support drop sma in tsdb

cmake/stub_CMakeLists.txt.in

 
 # stub
-ExternalProject_Add(stub
-        GIT_REPOSITORY https://github.com/coolxv/cpp-stub.git
-        GIT_SUBMODULES "src"
-        SOURCE_DIR "${CMAKE_CONTRIB_DIR}/cpp-stub"
-        BINARY_DIR "${CMAKE_CONTRIB_DIR}/cpp-stub/src"
-        CONFIGURE_COMMAND ""
-        BUILD_COMMAND ""
-        INSTALL_COMMAND ""
-        TEST_COMMAND ""
-)
+#ExternalProject_Add(stub
+#        GIT_REPOSITORY https://github.com/coolxv/cpp-stub.git
+#        GIT_SUBMODULES "src"
+#        SOURCE_DIR "${CMAKE_CONTRIB_DIR}/cpp-stub"
+#        BINARY_DIR "${CMAKE_CONTRIB_DIR}/cpp-stub/src"
+#        CONFIGURE_COMMAND ""
+#        BUILD_COMMAND ""
+#        INSTALL_COMMAND ""
+#        TEST_COMMAND ""
+#)

example/src/tstream.c

@@ -63,7 +63,7 @@ int32_t init_env() {
 }
 
 int32_t create_stream() {
-  printf("create topic\n");
+  printf("create stream\n");
   TAOS_RES* pRes;
   TAOS*     pConn = taos_connect("localhost", "root", "taosdata", NULL, 0);
   if (pConn == NULL) {
@@ -77,7 +77,9 @@ int32_t create_stream() {
   }
   taos_free_result(pRes);
 
-  const char* sql = "select ts,k from tu1";
+  /*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));

include/common/taosdef.h

@@ -69,6 +69,7 @@ typedef enum {
   TSDB_SMA_STAT_DROPPED = 2,   // sma dropped
 } ETsdbSmaStat;                // bit operation
 
+
 typedef enum {
   TSDB_SMA_TYPE_BLOCK = 0,       // Block-wise SMA
   TSDB_SMA_TYPE_TIME_RANGE = 1,  // Time-range-wise SMA
@@ -77,6 +78,8 @@ typedef enum {
 
 extern char *qtypeStr[];
 
+#define TSDB_PORT_DNODEDNODE 5
+#define TSDB_PORT_SYNC      10
 #define TSDB_PORT_HTTP      11
 
 #ifdef __cplusplus

include/common/tmsg.h

@@ -23,8 +23,8 @@
 #include "tencode.h"
 #include "thash.h"
 #include "tlist.h"
-#include "trow.h"
 #include "tname.h"
+#include "trow.h"
 #include "tuuid.h"
 
 #ifdef __cplusplus
@@ -472,10 +472,9 @@ 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);
+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];
@@ -1428,9 +1427,8 @@ typedef struct {
   SArray* rspList;  // SArray<SVCreateTbRsp>
 } SVCreateTbBatchRsp;
 
-int32_t tSerializeSVCreateTbBatchRsp(void *buf, int32_t bufLen, SVCreateTbBatchRsp *pRsp);
-int32_t tDeserializeSVCreateTbBatchRsp(void *buf, int32_t bufLen, SVCreateTbBatchRsp *pRsp);
-
+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;
@@ -2312,6 +2310,11 @@ enum {
   STREAM_TASK_STATUS__STOP,
 };
 
+enum {
+  STREAM_NEXT_OP_DST__VND = 1,
+  STREAM_NEXT_OP_DST__SND,
+};
+
 typedef struct {
   void* inputHandle;
   void* executor;
@@ -2326,6 +2329,7 @@ typedef struct {
   int8_t  pipeSink;
   int8_t  numOfRunners;
   int8_t  parallelizable;
+  int8_t  nextOpDst;  // vnode or snode
   SEpSet  NextOpEp;
   char*   qmsg;
   // not applied to encoder and decoder
@@ -2341,6 +2345,8 @@ static FORCE_INLINE SStreamTask* streamTaskNew(int64_t streamId, int32_t level)
     return NULL;
   }
   pTask->taskId = tGenIdPI32();
+  pTask->streamId = streamId;
+  pTask->level = level;
   pTask->status = STREAM_TASK_STATUS__RUNNING;
   pTask->qmsg = NULL;
   return pTask;
@@ -2368,6 +2374,13 @@ typedef struct {
   int32_t reserved;
 } SStreamTaskExecRsp;
 
+typedef struct {
+  SMsgHead head;
+  int64_t  streamId;
+  int64_t  version;
+  SArray*  res;  // SArray<SSDataBlock>
+} SStreamSmaSinkReq;
+
 #pragma pack(pop)
 
 #ifdef __cplusplus

include/common/tmsgdef.h

@@ -191,6 +191,7 @@ enum {
   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_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)

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);
@@ -157,10 +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    syncPropose(int64_t rid, const SRpcMsg* pMsg, bool isWeak);        // use this function
-int32_t    syncForwardToPeer(int64_t rid, const SRpcMsg* pMsg, bool isWeak);  // just for compatibility
+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/os/osSocket.h

@@ -51,10 +51,31 @@
 extern "C" {
 #endif
 
-#if (defined(_TD_WINDOWS_64) || defined(_TD_WINDOWS_32)) 
+#if defined(WINDOWS)
   #define htobe64 htonll
 #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)
 //  #define htobe64 htonll
 
@@ -83,6 +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;
@@ -91,6 +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);

source/common/src/tglobal.c

@@ -132,6 +132,9 @@ bool tsdbForceKeepFile = false;
 int32_t  tsDiskCfgNum = 0;
 SDiskCfg tsDiskCfg[TFS_MAX_DISKS] = {0};
 
+// stream scheduler
+bool tsStreamSchedV = true;
+
 /*
  * minimum scale for whole system, millisecond by default
  * for TSDB_TIME_PRECISION_MILLI: 86400000L

source/common/src/tmsg.c

@@ -318,7 +318,7 @@ int32_t tSerializeSVCreateTbReq(void **buf, SVCreateTbReq *pReq) {
         tlen += taosEncodeFixedU32(buf, (uint32_t)param->xFilesFactor);
         tlen += taosEncodeFixedI8(buf, param->delayUnit);
         tlen += taosEncodeFixedI8(buf, param->nFuncIds);
-        for(int8_t i=0; i< param->nFuncIds; ++i) {
+        for (int8_t i = 0; i < param->nFuncIds; ++i) {
           tlen += taosEncodeFixedI32(buf, param->pFuncIds[i]);
         }
         tlen += taosEncodeFixedI64(buf, param->delay);
@@ -345,7 +345,7 @@ int32_t tSerializeSVCreateTbReq(void **buf, SVCreateTbReq *pReq) {
         tlen += taosEncodeFixedU32(buf, (uint32_t)param->xFilesFactor);
         tlen += taosEncodeFixedI8(buf, param->delayUnit);
         tlen += taosEncodeFixedI8(buf, param->nFuncIds);
-        for(int8_t i=0; i< param->nFuncIds; ++i) {
+        for (int8_t i = 0; i < param->nFuncIds; ++i) {
           tlen += taosEncodeFixedI32(buf, param->pFuncIds[i]);
         }
         tlen += taosEncodeFixedI64(buf, param->delay);
@@ -385,7 +385,7 @@ void *tDeserializeSVCreateTbReq(void *buf, SVCreateTbReq *pReq) {
         buf = taosDecodeStringTo(buf, pReq->stbCfg.pTagSchema[i].name);
       }
       buf = taosDecodeFixedI16(buf, &(pReq->stbCfg.nBSmaCols));
-      if(pReq->stbCfg.nBSmaCols > 0) {
+      if (pReq->stbCfg.nBSmaCols > 0) {
         pReq->stbCfg.pBSmaCols = (col_id_t *)malloc(pReq->stbCfg.nBSmaCols * sizeof(col_id_t));
         for (col_id_t i = 0; i < pReq->stbCfg.nBSmaCols; ++i) {
           buf = taosDecodeFixedI16(buf, pReq->stbCfg.pBSmaCols + i);
@@ -393,14 +393,14 @@ void *tDeserializeSVCreateTbReq(void *buf, SVCreateTbReq *pReq) {
       } else {
         pReq->stbCfg.pBSmaCols = NULL;
       }
-      if(pReq->rollup) {
+      if (pReq->rollup) {
         pReq->stbCfg.pRSmaParam = (SRSmaParam *)malloc(sizeof(SRSmaParam));
         SRSmaParam *param = pReq->stbCfg.pRSmaParam;
-        buf = taosDecodeFixedU32(buf, (uint32_t*)&param->xFilesFactor);
+        buf = taosDecodeFixedU32(buf, (uint32_t *)&param->xFilesFactor);
         buf = taosDecodeFixedI8(buf, &param->delayUnit);
         buf = taosDecodeFixedI8(buf, &param->nFuncIds);
-        if(param->nFuncIds > 0) {
-          for (int8_t i = 0; i< param->nFuncIds; ++i) {
+        if (param->nFuncIds > 0) {
+          for (int8_t i = 0; i < param->nFuncIds; ++i) {
             buf = taosDecodeFixedI32(buf, param->pFuncIds + i);
           }
         } else {
@@ -439,8 +439,8 @@ void *tDeserializeSVCreateTbReq(void *buf, SVCreateTbReq *pReq) {
         buf = taosDecodeFixedU32(buf, (uint32_t*)&param->xFilesFactor);
         buf = taosDecodeFixedI8(buf, &param->delayUnit);
         buf = taosDecodeFixedI8(buf, &param->nFuncIds);
-        if(param->nFuncIds > 0) {
-          for (int8_t i = 0; i< param->nFuncIds; ++i) {
+        if (param->nFuncIds > 0) {
+          for (int8_t i = 0; i < param->nFuncIds; ++i) {
             buf = taosDecodeFixedI32(buf, param->pFuncIds + i);
           }
         } else {
@@ -2775,7 +2775,6 @@ int32_t tDeserializeSVCreateTbBatchRsp(void *buf, int32_t bufLen, SVCreateTbBatc
   return 0;
 }
 
-
 int32_t tSerializeSVCreateTSmaReq(void **buf, SVCreateTSmaReq *pReq) {
   int32_t tlen = 0;
 
@@ -2824,7 +2823,8 @@ int32_t tSerializeSCMCreateStreamReq(void *buf, int32_t bufLen, const SCMCreateS
   if (tEncodeCStr(&encoder, pReq->name) < 0) return -1;
   if (tEncodeCStr(&encoder, pReq->outputTbName) < 0) return -1;
   if (tEncodeI8(&encoder, pReq->igExists) < 0) return -1;
-  if (tEncodeCStr(&encoder, pReq->sql) < 0) return -1;
+  if (tEncodeI32(&encoder, sqlLen) < 0) return -1;
+  if (tEncodeI32(&encoder, astLen) < 0) return -1;
   if (sqlLen > 0 && tEncodeCStr(&encoder, pReq->sql) < 0) return -1;
   if (astLen > 0 && tEncodeCStr(&encoder, pReq->ast) < 0) return -1;
 
@@ -2872,30 +2872,36 @@ void tFreeSCMCreateStreamReq(SCMCreateStreamReq *pReq) {
 }
 
 int32_t tEncodeSStreamTask(SCoder *pEncoder, const SStreamTask *pTask) {
-  if (tStartEncode(pEncoder) < 0) return -1;
+  /*if (tStartEncode(pEncoder) < 0) return -1;*/
   if (tEncodeI64(pEncoder, pTask->streamId) < 0) return -1;
   if (tEncodeI32(pEncoder, pTask->taskId) < 0) return -1;
   if (tEncodeI32(pEncoder, pTask->level) < 0) return -1;
   if (tEncodeI8(pEncoder, pTask->status) < 0) return -1;
+  if (tEncodeI8(pEncoder, pTask->pipeSource) < 0) return -1;
   if (tEncodeI8(pEncoder, pTask->pipeSink) < 0) return -1;
+  if (tEncodeI8(pEncoder, pTask->parallelizable) < 0) return -1;
+  if (tEncodeI8(pEncoder, pTask->nextOpDst) < 0) return -1;
   // if (tEncodeI8(pEncoder, pTask->numOfRunners) < 0) return -1;
   if (tEncodeSEpSet(pEncoder, &pTask->NextOpEp) < 0) return -1;
   if (tEncodeCStr(pEncoder, pTask->qmsg) < 0) return -1;
-  tEndEncode(pEncoder);
+  /*tEndEncode(pEncoder);*/
   return pEncoder->pos;
 }
 
 int32_t tDecodeSStreamTask(SCoder *pDecoder, SStreamTask *pTask) {
-  if (tStartDecode(pDecoder) < 0) return -1;
+  /*if (tStartDecode(pDecoder) < 0) return -1;*/
   if (tDecodeI64(pDecoder, &pTask->streamId) < 0) return -1;
   if (tDecodeI32(pDecoder, &pTask->taskId) < 0) return -1;
   if (tDecodeI32(pDecoder, &pTask->level) < 0) return -1;
   if (tDecodeI8(pDecoder, &pTask->status) < 0) return -1;
+  if (tDecodeI8(pDecoder, &pTask->pipeSource) < 0) return -1;
   if (tDecodeI8(pDecoder, &pTask->pipeSink) < 0) return -1;
+  if (tDecodeI8(pDecoder, &pTask->parallelizable) < 0) return -1;
+  if (tDecodeI8(pDecoder, &pTask->nextOpDst) < 0) return -1;
   // if (tDecodeI8(pDecoder, &pTask->numOfRunners) < 0) return -1;
   if (tDecodeSEpSet(pDecoder, &pTask->NextOpEp) < 0) return -1;
   if (tDecodeCStrAlloc(pDecoder, &pTask->qmsg) < 0) return -1;
-  tEndDecode(pDecoder);
+  /*tEndDecode(pDecoder);*/
   return 0;
 }
 

source/dnode/mgmt/mnode/src/mmMsg.c

@@ -142,6 +142,8 @@ void mmInitMsgHandles(SMgmtWrapper *pWrapper) {
   dndSetMsgHandle(pWrapper, TDMT_MND_SUBSCRIBE, (NodeMsgFp)mmProcessWriteMsg, 0);
   dndSetMsgHandle(pWrapper, TDMT_MND_MQ_COMMIT_OFFSET, (NodeMsgFp)mmProcessWriteMsg, 0);
   dndSetMsgHandle(pWrapper, TDMT_MND_GET_SUB_EP, (NodeMsgFp)mmProcessReadMsg, 0);
+  dndSetMsgHandle(pWrapper, TDMT_MND_CREATE_STREAM, (NodeMsgFp)mmProcessWriteMsg, 0);
+  dndSetMsgHandle(pWrapper, TDMT_VND_TASK_DEPLOY_RSP, (NodeMsgFp)mmProcessWriteMsg, 0);
 
   // Requests handled by VNODE
   dndSetMsgHandle(pWrapper, TDMT_VND_MQ_SET_CONN_RSP, (NodeMsgFp)mmProcessWriteMsg, 0);

source/dnode/mgmt/vnode/src/vmMsg.c

@@ -277,6 +277,7 @@ void vmInitMsgHandles(SMgmtWrapper *pWrapper) {
   dndSetMsgHandle(pWrapper, TDMT_VND_TASK_DEPLOY, (NodeMsgFp)vmProcessWriteMsg, 0);
   dndSetMsgHandle(pWrapper, TDMT_VND_QUERY_HEARTBEAT, (NodeMsgFp)vmProcessFetchMsg, 0);
   dndSetMsgHandle(pWrapper, TDMT_VND_TASK_EXEC, (NodeMsgFp)vmProcessFetchMsg, 0);
+  dndSetMsgHandle(pWrapper, TDMT_VND_STREAM_TRIGGER, (NodeMsgFp)vmProcessFetchMsg, 0);
 
   dndSetMsgHandle(pWrapper, TDMT_DND_CREATE_VNODE, (NodeMsgFp)vmProcessMgmtMsg, 0);
   dndSetMsgHandle(pWrapper, TDMT_DND_ALTER_VNODE, (NodeMsgFp)vmProcessMgmtMsg, 0);

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

@@ -697,12 +697,12 @@ typedef struct {
   char*   logicalPlan;
   char*   physicalPlan;
   SArray* tasks;  // SArray<SArray<SStreamTask>>
+  SArray* outputName;
 } SStreamObj;
 
 int32_t tEncodeSStreamObj(SCoder* pEncoder, const SStreamObj* pObj);
 int32_t tDecodeSStreamObj(SCoder* pDecoder, SStreamObj* pObj);
 
-
 #ifdef __cplusplus
 }
 #endif

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

@@ -16,6 +16,7 @@
 #include "mndDef.h"
 
 int32_t tEncodeSStreamObj(SCoder *pEncoder, const SStreamObj *pObj) {
+  int32_t outputNameSz = 0;
   if (tEncodeCStr(pEncoder, pObj->name) < 0) return -1;
   if (tEncodeCStr(pEncoder, pObj->db) < 0) return -1;
   if (tEncodeI64(pEncoder, pObj->createTime) < 0) return -1;
@@ -43,6 +44,15 @@ int32_t tEncodeSStreamObj(SCoder *pEncoder, const SStreamObj *pObj) {
   } else {
     tEncodeI32(pEncoder, 0);
   }
+
+  if (pObj->outputName != NULL) {
+    outputNameSz = taosArrayGetSize(pObj->outputName);
+  }
+  if (tEncodeI32(pEncoder, outputNameSz) < 0) return -1;
+  for (int32_t i = 0; i < outputNameSz; i++) {
+    char *name = taosArrayGetP(pObj->outputName, i);
+    if (tEncodeCStr(pEncoder, name) < 0) return -1;
+  }
   return pEncoder->pos;
 }
 
@@ -76,5 +86,16 @@ int32_t tDecodeSStreamObj(SCoder *pDecoder, SStreamObj *pObj) {
   } else {
     pObj->tasks = NULL;
   }
+  int32_t outputNameSz;
+  if (tDecodeI32(pDecoder, &outputNameSz) < 0) return -1;
+  pObj->outputName = taosArrayInit(outputNameSz, sizeof(void *));
+  if (pObj->outputName == NULL) {
+    return -1;
+  }
+  for (int32_t i = 0; i < outputNameSz; i++) {
+    char *name;
+    if (tDecodeCStrAlloc(pDecoder, &name) < 0) return -1;
+    taosArrayPush(pObj->outputName, &name);
+  }
   return 0;
 }

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

@@ -32,20 +32,23 @@
 #include "tname.h"
 #include "tuuid.h"
 
-int32_t mndPersistTaskDeployReq(STrans* pTrans, SStreamTask* pTask, const SEpSet* pEpSet, tmsg_t type) {
+extern bool tsStreamSchedV;
+
+int32_t mndPersistTaskDeployReq(STrans* pTrans, SStreamTask* pTask, const SEpSet* pEpSet, tmsg_t type, int32_t nodeId) {
   SCoder encoder;
   tCoderInit(&encoder, TD_LITTLE_ENDIAN, NULL, 0, TD_ENCODER);
   tEncodeSStreamTask(&encoder, pTask);
-  int32_t tlen = sizeof(SMsgHead) + encoder.pos;
+  int32_t size = encoder.pos;
+  int32_t tlen = sizeof(SMsgHead) + size;
   tCoderClear(&encoder);
   void* buf = malloc(tlen);
   if (buf == NULL) {
     terrno = TSDB_CODE_OUT_OF_MEMORY;
     return -1;
   }
-  ((SMsgHead*)buf)->streamTaskId = pTask->taskId;
+  ((SMsgHead*)buf)->streamTaskId = htonl(nodeId);
   void* abuf = POINTER_SHIFT(buf, sizeof(SMsgHead));
-  tCoderInit(&encoder, TD_LITTLE_ENDIAN, abuf, tlen, TD_ENCODER);
+  tCoderInit(&encoder, TD_LITTLE_ENDIAN, abuf, size, TD_ENCODER);
   tEncodeSStreamTask(&encoder, pTask);
   tCoderClear(&encoder);
 
@@ -70,7 +73,7 @@ int32_t mndAssignTaskToVg(SMnode* pMnode, STrans* pTrans, SStreamTask* pTask, SS
     terrno = TSDB_CODE_QRY_INVALID_INPUT;
     return -1;
   }
-  mndPersistTaskDeployReq(pTrans, pTask, &plan->execNode.epSet, TDMT_VND_TASK_DEPLOY);
+  mndPersistTaskDeployReq(pTrans, pTask, &plan->execNode.epSet, TDMT_VND_TASK_DEPLOY, pVgroup->vgId);
   return 0;
 }
 
@@ -90,7 +93,7 @@ int32_t mndAssignTaskToSnode(SMnode* pMnode, STrans* pTrans, SStreamTask* pTask,
     terrno = TSDB_CODE_QRY_INVALID_INPUT;
     return -1;
   }
-  mndPersistTaskDeployReq(pTrans, pTask, &plan->execNode.epSet, TDMT_SND_TASK_DEPLOY);
+  mndPersistTaskDeployReq(pTrans, pTask, &plan->execNode.epSet, TDMT_SND_TASK_DEPLOY, 0);
   return 0;
 }
 
@@ -106,6 +109,7 @@ int32_t mndScheduleStream(SMnode* pMnode, STrans* pTrans, SStreamObj* pStream) {
 
   int32_t totLevel = LIST_LENGTH(pPlan->pSubplans);
   pStream->tasks = taosArrayInit(totLevel, sizeof(SArray));
+  int32_t lastUsedVgId = 0;
 
   for (int32_t level = 0; level < totLevel; level++) {
     SArray*        taskOneLevel = taosArrayInit(0, sizeof(SStreamTask));
@@ -113,9 +117,9 @@ int32_t mndScheduleStream(SMnode* pMnode, STrans* pTrans, SStreamObj* pStream) {
     int32_t        opNum = LIST_LENGTH(inner->pNodeList);
     ASSERT(opNum == 1);
 
-    SSubplan* plan = nodesListGetNode(inner->pNodeList, level);
+    SSubplan* plan = nodesListGetNode(inner->pNodeList, 0);
     if (level == 0) {
-      ASSERT(plan->type == SUBPLAN_TYPE_SCAN);
+      ASSERT(plan->subplanType == SUBPLAN_TYPE_SCAN);
       void* pIter = NULL;
       while (1) {
         pIter = sdbFetch(pSdb, SDB_VGROUP, pIter, (void**)&pVgroup);
@@ -125,11 +129,14 @@ int32_t mndScheduleStream(SMnode* pMnode, STrans* pTrans, SStreamObj* pStream) {
           continue;
         }
 
+        lastUsedVgId = pVgroup->vgId;
         pStream->vgNum++;
         // send to vnode
 
         SStreamTask* pTask = streamTaskNew(pStream->uid, level);
+        pTask->pipeSource = 1;
         pTask->pipeSink = level == totLevel - 1 ? 1 : 0;
+        pTask->parallelizable = 1;
         // TODO: set to
         if (mndAssignTaskToVg(pMnode, pTrans, pTask, plan, pVgroup) < 0) {
           sdbRelease(pSdb, pVgroup);
@@ -140,7 +147,21 @@ int32_t mndScheduleStream(SMnode* pMnode, STrans* pTrans, SStreamObj* pStream) {
       }
     } else {
       SStreamTask* pTask = streamTaskNew(pStream->uid, level);
+      pTask->pipeSource = 0;
       pTask->pipeSink = level == totLevel - 1 ? 1 : 0;
+      pTask->parallelizable = plan->subplanType == SUBPLAN_TYPE_SCAN;
+      pTask->nextOpDst = STREAM_NEXT_OP_DST__VND;
+
+      if (tsStreamSchedV) {
+        ASSERT(lastUsedVgId != 0);
+        SVgObj* pVg = mndAcquireVgroup(pMnode, lastUsedVgId);
+        if (mndAssignTaskToVg(pMnode, pTrans, pTask, plan, pVg) < 0) {
+          sdbRelease(pSdb, pVg);
+          qDestroyQueryPlan(pPlan);
+          return -1;
+        }
+        sdbRelease(pSdb, pVg);
+      } else {
         SSnodeObj* pSnode = mndSchedFetchSnode(pMnode);
         if (pSnode != NULL) {
           if (mndAssignTaskToSnode(pMnode, pTrans, pTask, plan, pSnode) < 0) {
@@ -153,6 +174,8 @@ int32_t mndScheduleStream(SMnode* pMnode, STrans* pTrans, SStreamObj* pStream) {
           // TODO: assign to one vg
           ASSERT(0);
         }
+      }
+
       taosArrayPush(taskOneLevel, pTask);
     }
     taosArrayPush(pStream->tasks, taskOneLevel);

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

@@ -21,6 +21,7 @@
 #include "mndScheduler.h"
 #include "mndShow.h"
 #include "mndStb.h"
+#include "mndTopic.h"
 #include "mndTrans.h"
 #include "mndUser.h"
 #include "mndVgroup.h"
@@ -33,6 +34,7 @@ static int32_t mndStreamActionInsert(SSdb *pSdb, SStreamObj *pStream);
 static int32_t mndStreamActionDelete(SSdb *pSdb, SStreamObj *pStream);
 static int32_t mndStreamActionUpdate(SSdb *pSdb, SStreamObj *pStream, SStreamObj *pNewStream);
 static int32_t mndProcessCreateStreamReq(SNodeMsg *pReq);
+static int32_t mndProcessTaskDeployInternalRsp(SNodeMsg *pRsp);
 /*static int32_t mndProcessDropStreamReq(SNodeMsg *pReq);*/
 /*static int32_t mndProcessDropStreamInRsp(SNodeMsg *pRsp);*/
 static int32_t mndProcessStreamMetaReq(SNodeMsg *pReq);
@@ -50,6 +52,8 @@ int32_t mndInitStream(SMnode *pMnode) {
                      .deleteFp = (SdbDeleteFp)mndStreamActionDelete};
 
   mndSetMsgHandle(pMnode, TDMT_MND_CREATE_STREAM, mndProcessCreateStreamReq);
+  mndSetMsgHandle(pMnode, TDMT_VND_TASK_DEPLOY_RSP, mndProcessTaskDeployInternalRsp);
+  mndSetMsgHandle(pMnode, TDMT_SND_TASK_DEPLOY_RSP, mndProcessTaskDeployInternalRsp);
   /*mndSetMsgHandle(pMnode, TDMT_MND_DROP_STREAM, mndProcessDropStreamReq);*/
   /*mndSetMsgHandle(pMnode, TDMT_MND_DROP_STREAM_RSP, mndProcessDropStreamInRsp);*/
 
@@ -68,7 +72,7 @@ SSdbRaw *mndStreamActionEncode(SStreamObj *pStream) {
 
   SCoder encoder;
   tCoderInit(&encoder, TD_LITTLE_ENDIAN, NULL, 0, TD_ENCODER);
-  if (tEncodeSStreamObj(NULL, pStream) < 0) {
+  if (tEncodeSStreamObj(&encoder, pStream) < 0) {
     tCoderClear(&encoder);
     goto STREAM_ENCODE_OVER;
   }
@@ -83,7 +87,7 @@ SSdbRaw *mndStreamActionEncode(SStreamObj *pStream) {
   if (buf == NULL) goto STREAM_ENCODE_OVER;
 
   tCoderInit(&encoder, TD_LITTLE_ENDIAN, buf, tlen, TD_ENCODER);
-  if (tEncodeSStreamObj(NULL, pStream) < 0) {
+  if (tEncodeSStreamObj(&encoder, pStream) < 0) {
     tCoderClear(&encoder);
     goto STREAM_ENCODE_OVER;
   }
@@ -135,7 +139,7 @@ SSdbRow *mndStreamActionDecode(SSdbRaw *pRaw) {
   SDB_GET_BINARY(pRaw, dataPos, buf, tlen, STREAM_DECODE_OVER);
 
   SCoder decoder;
-  tCoderInit(&decoder, TD_LITTLE_ENDIAN, NULL, 0, TD_DECODER);
+  tCoderInit(&decoder, TD_LITTLE_ENDIAN, buf, tlen + 1, TD_DECODER);
   if (tDecodeSStreamObj(&decoder, pStream) < 0) {
     goto STREAM_DECODE_OVER;
   }
@@ -191,6 +195,11 @@ void mndReleaseStream(SMnode *pMnode, SStreamObj *pStream) {
   sdbRelease(pSdb, pStream);
 }
 
+static int32_t mndProcessTaskDeployInternalRsp(SNodeMsg *pRsp) {
+  mndTransProcessRsp(pRsp);
+  return 0;
+}
+
 static SDbObj *mndAcquireDbByStream(SMnode *pMnode, char *streamName) {
   SName name = {0};
   tNameFromString(&name, streamName, T_NAME_ACCT | T_NAME_DB | T_NAME_TABLE);
@@ -209,6 +218,55 @@ static int32_t mndCheckCreateStreamReq(SCMCreateStreamReq *pCreate) {
   return 0;
 }
 
+static SArray *mndExtractNamesFromAst(const SNode *pAst) {
+  if (pAst->type != QUERY_NODE_SELECT_STMT) return NULL;
+
+  SArray *names = taosArrayInit(0, sizeof(void *));
+  if (names == NULL) {
+    return NULL;
+  }
+  SSelectStmt *pSelect = (SSelectStmt *)pAst;
+  SNodeList   *pNodes = pSelect->pProjectionList;
+  SListCell   *pCell = pNodes->pHead;
+  while (pCell != NULL) {
+    if (pCell->pNode->type != QUERY_NODE_FUNCTION) {
+      continue;
+    }
+    SFunctionNode *pFunction = (SFunctionNode *)pCell->pNode;
+    char          *name = strdup(pFunction->node.aliasName);
+    taosArrayPush(names, &name);
+    pCell = pCell->pNext;
+  }
+  return names;
+}
+
+static int32_t mndStreamGetPlanString(const SCMCreateStreamReq *pCreate, char **pStr) {
+  if (NULL == pCreate->ast) {
+    return TSDB_CODE_SUCCESS;
+  }
+
+  SNode  *pAst = NULL;
+  int32_t code = nodesStringToNode(pCreate->ast, &pAst);
+
+  SQueryPlan *pPlan = NULL;
+  if (TSDB_CODE_SUCCESS == code) {
+    SPlanContext cxt = {
+        .pAstRoot = pAst,
+        .topicQuery = false,
+        .streamQuery = true,
+    };
+    code = qCreateQueryPlan(&cxt, &pPlan, NULL);
+  }
+
+  if (TSDB_CODE_SUCCESS == code) {
+    code = nodesNodeToString(pPlan, false, pStr, NULL);
+  }
+  nodesDestroyNode(pAst);
+  nodesDestroyNode(pPlan);
+  terrno = code;
+  return code;
+}
+
 static int32_t mndCreateStream(SMnode *pMnode, SNodeMsg *pReq, SCMCreateStreamReq *pCreate, SDbObj *pDb) {
   mDebug("stream:%s to create", pCreate->name);
   SStreamObj streamObj = {0};
@@ -220,8 +278,26 @@ static int32_t mndCreateStream(SMnode *pMnode, SNodeMsg *pReq, SCMCreateStreamRe
   streamObj.dbUid = pDb->uid;
   streamObj.version = 1;
   streamObj.sql = pCreate->sql;
-  streamObj.physicalPlan = "";
-  streamObj.logicalPlan = "";
+  /*streamObj.physicalPlan = "";*/
+  streamObj.logicalPlan = "not implemented";
+
+  SNode *pAst = NULL;
+  if (nodesStringToNode(pCreate->ast, &pAst) < 0) {
+    return -1;
+  }
+  SArray *names = mndExtractNamesFromAst(pAst);
+  printf("|");
+  for (int i = 0; i < taosArrayGetSize(names); i++) {
+    printf(" %15s |", (char *)taosArrayGetP(names, i));
+  }
+  printf("\n=======================================================\n");
+
+  streamObj.outputName = names;
+
+  if (TSDB_CODE_SUCCESS != mndStreamGetPlanString(pCreate, &streamObj.physicalPlan)) {
+    mError("topic:%s, failed to get plan since %s", pCreate->name, terrstr());
+    return -1;
+  }
 
   STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_ROLLBACK, TRN_TYPE_CREATE_STREAM, &pReq->rpcMsg);
   if (pTrans == NULL) {

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

@@ -236,7 +236,7 @@ static int32_t mndCheckCreateTopicReq(SCMCreateTopicReq *pCreate) {
   return 0;
 }
 
-static int32_t mndGetPlanString(SCMCreateTopicReq *pCreate, char **pStr) {
+static int32_t mndGetPlanString(const SCMCreateTopicReq *pCreate, char **pStr) {
   if (NULL == pCreate->ast) {
     return TSDB_CODE_SUCCESS;
   }

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

@@ -167,6 +167,7 @@ struct STQ {
   STqMetaStore* tqMeta;
   STqPushMgr*   tqPushMgr;
   SHashObj*     pStreamTasks;
+  SVnode*       pVnode;
   SWal*         pWal;
   SMeta*        pVnodeMeta;
 };

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

@@ -77,6 +77,4 @@ static FORCE_INLINE int tsdbUnLockSma(SSmaEnv *pEnv) {
   return 0;
 }
 
-
-
 #endif /* _TD_TSDB_SMA_H_ */
\ No newline at end of file

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

@@ -82,12 +82,6 @@ struct SVnode {
 
 int vnodeScheduleTask(SVnodeTask* task);
 
-int32_t vnodePutToVQueryQ(SVnode* pVnode, struct SRpcMsg* pReq);
-int32_t vnodePutToVFetchQ(SVnode* pVnode, struct SRpcMsg* pReq);
-int32_t vnodeSendReq(SVnode* pVnode, struct SEpSet* epSet, struct SRpcMsg* pReq);
-int32_t vnodeSendMnodeReq(SVnode* pVnode, struct SRpcMsg* pReq);
-void    vnodeSendRsp(SVnode* pVnode, struct SEpSet* epSet, struct SRpcMsg* pRsp);
-
 #define vFatal(...)                                              \
   do {                                                           \
     if (vDebugFlag & DEBUG_FATAL) {                              \
@@ -177,19 +171,20 @@ int  tqInit();
 void tqCleanUp();
 
 // open in each vnode
-STQ* tqOpen(const char* path, SWal* pWal, SMeta* pMeta, STqCfg* tqConfig, SMemAllocatorFactory* allocFac);
+STQ* tqOpen(const char* path, SVnode* pVnode, SWal* pWal, SMeta* pMeta, STqCfg* tqConfig,
+            SMemAllocatorFactory* allocFac);
 void tqClose(STQ*);
 
 // required by vnode
-int tqPushMsg(STQ*, void* msg, tmsg_t msgType, int64_t version);
+int tqPushMsg(STQ*, void* msg, int32_t msgLen, tmsg_t msgType, int64_t version);
 int tqCommit(STQ*);
 
 int32_t tqProcessPollReq(STQ* pTq, SRpcMsg* pMsg);
 int32_t tqProcessSetConnReq(STQ* pTq, char* msg);
 int32_t tqProcessRebReq(STQ* pTq, char* msg);
 int32_t tqProcessTaskExec(STQ* pTq, SRpcMsg* msg);
-
 int32_t tqProcessTaskDeploy(STQ* pTq, char* msg, int32_t msgLen);
+int32_t tqProcessStreamTrigger(STQ* pTq, void* data, int32_t dataLen);
 
 #ifdef __cplusplus
 }

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

@@ -17,11 +17,14 @@
 #include "tqInt.h"
 #include "tqMetaStore.h"
 
+void tqDebugShowSSData(SArray* dataBlocks);
+
 int32_t tqInit() { return tqPushMgrInit(); }
 
 void tqCleanUp() { tqPushMgrCleanUp(); }
 
-STQ* tqOpen(const char* path, SWal* pWal, SMeta* pVnodeMeta, STqCfg* tqConfig, SMemAllocatorFactory* allocFac) {
+STQ* tqOpen(const char* path, SVnode* pVnode, SWal* pWal, SMeta* pVnodeMeta, STqCfg* tqConfig,
+            SMemAllocatorFactory* allocFac) {
   STQ* pTq = malloc(sizeof(STQ));
   if (pTq == NULL) {
     terrno = TSDB_CODE_TQ_OUT_OF_MEMORY;
@@ -29,6 +32,7 @@ STQ* tqOpen(const char* path, SWal* pWal, SMeta* pVnodeMeta, STqCfg* tqConfig, S
   }
   pTq->path = strdup(path);
   pTq->tqConfig = tqConfig;
+  pTq->pVnode = pVnode;
   pTq->pWal = pWal;
   pTq->pVnodeMeta = pVnodeMeta;
 #if 0
@@ -68,46 +72,19 @@ void tqClose(STQ* pTq) {
   // TODO
 }
 
-int tqPushMsg(STQ* pTq, void* msg, tmsg_t msgType, int64_t version) {
+int tqPushMsg(STQ* pTq, void* msg, int32_t msgLen, tmsg_t msgType, int64_t version) {
   if (msgType != TDMT_VND_SUBMIT) return 0;
-
-  void* pIter = NULL;
-
-  while (1) {
-    pIter = taosHashIterate(pTq->pStreamTasks, pIter);
-    if (pIter == NULL) break;
-    SStreamTask* pTask = (SStreamTask*)pIter;
-    if (!pTask->pipeSource) continue;
-
-    int32_t workerId = 0;
-    void*   exec = pTask->runner[workerId].executor;
-    qSetStreamInput(exec, msg, STREAM_DATA_TYPE_SUBMIT_BLOCK);
-    SArray* pRes = taosArrayInit(0, sizeof(SSDataBlock));
-    while (1) {
-      SSDataBlock* output;
-      uint64_t     ts;
-      if (qExecTask(exec, &output, &ts) < 0) {
-        ASSERT(false);
-      }
-      if (output == NULL) {
-        break;
-      }
-      taosArrayPush(pRes, output);
-    }
-    if (pTask->pipeSink) {
-      // write back
-    } else {
-      int32_t tlen = sizeof(SStreamExecMsgHead) + tEncodeDataBlocks(NULL, pRes);
-      void*   buf = rpcMallocCont(tlen);
-      if (buf == NULL) {
+  void* data = malloc(msgLen);
+  if (data == NULL) {
     return -1;
   }
-      void* abuf = POINTER_SHIFT(buf, sizeof(SStreamExecMsgHead));
-      tEncodeDataBlocks(abuf, pRes);
-      // serialize
-      // to next level
-    }
-  }
+  memcpy(data, msg, msgLen);
+  SRpcMsg req = {
+      .msgType = TDMT_VND_STREAM_TRIGGER,
+      .pCont = data,
+      .contLen = msgLen,
+  };
+  tmsgPutToQueue(&pTq->pVnode->msgCb, FETCH_QUEUE, &req);
 
 #if 0
   void* pIter = taosHashIterate(pTq->tqPushMgr->pHash, NULL);
@@ -483,7 +460,9 @@ int32_t tqProcessTaskDeploy(STQ* pTq, char* msg, int32_t msgLen) {
   }
   SCoder decoder;
   tCoderInit(&decoder, TD_LITTLE_ENDIAN, (uint8_t*)msg, msgLen, TD_DECODER);
-  tDecodeSStreamTask(&decoder, pTask);
+  if (tDecodeSStreamTask(&decoder, pTask) < 0) {
+    ASSERT(0);
+  }
   tCoderClear(&decoder);
 
   tqExpandTask(pTq, pTask, 8);
@@ -560,7 +539,11 @@ void tqDebugShowSSData(SArray* dataBlocks) {
             break;
           case TSDB_DATA_TYPE_INT:
           case TSDB_DATA_TYPE_UINT:
-            printf(" %15u |", *(uint32_t*)var);
+            printf(" %15d |", *(int32_t*)var);
+            break;
+          case TSDB_DATA_TYPE_BIGINT:
+          case TSDB_DATA_TYPE_UBIGINT:
+            printf(" %15ld |", *(int64_t*)var);
             break;
         }
       }
@@ -569,6 +552,62 @@ void tqDebugShowSSData(SArray* dataBlocks) {
   }
 }
 
+int32_t tqProcessStreamTrigger(STQ* pTq, void* data, int32_t dataLen) {
+  void* pIter = NULL;
+
+  while (1) {
+    pIter = taosHashIterate(pTq->pStreamTasks, pIter);
+    if (pIter == NULL) break;
+    SStreamTask* pTask = (SStreamTask*)pIter;
+    if (!pTask->pipeSource) continue;
+
+    int32_t workerId = 0;
+    void*   exec = pTask->runner[workerId].executor;
+    qSetStreamInput(exec, data, STREAM_DATA_TYPE_SUBMIT_BLOCK);
+    SArray* pRes = taosArrayInit(0, sizeof(SSDataBlock));
+    while (1) {
+      SSDataBlock* output;
+      uint64_t     ts;
+      if (qExecTask(exec, &output, &ts) < 0) {
+        ASSERT(false);
+      }
+      if (output == NULL) {
+        break;
+      }
+      taosArrayPush(pRes, output);
+    }
+    if (pTask->pipeSink) {
+      // write back
+      /*printf("reach end\n");*/
+      tqDebugShowSSData(pRes);
+    } else {
+      int32_t tlen = sizeof(SStreamExecMsgHead) + tEncodeDataBlocks(NULL, pRes);
+      void*   buf = rpcMallocCont(tlen);
+      if (buf == NULL) {
+        return -1;
+      }
+      void* abuf = POINTER_SHIFT(buf, sizeof(SStreamExecMsgHead));
+      tEncodeDataBlocks(abuf, pRes);
+      tmsg_t type;
+
+      if (pTask->nextOpDst == STREAM_NEXT_OP_DST__VND) {
+        type = TDMT_VND_TASK_EXEC;
+      } else {
+        type = TDMT_SND_TASK_EXEC;
+      }
+
+      SRpcMsg reqMsg = {
+          .pCont = buf,
+          .contLen = tlen,
+          .code = 0,
+          .msgType = type,
+      };
+      tmsgSendReq(&pTq->pVnode->msgCb, &pTask->NextOpEp, &reqMsg);
+    }
+  }
+  return 0;
+}
+
 int32_t tqProcessTaskExec(STQ* pTq, SRpcMsg* msg) {
   SStreamTaskExecReq* pReq = msg->pCont;
 

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

@@ -126,6 +126,36 @@ SArray* tqRetrieveDataBlock(STqReadHandle* pHandle) {
   if (pArray == NULL) {
     return NULL;
   }
+  int32_t colMeta = 0;
+  int32_t colNeed = 0;
+  while (colMeta < pSchemaWrapper->nCols && colNeed < colNumNeed) {
+    SSchema* pColSchema = &pSchemaWrapper->pSchema[colMeta];
+    int16_t  colIdSchema = pColSchema->colId;
+    int16_t  colIdNeed = *(int16_t*)taosArrayGet(pHandle->pColIdList, colNeed);
+    if (colIdSchema < colIdNeed) {
+      colMeta++;
+    } else if (colIdSchema > colIdNeed) {
+      colNeed++;
+    } else {
+      SColumnInfoData colInfo = {0};
+      int             sz = numOfRows * pColSchema->bytes;
+      colInfo.info.bytes = pColSchema->bytes;
+      colInfo.info.colId = pColSchema->colId;
+      colInfo.info.type = pColSchema->type;
+
+      colInfo.pData = calloc(1, sz);
+      if (colInfo.pData == NULL) {
+        // TODO free
+        taosArrayDestroy(pArray);
+        return NULL;
+      }
+
+      blockDataEnsureColumnCapacity(&colInfo, numOfRows);
+      taosArrayPush(pArray, &colInfo);
+      colMeta++;
+      colNeed++;
+    }
+  }
 
   int j = 0;
   for (int32_t i = 0; i < colNumNeed; i++) {
@@ -163,11 +193,23 @@ SArray* tqRetrieveDataBlock(STqReadHandle* pHandle) {
   while ((row = tGetSubmitBlkNext(&pHandle->blkIter)) != NULL) {
     tdSTSRowIterReset(&iter, row);
     // get all wanted col of that block
+    int32_t colTot = taosArrayGetSize(pArray);
+    for (int32_t i = 0; i < colTot; i++) {
+      SColumnInfoData* pColData = taosArrayGet(pArray, i);
+      SCellVal         sVal = {0};
+      if (!tdSTSRowIterNext(&iter, pColData->info.colId, pColData->info.type, &sVal)) {
+        break;
+      }
+      memcpy(POINTER_SHIFT(pColData->pData, curRow * pColData->info.bytes), sVal.val, pColData->info.bytes);
+    }
+#if 0
     for (int32_t i = 0; i < colNumNeed; i++) {
       SColumnInfoData* pColData = taosArrayGet(pArray, i);
       STColumn*        pCol = schemaColAt(pTschema, i);
       // TODO
-      ASSERT(pCol->colId == pColData->info.colId);
+      if(pCol->colId != pColData->info.colId) {
+        continue;
+      }
       // void* val = tdGetMemRowDataOfColEx(row, pCol->colId, pCol->type, TD_DATA_ROW_HEAD_SIZE + pCol->offset, &kvIdx);
       SCellVal sVal = {0};
       if (!tdSTSRowIterNext(&iter, pCol->colId, pCol->type, &sVal)) {
@@ -176,6 +218,7 @@ SArray* tqRetrieveDataBlock(STqReadHandle* pHandle) {
       }
       memcpy(POINTER_SHIFT(pColData->pData, curRow * pCol->bytes), sVal.val, pCol->bytes);
     }
+#endif
     curRow++;
   }
   return pArray;

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

@@ -403,18 +403,12 @@ static STsdbReadHandle* tsdbQueryTablesImpl(STsdb* tsdb, STsdbQueryCond* pCond,
       SColumnInfoData colInfo = {{0}, 0};
       colInfo.info = pCond->colList[i];
 
-      colInfo.pData = calloc(1, EXTRA_BYTES + pReadHandle->outputCapacity * pCond->colList[i].bytes);
-      if (!IS_VAR_DATA_TYPE(colInfo.info.type)) {
-        colInfo.nullbitmap = calloc(1, BitmapLen(pReadHandle->outputCapacity));
-      }
-
-      if (colInfo.pData == NULL || (colInfo.nullbitmap == NULL && (!IS_VAR_DATA_TYPE(colInfo.info.type)))) {
+      int32_t code = blockDataEnsureColumnCapacity(&colInfo, pReadHandle->outputCapacity);
+      if (code != TSDB_CODE_SUCCESS) {
         goto _end;
       }
 
       taosArrayPush(pReadHandle->pColumns, &colInfo);
-
-
       pReadHandle->statis[i].colId = colInfo.info.colId;
     }
 
@@ -1418,34 +1412,37 @@ static int32_t doCopyRowsFromFileBlock(STsdbReadHandle* pTsdbReadHandle, int32_t
     }
 
     if (!isAllRowsNull(src) && pColInfo->info.colId == src->colId) {
-      if (pColInfo->info.type != TSDB_DATA_TYPE_BINARY && pColInfo->info.type != TSDB_DATA_TYPE_NCHAR) {
-        memmove(pData, (char*)src->pData + bytes * start, bytes * num);
-      } else {  // handle the var-string
-        char* dst = pData;
+      if (!IS_VAR_DATA_TYPE(pColInfo->info.type)) {  // todo opt performance
+//        memmove(pData, (char*)src->pData + bytes * start, bytes * num);
+        for(int32_t k = start; k < num + start; ++k) {
+          SCellVal sVal = {0};
+          if (tdGetColDataOfRow(&sVal, src, k) < 0) {
+            TASSERT(0);
+          }
 
+          if (sVal.valType == TD_VTYPE_NULL) {
+            colDataAppend(pColInfo, k, NULL, true);
+          } else {
+            colDataAppend(pColInfo, k, sVal.val, false);
+          }
+        }
+      } else {  // handle the var-string
         // todo refactor, only copy one-by-one
         for (int32_t k = start; k < num + start; ++k) {
           SCellVal sVal = {0};
           if(tdGetColDataOfRow(&sVal, src, k) < 0){
             TASSERT(0);
           }
-          memcpy(dst, sVal.val, varDataTLen(sVal.val));
-          dst += bytes;
+
+          colDataAppend(pColInfo, k, sVal.val, false);
         }
       }
 
       j++;
       i++;
     } else { // pColInfo->info.colId < src->colId, it is a NULL data
-      if (pColInfo->info.type == TSDB_DATA_TYPE_BINARY || pColInfo->info.type == TSDB_DATA_TYPE_NCHAR) {
-        char* dst = pData;
-
-        for(int32_t k = start; k < num + start; ++k) {
-          setVardataNull(dst, pColInfo->info.type);
-          dst += bytes;
-        }
-      } else {
-        setNullN(pData, pColInfo->info.type, pColInfo->info.bytes, num);
+      for(int32_t k = start; k < num + start; ++k) {  // TODO opt performance
+        colDataAppend(pColInfo, k, NULL, true);
       }
       i++;
     }
@@ -1453,23 +1450,9 @@ static int32_t doCopyRowsFromFileBlock(STsdbReadHandle* pTsdbReadHandle, int32_t
 
   while (i < requiredNumOfCols) { // the remain columns are all null data
     SColumnInfoData* pColInfo = taosArrayGet(pTsdbReadHandle->pColumns, i);
-    if (ASCENDING_TRAVERSE(pTsdbReadHandle->order)) {
-      pData = (char*)pColInfo->pData + numOfRows * pColInfo->info.bytes;
-    } else {
-      pData = (char*)pColInfo->pData + (capacity - numOfRows - num) * pColInfo->info.bytes;
-    }
-
-    if (pColInfo->info.type == TSDB_DATA_TYPE_BINARY || pColInfo->info.type == TSDB_DATA_TYPE_NCHAR) {
-      char* dst = pData;
-
     for(int32_t k = start; k < num + start; ++k) {
-        setVardataNull(dst, pColInfo->info.type);
-        dst += pColInfo->info.bytes;
+      colDataAppend(pColInfo, k, NULL, true); // TODO add a fast version to set a number of consecutive NULL value.
     }
-    } else {
-      setNullN(pData, pColInfo->info.type, pColInfo->info.bytes, num);
-    }
-
     i++;
   }
 

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

@@ -493,6 +493,15 @@ static int32_t tsdbResetExpiredWindow(STsdb *pTsdb, SSmaStat *pStat, int64_t ind
                skey, indexUid);
       return TSDB_CODE_FAILED;
     }
+    // TODO: use a standalone interface to received state upate notification from stream computing module.
+    /**
+     * @brief state
+     *  - When SMA env init in TSDB, its status is TSDB_SMA_STAT_OK.
+     *  - In startup phase of stream computing module, it should notify the SMA env in TSDB to expired if needed(e.g.
+     * when batch data caculation not finised)
+     *  - When TSDB_SMA_STAT_OK, the stream computing module should also notify that to the SMA env in TSDB.
+     */
+    pItem->state = TSDB_SMA_STAT_OK;
   } else {
     // error handling
     tsdbUnRefSmaStat(pTsdb, pStat);
@@ -796,6 +805,7 @@ static int32_t tsdbInsertTSmaDataImpl(STsdb *pTsdb, char *msg) {
   if (pStat && pStat->smaStatItems) {
     pItem = taosHashGet(pStat->smaStatItems, &indexUid, sizeof(indexUid));
   }
+
   if ((pItem == NULL) || ((pItem = *(SSmaStatItem **)pItem) == NULL) || tsdbSmaStatIsDropped(pItem)) {
     terrno = TSDB_CODE_TDB_INVALID_SMA_STAT;
     tsdbUnRefSmaStat(pTsdb, pStat);
@@ -864,14 +874,14 @@ static int32_t tsdbDropTSmaDataImpl(STsdb *pTsdb, int64_t indexUid) {
     if ((pItem != NULL) || ((pItem = *(SSmaStatItem **)pItem) != NULL)) {
       if (tsdbSmaStatIsDropped(pItem)) {
         tsdbDebug("vgId:%d tSma stat is already dropped for %" PRIi64, REPO_ID(pTsdb), indexUid);
-        return TSDB_CODE_TDB_INVALID_ACTION;  // TODO: duplicate drop msg should be intercept by mnode
+        return TSDB_CODE_TDB_INVALID_ACTION;  // TODO: duplicate drop msg would be intercepted by mnode
       }
 
       tsdbWLockSma(pEnv);
       if (tsdbSmaStatIsDropped(pItem)) {
         tsdbUnLockSma(pEnv);
         tsdbDebug("vgId:%d tSma stat is already dropped for %" PRIi64, REPO_ID(pTsdb), indexUid);
-        return TSDB_CODE_TDB_INVALID_ACTION;  // TODO: duplicate drop msg should be intercept by mnode
+        return TSDB_CODE_TDB_INVALID_ACTION;  // TODO: duplicate drop msg would be intercepted by mnode
       }
       tsdbSmaStatSetDropped(pItem);
       tsdbUnLockSma(pEnv);
@@ -893,6 +903,7 @@ static int32_t tsdbDropTSmaDataImpl(STsdb *pTsdb, int64_t indexUid) {
   }
   // clear sma data files
   // TODO: 
+
 }
 
 static int32_t tsdbSetRSmaDataFile(STSmaWriteH *pSmaH, STSmaDataWrapper *pData, int32_t fid) {
@@ -1091,7 +1102,7 @@ static int32_t tsdbGetTSmaDataImpl(STsdb *pTsdb, STSmaDataWrapper *pData, int64_
   if (!tsdbSmaStatIsOK(pItem, &smaStat)) {  // TODO: multiple check for large scale sma query
     tsdbUnRefSmaStat(pTsdb, SMA_ENV_STAT(pTsdb->pTSmaEnv));
     terrno = TSDB_CODE_TDB_INVALID_SMA_STAT;
-    tsdbDebug("vgId:%d getTSmaDataImpl failed from index %" PRIi64 " since %s %" PRIi8, REPO_ID(pTsdb), indexUid,
+    tsdbWarn("vgId:%d getTSmaDataImpl failed from index %" PRIi64 " since %s %" PRIi8, REPO_ID(pTsdb), indexUid,
              tstrerror(terrno), smaStat);
     return TSDB_CODE_FAILED;
   }

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

@@ -115,7 +115,8 @@ static int vnodeOpenImpl(SVnode *pVnode) {
 
   // Open tsdb
   sprintf(dir, "%s/tsdb", pVnode->path);
-  pVnode->pTsdb = tsdbOpen(dir, pVnode->vgId, &(pVnode->config.tsdbCfg), vBufPoolGetMAF(pVnode), pVnode->pMeta, pVnode->pTfs);
+  pVnode->pTsdb =
+      tsdbOpen(dir, pVnode->vgId, &(pVnode->config.tsdbCfg), vBufPoolGetMAF(pVnode), pVnode->pMeta, pVnode->pTfs);
   if (pVnode->pTsdb == NULL) {
     // TODO: handle error
     return -1;
@@ -131,7 +132,7 @@ static int vnodeOpenImpl(SVnode *pVnode) {
 
   // Open TQ
   sprintf(dir, "%s/tq", pVnode->path);
-  pVnode->pTq = tqOpen(dir, pVnode->pWal, pVnode->pMeta, &(pVnode->config.tqCfg), vBufPoolGetMAF(pVnode));
+  pVnode->pTq = tqOpen(dir, pVnode, pVnode->pWal, pVnode->pMeta, &(pVnode->config.tqCfg), vBufPoolGetMAF(pVnode));
   if (pVnode->pTq == NULL) {
     // TODO: handle error
     return -1;

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

@@ -68,6 +68,8 @@ int vnodeProcessFetchMsg(SVnode *pVnode, SRpcMsg *pMsg) {
       return tqProcessPollReq(pVnode->pTq, pMsg);
     case TDMT_VND_TASK_EXEC:
       return tqProcessTaskExec(pVnode->pTq, pMsg);
+    case TDMT_VND_STREAM_TRIGGER:
+      return tqProcessStreamTrigger(pVnode->pTq, pMsg->pCont, pMsg->contLen);
     case TDMT_VND_QUERY_HEARTBEAT:
       return qWorkerProcessHbMsg(pVnode, pVnode->pQuery, pMsg);
     default:
@@ -163,7 +165,6 @@ static int vnodeGetTableMeta(SVnode *pVnode, SRpcMsg *pMsg) {
     memcpy(POINTER_SHIFT(metaRsp.pSchemas, sizeof(SSchema) * pSW->nCols), pTagSchema, sizeof(SSchema) * nTagCols);
   }
 
-
 _exit:
 
   rspLen = tSerializeSTableMetaRsp(NULL, 0, &metaRsp);
@@ -179,7 +180,6 @@ _exit:
   }
   tSerializeSTableMetaRsp(pRsp, rspLen, &metaRsp);
 
-
   tFreeSTableMetaRsp(&metaRsp);
   if (pSW != NULL) {
     tfree(pSW->pSchema);

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

@@ -59,7 +59,7 @@ int vnodeApplyWMsg(SVnode *pVnode, SRpcMsg *pMsg, SRpcMsg **pRsp) {
   // todo: change the interface here
   int64_t ver;
   taosDecodeFixedI64(POINTER_SHIFT(pMsg->pCont, sizeof(SMsgHead)), &ver);
-  if (tqPushMsg(pVnode->pTq, ptr, pMsg->msgType, ver) < 0) {
+  if (tqPushMsg(pVnode->pTq, pMsg->pCont, pMsg->contLen, pMsg->msgType, ver) < 0) {
     // TODO: handle error
   }
 

source/libs/executor/inc/executorimpl.h

@@ -541,10 +541,20 @@ typedef struct SFillOperatorInfo {
   bool              multigroupResult;
 } SFillOperatorInfo;
 
+typedef struct SGroupKeys {
+  char   *pData;
+  bool    isNull;
+  int16_t type;
+  int32_t bytes;
+}SGroupKeys;
+
 typedef struct SGroupbyOperatorInfo {
   SOptrBasicInfo binfo;
   SArray*        pGroupCols;
-  char*          prevData;  // previous group by value
+  SArray*        pGroupColVals; // current group column values, SArray<SGroupKeys>
+  bool           isInit;       // denote if current val is initialized or not
+  char*          keyBuf;       // group by keys for hash
+  int32_t        groupKeyLen;  // total group by column width
   SGroupResInfo  groupResInfo;
   SAggSupporter  aggSup;
 } SGroupbyOperatorInfo;

source/libs/executor/src/executor.c

@@ -30,7 +30,7 @@ static int32_t doSetStreamBlock(SOperatorInfo* pOperator, void* input, int32_t t
       qError("join not supported for stream block scan, %s" PRIx64, id);
       return TSDB_CODE_QRY_APP_ERROR;
     }
-
+    pOperator->status = OP_NOT_OPENED;
     return doSetStreamBlock(pOperator->pDownstream[0], input, type, id);
   } else {
     SStreamBlockScanInfo* pInfo = pOperator->info;

source/libs/executor/src/executorimpl.c

@@ -1623,67 +1623,142 @@ static void hashAllIntervalAgg(SOperatorInfo* pOperatorInfo, SResultRowInfo* pRe
 //  updateResultRowInfoActiveIndex(pResultRowInfo, pQueryAttr, pRuntimeEnv->current->lastKey);
 }
 
-static void doHashGroupbyAgg(SOperatorInfo* pOperator, SGroupbyOperatorInfo *pInfo, SSDataBlock *pBlock) {
-  SExecTaskInfo *pTaskInfo = pOperator->pTaskInfo;
-
-  // TODO multiple group by columns
-  SColumn* pCol = taosArrayGet(pInfo->pGroupCols, 0);
+static bool groupKeyCompare(SGroupbyOperatorInfo* pInfo, SSDataBlock* pBlock, int32_t rowIndex, int32_t numOfGroupCols) {
+  SColumnDataAgg* pColAgg = NULL;
+  for (int32_t i = 0; i < numOfGroupCols; ++i) {
+    SColumn*         pCol = taosArrayGet(pInfo->pGroupCols, i);
     SColumnInfoData* pColInfoData = taosArrayGet(pBlock->pDataBlock, pCol->slotId);
-
-  int16_t bytes = pColInfoData->info.bytes;
-  int16_t type = pColInfoData->info.type;
-  if (type == TSDB_DATA_TYPE_FLOAT || type == TSDB_DATA_TYPE_DOUBLE) {
-    //qError("QInfo:0x%"PRIx64" group by not supported on double/float columns, abort", GET_TASKID(pRuntimeEnv));
-    return;
+    if (pBlock->pBlockAgg != NULL) {
+      pColAgg = &pBlock->pBlockAgg[pCol->slotId];  // TODO is agg data matched?
     }
 
-  STimeWindow w = TSWINDOW_INITIALIZER;
+    bool isNull = colDataIsNull(pColInfoData, pBlock->info.rows, rowIndex, pColAgg);
 
-  int32_t num = 0;
-  for (int32_t j = 0; j < pBlock->info.rows; ++j) {
-    if (colDataIsNull(pColInfoData, pBlock->info.rows, j, NULL)) { // TODO
+    SGroupKeys* pkey = taosArrayGet(pInfo->pGroupColVals, i);
+    if (pkey->isNull && isNull) {
       continue;
     }
 
-    char* val = colDataGetData(pColInfoData, j);
+    if (isNull || pkey->isNull) {
+      return false;
+    }
 
-    // Compare with the previous row of this column, and do not set the output buffer again if they are identical.
-    if (pInfo->prevData == NULL) {
-      pInfo->prevData = malloc(bytes);
-      memcpy(pInfo->prevData, val, bytes);
-      num++;
+    char* val = colDataGetData(pColInfoData, rowIndex);
+
+    if (IS_VAR_DATA_TYPE(pkey->type)) {
+      int32_t len = varDataLen(val);
+      if (len == varDataLen(pkey->pData) && memcmp(varDataVal(pkey->pData), varDataVal(val), len) == 0) {
         continue;
+      } else {
+        return false;
+      }
+    } else {
+      if (memcmp(pkey->pData, val, pkey->bytes) != 0) {
+        return false;
+      }
+    }
   }
 
-    if (IS_VAR_DATA_TYPE(type)) {
-      int32_t len = varDataLen(val);
-      if(len == varDataLen(pInfo->prevData) && memcmp(varDataVal(pInfo->prevData), varDataVal(val), len) == 0) {
-        num++;
+  return true;
+}
+
+static void keepGroupKeys(SGroupbyOperatorInfo* pInfo, SSDataBlock* pBlock, int32_t rowIndex, int32_t numOfGroupCols) {
+  SColumnDataAgg* pColAgg = NULL;
+
+  for (int32_t i = 0; i < numOfGroupCols; ++i) {
+    SColumn*         pCol = taosArrayGet(pInfo->pGroupCols, i);
+    SColumnInfoData* pColInfoData = taosArrayGet(pBlock->pDataBlock, pCol->slotId);
+
+    if (pBlock->pBlockAgg != NULL) {
+      pColAgg = &pBlock->pBlockAgg[pCol->slotId];  // TODO is agg data matched?
+    }
+
+    SGroupKeys* pkey = taosArrayGet(pInfo->pGroupColVals, i);
+    if (colDataIsNull(pColInfoData, pBlock->info.rows, rowIndex, pColAgg)) {
+      pkey->isNull = true;
+    } else {
+      char* val = colDataGetData(pColInfoData, rowIndex);
+      if (IS_VAR_DATA_TYPE(pkey->type)) {
+        memcpy(pkey->pData, val, varDataTLen(val));
+      } else {
+        memcpy(pkey->pData, val, pkey->bytes);
+      }
+    }
+  }
+}
+
+static int32_t generatedHashKey(void* pKey, int32_t* length, SArray* pGroupColVals) {
+  ASSERT(pKey != NULL);
+  size_t numOfGroupCols = taosArrayGetSize(pGroupColVals);
+
+  char* isNull = (char*) pKey;
+  char* pStart = (char*) pKey + sizeof(int8_t) * numOfGroupCols;
+  for(int32_t i = 0; i < numOfGroupCols; ++i) {
+    SGroupKeys* pkey = taosArrayGet(pGroupColVals, i);
+    if (pkey->isNull) {
+      isNull[i] = 1;
       continue;
     }
+
+    isNull[i] = 0;
+    if (IS_VAR_DATA_TYPE(pkey->type)) {
+      varDataCopy(pStart, pkey->pData);
+      pStart += varDataTLen(pkey->pData);
+      ASSERT(varDataTLen(pkey->pData) <= pkey->bytes);
     } else {
-      if (memcmp(pInfo->prevData, val, bytes) == 0) {
+      memcpy(pStart, pkey->pData, pkey->bytes);
+      pStart += pkey->bytes;
+    }
+  }
+
+  *length = (pStart - (char*) pKey);
+  return 0;
+}
+
+static void doHashGroupbyAgg(SOperatorInfo* pOperator, SSDataBlock *pBlock) {
+  SExecTaskInfo *pTaskInfo = pOperator->pTaskInfo;
+  SGroupbyOperatorInfo *pInfo = pOperator->info;
+
+  int32_t numOfGroupCols = taosArrayGetSize(pInfo->pGroupCols);
+//  if (type == TSDB_DATA_TYPE_FLOAT || type == TSDB_DATA_TYPE_DOUBLE) {
+    //qError("QInfo:0x%"PRIx64" group by not supported on double/float columns, abort", GET_TASKID(pRuntimeEnv));
+//    return;
+//  }
+
+  int32_t len = 0;
+  STimeWindow w = TSWINDOW_INITIALIZER;
+
+  int32_t num = 0;
+  for (int32_t j = 0; j < pBlock->info.rows; ++j) {
+    // Compare with the previous row of this column, and do not set the output buffer again if they are identical.
+    if (!pInfo->isInit) {
+      keepGroupKeys(pInfo, pBlock, j, numOfGroupCols);
+      pInfo->isInit = true;
       num++;
       continue;
     }
+
+    bool equal = groupKeyCompare(pInfo, pBlock, j, numOfGroupCols);
+    if (equal) {
+      num++;
+      continue;
     }
 
-    int32_t ret = setGroupResultOutputBuf_rv(&(pInfo->binfo), pOperator->numOfOutput, pInfo->prevData, type, bytes, 0,
+    /*int32_t ret = */generatedHashKey(pInfo->keyBuf, &len, pInfo->pGroupColVals);
+    int32_t ret = setGroupResultOutputBuf_rv(&(pInfo->binfo), pOperator->numOfOutput, pInfo->keyBuf, TSDB_DATA_TYPE_VARCHAR, len, 0,
                                              pInfo->aggSup.pResultBuf, pTaskInfo, &pInfo->aggSup);
     if (ret != TSDB_CODE_SUCCESS) {  // null data, too many state code
       longjmp(pTaskInfo->env, TSDB_CODE_QRY_APP_ERROR);
     }
 
     doApplyFunctions(pInfo->binfo.pCtx, &w, j - num, num, NULL, pBlock->info.rows, pOperator->numOfOutput, TSDB_ORDER_ASC);
-
+    keepGroupKeys(pInfo, pBlock, j, numOfGroupCols);
     num = 1;
-    memcpy(pInfo->prevData, val, bytes);
   }
 
   if (num > 0) {
-    char* val = ((char*)pColInfoData->pData) + bytes * (pBlock->info.rows - num);
-    memcpy(pInfo->prevData, val, bytes);
-    int32_t ret = setGroupResultOutputBuf_rv(&(pInfo->binfo), pOperator->numOfOutput, pInfo->prevData, type, bytes, 0,
+    /*int32_t ret = */generatedHashKey(pInfo->keyBuf, &len, pInfo->pGroupColVals);
+    int32_t ret = setGroupResultOutputBuf_rv(&(pInfo->binfo), pOperator->numOfOutput, pInfo->keyBuf, TSDB_DATA_TYPE_VARCHAR, len, 0,
                                              pInfo->aggSup.pResultBuf, pTaskInfo, &pInfo->aggSup);
     if (ret != TSDB_CODE_SUCCESS) {
       longjmp(pTaskInfo->env, TSDB_CODE_QRY_APP_ERROR);
@@ -1691,8 +1766,6 @@ static void doHashGroupbyAgg(SOperatorInfo* pOperator, SGroupbyOperatorInfo *pIn
 
     doApplyFunctions(pInfo->binfo.pCtx, &w, pBlock->info.rows - num, num, NULL, pBlock->info.rows, pOperator->numOfOutput, TSDB_ORDER_ASC);
   }
-
-  tfree(pInfo->prevData);
 }
 
 static void doSessionWindowAggImpl(SOperatorInfo* pOperator, SSessionAggOperatorInfo *pInfo, SSDataBlock *pSDataBlock) {
@@ -1783,20 +1856,11 @@ static void setResultRowKey(SResultRow* pResultRow, char* pData, int16_t type) {
 
 static int32_t setGroupResultOutputBuf_rv(SOptrBasicInfo *binfo, int32_t numOfCols, char *pData, int16_t type, int16_t bytes, int32_t groupId,
     SDiskbasedBuf* pBuf, SExecTaskInfo* pTaskInfo, SAggSupporter* pAggSup) {
-  int32_t        *rowCellInfoOffset = binfo->rowCellInfoOffset;
   SResultRowInfo *pResultRowInfo    = &binfo->resultRowInfo;
   SqlFunctionCtx *pCtx              = binfo->pCtx;
 
-  // not assign result buffer yet, add new result buffer, TODO remove it
-  char* d = pData;
-  int16_t len = bytes;
-  if (IS_VAR_DATA_TYPE(type)) {
-    d = varDataVal(pData);
-    len = varDataLen(pData);
-  }
-
-  int64_t tid = 0;
-  SResultRow *pResultRow = doSetResultOutBufByKey_rv(pBuf, pResultRowInfo, groupId, (char *)pData, TSDB_KEYSIZE, true, groupId, pTaskInfo, true, pAggSup);
+  SResultRow *pResultRow = doSetResultOutBufByKey_rv(pBuf, pResultRowInfo, groupId, (char *)pData, bytes, true, groupId,
+      pTaskInfo, true, pAggSup);
   assert (pResultRow != NULL);
 
   setResultRowKey(pResultRow, pData, type);
@@ -6969,7 +7033,7 @@ static SSDataBlock* hashGroupbyAggregate(SOperatorInfo *pOperator, bool* newgrou
     // the pDataBlock are always the same one, no need to call this again
     setInputDataBlock(pOperator, pInfo->binfo.pCtx, pBlock, order);
 //    setTagValue(pOperator, pRuntimeEnv->current->pTable, pInfo->binfo.pCtx, pOperator->numOfOutput);
-    doHashGroupbyAgg(pOperator, pInfo, pBlock);
+    doHashGroupbyAgg(pOperator, pBlock);
   }
 
   pOperator->status = OP_RES_TO_RETURN;
@@ -7281,7 +7345,9 @@ void destroySFillOperatorInfo(void* param, int32_t numOfOutput) {
 void destroyGroupbyOperatorInfo(void* param, int32_t numOfOutput) {
   SGroupbyOperatorInfo* pInfo = (SGroupbyOperatorInfo*) param;
   doDestroyBasicInfo(&pInfo->binfo, numOfOutput);
-  tfree(pInfo->prevData);
+  tfree(pInfo->keyBuf);
+  taosArrayDestroy(pInfo->pGroupCols);
+  taosArrayDestroy(pInfo->pGroupColVals);
 }
 
 static void destroyProjectOperatorInfo(void* param, int32_t numOfOutput) {
@@ -7664,6 +7730,39 @@ SOperatorInfo* createAllMultiTableTimeIntervalOperatorInfo(STaskRuntimeEnv* pRun
   return pOperator;
 }
 
+static int32_t initGroupOptrInfo(SGroupbyOperatorInfo *pInfo, SArray* pGroupColList) {
+  pInfo->pGroupColVals = taosArrayInit(4, sizeof(SGroupKeys));
+  if (pInfo->pGroupColVals == NULL) {
+    return TSDB_CODE_OUT_OF_MEMORY;
+  }
+
+  int32_t numOfGroupCols = taosArrayGetSize(pGroupColList);
+  for(int32_t i = 0; i < numOfGroupCols; ++i) {
+    SColumn* pCol = taosArrayGet(pGroupColList, i);
+    pInfo->groupKeyLen += pCol->bytes;
+
+    struct SGroupKeys key = {0};
+    key.bytes  = pCol->bytes;
+    key.type   = pCol->type;
+    key.isNull = false;
+    key.pData  = calloc(1, pCol->bytes);
+    if (key.pData == NULL) {
+      return TSDB_CODE_OUT_OF_MEMORY;
+    }
+
+    taosArrayPush(pInfo->pGroupColVals, &key);
+  }
+
+  int32_t nullFlagSize = sizeof(int8_t) * numOfGroupCols;
+  pInfo->keyBuf = calloc(1, pInfo->groupKeyLen + nullFlagSize);
+
+  if (pInfo->keyBuf == NULL) {
+    return TSDB_CODE_OUT_OF_MEMORY;
+  }
+
+  return TSDB_CODE_SUCCESS;
+}
+
 SOperatorInfo* createGroupOperatorInfo(SOperatorInfo* downstream, SExprInfo* pExprInfo, int32_t numOfCols, SSDataBlock* pResultBlock,
                                        SArray* pGroupColList, SExecTaskInfo* pTaskInfo, const STableGroupInfo* pTableGroupInfo) {
   SGroupbyOperatorInfo* pInfo = calloc(1, sizeof(SGroupbyOperatorInfo));
@@ -7674,9 +7773,13 @@ SOperatorInfo* createGroupOperatorInfo(SOperatorInfo* downstream, SExprInfo* pEx
 
   pInfo->pGroupCols = pGroupColList;
   initAggInfo(&pInfo->binfo, &pInfo->aggSup, pExprInfo, numOfCols, 4096, pResultBlock, pTaskInfo->id.str);
-
   initResultRowInfo(&pInfo->binfo.resultRowInfo, 8);
 
+  int32_t code = initGroupOptrInfo(pInfo, pGroupColList);
+  if (code != TSDB_CODE_SUCCESS) {
+    goto _error;
+  }
+
   pOperator->name         = "GroupbyAggOperator";
   pOperator->blockingOptr = true;
   pOperator->status       = OP_NOT_OPENED;
@@ -7688,7 +7791,7 @@ SOperatorInfo* createGroupOperatorInfo(SOperatorInfo* downstream, SExprInfo* pEx
   pOperator->getNextFn    = hashGroupbyAggregate;
   pOperator->closeFn      = destroyGroupbyOperatorInfo;
 
-  int32_t code = appendDownstream(pOperator, &downstream, 1);
+  code = appendDownstream(pOperator, &downstream, 1);
   return pOperator;
 
   _error:
@@ -8345,7 +8448,9 @@ SOperatorInfo* doCreateOperatorTreeNode(SPhysiNode* pPhyNode, SExecTaskInfo* pTa
       SExprInfo* pExprInfo = createExprInfo(pIntervalPhyNode->pFuncs, NULL, &num);
       SSDataBlock* pResBlock = createOutputBuf_rv1(pPhyNode->pOutputDataBlockDesc);
 
-      SInterval interval = {.interval = pIntervalPhyNode->interval, .sliding = pIntervalPhyNode->sliding, .intervalUnit = 'a', .slidingUnit = 'a'};
+      SInterval interval = {.interval = pIntervalPhyNode->interval, .sliding = pIntervalPhyNode->sliding,
+                            .intervalUnit = pIntervalPhyNode->intervalUnit,
+                            .slidingUnit = pIntervalPhyNode->slidingUnit, .offset = pIntervalPhyNode->offset};
       return createIntervalOperatorInfo(op, pExprInfo, num, pResBlock, &interval, pTableGroupInfo, pTaskInfo);
     }
   } /*else if (pPhyNode->info.type == OP_MultiTableAggregate) {

source/libs/sync/inc/syncRaftLog.h

@@ -47,6 +47,8 @@ SyncIndex       logStoreGetCommitIndex(SSyncLogStore* pLogStore);
 SSyncRaftEntry* logStoreGetLastEntry(SSyncLogStore* pLogStore);
 cJSON*          logStore2Json(SSyncLogStore* pLogStore);
 char*           logStore2Str(SSyncLogStore* pLogStore);
+cJSON*          logStoreSimple2Json(SSyncLogStore* pLogStore);
+char*           logStoreSimple2Str(SSyncLogStore* pLogStore);
 
 // for debug
 void logStorePrint(SSyncLogStore* pLogStore);
@@ -54,6 +56,11 @@ void logStorePrint2(char* s, SSyncLogStore* pLogStore);
 void logStoreLog(SSyncLogStore* pLogStore);
 void logStoreLog2(char* s, SSyncLogStore* pLogStore);
 
+void logStoreSimplePrint(SSyncLogStore* pLogStore);
+void logStoreSimplePrint2(char* s, SSyncLogStore* pLogStore);
+void logStoreSimpleLog(SSyncLogStore* pLogStore);
+void logStoreSimpleLog2(char* s, SSyncLogStore* pLogStore);
+
 #ifdef __cplusplus
 }
 #endif

source/libs/sync/src/syncAppendEntries.c

@@ -120,6 +120,11 @@ int32_t syncNodeOnAppendEntriesCb(SSyncNode* ths, SyncAppendEntries* pMsg) {
   // reject request
   if ((pMsg->term < ths->pRaftStore->currentTerm) ||
       ((pMsg->term == ths->pRaftStore->currentTerm) && (ths->state == TAOS_SYNC_STATE_FOLLOWER) && !logOK)) {
+    sTrace(
+        "syncNodeOnAppendEntriesCb --> reject, pMsg->term:%lu, ths->pRaftStore->currentTerm:%lu, ths->state:%d, "
+        "logOK:%d",
+        pMsg->term, ths->pRaftStore->currentTerm, ths->state, logOK);
+
     SyncAppendEntriesReply* pReply = syncAppendEntriesReplyBuild();
     pReply->srcId = ths->myRaftId;
     pReply->destId = pMsg->srcId;
@@ -137,6 +142,11 @@ int32_t syncNodeOnAppendEntriesCb(SSyncNode* ths, SyncAppendEntries* pMsg) {
 
   // return to follower state
   if (pMsg->term == ths->pRaftStore->currentTerm && ths->state == TAOS_SYNC_STATE_CANDIDATE) {
+    sTrace(
+        "syncNodeOnAppendEntriesCb --> return to follower, pMsg->term:%lu, ths->pRaftStore->currentTerm:%lu, "
+        "ths->state:%d, logOK:%d",
+        pMsg->term, ths->pRaftStore->currentTerm, ths->state, logOK);
+
     syncNodeBecomeFollower(ths);
 
     // ret or reply?
@@ -145,29 +155,22 @@ int32_t syncNodeOnAppendEntriesCb(SSyncNode* ths, SyncAppendEntries* pMsg) {
 
   // accept request
   if (pMsg->term == ths->pRaftStore->currentTerm && ths->state == TAOS_SYNC_STATE_FOLLOWER && logOK) {
-    bool preMatch = false;
-    if (pMsg->prevLogIndex == SYNC_INDEX_INVALID &&
-        ths->pLogStore->getLastIndex(ths->pLogStore) == SYNC_INDEX_INVALID) {
-      preMatch = true;
-    }
-    if (pMsg->prevLogIndex >= SYNC_INDEX_BEGIN && pMsg->prevLogIndex <= ths->pLogStore->getLastIndex(ths->pLogStore)) {
-      SSyncRaftEntry* pPreEntry = logStoreGetEntry(ths->pLogStore, pMsg->prevLogIndex);
-      assert(pPreEntry != NULL);
-      if (pMsg->prevLogTerm == pPreEntry->term) {
-        preMatch = true;
-      }
-      syncEntryDestory(pPreEntry);
-    }
-
-    if (preMatch) {
-      // must has preIndex in local log
+    // preIndex = -1, or has preIndex entry in local log
     assert(pMsg->prevLogIndex <= ths->pLogStore->getLastIndex(ths->pLogStore));
 
+    // has extra entries (> preIndex) in local log
     bool hasExtraEntries = pMsg->prevLogIndex < ths->pLogStore->getLastIndex(ths->pLogStore);
+
+    // has entries in SyncAppendEntries msg
     bool hasAppendEntries = pMsg->dataLen > 0;
 
+    sTrace(
+        "syncNodeOnAppendEntriesCb --> accept, pMsg->term:%lu, ths->pRaftStore->currentTerm:%lu, ths->state:%d, "
+        "logOK:%d, hasExtraEntries:%d, hasAppendEntries:%d",
+        pMsg->term, ths->pRaftStore->currentTerm, ths->state, logOK, hasExtraEntries, hasAppendEntries);
+
     if (hasExtraEntries && hasAppendEntries) {
-        // conflict
+      // not conflict by default
       bool conflict = false;
 
       SyncIndex       extraIndex = pMsg->prevLogIndex + 1;
@@ -177,8 +180,9 @@ int32_t syncNodeOnAppendEntriesCb(SSyncNode* ths, SyncAppendEntries* pMsg) {
       SSyncRaftEntry* pAppendEntry = syncEntryDeserialize(pMsg->data, pMsg->dataLen);
       assert(pAppendEntry != NULL);
 
+      // log not match, conflict
       assert(extraIndex == pAppendEntry->index);
-        if (pExtraEntry->term == pAppendEntry->term) {
+      if (pExtraEntry->term != pAppendEntry->term) {
         conflict = true;
       }
 
@@ -187,6 +191,8 @@ int32_t syncNodeOnAppendEntriesCb(SSyncNode* ths, SyncAppendEntries* pMsg) {
         SyncIndex delBegin = ths->pLogStore->getLastIndex(ths->pLogStore);
         SyncIndex delEnd = extraIndex;
 
+        sTrace("syncNodeOnAppendEntriesCb --> conflict:%d, delBegin:%ld, delEnd:%ld", conflict, delBegin, delEnd);
+
         // notice! reverse roll back!
         for (SyncIndex index = delEnd; index >= delBegin; --index) {
           if (ths->pFsm->FpRollBackCb != NULL) {
@@ -195,7 +201,7 @@ int32_t syncNodeOnAppendEntriesCb(SSyncNode* ths, SyncAppendEntries* pMsg) {
 
             SRpcMsg rpcMsg;
             syncEntry2OriginalRpc(pRollBackEntry, &rpcMsg);
-              ths->pFsm->FpRollBackCb(ths->pFsm, &rpcMsg, pRollBackEntry->index, pRollBackEntry->isWeak, 0);
+            ths->pFsm->FpRollBackCb(ths->pFsm, &rpcMsg, pRollBackEntry->index, pRollBackEntry->isWeak, 0, ths->state);
             rpcFreeCont(rpcMsg.pCont);
             syncEntryDestory(pRollBackEntry);
           }
@@ -212,7 +218,7 @@ int32_t syncNodeOnAppendEntriesCb(SSyncNode* ths, SyncAppendEntries* pMsg) {
         syncEntry2OriginalRpc(pAppendEntry, &rpcMsg);
         if (ths->pFsm != NULL) {
           if (ths->pFsm->FpPreCommitCb != NULL) {
-              ths->pFsm->FpPreCommitCb(ths->pFsm, &rpcMsg, pAppendEntry->index, pAppendEntry->isWeak, 0);
+            ths->pFsm->FpPreCommitCb(ths->pFsm, &rpcMsg, pAppendEntry->index, pAppendEntry->isWeak, 2, ths->state);
           }
         }
         rpcFreeCont(rpcMsg.pCont);
@@ -237,7 +243,7 @@ int32_t syncNodeOnAppendEntriesCb(SSyncNode* ths, SyncAppendEntries* pMsg) {
       syncEntry2OriginalRpc(pAppendEntry, &rpcMsg);
       if (ths->pFsm != NULL) {
         if (ths->pFsm->FpPreCommitCb != NULL) {
-            ths->pFsm->FpPreCommitCb(ths->pFsm, &rpcMsg, pAppendEntry->index, pAppendEntry->isWeak, 0);
+          ths->pFsm->FpPreCommitCb(ths->pFsm, &rpcMsg, pAppendEntry->index, pAppendEntry->isWeak, 3, ths->state);
         }
       }
       rpcFreeCont(rpcMsg.pCont);
@@ -267,23 +273,8 @@ int32_t syncNodeOnAppendEntriesCb(SSyncNode* ths, SyncAppendEntries* pMsg) {
     SRpcMsg rpcMsg;
     syncAppendEntriesReply2RpcMsg(pReply, &rpcMsg);
     syncNodeSendMsgById(&pReply->destId, ths, &rpcMsg);
-
     syncAppendEntriesReplyDestroy(pReply);
 
-    } else {
-      SyncAppendEntriesReply* pReply = syncAppendEntriesReplyBuild();
-      pReply->srcId = ths->myRaftId;
-      pReply->destId = pMsg->srcId;
-      pReply->term = ths->pRaftStore->currentTerm;
-      pReply->success = false;
-      pReply->matchIndex = SYNC_INDEX_INVALID;
-
-      SRpcMsg rpcMsg;
-      syncAppendEntriesReply2RpcMsg(pReply, &rpcMsg);
-      syncNodeSendMsgById(&pReply->destId, ths, &rpcMsg);
-      syncAppendEntriesReplyDestroy(pReply);
-    }
-
     // maybe update commit index from leader
     if (pMsg->commitIndex > ths->commitIndex) {
       // has commit entry in local
@@ -308,7 +299,7 @@ int32_t syncNodeOnAppendEntriesCb(SSyncNode* ths, SyncAppendEntries* pMsg) {
               syncEntry2OriginalRpc(pEntry, &rpcMsg);
 
               if (ths->pFsm->FpCommitCb != NULL) {
-                ths->pFsm->FpCommitCb(ths->pFsm, &rpcMsg, pEntry->index, pEntry->isWeak, 0);
+                ths->pFsm->FpCommitCb(ths->pFsm, &rpcMsg, pEntry->index, pEntry->isWeak, 0, ths->state);
               }
 
               rpcFreeCont(rpcMsg.pCont);

source/libs/sync/src/syncAppendEntriesReply.c

@@ -48,6 +48,9 @@ int32_t syncNodeOnAppendEntriesReplyCb(SSyncNode* ths, SyncAppendEntriesReply* p
     return ret;
   }
 
+  syncIndexMgrLog2("==syncNodeOnAppendEntriesReplyCb== before pNextIndex", ths->pNextIndex);
+  syncIndexMgrLog2("==syncNodeOnAppendEntriesReplyCb== before pMatchIndex", ths->pMatchIndex);
+
   // no need this code, because if I receive reply.term, then I must have sent for that term.
   //  if (pMsg->term > ths->pRaftStore->currentTerm) {
   //    syncNodeUpdateTerm(ths, pMsg->term);
@@ -77,5 +80,8 @@ int32_t syncNodeOnAppendEntriesReplyCb(SSyncNode* ths, SyncAppendEntriesReply* p
     syncIndexMgrSetIndex(ths->pNextIndex, &(pMsg->srcId), nextIndex);
   }
 
+  syncIndexMgrLog2("==syncNodeOnAppendEntriesReplyCb== after pNextIndex", ths->pNextIndex);
+  syncIndexMgrLog2("==syncNodeOnAppendEntriesReplyCb== after pMatchIndex", ths->pMatchIndex);
+
   return ret;
 }

source/libs/sync/src/syncCommit.c

@@ -63,7 +63,14 @@ void syncMaybeAdvanceCommitIndex(SSyncNode* pSyncNode) {
       if (pEntry->term == pSyncNode->pRaftStore->currentTerm) {
         // update commit index
         newCommitIndex = index;
+        sTrace("syncMaybeAdvanceCommitIndex maybe to update, newCommitIndex:%ld commit, pSyncNode->commitIndex:%ld",
+               newCommitIndex, pSyncNode->commitIndex);
         break;
+      } else {
+        sTrace(
+            "syncMaybeAdvanceCommitIndex can not commit due to term not equal, pEntry->term:%lu, "
+            "pSyncNode->pRaftStore->currentTerm:%lu",
+            pEntry->term, pSyncNode->pRaftStore->currentTerm);
       }
     }
   }
@@ -91,7 +98,7 @@ void syncMaybeAdvanceCommitIndex(SSyncNode* pSyncNode) {
           syncEntry2OriginalRpc(pEntry, &rpcMsg);
 
           if (pSyncNode->pFsm->FpCommitCb != NULL) {
-            pSyncNode->pFsm->FpCommitCb(pSyncNode->pFsm, &rpcMsg, pEntry->index, pEntry->isWeak, 0);
+            pSyncNode->pFsm->FpCommitCb(pSyncNode->pFsm, &rpcMsg, pEntry->index, pEntry->isWeak, 0, pSyncNode->state);
           }
 
           rpcFreeCont(rpcMsg.pCont);

source/libs/sync/src/syncIO.c

@@ -257,13 +257,6 @@ static void *syncIOConsumerFunc(void *param) {
           assert(pSyncMsg != NULL);
           io->FpOnSyncPing(io->pSyncNode, pSyncMsg);
           syncPingDestroy(pSyncMsg);
-          /*
-          pSyncMsg = syncPingBuild(pRpcMsg->contLen);
-          syncPingFromRpcMsg(pRpcMsg, pSyncMsg);
-          // memcpy(pSyncMsg, tmpRpcMsg.pCont, tmpRpcMsg.contLen);
-          io->FpOnSyncPing(io->pSyncNode, pSyncMsg);
-          syncPingDestroy(pSyncMsg);
-          */
         }
 
       } else if (pRpcMsg->msgType == SYNC_PING_REPLY) {

source/libs/sync/src/syncMain.c

@@ -104,9 +104,26 @@ int32_t syncReconfig(int64_t rid, const SSyncCfg* pSyncCfg) {
 }
 
 int32_t syncPropose(int64_t rid, const SRpcMsg* pMsg, bool isWeak) {
-  int32_t ret = 0;
+  int32_t ret = syncPropose2(rid, pMsg, isWeak, 0);
+  return ret;
+}
+
+int32_t syncForwardToPeer(int64_t rid, const SRpcMsg* pMsg, bool isWeak) {
+  int32_t ret = syncPropose(rid, pMsg, isWeak);
+  return ret;
+}
 
-  // todo : get pointer from rid
+ESyncState syncGetMyRole(int64_t rid) {
+  SSyncNode* pSyncNode = (SSyncNode*)taosAcquireRef(tsNodeRefId, rid);
+  if (pSyncNode == NULL) {
+    return TAOS_SYNC_STATE_ERROR;
+  }
+  assert(rid == pSyncNode->rid);
+  return pSyncNode->state;
+}
+
+int32_t syncPropose2(int64_t rid, const SRpcMsg* pMsg, bool isWeak, uint64_t seqNum) {
+  int32_t    ret = 0;
   SSyncNode* pSyncNode = (SSyncNode*)taosAcquireRef(tsNodeRefId, rid);
   if (pSyncNode == NULL) {
     return -1;
@@ -114,7 +131,7 @@ int32_t syncPropose(int64_t rid, const SRpcMsg* pMsg, bool isWeak) {
   assert(rid == pSyncNode->rid);
 
   if (pSyncNode->state == TAOS_SYNC_STATE_LEADER) {
-    SyncClientRequest* pSyncMsg = syncClientRequestBuild2(pMsg, 0, isWeak);
+    SyncClientRequest* pSyncMsg = syncClientRequestBuild2(pMsg, seqNum, isWeak);
     SRpcMsg            rpcMsg;
     syncClientRequest2RpcMsg(pSyncMsg, &rpcMsg);
     pSyncNode->FpEqMsg(pSyncNode->queue, &rpcMsg);
@@ -122,7 +139,7 @@ int32_t syncPropose(int64_t rid, const SRpcMsg* pMsg, bool isWeak) {
     ret = 0;
 
   } else {
-    sTrace("syncForwardToPeer not leader, %s", syncUtilState2String(pSyncNode->state));
+    sTrace("syncPropose not leader, %s", syncUtilState2String(pSyncNode->state));
     ret = -1;  // todo : need define err code !!
   }
 
@@ -130,19 +147,6 @@ int32_t syncPropose(int64_t rid, const SRpcMsg* pMsg, bool isWeak) {
   return ret;
 }
 
-int32_t syncForwardToPeer(int64_t rid, const SRpcMsg* pMsg, bool isWeak) {
-  int32_t ret = syncPropose(rid, pMsg, isWeak);
-  return ret;
-}
-
-ESyncState syncGetMyRole(int64_t rid) {
-  // todo : get pointer from rid
-  SSyncNode* pSyncNode = NULL;
-  return pSyncNode->state;
-}
-
-void syncGetNodesRole(int64_t rid, SNodesRole* pNodeRole) {}
-
 // open/close --------------
 SSyncNode* syncNodeOpen(const SSyncInfo* pSyncInfo) {
   SSyncNode* pSyncNode = (SSyncNode*)malloc(sizeof(SSyncNode));
@@ -848,7 +852,7 @@ static int32_t syncNodeOnClientRequestCb(SSyncNode* ths, SyncClientRequest* pMsg
 
     if (ths->pFsm != NULL) {
       if (ths->pFsm->FpPreCommitCb != NULL) {
-        ths->pFsm->FpPreCommitCb(ths->pFsm, &rpcMsg, pEntry->index, pEntry->isWeak, 0);
+        ths->pFsm->FpPreCommitCb(ths->pFsm, &rpcMsg, pEntry->index, pEntry->isWeak, 0, ths->state);
       }
     }
     rpcFreeCont(rpcMsg.pCont);
@@ -863,7 +867,7 @@ static int32_t syncNodeOnClientRequestCb(SSyncNode* ths, SyncClientRequest* pMsg
 
     if (ths->pFsm != NULL) {
       if (ths->pFsm->FpPreCommitCb != NULL) {
-        ths->pFsm->FpPreCommitCb(ths->pFsm, &rpcMsg, pEntry->index, pEntry->isWeak, -2);
+        ths->pFsm->FpPreCommitCb(ths->pFsm, &rpcMsg, pEntry->index, pEntry->isWeak, 1, ths->state);
       }
     }
     rpcFreeCont(rpcMsg.pCont);

source/libs/sync/src/syncMessage.c

@@ -834,7 +834,7 @@ cJSON* syncRequestVote2Json(const SyncRequestVote* pMsg) {
 
     snprintf(u64buf, sizeof(u64buf), "%lu", pMsg->term);
     cJSON_AddStringToObject(pRoot, "term", u64buf);
-    snprintf(u64buf, sizeof(u64buf), "%lu", pMsg->lastLogIndex);
+    snprintf(u64buf, sizeof(u64buf), "%ld", pMsg->lastLogIndex);
     cJSON_AddStringToObject(pRoot, "lastLogIndex", u64buf);
     snprintf(u64buf, sizeof(u64buf), "%lu", pMsg->lastLogTerm);
     cJSON_AddStringToObject(pRoot, "lastLogTerm", u64buf);
@@ -1127,14 +1127,14 @@ cJSON* syncAppendEntries2Json(const SyncAppendEntries* pMsg) {
     snprintf(u64buf, sizeof(u64buf), "%lu", pMsg->term);
     cJSON_AddStringToObject(pRoot, "term", u64buf);
 
-    snprintf(u64buf, sizeof(u64buf), "%lu", pMsg->prevLogIndex);
-    cJSON_AddStringToObject(pRoot, "pre_log_index", u64buf);
+    snprintf(u64buf, sizeof(u64buf), "%ld", pMsg->prevLogIndex);
+    cJSON_AddStringToObject(pRoot, "prevLogIndex", u64buf);
 
     snprintf(u64buf, sizeof(u64buf), "%lu", pMsg->prevLogTerm);
     cJSON_AddStringToObject(pRoot, "pre_log_term", u64buf);
 
-    snprintf(u64buf, sizeof(u64buf), "%lu", pMsg->commitIndex);
-    cJSON_AddStringToObject(pRoot, "commit_index", u64buf);
+    snprintf(u64buf, sizeof(u64buf), "%ld", pMsg->commitIndex);
+    cJSON_AddStringToObject(pRoot, "commitIndex", u64buf);
 
     cJSON_AddNumberToObject(pRoot, "dataLen", pMsg->dataLen);
     char* s;
@@ -1288,7 +1288,7 @@ cJSON* syncAppendEntriesReply2Json(const SyncAppendEntriesReply* pMsg) {
     snprintf(u64buf, sizeof(u64buf), "%lu", pMsg->term);
     cJSON_AddStringToObject(pRoot, "term", u64buf);
     cJSON_AddNumberToObject(pRoot, "success", pMsg->success);
-    snprintf(u64buf, sizeof(u64buf), "%lu", pMsg->matchIndex);
+    snprintf(u64buf, sizeof(u64buf), "%ld", pMsg->matchIndex);
     cJSON_AddStringToObject(pRoot, "matchIndex", u64buf);
   }
 

source/libs/sync/src/syncRaftLog.c

@@ -165,6 +165,34 @@ char* logStore2Str(SSyncLogStore* pLogStore) {
   return serialized;
 }
 
+cJSON* logStoreSimple2Json(SSyncLogStore* pLogStore) {
+  char               u64buf[128];
+  SSyncLogStoreData* pData = (SSyncLogStoreData*)pLogStore->data;
+  cJSON*             pRoot = cJSON_CreateObject();
+
+  if (pData != NULL && pData->pWal != NULL) {
+    snprintf(u64buf, sizeof(u64buf), "%p", pData->pSyncNode);
+    cJSON_AddStringToObject(pRoot, "pSyncNode", u64buf);
+    snprintf(u64buf, sizeof(u64buf), "%p", pData->pWal);
+    cJSON_AddStringToObject(pRoot, "pWal", u64buf);
+    snprintf(u64buf, sizeof(u64buf), "%ld", logStoreLastIndex(pLogStore));
+    cJSON_AddStringToObject(pRoot, "LastIndex", u64buf);
+    snprintf(u64buf, sizeof(u64buf), "%lu", logStoreLastTerm(pLogStore));
+    cJSON_AddStringToObject(pRoot, "LastTerm", u64buf);
+  }
+
+  cJSON* pJson = cJSON_CreateObject();
+  cJSON_AddItemToObject(pJson, "SSyncLogStoreSimple", pRoot);
+  return pJson;
+}
+
+char* logStoreSimple2Str(SSyncLogStore* pLogStore) {
+  cJSON* pJson = logStoreSimple2Json(pLogStore);
+  char*  serialized = cJSON_Print(pJson);
+  cJSON_Delete(pJson);
+  return serialized;
+}
+
 // for debug -----------------
 void logStorePrint(SSyncLogStore* pLogStore) {
   char* serialized = logStore2Str(pLogStore);
@@ -191,3 +219,30 @@ void logStoreLog2(char* s, SSyncLogStore* pLogStore) {
   sTrace("logStorePrint | len:%lu | %s | %s", strlen(serialized), s, serialized);
   free(serialized);
 }
+
+// for debug -----------------
+void logStoreSimplePrint(SSyncLogStore* pLogStore) {
+  char* serialized = logStoreSimple2Str(pLogStore);
+  printf("logStoreSimplePrint | len:%lu | %s \n", strlen(serialized), serialized);
+  fflush(NULL);
+  free(serialized);
+}
+
+void logStoreSimplePrint2(char* s, SSyncLogStore* pLogStore) {
+  char* serialized = logStoreSimple2Str(pLogStore);
+  printf("logStoreSimplePrint2 | len:%lu | %s | %s \n", strlen(serialized), s, serialized);
+  fflush(NULL);
+  free(serialized);
+}
+
+void logStoreSimpleLog(SSyncLogStore* pLogStore) {
+  char* serialized = logStoreSimple2Str(pLogStore);
+  sTrace("logStoreSimpleLog | len:%lu | %s", strlen(serialized), serialized);
+  free(serialized);
+}
+
+void logStoreSimpleLog2(char* s, SSyncLogStore* pLogStore) {
+  char* serialized = logStoreSimple2Str(pLogStore);
+  sTrace("logStoreSimpleLog2 | len:%lu | %s | %s", strlen(serialized), s, serialized);
+  free(serialized);
+}
\ No newline at end of file

source/libs/sync/src/syncReplication.c

@@ -49,6 +49,10 @@
 int32_t syncNodeAppendEntriesPeers(SSyncNode* pSyncNode) {
   assert(pSyncNode->state == TAOS_SYNC_STATE_LEADER);
 
+  syncIndexMgrLog2("==syncNodeAppendEntriesPeers== pNextIndex", pSyncNode->pNextIndex);
+  syncIndexMgrLog2("==syncNodeAppendEntriesPeers== pMatchIndex", pSyncNode->pMatchIndex);
+  logStoreSimpleLog2("==syncNodeAppendEntriesPeers==", pSyncNode->pLogStore);
+
   int32_t ret = 0;
   for (int i = 0; i < pSyncNode->peersNum; ++i) {
     SRaftId* pDestId = &(pSyncNode->peersId[i]);
@@ -99,6 +103,8 @@ int32_t syncNodeAppendEntriesPeers(SSyncNode* pSyncNode) {
     pMsg->prevLogTerm = preLogTerm;
     pMsg->commitIndex = pSyncNode->commitIndex;
 
+    syncAppendEntriesLog2("==syncNodeAppendEntriesPeers==", pMsg);
+
     // send AppendEntries
     syncNodeAppendEntries(pSyncNode, pDestId, pMsg);
     syncAppendEntriesDestroy(pMsg);

source/libs/sync/test/CMakeLists.txt

@@ -29,10 +29,14 @@ add_executable(syncPingTimerTest2 "")
 add_executable(syncPingSelfTest "")
 add_executable(syncElectTest "")
 add_executable(syncElectTest2 "")
+add_executable(syncElectTest3 "")
 add_executable(syncEncodeTest "")
 add_executable(syncWriteTest "")
 add_executable(syncReplicateTest "")
+add_executable(syncReplicateTest2 "")
+add_executable(syncReplicateLoadTest "")
 add_executable(syncRefTest "")
+add_executable(syncLogStoreCheck "")
 
 
 target_sources(syncTest
@@ -159,6 +163,10 @@ target_sources(syncElectTest2
     PRIVATE
     "syncElectTest2.cpp"
 )
+target_sources(syncElectTest3
+    PRIVATE
+    "syncElectTest3.cpp"
+)
 target_sources(syncEncodeTest
     PRIVATE
     "syncEncodeTest.cpp"
@@ -171,10 +179,22 @@ target_sources(syncReplicateTest
     PRIVATE
     "syncReplicateTest.cpp"
 )
+target_sources(syncReplicateTest2
+    PRIVATE
+    "syncReplicateTest2.cpp"
+)
+target_sources(syncReplicateLoadTest
+    PRIVATE
+    "syncReplicateLoadTest.cpp"
+)
 target_sources(syncRefTest
     PRIVATE
     "syncRefTest.cpp"
 )
+target_sources(syncLogStoreCheck
+    PRIVATE
+    "syncLogStoreCheck.cpp"
+)
 
 
 target_include_directories(syncTest
@@ -332,6 +352,11 @@ target_include_directories(syncElectTest2
     "${CMAKE_SOURCE_DIR}/include/libs/sync"
     "${CMAKE_CURRENT_SOURCE_DIR}/../inc"
 )
+target_include_directories(syncElectTest3
+    PUBLIC
+    "${CMAKE_SOURCE_DIR}/include/libs/sync"
+    "${CMAKE_CURRENT_SOURCE_DIR}/../inc"
+)
 target_include_directories(syncEncodeTest
     PUBLIC
     "${CMAKE_SOURCE_DIR}/include/libs/sync"
@@ -347,11 +372,26 @@ target_include_directories(syncReplicateTest
     "${CMAKE_SOURCE_DIR}/include/libs/sync"
     "${CMAKE_CURRENT_SOURCE_DIR}/../inc"
 )
+target_include_directories(syncReplicateTest2
+    PUBLIC
+    "${CMAKE_SOURCE_DIR}/include/libs/sync"
+    "${CMAKE_CURRENT_SOURCE_DIR}/../inc"
+)
+target_include_directories(syncReplicateLoadTest
+    PUBLIC
+    "${CMAKE_SOURCE_DIR}/include/libs/sync"
+    "${CMAKE_CURRENT_SOURCE_DIR}/../inc"
+)
 target_include_directories(syncRefTest
     PUBLIC
     "${CMAKE_SOURCE_DIR}/include/libs/sync"
     "${CMAKE_CURRENT_SOURCE_DIR}/../inc"
 )
+target_include_directories(syncLogStoreCheck
+    PUBLIC
+    "${CMAKE_SOURCE_DIR}/include/libs/sync"
+    "${CMAKE_CURRENT_SOURCE_DIR}/../inc"
+)
 
 
 target_link_libraries(syncTest
@@ -478,6 +518,10 @@ target_link_libraries(syncElectTest2
     sync
     gtest_main
 )
+target_link_libraries(syncElectTest3
+    sync
+    gtest_main
+)
 target_link_libraries(syncEncodeTest
     sync
     gtest_main
@@ -490,10 +534,22 @@ target_link_libraries(syncReplicateTest
     sync
     gtest_main
 )
+target_link_libraries(syncReplicateTest2
+    sync
+    gtest_main
+)
+target_link_libraries(syncReplicateLoadTest
+    sync
+    gtest_main
+)
 target_link_libraries(syncRefTest
     sync
     gtest_main
 )
+target_link_libraries(syncLogStoreCheck
+    sync
+    gtest_main
+)
 
 
 enable_testing()

source/libs/sync/test/syncElectTest3.cpp

+#include <gtest/gtest.h>
+#include <stdio.h>
+#include "syncEnv.h"
+#include "syncIO.h"
+#include "syncInt.h"
+#include "syncRaftLog.h"
+#include "syncRaftStore.h"
+#include "syncUtil.h"
+#include "tref.h"
+
+void logTest() {
+  sTrace("--- sync log test: trace");
+  sDebug("--- sync log test: debug");
+  sInfo("--- sync log test: info");
+  sWarn("--- sync log test: warn");
+  sError("--- sync log test: error");
+  sFatal("--- sync log test: fatal");
+}
+
+uint16_t ports[] = {7010, 7110, 7210, 7310, 7410};
+int32_t  replicaNum = 3;
+int32_t  myIndex = 0;
+
+SRaftId   ids[TSDB_MAX_REPLICA];
+SSyncInfo syncInfo;
+SSyncFSM* pFsm;
+SWal*     pWal;
+
+int64_t syncNodeInit() {
+  syncInfo.vgId = 1234;
+  syncInfo.rpcClient = gSyncIO->clientRpc;
+  syncInfo.FpSendMsg = syncIOSendMsg;
+  syncInfo.queue = gSyncIO->pMsgQ;
+  syncInfo.FpEqMsg = syncIOEqMsg;
+  syncInfo.pFsm = pFsm;
+  snprintf(syncInfo.path, sizeof(syncInfo.path), "./elect3_test_%d", myIndex);
+
+  int code = walInit();
+  assert(code == 0);
+  SWalCfg walCfg;
+  memset(&walCfg, 0, sizeof(SWalCfg));
+  walCfg.vgId = syncInfo.vgId;
+  walCfg.fsyncPeriod = 1000;
+  walCfg.retentionPeriod = 1000;
+  walCfg.rollPeriod = 1000;
+  walCfg.retentionSize = 1000;
+  walCfg.segSize = 1000;
+  walCfg.level = TAOS_WAL_FSYNC;
+
+  char tmpdir[128];
+  snprintf(tmpdir, sizeof(tmpdir), "./elect3_test_wal_%d", myIndex);
+  pWal = walOpen(tmpdir, &walCfg);
+  assert(pWal != NULL);
+
+  syncInfo.pWal = pWal;
+
+  SSyncCfg* pCfg = &syncInfo.syncCfg;
+  pCfg->myIndex = myIndex;
+  pCfg->replicaNum = replicaNum;
+
+  for (int i = 0; i < replicaNum; ++i) {
+    pCfg->nodeInfo[i].nodePort = ports[i];
+    snprintf(pCfg->nodeInfo[i].nodeFqdn, sizeof(pCfg->nodeInfo[i].nodeFqdn), "%s", "127.0.0.1");
+    // taosGetFqdn(pCfg->nodeInfo[0].nodeFqdn);
+  }
+
+  int64_t rid = syncStart(&syncInfo);
+  assert(rid > 0);
+
+  SSyncNode* pSyncNode = (SSyncNode*)syncNodeAcquire(rid);
+  assert(pSyncNode != NULL);
+
+  pSyncNode->hbBaseLine = 500;
+  pSyncNode->electBaseLine = 1500;
+
+  gSyncIO->FpOnSyncPing = pSyncNode->FpOnPing;
+  gSyncIO->FpOnSyncPingReply = pSyncNode->FpOnPingReply;
+  gSyncIO->FpOnSyncRequestVote = pSyncNode->FpOnRequestVote;
+  gSyncIO->FpOnSyncRequestVoteReply = pSyncNode->FpOnRequestVoteReply;
+  gSyncIO->FpOnSyncAppendEntries = pSyncNode->FpOnAppendEntries;
+  gSyncIO->FpOnSyncAppendEntriesReply = pSyncNode->FpOnAppendEntriesReply;
+  gSyncIO->FpOnSyncPing = pSyncNode->FpOnPing;
+  gSyncIO->FpOnSyncPingReply = pSyncNode->FpOnPingReply;
+  gSyncIO->FpOnSyncTimeout = pSyncNode->FpOnTimeout;
+  gSyncIO->pSyncNode = pSyncNode;
+
+  syncNodeRelease(pSyncNode);
+
+  return rid;
+}
+
+void initRaftId(SSyncNode* pSyncNode) {
+  for (int i = 0; i < replicaNum; ++i) {
+    ids[i] = pSyncNode->replicasId[i];
+    char* s = syncUtilRaftId2Str(&ids[i]);
+    printf("raftId[%d] : %s\n", i, s);
+    free(s);
+  }
+}
+
+int main(int argc, char** argv) {
+  // taosInitLog((char *)"syncTest.log", 100000, 10);
+  tsAsyncLog = 0;
+  sDebugFlag = 143 + 64;
+
+  myIndex = 0;
+  if (argc >= 2) {
+    myIndex = atoi(argv[1]);
+  }
+
+  int32_t ret = syncIOStart((char*)"127.0.0.1", ports[myIndex]);
+  assert(ret == 0);
+
+  ret = syncInit();
+  assert(ret == 0);
+
+  int64_t rid = syncNodeInit();
+  assert(rid > 0);
+
+  SSyncNode* pSyncNode = (SSyncNode*)syncNodeAcquire(rid);
+  assert(pSyncNode != NULL);
+
+  syncNodePrint2((char*)"", pSyncNode);
+  initRaftId(pSyncNode);
+
+  //---------------------------
+  while (1) {
+    sTrace(
+        "elect sleep, state: %d, %s, term:%lu electTimerLogicClock:%lu, electTimerLogicClockUser:%lu, electTimerMS:%d",
+        pSyncNode->state, syncUtilState2String(pSyncNode->state), pSyncNode->pRaftStore->currentTerm,
+        pSyncNode->electTimerLogicClock, pSyncNode->electTimerLogicClockUser, pSyncNode->electTimerMS);
+    taosMsleep(1000);
+  }
+
+  syncNodeRelease(pSyncNode);
+
+  return 0;
+}

source/libs/sync/test/syncLogStoreCheck.cpp

+#include <gtest/gtest.h>
+#include <stdio.h>
+#include "syncEnv.h"
+#include "syncIO.h"
+#include "syncInt.h"
+#include "syncRaftLog.h"
+#include "syncRaftStore.h"
+#include "syncUtil.h"
+
+void logTest() {
+  sTrace("--- sync log test: trace");
+  sDebug("--- sync log test: debug");
+  sInfo("--- sync log test: info");
+  sWarn("--- sync log test: warn");
+  sError("--- sync log test: error");
+  sFatal("--- sync log test: fatal");
+}
+
+uint16_t ports[] = {7010, 7110, 7210, 7310, 7410};
+int32_t  replicaNum = 1;
+int32_t  myIndex = 0;
+
+SRaftId    ids[TSDB_MAX_REPLICA];
+SSyncInfo  syncInfo;
+SSyncFSM*  pFsm;
+SWal*      pWal;
+SSyncNode* pSyncNode;
+
+SSyncNode* syncNodeInit(const char* path) {
+  syncInfo.vgId = 1234;
+  syncInfo.rpcClient = gSyncIO->clientRpc;
+  syncInfo.FpSendMsg = syncIOSendMsg;
+  syncInfo.queue = gSyncIO->pMsgQ;
+  syncInfo.FpEqMsg = syncIOEqMsg;
+  syncInfo.pFsm = pFsm;
+  snprintf(syncInfo.path, sizeof(syncInfo.path), "%s", "./log_check");
+
+  int code = walInit();
+  assert(code == 0);
+  SWalCfg walCfg;
+  memset(&walCfg, 0, sizeof(SWalCfg));
+  walCfg.vgId = syncInfo.vgId;
+  walCfg.fsyncPeriod = 1000;
+  walCfg.retentionPeriod = 1000;
+  walCfg.rollPeriod = 1000;
+  walCfg.retentionSize = 1000;
+  walCfg.segSize = 1000;
+  walCfg.level = TAOS_WAL_FSYNC;
+  pWal = walOpen(path, &walCfg);
+  assert(pWal != NULL);
+
+  syncInfo.pWal = pWal;
+
+  SSyncCfg* pCfg = &syncInfo.syncCfg;
+  pCfg->myIndex = myIndex;
+  pCfg->replicaNum = replicaNum;
+
+  for (int i = 0; i < replicaNum; ++i) {
+    pCfg->nodeInfo[i].nodePort = ports[i];
+    snprintf(pCfg->nodeInfo[i].nodeFqdn, sizeof(pCfg->nodeInfo[i].nodeFqdn), "%s", "127.0.0.1");
+    // taosGetFqdn(pCfg->nodeInfo[0].nodeFqdn);
+  }
+
+  pSyncNode = syncNodeOpen(&syncInfo);
+  assert(pSyncNode != NULL);
+
+  gSyncIO->FpOnSyncPing = pSyncNode->FpOnPing;
+  gSyncIO->FpOnSyncPingReply = pSyncNode->FpOnPingReply;
+  gSyncIO->FpOnSyncRequestVote = pSyncNode->FpOnRequestVote;
+  gSyncIO->FpOnSyncRequestVoteReply = pSyncNode->FpOnRequestVoteReply;
+  gSyncIO->FpOnSyncAppendEntries = pSyncNode->FpOnAppendEntries;
+  gSyncIO->FpOnSyncAppendEntriesReply = pSyncNode->FpOnAppendEntriesReply;
+  gSyncIO->FpOnSyncPing = pSyncNode->FpOnPing;
+  gSyncIO->FpOnSyncPingReply = pSyncNode->FpOnPingReply;
+  gSyncIO->FpOnSyncTimeout = pSyncNode->FpOnTimeout;
+  gSyncIO->pSyncNode = pSyncNode;
+
+  return pSyncNode;
+}
+
+SSyncNode* logStoreCheck(const char* path) { return syncNodeInit(path); }
+
+int main(int argc, char** argv) {
+  // taosInitLog((char *)"syncTest.log", 100000, 10);
+  tsAsyncLog = 0;
+  sDebugFlag = 143 + 64;
+
+  myIndex = 0;
+  if (argc >= 2) {
+    myIndex = atoi(argv[1]);
+  }
+
+  int32_t ret = syncIOStart((char*)"127.0.0.1", ports[myIndex]);
+  assert(ret == 0);
+
+  ret = syncEnvStart();
+  assert(ret == 0);
+
+  pSyncNode = logStoreCheck(argv[1]);
+  assert(pSyncNode != NULL);
+
+  logStorePrint2((char*)"logStoreCheck", pSyncNode->pLogStore);
+
+  return 0;
+}

source/libs/sync/test/syncReplicateLoadTest.cpp

+#include <gtest/gtest.h>
+#include <stdio.h>
+#include "syncEnv.h"
+#include "syncIO.h"
+#include "syncInt.h"
+#include "syncMessage.h"
+#include "syncRaftEntry.h"
+#include "syncRaftLog.h"
+#include "syncRaftStore.h"
+#include "syncUtil.h"
+
+void logTest() {
+  sTrace("--- sync log test: trace");
+  sDebug("--- sync log test: debug");
+  sInfo("--- sync log test: info");
+  sWarn("--- sync log test: warn");
+  sError("--- sync log test: error");
+  sFatal("--- sync log test: fatal");
+}
+
+uint16_t ports[] = {7010, 7110, 7210, 7310, 7410};
+int32_t  replicaNum = 3;
+int32_t  myIndex = 0;
+
+SRaftId   ids[TSDB_MAX_REPLICA];
+SSyncInfo syncInfo;
+SSyncFSM *pFsm;
+SWal *    pWal;
+
+void CommitCb(struct SSyncFSM *pFsm, const SRpcMsg *pMsg, SyncIndex index, bool isWeak, int32_t code,
+              ESyncState state) {
+  char logBuf[256];
+  snprintf(logBuf, sizeof(logBuf), "==callback== ==CommitCb== pFsm:%p, index:%ld, isWeak:%d, code:%d, state:%d %s \n",
+           pFsm, index, isWeak, code, state, syncUtilState2String(state));
+  syncRpcMsgPrint2(logBuf, (SRpcMsg *)pMsg);
+}
+
+void PreCommitCb(struct SSyncFSM *pFsm, const SRpcMsg *pMsg, SyncIndex index, bool isWeak, int32_t code,
+                 ESyncState state) {
+  char logBuf[256];
+  snprintf(logBuf, sizeof(logBuf),
+           "==callback== ==PreCommitCb== pFsm:%p, index:%ld, isWeak:%d, code:%d, state:%d %s \n", pFsm, index, isWeak,
+           code, state, syncUtilState2String(state));
+  syncRpcMsgPrint2(logBuf, (SRpcMsg *)pMsg);
+}
+
+void RollBackCb(struct SSyncFSM *pFsm, const SRpcMsg *pMsg, SyncIndex index, bool isWeak, int32_t code,
+                ESyncState state) {
+  char logBuf[256];
+  snprintf(logBuf, sizeof(logBuf), "==callback== ==RollBackCb== pFsm:%p, index:%ld, isWeak:%d, code:%d, state:%d %s \n",
+           pFsm, index, isWeak, code, state, syncUtilState2String(state));
+  syncRpcMsgPrint2(logBuf, (SRpcMsg *)pMsg);
+}
+
+void initFsm() {
+  pFsm = (SSyncFSM *)malloc(sizeof(SSyncFSM));
+  pFsm->FpCommitCb = CommitCb;
+  pFsm->FpPreCommitCb = PreCommitCb;
+  pFsm->FpRollBackCb = RollBackCb;
+}
+
+int64_t syncNodeInit() {
+  syncInfo.vgId = 1234;
+  syncInfo.rpcClient = gSyncIO->clientRpc;
+  syncInfo.FpSendMsg = syncIOSendMsg;
+  syncInfo.queue = gSyncIO->pMsgQ;
+  syncInfo.FpEqMsg = syncIOEqMsg;
+  syncInfo.pFsm = pFsm;
+  snprintf(syncInfo.path, sizeof(syncInfo.path), "./replicate2_test_%d", myIndex);
+
+  int code = walInit();
+  assert(code == 0);
+  SWalCfg walCfg;
+  memset(&walCfg, 0, sizeof(SWalCfg));
+  walCfg.vgId = syncInfo.vgId;
+  walCfg.fsyncPeriod = 1000;
+  walCfg.retentionPeriod = 1000;
+  walCfg.rollPeriod = 1000;
+  walCfg.retentionSize = 1000;
+  walCfg.segSize = 1000;
+  walCfg.level = TAOS_WAL_FSYNC;
+
+  char tmpdir[128];
+  snprintf(tmpdir, sizeof(tmpdir), "./replicate2_test_wal_%d", myIndex);
+  pWal = walOpen(tmpdir, &walCfg);
+  assert(pWal != NULL);
+
+  syncInfo.pWal = pWal;
+
+  SSyncCfg *pCfg = &syncInfo.syncCfg;
+  pCfg->myIndex = myIndex;
+  pCfg->replicaNum = replicaNum;
+
+  for (int i = 0; i < replicaNum; ++i) {
+    pCfg->nodeInfo[i].nodePort = ports[i];
+    snprintf(pCfg->nodeInfo[i].nodeFqdn, sizeof(pCfg->nodeInfo[i].nodeFqdn), "%s", "127.0.0.1");
+    // taosGetFqdn(pCfg->nodeInfo[0].nodeFqdn);
+  }
+
+  int64_t rid = syncStart(&syncInfo);
+  assert(rid > 0);
+
+  SSyncNode *pSyncNode = (SSyncNode *)syncNodeAcquire(rid);
+  assert(pSyncNode != NULL);
+
+  // pSyncNode->hbBaseLine = 500;
+  // pSyncNode->electBaseLine = 1500;
+
+  gSyncIO->FpOnSyncPing = pSyncNode->FpOnPing;
+  gSyncIO->FpOnSyncPingReply = pSyncNode->FpOnPingReply;
+  gSyncIO->FpOnSyncRequestVote = pSyncNode->FpOnRequestVote;
+  gSyncIO->FpOnSyncRequestVoteReply = pSyncNode->FpOnRequestVoteReply;
+  gSyncIO->FpOnSyncAppendEntries = pSyncNode->FpOnAppendEntries;
+  gSyncIO->FpOnSyncAppendEntriesReply = pSyncNode->FpOnAppendEntriesReply;
+  gSyncIO->FpOnSyncTimeout = pSyncNode->FpOnTimeout;
+  gSyncIO->FpOnSyncClientRequest = pSyncNode->FpOnClientRequest;
+  gSyncIO->pSyncNode = pSyncNode;
+
+  syncNodeRelease(pSyncNode);
+
+  return rid;
+}
+
+void initRaftId(SSyncNode *pSyncNode) {
+  for (int i = 0; i < replicaNum; ++i) {
+    ids[i] = pSyncNode->replicasId[i];
+    char *s = syncUtilRaftId2Str(&ids[i]);
+    printf("raftId[%d] : %s\n", i, s);
+    free(s);
+  }
+}
+
+SRpcMsg *step0(int i) {
+  SRpcMsg *pMsg = (SRpcMsg *)malloc(sizeof(SRpcMsg));
+  memset(pMsg, 0, sizeof(SRpcMsg));
+  pMsg->msgType = 9999;
+  pMsg->contLen = 128;
+  pMsg->pCont = malloc(pMsg->contLen);
+  snprintf((char *)(pMsg->pCont), pMsg->contLen, "value-%u-%d", ports[myIndex], i);
+  return pMsg;
+}
+
+SyncClientRequest *step1(const SRpcMsg *pMsg) {
+  SyncClientRequest *pRetMsg = syncClientRequestBuild2(pMsg, 123, true);
+  return pRetMsg;
+}
+
+int main(int argc, char **argv) {
+  // taosInitLog((char *)"syncTest.log", 100000, 10);
+  tsAsyncLog = 0;
+  sDebugFlag = 143 + 64;
+  void logTest();
+
+  myIndex = 0;
+  if (argc >= 2) {
+    myIndex = atoi(argv[1]);
+  }
+
+  int32_t ret = syncIOStart((char *)"127.0.0.1", ports[myIndex]);
+  assert(ret == 0);
+
+  initFsm();
+
+  ret = syncInit();
+  assert(ret == 0);
+
+  int64_t rid = syncNodeInit();
+  assert(rid > 0);
+
+  SSyncNode *pSyncNode = (SSyncNode *)syncNodeAcquire(rid);
+  assert(pSyncNode != NULL);
+
+  syncNodePrint2((char *)"", pSyncNode);
+  initRaftId(pSyncNode);
+
+  // only load ...
+
+  while (1) {
+    sTrace(
+        "replicate sleep, state: %d, %s, term:%lu electTimerLogicClock:%lu, electTimerLogicClockUser:%lu, "
+        "electTimerMS:%d",
+        pSyncNode->state, syncUtilState2String(pSyncNode->state), pSyncNode->pRaftStore->currentTerm,
+        pSyncNode->electTimerLogicClock, pSyncNode->electTimerLogicClockUser, pSyncNode->electTimerMS);
+    taosMsleep(1000);
+  }
+
+  return 0;
+}

source/libs/sync/test/syncReplicateTest.cpp

@@ -28,19 +28,29 @@ SSyncFSM * pFsm;
 SWal *     pWal;
 SSyncNode *gSyncNode;
 
-void CommitCb(struct SSyncFSM *pFsm, const SRpcMsg *pBuf, SyncIndex index, bool isWeak, int32_t code) {
-  printf("==CommitCb== pFsm:%p, index:%ld, isWeak:%d, code:%d \n", pFsm, index, isWeak, code);
-  syncRpcMsgPrint2((char *)"==CommitCb==", (SRpcMsg *)pBuf);
+void CommitCb(struct SSyncFSM *pFsm, const SRpcMsg *pMsg, SyncIndex index, bool isWeak, int32_t code,
+              ESyncState state) {
+  char logBuf[256];
+  snprintf(logBuf, sizeof(logBuf), "==callback== ==CommitCb== pFsm:%p, index:%ld, isWeak:%d, code:%d, state:%d %s \n",
+           pFsm, index, isWeak, code, state, syncUtilState2String(state));
+  syncRpcMsgPrint2(logBuf, (SRpcMsg *)pMsg);
 }
 
-void PreCommitCb(struct SSyncFSM *pFsm, const SRpcMsg *pBuf, SyncIndex index, bool isWeak, int32_t code) {
-  printf("==PreCommitCb== pFsm:%p, index:%ld, isWeak:%d, code:%d \n", pFsm, index, isWeak, code);
-  syncRpcMsgPrint2((char *)"==PreCommitCb==", (SRpcMsg *)pBuf);
+void PreCommitCb(struct SSyncFSM *pFsm, const SRpcMsg *pMsg, SyncIndex index, bool isWeak, int32_t code,
+                 ESyncState state) {
+  char logBuf[256];
+  snprintf(logBuf, sizeof(logBuf),
+           "==callback== ==PreCommitCb== pFsm:%p, index:%ld, isWeak:%d, code:%d, state:%d %s \n", pFsm, index, isWeak,
+           code, state, syncUtilState2String(state));
+  syncRpcMsgPrint2(logBuf, (SRpcMsg *)pMsg);
 }
 
-void RollBackCb(struct SSyncFSM *pFsm, const SRpcMsg *pBuf, SyncIndex index, bool isWeak, int32_t code) {
-  printf("==RollBackCb== pFsm:%p, index:%ld, isWeak:%d, code:%d \n", pFsm, index, isWeak, code);
-  syncRpcMsgPrint2((char *)"==RollBackCb==", (SRpcMsg *)pBuf);
+void RollBackCb(struct SSyncFSM *pFsm, const SRpcMsg *pMsg, SyncIndex index, bool isWeak, int32_t code,
+                ESyncState state) {
+  char logBuf[256];
+  snprintf(logBuf, sizeof(logBuf), "==callback== ==RollBackCb== pFsm:%p, index:%ld, isWeak:%d, code:%d, state:%d %s \n",
+           pFsm, index, isWeak, code, state, syncUtilState2String(state));
+  syncRpcMsgPrint2(logBuf, (SRpcMsg *)pMsg);
 }
 
 void initFsm() {

source/libs/sync/test/syncReplicateTest2.cpp

+#include <gtest/gtest.h>
+#include <stdio.h>
+#include "syncEnv.h"
+#include "syncIO.h"
+#include "syncInt.h"
+#include "syncMessage.h"
+#include "syncRaftEntry.h"
+#include "syncRaftLog.h"
+#include "syncRaftStore.h"
+#include "syncUtil.h"
+
+void logTest() {
+  sTrace("--- sync log test: trace");
+  sDebug("--- sync log test: debug");
+  sInfo("--- sync log test: info");
+  sWarn("--- sync log test: warn");
+  sError("--- sync log test: error");
+  sFatal("--- sync log test: fatal");
+}
+
+uint16_t ports[] = {7010, 7110, 7210, 7310, 7410};
+int32_t  replicaNum = 3;
+int32_t  myIndex = 0;
+
+SRaftId   ids[TSDB_MAX_REPLICA];
+SSyncInfo syncInfo;
+SSyncFSM *pFsm;
+SWal *    pWal;
+
+void CommitCb(struct SSyncFSM *pFsm, const SRpcMsg *pMsg, SyncIndex index, bool isWeak, int32_t code,
+              ESyncState state) {
+  char logBuf[256];
+  snprintf(logBuf, sizeof(logBuf), "==callback== ==CommitCb== pFsm:%p, index:%ld, isWeak:%d, code:%d, state:%d %s \n",
+           pFsm, index, isWeak, code, state, syncUtilState2String(state));
+  syncRpcMsgPrint2(logBuf, (SRpcMsg *)pMsg);
+}
+
+void PreCommitCb(struct SSyncFSM *pFsm, const SRpcMsg *pMsg, SyncIndex index, bool isWeak, int32_t code,
+                 ESyncState state) {
+  char logBuf[256];
+  snprintf(logBuf, sizeof(logBuf),
+           "==callback== ==PreCommitCb== pFsm:%p, index:%ld, isWeak:%d, code:%d, state:%d %s \n", pFsm, index, isWeak,
+           code, state, syncUtilState2String(state));
+  syncRpcMsgPrint2(logBuf, (SRpcMsg *)pMsg);
+}
+
+void RollBackCb(struct SSyncFSM *pFsm, const SRpcMsg *pMsg, SyncIndex index, bool isWeak, int32_t code,
+                ESyncState state) {
+  char logBuf[256];
+  snprintf(logBuf, sizeof(logBuf), "==callback== ==RollBackCb== pFsm:%p, index:%ld, isWeak:%d, code:%d, state:%d %s \n",
+           pFsm, index, isWeak, code, state, syncUtilState2String(state));
+  syncRpcMsgPrint2(logBuf, (SRpcMsg *)pMsg);
+}
+
+void initFsm() {
+  pFsm = (SSyncFSM *)malloc(sizeof(SSyncFSM));
+  pFsm->FpCommitCb = CommitCb;
+  pFsm->FpPreCommitCb = PreCommitCb;
+  pFsm->FpRollBackCb = RollBackCb;
+}
+
+int64_t syncNodeInit() {
+  syncInfo.vgId = 1234;
+  syncInfo.rpcClient = gSyncIO->clientRpc;
+  syncInfo.FpSendMsg = syncIOSendMsg;
+  syncInfo.queue = gSyncIO->pMsgQ;
+  syncInfo.FpEqMsg = syncIOEqMsg;
+  syncInfo.pFsm = pFsm;
+  snprintf(syncInfo.path, sizeof(syncInfo.path), "./replicate2_test_%d", myIndex);
+
+  int code = walInit();
+  assert(code == 0);
+  SWalCfg walCfg;
+  memset(&walCfg, 0, sizeof(SWalCfg));
+  walCfg.vgId = syncInfo.vgId;
+  walCfg.fsyncPeriod = 1000;
+  walCfg.retentionPeriod = 1000;
+  walCfg.rollPeriod = 1000;
+  walCfg.retentionSize = 1000;
+  walCfg.segSize = 1000;
+  walCfg.level = TAOS_WAL_FSYNC;
+
+  char tmpdir[128];
+  snprintf(tmpdir, sizeof(tmpdir), "./replicate2_test_wal_%d", myIndex);
+  pWal = walOpen(tmpdir, &walCfg);
+  assert(pWal != NULL);
+
+  syncInfo.pWal = pWal;
+
+  SSyncCfg *pCfg = &syncInfo.syncCfg;
+  pCfg->myIndex = myIndex;
+  pCfg->replicaNum = replicaNum;
+
+  for (int i = 0; i < replicaNum; ++i) {
+    pCfg->nodeInfo[i].nodePort = ports[i];
+    snprintf(pCfg->nodeInfo[i].nodeFqdn, sizeof(pCfg->nodeInfo[i].nodeFqdn), "%s", "127.0.0.1");
+    // taosGetFqdn(pCfg->nodeInfo[0].nodeFqdn);
+  }
+
+  int64_t rid = syncStart(&syncInfo);
+  assert(rid > 0);
+
+  SSyncNode *pSyncNode = (SSyncNode *)syncNodeAcquire(rid);
+  assert(pSyncNode != NULL);
+
+  // pSyncNode->hbBaseLine = 500;
+  // pSyncNode->electBaseLine = 1500;
+
+  gSyncIO->FpOnSyncPing = pSyncNode->FpOnPing;
+  gSyncIO->FpOnSyncPingReply = pSyncNode->FpOnPingReply;
+  gSyncIO->FpOnSyncRequestVote = pSyncNode->FpOnRequestVote;
+  gSyncIO->FpOnSyncRequestVoteReply = pSyncNode->FpOnRequestVoteReply;
+  gSyncIO->FpOnSyncAppendEntries = pSyncNode->FpOnAppendEntries;
+  gSyncIO->FpOnSyncAppendEntriesReply = pSyncNode->FpOnAppendEntriesReply;
+  gSyncIO->FpOnSyncTimeout = pSyncNode->FpOnTimeout;
+  gSyncIO->FpOnSyncClientRequest = pSyncNode->FpOnClientRequest;
+  gSyncIO->pSyncNode = pSyncNode;
+
+  syncNodeRelease(pSyncNode);
+
+  return rid;
+}
+
+void initRaftId(SSyncNode *pSyncNode) {
+  for (int i = 0; i < replicaNum; ++i) {
+    ids[i] = pSyncNode->replicasId[i];
+    char *s = syncUtilRaftId2Str(&ids[i]);
+    printf("raftId[%d] : %s\n", i, s);
+    free(s);
+  }
+}
+
+SRpcMsg *step0(int i) {
+  SRpcMsg *pMsg = (SRpcMsg *)malloc(sizeof(SRpcMsg));
+  memset(pMsg, 0, sizeof(SRpcMsg));
+  pMsg->msgType = 9999;
+  pMsg->contLen = 128;
+  pMsg->pCont = malloc(pMsg->contLen);
+  snprintf((char *)(pMsg->pCont), pMsg->contLen, "value-%u-%d", ports[myIndex], i);
+  return pMsg;
+}
+
+SyncClientRequest *step1(const SRpcMsg *pMsg) {
+  SyncClientRequest *pRetMsg = syncClientRequestBuild2(pMsg, 123, true);
+  return pRetMsg;
+}
+
+int main(int argc, char **argv) {
+  // taosInitLog((char *)"syncTest.log", 100000, 10);
+  tsAsyncLog = 0;
+  sDebugFlag = 143 + 64;
+  void logTest();
+
+  myIndex = 0;
+  if (argc >= 2) {
+    myIndex = atoi(argv[1]);
+  }
+
+  int32_t ret = syncIOStart((char *)"127.0.0.1", ports[myIndex]);
+  assert(ret == 0);
+
+  initFsm();
+
+  ret = syncInit();
+  assert(ret == 0);
+
+  int64_t rid = syncNodeInit();
+  assert(rid > 0);
+
+  SSyncNode *pSyncNode = (SSyncNode *)syncNodeAcquire(rid);
+  assert(pSyncNode != NULL);
+
+  syncNodePrint2((char *)"", pSyncNode);
+  initRaftId(pSyncNode);
+
+  for (int i = 0; i < 30; ++i) {
+    // step0
+    SRpcMsg *pMsg0 = step0(i);
+    syncRpcMsgPrint2((char *)"==step0==", pMsg0);
+
+    syncPropose(rid, pMsg0, true);
+    taosMsleep(1000);
+
+    free(pMsg0);
+
+    sTrace(
+        "syncPropose sleep, state: %d, %s, term:%lu electTimerLogicClock:%lu, electTimerLogicClockUser:%lu, "
+        "electTimerMS:%d",
+        pSyncNode->state, syncUtilState2String(pSyncNode->state), pSyncNode->pRaftStore->currentTerm,
+        pSyncNode->electTimerLogicClock, pSyncNode->electTimerLogicClockUser, pSyncNode->electTimerMS);
+  }
+
+  while (1) {
+    sTrace(
+        "replicate sleep, state: %d, %s, term:%lu electTimerLogicClock:%lu, electTimerLogicClockUser:%lu, "
+        "electTimerMS:%d",
+        pSyncNode->state, syncUtilState2String(pSyncNode->state), pSyncNode->pRaftStore->currentTerm,
+        pSyncNode->electTimerLogicClock, pSyncNode->electTimerLogicClockUser, pSyncNode->electTimerMS);
+    taosMsleep(1000);
+  }
+
+  return 0;
+}

source/libs/sync/test/syncWriteTest.cpp

@@ -28,19 +28,29 @@ SSyncFSM * pFsm;
 SWal *     pWal;
 SSyncNode *gSyncNode;
 
-void CommitCb(struct SSyncFSM *pFsm, const SRpcMsg *pBuf, SyncIndex index, bool isWeak, int32_t code) {
-  printf("==CommitCb== pFsm:%p, index:%ld, isWeak:%d, code:%d \n", pFsm, index, isWeak, code);
-  syncRpcMsgPrint2((char *)"==CommitCb==", (SRpcMsg *)pBuf);
+void CommitCb(struct SSyncFSM *pFsm, const SRpcMsg *pMsg, SyncIndex index, bool isWeak, int32_t code,
+              ESyncState state) {
+  char logBuf[256];
+  snprintf(logBuf, sizeof(logBuf), "==callback== ==CommitCb== pFsm:%p, index:%ld, isWeak:%d, code:%d, state:%d %s \n",
+           pFsm, index, isWeak, code, state, syncUtilState2String(state));
+  syncRpcMsgPrint2(logBuf, (SRpcMsg *)pMsg);
 }
 
-void PreCommitCb(struct SSyncFSM *pFsm, const SRpcMsg *pBuf, SyncIndex index, bool isWeak, int32_t code) {
-  printf("==PreCommitCb== pFsm:%p, index:%ld, isWeak:%d, code:%d \n", pFsm, index, isWeak, code);
-  syncRpcMsgPrint2((char *)"==PreCommitCb==", (SRpcMsg *)pBuf);
+void PreCommitCb(struct SSyncFSM *pFsm, const SRpcMsg *pMsg, SyncIndex index, bool isWeak, int32_t code,
+                 ESyncState state) {
+  char logBuf[256];
+  snprintf(logBuf, sizeof(logBuf),
+           "==callback== ==PreCommitCb== pFsm:%p, index:%ld, isWeak:%d, code:%d, state:%d %s \n", pFsm, index, isWeak,
+           code, state, syncUtilState2String(state));
+  syncRpcMsgPrint2(logBuf, (SRpcMsg *)pMsg);
 }
 
-void RollBackCb(struct SSyncFSM *pFsm, const SRpcMsg *pBuf, SyncIndex index, bool isWeak, int32_t code) {
-  printf("==RollBackCb== pFsm:%p, index:%ld, isWeak:%d, code:%d \n", pFsm, index, isWeak, code);
-  syncRpcMsgPrint2((char *)"==RollBackCb==", (SRpcMsg *)pBuf);
+void RollBackCb(struct SSyncFSM *pFsm, const SRpcMsg *pMsg, SyncIndex index, bool isWeak, int32_t code,
+                ESyncState state) {
+  char logBuf[256];
+  snprintf(logBuf, sizeof(logBuf), "==callback== ==RollBackCb== pFsm:%p, index:%ld, isWeak:%d, code:%d, state:%d %s \n",
+           pFsm, index, isWeak, code, state, syncUtilState2String(state));
+  syncRpcMsgPrint2(logBuf, (SRpcMsg *)pMsg);
 }
 
 void initFsm() {

source/os/src/osSocket.c

@@ -47,9 +47,6 @@
 #endif
 #endif
 
-typedef int32_t SocketFd;
-typedef SocketFd  EpollFd;
-
 typedef struct TdSocketServer {
 #if SOCKET_WITH_LOCK
   TdThreadRwlock rwlock;
@@ -58,14 +55,6 @@ typedef struct TdSocketServer {
   SocketFd fd;
 } *TdSocketServerPtr, TdSocketServer;
 
-typedef struct TdSocket {
-#if SOCKET_WITH_LOCK
-  TdThreadRwlock rwlock;
-#endif
-  int      refId;
-  SocketFd fd;
-} *TdSocketPtr, TdSocket;
-
 typedef struct TdEpoll {
 #if SOCKET_WITH_LOCK
   TdThreadRwlock rwlock;

source/util/src/terror.c

@@ -336,6 +336,7 @@ TAOS_DEFINE_ERROR(TSDB_CODE_TDB_TDB_ENV_OPEN_ERROR,       "TDB env open error")
 TAOS_DEFINE_ERROR(TSDB_CODE_TDB_NO_SMA_INDEX_IN_META,     "No sma index in meta")
 TAOS_DEFINE_ERROR(TSDB_CODE_TDB_INVALID_SMA_STAT,         "Invalid sma state")
 
+
 // query
 TAOS_DEFINE_ERROR(TSDB_CODE_QRY_INVALID_QHANDLE,          "Invalid handle")
 TAOS_DEFINE_ERROR(TSDB_CODE_QRY_INVALID_MSG,              "Invalid message")    // failed to validate the sql expression msg by vnode

tests/script/jenkins/basic.txt

@@ -10,18 +10,20 @@
 ./test.sh -f tsim/db/basic7.sim
 ./test.sh -f tsim/db/error1.sim
 
-# ---- table
-./test.sh -f tsim/table/basic1.sim 
-
 # ---- dnode
 ./test.sh -f tsim/dnode/basic1.sim
 
 # ---- insert
 ./test.sh -f tsim/insert/basic0.sim
+./test.sh -f tsim/insert/null.sim
 
 # ---- query
 ./test.sh -f tsim/query/interval.sim
 
+# ---- table
+./test.sh -f tsim/table/basic1.sim 
+
 # ---- tmq
 ./test.sh -f tsim/tmq/basic.sim
+
 #======================b1-end===============

tests/script/tsim/insert/basic0.sim

@@ -36,14 +36,14 @@ endi
 
 print =============== insert data, mode1: one row one table in sql
 print =============== insert data, mode1: mulit rows one table in sql
-print =============== insert data, mode1: one rows mulit table in sql
-print =============== insert data, mode1: mulit rows mulit table in sql
+#print =============== insert data, mode1: one rows mulit table in sql
+#print =============== insert data, mode1: mulit rows mulit table in sql
 sql insert into ct1 values(now+0s, 10, 2.0, 3.0)
-sql insert into ct1 values(now+1s, 11, 2.1, 3.1)(now+2s, 12, 2.2, 3.2)(now+3s, 13, 2.3, 3.3)
+sql insert into ct1 values(now+1s, 11, 2.1, 3.1)(now+2s, -12, -2.2, -3.2)(now+3s, -13, -2.3, -3.3)
 sql insert into ct2 values(now+0s, 10, 2.0, 3.0)
-sql insert into ct2 values(now+1s, 11, 2.1, 3.1)(now+2s, 12, 2.2, 3.2)(now+3s, 13, 2.3, 3.3)
-sql insert into ct1 values(now+4s, -14, -2.4, -3.4) ct2 values(now+4s, -14, -2.4, -3.4)
-sql insert into ct1 values(now+5s, -15, -2.5, -3.5)(now+6s, -16, -2.6, -3.6) ct2 values(now+5s, -15, -2.5, -3.5)(now+6s, -16, -2.6, -3.6)
+sql insert into ct2 values(now+1s, 11, 2.1, 3.1)(now+2s, -12, -2.2, -3.2)(now+3s, -13, -2.3, -3.3)
+#sql insert into ct1 values(now+4s, -14, -2.4, -3.4) ct2 values(now+4s, -14, -2.4, -3.4)
+#sql insert into ct1 values(now+5s, -15, -2.5, -3.5)(now+6s, -16, -2.6, -3.6) ct2 values(now+5s, -15, -2.5, -3.5)(now+6s, -16, -2.6, -3.6)
 
 sql insert into ct3 values('2021-01-01 00:00:00.000', 10, 2.0, 3.0)
 
@@ -52,14 +52,11 @@ sql insert into ct3 values('2021-01-01 00:00:00.000', 10, 2.0, 3.0)
 print =============== query data from child table
 sql select * from ct1
 print rows: $rows 
-print $data00  $data01
-print $data10  $data11 
-print $data20  $data21
-print $data30  $data31
-print $data40  $data41
-print $data50  $data51
-print $data60  $data61
-if $rows != 7 then
+print $data00  $data01 $data02 $data03
+print $data10  $data11 $data12 $data13
+print $data20  $data21 $data22 $data23
+print $data30  $data31 $data32 $data33
+if $rows != 4 then
   return -1
 endi
 if $data01 != 10 then 
@@ -84,28 +81,29 @@ endi
 
 print =============== select count(*) from child table
 sql select count(*) from ct1
+print rows: $rows 
+print $data00  $data01 $data02 $data03
 if $rows != 1 then 
   return -1
 endi
-
-print $data00 $data01 $data02
-if $data00 != 7 then 
+if $data00 != 4 then 
   return -1
 endi
 
 print =============== select count(column) from child table
 sql select count(ts), count(c1), count(c2), count(c3) from ct1
+print rows: $rows 
 print $data00  $data01 $data02 $data03
-if $data00 != 7 then 
+if $data00 != 4 then 
   return -1
 endi
-if $data01 != 7 then 
+if $data01 != 4 then 
   return -1
 endi
-if $data02 != 7 then 
+if $data02 != 4 then 
   return -1
 endi
-if $data03 != 7 then 
+if $data03 != 4 then 
   return -1
 endi
 
@@ -115,17 +113,18 @@ endi
 
 print =============== select min(column) from child table
 sql select min(c1), min(c2), min(c3) from ct1
+print rows: $rows 
 print $data00  $data01 $data02 $data03
 if $rows != 1 then 
   return -1
 endi
-if $data00 != -16 then 
+if $data00 != -13 then 
   return -1
 endi
-if $data01 != -2.60000 then 
+if $data01 != -2.30000 then 
   return -1
 endi
-if $data02 != -3.600000000 then 
+if $data02 != -3.300000000 then 
   return -1
 endi
 
@@ -135,13 +134,13 @@ print $data00 $data01 $data02  $data03
 if $rows != 1 then 
   return -1
 endi
-if $data00 != 13 then 
+if $data00 != 11 then 
   return -1
 endi
-if $data01 != 2.30000 then 
+if $data01 != 2.10000 then 
   return -1
 endi
-if $data02 != 3.300000000 then 
+if $data02 != 3.100000000 then 
   return -1
 endi
 
@@ -151,49 +150,53 @@ print $data00 $data01 $data02  $data03
 if $rows != 1 then 
   return -1
 endi
-if $data00 != 1 then 
+if $data00 != -4 then 
   return -1
 endi
-if $data01 != 1.099999905 then 
+if $data01 != -0.400000095 then 
   return -1
 endi
-if $data02 != 2.100000000 then 
+if $data02 != -0.400000000 then 
   return -1
 endi
 
-print =============== select column, from child table
+print =============== select column without timestamp, from child table
 sql select c1, c2, c3 from ct1
+print rows: $rows
 print $data00 $data01 $data02
-#if $rows != 4 then 
-#  return -1
-#endi
-#if $data00 != 10 then 
-#  return -1
-#endi
-#if $data01 != 2.00000 then 
-#  return -1
-#endi
-#if $data02 != 3.000000000 then 
-#  return -1
-#endi
-#if $data10 != 11 then 
-#  return -1
-#endi
-#if $data11 != 2.10000 then 
-#  return -1
-#endi
-#if $data12 != 3.100000000 then 
-#  return -1
-#endi
-#if $data30 != 13 then 
-#  return -1
-#endi
-#if $data31 != 2.30000 then 
-#  return -1
-#endi
-#if $data32 != 3.300000000 then 
-#  return -1
-#endi
+print $data10 $data11 $data12
+print $data20 $data21 $data22
+print $data30 $data31 $data32
+if $rows != 4 then 
+  return -1
+endi
+if $data00 != 10 then 
+  return -1
+endi
+if $data01 != 2.00000 then 
+  return -1
+endi
+if $data02 != 3.000000000 then 
+  return -1
+endi
+if $data10 != 11 then 
+  return -1
+endi
+if $data11 != 2.10000 then 
+  return -1
+endi
+if $data12 != 3.100000000 then 
+  return -1
+endi
+if $data30 != -13 then 
+  return -1
+endi
+if $data31 != -2.30000 then 
+  return -1
+endi
+if $data32 != -3.300000000 then 
+  return -1
+endi
 #===================================================================
 #===================================================================
 
@@ -202,17 +205,17 @@ print $data00 $data01 $data02
 #if $rows != 4 then 
 #  return -1
 #endi
+
 #print =============== select count(*) from supter table
 #sql select count(*) from stb
+#print $data00 $data01 $data02
 #if $rows != 1 then 
 #  return -1
 #endi
-#
-#print $data00 $data01 $data02
-#if $data00 != 8 then 
+#if $data00 != 9 then 
 #  return -1
 #endi
-#
+
 #print =============== select count(column) from supter table
 #sql select count(ts), count(c1), count(c2), count(c3) from stb
 #print $data00 $data01 $data02  $data03
@@ -237,9 +240,31 @@ print =============== stop and restart taosd, then again do query above
 system sh/exec.sh -n dnode1 -s stop -x SIGINT
 system sh/exec.sh -n dnode1 -s start
 
-sleep 2000
+$loop_cnt = 0
+check_dnode_ready:
+	$loop_cnt = $loop_cnt + 1
+	sleep 200
+	if $loop_cnt == 10 then
+	  print ====> dnode not ready!
+		return -1
+	endi
+sql show dnodes
+print ===> $rows $data00 $data01 $data02 $data03 $data04 $data05
+if $data00 != 1 then
+  return -1
+endi
+if $data04 != ready then
+  goto check_dnode_ready
+endi
+
+print =============== query data from child table
 sql select * from ct1
-if $rows != 7 then
+print rows: $rows 
+print $data00  $data01 $data02 $data03
+print $data10  $data11 $data12 $data13
+print $data20  $data21 $data22 $data23
+print $data30  $data31 $data32 $data33
+if $rows != 4 then
   return -1
 endi
 if $data01 != 10 then 
@@ -264,28 +289,29 @@ endi
 
 print =============== select count(*) from child table
 sql select count(*) from ct1
+print rows: $rows 
+print $data00  $data01 $data02 $data03
 if $rows != 1 then 
   return -1
 endi
-
-print $data00 $data01 $data02
-if $data00 != 7 then 
+if $data00 != 4 then 
   return -1
 endi
 
 print =============== select count(column) from child table
 sql select count(ts), count(c1), count(c2), count(c3) from ct1
+print rows: $rows 
 print $data00  $data01 $data02 $data03
-if $data00 != 7 then 
+if $data00 != 4 then 
   return -1
 endi
-if $data01 != 7 then 
+if $data01 != 4 then 
   return -1
 endi
-if $data02 != 7 then 
+if $data02 != 4 then 
   return -1
 endi
-if $data03 != 7 then 
+if $data03 != 4 then 
   return -1
 endi
 
@@ -295,17 +321,18 @@ endi
 
 print =============== select min(column) from child table
 sql select min(c1), min(c2), min(c3) from ct1
+print rows: $rows 
 print $data00  $data01 $data02 $data03
 if $rows != 1 then 
   return -1
 endi
-if $data00 != -16 then 
+if $data00 != -13 then 
   return -1
 endi
-if $data01 != -2.60000 then 
+if $data01 != -2.30000 then 
   return -1
 endi
-if $data02 != -3.600000000 then 
+if $data02 != -3.300000000 then 
   return -1
 endi
 
@@ -315,13 +342,13 @@ print $data00 $data01 $data02  $data03
 if $rows != 1 then 
   return -1
 endi
-if $data00 != 13 then 
+if $data00 != 11 then 
   return -1
 endi
-if $data01 != 2.30000 then 
+if $data01 != 2.10000 then 
   return -1
 endi
-if $data02 != 3.300000000 then 
+if $data02 != 3.100000000 then 
   return -1
 endi
 
@@ -331,14 +358,87 @@ print $data00 $data01 $data02  $data03
 if $rows != 1 then 
   return -1
 endi
-if $data00 != 1 then 
+if $data00 != -4 then 
+  return -1
+endi
+if $data01 != -0.400000095 then 
+  return -1
+endi
+if $data02 != -0.400000000 then 
+  return -1
+endi
+
+print =============== select column without timestamp, from child table
+sql select c1, c2, c3 from ct1
+print rows: $rows
+print $data00 $data01 $data02
+print $data10 $data11 $data12
+print $data20 $data21 $data22
+print $data30 $data31 $data32
+if $rows != 4 then 
   return -1
 endi
-if $data01 != 1.099999905 then 
+if $data00 != 10 then 
   return -1
 endi
-if $data02 != 2.100000000 then 
+if $data01 != 2.00000 then 
   return -1
 endi
+if $data02 != 3.000000000 then 
+  return -1
+endi
+if $data10 != 11 then 
+  return -1
+endi
+if $data11 != 2.10000 then 
+  return -1
+endi
+if $data12 != 3.100000000 then 
+  return -1
+endi
+if $data30 != -13 then 
+  return -1
+endi
+if $data31 != -2.30000 then 
+  return -1
+endi
+if $data32 != -3.300000000 then 
+  return -1
+endi
+#===================================================================
+#===================================================================
+
+#print =============== query data from stb
+#sql select * from stb
+#if $rows != 4 then 
+#  return -1
+#endi      
+
+#print =============== select count(*) from supter table
+#sql select count(*) from stb
+#print $data00 $data01 $data02
+#if $rows != 1 then 
+#  return -1
+#endi
+#if $data00 != 9 then 
+#  return -1
+#endi
+
+#print =============== select count(column) from supter table
+#sql select count(ts), count(c1), count(c2), count(c3) from stb
+#print $data00 $data01 $data02  $data03
+#if $data00 != 8 then 
+#  return -1
+#endi
+#if $data01 != 8 then 
+#  return -1
+#endi
+#if $data02 != 8 then 
+#  return -1
+#endi
+#if $data03 != 8 then 
+#  return -1
+#endi
+
 
 #system sh/exec.sh -n dnode1 -s stop -x SIGINT

tests/script/tsim/insert/null.sim

@@ -7,7 +7,7 @@ sql connect
 print =============== create database
 sql create database d0
 sql show databases
-if $rows != 1 then 
+if $rows != 2 then 
   return -1
 endi
 
@@ -40,7 +40,7 @@ sql insert into ct1 values (now+7s, 14, NULL, 3.5, 95)
 sql insert into ct1 values (now+8s, 15, 2.5, NULL, 96)
 sql insert into ct1 values (now+9s, 16, 2.6, 3.6, NULL)
 sql insert into ct1 values (now+10s, NULL, NULL, NULL, NULL)
-sql insert into ct1 values (now+11s, -2147483648, 2.7, 3.7, 97)
+sql insert into ct1 values (now+11s, -2147483647, 2.7, 3.7, 97)
 
 #===================================================================
 #===================================================================
@@ -75,7 +75,6 @@ endi
 #  return -1
 #endi
 
-
 print =============== select count(*) from child table
 sql select count(*) from ct1
 print ===> select count(*) from ct1
@@ -84,9 +83,7 @@ print ===> rows0: $data00 $data01 $data02 $data03 $data04
 if $rows != 1 then 
   return -1
 endi
-
-print $data00 $data01 $data02
-if $data00 != 4 then 
+if $data00 != 12 then 
   return -1
 endi
 
@@ -95,17 +92,16 @@ sql select count(ts), count(c1), count(c2), count(c3) from ct1
 print ===> select count(ts), count(c1), count(c2), count(c3) from ct1
 print ===> rows: $rows
 print ===> rows0: $data00 $data01 $data02 $data03 $data04
-
-if $data00 != 4 then 
+if $data00 != 12 then 
   return -1
 endi
-if $data01 != 4 then 
+if $data01 != 8 then 
   return -1
 endi
-if $data02 != 4 then 
+if $data02 != 8 then 
   return -1
 endi
-if $data03 != 4 then 
+if $data03 != 8 then 
   return -1
 endi
 
@@ -121,7 +117,7 @@ print ===> rows0: $data00 $data01 $data02 $data03 $data04
 if $rows != 1 then 
   return -1
 endi
-if $data00 != 10 then 
+if $data00 != -2147483647 then 
   return -1
 endi
 if $data01 != 2.00000 then 
@@ -139,13 +135,13 @@ print ===> rows0: $data00 $data01 $data02 $data03 $data04
 if $rows != 1 then 
   return -1
 endi
-if $data00 != 13 then 
+if $data00 != 16 then 
   return -1
 endi
-if $data01 != 2.30000 then 
+if $data01 != 2.70000 then 
   return -1
 endi
-if $data02 != 3.300000000 then 
+if $data02 != 3.700000000 then 
   return -1
 endi
 
@@ -157,13 +153,13 @@ print ===> rows0: $data00 $data01 $data02 $data03 $data04
 if $rows != 1 then 
   return -1
 endi
-if $data00 != 46 then 
+if $data00 != -2147483556  then 
   return -1
 endi
-if $data01 != 8.599999905 then 
+if $data01 != 18.799999952 then 
   return -1
 endi
-if $data02 != 12.600000000 then 
+if $data02 != 26.800000000 then 
   return -1
 endi
 
@@ -172,42 +168,48 @@ sql select c1, c2, c3 from ct1
 print ===> select c1, c2, c3 from ct1
 print ===> rows: $rows
 print ===> rows0: $data00 $data01 $data02 $data03 $data04
-#if $rows != 4 then 
-#  return -1
-#endi
-#if $data00 != 10 then 
-#  return -1
-#endi
-#if $data01 != 2.00000 then 
-#  return -1
-#endi
-#if $data02 != 3.000000000 then 
-#  return -1
-#endi
-#if $data10 != 11 then 
-#  return -1
-#endi
-#if $data11 != 2.10000 then 
-#  return -1
-#endi
-#if $data12 != 3.100000000 then 
-#  return -1
-#endi
-#if $data30 != 13 then 
-#  return -1
-#endi
-#if $data31 != 2.30000 then 
-#  return -1
-#endi
-#if $data32 != 3.300000000 then 
-#  return -1
-#endi
+if $rows != 12 then 
+  return -1
+endi
+if $data00 != 10 then 
+  return -1
+endi
+if $data01 != 2.00000 then 
+  return -1
+endi
+if $data02 != 3.000000000 then 
+  return -1
+endi
+if $data10 != NULL then 
+  return -1
+endi
+if $data11 != NULL then 
+  return -1
+endi
+if $data12 != NULL then 
+  return -1
+endi
+if $data30 != 11 then 
+  return -1
+endi
+if $data31 != NULL then 
+  return -1
+endi
+if $data32 != 3.200000000 then 
+  return -1
+endi
+if $data90 != 16 then 
+  return -1
+endi
+if $data91 != 2.60000 then 
+  return -1
+endi
+if $data92 != 3.600000000 then 
+  return -1
+endi
 #===================================================================
 #===================================================================
 
-
-return 
-
 #print =============== query data from stb
 #sql select * from stb
 #print ===> 
@@ -218,19 +220,18 @@ return
 #endi
 #print =============== select count(*) from supter table
 #sql select count(*) from stb
+#print $data00 $data01 $data02
 #if $rows != 1 then 
 #  return -1
 #endi
-#
-#print $data00 $data01 $data02
-#if $data00 != 8 then 
+#if $data00 != 12 then 
 #  return -1
 #endi
-#
+
 #print =============== select count(column) from supter table
 #sql select count(ts), count(c1), count(c2), count(c3) from stb
 #print $data00 $data01 $data02  $data03
-#if $data00 != 8 then 
+#if $data00 != 12 then 
 #  return -1
 #endi
 #if $data01 != 8 then 
@@ -243,6 +244,7 @@ return
 #  return -1
 #endi
 
+return
 
 #===================================================================
 #===================================================================
@@ -251,9 +253,36 @@ print =============== stop and restart taosd, then again do query above
 system sh/exec.sh -n dnode1 -s stop -x SIGINT
 system sh/exec.sh -n dnode1 -s start
 
-sleep 2000
+print ===> waiting dnode ready
+$loop_cnt = 0
+check_dnode_ready:
+	$loop_cnt = $loop_cnt + 1
+	sleep 200
+	if $loop_cnt == 10 then
+	  print ====> dnode not ready!
+		return -1
+	endi
+sql show dnodes
+print ===> $rows $data00 $data01 $data02 $data03 $data04 $data05
+if $data00 != 1 then
+  return -1
+endi
+if $data04 != ready then
+  goto check_dnode_ready
+endi
+
+#===================================================================
+#===================================================================
+print =============== query data from child table
 sql select * from ct1
-if $rows != 4 then         # after fix bug, modify 4 to 7
+print ===> select * from ct1
+print ===> rows: $rows
+print ===> rows0: $data00 $data01 $data02 $data03 $data04
+print ===> rows1: $data10 $data11 $data12 $data13 $data14
+print ===> rows2: $data20 $data21 $data22 $data23 $data24
+print ===> rows3: $data30 $data31 $data32 $data33 $data34
+print ===> rows4: $data40 $data41 $data42 $data43 $data44
+if $rows != 12 then
   return -1
 endi
 if $data01 != 10 then 
@@ -275,31 +304,33 @@ endi
 #  return -1
 #endi
 
-
 print =============== select count(*) from child table
 sql select count(*) from ct1
+print ===> select count(*) from ct1
+print ===> rows: $rows
+print ===> rows0: $data00 $data01 $data02 $data03 $data04
 if $rows != 1 then 
   return -1
 endi
-
-print $data00 $data01 $data02
-if $data00 != 4 then 
+if $data00 != 12 then 
   return -1
 endi
 
 print =============== select count(column) from child table
 sql select count(ts), count(c1), count(c2), count(c3) from ct1
-print $data00 $data01 $data02  $data03
-if $data00 != 4 then 
+print ===> select count(ts), count(c1), count(c2), count(c3) from ct1
+print ===> rows: $rows
+print ===> rows0: $data00 $data01 $data02 $data03 $data04
+if $data00 != 12 then 
   return -1
 endi
-if $data01 != 4 then 
+if $data01 != 8 then 
   return -1
 endi
-if $data02 != 4 then 
+if $data02 != 8 then 
   return -1
 endi
-if $data03 != 4 then 
+if $data03 != 8 then 
   return -1
 endi
 
@@ -309,11 +340,13 @@ endi
 
 print =============== select min(column) from child table
 sql select min(c1), min(c2), min(c3) from ct1
-print $data00 $data01 $data02  $data03
+print ===> select min(c1), min(c2), min(c3) from ct1
+print ===> rows: $rows
+print ===> rows0: $data00 $data01 $data02 $data03 $data04
 if $rows != 1 then 
   return -1
 endi
-if $data00 != 10 then 
+if $data00 != -2147483647 then 
   return -1
 endi
 if $data01 != 2.00000 then 
@@ -325,34 +358,119 @@ endi
 
 print =============== select max(column) from child table
 sql select max(c1), max(c2), max(c3) from ct1
-print $data00 $data01 $data02  $data03
+print ===> select max(c1), max(c2), max(c3) from ct1
+print ===> rows: $rows
+print ===> rows0: $data00 $data01 $data02 $data03 $data04
 if $rows != 1 then 
   return -1
 endi
-if $data00 != 13 then 
+if $data00 != 16 then 
   return -1
 endi
-if $data01 != 2.30000 then 
+if $data01 != 2.70000 then 
   return -1
 endi
-if $data02 != 3.300000000 then 
+if $data02 != 3.700000000 then 
   return -1
 endi
 
 print =============== select sum(column) from child table
 sql select sum(c1), sum(c2), sum(c3) from ct1
-print $data00 $data01 $data02  $data03
+print ===> select sum(c1), sum(c2), sum(c3) from ct1
+print ===> rows: $rows
+print ===> rows0: $data00 $data01 $data02 $data03 $data04
 if $rows != 1 then 
   return -1
 endi
-if $data00 != 46 then 
+if $data00 != -2147483556  then 
+  return -1
+endi
+if $data01 != 18.799999952 then 
+  return -1
+endi
+if $data02 != 26.800000000 then 
+  return -1
+endi
+
+print =============== select column, from child table
+sql select c1, c2, c3 from ct1
+print ===> select c1, c2, c3 from ct1
+print ===> rows: $rows
+print ===> rows0: $data00 $data01 $data02 $data03 $data04
+if $rows != 12 then 
+  return -1
+endi
+if $data00 != 10 then 
+  return -1
+endi
+if $data01 != 2.00000 then 
+  return -1
+endi
+if $data02 != 3.000000000 then 
+  return -1
+endi
+if $data10 != NULL then 
+  return -1
+endi
+if $data11 != NULL then 
+  return -1
+endi
+if $data12 != NULL then 
   return -1
 endi
-if $data01 != 8.599999905 then 
+if $data30 != 11 then 
   return -1
 endi
-if $data02 != 12.600000000 then 
+if $data31 != NULL then 
   return -1
 endi
+if $data32 != 3.200000000 then 
+  return -1
+endi
+if $data90 != 16 then 
+  return -1
+endi
+if $data91 != 2.60000 then 
+  return -1
+endi
+if $data92 != 3.600000000 then 
+  return -1
+endi
+#===================================================================
+#===================================================================
+
+#print =============== query data from stb
+#sql select * from stb
+#print ===> 
+#print ===> rows: $rows
+#print ===> rows0: $data00 $data01 $data02 $data03 $data04
+#if $rows != 4 then 
+#  return -1
+#endi
+#print =============== select count(*) from supter table
+#sql select count(*) from stb
+#print $data00 $data01 $data02
+#if $rows != 1 then 
+#  return -1
+#endi
+#if $data00 != 12 then 
+#  return -1
+#endi
+
+#print =============== select count(column) from supter table
+#sql select count(ts), count(c1), count(c2), count(c3) from stb
+#print $data00 $data01 $data02  $data03
+#if $data00 != 12 then 
+#  return -1
+#endi
+#if $data01 != 8 then 
+#  return -1
+#endi
+#if $data02 != 8 then 
+#  return -1
+#endi
+#if $data03 != 8 then 
+#  return -1
+#endi
 
 #system sh/exec.sh -n dnode1 -s stop -x SIGINT

tests/script/tsim/table/basic1.sim

@@ -3,10 +3,33 @@ system sh/deploy.sh -n dnode1 -i 1
 system sh/exec.sh -n dnode1 -s start
 sql connect
 
+#=========== TD-14042 start
+sql create database db1;
+sql create database db2;
+sql create database db3;
+
+sql use db1;
+sql create table st1 (ts timestamp, i int) tags (j int);
+sql create table tb1 using st1 tags(1);
+
+sql use db2;
+sql create table st2 (ts timestamp, i int) tags (j int);
+sql create table tb2 using st2 tags(1);
+
+sql use db3;
+sql create table st3 (ts timestamp, i int) tags (j int);
+sql create table tb3 using st3 tags(1);
+
+sql show tables
+if $rows != 1 then 
+  return -1
+endi
+#=========== TD-14042 end
+
 print =============== create database
 sql create database d1
 sql show databases
-if $rows != 2 then 
+if $rows != 5 then 
   return -1
 endi
 
@@ -186,10 +209,26 @@ if $rows != 21 then
   return -1
 endi
 
-#system sh/exec.sh -n dnode1 -s stop -x SIGINT
-#system sh/exec.sh -n dnode1 -s start
+system sh/exec.sh -n dnode1 -s stop -x SIGINT
+system sh/exec.sh -n dnode1 -s start
+
+$loop_cnt = 0
+check_dnode_ready:
+	$loop_cnt = $loop_cnt + 1
+	sleep 200
+	if $loop_cnt == 10 then
+	  print ====> dnode not ready!
+		return -1
+	endi
+sql show dnodes
+print ===> $rows $data00 $data01 $data02 $data03 $data04 $data05
+if $data00 != 1 then
+  return -1
+endi
+if $data04 != ready then
+  goto check_dnode_ready
+endi
 
-sleep 2000
 print =============== query data
 sql select * from c1
 if $rows != 3 then 

tests/script/tsim/testCaseSuite.sim

 
+run tsim/user/basic1.sim
+
 run tsim/db/basic1.sim
 run tsim/db/basic6.sim
 run tsim/db/basic7.sim
@@ -7,12 +9,14 @@ run tsim/db/error1.sim
 run tsim/dnode/basic1.sim
 
 run tsim/insert/basic0.sim
-run tsim/insert/basic1.sim
-run tsim/insert/null.sim
+#run tsim/insert/basic1.sim      # TD-14246
+#run tsim/insert/backquote.sim    # TD-14261
+#run tsim/insert/null.sim
 
-run tsim/query/interval-offset.sim
 run tsim/query/interval.sim
+#run tsim/query/interval-offset.sim  # TD-14266
 
 run tsim/table/basic1.sim
 
-run tsim/user/basic1.sim
+run tsim/tmq/basic.sim
+

tools/CMakeLists.txt

@@ -4,6 +4,7 @@ IF (TD_TAOS_TOOLS)
     INCLUDE_DIRECTORIES(${CMAKE_SOURCE_DIR}/include/common)
     INCLUDE_DIRECTORIES(${CMAKE_SOURCE_DIR}/include/util)
     INCLUDE_DIRECTORIES(${CMAKE_SOURCE_DIR}/include/os)
+    INCLUDE_DIRECTORIES(${CMAKE_SOURCE_DIR}/include/libs/transport)
     ADD_SUBDIRECTORY(taos-tools)
 ENDIF ()
 

tools/shell/CMakeLists.txt

@@ -4,9 +4,7 @@ add_executable(shell ${SHELL_SRC})
 target_link_libraries(
     shell
     PUBLIC taos
-    PUBLIC util
-    PUBLIC common
-    PUBLIC os
+    PRIVATE os common transport util
 )
 target_include_directories(
     shell

tools/shell/inc/syncMsg.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 TDENGINE_SYNC_MSG_H
+#define TDENGINE_SYNC_MSG_H
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+#include "tsync.h"
+
+typedef enum {
+  TAOS_SMSG_START         = 0,
+  TAOS_SMSG_SYNC_DATA     = 1,
+  TAOS_SMSG_SYNC_DATA_RSP = 2,
+  TAOS_SMSG_SYNC_FWD      = 3,
+  TAOS_SMSG_SYNC_FWD_RSP  = 4,
+  TAOS_SMSG_SYNC_REQ      = 5,
+  TAOS_SMSG_SYNC_REQ_RSP  = 6,
+  TAOS_SMSG_SYNC_MUST     = 7,
+  TAOS_SMSG_SYNC_MUST_RSP = 8,
+  TAOS_SMSG_STATUS        = 9,
+  TAOS_SMSG_STATUS_RSP    = 10,
+  TAOS_SMSG_SETUP         = 11,
+  TAOS_SMSG_SETUP_RSP     = 12,
+  TAOS_SMSG_SYNC_FILE     = 13,
+  TAOS_SMSG_SYNC_FILE_RSP = 14,
+  TAOS_SMSG_TEST          = 15,
+  TAOS_SMSG_END           = 16
+} ESyncMsgType;
+
+typedef enum {
+  SYNC_STATUS_BROADCAST,
+  SYNC_STATUS_BROADCAST_RSP,
+  SYNC_STATUS_SETUP_CONN,
+  SYNC_STATUS_SETUP_CONN_RSP,
+  SYNC_STATUS_EXCHANGE_DATA,
+  SYNC_STATUS_EXCHANGE_DATA_RSP,
+  SYNC_STATUS_CHECK_ROLE,
+  SYNC_STATUS_CHECK_ROLE_RSP
+} ESyncStatusType;
+
+#pragma pack(push, 1)
+
+typedef struct {
+  int8_t   type;       // msg type
+  int8_t   protocol;   // protocol version
+  uint16_t signature;  // fixed value
+  int32_t  code;       //
+  int32_t  cId;        // cluster Id
+  int32_t  vgId;       // vg ID
+  int32_t  len;        // content length, does not include head
+  uint32_t cksum;
+} SSyncHead;
+
+typedef struct {
+  SSyncHead head;
+  uint16_t  port;
+  uint16_t  tranId;
+  int32_t   sourceId;  // only for arbitrator
+  char      fqdn[TSDB_FQDN_LEN];
+} SSyncMsg;
+
+typedef struct {
+  SSyncHead head;
+  int8_t    sync;
+  int8_t    reserved;
+  uint16_t  tranId;
+  int8_t    reserverd[4];
+} SSyncRsp;
+
+typedef struct {
+  int8_t    role;
+  uint64_t  version;
+} SPeerStatus;
+
+typedef struct {
+  SSyncHead   head;
+  int8_t      role;
+  int8_t      ack;
+  int8_t      type;
+  int8_t      reserved[3];
+  uint16_t    tranId;
+  uint64_t    version;
+  SPeerStatus peersStatus[TAOS_SYNC_MAX_REPLICA];
+} SPeersStatus;
+
+typedef struct {
+  SSyncHead head;
+  uint64_t  fversion;
+} SFileVersion;
+
+typedef struct {
+  SSyncHead head;
+  int8_t    ack;
+} SFileAck;
+
+typedef struct {
+  SSyncHead head;
+  uint64_t  version;
+  int32_t   code;
+} SFwdRsp;
+
+#pragma pack(pop)
+
+#define SYNC_PROTOCOL_VERSION 1
+#define SYNC_SIGNATURE ((uint16_t)(0xCDEF))
+
+extern char *statusType[];
+
+uint16_t syncGenTranId();
+int32_t  syncCheckHead(SSyncHead *pHead);
+
+void syncBuildSyncFwdMsg(SSyncHead *pHead, int32_t vgId, int32_t len);
+void syncBuildSyncFwdRsp(SFwdRsp *pMsg, int32_t vgId, uint64_t version, int32_t code);
+void syncBuildSyncReqMsg(SSyncMsg *pMsg, int32_t vgId);
+void syncBuildSyncDataMsg(SSyncMsg *pMsg, int32_t vgId);
+void syncBuildSyncSetupMsg(SSyncMsg *pMsg, int32_t vgId);
+void syncBuildPeersStatus(SPeersStatus *pMsg, int32_t vgId);
+void syncBuildSyncTestMsg(SSyncMsg *pMsg, int32_t vgId);
+
+void syncBuildFileAck(SFileAck *pMsg, int32_t vgId);
+void syncBuildFileVersion(SFileVersion *pMsg, int32_t vgId);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif  // TDENGINE_VNODEPEER_H

tools/shell/inc/tsync.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 TDENGINE_SYNC_H
+#define TDENGINE_SYNC_H
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#define TAOS_SYNC_MAX_REPLICA 5
+#define TAOS_SYNC_MAX_INDEX   0x7FFFFFFF
+
+typedef enum {
+  TAOS_SYNC_ROLE_OFFLINE  = 0,
+  TAOS_SYNC_ROLE_UNSYNCED = 1,
+  TAOS_SYNC_ROLE_SYNCING  = 2,
+  TAOS_SYNC_ROLE_SLAVE    = 3,
+  TAOS_SYNC_ROLE_MASTER   = 4
+} ESyncRole;
+
+typedef enum {
+  TAOS_SYNC_STATUS_INIT  = 0,
+  TAOS_SYNC_STATUS_START = 1,
+  TAOS_SYNC_STATUS_FILE  = 2,
+  TAOS_SYNC_STATUS_CACHE = 3
+} ESyncStatus;
+
+typedef struct {
+  uint32_t  nodeId;    // node ID assigned by TDengine
+  uint16_t  nodePort;  // node sync Port
+  char      nodeFqdn[TSDB_FQDN_LEN]; // node FQDN  
+} SNodeInfo;
+
+typedef struct {
+  int8_t     quorum;    // number of confirms required, >=1 
+  int8_t     replica;   // number of replications, >=1
+  SNodeInfo  nodeInfo[TAOS_SYNC_MAX_REPLICA];
+} SSyncCfg;
+
+typedef struct {
+  int32_t  selfIndex;
+  uint32_t nodeId[TAOS_SYNC_MAX_REPLICA];
+  int32_t  role[TAOS_SYNC_MAX_REPLICA];
+} SNodesRole;
+
+// get the wal file from index or after
+// return value, -1: error, 1:more wal files, 0:last WAL. if name[0]==0, no WAL file
+typedef int32_t  (*FGetWalInfo)(int32_t vgId, char *fileName, int64_t *fileId); 
+ 
+// when a forward pkt is received, call this to handle data
+typedef int32_t  (*FWriteToCache)(int32_t vgId, void *pHead, int32_t qtype, void *pMsg);
+
+// when forward is confirmed by peer, master call this API to notify app
+typedef void     (*FConfirmForward)(int32_t vgId, void *mhandle, int32_t code);
+
+// when role is changed, call this to notify app
+typedef void     (*FNotifyRole)(int32_t vgId, int8_t role);
+
+// if a number of retrieving data failed, call this to start flow control 
+typedef void     (*FNotifyFlowCtrl)(int32_t vgId, int32_t level);
+
+// when data file is synced successfully, notity app
+typedef void     (*FStartSyncFile)(int32_t vgId);
+typedef void     (*FStopSyncFile)(int32_t vgId, uint64_t fversion);
+
+// get file version
+typedef int32_t  (*FGetVersion)(int32_t vgId, uint64_t *fver, uint64_t *vver);
+
+typedef int32_t  (*FSendFile)(void *tsdb, SOCKET socketFd);
+typedef int32_t  (*FRecvFile)(void *tsdb, SOCKET socketFd);
+
+typedef struct {
+  int32_t  vgId;       // vgroup ID
+  uint64_t version;    // initial version
+  SSyncCfg syncCfg;    // configuration from mgmt
+  char     path[TSDB_FILENAME_LEN];  // path to the file
+  void *   pTsdb;
+  FGetWalInfo       getWalInfoFp;
+  FWriteToCache     writeToCacheFp;
+  FConfirmForward   confirmForward;
+  FNotifyRole       notifyRoleFp;
+  FNotifyFlowCtrl   notifyFlowCtrlFp;
+  FStartSyncFile    startSyncFileFp;
+  FStopSyncFile     stopSyncFileFp;
+  FGetVersion       getVersionFp;
+  FSendFile         sendFileFp;
+  FRecvFile         recvFileFp;
+} SSyncInfo;
+
+typedef void *tsync_h;
+
+int32_t syncInit();
+void    syncCleanUp();
+
+int64_t syncStart(const SSyncInfo *);
+void    syncStop(int64_t rid);
+int32_t syncReconfig(int64_t rid, const SSyncCfg *);
+int32_t syncForwardToPeer(int64_t rid, void *pHead, void *mhandle, int32_t qtype, bool force);
+void    syncConfirmForward(int64_t rid, uint64_t version, int32_t code, bool force);
+void    syncRecover(int64_t rid);  // recover from other nodes:
+int32_t syncGetNodesRole(int64_t rid, SNodesRole *);
+
+extern char *syncRole[];
+
+//global configurable parameters
+extern int32_t  sDebugFlag;
+extern char     tsArbitrator[];
+extern uint16_t tsSyncPort;
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif  // TDENGINE_SYNC_H

tools/shell/src/shellMain.c

@@ -32,6 +32,8 @@
 int indicator = 1;
 
 void insertChar(Command *cmd, char *c, int size);
+void taosNetTest(char *role, char *host, int32_t port, int32_t pkgLen,
+                 int32_t pkgNum, char *pkgType);
 const char *argp_program_version = version;
 const char *argp_program_bug_address = "<support@taosdata.com>";
 static char doc[] = "";
@@ -59,7 +61,8 @@ static struct argp_option options[] = {
   {"netrole",    'n', "NETROLE",    0,                   "Net role when network connectivity test, default is startup, options: client|server|rpc|startup|sync|speen|fqdn."},
   {"pktlen",     'l', "PKTLEN",     0,                   "Packet length used for net test, default is 1000 bytes."},
   {"pktnum",     'N', "PKTNUM",     0,                   "Packet numbers used for net test, default is 100."},
-  {"pkttype",    'S', "PKTTYPE",    0,                   "Packet type used for net test, default is TCP."},
+// Shuduo: 3.0 does not support UDP any more
+//  {"pkttype",    'S', "PKTTYPE",    0,                   "Packet type used for net test, default is TCP."},
   {0}};
 
 static error_t parse_opt(int key, char *arg, struct argp_state *state) {
@@ -629,16 +632,25 @@ int main(int argc, char *argv[]) {
     taosDumpGlobalCfg();
     exit(0);
   }
+#endif
 
   if (args.netTestRole && args.netTestRole[0] != 0) {
-    if (taos_init()) {
+    TAOS *con = NULL;
+    if (args.auth == NULL) {
+      con = taos_connect(args.host, args.user, args.password, args.database, args.port);
+    } else {
+      con = taos_connect_auth(args.host, args.user, args.auth, args.database, args.port);
+    }
+
+/*    if (taos_init()) {
       printf("Failed to init taos");
       exit(EXIT_FAILURE);
     }
+    */
     taosNetTest(args.netTestRole, args.host, args.port, args.pktLen, args.pktNum, args.pktType);
+    taos_close(con);
     exit(0);
   }
-#endif
 
   /* Initialize the shell */
   TAOS *con = shellInit(&args);

tools/shell/src/backup/tnettest.c → tools/shell/src/tnettest.c

@@ -14,18 +14,20 @@
  */
 
 #define _DEFAULT_SOURCE
+#define ALLOW_FORBID_FUNC
 #include "os.h"
 #include "taosdef.h"
 #include "tmsg.h"
 #include "taoserror.h"
 #include "tlog.h"
 #include "tglobal.h"
-#include "tsocket.h"
 #include "trpc.h"
 #include "rpcHead.h"
 #include "tchecksum.h"
 #include "syncMsg.h"
 
+#include "osSocket.h"
+
 #define MAX_PKG_LEN (64 * 1000)
 #define MAX_SPEED_PKG_LEN (1024 * 1024 * 1024)
 #define MIN_SPEED_PKG_LEN 1024
@@ -33,7 +35,7 @@
 #define MIN_SPEED_PKG_NUM 1
 #define BUFFER_SIZE (MAX_PKG_LEN + 1024)
 
-extern int32_t tsRpcMaxUdpSize;
+extern int tsRpcMaxUdpSize;
 
 typedef struct {
   char *   hostFqdn;
@@ -71,15 +73,23 @@ static void *taosNetBindUdpPort(void *sarg) {
     return NULL;
   }
 
-  if (taosSetSockOpt(serverSocket, SOL_SOCKET, SO_SNDBUF, (void *)&bufSize, sizeof(bufSize)) != 0) {
+  TdSocketPtr pSocket = (TdSocketPtr)malloc(sizeof(TdSocket));
+  if (pSocket == NULL) {
+    taosCloseSocketNoCheck1(serverSocket);
+    return NULL;
+  }
+  pSocket->fd = serverSocket;
+  pSocket->refId = 0;
+
+  if (taosSetSockOpt(pSocket, SOL_SOCKET, SO_SNDBUF, (void *)&bufSize, sizeof(bufSize)) != 0) {
     uError("failed to set the send buffer size for UDP socket\n");
-    taosCloseSocket(serverSocket);
+    taosCloseSocket(&pSocket);
     return NULL;
   }
 
-  if (taosSetSockOpt(serverSocket, SOL_SOCKET, SO_RCVBUF, (void *)&bufSize, sizeof(bufSize)) != 0) {
+  if (taosSetSockOpt(pSocket, SOL_SOCKET, SO_RCVBUF, (void *)&bufSize, sizeof(bufSize)) != 0) {
     uError("failed to set the receive buffer size for UDP socket\n");
-    taosCloseSocket(serverSocket);
+    taosCloseSocket(&pSocket);
     return NULL;
   }
 
@@ -98,13 +108,13 @@ static void *taosNetBindUdpPort(void *sarg) {
     uInfo("UDP: recv:%d bytes from %s at %d", iDataNum, taosInetNtoa(clientAddr.sin_addr), port);
 
     if (iDataNum > 0) {
-      iDataNum = taosSendto(serverSocket, buffer, iDataNum, 0, (struct sockaddr *)&clientAddr, (int32_t)sin_size);
+      iDataNum = taosSendto(pSocket, buffer, iDataNum, 0, (struct sockaddr *)&clientAddr, (int32_t)sin_size);
     }
 
     uInfo("UDP: send:%d bytes to %s at %d", iDataNum, taosInetNtoa(clientAddr.sin_addr), port);
   }
 
-  taosCloseSocket(serverSocket);
+  taosCloseSocket(&pSocket);
   return NULL;
 }
 
@@ -132,25 +142,35 @@ static void *taosNetBindTcpPort(void *sarg) {
   server_addr.sin_addr.s_addr = htonl(INADDR_ANY);
 
   int32_t reuse = 1;
-  if (taosSetSockOpt(serverSocket, SOL_SOCKET, SO_REUSEADDR, (void *)&reuse, sizeof(reuse)) < 0) {
+  TdSocketPtr pSocket = (TdSocketPtr)malloc(sizeof(TdSocket));
+  if (pSocket == NULL) {
+    taosCloseSocketNoCheck1(serverSocket);
+    return NULL;
+  }
+  pSocket->fd = serverSocket;
+  pSocket->refId = 0;
+
+  if (taosSetSockOpt(pSocket, SOL_SOCKET, SO_REUSEADDR, (void *)&reuse, sizeof(reuse)) < 0) {
     uError("setsockopt SO_REUSEADDR failed: %d (%s)", errno, strerror(errno));
-    taosCloseSocket(serverSocket);
+    taosCloseSocket(&pSocket);
     return NULL;
   }
 
   if (bind(serverSocket, (struct sockaddr *)&server_addr, sizeof(server_addr)) < 0) {
     uError("failed to bind TCP port:%d since %s", port, strerror(errno));
+    taosCloseSocket(&pSocket);
     return NULL;
   }
 
-  if (taosKeepTcpAlive(serverSocket) < 0) {
+  if (taosKeepTcpAlive(pSocket) < 0) {
     uError("failed to set tcp server keep-alive option since %s", strerror(errno));
-    taosCloseSocket(serverSocket);
+    taosCloseSocket(&pSocket);
     return NULL;
   }
 
   if (listen(serverSocket, 10) < 0) {
     uError("failed to listen TCP port:%d since %s", port, strerror(errno));
+    taosCloseSocket(&pSocket);
     return NULL;
   }
 
@@ -163,26 +183,26 @@ static void *taosNetBindTcpPort(void *sarg) {
       continue;
     }
 
-    int32_t ret = taosReadMsg(client, buffer, pinfo->pktLen);
+    int32_t ret = taosReadMsg(pSocket, buffer, pinfo->pktLen);
     if (ret < 0 || ret != pinfo->pktLen) {
       uError("TCP: failed to read %d bytes at port:%d since %s", pinfo->pktLen, port, strerror(errno));
-      taosCloseSocket(serverSocket);
+      taosCloseSocket(&pSocket);
       return NULL;
     }
 
     uInfo("TCP: read:%d bytes from %s at %d", pinfo->pktLen, taosInetNtoa(clientAddr.sin_addr), port);
 
-    ret = taosWriteMsg(client, buffer, pinfo->pktLen);
+    ret = taosWriteMsg(pSocket, buffer, pinfo->pktLen);
     if (ret < 0) {
       uError("TCP: failed to write %d bytes at %d since %s", pinfo->pktLen, port, strerror(errno));
-      taosCloseSocket(serverSocket);
+      taosCloseSocket(&pSocket);
       return NULL;
     }
 
     uInfo("TCP: write:%d bytes to %s at %d", pinfo->pktLen, taosInetNtoa(clientAddr.sin_addr), port);
   }
 
-  taosCloseSocket(serverSocket);
+  taosCloseSocket(&pSocket);
   return NULL;
 }
 
@@ -196,9 +216,17 @@ static int32_t taosNetCheckTcpPort(STestInfo *info) {
   }
 
   int32_t reuse = 1;
-  if (taosSetSockOpt(clientSocket, SOL_SOCKET, SO_REUSEADDR, (void *)&reuse, sizeof(reuse)) < 0) {
+  TdSocketPtr pSocket = (TdSocketPtr)malloc(sizeof(TdSocket));
+  if (pSocket == NULL) {
+    taosCloseSocketNoCheck1(clientSocket);
+    return -1;
+  }
+  pSocket->fd = clientSocket;
+  pSocket->refId = 0;
+
+  if (taosSetSockOpt(pSocket, SOL_SOCKET, SO_REUSEADDR, (void *)&reuse, sizeof(reuse)) < 0) {
     uError("setsockopt SO_REUSEADDR failed: %d (%s)", errno, strerror(errno));
-    taosCloseSocket(clientSocket);
+    taosCloseSocket(&pSocket);
     return -1;
   }
 
@@ -210,27 +238,30 @@ static int32_t taosNetCheckTcpPort(STestInfo *info) {
 
   if (connect(clientSocket, (struct sockaddr *)&serverAddr, sizeof(serverAddr)) < 0) {
     uError("TCP: failed to connect port %s:%d since %s", taosIpStr(info->hostIp), info->port, strerror(errno));
+    taosCloseSocket(&pSocket);
     return -1;
   }
 
-  taosKeepTcpAlive(clientSocket);
+  taosKeepTcpAlive(pSocket);
 
   sprintf(buffer, "client send TCP pkg to %s:%d, content: 1122334455", taosIpStr(info->hostIp), info->port);
   sprintf(buffer + info->pktLen - 16, "1122334455667788");
 
-  int32_t ret = taosWriteMsg(clientSocket, buffer, info->pktLen);
+  int32_t ret = taosWriteMsg(pSocket, buffer, info->pktLen);
   if (ret < 0) {
     uError("TCP: failed to write msg to %s:%d since %s", taosIpStr(info->hostIp), info->port, strerror(errno));
+    taosCloseSocket(&pSocket);
     return -1;
   }
 
-  ret = taosReadMsg(clientSocket, buffer, info->pktLen);
+  ret = taosReadMsg(pSocket, buffer, info->pktLen);
   if (ret < 0) {
     uError("TCP: failed to read msg from %s:%d since %s", taosIpStr(info->hostIp), info->port, strerror(errno));
+    taosCloseSocket(&pSocket);
     return -1;
   }
 
-  taosCloseSocket(clientSocket);
+  taosCloseSocket(&pSocket);
   return 0;
 }
 
@@ -247,13 +278,23 @@ static int32_t taosNetCheckUdpPort(STestInfo *info) {
     return -1;
   }
 
-  if (taosSetSockOpt(clientSocket, SOL_SOCKET, SO_SNDBUF, (void *)&bufSize, sizeof(bufSize)) != 0) {
+  TdSocketPtr pSocket = (TdSocketPtr)malloc(sizeof(TdSocket));
+  if (pSocket == NULL) {
+    taosCloseSocketNoCheck1(clientSocket);
+    return -1;
+  }
+  pSocket->fd = clientSocket;
+  pSocket->refId = 0;
+
+  if (taosSetSockOpt(pSocket, SOL_SOCKET, SO_SNDBUF, (void *)&bufSize, sizeof(bufSize)) != 0) {
     uError("failed to set the send buffer size for UDP socket\n");
+    taosCloseSocket(&pSocket);
     return -1;
   }
 
-  if (taosSetSockOpt(clientSocket, SOL_SOCKET, SO_RCVBUF, (void *)&bufSize, sizeof(bufSize)) != 0) {
+  if (taosSetSockOpt(pSocket, SOL_SOCKET, SO_RCVBUF, (void *)&bufSize, sizeof(bufSize)) != 0) {
     uError("failed to set the receive buffer size for UDP socket\n");
+    taosCloseSocket(&pSocket);
     return -1;
   }
 
@@ -268,9 +309,10 @@ static int32_t taosNetCheckUdpPort(STestInfo *info) {
 
   socklen_t sin_size = sizeof(*(struct sockaddr *)&serverAddr);
 
-  iDataNum = taosSendto(clientSocket, buffer, info->pktLen, 0, (struct sockaddr *)&serverAddr, (int32_t)sin_size);
+  iDataNum = taosSendto(pSocket, buffer, info->pktLen, 0, (struct sockaddr *)&serverAddr, (int32_t)sin_size);
   if (iDataNum < 0 || iDataNum != info->pktLen) {
     uError("UDP: failed to perform sendto func since %s", strerror(errno));
+    taosCloseSocket(&pSocket);
     return -1;
   }
 
@@ -280,10 +322,11 @@ static int32_t taosNetCheckUdpPort(STestInfo *info) {
 
   if (iDataNum < 0 || iDataNum != info->pktLen) {
     uError("UDP: received ack:%d bytes(expect:%d) from port:%d since %s", iDataNum, info->pktLen, info->port, strerror(errno));
+    taosCloseSocket(&pSocket);
     return -1;
   }
 
-  taosCloseSocket(clientSocket);
+  taosCloseSocket(&pSocket);
   return 0;
 }
 
@@ -339,7 +382,7 @@ void *taosNetInitRpc(char *secretEncrypt, char spi) {
 }
 
 static int32_t taosNetCheckRpc(const char* serverFqdn, uint16_t port, uint16_t pktLen, char spi, SStartupReq *pStep) {
-  SRpcEpSet epSet;
+  SEpSet epSet;
   SRpcMsg   reqMsg;
   SRpcMsg   rspMsg;
   void *    pRpcConn;
@@ -352,11 +395,10 @@ static int32_t taosNetCheckRpc(const char* serverFqdn, uint16_t port, uint16_t p
     return TSDB_CODE_RPC_NETWORK_UNAVAIL;
   }
 
-  memset(&epSet, 0, sizeof(SRpcEpSet));
-  epSet.inUse = 0;
+  memset(&epSet, 0, sizeof(SEpSet));
+  strcpy(epSet.eps[0].fqdn, serverFqdn);
+  epSet.eps[0].port = port;
   epSet.numOfEps = 1;
-  epSet.port[0] = port;
-  strcpy(epSet.fqdn[0], serverFqdn);
 
   reqMsg.msgType = TDMT_DND_NETWORK_TEST;
   reqMsg.pCont = rpcMallocCont(pktLen);
@@ -425,8 +467,8 @@ static void taosNetCheckSync(char *host, int32_t port) {
     return;
   }
 
-  SOCKET connFd = taosOpenTcpClientSocket(ip, (uint16_t)port, 0);
-  if (connFd < 0) {
+  TdSocketPtr pSocket = taosOpenTcpClientSocket(ip, (uint16_t)port, 0);
+  if (pSocket == NULL) {
     uError("failed to create socket while test port:%d since %s", port, strerror(errno));
     return;
   }
@@ -443,17 +485,17 @@ static void taosNetCheckSync(char *host, int32_t port) {
   pHead->len = sizeof(SSyncMsg) - sizeof(SSyncHead);
   taosCalcChecksumAppend(0, (uint8_t *)pHead, sizeof(SSyncHead));
 
-  if (taosWriteMsg(connFd, &msg, sizeof(SSyncMsg)) != sizeof(SSyncMsg)) {
+  if (taosWriteMsg(pSocket, &msg, sizeof(SSyncMsg)) != sizeof(SSyncMsg)) {
     uError("failed to test port:%d while send msg since %s", port, strerror(errno));
     return;
   }
 
-  if (taosReadMsg(connFd, &msg, sizeof(SSyncMsg)) != sizeof(SSyncMsg)) {
+  if (taosReadMsg(pSocket, &msg, sizeof(SSyncMsg)) != sizeof(SSyncMsg)) {
     uError("failed to test port:%d while recv msg since %s", port, strerror(errno));
   }
 
   uInfo("successed to test TCP port:%d", port);
-  taosCloseSocket(connFd);
+  taosCloseSocket(&pSocket);
 }
 
 static void taosNetTestRpc(char *host, int32_t startPort, int32_t pkgLen) {
@@ -494,7 +536,6 @@ static void taosNetTestRpc(char *host, int32_t startPort, int32_t pkgLen) {
   }
 
   taosNetCheckSync(host, startPort + TSDB_PORT_SYNC);
-  taosNetCheckSync(host, startPort + TSDB_PORT_ARBITRATOR);
 }
 
 static void taosNetTestClient(char *host, int32_t startPort, int32_t pkgLen) {
@@ -578,7 +619,7 @@ static void taosNetCheckSpeed(char *host, int32_t port, int32_t pkgLen,
   }
   tsCompressMsgSize = -1;
 
-  SRpcEpSet epSet;
+  SEpSet epSet;
   SRpcMsg   reqMsg;
   SRpcMsg   rspMsg;
   void *    pRpcConn;
@@ -596,11 +637,10 @@ static void taosNetCheckSpeed(char *host, int32_t port, int32_t pkgLen,
   for (int32_t i = 1; i <= pkgNum; i++) {
     uint64_t startTime = taosGetTimestampUs();
 
-    memset(&epSet, 0, sizeof(SRpcEpSet));
-    epSet.inUse = 0;
+    memset(&epSet, 0, sizeof(SEpSet));
+    strcpy(epSet.eps[0].fqdn, host);
+    epSet.eps[0].port = port;
     epSet.numOfEps = 1;
-    epSet.port[0] = port;
-    strcpy(epSet.fqdn[0], host);
 
     reqMsg.msgType = TDMT_DND_NETWORK_TEST;
     reqMsg.pCont = rpcMallocCont(pkgLen);
@@ -641,7 +681,7 @@ static void taosNetCheckSpeed(char *host, int32_t port, int32_t pkgLen,
 
 void taosNetTest(char *role, char *host, int32_t port, int32_t pkgLen,
                  int32_t pkgNum, char *pkgType) {
-  tscEmbedded = 1;
+  tsLogEmbedded = 1;
   if (host == NULL) host = tsLocalFqdn;
   if (port == 0) port = tsServerPort;
   if (0 == strcmp("speed", role)){
@@ -659,14 +699,14 @@ void taosNetTest(char *role, char *host, int32_t port, int32_t pkgLen,
   } else if (0 == strcmp("server", role)) {
     taosNetTestServer(host, port, pkgLen);
   } else if (0 == strcmp("rpc", role)) {
-    tscEmbedded = 0;
+    tsLogEmbedded = 0;
     taosNetTestRpc(host, port, pkgLen);
   } else if (0 == strcmp("sync", role)) {
     taosNetCheckSync(host, port);
   } else if (0 == strcmp("startup", role)) {
     taosNetTestStartup(host, port);
   } else if (0 == strcmp("speed", role)) {
-    tscEmbedded = 0;
+    tsLogEmbedded = 0;
     char type[10] = {0};
     taosNetCheckSpeed(host, port, pkgLen, pkgNum, strtolower(type, pkgType));
   }else if (0 == strcmp("fqdn", role)) {
@@ -675,5 +715,5 @@ void taosNetTest(char *role, char *host, int32_t port, int32_t pkgLen,
     taosNetTestStartup(host, port);
   }
 
-  tscEmbedded = 0;
+  tsLogEmbedded = 0;
 }