +93,7 @@ int32_t tsdbMemTableCreate2(STsdb *pTsdb, SMemTable **ppMemTb) {
   pMemTb->pBuckets = taosMemoryCalloc(pMemTb->nBucket, sizeof(*pMemTb->pBuckets));
   if (pMemTb->pBuckets == NULL) {
     terrno = TSDB_CODE_OUT_OF_MEMORY;
+    taosMemoryFree(pMemTb);
     return -1;
   }
 
@@ -83,8 +101,121 @@ int32_t tsdbMemTableCreate2(STsdb *pTsdb, SMemTable **ppMemTb) {
   return 0;
 }
 
-int32_t tsdbMemTableDestroy2(STsdb *pTsdb, SMemTable *pMT) {
+int32_t tsdbMemTableDestroy2(STsdb *pTsdb, SMemTable *pMemTb) {
-  // TODO
+  if (pMemTb) {
+    // loop to destroy the contents (todo)
+    taosMemoryFree(pMemTb->pBuckets);
+    taosMemoryFree(pMemTb);
+  }
+  return 0;
+}
+
+int32_t tsdbInsertData2(SMemTable *pMemTb, int64_t version, const SVSubmitBlk *pSubmitBlk) {
+  SMemData          *pMemData;
+  STsdb             *pTsdb = pMemTb->pTsdb;
+  SVnode            *pVnode = pTsdb->pVnode;
+  SVBufPool         *pPool = pVnode->inUse;
+  int32_t            hash;
+  int32_t            tlen;
+  uint8_t            buf[16];
+  int32_t            rlen;
+  const uint8_t     *p;
+  SMemSkipListNode  *pSlNode;
+  const STSRow      *pTSRow;
+  SMemSkipListCurosr slc = {0};
+
+  // search hash
+  hash = (pSubmitBlk->suid + pSubmitBlk->uid) % pMemTb->nBucket;
+  for (pMemData = pMemTb->pBuckets[hash]; pMemData; pMemData = pMemData->pHashNext) {
+    if (pMemData->suid == pSubmitBlk->suid && pMemData->uid == pSubmitBlk->uid) break;
+  }
+
+  // create pMemData if need
+  if (pMemData == NULL) {
+    int8_t            maxLevel = pVnode->config.tsdbCfg.slLevel;
+    int32_t           tsize = sizeof(*pMemData) + SL_NODE_HALF_SIZE(maxLevel) * 2;
+    SMemSkipListNode *pHead, *pTail;
+
+    pMemData = vnodeBufPoolMalloc(pPool, tsize);
+    if (pMemData == NULL) {
+      terrno = TSDB_CODE_OUT_OF_MEMORY;
+      return -1;
+    }
+
+    pMemData->pHashNext = NULL;
+    pMemData->suid = pSubmitBlk->suid;
+    pMemData->uid = pSubmitBlk->uid;
+    pMemData->minKey = TSKEY_MAX;
+    pMemData->maxKey = TSKEY_MIN;
+    pMemData->minVer = -1;
+    pMemData->maxVer = -1;
+    pMemData->nRows = 0;
+    pMemData->sl.seed = taosRand();
+    pMemData->sl.maxLevel = maxLevel;
+    pMemData->sl.level = 0;
+    pMemData->sl.size = 0;
+    pHead = SL_HEAD_NODE(&pMemData->sl);
+    pTail = SL_TAIL_NODE(&pMemData->sl);
+    pHead->level = maxLevel;
+    pTail->level = maxLevel;
+    for (int iLevel = 0; iLevel < maxLevel; iLevel++) {
+      SL_NODE_FORWARD(pHead, iLevel) = pTail;
+      SL_NODE_FORWARD(pTail, iLevel) = pHead;
+    }
+
+    // add to MemTable
+    hash = (pMemData->suid + pMemData->uid) % pMemTb->nBucket;
+    pMemData->pHashNext = pMemTb->pBuckets[hash];
+    pMemTb->pBuckets[hash] = pMemData;
+    pMemTb->nHash++;
+  }
+
+// loop to insert data to skiplist
+#if 0
+  tsdbMemSkipListCursorOpen(&slc, &pMemData->sl);
+  p = pSubmitBlk->pData;
+  for (;;) {
+    if (p - (uint8_t *)pSubmitBlk->pData >= pSubmitBlk->nData) break;
+
+    const uint8_t *pt = p;
+    p = tGetBinary(p, &pTSRow, &rlen);
+
+    // check the row (todo)
+
+    // move the cursor to position to write (todo)
+    int32_t c;
+    tsdbMemSkipListCursorMoveTo(&slc, pTSRow, version, &c);
+    ASSERT(c);
+
+    // encode row
+    int8_t  level = tsdbMemSkipListRandLevel(&pMemData->sl);
+    int32_t tsize = SL_NODE_SIZE(level) + sizeof(version) + (p - pt);
+    pSlNode = vnodeBufPoolMalloc(pPool, tsize);
+    pSlNode->level = level;
+
+    uint8_t *pData = SL_NODE_DATA(pSlNode);
+    *(int64_t *)pData = version;
+    pData += sizeof(version);
+    memcpy(pData, pt, p - pt);
+
+    // insert row
+    tsdbMemSkipListCursorPut(&slc, pSlNode);
+
+    // update status
+    if (pTSRow->ts < pMemData->minKey) pMemData->minKey = pTSRow->ts;
+    if (pTSRow->ts > pMemData->maxKey) pMemData->maxKey = pTSRow->ts;
+  }
+  tsdbMemSkipListCursorClose(&slc);
+#endif
+
+  if (pMemData->minVer == -1) pMemData->minVer = version;
+  if (pMemData->maxVer == -1 || pMemData->maxVer < version) pMemData->maxVer = version;
+
+  if (pMemTb->minKey < pMemData->minKey) pMemTb->minKey = pMemData->minKey;
+  if (pMemTb->maxKey < pMemData->maxKey) pMemTb->maxKey = pMemData->maxKey;
+  if (pMemTb->minVer == -1) pMemTb->minVer = version;
+  if (pMemTb->maxVer == -1 || pMemTb->maxVer < version) pMemTb->maxVer = version;
+
   return 0;
 }
 

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

@@ -163,7 +163,7 @@ static int32_t tsdbCheckInfoCompar(const void* key1, const void* key2);
 // static void*   destroyTableCheckInfo(SArray* pTableCheckInfo);
 static bool tsdbGetExternalRow(tsdbReaderT pHandle);
 
-static STsdb* getTsdbByRetentions(SVnode* pVnode, TSKEY winSKey, SRetention* retentions);
+static STsdb* getTsdbByRetentions(SVnode* pVnode, STsdbReadHandle* pReadHandle, TSKEY winSKey, SRetention* retentions);
 
 static void tsdbInitDataBlockLoadInfo(SDataBlockLoadInfo* pBlockLoadInfo) {
   pBlockLoadInfo->slot = -1;
@@ -351,36 +351,43 @@ static void setQueryTimewindow(STsdbReadHandle* pTsdbReadHandle, SQueryTableData
               pTsdbReadHandle->window.ekey, pTsdbReadHandle->idStr);
   }
 }
-#if 0
+#if 1
-int           nQUERY = 0;
+int nQUERY = 0;
 #endif
-static STsdb* getTsdbByRetentions(SVnode* pVnode, TSKEY winSKey, SRetention* retentions) {
+static STsdb* getTsdbByRetentions(SVnode* pVnode, STsdbReadHandle* pReadHandle, TSKEY winSKey, SRetention* retentions) {
   if (vnodeIsRollup(pVnode)) {
     int level = 0;
-#if 1
+#if 0
     int64_t now = taosGetTimestamp(pVnode->config.tsdbCfg.precision);
     for (int i = 0; i < TSDB_RETENTION_MAX; ++i) {
       SRetention* pRetention = retentions + i;
       if (pRetention->keep <= 0 || (now - pRetention->keep) >= winSKey) {
         break;
       }
+      ++level;
     }
 #endif
-#if 0
+#if 1
-    ++nQUERY;
+    switch ((nQUERY++) % 3) {
-    if(nQUERY%3 == 0) {
+      case 0:
-      level = 2;
+        level = 0;
-    } else if(nQUERY%2 == 0) {
+        break;
-      level = 1;
+      case 1:
-    } else {
+        level = 1;
-      level = 0;
+        break;
+      default:
+        level = 2;
+        break;
     }
 #endif
     if (level == TSDB_RETENTION_L0) {
+      tsdbDebug("%p rsma level %d is selected to query\n", pReadHandle, level);
       return VND_RSMA0(pVnode);
     } else if (level == TSDB_RETENTION_L1) {
+      tsdbDebug("%p rsma level %d is selected to query\n", pReadHandle, level);
       return VND_RSMA1(pVnode);
     } else {
+      tsdbDebug("%p rsma level %d is selected to query\n", pReadHandle, level);
       return VND_RSMA2(pVnode);
     }
   }
@@ -393,7 +400,7 @@ static STsdbReadHandle* tsdbQueryTablesImpl(SVnode* pVnode, SQueryTableDataCond*
     goto _end;
   }
 
-  STsdb* pTsdb = getTsdbByRetentions(pVnode, pCond->twindow.skey, pVnode->config.tsdbCfg.retentions);
+  STsdb* pTsdb = getTsdbByRetentions(pVnode, pReadHandle, pCond->twindow.skey, pVnode->config.tsdbCfg.retentions);
 
   pReadHandle->order = pCond->order;
   pReadHandle->pTsdb = pTsdb;
@@ -803,12 +810,16 @@ static void destroyTableMemIterator(STableCheckInfo* pCheckInfo) {
   tSkipListDestroyIter(pCheckInfo->iiter);
 }
 
-static TSKEY extractFirstTraverseKey(STableCheckInfo* pCheckInfo, int32_t order, int32_t update) {
+static TSKEY extractFirstTraverseKey(STableCheckInfo* pCheckInfo, int32_t order, int32_t update, TDRowVerT maxVer) {
   STSRow *rmem = NULL, *rimem = NULL;
   if (pCheckInfo->iter) {
     SSkipListNode* node = tSkipListIterGet(pCheckInfo->iter);
     if (node != NULL) {
       rmem = (STSRow*)SL_GET_NODE_DATA(node);
+      // TODO: filter max version
+      // if (TD_ROW_VER(rmem) > maxVer) {
+      //   rmem = NULL;
+      // }
     }
   }
 
@@ -816,6 +827,10 @@ static TSKEY extractFirstTraverseKey(STableCheckInfo* pCheckInfo, int32_t order,
     SSkipListNode* node = tSkipListIterGet(pCheckInfo->iiter);
     if (node != NULL) {
       rimem = (STSRow*)SL_GET_NODE_DATA(node);
+      // TODO: filter max version
+      // if (TD_ROW_VER(rimem) > maxVer) {
+      //   rimem = NULL;
+      // }
     }
   }
 
@@ -837,6 +852,7 @@ static TSKEY extractFirstTraverseKey(STableCheckInfo* pCheckInfo, int32_t order,
   TSKEY r2 = TD_ROW_KEY(rimem);
 
   if (r1 == r2) {
+#if 0
     if (update == TD_ROW_DISCARD_UPDATE) {
       pCheckInfo->chosen = CHECKINFO_CHOSEN_IMEM;
       tSkipListIterNext(pCheckInfo->iter);
@@ -846,6 +862,13 @@ static TSKEY extractFirstTraverseKey(STableCheckInfo* pCheckInfo, int32_t order,
     } else {
       pCheckInfo->chosen = CHECKINFO_CHOSEN_BOTH;
     }
+#endif
+    if (TD_SUPPORT_UPDATE(update)) {
+      pCheckInfo->chosen = CHECKINFO_CHOSEN_BOTH;
+    } else {
+      pCheckInfo->chosen = CHECKINFO_CHOSEN_IMEM;
+      tSkipListIterNext(pCheckInfo->iter);
+    }
     return r1;
   } else if (r1 < r2 && ASCENDING_TRAVERSE(order)) {
     pCheckInfo->chosen = CHECKINFO_CHOSEN_MEM;
@@ -856,12 +879,18 @@ static TSKEY extractFirstTraverseKey(STableCheckInfo* pCheckInfo, int32_t order,
   }
 }
 
-static STSRow* getSRowInTableMem(STableCheckInfo* pCheckInfo, int32_t order, int32_t update, STSRow** extraRow) {
+
+static STSRow* getSRowInTableMem(STableCheckInfo* pCheckInfo, int32_t order, int32_t update, STSRow** extraRow, TDRowVerT maxVer) {
   STSRow *rmem = NULL, *rimem = NULL;
   if (pCheckInfo->iter) {
     SSkipListNode* node = tSkipListIterGet(pCheckInfo->iter);
     if (node != NULL) {
       rmem = (STSRow*)SL_GET_NODE_DATA(node);
+#if 0 // TODO: skiplist refactor
+      if (TD_ROW_VER(rmem) > maxVer) {
+        rmem = NULL;
+      }
+#endif
     }
   }
 
@@ -869,6 +898,11 @@ static STSRow* getSRowInTableMem(STableCheckInfo* pCheckInfo, int32_t order, int
     SSkipListNode* node = tSkipListIterGet(pCheckInfo->iiter);
     if (node != NULL) {
       rimem = (STSRow*)SL_GET_NODE_DATA(node);
+#if 0 // TODO: skiplist refactor
+      if (TD_ROW_VER(rimem) > maxVer) {
+        rimem = NULL;
+      }
+#endif
     }
   }
 
@@ -890,6 +924,7 @@ static STSRow* getSRowInTableMem(STableCheckInfo* pCheckInfo, int32_t order, int
   TSKEY r2 = TD_ROW_KEY(rimem);
 
   if (r1 == r2) {
+#if 0
     if (update == TD_ROW_DISCARD_UPDATE) {
       tSkipListIterNext(pCheckInfo->iter);
       pCheckInfo->chosen = CHECKINFO_CHOSEN_IMEM;
@@ -903,6 +938,16 @@ static STSRow* getSRowInTableMem(STableCheckInfo* pCheckInfo, int32_t order, int
       *extraRow = rimem;
       return rmem;
     }
+#endif
+    if (TD_SUPPORT_UPDATE(update)) {
+      pCheckInfo->chosen = CHECKINFO_CHOSEN_BOTH;
+      *extraRow = rimem;
+      return rmem;
+    } else {
+      tSkipListIterNext(pCheckInfo->iter);
+      pCheckInfo->chosen = CHECKINFO_CHOSEN_IMEM;
+      return rimem;
+    }
   } else {
     if (ASCENDING_TRAVERSE(order)) {
       if (r1 < r2) {
@@ -973,7 +1018,7 @@ static bool hasMoreDataInCache(STsdbReadHandle* pHandle) {
     initTableMemIterator(pHandle, pCheckInfo);
   }
 
-  STSRow* row = getSRowInTableMem(pCheckInfo, pHandle->order, pCfg->update, NULL);
+  STSRow* row = getSRowInTableMem(pCheckInfo, pHandle->order, pCfg->update, NULL, TD_VER_MAX);
   if (row == NULL) {
     return false;
   }
@@ -1250,7 +1295,7 @@ static int32_t handleDataMergeIfNeeded(STsdbReadHandle* pTsdbReadHandle, SBlock*
   /*bool hasData = */ initTableMemIterator(pTsdbReadHandle, pCheckInfo);
   assert(cur->pos >= 0 && cur->pos <= binfo.rows);
 
-  key = extractFirstTraverseKey(pCheckInfo, pTsdbReadHandle->order, pCfg->update);
+  key = extractFirstTraverseKey(pCheckInfo, pTsdbReadHandle->order, pCfg->update, TD_VER_MAX);
 
   if (key != TSKEY_INITIAL_VAL) {
     tsdbDebug("%p key in mem:%" PRId64 ", %s", pTsdbReadHandle, key, pTsdbReadHandle->idStr);
@@ -1497,10 +1542,10 @@ static int32_t doCopyRowsFromFileBlock(STsdbReadHandle* pTsdbReadHandle, int32_t
             TASSERT(0);
           }
 
-          if (sVal.valType == TD_VTYPE_NULL) {
+          if (sVal.valType == TD_VTYPE_NORM) {
-            colDataAppendNULL(pColInfo, rowIndex);
-          } else {
             colDataAppend(pColInfo, rowIndex, sVal.val, false);
+          } else {
+            colDataAppendNULL(pColInfo, rowIndex);
           }
         }
       } else {  // handle the var-string
@@ -1513,10 +1558,10 @@ static int32_t doCopyRowsFromFileBlock(STsdbReadHandle* pTsdbReadHandle, int32_t
             TASSERT(0);
           }
 
-          if (sVal.valType == TD_VTYPE_NULL) {
+          if (sVal.valType == TD_VTYPE_NORM) {
-            colDataAppendNULL(pColInfo, rowIndex);
-          } else {
             colDataAppend(pColInfo, rowIndex, sVal.val, false);
+          } else {
+            colDataAppendNULL(pColInfo, rowIndex);
           }
         }
       }
@@ -1541,11 +1586,26 @@ static int32_t doCopyRowsFromFileBlock(STsdbReadHandle* pTsdbReadHandle, int32_t
   return numOfRows + num;
 }
 
-// TODO fix bug for reverse copy data problem
+/**
-// Note: row1 always has high priority
+ * @brief  // TODO fix bug for reverse copy data problem
-static void mergeTwoRowFromMem(STsdbReadHandle* pTsdbReadHandle, int32_t capacity, int32_t numOfRows, STSRow* row1,
+ *        Note: row1 always has high priority
-                               STSRow* row2, int32_t numOfCols, uint64_t uid, STSchema* pSchema1, STSchema* pSchema2,
+ *
-                               bool forceSetNull) {
+ * @param pTsdbReadHandle
+ * @param capacity
+ * @param curRow
+ * @param row1
+ * @param row2
+ * @param numOfCols
+ * @param uid
+ * @param pSchema1
+ * @param pSchema2
+ * @param update
+ * @param lastRowKey
+ * @return int32_t The quantity of rows appended
+ */
+static int32_t mergeTwoRowFromMem(STsdbReadHandle* pTsdbReadHandle, int32_t capacity, int32_t* curRow, STSRow* row1,
+                                  STSRow* row2, int32_t numOfCols, uint64_t uid, STSchema* pSchema1, STSchema* pSchema2,
+                                  bool update, TSKEY* lastRowKey) {
 #if 1
   STSchema* pSchema;
   STSRow*   row;
@@ -1557,12 +1617,17 @@ static void mergeTwoRowFromMem(STsdbReadHandle* pTsdbReadHandle, int32_t capacit
   bool     isChosenRowDataRow;
   int32_t  chosen_itr;
   SCellVal sVal = {0};
+  TSKEY    rowKey = TSKEY_INITIAL_VAL;
+  int32_t  nResult = 0;
+  int32_t  mergeOption = 0;  // 0 discard 1 overwrite 2 merge
 
   // the schema version info is embeded in STSRow
   int32_t numOfColsOfRow1 = 0;
 
   if (pSchema1 == NULL) {
-    pSchema1 = metaGetTbTSchema(REPO_META(pTsdbReadHandle->pTsdb), uid, TD_ROW_SVER(row1));
+    // pSchema1 = metaGetTbTSchema(REPO_META(pTsdbReadHandle->pTsdb), uid, TD_ROW_SVER(row1));
+    // TODO: use the real schemaVersion
+    pSchema1 = metaGetTbTSchema(REPO_META(pTsdbReadHandle->pTsdb), uid, 0);
   }
 
 #ifdef TD_DEBUG_PRINT_ROW
@@ -1579,7 +1644,9 @@ static void mergeTwoRowFromMem(STsdbReadHandle* pTsdbReadHandle, int32_t capacit
   if (row2) {
     isRow2DataRow = TD_IS_TP_ROW(row2);
     if (pSchema2 == NULL) {
-      pSchema2 = metaGetTbTSchema(REPO_META(pTsdbReadHandle->pTsdb), uid, TD_ROW_SVER(row2));
+      // pSchema2 = metaGetTbTSchema(REPO_META(pTsdbReadHandle->pTsdb), uid, TD_ROW_SVER(row2));
+      // TODO: use the real schemaVersion
+      pSchema2 = metaGetTbTSchema(REPO_META(pTsdbReadHandle->pTsdb), uid, 0);
     }
     if (isRow2DataRow) {
       numOfColsOfRow2 = schemaNCols(pSchema2);
@@ -1610,19 +1677,19 @@ static void mergeTwoRowFromMem(STsdbReadHandle* pTsdbReadHandle, int32_t capacit
       colIdOfRow2 = tdKvRowColIdAt(row2, k);
     }
 
-    if (colIdOfRow1 == colIdOfRow2) {
+    if (colIdOfRow1 < colIdOfRow2) { // the most probability
       if (colIdOfRow1 < pColInfo->info.colId) {
-        j++;
+        ++j;
-        k++;
         continue;
       }
       row = row1;
       pSchema = pSchema1;
       isChosenRowDataRow = isRow1DataRow;
       chosen_itr = j;
-    } else if (colIdOfRow1 < colIdOfRow2) {
+    } else if (colIdOfRow1 == colIdOfRow2) {
       if (colIdOfRow1 < pColInfo->info.colId) {
-        j++;
+        ++j;
+        ++k;
         continue;
       }
       row = row1;
@@ -1631,7 +1698,7 @@ static void mergeTwoRowFromMem(STsdbReadHandle* pTsdbReadHandle, int32_t capacit
       chosen_itr = j;
     } else {
       if (colIdOfRow2 < pColInfo->info.colId) {
-        k++;
+        ++k;
         continue;
       }
       row = row2;
@@ -1639,16 +1706,48 @@ static void mergeTwoRowFromMem(STsdbReadHandle* pTsdbReadHandle, int32_t capacit
       chosen_itr = k;
       isChosenRowDataRow = isRow2DataRow;
     }
+
     if (isChosenRowDataRow) {
       colId = pSchema->columns[chosen_itr].colId;
       offset = pSchema->columns[chosen_itr].offset;
       // TODO: use STSRowIter
       tdSTpRowGetVal(row, colId, pSchema->columns[chosen_itr].type, pSchema->flen, offset, chosen_itr - 1, &sVal);
+      if (colId == PRIMARYKEY_TIMESTAMP_COL_ID) {
+        rowKey = *(TSKEY*)sVal.val;
+        if (rowKey != *lastRowKey) {
+          mergeOption = 1;
+          if (*lastRowKey != TSKEY_INITIAL_VAL) {
+            ++(*curRow);
+          }
+          ++nResult;
+        } else if (update){
+          mergeOption = 2;
+        } else {
+          mergeOption = 0;
+          break;
+        }
+
+        *lastRowKey = rowKey;
+      }
     } else {
       // TODO: use STSRowIter
       if (chosen_itr == 0) {
         colId = PRIMARYKEY_TIMESTAMP_COL_ID;
         tdSKvRowGetVal(row, PRIMARYKEY_TIMESTAMP_COL_ID, -1, -1, &sVal);
+        rowKey = *(TSKEY*)sVal.val;
+        if (rowKey != *lastRowKey) {
+          mergeOption = 1;
+          if (*lastRowKey != TSKEY_INITIAL_VAL) {
+            ++(*curRow);
+          }
+          ++nResult;
+        } else if(update) {
+          mergeOption = 2;
+        } else {
+          mergeOption = 0;
+          break;
+        }
+        *lastRowKey = rowKey;
       } else {
         SKvRowIdx* pColIdx = tdKvRowColIdxAt(row, chosen_itr - 1);
         colId = pColIdx->colId;
@@ -1657,35 +1756,46 @@ static void mergeTwoRowFromMem(STsdbReadHandle* pTsdbReadHandle, int32_t capacit
       }
     }
 
+    ASSERT(rowKey != TSKEY_INITIAL_VAL);
+
     if (colId == pColInfo->info.colId) {
       if (tdValTypeIsNorm(sVal.valType)) {
-        colDataAppend(pColInfo, numOfRows, sVal.val, false);
+        colDataAppend(pColInfo, *curRow, sVal.val, false);
-      } else if (forceSetNull) {
+      } else if (tdValTypeIsNull(sVal.valType)) {
-        colDataAppend(pColInfo, numOfRows, NULL, true);
+        colDataAppend(pColInfo, *curRow, NULL, true);
+      } else if (tdValTypeIsNone(sVal.valType)) {
+        // TODO: Set null if nothing append for this row
+        if (mergeOption == 1) {
+          colDataAppend(pColInfo, *curRow, NULL, true);
+        }
+      } else {
+        ASSERT(0);
       }
 
-      i++;
+      ++i;
 
       if (row == row1) {
-        j++;
+        ++j;
       } else {
-        k++;
+        ++k;
       }
     } else {
-      if (forceSetNull) {
+      if (mergeOption == 1) {
-        colDataAppend(pColInfo, numOfRows, NULL, true);
+        colDataAppend(pColInfo, *curRow, NULL, true);
       }
-      i++;
+      ++i;
     }
   }
 
-  if (forceSetNull) {
+  if (mergeOption == 1) {
     while (i < numOfCols) {  // the remain columns are all null data
       SColumnInfoData* pColInfo = taosArrayGet(pTsdbReadHandle->pColumns, i);
-      colDataAppend(pColInfo, numOfRows, NULL, true);
+      colDataAppend(pColInfo, *curRow, NULL, true);
-      i++;
+      ++i;
     }
   }
+
+  return nResult;
 #endif
 }
 
@@ -1859,6 +1969,7 @@ static void doMergeTwoLevelData(STsdbReadHandle* pTsdbReadHandle, STableCheckInf
 
   // compared with the data from in-memory buffer, to generate the correct timestamp array list
   int32_t numOfRows = 0;
+  int32_t curRow = 0;
 
   int16_t   rv1 = -1;
   int16_t   rv2 = -1;
@@ -1874,9 +1985,11 @@ static void doMergeTwoLevelData(STsdbReadHandle* pTsdbReadHandle, STableCheckInf
     return;
   } else if (pCheckInfo->iter != NULL || pCheckInfo->iiter != NULL) {
     SSkipListNode* node = NULL;
+    TSKEY          lastRowKey = TSKEY_INITIAL_VAL;
+
     do {
       STSRow* row2 = NULL;
-      STSRow* row1 = getSRowInTableMem(pCheckInfo, pTsdbReadHandle->order, pCfg->update, &row2);
+      STSRow* row1 = getSRowInTableMem(pCheckInfo, pTsdbReadHandle->order, pCfg->update, &row2, TD_VER_MAX);
       if (row1 == NULL) {
         break;
       }
@@ -1905,9 +2018,9 @@ static void doMergeTwoLevelData(STsdbReadHandle* pTsdbReadHandle, STableCheckInf
           rv2 = TD_ROW_SVER(row2);
         }
 
-        mergeTwoRowFromMem(pTsdbReadHandle, pTsdbReadHandle->outputCapacity, numOfRows, row1, row2, numOfCols,
+        numOfRows += mergeTwoRowFromMem(pTsdbReadHandle, pTsdbReadHandle->outputCapacity, &curRow, row1, row2, numOfCols,
-                           pCheckInfo->tableId, pSchema1, pSchema2, true);
+                           pCheckInfo->tableId, pSchema1, pSchema2, pCfg->update, &lastRowKey);
-        numOfRows += 1;
+        // numOfRows += 1;
         if (cur->win.skey == TSKEY_INITIAL_VAL) {
           cur->win.skey = key;
         }
@@ -1918,6 +2031,7 @@ static void doMergeTwoLevelData(STsdbReadHandle* pTsdbReadHandle, STableCheckInf
 
         moveToNextRowInMem(pCheckInfo);
       } else if (key == tsArray[pos]) {  // data in buffer has the same timestamp of data in file block, ignore it
+#if 0
         if (pCfg->update) {
           if (pCfg->update == TD_ROW_PARTIAL_UPDATE) {
             doCopyRowsFromFileBlock(pTsdbReadHandle, pTsdbReadHandle->outputCapacity, numOfRows, pos, pos);
@@ -1933,7 +2047,7 @@ static void doMergeTwoLevelData(STsdbReadHandle* pTsdbReadHandle, STableCheckInf
 
           bool forceSetNull = pCfg->update != TD_ROW_PARTIAL_UPDATE;
           mergeTwoRowFromMem(pTsdbReadHandle, pTsdbReadHandle->outputCapacity, numOfRows, row1, row2, numOfCols,
-                             pCheckInfo->tableId, pSchema1, pSchema2, forceSetNull);
+                             pCheckInfo->tableId, pSchema1, pSchema2, forceSetNull, &lastRowKey);
           numOfRows += 1;
           if (cur->win.skey == TSKEY_INITIAL_VAL) {
             cur->win.skey = key;
@@ -1943,6 +2057,35 @@ static void doMergeTwoLevelData(STsdbReadHandle* pTsdbReadHandle, STableCheckInf
           cur->lastKey = key + step;
           cur->mixBlock = true;
 
+          moveToNextRowInMem(pCheckInfo);
+          pos += step;
+        } else {
+          moveToNextRowInMem(pCheckInfo);
+        }
+#endif
+        if (TD_SUPPORT_UPDATE(pCfg->update)) {
+          doCopyRowsFromFileBlock(pTsdbReadHandle, pTsdbReadHandle->outputCapacity, curRow, pos, pos);
+
+          if (rv1 != TD_ROW_SVER(row1)) {
+            //            pSchema1 = tsdbGetTableSchemaByVersion(pTable, memRowVersion(row1));
+            rv1 = TD_ROW_SVER(row1);
+          }
+          if (row2 && rv2 != TD_ROW_SVER(row2)) {
+            //            pSchema2 = tsdbGetTableSchemaByVersion(pTable, memRowVersion(row2));
+            rv2 = TD_ROW_SVER(row2);
+          }
+
+          numOfRows += mergeTwoRowFromMem(pTsdbReadHandle, pTsdbReadHandle->outputCapacity, &curRow, row1, row2, numOfCols,
+                             pCheckInfo->tableId, pSchema1, pSchema2, pCfg->update, &lastRowKey);
+          // ++numOfRows;
+          if (cur->win.skey == TSKEY_INITIAL_VAL) {
+            cur->win.skey = key;
+          }
+
+          cur->win.ekey = key;
+          cur->lastKey = key + step;
+          cur->mixBlock = true;
+
           moveToNextRowInMem(pCheckInfo);
           pos += step;
         } else {
@@ -1958,11 +2101,18 @@ static void doMergeTwoLevelData(STsdbReadHandle* pTsdbReadHandle, STableCheckInf
         assert(end != -1);
 
         if (tsArray[end] == key) {  // the value of key in cache equals to the end timestamp value, ignore it
+#if 0
           if (pCfg->update == TD_ROW_DISCARD_UPDATE) {
             moveToNextRowInMem(pCheckInfo);
           } else {
             end -= step;
           }
+#endif
+          if (!TD_SUPPORT_UPDATE(pCfg->update)) {
+            moveToNextRowInMem(pCheckInfo);
+          } else {
+            end -= step;
+          }
         }
 
         int32_t qstart = 0, qend = 0;
@@ -2572,6 +2722,7 @@ static UNUSED_FUNC void changeQueryHandleForInterpQuery(tsdbReaderT pHandle) {
 static int tsdbReadRowsFromCache(STableCheckInfo* pCheckInfo, TSKEY maxKey, int maxRowsToRead, STimeWindow* win,
                                  STsdbReadHandle* pTsdbReadHandle) {
   int       numOfRows = 0;
+  int       curRows = 0;
   int32_t   numOfCols = (int32_t)taosArrayGetSize(pTsdbReadHandle->pColumns);
   STsdbCfg* pCfg = REPO_CFG(pTsdbReadHandle->pTsdb);
   win->skey = TSKEY_INITIAL_VAL;
@@ -2579,9 +2730,11 @@ static int tsdbReadRowsFromCache(STableCheckInfo* pCheckInfo, TSKEY maxKey, int
   int64_t   st = taosGetTimestampUs();
   int16_t   rv = -1;
   STSchema* pSchema = NULL;
+  TSKEY     lastRowKey = TSKEY_INITIAL_VAL;
+
 
   do {
-    STSRow* row = getSRowInTableMem(pCheckInfo, pTsdbReadHandle->order, pCfg->update, NULL);
+    STSRow* row = getSRowInTableMem(pCheckInfo, pTsdbReadHandle->order, pCfg->update, NULL, TD_VER_MAX);
     if (row == NULL) {
       break;
     }
@@ -2604,16 +2757,18 @@ static int tsdbReadRowsFromCache(STableCheckInfo* pCheckInfo, TSKEY maxKey, int
       pSchema = metaGetTbTSchema(REPO_META(pTsdbReadHandle->pTsdb), pCheckInfo->tableId, 0);
       rv = TD_ROW_SVER(row);
     }
-    mergeTwoRowFromMem(pTsdbReadHandle, maxRowsToRead, numOfRows, row, NULL, numOfCols, pCheckInfo->tableId, pSchema,
+    numOfRows += mergeTwoRowFromMem(pTsdbReadHandle, maxRowsToRead, &curRows, row, NULL, numOfCols, pCheckInfo->tableId, pSchema,
-                       NULL, true);
+                       NULL, pCfg->update, &lastRowKey);
 
-    if (++numOfRows >= maxRowsToRead) {
+    if (numOfRows >= maxRowsToRead) {
       moveToNextRowInMem(pCheckInfo);
       break;
     }
 
   } while (moveToNextRowInMem(pCheckInfo));
 
+  taosMemoryFreeClear(pSchema); // free the STSChema
+
   assert(numOfRows <= maxRowsToRead);
 
   // if the buffer is not full in case of descending order query, move the data in the front of the buffer
@@ -2731,6 +2886,8 @@ static bool loadCachedLastRow(STsdbReadHandle* pTsdbReadHandle) {
   STSRow* pRow = NULL;
   TSKEY   key = TSKEY_INITIAL_VAL;
   int32_t step = ASCENDING_TRAVERSE(pTsdbReadHandle->order) ? 1 : -1;
+  TSKEY   lastRowKey = TSKEY_INITIAL_VAL;
+  int32_t curRow = 0;
 
   if (++pTsdbReadHandle->activeIndex < numOfTables) {
     STableCheckInfo* pCheckInfo = taosArrayGet(pTsdbReadHandle->pTableCheckInfo, pTsdbReadHandle->activeIndex);
@@ -2738,8 +2895,8 @@ static bool loadCachedLastRow(STsdbReadHandle* pTsdbReadHandle) {
     //    if (ret != TSDB_CODE_SUCCESS) {
     //      return false;
     //    }
-    mergeTwoRowFromMem(pTsdbReadHandle, pTsdbReadHandle->outputCapacity, 0, pRow, NULL, numOfCols, pCheckInfo->tableId,
+    mergeTwoRowFromMem(pTsdbReadHandle, pTsdbReadHandle->outputCapacity, &curRow, pRow, NULL, numOfCols, pCheckInfo->tableId,
-                       NULL, NULL, true);
+                       NULL, NULL, true, &lastRowKey);
     taosMemoryFreeClear(pRow);
 
     // update the last key value

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

@@ -246,6 +246,12 @@ int tsdbLoadBlockInfo(SReadH *pReadh, void *pTarget) {
   return 0;
 }
 
+static FORCE_INLINE void tsdbSwapDataCols(SDataCols *pDest, SDataCols *pSrc) {
+  SDataCol *pCols = pDest->cols;
+  memcpy(pDest, pSrc, sizeof(SDataCols));
+  pSrc->cols = pCols;
+}
+
 int tsdbLoadBlockData(SReadH *pReadh, SBlock *pBlock, SBlockInfo *pBlkInfo) {
   ASSERT(pBlock->numOfSubBlocks > 0);
   STsdbCfg *pCfg = REPO_CFG(pReadh->pRepo);
@@ -266,17 +272,29 @@ int tsdbLoadBlockData(SReadH *pReadh, SBlock *pBlock, SBlockInfo *pBlkInfo) {
     if (tsdbLoadBlockDataImpl(pReadh, iBlock, pReadh->pDCols[1]) < 0) return -1;
     // TODO: use the real maxVersion to replace the UINT64_MAX to support Multi-Version
     if (tdMergeDataCols(pReadh->pDCols[0], pReadh->pDCols[1], pReadh->pDCols[1]->numOfRows, NULL,
-                        update != TD_ROW_PARTIAL_UPDATE, UINT64_MAX) < 0)
+                        TD_SUPPORT_UPDATE(update), TD_VER_MAX) < 0)
       return -1;
   }
+  // if ((pBlock->numOfSubBlocks == 1) && (iBlock->hasDupKey)) { // TODO: use this line
+  if (pBlock->numOfSubBlocks == 1) {
+    tdResetDataCols(pReadh->pDCols[1]);
+    pReadh->pDCols[1]->bitmapMode = pReadh->pDCols[0]->bitmapMode;
+    if (tdMergeDataCols(pReadh->pDCols[1], pReadh->pDCols[0], pReadh->pDCols[0]->numOfRows, NULL,
+                        TD_SUPPORT_UPDATE(update), TD_VER_MAX) < 0) {
+      return -1;
+    }
+    tsdbSwapDataCols(pReadh->pDCols[0], pReadh->pDCols[1]);
+    ASSERT(pReadh->pDCols[0]->bitmapMode != 0);
+  }
 
-  ASSERT(pReadh->pDCols[0]->numOfRows == pBlock->numOfRows);
+  ASSERT(pReadh->pDCols[0]->numOfRows <= pBlock->numOfRows);
   ASSERT(dataColsKeyFirst(pReadh->pDCols[0]) == pBlock->keyFirst);
   ASSERT(dataColsKeyLast(pReadh->pDCols[0]) == pBlock->keyLast);
 
   return 0;
 }
 
+// TODO: filter by Multi-Version
 int tsdbLoadBlockDataCols(SReadH *pReadh, SBlock *pBlock, SBlockInfo *pBlkInfo, const int16_t *colIds, int numOfColsIds,
                           bool mergeBitmap) {
   ASSERT(pBlock->numOfSubBlocks > 0);
@@ -297,9 +315,21 @@ int tsdbLoadBlockDataCols(SReadH *pReadh, SBlock *pBlock, SBlockInfo *pBlkInfo,
     if (tsdbLoadBlockDataColsImpl(pReadh, iBlock, pReadh->pDCols[1], colIds, numOfColsIds) < 0) return -1;
     // TODO: use the real maxVersion to replace the UINT64_MAX to support Multi-Version
     if (tdMergeDataCols(pReadh->pDCols[0], pReadh->pDCols[1], pReadh->pDCols[1]->numOfRows, NULL,
-                        update != TD_ROW_PARTIAL_UPDATE, UINT64_MAX) < 0)
+                        TD_SUPPORT_UPDATE(update), TD_VER_MAX) < 0)
       return -1;
   }
+  // if ((pBlock->numOfSubBlocks == 1) && (iBlock->hasDupKey)) { // TODO: use this line
+  if (pBlock->numOfSubBlocks == 1) {
+    tdResetDataCols(pReadh->pDCols[1]);
+    pReadh->pDCols[1]->bitmapMode = pReadh->pDCols[0]->bitmapMode;
+    if (tdMergeDataCols(pReadh->pDCols[1], pReadh->pDCols[0], pReadh->pDCols[0]->numOfRows, NULL,
+                        TD_SUPPORT_UPDATE(update), TD_VER_MAX) < 0) {
+      return -1;
+    }
+    tsdbSwapDataCols(pReadh->pDCols[0], pReadh->pDCols[1]);
+    ASSERT(pReadh->pDCols[0]->bitmapMode != 0);
+  }
+
 
   if (mergeBitmap && !tdDataColsIsBitmapI(pReadh->pDCols[0])) {
     for (int i = 0; i < numOfColsIds; ++i) {
@@ -312,7 +342,7 @@ int tsdbLoadBlockDataCols(SReadH *pReadh, SBlock *pBlock, SBlockInfo *pBlkInfo,
     }
   }
 
-  ASSERT(pReadh->pDCols[0]->numOfRows == pBlock->numOfRows);
+  ASSERT(pReadh->pDCols[0]->numOfRows <= pBlock->numOfRows);
   ASSERT(dataColsKeyFirst(pReadh->pDCols[0]) == pBlock->keyFirst);
   ASSERT(dataColsKeyLast(pReadh->pDCols[0]) == pBlock->keyLast);
 
@@ -586,7 +616,7 @@ static int tsdbLoadBlockDataImpl(SReadH *pReadh, SBlock *pBlock, SDataCols *pDat
       tcolId = pBlockCol->colId;
       toffset = tsdbGetBlockColOffset(pBlockCol);
       tlen = pBlockCol->len;
-      pDataCol->bitmap = pBlockCol->bitmap;
+      pDataCol->bitmap = pBlockCol->blen > 0 ? 1 : 0;
     } else {
       ASSERT(pDataCol->colId == tcolId);
       TD_SET_COL_ROWS_NORM(pDataCol);
@@ -594,17 +624,8 @@ static int tsdbLoadBlockDataImpl(SReadH *pReadh, SBlock *pBlock, SDataCols *pDat
 
     // int32_t tBitmaps = 0;
     int32_t tLenBitmap = 0;
-    if ((dcol != 0) && !TD_COL_ROWS_NORM(pBlockCol)) {
+    if ((dcol != 0) && (pBlockCol->blen > 0)) {
       tLenBitmap = nBitmaps;
-#if 0
-      if (IS_VAR_DATA_TYPE(pDataCol->type)) {
-        tBitmaps = nBitmaps;
-        tLenBitmap = tBitmaps;
-      } else {
-        tBitmaps = (int32_t)ceil((double)nBitmaps / TYPE_BYTES[pDataCol->type]);
-        tLenBitmap = tBitmaps * TYPE_BYTES[pDataCol->type];
-      }
-#endif
     }
 
     if (tcolId == pDataCol->colId) {
@@ -623,15 +644,15 @@ static int tsdbLoadBlockDataImpl(SReadH *pReadh, SBlock *pBlock, SDataCols *pDat
       }
 
       if (dcol != 0) {
-        ccol++;
+        ++ccol;
       }
-      dcol++;
+      ++dcol;
     } else if (tcolId < pDataCol->colId) {
-      ccol++;
+      ++ccol;
     } else {
       // Set current column as NULL and forward
       dataColReset(pDataCol);
-      dcol++;
+      ++dcol;
     }
   }
 
@@ -754,8 +775,7 @@ static int tsdbLoadBlockDataColsImpl(SReadH *pReadh, SBlock *pBlock, SDataCols *
 
     if (colId == PRIMARYKEY_TIMESTAMP_COL_ID) {  // load the key row
       blockCol.colId = colId;
-      TD_SET_COL_ROWS_NORM(&blockCol);  // default is NORM for the primary key column
+      blockCol.blen = 0;  // default is NORM for the primary key column
-      blockCol.blen = 0;
       blockCol.len = pBlock->keyLen;
       blockCol.type = pDataCol->type;
       blockCol.offset = TSDB_KEY_COL_OFFSET;
@@ -785,7 +805,7 @@ static int tsdbLoadBlockDataColsImpl(SReadH *pReadh, SBlock *pBlock, SDataCols *
       ASSERT(pBlockCol->colId == pDataCol->colId);
     }
     // set the bitmap
-    pDataCol->bitmap = pBlockCol->bitmap;
+    pDataCol->bitmap = pBlockCol->blen > 0 ? 1 : 0;
 
     if (tsdbLoadColData(pReadh, pDFile, pBlock, pBlockCol, pDataCol) < 0) return -1;
   }
@@ -803,17 +823,8 @@ static int tsdbLoadColData(SReadH *pReadh, SDFile *pDFile, SBlock *pBlock, SBloc
   // int32_t tBitmaps = 0;
   int32_t tLenBitmap = 0;
 
-  if (!TD_COL_ROWS_NORM(pBlockCol)) {
+  if (pBlockCol->blen) {
     tLenBitmap = nBitmaps;
-#if 0
-    if (IS_VAR_DATA_TYPE(pDataCol->type)) {
-      tBitmaps = nBitmaps;
-      tLenBitmap = tBitmaps;
-    } else {
-      tBitmaps = (int32_t)ceil((double)nBitmaps / TYPE_BYTES[pDataCol->type]);
-      tLenBitmap = tBitmaps * TYPE_BYTES[pDataCol->type];
-    }
-#endif
   }
 
   int tsize = pDataCol->bytes * pBlock->numOfRows + tLenBitmap + 2 * COMP_OVERFLOW_BYTES;

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

@@ -1635,17 +1635,20 @@ int32_t tsdbCreateTSma(STsdb *pTsdb, char *pMsg) {
   SSmaCfg vCreateSmaReq = {0};
   if (!tDeserializeSVCreateTSmaReq(pMsg, &vCreateSmaReq)) {
     terrno = TSDB_CODE_OUT_OF_MEMORY;
-    tsdbWarn("vgId:%d TDMT_VND_CREATE_SMA received but deserialize failed since %s", REPO_ID(pTsdb), terrstr(terrno));
+    tsdbWarn("vgId:%d tsma create msg received but deserialize failed since %s", REPO_ID(pTsdb), terrstr(terrno));
     return -1;
   }
-  tsdbDebug("vgId:%d TDMT_VND_CREATE_SMA msg received for %s:%" PRIi64, REPO_ID(pTsdb), vCreateSmaReq.tSma.indexName,
+  
-            vCreateSmaReq.tSma.indexUid);
+  tsdbDebug("vgId:%d tsma create msg %s:%" PRIi64 " for table %" PRIi64 " received", REPO_ID(pTsdb),
+            vCreateSmaReq.tSma.indexName, vCreateSmaReq.tSma.indexUid, vCreateSmaReq.tSma.tableUid);
 
   // record current timezone of server side
   vCreateSmaReq.tSma.timezoneInt = tsTimezone;
 
   if (metaCreateTSma(REPO_META(pTsdb), &vCreateSmaReq) < 0) {
     // TODO: handle error
+    tsdbWarn("vgId:%d tsma %s:%" PRIi64 " create failed for table %" PRIi64 " since %s", REPO_ID(pTsdb),
+             vCreateSmaReq.tSma.indexName, vCreateSmaReq.tSma.indexUid, vCreateSmaReq.tSma.tableUid, terrstr(terrno));
     tdDestroyTSma(&vCreateSmaReq.tSma);
     return -1;
   }

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

@@ -360,7 +360,11 @@ static int vnodeProcessCreateTbReq(SVnode *pVnode, int64_t version, void *pReq,
 
     // do create table
     if (metaCreateTable(pVnode->pMeta, version, pCreateReq) < 0) {
-      cRsp.code = terrno;
+      if (pCreateReq->flags & TD_CREATE_IF_NOT_EXISTS && terrno == TSDB_CODE_TDB_TABLE_ALREADY_EXIST) {
+        cRsp.code = TSDB_CODE_SUCCESS;
+      } else {
+        cRsp.code = terrno;
+      }
     } else {
       cRsp.code = TSDB_CODE_SUCCESS;
       tsdbFetchTbUidList(pVnode->pTsdb, &pStore, pCreateReq->ctb.suid, pCreateReq->uid);
@@ -529,6 +533,13 @@ static int vnodeProcessSubmitReq(SVnode *pVnode, int64_t version, void *pReq, in
         }
       }
 
+      msgIter.uid = createTbReq.uid;
+      if (createTbReq.type == TSDB_CHILD_TABLE) {
+        msgIter.suid = createTbReq.ctb.suid;
+      } else {
+        msgIter.suid = 0;
+      }
+
       tCoderClear(&coder);
     }
 

source/libs/CMakeLists.txt

@@ -17,4 +17,4 @@ add_subdirectory(tfs)
 add_subdirectory(monitor)
 add_subdirectory(nodes)
 add_subdirectory(scalar)
-add_subdirectory(command)
+add_subdirectory(command)

source/libs/executor/src/indexoperator.c

@@ -252,8 +252,8 @@ static int32_t sifExecFunction(SFunctionNode *node, SIFCtx *ctx, SIFParam *outpu
   return TSDB_CODE_QRY_INVALID_INPUT;
 }
 static int32_t sifDoIndex(SIFParam *left, SIFParam *right, int8_t operType, SIFParam *output) {
-  SIndexTerm *tm = indexTermCreate(left->suid, DEFAULT, operType, left->colValType, left->colName,
+  SIndexTerm *tm = indexTermCreate(left->suid, DEFAULT, left->colValType, left->colName, strlen(left->colName),
-                                   strlen(left->colName), right->condValue, strlen(right->condValue));
+                                   right->condValue, strlen(right->condValue));
   if (tm == NULL) {
     return TSDB_CODE_QRY_OUT_OF_MEMORY;
   }

source/libs/executor/src/scanoperator.c

@@ -539,10 +539,12 @@ static SSDataBlock* doStreamBlockScan(SOperatorInfo* pOperator) {
 
     while (tqNextDataBlock(pInfo->readerHandle)) {
       SArray*  pCols = NULL;
-      uint64_t groupId;
+      uint64_t groupId = 0;
-      int32_t  numOfRows;
+      uint64_t uid = 0;
-      int16_t  outputCol;
+      int32_t  numOfRows = 0;
-      int32_t  code = tqRetrieveDataBlock(&pCols, pInfo->readerHandle, &groupId, &numOfRows, &outputCol);
+      int16_t  outputCol = 0;
+
+      int32_t  code = tqRetrieveDataBlock(&pCols, pInfo->readerHandle, &groupId, &uid, &numOfRows, &outputCol);
 
       if (code != TSDB_CODE_SUCCESS || numOfRows == 0) {
         pTaskInfo->code = code;
@@ -551,6 +553,7 @@ static SSDataBlock* doStreamBlockScan(SOperatorInfo* pOperator) {
 
       pInfo->pRes->info.groupId = groupId;
       pInfo->pRes->info.rows = numOfRows;
+      pInfo->pRes->info.uid = uid;
 
       int32_t numOfCols = pInfo->pRes->info.numOfCols;
       for (int32_t i = 0; i < numOfCols; ++i) {
@@ -606,10 +609,8 @@ SOperatorInfo* createStreamScanOperatorInfo(void* streamReadHandle, SSDataBlock*
   SStreamBlockScanInfo* pInfo = taosMemoryCalloc(1, sizeof(SStreamBlockScanInfo));
   SOperatorInfo*        pOperator = taosMemoryCalloc(1, sizeof(SOperatorInfo));
   if (pInfo == NULL || pOperator == NULL) {
-    taosMemoryFreeClear(pInfo);
-    taosMemoryFreeClear(pOperator);
     terrno = TSDB_CODE_QRY_OUT_OF_MEMORY;
-    return NULL;
+    goto _error;
   }
 
   int32_t numOfOutput = taosArrayGetSize(pColList);
@@ -626,16 +627,13 @@ SOperatorInfo* createStreamScanOperatorInfo(void* streamReadHandle, SSDataBlock*
   tqReadHandleSetColIdList((STqReadHandle*)streamReadHandle, pColIds);
   int32_t code = tqReadHandleSetTbUidList(streamReadHandle, pTableIdList);
   if (code != 0) {
-    taosMemoryFreeClear(pInfo);
+    goto _error;
-    taosMemoryFreeClear(pOperator);
-    return NULL;
   }
 
   pInfo->pBlockLists = taosArrayInit(4, POINTER_BYTES);
   if (pInfo->pBlockLists == NULL) {
-    taosMemoryFreeClear(pInfo);
+    terrno = TSDB_CODE_OUT_OF_MEMORY;
-    taosMemoryFreeClear(pOperator);
+    goto _error;
-    return NULL;
   }
 
   pInfo->readerHandle = streamReadHandle;
@@ -647,7 +645,7 @@ SOperatorInfo* createStreamScanOperatorInfo(void* streamReadHandle, SSDataBlock*
   pOperator->blocking        = false;
   pOperator->status          = OP_NOT_OPENED;
   pOperator->info            = pInfo;
-  pOperator->numOfExprs     = pResBlock->info.numOfCols;
+  pOperator->numOfExprs      = pResBlock->info.numOfCols;
   pOperator->fpSet._openFn   = operatorDummyOpenFn;
   pOperator->fpSet.getNextFn = doStreamBlockScan;
   pOperator->fpSet.closeFn   = operatorDummyCloseFn;
@@ -656,6 +654,11 @@ SOperatorInfo* createStreamScanOperatorInfo(void* streamReadHandle, SSDataBlock*
   pOperator->fpSet = createOperatorFpSet(operatorDummyOpenFn, doStreamBlockScan, NULL, NULL, operatorDummyCloseFn, NULL, NULL, NULL);
 
   return pOperator;
+
+  _error:
+  taosMemoryFreeClear(pInfo);
+  taosMemoryFreeClear(pOperator);
+  return NULL;
 }
 
 static void destroySysScanOperator(void* param, int32_t numOfOutput) {

source/libs/executor/src/sortoperator.c

@@ -67,6 +67,10 @@ SSDataBlock* getSortedBlockData(SSortHandle* pHandle, SSDataBlock* pDataBlock, i
   ASSERT(taosArrayGetSize(pColMatchInfo) == pDataBlock->info.numOfCols);
 
   SSDataBlock* p = tsortGetSortedDataBlock(pHandle);
+  if (p == NULL) {
+    return NULL;
+  }
+
   blockDataEnsureCapacity(p, capacity);
 
   while (1) {

source/libs/executor/src/tsort.c

@@ -553,7 +553,7 @@ static int32_t createInitialSortedMultiSources(SSortHandle* pHandle) {
       }
     }
 
-    if (pHandle->pDataBlock->info.rows > 0) {
+    if (pHandle->pDataBlock != NULL && pHandle->pDataBlock->info.rows > 0) {
       size_t size = blockDataGetSize(pHandle->pDataBlock);
 
       // Perform the in-memory sort and then flush data in the buffer into disk.

source/libs/function/inc/builtinsimpl.h

@@ -73,6 +73,11 @@ bool spreadFunctionSetup(SqlFunctionCtx *pCtx, SResultRowEntryInfo* pResultInfo)
 int32_t spreadFunction(SqlFunctionCtx* pCtx);
 int32_t spreadFinalize(SqlFunctionCtx* pCtx, SSDataBlock* pBlock);
 
+bool getHistogramFuncEnv(struct SFunctionNode* pFunc, SFuncExecEnv* pEnv);
+bool histogramFunctionSetup(SqlFunctionCtx *pCtx, SResultRowEntryInfo* pResultInfo);
+int32_t histogramFunction(SqlFunctionCtx* pCtx);
+int32_t histogramFinalize(SqlFunctionCtx* pCtx, SSDataBlock* pBlock);
+
 #ifdef __cplusplus
 }
 #endif

source/libs/function/src/builtins.c

@@ -225,6 +225,26 @@ static int32_t translateSpread(SFunctionNode* pFunc, char* pErrBuf, int32_t len)
   return TSDB_CODE_SUCCESS;
 }
 
+static int32_t translateHistogram(SFunctionNode* pFunc, char* pErrBuf, int32_t len) {
+  if (4 != LIST_LENGTH(pFunc->pParameterList)) {
+    return invaildFuncParaNumErrMsg(pErrBuf, len, pFunc->functionName);
+  }
+
+  uint8_t colType = ((SExprNode*)nodesListGetNode(pFunc->pParameterList, 0))->resType.type;
+  if (!IS_NUMERIC_TYPE(colType)) {
+    return invaildFuncParaTypeErrMsg(pErrBuf, len, pFunc->functionName);
+  }
+
+  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);
+  }
+
+  pFunc->node.resType = (SDataType) { .bytes = tDataTypes[TSDB_DATA_TYPE_DOUBLE].bytes, .type = TSDB_DATA_TYPE_DOUBLE };
+  return TSDB_CODE_SUCCESS;
+}
+
 static int32_t translateLastRow(SFunctionNode* pFunc, char* pErrBuf, int32_t len) {
   // todo
   return TSDB_CODE_SUCCESS;
@@ -599,6 +619,16 @@ const SBuiltinFuncDefinition funcMgtBuiltins[] = {
     .processFunc  = diffFunction,
     .finalizeFunc = functionFinalize
   },
+  {
+    .name = "histogram",
+    .type = FUNCTION_TYPE_HISTOGRAM,
+    .classification = FUNC_MGT_AGG_FUNC,
+    .translateFunc = translateHistogram,
+    .getEnvFunc   = getHistogramFuncEnv,
+    .initFunc     = histogramFunctionSetup,
+    .processFunc  = histogramFunction,
+    .finalizeFunc = histogramFinalize
+  },
   {
     .name = "abs",
     .type = FUNCTION_TYPE_ABS,

source/libs/function/src/builtinsimpl.c

@@ -20,6 +20,8 @@
 #include "tdatablock.h"
 #include "tpercentile.h"
 
+#define HISTOGRAM_MAX_BINS_NUM 100
+
 typedef struct SSumRes {
   union {
     int64_t  isum;
@@ -89,6 +91,22 @@ typedef struct SSpreadInfo {
   double max;
 } SSpreadInfo;
 
+typedef struct SHistoFuncBin {
+  double lower;
+  double upper;
+  union {
+    int64_t count;
+    double  percentage;
+  };
+} SHistoFuncBin;
+
+typedef struct SHistoFuncInfo {
+  int32_t numOfBins;
+  bool    normalized;
+  SHistoFuncBin bins[];
+} SHistoFuncInfo;
+
+
 #define SET_VAL(_info, numOfElem, res) \
   do {                                 \
     if ((numOfElem) <= 0) {            \
@@ -1777,3 +1795,34 @@ int32_t spreadFinalize(SqlFunctionCtx* pCtx, SSDataBlock* pBlock) {
   }
   return functionFinalize(pCtx, pBlock);
 }
+
+bool getHistogramFuncEnv(SFunctionNode* UNUSED_PARAM(pFunc), SFuncExecEnv* pEnv) {
+  pEnv->calcMemSize = sizeof(SHistoFuncInfo) + HISTOGRAM_MAX_BINS_NUM * sizeof(SHistoFuncBin);
+  return true;
+}
+
+bool histogramFunctionSetup(SqlFunctionCtx *pCtx, SResultRowEntryInfo* pResultInfo) {
+  if (!functionSetup(pCtx, pResultInfo)) {
+    return false;
+  }
+
+  SHistoFuncInfo* pInfo = GET_ROWCELL_INTERBUF(pResultInfo);
+  char* binType = pCtx->param[1].param.pz;
+  char* binDesc = pCtx->param[2].param.pz;
+  int64_t nornalized = pCtx->param[3].param.i;
+
+
+  return true;
+}
+
+int32_t histogramFunction(SqlFunctionCtx *pCtx) {
+  return TSDB_CODE_SUCCESS;
+}
+
+int32_t histogramFinalize(SqlFunctionCtx* pCtx, SSDataBlock* pBlock) {
+  SHistoFuncInfo* pInfo = GET_ROWCELL_INTERBUF(GET_RES_INFO(pCtx));
+  //if (pInfo->hasResult == true) {
+  //  SET_DOUBLE_VAL(&pInfo->result, pInfo->max - pInfo->min);
+  //}
+  return functionFinalize(pCtx, pBlock);
+}

source/libs/index/inc/indexCache.h

@@ -31,6 +31,7 @@ extern "C" {
 typedef struct MemTable {
   T_REF_DECLARE()
   SSkipList* mem;
+  void*      pCache;
 } MemTable;
 typedef struct IndexCache {
   T_REF_DECLARE()
@@ -47,6 +48,7 @@ typedef struct IndexCache {
 } IndexCache;
 
 #define CACHE_VERSION(cache) atomic_load_32(&cache->version)
+
 typedef struct CacheTerm {
   // key
   char*   colVal;

source/libs/index/inc/indexComm.h

@@ -24,7 +24,7 @@ extern char JSON_COLUMN[];
 extern char JSON_VALUE_DELIM;
 
 char* indexPackJsonData(SIndexTerm* itm);
-
+char* indexPackJsonDataPrefix(SIndexTerm* itm, int32_t* skip);
 #ifdef __cplusplus
 }
 #endif

source/libs/index/inc/indexInt.h

@@ -166,7 +166,9 @@ int32_t indexSerialCacheKey(ICacheKey* key, char* buf);
   } while (0)
 
 #define INDEX_TYPE_CONTAIN_EXTERN_TYPE(ty, exTy) (((ty >> 4) & (exTy)) != 0)
-#define INDEX_TYPE_GET_TYPE(ty)                  (ty & 0x0F)
+
+#define INDEX_TYPE_GET_TYPE(ty) (ty & 0x0F)
+
 #define INDEX_TYPE_ADD_EXTERN_TYPE(ty, exTy) \
   do {                                       \
     uint8_t oldTy = ty;                      \

source/libs/index/src/index.c

@@ -244,8 +244,8 @@ int indexMultiTermQueryAdd(SIndexMultiTermQuery* pQuery, SIndexTerm* term, EInde
   return 0;
 }
 
-SIndexTerm* indexTermCreate(int64_t suid, SIndexOperOnColumn oper, int8_t queryType, uint8_t colType,
+SIndexTerm* indexTermCreate(int64_t suid, SIndexOperOnColumn oper, uint8_t colType, const char* colName,
-                            const char* colName, int32_t nColName, const char* colVal, int32_t nColVal) {
+                            int32_t nColName, const char* colVal, int32_t nColVal) {
   SIndexTerm* tm = (SIndexTerm*)taosMemoryCalloc(1, (sizeof(SIndexTerm)));
   if (tm == NULL) {
     return NULL;

source/libs/index/src/indexCache.c

@@ -34,44 +34,96 @@ static char*   indexCacheTermGet(const void* pData);
 
 static MemTable* indexInternalCacheCreate(int8_t type);
 
-static int32_t cacheSearchTerm(void* cache, CacheTerm* ct, SIdxTempResult* tr, STermValueType* s);
+static int32_t cacheSearchTerm(void* cache, SIndexTerm* ct, SIdxTempResult* tr, STermValueType* s);
-static int32_t cacheSearchPrefix(void* cache, CacheTerm* ct, SIdxTempResult* tr, STermValueType* s);
+static int32_t cacheSearchPrefix(void* cache, SIndexTerm* ct, SIdxTempResult* tr, STermValueType* s);
-static int32_t cacheSearchSuffix(void* cache, CacheTerm* ct, SIdxTempResult* tr, STermValueType* s);
+static int32_t cacheSearchSuffix(void* cache, SIndexTerm* ct, SIdxTempResult* tr, STermValueType* s);
-static int32_t cacheSearchRegex(void* cache, CacheTerm* ct, SIdxTempResult* tr, STermValueType* s);
+static int32_t cacheSearchRegex(void* cache, SIndexTerm* ct, SIdxTempResult* tr, STermValueType* s);
-static int32_t cacheSearchLessThan(void* cache, CacheTerm* ct, SIdxTempResult* tr, STermValueType* s);
+static int32_t cacheSearchLessThan(void* cache, SIndexTerm* ct, SIdxTempResult* tr, STermValueType* s);
-static int32_t cacheSearchLessEqual(void* cache, CacheTerm* ct, SIdxTempResult* tr, STermValueType* s);
+static int32_t cacheSearchLessEqual(void* cache, SIndexTerm* ct, SIdxTempResult* tr, STermValueType* s);
-static int32_t cacheSearchGreaterThan(void* cache, CacheTerm* ct, SIdxTempResult* tr, STermValueType* s);
+static int32_t cacheSearchGreaterThan(void* cache, SIndexTerm* ct, SIdxTempResult* tr, STermValueType* s);
-static int32_t cacheSearchGreaterEqual(void* cache, CacheTerm* ct, SIdxTempResult* tr, STermValueType* s);
+static int32_t cacheSearchGreaterEqual(void* cache, SIndexTerm* ct, SIdxTempResult* tr, STermValueType* s);
-static int32_t cacheSearchRange(void* cache, CacheTerm* ct, SIdxTempResult* tr, STermValueType* s);
+static int32_t cacheSearchRange(void* cache, SIndexTerm* ct, SIdxTempResult* tr, STermValueType* s);
 /*comm func of compare, used in (LE/LT/GE/GT compare)*/
-static int32_t cacheSearchCompareFunc(void* cache, CacheTerm* ct, SIdxTempResult* tr, STermValueType* s,
+static int32_t cacheSearchCompareFunc(void* cache, SIndexTerm* ct, SIdxTempResult* tr, STermValueType* s,
                                       RangeType type);
+static int32_t cacheSearchTerm_JSON(void* cache, SIndexTerm* ct, SIdxTempResult* tr, STermValueType* s);
+static int32_t cacheSearchPrefix_JSON(void* cache, SIndexTerm* ct, SIdxTempResult* tr, STermValueType* s);
+static int32_t cacheSearchSuffix_JSON(void* cache, SIndexTerm* ct, SIdxTempResult* tr, STermValueType* s);
+static int32_t cacheSearchRegex_JSON(void* cache, SIndexTerm* ct, SIdxTempResult* tr, STermValueType* s);
+static int32_t cacheSearchLessThan_JSON(void* cache, SIndexTerm* ct, SIdxTempResult* tr, STermValueType* s);
+static int32_t cacheSearchLessEqual_JSON(void* cache, SIndexTerm* ct, SIdxTempResult* tr, STermValueType* s);
+static int32_t cacheSearchGreaterThan_JSON(void* cache, SIndexTerm* ct, SIdxTempResult* tr, STermValueType* s);
+static int32_t cacheSearchGreaterEqual_JSON(void* cache, SIndexTerm* ct, SIdxTempResult* tr, STermValueType* s);
+static int32_t cacheSearchRange_JSON(void* cache, SIndexTerm* ct, SIdxTempResult* tr, STermValueType* s);
+
+static int32_t cacheSearchCompareFunc_JSON(void* cache, SIndexTerm* term, SIdxTempResult* tr, STermValueType* s,
+                                           RangeType type);
 
 typedef enum { MATCH, CONTINUE, BREAK } TExeCond;
 typedef TExeCond (*_cache_range_compare)(void* a, void* b, int8_t type);
 
-static TExeCond tCompareLessThan(void* a, void* b, int8_t type) { return MATCH; }
+static TExeCond tDoCommpare(__compar_fn_t func, int8_t comType, void* a, void* b) {
-static TExeCond tCompareLessEqual(void* a, void* b, int8_t type) { return MATCH; }
+  int32_t ret = func(a, b);
-static TExeCond tCompareGreaterThan(void* a, void* b, int8_t type) { return MATCH; }
+  switch (comType) {
-static TExeCond tCompareGreaterEqual(void* a, void* b, int8_t type) { return MATCH; }
+    case QUERY_LESS_THAN: {
+      if (ret < 0) return MATCH;
+    } break;
+    case QUERY_LESS_EQUAL: {
+      if (ret <= 0) return MATCH;
+      break;
+    }
+    case QUERY_GREATER_THAN: {
+      if (ret > 0) return MATCH;
+      break;
+    }
+    case QUERY_GREATER_EQUAL: {
+      if (ret >= 0) return MATCH;
+    }
+  }
+  return CONTINUE;
+}
+static TExeCond tCompareLessThan(void* a, void* b, int8_t type) {
+  __compar_fn_t func = getComparFunc(type, 0);
+  return tDoCommpare(func, QUERY_LESS_THAN, a, b);
+}
+static TExeCond tCompareLessEqual(void* a, void* b, int8_t type) {
+  __compar_fn_t func = getComparFunc(type, 0);
+  return tDoCommpare(func, QUERY_LESS_EQUAL, a, b);
+}
+static TExeCond tCompareGreaterThan(void* a, void* b, int8_t type) {
+  __compar_fn_t func = getComparFunc(type, 0);
+  return tDoCommpare(func, QUERY_GREATER_THAN, a, b);
+}
+static TExeCond tCompareGreaterEqual(void* a, void* b, int8_t type) {
+  __compar_fn_t func = getComparFunc(type, 0);
+  return tDoCommpare(func, QUERY_GREATER_EQUAL, a, b);
+}
 
 static TExeCond (*rangeCompare[])(void* a, void* b, int8_t type) = {tCompareLessThan, tCompareLessEqual,
                                                                     tCompareGreaterThan, tCompareGreaterEqual};
 
-static int32_t (*cacheSearch[])(void* cache, CacheTerm* ct, SIdxTempResult* tr, STermValueType* s) = {
+static int32_t (*cacheSearch[][QUERY_MAX])(void* cache, SIndexTerm* ct, SIdxTempResult* tr, STermValueType* s) = {
-    cacheSearchTerm,      cacheSearchPrefix,      cacheSearchSuffix,       cacheSearchRegex, cacheSearchLessThan,
+    {cacheSearchTerm, cacheSearchPrefix, cacheSearchSuffix, cacheSearchRegex, cacheSearchLessThan, cacheSearchLessEqual,
-    cacheSearchLessEqual, cacheSearchGreaterThan, cacheSearchGreaterEqual, cacheSearchRange};
+     cacheSearchGreaterThan, cacheSearchGreaterEqual, cacheSearchRange},
+    {cacheSearchTerm_JSON, cacheSearchPrefix_JSON, cacheSearchSuffix_JSON, cacheSearchRegex_JSON,
+     cacheSearchLessThan_JSON, cacheSearchLessEqual_JSON, cacheSearchGreaterThan_JSON, cacheSearchGreaterEqual_JSON,
+     cacheSearchRange_JSON}};
 
 static void doMergeWork(SSchedMsg* msg);
 static bool indexCacheIteratorNext(Iterate* itera);
 
-static int32_t cacheSearchTerm(void* cache, CacheTerm* ct, SIdxTempResult* tr, STermValueType* s) {
+static int32_t cacheSearchTerm(void* cache, SIndexTerm* term, SIdxTempResult* tr, STermValueType* s) {
   if (cache == NULL) {
     return 0;
   }
+  MemTable*   mem = cache;
+  IndexCache* pCache = mem->pCache;
 
-  MemTable* mem = cache;
+  CacheTerm* pCt = taosMemoryCalloc(1, sizeof(CacheTerm));
-  char*     key = indexCacheTermGet(ct);
+  pCt->colVal = term->colVal;
+  pCt->version = atomic_load_32(&pCache->version);
+
+  char* key = indexCacheTermGet(pCt);
 
   SSkipListIterator* iter = tSkipListCreateIterFromVal(mem->mem, key, TSDB_DATA_TYPE_BINARY, TSDB_ORDER_ASC);
   while (tSkipListIterNext(iter)) {
@@ -80,7 +132,7 @@ static int32_t cacheSearchTerm(void* cache, CacheTerm* ct, SIdxTempResult* tr, S
       break;
     }
     CacheTerm* c = (CacheTerm*)SL_GET_NODE_DATA(node);
-    if (0 == strcmp(c->colVal, ct->colVal)) {
+    if (0 == strcmp(c->colVal, pCt->colVal)) {
       if (c->operaType == ADD_VALUE) {
         INDEX_MERGE_ADD_DEL(tr->deled, tr->added, c->uid)
         // taosArrayPush(result, &c->uid);
@@ -92,30 +144,39 @@ static int32_t cacheSearchTerm(void* cache, CacheTerm* ct, SIdxTempResult* tr, S
       break;
     }
   }
+
+  taosMemoryFree(pCt);
   tSkipListDestroyIter(iter);
   return 0;
 }
-static int32_t cacheSearchPrefix(void* cache, CacheTerm* ct, SIdxTempResult* tr, STermValueType* s) {
+static int32_t cacheSearchPrefix(void* cache, SIndexTerm* term, SIdxTempResult* tr, STermValueType* s) {
   // impl later
   return 0;
 }
-static int32_t cacheSearchSuffix(void* cache, CacheTerm* ct, SIdxTempResult* tr, STermValueType* s) {
+static int32_t cacheSearchSuffix(void* cache, SIndexTerm* term, SIdxTempResult* tr, STermValueType* s) {
   // impl later
   return 0;
 }
-static int32_t cacheSearchRegex(void* cache, CacheTerm* ct, SIdxTempResult* tr, STermValueType* s) {
+static int32_t cacheSearchRegex(void* cache, SIndexTerm* term, SIdxTempResult* tr, STermValueType* s) {
   // impl later
   return 0;
 }
-static int32_t cacheSearchCompareFunc(void* cache, CacheTerm* ct, SIdxTempResult* tr, STermValueType* s,
+static int32_t cacheSearchCompareFunc(void* cache, SIndexTerm* term, SIdxTempResult* tr, STermValueType* s,
                                       RangeType type) {
   if (cache == NULL) {
     return 0;
   }
+
   _cache_range_compare cmpFn = rangeCompare[type];
 
-  MemTable* mem = cache;
+  MemTable*   mem = cache;
-  char*     key = indexCacheTermGet(ct);
+  IndexCache* pCache = mem->pCache;
+
+  CacheTerm* pCt = taosMemoryCalloc(1, sizeof(CacheTerm));
+  pCt->colVal = term->colVal;
+  pCt->version = atomic_load_32(&pCache->version);
+
+  char* key = indexCacheTermGet(pCt);
 
   SSkipListIterator* iter = tSkipListCreateIter(mem->mem);
   while (tSkipListIterNext(iter)) {
@@ -124,7 +185,7 @@ static int32_t cacheSearchCompareFunc(void* cache, CacheTerm* ct, SIdxTempResult
       break;
     }
     CacheTerm* c = (CacheTerm*)SL_GET_NODE_DATA(node);
-    TExeCond   cond = cmpFn(c->colVal, ct->colVal, ct->colType);
+    TExeCond   cond = cmpFn(c->colVal, pCt->colVal, pCt->colType);
     if (cond == MATCH) {
       if (c->operaType == ADD_VALUE) {
         INDEX_MERGE_ADD_DEL(tr->deled, tr->added, c->uid)
@@ -134,26 +195,153 @@ static int32_t cacheSearchCompareFunc(void* cache, CacheTerm* ct, SIdxTempResult
         INDEX_MERGE_ADD_DEL(tr->added, tr->deled, c->uid)
       }
     } else if (cond == CONTINUE) {
+      continue;
     } else if (cond == BREAK) {
       break;
     }
   }
+  taosMemoryFree(pCt);
   tSkipListDestroyIter(iter);
   return TSDB_CODE_SUCCESS;
 }
-static int32_t cacheSearchLessThan(void* cache, CacheTerm* ct, SIdxTempResult* tr, STermValueType* s) {
+static int32_t cacheSearchLessThan(void* cache, SIndexTerm* term, SIdxTempResult* tr, STermValueType* s) {
-  return cacheSearchCompareFunc(cache, ct, tr, s, LT);
+  return cacheSearchCompareFunc(cache, term, tr, s, LT);
 }
-static int32_t cacheSearchLessEqual(void* cache, CacheTerm* ct, SIdxTempResult* tr, STermValueType* s) {
+static int32_t cacheSearchLessEqual(void* cache, SIndexTerm* term, SIdxTempResult* tr, STermValueType* s) {
-  return cacheSearchCompareFunc(cache, ct, tr, s, LE);
+  return cacheSearchCompareFunc(cache, term, tr, s, LE);
 }
-static int32_t cacheSearchGreaterThan(void* cache, CacheTerm* ct, SIdxTempResult* tr, STermValueType* s) {
+static int32_t cacheSearchGreaterThan(void* cache, SIndexTerm* term, SIdxTempResult* tr, STermValueType* s) {
-  return cacheSearchCompareFunc(cache, ct, tr, s, GT);
+  return cacheSearchCompareFunc(cache, term, tr, s, GT);
 }
-static int32_t cacheSearchGreaterEqual(void* cache, CacheTerm* ct, SIdxTempResult* tr, STermValueType* s) {
+static int32_t cacheSearchGreaterEqual(void* cache, SIndexTerm* term, SIdxTempResult* tr, STermValueType* s) {
-  return cacheSearchCompareFunc(cache, ct, tr, s, GE);
+  return cacheSearchCompareFunc(cache, term, tr, s, GE);
 }
-static int32_t cacheSearchRange(void* cache, CacheTerm* ct, SIdxTempResult* tr, STermValueType* s) {
+
+static int32_t cacheSearchTerm_JSON(void* cache, SIndexTerm* term, SIdxTempResult* tr, STermValueType* s) {
+  if (cache == NULL) {
+    return 0;
+  }
+  MemTable*   mem = cache;
+  IndexCache* pCache = mem->pCache;
+
+  CacheTerm* pCt = taosMemoryCalloc(1, sizeof(CacheTerm));
+  pCt->colVal = term->colVal;
+  pCt->version = atomic_load_32(&pCache->version);
+
+  char* exBuf = NULL;
+  if (INDEX_TYPE_CONTAIN_EXTERN_TYPE(term->colType, TSDB_DATA_TYPE_JSON)) {
+    exBuf = indexPackJsonData(term);
+    pCt->colVal = exBuf;
+  }
+  char* key = indexCacheTermGet(pCt);
+
+  SSkipListIterator* iter = tSkipListCreateIterFromVal(mem->mem, key, TSDB_DATA_TYPE_BINARY, TSDB_ORDER_ASC);
+  while (tSkipListIterNext(iter)) {
+    SSkipListNode* node = tSkipListIterGet(iter);
+    if (node == NULL) {
+      break;
+    }
+    CacheTerm* c = (CacheTerm*)SL_GET_NODE_DATA(node);
+    if (0 == strcmp(c->colVal, pCt->colVal)) {
+      if (c->operaType == ADD_VALUE) {
+        INDEX_MERGE_ADD_DEL(tr->deled, tr->added, c->uid)
+        // taosArrayPush(result, &c->uid);
+        *s = kTypeValue;
+      } else if (c->operaType == DEL_VALUE) {
+        INDEX_MERGE_ADD_DEL(tr->added, tr->deled, c->uid)
+      }
+    } else {
+      break;
+    }
+  }
+
+  taosMemoryFree(pCt);
+  taosMemoryFree(exBuf);
+  tSkipListDestroyIter(iter);
+  return 0;
+
+  return TSDB_CODE_SUCCESS;
+}
+static int32_t cacheSearchPrefix_JSON(void* cache, SIndexTerm* term, SIdxTempResult* tr, STermValueType* s) {
+  return TSDB_CODE_SUCCESS;
+}
+static int32_t cacheSearchSuffix_JSON(void* cache, SIndexTerm* term, SIdxTempResult* tr, STermValueType* s) {
+  return TSDB_CODE_SUCCESS;
+}
+static int32_t cacheSearchRegex_JSON(void* cache, SIndexTerm* term, SIdxTempResult* tr, STermValueType* s) {
+  return TSDB_CODE_SUCCESS;
+}
+static int32_t cacheSearchLessThan_JSON(void* cache, SIndexTerm* term, SIdxTempResult* tr, STermValueType* s) {
+  return cacheSearchCompareFunc_JSON(cache, term, tr, s, LT);
+}
+static int32_t cacheSearchLessEqual_JSON(void* cache, SIndexTerm* term, SIdxTempResult* tr, STermValueType* s) {
+  return cacheSearchCompareFunc_JSON(cache, term, tr, s, LE);
+}
+static int32_t cacheSearchGreaterThan_JSON(void* cache, SIndexTerm* term, SIdxTempResult* tr, STermValueType* s) {
+  return cacheSearchCompareFunc_JSON(cache, term, tr, s, GT);
+}
+static int32_t cacheSearchGreaterEqual_JSON(void* cache, SIndexTerm* term, SIdxTempResult* tr, STermValueType* s) {
+  return cacheSearchCompareFunc_JSON(cache, term, tr, s, GE);
+}
+static int32_t cacheSearchRange_JSON(void* cache, SIndexTerm* term, SIdxTempResult* tr, STermValueType* s) {
+  return TSDB_CODE_SUCCESS;
+}
+
+static int32_t cacheSearchCompareFunc_JSON(void* cache, SIndexTerm* term, SIdxTempResult* tr, STermValueType* s,
+                                           RangeType type) {
+  if (cache == NULL) {
+    return 0;
+  }
+  _cache_range_compare cmpFn = rangeCompare[type];
+
+  MemTable*   mem = cache;
+  IndexCache* pCache = mem->pCache;
+
+  CacheTerm* pCt = taosMemoryCalloc(1, sizeof(CacheTerm));
+  pCt->colVal = term->colVal;
+  pCt->version = atomic_load_32(&pCache->version);
+
+  int8_t dType = INDEX_TYPE_GET_TYPE(term->colType);
+  int    skip = 0;
+  char*  exBuf = NULL;
+
+  if (INDEX_TYPE_CONTAIN_EXTERN_TYPE(term->colType, TSDB_DATA_TYPE_JSON)) {
+    exBuf = indexPackJsonDataPrefix(term, &skip);
+    pCt->colVal = exBuf;
+  }
+  char* key = indexCacheTermGet(pCt);
+
+  SSkipListIterator* iter = tSkipListCreateIterFromVal(mem->mem, key, TSDB_DATA_TYPE_BINARY, TSDB_ORDER_ASC);
+  while (tSkipListIterNext(iter)) {
+    SSkipListNode* node = tSkipListIterGet(iter);
+    if (node == NULL) {
+      break;
+    }
+    CacheTerm* c = (CacheTerm*)SL_GET_NODE_DATA(node);
+
+    TExeCond cond = cmpFn(c->colVal + skip, term->colVal, dType);
+    if (cond == MATCH) {
+      if (c->operaType == ADD_VALUE) {
+        INDEX_MERGE_ADD_DEL(tr->deled, tr->added, c->uid)
+        // taosArrayPush(result, &c->uid);
+        *s = kTypeValue;
+      } else if (c->operaType == DEL_VALUE) {
+        INDEX_MERGE_ADD_DEL(tr->added, tr->deled, c->uid)
+      }
+    } else if (cond == CONTINUE) {
+      continue;
+    } else if (cond == BREAK) {
+      break;
+    }
+  }
+
+  taosMemoryFree(pCt);
+  taosMemoryFree(exBuf);
+  tSkipListDestroyIter(iter);
+
+  return TSDB_CODE_SUCCESS;
+}
+static int32_t cacheSearchRange(void* cache, SIndexTerm* term, SIdxTempResult* tr, STermValueType* s) {
   // impl later
   return 0;
 }
@@ -167,6 +355,7 @@ IndexCache* indexCacheCreate(SIndex* idx, uint64_t suid, const char* colName, in
   };
 
   cache->mem = indexInternalCacheCreate(type);
+  cache->mem->pCache = cache;
   cache->colName = INDEX_TYPE_CONTAIN_EXTERN_TYPE(type, TSDB_DATA_TYPE_JSON) ? tstrdup(JSON_COLUMN) : tstrdup(colName);
   cache->type = type;
   cache->index = idx;
@@ -325,6 +514,7 @@ static void indexCacheMakeRoomForWrite(IndexCache* cache) {
       indexCacheRef(cache);
       cache->imm = cache->mem;
       cache->mem = indexInternalCacheCreate(cache->type);
+      cache->mem->pCache = cache;
       cache->occupiedMem = 0;
       // sched to merge
       // unref cache in bgwork
@@ -379,11 +569,19 @@ int indexCacheDel(void* cache, const char* fieldValue, int32_t fvlen, uint64_t u
   return 0;
 }
 
-static int indexQueryMem(MemTable* mem, CacheTerm* ct, EIndexQueryType qtype, SIdxTempResult* tr, STermValueType* s) {
+static int32_t indexQueryMem(MemTable* mem, SIndexTermQuery* query, SIdxTempResult* tr, STermValueType* s) {
   if (mem == NULL) {
     return 0;
   }
-  return cacheSearch[qtype](mem, ct, tr, s);
+
+  SIndexTerm*     term = query->term;
+  EIndexQueryType qtype = query->qType;
+
+  if (INDEX_TYPE_CONTAIN_EXTERN_TYPE(term->colType, TSDB_DATA_TYPE_JSON)) {
+    return cacheSearch[1][qtype](mem, term, tr, s);
+  } else {
+    return cacheSearch[0][qtype](mem, term, tr, s);
+  }
 }
 int indexCacheSearch(void* cache, SIndexTermQuery* query, SIdxTempResult* result, STermValueType* s) {
   int64_t st = taosGetTimestampUs();
@@ -400,25 +598,24 @@ int indexCacheSearch(void* cache, SIndexTermQuery* query, SIdxTempResult* result
   indexMemRef(imm);
   taosThreadMutexUnlock(&pCache->mtx);
 
-  SIndexTerm*     term = query->term;
+  // SIndexTerm*     term = query->term;
-  EIndexQueryType qtype = query->qType;
+  // EIndexQueryType qtype = query->qType;
 
-  bool  hasJson = INDEX_TYPE_CONTAIN_EXTERN_TYPE(term->colType, TSDB_DATA_TYPE_JSON);
+  // bool  isJson = INDEX_TYPE_CONTAIN_EXTERN_TYPE(term->colType, TSDB_DATA_TYPE_JSON);
-  char* p = term->colVal;
+  // char* p = term->colVal;
-  if (hasJson) {
+  // if (isJson) {
-    p = indexPackJsonData(term);
+  //  p = indexPackJsonData(term);
-  }
+  //}
-  CacheTerm ct = {.colVal = p, .version = atomic_load_32(&pCache->version)};
+  // CacheTerm ct = {.colVal = p, .version = atomic_load_32(&pCache->version)};
 
-  int ret = indexQueryMem(mem, &ct, qtype, result, s);
+  int ret = indexQueryMem(mem, query, result, s);
   if (ret == 0 && *s != kTypeDeletion) {
     // continue search in imm
-    ret = indexQueryMem(imm, &ct, qtype, result, s);
+    ret = indexQueryMem(imm, query, result, s);
-  }
-
-  if (hasJson) {
-    taosMemoryFreeClear(p);
   }
+  // if (isJson) {
+  //  taosMemoryFreeClear(p);
+  //}
 
   indexMemUnRef(mem);
   indexMemUnRef(imm);

source/libs/index/src/indexComm.c

@@ -46,3 +46,30 @@ char* indexPackJsonData(SIndexTerm* itm) {
 
   return buf;
 }
+char* indexPackJsonDataPrefix(SIndexTerm* itm, int32_t* skip) {
+  /*
+   * |<-----colname---->|<-----dataType---->|<--------colVal---------->|
+   * |<-----string----->|<-----uint8_t----->|<----depend on dataType-->|
+   */
+  uint8_t ty = INDEX_TYPE_GET_TYPE(itm->colType);
+
+  int32_t sz = itm->nColName + itm->nColVal + sizeof(uint8_t) + sizeof(JSON_VALUE_DELIM) * 2 + 1;
+  char*   buf = (char*)taosMemoryCalloc(1, sz);
+  char*   p = buf;
+
+  memcpy(p, itm->colName, itm->nColName);
+  p += itm->nColName;
+
+  memcpy(p, &JSON_VALUE_DELIM, sizeof(JSON_VALUE_DELIM));
+  p += sizeof(JSON_VALUE_DELIM);
+
+  memcpy(p, &ty, sizeof(ty));
+  p += sizeof(ty);
+
+  memcpy(p, &JSON_VALUE_DELIM, sizeof(JSON_VALUE_DELIM));
+  p += sizeof(JSON_VALUE_DELIM);
+
+  *skip = p - buf;
+
+  return buf;
+}

source/libs/index/src/indexTfile.c

@@ -308,20 +308,20 @@ static int32_t tfSearchRegex(void* reader, SIndexTerm* tem, SIdxTempResult* tr)
 }
 
 static int32_t tfSearchCompareFunc(void* reader, SIndexTerm* tem, SIdxTempResult* tr, RangeType type) {
-  bool     hasJson = INDEX_TYPE_CONTAIN_EXTERN_TYPE(tem->colType, TSDB_DATA_TYPE_JSON);
+  bool  hasJson = INDEX_TYPE_CONTAIN_EXTERN_TYPE(tem->colType, TSDB_DATA_TYPE_JSON);
-  int      ret = 0;
+  int   ret = 0;
-  char*    p = tem->colVal;
+  char* p = tem->colVal;
-  uint64_t sz = tem->nColVal;
+  int   skip = 0;
+
   if (hasJson) {
-    p = indexPackJsonData(tem);
+    p = indexPackJsonDataPrefix(tem, &skip);
-    sz = strlen(p);
   }
   SArray* offsets = taosArrayInit(16, sizeof(uint64_t));
 
   AutomationCtx*    ctx = automCtxCreate((void*)p, AUTOMATION_ALWAYS);
   FstStreamBuilder* sb = fstSearch(((TFileReader*)reader)->fst, ctx);
 
-  FstSlice h = fstSliceCreate((uint8_t*)p, sz);
+  FstSlice h = fstSliceCreate((uint8_t*)p, skip);
   fstStreamBuilderSetRange(sb, &h, type);
   fstSliceDestroy(&h);
 
@@ -606,16 +606,16 @@ static bool tfileIteratorNext(Iterate* iiter) {
 static IterateValue* tifileIterateGetValue(Iterate* iter) { return &iter->val; }
 
 static TFileFstIter* tfileFstIteratorCreate(TFileReader* reader) {
-  TFileFstIter* tIter = taosMemoryCalloc(1, sizeof(TFileFstIter));
+  TFileFstIter* iter = taosMemoryCalloc(1, sizeof(TFileFstIter));
-  if (tIter == NULL) {
+  if (iter == NULL) {
     return NULL;
   }
 
-  tIter->ctx = automCtxCreate(NULL, AUTOMATION_ALWAYS);
+  iter->ctx = automCtxCreate(NULL, AUTOMATION_ALWAYS);
-  tIter->fb = fstSearch(reader->fst, tIter->ctx);
+  iter->fb = fstSearch(reader->fst, iter->ctx);
-  tIter->st = streamBuilderIntoStream(tIter->fb);
+  iter->st = streamBuilderIntoStream(iter->fb);
-  tIter->rdr = reader;
+  iter->rdr = reader;
-  return tIter;
+  return iter;
 }
 
 Iterate* tfileIteratorCreate(TFileReader* reader) {

source/libs/index/test/indexTests.cc

@@ -483,7 +483,7 @@ TEST_F(IndexTFileEnv, test_tfile_write) {
 
   std::string     colName("voltage");
   std::string     colVal("ab");
-  SIndexTerm*     term = indexTermCreate(1, ADD_VALUE, 0, TSDB_DATA_TYPE_BINARY, colName.c_str(), colName.size(),
+  SIndexTerm*     term = indexTermCreate(1, ADD_VALUE, TSDB_DATA_TYPE_BINARY, colName.c_str(), colName.size(),
                                      colVal.c_str(), colVal.size());
   SIndexTermQuery query = {term, QUERY_TERM};
 
@@ -557,7 +557,7 @@ TEST_F(IndexCacheEnv, cache_test) {
   std::string colName("voltage");
   {
     std::string colVal("v1");
-    SIndexTerm* term = indexTermCreate(0, ADD_VALUE, 0, TSDB_DATA_TYPE_BINARY, colName.c_str(), colName.size(),
+    SIndexTerm* term = indexTermCreate(0, ADD_VALUE, TSDB_DATA_TYPE_BINARY, colName.c_str(), colName.size(),
                                        colVal.c_str(), colVal.size());
     coj->Put(term, colId, version++, suid++);
     indexTermDestroy(term);
@@ -565,28 +565,28 @@ TEST_F(IndexCacheEnv, cache_test) {
   }
   {
     std::string colVal("v3");
-    SIndexTerm* term = indexTermCreate(0, ADD_VALUE, 0, TSDB_DATA_TYPE_BINARY, colName.c_str(), colName.size(),
+    SIndexTerm* term = indexTermCreate(0, ADD_VALUE, TSDB_DATA_TYPE_BINARY, colName.c_str(), colName.size(),
                                        colVal.c_str(), colVal.size());
     coj->Put(term, colId, version++, suid++);
     indexTermDestroy(term);
   }
   {
     std::string colVal("v2");
-    SIndexTerm* term = indexTermCreate(0, ADD_VALUE, 0, TSDB_DATA_TYPE_BINARY, colName.c_str(), colName.size(),
+    SIndexTerm* term = indexTermCreate(0, ADD_VALUE, TSDB_DATA_TYPE_BINARY, colName.c_str(), colName.size(),
                                        colVal.c_str(), colVal.size());
     coj->Put(term, colId, version++, suid++);
     indexTermDestroy(term);
   }
   {
     std::string colVal("v3");
-    SIndexTerm* term = indexTermCreate(0, ADD_VALUE, 0, TSDB_DATA_TYPE_BINARY, colName.c_str(), colName.size(),
+    SIndexTerm* term = indexTermCreate(0, ADD_VALUE, TSDB_DATA_TYPE_BINARY, colName.c_str(), colName.size(),
                                        colVal.c_str(), colVal.size());
     coj->Put(term, colId, version++, suid++);
     indexTermDestroy(term);
   }
   {
     std::string colVal("v3");
-    SIndexTerm* term = indexTermCreate(0, ADD_VALUE, 0, TSDB_DATA_TYPE_BINARY, colName.c_str(), colName.size(),
+    SIndexTerm* term = indexTermCreate(0, ADD_VALUE, TSDB_DATA_TYPE_BINARY, colName.c_str(), colName.size(),
                                        colVal.c_str(), colVal.size());
     coj->Put(term, colId, version++, suid++);
     indexTermDestroy(term);
@@ -595,14 +595,14 @@ TEST_F(IndexCacheEnv, cache_test) {
   std::cout << "--------first----------" << std::endl;
   {
     std::string colVal("v3");
-    SIndexTerm* term = indexTermCreate(0, ADD_VALUE, 0, TSDB_DATA_TYPE_BINARY, colName.c_str(), colName.size(),
+    SIndexTerm* term = indexTermCreate(0, ADD_VALUE, TSDB_DATA_TYPE_BINARY, colName.c_str(), colName.size(),
                                        colVal.c_str(), colVal.size());
     coj->Put(term, othColId, version++, suid++);
     indexTermDestroy(term);
   }
   {
     std::string colVal("v4");
-    SIndexTerm* term = indexTermCreate(0, ADD_VALUE, 0, TSDB_DATA_TYPE_BINARY, colName.c_str(), colName.size(),
+    SIndexTerm* term = indexTermCreate(0, ADD_VALUE, TSDB_DATA_TYPE_BINARY, colName.c_str(), colName.size(),
                                        colVal.c_str(), colVal.size());
     coj->Put(term, othColId, version++, suid++);
     indexTermDestroy(term);
@@ -613,7 +613,7 @@ TEST_F(IndexCacheEnv, cache_test) {
     std::string colVal("v4");
     for (size_t i = 0; i < 10; i++) {
       colVal[colVal.size() - 1] = 'a' + i;
-      SIndexTerm* term = indexTermCreate(0, ADD_VALUE, 0, TSDB_DATA_TYPE_BINARY, colName.c_str(), colName.size(),
+      SIndexTerm* term = indexTermCreate(0, ADD_VALUE, TSDB_DATA_TYPE_BINARY, colName.c_str(), colName.size(),
                                          colVal.c_str(), colVal.size());
       coj->Put(term, colId, version++, suid++);
       indexTermDestroy(term);
@@ -623,7 +623,7 @@ TEST_F(IndexCacheEnv, cache_test) {
   // begin query
   {
     std::string     colVal("v3");
-    SIndexTerm*     term = indexTermCreate(0, ADD_VALUE, 0, TSDB_DATA_TYPE_BINARY, colName.c_str(), colName.size(),
+    SIndexTerm*     term = indexTermCreate(0, ADD_VALUE, TSDB_DATA_TYPE_BINARY, colName.c_str(), colName.size(),
                                        colVal.c_str(), colVal.size());
     SIndexTermQuery query = {term, QUERY_TERM};
     SArray*         ret = (SArray*)taosArrayInit(4, sizeof(suid));
@@ -638,7 +638,7 @@ TEST_F(IndexCacheEnv, cache_test) {
   }
   {
     std::string     colVal("v2");
-    SIndexTerm*     term = indexTermCreate(0, ADD_VALUE, 0, TSDB_DATA_TYPE_BINARY, colName.c_str(), colName.size(),
+    SIndexTerm*     term = indexTermCreate(0, ADD_VALUE, TSDB_DATA_TYPE_BINARY, colName.c_str(), colName.size(),
                                        colVal.c_str(), colVal.size());
     SIndexTermQuery query = {term, QUERY_TERM};
     SArray*         ret = (SArray*)taosArrayInit(4, sizeof(suid));
@@ -670,7 +670,7 @@ class IndexObj {
     return ret;
   }
   void Del(const std::string& colName, const std::string& colVal, uint64_t uid) {
-    SIndexTerm*      term = indexTermCreate(0, DEL_VALUE, 0, TSDB_DATA_TYPE_BINARY, colName.c_str(), colName.size(),
+    SIndexTerm*      term = indexTermCreate(0, DEL_VALUE, TSDB_DATA_TYPE_BINARY, colName.c_str(), colName.size(),
                                        colVal.c_str(), colVal.size());
     SIndexMultiTerm* terms = indexMultiTermCreate();
     indexMultiTermAdd(terms, term);
@@ -679,7 +679,7 @@ class IndexObj {
   }
   int WriteMillonData(const std::string& colName, const std::string& colVal = "Hello world",
                       size_t numOfTable = 100 * 10000) {
-    SIndexTerm*      term = indexTermCreate(0, ADD_VALUE, 0, TSDB_DATA_TYPE_BINARY, colName.c_str(), colName.size(),
+    SIndexTerm*      term = indexTermCreate(0, ADD_VALUE, TSDB_DATA_TYPE_BINARY, colName.c_str(), colName.size(),
                                        colVal.c_str(), colVal.size());
     SIndexMultiTerm* terms = indexMultiTermCreate();
     indexMultiTermAdd(terms, term);
@@ -701,7 +701,7 @@ class IndexObj {
         // opt
         tColVal[taosRand() % colValSize] = 'a' + k % 26;
       }
-      SIndexTerm*      term = indexTermCreate(0, ADD_VALUE, 0, TSDB_DATA_TYPE_BINARY, colName.c_str(), colName.size(),
+      SIndexTerm*      term = indexTermCreate(0, ADD_VALUE, TSDB_DATA_TYPE_BINARY, colName.c_str(), colName.size(),
                                          tColVal.c_str(), tColVal.size());
       SIndexMultiTerm* terms = indexMultiTermCreate();
       indexMultiTermAdd(terms, term);
@@ -737,7 +737,7 @@ class IndexObj {
 
   int SearchOne(const std::string& colName, const std::string& colVal) {
     SIndexMultiTermQuery* mq = indexMultiTermQueryCreate(MUST);
-    SIndexTerm* term = indexTermCreate(0, ADD_VALUE, 0, TSDB_DATA_TYPE_BINARY, colName.c_str(), colName.size(),
+    SIndexTerm*           term = indexTermCreate(0, ADD_VALUE, TSDB_DATA_TYPE_BINARY, colName.c_str(), colName.size(),
                                        colVal.c_str(), colVal.size());
     indexMultiTermQueryAdd(mq, term, QUERY_TERM);
 
@@ -759,7 +759,7 @@ class IndexObj {
   }
   int SearchOneTarget(const std::string& colName, const std::string& colVal, uint64_t val) {
     SIndexMultiTermQuery* mq = indexMultiTermQueryCreate(MUST);
-    SIndexTerm* term = indexTermCreate(0, ADD_VALUE, 0, TSDB_DATA_TYPE_BINARY, colName.c_str(), colName.size(),
+    SIndexTerm*           term = indexTermCreate(0, ADD_VALUE, TSDB_DATA_TYPE_BINARY, colName.c_str(), colName.size(),
                                        colVal.c_str(), colVal.size());
     indexMultiTermQueryAdd(mq, term, QUERY_TERM);
 
@@ -784,7 +784,7 @@ class IndexObj {
 
   void PutOne(const std::string& colName, const std::string& colVal) {
     SIndexMultiTerm* terms = indexMultiTermCreate();
-    SIndexTerm*      term = indexTermCreate(0, ADD_VALUE, 0, TSDB_DATA_TYPE_BINARY, colName.c_str(), colName.size(),
+    SIndexTerm*      term = indexTermCreate(0, ADD_VALUE, TSDB_DATA_TYPE_BINARY, colName.c_str(), colName.size(),
                                        colVal.c_str(), colVal.size());
     indexMultiTermAdd(terms, term);
     Put(terms, 10);
@@ -792,7 +792,7 @@ class IndexObj {
   }
   void PutOneTarge(const std::string& colName, const std::string& colVal, uint64_t val) {
     SIndexMultiTerm* terms = indexMultiTermCreate();
-    SIndexTerm*      term = indexTermCreate(0, ADD_VALUE, 0, TSDB_DATA_TYPE_BINARY, colName.c_str(), colName.size(),
+    SIndexTerm*      term = indexTermCreate(0, ADD_VALUE, TSDB_DATA_TYPE_BINARY, colName.c_str(), colName.size(),
                                        colVal.c_str(), colVal.size());
     indexMultiTermAdd(terms, term);
     Put(terms, val);
@@ -832,7 +832,7 @@ TEST_F(IndexEnv2, testIndexOpen) {
   {
     std::string colName("tag1"), colVal("Hello");
 
-    SIndexTerm*      term = indexTermCreate(0, ADD_VALUE, 0, TSDB_DATA_TYPE_BINARY, colName.c_str(), colName.size(),
+    SIndexTerm*      term = indexTermCreate(0, ADD_VALUE, TSDB_DATA_TYPE_BINARY, colName.c_str(), colName.size(),
                                        colVal.c_str(), colVal.size());
     SIndexMultiTerm* terms = indexMultiTermCreate();
     indexMultiTermAdd(terms, term);
@@ -847,7 +847,7 @@ TEST_F(IndexEnv2, testIndexOpen) {
     size_t      size = 200;
     std::string colName("tag1"), colVal("hello");
 
-    SIndexTerm*      term = indexTermCreate(0, ADD_VALUE, 0, TSDB_DATA_TYPE_BINARY, colName.c_str(), colName.size(),
+    SIndexTerm*      term = indexTermCreate(0, ADD_VALUE, TSDB_DATA_TYPE_BINARY, colName.c_str(), colName.size(),
                                        colVal.c_str(), colVal.size());
     SIndexMultiTerm* terms = indexMultiTermCreate();
     indexMultiTermAdd(terms, term);
@@ -862,7 +862,7 @@ TEST_F(IndexEnv2, testIndexOpen) {
     size_t      size = 200;
     std::string colName("tag1"), colVal("Hello");
 
-    SIndexTerm*      term = indexTermCreate(0, ADD_VALUE, 0, TSDB_DATA_TYPE_BINARY, colName.c_str(), colName.size(),
+    SIndexTerm*      term = indexTermCreate(0, ADD_VALUE, TSDB_DATA_TYPE_BINARY, colName.c_str(), colName.size(),
                                        colVal.c_str(), colVal.size());
     SIndexMultiTerm* terms = indexMultiTermCreate();
     indexMultiTermAdd(terms, term);
@@ -877,7 +877,7 @@ TEST_F(IndexEnv2, testIndexOpen) {
   {
     std::string           colName("tag1"), colVal("Hello");
     SIndexMultiTermQuery* mq = indexMultiTermQueryCreate(MUST);
-    SIndexTerm* term = indexTermCreate(0, ADD_VALUE, 0, TSDB_DATA_TYPE_BINARY, colName.c_str(), colName.size(),
+    SIndexTerm*           term = indexTermCreate(0, ADD_VALUE, TSDB_DATA_TYPE_BINARY, colName.c_str(), colName.size(),
                                        colVal.c_str(), colVal.size());
     indexMultiTermQueryAdd(mq, term, QUERY_TERM);
 

source/libs/index/test/jsonUT.cc

@@ -40,7 +40,7 @@ TEST_F(JsonEnv, testWrite) {
   {
     std::string colName("test");
     std::string colVal("ab");
-    SIndexTerm* term = indexTermCreate(1, ADD_VALUE, 0, TSDB_DATA_TYPE_BINARY, colName.c_str(), colName.size(),
+    SIndexTerm* term = indexTermCreate(1, ADD_VALUE, TSDB_DATA_TYPE_BINARY, colName.c_str(), colName.size(),
                                        colVal.c_str(), colVal.size());
 
     SIndexMultiTerm* terms = indexMultiTermCreate();
@@ -53,7 +53,7 @@ TEST_F(JsonEnv, testWrite) {
   {
     std::string colName("voltage");
     std::string colVal("ab1");
-    SIndexTerm* term = indexTermCreate(1, ADD_VALUE, 0, TSDB_DATA_TYPE_BINARY, colName.c_str(), colName.size(),
+    SIndexTerm* term = indexTermCreate(1, ADD_VALUE, TSDB_DATA_TYPE_BINARY, colName.c_str(), colName.size(),
                                        colVal.c_str(), colVal.size());
 
     SIndexMultiTerm* terms = indexMultiTermCreate();
@@ -66,7 +66,7 @@ TEST_F(JsonEnv, testWrite) {
   {
     std::string colName("voltage");
     std::string colVal("123");
-    SIndexTerm* term = indexTermCreate(1, ADD_VALUE, 0, TSDB_DATA_TYPE_BINARY, colName.c_str(), colName.size(),
+    SIndexTerm* term = indexTermCreate(1, ADD_VALUE, TSDB_DATA_TYPE_BINARY, colName.c_str(), colName.size(),
                                        colVal.c_str(), colVal.size());
 
     SIndexMultiTerm* terms = indexMultiTermCreate();
@@ -81,7 +81,7 @@ TEST_F(JsonEnv, testWrite) {
     std::string colVal("ab");
 
     SIndexMultiTermQuery* mq = indexMultiTermQueryCreate(MUST);
-    SIndexTerm*           q = indexTermCreate(1, ADD_VALUE, 0, TSDB_DATA_TYPE_BINARY, colName.c_str(), colName.size(),
+    SIndexTerm*           q = indexTermCreate(1, ADD_VALUE, TSDB_DATA_TYPE_BINARY, colName.c_str(), colName.size(),
                                     colVal.c_str(), colVal.size());
 
     SArray* result = taosArrayInit(1, sizeof(uint64_t));
@@ -95,7 +95,7 @@ TEST_F(JsonEnv, testWriteMillonData) {
   {
     std::string colName("test");
     std::string colVal("ab");
-    SIndexTerm* term = indexTermCreate(1, ADD_VALUE, 0, TSDB_DATA_TYPE_BINARY, colName.c_str(), colName.size(),
+    SIndexTerm* term = indexTermCreate(1, ADD_VALUE, TSDB_DATA_TYPE_BINARY, colName.c_str(), colName.size(),
                                        colVal.c_str(), colVal.size());
 
     SIndexMultiTerm* terms = indexMultiTermCreate();
@@ -110,7 +110,7 @@ TEST_F(JsonEnv, testWriteMillonData) {
     std::string colVal("abxxxxxxxxxxxx");
     for (int i = 0; i < 1000; i++) {
       colVal[i % colVal.size()] = '0' + i % 128;
-      SIndexTerm* term = indexTermCreate(1, ADD_VALUE, 0, TSDB_DATA_TYPE_BINARY, colName.c_str(), colName.size(),
+      SIndexTerm* term = indexTermCreate(1, ADD_VALUE, TSDB_DATA_TYPE_BINARY, colName.c_str(), colName.size(),
                                          colVal.c_str(), colVal.size());
 
       SIndexMultiTerm* terms = indexMultiTermCreate();
@@ -124,7 +124,7 @@ TEST_F(JsonEnv, testWriteMillonData) {
   {
     std::string colName("voltagefdadfa");
     std::string colVal("abxxxxxxxxxxxx");
-    SIndexTerm* term = indexTermCreate(1, ADD_VALUE, 0, TSDB_DATA_TYPE_BINARY, colName.c_str(), colName.size(),
+    SIndexTerm* term = indexTermCreate(1, ADD_VALUE, TSDB_DATA_TYPE_BINARY, colName.c_str(), colName.size(),
                                        colVal.c_str(), colVal.size());
 
     SIndexMultiTerm* terms = indexMultiTermCreate();
@@ -139,7 +139,7 @@ TEST_F(JsonEnv, testWriteMillonData) {
     std::string colVal("ab");
 
     SIndexMultiTermQuery* mq = indexMultiTermQueryCreate(MUST);
-    SIndexTerm*           q = indexTermCreate(1, ADD_VALUE, 0, TSDB_DATA_TYPE_BINARY, colName.c_str(), colName.size(),
+    SIndexTerm*           q = indexTermCreate(1, ADD_VALUE, TSDB_DATA_TYPE_BINARY, colName.c_str(), colName.size(),
                                     colVal.c_str(), colVal.size());
 
     SArray* result = taosArrayInit(1, sizeof(uint64_t));

source/libs/parser/inc/parInsertData.h

@@ -133,15 +133,15 @@ static FORCE_INLINE int32_t setBlockInfo(SSubmitBlk *pBlocks, STableDataBlocks *
 int32_t schemaIdxCompar(const void *lhs, const void *rhs);
 int32_t boundIdxCompar(const void *lhs, const void *rhs);
 void    setBoundColumnInfo(SParsedDataColInfo *pColList, SSchema *pSchema, col_id_t numOfCols);
-void    destroyBlockArrayList(SArray *pDataBlockList);
+void destroyBlockArrayList(SArray* pDataBlockList);
-void    destroyBlockHashmap(SHashObj *pDataBlockHash);
+void destroyBlockHashmap(SHashObj* pDataBlockHash);
-int     initRowBuilder(SRowBuilder *pBuilder, int16_t schemaVer, SParsedDataColInfo *pColInfo);
+int  initRowBuilder(SRowBuilder *pBuilder, int16_t schemaVer, SParsedDataColInfo *pColInfo);
-int32_t allocateMemIfNeed(STableDataBlocks *pDataBlock, int32_t rowSize, int32_t *numOfRows);
+int32_t allocateMemIfNeed(STableDataBlocks *pDataBlock, int32_t rowSize, int32_t * numOfRows);
-int32_t getDataBlockFromList(SHashObj *pHashList, int64_t id, int32_t size, int32_t startOffset, int32_t rowSize,
+int32_t getDataBlockFromList(SHashObj* pHashList, int64_t id, int32_t size, int32_t startOffset, int32_t rowSize,
-                             const STableMeta *pTableMeta, STableDataBlocks **dataBlocks, SArray *pBlockList,
+     STableMeta* pTableMeta, STableDataBlocks** dataBlocks, SArray* pBlockList, SVCreateTbReq* pCreateTbReq);
-                             SVCreateTbReq *pCreateTbReq);
+int32_t mergeTableDataBlocks(SHashObj* pHashObj, uint8_t payloadType, SArray** pVgDataBlocks);
-int32_t mergeTableDataBlocks(SHashObj *pHashObj, uint8_t payloadType, SArray **pVgDataBlocks);
+int32_t buildCreateTbMsg(STableDataBlocks* pBlocks, SVCreateTbReq* pCreateTbReq);
-int32_t buildCreateTbMsg(STableDataBlocks *pBlocks, SVCreateTbReq *pCreateTbReq);
+
 int32_t allocateMemForSize(STableDataBlocks *pDataBlock, int32_t allSize);
 
 #endif  // TDENGINE_DATABLOCKMGT_H

source/libs/parser/inc/parUtil.h

@@ -41,7 +41,6 @@ int32_t generateSyntaxErrMsg(SMsgBuf* pBuf, int32_t errCode, ...);
 int32_t buildInvalidOperationMsg(SMsgBuf* pMsgBuf, const char* msg);
 int32_t buildSyntaxErrMsg(SMsgBuf* pBuf, const char* additionalInfo, const char* sourceStr);
 
-STableMeta*   tableMetaDup(const STableMeta* pTableMeta);
 SSchema*      getTableColumnSchema(const STableMeta* pTableMeta);
 SSchema*      getTableTagSchema(const STableMeta* pTableMeta);
 int32_t       getNumOfColumns(const STableMeta* pTableMeta);

source/libs/parser/src/parInsert.c

@@ -279,6 +279,7 @@ static void buildMsgHeader(STableDataBlocks* src, SVgDataBlocks* blocks) {
   int32_t     numOfBlocks = blocks->numOfTables;
   while (numOfBlocks--) {
     int32_t dataLen = blk->dataLen;
+    int32_t schemaLen = blk->schemaLen;
     blk->uid = htobe64(blk->uid);
     blk->suid = htobe64(blk->suid);
     blk->padding = htonl(blk->padding);
@@ -286,7 +287,7 @@ static void buildMsgHeader(STableDataBlocks* src, SVgDataBlocks* blocks) {
     blk->dataLen = htonl(blk->dataLen);
     blk->schemaLen = htonl(blk->schemaLen);
     blk->numOfRows = htons(blk->numOfRows);
-    blk = (SSubmitBlk*)(blk->data + dataLen);
+    blk = (SSubmitBlk*)(blk->data + schemaLen + dataLen);
   }
 }
 
@@ -761,11 +762,9 @@ static int32_t KvRowAppend(SMsgBuf* pMsgBuf, const void* value, int32_t len, voi
   return TSDB_CODE_SUCCESS;
 }
 
-static int32_t buildCreateTbReq(SVCreateTbReq* pTbReq, const SName* pName, SKVRow row, int64_t suid) {
+static int32_t buildCreateTbReq(SVCreateTbReq *pTbReq, const char* tname, SKVRow row, int64_t suid) {
-  char dbFName[TSDB_DB_FNAME_LEN] = {0};
-  tNameGetFullDbName(pName, dbFName);
   pTbReq->type = TD_CHILD_TABLE;
-  pTbReq->name = strdup(pName->tname);
+  pTbReq->name = strdup(tname);
   pTbReq->ctb.suid = suid;
   pTbReq->ctb.pTag = row;
 
@@ -773,7 +772,7 @@ static int32_t buildCreateTbReq(SVCreateTbReq* pTbReq, const SName* pName, SKVRo
 }
 
 // pSql -> tag1_value, ...)
-static int32_t parseTagsClause(SInsertParseContext* pCxt, SSchema* pSchema, uint8_t precision, const SName* pName) {
+static int32_t parseTagsClause(SInsertParseContext* pCxt, SSchema* pSchema, uint8_t precision, const char* tName) {
   if (tdInitKVRowBuilder(&pCxt->tagsBuilder) < 0) {
     return TSDB_CODE_TSC_OUT_OF_MEMORY;
   }
@@ -814,7 +813,7 @@ static int32_t parseTagsClause(SInsertParseContext* pCxt, SSchema* pSchema, uint
   }
   tdSortKVRowByColIdx(row);
 
-  return buildCreateTbReq(&pCxt->createTblReq, pName, row, pCxt->pTableMeta->suid);
+  return buildCreateTbReq(&pCxt->createTblReq, tName, row, pCxt->pTableMeta->suid);
 }
 
 static int32_t cloneTableMeta(STableMeta* pSrc, STableMeta** pDst) {
@@ -883,7 +882,7 @@ static int32_t parseUsingClause(SInsertParseContext* pCxt, SToken* pTbnameToken)
   if (TK_NK_LP != sToken.type) {
     return buildSyntaxErrMsg(&pCxt->msg, "( is expected", sToken.z);
   }
-  CHECK_CODE(parseTagsClause(pCxt, pCxt->pTableMeta->schema, getTableInfo(pCxt->pTableMeta).precision, &name));
+  CHECK_CODE(parseTagsClause(pCxt, pCxt->pTableMeta->schema, getTableInfo(pCxt->pTableMeta).precision, name.tname));
   NEXT_TOKEN(pCxt->pSql, sToken);
   if (TK_NK_RP != sToken.type) {
     return buildSyntaxErrMsg(&pCxt->msg, ") is expected", sToken.z);
@@ -1037,11 +1036,6 @@ static void destroyDataBlock(STableDataBlocks* pDataBlock) {
 
   taosMemoryFreeClear(pDataBlock->pData);
   if (!pDataBlock->cloned) {
-    // free the refcount for metermeta
-    if (pDataBlock->pTableMeta != NULL) {
-      taosMemoryFreeClear(pDataBlock->pTableMeta);
-    }
-
     destroyBoundColumnInfo(&pDataBlock->boundColumnInfo);
   }
   taosMemoryFreeClear(pDataBlock);
@@ -1269,10 +1263,9 @@ int32_t qBuildStmtOutput(SQuery* pQuery, SHashObj* pVgHash, SHashObj* pBlockHash
   return TSDB_CODE_SUCCESS;
 }
 
-int32_t qBindStmtTagsValue(void* pBlock, void* boundTags, int64_t suid, SName* pName, TAOS_MULTI_BIND* bind,
+int32_t qBindStmtTagsValue(void *pBlock, void *boundTags, int64_t suid, char *tName, TAOS_MULTI_BIND *bind, char *msgBuf, int32_t msgBufLen){
-                           char* msgBuf, int32_t msgBufLen) {
+  STableDataBlocks *pDataBlock = (STableDataBlocks *)pBlock;
-  STableDataBlocks*   pDataBlock = (STableDataBlocks*)pBlock;
+  SMsgBuf pBuf = {.buf = msgBuf, .len = msgBufLen};
-  SMsgBuf             pBuf = {.buf = msgBuf, .len = msgBufLen};
   SParsedDataColInfo* tags = (SParsedDataColInfo*)boundTags;
   if (NULL == tags) {
     return TSDB_CODE_QRY_APP_ERROR;
@@ -1311,7 +1304,7 @@ int32_t qBindStmtTagsValue(void* pBlock, void* boundTags, int64_t suid, SName* p
   tdSortKVRowByColIdx(row);
 
   SVCreateTbReq tbReq = {0};
-  CHECK_CODE(buildCreateTbReq(&tbReq, pName, row, suid));
+  CHECK_CODE(buildCreateTbReq(&tbReq, tName, row, suid));
   CHECK_CODE(buildCreateTbMsg(pDataBlock, &tbReq));
 
   destroyCreateSubTbReq(&tbReq);
@@ -1543,3 +1536,265 @@ int32_t qBuildStmtColFields(void* pBlock, int32_t* fieldNum, TAOS_FIELD** fields
 
   return TSDB_CODE_SUCCESS;
 }
+
+// schemaless logic start
+
+typedef struct SmlExecHandle {
+  SHashObj*    pBlockHash;
+
+  SParsedDataColInfo tags;      // each table
+  SKVRowBuilder tagsBuilder;    // each table
+  SVCreateTbReq createTblReq;   // each table
+
+  SQuery* pQuery;
+} SSmlExecHandle;
+
+static int32_t smlBoundColumns(SArray *cols, SParsedDataColInfo* pColList, SSchema* pSchema) {
+  col_id_t nCols = pColList->numOfCols;
+
+  pColList->numOfBound = 0;
+  pColList->boundNullLen = 0;
+  memset(pColList->boundColumns, 0, sizeof(col_id_t) * nCols);
+  for (col_id_t i = 0; i < nCols; ++i) {
+    pColList->cols[i].valStat = VAL_STAT_NONE;
+  }
+
+  bool     isOrdered = true;
+  col_id_t lastColIdx = -1;  // last column found
+  for (int i = 0; i < taosArrayGetSize(cols); ++i) {
+    SSmlKv *kv = taosArrayGetP(cols, i);
+    SToken sToken = {.n=kv->keyLen, .z=(char*)kv->key};
+    col_id_t t = lastColIdx + 1;
+    col_id_t index = findCol(&sToken, t, nCols, pSchema);
+    if (index < 0 && t > 0) {
+      index = findCol(&sToken, 0, t, pSchema);
+      isOrdered = false;
+    }
+    if (index < 0) {
+      return TSDB_CODE_SML_INVALID_DATA;
+    }
+    if (pColList->cols[index].valStat == VAL_STAT_HAS) {
+      return TSDB_CODE_SML_INVALID_DATA;
+    }
+    lastColIdx = index;
+    pColList->cols[index].valStat = VAL_STAT_HAS;
+    pColList->boundColumns[pColList->numOfBound] = index + PRIMARYKEY_TIMESTAMP_COL_ID;
+    ++pColList->numOfBound;
+    switch (pSchema[t].type) {
+      case TSDB_DATA_TYPE_BINARY:
+        pColList->boundNullLen += (sizeof(VarDataOffsetT) + VARSTR_HEADER_SIZE + CHAR_BYTES);
+        break;
+      case TSDB_DATA_TYPE_NCHAR:
+        pColList->boundNullLen += (sizeof(VarDataOffsetT) + VARSTR_HEADER_SIZE + TSDB_NCHAR_SIZE);
+        break;
+      default:
+        pColList->boundNullLen += TYPE_BYTES[pSchema[t].type];
+        break;
+    }
+  }
+
+  pColList->orderStatus = isOrdered ? ORDER_STATUS_ORDERED : ORDER_STATUS_DISORDERED;
+
+  if (!isOrdered) {
+    pColList->colIdxInfo = taosMemoryCalloc(pColList->numOfBound, sizeof(SBoundIdxInfo));
+    if (NULL == pColList->colIdxInfo) {
+      return TSDB_CODE_TSC_OUT_OF_MEMORY;
+    }
+    SBoundIdxInfo* pColIdx = pColList->colIdxInfo;
+    for (col_id_t i = 0; i < pColList->numOfBound; ++i) {
+      pColIdx[i].schemaColIdx = pColList->boundColumns[i];
+      pColIdx[i].boundIdx = i;
+    }
+    qsort(pColIdx, pColList->numOfBound, sizeof(SBoundIdxInfo), schemaIdxCompar);
+    for (col_id_t i = 0; i < pColList->numOfBound; ++i) {
+      pColIdx[i].finalIdx = i;
+    }
+    qsort(pColIdx, pColList->numOfBound, sizeof(SBoundIdxInfo), boundIdxCompar);
+  }
+
+  if(pColList->numOfCols > pColList->numOfBound){
+    memset(&pColList->boundColumns[pColList->numOfBound], 0,
+           sizeof(col_id_t) * (pColList->numOfCols - pColList->numOfBound));
+  }
+
+  return TSDB_CODE_SUCCESS;
+}
+
+static int32_t smlBoundTags(SArray *cols, SKVRowBuilder *tagsBuilder, SParsedDataColInfo* tags, SSchema* pSchema, SKVRow *row, SMsgBuf *msg) {
+  if (tdInitKVRowBuilder(tagsBuilder) < 0) {
+    return TSDB_CODE_TSC_OUT_OF_MEMORY;
+  }
+
+  SKvParam param = {.builder = tagsBuilder};
+  for (int i = 0; i < tags->numOfBound; ++i) {
+    SSchema* pTagSchema = &pSchema[tags->boundColumns[i] - 1]; // colId starts with 1
+    param.schema = pTagSchema;
+    SSmlKv *kv = taosArrayGetP(cols, i);
+    KvRowAppend(msg, kv->value, kv->valueLen, &param) ;
+  }
+
+
+  *row = tdGetKVRowFromBuilder(tagsBuilder);
+  if(*row == NULL){
+    return TSDB_CODE_SML_INVALID_DATA;
+  }
+  tdSortKVRowByColIdx(*row);
+  return TSDB_CODE_SUCCESS;
+}
+
+int32_t smlBindData(void *handle, SArray *tags, SArray *colsFormat, SHashObj *colsHash, SArray *cols, bool format,
+                    STableMeta *pTableMeta, char *tableName, char *msgBuf, int16_t msgBufLen) {
+  SMsgBuf pBuf = {.buf = msgBuf, .len = msgBufLen};
+
+  SSmlExecHandle *smlHandle = (SSmlExecHandle *)handle;
+  SSchema* pTagsSchema = getTableTagSchema(pTableMeta);
+  setBoundColumnInfo(&smlHandle->tags, pTagsSchema, getNumOfTags(pTableMeta));
+  int ret = smlBoundColumns(tags, &smlHandle->tags, pTagsSchema);
+  if(ret != TSDB_CODE_SUCCESS){
+    buildInvalidOperationMsg(&pBuf, "bound tags error");
+    return ret;
+  }
+  SKVRow row = NULL;
+  ret = smlBoundTags(tags, &smlHandle->tagsBuilder, &smlHandle->tags, pTagsSchema, &row, &pBuf);
+  if(ret != TSDB_CODE_SUCCESS){
+    return ret;
+  }
+
+  buildCreateTbReq(&smlHandle->createTblReq, tableName, row, pTableMeta->suid);
+
+  STableDataBlocks* pDataBlock = NULL;
+  ret = getDataBlockFromList(smlHandle->pBlockHash, pTableMeta->uid, TSDB_DEFAULT_PAYLOAD_SIZE,
+                                  sizeof(SSubmitBlk), getTableInfo(pTableMeta).rowSize, pTableMeta,
+                                  &pDataBlock, NULL, &smlHandle->createTblReq);
+  if(ret != TSDB_CODE_SUCCESS){
+    buildInvalidOperationMsg(&pBuf, "create data block error");
+    return ret;
+  }
+
+  SSchema* pSchema = getTableColumnSchema(pTableMeta);
+
+
+  if(format){
+    ret = smlBoundColumns(taosArrayGetP(colsFormat, 0), &pDataBlock->boundColumnInfo, pSchema);
+  }else{
+    SArray *columns = taosArrayInit(16, POINTER_BYTES);
+    void **p1 = taosHashIterate(colsHash, NULL);
+    while (p1) {
+      SSmlKv* kv = *p1;
+      taosArrayPush(columns, &kv);
+      p1 = taosHashIterate(colsHash, p1);
+    }
+    ret = smlBoundColumns(columns, &pDataBlock->boundColumnInfo, pSchema);
+    taosArrayDestroy(columns);
+  }
+
+  if(ret != TSDB_CODE_SUCCESS){
+    buildInvalidOperationMsg(&pBuf, "bound cols error");
+    return ret;
+  }
+  int32_t extendedRowSize = getExtendedRowSize(pDataBlock);
+  SParsedDataColInfo* spd = &pDataBlock->boundColumnInfo;
+  SRowBuilder*        pBuilder = &pDataBlock->rowBuilder;
+  SMemParam param = {.rb = pBuilder};
+
+  initRowBuilder(&pDataBlock->rowBuilder, pDataBlock->pTableMeta->sversion, &pDataBlock->boundColumnInfo);
+
+  int32_t rowNum = format ? taosArrayGetSize(colsFormat) : taosArrayGetSize(cols);
+  if(rowNum <= 0) {
+    return buildInvalidOperationMsg(&pBuf, "cols size <= 0");
+  }
+  ret = allocateMemForSize(pDataBlock, extendedRowSize * rowNum);
+  if(ret != TSDB_CODE_SUCCESS){
+    buildInvalidOperationMsg(&pBuf, "allocate memory error");
+    return ret;
+  }
+  for (int32_t r = 0; r < rowNum; ++r) {
+    STSRow* row = (STSRow*)(pDataBlock->pData + pDataBlock->size);  // skip the SSubmitBlk header
+    tdSRowResetBuf(pBuilder, row);
+    void *rowData = NULL;
+    if(format){
+      rowData = taosArrayGetP(colsFormat, r);
+    }else{
+      rowData = taosArrayGetP(cols, r);
+    }
+
+    // 1. set the parsed value from sql string
+    for (int c = 0; c < spd->numOfBound; ++c) {
+      SSchema* pColSchema = &pSchema[spd->boundColumns[c] - 1];
+
+      param.schema = pColSchema;
+      getSTSRowAppendInfo(pBuilder->rowType, spd, c, &param.toffset, &param.colIdx);
+
+      SSmlKv *kv = NULL;
+      if(format){
+        kv = taosArrayGetP(rowData, c);
+        if (!kv){
+          char msg[64] = {0};
+          sprintf(msg, "cols num not the same like before:%d", r);
+          return buildInvalidOperationMsg(&pBuf, msg);
+        }
+      }else{
+        void **p =taosHashGet(rowData, pColSchema->name, strlen(pColSchema->name));
+        kv = *p;
+      }
+
+      if (kv->length == 0) {
+        MemRowAppend(&pBuf, NULL, 0, &param);
+      } else {
+        int32_t colLen = pColSchema->bytes;
+        if (IS_VAR_DATA_TYPE(pColSchema->type)) {
+          colLen = kv->length;
+        }
+
+        MemRowAppend(&pBuf, &(kv->value), colLen, &param);
+      }
+
+      if (PRIMARYKEY_TIMESTAMP_COL_ID == pColSchema->colId) {
+        TSKEY tsKey = TD_ROW_KEY(row);
+        checkTimestamp(pDataBlock, (const char *)&tsKey);
+      }
+    }
+
+    // set the null value for the columns that do not assign values
+    if ((spd->numOfBound < spd->numOfCols) && TD_IS_TP_ROW(row)) {
+      for (int32_t i = 0; i < spd->numOfCols; ++i) {
+        if (spd->cols[i].valStat == VAL_STAT_NONE) {  // the primary TS key is not VAL_STAT_NONE
+          tdAppendColValToTpRow(pBuilder, TD_VTYPE_NONE, getNullValue(pSchema[i].type), true, pSchema[i].type, i,
+                                spd->cols[i].toffset);
+        }
+      }
+    }
+
+    pDataBlock->size += extendedRowSize;
+  }
+
+  SSubmitBlk *pBlocks = (SSubmitBlk *)(pDataBlock->pData);
+  if (TSDB_CODE_SUCCESS != setBlockInfo(pBlocks, pDataBlock, rowNum)) {
+    return buildInvalidOperationMsg(&pBuf, "too many rows in sql, total number of rows should be less than 32767");
+  }
+
+  return TSDB_CODE_SUCCESS;
+}
+
+void* smlInitHandle(SQuery *pQuery){
+  SSmlExecHandle *handle = taosMemoryCalloc(1, sizeof(SSmlExecHandle));
+  if(!handle) return NULL;
+  handle->pBlockHash = taosHashInit(16, taosGetDefaultHashFunction(TSDB_DATA_TYPE_BIGINT), true, false);
+  handle->pQuery = pQuery;
+
+  return handle;
+}
+
+void smlDestroyHandle(void *pHandle){
+  if(!pHandle) return;
+  SSmlExecHandle *handle = (SSmlExecHandle *)pHandle;
+  destroyBlockHashmap(handle->pBlockHash);
+  taosMemoryFree(handle);
+}
+
+int32_t smlBuildOutput(void* handle, SHashObj* pVgHash) {
+  SSmlExecHandle *smlHandle = (SSmlExecHandle *)handle;
+  return qBuildStmtOutput(smlHandle->pQuery, pVgHash, smlHandle->pBlockHash);
+}
+// schemaless logic end
+

source/libs/parser/src/parInsertData.c

@@ -115,7 +115,7 @@ void destroyBoundColumnInfo(void* pBoundInfo) {
   taosMemoryFreeClear(pColList->colIdxInfo);
 }
 
-static int32_t createDataBlock(size_t defaultSize, int32_t rowSize, int32_t startOffset, const STableMeta* pTableMeta,
+static int32_t createDataBlock(size_t defaultSize, int32_t rowSize, int32_t startOffset, STableMeta* pTableMeta,
                                STableDataBlocks** dataBlocks) {
   STableDataBlocks* dataBuf = (STableDataBlocks*)taosMemoryCalloc(1, sizeof(STableDataBlocks));
   if (dataBuf == NULL) {
@@ -137,8 +137,7 @@ static int32_t createDataBlock(size_t defaultSize, int32_t rowSize, int32_t star
   }
   memset(dataBuf->pData, 0, sizeof(SSubmitBlk));
 
-  // Here we keep the tableMeta to avoid it to be remove by other threads.
+  dataBuf->pTableMeta = pTableMeta;
-  dataBuf->pTableMeta = tableMetaDup(pTableMeta);
 
   SParsedDataColInfo* pColInfo = &dataBuf->boundColumnInfo;
   SSchema*            pSchema = getTableColumnSchema(dataBuf->pTableMeta);
@@ -187,7 +186,7 @@ int32_t buildCreateTbMsg(STableDataBlocks* pBlocks, SVCreateTbReq* pCreateTbReq)
 }
 
 int32_t getDataBlockFromList(SHashObj* pHashList, int64_t id, int32_t size, int32_t startOffset, int32_t rowSize,
-                             const STableMeta* pTableMeta, STableDataBlocks** dataBlocks, SArray* pBlockList,
+                             STableMeta* pTableMeta, STableDataBlocks** dataBlocks, SArray* pBlockList,
                              SVCreateTbReq* pCreateTbReq) {
   *dataBlocks = NULL;
   STableDataBlocks** t1 = (STableDataBlocks**)taosHashGet(pHashList, (const char*)&id, sizeof(id));
@@ -238,9 +237,9 @@ static void destroyDataBlock(STableDataBlocks* pDataBlock) {
   taosMemoryFreeClear(pDataBlock->pData);
   if (!pDataBlock->cloned) {
     // free the refcount for metermeta
-    if (pDataBlock->pTableMeta != NULL) {
+//    if (pDataBlock->pTableMeta != NULL) {
-      taosMemoryFreeClear(pDataBlock->pTableMeta);
+//      taosMemoryFreeClear(pDataBlock->pTableMeta);
-    }
+//    }
 
     destroyBoundColumnInfo(&pDataBlock->boundColumnInfo);
   }
@@ -456,6 +455,7 @@ int32_t mergeTableDataBlocks(SHashObj* pHashObj, uint8_t payloadType, SArray** p
     SSubmitBlk* pBlocks = (SSubmitBlk*)pOneTableBlock->pData;
     if (pBlocks->numOfRows > 0) {
       STableDataBlocks* dataBuf = NULL;
+      pOneTableBlock->pTableMeta->vgId = pOneTableBlock->vgId;    // for schemaless, restore origin vgId
       int32_t           ret =
           getDataBlockFromList(pVnodeDataBlockHashList, pOneTableBlock->vgId, TSDB_PAYLOAD_SIZE, INSERT_HEAD_SIZE, 0,
                                pOneTableBlock->pTableMeta, &dataBuf, pVnodeDataBlockList, NULL);

source/libs/parser/src/parUtil.c

@@ -191,27 +191,7 @@ int32_t buildSyntaxErrMsg(SMsgBuf* pBuf, const char* additionalInfo, const char*
   return TSDB_CODE_TSC_SQL_SYNTAX_ERROR;
 }
 
-static uint32_t getTableMetaSize(const STableMeta* pTableMeta) {
+SSchema *getTableColumnSchema(const STableMeta *pTableMeta) {
-  assert(pTableMeta != NULL);
-
-  int32_t totalCols = 0;
-  if (pTableMeta->tableInfo.numOfColumns >= 0) {
-    totalCols = pTableMeta->tableInfo.numOfColumns + pTableMeta->tableInfo.numOfTags;
-  }
-
-  return sizeof(STableMeta) + totalCols * sizeof(SSchema);
-}
-
-STableMeta* tableMetaDup(const STableMeta* pTableMeta) {
-  assert(pTableMeta != NULL);
-  size_t size = getTableMetaSize(pTableMeta);
-
-  STableMeta* p = taosMemoryMalloc(size);
-  memcpy(p, pTableMeta, size);
-  return p;
-}
-
-SSchema* getTableColumnSchema(const STableMeta* pTableMeta) {
   assert(pTableMeta != NULL);
   return (SSchema*)pTableMeta->schema;
 }

source/libs/scalar/test/scalar/scalarTests.cpp

@@ -166,7 +166,6 @@ void scltMakeColumnNode(SNode **pNode, SSDataBlock **block, int32_t dataType, in
     idata.pData = (char *)taosMemoryCalloc(1, size);
     colInfoDataEnsureCapacity(&idata, 0, rowNum);
     taosArrayPush(res->pDataBlock, &idata);
-    
     SColumnInfoData *pColumn = (SColumnInfoData *)taosArrayGetLast(res->pDataBlock);
     for (int32_t i = 0; i < rowNum; ++i) {
       colDataAppend(pColumn, i, (const char *)value, false);

source/libs/sync/src/syncRaftLog.c

@@ -57,7 +57,13 @@ int32_t logStoreAppendEntry(SSyncLogStore* pLogStore, SSyncRaftEntry* pEntry) {
   syncMeta.seqNum = pEntry->seqNum;
   syncMeta.term = pEntry->term;
   code = walWriteWithSyncInfo(pWal, pEntry->index, pEntry->originalRpcType, syncMeta, pEntry->data, pEntry->dataLen);
-  assert(code == 0);
+  if (code != 0) {
+    int32_t err = terrno;
+    const char *errStr = tstrerror(err);
+    sError("walWriteWithSyncInfo error, err:%d, msg:%s", err, errStr);
+    ASSERT(0);
+  }    
+  //assert(code == 0);
 
   walFsync(pWal, true);
   return code;
@@ -69,7 +75,14 @@ SSyncRaftEntry* logStoreGetEntry(SSyncLogStore* pLogStore, SyncIndex index) {
 
   if (index >= SYNC_INDEX_BEGIN && index <= logStoreLastIndex(pLogStore)) {
     SWalReadHandle* pWalHandle = walOpenReadHandle(pWal);
-    assert(walReadWithHandle(pWalHandle, index) == 0);
+    int32_t code = walReadWithHandle(pWalHandle, index);
+    if (code != 0) {
+      int32_t err = terrno;
+      const char *errStr = tstrerror(err);
+      sError("walReadWithHandle error, err:%d, msg:%s", err, errStr);
+      ASSERT(0);
+    }    
+    //assert(walReadWithHandle(pWalHandle, index) == 0);
 
     SSyncRaftEntry* pEntry = syncEntryBuild(pWalHandle->pHead->head.bodyLen);
     assert(pEntry != NULL);

source/libs/transport/src/transSrv.c

@@ -784,6 +784,11 @@ static void uvDestroyConn(uv_handle_t* handle) {
   tDebug("server conn %p destroy", conn);
   // uv_timer_stop(&conn->pTimer);
   transQueueDestroy(&conn->srvMsgs);
+
+  if (conn->regArg.init == 1) {
+    transFreeMsg(conn->regArg.msg.pCont);
+    conn->regArg.init = 0;
+  }
   QUEUE_REMOVE(&conn->queue);
   taosMemoryFree(conn->pTcp);
   taosMemoryFree(conn);

source/libs/wal/src/walRead.c

@@ -49,7 +49,10 @@ void walCloseReadHandle(SWalReadHandle *pRead) {
   taosMemoryFree(pRead);
 }
 
-int32_t walRegisterRead(SWalReadHandle *pRead, int64_t ver) { return 0; }
+int32_t walRegisterRead(SWalReadHandle *pRead, int64_t ver) {
+  // TODO
+  return 0;
+}
 
 static int32_t walReadSeekFilePos(SWalReadHandle *pRead, int64_t fileFirstVer, int64_t ver) {
   int code = 0;

source/libs/wal/src/walWrite.c

@@ -198,12 +198,14 @@ int walRoll(SWal *pWal) {
   if (pWal->pWriteIdxTFile != NULL) {
     code = taosCloseFile(&pWal->pWriteIdxTFile);
     if (code != 0) {
+      terrno = TAOS_SYSTEM_ERROR(errno);
       return -1;
     }
   }
   if (pWal->pWriteLogTFile != NULL) {
     code = taosCloseFile(&pWal->pWriteLogTFile);
     if (code != 0) {
+      terrno = TAOS_SYSTEM_ERROR(errno);
       return -1;
     }
   }
@@ -263,14 +265,19 @@ int64_t walWriteWithSyncInfo(SWal *pWal, int64_t index, tmsg_t msgType, SSyncLog
   if (index == pWal->vers.lastVer + 1) {
     if (taosArrayGetSize(pWal->fileInfoSet) == 0) {
       pWal->vers.firstVer = index;
-      code = walRoll(pWal);
+      if (walRoll(pWal) < 0) {
-      ASSERT(code == 0);
+        return -1;
+      }
     } else {
       int64_t passed = walGetSeq() - pWal->lastRollSeq;
       if (pWal->cfg.rollPeriod != -1 && pWal->cfg.rollPeriod != 0 && passed > pWal->cfg.rollPeriod) {
-        walRoll(pWal);
+        if (walRoll(pWal) < 0) {
+          return -1;
+        }
       } else if (pWal->cfg.segSize != -1 && pWal->cfg.segSize != 0 && walGetLastFileSize(pWal) > pWal->cfg.segSize) {
-        walRoll(pWal);
+        if (walRoll(pWal) < 0) {
+          return -1;
+        }
       }
     }
   } else {

source/util/src/terror.c

@@ -449,6 +449,11 @@ TAOS_DEFINE_ERROR(TSDB_CODE_PAR_TABLE_NOT_EXIST,          "Table does not exist"
 //planner
 TAOS_DEFINE_ERROR(TSDB_CODE_PLAN_INTERNAL_ERROR,          "planner internal error")
 
+//schemaless
+TAOS_DEFINE_ERROR(TSDB_CODE_SML_INVALID_PROTOCOL_TYPE,    "Invalid line protocol type")
+TAOS_DEFINE_ERROR(TSDB_CODE_SML_INVALID_PRECISION_TYPE,   "Invalid timestamp precision type")
+TAOS_DEFINE_ERROR(TSDB_CODE_SML_INVALID_DATA,             "Invalid data type")
+
 #ifdef TAOS_ERROR_C
 };
 #endif

tests/script/tsim/query/scalarFunction.sim

@@ -102,20 +102,26 @@ print ====> $data60 $data61 $data62 $data63 $data64 $data65
 print ====> $data70 $data71 $data72 $data73 $data74 $data75
 print ====> $data80 $data81 $data82 $data83 $data84 $data85
 print ====> $data90 $data91 $data92 $data93 $data94 $data95
+print ====> rows = $rows and rowNum = $rowNum for ct1
 if $rows != $rowNum then
   return -1
 endi
+
 print ====> select c1, abs(c1), c2, abs(c2), c3, abs(c3) from stb
 sql select c1, abs(c1), c2, abs(c2), c3, abs(c3) from stb
+print ====> rows = $rows and totalRows = $totalRows for stb
 if $rows != $totalRows then
   return -1
 endi
+
 print ====> select c1, abs(c1), c2, abs(c2), c3, abs(c3) from ntb
 sql select c1, abs(c1), c2, abs(c2), c3, abs(c3) from ntb
+print ====> rows = $rows and rowNum = $rowNum for ntb
 if $rows != $rowNum then
   return -1
 endi
 
+
 print ====> log
 sql select c1, log(c1, 10), c2, log(c2, 10), c3, log(c3, 10) from ct1
 print ====> select c1, log(c1, 10), c2, log(c2, 10), c3, log(c3, 10) from ct1

tools/shell/src/shellCommand.c

@@ -151,8 +151,8 @@ void shellGetPrevCharSize(const char *str, int32_t pos, int32_t *size, int32_t *
     if (str[pos] > 0 || shellCountPrefixOnes((uint8_t)str[pos]) > 1) break;
   }
 
-  int32_t rc = taosMbToWchar(&wc, str + pos, MB_CUR_MAX);
+  taosMbToWchar(&wc, str + pos, MB_CUR_MAX);
-  assert(rc == *size);
+  // assert(rc == *size); // it will be core, if str is encode by utf8 and taos charset is gbk
 
   *width = taosWcharWidth(wc);
 }

tools/shell/src/shellEngine.c

@@ -118,66 +118,27 @@ int32_t shellRunCommand(char *command) {
     }
   }
 
-  bool esc = false;
+  char quote = 0, *cmd = command;
-  char quote = 0, *cmd = command, *p = command;
   for (char c = *command++; c != 0; c = *command++) {
-    if (esc) {
+    if (c == '\\' && (*command == '\'' || *command == '"' || *command == '`')) {
-      switch (c) {
+      command ++;
-        case 'n':
-          c = '\n';
-          break;
-        case 'r':
-          c = '\r';
-          break;
-        case 't':
-          c = '\t';
-          break;
-        case 'G':
-          *p++ = '\\';
-          break;
-        case '\'':
-        case '"':
-          if (quote) {
-            *p++ = '\\';
-          }
-          break;
-        default:
-          *p++ = '\\';
-          break;
-      }
-      *p++ = c;
-      esc = false;
       continue;
     }
 
-    if (c == '\\') {
-      if (quote != 0 && (*command == '_' || *command == '\\')) {
-        // DO nothing
-      } else {
-        esc = true;
-        continue;
-      }
-    }
-
     if (quote == c) {
       quote = 0;
-    } else if (quote == 0 && (c == '\'' || c == '"')) {
+    } else if (quote == 0 && (c == '\'' || c == '"' || c == '`')) {
       quote = c;
-    }
+    } else if (c == ';' && quote == 0) {
-
+      c = *command;
-    *p++ = c;
+      *command = 0;
-    if (c == ';' && quote == 0) {
-      c = *p;
-      *p = 0;
       if (shellRunSingleCommand(cmd) < 0) {
         return -1;
       }
-      *p = c;
+      *command = c;
-      p = cmd;
+      cmd = command;
     }
   }
-
-  *p = 0;
   return shellRunSingleCommand(cmd);
 }
 

tools/taos-tools

@@ -1 +1 @@
-Subproject commit 59e0ebaf4905e4cb6d95a01c58b3fa507abc5a20
+Subproject commit 2f3dfddd4d9a869e706ba3cf98fb6d769404cd7c