Merge branch '3.0' into feature/tdb-merge

.gitignore

@@ -88,6 +88,7 @@ tests/examples/JDBC/JDBCDemo/.classpath
 tests/examples/JDBC/JDBCDemo/.project
 tests/examples/JDBC/JDBCDemo/.settings/
 source/libs/parser/inc/sql.*
+tests/script/tmqResult.txt
 
 # Emacs
 # -*- mode: gitignore; -*-

contrib/CMakeLists.txt

@@ -83,11 +83,6 @@ if(${BUILD_WITH_NURAFT})
     cat("${CMAKE_SUPPORT_DIR}/nuraft_CMakeLists.txt.in" ${CONTRIB_TMP_FILE})
 endif(${BUILD_WITH_NURAFT})
 
-# iconv
-if(${BUILD_WITH_ICONV})
-    cat("${CMAKE_SUPPORT_DIR}/iconv_CMakeLists.txt.in" ${CONTRIB_TMP_FILE})
-endif(${BUILD_WITH_ICONV})
-
 # download dependencies
 configure_file(${CONTRIB_TMP_FILE} "${CMAKE_CONTRIB_DIR}/deps-download/CMakeLists.txt")
 execute_process(COMMAND "${CMAKE_COMMAND}" -G "${CMAKE_GENERATOR}" .
@@ -213,10 +208,9 @@ endif(${BUILD_WITH_TRAFT})
 
 # LIBUV
 if(${BUILD_WITH_UV})
-    if (${TD_WINDOWS})
-        file(READ "libuv/include/uv.h" CONTENTS)
-        string(REGEX REPLACE "/([\r]*)\nstruct uv_tcp_s {" "/\\1\ntypedef BOOL (PASCAL *LPFN_CONNECTEX) (SOCKET s, const struct sockaddr* name, int namelen, PVOID lpSendBuffer, DWORD dwSendDataLength,LPDWORD lpdwBytesSent, LPOVERLAPPED lpOverlapped);\\1\nstruct uv_tcp_s {" CONTENTS_NEW "${CONTENTS}")
-        file(WRITE "libuv/include/uv.h" "${CONTENTS_NEW}")
+    if (NOT ${CMAKE_SYSTEM_NAME} MATCHES "Windows")
+        MESSAGE("Windows need set no-sign-compare")
+        add_compile_options(-Wno-sign-compare)
     endif ()
     add_subdirectory(libuv)
 endif(${BUILD_WITH_UV})
@@ -249,15 +243,7 @@ if(${BUILD_WITH_SQLITE})
 endif(${BUILD_WITH_SQLITE})
 
 # pthread
-if(${BUILD_PTHREAD})
-    add_definitions(-DPTW32_STATIC_LIB)
-    add_subdirectory(pthread)
-endif(${BUILD_PTHREAD})
-
-# iconv
-if(${BUILD_WITH_ICONV})
-    add_subdirectory(iconv)
-endif(${BUILD_WITH_ICONV})
+
 
 # ================================================================================================
 # Build test

example/CMakeLists.txt

-aux_source_directory(src TMQ_DEMO_SRC)
+add_executable(tmq "")
+add_executable(tstream "")
 
-add_executable(tmq ${TMQ_DEMO_SRC})
-target_link_libraries(
-    tmq taos
+target_sources(tmq
+    PRIVATE
+    "src/tmq.c"
 )
-target_include_directories(
-    tmq
+
+target_sources(tstream
+    PRIVATE
+    "src/tstream.c"
+)
+target_link_libraries(tmq
+    taos
+)
+
+target_link_libraries(tstream
+    taos
+)
+
+target_include_directories(tmq
+    PRIVATE "${CMAKE_CURRENT_SOURCE_DIR}/inc"
+)
+
+target_include_directories(tstream
     PRIVATE "${CMAKE_CURRENT_SOURCE_DIR}/inc"
 )
 
 SET_TARGET_PROPERTIES(tmq PROPERTIES OUTPUT_NAME tmq)
+SET_TARGET_PROPERTIES(tstream PROPERTIES OUTPUT_NAME tstream)

example/src/tstream.c

+/*
+ * 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 <assert.h>
+#include <stdio.h>
+#include <string.h>
+#include <time.h>
+#include "taos.h"
+
+int32_t init_env() {
+  TAOS* pConn = taos_connect("localhost", "root", "taosdata", NULL, 0);
+  if (pConn == NULL) {
+    return -1;
+  }
+
+  TAOS_RES* pRes = taos_query(pConn, "create database if not exists abc1 vgroups 2");
+  if (taos_errno(pRes) != 0) {
+    printf("error in create db, reason:%s\n", taos_errstr(pRes));
+    return -1;
+  }
+  taos_free_result(pRes);
+
+  pRes = taos_query(pConn, "use abc1");
+  if (taos_errno(pRes) != 0) {
+    printf("error in use db, reason:%s\n", taos_errstr(pRes));
+    return -1;
+  }
+  taos_free_result(pRes);
+
+  pRes = taos_query(pConn, "create stable if not exists st1 (ts timestamp, k int) tags(a int)");
+  if (taos_errno(pRes) != 0) {
+    printf("failed to create super table st1, reason:%s\n", taos_errstr(pRes));
+    return -1;
+  }
+  taos_free_result(pRes);
+
+  pRes = taos_query(pConn, "create table if not exists tu1 using st1 tags(1)");
+  if (taos_errno(pRes) != 0) {
+    printf("failed to create child table tu1, reason:%s\n", taos_errstr(pRes));
+    return -1;
+  }
+  taos_free_result(pRes);
+
+  pRes = taos_query(pConn, "create table if not exists tu2 using st1 tags(2)");
+  if (taos_errno(pRes) != 0) {
+    printf("failed to create child table tu2, reason:%s\n", taos_errstr(pRes));
+    return -1;
+  }
+  taos_free_result(pRes);
+  return 0;
+}
+
+int32_t create_stream() {
+  printf("create stream\n");
+  TAOS_RES* pRes;
+  TAOS*     pConn = taos_connect("localhost", "root", "taosdata", NULL, 0);
+  if (pConn == NULL) {
+    return -1;
+  }
+
+  pRes = taos_query(pConn, "use abc1");
+  if (taos_errno(pRes) != 0) {
+    printf("error in use db, reason:%s\n", taos_errstr(pRes));
+    return -1;
+  }
+  taos_free_result(pRes);
+
+  /*const char* sql = "select min(k), max(k), sum(k) from tu1";*/
+  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);
+  if (taos_errno(pRes) != 0) {
+    printf("failed to create stream out1, reason:%s\n", taos_errstr(pRes));
+    return -1;
+  }
+  taos_free_result(pRes);
+  taos_close(pConn);
+  return 0;
+}
+
+int main(int argc, char* argv[]) {
+  int code;
+  if (argc > 1) {
+    printf("env init\n");
+    code = init_env();
+  }
+  create_stream();
+#if 0
+  tmq_t*      tmq = build_consumer();
+  tmq_list_t* topic_list = build_topic_list();
+  /*perf_loop(tmq, topic_list);*/
+  /*basic_consume_loop(tmq, topic_list);*/
+  sync_consume_loop(tmq, topic_list);
+#endif
+}

include/client/taos.h

@@ -85,11 +85,7 @@ typedef struct taosField {
   int32_t bytes;
 } TAOS_FIELD;
 
-#ifdef _TD_GO_DLL_
-#define DLL_EXPORT __declspec(dllexport)
-#else
 #define DLL_EXPORT
-#endif
 
 typedef void (*__taos_async_fn_t)(void *param, TAOS_RES *, int code);
 
@@ -218,7 +214,6 @@ typedef void(tmq_commit_cb(tmq_t *, tmq_resp_err_t, tmq_topic_vgroup_list_t *, v
 DLL_EXPORT tmq_list_t *tmq_list_new();
 DLL_EXPORT int32_t     tmq_list_append(tmq_list_t *, const char *);
 
-DLL_EXPORT TAOS_RES   *tmq_create_topic(TAOS *taos, const char *name, const char *sql, int sqlLen);
 DLL_EXPORT tmq_t      *tmq_consumer_new(void *conn, tmq_conf_t *conf, char *errstr, int32_t errstrLen);
 DLL_EXPORT void        tmq_message_destroy(tmq_message_t *tmq_message);
 DLL_EXPORT const char *tmq_err2str(tmq_resp_err_t);
@@ -262,7 +257,12 @@ int32_t tmqGetSkipLogNum(tmq_message_t *tmq_message);
 DLL_EXPORT TAOS_ROW tmq_get_row(tmq_message_t *message);
 DLL_EXPORT char    *tmq_get_topic_name(tmq_message_t *message);
 
-/* ---------------------- OTHER ---------------------------- */
+/* --------------------TMPORARY INTERFACE FOR TESTING--------------------- */
+DLL_EXPORT TAOS_RES *tmq_create_topic(TAOS *taos, const char *name, const char *sql, int sqlLen);
+
+DLL_EXPORT TAOS_RES *tmq_create_stream(TAOS *taos, const char *streamName, const char *tbName, const char *sql);
+
+/* -------------------------------- OTHER -------------------------------- */
 typedef void (*TAOS_SUBSCRIBE_CALLBACK)(TAOS_SUB *tsub, TAOS_RES *res, void *param, int code);
 
 DLL_EXPORT int taos_stmt_affected_rows(TAOS_STMT *stmt);

include/common/taosdef.h

@@ -30,12 +30,13 @@ typedef int64_t tb_uid_t;
 #define IS_TSWINDOW_SPECIFIED(win) (((win).skey != INT64_MIN) || ((win).ekey != INT64_MAX))
 
 typedef enum {
-  TSDB_SUPER_TABLE = 1,   // super table
-  TSDB_CHILD_TABLE = 2,   // table created from super table
-  TSDB_NORMAL_TABLE = 3,  // ordinary table
-  TSDB_STREAM_TABLE = 4,  // table created by stream processing
-  TSDB_TEMP_TABLE = 5,    // temp table created by nest query
-  TSDB_TABLE_MAX = 6
+  TSDB_SUPER_TABLE  = 1,    // super table
+  TSDB_CHILD_TABLE  = 2,    // table created from super table
+  TSDB_NORMAL_TABLE = 3,    // ordinary table
+  TSDB_STREAM_TABLE = 4,    // table created from stream computing
+  TSDB_TEMP_TABLE   = 5,    // temp table created by nest query
+  TSDB_SYSTEM_TABLE = 6,
+  TSDB_TABLE_MAX    = 7
 } ETableType;
 
 typedef enum {
@@ -62,9 +63,12 @@ typedef enum {
 } ETsdbStatisStatus;
 
 typedef enum {
-  TSDB_SMA_STAT_OK = 0,       // ready to provide service
-  TSDB_SMA_STAT_EXPIRED = 1,  // not ready or expired
-} ETsdbSmaStat;
+  TSDB_SMA_STAT_UNKNOWN = -1,  // unknown
+  TSDB_SMA_STAT_OK = 0,        // ready to provide service
+  TSDB_SMA_STAT_EXPIRED = 1,   // not ready or expired
+  TSDB_SMA_STAT_DROPPED = 2,   // sma dropped
+} ETsdbSmaStat;                // bit operation
+
 
 typedef enum {
   TSDB_SMA_TYPE_BLOCK = 0,       // Block-wise SMA

include/common/tcommon.h

@@ -54,25 +54,21 @@ typedef struct SColumnDataAgg {
 } SColumnDataAgg;
 
 typedef struct SDataBlockInfo {
-  STimeWindow    window;
-  int32_t        rows;
-  int32_t        rowSize;
-  int16_t        numOfCols;
-  int16_t        hasVarCol;
-  union {int64_t uid; int64_t blockId;};
+  STimeWindow window;
+  int32_t     rows;
+  int32_t     rowSize;
+  int16_t     numOfCols;
+  int16_t     hasVarCol;
+  union {
+    int64_t uid;
+    int64_t blockId;
+  };
+  int64_t groupId;  // no need to serialize
 } SDataBlockInfo;
 
-//typedef struct SConstantItem {
-//  SColumnInfo info;
-//  int32_t     startRow;  // run-length-encoding to save the space for multiple rows
-//  int32_t     endRow;
-//  SVariant    value;
-//} SConstantItem;
-
-// info.numOfCols = taosArrayGetSize(pDataBlock) + taosArrayGetSize(pConstantList);
 typedef struct SSDataBlock {
-  SColumnDataAgg *pBlockAgg;
-  SArray         *pDataBlock;    // SArray<SColumnInfoData>
+  SColumnDataAgg* pBlockAgg;
+  SArray*         pDataBlock;  // SArray<SColumnInfoData>
   SDataBlockInfo  info;
 } SSDataBlock;
 
@@ -98,23 +94,40 @@ void*   blockDataDestroy(SSDataBlock* pBlock);
 int32_t tEncodeDataBlock(void** buf, const SSDataBlock* pBlock);
 void*   tDecodeDataBlock(const void* buf, SSDataBlock* pBlock);
 
-static FORCE_INLINE void tDeleteSSDataBlock(SSDataBlock* pBlock) {
-  if (pBlock == NULL) {
-    return;
+int32_t tEncodeDataBlocks(void** buf, const SArray* blocks);
+void*   tDecodeDataBlocks(const void* buf, SArray** blocks);
+
+static FORCE_INLINE void blockDestroyInner(SSDataBlock* pBlock) {
+  // WARNING: do not use info.numOfCols,
+  // sometimes info.numOfCols != array size
+  int32_t numOfOutput = taosArrayGetSize(pBlock->pDataBlock);
+  for (int32_t i = 0; i < numOfOutput; ++i) {
+    SColumnInfoData* pColInfoData = (SColumnInfoData*)taosArrayGet(pBlock->pDataBlock, i);
+    if (IS_VAR_DATA_TYPE(pColInfoData->info.type)) {
+      taosMemoryFreeClear(pColInfoData->varmeta.offset);
+    } else {
+      taosMemoryFreeClear(pColInfoData->nullbitmap);
+    }
+
+    taosMemoryFreeClear(pColInfoData->pData);
   }
-  blockDataDestroy(pBlock);
+
+  taosArrayDestroy(pBlock->pDataBlock);
+  taosMemoryFreeClear(pBlock->pBlockAgg);
 }
 
+static FORCE_INLINE void tDeleteSSDataBlock(SSDataBlock* pBlock) { blockDestroyInner(pBlock); }
+
 static FORCE_INLINE int32_t tEncodeSMqPollRsp(void** buf, const SMqPollRsp* pRsp) {
   int32_t tlen = 0;
   int32_t sz = 0;
-  tlen += taosEncodeFixedI64(buf, pRsp->consumerId);
+  // tlen += taosEncodeFixedI64(buf, pRsp->consumerId);
   tlen += taosEncodeFixedI64(buf, pRsp->reqOffset);
   tlen += taosEncodeFixedI64(buf, pRsp->rspOffset);
   tlen += taosEncodeFixedI32(buf, pRsp->skipLogNum);
   tlen += taosEncodeFixedI32(buf, pRsp->numOfTopics);
   if (pRsp->numOfTopics == 0) return tlen;
-  tlen += tEncodeSSchemaWrapper(buf, pRsp->schemas);
+  tlen += tEncodeSSchemaWrapper(buf, pRsp->schema);
   if (pRsp->pBlockData) {
     sz = taosArrayGetSize(pRsp->pBlockData);
   }
@@ -128,15 +141,15 @@ static FORCE_INLINE int32_t tEncodeSMqPollRsp(void** buf, const SMqPollRsp* pRsp
 
 static FORCE_INLINE void* tDecodeSMqPollRsp(void* buf, SMqPollRsp* pRsp) {
   int32_t sz;
-  buf = taosDecodeFixedI64(buf, &pRsp->consumerId);
+  // buf = taosDecodeFixedI64(buf, &pRsp->consumerId);
   buf = taosDecodeFixedI64(buf, &pRsp->reqOffset);
   buf = taosDecodeFixedI64(buf, &pRsp->rspOffset);
   buf = taosDecodeFixedI32(buf, &pRsp->skipLogNum);
   buf = taosDecodeFixedI32(buf, &pRsp->numOfTopics);
   if (pRsp->numOfTopics == 0) return buf;
-  pRsp->schemas = (SSchemaWrapper*)calloc(1, sizeof(SSchemaWrapper));
-  if (pRsp->schemas == NULL) return NULL;
-  buf = tDecodeSSchemaWrapper(buf, pRsp->schemas);
+  pRsp->schema = (SSchemaWrapper*)taosMemoryCalloc(1, sizeof(SSchemaWrapper));
+  if (pRsp->schema == NULL) return NULL;
+  buf = tDecodeSSchemaWrapper(buf, pRsp->schema);
   buf = taosDecodeFixedI32(buf, &sz);
   pRsp->pBlockData = taosArrayInit(sz, sizeof(SSDataBlock));
   for (int32_t i = 0; i < sz; i++) {
@@ -148,13 +161,13 @@ static FORCE_INLINE void* tDecodeSMqPollRsp(void* buf, SMqPollRsp* pRsp) {
 }
 
 static FORCE_INLINE void tDeleteSMqConsumeRsp(SMqPollRsp* pRsp) {
-  if (pRsp->schemas) {
-    if (pRsp->schemas->nCols) {
-      tfree(pRsp->schemas->pSchema);
+  if (pRsp->schema) {
+    if (pRsp->schema->nCols) {
+      taosMemoryFreeClear(pRsp->schema->pSchema);
     }
-    free(pRsp->schemas);
+    taosMemoryFree(pRsp->schema);
   }
-  taosArrayDestroyEx(pRsp->pBlockData, (void (*)(void*))tDeleteSSDataBlock);
+  taosArrayDestroyEx(pRsp->pBlockData, (void (*)(void*))blockDestroyInner);
   pRsp->pBlockData = NULL;
 }
 
@@ -166,10 +179,8 @@ typedef struct SColumn {
     int64_t  dataBlockId;
   };
 
-  union {
-    int16_t colId;
-    int16_t slotId;
-  };
+  int16_t colId;
+  int16_t slotId;
 
   char    name[TSDB_COL_NAME_LEN];
   int8_t  flag;  // column type: normal column, tag, or user-input column (integer/float/string)
@@ -196,7 +207,7 @@ typedef struct SGroupbyExpr {
 
 typedef struct SFunctParam {
   int32_t  type;
-  SColumn *pCol;
+  SColumn* pCol;
   SVariant param;
 } SFunctParam;
 
@@ -214,12 +225,12 @@ typedef struct SResSchame {
 typedef struct SExprBasicInfo {
   SResSchema   resSchema;
   int16_t      numOfParams;  // argument value of each function
-  SFunctParam *pParam;
+  SFunctParam* pParam;
 } SExprBasicInfo;
 
 typedef struct SExprInfo {
-  struct SExprBasicInfo  base;
-  struct tExprNode      *pExpr;
+  struct SExprBasicInfo base;
+  struct tExprNode*     pExpr;
 } SExprInfo;
 
 typedef struct SStateWindow {

include/common/tdatablock.h

@@ -104,6 +104,7 @@ static FORCE_INLINE bool colDataIsNull(const SColumnInfoData* pColumnInfoData, u
 int32_t colDataAppend(SColumnInfoData* pColumnInfoData, uint32_t currentRow, const char* pData, bool isNull);
 int32_t colDataMergeCol(SColumnInfoData* pColumnInfoData, uint32_t numOfRow1, const SColumnInfoData* pSource,
                         uint32_t numOfRow2);
+int32_t colDataAssign(SColumnInfoData* pColumnInfoData, const SColumnInfoData* pSource, int32_t numOfRows);
 int32_t blockDataUpdateTsWindow(SSDataBlock* pDataBlock);
 
 int32_t colDataGetLength(const SColumnInfoData* pColumnInfoData, int32_t numOfRows);
@@ -112,16 +113,16 @@ void    colDataTrim(SColumnInfoData* pColumnInfoData);
 size_t blockDataGetNumOfCols(const SSDataBlock* pBlock);
 size_t blockDataGetNumOfRows(const SSDataBlock* pBlock);
 
-int32_t      blockDataMerge(SSDataBlock* pDest, const SSDataBlock* pSrc);
-int32_t      blockDataSplitRows(SSDataBlock* pBlock, bool hasVarCol, int32_t startIndex, int32_t* stopIndex,
-                                int32_t pageSize);
-SSDataBlock* blockDataExtractBlock(SSDataBlock* pBlock, int32_t startIndex, int32_t rowCount);
-
+int32_t blockDataMerge(SSDataBlock* pDest, const SSDataBlock* pSrc);
+int32_t blockDataSplitRows(SSDataBlock* pBlock, bool hasVarCol, int32_t startIndex, int32_t* stopIndex,
+                           int32_t pageSize);
 int32_t blockDataToBuf(char* buf, const SSDataBlock* pBlock);
 int32_t blockDataFromBuf(SSDataBlock* pBlock, const char* buf);
 
+SSDataBlock* blockDataExtractBlock(SSDataBlock* pBlock, int32_t startIndex, int32_t rowCount);
+
 size_t blockDataGetSize(const SSDataBlock* pBlock);
-size_t blockDataGetRowSize(const SSDataBlock* pBlock);
+size_t blockDataGetRowSize(SSDataBlock* pBlock);
 double blockDataGetSerialRowSize(const SSDataBlock* pBlock);
 size_t blockDataGetSerialMetaSize(const SSDataBlock* pBlock);
 
@@ -132,11 +133,13 @@ int32_t blockDataSort_rv(SSDataBlock* pDataBlock, SArray* pOrderInfo, bool nullF
 
 int32_t      blockDataEnsureColumnCapacity(SColumnInfoData* pColumn, uint32_t numOfRows);
 int32_t      blockDataEnsureCapacity(SSDataBlock* pDataBlock, uint32_t numOfRows);
-void         blockDataClearup(SSDataBlock* pDataBlock);
+void         blockDataCleanup(SSDataBlock* pDataBlock);
 SSDataBlock* createOneDataBlock(const SSDataBlock* pDataBlock);
 size_t       blockDataGetCapacityInRow(const SSDataBlock* pBlock, size_t pageSize);
 void*        blockDataDestroy(SSDataBlock* pBlock);
 
+void blockDebugShowData(const SArray* dataBlocks);
+
 #ifdef __cplusplus
 }
 #endif

include/common/tdataformat.h

@@ -93,7 +93,7 @@ typedef struct {
 #define schemaFLen(s)     ((s)->flen)
 #define schemaVLen(s)     ((s)->vlen)
 #define schemaColAt(s, i) ((s)->columns + i)
-#define tdFreeSchema(s)   tfree((s))
+#define tdFreeSchema(s)   taosMemoryFreeClear((s))
 
 STSchema *tdDupSchema(const STSchema *pSchema);
 int32_t   tdEncodeSchema(void **buf, STSchema *pSchema);
@@ -493,7 +493,7 @@ typedef struct {
 #define kvRowCpy(dst, r)       memcpy((dst), (r), kvRowLen(r))
 #define kvRowColVal(r, colIdx) POINTER_SHIFT(kvRowValues(r), (colIdx)->offset)
 #define kvRowColIdxAt(r, i)    (kvRowColIdx(r) + (i))
-#define kvRowFree(r)           tfree(r)
+#define kvRowFree(r)           taosMemoryFreeClear(r)
 #define kvRowEnd(r)            POINTER_SHIFT(r, kvRowLen(r))
 #define kvRowValLen(r)         (kvRowLen(r) - TD_KV_ROW_HEAD_SIZE - sizeof(SColIdx) * kvRowNCols(r))
 #define kvRowTKey(r)           (*(TKEY *)(kvRowValues(r)))
@@ -593,7 +593,7 @@ SKVRow  tdGetKVRowFromBuilder(SKVRowBuilder *pBuilder);
 static FORCE_INLINE int32_t tdAddColToKVRow(SKVRowBuilder *pBuilder, int16_t colId, int8_t type, const void *value) {
   if (pBuilder->nCols >= pBuilder->tCols) {
     pBuilder->tCols *= 2;
-    SColIdx *pColIdx = (SColIdx *)realloc((void *)(pBuilder->pColIdx), sizeof(SColIdx) * pBuilder->tCols);
+    SColIdx *pColIdx = (SColIdx *)taosMemoryRealloc((void *)(pBuilder->pColIdx), sizeof(SColIdx) * pBuilder->tCols);
     if (pColIdx == NULL) return -1;
     pBuilder->pColIdx = pColIdx;
   }
@@ -608,7 +608,7 @@ static FORCE_INLINE int32_t tdAddColToKVRow(SKVRowBuilder *pBuilder, int16_t col
     while (tlen > pBuilder->alloc - pBuilder->size) {
       pBuilder->alloc *= 2;
     }
-    void *buf = realloc(pBuilder->buf, pBuilder->alloc);
+    void *buf = taosMemoryRealloc(pBuilder->buf, pBuilder->alloc);
     if (buf == NULL) return -1;
     pBuilder->buf = buf;
   }

include/common/tglobal.h

@@ -51,6 +51,7 @@ extern int32_t tsCompatibleModel;
 extern bool    tsEnableSlaveQuery;
 extern bool    tsPrintAuth;
 extern int64_t tsTickPerDay[3];
+extern int32_t tsMultiProcess;
 
 // monitor
 extern bool     tsEnableMonitor;

include/common/tmsg.h

@@ -23,6 +23,7 @@
 #include "tencode.h"
 #include "thash.h"
 #include "tlist.h"
+#include "tname.h"
 #include "trow.h"
 #include "tuuid.h"
 
@@ -108,6 +109,7 @@ typedef enum _mgmt_table {
   TSDB_MGMT_TABLE_STREAMTABLES,
   TSDB_MGMT_TABLE_TP,
   TSDB_MGMT_TABLE_FUNC,
+  TSDB_MGMT_TABLE_INDEX,
   TSDB_MGMT_TABLE_MAX,
 } EShowType;
 
@@ -182,6 +184,13 @@ typedef struct SField {
   int32_t bytes;
 } SField;
 
+typedef struct SRetention {
+  int32_t freq;
+  int32_t keep;
+  int8_t  freqUnit;
+  int8_t  keepUnit;
+} SRetention;
+
 #pragma pack(push, 1)
 
 // null-terminated string instead of char array to avoid too many memory consumption in case of more than 1M tableMeta
@@ -192,21 +201,13 @@ typedef struct SEp {
 
 typedef struct {
   int32_t contLen;
-  union {
-    int32_t vgId;
-    int32_t streamTaskId;
-  };
+  int32_t vgId;
 } SMsgHead;
 
-typedef struct {
-  int32_t workerType;
-  int32_t streamTaskId;
-} SStreamExecMsgHead;
-
 // Submit message for one table
 typedef struct SSubmitBlk {
   int64_t uid;        // table unique id
-  int32_t tid;        // table id
+  int64_t suid;       // stable id
   int32_t padding;    // TODO just for padding here
   int32_t sversion;   // data schema version
   int32_t dataLen;    // data part length, not including the SSubmitBlk head
@@ -231,12 +232,12 @@ typedef struct {
 } SSubmitBlkIter;
 
 typedef struct {
-  int32_t totalLen;
-  int32_t len;
-  void*   pMsg;
+  int32_t     totalLen;
+  int32_t     len;
+  const void* pMsg;
 } SSubmitMsgIter;
 
-int32_t tInitSubmitMsgIter(SSubmitReq* pMsg, SSubmitMsgIter* pIter);
+int32_t tInitSubmitMsgIter(const SSubmitReq* pMsg, SSubmitMsgIter* pIter);
 int32_t tGetSubmitMsgNext(SSubmitMsgIter* pIter, SSubmitBlk** pPBlock);
 int32_t tInitSubmitBlkIter(SSubmitBlk* pBlock, SSubmitBlkIter* pIter);
 STSRow* tGetSubmitBlkNext(SSubmitBlkIter* pIter);
@@ -267,11 +268,18 @@ typedef struct SSchema {
 typedef struct {
   char    name[TSDB_TABLE_FNAME_LEN];
   int8_t  igExists;
+  float   xFilesFactor;
+  int32_t aggregationMethod;
+  int32_t delay;
+  int32_t ttl;
   int32_t numOfColumns;
   int32_t numOfTags;
-  SArray* pColumns;
-  SArray* pTags;
-  char    comment[TSDB_STB_COMMENT_LEN];
+  int32_t numOfSmas;
+  int32_t commentLen;
+  SArray* pColumns;  // array of SField
+  SArray* pTags;     // array of SField
+  SArray* pSmas;     // array of SField
+  char*   comment;
 } SMCreateStbReq;
 
 int32_t tSerializeSMCreateStbReq(void* buf, int32_t bufLen, SMCreateStbReq* pReq);
@@ -461,7 +469,9 @@ typedef struct {
   int32_t tz;  // query client timezone
   char    intervalUnit;
   char    slidingUnit;
-  char    offsetUnit;
+  char
+      offsetUnit;  // TODO Remove it, the offset is the number of precision tickle, and it must be a immutable duration.
+  int8_t  precision;
   int64_t interval;
   int64_t sliding;
   int64_t offset;
@@ -471,6 +481,10 @@ typedef struct {
   int32_t code;
 } SQueryTableRsp;
 
+int32_t tSerializeSQueryTableRsp(void* buf, int32_t bufLen, SQueryTableRsp* pRsp);
+
+int32_t tDeserializeSQueryTableRsp(void* buf, int32_t bufLen, SQueryTableRsp* pRsp);
+
 typedef struct {
   char    db[TSDB_DB_FNAME_LEN];
   int32_t numOfVgroups;
@@ -493,10 +507,13 @@ typedef struct {
   int8_t  cacheLastRow;
   int8_t  ignoreExist;
   int8_t  streamMode;
+  int32_t numOfRetensions;
+  SArray* pRetensions;  // SRetention
 } SCreateDbReq;
 
 int32_t tSerializeSCreateDbReq(void* buf, int32_t bufLen, SCreateDbReq* pReq);
 int32_t tDeserializeSCreateDbReq(void* buf, int32_t bufLen, SCreateDbReq* pReq);
+void    tFreeSCreateDbReq(SCreateDbReq* pReq);
 
 typedef struct {
   char    db[TSDB_DB_FNAME_LEN];
@@ -697,6 +714,7 @@ typedef struct {
 
 int32_t tSerializeSStatusRsp(void* buf, int32_t bufLen, SStatusRsp* pRsp);
 int32_t tDeserializeSStatusRsp(void* buf, int32_t bufLen, SStatusRsp* pRsp);
+void    tFreeSStatusRsp(SStatusRsp* pRsp);
 
 typedef struct {
   int32_t reserved;
@@ -740,11 +758,13 @@ typedef struct {
   int8_t   selfIndex;
   int8_t   streamMode;
   SReplica replicas[TSDB_MAX_REPLICA];
-
+  int32_t  numOfRetensions;
+  SArray*  pRetensions;  // SRetention
 } SCreateVnodeReq, SAlterVnodeReq;
 
 int32_t tSerializeSCreateVnodeReq(void* buf, int32_t bufLen, SCreateVnodeReq* pReq);
 int32_t tDeserializeSCreateVnodeReq(void* buf, int32_t bufLen, SCreateVnodeReq* pReq);
+int32_t tFreeSCreateVnodeReq(SCreateVnodeReq* pReq);
 
 typedef struct {
   int32_t vgId;
@@ -862,6 +882,7 @@ void    tFreeSShowRsp(SShowRsp* pRsp);
 typedef struct {
   int32_t type;
   char    db[TSDB_DB_FNAME_LEN];
+  char    tb[TSDB_TABLE_NAME_LEN];
   int64_t showId;
   int8_t  free;
 } SRetrieveTableReq;
@@ -879,6 +900,17 @@ typedef struct {
   char    data[];
 } SRetrieveTableRsp;
 
+typedef struct {
+  int64_t handle;
+  int64_t useconds;
+  int8_t  completed;  // all results are returned to client
+  int8_t  precision;
+  int8_t  compressed;
+  int32_t compLen;
+  int32_t numOfRows;
+  char    data[];
+} SRetrieveMetaTableRsp;
+
 typedef struct {
   char    fqdn[TSDB_FQDN_LEN];  // end point, hostname:port
   int32_t port;
@@ -1093,7 +1125,6 @@ int32_t tDeserializeSSchedulerHbReq(void* buf, int32_t bufLen, SSchedulerHbReq*
 void    tFreeSSchedulerHbReq(SSchedulerHbReq* pReq);
 
 typedef struct {
-  uint64_t       seqId;
   SQueryNodeEpId epId;
   SArray*        taskStatus;  // SArray<STaskStatus>
 } SSchedulerHbRsp;
@@ -1128,10 +1159,10 @@ typedef struct {
 
 typedef struct {
   char   name[TSDB_TOPIC_FNAME_LEN];
+  char   outputSTbName[TSDB_TABLE_FNAME_LEN];
   int8_t igExists;
   char*  sql;
-  char*  physicalPlan;
-  char*  logicalPlan;
+  char*  ast;
 } SCMCreateStreamReq;
 
 typedef struct {
@@ -1277,11 +1308,11 @@ typedef struct {
 } SMqRebSubscribe;
 
 static FORCE_INLINE SMqRebSubscribe* tNewSMqRebSubscribe(const char* key) {
-  SMqRebSubscribe* pRebSub = (SMqRebSubscribe*)calloc(1, sizeof(SMqRebSubscribe));
+  SMqRebSubscribe* pRebSub = (SMqRebSubscribe*)taosMemoryCalloc(1, sizeof(SMqRebSubscribe));
   if (pRebSub == NULL) {
     goto _err;
   }
-  pRebSub->key = key;
+  pRebSub->key = strdup(key);
   pRebSub->lostConsumers = taosArrayInit(0, sizeof(int64_t));
   if (pRebSub->lostConsumers == NULL) {
     goto _err;
@@ -1299,7 +1330,7 @@ _err:
   taosArrayDestroy(pRebSub->lostConsumers);
   taosArrayDestroy(pRebSub->removedConsumers);
   taosArrayDestroy(pRebSub->newConsumers);
-  tfree(pRebSub);
+  taosMemoryFreeClear(pRebSub);
   return NULL;
 }
 
@@ -1344,33 +1375,54 @@ typedef struct {
   int64_t  tuid;
 } SDDropTopicReq;
 
+typedef struct {
+  float    xFilesFactor;
+  int8_t   delayUnit;
+  int8_t   nFuncIds;
+  int32_t* pFuncIds;
+  int64_t  delay;
+} SRSmaParam;
+
 typedef struct SVCreateTbReq {
   int64_t  ver;  // use a general definition
+  char*    dbFName;
   char*    name;
   uint32_t ttl;
   uint32_t keep;
-  uint8_t  type;
   union {
+    uint8_t info;
     struct {
-      tb_uid_t suid;
-      uint32_t nCols;
-      SSchema* pSchema;
-      uint32_t nTagCols;
-      SSchema* pTagSchema;
+      uint8_t rollup : 1;  // 1 means rollup sma
+      uint8_t type : 7;
+    };
+  };
+  union {
+    struct {
+      tb_uid_t    suid;
+      uint32_t    nCols;
+      SSchema*    pSchema;
+      uint32_t    nTagCols;
+      SSchema*    pTagSchema;
+      col_id_t    nBSmaCols;
+      col_id_t*   pBSmaCols;
+      SRSmaParam* pRSmaParam;
     } stbCfg;
     struct {
       tb_uid_t suid;
       SKVRow   pTag;
     } ctbCfg;
     struct {
-      uint32_t nCols;
-      SSchema* pSchema;
+      uint32_t    nCols;
+      SSchema*    pSchema;
+      col_id_t    nBSmaCols;
+      col_id_t*   pBSmaCols;
+      SRSmaParam* pRSmaParam;
     } ntbCfg;
   };
 } SVCreateTbReq, SVUpdateTbReq;
 
 typedef struct {
-  int tmp;  // TODO: to avoid compile error
+  int32_t code;
 } SVCreateTbRsp, SVUpdateTbRsp;
 
 int32_t tSerializeSVCreateTbReq(void** buf, SVCreateTbReq* pReq);
@@ -1381,12 +1433,15 @@ typedef struct {
   SArray* pArray;
 } SVCreateTbBatchReq;
 
+int32_t tSerializeSVCreateTbBatchReq(void** buf, SVCreateTbBatchReq* pReq);
+void*   tDeserializeSVCreateTbBatchReq(void* buf, SVCreateTbBatchReq* pReq);
+
 typedef struct {
-  int tmp;  // TODO: to avoid compile error
+  SArray* rspList;  // SArray<SVCreateTbRsp>
 } SVCreateTbBatchRsp;
 
-int32_t tSerializeSVCreateTbBatchReq(void** buf, SVCreateTbBatchReq* pReq);
-void*   tDeserializeSVCreateTbBatchReq(void* buf, SVCreateTbBatchReq* pReq);
+int32_t tSerializeSVCreateTbBatchRsp(void* buf, int32_t bufLen, SVCreateTbBatchRsp* pRsp);
+int32_t tDeserializeSVCreateTbBatchRsp(void* buf, int32_t bufLen, SVCreateTbBatchRsp* pRsp);
 
 typedef struct {
   int64_t  ver;
@@ -1576,7 +1631,7 @@ static FORCE_INLINE void tFreeReqKvHash(SHashObj* info) {
   void* pIter = taosHashIterate(info, NULL);
   while (pIter != NULL) {
     SKv* kv = (SKv*)pIter;
-    tfree(kv->value);
+    taosMemoryFreeClear(kv->value);
     pIter = taosHashIterate(info, pIter);
   }
 }
@@ -1599,13 +1654,13 @@ static FORCE_INLINE void tFreeClientHbBatchReq(void* pReq, bool deep) {
   } else {
     taosArrayDestroy(req->reqs);
   }
-  free(pReq);
+  taosMemoryFree(pReq);
 }
 
 static FORCE_INLINE void tFreeClientKv(void* pKv) {
   SKv* kv = (SKv*)pKv;
   if (kv) {
-    tfree(kv->value);
+    taosMemoryFreeClear(kv->value);
   }
 }
 
@@ -1632,7 +1687,7 @@ static FORCE_INLINE int32_t tEncodeSKv(SCoder* pEncoder, const SKv* pKv) {
 static FORCE_INLINE int32_t tDecodeSKv(SCoder* pDecoder, SKv* pKv) {
   if (tDecodeI32(pDecoder, &pKv->key) < 0) return -1;
   if (tDecodeI32(pDecoder, &pKv->valueLen) < 0) return -1;
-  pKv->value = malloc(pKv->valueLen + 1);
+  pKv->value = taosMemoryMalloc(pKv->valueLen + 1);
   if (pKv->value == NULL) return -1;
   if (tDecodeCStrTo(pDecoder, (char*)pKv->value) < 0) return -1;
   return 0;
@@ -1886,7 +1941,7 @@ static FORCE_INLINE int32_t tEncodeSSchemaWrapper(void** buf, const SSchemaWrapp
 
 static FORCE_INLINE void* tDecodeSSchemaWrapper(void* buf, SSchemaWrapper* pSW) {
   buf = taosDecodeFixedU32(buf, &pSW->nCols);
-  pSW->pSchema = (SSchema*)calloc(pSW->nCols, sizeof(SSchema));
+  pSW->pSchema = (SSchema*)taosMemoryCalloc(pSW->nCols, sizeof(SSchema));
   if (pSW->pSchema == NULL) {
     return NULL;
   }
@@ -1896,12 +1951,47 @@ static FORCE_INLINE void* tDecodeSSchemaWrapper(void* buf, SSchemaWrapper* pSW)
   }
   return buf;
 }
+
+typedef struct {
+  char    name[TSDB_TABLE_FNAME_LEN];
+  char    stb[TSDB_TABLE_FNAME_LEN];
+  int8_t  igExists;
+  int8_t  intervalUnit;
+  int8_t  slidingUnit;
+  int8_t  timezone;
+  int32_t dstVgId;  // for stream
+  int64_t interval;
+  int64_t offset;
+  int64_t sliding;
+  int32_t exprLen;        // strlen + 1
+  int32_t tagsFilterLen;  // strlen + 1
+  int32_t sqlLen;         // strlen + 1
+  int32_t astLen;         // strlen + 1
+  char*   expr;
+  char*   tagsFilter;
+  char*   sql;
+  char*   ast;
+} SMCreateSmaReq;
+
+int32_t tSerializeSMCreateSmaReq(void* buf, int32_t bufLen, SMCreateSmaReq* pReq);
+int32_t tDeserializeSMCreateSmaReq(void* buf, int32_t bufLen, SMCreateSmaReq* pReq);
+void    tFreeSMCreateSmaReq(SMCreateSmaReq* pReq);
+
+typedef struct {
+  char   name[TSDB_TABLE_FNAME_LEN];
+  int8_t igNotExists;
+} SMDropSmaReq;
+
+int32_t tSerializeSMDropSmaReq(void* buf, int32_t bufLen, SMDropSmaReq* pReq);
+int32_t tDeserializeSMDropSmaReq(void* buf, int32_t bufLen, SMDropSmaReq* pReq);
+
 typedef struct {
   int8_t   version;       // for compatibility(default 0)
   int8_t   intervalUnit;  // MACRO: TIME_UNIT_XXX
   int8_t   slidingUnit;   // MACRO: TIME_UNIT_XXX
+  int8_t   timezoneInt;   // sma data expired if timezone changes.
   char     indexName[TSDB_INDEX_NAME_LEN];
-  char     timezone[TD_TIMEZONE_LEN];  // sma data expired if timezone changes.
+  char     timezone[TD_TIMEZONE_LEN];
   int32_t  exprLen;
   int32_t  tagsFilterLen;
   int64_t  indexUid;
@@ -1919,13 +2009,14 @@ typedef struct {
 } SVCreateTSmaReq;
 
 typedef struct {
-  int8_t      type;                            // 0 status report, 1 update data
-  char        indexName[TSDB_INDEX_NAME_LEN];  //
-  STimeWindow windows;
+  int8_t  type;  // 0 status report, 1 update data
+  int64_t indexUid;
+  int64_t skey;  // start TS key of interval/sliding window
 } STSmaMsg;
 
 typedef struct {
   int64_t ver;  // use a general definition
+  int64_t indexUid;
   char    indexName[TSDB_INDEX_NAME_LEN];
 } SVDropTSmaReq;
 
@@ -1985,8 +2076,8 @@ typedef struct {
 
 static FORCE_INLINE void tdDestroyTSma(STSma* pSma) {
   if (pSma) {
-    tfree(pSma->expr);
-    tfree(pSma->tagsFilter);
+    taosMemoryFreeClear(pSma->expr);
+    taosMemoryFreeClear(pSma->tagsFilter);
   }
 }
 
@@ -1996,19 +2087,24 @@ static FORCE_INLINE void tdDestroyTSmaWrapper(STSmaWrapper* pSW) {
       for (uint32_t i = 0; i < pSW->number; ++i) {
         tdDestroyTSma(pSW->tSma + i);
       }
-      tfree(pSW->tSma);
+      taosMemoryFreeClear(pSW->tSma);
     }
   }
 }
 
+static FORCE_INLINE void tdFreeTSmaWrapper(STSmaWrapper* pSW) {
+  tdDestroyTSmaWrapper(pSW);
+  taosMemoryFreeClear(pSW);
+}
+
 static FORCE_INLINE int32_t tEncodeTSma(void** buf, const STSma* pSma) {
   int32_t tlen = 0;
 
   tlen += taosEncodeFixedI8(buf, pSma->version);
   tlen += taosEncodeFixedI8(buf, pSma->intervalUnit);
   tlen += taosEncodeFixedI8(buf, pSma->slidingUnit);
+  tlen += taosEncodeFixedI8(buf, pSma->timezoneInt);
   tlen += taosEncodeString(buf, pSma->indexName);
-  tlen += taosEncodeString(buf, pSma->timezone);
   tlen += taosEncodeFixedI32(buf, pSma->exprLen);
   tlen += taosEncodeFixedI32(buf, pSma->tagsFilterLen);
   tlen += taosEncodeFixedI64(buf, pSma->indexUid);
@@ -2042,8 +2138,8 @@ static FORCE_INLINE void* tDecodeTSma(void* buf, STSma* pSma) {
   buf = taosDecodeFixedI8(buf, &pSma->version);
   buf = taosDecodeFixedI8(buf, &pSma->intervalUnit);
   buf = taosDecodeFixedI8(buf, &pSma->slidingUnit);
+  buf = taosDecodeFixedI8(buf, &pSma->timezoneInt);
   buf = taosDecodeStringTo(buf, pSma->indexName);
-  buf = taosDecodeStringTo(buf, pSma->timezone);
   buf = taosDecodeFixedI32(buf, &pSma->exprLen);
   buf = taosDecodeFixedI32(buf, &pSma->tagsFilterLen);
   buf = taosDecodeFixedI64(buf, &pSma->indexUid);
@@ -2076,7 +2172,7 @@ static FORCE_INLINE void* tDecodeTSma(void* buf, STSma* pSma) {
 static FORCE_INLINE void* tDecodeTSmaWrapper(void* buf, STSmaWrapper* pSW) {
   buf = taosDecodeFixedU32(buf, &pSW->number);
 
-  pSW->tSma = (STSma*)calloc(pSW->number, sizeof(STSma));
+  pSW->tSma = (STSma*)taosMemoryCalloc(pSW->number, sizeof(STSma));
   if (pSW->tSma == NULL) {
     return NULL;
   }
@@ -2086,7 +2182,7 @@ static FORCE_INLINE void* tDecodeTSmaWrapper(void* buf, STSmaWrapper* pSW) {
       for (uint32_t j = i; j >= 0; --i) {
         tdDestroyTSma(pSW->tSma + j);
       }
-      free(pSW->tSma);
+      taosMemoryFree(pSW->tSma);
       return NULL;
     }
   }
@@ -2114,25 +2210,16 @@ typedef struct {
   int8_t  mqMsgType;
   int32_t code;
   int32_t epoch;
+  int64_t consumerId;
 } SMqRspHead;
 
-typedef struct {
-  int64_t         consumerId;
-  SSchemaWrapper* schemas;
-  int64_t         reqOffset;
-  int64_t         rspOffset;
-  int32_t         skipLogNum;
-  int32_t         numOfTopics;
-  SArray*         pBlockData;  // SArray<SSDataBlock>
-} SMqPollRsp;
-
-// one req for one vg+topic
 typedef struct {
   SMsgHead head;
 
   int64_t consumerId;
   int64_t blockingTime;
   int32_t epoch;
+  int8_t  withSchema;
   char    cgroup[TSDB_CGROUP_LEN];
 
   int64_t currentOffset;
@@ -2151,19 +2238,22 @@ typedef struct {
 } SMqSubTopicEp;
 
 typedef struct {
-  int64_t consumerId;
-  char    cgroup[TSDB_CGROUP_LEN];
-  SArray* topics;  // SArray<SMqSubTopicEp>
-} SMqCMGetSubEpRsp;
+  SMqRspHead head;
+  int64_t    reqOffset;
+  int64_t    rspOffset;
+  int32_t    skipLogNum;
+  // TODO: replace with topic name
+  int32_t numOfTopics;
+  // TODO: remove from msg
+  SSchemaWrapper* schema;
+  SArray*         pBlockData;  // SArray<SSDataBlock>
+} SMqPollRsp;
 
 typedef struct {
   SMqRspHead head;
-  union {
-    SMqPollRsp       consumeRsp;
-    SMqCMGetSubEpRsp getEpRsp;
-  };
-  void* extra;
-} SMqMsgWrapper;
+  char       cgroup[TSDB_CGROUP_LEN];
+  SArray*    topics;  // SArray<SMqSubTopicEp>
+} SMqCMGetSubEpRsp;
 
 typedef struct {
   int32_t curBlock;
@@ -2171,11 +2261,13 @@ typedef struct {
   void**  uData;
 } SMqRowIter;
 
-struct tmq_message_t_v1 {
-  SMqPollRsp rsp;
+struct tmq_message_t {
+  SMqPollRsp msg;
+  void*      vg;
   SMqRowIter iter;
 };
 
+#if 0
 struct tmq_message_t {
   SMqRspHead head;
   union {
@@ -2187,6 +2279,7 @@ struct tmq_message_t {
   int32_t curRow;
   void**  uData;
 };
+#endif
 
 static FORCE_INLINE void tDeleteSMqSubTopicEp(SMqSubTopicEp* pSubTopicEp) { taosArrayDestroy(pSubTopicEp->vgs); }
 
@@ -2239,8 +2332,7 @@ static FORCE_INLINE void* tDecodeSMqSubTopicEp(void* buf, SMqSubTopicEp* pTopicE
 
 static FORCE_INLINE int32_t tEncodeSMqCMGetSubEpRsp(void** buf, const SMqCMGetSubEpRsp* pRsp) {
   int32_t tlen = 0;
-  tlen += taosEncodeFixedI64(buf, pRsp->consumerId);
-  tlen += taosEncodeString(buf, pRsp->cgroup);
+  // tlen += taosEncodeString(buf, pRsp->cgroup);
   int32_t sz = taosArrayGetSize(pRsp->topics);
   tlen += taosEncodeFixedI32(buf, sz);
   for (int32_t i = 0; i < sz; i++) {
@@ -2251,8 +2343,7 @@ static FORCE_INLINE int32_t tEncodeSMqCMGetSubEpRsp(void** buf, const SMqCMGetSu
 }
 
 static FORCE_INLINE void* tDecodeSMqCMGetSubEpRsp(void* buf, SMqCMGetSubEpRsp* pRsp) {
-  buf = taosDecodeFixedI64(buf, &pRsp->consumerId);
-  buf = taosDecodeStringTo(buf, pRsp->cgroup);
+  // buf = taosDecodeStringTo(buf, pRsp->cgroup);
   int32_t sz;
   buf = taosDecodeFixedI32(buf, &sz);
   pRsp->topics = taosArrayInit(sz, sizeof(SMqSubTopicEp));
@@ -2266,56 +2357,6 @@ static FORCE_INLINE void* tDecodeSMqCMGetSubEpRsp(void* buf, SMqCMGetSubEpRsp* p
   }
   return buf;
 }
-
-enum {
-  STREAM_TASK_STATUS__RUNNING = 1,
-  STREAM_TASK_STATUS__STOP,
-};
-
-typedef struct {
-  int64_t streamId;
-  int32_t taskId;
-  int32_t level;
-  int8_t  status;
-  char*   qmsg;
-  void*   executor;
-  // void*   stateStore;
-  //  storage handle
-} SStreamTask;
-
-static FORCE_INLINE SStreamTask* streamTaskNew(int64_t streamId, int32_t level) {
-  SStreamTask* pTask = (SStreamTask*)calloc(1, sizeof(SStreamTask));
-  if (pTask == NULL) {
-    return NULL;
-  }
-  pTask->taskId = tGenIdPI32();
-  pTask->status = STREAM_TASK_STATUS__RUNNING;
-  pTask->qmsg = NULL;
-  return pTask;
-}
-
-int32_t tEncodeSStreamTask(SCoder* pEncoder, const SStreamTask* pTask);
-int32_t tDecodeSStreamTask(SCoder* pDecoder, SStreamTask* pTask);
-void    tFreeSStreamTask(SStreamTask* pTask);
-
-typedef struct {
-  SMsgHead     head;
-  SStreamTask* task;
-} SStreamTaskDeployReq;
-
-typedef struct {
-  int32_t reserved;
-} SStreamTaskDeployRsp;
-
-typedef struct {
-  SStreamExecMsgHead head;
-  // TODO: other info needed by task
-} SStreamTaskExecReq;
-
-typedef struct {
-  int32_t reserved;
-} SStreamTaskExecRsp;
-
 #pragma pack(pop)
 
 #ifdef __cplusplus

include/common/tmsgcb.h

+/*
+ * Copyright (c) 2019 TAOS Data, Inc. <jhtao@taosdata.com>
+ *
+ * This program is free software: you can use, redistribute, and/or modify
+ * it under the terms of the GNU Affero General Public License, version 3
+ * or later ("AGPL"), as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.
+ *
+ * You should have received a copy of the GNU Affero General Public License
+ * along with this program. If not, see <http://www.gnu.org/licenses/>.
+ */
+
+#ifndef _TD_COMMON_MSG_CB_H_
+#define _TD_COMMON_MSG_CB_H_
+
+#include "os.h"
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+typedef struct SRpcMsg      SRpcMsg;
+typedef struct SEpSet       SEpSet;
+typedef struct SMgmtWrapper SMgmtWrapper;
+typedef enum {
+  QUERY_QUEUE,
+  FETCH_QUEUE,
+  READ_QUEUE,
+  WRITE_QUEUE,
+  APPLY_QUEUE,
+  SYNC_QUEUE,
+  MERGE_QUEUE,
+  QUEUE_MAX,
+} EQueueType;
+
+typedef int32_t (*PutToQueueFp)(SMgmtWrapper* pWrapper, SRpcMsg* pReq);
+typedef int32_t (*GetQueueSizeFp)(SMgmtWrapper* pWrapper, int32_t vgId, EQueueType qtype);
+typedef int32_t (*SendReqFp)(SMgmtWrapper* pWrapper, SEpSet* epSet, SRpcMsg* pReq);
+typedef int32_t (*SendMnodeReqFp)(SMgmtWrapper* pWrapper, SRpcMsg* pReq);
+typedef void (*SendRspFp)(SMgmtWrapper* pWrapper, SRpcMsg* pRsp);
+
+typedef struct {
+  SMgmtWrapper*  pWrapper;
+  PutToQueueFp   queueFps[QUEUE_MAX];
+  GetQueueSizeFp qsizeFp;
+  SendReqFp      sendReqFp;
+  SendMnodeReqFp sendMnodeReqFp;
+  SendRspFp      sendRspFp;
+} SMsgCb;
+
+int32_t tmsgPutToQueue(const SMsgCb* pMsgCb, EQueueType qtype, SRpcMsg* pReq);
+int32_t tmsgGetQueueSize(const SMsgCb* pMsgCb, int32_t vgId, EQueueType qtype);
+int32_t tmsgSendReq(const SMsgCb* pMsgCb, SEpSet* epSet, SRpcMsg* pReq);
+int32_t tmsgSendMnodeReq(const SMsgCb* pMsgCb, SRpcMsg* pReq);
+void    tmsgSendRsp(const SMsgCb* pMsgCb, SRpcMsg* pRsp);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /*_TD_COMMON_MSG_CB_H_*/

include/common/tmsgdef.h

@@ -127,6 +127,8 @@ enum {
   TD_DEF_MSG_TYPE(TDMT_MND_CREATE_STB, "mnode-create-stb", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_MND_ALTER_STB, "mnode-alter-stb", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_MND_DROP_STB, "mnode-drop-stb", NULL, NULL)
+  TD_DEF_MSG_TYPE(TDMT_MND_CREATE_SMA, "mnode-create-sma", NULL, NULL)
+  TD_DEF_MSG_TYPE(TDMT_MND_DROP_SMA, "mnode-drop-sma", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_MND_TABLE_META, "mnode-table-meta", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_MND_VGROUP_LIST, "mnode-vgroup-list", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_MND_QNODE_LIST, "mnode-qnode-list", NULL, NULL)
@@ -190,6 +192,11 @@ enum {
   TD_DEF_MSG_TYPE(TDMT_VND_SUBSCRIBE, "vnode-subscribe", SMVSubscribeReq, SMVSubscribeRsp)
   TD_DEF_MSG_TYPE(TDMT_VND_CONSUME, "vnode-consume", SMqCVConsumeReq, SMqCVConsumeRsp)
   TD_DEF_MSG_TYPE(TDMT_VND_TASK_DEPLOY, "vnode-task-deploy", SStreamTaskDeployReq, SStreamTaskDeployRsp)
+  TD_DEF_MSG_TYPE(TDMT_VND_TASK_EXEC, "vnode-task-exec", SStreamTaskExecReq, SStreamTaskExecRsp)
+  TD_DEF_MSG_TYPE(TDMT_VND_TASK_PIPE_EXEC, "vnode-task-pipe-exec", SStreamTaskExecReq, SStreamTaskExecRsp)
+  TD_DEF_MSG_TYPE(TDMT_VND_TASK_MERGE_EXEC, "vnode-task-merge-exec", SStreamTaskExecReq, SStreamTaskExecRsp)
+  TD_DEF_MSG_TYPE(TDMT_VND_TASK_WRITE_EXEC, "vnode-task-write-exec", SStreamTaskExecReq, SStreamTaskExecRsp)
+  TD_DEF_MSG_TYPE(TDMT_VND_STREAM_TRIGGER, "vnode-stream-trigger", NULL, NULL)
 
   TD_DEF_MSG_TYPE(TDMT_VND_CREATE_SMA, "vnode-create-sma", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_VND_CANCEL_SMA, "vnode-cancel-sma", NULL, NULL)
@@ -202,7 +209,13 @@ enum {
   TD_NEW_MSG_SEG(TDMT_SND_MSG)
   TD_DEF_MSG_TYPE(TDMT_SND_TASK_DEPLOY, "snode-task-deploy", SStreamTaskDeployReq, SStreamTaskDeployRsp)
   TD_DEF_MSG_TYPE(TDMT_SND_TASK_EXEC, "snode-task-exec", SStreamTaskExecReq, SStreamTaskExecRsp)
+  TD_DEF_MSG_TYPE(TDMT_SND_TASK_PIPE_EXEC, "snode-task-pipe-exec", SStreamTaskExecReq, SStreamTaskExecRsp)
+  TD_DEF_MSG_TYPE(TDMT_SND_TASK_MERGE_EXEC, "snode-task-merge-exec", SStreamTaskExecReq, SStreamTaskExecRsp)
 
+  // Requests handled by SCHEDULER
+  TD_NEW_MSG_SEG(TDMT_SCH_MSG)
+  TD_DEF_MSG_TYPE(TDMT_SCH_LINK_BROKEN, "scheduler-link-broken", NULL, NULL)
+  
 #if defined(TD_MSG_NUMBER_)
   TDMT_MAX
 #endif

include/common/tname.h

@@ -17,7 +17,6 @@
 #define _TD_COMMON_NAME_H_
 
 #include "tdef.h"
-#include "tmsg.h"
 
 #ifdef __cplusplus
 extern "C" {
@@ -61,7 +60,8 @@ int32_t tNameFromString(SName* dst, const char* str, uint32_t type);
 
 int32_t tNameSetAcctId(SName* dst, int32_t acctId);
 
-SSchema createSchema(uint8_t type, int32_t bytes, int32_t colId, const char* name);
+bool tNameDBNameEqual(SName* left, SName* right);
+
 
 #ifdef __cplusplus
 }

include/common/trow.h

@@ -166,6 +166,7 @@ typedef struct {
 #define TD_ROW_HEAD_LEN  (sizeof(STSRow))
 #define TD_ROW_NCOLS_LEN (sizeof(col_id_t))
 
+#define TD_ROW_INFO(r)     ((r)->info)
 #define TD_ROW_TYPE(r)     ((r)->type)
 #define TD_ROW_DELETE(r)   ((r)->del)
 #define TD_ROW_ENDIAN(r)   ((r)->endian)
@@ -180,6 +181,7 @@ typedef struct {
 // (int32_t)ceil((double)nCols/TD_VTYPE_PARTS) should be added if TD_SUPPORT_BITMAP defined.
 #define TD_ROW_MAX_BYTES_FROM_SCHEMA(s) (schemaTLen(s) + TD_ROW_HEAD_LEN)
 
+#define TD_ROW_SET_INFO(r, i)  (TD_ROW_INFO(r) = (i))
 #define TD_ROW_SET_TYPE(r, t)  (TD_ROW_TYPE(r) = (t))
 #define TD_ROW_SET_DELETE(r)   (TD_ROW_DELETE(r) = 1)
 #define TD_ROW_SET_SVER(r, v)  (TD_ROW_SVER(r) = (v))
@@ -473,6 +475,7 @@ static int32_t tdSRowResetBuf(SRowBuilder *pBuilder, void *pBuf) {
     return terrno;
   }
 
+  TD_ROW_SET_INFO(pBuilder->pBuf, 0);
   TD_ROW_SET_TYPE(pBuilder->pBuf, pBuilder->rowType);
 
   uint32_t len = 0;
@@ -968,10 +971,14 @@ static FORCE_INLINE int32_t tdGetColDataOfRow(SCellVal *pVal, SDataCol *pCol, in
 #endif
     return TSDB_CODE_SUCCESS;
   }
-  if (tdGetBitmapValType(pCol->pBitmap, row, &(pVal->valType)) < 0) {
+
+  if (TD_COL_ROWS_NORM(pCol)) {
+    pVal->valType = TD_VTYPE_NORM;
+  } else if (tdGetBitmapValType(pCol->pBitmap, row, &(pVal->valType)) < 0) {
     return terrno;
   }
-  if (TD_COL_ROWS_NORM(pCol) || tdValTypeIsNorm(pVal->valType)) {
+
+  if (tdValTypeIsNorm(pVal->valType)) {
     if (IS_VAR_DATA_TYPE(pCol->type)) {
       pVal->val = POINTER_SHIFT(pCol->pData, pCol->dataOff[row]);
     } else {

include/common/ttokendef.h

@@ -51,130 +51,133 @@
 #define TK_USER                            33
 #define TK_PRIVILEGE                       34
 #define TK_DROP                            35
-#define TK_SHOW                            36
-#define TK_DNODE                           37
-#define TK_PORT                            38
-#define TK_NK_INTEGER                      39
-#define TK_DNODES                          40
-#define TK_NK_IPTOKEN                      41
-#define TK_LOCAL                           42
-#define TK_QNODE                           43
-#define TK_ON                              44
-#define TK_QNODES                          45
-#define TK_DATABASE                        46
-#define TK_DATABASES                       47
-#define TK_USE                             48
-#define TK_IF                              49
-#define TK_NOT                             50
-#define TK_EXISTS                          51
-#define TK_BLOCKS                          52
-#define TK_CACHE                           53
-#define TK_CACHELAST                       54
-#define TK_COMP                            55
-#define TK_DAYS                            56
-#define TK_FSYNC                           57
-#define TK_MAXROWS                         58
-#define TK_MINROWS                         59
-#define TK_KEEP                            60
-#define TK_PRECISION                       61
-#define TK_QUORUM                          62
-#define TK_REPLICA                         63
-#define TK_TTL                             64
-#define TK_WAL                             65
-#define TK_VGROUPS                         66
-#define TK_SINGLE_STABLE                   67
-#define TK_STREAM_MODE                     68
-#define TK_RETENTIONS                      69
-#define TK_FILE_FACTOR                     70
-#define TK_NK_FLOAT                        71
-#define TK_TABLE                           72
-#define TK_NK_LP                           73
-#define TK_NK_RP                           74
-#define TK_STABLE                          75
-#define TK_TABLES                          76
-#define TK_STABLES                         77
-#define TK_ADD                             78
-#define TK_COLUMN                          79
-#define TK_MODIFY                          80
-#define TK_RENAME                          81
-#define TK_TAG                             82
-#define TK_SET                             83
-#define TK_NK_EQ                           84
-#define TK_USING                           85
-#define TK_TAGS                            86
-#define TK_NK_DOT                          87
-#define TK_NK_COMMA                        88
-#define TK_COMMENT                         89
-#define TK_BOOL                            90
-#define TK_TINYINT                         91
-#define TK_SMALLINT                        92
-#define TK_INT                             93
-#define TK_INTEGER                         94
-#define TK_BIGINT                          95
-#define TK_FLOAT                           96
-#define TK_DOUBLE                          97
-#define TK_BINARY                          98
-#define TK_TIMESTAMP                       99
-#define TK_NCHAR                          100
-#define TK_UNSIGNED                       101
-#define TK_JSON                           102
-#define TK_VARCHAR                        103
-#define TK_MEDIUMBLOB                     104
-#define TK_BLOB                           105
-#define TK_VARBINARY                      106
-#define TK_DECIMAL                        107
-#define TK_SMA                            108
-#define TK_ROLLUP                         109
-#define TK_INDEX                          110
-#define TK_FULLTEXT                       111
-#define TK_FUNCTION                       112
-#define TK_INTERVAL                       113
-#define TK_TOPIC                          114
-#define TK_AS                             115
-#define TK_MNODES                         116
-#define TK_NK_BOOL                        117
-#define TK_NK_VARIABLE                    118
-#define TK_BETWEEN                        119
-#define TK_IS                             120
-#define TK_NULL                           121
-#define TK_NK_LT                          122
-#define TK_NK_GT                          123
-#define TK_NK_LE                          124
-#define TK_NK_GE                          125
-#define TK_NK_NE                          126
-#define TK_LIKE                           127
-#define TK_MATCH                          128
-#define TK_NMATCH                         129
-#define TK_IN                             130
-#define TK_FROM                           131
-#define TK_JOIN                           132
-#define TK_INNER                          133
-#define TK_SELECT                         134
-#define TK_DISTINCT                       135
-#define TK_WHERE                          136
-#define TK_PARTITION                      137
-#define TK_BY                             138
-#define TK_SESSION                        139
-#define TK_STATE_WINDOW                   140
-#define TK_SLIDING                        141
-#define TK_FILL                           142
-#define TK_VALUE                          143
-#define TK_NONE                           144
-#define TK_PREV                           145
-#define TK_LINEAR                         146
-#define TK_NEXT                           147
-#define TK_GROUP                          148
-#define TK_HAVING                         149
-#define TK_ORDER                          150
-#define TK_SLIMIT                         151
-#define TK_SOFFSET                        152
-#define TK_LIMIT                          153
-#define TK_OFFSET                         154
-#define TK_ASC                            155
-#define TK_DESC                           156
-#define TK_NULLS                          157
-#define TK_FIRST                          158
-#define TK_LAST                           159
+#define TK_DNODE                           36
+#define TK_PORT                            37
+#define TK_NK_INTEGER                      38
+#define TK_DNODES                          39
+#define TK_NK_IPTOKEN                      40
+#define TK_LOCAL                           41
+#define TK_QNODE                           42
+#define TK_ON                              43
+#define TK_DATABASE                        44
+#define TK_USE                             45
+#define TK_IF                              46
+#define TK_NOT                             47
+#define TK_EXISTS                          48
+#define TK_BLOCKS                          49
+#define TK_CACHE                           50
+#define TK_CACHELAST                       51
+#define TK_COMP                            52
+#define TK_DAYS                            53
+#define TK_FSYNC                           54
+#define TK_MAXROWS                         55
+#define TK_MINROWS                         56
+#define TK_KEEP                            57
+#define TK_PRECISION                       58
+#define TK_QUORUM                          59
+#define TK_REPLICA                         60
+#define TK_TTL                             61
+#define TK_WAL                             62
+#define TK_VGROUPS                         63
+#define TK_SINGLE_STABLE                   64
+#define TK_STREAM_MODE                     65
+#define TK_RETENTIONS                      66
+#define TK_FILE_FACTOR                     67
+#define TK_NK_FLOAT                        68
+#define TK_TABLE                           69
+#define TK_NK_LP                           70
+#define TK_NK_RP                           71
+#define TK_STABLE                          72
+#define TK_ADD                             73
+#define TK_COLUMN                          74
+#define TK_MODIFY                          75
+#define TK_RENAME                          76
+#define TK_TAG                             77
+#define TK_SET                             78
+#define TK_NK_EQ                           79
+#define TK_USING                           80
+#define TK_TAGS                            81
+#define TK_NK_DOT                          82
+#define TK_NK_COMMA                        83
+#define TK_COMMENT                         84
+#define TK_BOOL                            85
+#define TK_TINYINT                         86
+#define TK_SMALLINT                        87
+#define TK_INT                             88
+#define TK_INTEGER                         89
+#define TK_BIGINT                          90
+#define TK_FLOAT                           91
+#define TK_DOUBLE                          92
+#define TK_BINARY                          93
+#define TK_TIMESTAMP                       94
+#define TK_NCHAR                           95
+#define TK_UNSIGNED                        96
+#define TK_JSON                            97
+#define TK_VARCHAR                         98
+#define TK_MEDIUMBLOB                      99
+#define TK_BLOB                           100
+#define TK_VARBINARY                      101
+#define TK_DECIMAL                        102
+#define TK_SMA                            103
+#define TK_ROLLUP                         104
+#define TK_SHOW                           105
+#define TK_DATABASES                      106
+#define TK_TABLES                         107
+#define TK_STABLES                        108
+#define TK_MNODES                         109
+#define TK_MODULES                        110
+#define TK_QNODES                         111
+#define TK_FUNCTIONS                      112
+#define TK_INDEXES                        113
+#define TK_FROM                           114
+#define TK_LIKE                           115
+#define TK_INDEX                          116
+#define TK_FULLTEXT                       117
+#define TK_FUNCTION                       118
+#define TK_INTERVAL                       119
+#define TK_TOPIC                          120
+#define TK_AS                             121
+#define TK_NK_BOOL                        122
+#define TK_NK_VARIABLE                    123
+#define TK_BETWEEN                        124
+#define TK_IS                             125
+#define TK_NULL                           126
+#define TK_NK_LT                          127
+#define TK_NK_GT                          128
+#define TK_NK_LE                          129
+#define TK_NK_GE                          130
+#define TK_NK_NE                          131
+#define TK_MATCH                          132
+#define TK_NMATCH                         133
+#define TK_IN                             134
+#define TK_JOIN                           135
+#define TK_INNER                          136
+#define TK_SELECT                         137
+#define TK_DISTINCT                       138
+#define TK_WHERE                          139
+#define TK_PARTITION                      140
+#define TK_BY                             141
+#define TK_SESSION                        142
+#define TK_STATE_WINDOW                   143
+#define TK_SLIDING                        144
+#define TK_FILL                           145
+#define TK_VALUE                          146
+#define TK_NONE                           147
+#define TK_PREV                           148
+#define TK_LINEAR                         149
+#define TK_NEXT                           150
+#define TK_GROUP                          151
+#define TK_HAVING                         152
+#define TK_ORDER                          153
+#define TK_SLIMIT                         154
+#define TK_SOFFSET                        155
+#define TK_LIMIT                          156
+#define TK_OFFSET                         157
+#define TK_ASC                            158
+#define TK_DESC                           159
+#define TK_NULLS                          160
+#define TK_FIRST                          161
+#define TK_LAST                           162
 
 #define TK_NK_SPACE                       300
 #define TK_NK_COMMENT                     301

include/common/ttypes.h

@@ -46,7 +46,7 @@ typedef struct {
 #define varDataCopy(dst, v)    memcpy((dst), (void *)(v), varDataTLen(v))
 #define varDataLenByData(v)    (*(VarDataLenT *)(((char *)(v)) - VARSTR_HEADER_SIZE))
 #define varDataSetLen(v, _len) (((VarDataLenT *)(v))[0] = (VarDataLenT)(_len))
-#define IS_VAR_DATA_TYPE(t)    (((t) == TSDB_DATA_TYPE_BINARY) || ((t) == TSDB_DATA_TYPE_NCHAR))
+#define IS_VAR_DATA_TYPE(t)    (((t) == TSDB_DATA_TYPE_VARCHAR) || ((t) == TSDB_DATA_TYPE_NCHAR))
 
 #define varDataNetLen(v)  (htons(((VarDataLenT *)(v))[0]))
 #define varDataNetTLen(v) (sizeof(VarDataLenT) + varDataNetLen(v))
@@ -159,6 +159,7 @@ typedef struct {
   (IS_SIGNED_NUMERIC_TYPE(_t) || (_t) == (TSDB_DATA_TYPE_BOOL) || (_t) == (TSDB_DATA_TYPE_TIMESTAMP))
 #define IS_CONVERT_AS_UNSIGNED(_t) (IS_UNSIGNED_NUMERIC_TYPE(_t) || (_t) == (TSDB_DATA_TYPE_BOOL))
 
+// TODO remove this function
 static FORCE_INLINE bool isNull(const void *val, int32_t type) {
   switch (type) {
     case TSDB_DATA_TYPE_BOOL:

include/dnode/bnode/bnode.h

@@ -16,29 +16,20 @@
 #ifndef _TD_BNODE_H_
 #define _TD_BNODE_H_
 
+#include "tmsgcb.h"
+
 #ifdef __cplusplus
 extern "C" {
 #endif
 
 /* ------------------------ TYPES EXPOSED ------------------------ */
-typedef struct SDnode SDnode;
 typedef struct SBnode SBnode;
-typedef int32_t (*SendReqToDnodeFp)(SDnode *pDnode, struct SEpSet *epSet, struct SRpcMsg *pMsg);
-typedef int32_t (*SendReqToMnodeFp)(SDnode *pDnode, struct SRpcMsg *pMsg);
-typedef void (*SendRedirectRspFp)(SDnode *pDnode, struct SRpcMsg *pMsg);
 
 typedef struct {
-  int64_t numOfErrors;
 } SBnodeLoad;
 
 typedef struct {
-  int32_t           sver;
-  int32_t           dnodeId;
-  int64_t           clusterId;
-  SDnode           *pDnode;
-  SendReqToDnodeFp  sendReqToDnodeFp;
-  SendReqToMnodeFp  sendReqToMnodeFp;
-  SendRedirectRspFp sendRedirectRspFp;
+  SMsgCb msgCb;
 } SBnodeOpt;
 
 /* ------------------------ SBnode ------------------------ */
@@ -76,13 +67,6 @@ int32_t bndGetLoad(SBnode *pBnode, SBnodeLoad *pLoad);
  */
 int32_t bndProcessWMsgs(SBnode *pBnode, SArray *pMsgs);
 
-/**
- * @brief Drop a bnode.
- *
- * @param path Path of the bnode.
- */
-void bndDestroy(const char *path);
-
 #ifdef __cplusplus
 }
 #endif

include/dnode/mgmt/dnode.h

@@ -33,8 +33,7 @@ typedef struct SDnode SDnode;
 int32_t dndInit();
 
 /**
- * @brief clear the environment
- *
+ * @brief Clear the environment
  */
 void dndCleanup();
 
@@ -42,22 +41,24 @@ void dndCleanup();
 typedef struct {
   int32_t   numOfSupportVnodes;
   uint16_t  serverPort;
-  char      dataDir[TSDB_FILENAME_LEN];
+  char      dataDir[PATH_MAX];
   char      localEp[TSDB_EP_LEN];
   char      localFqdn[TSDB_FQDN_LEN];
   char      firstEp[TSDB_EP_LEN];
   char      secondEp[TSDB_EP_LEN];
   SDiskCfg *pDisks;
   int32_t   numOfDisks;
-} SDnodeObjCfg;
+} SDnodeOpt;
+
+typedef enum { DND_EVENT_START, DND_EVENT_STOP = 1, DND_EVENT_RELOAD } EDndEvent;
 
 /**
  * @brief Initialize and start the dnode.
  *
- * @param pCfg Config of the dnode.
+ * @param pOption Option of the dnode.
  * @return SDnode* The dnode object.
  */
-SDnode *dndCreate(SDnodeObjCfg *pCfg);
+SDnode *dndCreate(const SDnodeOpt *pOption);
 
 /**
  * @brief Stop and cleanup the dnode.
@@ -66,6 +67,21 @@ SDnode *dndCreate(SDnodeObjCfg *pCfg);
  */
 void dndClose(SDnode *pDnode);
 
+/**
+ * @brief Run dnode until specific event is receive.
+ *
+ * @param pDnode The dnode object to run.
+ */
+int32_t dndRun(SDnode *pDnode);
+
+/**
+ * @brief Handle event in the dnode.
+ *
+ * @param pDnode The dnode object to close.
+ * @param event The event to handle.
+ */
+void dndHandleEvent(SDnode *pDnode, EDndEvent event);
+
 #ifdef __cplusplus
 }
 #endif

include/dnode/mnode/mnode.h

@@ -17,33 +17,24 @@
 #define _TD_MND_H_
 
 #include "monitor.h"
+#include "tmsg.h"
+#include "tmsgcb.h"
+#include "trpc.h"
 
 #ifdef __cplusplus
 extern "C" {
 #endif
 
 /* ------------------------ TYPES EXPOSED ------------------------ */
-typedef struct SDnode    SDnode;
-typedef struct SMnode    SMnode;
-typedef struct SMnodeMsg SMnodeMsg;
-typedef int32_t (*SendReqToDnodeFp)(SDnode *pDnode, struct SEpSet *epSet, struct SRpcMsg *rpcMsg);
-typedef int32_t (*SendReqToMnodeFp)(SDnode *pDnode, struct SRpcMsg *rpcMsg);
-typedef int32_t (*PutReqToMWriteQFp)(SDnode *pDnode, struct SRpcMsg *rpcMsg);
-typedef int32_t (*PutReqToMReadQFp)(SDnode *pDnode, struct SRpcMsg *rpcMsg);
-typedef void (*SendRedirectRspFp)(SDnode *pDnode, struct SRpcMsg *rpcMsg);
+typedef struct SMnode SMnode;
 
 typedef struct {
-  int32_t           dnodeId;
-  int64_t           clusterId;
-  int8_t            replica;
-  int8_t            selfIndex;
-  SReplica          replicas[TSDB_MAX_REPLICA];
-  SDnode           *pDnode;
-  PutReqToMWriteQFp putReqToMWriteQFp;
-  PutReqToMReadQFp  putReqToMReadQFp;
-  SendReqToDnodeFp  sendReqToDnodeFp;
-  SendReqToMnodeFp  sendReqToMnodeFp;
-  SendRedirectRspFp sendRedirectRspFp;
+  int32_t  dnodeId;
+  int64_t  clusterId;
+  int8_t   replica;
+  int8_t   selfIndex;
+  SReplica replicas[TSDB_MAX_REPLICA];
+  SMsgCb   msgCb;
 } SMnodeOpt;
 
 /* ------------------------ SMnode ------------------------ */
@@ -73,11 +64,11 @@ void mndClose(SMnode *pMnode);
 int32_t mndAlter(SMnode *pMnode, const SMnodeOpt *pOption);
 
 /**
- * @brief Drop a mnode.
+ * @brief Start mnode
  *
- * @param path Path of the mnode.
+ * @param pMnode The mnode object.
  */
-void mndDestroy(const char *path);
+int32_t mndStart(SMnode *pMnode);
 
 /**
  * @brief Get mnode monitor info.
@@ -104,37 +95,13 @@ int32_t mndGetMonitorInfo(SMnode *pMnode, SMonClusterInfo *pClusterInfo, SMonVgr
  */
 int32_t mndRetriveAuth(SMnode *pMnode, char *user, char *spi, char *encrypt, char *secret, char *ckey);
 
-/**
- * @brief Initialize mnode msg.
- *
- * @param pMnode The mnode object.
- * @param pMsg The request rpc msg.
- * @return int32_t The created mnode msg.
- */
-SMnodeMsg *mndInitMsg(SMnode *pMnode, SRpcMsg *pRpcMsg);
-
-/**
- * @brief Cleanup mnode msg.
- *
- * @param pMsg The request msg.
- */
-void mndCleanupMsg(SMnodeMsg *pMsg);
-
-/**
- * @brief Cleanup mnode msg.
- *
- * @param pMsg The request msg.
- * @param code The error code.
- */
-void mndSendRsp(SMnodeMsg *pMsg, int32_t code);
-
 /**
  * @brief Process the read, write, sync request.
  *
  * @param pMsg The request msg.
  * @return int32_t 0 for success, -1 for failure.
  */
-void mndProcessMsg(SMnodeMsg *pMsg);
+int32_t mndProcessMsg(SNodeMsg *pMsg);
 
 #ifdef __cplusplus
 }

include/dnode/mnode/sdb/sdb.h

@@ -119,10 +119,11 @@ typedef enum {
   SDB_CONSUMER = 13,
   SDB_TOPIC = 14,
   SDB_VGROUP = 15,
-  SDB_STB = 16,
-  SDB_DB = 17,
-  SDB_FUNC = 18,
-  SDB_MAX = 19
+  SDB_SMA = 16,
+  SDB_STB = 17,
+  SDB_DB = 18,
+  SDB_FUNC = 19,
+  SDB_MAX = 20
 } ESdbType;
 
 typedef struct SSdb SSdb;

include/dnode/qnode/qnode.h

@@ -16,16 +16,14 @@
 #ifndef _TD_QNODE_H_
 #define _TD_QNODE_H_
 
+#include "tmsgcb.h"
+
 #ifdef __cplusplus
 extern "C" {
 #endif
 
 /* ------------------------ TYPES EXPOSED ------------------------ */
-typedef struct SDnode SDnode;
 typedef struct SQnode SQnode;
-typedef int32_t (*SendReqToDnodeFp)(SDnode *pDnode, struct SEpSet *epSet, struct SRpcMsg *pMsg);
-typedef int32_t (*SendReqToMnodeFp)(SDnode *pDnode, struct SRpcMsg *pMsg);
-typedef void (*SendRedirectRspFp)(SDnode *pDnode, struct SRpcMsg *pMsg);
 
 typedef struct {
   int64_t numOfStartTask;
@@ -39,13 +37,7 @@ typedef struct {
 } SQnodeLoad;
 
 typedef struct {
-  int32_t           sver;
-  int32_t           dnodeId;
-  int64_t           clusterId;
-  SDnode           *pDnode;
-  SendReqToDnodeFp  sendReqToDnodeFp;
-  SendReqToMnodeFp  sendReqToMnodeFp;
-  SendRedirectRspFp sendRedirectRspFp;
+  SMsgCb msgCb;
 } SQnodeOpt;
 
 /* ------------------------ SQnode ------------------------ */
@@ -78,10 +70,9 @@ int32_t qndGetLoad(SQnode *pQnode, SQnodeLoad *pLoad);
  *
  * @param pQnode The qnode object.
  * @param pMsg The request message
- * @param pRsp The response message
- * @return int32_t 0 for success, -1 for failure
  */
-int32_t qndProcessMsg(SQnode *pQnode, SRpcMsg *pMsg, SRpcMsg **pRsp);
+int32_t qndProcessQueryMsg(SQnode *pQnode, SRpcMsg *pMsg);
+int32_t qndProcessFetchMsg(SQnode *pQnode, SRpcMsg *pMsg);
 
 #ifdef __cplusplus
 }

include/dnode/snode/snode.h

@@ -16,7 +16,7 @@
 #ifndef _TD_SNODE_H_
 #define _TD_SNODE_H_
 
-#include "tcommon.h"
+#include "tmsgcb.h"
 #include "tmsg.h"
 #include "trpc.h"
 
@@ -25,24 +25,14 @@ extern "C" {
 #endif
 
 /* ------------------------ TYPES EXPOSED ------------------------ */
-typedef struct SDnode SDnode;
 typedef struct SSnode SSnode;
-typedef int32_t (*SendReqToDnodeFp)(SDnode *pDnode, struct SEpSet *epSet, struct SRpcMsg *pMsg);
-typedef int32_t (*SendReqToMnodeFp)(SDnode *pDnode, struct SRpcMsg *pMsg);
-typedef void (*SendRedirectRspFp)(SDnode *pDnode, struct SRpcMsg *pMsg);
 
 typedef struct {
-  int64_t numOfErrors;
+  int32_t reserved;
 } SSnodeLoad;
 
 typedef struct {
-  int32_t           sver;
-  int32_t           dnodeId;
-  int64_t           clusterId;
-  SDnode           *pDnode;
-  SendReqToDnodeFp  sendReqToDnodeFp;
-  SendReqToMnodeFp  sendReqToMnodeFp;
-  SendRedirectRspFp sendRedirectRspFp;
+  SMsgCb msgCb;
 } SSnodeOpt;
 
 /* ------------------------ SSnode ------------------------ */
@@ -77,20 +67,9 @@ int32_t sndGetLoad(SSnode *pSnode, SSnodeLoad *pLoad);
  * @param pSnode The snode object.
  * @param pMsg The request message
  * @param pRsp The response message
- * @return int32_t 0 for success, -1 for failure
  */
-// int32_t sndProcessMsg(SSnode *pSnode, SRpcMsg *pMsg, SRpcMsg **pRsp);
-
-int32_t sndProcessUMsg(SSnode *pSnode, SRpcMsg *pMsg);
-
-int32_t sndProcessSMsg(SSnode *pSnode, SRpcMsg *pMsg);
-
-/**
- * @brief Drop a snode.
- *
- * @param path Path of the snode.
- */
-void sndDestroy(const char *path);
+void sndProcessUMsg(SSnode *pSnode, SRpcMsg *pMsg);
+void sndProcessSMsg(SSnode *pSnode, SRpcMsg *pMsg);
 
 #ifdef __cplusplus
 }

include/libs/catalog/catalog.h

@@ -103,16 +103,17 @@ int32_t catalogGetDBVgVersion(SCatalog* pCtg, const char* dbFName, int32_t* vers
  * @param pTransporter (input, rpc object)
  * @param pMgmtEps (input, mnode EPs)
  * @param pDBName (input, full db name)
- * @param forceUpdate (input, force update db vgroup info from mnode) 
  * @param pVgroupList (output, vgroup info list, element is SVgroupInfo, NEED to simply free the array by caller)
  * @return error code
  */
-int32_t catalogGetDBVgInfo(SCatalog* pCatalog, void *pTransporter, const SEpSet* pMgmtEps, const char* pDBName, bool forceUpdate, SArray** pVgroupList);
+int32_t catalogGetDBVgInfo(SCatalog* pCatalog, void *pTransporter, const SEpSet* pMgmtEps, const char* pDBName, SArray** pVgroupList);
 
 int32_t catalogUpdateDBVgInfo(SCatalog* pCatalog, const char* dbName, uint64_t dbId, SDBVgInfo* dbInfo);
 
 int32_t catalogRemoveDB(SCatalog* pCatalog, const char* dbName, uint64_t dbId);
 
+int32_t catalogRemoveTableMeta(SCatalog* pCtg, const SName* pTableName);
+
 int32_t catalogRemoveStbMeta(SCatalog* pCtg, const char* dbFName, uint64_t dbId, const char* stbName, uint64_t suid);
 
 /**
@@ -120,7 +121,7 @@ int32_t catalogRemoveStbMeta(SCatalog* pCtg, const char* dbFName, uint64_t dbId,
  * @param pCatalog (input, got with catalogGetHandle)
  * @param pTransporter (input, rpc object)
  * @param pMgmtEps (input, mnode EPs)
- * @param pTableName (input, table name, NOT including db name)
+ * @param pTableName (input, table name)
  * @param pTableMeta(output, table meta data, NEED to free it by calller)
  * @return error code
  */
@@ -131,7 +132,7 @@ int32_t catalogGetTableMeta(SCatalog* pCatalog, void * pTransporter, const SEpSe
  * @param pCatalog (input, got with catalogGetHandle)
  * @param pTransporter (input, rpc object)
  * @param pMgmtEps (input, mnode EPs)
- * @param pTableName (input, table name, NOT including db name)
+ * @param pTableName (input, table name)
  * @param pTableMeta(output, table meta data, NEED to free it by calller)
  * @return error code
  */
@@ -140,28 +141,38 @@ int32_t catalogGetSTableMeta(SCatalog* pCatalog, void * pTransporter, const SEpS
 int32_t catalogUpdateSTableMeta(SCatalog* pCatalog, STableMetaRsp *rspMsg);
 
 
+/**
+ * Force refresh DB's local cached vgroup info. 
+ * @param pCtg (input, got with catalogGetHandle)
+ * @param pTrans (input, rpc object)
+ * @param pMgmtEps (input, mnode EPs)
+ * @param dbFName (input, db full name)
+ * @return error code
+ */
+int32_t catalogRefreshDBVgInfo(SCatalog* pCtg, void *pTrans, const SEpSet* pMgmtEps, const char* dbFName);
+
 /**
  * Force refresh a table's local cached meta data. 
  * @param pCatalog (input, got with catalogGetHandle)
  * @param pTransporter (input, rpc object)
  * @param pMgmtEps (input, mnode EPs)
- * @param pTableName (input, table name, NOT including db name)
+ * @param pTableName (input, table name)
  * @param isSTable (input, is super table or not, 1:supposed to be stable, 0: supposed not to be stable, -1:not sure) 
  * @return error code
  */
-  int32_t catalogRefreshTableMeta(SCatalog* pCatalog, void *pTransporter, const SEpSet* pMgmtEps, const SName* pTableName, int32_t isSTable);
+int32_t catalogRefreshTableMeta(SCatalog* pCatalog, void *pTransporter, const SEpSet* pMgmtEps, const SName* pTableName, int32_t isSTable);
 
 /**
  * Force refresh a table's local cached meta data and get the new one. 
  * @param pCatalog (input, got with catalogGetHandle)
  * @param pTransporter (input, rpc object)
  * @param pMgmtEps (input, mnode EPs)
- * @param pTableName (input, table name, NOT including db name)
+ * @param pTableName (input, table name)
  * @param pTableMeta(output, table meta data, NEED to free it by calller) 
  * @param isSTable (input, is super table or not, 1:supposed to be stable, 0: supposed not to be stable, -1:not sure) 
  * @return error code
  */
-  int32_t catalogRefreshGetTableMeta(SCatalog* pCatalog, void *pTransporter, const SEpSet* pMgmtEps, const SName* pTableName, STableMeta** pTableMeta, int32_t isSTable);
+int32_t catalogRefreshGetTableMeta(SCatalog* pCatalog, void *pTransporter, const SEpSet* pMgmtEps, const SName* pTableName, STableMeta** pTableMeta, int32_t isSTable);
 
 
 
@@ -170,7 +181,7 @@ int32_t catalogUpdateSTableMeta(SCatalog* pCatalog, STableMetaRsp *rspMsg);
  * @param pCatalog (input, got with catalogGetHandle)
  * @param pTransporter (input, rpc object)
  * @param pMgmtEps (input, mnode EPs)
- * @param pTableName (input, table name, NOT including db name)
+ * @param pTableName (input, table name)
  * @param pVgroupList (output, vgroup info list, element is SVgroupInfo, NEED to simply free the array by caller)
  * @return error code
  */
@@ -181,7 +192,7 @@ int32_t catalogGetTableDistVgInfo(SCatalog* pCatalog, void *pTransporter, const
  * @param pCatalog (input, got with catalogGetHandle)
  * @param pTransporter (input, rpc object)
  * @param pMgmtEps (input, mnode EPs)
- * @param pTableName (input, table name, NOT including db name)
+ * @param pTableName (input, table name)
  * @param vgInfo (output, vgroup info)
  * @return error code
  */

include/libs/executor/executor.h

@@ -21,6 +21,7 @@ extern "C" {
 #endif
 
 #include "tcommon.h"
+#include "query.h"
 
 typedef void* qTaskInfo_t;
 typedef void* DataSinkHandle;
@@ -30,23 +31,28 @@ struct SSubplan;
 typedef struct SReadHandle {
   void* reader;
   void* meta;
+  void* config;
 } SReadHandle;
 
- /**
-  * Create the exec task for streaming mode
-  * @param pMsg
-  * @param streamReadHandle
-  * @return
-  */
-qTaskInfo_t qCreateStreamExecTaskInfo(void *msg, void* streamReadHandle);
+#define STREAM_DATA_TYPE_SUBMIT_BLOCK 0x1
+#define STREAM_DATA_TYPE_SSDATA_BLOCK 0x2
+  
+/**
+ * Create the exec task for streaming mode
+ * @param pMsg
+ * @param streamReadHandle
+ * @return
+ */
+qTaskInfo_t qCreateStreamExecTaskInfo(void* msg, void* streamReadHandle);
 
 /**
  * Set the input data block for the stream scan.
  * @param tinfo
  * @param input
+ * @param type
  * @return
  */
-int32_t qSetStreamInput(qTaskInfo_t tinfo, const void* input);
+int32_t qSetStreamInput(qTaskInfo_t tinfo, const void* input, int32_t type);
 
 /**
  * Update the table id list, add or remove.
@@ -58,16 +64,17 @@ int32_t qSetStreamInput(qTaskInfo_t tinfo, const void* input);
  */
 int32_t qUpdateQualifiedTableId(qTaskInfo_t tinfo, SArray* tableIdList, bool isAdd);
 
- /**
-  * Create the exec task object according to task json
-  * @param readHandle
-  * @param vgId
-  * @param pTaskInfoMsg
-  * @param pTaskInfo
-  * @param qId
-  * @return
-  */
-int32_t qCreateExecTask(SReadHandle* readHandle, int32_t vgId, uint64_t taskId, struct SSubplan* pPlan, qTaskInfo_t* pTaskInfo, DataSinkHandle* handle);
+/**
+ * Create the exec task object according to task json
+ * @param readHandle
+ * @param vgId
+ * @param pTaskInfoMsg
+ * @param pTaskInfo
+ * @param qId
+ * @return
+ */
+int32_t qCreateExecTask(SReadHandle* readHandle, int32_t vgId, uint64_t taskId, struct SSubplan* pPlan,
+                        qTaskInfo_t* pTaskInfo, DataSinkHandle* handle);
 
 /**
  * The main task execution function, including query on both table and multiple tables,
@@ -77,7 +84,7 @@ int32_t qCreateExecTask(SReadHandle* readHandle, int32_t vgId, uint64_t taskId,
  * @param handle
  * @return
  */
-int32_t qExecTask(qTaskInfo_t tinfo, SSDataBlock** pRes, uint64_t *useconds);
+int32_t qExecTask(qTaskInfo_t tinfo, SSDataBlock** pRes, uint64_t* useconds);
 
 /**
  * Retrieve the produced results information, if current query is not paused or completed,
@@ -140,7 +147,8 @@ int32_t qGetQualifiedTableIdList(void* pTableList, const char* tagCond, int32_t
  * @param numOfIndex
  * @return
  */
-//int32_t qCreateTableGroupByGroupExpr(SArray* pTableIdList, TSKEY skey, STableGroupInfo groupInfo, SColIndex* groupByIndex, int32_t numOfIndex);
+// int32_t qCreateTableGroupByGroupExpr(SArray* pTableIdList, TSKEY skey, STableGroupInfo groupInfo, SColIndex*
+// groupByIndex, int32_t numOfIndex);
 
 /**
  * Update the table id list of a given query.
@@ -163,19 +171,19 @@ void* qOpenTaskMgmt(int32_t vgId);
  * broadcast the close information and wait for all query stop.
  * @param pExecutor
  */
-void  qTaskMgmtNotifyClosing(void* pExecutor);
+void qTaskMgmtNotifyClosing(void* pExecutor);
 
 /**
  * Re-open the query handle management module when opening the vnode again.
  * @param pExecutor
  */
-void  qQueryMgmtReOpen(void *pExecutor);
+void qQueryMgmtReOpen(void* pExecutor);
 
 /**
  * Close query mgmt and clean up resources.
  * @param pExecutor
  */
-void  qCleanupTaskMgmt(void* pExecutor);
+void qCleanupTaskMgmt(void* pExecutor);
 
 /**
  * Add the query into the query mgmt object
@@ -184,7 +192,7 @@ void  qCleanupTaskMgmt(void* pExecutor);
  * @param qInfo
  * @return
  */
-void** qRegisterTask(void* pMgmt, uint64_t qId, void *qInfo);
+void** qRegisterTask(void* pMgmt, uint64_t qId, void* qInfo);
 
 /**
  * acquire the query handle according to the key from query mgmt object.

include/libs/function/function.h

@@ -163,7 +163,7 @@ typedef struct SInputColumnInfoData {
 typedef struct SqlFunctionCtx {
   SInputColumnInfoData input;
   SResultDataInfo      resDataInfo;
-  uint32_t             order;  // asc|desc
+  uint32_t             order;  // data block scanner order: asc|desc
   ////////////////////////////////////////////////////////////////
   int32_t          startRow;   // start row index
   int32_t          size;       // handled processed row number
@@ -327,7 +327,7 @@ bool taosFillHasMoreResults(struct SFillInfo* pFillInfo);
 
 struct SFillInfo* taosCreateFillInfo(int32_t order, TSKEY skey, int32_t numOfTags, int32_t capacity, int32_t numOfCols,
                               int64_t slidingTime, int8_t slidingUnit, int8_t precision, int32_t fillType,
-                              struct SFillColInfo* pFillCol, void* handle);
+                              struct SFillColInfo* pFillCol, const char* id);
 
 void* taosDestroyFillInfo(struct SFillInfo *pFillInfo);
 int64_t taosFillResultDataBlock(struct SFillInfo* pFillInfo, void** output, int32_t capacity);

include/libs/function/functionMgt.h

@@ -57,18 +57,19 @@ typedef enum EFunctionType {
 
   // math function
   FUNCTION_TYPE_ABS = 1000,
-  FUNCTION_TYPE_ACOS,
-  FUNCTION_TYPE_ASION,
-  FUNCTION_TYPE_ATAN,
-  FUNCTION_TYPE_CEIL,
-  FUNCTION_TYPE_COS,
-  FUNCTION_TYPE_FLOOR,
   FUNCTION_TYPE_LOG,
   FUNCTION_TYPE_POW,
+  FUNCTION_TYPE_SQRT,
+  FUNCTION_TYPE_CEIL,
+  FUNCTION_TYPE_FLOOR,
   FUNCTION_TYPE_ROUND,
+
   FUNCTION_TYPE_SIN,
-  FUNCTION_TYPE_SQRT,
+  FUNCTION_TYPE_COS,
   FUNCTION_TYPE_TAN,
+  FUNCTION_TYPE_ASIN,
+  FUNCTION_TYPE_ACOS,
+  FUNCTION_TYPE_ATAN,
 
   // string function
   FUNCTION_TYPE_CHAR_LENGTH = 1500,

include/libs/nodes/cmdnodes.h

@@ -179,7 +179,8 @@ typedef struct SAlterDnodeStmt {
 
 typedef struct SShowStmt {
   ENodeType type;
-  char dbName[TSDB_DB_NAME_LEN];
+  SNode* pDbName;        // SValueNode
+  SNode* pTbNamePattern; // SValueNode
 } SShowStmt;
 
 typedef enum EIndexType {
@@ -198,6 +199,7 @@ typedef struct SIndexOptions {
 typedef struct SCreateIndexStmt {
   ENodeType type;
   EIndexType indexType;
+  bool ignoreExists;
   char indexName[TSDB_INDEX_NAME_LEN];
   char tableName[TSDB_TABLE_NAME_LEN];
   SNodeList* pCols;
@@ -206,6 +208,7 @@ typedef struct SCreateIndexStmt {
 
 typedef struct SDropIndexStmt {
   ENodeType type;
+  bool ignoreNotExists;
   char indexName[TSDB_INDEX_NAME_LEN];
   char tableName[TSDB_TABLE_NAME_LEN];
 } SDropIndexStmt;

include/libs/nodes/nodes.h

@@ -78,7 +78,6 @@ typedef enum ENodeType {
   QUERY_NODE_CREATE_DATABASE_STMT,
   QUERY_NODE_DROP_DATABASE_STMT,
   QUERY_NODE_ALTER_DATABASE_STMT,
-  QUERY_NODE_SHOW_DATABASES_STMT, // temp
   QUERY_NODE_CREATE_TABLE_STMT,
   QUERY_NODE_CREATE_SUBTABLE_CLAUSE,
   QUERY_NODE_CREATE_MULTI_TABLE_STMT,
@@ -86,20 +85,13 @@ typedef enum ENodeType {
   QUERY_NODE_DROP_TABLE_STMT,
   QUERY_NODE_DROP_SUPER_TABLE_STMT,
   QUERY_NODE_ALTER_TABLE_STMT,
-  QUERY_NODE_SHOW_TABLES_STMT, // temp
-  QUERY_NODE_SHOW_STABLES_STMT,
   QUERY_NODE_CREATE_USER_STMT,
   QUERY_NODE_ALTER_USER_STMT,
   QUERY_NODE_DROP_USER_STMT,
-  QUERY_NODE_SHOW_USERS_STMT,
   QUERY_NODE_USE_DATABASE_STMT,
   QUERY_NODE_CREATE_DNODE_STMT,
   QUERY_NODE_DROP_DNODE_STMT,
   QUERY_NODE_ALTER_DNODE_STMT,
-  QUERY_NODE_SHOW_DNODES_STMT,
-  QUERY_NODE_SHOW_VGROUPS_STMT,
-  QUERY_NODE_SHOW_MNODES_STMT,
-  QUERY_NODE_SHOW_QNODES_STMT,
   QUERY_NODE_CREATE_INDEX_STMT,
   QUERY_NODE_DROP_INDEX_STMT,
   QUERY_NODE_CREATE_QNODE_STMT,
@@ -107,6 +99,18 @@ typedef enum ENodeType {
   QUERY_NODE_CREATE_TOPIC_STMT,
   QUERY_NODE_DROP_TOPIC_STMT,
   QUERY_NODE_ALTER_LOCAL_STMT,
+  QUERY_NODE_SHOW_DATABASES_STMT,
+  QUERY_NODE_SHOW_TABLES_STMT,
+  QUERY_NODE_SHOW_STABLES_STMT,
+  QUERY_NODE_SHOW_USERS_STMT,
+  QUERY_NODE_SHOW_DNODES_STMT,
+  QUERY_NODE_SHOW_VGROUPS_STMT,
+  QUERY_NODE_SHOW_MNODES_STMT,
+  QUERY_NODE_SHOW_MODULES_STMT,
+  QUERY_NODE_SHOW_QNODES_STMT,
+  QUERY_NODE_SHOW_FUNCTIONS_STMT,
+  QUERY_NODE_SHOW_INDEXES_STMT,
+  QUERY_NODE_SHOW_STREAMS_STMT,
 
   // logic plan node
   QUERY_NODE_LOGIC_PLAN_SCAN,
@@ -131,6 +135,7 @@ typedef enum ENodeType {
   QUERY_NODE_PHYSICAL_PLAN_EXCHANGE,
   QUERY_NODE_PHYSICAL_PLAN_SORT,
   QUERY_NODE_PHYSICAL_PLAN_INTERVAL,
+  QUERY_NODE_PHYSICAL_PLAN_SESSION_WINDOW,
   QUERY_NODE_PHYSICAL_PLAN_DISPATCH,
   QUERY_NODE_PHYSICAL_PLAN_INSERT,
   QUERY_NODE_PHYSICAL_SUBPLAN,
@@ -165,6 +170,7 @@ void nodesDestroyNode(SNodeptr pNode);
 SNodeList* nodesMakeList();
 int32_t nodesListAppend(SNodeList* pList, SNodeptr pNode);
 int32_t nodesListStrictAppend(SNodeList* pList, SNodeptr pNode);
+int32_t nodesListMakeAppend(SNodeList** pList, SNodeptr pNode);
 int32_t nodesListAppendList(SNodeList* pTarget, SNodeList* pSrc);
 int32_t nodesListStrictAppendList(SNodeList* pTarget, SNodeList* pSrc);
 SListCell* nodesListErase(SNodeList* pList, SListCell* pCell);

include/libs/nodes/plannodes.h

@@ -35,7 +35,7 @@ typedef struct SLogicNode {
 typedef enum EScanType {
   SCAN_TYPE_TAG,
   SCAN_TYPE_TABLE,
-  SCAN_TYPE_STABLE,
+  SCAN_TYPE_SYSTEM_TABLE,
   SCAN_TYPE_STREAM
 } EScanType;
 
@@ -48,6 +48,7 @@ typedef struct SScanLogicNode {
   uint8_t scanFlag;         // denotes reversed scan of data or not
   STimeWindow scanRange;
   SName tableName;
+  bool showRewrite;
 } SScanLogicNode;
 
 typedef struct SJoinLogicNode {
@@ -95,6 +96,7 @@ typedef struct SWindowLogicNode {
   int8_t  intervalUnit;
   int8_t  slidingUnit;
   SFillNode* pFill;
+  int64_t sessionGap;
 } SWindowLogicNode;
 
 typedef enum ESubplanType {
@@ -110,7 +112,7 @@ typedef struct SSubplanId {
   int32_t subplanId;
 } SSubplanId;
 
-typedef struct SSubLogicPlan {
+typedef struct SLogicSubplan {
   ENodeType type;
   SSubplanId id;
   SNodeList* pChildren;
@@ -120,7 +122,7 @@ typedef struct SSubLogicPlan {
   SVgroupsInfo* pVgroupList;
   int32_t level;
   int32_t splitFlag;
-} SSubLogicPlan;
+} SLogicSubplan;
 
 typedef struct SQueryLogicPlan {
   ENodeType type;
@@ -154,7 +156,7 @@ typedef struct SPhysiNode {
 } SPhysiNode;
 
 typedef struct SScanPhysiNode {
-  SPhysiNode  node;
+  SPhysiNode node;
   SNodeList* pScanCols;
   uint64_t uid;           // unique id of the table
   int8_t tableType;
@@ -164,10 +166,16 @@ typedef struct SScanPhysiNode {
   SName tableName;
 } SScanPhysiNode;
 
-typedef SScanPhysiNode SSystemTableScanPhysiNode;
 typedef SScanPhysiNode STagScanPhysiNode;
 typedef SScanPhysiNode SStreamScanPhysiNode;
 
+typedef struct SSystemTableScanPhysiNode {
+  SScanPhysiNode scan;
+  SEpSet mgmtEpSet;
+  bool showRewrite;
+  int32_t accountId;
+} SSystemTableScanPhysiNode;
+
 typedef struct STableScanPhysiNode {
   SScanPhysiNode scan;
   uint8_t scanFlag;         // denotes reversed scan of data or not
@@ -209,10 +217,14 @@ typedef struct SExchangePhysiNode {
   SNodeList* pSrcEndPoints;  // element is SDownstreamSource, scheduler fill by calling qSetSuplanExecutionNode
 } SExchangePhysiNode;
 
-typedef struct SIntervalPhysiNode {
+typedef struct SWinodwPhysiNode {
   SPhysiNode node;
   SNodeList* pExprs;   // these are expression list of parameter expression of function
   SNodeList* pFuncs;
+} SWinodwPhysiNode;
+
+typedef struct SIntervalPhysiNode {
+  SWinodwPhysiNode window;
   int64_t    interval;
   int64_t    offset;
   int64_t    sliding;
@@ -221,6 +233,11 @@ typedef struct SIntervalPhysiNode {
   SFillNode* pFill;
 } SIntervalPhysiNode;
 
+typedef struct SSessionWinodwPhysiNode {
+  SWinodwPhysiNode window;
+  int64_t    gap;
+} SSessionWinodwPhysiNode;
+
 typedef struct SDataSinkNode {
   ENodeType type;
   SDataBlockDescNode* pInputDataBlockDesc;
@@ -243,6 +260,7 @@ typedef struct SSubplan {
   ESubplanType subplanType;
   int32_t msgType;      // message type for subplan, used to denote the send message type to vnode.
   int32_t level;        // the execution level of current subplan, starting from 0 in a top-down manner.
+  char dbFName[TSDB_DB_FNAME_LEN];
   SQueryNodeAddr execNode;    // for the scan/modify subplan, the optional execution node
   SQueryNodeStat execNodeStat; // only for scan subplan
   SNodeList* pChildren;    // the datasource subplan,from which to fetch the result

include/libs/nodes/querynodes.h

@@ -23,7 +23,8 @@ extern "C" {
 #include "nodes.h"
 #include "tmsg.h"
 
-#define TABLE_META_SIZE(pMeta) (NULL == (pMeta) ? 0 : (sizeof(STableMeta) + ((pMeta)->tableInfo.numOfColumns + (pMeta)->tableInfo.numOfTags) * sizeof(SSchema)))
+#define TABLE_TOTAL_COL_NUM(pMeta) ((pMeta)->tableInfo.numOfColumns + (pMeta)->tableInfo.numOfTags)
+#define TABLE_META_SIZE(pMeta) (NULL == (pMeta) ? 0 : (sizeof(STableMeta) + TABLE_TOTAL_COL_NUM((pMeta)) * sizeof(SSchema)))
 #define VGROUPS_INFO_SIZE(pInfo) (NULL == (pInfo) ? 0 : (sizeof(SVgroupsInfo) + (pInfo)->numOfVgroups * sizeof(SVgroupInfo)))
 
 typedef struct SRawExprNode {
@@ -129,6 +130,7 @@ typedef struct SRealTableNode {
   STableNode table; // QUERY_NODE_REAL_TABLE
   struct STableMeta* pMeta;
   SVgroupsInfo* pVgroupList;
+  char useDbName[TSDB_DB_NAME_LEN];
 } SRealTableNode;
 
 typedef struct STempTableNode {
@@ -189,8 +191,8 @@ typedef struct SStateWindowNode {
 
 typedef struct SSessionWindowNode {
   ENodeType type; // QUERY_NODE_SESSION_WINDOW
-  int64_t gap;             // gap between two session window(in microseconds)
   SNode* pCol;
+  SNode* pGap; // gap between two session window(in microseconds)
 } SSessionWindowNode;
 
 typedef struct SIntervalWindowNode {

include/libs/parser/parser.h

@@ -26,7 +26,7 @@ typedef struct SParseContext {
   uint64_t         requestId;
   int32_t          acctId;
   const char      *db;
-  bool             streamQuery;
+  bool             topicQuery;
   void            *pTransporter;
   SEpSet           mgmtEpSet;
   const char      *pSql;           // sql string
@@ -52,13 +52,17 @@ typedef struct SQuery {
   SSchema* pResSchema;
   SCmdMsgInfo* pCmdMsg;
   int32_t msgType;
-  bool streamQuery;
+  SArray* pDbList;
+  SArray* pTableList;
+  bool showRewrite;
 } SQuery;
 
 int32_t qParseQuerySql(SParseContext* pCxt, SQuery** pQuery);
 
 void qDestroyQuery(SQuery* pQueryNode);
 
+int32_t qExtractResultSchema(const SNode* pRoot, int32_t* numOfCols, SSchema** pSchema);
+
 #ifdef __cplusplus
 }
 #endif

include/libs/planner/planner.h

@@ -25,8 +25,11 @@ extern "C" {
 typedef struct SPlanContext {
   uint64_t queryId;
   int32_t acctId;
+  SEpSet mgmtEpSet;
   SNode* pAstRoot;
+  bool topicQuery;
   bool streamQuery;
+  bool showRewrite;
 } SPlanContext;
 
 // Create the physical plan for the query, according to the AST.

include/libs/qcom/query.h

@@ -25,7 +25,7 @@ extern "C" {
 #include "tlog.h"
 #include "tmsg.h"
 
-enum {
+typedef enum {
   JOB_TASK_STATUS_NULL = 0,
   JOB_TASK_STATUS_NOT_START = 1,
   JOB_TASK_STATUS_EXECUTING,
@@ -35,12 +35,12 @@ enum {
   JOB_TASK_STATUS_CANCELLING,
   JOB_TASK_STATUS_CANCELLED,
   JOB_TASK_STATUS_DROPPING,
-};
+} EJobTaskType;
 
-enum {
+typedef enum {
   TASK_TYPE_PERSISTENT = 1,
   TASK_TYPE_TEMP,
-};
+} ETaskType;
 
 typedef struct STableComInfo {
   uint8_t numOfTags;     // the number of tags in schema
@@ -150,6 +150,8 @@ int32_t cleanupTaskQueue();
  */
 int32_t taosAsyncExec(__async_exec_fn_t execFn, void* execParam, int32_t* code);
 
+int32_t asyncSendMsgToServerExt(void* pTransporter, SEpSet* epSet, int64_t* pTransporterId, const SMsgSendInfo* pInfo, bool persistHandle, void *ctx);
+
 /**
  * Asynchronously send message to server, after the response received, the callback will be incured.
  *
@@ -169,6 +171,9 @@ const SSchema* tGetTbnameColumnSchema();
 bool           tIsValidSchema(struct SSchema* pSchema, int32_t numOfCols, int32_t numOfTags);
 
 int32_t queryCreateTableMetaFromMsg(STableMetaRsp* msg, bool isSuperTable, STableMeta** pMeta);
+char *jobTaskStatusStr(int32_t status);
+
+SSchema createSchema(uint8_t type, int32_t bytes, int32_t colId, const char* name);
 
 extern int32_t (*queryBuildMsg[TDMT_MAX])(void* input, char** msg, int32_t msgSize, int32_t* msgLen);
 extern int32_t (*queryProcessMsgRsp[TDMT_MAX])(void* output, char* msg, int32_t msgSize);
@@ -178,6 +183,15 @@ extern int32_t (*queryProcessMsgRsp[TDMT_MAX])(void* output, char* msg, int32_t
 #define SET_META_TYPE_TABLE(t)      (t) = META_TYPE_TABLE
 #define SET_META_TYPE_BOTH_TABLE(t) (t) = META_TYPE_BOTH_TABLE
 
+#define NEED_CLIENT_RM_TBLMETA_ERROR(_code) ((_code) == TSDB_CODE_TDB_INVALID_TABLE_ID || (_code) == TSDB_CODE_VND_TB_NOT_EXIST)
+#define NEED_CLIENT_REFRESH_VG_ERROR(_code) ((_code) == TSDB_CODE_VND_HASH_MISMATCH || (_code) == TSDB_CODE_VND_INVALID_VGROUP_ID)
+#define NEED_CLIENT_REFRESH_TBLMETA_ERROR(_code) ((_code) == TSDB_CODE_TDB_TABLE_RECREATED)
+#define NEED_CLIENT_HANDLE_ERROR(_code) (NEED_CLIENT_RM_TBLMETA_ERROR(_code) || NEED_CLIENT_REFRESH_VG_ERROR(_code) || NEED_CLIENT_REFRESH_TBLMETA_ERROR(_code))
+
+#define NEED_SCHEDULER_RETRY_ERROR(_code) ((_code) == TSDB_CODE_RPC_REDIRECT || (_code) == TSDB_CODE_RPC_NETWORK_UNAVAIL)
+
+#define REQUEST_MAX_TRY_TIMES 5
+
 #define qFatal(...)                                                     \
   do {                                                                  \
     if (qDebugFlag & DEBUG_FATAL) {                                     \

include/libs/qworker/qworker.h

@@ -20,6 +20,7 @@
 extern "C" {
 #endif
 
+#include "tmsgcb.h"
 #include "trpc.h"
 
 
@@ -27,6 +28,7 @@ enum {
   NODE_TYPE_VNODE = 1,
   NODE_TYPE_QNODE,
   NODE_TYPE_SNODE,
+  NODE_TYPE_MNODE,
 };
 
 
@@ -48,11 +50,7 @@ typedef struct {
   uint64_t numOfErrors;
 } SQWorkerStat;
 
-typedef int32_t (*putReqToQueryQFp)(void *, struct SRpcMsg *);
-typedef int32_t (*sendReqToDnodeFp)(void *, struct SEpSet *, struct SRpcMsg *);
-
-int32_t qWorkerInit(int8_t nodeType, int32_t nodeId, SQWorkerCfg *cfg, void **qWorkerMgmt, void *nodeObj,
-                    putReqToQueryQFp fp1, sendReqToDnodeFp fp2);
+int32_t qWorkerInit(int8_t nodeType, int32_t nodeId, SQWorkerCfg *cfg, void **qWorkerMgmt, const SMsgCb *pMsgCb);
 
 int32_t qWorkerProcessQueryMsg(void *node, void *qWorkerMgmt, SRpcMsg *pMsg);
 

include/libs/scalar/scalar.h

@@ -42,6 +42,21 @@ int32_t scalarGenerateSetFromList(void **data, void *pNode, uint32_t type);
 int32_t vectorGetConvertType(int32_t type1, int32_t type2);
 int32_t vectorConvertImpl(SScalarParam* pIn, SScalarParam* pOut);
 
+int32_t absFunction(SScalarParam *pInput, int32_t inputNum, SScalarParam *pOutput);
+int32_t logFunction(SScalarParam *pInput, int32_t inputNum, SScalarParam *pOutput);
+int32_t powFunction(SScalarParam *pInput, int32_t inputNum, SScalarParam *pOutput);
+int32_t sqrtFunction(SScalarParam *pInput, int32_t inputNum, SScalarParam *pOutput);
+
+int32_t sinFunction(SScalarParam *pInput, int32_t inputNum, SScalarParam *pOutput);
+int32_t cosFunction(SScalarParam *pInput, int32_t inputNum, SScalarParam *pOutput);
+int32_t tanFunction(SScalarParam *pInput, int32_t inputNum, SScalarParam *pOutput);
+int32_t asinFunction(SScalarParam *pInput, int32_t inputNum, SScalarParam *pOutput);
+int32_t acosFunction(SScalarParam *pInput, int32_t inputNum, SScalarParam *pOutput);
+int32_t atanFunction(SScalarParam *pInput, int32_t inputNum, SScalarParam *pOutput);
+
+int32_t ceilFunction(SScalarParam *pInput, int32_t inputNum, SScalarParam *pOutput);
+int32_t floorFunction(SScalarParam *pInput, int32_t inputNum, SScalarParam *pOutput);
+int32_t roundFunction(SScalarParam *pInput, int32_t inputNum, SScalarParam *pOutput);
 
 #ifdef __cplusplus
 }

include/libs/scheduler/scheduler.h

@@ -52,10 +52,10 @@ typedef struct SQueryProfileSummary {
 } SQueryProfileSummary;
 
 typedef struct SQueryResult {
-  int32_t  code;
-  uint64_t numOfRows;
-  int32_t  msgSize;
-  char    *msg;
+  int32_t         code;
+  uint64_t        numOfRows;
+  int32_t         msgSize;
+  char           *msg;
 } SQueryResult;
 
 typedef struct STaskInfo {

include/libs/stream/tstream.h

+/*
+ * Copyright (c) 2019 TAOS Data, Inc. <jhtao@taosdata.com>
+ *
+ * This program is free software: you can use, redistribute, and/or modify
+ * it under the terms of the GNU Affero General Public License, version 3
+ * or later ("AGPL"), as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.
+ *
+ * You should have received a copy of the GNU Affero General Public License
+ * along with this program. If not, see <http://www.gnu.org/licenses/>.
+ */
+
+#include "tdatablock.h"
+#include "tmsg.h"
+#include "tmsgcb.h"
+#include "trpc.h"
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#ifndef _TSTREAM_H_
+#define _TSTREAM_H_
+
+enum {
+  STREAM_TASK_STATUS__RUNNING = 1,
+  STREAM_TASK_STATUS__STOP,
+};
+
+#if 0
+// pipe  -> fetch/pipe queue
+// merge -> merge      queue
+// write -> write      queue
+enum {
+  TASK_DISPATCH_MSG__SND_PIPE = 1,
+  TASK_DISPATCH_MSG__SND_MERGE,
+  TASK_DISPATCH_MSG__VND_PIPE,
+  TASK_DISPATCH_MSG__VND_MERGE,
+  TASK_DISPATCH_MSG__VND_WRITE,
+};
+#endif
+
+typedef struct {
+  int32_t nodeId;  // 0 for snode
+  SEpSet  epSet;
+} SStreamTaskEp;
+
+typedef struct {
+  void* inputHandle;
+  void* executor;
+} SStreamRunner;
+
+typedef struct {
+  int8_t parallelizable;
+  char*  qmsg;
+  // followings are not applicable to encoder and decoder
+  int8_t         numOfRunners;
+  SStreamRunner* runners;
+} STaskExec;
+
+typedef struct {
+  int8_t reserved;
+} STaskDispatcherInplace;
+
+typedef struct {
+  int32_t nodeId;
+  SEpSet  epSet;
+} STaskDispatcherFixedEp;
+
+typedef struct {
+  int8_t  hashMethod;
+  SArray* info;
+} STaskDispatcherShuffle;
+
+typedef struct {
+  int8_t reserved;
+  // not applicable to encoder and decoder
+  SHashObj* pHash;  // groupId to tbuid
+} STaskSinkTb;
+
+typedef struct {
+  int8_t reserved;
+} STaskSinkSma;
+
+typedef struct {
+  int8_t reserved;
+} STaskSinkFetch;
+
+typedef struct {
+  int8_t reserved;
+} STaskSinkShow;
+
+enum {
+  TASK_SOURCE__SCAN = 1,
+  TASK_SOURCE__PIPE,
+  TASK_SOURCE__MERGE,
+};
+
+enum {
+  TASK_EXEC__NONE = 1,
+  TASK_EXEC__PIPE,
+  TASK_EXEC__MERGE,
+};
+
+enum {
+  TASK_DISPATCH__NONE = 1,
+  TASK_DISPATCH__INPLACE,
+  TASK_DISPATCH__FIXED,
+  TASK_DISPATCH__SHUFFLE,
+};
+
+enum {
+  TASK_SINK__NONE = 1,
+  TASK_SINK__TABLE,
+  TASK_SINK__SMA,
+  TASK_SINK__FETCH,
+  TASK_SINK__SHOW,
+};
+
+typedef struct {
+  int64_t streamId;
+  int32_t taskId;
+  int8_t  status;
+
+  int8_t  sourceType;
+  int8_t  execType;
+  int8_t  sinkType;
+  int8_t  dispatchType;
+  int16_t dispatchMsgType;
+  int32_t downstreamTaskId;
+
+  int32_t nodeId;
+  SEpSet  epSet;
+
+  // source preprocess
+
+  // exec
+  STaskExec exec;
+
+  // local sink
+  union {
+    STaskSinkTb    tbSink;
+    STaskSinkSma   smaSink;
+    STaskSinkFetch fetchSink;
+    STaskSinkShow  showSink;
+  };
+
+  // dispatch
+  union {
+    STaskDispatcherInplace inplaceDispatcher;
+    STaskDispatcherFixedEp fixedEpDispatcher;
+    STaskDispatcherShuffle shuffleDispatcher;
+  };
+
+  // state storage
+
+} SStreamTask;
+
+SStreamTask* tNewSStreamTask(int64_t streamId);
+int32_t      tEncodeSStreamTask(SCoder* pEncoder, const SStreamTask* pTask);
+int32_t      tDecodeSStreamTask(SCoder* pDecoder, SStreamTask* pTask);
+void         tFreeSStreamTask(SStreamTask* pTask);
+
+typedef struct {
+  // SMsgHead     head;
+  SStreamTask* task;
+} SStreamTaskDeployReq;
+
+typedef struct {
+  int32_t reserved;
+} SStreamTaskDeployRsp;
+
+typedef struct {
+  // SMsgHead head;
+  int64_t streamId;
+  int32_t taskId;
+  SArray* data;  // SArray<SSDataBlock>
+} SStreamTaskExecReq;
+
+int32_t tEncodeSStreamTaskExecReq(void** buf, const SStreamTaskExecReq* pReq);
+void*   tDecodeSStreamTaskExecReq(const void* buf, SStreamTaskExecReq* pReq);
+void    tFreeSStreamTaskExecReq(SStreamTaskExecReq* pReq);
+
+typedef struct {
+  int32_t reserved;
+} SStreamTaskExecRsp;
+
+typedef struct {
+  // SMsgHead head;
+  int64_t streamId;
+  int64_t version;
+  SArray* res;  // SArray<SSDataBlock>
+} SStreamSinkReq;
+
+int32_t streamExecTask(SStreamTask* pTask, SMsgCb* pMsgCb, const void* input, int32_t inputType, int32_t workId);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* ifndef _TSTREAM_H_ */

include/libs/sync/sync.h

@@ -35,6 +35,7 @@ typedef enum {
   TAOS_SYNC_STATE_FOLLOWER = 100,
   TAOS_SYNC_STATE_CANDIDATE = 101,
   TAOS_SYNC_STATE_LEADER = 102,
+  TAOS_SYNC_STATE_ERROR = 103,
 } ESyncState;
 
 typedef struct SSyncBuffer {
@@ -68,17 +69,20 @@ typedef struct SSnapshot {
 typedef struct SSyncFSM {
   void* data;
 
-  // when value in pBuf finish a raft flow, FpCommitCb is called, code indicates the result
+  // when value in pMsg finish a raft flow, FpCommitCb is called, code indicates the result
   // user can do something according to the code and isWeak. for example, write data into tsdb
-  void (*FpCommitCb)(struct SSyncFSM* pFsm, const SRpcMsg* pBuf, SyncIndex index, bool isWeak, int32_t code);
+  void (*FpCommitCb)(struct SSyncFSM* pFsm, const SRpcMsg* pMsg, SyncIndex index, bool isWeak, int32_t code,
+                     ESyncState state);
 
-  // when value in pBuf has been written into local log store, FpPreCommitCb is called, code indicates the result
+  // when value in pMsg has been written into local log store, FpPreCommitCb is called, code indicates the result
   // user can do something according to the code and isWeak. for example, write data into tsdb
-  void (*FpPreCommitCb)(struct SSyncFSM* pFsm, const SRpcMsg* pBuf, SyncIndex index, bool isWeak, int32_t code);
+  void (*FpPreCommitCb)(struct SSyncFSM* pFsm, const SRpcMsg* pMsg, SyncIndex index, bool isWeak, int32_t code,
+                        ESyncState state);
 
   // when log entry is updated by a new one, FpRollBackCb is called
   // user can do something to roll back. for example, delete data from tsdb, or just ignore it
-  void (*FpRollBackCb)(struct SSyncFSM* pFsm, const SRpcMsg* pBuf, SyncIndex index, bool isWeak, int32_t code);
+  void (*FpRollBackCb)(struct SSyncFSM* pFsm, const SRpcMsg* pMsg, SyncIndex index, bool isWeak, int32_t code,
+                       ESyncState state);
 
   // user should implement this function, use "data" to take snapshot into "snapshot"
   int32_t (*FpTakeSnapshot)(SSnapshot* snapshot);
@@ -157,9 +161,14 @@ void    syncCleanUp();
 int64_t    syncStart(const SSyncInfo* pSyncInfo);
 void       syncStop(int64_t rid);
 int32_t    syncReconfig(int64_t rid, const SSyncCfg* pSyncCfg);
-int32_t    syncForwardToPeer(int64_t rid, const SRpcMsg* pMsg, bool isWeak);
+int32_t    syncPropose(int64_t rid, const SRpcMsg* pMsg, bool isWeak);
 ESyncState syncGetMyRole(int64_t rid);
-void       syncGetNodesRole(int64_t rid, SNodesRole* pNodeRole);
+
+// propose with sequence number, to implement linearizable semantics
+int32_t syncPropose2(int64_t rid, const SRpcMsg* pMsg, bool isWeak, uint64_t seqNum);
+
+// for compatibility, the same as syncPropose
+int32_t syncForwardToPeer(int64_t rid, const SRpcMsg* pMsg, bool isWeak);
 
 extern int32_t sDebugFlag;
 

include/libs/transport/trpc.h

@@ -38,16 +38,24 @@ typedef struct SRpcConnInfo {
 
 typedef struct SRpcMsg {
   tmsg_t  msgType;
-  tmsg_t  expectMsgType;
   void *  pCont;
   int     contLen;
   int32_t code;
-  void *  handle;   // rpc handle returned to app
-  void *  ahandle;  // app handle set by client
-  int     noResp;   // has response or not(default 0 indicate resp);
+  void *  handle;         // rpc handle returned to app
+  void *  ahandle;        // app handle set by client
+  int     noResp;         // has response or not(default 0, 0: resp, 1: no resp);
+  int     persistHandle;  // persist handle or not
 
 } SRpcMsg;
 
+typedef struct {
+  char    user[TSDB_USER_LEN];
+  SRpcMsg rpcMsg;
+  int32_t rspLen;
+  void *  pRsp;
+  void *  pNode;
+} SNodeMsg;
+
 typedef struct SRpcInit {
   uint16_t localPort;     // local port
   char *   label;         // for debug purpose
@@ -69,17 +77,26 @@ typedef struct SRpcInit {
   // call back to retrieve the client auth info, for server app only
   int (*afp)(void *parent, char *tableId, char *spi, char *encrypt, char *secret, char *ckey);
 
-  // call back to keep conn or not
-  bool (*pfp)(void *parent, tmsg_t msgType);
+  void *parent;
+} SRpcInit;
 
-  // to support Send messages multiple times on a link
-  void *(*mfp)(void *parent, tmsg_t msgType);
+typedef struct {
+  void     *val;
+  int32_t (*clone)(void *src, void **dst);
+  void    (*freeFunc)(const void *arg);
+} SRpcCtxVal;
 
-  // call back  to handle except when query/fetch in progress
-  bool (*efp)(void *parent, tmsg_t msgType);
+typedef struct {
+  int32_t   msgType;
+  void     *val;
+  int32_t (*clone)(void *src, void **dst);
+  void    (*freeFunc)(const void *arg);
+} SRpcBrokenlinkVal;
 
-  void *parent;
-} SRpcInit;
+typedef struct {
+  SHashObj *        args;
+  SRpcBrokenlinkVal brokenVal;
+} SRpcCtx;
 
 int32_t rpcInit();
 void    rpcCleanup();
@@ -89,16 +106,17 @@ void *  rpcMallocCont(int contLen);
 void    rpcFreeCont(void *pCont);
 void *  rpcReallocCont(void *ptr, int contLen);
 void    rpcSendRequest(void *thandle, const SEpSet *pEpSet, SRpcMsg *pMsg, int64_t *rid);
-void    rpcSendResponse(const SRpcMsg *pMsg);
-void    rpcSendRedirectRsp(void *pConn, const SEpSet *pEpSet);
-int     rpcGetConnInfo(void *thandle, SRpcConnInfo *pInfo);
-void    rpcSendRecv(void *shandle, SEpSet *pEpSet, SRpcMsg *pReq, SRpcMsg *pRsp);
-int     rpcReportProgress(void *pConn, char *pCont, int contLen);
-void    rpcCancelRequest(int64_t rid);
-
+void    rpcSendRequestWithCtx(void *thandle, const SEpSet *pEpSet, SRpcMsg *pMsg, int64_t *rid, SRpcCtx *ctx);
+
+void rpcSendResponse(const SRpcMsg *pMsg);
+void rpcSendRedirectRsp(void *pConn, const SEpSet *pEpSet);
+int  rpcGetConnInfo(void *thandle, SRpcConnInfo *pInfo);
+void rpcSendRecv(void *shandle, SEpSet *pEpSet, SRpcMsg *pReq, SRpcMsg *pRsp);
+int  rpcReportProgress(void *pConn, char *pCont, int contLen);
+void rpcCancelRequest(int64_t rid);
+void rpcRegisterBrokenLinkArg(SRpcMsg *msg);
 // just release client conn to rpc instance, no close sock
-void rpcReleaseHandle(void *handle, int8_t type);
-
+void rpcReleaseHandle(void *handle, int8_t type);  //
 void rpcRefHandle(void *handle, int8_t type);
 void rpcUnrefHandle(void *handle, int8_t type);
 

include/libs/wal/wal.h

@@ -61,30 +61,45 @@ extern "C" {
     }                                                             \
   }
 
-#define WAL_HEAD_VER 0
+#define WAL_HEAD_VER     0
 #define WAL_NOSUFFIX_LEN 20
-#define WAL_SUFFIX_AT (WAL_NOSUFFIX_LEN + 1)
-#define WAL_LOG_SUFFIX "log"
+#define WAL_SUFFIX_AT    (WAL_NOSUFFIX_LEN + 1)
+#define WAL_LOG_SUFFIX   "log"
 #define WAL_INDEX_SUFFIX "idx"
-#define WAL_REFRESH_MS 1000
-#define WAL_MAX_SIZE (TSDB_MAX_WAL_SIZE + sizeof(SWalHead))
-#define WAL_PATH_LEN (TSDB_FILENAME_LEN + 12)
-#define WAL_FILE_LEN (WAL_PATH_LEN + 32)
-#define WAL_MAGIC 0xFAFBFCFDULL
+#define WAL_REFRESH_MS   1000
+#define WAL_MAX_SIZE     (TSDB_MAX_WAL_SIZE + sizeof(SWalHead))
+#define WAL_PATH_LEN     (TSDB_FILENAME_LEN + 12)
+#define WAL_FILE_LEN     (WAL_PATH_LEN + 32)
+#define WAL_MAGIC        0xFAFBFCFDULL
 
 #define WAL_CUR_FAILED 1
 
 #pragma pack(push, 1)
-typedef enum { TAOS_WAL_NOLOG = 0, TAOS_WAL_WRITE = 1, TAOS_WAL_FSYNC = 2 } EWalType;
+typedef enum {
+  TAOS_WAL_NOLOG = 0,
+  TAOS_WAL_WRITE = 1,
+  TAOS_WAL_FSYNC = 2,
+} EWalType;
+
+// used by sync module
+typedef struct {
+  int8_t   isWeek;
+  uint64_t seqNum;
+  uint64_t term;
+} SSyncLogMeta;
 
 typedef struct SWalReadHead {
   int8_t  headVer;
-  int16_t msgType;
   int8_t  reserved;
+  int16_t msgType;
   int32_t len;
   int64_t ingestTs;  // not implemented
   int64_t version;
-  char    body[];
+
+  // sync meta
+  SSyncLogMeta syncMeta;
+
+  char body[];
 } SWalReadHead;
 
 typedef struct {
@@ -117,17 +132,17 @@ typedef struct SWal {
   SWalCfg cfg;
   int32_t fsyncSeq;
   // meta
-  SWalVer vers;
+  SWalVer   vers;
   TdFilePtr pWriteLogTFile;
   TdFilePtr pWriteIdxTFile;
-  int32_t writeCur;
-  SArray *fileInfoSet;
+  int32_t   writeCur;
+  SArray   *fileInfoSet;
   // status
   int64_t totSize;
   int64_t lastRollSeq;
   // ctl
-  int64_t         refId;
-  pthread_mutex_t mutex;
+  int64_t       refId;
+  TdThreadMutex mutex;
   // path
   char path[WAL_PATH_LEN];
   // reusable write head
@@ -158,6 +173,8 @@ int32_t walAlter(SWal *, SWalCfg *pCfg);
 void    walClose(SWal *);
 
 // write
+int64_t walWriteWithSyncInfo(SWal *, int64_t index, tmsg_t msgType, SSyncLogMeta syncMeta, const void *body,
+                             int32_t bodyLen);
 int64_t walWrite(SWal *, int64_t index, tmsg_t msgType, const void *body, int32_t bodyLen);
 void    walFsync(SWal *, bool force);
 

include/os/os.h

@@ -22,7 +22,6 @@ extern "C" {
 
 #include <assert.h>
 #include <ctype.h>
-#include <pthread.h>
 #include <semaphore.h>
 
 #include <regex.h>
@@ -83,6 +82,7 @@ extern "C" {
 #include "osMath.h"
 #include "osMemory.h"
 #include "osRand.h"
+#include "osThread.h"
 #include "osSemaphore.h"
 #include "osSignal.h"
 #include "osSleep.h"
@@ -90,7 +90,6 @@ extern "C" {
 #include "osString.h"
 #include "osSysinfo.h"
 #include "osSystem.h"
-#include "osThread.h"
 #include "osTime.h"
 #include "osTimer.h"
 #include "osTimezone.h"

include/os/osDir.h

@@ -17,6 +17,7 @@
 #define _TD_OS_DIR_H_
 
 // If the error is in a third-party library, place this header file under the third-party library header file.
+// When you want to use this feature, you should find or add the same function in the following section.
 #ifndef ALLOW_FORBID_FUNC
     #define opendir OPENDIR_FUNC_TAOS_FORBID
     #define readdir READDIR_FUNC_TAOS_FORBID

include/os/osEnv.h

@@ -45,6 +45,7 @@ extern SDiskSpace tsTempSpace;
 
 void osInit();
 void osUpdate();
+void osCleanup();
 bool osLogSpaceAvailable();
 void osSetTimezone(const char *timezone);
 

include/os/osFile.h

@@ -23,6 +23,7 @@ extern "C" {
 #include "osSocket.h"
 
 // If the error is in a third-party library, place this header file under the third-party library header file.
+// When you want to use this feature, you should find or add the same function in the following sectio
 #ifndef ALLOW_FORBID_FUNC
     #define open OPEN_FUNC_TAOS_FORBID
     #define fopen FOPEN_FUNC_TAOS_FORBID

include/os/osLocale.h

@@ -23,6 +23,7 @@ extern "C" {
 #endif
 
 // If the error is in a third-party library, place this header file under the third-party library header file.
+// When you want to use this feature, you should find or add the same function in the following section.
 #ifndef ALLOW_FORBID_FUNC
     #define setlocale SETLOCALE_FUNC_TAOS_FORBID
 #endif

include/os/osMemory.h

@@ -20,12 +20,25 @@
 extern "C" {
 #endif
 
-#define tfree(x)         \
-  do {                   \
-    if (x) {             \
-      free((void *)(x)); \
-      (x) = 0;           \
-    }                    \
+// If the error is in a third-party library, place this header file under the third-party library header file.
+// When you want to use this feature, you should find or add the same function in the following sectio
+#ifndef ALLOW_FORBID_FUNC
+    #define malloc MALLOC_FUNC_TAOS_FORBID
+    #define calloc CALLOC_FUNC_TAOS_FORBID
+    #define realloc REALLOC_FUNC_TAOS_FORBID
+    #define free FREE_FUNC_TAOS_FORBID
+#endif
+
+void *taosMemoryMalloc(int32_t size);
+void *taosMemoryCalloc(int32_t num, int32_t size);
+void *taosMemoryRealloc(void *ptr, int32_t size);
+void taosMemoryFree(const void *ptr);
+int32_t taosMemorySize(void *ptr);
+
+#define taosMemoryFreeClear(ptr) \
+  do {                           \
+    taosMemoryFree(ptr);         \
+    (ptr)=NULL;                  \
   } while (0)
 
 #ifdef __cplusplus

include/os/osRand.h

@@ -21,6 +21,7 @@ extern "C" {
 #endif
 
 // If the error is in a third-party library, place this header file under the third-party library header file.
+// When you want to use this feature, you should find or add the same function in the following section.
 #ifndef ALLOW_FORBID_FUNC
     #define rand RAND_FUNC_TAOS_FORBID
     #define srand SRAND_FUNC_TAOS_FORBID

include/os/osSemaphore.h

@@ -20,7 +20,6 @@
 extern "C" {
 #endif
 
-#include <pthread.h>
 #include <semaphore.h>
 
 #if defined (_TD_DARWIN_64)
@@ -38,25 +37,25 @@ extern "C" {
 #endif
 
 #if defined (_TD_DARWIN_64)
-//  #define pthread_rwlock_t pthread_mutex_t
-//  #define pthread_rwlock_init(lock, NULL) pthread_mutex_init(lock, NULL)
-//  #define pthread_rwlock_destroy(lock) pthread_mutex_destroy(lock)
-// #define pthread_rwlock_wrlock(lock) pthread_mutex_lock(lock)
-//  #define pthread_rwlock_rdlock(lock) pthread_mutex_lock(lock)
-//  #define pthread_rwlock_unlock(lock) pthread_mutex_unlock(lock)
+//  #define TdThreadRwlock TdThreadMutex
+//  #define taosThreadRwlockInit(lock, NULL) taosThreadMutexInit(lock, NULL)
+//  #define taosThreadRwlockDestroy(lock) taosThreadMutexDestroy(lock)
+// #define taosThreadRwlockWrlock(lock) taosThreadMutexLock(lock)
+//  #define taosThreadRwlockRdlock(lock) taosThreadMutexLock(lock)
+//  #define taosThreadRwlockUnlock(lock) taosThreadMutexUnlock(lock)
 
-  #define pthread_spinlock_t pthread_mutex_t
-  #define pthread_spin_init(lock, NULL) pthread_mutex_init(lock, NULL)
-  #define pthread_spin_destroy(lock) pthread_mutex_destroy(lock)
-  #define pthread_spin_lock(lock) pthread_mutex_lock(lock)
-  #define pthread_spin_unlock(lock) pthread_mutex_unlock(lock)
+  #define TdThreadSpinlock TdThreadMutex
+  #define taosThreadSpinInit(lock, NULL) taosThreadMutexInit(lock, NULL)
+  #define taosThreadSpinDestroy(lock) taosThreadMutexDestroy(lock)
+  #define taosThreadSpinLock(lock) taosThreadMutexLock(lock)
+  #define taosThreadSpinUnlock(lock) taosThreadMutexUnlock(lock)
 #endif
 
-bool    taosCheckPthreadValid(pthread_t thread);
+bool    taosCheckPthreadValid(TdThread thread);
 int64_t taosGetSelfPthreadId();
-int64_t taosGetPthreadId(pthread_t thread);
-void    taosResetPthread(pthread_t* thread);
-bool    taosComparePthread(pthread_t first, pthread_t second);
+int64_t taosGetPthreadId(TdThread thread);
+void    taosResetPthread(TdThread* thread);
+bool    taosComparePthread(TdThread first, TdThread second);
 int32_t taosGetPId();
 int32_t taosGetAppName(char* name, int32_t* len);
 

include/os/osSleep.h

@@ -21,6 +21,7 @@ extern "C" {
 #endif
 
 // If the error is in a third-party library, place this header file under the third-party library header file.
+// When you want to use this feature, you should find or add the same function in the following section.
 #ifndef ALLOW_FORBID_FUNC
     #define Sleep SLEEP_FUNC_TAOS_FORBID
     #define sleep SLEEP_FUNC_TAOS_FORBID

include/os/osSocket.h

@@ -17,6 +17,7 @@
 #define _TD_OS_SOCKET_H_
 
 // If the error is in a third-party library, place this header file under the third-party library header file.
+// When you want to use this feature, you should find or add the same function in the following section.
 #ifndef ALLOW_FORBID_FUNC
     #define socket SOCKET_FUNC_TAOS_FORBID
     #define bind   BIND_FUNC_TAOS_FORBID
@@ -50,11 +51,29 @@
 extern "C" {
 #endif
 
-#if (defined(_TD_WINDOWS_64) || defined(_TD_WINDOWS_32)) 
+#if defined(WINDOWS)
   #define htobe64 htonll
-  #if defined(_TD_GO_DLL_)
-    uint64_t htonll(uint64_t val);
+#endif
+
+#if defined(WINDOWS)
+  #define TAOS_EPOLL_WAIT_TIME 100
+  typedef SOCKET eventfd_t;
+  #define eventfd(a, b) -1
+  typedef SOCKET EpollFd;
+  #define EpollClose(pollFd) epoll_close(pollFd)
+  #ifndef EPOLLWAKEUP
+    #define EPOLLWAKEUP (1u << 29)
   #endif
+#elif defined(_TD_DARWIN_64)
+  #define TAOS_EPOLL_WAIT_TIME 500
+  typedef int32_t SOCKET;
+  typedef SOCKET EpollFd;
+  #define EpollClose(pollFd) epoll_close(pollFd)
+#else
+  #define TAOS_EPOLL_WAIT_TIME 500
+  typedef int32_t SOCKET;
+  typedef SOCKET EpollFd;
+  #define EpollClose(pollFd) taosCloseSocket(pollFd)
 #endif
 
 #if defined(_TD_DARWIN_64)
@@ -66,12 +85,12 @@ extern "C" {
 #	define htole16(x) OSSwapHostToLittleInt16(x)
 #	define be16toh(x) OSSwapBigToHostInt16(x)
 #	define le16toh(x) OSSwapLittleToHostInt16(x)
- 
+
 #	define htobe32(x) OSSwapHostToBigInt32(x)
 #	define htole32(x) OSSwapHostToLittleInt32(x)
 #	define be32toh(x) OSSwapBigToHostInt32(x)
 #	define le32toh(x) OSSwapLittleToHostInt32(x)
- 
+
 #	define htobe64(x) OSSwapHostToBigInt64(x)
 #	define htole64(x) OSSwapHostToLittleInt64(x)
 #	define be64toh(x) OSSwapBigToHostInt64(x)
@@ -85,6 +104,17 @@ extern "C" {
 
 #define TAOS_EPOLL_WAIT_TIME 500
 
+typedef int32_t SocketFd;
+typedef SocketFd  EpollFd;
+
+typedef struct TdSocket {
+#if SOCKET_WITH_LOCK
+  TdThreadRwlock rwlock;
+#endif
+  int      refId;
+  SocketFd fd;
+} *TdSocketPtr, TdSocket;
+
 typedef struct TdSocketServer *TdSocketServerPtr;
 typedef struct TdSocket *TdSocketPtr;
 typedef struct TdEpoll *TdEpollPtr;
@@ -93,6 +123,7 @@ int32_t taosSendto(TdSocketPtr pSocket, void * msg, int len, unsigned int flags,
 int32_t taosWriteSocket(TdSocketPtr pSocket, void *msg, int len);
 int32_t taosReadSocket(TdSocketPtr pSocket, void *msg, int len);
 int32_t taosReadFromSocket(TdSocketPtr pSocket, void *buf, int32_t len, int32_t flags, struct sockaddr *destAddr, socklen_t *addrLen);
+int32_t taosCloseSocketNoCheck1(SocketFd fd);
 int32_t taosCloseSocket(TdSocketPtr *ppSocket);
 int32_t taosCloseSocketServer(TdSocketServerPtr *ppSocketServer);
 int32_t taosShutDownSocketRD(TdSocketPtr pSocket);

include/os/osString.h

@@ -24,6 +24,7 @@ typedef wchar_t TdWchar;
 typedef int32_t TdUcs4;
 
 // If the error is in a third-party library, place this header file under the third-party library header file.
+// When you want to use this feature, you should find or add the same function in the following section.
 #ifndef ALLOW_FORBID_FUNC
     #define iconv_open ICONV_OPEN_FUNC_TAOS_FORBID
     #define iconv_close ICONV_CLOSE_FUNC_TAOS_FORBID
@@ -35,7 +36,7 @@ typedef int32_t TdUcs4;
     #define wctomb WCTOMB_FUNC_TAOS_FORBID
     #define wcstombs WCSTOMBS_FUNC_TAOS_FORBID
     #define wcsncpy WCSNCPY_FUNC_TAOS_FORBID
-    #define wchar_t WCHAR_T_FUNC_TAOS_FORBID
+    #define wchar_t WCHAR_T_TYPE_TAOS_FORBID
 #endif
 
 #if defined(_TD_WINDOWS_64) || defined(_TD_WINDOWS_32)

include/os/osSystem.h

@@ -21,6 +21,7 @@ extern "C" {
 #endif
 
 // If the error is in a third-party library, place this header file under the third-party library header file.
+// When you want to use this feature, you should find or add the same function in the following section.
 #ifndef ALLOW_FORBID_FUNC
     #define popen POPEN_FUNC_TAOS_FORBID
     #define pclose PCLOSE_FUNC_TAOS_FORBID
@@ -28,7 +29,6 @@ extern "C" {
     #define tcgetattr TCGETATTR_FUNC_TAOS_FORBID
 #endif
 
-int32_t taosSystem(const char *cmd, char *buf, int32_t bufSize);
 void* taosLoadDll(const char* filename);
 void* taosLoadSym(void* handle, char* name);
 void  taosCloseDll(void* handle);

include/os/osThread.h

@@ -22,6 +22,95 @@
 extern "C" {
 #endif
 
+typedef pthread_t TdThread;
+typedef pthread_spinlock_t TdThreadSpinlock;
+typedef pthread_mutex_t TdThreadMutex;
+typedef pthread_mutexattr_t TdThreadMutexAttr;
+typedef pthread_rwlock_t TdThreadRwlock;
+typedef pthread_attr_t TdThreadAttr;
+typedef pthread_once_t TdThreadOnce;
+typedef pthread_rwlockattr_t TdThreadRwlockAttr;
+typedef pthread_cond_t TdThreadCond;
+typedef pthread_condattr_t TdThreadCondAttr;
+
+#define taosThreadCleanupPush pthread_cleanup_push
+#define taosThreadCleanupPop pthread_cleanup_pop
+
+// If the error is in a third-party library, place this header file under the third-party library header file.
+// When you want to use this feature, you should find or add the same function in the following section.
+#ifndef ALLOW_FORBID_FUNC
+    #define pthread_t PTHREAD_T_TYPE_TAOS_FORBID
+    #define pthread_spinlock_t PTHREAD_SPINLOCK_T_TYPE_TAOS_FORBID
+    #define pthread_mutex_t PTHREAD_MUTEX_T_TYPE_TAOS_FORBID
+    #define pthread_mutexattr_t PTHREAD_MUTEXATTR_T_TYPE_TAOS_FORBID
+    #define pthread_rwlock_t PTHREAD_RWLOCK_T_TYPE_TAOS_FORBID
+    #define pthread_attr_t PTHREAD_ATTR_T_TYPE_TAOS_FORBID
+    #define pthread_once_t PTHREAD_ONCE_T_TYPE_TAOS_FORBID
+    #define pthread_rwlockattr_t PTHREAD_RWLOCKATTR_T_TYPE_TAOS_FORBID
+    #define pthread_cond_t PTHREAD_COND_T_TYPE_TAOS_FORBID
+    #define pthread_condattr_t PTHREAD_CONDATTR_T_TYPE_TAOS_FORBID
+    #define pthread_spin_init PTHREAD_SPIN_INIT_FUNC_TAOS_FORBID
+    #define pthread_mutex_init PTHREAD_MUTEX_INIT_FUNC_TAOS_FORBID
+    #define pthread_spin_destroy PTHREAD_SPIN_DESTROY_FUNC_TAOS_FORBID
+    #define pthread_mutex_destroy PTHREAD_MUTEX_DESTROY_FUNC_TAOS_FORBID
+    #define pthread_spin_lock PTHREAD_SPIN_LOCK_FUNC_TAOS_FORBID
+    #define pthread_mutex_lock PTHREAD_MUTEX_LOCK_FUNC_TAOS_FORBID
+    #define pthread_spin_unlock PTHREAD_SPIN_UNLOCK_FUNC_TAOS_FORBID
+    #define pthread_mutex_unlock PTHREAD_MUTEX_UNLOCK_FUNC_TAOS_FORBID
+    #define pthread_rwlock_rdlock PTHREAD_RWLOCK_RDLOCK_FUNC_TAOS_FORBID
+    #define pthread_rwlock_wrlock PTHREAD_RWLOCK_WRLOCK_FUNC_TAOS_FORBID
+    #define pthread_rwlock_unlock PTHREAD_RWLOCK_UNLOCK_FUNC_TAOS_FORBID
+    #define pthread_testcancel PTHREAD_TESTCANCEL_FUNC_TAOS_FORBID
+    #define pthread_attr_init PTHREAD_ATTR_INIT_FUNC_TAOS_FORBID
+    #define pthread_create PTHREAD_CREATE_FUNC_TAOS_FORBID
+    #define pthread_once PTHREAD_ONCE_FUNC_TAOS_FORBID
+    #define pthread_attr_setdetachstate PTHREAD_ATTR_SETDETACHSTATE_FUNC_TAOS_FORBID
+    #define pthread_attr_destroy PTHREAD_ATTR_DESTROY_FUNC_TAOS_FORBID
+    #define pthread_join PTHREAD_JOIN_FUNC_TAOS_FORBID
+    #define pthread_rwlock_init PTHREAD_RWLOCK_INIT_FUNC_TAOS_FORBID
+    #define pthread_rwlock_destroy PTHREAD_RWLOCK_DESTROY_FUNC_TAOS_FORBID
+    #define pthread_cond_signal PTHREAD_COND_SIGNAL_FUNC_TAOS_FORBID
+    #define pthread_cond_init PTHREAD_COND_INIT_FUNC_TAOS_FORBID
+    #define pthread_cond_broadcast PTHREAD_COND_BROADCAST_FUNC_TAOS_FORBID
+    #define pthread_cond_destroy PTHREAD_COND_DESTROY_FUNC_TAOS_FORBID
+    #define pthread_cond_wait PTHREAD_COND_WAIT_FUNC_TAOS_FORBID
+    #define pthread_self PTHREAD_SELF_FUNC_TAOS_FORBID
+    #define pthread_equal PTHREAD_EQUAL_FUNC_TAOS_FORBID
+    #define pthread_sigmask PTHREAD_SIGMASK_FUNC_TAOS_FORBID
+    #define pthread_cancel PTHREAD_CANCEL_FUNC_TAOS_FORBID
+    #define pthread_kill PTHREAD_KILL_FUNC_TAOS_FORBID
+#endif
+
+int32_t taosThreadSpinInit(TdThreadSpinlock *lock, int pshared);
+int32_t taosThreadMutexInit(TdThreadMutex *mutex, const TdThreadMutexAttr *attr);
+int32_t taosThreadSpinDestroy(TdThreadSpinlock *lock);
+int32_t taosThreadMutexDestroy(TdThreadMutex * mutex);
+int32_t taosThreadSpinLock(TdThreadSpinlock *lock);
+int32_t taosThreadMutexLock(TdThreadMutex *mutex);
+int32_t taosThreadRwlockRdlock(TdThreadRwlock *rwlock);
+int32_t taosThreadSpinUnlock(TdThreadSpinlock *lock);
+int32_t taosThreadMutexUnlock(TdThreadMutex *mutex);
+int32_t taosThreadRwlockWrlock(TdThreadRwlock *rwlock);
+int32_t taosThreadRwlockUnlock(TdThreadRwlock *rwlock);
+void taosThreadTestCancel(void);
+int32_t taosThreadAttrInit(TdThreadAttr *attr);
+int32_t taosThreadCreate(TdThread *tid, const TdThreadAttr *attr, void*(*start)(void*), void *arg);
+int32_t taosThreadOnce(TdThreadOnce *onceControl, void(*initRoutine)(void));
+int32_t taosThreadAttrSetDetachState(TdThreadAttr *attr, int32_t detachState);
+int32_t taosThreadAttrDestroy(TdThreadAttr *attr);
+int32_t taosThreadJoin(TdThread thread, void **pValue);
+int32_t taosThreadRwlockInit(TdThreadRwlock *rwlock, const TdThreadRwlockAttr *attr);
+int32_t taosThreadRwlockDestroy(TdThreadRwlock *rwlock);
+int32_t taosThreadCondSignal(TdThreadCond *cond);
+int32_t taosThreadCondInit(TdThreadCond *cond, const TdThreadCondAttr *attr);
+int32_t taosThreadCondBroadcast(TdThreadCond *cond);
+int32_t taosThreadCondDestroy(TdThreadCond *cond);
+int32_t taosThreadCondWait(TdThreadCond *cond, TdThreadMutex *mutex);
+TdThread taosThreadSelf(void);
+int32_t taosThreadEqual(TdThread t1, TdThread t2);
+int32_t taosThreadSigmask(int how, sigset_t const *set, sigset_t *oset);
+int32_t taosThreadCancel(TdThread thread);
+int32_t taosThreadKill(TdThread thread, int sig);
 #ifdef __cplusplus
 }
 #endif

include/os/osTime.h

@@ -21,6 +21,7 @@ extern "C" {
 #endif
 
 // If the error is in a third-party library, place this header file under the third-party library header file.
+// When you want to use this feature, you should find or add the same function in the following section.
 #ifndef ALLOW_FORBID_FUNC
     #define strptime STRPTIME_FUNC_TAOS_FORBID
     #define gettimeofday GETTIMEOFDAY_FUNC_TAOS_FORBID
@@ -33,11 +34,7 @@ extern "C" {
 
   #define CLOCK_REALTIME 	0
 
-  #ifdef _TD_GO_DLL_
-    #define MILLISECOND_PER_SECOND (1000LL)
-  #else
-    #define MILLISECOND_PER_SECOND (1000i64)
-  #endif
+  #define MILLISECOND_PER_SECOND (1000i64)
 #else
   #define MILLISECOND_PER_SECOND ((int64_t)1000L)
 #endif

include/os/osTimer.h

@@ -21,6 +21,7 @@ extern "C" {
 #endif
 
 // If the error is in a third-party library, place this header file under the third-party library header file.
+// When you want to use this feature, you should find or add the same function in the following section.
 #ifndef ALLOW_FORBID_FUNC
     #define timer_create TIMER_CREATE_FUNC_TAOS_FORBID
     #define timer_settime TIMER_SETTIME_FUNC_TAOS_FORBID

include/os/osTimezone.h

@@ -20,6 +20,12 @@
 extern "C" {
 #endif
 
+// If the error is in a third-party library, place this header file under the third-party library header file.
+// When you want to use this feature, you should find or add the same function in the following section.
+#ifndef ALLOW_FORBID_FUNC
+    #define tzset TZSET_FUNC_TAOS_FORBID
+#endif
+
 void taosGetSystemTimezone(char *outTimezone);
 void taosSetSystemTimezone(const char *inTimezone, char *outTimezone, int8_t *outDaylight);
 

include/util/taoserror.h

@@ -75,6 +75,7 @@ int32_t* taosGetErrno();
 #define TSDB_CODE_REPEAT_INIT                   TAOS_DEF_ERROR_CODE(0, 0x010B)
 #define TSDB_CODE_CFG_NOT_FOUND                 TAOS_DEF_ERROR_CODE(0, 0x010C)
 #define TSDB_CODE_INVALID_CFG                   TAOS_DEF_ERROR_CODE(0, 0x010D)
+#define TSDB_CODE_OUT_OF_SHM_MEM                TAOS_DEF_ERROR_CODE(0, 0x010E)
 #define TSDB_CODE_REF_NO_MEMORY                 TAOS_DEF_ERROR_CODE(0, 0x0110)
 #define TSDB_CODE_REF_FULL                      TAOS_DEF_ERROR_CODE(0, 0x0111)
 #define TSDB_CODE_REF_ID_REMOVED                TAOS_DEF_ERROR_CODE(0, 0x0112)
@@ -273,38 +274,23 @@ int32_t* taosGetErrno();
 #define TSDB_CODE_MND_STREAM_NOT_EXIST          TAOS_DEF_ERROR_CODE(0, 0x03F1)
 #define TSDB_CODE_MND_INVALID_STREAM_OPTION     TAOS_DEF_ERROR_CODE(0, 0x03F2)
 
+// mnode-sma
+#define TSDB_CODE_MND_SMA_ALREADY_EXIST         TAOS_DEF_ERROR_CODE(0, 0x0400)
+#define TSDB_CODE_MND_SMA_NOT_EXIST             TAOS_DEF_ERROR_CODE(0, 0x0401)
+#define TSDB_CODE_MND_INVALID_SMA_OPTION        TAOS_DEF_ERROR_CODE(0, 0x0402)
+
 // dnode
-#define TSDB_CODE_DND_ACTION_IN_PROGRESS        TAOS_DEF_ERROR_CODE(0, 0x0400)
-#define TSDB_CODE_DND_OFFLINE                   TAOS_DEF_ERROR_CODE(0, 0x0401)
-#define TSDB_CODE_DND_INVALID_MSG_LEN           TAOS_DEF_ERROR_CODE(0, 0x0402)
-#define TSDB_CODE_DND_DNODE_READ_FILE_ERROR     TAOS_DEF_ERROR_CODE(0, 0x0410)
-#define TSDB_CODE_DND_DNODE_WRITE_FILE_ERROR    TAOS_DEF_ERROR_CODE(0, 0x0411)
-#define TSDB_CODE_DND_MNODE_ALREADY_DEPLOYED    TAOS_DEF_ERROR_CODE(0, 0x0420)
-#define TSDB_CODE_DND_MNODE_NOT_DEPLOYED        TAOS_DEF_ERROR_CODE(0, 0x0421)
-#define TSDB_CODE_DND_MNODE_INVALID_OPTION      TAOS_DEF_ERROR_CODE(0, 0x0422)
-#define TSDB_CODE_DND_MNODE_READ_FILE_ERROR     TAOS_DEF_ERROR_CODE(0, 0x0423)
-#define TSDB_CODE_DND_MNODE_WRITE_FILE_ERROR    TAOS_DEF_ERROR_CODE(0, 0x0424)
-#define TSDB_CODE_DND_QNODE_ALREADY_DEPLOYED    TAOS_DEF_ERROR_CODE(0, 0x0430)
-#define TSDB_CODE_DND_QNODE_NOT_DEPLOYED        TAOS_DEF_ERROR_CODE(0, 0x0431)
-#define TSDB_CODE_DND_QNODE_INVALID_OPTION      TAOS_DEF_ERROR_CODE(0, 0x0432)
-#define TSDB_CODE_DND_QNODE_READ_FILE_ERROR     TAOS_DEF_ERROR_CODE(0, 0x0433)
-#define TSDB_CODE_DND_QNODE_WRITE_FILE_ERROR    TAOS_DEF_ERROR_CODE(0, 0x0434)
-#define TSDB_CODE_DND_SNODE_ALREADY_DEPLOYED    TAOS_DEF_ERROR_CODE(0, 0x0440)
-#define TSDB_CODE_DND_SNODE_NOT_DEPLOYED        TAOS_DEF_ERROR_CODE(0, 0x0441)
-#define TSDB_CODE_DND_SNODE_INVALID_OPTION      TAOS_DEF_ERROR_CODE(0, 0x0442)
-#define TSDB_CODE_DND_SNODE_READ_FILE_ERROR     TAOS_DEF_ERROR_CODE(0, 0x0443)
-#define TSDB_CODE_DND_SNODE_WRITE_FILE_ERROR    TAOS_DEF_ERROR_CODE(0, 0x0444)
-#define TSDB_CODE_DND_BNODE_ALREADY_DEPLOYED    TAOS_DEF_ERROR_CODE(0, 0x0450)
-#define TSDB_CODE_DND_BNODE_NOT_DEPLOYED        TAOS_DEF_ERROR_CODE(0, 0x0451)
-#define TSDB_CODE_DND_BNODE_INVALID_OPTION      TAOS_DEF_ERROR_CODE(0, 0x0452)
-#define TSDB_CODE_DND_BNODE_READ_FILE_ERROR     TAOS_DEF_ERROR_CODE(0, 0x0453)
-#define TSDB_CODE_DND_BNODE_WRITE_FILE_ERROR    TAOS_DEF_ERROR_CODE(0, 0x0454)
-#define TSDB_CODE_DND_VNODE_ALREADY_DEPLOYED    TAOS_DEF_ERROR_CODE(0, 0x0460)
-#define TSDB_CODE_DND_VNODE_NOT_DEPLOYED        TAOS_DEF_ERROR_CODE(0, 0x0461)
-#define TSDB_CODE_DND_VNODE_INVALID_OPTION      TAOS_DEF_ERROR_CODE(0, 0x0462)
-#define TSDB_CODE_DND_VNODE_READ_FILE_ERROR     TAOS_DEF_ERROR_CODE(0, 0x0463)
-#define TSDB_CODE_DND_VNODE_WRITE_FILE_ERROR    TAOS_DEF_ERROR_CODE(0, 0x0464)
-#define TSDB_CODE_DND_VNODE_TOO_MANY_VNODES     TAOS_DEF_ERROR_CODE(0, 0x0465)
+#define TSDB_CODE_DND_ACTION_IN_PROGRESS        TAOS_DEF_ERROR_CODE(0, 0x04A0)
+#define TSDB_CODE_DND_OFFLINE                   TAOS_DEF_ERROR_CODE(0, 0x04A1)
+#define TSDB_CODE_DND_INVALID_MSG_LEN           TAOS_DEF_ERROR_CODE(0, 0x04A2)
+#define TSDB_CODE_NODE_ALREADY_DEPLOYED         TAOS_DEF_ERROR_CODE(0, 0x04A3)
+#define TSDB_CODE_NODE_NOT_DEPLOYED             TAOS_DEF_ERROR_CODE(0, 0x04A4)
+#define TSDB_CODE_NODE_PARSE_FILE_ERROR         TAOS_DEF_ERROR_CODE(0, 0x04A5)
+#define TSDB_CODE_NODE_INVALID_OPTION           TAOS_DEF_ERROR_CODE(0, 0x04A6)
+#define TSDB_CODE_DND_VNODE_ALREADY_DEPLOYED    TAOS_DEF_ERROR_CODE(0, 0x04A7)
+#define TSDB_CODE_DND_VNODE_NOT_DEPLOYED        TAOS_DEF_ERROR_CODE(0, 0x04A8)
+#define TSDB_CODE_DND_VNODE_INVALID_OPTION      TAOS_DEF_ERROR_CODE(0, 0x04A9)
+#define TSDB_CODE_DND_VNODE_TOO_MANY_VNODES     TAOS_DEF_ERROR_CODE(0, 0x04AA)
 
 // vnode
 #define TSDB_CODE_VND_ACTION_IN_PROGRESS        TAOS_DEF_ERROR_CODE(0, 0x0500)
@@ -328,6 +314,8 @@ int32_t* taosGetErrno();
 #define TSDB_CODE_VND_IS_SYNCING                TAOS_DEF_ERROR_CODE(0, 0x0513)
 #define TSDB_CODE_VND_INVALID_TSDB_STATE        TAOS_DEF_ERROR_CODE(0, 0x0514)
 #define TSDB_CODE_VND_TB_NOT_EXIST              TAOS_DEF_ERROR_CODE(0, 0x0515)
+#define TSDB_CODE_VND_SMA_NOT_EXIST             TAOS_DEF_ERROR_CODE(0, 0x0516)
+#define TSDB_CODE_VND_HASH_MISMATCH             TAOS_DEF_ERROR_CODE(0, 0x0517)
 
 // tsdb
 #define TSDB_CODE_TDB_INVALID_TABLE_ID          TAOS_DEF_ERROR_CODE(0, 0x0600)
@@ -353,8 +341,10 @@ int32_t* taosGetErrno();
 #define TSDB_CODE_TDB_MESSED_MSG                TAOS_DEF_ERROR_CODE(0, 0x0614)
 #define TSDB_CODE_TDB_IVLD_TAG_VAL              TAOS_DEF_ERROR_CODE(0, 0x0615)
 #define TSDB_CODE_TDB_NO_CACHE_LAST_ROW         TAOS_DEF_ERROR_CODE(0, 0x0616)
-#define TSDB_CODE_TDB_NO_SMA_INDEX_IN_META      TAOS_DEF_ERROR_CODE(0, 0x0617)
+#define TSDB_CODE_TDB_TABLE_RECREATED           TAOS_DEF_ERROR_CODE(0, 0x0617)
 #define TSDB_CODE_TDB_TDB_ENV_OPEN_ERROR        TAOS_DEF_ERROR_CODE(0, 0x0618)
+#define TSDB_CODE_TDB_NO_SMA_INDEX_IN_META      TAOS_DEF_ERROR_CODE(0, 0x0619)
+#define TSDB_CODE_TDB_INVALID_SMA_STAT          TAOS_DEF_ERROR_CODE(0, 0x0620)
 
 // query
 #define TSDB_CODE_QRY_INVALID_QHANDLE           TAOS_DEF_ERROR_CODE(0, 0x0700)
@@ -456,10 +446,12 @@ int32_t* taosGetErrno();
 #define TSDB_CODE_CTG_SYS_ERROR                 TAOS_DEF_ERROR_CODE(0, 0x2404)
 #define TSDB_CODE_CTG_DB_DROPPED                TAOS_DEF_ERROR_CODE(0, 0x2405)
 #define TSDB_CODE_CTG_OUT_OF_SERVICE            TAOS_DEF_ERROR_CODE(0, 0x2406)
+#define TSDB_CODE_CTG_VG_META_MISMATCH          TAOS_DEF_ERROR_CODE(0, 0x2407)
 
-//scheduler
+//scheduler&qworker
 #define TSDB_CODE_SCH_STATUS_ERROR              TAOS_DEF_ERROR_CODE(0, 0x2501)
 #define TSDB_CODE_SCH_INTERNAL_ERROR            TAOS_DEF_ERROR_CODE(0, 0x2502)
+#define TSDB_CODE_QW_MSG_ERROR                  TAOS_DEF_ERROR_CODE(0, 0x2503)
 
 //parser
 #define TSDB_CODE_PAR_SYNTAX_ERROR              TAOS_DEF_ERROR_CODE(0, 0x2600)
@@ -483,6 +475,7 @@ int32_t* taosGetErrno();
 #define TSDB_CODE_PAR_INVALID_PORT              TAOS_DEF_ERROR_CODE(0, 0x2612)
 #define TSDB_CODE_PAR_INVALID_ENDPOINT          TAOS_DEF_ERROR_CODE(0, 0x2613)
 #define TSDB_CODE_PAR_EXPRIE_STATEMENT          TAOS_DEF_ERROR_CODE(0, 0x2614)
+#define TSDB_CODE_PAR_INTERVAL_VALUE_TOO_SMALL  TAOS_DEF_ERROR_CODE(0, 0x2615)
 
 #ifdef __cplusplus
 }

include/util/tcoding.h

@@ -351,7 +351,7 @@ static FORCE_INLINE void *taosDecodeString(const void *buf, char **value) {
   uint64_t size = 0;
 
   buf = taosDecodeVariantU64(buf, &size);
-  *value = (char *)malloc((size_t)size + 1);
+  *value = (char *)taosMemoryMalloc((size_t)size + 1);
 
   if (*value == NULL) return NULL;
   memcpy(*value, buf, (size_t)size);
@@ -386,7 +386,7 @@ static FORCE_INLINE int32_t taosEncodeBinary(void **buf, const void *value, int3
 }
 
 static FORCE_INLINE void *taosDecodeBinary(const void *buf, void **value, int32_t valueLen) {
-  *value = malloc((size_t)valueLen);
+  *value = taosMemoryMalloc((size_t)valueLen);
   if (*value == NULL) return NULL;
   memcpy(*value, buf, (size_t)valueLen);
 

include/util/tdef.h

@@ -99,7 +99,7 @@ extern const int32_t TYPE_BYTES[15];
 #define TSDB_INS_TABLE_MNODES                 "mnodes"
 #define TSDB_INS_TABLE_MODULES                "modules"
 #define TSDB_INS_TABLE_QNODES                 "qnodes"
-#define TSDB_INS_TABLE_USER_DATABASE          "user_database"
+#define TSDB_INS_TABLE_USER_DATABASES         "user_databases"
 #define TSDB_INS_TABLE_USER_FUNCTIONS         "user_functions"
 #define TSDB_INS_TABLE_USER_INDEXES           "user_indexes"
 #define TSDB_INS_TABLE_USER_STABLES           "user_stables"
@@ -109,6 +109,8 @@ extern const int32_t TYPE_BYTES[15];
 #define TSDB_INS_TABLE_USER_USERS             "user_users"
 #define TSDB_INS_TABLE_VGROUPS                "vgroups"
 
+#define TSDB_INS_USER_STABLES_DBNAME_COLID    2
+
 #define TSDB_TICK_PER_SECOND(precision)                      \
   ((int64_t)((precision) == TSDB_TIME_PRECISION_MILLI ? 1e3L \
                                                       : ((precision) == TSDB_TIME_PRECISION_MICRO ? 1e6L : 1e9L)))
@@ -453,6 +455,10 @@ enum {
   SND_WORKER_TYPE__UNIQUE,
 };
 
+#define MND_VGID -1
+#define QND_VGID 1
+#define VND_VGID 0
+
 #ifdef __cplusplus
 }
 #endif

include/util/tencode.h

@@ -406,7 +406,7 @@ static FORCE_INLINE int32_t tDecodeBinaryAlloc(SCoder* pDecoder, void** val, uin
   if (tDecodeU64v(pDecoder, len) < 0) return -1;
 
   if (TD_CODER_CHECK_CAPACITY_FAILED(pDecoder, *len)) return -1;
-  *val = malloc(*len);
+  *val = taosMemoryMalloc(*len);
   if (*val == NULL) return -1;
   memcpy(*val, TD_CODER_CURRENT(pDecoder), *len);
 

include/util/tfreelist.h

@@ -31,7 +31,7 @@ typedef TD_SLIST(SFreeListNode) SFreeList;
 
 #define TFL_MALLOC(PTR, TYPE, SIZE, LIST)                      \
   do {                                                         \
-    void *ptr = malloc((SIZE) + sizeof(struct SFreeListNode)); \
+    void *ptr = taosMemoryMalloc((SIZE) + sizeof(struct SFreeListNode)); \
     if (ptr) {                                                 \
       TD_SLIST_PUSH((LIST), (struct SFreeListNode *)ptr);      \
       ptr = ((struct SFreeListNode *)ptr)->payload;            \
@@ -49,7 +49,7 @@ static FORCE_INLINE void tFreeListClear(SFreeList *pFL) {
     pNode = TD_SLIST_HEAD(pFL);
     if (pNode == NULL) break;
     TD_SLIST_POP(pFL);
-    free(pNode);
+    taosMemoryFree(pNode);
   }
 }
 

include/util/thash.h

@@ -86,7 +86,7 @@ int32_t taosHashGetSize(const SHashObj *pHashObj);
  * @param size
  * @return
  */
-int32_t taosHashPut(SHashObj *pHashObj, const void *key, size_t keyLen, void *data, size_t size);
+int32_t taosHashPut(SHashObj *pHashObj, const void *key, size_t keyLen, const void *data, size_t size);
 
 /**
  * return the payload data with the specified key

include/util/tidpool.h

@@ -27,7 +27,7 @@ typedef struct {
   int32_t         numOfFree;
   int32_t         freeSlot;
   bool           *freeList;
-  pthread_mutex_t mutex;
+  TdThreadMutex mutex;
 } id_pool_t;
 
 void   *taosInitIdPool(int32_t maxId);

include/util/tjson.h

@@ -22,6 +22,14 @@
 extern "C" {
 #endif
 
+#define tjsonGetNumberValue(pJson, pName, val) \
+  ({ \
+    uint64_t _tmp = 0; \
+    int32_t _code = tjsonGetUBigIntValue(pJson, pName, &_tmp); \
+    val = _tmp; \
+    _code; \
+  })
+
 typedef void SJson;
 
 SJson* tjsonCreateObject();

include/util/tlist.h

@@ -216,7 +216,7 @@ typedef struct {
 #define listNEles(l)    TD_DLIST_NELES(l)
 #define listEleSize(l)  ((l)->eleSize)
 #define isListEmpty(l)  (TD_DLIST_NELES(l) == 0)
-#define listNodeFree(n) free(n)
+#define listNodeFree(n) taosMemoryFree(n)
 
 void       tdListInit(SList *list, int32_t eleSize);
 void       tdListEmpty(SList *list);

source/dnode/mgmt/impl/inc/dndMgmt.h → include/util/tprocess.h

@@ -13,30 +13,50 @@
  * along with this program. If not, see <http://www.gnu.org/licenses/>.
  */
 
-#ifndef _TD_DND_DNODE_H_
-#define _TD_DND_DNODE_H_
+#ifndef _TD_UTIL_PROCESS_H_
+#define _TD_UTIL_PROCESS_H_
+
+#include "os.h"
 
 #ifdef __cplusplus
 extern "C" {
 #endif
-#include "dndEnv.h"
 
-int32_t dndInitMgmt(SDnode *pDnode);
-void    dndStopMgmt(SDnode *pDnode);
-void    dndCleanupMgmt(SDnode *pDnode);
+typedef struct SProcQueue SProcQueue;
+typedef struct SProcObj   SProcObj;
+typedef void *(*ProcMallocFp)(int32_t contLen);
+typedef void *(*ProcFreeFp)(void *pCont);
+typedef void *(*ProcConsumeFp)(void *pParent, void *pHead, int32_t headLen, void *pBody, int32_t bodyLen);
+
+typedef struct {
+  int32_t       childQueueSize;
+  ProcConsumeFp childConsumeFp;
+  ProcMallocFp  childMallocHeadFp;
+  ProcFreeFp    childFreeHeadFp;
+  ProcMallocFp  childMallocBodyFp;
+  ProcFreeFp    childFreeBodyFp;
+  int32_t       parentQueueSize;
+  ProcConsumeFp parentConsumeFp;
+  ProcMallocFp  parentdMallocHeadFp;
+  ProcFreeFp    parentFreeHeadFp;
+  ProcMallocFp  parentMallocBodyFp;
+  ProcFreeFp    parentFreeBodyFp;
+  bool          testFlag;
+  void         *pParent;
+  const char   *name;
+} SProcCfg;
 
-int32_t dndGetDnodeId(SDnode *pDnode);
-int64_t dndGetClusterId(SDnode *pDnode);
-void    dndGetDnodeEp(SDnode *pDnode, int32_t dnodeId, char *pEp, char *pFqdn, uint16_t *pPort);
-void    dndGetMnodeEpSet(SDnode *pDnode, SEpSet *pEpSet);
+SProcObj *taosProcInit(const SProcCfg *pCfg);
+void      taosProcCleanup(SProcObj *pProc);
+int32_t   taosProcRun(SProcObj *pProc);
+void      taosProcStop(SProcObj *pProc);
+bool      taosProcIsChild(SProcObj *pProc);
 
-void dndSendRedirectRsp(SDnode *pDnode, SRpcMsg *pMsg);
-void dndSendStatusReq(SDnode *pDnode);
-void dndProcessMgmtMsg(SDnode *pDnode, SRpcMsg *pRpcMsg, SEpSet *pEpSet);
-void dndProcessStartupReq(SDnode *pDnode, SRpcMsg *pMsg);
+int32_t taosProcPutToChildQueue(SProcObj *pProc, void *pHead, int32_t headLen, void *pBody, int32_t bodyLen);
+int32_t taosProcPutToParentQueue(SProcObj *pProc, void *pHead, int32_t headLen, void *pBody, int32_t bodyLen);
 
 #ifdef __cplusplus
 }
 #endif
 
-#endif /*_TD_DND_DNODE_H_*/
+#endif /*_TD_UTIL_PROCESS_H_*/
\ No newline at end of file

include/util/tqueue.h

@@ -42,8 +42,14 @@ shall be used to set up the protection.
 typedef struct STaosQueue STaosQueue;
 typedef struct STaosQset  STaosQset;
 typedef struct STaosQall  STaosQall;
-typedef void (*FItem)(void *ahandle, void *pItem);
-typedef void (*FItems)(void *ahandle, STaosQall *qall, int32_t numOfItems);
+typedef struct {
+  void   *ahandle;
+  int32_t workerId;
+  int32_t threadNum;
+} SQueueInfo;
+
+typedef void (*FItem)(SQueueInfo *pInfo, void *pItem);
+typedef void (*FItems)(SQueueInfo *pInfo, STaosQall *qall, int32_t numOfItems);
 
 STaosQueue *taosOpenQueue();
 void        taosCloseQueue(STaosQueue *queue);
@@ -70,10 +76,7 @@ int32_t    taosGetQueueNumber(STaosQset *qset);
 
 int32_t taosReadQitemFromQset(STaosQset *qset, void **ppItem, void **ahandle, FItem *itemFp);
 int32_t taosReadAllQitemsFromQset(STaosQset *qset, STaosQall *qall, void **ahandle, FItems *itemsFp);
-
-int32_t taosReadQitemFromQsetByThread(STaosQset *qset, void **ppItem, void **ahandle, FItem *itemFp, int32_t threadId);
 void    taosResetQsetThread(STaosQset *qset, void *pItem);
-
 int32_t taosGetQueueItemsNumber(STaosQueue *queue);
 int32_t taosGetQsetItemsNumber(STaosQset *qset);
 

include/util/tskiplist.h

@@ -57,7 +57,7 @@ typedef struct SSkipListNode {
  * @date   2017/11/12
  * the simple version of skip list.
  *
- * for multi-thread safe purpose, we employ pthread_rwlock_t to guarantee to generate
+ * for multi-thread safe purpose, we employ TdThreadRwlock to guarantee to generate
  * deterministic result. Later, we will remove the lock in SkipList to further enhance the performance.
  * In this case, one should use the concurrent skip list (by using michael-scott algorithm) instead of
  * this simple version in a multi-thread environment, to achieve higher performance of read/write operations.
@@ -106,7 +106,7 @@ typedef struct SSkipList {
   uint32_t          seed;
   __compar_fn_t     comparFn;
   __sl_key_fn_t     keyFn;
-  pthread_rwlock_t *lock;
+  TdThreadRwlock *lock;
   uint16_t          len;
   uint8_t           maxLevel;
   uint8_t           flags;

include/util/tthread.h

@@ -22,9 +22,11 @@
 extern "C" {
 #endif
 
-pthread_t* taosCreateThread(void* (*__start_routine)(void*), void* param);
-bool       taosDestoryThread(pthread_t* pthread);
-bool       taosThreadRunning(pthread_t* pthread);
+TdThread* taosCreateThread(void* (*__start_routine)(void*), void* param);
+bool       taosDestoryThread(TdThread* pthread);
+bool       taosThreadRunning(TdThread* pthread);
+
+typedef void *(*ThreadFp)(void *param);
 
 #ifdef __cplusplus
 }

include/util/tworker.h

@@ -27,34 +27,35 @@ typedef struct SWWorkerPool SWWorkerPool;
 
 typedef struct SQWorker {
   int32_t       id;      // worker ID
-  pthread_t     thread;  // thread
+  TdThread      thread;  // thread
   SQWorkerPool *pool;
-} SQWorker, SFWorker;
+} SQWorker;
 
 typedef struct SQWorkerPool {
-  int32_t         max;  // max number of workers
-  int32_t         min;  // min number of workers
-  int32_t         num;  // current number of workers
-  STaosQset      *qset;
-  const char     *name;
-  SQWorker       *workers;
-  pthread_mutex_t mutex;
-} SQWorkerPool, SFWorkerPool;
+  int32_t       max;  // max number of workers
+  int32_t       min;  // min number of workers
+  int32_t       num;  // current number of workers
+  STaosQset    *qset;
+  const char   *name;
+  SQWorker     *workers;
+  TdThreadMutex mutex;
+} SQWorkerPool;
 
 typedef struct SWWorker {
   int32_t       id;      // worker id
-  pthread_t     thread;  // thread
+  TdThread      thread;  // thread
   STaosQall    *qall;
   STaosQset    *qset;  // queue set
   SWWorkerPool *pool;
 } SWWorker;
 
 typedef struct SWWorkerPool {
-  int32_t         max;     // max number of workers
-  int32_t         nextId;  // from 0 to max-1, cyclic
-  const char     *name;
-  SWWorker       *workers;
-  pthread_mutex_t mutex;
+  int32_t       max;  // max number of workers
+  int32_t       num;
+  int32_t       nextId;  // from 0 to max-1, cyclic
+  const char   *name;
+  SWWorker     *workers;
+  TdThreadMutex mutex;
 } SWWorkerPool;
 
 int32_t     tQWorkerInit(SQWorkerPool *pool);
@@ -62,16 +63,43 @@ void        tQWorkerCleanup(SQWorkerPool *pool);
 STaosQueue *tQWorkerAllocQueue(SQWorkerPool *pool, void *ahandle, FItem fp);
 void        tQWorkerFreeQueue(SQWorkerPool *pool, STaosQueue *queue);
 
-int32_t     tFWorkerInit(SFWorkerPool *pool);
-void        tFWorkerCleanup(SFWorkerPool *pool);
-STaosQueue *tFWorkerAllocQueue(SFWorkerPool *pool, void *ahandle, FItem fp);
-void        tFWorkerFreeQueue(SFWorkerPool *pool, STaosQueue *queue);
-
 int32_t     tWWorkerInit(SWWorkerPool *pool);
 void        tWWorkerCleanup(SWWorkerPool *pool);
 STaosQueue *tWWorkerAllocQueue(SWWorkerPool *pool, void *ahandle, FItems fp);
 void        tWWorkerFreeQueue(SWWorkerPool *pool, STaosQueue *queue);
 
+typedef struct {
+  const char *name;
+  int32_t     minNum;
+  int32_t     maxNum;
+  FItem       fp;
+  void       *param;
+} SSingleWorkerCfg;
+
+typedef struct {
+  const char  *name;
+  STaosQueue  *queue;
+  SQWorkerPool pool;
+} SSingleWorker;
+
+typedef struct {
+  const char *name;
+  int32_t     maxNum;
+  FItems      fp;
+  void       *param;
+} SMultiWorkerCfg;
+
+typedef struct {
+  const char  *name;
+  STaosQueue  *queue;
+  SWWorkerPool pool;
+} SMultiWorker;
+
+int32_t tSingleWorkerInit(SSingleWorker *pWorker, const SSingleWorkerCfg *pCfg);
+void    tSingleWorkerCleanup(SSingleWorker *pWorker);
+int32_t tMultiWorkerInit(SMultiWorker *pWorker, const SMultiWorkerCfg *pCfg);
+void    tMultiWorkerCleanup(SMultiWorker *pWorker);
+
 #ifdef __cplusplus
 }
 #endif

include/util/types.h

@@ -84,6 +84,8 @@ typedef uint16_t VarDataLenT;  // maxVarDataLen: 32767
 #define varDataLen(v) ((VarDataLenT *)(v))[0]
 #define varDataVal(v) ((char *)(v) + VARSTR_HEADER_SIZE)
 
+#define NCHAR_WIDTH_TO_BYTES(n)  ((n) * TSDB_NCHAR_SIZE + VARSTR_HEADER_SIZE)
+
 typedef int32_t VarDataOffsetT;
 
 typedef struct tstr {

source/client/inc/clientInt.h

@@ -37,9 +37,8 @@ extern "C" {
 
 #define CHECK_CODE_GOTO(expr, label) \
   do {                               \
-    int32_t code = expr;             \
+    code = expr;                     \
     if (TSDB_CODE_SUCCESS != code) { \
-      terrno = code;                 \
       goto label;                    \
     }                                \
   } while (0)
@@ -77,8 +76,8 @@ typedef struct {
   int8_t inited;
   // ctl
   int8_t          threadStop;
-  pthread_t       thread;
-  pthread_mutex_t lock;       // used when app init and cleanup
+  TdThread       thread;
+  TdThreadMutex lock;       // used when app init and cleanup
   SArray*         appHbMgrs;  // SArray<SAppHbMgr*> one for each cluster
   FHbReqHandle    reqHandle[HEARTBEAT_TYPE_MAX];
   FHbRspHandle    rspHandle[HEARTBEAT_TYPE_MAX];
@@ -125,7 +124,7 @@ typedef struct SAppInfo {
   int32_t         pid;
   int32_t         numOfThreads;
   SHashObj*       pInstMap;
-  pthread_mutex_t mutex;
+  TdThreadMutex mutex;
 } SAppInfo;
 
 typedef struct STscObj {
@@ -137,23 +136,33 @@ typedef struct STscObj {
   uint32_t        connId;
   int32_t         connType;
   uint64_t        id;         // ref ID returned by taosAddRef
-  pthread_mutex_t mutex;      // used to protect the operation on db
+  TdThreadMutex mutex;      // used to protect the operation on db
   int32_t         numOfReqs;  // number of sqlObj bound to this connection
   SAppInstInfo*   pAppInfo;
 } STscObj;
 
+typedef struct SResultColumn {
+  union {
+    char*       nullbitmap;  // bitmap, one bit for each item in the list
+    int32_t*    offset;
+  };
+  char*         pData;
+} SResultColumn;
+
 typedef struct SReqResultInfo {
-  const char* pRspMsg;
-  const char* pData;
-  TAOS_FIELD* fields;
-  uint32_t    numOfCols;
-  int32_t*    length;
-  TAOS_ROW    row;
-  char**      pCol;
-  uint32_t    numOfRows;
-  uint64_t    totalRows;
-  uint32_t    current;
-  bool        completed;
+  const char*    pRspMsg;
+  const char*    pData;
+  TAOS_FIELD*    fields;
+  uint32_t       numOfCols;
+  int32_t*       length;
+  char**         convertBuf;
+  TAOS_ROW       row;
+  SResultColumn* pCol;
+  uint32_t       numOfRows;
+  uint64_t       totalRows;
+  uint32_t       current;
+  bool           completed;
+  int32_t        payloadLen;
 } SReqResultInfo;
 
 typedef struct SShowReqInfo {
@@ -186,6 +195,8 @@ typedef struct SRequestObj {
   char*                msgBuf;
   void*                pInfo;  // sql parse info, generated by parser module
   int32_t              code;
+  SArray*              dbList;
+  SArray*              tableList;
   SQueryExecMetric     metric;
   SRequestSendRecvBody body;
 } SRequestObj;
@@ -226,11 +237,11 @@ TAOS* taos_connect_internal(const char* ip, const char* user, const char* pass,
 
 void* doFetchRow(SRequestObj* pRequest);
 
-void setResultDataPtr(SReqResultInfo* pResultInfo, TAOS_FIELD* pFields, int32_t numOfCols, int32_t numOfRows);
+int32_t setResultDataPtr(SReqResultInfo* pResultInfo, TAOS_FIELD* pFields, int32_t numOfCols, int32_t numOfRows);
 
 int32_t buildRequest(STscObj* pTscObj, const char* sql, int sqlLen, SRequestObj** pRequest);
 
-int32_t parseSql(SRequestObj* pRequest, bool streamQuery, SQuery** pQuery);
+int32_t parseSql(SRequestObj* pRequest, bool topicQuery, SQuery** pQuery);
 int32_t getPlan(SRequestObj* pRequest, SQuery* pQuery, SQueryPlan** pPlan, SArray* pNodeList);
 
 // --- heartbeat

source/client/src/clientEnv.c

@@ -33,7 +33,7 @@ SAppInfo appInfo;
 int32_t  clientReqRefPool = -1;
 int32_t  clientConnRefPool = -1;
 
-static pthread_once_t tscinit = PTHREAD_ONCE_INIT;
+static TdThreadOnce tscinit = PTHREAD_ONCE_INIT;
 volatile int32_t      tscInitRes = 0;
 
 static void registerRequest(SRequestObj *pRequest) {
@@ -48,8 +48,8 @@ static void registerRequest(SRequestObj *pRequest) {
   if (pTscObj->pAppInfo) {
     SInstanceSummary *pSummary = &pTscObj->pAppInfo->summary;
 
-    int32_t total = atomic_add_fetch_32(&pSummary->totalRequests, 1);
-    int32_t currentInst = atomic_add_fetch_32(&pSummary->currentRequests, 1);
+    int32_t total = atomic_add_fetch_64(&pSummary->totalRequests, 1);
+    int32_t currentInst = atomic_add_fetch_64(&pSummary->currentRequests, 1);
     tscDebug("0x%" PRIx64 " new Request from connObj:0x%" PRIx64
              ", current:%d, app current:%d, total:%d, reqId:0x%" PRIx64,
              pRequest->self, pRequest->pTscObj->id, num, currentInst, total, pRequest->requestId);
@@ -62,7 +62,7 @@ static void deregisterRequest(SRequestObj *pRequest) {
   STscObj *         pTscObj = pRequest->pTscObj;
   SInstanceSummary *pActivity = &pTscObj->pAppInfo->summary;
 
-  int32_t currentInst = atomic_sub_fetch_32(&pActivity->currentRequests, 1);
+  int32_t currentInst = atomic_sub_fetch_64(&pActivity->currentRequests, 1);
   int32_t num = atomic_sub_fetch_32(&pTscObj->numOfReqs, 1);
 
   int64_t duration = taosGetTimestampMs() - pRequest->metric.start;
@@ -90,7 +90,6 @@ void *openTransporter(const char *user, const char *auth, int32_t numOfThread) {
   rpcInit.label = "TSC";
   rpcInit.numOfThreads = numOfThread;
   rpcInit.cfp = processMsgFromServer;
-  rpcInit.pfp = persistConnForSpecificMsg;
   rpcInit.sessions = tsMaxConnections;
   rpcInit.connType = TAOS_CONN_CLIENT;
   rpcInit.user = (char *)user;
@@ -115,12 +114,12 @@ void destroyTscObj(void *pObj) {
   hbDeregisterConn(pTscObj->pAppInfo->pAppHbMgr, connKey);
   atomic_sub_fetch_64(&pTscObj->pAppInfo->numOfConns, 1);
   tscDebug("connObj 0x%" PRIx64 " destroyed, totalConn:%" PRId64, pTscObj->id, pTscObj->pAppInfo->numOfConns);
-  pthread_mutex_destroy(&pTscObj->mutex);
-  tfree(pTscObj);
+  taosThreadMutexDestroy(&pTscObj->mutex);
+  taosMemoryFreeClear(pTscObj);
 }
 
 void *createTscObj(const char *user, const char *auth, const char *db, SAppInstInfo *pAppInfo) {
-  STscObj *pObj = (STscObj *)calloc(1, sizeof(STscObj));
+  STscObj *pObj = (STscObj *)taosMemoryCalloc(1, sizeof(STscObj));
   if (NULL == pObj) {
     terrno = TSDB_CODE_TSC_OUT_OF_MEMORY;
     return NULL;
@@ -134,7 +133,7 @@ void *createTscObj(const char *user, const char *auth, const char *db, SAppInstI
     tstrncpy(pObj->db, db, tListLen(pObj->db));
   }
 
-  pthread_mutex_init(&pObj->mutex, NULL);
+  taosThreadMutexInit(&pObj->mutex, NULL);
   pObj->id = taosAddRef(clientConnRefPool, pObj);
 
   tscDebug("connObj created, 0x%" PRIx64, pObj->id);
@@ -144,7 +143,7 @@ void *createTscObj(const char *user, const char *auth, const char *db, SAppInstI
 void *createRequest(STscObj *pObj, __taos_async_fn_t fp, void *param, int32_t type) {
   assert(pObj != NULL);
 
-  SRequestObj *pRequest = (SRequestObj *)calloc(1, sizeof(SRequestObj));
+  SRequestObj *pRequest = (SRequestObj *)taosMemoryCalloc(1, sizeof(SRequestObj));
   if (NULL == pRequest) {
     terrno = TSDB_CODE_TSC_OUT_OF_MEMORY;
     return NULL;
@@ -157,7 +156,7 @@ void *createRequest(STscObj *pObj, __taos_async_fn_t fp, void *param, int32_t ty
   pRequest->type = type;
   pRequest->pTscObj = pObj;
   pRequest->body.fp = fp;  // not used it yet
-  pRequest->msgBuf = calloc(1, ERROR_MSG_BUF_DEFAULT_SIZE);
+  pRequest->msgBuf = taosMemoryCalloc(1, ERROR_MSG_BUF_DEFAULT_SIZE);
   tsem_init(&pRequest->body.rspSem, 0, 0);
 
   registerRequest(pRequest);
@@ -165,11 +164,18 @@ void *createRequest(STscObj *pObj, __taos_async_fn_t fp, void *param, int32_t ty
 }
 
 static void doFreeReqResultInfo(SReqResultInfo *pResInfo) {
-  tfree(pResInfo->pRspMsg);
-  tfree(pResInfo->length);
-  tfree(pResInfo->row);
-  tfree(pResInfo->pCol);
-  tfree(pResInfo->fields);
+  taosMemoryFreeClear(pResInfo->pRspMsg);
+  taosMemoryFreeClear(pResInfo->length);
+  taosMemoryFreeClear(pResInfo->row);
+  taosMemoryFreeClear(pResInfo->pCol);
+  taosMemoryFreeClear(pResInfo->fields);
+
+  if (pResInfo->convertBuf != NULL) {
+    for (int32_t i = 0; i < pResInfo->numOfCols; ++i) {
+      taosMemoryFreeClear(pResInfo->convertBuf[i]);
+    }
+    taosMemoryFreeClear(pResInfo->convertBuf);
+  }
 }
 
 static void doDestroyRequest(void *p) {
@@ -178,20 +184,24 @@ static void doDestroyRequest(void *p) {
 
   assert(RID_VALID(pRequest->self));
 
-  tfree(pRequest->msgBuf);
-  tfree(pRequest->sqlstr);
-  tfree(pRequest->pInfo);
-  tfree(pRequest->pDb);
+  taosMemoryFreeClear(pRequest->msgBuf);
+  taosMemoryFreeClear(pRequest->sqlstr);
+  taosMemoryFreeClear(pRequest->pInfo);
+  taosMemoryFreeClear(pRequest->pDb);
 
   doFreeReqResultInfo(&pRequest->body.resInfo);
   qDestroyQueryPlan(pRequest->body.pDag);
 
+  if (pRequest->body.queryJob != 0) {
+    schedulerFreeJob(pRequest->body.queryJob);
+  }
+
   if (pRequest->body.showInfo.pArray != NULL) {
     taosArrayDestroy(pRequest->body.showInfo.pArray);
   }
 
   deregisterRequest(pRequest);
-  tfree(pRequest);
+  taosMemoryFreeClear(pRequest);
 }
 
 void destroyRequest(SRequestObj *pRequest) {
@@ -243,7 +253,7 @@ void taos_init_imp(void) {
 
   // transDestroyBuffer(&conn->readBuf);
   taosGetAppName(appInfo.appName, NULL);
-  pthread_mutex_init(&appInfo.mutex, NULL);
+  taosThreadMutexInit(&appInfo.mutex, NULL);
 
   appInfo.pid = taosGetPId();
   appInfo.startTime = taosGetTimestampMs();
@@ -252,7 +262,7 @@ void taos_init_imp(void) {
 }
 
 int taos_init() {
-  pthread_once(&tscinit, taos_init_imp);
+  taosThreadOnce(&tscinit, taos_init_imp);
   return tscInitRes;
 }
 
@@ -350,7 +360,7 @@ int taos_options_imp(TSDB_OPTION option, const char *str) {
             tscInfo("charset:%s is not valid in locale, charset remains:%s", charset, tsCharset);
           }
 
-          free(charset);
+          taosMemoryFree(charset);
         } else {  // it may be windows system
           tscInfo("charset remains:%s", tsCharset);
         }
@@ -508,9 +518,9 @@ static setConfRet taos_set_config_imp(const char *config){
 }
 
 setConfRet taos_set_config(const char *config){
-  pthread_mutex_lock(&setConfMutex);
+  taosThreadMutexLock(&setConfMutex);
   setConfRet ret = taos_set_config_imp(config);
-  pthread_mutex_unlock(&setConfMutex);
+  taosThreadMutexUnlock(&setConfMutex);
   return ret;
 }
 #endif

source/client/src/clientHb.c

@@ -166,7 +166,7 @@ static int32_t hbQueryHbRspHandle(SAppHbMgr *pAppHbMgr, SClientHbRsp *pRsp) {
 static int32_t hbAsyncCallBack(void *param, const SDataBuf *pMsg, int32_t code) {
   static int32_t emptyRspNum = 0;
   if (code != 0) {
-    tfree(param);
+    taosMemoryFreeClear(param);
     return -1;
   }
 
@@ -179,12 +179,12 @@ static int32_t hbAsyncCallBack(void *param, const SDataBuf *pMsg, int32_t code)
   SAppInstInfo **pInst = taosHashGet(appInfo.pInstMap, key, strlen(key));
   if (pInst == NULL || NULL == *pInst) {
     tscError("cluster not exist, key:%s", key);
-    tfree(param);
+    taosMemoryFreeClear(param);
     tFreeClientHbBatchRsp(&pRsp);
     return -1;
   }
 
-  tfree(param);
+  taosMemoryFreeClear(param);
 
   if (rspNum) {
     tscDebug("hb got %d rsp, %d empty rsp received before", rspNum,
@@ -317,7 +317,7 @@ void hbFreeReq(void *req) {
 }
 
 SClientHbBatchReq *hbGatherAllInfo(SAppHbMgr *pAppHbMgr) {
-  SClientHbBatchReq *pBatchReq = calloc(1, sizeof(SClientHbBatchReq));
+  SClientHbBatchReq *pBatchReq = taosMemoryCalloc(1, sizeof(SClientHbBatchReq));
   if (pBatchReq == NULL) {
     terrno = TSDB_CODE_TSC_OUT_OF_MEMORY;
     return NULL;
@@ -346,7 +346,7 @@ SClientHbBatchReq *hbGatherAllInfo(SAppHbMgr *pAppHbMgr) {
 
   if (code) {
     taosArrayDestroyEx(pBatchReq->reqs, hbFreeReq);
-    tfree(pBatchReq);
+    taosMemoryFreeClear(pBatchReq);
   }
 
   return pBatchReq;
@@ -372,7 +372,7 @@ static void *hbThreadFunc(void *param) {
       break;
     }
 
-    pthread_mutex_lock(&clientHbMgr.lock);
+    taosThreadMutexLock(&clientHbMgr.lock);
 
     int sz = taosArrayGetSize(clientHbMgr.appHbMgrs);
     for (int i = 0; i < sz; i++) {
@@ -387,7 +387,7 @@ static void *hbThreadFunc(void *param) {
         continue;
       }
       int   tlen = tSerializeSClientHbBatchReq(NULL, 0, pReq);
-      void *buf = malloc(tlen);
+      void *buf = taosMemoryMalloc(tlen);
       if (buf == NULL) {
         terrno = TSDB_CODE_TSC_OUT_OF_MEMORY;
         tFreeClientHbBatchReq(pReq, false);
@@ -396,13 +396,13 @@ static void *hbThreadFunc(void *param) {
       }
 
       tSerializeSClientHbBatchReq(buf, tlen, pReq);
-      SMsgSendInfo *pInfo = calloc(1, sizeof(SMsgSendInfo));
+      SMsgSendInfo *pInfo = taosMemoryCalloc(1, sizeof(SMsgSendInfo));
 
       if (pInfo == NULL) {
         terrno = TSDB_CODE_TSC_OUT_OF_MEMORY;
         tFreeClientHbBatchReq(pReq, false);
         hbClearReqInfo(pAppHbMgr);
-        free(buf);
+        taosMemoryFree(buf);
         break;
       }
       pInfo->fp = hbAsyncCallBack;
@@ -423,7 +423,7 @@ static void *hbThreadFunc(void *param) {
       atomic_add_fetch_32(&pAppHbMgr->reportCnt, 1);
     }
 
-    pthread_mutex_unlock(&clientHbMgr.lock);
+    taosThreadMutexUnlock(&clientHbMgr.lock);
 
     taosMsleep(HEARTBEAT_INTERVAL);
   }
@@ -431,15 +431,15 @@ static void *hbThreadFunc(void *param) {
 }
 
 static int32_t hbCreateThread() {
-  pthread_attr_t thAttr;
-  pthread_attr_init(&thAttr);
-  pthread_attr_setdetachstate(&thAttr, PTHREAD_CREATE_JOINABLE);
+  TdThreadAttr thAttr;
+  taosThreadAttrInit(&thAttr);
+  taosThreadAttrSetDetachState(&thAttr, PTHREAD_CREATE_JOINABLE);
 
-//  if (pthread_create(&clientHbMgr.thread, &thAttr, hbThreadFunc, NULL) != 0) {
+//  if (taosThreadCreate(&clientHbMgr.thread, &thAttr, hbThreadFunc, NULL) != 0) {
 //    terrno = TAOS_SYSTEM_ERROR(errno);
 //    return -1;
 //  }
-//  pthread_attr_destroy(&thAttr);
+//  taosThreadAttrDestroy(&thAttr);
   return 0;
 }
 
@@ -458,7 +458,7 @@ static void hbStopThread() {
 
 SAppHbMgr *appHbMgrInit(SAppInstInfo *pAppInstInfo, char *key) {
   hbMgrInit();
-  SAppHbMgr *pAppHbMgr = malloc(sizeof(SAppHbMgr));
+  SAppHbMgr *pAppHbMgr = taosMemoryMalloc(sizeof(SAppHbMgr));
   if (pAppHbMgr == NULL) {
     terrno = TSDB_CODE_OUT_OF_MEMORY;
     return NULL;
@@ -478,7 +478,7 @@ SAppHbMgr *appHbMgrInit(SAppInstInfo *pAppInstInfo, char *key) {
 
   if (pAppHbMgr->activeInfo == NULL) {
     terrno = TSDB_CODE_OUT_OF_MEMORY;
-    free(pAppHbMgr);
+    taosMemoryFree(pAppHbMgr);
     return NULL;
   }
 
@@ -488,19 +488,19 @@ SAppHbMgr *appHbMgrInit(SAppInstInfo *pAppInstInfo, char *key) {
 
   if (pAppHbMgr->connInfo == NULL) {
     terrno = TSDB_CODE_OUT_OF_MEMORY;
-    free(pAppHbMgr);
+    taosMemoryFree(pAppHbMgr);
     return NULL;
   }
 
-  pthread_mutex_lock(&clientHbMgr.lock);
+  taosThreadMutexLock(&clientHbMgr.lock);
   taosArrayPush(clientHbMgr.appHbMgrs, &pAppHbMgr);
-  pthread_mutex_unlock(&clientHbMgr.lock);
+  taosThreadMutexUnlock(&clientHbMgr.lock);
 
   return pAppHbMgr;
 }
 
 void appHbMgrCleanup(void) {
-  pthread_mutex_lock(&clientHbMgr.lock);
+  taosThreadMutexLock(&clientHbMgr.lock);
 
   int sz = taosArrayGetSize(clientHbMgr.appHbMgrs);
   for (int i = 0; i < sz; i++) {
@@ -511,7 +511,7 @@ void appHbMgrCleanup(void) {
     pTarget->connInfo = NULL;
   }
 
-  pthread_mutex_unlock(&clientHbMgr.lock);
+  taosThreadMutexUnlock(&clientHbMgr.lock);
 }
 
 int hbMgrInit() {
@@ -520,7 +520,7 @@ int hbMgrInit() {
   if (old == 1) return 0;
 
   clientHbMgr.appHbMgrs = taosArrayInit(0, sizeof(void *));
-  pthread_mutex_init(&clientHbMgr.lock, NULL);
+  taosThreadMutexInit(&clientHbMgr.lock, NULL);
 
   // init handle funcs
   hbMgrInitHandle();
@@ -539,10 +539,10 @@ void hbMgrCleanUp() {
   int8_t old = atomic_val_compare_exchange_8(&clientHbMgr.inited, 1, 0);
   if (old == 0) return;
 
-  pthread_mutex_lock(&clientHbMgr.lock);
+  taosThreadMutexLock(&clientHbMgr.lock);
   appHbMgrCleanup();
   taosArrayDestroy(clientHbMgr.appHbMgrs);
-  pthread_mutex_unlock(&clientHbMgr.lock);
+  taosThreadMutexUnlock(&clientHbMgr.lock);
 
   clientHbMgr.appHbMgrs = NULL;
 #endif
@@ -580,7 +580,7 @@ int hbRegisterConn(SAppHbMgr *pAppHbMgr, int32_t connId, int64_t clusterId, int3
 
   switch (hbType) {
     case HEARTBEAT_TYPE_QUERY: {
-      int64_t *pClusterId = malloc(sizeof(int64_t));
+      int64_t *pClusterId = taosMemoryMalloc(sizeof(int64_t));
       *pClusterId = clusterId;
 
       info.param = pClusterId;

source/client/src/clientImpl.c

@@ -12,7 +12,7 @@
 static int32_t       initEpSetFromCfg(const char* firstEp, const char* secondEp, SCorEpSet* pEpSet);
 static SMsgSendInfo* buildConnectMsg(SRequestObj* pRequest);
 static void          destroySendMsgInfo(SMsgSendInfo* pMsgBody);
-static void          setQueryResultFromRsp(SReqResultInfo* pResultInfo, const SRetrieveTableRsp* pRsp);
+static int32_t       setQueryResultFromRsp(SReqResultInfo* pResultInfo, const SRetrieveTableRsp* pRsp);
 
 static bool stringLengthCheck(const char* str, size_t maxsize) {
   if (str == NULL) {
@@ -95,12 +95,12 @@ TAOS* taos_connect_internal(const char* ip, const char* user, const char* pass,
   char*          key = getClusterKey(user, secretEncrypt, ip, port);
   SAppInstInfo** pInst = NULL;
 
-  pthread_mutex_lock(&appInfo.mutex);
+  taosThreadMutexLock(&appInfo.mutex);
 
   pInst = taosHashGet(appInfo.pInstMap, key, strlen(key));
   SAppInstInfo* p = NULL;
   if (pInst == NULL) {
-    p = calloc(1, sizeof(struct SAppInstInfo));
+    p = taosMemoryCalloc(1, sizeof(struct SAppInstInfo));
     p->mgmtEp = epSet;
     p->pTransporter = openTransporter(user, secretEncrypt, tsNumOfCores);
     p->pAppHbMgr = appHbMgrInit(p, key);
@@ -109,9 +109,9 @@ TAOS* taos_connect_internal(const char* ip, const char* user, const char* pass,
     pInst = &p;
   }
 
-  pthread_mutex_unlock(&appInfo.mutex);
+  taosThreadMutexUnlock(&appInfo.mutex);
 
-  tfree(key);
+  taosMemoryFreeClear(key);
   return taosConnectImpl(user, &secretEncrypt[0], localDb, NULL, NULL, *pInst);
 }
 
@@ -122,7 +122,7 @@ int32_t buildRequest(STscObj* pTscObj, const char* sql, int sqlLen, SRequestObj*
     return TSDB_CODE_TSC_OUT_OF_MEMORY;
   }
 
-  (*pRequest)->sqlstr = malloc(sqlLen + 1);
+  (*pRequest)->sqlstr = taosMemoryMalloc(sqlLen + 1);
   if ((*pRequest)->sqlstr == NULL) {
     tscError("0x%" PRIx64 " failed to prepare sql string buffer", (*pRequest)->self);
     (*pRequest)->msgBuf = strdup("failed to prepare sql string buffer");
@@ -137,14 +137,14 @@ int32_t buildRequest(STscObj* pTscObj, const char* sql, int sqlLen, SRequestObj*
   return TSDB_CODE_SUCCESS;
 }
 
-int32_t parseSql(SRequestObj* pRequest, bool streamQuery, SQuery** pQuery) {
+int32_t parseSql(SRequestObj* pRequest, bool topicQuery, SQuery** pQuery) {
   STscObj* pTscObj = pRequest->pTscObj;
 
   SParseContext cxt = {
       .requestId = pRequest->requestId,
       .acctId = pTscObj->acctId,
       .db = pRequest->pDb,
-      .streamQuery = streamQuery,
+      .topicQuery = topicQuery,
       .pSql = pRequest->sqlstr,
       .sqlLen = pRequest->sqlLen,
       .pMsg = pRequest->msgBuf,
@@ -159,8 +159,12 @@ int32_t parseSql(SRequestObj* pRequest, bool streamQuery, SQuery** pQuery) {
   }
 
   code = qParseQuerySql(&cxt, pQuery);
-  if (TSDB_CODE_SUCCESS == code && ((*pQuery)->haveResultSet)) {
-    setResSchemaInfo(&pRequest->body.resInfo, (*pQuery)->pResSchema, (*pQuery)->numOfResCols);
+  if (TSDB_CODE_SUCCESS == code) {
+    if ((*pQuery)->haveResultSet) {
+      setResSchemaInfo(&pRequest->body.resInfo, (*pQuery)->pResSchema, (*pQuery)->numOfResCols);
+    }
+    TSWAP(pRequest->dbList, (*pQuery)->pDbList, SArray*);
+    TSWAP(pRequest->tableList, (*pQuery)->pTableList, SArray*);
   }
 
   return code;
@@ -191,15 +195,25 @@ int32_t execDdlQuery(SRequestObj* pRequest, SQuery* pQuery) {
 
 int32_t getPlan(SRequestObj* pRequest, SQuery* pQuery, SQueryPlan** pPlan, SArray* pNodeList) {
   pRequest->type = pQuery->msgType;
-  SPlanContext cxt = { .queryId = pRequest->requestId, .pAstRoot = pQuery->pRoot, .acctId = pRequest->pTscObj->acctId };
-  return qCreateQueryPlan(&cxt, pPlan, pNodeList);
+  SPlanContext cxt = {
+    .queryId = pRequest->requestId,
+    .acctId = pRequest->pTscObj->acctId,
+    .mgmtEpSet = getEpSet_s(&pRequest->pTscObj->pAppInfo->mgmtEp),
+    .pAstRoot = pQuery->pRoot,
+    .showRewrite = pQuery->showRewrite
+  };
+  int32_t  code = qCreateQueryPlan(&cxt, pPlan, pNodeList);
+  if (code != 0) {
+    return code;
+  }
+  return code;
 }
 
 void setResSchemaInfo(SReqResultInfo* pResInfo, const SSchema* pSchema, int32_t numOfCols) {
   assert(pSchema != NULL && numOfCols > 0);
 
   pResInfo->numOfCols = numOfCols;
-  pResInfo->fields = calloc(numOfCols, sizeof(pSchema[0]));
+  pResInfo->fields = taosMemoryCalloc(numOfCols, sizeof(pSchema[0]));
 
   for (int32_t i = 0; i < pResInfo->numOfCols; ++i) {
     pResInfo->fields[i].bytes = pSchema[i].bytes;
@@ -219,6 +233,7 @@ int32_t scheduleQuery(SRequestObj* pRequest, SQueryPlan* pDag, SArray* pNodeList
     }
 
     pRequest->code = code;
+    terrno = code;
     return pRequest->code;
   }
 
@@ -231,22 +246,16 @@ int32_t scheduleQuery(SRequestObj* pRequest, SQueryPlan* pDag, SArray* pNodeList
   }
 
   pRequest->code = res.code;
+  terrno = res.code;  
   return pRequest->code;
 }
 
-TAOS_RES* taos_query_l(TAOS* taos, const char* sql, int sqlLen) {
-  STscObj* pTscObj = (STscObj*)taos;
-  if (sqlLen > (size_t)TSDB_MAX_ALLOWED_SQL_LEN) {
-    tscError("sql string exceeds max length:%d", TSDB_MAX_ALLOWED_SQL_LEN);
-    terrno = TSDB_CODE_TSC_EXCEED_SQL_LIMIT;
-    return NULL;
-  }
-
+SRequestObj* execQueryImpl(STscObj* pTscObj, const char* sql, int sqlLen) {
   SRequestObj* pRequest = NULL;
   SQuery* pQuery = NULL;
+  int32_t code = 0;
   SArray* pNodeList = taosArrayInit(4, sizeof(struct SQueryNodeAddr));
 
-  terrno = TSDB_CODE_SUCCESS;
   CHECK_CODE_GOTO(buildRequest(pTscObj, sql, sqlLen, &pRequest), _return);
   CHECK_CODE_GOTO(parseSql(pRequest, false, &pQuery), _return);
 
@@ -255,19 +264,91 @@ TAOS_RES* taos_query_l(TAOS* taos, const char* sql, int sqlLen) {
   } else {
     CHECK_CODE_GOTO(getPlan(pRequest, pQuery, &pRequest->body.pDag, pNodeList), _return);
     CHECK_CODE_GOTO(scheduleQuery(pRequest, pRequest->body.pDag, pNodeList), _return);
-    pRequest->code = terrno;
   }
 
 _return:
   taosArrayDestroy(pNodeList);
   qDestroyQuery(pQuery);
-  if (NULL != pRequest && TSDB_CODE_SUCCESS != terrno) {
+  if (NULL != pRequest && TSDB_CODE_SUCCESS != code) {
     pRequest->code = terrno;
   }
 
   return pRequest;
 }
 
+int32_t refreshMeta(STscObj* pTscObj, SRequestObj* pRequest) {
+  SCatalog *pCatalog = NULL;
+  int32_t code = 0;
+  int32_t dbNum = taosArrayGetSize(pRequest->dbList);
+  int32_t tblNum = taosArrayGetSize(pRequest->tableList);
+
+  if (dbNum <= 0 && tblNum <= 0) {
+    return TSDB_CODE_QRY_APP_ERROR;
+  }
+  
+  code = catalogGetHandle(pTscObj->pAppInfo->clusterId, &pCatalog);
+  if (code != TSDB_CODE_SUCCESS) {
+    return code;
+  }
+
+  SEpSet epset = getEpSet_s(&pTscObj->pAppInfo->mgmtEp);
+
+  for (int32_t i = 0; i < dbNum; ++i) {
+    char *dbFName = taosArrayGet(pRequest->dbList, i);
+    
+    code = catalogRefreshDBVgInfo(pCatalog, pTscObj->pAppInfo->pTransporter, &epset, dbFName);
+    if (code != TSDB_CODE_SUCCESS) {
+      return code;
+    }
+  }
+
+  for (int32_t i = 0; i < tblNum; ++i) {
+    SName *tableName = taosArrayGet(pRequest->tableList, i);
+
+    code = catalogRefreshTableMeta(pCatalog, pTscObj->pAppInfo->pTransporter, &epset, tableName, -1);
+    if (code != TSDB_CODE_SUCCESS) {
+      return code;
+    }
+  }
+
+  return code;
+}
+
+
+SRequestObj* execQuery(STscObj* pTscObj, const char* sql, int sqlLen) {
+  SRequestObj* pRequest = NULL;
+  int32_t retryNum = 0;
+  int32_t code = 0;
+
+  while (retryNum++ < REQUEST_MAX_TRY_TIMES) {
+    pRequest = execQueryImpl(pTscObj, sql, sqlLen);
+    if (TSDB_CODE_SUCCESS == pRequest->code || !NEED_CLIENT_HANDLE_ERROR(pRequest->code)) {
+      break;
+    }
+
+    code = refreshMeta(pTscObj, pRequest);
+    if (code) {
+      pRequest->code = code;
+      break;
+    }
+
+    destroyRequest(pRequest);
+  }
+  
+  return pRequest;
+}
+
+TAOS_RES* taos_query_l(TAOS* taos, const char* sql, int sqlLen) {
+  STscObj* pTscObj = (STscObj*)taos;
+  if (sqlLen > (size_t)TSDB_MAX_ALLOWED_SQL_LEN) {
+    tscError("sql string exceeds max length:%d", TSDB_MAX_ALLOWED_SQL_LEN);
+    terrno = TSDB_CODE_TSC_EXCEED_SQL_LIMIT;
+    return NULL;
+  }
+
+  return execQuery(pTscObj, sql, sqlLen);
+}
+
 int initEpSetFromCfg(const char* firstEp, const char* secondEp, SCorEpSet* pEpSet) {
   pEpSet->version = 0;
 
@@ -343,7 +424,7 @@ STscObj* taosConnectImpl(const char* user, const char* auth, const char* db, __t
 }
 
 static SMsgSendInfo* buildConnectMsg(SRequestObj* pRequest) {
-  SMsgSendInfo* pMsgSendInfo = calloc(1, sizeof(SMsgSendInfo));
+  SMsgSendInfo* pMsgSendInfo = taosMemoryCalloc(1, sizeof(SMsgSendInfo));
   if (pMsgSendInfo == NULL) {
     terrno = TSDB_CODE_TSC_OUT_OF_MEMORY;
     return NULL;
@@ -363,14 +444,14 @@ static SMsgSendInfo* buildConnectMsg(SRequestObj* pRequest) {
   if (db != NULL) {
     tstrncpy(connectReq.db, db, sizeof(connectReq.db));
   }
-  tfree(db);
+  taosMemoryFreeClear(db);
 
   connectReq.pid = htonl(appInfo.pid);
   connectReq.startTime = htobe64(appInfo.startTime);
   tstrncpy(connectReq.app, appInfo.appName, sizeof(connectReq.app));
 
   int32_t contLen = tSerializeSConnectReq(NULL, 0, &connectReq);
-  void*   pReq = malloc(contLen);
+  void*   pReq = taosMemoryMalloc(contLen);
   tSerializeSConnectReq(pReq, contLen, &connectReq);
 
   pMsgSendInfo->msgInfo.len = contLen;
@@ -380,11 +461,11 @@ static SMsgSendInfo* buildConnectMsg(SRequestObj* pRequest) {
 
 static void destroySendMsgInfo(SMsgSendInfo* pMsgBody) {
   assert(pMsgBody != NULL);
-  tfree(pMsgBody->msgInfo.pData);
-  tfree(pMsgBody);
+  taosMemoryFreeClear(pMsgBody->msgInfo.pData);
+  taosMemoryFreeClear(pMsgBody);
 }
 bool persistConnForSpecificMsg(void* parenct, tmsg_t msgType) {
-  return msgType == TDMT_VND_QUERY_RSP || msgType == TDMT_VND_FETCH_RSP || msgType == TDMT_VND_RES_READY_RSP;
+  return msgType == TDMT_VND_QUERY_RSP || msgType == TDMT_VND_FETCH_RSP || msgType == TDMT_VND_RES_READY_RSP || msgType == TDMT_VND_QUERY_HEARTBEAT_RSP;
 }
 void processMsgFromServer(void* parent, SRpcMsg* pMsg, SEpSet* pEpSet) {
   SMsgSendInfo* pSendInfo = (SMsgSendInfo*)pMsg->ahandle;
@@ -395,7 +476,6 @@ void processMsgFromServer(void* parent, SRpcMsg* pMsg, SEpSet* pEpSet) {
     assert(pRequest->self == pSendInfo->requestObjRefId);
 
     pRequest->metric.rsp = taosGetTimestampMs();
-    pRequest->code = pMsg->code;
 
     STscObj* pTscObj = pRequest->pTscObj;
     if (pEpSet) {
@@ -423,7 +503,7 @@ void processMsgFromServer(void* parent, SRpcMsg* pMsg, SEpSet* pEpSet) {
   SDataBuf buf = {.len = pMsg->contLen, .pData = NULL, .handle = pMsg->handle};
 
   if (pMsg->contLen > 0) {
-    buf.pData = calloc(1, pMsg->contLen);
+    buf.pData = taosMemoryCalloc(1, pMsg->contLen);
     if (buf.pData == NULL) {
       terrno = TSDB_CODE_OUT_OF_MEMORY;
       pMsg->code = TSDB_CODE_OUT_OF_MEMORY;
@@ -479,13 +559,16 @@ void* doFetchRow(SRequestObj* pRequest) {
       }
 
       SReqResultInfo* pResInfo = &pRequest->body.resInfo;
-      int32_t         code = schedulerFetchRows(pRequest->body.queryJob, (void**)&pResInfo->pData);
-      if (code != TSDB_CODE_SUCCESS) {
-        pRequest->code = code;
+      pRequest->code = schedulerFetchRows(pRequest->body.queryJob, (void**)&pResInfo->pData);
+      if (pRequest->code != TSDB_CODE_SUCCESS) {
+        return NULL;
+      }
+
+      pRequest->code = setQueryResultFromRsp(&pRequest->body.resInfo, (SRetrieveTableRsp*)pResInfo->pData);
+      if (pRequest->code != TSDB_CODE_SUCCESS) {
         return NULL;
       }
 
-      setQueryResultFromRsp(&pRequest->body.resInfo, (SRetrieveTableRsp*)pResInfo->pData);
       tscDebug("0x%" PRIx64 " fetch results, numOfRows:%d total Rows:%" PRId64 ", complete:%d, reqId:0x%" PRIx64,
                pRequest->self, pResInfo->numOfRows, pResInfo->totalRows, pResInfo->completed, pRequest->requestId);
 
@@ -512,7 +595,7 @@ void* doFetchRow(SRequestObj* pRequest) {
       }
 
       SVgroupInfo*     pVgroupInfo = taosArrayGet(pShowReqInfo->pArray, pShowReqInfo->currentIndex);
-      SVShowTablesReq* pShowReq = calloc(1, sizeof(SVShowTablesReq));
+      SVShowTablesReq* pShowReq = taosMemoryCalloc(1, sizeof(SVShowTablesReq));
       pShowReq->head.vgId = htonl(pVgroupInfo->vgId);
 
       pRequest->body.requestMsg.len = sizeof(SVShowTablesReq);
@@ -552,10 +635,35 @@ void* doFetchRow(SRequestObj* pRequest) {
 _return:
 
   for (int32_t i = 0; i < pResultInfo->numOfCols; ++i) {
-    pResultInfo->row[i] = pResultInfo->pCol[i] + pResultInfo->fields[i].bytes * pResultInfo->current;
-    if (IS_VAR_DATA_TYPE(pResultInfo->fields[i].type)) {
-      pResultInfo->length[i] = varDataLen(pResultInfo->row[i]);
-      pResultInfo->row[i] = varDataVal(pResultInfo->row[i]);
+    SResultColumn* pCol = &pResultInfo->pCol[i];
+
+    int32_t type = pResultInfo->fields[i].type;
+    int32_t bytes = pResultInfo->fields[i].bytes;
+
+    if (IS_VAR_DATA_TYPE(type)) {
+      if (pCol->offset[pResultInfo->current] != -1) {
+        char* pStart = pResultInfo->pCol[i].offset[pResultInfo->current] + pResultInfo->pCol[i].pData;
+
+        pResultInfo->length[i] = varDataLen(pStart);
+        pResultInfo->row[i] = varDataVal(pStart);
+
+        if (type == TSDB_DATA_TYPE_NCHAR) {
+          int32_t len = taosUcs4ToMbs((TdUcs4*)varDataVal(pStart), varDataLen(pStart), varDataVal(pResultInfo->convertBuf[i]));
+          ASSERT(len <= bytes);
+
+          pResultInfo->row[i] = varDataVal(pResultInfo->convertBuf[i]);
+          varDataSetLen(pResultInfo->convertBuf[i], len);
+          pResultInfo->length[i] = len;
+        }
+      } else {
+        pResultInfo->row[i] = NULL;
+      }
+    } else {
+      if (!colDataIsNull_f(pCol->nullbitmap, pResultInfo->current)) {
+        pResultInfo->row[i] = pResultInfo->pCol[i].pData + bytes * pResultInfo->current;
+      } else {
+        pResultInfo->row[i] = NULL;
+      }
     }
   }
 
@@ -563,60 +671,91 @@ _return:
   return pResultInfo->row;
 }
 
-static void doPrepareResPtr(SReqResultInfo* pResInfo) {
+static int32_t doPrepareResPtr(SReqResultInfo* pResInfo) {
   if (pResInfo->row == NULL) {
-    pResInfo->row = calloc(pResInfo->numOfCols, POINTER_BYTES);
-    pResInfo->pCol = calloc(pResInfo->numOfCols, POINTER_BYTES);
-    pResInfo->length = calloc(pResInfo->numOfCols, sizeof(int32_t));
+    pResInfo->row    = taosMemoryCalloc(pResInfo->numOfCols, POINTER_BYTES);
+    pResInfo->pCol   = taosMemoryCalloc(pResInfo->numOfCols, sizeof(SResultColumn));
+    pResInfo->length = taosMemoryCalloc(pResInfo->numOfCols, sizeof(int32_t));
+    pResInfo->convertBuf = taosMemoryCalloc(pResInfo->numOfCols, POINTER_BYTES);
+
+    if (pResInfo->row == NULL || pResInfo->pCol == NULL || pResInfo->length == NULL || pResInfo->convertBuf == NULL) {
+      return TSDB_CODE_OUT_OF_MEMORY;
+    }
+
+    for(int32_t i = 0; i < pResInfo->numOfCols; ++i) {
+      if(pResInfo->fields[i].type == TSDB_DATA_TYPE_NCHAR) {
+        pResInfo->convertBuf[i] = taosMemoryCalloc(1, NCHAR_WIDTH_TO_BYTES(pResInfo->fields[i].bytes));
+      }
+    }
   }
+
+  return TSDB_CODE_SUCCESS;
 }
 
-void setResultDataPtr(SReqResultInfo* pResultInfo, TAOS_FIELD* pFields, int32_t numOfCols, int32_t numOfRows) {
+int32_t setResultDataPtr(SReqResultInfo* pResultInfo, TAOS_FIELD* pFields, int32_t numOfCols, int32_t numOfRows) {
   assert(numOfCols > 0 && pFields != NULL && pResultInfo != NULL);
   if (numOfRows == 0) {
-    return;
+    return TSDB_CODE_SUCCESS;
   }
 
-  // todo check for the failure of malloc
-  doPrepareResPtr(pResultInfo);
+  int32_t code = doPrepareResPtr(pResultInfo);
+  if (code != TSDB_CODE_SUCCESS) {
+    return code;
+  }
 
-  int32_t offset = 0;
+  int32_t* colLength = (int32_t*)pResultInfo->pData;
+  char*    pStart = ((char*)pResultInfo->pData) + sizeof(int32_t) * numOfCols;
   for (int32_t i = 0; i < numOfCols; ++i) {
+    colLength[i] = htonl(colLength[i]);
+
+    if (IS_VAR_DATA_TYPE(pResultInfo->fields[i].type)) {
+      pResultInfo->pCol[i].offset = (int32_t*)pStart;
+      pStart += numOfRows * sizeof(int32_t);
+    } else {
+      pResultInfo->pCol[i].nullbitmap = pStart;
+      pStart += BitmapLen(pResultInfo->numOfRows);
+    }
+
+    pResultInfo->pCol[i].pData = pStart;
     pResultInfo->length[i] = pResultInfo->fields[i].bytes;
-    pResultInfo->row[i] = (char*)(pResultInfo->pData + offset * pResultInfo->numOfRows);
-    pResultInfo->pCol[i] = pResultInfo->row[i];
-    offset += pResultInfo->fields[i].bytes;
+    pResultInfo->row[i] = pResultInfo->pCol[i].pData;
+
+    pStart += colLength[i];
   }
+
+  return TSDB_CODE_SUCCESS;
 }
 
 char* getDbOfConnection(STscObj* pObj) {
   char* p = NULL;
-  pthread_mutex_lock(&pObj->mutex);
+  taosThreadMutexLock(&pObj->mutex);
   size_t len = strlen(pObj->db);
   if (len > 0) {
     p = strndup(pObj->db, tListLen(pObj->db));
   }
 
-  pthread_mutex_unlock(&pObj->mutex);
+  taosThreadMutexUnlock(&pObj->mutex);
   return p;
 }
 
 void setConnectionDB(STscObj* pTscObj, const char* db) {
   assert(db != NULL && pTscObj != NULL);
-  pthread_mutex_lock(&pTscObj->mutex);
+  taosThreadMutexLock(&pTscObj->mutex);
   tstrncpy(pTscObj->db, db, tListLen(pTscObj->db));
-  pthread_mutex_unlock(&pTscObj->mutex);
+  taosThreadMutexUnlock(&pTscObj->mutex);
 }
 
-void setQueryResultFromRsp(SReqResultInfo* pResultInfo, const SRetrieveTableRsp* pRsp) {
+int32_t setQueryResultFromRsp(SReqResultInfo* pResultInfo, const SRetrieveTableRsp* pRsp) {
   assert(pResultInfo != NULL && pRsp != NULL);
 
-  pResultInfo->pRspMsg = (const char*)pRsp;
-  pResultInfo->pData = (void*)pRsp->data;
-  pResultInfo->numOfRows = htonl(pRsp->numOfRows);
-  pResultInfo->current = 0;
-  pResultInfo->completed = (pRsp->completed == 1);
+  pResultInfo->pRspMsg    = (const char*)pRsp;
+  pResultInfo->pData      = (void*)pRsp->data;
+  pResultInfo->numOfRows  = htonl(pRsp->numOfRows);
+  pResultInfo->current    = 0;
+  pResultInfo->completed  = (pRsp->completed == 1);
+  pResultInfo->payloadLen = htonl(pRsp->compLen);
 
+  // TODO handle the compressed case
   pResultInfo->totalRows += pResultInfo->numOfRows;
-  setResultDataPtr(pResultInfo, pResultInfo->fields, pResultInfo->numOfCols, pResultInfo->numOfRows);
+  return setResultDataPtr(pResultInfo, pResultInfo->fields, pResultInfo->numOfCols, pResultInfo->numOfRows);
 }

source/dnode/bnode/inc/bndInt.h

@@ -30,7 +30,7 @@ extern "C" {
 #endif
 
 typedef struct SBnode {
-  SBnodeOpt opt;
+  SMsgCb msgCb;
 } SBnode;
 
 #ifdef __cplusplus