[td-11818] merge 3.0

include/common/tmsg.h

@@ -1232,6 +1232,12 @@ typedef struct {
   char    data[];
 } SVShowTablesFetchRsp;
 
+typedef struct SMqCMGetSubEpReq {
+  int64_t consumerId;
+  int32_t epoch;
+  char    cgroup[TSDB_CONSUMER_GROUP_LEN];
+} SMqCMGetSubEpReq;
+
 #pragma pack(pop)
 
 static FORCE_INLINE int32_t tEncodeSMsgHead(void** buf, const SMsgHead* pMsg) {
@@ -1518,9 +1524,7 @@ typedef struct SMqSetCVgReq {
   char*    sql;
   char*    logicalPlan;
   char*    physicalPlan;
-  uint32_t qmsgLen;
-  void*    qmsg;
-  //SSubQueryMsg msg;
+  char*    qmsg;
 } SMqSetCVgReq;
 
 static FORCE_INLINE int32_t tEncodeSSubQueryMsg(void** buf, const SSubQueryMsg* pMsg) {
@@ -1552,8 +1556,7 @@ static FORCE_INLINE int32_t tEncodeSMqSetCVgReq(void** buf, const SMqSetCVgReq*
   tlen += taosEncodeString(buf, pReq->sql);
   tlen += taosEncodeString(buf, pReq->logicalPlan);
   tlen += taosEncodeString(buf, pReq->physicalPlan);
-  tlen += taosEncodeFixedU32(buf, pReq->qmsgLen);
-  tlen += taosEncodeBinary(buf, pReq->qmsg, pReq->qmsgLen);
+  tlen += taosEncodeString(buf, (char*)pReq->qmsg);
   //tlen += tEncodeSSubQueryMsg(buf, &pReq->msg);
   return tlen;
 }
@@ -1567,8 +1570,7 @@ static FORCE_INLINE void* tDecodeSMqSetCVgReq(void* buf, SMqSetCVgReq* pReq) {
   buf = taosDecodeString(buf, &pReq->sql);
   buf = taosDecodeString(buf, &pReq->logicalPlan);
   buf = taosDecodeString(buf, &pReq->physicalPlan);
-  buf = taosDecodeFixedU32(buf, &pReq->qmsgLen);
-  buf = taosDecodeBinary(buf, &pReq->qmsg, pReq->qmsgLen);
+  buf = taosDecodeString(buf, (char**)&pReq->qmsg);
   //buf = tDecodeSSubQueryMsg(buf, &pReq->msg);
   return buf;
 }
@@ -1585,37 +1587,41 @@ typedef struct SMqColData {
   int16_t colId;
   int16_t type;
   int16_t bytes;
-  char    data[];
-} SMqColData;
+} SMqColMeta;
 
 typedef struct SMqTbData {
   int64_t    uid;
-  int32_t    numOfCols;
   int32_t    numOfRows;
-  SMqColData colData[];
+  char       colData[];
 } SMqTbData;
 
 typedef struct SMqTopicBlk {
-  char      topicName[TSDB_TOPIC_FNAME_LEN];
-  int64_t   committedOffset;
-  int64_t   reqOffset;
-  int64_t   rspOffset;
-  int32_t   skipLogNum;
-  int32_t   bodyLen;
-  int32_t   numOfTb;
-  SMqTbData tbData[];
+  char       topicName[TSDB_TOPIC_FNAME_LEN];
+  int64_t    committedOffset;
+  int64_t    reqOffset;
+  int64_t    rspOffset;
+  int32_t    skipLogNum;
+  int32_t    bodyLen;
+  int32_t    numOfTb;
+  SMqTbData* tbData;
 } SMqTopicData;
 
 typedef struct SMqConsumeRsp {
-  int64_t      reqId;
-  int64_t      consumerId;
-  int32_t      bodyLen;
-  int32_t      numOfTopics;
-  SMqTopicData data[];
+  int64_t       consumerId;
+  int32_t       numOfCols;
+  SMqColMeta*   meta;
+  int32_t       numOfTopics;
+  SMqTopicData* data;
 } SMqConsumeRsp;
 
+static FORCE_INLINE int32_t tEncodeSMqConsumeRsp(void** buf, const SMqConsumeRsp* pRsp) {
+  int32_t tlen = 0;
+  return tlen;
+}
+
 // one req for one vg+topic
 typedef struct SMqConsumeReq {
+  SMsgHead       head;
   //0: commit only, current offset
   //1: consume only, poll next offset
   //2: commit current and consume next offset
@@ -1630,6 +1636,92 @@ typedef struct SMqConsumeReq {
   char           topic[TSDB_TOPIC_FNAME_LEN];
 } SMqConsumeReq;
 
+typedef struct SMqSubVgEp {
+  int32_t vgId;
+  SEpSet  epSet;
+} SMqSubVgEp;
+
+typedef struct SMqSubTopicEp {
+  char    topic[TSDB_TOPIC_FNAME_LEN];
+  SArray* vgs;   // SArray<SMqSubVgEp>
+} SMqSubTopicEp;
+
+typedef struct SMqCMGetSubEpRsp {
+  int64_t consumerId;
+  char    cgroup[TSDB_CONSUMER_GROUP_LEN];
+  SArray* topics;  // SArray<SMqSubTopicEp>
+} SMqCMGetSubEpRsp;
+
+static FORCE_INLINE int32_t tEncodeSMqSubVgEp(void** buf, const SMqSubVgEp* pVgEp) {
+  int32_t tlen = 0;
+  tlen += taosEncodeFixedI32(buf, pVgEp->vgId);
+  tlen += taosEncodeSEpSet(buf, &pVgEp->epSet);
+  return tlen;
+}
+
+static FORCE_INLINE void* tDecodeSMqSubVgEp(void* buf, SMqSubVgEp* pVgEp) {
+  buf = taosDecodeFixedI32(buf, &pVgEp->vgId);
+  buf = taosDecodeSEpSet(buf, &pVgEp->epSet);
+  return buf;
+}
+
+static FORCE_INLINE int32_t tEncodeSMqSubTopicEp(void** buf, const SMqSubTopicEp* pTopicEp) {
+  int32_t tlen = 0;
+  tlen += taosEncodeString(buf, pTopicEp->topic);
+  int32_t sz = taosArrayGetSize(pTopicEp->vgs);
+  tlen += taosEncodeFixedI32(buf, sz);
+  for (int32_t i = 0; i < sz; i++) {
+    SMqSubVgEp* pVgEp = (SMqSubVgEp*)taosArrayGet(pTopicEp->vgs, i);
+    tlen += tEncodeSMqSubVgEp(buf, pVgEp);
+  }
+  return tlen;
+}
+
+static FORCE_INLINE void* tDecodeSMqSubTopicEp(void* buf, SMqSubTopicEp* pTopicEp) {
+  buf = taosDecodeStringTo(buf, pTopicEp->topic);
+  int32_t sz;
+  buf = taosDecodeFixedI32(buf, &sz);
+  pTopicEp->vgs = taosArrayInit(sz, sizeof(SMqSubVgEp));
+  if (pTopicEp->vgs == NULL) {
+    return NULL;
+  }
+  for (int32_t i = 0; i < sz; i++) {
+    SMqSubVgEp vgEp;
+    buf = tDecodeSMqSubVgEp(buf, &vgEp);
+    taosArrayPush(pTopicEp->vgs, &vgEp);
+  }
+  return buf;
+}
+
+static FORCE_INLINE int32_t tEncodeSMqCMGetSubEpRsp(void** buf, const SMqCMGetSubEpRsp* pRsp) {
+  int32_t tlen = 0;
+  tlen += taosEncodeFixedI64(buf, pRsp->consumerId);
+  tlen += taosEncodeString(buf, pRsp->cgroup);
+  int32_t sz = taosArrayGetSize(pRsp->topics);
+  tlen += taosEncodeFixedI32(buf, sz);
+  for (int32_t i = 0; i < sz; i++) {
+    SMqSubTopicEp* pVgEp = (SMqSubTopicEp*)taosArrayGet(pRsp->topics, i);
+    tlen += tEncodeSMqSubTopicEp(buf, pVgEp);
+  }
+  return tlen;
+}
+
+static FORCE_INLINE void* tDecodeSMqCMGetSubEpRsp(void* buf, SMqCMGetSubEpRsp* pRsp) {
+  buf = taosDecodeFixedI64(buf, &pRsp->consumerId);
+  buf = taosDecodeStringTo(buf, pRsp->cgroup);
+  int32_t sz;
+  buf = taosDecodeFixedI32(buf, &sz);
+  pRsp->topics = taosArrayInit(sz, sizeof(SMqSubTopicEp));
+  if (pRsp->topics == NULL) {
+    return NULL;
+  }
+  for (int32_t i = 0; i < sz; i++) {
+    SMqSubTopicEp topicEp;
+    buf = tDecodeSMqSubTopicEp(buf, &topicEp);
+    taosArrayPush(pRsp->topics, &topicEp);
+  }
+  return buf;
+}
 
 #ifdef __cplusplus
 }

include/common/tmsgdef.h

@@ -140,6 +140,7 @@ enum {
   TD_DEF_MSG_TYPE(TDMT_MND_ALTER_TOPIC, "mnode-alter-topic", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_MND_DROP_TOPIC, "mnode-drop-topic", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_MND_SUBSCRIBE, "mnode-subscribe", SCMSubscribeReq, SCMSubscribeRsp)
+  TD_DEF_MSG_TYPE(TDMT_MND_GET_SUB_EP, "mnode-get-sub-ep", SMqCMGetSubEpReq, SMqCMGetSubEpRsp)
   TD_DEF_MSG_TYPE(TDMT_MND_MQ_TIMER, "mnode-timer", SMqTmrMsg, SMqTmrMsg)
 
   // Requests handled by VNODE

include/libs/executor/executor.h

@@ -24,6 +24,7 @@ extern "C" {
 
 typedef void* qTaskInfo_t;
 typedef void* DataSinkHandle;
+struct SRpcMsg;
 struct SSubplan;
 
  /**
@@ -208,6 +209,8 @@ void** qReleaseTask(void* pMgmt, void* pQInfo, bool freeHandle);
  */
 void** qDeregisterQInfo(void* pMgmt, void* pQInfo);
 
+void qProcessFetchRsp(void* parent, struct SRpcMsg* pMsg, struct SEpSet* pEpSet);
+
 #ifdef __cplusplus
 }
 #endif

include/libs/qworker/qworker.h

@@ -49,9 +49,10 @@ typedef struct {
 } SQWorkerStat;
 
 typedef int32_t (*putReqToQueryQFp)(void *, struct SRpcMsg *);
+typedef int32_t (*sendReqToDnodeFp)(void *, struct SEpSet *, struct SRpcMsg *);
 
-
-int32_t qWorkerInit(int8_t nodeType, int32_t nodeId, SQWorkerCfg *cfg, void **qWorkerMgmt, void *nodeObj, putReqToQueryQFp fp);
+int32_t qWorkerInit(int8_t nodeType, int32_t nodeId, SQWorkerCfg *cfg, void **qWorkerMgmt, void *nodeObj,
+                    putReqToQueryQFp fp1, sendReqToDnodeFp fp2);
 
 int32_t qWorkerProcessQueryMsg(void *node, void *qWorkerMgmt, SRpcMsg *pMsg);
 
@@ -65,6 +66,8 @@ int32_t qWorkerProcessStatusMsg(void *node, void *qWorkerMgmt, SRpcMsg *pMsg);
 
 int32_t qWorkerProcessFetchMsg(void *node, void *qWorkerMgmt, SRpcMsg *pMsg);
 
+int32_t qWorkerProcessFetchRsp(void *node, void *qWorkerMgmt, SRpcMsg *pMsg);
+
 int32_t qWorkerProcessCancelMsg(void *node, void *qWorkerMgmt, SRpcMsg *pMsg);
 
 int32_t qWorkerProcessDropMsg(void *node, void *qWorkerMgmt, SRpcMsg *pMsg);

include/libs/wal/wal.h

@@ -19,6 +19,7 @@
 #include "tarray.h"
 #include "tdef.h"
 #include "tlog.h"
+#include "tmsg.h"
 #ifdef __cplusplus
 extern "C" {
 #endif
@@ -159,7 +160,7 @@ int32_t walAlter(SWal *, SWalCfg *pCfg);
 void    walClose(SWal *);
 
 // write
-int64_t walWrite(SWal *, int64_t index, uint8_t msgType, const void *body, int32_t bodyLen);
+int64_t walWrite(SWal *, int64_t index, tmsg_t msgType, const void *body, int32_t bodyLen);
 void    walFsync(SWal *, bool force);
 
 // apis for lifecycle management

include/util/taoserror.h

@@ -253,6 +253,7 @@ int32_t* taosGetErrno();
 #define TSDB_CODE_MND_INVALID_TOPIC_OPTION      TAOS_DEF_ERROR_CODE(0, 0x03E4)
 #define TSDB_CODE_MND_TOPIC_OPTION_UNCHNAGED    TAOS_DEF_ERROR_CODE(0, 0x03E5)
 #define TSDB_CODE_MND_NAME_CONFLICT_WITH_STB    TAOS_DEF_ERROR_CODE(0, 0x03E6)
+#define TSDB_CODE_MND_CONSUMER_NOT_EXIST        TAOS_DEF_ERROR_CODE(0, 0x03E7)
 
 // dnode
 #define TSDB_CODE_DND_ACTION_IN_PROGRESS        TAOS_DEF_ERROR_CODE(0, 0x0400)

include/util/tqueue.h

@@ -15,6 +15,7 @@
 
 #ifndef _TD_UTIL_QUEUE_H
 #define _TD_UTIL_QUEUE_H
+#include "os.h"
 
 #ifdef __cplusplus
 extern "C" {
@@ -40,13 +41,13 @@ shall be used to set up the protection.
 typedef struct STaosQueue STaosQueue;
 typedef struct STaosQset  STaosQset;
 typedef struct STaosQall  STaosQall;
-typedef void (*FProcessItem)(void *ahandle, void *pItem);
-typedef void (*FProcessItems)(void *ahandle, STaosQall *qall, int32_t numOfItems);
+typedef void (*FItem)(void *ahandle, void *pItem);
+typedef void (*FItems)(void *ahandle, STaosQall *qall, int32_t numOfItems);
 
 STaosQueue *taosOpenQueue();
 void        taosCloseQueue(STaosQueue *queue);
-void        taosSetQueueFp(STaosQueue *queue, FProcessItem itemFp, FProcessItems itemsFp);
-void       *taosAllocateQitem(int32_t size);
+void        taosSetQueueFp(STaosQueue *queue, FItem itemFp, FItems itemsFp);
+void *      taosAllocateQitem(int32_t size);
 void        taosFreeQitem(void *pItem);
 int32_t     taosWriteQitem(STaosQueue *queue, void *pItem);
 int32_t     taosReadQitem(STaosQueue *queue, void **ppItem);
@@ -66,8 +67,11 @@ int32_t    taosAddIntoQset(STaosQset *qset, STaosQueue *queue, void *ahandle);
 void       taosRemoveFromQset(STaosQset *qset, STaosQueue *queue);
 int32_t    taosGetQueueNumber(STaosQset *qset);
 
-int32_t taosReadQitemFromQset(STaosQset *qset, void **ppItem, void **ahandle, FProcessItem *itemFp);
-int32_t taosReadAllQitemsFromQset(STaosQset *qset, STaosQall *qall, void **ahandle, FProcessItems *itemsFp);
+int32_t taosReadQitemFromQset(STaosQset *qset, void **ppItem, void **ahandle, FItem *itemFp);
+int32_t taosReadAllQitemsFromQset(STaosQset *qset, STaosQall *qall, void **ahandle, FItems *itemsFp);
+
+int32_t taosReadQitemFromQsetByThread(STaosQset *qset, void **ppItem, void **ahandle, FItem *itemFp, int32_t threadId);
+void    taosResetQsetThread(STaosQset *qset, void *pItem);
 
 int32_t taosGetQueueItemsNumber(STaosQueue *queue);
 int32_t taosGetQsetItemsNumber(STaosQset *qset);

include/util/tworker.h

@@ -15,57 +15,61 @@
 
 #ifndef _TD_UTIL_WORKER_H
 #define _TD_UTIL_WORKER_H
-
 #include "tqueue.h"
 
 #ifdef __cplusplus
 extern "C" {
 #endif
 
-typedef struct SWorkerPool  SWorkerPool;
-typedef struct SMWorkerPool SMWorkerPool;
+typedef struct SQWorkerPool SQWorkerPool;
+typedef struct SWWorkerPool SWWorkerPool;
 
-typedef struct SWorker {
-  int32_t      id;      // worker ID
-  pthread_t    thread;  // thread
-  SWorkerPool *pool;
-} SWorker;
+typedef struct SQWorker {
+  int32_t       id;      // worker ID
+  pthread_t     thread;  // thread
+  SQWorkerPool *pool;
+} SQWorker, SFWorker;
 
-typedef struct SWorkerPool {
+typedef struct SQWorkerPool {
   int32_t         max;  // max number of workers
   int32_t         min;  // min number of workers
   int32_t         num;  // current number of workers
-  STaosQset      *qset;
-  const char     *name;
-  SWorker        *workers;
+  STaosQset *     qset;
+  const char *    name;
+  SQWorker *      workers;
   pthread_mutex_t mutex;
-} SWorkerPool;
+} SQWorkerPool, SFWorkerPool;
 
-typedef struct SMWorker {
+typedef struct SWWorker {
   int32_t       id;      // worker id
   pthread_t     thread;  // thread
-  STaosQall    *qall;
-  STaosQset    *qset;  // queue set
-  SMWorkerPool *pool;
-} SMWorker;
+  STaosQall *   qall;
+  STaosQset *   qset;  // queue set
+  SWWorkerPool *pool;
+} SWWorker;
 
-typedef struct SMWorkerPool {
+typedef struct SWWorkerPool {
   int32_t         max;     // max number of workers
   int32_t         nextId;  // from 0 to max-1, cyclic
-  const char     *name;
-  SMWorker       *workers;
+  const char *    name;
+  SWWorker *      workers;
   pthread_mutex_t mutex;
-} SMWorkerPool;
+} SWWorkerPool;
+
+int32_t     tQWorkerInit(SQWorkerPool *pool);
+void        tQWorkerCleanup(SQWorkerPool *pool);
+STaosQueue *tQWorkerAllocQueue(SQWorkerPool *pool, void *ahandle, FItem fp);
+void        tQWorkerFreeQueue(SQWorkerPool *pool, STaosQueue *queue);
 
-int32_t     tWorkerInit(SWorkerPool *pool);
-void        tWorkerCleanup(SWorkerPool *pool);
-STaosQueue *tWorkerAllocQueue(SWorkerPool *pool, void *ahandle, FProcessItem fp);
-void        tWorkerFreeQueue(SWorkerPool *pool, STaosQueue *queue);
+int32_t     tFWorkerInit(SFWorkerPool *pool);
+void        tFWorkerCleanup(SFWorkerPool *pool);
+STaosQueue *tFWorkerAllocQueue(SFWorkerPool *pool, void *ahandle, FItem fp);
+void        tFWorkerFreeQueue(SFWorkerPool *pool, STaosQueue *queue);
 
-int32_t     tMWorkerInit(SMWorkerPool *pool);
-void        tMWorkerCleanup(SMWorkerPool *pool);
-STaosQueue *tMWorkerAllocQueue(SMWorkerPool *pool, void *ahandle, FProcessItems fp);
-void        tMWorkerFreeQueue(SMWorkerPool *pool, STaosQueue *queue);
+int32_t     tWWorkerInit(SWWorkerPool *pool);
+void        tWWorkerCleanup(SWWorkerPool *pool);
+STaosQueue *tWWorkerAllocQueue(SWWorkerPool *pool, void *ahandle, FItems fp);
+void        tWWorkerFreeQueue(SWWorkerPool *pool, STaosQueue *queue);
 
 #ifdef __cplusplus
 }

source/client/src/clientImpl.c

@@ -263,7 +263,7 @@ int32_t scheduleQuery(SRequestObj* pRequest, SQueryDag* pDag, SArray* pNodeList)
 
 typedef struct SMqClientVg {
   // statistics
-  int64_t consumeCnt;
+  int64_t pollCnt;
   // offset
   int64_t committedOffset;
   int64_t currentOffset;
@@ -328,6 +328,7 @@ struct tmq_t {
   char           clientId[256];
   int64_t        consumerId;
   int64_t        status;
+  tsem_t         rspSem;
   STscObj*       pTscObj;
   tmq_commit_cb* commit_cb;
   int32_t        nextTopicIdx;
@@ -344,7 +345,9 @@ tmq_t* taos_consumer_new(void* conn, tmq_conf_t* conf, char* errstr, int32_t err
   strcpy(pTmq->clientId, conf->clientId);
   strcpy(pTmq->groupId, conf->groupId);
   pTmq->commit_cb = conf->commit_cb;
+  tsem_init(&pTmq->rspSem, 0, 0);
   pTmq->consumerId = generateRequestId() & ((uint64_t)-1 >> 1);
+  pTmq->clientTopics = taosArrayInit(0, sizeof(SMqClientTopic));
   return pTmq;
 }
 
@@ -371,11 +374,27 @@ int32_t tmq_list_append(tmq_list_t* ptr, char* src) {
 }
 
 
+int32_t tmq_null_cb(void* param, const SDataBuf* pMsg, int32_t code)  {
+  if (code == 0) {
+    //
+  }
+  //
+  return 0;
+}
+
 TAOS_RES* tmq_subscribe(tmq_t* tmq, tmq_list_t* topic_list) {
   SRequestObj *pRequest = NULL;
-  tmq->status = 1;
   int32_t sz = topic_list->cnt;
-  tmq->clientTopics = taosArrayInit(sz, sizeof(void*));
+  //destroy ex
+  taosArrayDestroy(tmq->clientTopics);
+  tmq->clientTopics = taosArrayInit(sz, sizeof(SMqClientTopic));
+
+  SCMSubscribeReq req;
+  req.topicNum = sz;
+  req.consumerId = tmq->consumerId;
+  req.consumerGroup = strdup(tmq->groupId);
+  req.topicNames = taosArrayInit(sz, sizeof(void*));
+
   for (int i = 0; i < sz; i++) {
     char* topicName = topic_list->elems[i];
 
@@ -390,16 +409,21 @@ TAOS_RES* tmq_subscribe(tmq_t* tmq, tmq_list_t* topic_list) {
     }
     tNameExtractFullName(&name, topicFname);
     tscDebug("subscribe topic: %s", topicFname);
-    taosArrayPush(tmq->clientTopics, &topicFname); 
+    SMqClientTopic topic = {
+      .nextVgIdx = 0,
+      .sql = NULL,
+      .sqlLen = 0,
+      .topicId = 0,
+      .topicName = topicFname,
+      .vgs = NULL
+    };
+    topic.vgs = taosArrayInit(0, sizeof(SMqClientVg));
+    taosArrayPush(tmq->clientTopics, &topic); 
     /*SMqClientTopic topic = {*/
       /*.*/
     /*};*/
+    taosArrayPush(req.topicNames, &topicFname);
   }
-  SCMSubscribeReq req;
-  req.topicNum = taosArrayGetSize(tmq->clientTopics);
-  req.consumerId = tmq->consumerId;
-  req.consumerGroup = strdup(tmq->groupId);
-  req.topicNames = tmq->clientTopics;
 
   int tlen = tSerializeSCMSubscribeReq(NULL, &req);
   void* buf = malloc(tlen);
@@ -411,26 +435,26 @@ TAOS_RES* tmq_subscribe(tmq_t* tmq, tmq_list_t* topic_list) {
   tSerializeSCMSubscribeReq(&abuf, &req);
   /*printf("formatted: %s\n", dagStr);*/
 
-  pRequest = createRequest(tmq->pTscObj, NULL, NULL, TSDB_SQL_SELECT);
+  pRequest = createRequest(tmq->pTscObj, NULL, NULL, TDMT_MND_SUBSCRIBE);
   if (pRequest == NULL) {
     tscError("failed to malloc sqlObj");
   }
 
   pRequest->body.requestMsg = (SDataBuf){ .pData = buf, .len = tlen };
-  pRequest->type = TDMT_MND_SUBSCRIBE;
 
-  SMsgSendInfo* body = buildMsgInfoImpl(pRequest);
+  SMsgSendInfo* sendInfo = buildMsgInfoImpl(pRequest);
+  /*sendInfo->fp*/
   SEpSet epSet = getEpSet_s(&tmq->pTscObj->pAppInfo->mgmtEp);
 
   int64_t transporterId = 0;
-  asyncSendMsgToServer(tmq->pTscObj->pAppInfo->pTransporter, &epSet, &transporterId, body);
+  asyncSendMsgToServer(tmq->pTscObj->pAppInfo->pTransporter, &epSet, &transporterId, sendInfo);
 
   tsem_wait(&pRequest->body.rspSem);
 
 _return:
-  if (body != NULL) {
-    destroySendMsgInfo(body);
-  }
+  /*if (sendInfo != NULL) {*/
+    /*destroySendMsgInfo(sendInfo);*/
+  /*}*/
 
   if (pRequest != NULL && terrno != TSDB_CODE_SUCCESS) {
     pRequest->code = terrno;
@@ -569,19 +593,19 @@ TAOS_RES *taos_create_topic(TAOS* taos, const char* topicName, const char* sql,
   pRequest->body.requestMsg = (SDataBuf){ .pData = buf, .len = tlen };
   pRequest->type = TDMT_MND_CREATE_TOPIC;
 
-  SMsgSendInfo* body = buildMsgInfoImpl(pRequest);
+  SMsgSendInfo* sendInfo = buildMsgInfoImpl(pRequest);
   SEpSet epSet = getEpSet_s(&pTscObj->pAppInfo->mgmtEp);
 
   int64_t transporterId = 0;
-  asyncSendMsgToServer(pTscObj->pAppInfo->pTransporter, &epSet, &transporterId, body);
+  asyncSendMsgToServer(pTscObj->pAppInfo->pTransporter, &epSet, &transporterId, sendInfo);
 
   tsem_wait(&pRequest->body.rspSem);
 
 _return:
   qDestroyQuery(pQueryNode);
-  if (body != NULL) {
-    destroySendMsgInfo(body);
-  }
+  /*if (sendInfo != NULL) {*/
+    /*destroySendMsgInfo(sendInfo);*/
+  /*}*/
 
   if (pRequest != NULL && terrno != TSDB_CODE_SUCCESS) {
     pRequest->code = terrno;
@@ -596,36 +620,117 @@ struct tmq_message_t {
   SMqConsumeRsp rsp;
 };
 
+int32_t tmq_poll_cb_inner(void* param, const SDataBuf* pMsg, int32_t code) {
+  return 0;
+}
+
+int32_t tmq_ask_ep_cb(void* param, const SDataBuf* pMsg, int32_t code) {
+  tmq_t* tmq = (tmq_t*)param;
+  if (code != 0) {
+    tsem_post(&tmq->rspSem);
+    return 0;
+  }
+  tscDebug("tmq ask ep cb called");
+  bool set = false;
+  SMqCMGetSubEpRsp rsp;
+  tDecodeSMqCMGetSubEpRsp(pMsg->pData, &rsp);
+  int32_t sz = taosArrayGetSize(rsp.topics);
+  // TODO: lock
+  tmq->clientTopics = taosArrayInit(sz, sizeof(SMqClientTopic));
+  for (int32_t i = 0; i < sz; i++) {
+    SMqClientTopic topic = {0};
+    SMqSubTopicEp* pTopicEp = taosArrayGet(rsp.topics, i);
+    topic.topicName = strdup(pTopicEp->topic);
+    int32_t vgSz = taosArrayGetSize(pTopicEp->vgs);
+    topic.vgs = taosArrayInit(vgSz, sizeof(SMqClientVg));
+    for (int32_t j = 0; j < vgSz; j++) {
+      SMqSubVgEp* pVgEp = taosArrayGet(pTopicEp->vgs, j);
+      SMqClientVg clientVg = {
+        .pollCnt = 0,
+        .committedOffset = -1,
+        .currentOffset = -1,
+        .vgId = pVgEp->vgId,
+        .epSet = pVgEp->epSet
+      };
+      taosArrayPush(topic.vgs, &clientVg);
+      set = true;
+    }
+    taosArrayPush(tmq->clientTopics, &topic);
+  }
+  if(set) tmq->status = 1;
+  // unlock
+  tsem_post(&tmq->rspSem);
+  return 0;
+}
+
 tmq_message_t* tmq_consume_poll(tmq_t* tmq, int64_t blocking_time) {
-  if (tmq->clientTopics == NULL || taosArrayGetSize(tmq->clientTopics) == 0) {
+
+  if (taosArrayGetSize(tmq->clientTopics) == 0 || tmq->status == 0) {
+    int32_t tlen = sizeof(SMqCMGetSubEpReq);
+    SMqCMGetSubEpReq* buf = malloc(tlen);
+    if (buf == NULL) {
+      tscError("failed to malloc get subscribe ep buf");
+    }
+    buf->consumerId = htobe64(tmq->consumerId);
+    strcpy(buf->cgroup, tmq->groupId);
+    
+    SRequestObj *pRequest = createRequest(tmq->pTscObj, NULL, NULL, TDMT_MND_GET_SUB_EP);
+    if (pRequest == NULL) {
+      tscError("failed to malloc subscribe ep request");
+    }
+
+    pRequest->body.requestMsg = (SDataBuf){ .pData = buf, .len = tlen };
+
+    SMsgSendInfo* sendInfo = buildMsgInfoImpl(pRequest);
+    sendInfo->requestObjRefId = 0;
+    sendInfo->param = tmq;
+    sendInfo->fp = tmq_ask_ep_cb;
+
+    SEpSet epSet = getEpSet_s(&tmq->pTscObj->pAppInfo->mgmtEp);
+
+    int64_t transporterId = 0;
+    asyncSendMsgToServer(tmq->pTscObj->pAppInfo->pTransporter, &epSet, &transporterId, sendInfo);
+
+    tsem_wait(&tmq->rspSem);
+  }
+
+  if (taosArrayGetSize(tmq->clientTopics) == 0) {
+    tscDebug("consumer:%ld poll but not assigned", tmq->consumerId);
     return NULL;
   }
-  SRequestObj *pRequest = NULL;
-  SMqConsumeReq req = {0};
-  req.reqType = 1;
-  req.blockingTime = blocking_time;
-  req.consumerId = tmq->consumerId;
-  strcpy(req.cgroup, tmq->groupId);
+
+  SMqConsumeReq* pReq = malloc(sizeof(SMqConsumeReq));
+  pReq->reqType = 1;
+  pReq->blockingTime = blocking_time;
+  pReq->consumerId = tmq->consumerId;
+  tmq_message_t* tmq_message = NULL;
+  strcpy(pReq->cgroup, tmq->groupId);
 
   SMqClientTopic* pTopic = taosArrayGet(tmq->clientTopics, tmq->nextTopicIdx);
   tmq->nextTopicIdx = (tmq->nextTopicIdx + 1) % taosArrayGetSize(tmq->clientTopics);
-  strcpy(req.topic, pTopic->topicName);
+  strcpy(pReq->topic, pTopic->topicName);
   int32_t nextVgIdx = pTopic->nextVgIdx;
   pTopic->nextVgIdx = (nextVgIdx + 1) % taosArrayGetSize(pTopic->vgs);
   SMqClientVg* pVg = taosArrayGet(pTopic->vgs, nextVgIdx);
-  req.offset = pVg->currentOffset;
+  pReq->offset = pVg->currentOffset;
+
+  pReq->head.vgId = htonl(pVg->vgId);
+  pReq->head.contLen = htonl(sizeof(SMqConsumeReq));
 
-  pRequest->body.requestMsg = (SDataBuf){ .pData = &req, .len = sizeof(SMqConsumeReq) };
-  pRequest->type = TDMT_VND_CONSUME;
+  SRequestObj* pRequest = createRequest(tmq->pTscObj, NULL, NULL, TDMT_VND_CONSUME);
+  pRequest->body.requestMsg = (SDataBuf){ .pData = pReq, .len = sizeof(SMqConsumeReq) };
 
-  SMsgSendInfo* body = buildMsgInfoImpl(pRequest);
+  SMsgSendInfo* sendInfo = buildMsgInfoImpl(pRequest);
+  /*sendInfo->requestObjRefId = 0;*/
+  /*sendInfo->param = &tmq_message;*/
+  /*sendInfo->fp = tmq_poll_cb_inner;*/
 
   int64_t transporterId = 0;
-  asyncSendMsgToServer(tmq->pTscObj->pAppInfo->pTransporter, &pVg->epSet, &transporterId, body);
+  asyncSendMsgToServer(tmq->pTscObj->pAppInfo->pTransporter, &pVg->epSet, &transporterId, sendInfo);
 
   tsem_wait(&pRequest->body.rspSem);
 
-  return (tmq_message_t*)pRequest->body.resInfo.pData;
+  return tmq_message;
 
   /*tsem_wait(&pRequest->body.rspSem);*/
 

source/client/test/clientTests.cpp

@@ -571,7 +571,7 @@ TEST(testCase, create_topic_Test) {
   if (taos_errno(pRes) != 0) {
     printf("error in use db, reason:%s\n", taos_errstr(pRes));
   }
-  taos_free_result(pRes);
+  //taos_free_result(pRes);
 
   TAOS_FIELD* pFields = taos_fetch_fields(pRes);
   ASSERT_TRUE(pFields == nullptr);
@@ -592,6 +592,12 @@ TEST(testCase, tmq_subscribe_Test) {
   TAOS* pConn = taos_connect("localhost", "root", "taosdata", NULL, 0);
   assert(pConn != NULL);
 
+  TAOS_RES* pRes = taos_query(pConn, "use abc1");
+  if (taos_errno(pRes) != 0) {
+    printf("error in use db, reason:%s\n", taos_errstr(pRes));
+  }
+  taos_free_result(pRes);
+
   tmq_conf_t* conf = tmq_conf_new();
   tmq_conf_set(conf, "group.id", "tg1");
   tmq_t* tmq = taos_consumer_new(pConn, conf, NULL, 0);
@@ -603,16 +609,16 @@ TEST(testCase, tmq_subscribe_Test) {
   while (1) {
     tmq_message_t* msg = tmq_consume_poll(tmq, 0);
     printf("get msg\n");
-    if (msg == NULL) break;
+    //if (msg == NULL) break;
   }
 }
+#endif
 
 TEST(testCase, tmq_consume_Test) {
 }
 
 TEST(testCase, tmq_commit_TEST) {
 }
-#endif
 
 TEST(testCase, projection_query_tables) {
   TAOS* pConn = taos_connect("localhost", "root", "taosdata", NULL, 0);

source/dnode/mgmt/impl/inc/dndEnv.h

@@ -31,8 +31,8 @@ typedef struct {
   SDnode     *pDnode;
   STaosQueue *queue;
   union {
-    SWorkerPool  pool;
-    SMWorkerPool mpool;
+    SQWorkerPool  pool;
+    SWWorkerPool mpool;
   };
 } SDnodeWorker;
 
@@ -109,10 +109,10 @@ typedef struct {
   int32_t      openVnodes;
   int32_t      totalVnodes;
   SRWLatch     latch;
-  SWorkerPool  queryPool;
-  SWorkerPool  fetchPool;
-  SMWorkerPool syncPool;
-  SMWorkerPool writePool;
+  SQWorkerPool queryPool;
+  SFWorkerPool fetchPool;
+  SWWorkerPool syncPool;
+  SWWorkerPool writePool;
 } SVnodesMgmt;
 
 typedef struct {

source/dnode/mgmt/impl/src/dndEnv.c

@@ -289,6 +289,7 @@ int32_t dndInit(const SDnodeEnvCfg *pCfg) {
       .charset = pCfg->charset,
       .nthreads = pCfg->numOfCommitThreads,
       .putReqToVQueryQFp = dndPutReqToVQueryQ,
+      .sendReqToDnodeFp = dndSendReqToDnode
   };
 
   if (vnodeInit(&vnodeOpt) != 0) {

source/dnode/mgmt/impl/src/dndMgmt.c

@@ -532,7 +532,7 @@ int32_t dndInitMgmt(SDnode *pDnode) {
   }
 
   if (pMgmt->dropped) {
-    dError("dnode will not start for its already dropped");
+    dError("dnode not start since its already dropped");
     return -1;
   }
 

source/dnode/mgmt/impl/src/dndTransport.c

@@ -25,9 +25,9 @@
 #include "dndMnode.h"
 #include "dndVnodes.h"
 
-#define INTERNAL_USER "_internal"
+#define INTERNAL_USER "_dnd"
 #define INTERNAL_CKEY "_key"
-#define INTERNAL_SECRET "_secret"
+#define INTERNAL_SECRET "_pwd"
 
 static void dndInitMsgFp(STransMgmt *pMgmt) {
   // Requests handled by DNODE
@@ -115,12 +115,14 @@ static void dndInitMsgFp(STransMgmt *pMgmt) {
   pMgmt->msgFp[TMSG_INDEX(TDMT_MND_SUBSCRIBE)] = dndProcessMnodeWriteMsg;
   /*pMgmt->msgFp[TMSG_INDEX(TDMT_VND_SUBSCRIBE_RSP)] = dndProcessMnodeWriteMsg;*/
   pMgmt->msgFp[TMSG_INDEX(TDMT_VND_MQ_SET_CONN_RSP)] = dndProcessMnodeWriteMsg;
+  pMgmt->msgFp[TMSG_INDEX(TDMT_MND_GET_SUB_EP)] = dndProcessMnodeReadMsg;
 
   // Requests handled by VNODE
   pMgmt->msgFp[TMSG_INDEX(TDMT_VND_SUBMIT)] = dndProcessVnodeWriteMsg;
   pMgmt->msgFp[TMSG_INDEX(TDMT_VND_QUERY)] = dndProcessVnodeQueryMsg;
-  pMgmt->msgFp[TMSG_INDEX(TDMT_VND_QUERY_CONTINUE)] = dndProcessVnodeQueryMsg;  
+  pMgmt->msgFp[TMSG_INDEX(TDMT_VND_QUERY_CONTINUE)] = dndProcessVnodeQueryMsg;
   pMgmt->msgFp[TMSG_INDEX(TDMT_VND_FETCH)] = dndProcessVnodeFetchMsg;
+  pMgmt->msgFp[TMSG_INDEX(TDMT_VND_FETCH_RSP)] = dndProcessVnodeFetchMsg;
   pMgmt->msgFp[TMSG_INDEX(TDMT_VND_ALTER_TABLE)] = dndProcessVnodeWriteMsg;
   pMgmt->msgFp[TMSG_INDEX(TDMT_VND_UPDATE_TAG_VAL)] = dndProcessVnodeWriteMsg;
   pMgmt->msgFp[TMSG_INDEX(TDMT_VND_TABLE_META)] = dndProcessVnodeFetchMsg;
@@ -147,6 +149,7 @@ static void dndInitMsgFp(STransMgmt *pMgmt) {
   pMgmt->msgFp[TMSG_INDEX(TDMT_VND_SHOW_TABLES_FETCH)] = dndProcessVnodeFetchMsg;
   pMgmt->msgFp[TMSG_INDEX(TDMT_VND_MQ_SET_CONN)] = dndProcessVnodeWriteMsg;
   pMgmt->msgFp[TMSG_INDEX(TDMT_VND_MQ_SET_CUR)] = dndProcessVnodeFetchMsg;
+  pMgmt->msgFp[TMSG_INDEX(TDMT_VND_CONSUME)] = dndProcessVnodeFetchMsg;
 }
 
 static void dndProcessResponse(void *parent, SRpcMsg *pRsp, SEpSet *pEpSet) {
@@ -157,17 +160,18 @@ static void dndProcessResponse(void *parent, SRpcMsg *pRsp, SEpSet *pEpSet) {
 
   if (dndGetStat(pDnode) == DND_STAT_STOPPED) {
     if (pRsp == NULL || pRsp->pCont == NULL) return;
-    dTrace("RPC %p, rsp:%s is ignored since dnode is stopping", pRsp->handle, TMSG_INFO(msgType));
+    dTrace("RPC %p, rsp:%s ignored since dnode exiting, app:%p", pRsp->handle, TMSG_INFO(msgType), pRsp->ahandle);
     rpcFreeCont(pRsp->pCont);
     return;
   }
 
   DndMsgFp fp = pMgmt->msgFp[TMSG_INDEX(msgType)];
   if (fp != NULL) {
-    dTrace("RPC %p, rsp:%s will be processed, code:0x%x", pRsp->handle, TMSG_INFO(msgType), pRsp->code & 0XFFFF);
+    dTrace("RPC %p, rsp:%s will be processed, code:0x%x app:%p", pRsp->handle, TMSG_INFO(msgType), pRsp->code & 0XFFFF,
+           pRsp->ahandle);
     (*fp)(pDnode, pRsp, pEpSet);
   } else {
-    dError("RPC %p, rsp:%s not processed", pRsp->handle, TMSG_INFO(msgType));
+    dError("RPC %p, rsp:%s not processed, app:%p", pRsp->handle, TMSG_INFO(msgType), pRsp->ahandle);
     rpcFreeCont(pRsp->pCont);
   }
 }
@@ -194,7 +198,7 @@ static int32_t dndInitClient(SDnode *pDnode) {
 
   pMgmt->clientRpc = rpcOpen(&rpcInit);
   if (pMgmt->clientRpc == NULL) {
-    dError("failed to init rpc client");
+    dError("failed to init dnode rpc client");
     return -1;
   }
 
@@ -217,40 +221,39 @@ static void dndProcessRequest(void *param, SRpcMsg *pReq, SEpSet *pEpSet) {
 
   tmsg_t msgType = pReq->msgType;
   if (msgType == TDMT_DND_NETWORK_TEST) {
-    dTrace("RPC %p, network test req, app:%p will be processed, code:0x%x", pReq->handle, pReq->ahandle, pReq->code);
+    dTrace("RPC %p, network test req will be processed, app:%p", pReq->handle, pReq->ahandle);
     dndProcessStartupReq(pDnode, pReq);
     return;
   }
 
   if (dndGetStat(pDnode) == DND_STAT_STOPPED) {
-    dError("RPC %p, req:%s app:%p is ignored since dnode exiting", pReq->handle, TMSG_INFO(msgType), pReq->ahandle);
-    SRpcMsg rspMsg = {.handle = pReq->handle, .code = TSDB_CODE_DND_OFFLINE};
+    dError("RPC %p, req:%s ignored since dnode exiting, app:%p", pReq->handle, TMSG_INFO(msgType), pReq->ahandle);
+    SRpcMsg rspMsg = {.handle = pReq->handle, .code = TSDB_CODE_DND_OFFLINE, .ahandle = pReq->ahandle};
     rpcSendResponse(&rspMsg);
     rpcFreeCont(pReq->pCont);
     return;
   } else if (dndGetStat(pDnode) != DND_STAT_RUNNING) {
-    dError("RPC %p, req:%s app:%p is ignored since dnode not running", pReq->handle, TMSG_INFO(msgType), pReq->ahandle);
-    SRpcMsg rspMsg = {.handle = pReq->handle, .code = TSDB_CODE_APP_NOT_READY};
+    dError("RPC %p, req:%s ignored since dnode not running, app:%p", pReq->handle, TMSG_INFO(msgType), pReq->ahandle);
+    SRpcMsg rspMsg = {.handle = pReq->handle, .code = TSDB_CODE_APP_NOT_READY, .ahandle = pReq->ahandle};
     rpcSendResponse(&rspMsg);
     rpcFreeCont(pReq->pCont);
     return;
   }
 
   if (pReq->pCont == NULL) {
-    dTrace("RPC %p, req:%s app:%p not processed since content is null", pReq->handle, TMSG_INFO(msgType),
-           pReq->ahandle);
-    SRpcMsg rspMsg = {.handle = pReq->handle, .code = TSDB_CODE_DND_INVALID_MSG_LEN};
+    dTrace("RPC %p, req:%s not processed since its empty, app:%p", pReq->handle, TMSG_INFO(msgType), pReq->ahandle);
+    SRpcMsg rspMsg = {.handle = pReq->handle, .code = TSDB_CODE_DND_INVALID_MSG_LEN, .ahandle = pReq->ahandle};
     rpcSendResponse(&rspMsg);
     return;
   }
 
   DndMsgFp fp = pMgmt->msgFp[TMSG_INDEX(msgType)];
   if (fp != NULL) {
-    dTrace("RPC %p, req:%s app:%p will be processed", pReq->handle, TMSG_INFO(msgType), pReq->ahandle);
+    dTrace("RPC %p, req:%s will be processed, app:%p", pReq->handle, TMSG_INFO(msgType), pReq->ahandle);
     (*fp)(pDnode, pReq, pEpSet);
   } else {
-    dError("RPC %p, req:%s app:%p is not processed since no handle", pReq->handle, TMSG_INFO(msgType), pReq->ahandle);
-    SRpcMsg rspMsg = {.handle = pReq->handle, .code = TSDB_CODE_MSG_NOT_PROCESSED};
+    dError("RPC %p, req:%s not processed since no handle, app:%p", pReq->handle, TMSG_INFO(msgType), pReq->ahandle);
+    SRpcMsg rspMsg = {.handle = pReq->handle, .code = TSDB_CODE_MSG_NOT_PROCESSED, .ahandle = pReq->ahandle};
     rpcSendResponse(&rspMsg);
     rpcFreeCont(pReq->pCont);
   }
@@ -290,24 +293,24 @@ static int32_t dndRetrieveUserAuthInfo(void *parent, char *user, char *spi, char
   SDnode *pDnode = parent;
 
   if (dndAuthInternalReq(parent, user, spi, encrypt, secret, ckey) == 0) {
-    dTrace("user:%s, get auth from internal mnode, spi:%d encrypt:%d", user, *spi, *encrypt);
+    dTrace("user:%s, get auth from mnode, spi:%d encrypt:%d", user, *spi, *encrypt);
     return 0;
   }
 
   if (dndGetUserAuthFromMnode(pDnode, user, spi, encrypt, secret, ckey) == 0) {
-    dTrace("user:%s, get auth from internal mnode, spi:%d encrypt:%d", user, *spi, *encrypt);
+    dTrace("user:%s, get auth from mnode, spi:%d encrypt:%d", user, *spi, *encrypt);
     return 0;
   }
 
   if (terrno != TSDB_CODE_APP_NOT_READY) {
-    dTrace("failed to get user auth from internal mnode since %s", terrstr());
+    dTrace("failed to get user auth from mnode since %s", terrstr());
     return -1;
   }
 
   SAuthReq *pReq = rpcMallocCont(sizeof(SAuthReq));
   tstrncpy(pReq->user, user, TSDB_USER_LEN);
 
-  SRpcMsg rpcMsg = {.pCont = pReq, .contLen = sizeof(SAuthReq), .msgType = TDMT_MND_AUTH};
+  SRpcMsg rpcMsg = {.pCont = pReq, .contLen = sizeof(SAuthReq), .msgType = TDMT_MND_AUTH, .ahandle = (void *)9528};
   SRpcMsg rpcRsp = {0};
   dTrace("user:%s, send user auth req to other mnodes, spi:%d encrypt:%d", user, pReq->spi, pReq->encrypt);
   dndSendMsgToMnodeRecv(pDnode, &rpcMsg, &rpcRsp);
@@ -351,7 +354,7 @@ static int32_t dndInitServer(SDnode *pDnode) {
 
   pMgmt->serverRpc = rpcOpen(&rpcInit);
   if (pMgmt->serverRpc == NULL) {
-    dError("failed to init rpc server");
+    dError("failed to init dnode rpc server");
     return -1;
   }
 

source/dnode/mgmt/impl/src/dndVnodes.c

@@ -253,7 +253,7 @@ static int32_t dndGetVnodesFromFile(SDnode *pDnode, SWrapperCfg **ppCfgs, int32_
     }
 
     for (int32_t i = 0; i < vnodesNum; ++i) {
-      cJSON       *vnode = cJSON_GetArrayItem(vnodes, i);
+      cJSON *      vnode = cJSON_GetArrayItem(vnodes, i);
       SWrapperCfg *pCfg = &pCfgs[i];
 
       cJSON *vgId = cJSON_GetObjectItem(vnode, "vgId");
@@ -382,7 +382,7 @@ static void *dnodeOpenVnodeFunc(void *param) {
     dndReportStartup(pDnode, "open-vnodes", stepDesc);
 
     SVnodeCfg cfg = {.pDnode = pDnode, .pTfs = pDnode->pTfs, .vgId = pCfg->vgId};
-    SVnode   *pImpl = vnodeOpen(pCfg->path, &cfg);
+    SVnode *  pImpl = vnodeOpen(pCfg->path, &cfg);
     if (pImpl == NULL) {
       dError("vgId:%d, failed to open vnode by thread:%d", pCfg->vgId, pThread->threadIndex);
       pThread->failed++;
@@ -527,7 +527,6 @@ static void dndGenerateVnodeCfg(SCreateVnodeReq *pCreate, SVnodeCfg *pCfg) {
   pCfg->vgId = pCreate->vgId;
   pCfg->wsize = pCreate->cacheBlockSize;
   pCfg->ssize = pCreate->cacheBlockSize;
-  pCfg->wsize = pCreate->cacheBlockSize;
   pCfg->lsize = pCreate->cacheBlockSize;
   pCfg->isHeapAllocator = true;
   pCfg->ttl = 4;
@@ -737,15 +736,9 @@ int32_t dndProcessCompactVnodeReq(SDnode *pDnode, SRpcMsg *pReq) {
   return 0;
 }
 
-static void dndProcessVnodeQueryQueue(SVnodeObj *pVnode, SRpcMsg *pMsg) {
-  SRpcMsg *pRsp = NULL;
-  vnodeProcessQueryReq(pVnode->pImpl, pMsg, &pRsp);
-}
+static void dndProcessVnodeQueryQueue(SVnodeObj *pVnode, SRpcMsg *pMsg) { vnodeProcessQueryMsg(pVnode->pImpl, pMsg); }
 
-static void dndProcessVnodeFetchQueue(SVnodeObj *pVnode, SRpcMsg *pMsg) {
-  SRpcMsg *pRsp = NULL;
-  vnodeProcessFetchReq(pVnode->pImpl, pMsg, &pRsp);
-}
+static void dndProcessVnodeFetchQueue(SVnodeObj *pVnode, SRpcMsg *pMsg) { vnodeProcessFetchMsg(pVnode->pImpl, pMsg); }
 
 static void dndProcessVnodeWriteQueue(SVnodeObj *pVnode, STaosQall *qall, int32_t numOfMsgs) {
   SArray *pArray = taosArrayInit(numOfMsgs, sizeof(SRpcMsg *));
@@ -916,27 +909,27 @@ static int32_t dndInitVnodeWorkers(SDnode *pDnode) {
   int32_t maxWriteThreads = TMAX(pDnode->env.numOfCores, 1);
   int32_t maxSyncThreads = TMAX(pDnode->env.numOfCores / 2, 1);
 
-  SWorkerPool *pPool = &pMgmt->queryPool;
-  pPool->name = "vnode-query";
-  pPool->min = minQueryThreads;
-  pPool->max = maxQueryThreads;
-  if (tWorkerInit(pPool) != 0) return -1;
+  SQWorkerPool *pQPool = &pMgmt->queryPool;
+  pQPool->name = "vnode-query";
+  pQPool->min = minQueryThreads;
+  pQPool->max = maxQueryThreads;
+  if (tQWorkerInit(pQPool) != 0) return -1;
 
-  pPool = &pMgmt->fetchPool;
-  pPool->name = "vnode-fetch";
-  pPool->min = minFetchThreads;
-  pPool->max = maxFetchThreads;
-  if (tWorkerInit(pPool) != 0) return -1;
+  SFWorkerPool *pFPool = &pMgmt->fetchPool;
+  pFPool->name = "vnode-fetch";
+  pFPool->min = minFetchThreads;
+  pFPool->max = maxFetchThreads;
+  if (tFWorkerInit(pFPool) != 0) return -1;
 
-  SMWorkerPool *pMPool = &pMgmt->writePool;
-  pMPool->name = "vnode-write";
-  pMPool->max = maxWriteThreads;
-  if (tMWorkerInit(pMPool) != 0) return -1;
+  SWWorkerPool *pWPool = &pMgmt->writePool;
+  pWPool->name = "vnode-write";
+  pWPool->max = maxWriteThreads;
+  if (tWWorkerInit(pWPool) != 0) return -1;
 
-  pMPool = &pMgmt->syncPool;
-  pMPool->name = "vnode-sync";
-  pMPool->max = maxSyncThreads;
-  if (tMWorkerInit(pMPool) != 0) return -1;
+  pWPool = &pMgmt->syncPool;
+  pWPool->name = "vnode-sync";
+  pWPool->max = maxSyncThreads;
+  if (tWWorkerInit(pWPool) != 0) return -1;
 
   dDebug("vnode workers is initialized");
   return 0;
@@ -944,21 +937,21 @@ static int32_t dndInitVnodeWorkers(SDnode *pDnode) {
 
 static void dndCleanupVnodeWorkers(SDnode *pDnode) {
   SVnodesMgmt *pMgmt = &pDnode->vmgmt;
-  tWorkerCleanup(&pMgmt->fetchPool);
-  tWorkerCleanup(&pMgmt->queryPool);
-  tMWorkerCleanup(&pMgmt->writePool);
-  tMWorkerCleanup(&pMgmt->syncPool);
+  tFWorkerCleanup(&pMgmt->fetchPool);
+  tQWorkerCleanup(&pMgmt->queryPool);
+  tWWorkerCleanup(&pMgmt->writePool);
+  tWWorkerCleanup(&pMgmt->syncPool);
   dDebug("vnode workers is closed");
 }
 
 static int32_t dndAllocVnodeQueue(SDnode *pDnode, SVnodeObj *pVnode) {
   SVnodesMgmt *pMgmt = &pDnode->vmgmt;
 
-  pVnode->pWriteQ = tMWorkerAllocQueue(&pMgmt->writePool, pVnode, (FProcessItems)dndProcessVnodeWriteQueue);
-  pVnode->pApplyQ = tMWorkerAllocQueue(&pMgmt->writePool, pVnode, (FProcessItems)dndProcessVnodeApplyQueue);
-  pVnode->pSyncQ = tMWorkerAllocQueue(&pMgmt->syncPool, pVnode, (FProcessItems)dndProcessVnodeSyncQueue);
-  pVnode->pFetchQ = tWorkerAllocQueue(&pMgmt->fetchPool, pVnode, (FProcessItem)dndProcessVnodeFetchQueue);
-  pVnode->pQueryQ = tWorkerAllocQueue(&pMgmt->queryPool, pVnode, (FProcessItem)dndProcessVnodeQueryQueue);
+  pVnode->pWriteQ = tWWorkerAllocQueue(&pMgmt->writePool, pVnode, (FItems)dndProcessVnodeWriteQueue);
+  pVnode->pApplyQ = tWWorkerAllocQueue(&pMgmt->writePool, pVnode, (FItems)dndProcessVnodeApplyQueue);
+  pVnode->pSyncQ = tWWorkerAllocQueue(&pMgmt->syncPool, pVnode, (FItems)dndProcessVnodeSyncQueue);
+  pVnode->pFetchQ = tFWorkerAllocQueue(&pMgmt->fetchPool, pVnode, (FItem)dndProcessVnodeFetchQueue);
+  pVnode->pQueryQ = tQWorkerAllocQueue(&pMgmt->queryPool, pVnode, (FItem)dndProcessVnodeQueryQueue);
 
   if (pVnode->pApplyQ == NULL || pVnode->pWriteQ == NULL || pVnode->pSyncQ == NULL || pVnode->pFetchQ == NULL ||
       pVnode->pQueryQ == NULL) {
@@ -971,11 +964,11 @@ static int32_t dndAllocVnodeQueue(SDnode *pDnode, SVnodeObj *pVnode) {
 
 static void dndFreeVnodeQueue(SDnode *pDnode, SVnodeObj *pVnode) {
   SVnodesMgmt *pMgmt = &pDnode->vmgmt;
-  tWorkerFreeQueue(&pMgmt->queryPool, pVnode->pQueryQ);
-  tWorkerFreeQueue(&pMgmt->fetchPool, pVnode->pFetchQ);
-  tMWorkerFreeQueue(&pMgmt->writePool, pVnode->pWriteQ);
-  tMWorkerFreeQueue(&pMgmt->writePool, pVnode->pApplyQ);
-  tMWorkerFreeQueue(&pMgmt->syncPool, pVnode->pSyncQ);
+  tQWorkerFreeQueue(&pMgmt->queryPool, pVnode->pQueryQ);
+  tFWorkerFreeQueue(&pMgmt->fetchPool, pVnode->pFetchQ);
+  tWWorkerFreeQueue(&pMgmt->writePool, pVnode->pWriteQ);
+  tWWorkerFreeQueue(&pMgmt->writePool, pVnode->pApplyQ);
+  tWWorkerFreeQueue(&pMgmt->syncPool, pVnode->pSyncQ);
   pVnode->pWriteQ = NULL;
   pVnode->pApplyQ = NULL;
   pVnode->pSyncQ = NULL;

source/dnode/mgmt/impl/src/dndWorker.c

@@ -31,28 +31,28 @@ int32_t dndInitWorker(SDnode *pDnode, SDnodeWorker *pWorker, EWorkerType type, c
   pWorker->pDnode = pDnode;
 
   if (pWorker->type == DND_WORKER_SINGLE) {
-    SWorkerPool *pPool = &pWorker->pool;
+    SQWorkerPool *pPool = &pWorker->pool;
     pPool->name = name;
     pPool->min = minNum;
     pPool->max = maxNum;
-    if (tWorkerInit(pPool) != 0) {
+    if (tQWorkerInit(pPool) != 0) {
       terrno = TSDB_CODE_OUT_OF_MEMORY;
       return -1;
     }
-    pWorker->queue = tWorkerAllocQueue(pPool, pDnode, (FProcessItem)queueFp);
+    pWorker->queue = tQWorkerAllocQueue(pPool, pDnode, (FItem)queueFp);
     if (pWorker->queue == NULL) {
       terrno = TSDB_CODE_OUT_OF_MEMORY;
       return -1;
     }
   } else if (pWorker->type == DND_WORKER_MULTI) {
-    SMWorkerPool *pPool = &pWorker->mpool;
+    SWWorkerPool *pPool = &pWorker->mpool;
     pPool->name = name;
     pPool->max = maxNum;
-    if (tMWorkerInit(pPool) != 0) {
+    if (tWWorkerInit(pPool) != 0) {
       terrno = TSDB_CODE_OUT_OF_MEMORY;
       return -1;
     }
-    pWorker->queue = tMWorkerAllocQueue(pPool, pDnode, (FProcessItems)queueFp);
+    pWorker->queue = tWWorkerAllocQueue(pPool, pDnode, (FItems)queueFp);
     if (pWorker->queue == NULL) {
       terrno = TSDB_CODE_OUT_OF_MEMORY;
       return -1;
@@ -70,11 +70,11 @@ void dndCleanupWorker(SDnodeWorker *pWorker) {
   }
 
   if (pWorker->type == DND_WORKER_SINGLE) {
-    tWorkerCleanup(&pWorker->pool);
-    tWorkerFreeQueue(&pWorker->pool, pWorker->queue);
+    tQWorkerCleanup(&pWorker->pool);
+    tQWorkerFreeQueue(&pWorker->pool, pWorker->queue);
   } else if (pWorker->type == DND_WORKER_MULTI) {
-    tMWorkerCleanup(&pWorker->mpool);
-    tMWorkerFreeQueue(&pWorker->mpool, pWorker->queue);
+    tWWorkerCleanup(&pWorker->mpool);
+    tWWorkerFreeQueue(&pWorker->mpool, pWorker->queue);
   } else {
   }
 }

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

@@ -25,7 +25,7 @@ extern "C" {
 int32_t mndInitConsumer(SMnode *pMnode);
 void    mndCleanupConsumer(SMnode *pMnode);
 
-SMqConsumerObj *mndAcquireConsumer(SMnode *pMnode, int32_t consumerId);
+SMqConsumerObj *mndAcquireConsumer(SMnode *pMnode, int64_t consumerId);
 void            mndReleaseConsumer(SMnode *pMnode, SMqConsumerObj *pConsumer);
 
 SSdbRaw *mndConsumerActionEncode(SMqConsumerObj *pConsumer);

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

@@ -363,9 +363,7 @@ typedef struct SMqConsumerEp {
   int64_t      consumerId;  // -1 for unassigned
   int64_t      lastConsumerHbTs;
   int64_t      lastVgHbTs;
-  uint32_t     qmsgLen;
   char*        qmsg;
-  //SSubQueryMsg qExec;
 } SMqConsumerEp;
 
 static FORCE_INLINE int32_t tEncodeSMqConsumerEp(void** buf, SMqConsumerEp* pConsumerEp) {
@@ -374,9 +372,10 @@ static FORCE_INLINE int32_t tEncodeSMqConsumerEp(void** buf, SMqConsumerEp* pCon
   tlen += taosEncodeFixedI32(buf, pConsumerEp->status);
   tlen += taosEncodeSEpSet(buf, &pConsumerEp->epSet);
   tlen += taosEncodeFixedI64(buf, pConsumerEp->consumerId);
+  tlen += taosEncodeFixedI64(buf, pConsumerEp->lastConsumerHbTs);
+  tlen += taosEncodeFixedI64(buf, pConsumerEp->lastVgHbTs);
   //tlen += tEncodeSSubQueryMsg(buf, &pConsumerEp->qExec);
-  tlen += taosEncodeFixedU32(buf, pConsumerEp->qmsgLen);
-  tlen += taosEncodeBinary(buf, pConsumerEp->qmsg, pConsumerEp->qmsgLen);
+  tlen += taosEncodeString(buf, pConsumerEp->qmsg);
   return tlen;
 }
 
@@ -385,9 +384,10 @@ static FORCE_INLINE void* tDecodeSMqConsumerEp(void** buf, SMqConsumerEp* pConsu
   buf = taosDecodeFixedI32(buf, &pConsumerEp->status);
   buf = taosDecodeSEpSet(buf, &pConsumerEp->epSet);
   buf = taosDecodeFixedI64(buf, &pConsumerEp->consumerId);
+  buf = taosDecodeFixedI64(buf, &pConsumerEp->lastConsumerHbTs);
+  buf = taosDecodeFixedI64(buf, &pConsumerEp->lastVgHbTs);
   //buf = tDecodeSSubQueryMsg(buf, &pConsumerEp->qExec);
-  buf = taosDecodeFixedU32(buf, &pConsumerEp->qmsgLen);
-  buf = taosDecodeBinary(buf, (void**)&pConsumerEp->qmsg, pConsumerEp->qmsgLen);
+  buf = taosDecodeString(buf, &pConsumerEp->qmsg);
   return buf;
 }
 
@@ -423,18 +423,27 @@ static FORCE_INLINE SMqSubscribeObj* tNewSubscribeObj() {
     free(pSub);
     return NULL;
   }
+  pSub->lostConsumer = taosArrayInit(0, sizeof(SMqConsumerEp));
+  if (pSub->lostConsumer == NULL) {
+    taosArrayDestroy(pSub->availConsumer);
+    taosArrayDestroy(pSub->assigned);
+    free(pSub);
+    return NULL;
+  }
   pSub->idleConsumer = taosArrayInit(0, sizeof(SMqConsumerEp));
-  if (pSub->assigned == NULL) {
+  if (pSub->idleConsumer == NULL) {
     taosArrayDestroy(pSub->availConsumer);
-    taosArrayDestroy(pSub->idleConsumer);
+    taosArrayDestroy(pSub->assigned);
+    taosArrayDestroy(pSub->lostConsumer);
     free(pSub);
     return NULL;
   }
   pSub->unassignedVg = taosArrayInit(0, sizeof(SMqConsumerEp));
-  if (pSub->assigned == NULL) {
+  if (pSub->unassignedVg == NULL) {
     taosArrayDestroy(pSub->availConsumer);
+    taosArrayDestroy(pSub->assigned);
+    taosArrayDestroy(pSub->lostConsumer);
     taosArrayDestroy(pSub->idleConsumer);
-    taosArrayDestroy(pSub->unassignedVg);
     free(pSub);
     return NULL;
   }
@@ -461,6 +470,13 @@ static FORCE_INLINE int32_t tEncodeSubscribeObj(void** buf, const SMqSubscribeOb
     tlen += tEncodeSMqConsumerEp(buf, pCEp);
   }
 
+  sz = taosArrayGetSize(pSub->lostConsumer);
+  tlen += taosEncodeFixedI32(buf, sz);
+  for (int32_t i = 0; i < sz; i++) {
+    SMqConsumerEp* pCEp = taosArrayGet(pSub->lostConsumer, i);
+    tlen += tEncodeSMqConsumerEp(buf, pCEp);
+  }
+
   sz = taosArrayGetSize(pSub->idleConsumer);
   tlen += taosEncodeFixedI32(buf, sz);
   for (int32_t i = 0; i < sz; i++) {
@@ -485,20 +501,47 @@ static FORCE_INLINE void* tDecodeSubscribeObj(void* buf, SMqSubscribeObj* pSub)
   int32_t sz;
 
   buf = taosDecodeFixedI32(buf, &sz);
-  pSub->assigned = taosArrayInit(sz, sizeof(int64_t));
-  if (pSub->assigned == NULL) {
+  pSub->availConsumer = taosArrayInit(sz, sizeof(int64_t));
+  if (pSub->availConsumer == NULL) {
     return NULL;
   }
   for (int32_t i = 0; i < sz; i++) {
     int64_t consumerId;
     buf = taosDecodeFixedI64(buf, &consumerId);
-    taosArrayPush(pSub->assigned, &consumerId);
+    taosArrayPush(pSub->availConsumer, &consumerId);
+  }
+
+  buf = taosDecodeFixedI32(buf, &sz);
+  pSub->assigned = taosArrayInit(sz, sizeof(SMqConsumerEp));
+  if (pSub->assigned == NULL) {
+    taosArrayDestroy(pSub->availConsumer);
+    return NULL;
+  }
+  for (int32_t i = 0; i < sz; i++) {
+    SMqConsumerEp cEp;
+    buf = tDecodeSMqConsumerEp(buf, &cEp);
+    taosArrayPush(pSub->assigned, &cEp);
+  }
+
+  buf = taosDecodeFixedI32(buf, &sz);
+  pSub->lostConsumer = taosArrayInit(sz, sizeof(SMqConsumerEp));
+  if (pSub->lostConsumer == NULL) {
+    taosArrayDestroy(pSub->availConsumer);
+    taosArrayDestroy(pSub->assigned);
+    return NULL;
+  }
+  for (int32_t i = 0; i < sz; i++) {
+    SMqConsumerEp cEp;
+    buf = tDecodeSMqConsumerEp(buf, &cEp);
+    taosArrayPush(pSub->lostConsumer, &cEp);
   }
 
   buf = taosDecodeFixedI32(buf, &sz);
   pSub->idleConsumer = taosArrayInit(sz, sizeof(SMqConsumerEp));
   if (pSub->idleConsumer == NULL) {
+    taosArrayDestroy(pSub->availConsumer);
     taosArrayDestroy(pSub->assigned);
+    taosArrayDestroy(pSub->lostConsumer);
     return NULL;
   }
   for (int32_t i = 0; i < sz; i++) {
@@ -507,10 +550,13 @@ static FORCE_INLINE void* tDecodeSubscribeObj(void* buf, SMqSubscribeObj* pSub)
     taosArrayPush(pSub->idleConsumer, &cEp);
   }
 
+
   buf = taosDecodeFixedI32(buf, &sz);
   pSub->unassignedVg = taosArrayInit(sz, sizeof(SMqConsumerEp));
   if (pSub->unassignedVg == NULL) {
+    taosArrayDestroy(pSub->availConsumer);
     taosArrayDestroy(pSub->assigned);
+    taosArrayDestroy(pSub->lostConsumer);
     taosArrayDestroy(pSub->idleConsumer);
     return NULL;
   }
@@ -580,7 +626,10 @@ static FORCE_INLINE int32_t tEncodeSMqConsumerTopic(void** buf, SMqConsumerTopic
   int32_t tlen = 0;
   tlen += taosEncodeString(buf, pConsumerTopic->name);
   tlen += taosEncodeFixedI32(buf, pConsumerTopic->epoch);
-  int32_t sz = taosArrayGetSize(pConsumerTopic->pVgInfo);
+  int32_t sz = 0;
+  if (pConsumerTopic->pVgInfo != NULL) {
+    sz = taosArrayGetSize(pConsumerTopic->pVgInfo);
+  }
   tlen += taosEncodeFixedI32(buf, sz);
   for (int32_t i = 0; i < sz; i++) {
     int32_t* pVgInfo = taosArrayGet(pConsumerTopic->pVgInfo, i);

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

@@ -30,24 +30,23 @@
 #define MND_CONSUMER_VER_NUMBER 1
 #define MND_CONSUMER_RESERVE_SIZE 64
 
-static int32_t  mndConsumerActionInsert(SSdb *pSdb, SMqConsumerObj *pConsumer);
-static int32_t  mndConsumerActionDelete(SSdb *pSdb, SMqConsumerObj *pConsumer);
-static int32_t  mndConsumerActionUpdate(SSdb *pSdb, SMqConsumerObj *pConsumer, SMqConsumerObj *pNewConsumer);
-static int32_t  mndProcessConsumerMetaMsg(SMnodeMsg *pMsg);
-static int32_t  mndGetConsumerMeta(SMnodeMsg *pMsg, SShowObj *pShow, STableMetaRsp *pMeta);
-static int32_t  mndRetrieveConsumer(SMnodeMsg *pMsg, SShowObj *pShow, char *data, int32_t rows);
-static void     mndCancelGetNextConsumer(SMnode *pMnode, void *pIter);
+static int32_t mndConsumerActionInsert(SSdb *pSdb, SMqConsumerObj *pConsumer);
+static int32_t mndConsumerActionDelete(SSdb *pSdb, SMqConsumerObj *pConsumer);
+static int32_t mndConsumerActionUpdate(SSdb *pSdb, SMqConsumerObj *pConsumer, SMqConsumerObj *pNewConsumer);
+static int32_t mndProcessConsumerMetaMsg(SMnodeMsg *pMsg);
+static int32_t mndGetConsumerMeta(SMnodeMsg *pMsg, SShowObj *pShow, STableMetaRsp *pMeta);
+static int32_t mndRetrieveConsumer(SMnodeMsg *pMsg, SShowObj *pShow, char *data, int32_t rows);
+static void    mndCancelGetNextConsumer(SMnode *pMnode, void *pIter);
 
 int32_t mndInitConsumer(SMnode *pMnode) {
   SSdbTable table = {.sdbType = SDB_CONSUMER,
-                     .keyType = SDB_KEY_BINARY,
+                     .keyType = SDB_KEY_INT64,
                      .encodeFp = (SdbEncodeFp)mndConsumerActionEncode,
                      .decodeFp = (SdbDecodeFp)mndConsumerActionDecode,
                      .insertFp = (SdbInsertFp)mndConsumerActionInsert,
                      .updateFp = (SdbUpdateFp)mndConsumerActionUpdate,
                      .deleteFp = (SdbDeleteFp)mndConsumerActionDelete};
 
-
   return sdbSetTable(pMnode->pSdb, table);
 }
 
@@ -61,10 +60,10 @@ SSdbRaw *mndConsumerActionEncode(SMqConsumerObj *pConsumer) {
   SSdbRaw *pRaw = sdbAllocRaw(SDB_CONSUMER, MND_CONSUMER_VER_NUMBER, size);
   if (pRaw == NULL) goto CM_ENCODE_OVER;
 
-  void* buf = malloc(tlen);
+  void *buf = malloc(tlen);
   if (buf == NULL) goto CM_ENCODE_OVER;
 
-  void* abuf = buf;
+  void *abuf = buf;
   tEncodeSMqConsumerObj(&abuf, pConsumer);
 
   int32_t dataPos = 0;
@@ -106,7 +105,7 @@ SSdbRow *mndConsumerActionDecode(SSdbRaw *pRaw) {
   int32_t dataPos = 0;
   int32_t len;
   SDB_GET_INT32(pRaw, dataPos, &len, CM_DECODE_OVER);
-  void* buf = malloc(len);
+  void *buf = malloc(len);
   if (buf == NULL) goto CM_DECODE_OVER;
   SDB_GET_BINARY(pRaw, dataPos, buf, len, CM_DECODE_OVER);
   SDB_GET_RESERVE(pRaw, dataPos, MND_CONSUMER_RESERVE_SIZE, CM_DECODE_OVER);
@@ -147,7 +146,7 @@ static int32_t mndConsumerActionUpdate(SSdb *pSdb, SMqConsumerObj *pOldConsumer,
   return 0;
 }
 
-SMqConsumerObj *mndAcquireConsumer(SMnode *pMnode, int32_t consumerId) {
+SMqConsumerObj *mndAcquireConsumer(SMnode *pMnode, int64_t consumerId) {
   SSdb           *pSdb = pMnode->pSdb;
   SMqConsumerObj *pConsumer = sdbAcquire(pSdb, SDB_CONSUMER, &consumerId);
   if (pConsumer == NULL) {

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

@@ -226,10 +226,10 @@ static int32_t mndCheckDbCfg(SMnode *pMnode, SDbCfg *pCfg) {
   if (pCfg->cacheBlockSize < TSDB_MIN_CACHE_BLOCK_SIZE || pCfg->cacheBlockSize > TSDB_MAX_CACHE_BLOCK_SIZE) return -1;
   if (pCfg->totalBlocks < TSDB_MIN_TOTAL_BLOCKS || pCfg->totalBlocks > TSDB_MAX_TOTAL_BLOCKS) return -1;
   if (pCfg->daysPerFile < TSDB_MIN_DAYS_PER_FILE || pCfg->daysPerFile > TSDB_MAX_DAYS_PER_FILE) return -1;
-  if (pCfg->daysToKeep0 < pCfg->daysPerFile) return -1;
   if (pCfg->daysToKeep0 < TSDB_MIN_KEEP || pCfg->daysToKeep0 > TSDB_MAX_KEEP) return -1;
   if (pCfg->daysToKeep1 < TSDB_MIN_KEEP || pCfg->daysToKeep1 > TSDB_MAX_KEEP) return -1;
   if (pCfg->daysToKeep2 < TSDB_MIN_KEEP || pCfg->daysToKeep2 > TSDB_MAX_KEEP) return -1;
+  if (pCfg->daysToKeep0 < pCfg->daysPerFile) return -1;
   if (pCfg->daysToKeep0 > pCfg->daysToKeep1) return -1;
   if (pCfg->daysToKeep1 > pCfg->daysToKeep2) return -1;
   if (pCfg->minRows < TSDB_MIN_MIN_ROW_FBLOCK || pCfg->minRows > TSDB_MAX_MIN_ROW_FBLOCK) return -1;
@@ -498,7 +498,7 @@ static int32_t mndProcessCreateDbReq(SMnodeMsg *pReq) {
   return TSDB_CODE_MND_ACTION_IN_PROGRESS;
 }
 
-static int32_t mndSetDbCfgFromAlterDbMsg(SDbObj *pDb, SAlterDbReq *pAlter) {
+static int32_t mndSetDbCfgFromAlterDbReq(SDbObj *pDb, SAlterDbReq *pAlter) {
   terrno = TSDB_CODE_MND_DB_OPTION_UNCHANGED;
 
   if (pAlter->totalBlocks >= 0 && pAlter->totalBlocks != pDb->cfg.totalBlocks) {
@@ -649,7 +649,7 @@ static int32_t mndProcessAlterDbReq(SMnodeMsg *pReq) {
   SDbObj dbObj = {0};
   memcpy(&dbObj, pDb, sizeof(SDbObj));
 
-  int32_t code = mndSetDbCfgFromAlterDbMsg(&dbObj, pAlter);
+  int32_t code = mndSetDbCfgFromAlterDbReq(&dbObj, pAlter);
   if (code != 0) {
     mndReleaseDb(pMnode, pDb);
     mError("db:%s, failed to alter since %s", pAlter->db, tstrerror(code));
@@ -1143,7 +1143,7 @@ static int32_t mndRetrieveDbs(SMnodeMsg *pReq, SShowObj *pShow, char *data, int3
         prec = TSDB_TIME_PRECISION_NANO_STR;
         break;
       default:
-        assert(false);
+        prec = "none";
         break;
     }
     STR_WITH_SIZE_TO_VARSTR(pWrite, prec, 2);

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

@@ -20,6 +20,7 @@
 #include "mndTrans.h"
 
 #define SDB_FUNC_VER 1
+#define SDB_FUNC_RESERVE_SIZE 64
 
 static SSdbRaw *mndFuncActionEncode(SFuncObj *pFunc);
 static SSdbRow *mndFuncActionDecode(SSdbRaw *pRaw);
@@ -60,7 +61,7 @@ void mndCleanupFunc(SMnode *pMnode) {}
 static SSdbRaw *mndFuncActionEncode(SFuncObj *pFunc) {
   terrno = TSDB_CODE_OUT_OF_MEMORY;
 
-  int32_t  size = pFunc->commentSize + pFunc->codeSize + sizeof(SFuncObj);
+  int32_t  size = pFunc->commentSize + pFunc->codeSize + sizeof(SFuncObj) + SDB_FUNC_RESERVE_SIZE;
   SSdbRaw *pRaw = sdbAllocRaw(SDB_FUNC, SDB_FUNC_VER, size);
   if (pRaw == NULL) goto FUNC_ENCODE_OVER;
 
@@ -78,6 +79,7 @@ static SSdbRaw *mndFuncActionEncode(SFuncObj *pFunc) {
   SDB_SET_INT32(pRaw, dataPos, pFunc->codeSize, FUNC_ENCODE_OVER)
   SDB_SET_BINARY(pRaw, dataPos, pFunc->pComment, pFunc->commentSize, FUNC_ENCODE_OVER)
   SDB_SET_BINARY(pRaw, dataPos, pFunc->pCode, pFunc->codeSize, FUNC_ENCODE_OVER)
+  SDB_SET_RESERVE(pRaw, dataPos, SDB_FUNC_RESERVE_SIZE, FUNC_ENCODE_OVER)
   SDB_SET_DATALEN(pRaw, dataPos, FUNC_ENCODE_OVER);
 
   terrno = 0;
@@ -131,6 +133,7 @@ static SSdbRow *mndFuncActionDecode(SSdbRaw *pRaw) {
 
   SDB_GET_BINARY(pRaw, dataPos, pFunc->pComment, pFunc->commentSize, FUNC_DECODE_OVER)
   SDB_GET_BINARY(pRaw, dataPos, pFunc->pCode, pFunc->codeSize, FUNC_DECODE_OVER)
+  SDB_GET_RESERVE(pRaw, dataPos, SDB_FUNC_RESERVE_SIZE, FUNC_DECODE_OVER)
 
   terrno = 0;
 

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

@@ -30,6 +30,8 @@
 #define MND_SUBSCRIBE_VER_NUMBER 1
 #define MND_SUBSCRIBE_RESERVE_SIZE 64
 
+static char *mndMakeSubscribeKey(char *cgroup, char *topicName);
+
 static SSdbRaw *mndSubActionEncode(SMqSubscribeObj *);
 static SSdbRow *mndSubActionDecode(SSdbRaw *pRaw);
 static int32_t  mndSubActionInsert(SSdb *pSdb, SMqSubscribeObj *);
@@ -41,9 +43,10 @@ static int32_t mndProcessSubscribeRsp(SMnodeMsg *pMsg);
 static int32_t mndProcessSubscribeInternalReq(SMnodeMsg *pMsg);
 static int32_t mndProcessSubscribeInternalRsp(SMnodeMsg *pMsg);
 static int32_t mndProcessMqTimerMsg(SMnodeMsg *pMsg);
+static int32_t mndProcessGetSubEpReq(SMnodeMsg *pMsg);
 
 static int mndBuildMqSetConsumerVgReq(SMnode *pMnode, STrans *pTrans, SMqConsumerObj *pConsumer,
-                                      SMqConsumerTopic *pConsumerTopic, SMqTopicObj *pTopic);
+                                      SMqConsumerTopic *pConsumerTopic, SMqTopicObj *pTopic, SMqConsumerEp *pSub);
 
 int32_t mndInitSubscribe(SMnode *pMnode) {
   SSdbTable table = {.sdbType = SDB_SUBSCRIBE,
@@ -57,9 +60,64 @@ int32_t mndInitSubscribe(SMnode *pMnode) {
   mndSetMsgHandle(pMnode, TDMT_MND_SUBSCRIBE, mndProcessSubscribeReq);
   mndSetMsgHandle(pMnode, TDMT_VND_MQ_SET_CONN_RSP, mndProcessSubscribeInternalRsp);
   mndSetMsgHandle(pMnode, TDMT_MND_MQ_TIMER, mndProcessMqTimerMsg);
+  mndSetMsgHandle(pMnode, TDMT_MND_GET_SUB_EP, mndProcessGetSubEpReq);
   return sdbSetTable(pMnode->pSdb, table);
 }
 
+static int32_t mndProcessGetSubEpReq(SMnodeMsg *pMsg) {
+  SMnode           *pMnode = pMsg->pMnode;
+  SMqCMGetSubEpReq *pReq = (SMqCMGetSubEpReq *)pMsg->rpcMsg.pCont;
+  SMqCMGetSubEpRsp  rsp;
+  int64_t           consumerId = be64toh(pReq->consumerId);
+
+  SMqConsumerObj *pConsumer = mndAcquireConsumer(pMsg->pMnode, consumerId);
+  if (pConsumer == NULL) {
+    terrno = TSDB_CODE_MND_CONSUMER_NOT_EXIST;
+    return -1;
+  }
+  ASSERT(strcmp(pReq->cgroup, pConsumer->cgroup) == 0);
+
+  strcpy(rsp.cgroup, pReq->cgroup);
+  rsp.consumerId = consumerId;
+  SArray *pTopics = pConsumer->topics;
+  int32_t sz = taosArrayGetSize(pTopics);
+  rsp.topics = taosArrayInit(sz, sizeof(SMqSubTopicEp));
+  for (int32_t i = 0; i < sz; i++) {
+    SMqSubTopicEp     topicEp;
+    SMqConsumerTopic *pConsumerTopic = taosArrayGet(pTopics, i);
+    strcpy(topicEp.topic, pConsumerTopic->name);
+
+    SMqSubscribeObj *pSub = mndAcquireSubscribe(pMnode, pConsumer->cgroup, pConsumerTopic->name);
+    int32_t          assignedSz = taosArrayGetSize(pSub->assigned);
+    topicEp.vgs = taosArrayInit(assignedSz, sizeof(SMqSubVgEp));
+    for (int32_t j = 0; j < assignedSz; j++) {
+      SMqConsumerEp *pCEp = taosArrayGet(pSub->assigned, j);
+      if (pCEp->consumerId == consumerId) {
+        SMqSubVgEp vgEp = {
+          .epSet = pCEp->epSet,
+          .vgId = pCEp->vgId
+        };
+        taosArrayPush(topicEp.vgs, &vgEp);
+      }
+    }
+    if (taosArrayGetSize(topicEp.vgs) != 0) {
+      taosArrayPush(rsp.topics, &topicEp);
+    }
+  }
+  int32_t tlen = tEncodeSMqCMGetSubEpRsp(NULL, &rsp);
+  void   *buf = rpcMallocCont(tlen);
+  if (buf == NULL) {
+    terrno = TSDB_CODE_OUT_OF_MEMORY;
+    return -1;
+  }
+  void *abuf = buf;
+  tEncodeSMqCMGetSubEpRsp(&abuf, &rsp);
+  //TODO: free rsp
+  pMsg->pCont = buf;
+  pMsg->contLen = tlen;
+  return 0;
+}
+
 static int32_t mndSplitSubscribeKey(char *key, char **topic, char **cgroup) {
   int i = 0;
   while (key[i] != ':') {
@@ -97,7 +155,7 @@ static int32_t mndProcessMqTimerMsg(SMnodeMsg *pMsg) {
 
         // build msg
 
-        SMqSetCVgReq* pReq = malloc(sizeof(SMqSetCVgReq) + pCEp->qmsgLen);
+        SMqSetCVgReq *pReq = malloc(sizeof(SMqSetCVgReq));
         if (pReq == NULL) {
           terrno = TSDB_CODE_OUT_OF_MEMORY;
           return -1;
@@ -107,8 +165,7 @@ static int32_t mndProcessMqTimerMsg(SMnodeMsg *pMsg) {
         pReq->sql = strdup(pTopic->sql);
         pReq->logicalPlan = strdup(pTopic->logicalPlan);
         pReq->physicalPlan = strdup(pTopic->physicalPlan);
-        pReq->qmsgLen = pCEp->qmsgLen;
-        memcpy(pReq->qmsg, pCEp->qmsg, pCEp->qmsgLen);
+        pReq->qmsg = strdup(pCEp->qmsg);
         int32_t tlen = tEncodeSMqSetCVgReq(NULL, pReq);
         void   *reqStr = malloc(tlen);
         if (reqStr == NULL) {
@@ -137,7 +194,7 @@ static int32_t mndProcessMqTimerMsg(SMnodeMsg *pMsg) {
       if (mndTransPrepare(pMnode, pTrans) != 0) {
         mError("trans:%d, failed to prepare since %s", pTrans->id, terrstr());
       }
-      mndReleaseTopic(pMnode, pTopic);
+      /*mndReleaseTopic(pMnode, pTopic);*/
       mndTransDrop(pTrans);
     }
     pIter = sdbFetch(pSdb, SDB_SUBSCRIBE, NULL, (void **)&pSub);
@@ -146,12 +203,14 @@ static int32_t mndProcessMqTimerMsg(SMnodeMsg *pMsg) {
 }
 
 static int mndInitUnassignedVg(SMnode *pMnode, SMqTopicObj *pTopic, SArray *unassignedVg) {
-  //convert phyplan to dag
+  // convert phyplan to dag
   SQueryDag *pDag = qStringToDag(pTopic->physicalPlan);
-  SArray *pArray;
-  SArray* inner = taosArrayGet(pDag->pSubplans, 0);
-  SSubplan *plan = taosArrayGetP(inner, 0);
-
+  SArray    *pArray;
+  SArray    *inner = taosArrayGet(pDag->pSubplans, 0);
+  SSubplan  *plan = taosArrayGetP(inner, 0);
+  plan->execNode.inUse = 0;
+  strcpy(plan->execNode.epAddr[0].fqdn, "localhost");
+  plan->execNode.epAddr[0].port = 6030;
   plan->execNode.nodeId = 2;
   SEpSet* pEpSet = &plan->execNode.epset;
 
@@ -163,21 +222,18 @@ static int mndInitUnassignedVg(SMnode *pMnode, SMqTopicObj *pTopic, SArray *unas
     return -1;
   }
   int32_t sz = taosArrayGetSize(pArray);
-  //convert dag to msg
+  // convert dag to msg
   for (int32_t i = 0; i < sz; i++) {
     SMqConsumerEp CEp;
     CEp.status = 0;
+    CEp.consumerId = -1;
     CEp.lastConsumerHbTs = CEp.lastVgHbTs = -1;
-    STaskInfo* pTaskInfo = taosArrayGet(pArray, i);
-    CEp.epSet = pTaskInfo->addr.epset;
-    /*mDebug("subscribe convert ep %d %s %s %s %s %s\n", CEp.epSet.numOfEps, CEp.epSet.fqdn[0], CEp.epSet.fqdn[1], CEp.epSet.fqdn[2], CEp.epSet.fqdn[3], CEp.epSet.fqdn[4]);*/
+    STaskInfo *pTaskInfo = taosArrayGet(pArray, i);
+    tConvertQueryAddrToEpSet(&CEp.epSet, &pTaskInfo->addr);
+    /*mDebug("subscribe convert ep %d %s %s %s %s %s\n", CEp.epSet.numOfEps, CEp.epSet.fqdn[0], CEp.epSet.fqdn[1],
+     * CEp.epSet.fqdn[2], CEp.epSet.fqdn[3], CEp.epSet.fqdn[4]);*/
     CEp.vgId = pTaskInfo->addr.nodeId;
-    CEp.qmsgLen = pTaskInfo->msg->contentLen;
-    CEp.qmsg = malloc(CEp.qmsgLen);
-    if (CEp.qmsg == NULL) {
-      return -1;
-    }
-    memcpy(CEp.qmsg, pTaskInfo->msg->msg, pTaskInfo->msg->contentLen);
+    CEp.qmsg = strdup(pTaskInfo->msg->msg);
     taosArrayPush(unassignedVg, &CEp);
   }
 
@@ -186,7 +242,7 @@ static int mndInitUnassignedVg(SMnode *pMnode, SMqTopicObj *pTopic, SArray *unas
 }
 
 static int mndBuildMqSetConsumerVgReq(SMnode *pMnode, STrans *pTrans, SMqConsumerObj *pConsumer,
-                                      SMqConsumerTopic *pConsumerTopic, SMqTopicObj *pTopic) {
+                                      SMqConsumerTopic *pConsumerTopic, SMqTopicObj *pTopic, SMqConsumerEp *pCEp) {
   int32_t sz = taosArrayGetSize(pConsumerTopic->pVgInfo);
   for (int32_t i = 0; i < sz; i++) {
     int32_t      vgId = *(int32_t *)taosArrayGet(pConsumerTopic->pVgInfo, i);
@@ -201,6 +257,7 @@ static int mndBuildMqSetConsumerVgReq(SMnode *pMnode, STrans *pTrans, SMqConsume
     req.sql = pTopic->sql;
     req.logicalPlan = pTopic->logicalPlan;
     req.physicalPlan = pTopic->physicalPlan;
+    req.qmsg = pCEp->qmsg;
     int32_t tlen = tEncodeSMqSetCVgReq(NULL, &req);
     void   *buf = malloc(sizeof(SMsgHead) + tlen);
     if (buf == NULL) {
@@ -208,18 +265,18 @@ static int mndBuildMqSetConsumerVgReq(SMnode *pMnode, STrans *pTrans, SMqConsume
       return -1;
     }
 
-    SMsgHead* pMsgHead = (SMsgHead*)buf;
+    SMsgHead *pMsgHead = (SMsgHead *)buf;
 
     pMsgHead->contLen = htonl(sizeof(SMsgHead) + tlen);
     pMsgHead->vgId = htonl(vgId);
 
-    void* abuf = POINTER_SHIFT(buf, sizeof(SMsgHead));
+    void *abuf = POINTER_SHIFT(buf, sizeof(SMsgHead));
     tEncodeSMqSetCVgReq(&abuf, &req);
 
     STransAction action = {0};
     action.epSet = mndGetVgroupEpset(pMnode, pVgObj);
     action.pCont = buf;
-    action.contLen = tlen;
+    action.contLen = sizeof(SMsgHead) + tlen;
     action.msgType = TDMT_VND_MQ_SET_CONN;
 
     mndReleaseVgroup(pMnode, pVgObj);
@@ -287,7 +344,7 @@ static SSdbRow *mndSubActionDecode(SSdbRaw *pRaw) {
   int32_t dataPos = 0;
   int32_t tlen;
   SDB_GET_INT32(pRaw, dataPos, &tlen, SUB_DECODE_OVER);
-  void   *buf = malloc(tlen + 1);
+  void *buf = malloc(tlen + 1);
   if (buf == NULL) goto SUB_DECODE_OVER;
   SDB_GET_BINARY(pRaw, dataPos, buf, tlen, SUB_DECODE_OVER);
   SDB_GET_RESERVE(pRaw, dataPos, MND_SUBSCRIBE_RESERVE_SIZE, SUB_DECODE_OVER);
@@ -495,11 +552,11 @@ static int32_t mndProcessSubscribeReq(SMnodeMsg *pMsg) {
           terrno = TSDB_CODE_OUT_OF_MEMORY;
           return -1;
         }
-        char* key = mndMakeSubscribeKey(consumerGroup, newTopicName);
+        char *key = mndMakeSubscribeKey(consumerGroup, newTopicName);
         strcpy(pSub->key, key);
         // set unassigned vg
         mndInitUnassignedVg(pMnode, pTopic, pSub->unassignedVg);
-        //TODO: disable alter
+        // TODO: disable alter
       }
       taosArrayPush(pSub->availConsumer, &consumerId);
 
@@ -508,12 +565,15 @@ static int32_t mndProcessSubscribeReq(SMnodeMsg *pMsg) {
 
       if (taosArrayGetSize(pConsumerTopic->pVgInfo) > 0) {
         ASSERT(taosArrayGetSize(pConsumerTopic->pVgInfo) == 1);
-        int32_t vgId = *(int32_t *)taosArrayGetLast(pConsumerTopic->pVgInfo);
-        // send setmsg to vnode
-        if (mndBuildMqSetConsumerVgReq(pMnode, pTrans, pConsumer, pConsumerTopic, pTopic) < 0) {
-          // TODO
-          return -1;
+        int32_t        vgId = *(int32_t *)taosArrayGetLast(pConsumerTopic->pVgInfo);
+        SMqConsumerEp *pCEp = taosArrayGetLast(pSub->assigned);
+        if (pCEp->vgId == vgId) {
+          if (mndBuildMqSetConsumerVgReq(pMnode, pTrans, pConsumer, pConsumerTopic, pTopic, pCEp) < 0) {
+            // TODO
+            return -1;
+          }
         }
+        // send setmsg to vnode
       }
 
       SSdbRaw *pRaw = mndSubActionEncode(pSub);
@@ -570,14 +630,14 @@ static int32_t mndProcessSubscribeReq(SMnodeMsg *pMsg) {
     mError("trans:%d, failed to prepare since %s", pTrans->id, terrstr());
     if (newSub) taosArrayDestroy(newSub);
     mndTransDrop(pTrans);
-    mndReleaseConsumer(pMnode, pConsumer);
+    /*mndReleaseConsumer(pMnode, pConsumer);*/
     return -1;
   }
 
   if (newSub) taosArrayDestroy(newSub);
   mndTransDrop(pTrans);
-  mndReleaseConsumer(pMnode, pConsumer);
-  return 0;
+  /*mndReleaseConsumer(pMnode, pConsumer);*/
+  return TSDB_CODE_MND_ACTION_IN_PROGRESS;
 }
 
 static int32_t mndProcessSubscribeInternalRsp(SMnodeMsg *pRsp) {

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

@@ -237,7 +237,7 @@ static int32_t mndCreateTopic(SMnode *pMnode, SMnodeMsg *pMsg, SCMCreateTopicReq
   tstrncpy(topicObj.db, pDb->name, TSDB_DB_FNAME_LEN);
   topicObj.createTime = taosGetTimestampMs();
   topicObj.updateTime = topicObj.createTime;
-  topicObj.uid = mndGenerateUid(pCreate->name, TSDB_TABLE_FNAME_LEN);
+  topicObj.uid = mndGenerateUid(pCreate->name, strlen(pCreate->name));
   topicObj.dbUid = pDb->uid;
   topicObj.version = 1;
   topicObj.sql = strdup(pCreate->sql);

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

@@ -921,7 +921,7 @@ static int32_t mndProcessTransMsg(SMnodeMsg *pMsg) {
 
 void mndTransPullup(SMnode *pMnode) {
   STrans *pTrans = NULL;
-  void   *pIter = NULL;
+  void *  pIter = NULL;
 
   while (1) {
     pIter = sdbFetch(pMnode->pSdb, SDB_TRANS, pIter, (void **)&pTrans);
@@ -930,4 +930,6 @@ void mndTransPullup(SMnode *pMnode) {
     mndTransExecute(pMnode, pTrans);
     sdbRelease(pMnode->pSdb, pTrans);
   }
+
+  sdbWriteFile(pMnode->pSdb);
 }

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

@@ -178,7 +178,7 @@ static int32_t mndVgroupActionUpdate(SSdb *pSdb, SVgObj *pOld, SVgObj *pNew) {
 SVgObj *mndAcquireVgroup(SMnode *pMnode, int32_t vgId) {
   SSdb   *pSdb = pMnode->pSdb;
   SVgObj *pVgroup = sdbAcquire(pSdb, SDB_VGROUP, &vgId);
-  if (pVgroup == NULL) {
+  if (pVgroup == NULL && terrno == TSDB_CODE_SDB_OBJ_NOT_THERE) {
     terrno = TSDB_CODE_MND_VGROUP_NOT_EXIST;
   }
   return pVgroup;

source/dnode/mnode/sdb/src/sdb.c

@@ -64,11 +64,7 @@ SSdb *sdbInit(SSdbOpt *pOption) {
 void sdbCleanup(SSdb *pSdb) {
   mDebug("start to cleanup sdb");
 
-  if (pSdb->curVer > pSdb->lastCommitVer) {
-    mDebug("write sdb file for current ver:%" PRId64 " larger than last commit ver:%" PRId64, pSdb->curVer,
-           pSdb->lastCommitVer);
-    sdbWriteFile(pSdb);
-  }
+  sdbWriteFile(pSdb);
 
   if (pSdb->currDir != NULL) {
     tfree(pSdb->currDir);

source/dnode/mnode/sdb/src/sdbFile.c

@@ -221,7 +221,7 @@ PARSE_SDB_DATA_ERROR:
   return code;
 }
 
-int32_t sdbWriteFile(SSdb *pSdb) {
+static int32_t sdbWriteFileImp(SSdb *pSdb) {
   int32_t code = 0;
 
   char tmpfile[PATH_MAX] = {0};
@@ -229,7 +229,8 @@ int32_t sdbWriteFile(SSdb *pSdb) {
   char curfile[PATH_MAX] = {0};
   snprintf(curfile, sizeof(curfile), "%s%ssdb.data", pSdb->currDir, TD_DIRSEP);
 
-  mDebug("start to write file:%s", curfile);
+  mDebug("start to write file:%s, current ver:%" PRId64 ", commit ver:%" PRId64, curfile, pSdb->curVer,
+         pSdb->lastCommitVer);
 
   FileFd fd = taosOpenFileCreateWriteTrunc(tmpfile);
   if (fd <= 0) {
@@ -323,12 +324,20 @@ int32_t sdbWriteFile(SSdb *pSdb) {
   return code;
 }
 
+int32_t sdbWriteFile(SSdb *pSdb) {
+  if (pSdb->curVer == pSdb->lastCommitVer) {
+    return 0;
+  }
+
+  return sdbWriteFileImp(pSdb);
+}
+
 int32_t sdbDeploy(SSdb *pSdb) {
   if (sdbRunDeployFp(pSdb) != 0) {
     return -1;
   }
 
-  if (sdbWriteFile(pSdb) != 0) {
+  if (sdbWriteFileImp(pSdb) != 0) {
     return -1;
   }
 

source/dnode/vnode/inc/tq.h

@@ -318,8 +318,8 @@ int       tqRegisterContext(STqGroup*, void* ahandle);
 int       tqSendLaunchQuery(STqMsgItem*, int64_t offset);
 #endif
 
-int32_t tqProcessConsumeReq(STQ* pTq, SRpcMsg* pMsg, SRpcMsg** ppRsp);
-int32_t tqProcessSetConnReq(STQ* pTq, char* msg, SRpcMsg** ppRsp);
+int32_t tqProcessConsumeReq(STQ* pTq, SRpcMsg* pMsg);
+int32_t tqProcessSetConnReq(STQ* pTq, char* msg);
 
 #ifdef __cplusplus
 }

source/dnode/vnode/inc/vnode.h

@@ -33,6 +33,7 @@ extern "C" {
 typedef struct SVnode SVnode;
 typedef struct SDnode SDnode;
 typedef int32_t (*PutReqToVQueryQFp)(SDnode *pDnode, struct SRpcMsg *pReq);
+typedef int32_t (*SendReqToDnodeFp)(SDnode *pDnode, struct SEpSet *epSet, struct SRpcMsg *rpcMsg);
 
 typedef struct STqCfg {
   // TODO
@@ -64,6 +65,7 @@ typedef struct {
   const char       *charset;
   uint16_t          nthreads;  // number of commit threads. 0 for no threads and a schedule queue should be given (TODO)
   PutReqToVQueryQFp putReqToVQueryQFp;
+  SendReqToDnodeFp  sendReqToDnodeFp;
 } SVnodeOpt;
 
 typedef struct STqReadHandle {
@@ -159,20 +161,18 @@ int vnodeProcessSyncReq(SVnode *pVnode, SRpcMsg *pMsg, SRpcMsg **pRsp);
  *
  * @param pVnode The vnode object.
  * @param pMsg The request message
- * @param pRsp The response message
  * @return int 0 for success, -1 for failure
  */
-int vnodeProcessQueryReq(SVnode *pVnode, SRpcMsg *pMsg, SRpcMsg **pRsp);
+int vnodeProcessQueryMsg(SVnode *pVnode, SRpcMsg *pMsg);
 
 /**
  * @brief Process a fetch message.
  *
  * @param pVnode The vnode object.
  * @param pMsg The request message
- * @param pRsp The response message
  * @return int 0 for success, -1 for failure
  */
-int vnodeProcessFetchReq(SVnode *pVnode, SRpcMsg *pMsg, SRpcMsg **pRsp);
+int vnodeProcessFetchMsg(SVnode *pVnode, SRpcMsg *pMsg);
 
 /* ------------------------ SVnodeCfg ------------------------ */
 /**

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

@@ -55,6 +55,7 @@ typedef struct SVnodeMgr {
   // For vnode Mgmt
   SDnode*           pDnode;
   PutReqToVQueryQFp putReqToVQueryQFp;
+  SendReqToDnodeFp  sendReqToDnodeFp;
 } SVnodeMgr;
 
 extern SVnodeMgr vnodeMgr;
@@ -85,6 +86,7 @@ struct SVnode {
 int vnodeScheduleTask(SVnodeTask* task);
 
 int32_t vnodePutReqToVQueryQ(SVnode* pVnode, struct SRpcMsg* pReq);
+void    vnodeSendReqToDnode(SVnode* pVnode, struct SEpSet* epSet, struct SRpcMsg* pReq);
 
 // For Log
 extern int32_t vDebugFlag;

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

@@ -667,8 +667,8 @@ int tqItemSSize() {
 }
 #endif
 
-int32_t tqProcessConsumeReq(STQ* pTq, SRpcMsg* pMsg, SRpcMsg** ppRsp) {
-  SMqConsumeReq* pReq = pMsg->pCont;
+int32_t tqProcessConsumeReq(STQ* pTq, SRpcMsg* pMsg) {
+  SMqConsumeReq*   pReq = pMsg->pCont;
   SRpcMsg          rpcMsg;
   int64_t          reqId = pReq->reqId;
   int64_t          consumerId = pReq->consumerId;
@@ -679,6 +679,7 @@ int32_t tqProcessConsumeReq(STQ* pTq, SRpcMsg* pMsg, SRpcMsg** ppRsp) {
   int              rspLen = 0;
 
   STqConsumerHandle* pConsumer = tqHandleGet(pTq->tqMeta, consumerId);
+  ASSERT(pConsumer);
   int                sz = taosArrayGetSize(pConsumer->topics);
 
   for (int i = 0; i < sz; i++) {
@@ -735,23 +736,6 @@ int32_t tqProcessConsumeReq(STQ* pTq, SRpcMsg* pMsg, SRpcMsg** ppRsp) {
         break;
       }
       if (pDataBlock != NULL) {
-        SMqTbData tbData = {
-          .uid = pDataBlock->info.uid,
-          .numOfCols = pDataBlock->info.numOfCols,
-          .numOfRows = pDataBlock->info.rows,
-        };
-        for (int i = 0; i < pDataBlock->info.numOfCols; i++) {
-          SColumnInfoData* pColData = taosArrayGet(pDataBlock->pDataBlock, i);
-          int32_t sz = pColData->info.bytes * pDataBlock->info.rows;
-          SMqColData colData = {
-            .bytes = pColData->info.bytes,
-            .colId = pColData->info.colId,
-            .type = pColData->info.type,
-          };
-          memcpy(colData.data, pColData->pData, colData.bytes * pDataBlock->info.rows);
-          memcpy(&tbData.colData[i], &colData, sz);
-        }
-        /*pDataBlock->info.*/
         taosArrayPush(pRes, pDataBlock);
       } else {
         break;
@@ -773,17 +757,17 @@ int32_t tqProcessConsumeReq(STQ* pTq, SRpcMsg* pMsg, SRpcMsg** ppRsp) {
     if (pTopic->buffer.lastOffset == -1 || pReq->offset > pTopic->buffer.lastOffset) {
       pTopic->buffer.lastOffset = pReq->offset;
     }
-    // put output into rsp
-    SMqConsumeRsp rsp = {
-      .consumerId = consumerId,
-      .numOfTopics = 1
-    };
   }
+    // put output into rsp
+  SMqConsumeRsp rsp = {
+    .consumerId = consumerId,
+    .numOfTopics = 1
+  };
 
   return 0;
 }
 
-int32_t tqProcessSetConnReq(STQ* pTq, char* msg, SRpcMsg** ppRsp) {
+int32_t tqProcessSetConnReq(STQ* pTq, char* msg) {
   SMqSetCVgReq req;
   tDecodeSMqSetCVgReq(msg, &req);
   STqConsumerHandle* pConsumer = calloc(sizeof(STqConsumerHandle), 1);
@@ -792,6 +776,8 @@ int32_t tqProcessSetConnReq(STQ* pTq, char* msg, SRpcMsg** ppRsp) {
   }
   strcpy(pConsumer->cgroup, req.cgroup);
   pConsumer->topics = taosArrayInit(0, sizeof(STqTopicHandle));
+  pConsumer->consumerId = req.newConsumerId;
+  pConsumer->epoch = 0;
 
   STqTopicHandle* pTopic = calloc(sizeof(STqTopicHandle), 1);
   if (pTopic == NULL) {
@@ -802,6 +788,8 @@ int32_t tqProcessSetConnReq(STQ* pTq, char* msg, SRpcMsg** ppRsp) {
   pTopic->sql = strdup(req.sql);
   pTopic->logicalPlan = strdup(req.logicalPlan);
   pTopic->physicalPlan = strdup(req.physicalPlan);
+  pTopic->committedOffset = -1;
+  pTopic->currentOffset = -1;
 
   pTopic->buffer.firstOffset = -1;
   pTopic->buffer.lastOffset = -1;
@@ -811,9 +799,11 @@ int32_t tqProcessSetConnReq(STQ* pTq, char* msg, SRpcMsg** ppRsp) {
   for (int i = 0; i < TQ_BUFFER_SIZE; i++) {
     pTopic->buffer.output[i].status = 0;
     STqReadHandle* pReadHandle = tqInitSubmitMsgScanner(pTq->pMeta);
-    pTopic->buffer.output[i].task = qCreateStreamExecTaskInfo(&req.qmsg, pReadHandle);
+    pTopic->buffer.output[i].task = qCreateStreamExecTaskInfo(req.qmsg, pReadHandle);
   }
   taosArrayPush(pConsumer->topics, pTopic);
+  tqHandleMovePut(pTq->tqMeta, req.newConsumerId, pConsumer);
+  tqHandleCommit(pTq->tqMeta, req.newConsumerId);
   terrno = TSDB_CODE_SUCCESS;
   return 0;
 }
@@ -832,14 +822,29 @@ STqReadHandle* tqInitSubmitMsgScanner(SMeta* pMeta) {
 
 void tqReadHandleSetMsg(STqReadHandle* pReadHandle, SSubmitMsg* pMsg, int64_t ver) {
   pReadHandle->pMsg = pMsg;
+  pMsg->length = htonl(pMsg->length);
+  pMsg->numOfBlocks = htonl(pMsg->numOfBlocks);
   tInitSubmitMsgIter(pMsg, &pReadHandle->msgIter);
   pReadHandle->ver = ver;
   memset(&pReadHandle->blkIter, 0, sizeof(SSubmitBlkIter));
 }
 
 bool tqNextDataBlock(STqReadHandle* pHandle) {
-  while (tGetSubmitMsgNext(&pHandle->msgIter, &pHandle->pBlock) >= 0) {
-    if (pHandle->tbUid == pHandle->pBlock->uid) return true;
+  while (1) {
+    if (tGetSubmitMsgNext(&pHandle->msgIter, &pHandle->pBlock) < 0) {
+      return false;
+    }
+    if (pHandle->pBlock == NULL) return false;
+
+    pHandle->pBlock->uid = htobe64(pHandle->pBlock->uid);
+    if (pHandle->tbUid == pHandle->pBlock->uid){
+      pHandle->pBlock->tid = htonl(pHandle->pBlock->tid);
+      pHandle->pBlock->sversion = htonl(pHandle->pBlock->sversion);
+      pHandle->pBlock->dataLen = htonl(pHandle->pBlock->dataLen);
+      pHandle->pBlock->schemaLen = htonl(pHandle->pBlock->schemaLen);
+      pHandle->pBlock->numOfRows = htons(pHandle->pBlock->numOfRows);
+     return true;
+    }
   }
   return false;
 }
@@ -855,8 +860,18 @@ int tqRetrieveDataBlockInfo(STqReadHandle* pHandle, SDataBlockInfo* pBlockInfo)
 
 SArray* tqRetrieveDataBlock(STqReadHandle* pHandle) {
   int32_t         sversion = pHandle->pBlock->sversion;
-  SSchemaWrapper* pSchemaWrapper = metaGetTableSchema(pHandle->pMeta, pHandle->pBlock->uid, sversion, true);
-  STSchema*       pTschema = metaGetTbTSchema(pHandle->pMeta, pHandle->pBlock->uid, sversion);
+  //TODO : change sversion
+  STSchema*       pTschema = metaGetTbTSchema(pHandle->pMeta, pHandle->pBlock->uid, 0);
+
+  tb_uid_t quid;
+  STbCfg* pTbCfg = metaGetTbInfoByUid(pHandle->pMeta, pHandle->pBlock->uid);
+  if (pTbCfg->type == META_CHILD_TABLE) {
+    quid = pTbCfg->ctbCfg.suid;
+  } else {
+    quid = pHandle->pBlock->uid;
+  }
+
+  SSchemaWrapper* pSchemaWrapper = metaGetTableSchema(pHandle->pMeta, quid, 0, true);
   SArray*         pArray = taosArrayInit(pSchemaWrapper->nCols, sizeof(SColumnInfoData));
   if (pArray == NULL) {
     return NULL;

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

@@ -26,6 +26,7 @@ int vnodeInit(const SVnodeOpt *pOption) {
 
   vnodeMgr.stop = false;
   vnodeMgr.putReqToVQueryQFp = pOption->putReqToVQueryQFp;
+  vnodeMgr.sendReqToDnodeFp = pOption->sendReqToDnodeFp;
 
   // Start commit handers
   if (pOption->nthreads > 0) {
@@ -96,6 +97,10 @@ int32_t vnodePutReqToVQueryQ(SVnode* pVnode, struct SRpcMsg* pReq) {
   return (*vnodeMgr.putReqToVQueryQFp)(pVnode->pDnode, pReq);
 }
 
+void vnodeSendReqToDnode(SVnode* pVnode, struct SEpSet* epSet, struct SRpcMsg* pReq) {
+  (*vnodeMgr.sendReqToDnodeFp)(pVnode->pDnode, epSet, pReq);
+}
+
 /* ------------------------ STATIC METHODS ------------------------ */
 static void* loop(void* arg) {
   setThreadName("vnode-commit");

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

@@ -17,14 +17,14 @@
 #include "vnd.h"
 
 static int32_t vnodeGetTableList(SVnode *pVnode, SRpcMsg *pMsg);
-static int     vnodeGetTableMeta(SVnode *pVnode, SRpcMsg *pMsg, SRpcMsg **pRsp);
+static int     vnodeGetTableMeta(SVnode *pVnode, SRpcMsg *pMsg);
 
 int vnodeQueryOpen(SVnode *pVnode) {
   return qWorkerInit(NODE_TYPE_VNODE, pVnode->vgId, NULL, (void **)&pVnode->pQuery, pVnode,
-                     (putReqToQueryQFp)vnodePutReqToVQueryQ);
+                     (putReqToQueryQFp)vnodePutReqToVQueryQ, (sendReqToDnodeFp)vnodeSendReqToDnode);
 }
 
-int vnodeProcessQueryReq(SVnode *pVnode, SRpcMsg *pMsg, SRpcMsg **pRsp) {
+int vnodeProcessQueryMsg(SVnode *pVnode, SRpcMsg *pMsg) {
   vTrace("message in query queue is processing");
 
   switch (pMsg->msgType) {
@@ -38,11 +38,13 @@ int vnodeProcessQueryReq(SVnode *pVnode, SRpcMsg *pMsg, SRpcMsg **pRsp) {
   }
 }
 
-int vnodeProcessFetchReq(SVnode *pVnode, SRpcMsg *pMsg, SRpcMsg **pRsp) {
+int vnodeProcessFetchMsg(SVnode *pVnode, SRpcMsg *pMsg) {
   vTrace("message in fetch queue is processing");
   switch (pMsg->msgType) {
     case TDMT_VND_FETCH:
       return qWorkerProcessFetchMsg(pVnode, pVnode->pQuery, pMsg);
+    case TDMT_VND_FETCH_RSP:
+      return qWorkerProcessFetchRsp(pVnode, pVnode->pQuery, pMsg);
     case TDMT_VND_RES_READY:
       return qWorkerProcessReadyMsg(pVnode, pVnode->pQuery, pMsg);
     case TDMT_VND_TASKS_STATUS:
@@ -57,16 +59,16 @@ int vnodeProcessFetchReq(SVnode *pVnode, SRpcMsg *pMsg, SRpcMsg **pRsp) {
       return vnodeGetTableList(pVnode, pMsg);
       //      return qWorkerProcessShowFetchMsg(pVnode->pMeta, pVnode->pQuery, pMsg);
     case TDMT_VND_TABLE_META:
-      return vnodeGetTableMeta(pVnode, pMsg, pRsp);
+      return vnodeGetTableMeta(pVnode, pMsg);
     case TDMT_VND_CONSUME:
-      return tqProcessConsumeReq(pVnode->pTq, pMsg, pRsp);
+      return tqProcessConsumeReq(pVnode->pTq, pMsg);
     default:
       vError("unknown msg type:%d in fetch queue", pMsg->msgType);
       return TSDB_CODE_VND_APP_ERROR;
   }
 }
 
-static int vnodeGetTableMeta(SVnode *pVnode, SRpcMsg *pMsg, SRpcMsg **pRsp) {
+static int vnodeGetTableMeta(SVnode *pVnode, SRpcMsg *pMsg) {
   STableInfoReq * pReq = (STableInfoReq *)(pMsg->pCont);
   STbCfg *        pTbCfg = NULL;
   STbCfg *        pStbCfg = NULL;

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

@@ -115,7 +115,7 @@ int vnodeApplyWMsg(SVnode *pVnode, SRpcMsg *pMsg, SRpcMsg **pRsp) {
       }
       break;
     case TDMT_VND_MQ_SET_CONN: {
-      if (tqProcessSetConnReq(pVnode->pTq, POINTER_SHIFT(ptr, sizeof(SMsgHead)), NULL) < 0) {
+      if (tqProcessSetConnReq(pVnode->pTq, POINTER_SHIFT(ptr, sizeof(SMsgHead))) < 0) {
         // TODO: handle error
       }
     } break;

source/libs/executor/src/executorMain.c

@@ -83,8 +83,9 @@ int32_t qCreateExecTask(void* readHandle, int32_t vgId, uint64_t taskId, SSubpla
   if (code != TSDB_CODE_SUCCESS) {
     goto _error;
   }
-
-  code = dsCreateDataSinker(pSubplan->pDataSink, handle);
+  if (handle) {
+    code = dsCreateDataSinker(pSubplan->pDataSink, handle);
+  }
 
   _error:
   // if failed to add ref for all tables in this query, abort current query
@@ -281,4 +282,4 @@ int32_t qKillQueryByQId(void* pMgmt, int64_t qId, int32_t waitMs, int32_t waitCo
   return error;
 }
 
-#endif
\ No newline at end of file
+#endif

source/libs/executor/src/executorimpl.c

@@ -5115,7 +5115,7 @@ static void destroySendMsgInfo(SMsgSendInfo* pMsgBody) {
   tfree(pMsgBody);
 }
 
-void processRspMsg(void* parent, SRpcMsg* pMsg, SEpSet* pEpSet) {
+void qProcessFetchRsp(void* parent, SRpcMsg* pMsg, SEpSet* pEpSet) {
   SMsgSendInfo *pSendInfo = (SMsgSendInfo *) pMsg->ahandle;
   assert(pMsg->ahandle != NULL);
 
@@ -5291,13 +5291,14 @@ SOperatorInfo* createExchangeOperatorInfo(const SArray* pSources, const SArray*
   pOperator->exec         = doLoadRemoteData;
   pOperator->pTaskInfo    = pTaskInfo;
 
+#if 1
   { // todo refactor
     SRpcInit rpcInit;
     memset(&rpcInit, 0, sizeof(rpcInit));
     rpcInit.localPort = 0;
     rpcInit.label = "EX";
     rpcInit.numOfThreads = 1;
-    rpcInit.cfp = processRspMsg;
+    rpcInit.cfp = qProcessFetchRsp;
     rpcInit.sessions = tsMaxConnections;
     rpcInit.connType = TAOS_CONN_CLIENT;
     rpcInit.user = (char *)"root";
@@ -5311,7 +5312,7 @@ SOperatorInfo* createExchangeOperatorInfo(const SArray* pSources, const SArray*
       return NULL; // todo
     }
   }
-
+#endif
   return pOperator;
 }
 

source/libs/qworker/inc/qworkerInt.h

@@ -137,6 +137,7 @@ typedef struct SQWorkerMgmt {
   SHashObj        *ctxHash;       //key: queryId+taskId, value: SQWTaskCtx
   void            *nodeObj;
   putReqToQueryQFp putToQueueFp;
+  sendReqToDnodeFp sendReqFp;
 } SQWorkerMgmt;
 
 #define QW_FPARAMS_DEF SQWorkerMgmt *mgmt, uint64_t sId, uint64_t qId, uint64_t tId

source/libs/qworker/src/qworker.c

@@ -1313,12 +1313,13 @@ _return:
   QW_RET(code);
 }
 
-int32_t qWorkerInit(int8_t nodeType, int32_t nodeId, SQWorkerCfg *cfg, void **qWorkerMgmt, void *nodeObj, putReqToQueryQFp fp) {
-  if (NULL == qWorkerMgmt || NULL == nodeObj || NULL == fp) {
+int32_t qWorkerInit(int8_t nodeType, int32_t nodeId, SQWorkerCfg *cfg, void **qWorkerMgmt, void *nodeObj,
+                    putReqToQueryQFp fp1, sendReqToDnodeFp fp2) {
+  if (NULL == qWorkerMgmt || NULL == nodeObj || NULL == fp1 || NULL == fp2) {
     qError("invalid param to init qworker");
     QW_RET(TSDB_CODE_QRY_INVALID_INPUT);
   }
-  
+
   SQWorkerMgmt *mgmt = calloc(1, sizeof(SQWorkerMgmt));
   if (NULL == mgmt) {
     qError("calloc %d failed", (int32_t)sizeof(SQWorkerMgmt));
@@ -1361,7 +1362,8 @@ int32_t qWorkerInit(int8_t nodeType, int32_t nodeId, SQWorkerCfg *cfg, void **qW
   mgmt->nodeType = nodeType;
   mgmt->nodeId = nodeId;
   mgmt->nodeObj = nodeObj;
-  mgmt->putToQueueFp = fp;
+  mgmt->putToQueueFp = fp1;
+  mgmt->sendReqFp = fp2;
 
   *qWorkerMgmt = mgmt;
 

source/libs/qworker/src/qworkerMsg.c

@@ -421,6 +421,11 @@ int32_t qWorkerProcessFetchMsg(void *node, void *qWorkerMgmt, SRpcMsg *pMsg) {
   return TSDB_CODE_SUCCESS;  
 }
 
+int32_t qWorkerProcessFetchRsp(void *node, void *qWorkerMgmt, SRpcMsg *pMsg) {
+  qProcessFetchRsp(NULL, pMsg, NULL);
+  return TSDB_CODE_SUCCESS;
+}
+
 int32_t qWorkerProcessCancelMsg(void *node, void *qWorkerMgmt, SRpcMsg *pMsg) {
   if (NULL == node || NULL == qWorkerMgmt || NULL == pMsg) {
     return TSDB_CODE_QRY_INVALID_INPUT;

source/libs/qworker/test/qworkerTests.cpp

@@ -1081,7 +1081,7 @@ TEST(rcTest, shortExecshortDelay) {
   qwtTestStop = false;
   qwtTestQuitThreadNum = 0;
 
-  code = qWorkerInit(NODE_TYPE_VNODE, 1, NULL, &mgmt, mockPointer, qwtPutReqToQueue);
+  code = qWorkerInit(NODE_TYPE_VNODE, 1, NULL, &mgmt, mockPointer, qwtPutReqToQueue, NULL);
   ASSERT_EQ(code, 0);
 
   qwtTestMaxExecTaskUsec = 0;
@@ -1162,7 +1162,7 @@ TEST(rcTest, longExecshortDelay) {
   qwtTestStop = false;
   qwtTestQuitThreadNum = 0;
 
-  code = qWorkerInit(NODE_TYPE_VNODE, 1, NULL, &mgmt, mockPointer, qwtPutReqToQueue);
+  code = qWorkerInit(NODE_TYPE_VNODE, 1, NULL, &mgmt, mockPointer, qwtPutReqToQueue, NULL);
   ASSERT_EQ(code, 0);
 
   qwtTestMaxExecTaskUsec = 1000000;
@@ -1245,7 +1245,7 @@ TEST(rcTest, shortExeclongDelay) {
   qwtTestStop = false;
   qwtTestQuitThreadNum = 0;
 
-  code = qWorkerInit(NODE_TYPE_VNODE, 1, NULL, &mgmt, mockPointer, qwtPutReqToQueue);
+  code = qWorkerInit(NODE_TYPE_VNODE, 1, NULL, &mgmt, mockPointer, qwtPutReqToQueue, NULL);
   ASSERT_EQ(code, 0);
 
   qwtTestMaxExecTaskUsec = 0;

source/libs/scheduler/src/scheduler.c

@@ -1468,13 +1468,14 @@ int32_t schedulerConvertDagToTaskList(SQueryDag* pDag, SArray **pTasks) {
     }
 
     int32_t msgSize = sizeof(SSubQueryMsg) + msgLen;
-    msg = calloc(1, msgSize);
     if (NULL == msg) {
       qError("calloc %d failed", msgSize);
       SCH_ERR_JRET(TSDB_CODE_QRY_OUT_OF_MEMORY);
     }
     
-    SSubQueryMsg *pMsg = (SSubQueryMsg*) msg;
+    SSubQueryMsg* pMsg = calloc(1, msgSize);
+    /*SSubQueryMsg *pMsg = (SSubQueryMsg*) msg;*/
+    memcpy(pMsg->msg, msg, msgLen);
     
     pMsg->header.vgId = tInfo.addr.nodeId;
     
@@ -1483,7 +1484,7 @@ int32_t schedulerConvertDagToTaskList(SQueryDag* pDag, SArray **pTasks) {
     pMsg->taskId = schGenUUID();
     pMsg->taskType = TASK_TYPE_PERSISTENT;
     pMsg->contentLen = msgLen;
-    memcpy(pMsg->msg, msg, msgLen);
+    /*memcpy(pMsg->msg, ((SSubQueryMsg*)msg)->msg, msgLen);*/
 
     tInfo.msg = pMsg;
 

source/libs/transport/inc/transComm.h

@@ -202,6 +202,8 @@ bool transDecompressMsg(char* msg, int32_t len, int32_t* flen);
 void transConnCtxDestroy(STransConnCtx* ctx);
 
 void transFreeMsg(void* msg);
+
+//
 typedef struct SConnBuffer {
   char* buf;
   int   len;
@@ -209,4 +211,9 @@ typedef struct SConnBuffer {
   int   left;
 } SConnBuffer;
 
+int transInitBuffer(SConnBuffer* buf);
+int transClearBuffer(SConnBuffer* buf);
+int transDestroyBuffer(SConnBuffer* buf);
+int transAllocBuffer(SConnBuffer* connBuf, uv_buf_t* uvBuf);
+
 #endif

source/libs/transport/src/trans.c

@@ -30,7 +30,8 @@ void* rpcOpen(const SRpcInit* pInit) {
     tstrncpy(pRpc->label, pInit->label, strlen(pInit->label));
   }
   pRpc->cfp = pInit->cfp;
-  pRpc->numOfThreads = pInit->numOfThreads > TSDB_MAX_RPC_THREADS ? TSDB_MAX_RPC_THREADS : pInit->numOfThreads;
+  // pRpc->numOfThreads = pInit->numOfThreads > TSDB_MAX_RPC_THREADS ? TSDB_MAX_RPC_THREADS : pInit->numOfThreads;
+  pRpc->numOfThreads = pInit->numOfThreads;
   pRpc->connType = pInit->connType;
   pRpc->idleTime = pInit->idleTime;
   pRpc->tcphandle = (*taosInitHandle[pRpc->connType])(0, pInit->localPort, pRpc->label, pRpc->numOfThreads, NULL, pRpc);
@@ -55,7 +56,13 @@ void* rpcMallocCont(int contLen) {
   }
   return start + sizeof(STransMsgHead);
 }
-void  rpcFreeCont(void* cont) { return; }
+void rpcFreeCont(void* cont) {
+  // impl
+  if (cont == NULL) {
+    return;
+  }
+  free((char*)cont - TRANS_MSG_OVERHEAD);
+}
 void* rpcReallocCont(void* ptr, int contLen) { return NULL; }
 
 void rpcSendRedirectRsp(void* pConn, const SEpSet* pEpSet) {}

source/libs/transport/src/transCli.c

@@ -31,6 +31,7 @@ typedef struct SCliConn {
   char         secured;
   uint64_t     expireTime;
   int8_t       notifyCount;  // timers already notify to client
+  int32_t      ref;
 } SCliConn;
 
 typedef struct SCliMsg {
@@ -94,8 +95,10 @@ static void clientHandleExcept(SCliConn* conn);
 // handle req from app
 static void clientHandleReq(SCliMsg* pMsg, SCliThrdObj* pThrd);
 
-static void clientMsgDestroy(SCliMsg* pMsg);
-static void destroyTransConnCtx(STransConnCtx* ctx);
+static void destroyUserdata(SRpcMsg* userdata);
+
+static void destroyCmsg(SCliMsg* cmsg);
+static void transDestroyConnCtx(STransConnCtx* ctx);
 // thread obj
 static SCliThrdObj* createThrdObj();
 static void         destroyThrdObj(SCliThrdObj* pThrd);
@@ -103,7 +106,8 @@ static void         destroyThrdObj(SCliThrdObj* pThrd);
 static void* clientThread(void* arg);
 
 static void clientHandleResp(SCliConn* conn) {
-  STransConnCtx* pCtx = ((SCliMsg*)conn->data)->ctx;
+  SCliMsg*       pMsg = conn->data;
+  STransConnCtx* pCtx = pMsg->ctx;
   SRpcInfo*      pRpc = pCtx->pTransInst;
 
   STransMsgHead* pHead = (STransMsgHead*)(conn->readBuf.buf);
@@ -112,41 +116,53 @@ static void clientHandleResp(SCliConn* conn) {
 
   SRpcMsg rpcMsg;
   rpcMsg.contLen = transContLenFromMsg(pHead->msgLen);
-  rpcMsg.pCont = transContFromHead(pHead);
+  rpcMsg.pCont = transContFromHead((char*)pHead);
   rpcMsg.code = pHead->code;
   rpcMsg.msgType = pHead->msgType;
   rpcMsg.ahandle = pCtx->ahandle;
 
+  tDebug("conn %p handle resp", conn);
   (pRpc->cfp)(NULL, &rpcMsg, NULL);
   conn->notifyCount += 1;
 
+  // buf's mem alread translated to rpcMsg.pCont
+  transClearBuffer(&conn->readBuf);
+
+  uv_read_start((uv_stream_t*)conn->stream, clientAllocBufferCb, clientReadCb);
+
   SCliThrdObj* pThrd = conn->hostThrd;
-  tfree(conn->data);
   addConnToPool(pThrd->pool, pCtx->ip, pCtx->port, conn);
 
+  destroyCmsg(pMsg);
+  conn->data = NULL;
   // start thread's timer of conn pool if not active
   if (!uv_is_active((uv_handle_t*)pThrd->pTimer) && pRpc->idleTime > 0) {
     uv_timer_start((uv_timer_t*)pThrd->pTimer, clientTimeoutCb, CONN_PERSIST_TIME(pRpc->idleTime) / 2, 0);
   }
-  destroyTransConnCtx(pCtx);
 }
 static void clientHandleExcept(SCliConn* pConn) {
+  if (pConn->data == NULL) {
+    // handle conn except in conn pool
+    clientConnDestroy(pConn, true);
+    return;
+  }
+  tDebug("conn %p start to destroy", pConn);
   SCliMsg* pMsg = pConn->data;
 
-  STransConnCtx* pCtx = pMsg->ctx;
-  SRpcInfo*      pRpc = pCtx->pTransInst;
+  destroyUserdata(&pMsg->msg);
 
-  transFreeMsg((pMsg->msg.pCont));
-  pMsg->msg.pCont = NULL;
+  STransConnCtx* pCtx = pMsg->ctx;
 
   SRpcMsg rpcMsg = {0};
   rpcMsg.ahandle = pCtx->ahandle;
   rpcMsg.code = -1;
   // SRpcInfo* pRpc = pMsg->ctx->pRpc;
-  (pRpc->cfp)(NULL, &rpcMsg, NULL);
-  tfree(pConn->data);
+  (pCtx->pTransInst->cfp)(NULL, &rpcMsg, NULL);
   pConn->notifyCount += 1;
-  destroyTransConnCtx(pCtx);
+
+  destroyCmsg(pMsg);
+  pConn->data = NULL;
+  // transDestroyConnCtx(pCtx);
   clientConnDestroy(pConn, true);
 }
 
@@ -213,14 +229,20 @@ static SCliConn* getConnFromPool(void* pool, char* ip, uint32_t port) {
   }
   queue* h = QUEUE_HEAD(&plist->conn);
   QUEUE_REMOVE(h);
-  return QUEUE_DATA(h, SCliConn, conn);
+
+  SCliConn* conn = QUEUE_DATA(h, SCliConn, conn);
+  QUEUE_INIT(&conn->conn);
+  return conn;
 }
 static void addConnToPool(void* pool, char* ip, uint32_t port, SCliConn* conn) {
   char key[128] = {0};
+
   tstrncpy(key, ip, strlen(ip));
   tstrncpy(key + strlen(key), (char*)(&port), sizeof(port));
+  tDebug("conn %p added to conn pool, read buf cap: %d", conn, conn->readBuf.cap);
 
   SRpcInfo* pRpc = ((SCliThrdObj*)conn->hostThrd)->pTransInst;
+
   conn->expireTime = taosGetTimestampMs() + CONN_PERSIST_TIME(pRpc->idleTime);
   SConnList* plist = taosHashGet((SHashObj*)pool, key, strlen(key));
   conn->notifyCount = 0;
@@ -237,40 +259,18 @@ static bool clientReadComplete(SConnBuffer* data) {
     if (msgLen > data->len) {
       data->left = msgLen - data->len;
       return false;
-    } else {
+    } else if (msgLen == data->len) {
+      data->left = 0;
       return true;
     }
   } else {
     return false;
   }
 }
-static void clientAllocReadBufferCb(uv_handle_t* handle, size_t suggested_size, uv_buf_t* buf) {
-  // impl later
-  static const int CAPACITY = 512;
-
+static void clientAllocBufferCb(uv_handle_t* handle, size_t suggested_size, uv_buf_t* buf) {
   SCliConn*    conn = handle->data;
   SConnBuffer* pBuf = &conn->readBuf;
-  if (pBuf->cap == 0) {
-    pBuf->buf = (char*)calloc(1, CAPACITY * sizeof(char));
-    pBuf->len = 0;
-    pBuf->cap = CAPACITY;
-    pBuf->left = -1;
-
-    buf->base = pBuf->buf;
-    buf->len = CAPACITY;
-  } else {
-    if (pBuf->len >= pBuf->cap) {
-      if (pBuf->left == -1) {
-        pBuf->cap *= 2;
-        pBuf->buf = realloc(pBuf->buf, pBuf->cap);
-      } else if (pBuf->len + pBuf->left > pBuf->cap) {
-        pBuf->cap = pBuf->len + pBuf->left;
-        pBuf->buf = realloc(pBuf->buf, pBuf->cap);
-      }
-    }
-    buf->base = pBuf->buf + pBuf->len;
-    buf->len = pBuf->cap - pBuf->len;
-  }
+  transAllocBuffer(pBuf, buf);
 }
 static void clientReadCb(uv_stream_t* handle, ssize_t nread, const uv_buf_t* buf) {
   // impl later
@@ -279,6 +279,7 @@ static void clientReadCb(uv_stream_t* handle, ssize_t nread, const uv_buf_t* buf
   if (nread > 0) {
     pBuf->len += nread;
     if (clientReadComplete(pBuf)) {
+      uv_read_stop((uv_stream_t*)conn->stream);
       tDebug("conn %p read complete", conn);
       clientHandleResp(conn);
     } else {
@@ -288,10 +289,12 @@ static void clientReadCb(uv_stream_t* handle, ssize_t nread, const uv_buf_t* buf
   }
   assert(nread <= 0);
   if (nread == 0) {
-    tError("conn %p closed", conn);
+    // ref http://docs.libuv.org/en/v1.x/stream.html?highlight=uv_read_start#c.uv_read_cb
+    // nread might be 0, which does not indicate an error or EOF. This is equivalent to EAGAIN or EWOULDBLOCK under
+    // read(2).
     return;
   }
-  if (nread < 0) {
+  if (nread < 0 || nread == UV_EOF) {
     tError("conn %p read error: %s", conn, uv_err_name(nread));
     clientHandleExcept(conn);
   }
@@ -300,43 +303,47 @@ static void clientReadCb(uv_stream_t* handle, ssize_t nread, const uv_buf_t* buf
 }
 
 static void clientConnDestroy(SCliConn* conn, bool clear) {
-  tDebug("conn %p destroy", conn);
-  if (clear) {
-    uv_close((uv_handle_t*)conn->stream, NULL);
+  //
+  conn->ref--;
+  if (conn->ref == 0) {
+    tDebug("conn %p remove from conn pool", conn);
+    QUEUE_REMOVE(&conn->conn);
+    tDebug("conn %p remove from conn pool successfully", conn);
+    if (clear) {
+      uv_close((uv_handle_t*)conn->stream, clientDestroy);
+    }
   }
-  free(conn->stream);
-  free(conn->readBuf.buf);
-  free(conn->writeReq);
-  free(conn);
 }
 static void clientDestroy(uv_handle_t* handle) {
   SCliConn* conn = handle->data;
-  // QUEUE_REMOVE(&conn->conn);
-  clientConnDestroy(conn, false);
+  // transDestroyBuffer(&conn->readBuf);
+
+  free(conn->stream);
+  free(conn->writeReq);
+  tDebug("conn %p destroy successfully", conn);
+  free(conn);
+
+  // clientConnDestroy(conn, false);
 }
 
 static void clientWriteCb(uv_write_t* req, int status) {
   SCliConn* pConn = req->data;
-
-  SCliMsg* pMsg = pConn->data;
-  transFreeMsg((pMsg->msg.pCont));
-  pMsg->msg.pCont = NULL;
-
   if (status == 0) {
     tDebug("conn %p data already was written out", pConn);
+    SCliMsg* pMsg = pConn->data;
+    if (pMsg == NULL) {
+      destroy
+          // handle
+          return;
+    }
+    destroyUserdata(&pMsg->msg);
   } else {
     tError("conn %p failed to write: %s", pConn, uv_err_name(status));
     clientHandleExcept(pConn);
     return;
   }
   SCliThrdObj* pThrd = pConn->hostThrd;
-  // if (pConn->stream == NULL) {
-  //  pConn->stream = (uv_stream_t*)malloc(sizeof(uv_tcp_t));
-  //  uv_tcp_init(pThrd->loop, (uv_tcp_t*)pConn->stream);
-  //  pConn->stream->data = pConn;
-  //}
-  uv_read_start((uv_stream_t*)pConn->stream, clientAllocReadBufferCb, clientReadCb);
-  // impl later
+  uv_read_start((uv_stream_t*)pConn->stream, clientAllocBufferCb, clientReadCb);
 }
 
 static void clientWrite(SCliConn* pConn) {
@@ -381,14 +388,11 @@ static void clientHandleReq(SCliMsg* pMsg, SCliThrdObj* pThrd) {
     tDebug("conn %p get from conn pool", conn);
     conn->data = pMsg;
     conn->writeReq->data = conn;
-
-    conn->readBuf.len = 0;
-    memset(conn->readBuf.buf, 0, conn->readBuf.cap);
-    conn->readBuf.left = -1;
+    transDestroyBuffer(&conn->readBuf);
     clientWrite(conn);
   } else {
     SCliConn* conn = calloc(1, sizeof(SCliConn));
-
+    conn->ref++;
     // read/write stream handle
     conn->stream = (uv_stream_t*)malloc(sizeof(uv_tcp_t));
     uv_tcp_init(pThrd->loop, (uv_tcp_t*)(conn->stream));
@@ -397,6 +401,7 @@ static void clientHandleReq(SCliMsg* pMsg, SCliThrdObj* pThrd) {
     // write req handle
     conn->writeReq = malloc(sizeof(uv_write_t));
     conn->writeReq->data = conn;
+
     QUEUE_INIT(&conn->conn);
 
     conn->connReq.data = conn;
@@ -459,8 +464,20 @@ void* taosInitClient(uint32_t ip, uint32_t port, char* label, int numOfThreads,
   }
   return cli;
 }
-static void clientMsgDestroy(SCliMsg* pMsg) {
-  // impl later
+
+static void destroyUserdata(SRpcMsg* userdata) {
+  if (userdata->pCont == NULL) {
+    return;
+  }
+  transFreeMsg(userdata->pCont);
+  userdata->pCont = NULL;
+}
+static void destroyCmsg(SCliMsg* pMsg) {
+  if (pMsg == NULL) {
+    return;
+  }
+  transDestroyConnCtx(pMsg->ctx);
+  destroyUserdata(&pMsg->msg);
   free(pMsg);
 }
 static SCliThrdObj* createThrdObj() {
@@ -493,7 +510,7 @@ static void destroyThrdObj(SCliThrdObj* pThrd) {
   free(pThrd);
 }
 
-static void destroyTransConnCtx(STransConnCtx* ctx) {
+static void transDestroyConnCtx(STransConnCtx* ctx) {
   if (ctx != NULL) {
     free(ctx->ip);
   }

source/libs/transport/src/transComm.c

@@ -198,4 +198,51 @@ void transFreeMsg(void* msg) {
   }
   free((char*)msg - sizeof(STransMsgHead));
 }
+
+int transInitBuffer(SConnBuffer* buf) {
+  transClearBuffer(buf);
+  return 0;
+}
+int transClearBuffer(SConnBuffer* buf) {
+  memset(buf, 0, sizeof(*buf));
+  return 0;
+}
+int transAllocBuffer(SConnBuffer* connBuf, uv_buf_t* uvBuf) {
+  /*
+   * formate of data buffer:
+   * |<--------------------------data from socket------------------------------->|
+   * |<------STransMsgHead------->|<-------------------other data--------------->|
+   */
+  static const int CAPACITY = 1024;
+
+  SConnBuffer* p = connBuf;
+  if (p->cap == 0) {
+    p->buf = (char*)calloc(CAPACITY, sizeof(char));
+    p->len = 0;
+    p->cap = CAPACITY;
+    p->left = -1;
+
+    uvBuf->base = p->buf;
+    uvBuf->len = CAPACITY;
+  } else {
+    if (p->len >= p->cap) {
+      if (p->left == -1) {
+        p->cap *= 2;
+        p->buf = realloc(p->buf, p->cap);
+      } else if (p->len + p->left > p->cap) {
+        p->cap = p->len + p->left;
+        p->buf = realloc(p->buf, p->len + p->left);
+      }
+    }
+    uvBuf->base = p->buf + p->len;
+    uvBuf->len = p->cap - p->len;
+  }
+  return 0;
+}
+int transDestroyBuffer(SConnBuffer* buf) {
+  if (buf->cap > 0) {
+    tfree(buf->buf);
+  }
+  transClearBuffer(buf);
+}
 #endif

source/libs/transport/src/transSrv.c

@@ -17,7 +17,7 @@
 
 #include "transComm.h"
 
-typedef struct SConn {
+typedef struct SSrvConn {
   uv_tcp_t*   pTcp;
   uv_write_t* pWriter;
   uv_timer_t* pTimer;
@@ -26,13 +26,14 @@ typedef struct SConn {
   queue       queue;
   int         ref;
   int         persist;  // persist connection or not
-  SConnBuffer connBuf;  // read buf,
+  SConnBuffer readBuf;  // read buf,
   int         inType;
   void*       pTransInst;  // rpc init
   void*       ahandle;     //
   void*       hostThrd;
+  void*       pSrvMsg;
 
-  SRpcMsg sendMsg;
+  // SRpcMsg sendMsg;
   // del later
   char secured;
   int  spi;
@@ -40,7 +41,13 @@ typedef struct SConn {
   char user[TSDB_UNI_LEN];  // user ID for the link
   char secret[TSDB_PASSWORD_LEN];
   char ckey[TSDB_PASSWORD_LEN];  // ciphering key
-} SConn;
+} SSrvConn;
+
+typedef struct SSrvMsg {
+  SSrvConn* pConn;
+  SRpcMsg   msg;
+  queue     q;
+} SSrvMsg;
 
 typedef struct SWorkThrdObj {
   pthread_t       thread;
@@ -48,8 +55,8 @@ typedef struct SWorkThrdObj {
   int             fd;
   uv_loop_t*      loop;
   uv_async_t*     workerAsync;  //
-  queue           conn;
-  pthread_mutex_t connMtx;
+  queue           msg;
+  pthread_mutex_t msgMtx;
   void*           pTransInst;
 } SWorkThrdObj;
 
@@ -68,9 +75,9 @@ typedef struct SServerObj {
 static const char* notify = "a";
 
 // refactor later
-static int transAddAuthPart(SConn* pConn, char* msg, int msgLen);
+static int transAddAuthPart(SSrvConn* pConn, char* msg, int msgLen);
 
-static int uvAuthMsg(SConn* pConn, char* msg, int msgLen);
+static int uvAuthMsg(SSrvConn* pConn, char* msg, int msgLen);
 
 static void uvAllocConnBufferCb(uv_handle_t* handle, size_t suggested_size, uv_buf_t* buf);
 static void uvAllocReadBufferCb(uv_handle_t* handle, size_t suggested_size, uv_buf_t* buf);
@@ -82,12 +89,13 @@ static void uvOnAcceptCb(uv_stream_t* stream, int status);
 static void uvOnConnectionCb(uv_stream_t* q, ssize_t nread, const uv_buf_t* buf);
 static void uvWorkerAsyncCb(uv_async_t* handle);
 
-static void uvPrepareSendData(SConn* conn, uv_buf_t* wb);
-
+static void uvPrepareSendData(SSrvMsg* msg, uv_buf_t* wb);
+static void uvStartSendResp(SSrvMsg* msg);
+static void destroySmsg(SSrvMsg* smsg);
 // check whether already read complete packet
-static bool   readComplete(SConnBuffer* buf);
-static SConn* createConn();
-static void   destroyConn(SConn* conn, bool clear /*clear handle or not*/);
+static bool      readComplete(SConnBuffer* buf);
+static SSrvConn* createConn();
+static void      destroyConn(SSrvConn* conn, bool clear /*clear handle or not*/);
 
 static void uvDestroyConn(uv_handle_t* handle);
 
@@ -105,31 +113,9 @@ void uvAllocReadBufferCb(uv_handle_t* handle, size_t suggested_size, uv_buf_t* b
    * |<--------------------------data from socket------------------------------->|
    * |<------STransMsgHead------->|<-------------------other data--------------->|
    */
-  static const int CAPACITY = 1024;
-
-  SConn*       conn = handle->data;
-  SConnBuffer* pBuf = &conn->connBuf;
-  if (pBuf->cap == 0) {
-    pBuf->buf = (char*)calloc(CAPACITY, sizeof(char));
-    pBuf->len = 0;
-    pBuf->cap = CAPACITY;
-    pBuf->left = -1;
-
-    buf->base = pBuf->buf;
-    buf->len = CAPACITY;
-  } else {
-    if (pBuf->len >= pBuf->cap) {
-      if (pBuf->left == -1) {
-        pBuf->cap *= 2;
-        pBuf->buf = realloc(pBuf->buf, pBuf->cap);
-      } else if (pBuf->len + pBuf->left > pBuf->cap) {
-        pBuf->cap = pBuf->len + pBuf->left;
-        pBuf->buf = realloc(pBuf->buf, pBuf->len + pBuf->left);
-      }
-    }
-    buf->base = pBuf->buf + pBuf->len;
-    buf->len = pBuf->cap - pBuf->len;
-  }
+  SSrvConn*    conn = handle->data;
+  SConnBuffer* pBuf = &conn->readBuf;
+  transAllocBuffer(pBuf, buf);
 }
 
 // check data read from socket completely or not
@@ -159,7 +145,7 @@ static bool readComplete(SConnBuffer* data) {
 //  // impl later
 //  STransMsgHead* pHead = (STransMsgHead*)pRecv->msg;
 //  SRpcInfo*      pRpc = (SRpcInfo*)pRecv->shandle;
-//  SConn*         pConn = pRecv->thandle;
+//  SSrvConn*         pConn = pRecv->thandle;
 //  tDump(pRecv->msg, pRecv->msgLen);
 //  terrno = 0;
 //  // SRpcReqContext* pContest;
@@ -167,7 +153,7 @@ static bool readComplete(SConnBuffer* data) {
 //  // do auth and check
 //}
 
-static int uvAuthMsg(SConn* pConn, char* msg, int len) {
+static int uvAuthMsg(SSrvConn* pConn, char* msg, int len) {
   STransMsgHead* pHead = (STransMsgHead*)msg;
 
   int code = 0;
@@ -222,14 +208,14 @@ static int uvAuthMsg(SConn* pConn, char* msg, int len) {
 
 // refers specifically to query or insert timeout
 static void uvHandleActivityTimeout(uv_timer_t* handle) {
-  SConn* conn = handle->data;
+  SSrvConn* conn = handle->data;
   tDebug("%p timeout since no activity", conn);
 }
 
-static void uvHandleReq(SConn* pConn) {
+static void uvHandleReq(SSrvConn* pConn) {
   SRecvInfo    info;
   SRecvInfo*   p = &info;
-  SConnBuffer* pBuf = &pConn->connBuf;
+  SConnBuffer* pBuf = &pConn->readBuf;
   p->msg = pBuf->buf;
   p->msgLen = pBuf->len;
   p->ip = 0;
@@ -255,7 +241,6 @@ static void uvHandleReq(SConn* pConn) {
   pHead->code = htonl(pHead->code);
 
   int32_t dlen = 0;
-  SRpcMsg rpcMsg;
   if (transDecompressMsg(NULL, 0, NULL)) {
     // add compress later
     // pHead = rpcDecompressRpcMsg(pHead);
@@ -264,6 +249,8 @@ static void uvHandleReq(SConn* pConn) {
     // impl later
     //
   }
+
+  SRpcMsg rpcMsg;
   rpcMsg.contLen = transContLenFromMsg(pHead->msgLen);
   rpcMsg.pCont = pHead->content;
   rpcMsg.msgType = pHead->msgType;
@@ -271,6 +258,7 @@ static void uvHandleReq(SConn* pConn) {
   rpcMsg.ahandle = NULL;
   rpcMsg.handle = pConn;
 
+  transClearBuffer(&pConn->readBuf);
   pConn->ref++;
   (*(pRpc->cfp))(pRpc->parent, &rpcMsg, NULL);
   // uv_timer_start(pConn->pTimer, uvHandleActivityTimeout, pRpc->idleTime * 10000, 0);
@@ -280,8 +268,8 @@ static void uvHandleReq(SConn* pConn) {
 
 void uvOnReadCb(uv_stream_t* cli, ssize_t nread, const uv_buf_t* buf) {
   // opt
-  SConn*       conn = cli->data;
-  SConnBuffer* pBuf = &conn->connBuf;
+  SSrvConn*    conn = cli->data;
+  SConnBuffer* pBuf = &conn->readBuf;
   if (nread > 0) {
     pBuf->len += nread;
     tDebug("conn %p read summroy, total read: %d, current read: %d", conn, pBuf->len, (int)nread);
@@ -294,11 +282,12 @@ void uvOnReadCb(uv_stream_t* cli, ssize_t nread, const uv_buf_t* buf) {
     return;
   }
   if (nread == 0) {
-    tDebug("conn %p except read", conn);
-    // destroyConn(conn, true);
     return;
   }
-  if (nread != UV_EOF) {
+  if (nread < 0 || nread != UV_EOF) {
+    if (conn->ref > 1) {
+      conn->ref++;  // ref > 1 signed that write is in progress
+    }
     tDebug("conn %p read error: %s", conn, uv_err_name(nread));
     destroyConn(conn, true);
   }
@@ -310,25 +299,23 @@ void uvAllocConnBufferCb(uv_handle_t* handle, size_t suggested_size, uv_buf_t* b
 
 void uvOnTimeoutCb(uv_timer_t* handle) {
   // opt
-  SConn* pConn = handle->data;
+  SSrvConn* pConn = handle->data;
   tDebug("conn %p time out", pConn);
 }
 
 void uvOnWriteCb(uv_write_t* req, int status) {
-  SConn* conn = req->data;
+  SSrvConn* conn = req->data;
 
-  SConnBuffer* buf = &conn->connBuf;
-  buf->len = 0;
-  memset(buf->buf, 0, buf->cap);
-  buf->left = -1;
-
-  SRpcMsg* pMsg = &conn->sendMsg;
-  transFreeMsg(pMsg->pCont);
+  SSrvMsg* smsg = conn->pSrvMsg;
+  destroySmsg(smsg);
+  conn->pSrvMsg = NULL;
 
+  transClearBuffer(&conn->readBuf);
   if (status == 0) {
     tDebug("conn %p data already was written on stream", conn);
   } else {
     tDebug("conn %p failed to write data, %s", conn, uv_err_name(status));
+    //
     destroyConn(conn, true);
   }
   // opt
@@ -341,16 +328,16 @@ static void uvOnPipeWriteCb(uv_write_t* req, int status) {
   }
 }
 
-static void uvPrepareSendData(SConn* conn, uv_buf_t* wb) {
+static void uvPrepareSendData(SSrvMsg* smsg, uv_buf_t* wb) {
   // impl later;
-  tDebug("conn %p prepare to send resp", conn);
-  SRpcMsg* pMsg = &conn->sendMsg;
+  tDebug("conn %p prepare to send resp", smsg->pConn);
+  SRpcMsg* pMsg = &smsg->msg;
   if (pMsg->pCont == 0) {
     pMsg->pCont = (void*)rpcMallocCont(0);
     pMsg->contLen = 0;
   }
   STransMsgHead* pHead = transHeadFromCont(pMsg->pCont);
-  pHead->msgType = conn->inType + 1;
+  pHead->msgType = smsg->pConn->inType + 1;
   // add more info
   char*   msg = (char*)pHead;
   int32_t len = transMsgLenFromCont(pMsg->contLen);
@@ -361,28 +348,53 @@ static void uvPrepareSendData(SConn* conn, uv_buf_t* wb) {
   wb->base = msg;
   wb->len = len;
 }
+static void uvStartSendResp(SSrvMsg* smsg) {
+  // impl
+  uv_buf_t wb;
+  uvPrepareSendData(smsg, &wb);
+
+  SSrvConn* pConn = smsg->pConn;
+  uv_timer_stop(pConn->pTimer);
+
+  pConn->pSrvMsg = smsg;
+  // conn->pWriter->data = smsg;
+  uv_write(pConn->pWriter, (uv_stream_t*)pConn->pTcp, &wb, 1, uvOnWriteCb);
+
+  // SRpcMsg* rpcMsg = smsg->msg;
+
+  return;
+}
+static void destroySmsg(SSrvMsg* smsg) {
+  if (smsg == NULL) {
+    return;
+  }
+  transFreeMsg(smsg->msg.pCont);
+  free(smsg);
+}
 void uvWorkerAsyncCb(uv_async_t* handle) {
   SWorkThrdObj* pThrd = handle->data;
-  SConn*        conn = NULL;
+  SSrvConn*     conn = NULL;
   queue         wq;
   // batch process to avoid to lock/unlock frequently
-  pthread_mutex_lock(&pThrd->connMtx);
-  QUEUE_MOVE(&pThrd->conn, &wq);
-  pthread_mutex_unlock(&pThrd->connMtx);
+  pthread_mutex_lock(&pThrd->msgMtx);
+  QUEUE_MOVE(&pThrd->msg, &wq);
+  pthread_mutex_unlock(&pThrd->msgMtx);
 
   while (!QUEUE_IS_EMPTY(&wq)) {
     queue* head = QUEUE_HEAD(&wq);
     QUEUE_REMOVE(head);
-    SConn* conn = QUEUE_DATA(head, SConn, queue);
-    if (conn == NULL) {
-      tError("except occurred, do nothing");
-      return;
+
+    SSrvMsg* msg = QUEUE_DATA(head, SSrvMsg, q);
+    if (msg == NULL) {
+      tError("except occurred, continue");
+      continue;
     }
-    uv_buf_t wb;
-    uvPrepareSendData(conn, &wb);
-    uv_timer_stop(conn->pTimer);
+    uvStartSendResp(msg);
+    // uv_buf_t wb;
+    // uvPrepareSendData(msg, &wb);
+    // uv_timer_stop(conn->pTimer);
 
-    uv_write(conn->pWriter, (uv_stream_t*)conn->pTcp, &wb, 1, uvOnWriteCb);
+    // uv_write(conn->pWriter, (uv_stream_t*)conn->pTcp, &wb, 1, uvOnWriteCb);
   }
 }
 
@@ -435,7 +447,7 @@ void uvOnConnectionCb(uv_stream_t* q, ssize_t nread, const uv_buf_t* buf) {
   uv_handle_type pending = uv_pipe_pending_type(pipe);
   assert(pending == UV_TCP);
 
-  SConn* pConn = createConn();
+  SSrvConn* pConn = createConn();
 
   pConn->pTransInst = pThrd->pTransInst;
   /* init conn timer*/
@@ -484,8 +496,8 @@ static bool addHandleToWorkloop(void* arg) {
 
   pThrd->pipe->data = pThrd;
 
-  QUEUE_INIT(&pThrd->conn);
-  pthread_mutex_init(&pThrd->connMtx, NULL);
+  QUEUE_INIT(&pThrd->msg);
+  pthread_mutex_init(&pThrd->msgMtx, NULL);
 
   pThrd->workerAsync = malloc(sizeof(uv_async_t));
   uv_async_init(pThrd->loop, pThrd->workerAsync, uvWorkerAsyncCb);
@@ -523,34 +535,43 @@ void* workerThread(void* arg) {
   uv_run(pThrd->loop, UV_RUN_DEFAULT);
 }
 
-static SConn* createConn() {
-  SConn* pConn = (SConn*)calloc(1, sizeof(SConn));
+static SSrvConn* createConn() {
+  SSrvConn* pConn = (SSrvConn*)calloc(1, sizeof(SSrvConn));
+  tDebug("conn %p created", pConn);
   ++pConn->ref;
   return pConn;
 }
 
-static void destroyConn(SConn* conn, bool clear) {
+static void destroyConn(SSrvConn* conn, bool clear) {
   if (conn == NULL) {
     return;
   }
-  if (--conn->ref == 0) {
+  // SRpcMsg* pMsg = &conn->sendMsg;
+  // transFreeMsg(pMsg->pCont);
+  // pMsg->pCont = NULL;
+
+  tDebug("conn %p try to destroy", conn);
+  if (--conn->ref > 0) {
     return;
   }
+  transDestroyBuffer(&conn->readBuf);
+  destroySmsg(conn->pSrvMsg);
+  conn->pSrvMsg = NULL;
+
   if (clear) {
-    uv_close((uv_handle_t*)conn->pTcp, NULL);
+    uv_close((uv_handle_t*)conn->pTcp, uvDestroyConn);
   }
+}
+static void uvDestroyConn(uv_handle_t* handle) {
+  SSrvConn* conn = handle->data;
+  tDebug("conn %p destroy", conn);
   uv_timer_stop(conn->pTimer);
   free(conn->pTimer);
-  free(conn->pTcp);
-  free(conn->connBuf.buf);
+  // free(conn->pTcp);
   free(conn->pWriter);
   free(conn);
 }
-static void uvDestroyConn(uv_handle_t* handle) {
-  SConn* conn = handle->data;
-  destroyConn(conn, false);
-}
-static int transAddAuthPart(SConn* pConn, char* msg, int msgLen) {
+static int transAddAuthPart(SSrvConn* pConn, char* msg, int msgLen) {
   STransMsgHead* pHead = (STransMsgHead*)msg;
 
   if (pConn->spi && pConn->secured == 0) {
@@ -632,6 +653,7 @@ void destroyWorkThrd(SWorkThrdObj* pThrd) {
   pthread_join(pThrd->thread, NULL);
   // free(srv->pipe[i]);
   free(pThrd->loop);
+  pthread_mutex_destroy(&pThrd->msgMtx);
   free(pThrd);
 }
 void taosCloseServer(void* arg) {
@@ -648,17 +670,20 @@ void taosCloseServer(void* arg) {
 }
 
 void rpcSendResponse(const SRpcMsg* pMsg) {
-  SConn*        pConn = pMsg->handle;
+  SSrvConn*     pConn = pMsg->handle;
   SWorkThrdObj* pThrd = pConn->hostThrd;
 
-  // opt later
-  pConn->sendMsg = *pMsg;
-  pthread_mutex_lock(&pThrd->connMtx);
-  QUEUE_PUSH(&pThrd->conn, &pConn->queue);
-  pthread_mutex_unlock(&pThrd->connMtx);
+  SSrvMsg* srvMsg = calloc(1, sizeof(SSrvMsg));
+  srvMsg->pConn = pConn;
+  srvMsg->msg = *pMsg;
+
+  pthread_mutex_lock(&pThrd->msgMtx);
+  QUEUE_PUSH(&pThrd->msg, &srvMsg->q);
+  pthread_mutex_unlock(&pThrd->msgMtx);
+
   tDebug("conn %p start to send resp", pConn);
 
-  uv_async_send(pConn->pWorkerAsync);
+  uv_async_send(pThrd->workerAsync);
 }
 
 #endif

source/libs/transport/test/rclient.c

@@ -65,6 +65,7 @@ static void *sendRequest(void *param) {
     // tsem_wait(&pInfo->rspSem);
     tsem_wait(&pInfo->rspSem);
     tDebug("recv response succefully");
+
     // usleep(100000000);
   }
 

source/libs/transport/test/rserver.c

@@ -165,6 +165,7 @@ int main(int argc, char *argv[]) {
     tError("failed to start RPC server");
     return -1;
   }
+  // sleep(5);
 
   tInfo("RPC server is running, ctrl-c to exit");
 
@@ -172,7 +173,6 @@ int main(int argc, char *argv[]) {
     dataFd = open(dataName, O_APPEND | O_CREAT | O_WRONLY, S_IRWXU | S_IRWXG | S_IRWXO);
     if (dataFd < 0) tInfo("failed to open data file, reason:%s", strerror(errno));
   }
-
   qhandle = taosOpenQueue();
   qset = taosOpenQset();
   taosAddIntoQset(qset, qhandle, NULL);

source/libs/wal/CMakeLists.txt

@@ -11,6 +11,7 @@ target_link_libraries(
     PUBLIC cjson
     PUBLIC os
     PUBLIC util
+    PUBLIC common
 )
 
 if(${BUILD_TEST})

source/libs/wal/src/walWrite.c

@@ -257,7 +257,7 @@ static int walWriteIndex(SWal *pWal, int64_t ver, int64_t offset) {
   return 0;
 }
 
-int64_t walWrite(SWal *pWal, int64_t index, uint8_t msgType, const void *body, int32_t bodyLen) {
+int64_t walWrite(SWal *pWal, int64_t index, tmsg_t msgType, const void *body, int32_t bodyLen) {
   if (pWal == NULL) return -1;
   int code = 0;
 

source/util/src/tqueue.c

@@ -13,35 +13,36 @@
  * along with this program. If not, see <http://www.gnu.org/licenses/>.
  */
 
-#include "os.h"
-
-#include "taoserror.h"
+#define _DEFAULT_SOURCE
 #include "tqueue.h"
+#include "taoserror.h"
 #include "ulog.h"
 
 typedef struct STaosQnode STaosQnode;
 
 typedef struct STaosQnode {
   STaosQnode *next;
+  STaosQueue *queue;
   char        item[];
 } STaosQnode;
 
 typedef struct STaosQueue {
   int32_t         itemSize;
   int32_t         numOfItems;
-  STaosQnode     *head;
-  STaosQnode     *tail;
-  STaosQueue     *next;     // for queue set
-  STaosQset      *qset;     // for queue set
-  void           *ahandle;  // for queue set
-  FProcessItem    itemFp;
-  FProcessItems   itemsFp;
+  int32_t         threadId;
+  STaosQnode *    head;
+  STaosQnode *    tail;
+  STaosQueue *    next;     // for queue set
+  STaosQset *     qset;     // for queue set
+  void *          ahandle;  // for queue set
+  FItem           itemFp;
+  FItems          itemsFp;
   pthread_mutex_t mutex;
 } STaosQueue;
 
 typedef struct STaosQset {
-  STaosQueue     *head;
-  STaosQueue     *current;
+  STaosQueue *    head;
+  STaosQueue *    current;
   pthread_mutex_t mutex;
   int32_t         numOfQueues;
   int32_t         numOfItems;
@@ -56,19 +57,23 @@ typedef struct STaosQall {
 } STaosQall;
 
 STaosQueue *taosOpenQueue() {
-  STaosQueue *queue = calloc(sizeof(STaosQueue), 1);
+  STaosQueue *queue = calloc(1, sizeof(STaosQueue));
   if (queue == NULL) {
     terrno = TSDB_CODE_OUT_OF_MEMORY;
     return NULL;
   }
 
-  pthread_mutex_init(&queue->mutex, NULL);
+  if (pthread_mutex_init(&queue->mutex, NULL) != 0) {
+    terrno = TSDB_CODE_OUT_OF_MEMORY;
+    return NULL;
+  }
 
-  uTrace("queue:%p is opened", queue);
+  queue->threadId = -1;
+  uDebug("queue:%p is opened", queue);
   return queue;
 }
 
-void taosSetQueueFp(STaosQueue *queue, FProcessItem itemFp, FProcessItems itemsFp) {
+void taosSetQueueFp(STaosQueue *queue, FItem itemFp, FItems itemsFp) {
   if (queue == NULL) return;
   queue->itemFp = itemFp;
   queue->itemsFp = itemsFp;
@@ -77,7 +82,7 @@ void taosSetQueueFp(STaosQueue *queue, FProcessItem itemFp, FProcessItems itemsF
 void taosCloseQueue(STaosQueue *queue) {
   if (queue == NULL) return;
   STaosQnode *pTemp;
-  STaosQset  *qset;
+  STaosQset * qset;
 
   pthread_mutex_lock(&queue->mutex);
   STaosQnode *pNode = queue->head;
@@ -85,7 +90,9 @@ void taosCloseQueue(STaosQueue *queue) {
   qset = queue->qset;
   pthread_mutex_unlock(&queue->mutex);
 
-  if (queue->qset) taosRemoveFromQset(qset, queue);
+  if (queue->qset) {
+    taosRemoveFromQset(qset, queue);
+  }
 
   while (pNode) {
     pTemp = pNode;
@@ -96,7 +103,7 @@ void taosCloseQueue(STaosQueue *queue) {
   pthread_mutex_destroy(&queue->mutex);
   free(queue);
 
-  uTrace("queue:%p is closed", queue);
+  uDebug("queue:%p is closed", queue);
 }
 
 bool taosQueueEmpty(STaosQueue *queue) {
@@ -120,19 +127,23 @@ int32_t taosQueueSize(STaosQueue *queue) {
 }
 
 void *taosAllocateQitem(int32_t size) {
-  STaosQnode *pNode = (STaosQnode *)calloc(sizeof(STaosQnode) + size, 1);
+  STaosQnode *pNode = calloc(1, sizeof(STaosQnode) + size);
+
+  if (pNode == NULL) {
+    terrno = TSDB_CODE_OUT_OF_MEMORY;
+    return NULL;
+  }
 
-  if (pNode == NULL) return NULL;
   uTrace("item:%p, node:%p is allocated", pNode->item, pNode);
   return (void *)pNode->item;
 }
 
-void taosFreeQitem(void *param) {
-  if (param == NULL) return;
+void taosFreeQitem(void *pItem) {
+  if (pItem == NULL) return;
 
-  char *temp = (char *)param;
+  char *temp = pItem;
   temp -= sizeof(STaosQnode);
-  uTrace("item:%p, node:%p is freed", param, temp);
+  uTrace("item:%p, node:%p is freed", pItem, temp);
   free(temp);
 }
 
@@ -175,7 +186,7 @@ int32_t taosReadQitem(STaosQueue *queue, void **ppItem) {
     queue->numOfItems--;
     if (queue->qset) atomic_sub_fetch_32(&queue->qset->numOfItems, 1);
     code = 1;
-    uDebug("item:%p is read out from queue:%p, items:%d", *ppItem, queue, queue->numOfItems);
+    uTrace("item:%p is read out from queue:%p, items:%d", *ppItem, queue, queue->numOfItems);
   }
 
   pthread_mutex_unlock(&queue->mutex);
@@ -183,7 +194,7 @@ int32_t taosReadQitem(STaosQueue *queue, void **ppItem) {
   return code;
 }
 
-STaosQall *taosAllocateQall() { return calloc(sizeof(STaosQall), 1); }
+STaosQall *taosAllocateQall() { return calloc(1, sizeof(STaosQall)); }
 
 void taosFreeQall(STaosQall *qall) { free(qall); }
 
@@ -238,7 +249,7 @@ int32_t taosGetQitem(STaosQall *qall, void **ppItem) {
 void taosResetQitems(STaosQall *qall) { qall->current = qall->start; }
 
 STaosQset *taosOpenQset() {
-  STaosQset *qset = (STaosQset *)calloc(sizeof(STaosQset), 1);
+  STaosQset *qset = calloc(sizeof(STaosQset), 1);
   if (qset == NULL) {
     terrno = TSDB_CODE_OUT_OF_MEMORY;
     return NULL;
@@ -247,7 +258,7 @@ STaosQset *taosOpenQset() {
   pthread_mutex_init(&qset->mutex, NULL);
   tsem_init(&qset->sem, 0, 0);
 
-  uTrace("qset:%p is opened", qset);
+  uDebug("qset:%p is opened", qset);
   return qset;
 }
 
@@ -268,7 +279,7 @@ void taosCloseQset(STaosQset *qset) {
   pthread_mutex_destroy(&qset->mutex);
   tsem_destroy(&qset->sem);
   free(qset);
-  uTrace("qset:%p is closed", qset);
+  uDebug("qset:%p is closed", qset);
 }
 
 // tsem_post 'qset->sem', so that reader threads waiting for it
@@ -338,12 +349,12 @@ void taosRemoveFromQset(STaosQset *qset, STaosQueue *queue) {
 
   pthread_mutex_unlock(&qset->mutex);
 
-  uTrace("queue:%p is removed from qset:%p", queue, qset);
+  uDebug("queue:%p is removed from qset:%p", queue, qset);
 }
 
 int32_t taosGetQueueNumber(STaosQset *qset) { return qset->numOfQueues; }
 
-int32_t taosReadQitemFromQset(STaosQset *qset, void **ppItem, void **ahandle, FProcessItem *itemFp) {
+int32_t taosReadQitemFromQset(STaosQset *qset, void **ppItem, void **ahandle, FItem *itemFp) {
   STaosQnode *pNode = NULL;
   int32_t     code = 0;
 
@@ -365,6 +376,7 @@ int32_t taosReadQitemFromQset(STaosQset *qset, void **ppItem, void **ahandle, FP
       *ppItem = pNode->item;
       if (ahandle) *ahandle = queue->ahandle;
       if (itemFp) *itemFp = queue->itemFp;
+
       queue->head = pNode->next;
       if (queue->head == NULL) queue->tail = NULL;
       queue->numOfItems--;
@@ -382,7 +394,7 @@ int32_t taosReadQitemFromQset(STaosQset *qset, void **ppItem, void **ahandle, FP
   return code;
 }
 
-int32_t taosReadAllQitemsFromQset(STaosQset *qset, STaosQall *qall, void **ahandle, FProcessItems *itemsFp) {
+int32_t taosReadAllQitemsFromQset(STaosQset *qset, STaosQall *qall, void **ahandle, FItems *itemsFp) {
   STaosQueue *queue;
   int32_t     code = 0;
 
@@ -411,7 +423,9 @@ int32_t taosReadAllQitemsFromQset(STaosQset *qset, STaosQall *qall, void **ahand
       queue->tail = NULL;
       queue->numOfItems = 0;
       atomic_sub_fetch_32(&qset->numOfItems, qall->numOfItems);
-      for (int32_t j = 1; j < qall->numOfItems; ++j) tsem_wait(&qset->sem);
+      for (int32_t j = 1; j < qall->numOfItems; ++j) {
+        tsem_wait(&qset->sem);
+      }
     }
 
     pthread_mutex_unlock(&queue->mutex);
@@ -423,6 +437,65 @@ int32_t taosReadAllQitemsFromQset(STaosQset *qset, STaosQall *qall, void **ahand
   return code;
 }
 
+int32_t taosReadQitemFromQsetByThread(STaosQset *qset, void **ppItem, void **ahandle, FItem *itemFp, int32_t threadId) {
+  STaosQnode *pNode = NULL;
+  int32_t     code = -1;
+
+  tsem_wait(&qset->sem);
+
+  pthread_mutex_lock(&qset->mutex);
+
+  for (int32_t i = 0; i < qset->numOfQueues; ++i) {
+    if (qset->current == NULL) qset->current = qset->head;
+    STaosQueue *queue = qset->current;
+    if (queue) qset->current = queue->next;
+    if (queue == NULL) break;
+    if (queue->head == NULL) continue;
+    if (queue->threadId != -1 && queue->threadId != threadId) {
+      code = 0;
+      continue;
+    }
+
+    pthread_mutex_lock(&queue->mutex);
+
+    if (queue->head) {
+      pNode = queue->head;
+      pNode->queue = queue;
+      queue->threadId = threadId;
+      *ppItem = pNode->item;
+
+      if (ahandle) *ahandle = queue->ahandle;
+      if (itemFp) *itemFp = queue->itemFp;
+
+      queue->head = pNode->next;
+      if (queue->head == NULL) queue->tail = NULL;
+      queue->numOfItems--;
+      atomic_sub_fetch_32(&qset->numOfItems, 1);
+      code = 1;
+      uTrace("item:%p is read out from queue:%p, items:%d", *ppItem, queue, queue->numOfItems);
+    }
+
+    pthread_mutex_unlock(&queue->mutex);
+    if (pNode) break;
+  }
+
+  pthread_mutex_unlock(&qset->mutex);
+
+  return code;
+}
+
+void taosResetQsetThread(STaosQset *qset, void *pItem) {
+  if (pItem == NULL) return;
+  STaosQnode *pNode = (STaosQnode *)((char *)pItem - sizeof(STaosQnode));
+
+  pthread_mutex_lock(&qset->mutex);
+  pNode->queue->threadId = -1;
+  for (int32_t i = 0; i < pNode->queue->numOfItems; ++i) {
+    tsem_post(&qset->sem);
+  }
+  pthread_mutex_unlock(&qset->mutex);
+}
+
 int32_t taosGetQueueItemsNumber(STaosQueue *queue) {
   if (!queue) return 0;
 

source/util/src/tworker.c

@@ -14,38 +14,46 @@
  */
 
 #define _DEFAULT_SOURCE
-#include "os.h"
-#include "ulog.h"
-#include "tqueue.h"
 #include "tworker.h"
+#include "taoserror.h"
+#include "ulog.h"
 
-typedef void* (*ThreadFp)(void *param);
+typedef void *(*ThreadFp)(void *param);
 
-int32_t tWorkerInit(SWorkerPool *pool) {
+int32_t tQWorkerInit(SQWorkerPool *pool) {
   pool->qset = taosOpenQset();
-  pool->workers = calloc(sizeof(SWorker), pool->max);
-  pthread_mutex_init(&pool->mutex, NULL);
-  for (int i = 0; i < pool->max; ++i) {
-    SWorker *worker = pool->workers + i;
+  pool->workers = calloc(sizeof(SQWorker), pool->max);
+  if (pool->workers == NULL) {
+    terrno = TSDB_CODE_OUT_OF_MEMORY;
+    return -1;
+  }
+
+  if (pthread_mutex_init(&pool->mutex, NULL)) {
+    terrno = TSDB_CODE_OUT_OF_MEMORY;
+    return -1;
+  }
+
+  for (int32_t i = 0; i < pool->max; ++i) {
+    SQWorker *worker = pool->workers + i;
     worker->id = i;
     worker->pool = pool;
   }
 
-  uInfo("worker:%s is initialized, min:%d max:%d", pool->name, pool->min, pool->max);
+  uDebug("worker:%s is initialized, min:%d max:%d", pool->name, pool->min, pool->max);
   return 0;
 }
 
-void tWorkerCleanup(SWorkerPool *pool) {
-  for (int i = 0; i < pool->max; ++i) {
-    SWorker *worker = pool->workers + i;
+void tQWorkerCleanup(SQWorkerPool *pool) {
+  for (int32_t i = 0; i < pool->max; ++i) {
+    SQWorker *worker = pool->workers + i;
     if (worker == NULL) continue;
     if (taosCheckPthreadValid(worker->thread)) {
       taosQsetThreadResume(pool->qset);
     }
   }
 
-  for (int i = 0; i < pool->max; ++i) {
-    SWorker *worker = pool->workers + i;
+  for (int32_t i = 0; i < pool->max; ++i) {
+    SQWorker *worker = pool->workers + i;
     if (worker == NULL) continue;
     if (taosCheckPthreadValid(worker->thread)) {
       pthread_join(worker->thread, NULL);
@@ -56,15 +64,15 @@ void tWorkerCleanup(SWorkerPool *pool) {
   taosCloseQset(pool->qset);
   pthread_mutex_destroy(&pool->mutex);
 
-  uInfo("worker:%s is closed", pool->name);
+  uDebug("worker:%s is closed", pool->name);
 }
 
-static void *tWorkerThreadFp(SWorker *worker) {
-  SWorkerPool *pool = worker->pool;
-  FProcessItem fp = NULL;
+static void *tQWorkerThreadFp(SQWorker *worker) {
+  SQWorkerPool *pool = worker->pool;
+  FItem         fp = NULL;
 
-  void   *msg = NULL;
-  void   *ahandle = NULL;
+  void *  msg = NULL;
+  void *  ahandle = NULL;
   int32_t code = 0;
 
   taosBlockSIGPIPE();
@@ -77,7 +85,7 @@ static void *tWorkerThreadFp(SWorker *worker) {
       break;
     }
 
-    if (fp) {
+    if (fp != NULL) {
       (*fp)(ahandle, msg);
     }
   }
@@ -85,11 +93,12 @@ static void *tWorkerThreadFp(SWorker *worker) {
   return NULL;
 }
 
-STaosQueue *tWorkerAllocQueue(SWorkerPool *pool, void *ahandle, FProcessItem fp) {
+STaosQueue *tWorkerAllocQueue(SQWorkerPool *pool, void *ahandle, FItem fp, ThreadFp threadFp) {
   pthread_mutex_lock(&pool->mutex);
   STaosQueue *queue = taosOpenQueue();
   if (queue == NULL) {
     pthread_mutex_unlock(&pool->mutex);
+    terrno = TSDB_CODE_OUT_OF_MEMORY;
     return NULL;
   }
 
@@ -99,14 +108,18 @@ STaosQueue *tWorkerAllocQueue(SWorkerPool *pool, void *ahandle, FProcessItem fp)
   // spawn a thread to process queue
   if (pool->num < pool->max) {
     do {
-      SWorker *worker = pool->workers + pool->num;
+      SQWorker *worker = pool->workers + pool->num;
 
       pthread_attr_t thAttr;
       pthread_attr_init(&thAttr);
       pthread_attr_setdetachstate(&thAttr, PTHREAD_CREATE_JOINABLE);
 
-      if (pthread_create(&worker->thread, &thAttr, (ThreadFp)tWorkerThreadFp, worker) != 0) {
+      if (pthread_create(&worker->thread, &thAttr, threadFp, worker) != 0) {
         uError("worker:%s:%d failed to create thread to process since %s", pool->name, worker->id, strerror(errno));
+        taosCloseQueue(queue);
+        terrno = TSDB_CODE_OUT_OF_MEMORY;
+        queue = NULL;
+        break;
       }
 
       pthread_attr_destroy(&thAttr);
@@ -121,19 +134,73 @@ STaosQueue *tWorkerAllocQueue(SWorkerPool *pool, void *ahandle, FProcessItem fp)
   return queue;
 }
 
-void tWorkerFreeQueue(SWorkerPool *pool, STaosQueue *queue) {
+STaosQueue *tQWorkerAllocQueue(SQWorkerPool *pool, void *ahandle, FItem fp) {
+  return tWorkerAllocQueue(pool, ahandle, fp, (ThreadFp)tQWorkerThreadFp);
+}
+
+void tQWorkerFreeQueue(SQWorkerPool *pool, STaosQueue *queue) {
   taosCloseQueue(queue);
   uDebug("worker:%s, queue:%p is freed", pool->name, queue);
 }
 
-int32_t tMWorkerInit(SMWorkerPool *pool) {
+int32_t tFWorkerInit(SFWorkerPool *pool) { return tQWorkerInit((SQWorkerPool *)pool); }
+
+void tFWorkerCleanup(SFWorkerPool *pool) { tQWorkerCleanup(pool); }
+
+static void *tFWorkerThreadFp(SQWorker *worker) {
+  SQWorkerPool *pool = worker->pool;
+
+  FItem   fp = NULL;
+  void *  msg = NULL;
+  void *  ahandle = NULL;
+  int32_t code = 0;
+
+  taosBlockSIGPIPE();
+  setThreadName(pool->name);
+  uDebug("worker:%s:%d is running", pool->name, worker->id);
+
+  while (1) {
+    code = taosReadQitemFromQsetByThread(pool->qset, (void **)&msg, &ahandle, &fp, worker->id);
+
+    if (code < 0) {
+      uDebug("worker:%s:%d qset:%p, got no message and exiting", pool->name, worker->id, pool->qset);
+      break;
+    } else if (code == 0) {
+      // uTrace("worker:%s:%d qset:%p, got no message and continue", pool->name, worker->id, pool->qset);
+      continue;
+    }
+
+    if (fp != NULL) {
+      (*fp)(ahandle, msg);
+    }
+
+    taosResetQsetThread(pool->qset, msg);
+  }
+
+  return NULL;
+}
+
+STaosQueue *tFWorkerAllocQueue(SQWorkerPool *pool, void *ahandle, FItem fp) {
+  return tWorkerAllocQueue(pool, ahandle, fp, (ThreadFp)tFWorkerThreadFp);
+}
+
+void tFWorkerFreeQueue(SFWorkerPool *pool, STaosQueue *queue) { tQWorkerFreeQueue(pool, queue); }
+
+int32_t tWWorkerInit(SWWorkerPool *pool) {
   pool->nextId = 0;
-  pool->workers = calloc(sizeof(SMWorker), pool->max);
-  if (pool->workers == NULL) return -1;
+  pool->workers = calloc(sizeof(SWWorker), pool->max);
+  if (pool->workers == NULL) {
+    terrno = TSDB_CODE_OUT_OF_MEMORY;
+    return -1;
+  }
+
+  if (pthread_mutex_init(&pool->mutex, NULL) != 0) {
+    terrno = TSDB_CODE_OUT_OF_MEMORY;
+    return -1;
+  }
 
-  pthread_mutex_init(&pool->mutex, NULL);
   for (int32_t i = 0; i < pool->max; ++i) {
-    SMWorker *worker = pool->workers + i;
+    SWWorker *worker = pool->workers + i;
     worker->id = i;
     worker->qall = NULL;
     worker->qset = NULL;
@@ -144,16 +211,18 @@ int32_t tMWorkerInit(SMWorkerPool *pool) {
   return 0;
 }
 
-void tMWorkerCleanup(SMWorkerPool *pool) {
+void tWWorkerCleanup(SWWorkerPool *pool) {
   for (int32_t i = 0; i < pool->max; ++i) {
-    SMWorker *worker = pool->workers + i;
+    SWWorker *worker = pool->workers + i;
     if (taosCheckPthreadValid(worker->thread)) {
-      if (worker->qset) taosQsetThreadResume(worker->qset);
+      if (worker->qset) {
+        taosQsetThreadResume(worker->qset);
+      }
     }
   }
 
   for (int32_t i = 0; i < pool->max; ++i) {
-    SMWorker *worker = pool->workers + i;
+    SWWorker *worker = pool->workers + i;
     if (taosCheckPthreadValid(worker->thread)) {
       pthread_join(worker->thread, NULL);
       taosFreeQall(worker->qall);
@@ -167,12 +236,12 @@ void tMWorkerCleanup(SMWorkerPool *pool) {
   uInfo("worker:%s is closed", pool->name);
 }
 
-static void *tWriteWorkerThreadFp(SMWorker *worker) {
-  SMWorkerPool *pool = worker->pool;
-  FProcessItems fp = NULL;
+static void *tWWorkerThreadFp(SWWorker *worker) {
+  SWWorkerPool *pool = worker->pool;
+  FItems        fp = NULL;
 
-  void   *msg = NULL;
-  void   *ahandle = NULL;
+  void *  msg = NULL;
+  void *  ahandle = NULL;
   int32_t numOfMsgs = 0;
   int32_t qtype = 0;
 
@@ -187,7 +256,7 @@ static void *tWriteWorkerThreadFp(SMWorker *worker) {
       break;
     }
 
-    if (fp) {
+    if (fp != NULL) {
       (*fp)(ahandle, worker->qall, numOfMsgs);
     }
   }
@@ -195,13 +264,14 @@ static void *tWriteWorkerThreadFp(SMWorker *worker) {
   return NULL;
 }
 
-STaosQueue *tMWorkerAllocQueue(SMWorkerPool *pool, void *ahandle, FProcessItems fp) {
+STaosQueue *tWWorkerAllocQueue(SWWorkerPool *pool, void *ahandle, FItems fp) {
   pthread_mutex_lock(&pool->mutex);
-  SMWorker *worker = pool->workers + pool->nextId;
+  SWWorker *worker = pool->workers + pool->nextId;
 
   STaosQueue *queue = taosOpenQueue();
   if (queue == NULL) {
     pthread_mutex_unlock(&pool->mutex);
+    terrno = TSDB_CODE_OUT_OF_MEMORY;
     return NULL;
   }
 
@@ -221,17 +291,19 @@ STaosQueue *tMWorkerAllocQueue(SMWorkerPool *pool, void *ahandle, FProcessItems
       taosCloseQset(worker->qset);
       taosCloseQueue(queue);
       pthread_mutex_unlock(&pool->mutex);
+      terrno = TSDB_CODE_OUT_OF_MEMORY;
       return NULL;
     }
     pthread_attr_t thAttr;
     pthread_attr_init(&thAttr);
     pthread_attr_setdetachstate(&thAttr, PTHREAD_CREATE_JOINABLE);
 
-    if (pthread_create(&worker->thread, &thAttr, (ThreadFp)tWriteWorkerThreadFp, worker) != 0) {
+    if (pthread_create(&worker->thread, &thAttr, (ThreadFp)tWWorkerThreadFp, worker) != 0) {
       uError("worker:%s:%d failed to create thread to process since %s", pool->name, worker->id, strerror(errno));
       taosFreeQall(worker->qall);
       taosCloseQset(worker->qset);
       taosCloseQueue(queue);
+      terrno = TSDB_CODE_OUT_OF_MEMORY;
       queue = NULL;
     } else {
       uDebug("worker:%s:%d is launched, max:%d", pool->name, worker->id, pool->max);
@@ -250,7 +322,7 @@ STaosQueue *tMWorkerAllocQueue(SMWorkerPool *pool, void *ahandle, FProcessItems
   return queue;
 }
 
-void tMWorkerFreeQueue(SMWorkerPool *pool, STaosQueue *queue) {
+void tWWorkerFreeQueue(SWWorkerPool *pool, STaosQueue *queue) {
   taosCloseQueue(queue);
   uDebug("worker:%s, queue:%p is freed", pool->name, queue);
 }

source/util/test/CMakeLists.txt

@@ -13,12 +13,12 @@ IF (HEADER_GTEST_INCLUDE_DIR AND (LIB_GTEST_STATIC_DIR OR LIB_GTEST_SHARED_DIR))
 
     LIST(REMOVE_ITEM SOURCE_LIST ${CMAKE_CURRENT_SOURCE_DIR}/trefTest.c)
     ADD_EXECUTABLE(utilTest ${SOURCE_LIST})
-    TARGET_LINK_LIBRARIES(utilTest util common os gtest pthread gcov)
+    TARGET_LINK_LIBRARIES(utilTest util common os gtest pthread)
 
     LIST(REMOVE_ITEM SOURCE_LIST ${CMAKE_CURRENT_SOURCE_DIR}/cacheTest.cpp)
     LIST(APPEND SOURCE_LIST ${CMAKE_CURRENT_SOURCE_DIR}/hashTest.cpp)
     ADD_EXECUTABLE(hashTest ${SOURCE_LIST})
-    TARGET_LINK_LIBRARIES(hashTest util common os gtest pthread gcov)
+    TARGET_LINK_LIBRARIES(hashTest util common os gtest pthread)
     
     LIST(APPEND BIN_SRC ${CMAKE_CURRENT_SOURCE_DIR}/trefTest.c)
     ADD_EXECUTABLE(trefTest ${BIN_SRC})

tests/script/tsim/dnode/basic1.sim

@@ -178,9 +178,9 @@ if $rows != 3 then
 endi
 
 sql select * from st
-#if $rows != 15 then 
-#  return -1
-#endi
+if $rows != 15 then 
+  return -1
+endi
 
 print =============== drop dnode
 sql drop dnode 2;

tests/script/tsim/table/basic1.sim

@@ -139,9 +139,9 @@ endi
 
 print =============== query data frpm st
 sql select * from st
-#if $rows != 21 then 
-#  return -1
-#endi
+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
@@ -200,8 +200,8 @@ endi
 
 print =============== query data frpm st
 sql select * from st
-#if $rows != 21 then 
-#  return -1
-#endi
+if $rows != 21 then 
+  return -1
+endi
 
 system sh/exec.sh -n dnode1 -s stop -x SIGINT
\ No newline at end of file