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

include/common/tmsg.h

@@ -2100,6 +2100,18 @@ enum {
   TOPIC_SUB_TYPE__TABLE,
 };
 
+typedef struct {
+  SMsgHead head;
+  int64_t  leftForVer;
+  int32_t  vgId;
+  int64_t  consumerId;
+  char     subKey[TSDB_SUBSCRIBE_KEY_LEN];
+} SMqVDeleteReq;
+
+typedef struct {
+  int8_t reserved;
+} SMqVDeleteRsp;
+
 typedef struct {
   int64_t leftForVer;
   int32_t vgId;
@@ -2258,20 +2270,22 @@ static FORCE_INLINE void tdDestroyTSma(STSma* pSma) {
   }
 }
 
-static FORCE_INLINE void tdDestroyTSmaWrapper(STSmaWrapper* pSW) {
+static FORCE_INLINE void tdDestroyTSmaWrapper(STSmaWrapper* pSW, bool deepCopy) {
   if (pSW) {
     if (pSW->tSma) {
+      if (deepCopy) {
         for (uint32_t i = 0; i < pSW->number; ++i) {
           tdDestroyTSma(pSW->tSma + i);
         }
+      }
       taosMemoryFreeClear(pSW->tSma);
     }
   }
 }
 
-static FORCE_INLINE void* tdFreeTSmaWrapper(STSmaWrapper* pSW) {
+static FORCE_INLINE void* tdFreeTSmaWrapper(STSmaWrapper* pSW, bool deepCopy) {
-  tdDestroyTSmaWrapper(pSW);
+  tdDestroyTSmaWrapper(pSW, deepCopy);
-  taosMemoryFree(pSW);
+  taosMemoryFreeClear(pSW);
   return NULL;
 }
 
@@ -2535,11 +2549,15 @@ static FORCE_INLINE void tDeleteSMqAskEpRsp(SMqAskEpRsp* pRsp) {
 }
 
 typedef struct {
-  void* data;
+  int64_t streamId;
+  int32_t taskId;
+  int32_t sourceVg;
+  int64_t sourceVer;
+  SArray* data;  // SArray<SSDataBlock>
 } SStreamDispatchReq;
 
 typedef struct {
-  int8_t status;
+  int8_t inputStatus;
 } SStreamDispatchRsp;
 
 #define TD_AUTO_CREATE_TABLE 0x1

include/common/tmsgdef.h

@@ -178,6 +178,7 @@ enum {
   TD_DEF_MSG_TYPE(TDMT_VND_MQ_CONNECT, "vnode-mq-connect", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_VND_MQ_DISCONNECT, "vnode-mq-disconnect", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_VND_MQ_VG_CHANGE, "vnode-mq-vg-change", SMqRebVgReq, SMqRebVgRsp)
+  TD_DEF_MSG_TYPE(TDMT_VND_MQ_VG_DELETE, "vnode-mq-vg-delete", SMqVDeleteReq, SMqVDeleteRsp)
   TD_DEF_MSG_TYPE(TDMT_VND_RES_READY, "vnode-res-ready", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_VND_TASKS_STATUS, "vnode-tasks-status", NULL, NULL)
   TD_DEF_MSG_TYPE(TDMT_VND_CANCEL_TASK, "vnode-cancel-task", NULL, NULL)

include/libs/function/functionMgt.h

@@ -41,6 +41,7 @@ typedef enum EFunctionType {
   FUNCTION_TYPE_SUM,
   FUNCTION_TYPE_TWA,
   FUNCTION_TYPE_HISTOGRAM,
+  FUNCTION_TYPE_HYPERLOGLOG,
 
   // nonstandard SQL function
   FUNCTION_TYPE_BOTTOM = 500,

include/libs/function/tudf.h

@@ -87,6 +87,7 @@ typedef struct SUdfInterBuf {
 } SUdfInterBuf;
 typedef void *UdfcFuncHandle;
 
+//low level APIs
 /**
  * setup udf
  * @param udf, in
@@ -115,6 +116,9 @@ int32_t doCallUdfScalarFunc(UdfcFuncHandle handle, SScalarParam *input, int32_t
  */
 int32_t doTeardownUdf(UdfcFuncHandle handle);
 
+void freeUdfInterBuf(SUdfInterBuf *buf);
+
+//high level APIs
 bool udfAggGetEnv(struct SFunctionNode* pFunc, SFuncExecEnv* pEnv);
 bool udfAggInit(struct SqlFunctionCtx *pCtx, struct SResultRowEntryInfo* pResultCellInfo);
 int32_t udfAggProcess(struct SqlFunctionCtx *pCtx);

include/util/taoserror.h

@@ -281,6 +281,7 @@ int32_t* taosGetErrno();
 #define TSDB_CODE_MND_SUBSCRIBE_NOT_EXIST       TAOS_DEF_ERROR_CODE(0, 0x03E8)
 #define TSDB_CODE_MND_OFFSET_NOT_EXIST          TAOS_DEF_ERROR_CODE(0, 0x03E9)
 #define TSDB_CODE_MND_CONSUMER_NOT_READY        TAOS_DEF_ERROR_CODE(0, 0x03EA)
+#define TSDB_CODE_MND_TOPIC_SUBSCRIBED          TAOS_DEF_ERROR_CODE(0, 0x03EB)
 
 // mnode-stream
 #define TSDB_CODE_MND_STREAM_ALREADY_EXIST      TAOS_DEF_ERROR_CODE(0, 0x03F0)

include/util/thash.h

@@ -40,6 +40,7 @@ typedef void (*_hash_free_fn_t)(void *);
  */
 uint32_t MurmurHash3_32(const char *key, uint32_t len);
 
+uint64_t MurmurHash3_64(const char *key, uint32_t len);
 /**
  *
  * @param key

source/dnode/mgmt/mgmt_mnode/src/mmHandle.c

@@ -216,6 +216,7 @@ SArray *mmGetMsgHandles() {
   if (dmSetMgmtHandle(pArray, TDMT_MND_MQ_COMMIT_OFFSET, mmPutNodeMsgToWriteQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_MND_MQ_ASK_EP, mmPutNodeMsgToReadQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_VND_MQ_VG_CHANGE_RSP, mmPutNodeMsgToWriteQueue, 0) == NULL) goto _OVER;
+  if (dmSetMgmtHandle(pArray, TDMT_VND_MQ_VG_DELETE_RSP, mmPutNodeMsgToWriteQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_MND_CREATE_STREAM, mmPutNodeMsgToWriteQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_VND_TASK_DEPLOY_RSP, mmPutNodeMsgToWriteQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_MND_GET_DB_CFG, mmPutNodeMsgToReadQueue, 0) == NULL) goto _OVER;

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

@@ -308,6 +308,7 @@ SArray *vmGetMsgHandles() {
   if (dmSetMgmtHandle(pArray, TDMT_VND_DROP_SMA, vmPutNodeMsgToWriteQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_VND_SUBMIT_RSMA, vmPutNodeMsgToWriteQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_VND_MQ_VG_CHANGE, vmPutNodeMsgToWriteQueue, 0) == NULL) goto _OVER;
+  if (dmSetMgmtHandle(pArray, TDMT_VND_MQ_VG_DELETE, vmPutNodeMsgToWriteQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_VND_CONSUME, vmPutNodeMsgToFetchQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_VND_TASK_DEPLOY, vmPutNodeMsgToWriteQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_VND_QUERY_HEARTBEAT, vmPutNodeMsgToFetchQueue, 0) == NULL) goto _OVER;

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

@@ -459,6 +459,7 @@ typedef struct {
   char*          ast;
   char*          physicalPlan;
   SSchemaWrapper schema;
+  int32_t        refConsumerCnt;
 } SMqTopicObj;
 
 typedef struct {

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

@@ -38,6 +38,9 @@ static FORCE_INLINE int32_t mndMakePartitionKey(char *key, const char *cgroup, c
 }
 
 int32_t mndDropOffsetByDB(SMnode *pMnode, STrans *pTrans, SDbObj *pDb);
+int32_t mndDropOffsetByTopic(SMnode *pMnode, STrans *pTrans, const char *topic);
+
+bool mndOffsetFromTopic(SMqOffsetObj *pOffset, const char *topic);
 
 #ifdef __cplusplus
 }

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

@@ -32,6 +32,7 @@ void             mndReleaseSubscribe(SMnode *pMnode, SMqSubscribeObj *pSub);
 int32_t mndMakeSubscribeKey(char *key, const char *cgroup, const char *topicName);
 
 int32_t mndDropSubByDB(SMnode *pMnode, STrans *pTrans, SDbObj *pDb);
+int32_t mndDropSubByTopic(SMnode *pMnode, STrans *pTrans, const char *topic);
 
 #ifdef __cplusplus
 }

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

@@ -35,6 +35,8 @@ int32_t mndDropTopicByDB(SMnode *pMnode, STrans *pTrans, SDbObj *pDb);
 
 const char *mndTopicGetShowName(const char topic[TSDB_TOPIC_FNAME_LEN]);
 
+int32_t mndSetTopicRedoLogs(SMnode *pMnode, STrans *pTrans, SMqTopicObj *pTopic);
+
 #ifdef __cplusplus
 }
 #endif

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

@@ -399,6 +399,9 @@ static int32_t mndProcessSubscribeReq(SNodeMsg *pMsg) {
 
   int32_t newTopicNum = taosArrayGetSize(newSub);
   // check topic existance
+  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_RETRY, TRN_TYPE_SUBSCRIBE, &pMsg->rpcMsg);
+  if (pTrans == NULL) goto SUBSCRIBE_OVER;
+
   for (int32_t i = 0; i < newTopicNum; i++) {
     char        *topic = taosArrayGetP(newSub, i);
     SMqTopicObj *pTopic = mndAcquireTopic(pMnode, topic);
@@ -406,7 +409,14 @@ static int32_t mndProcessSubscribeReq(SNodeMsg *pMsg) {
       terrno = TSDB_CODE_MND_TOPIC_NOT_EXIST;
       goto SUBSCRIBE_OVER;
     }
-    // TODO lock topic to prevent drop
+
+    // ref topic to prevent drop
+    // TODO make topic complete
+    SMqTopicObj topicObj = {0};
+    memcpy(&topicObj, pTopic, sizeof(SMqTopicObj));
+    topicObj.refConsumerCnt = pTopic->refConsumerCnt + 1;
+    if (mndSetTopicRedoLogs(pMnode, pTrans, &topicObj) != 0) goto SUBSCRIBE_OVER;
+
     mndReleaseTopic(pMnode, pTopic);
   }
 
@@ -422,8 +432,6 @@ static int32_t mndProcessSubscribeReq(SNodeMsg *pMsg) {
       taosArrayPush(pConsumerNew->assignedTopics, &newTopicCopy);
     }
 
-    STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_RETRY, TRN_TYPE_SUBSCRIBE, &pMsg->rpcMsg);
-    if (pTrans == NULL) goto SUBSCRIBE_OVER;
     if (mndSetConsumerCommitLogs(pMnode, pTrans, pConsumerNew) != 0) goto SUBSCRIBE_OVER;
     if (mndTransPrepare(pMnode, pTrans) != 0) goto SUBSCRIBE_OVER;
 
@@ -494,8 +502,6 @@ static int32_t mndProcessSubscribeReq(SNodeMsg *pMsg) {
       goto SUBSCRIBE_OVER;
     }
 
-    STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_RETRY, TRN_TYPE_SUBSCRIBE, &pMsg->rpcMsg);
-    if (pTrans == NULL) goto SUBSCRIBE_OVER;
     if (mndSetConsumerCommitLogs(pMnode, pTrans, pConsumerNew) != 0) goto SUBSCRIBE_OVER;
     if (mndTransPrepare(pMnode, pTrans) != 0) goto SUBSCRIBE_OVER;
   }
@@ -503,6 +509,8 @@ static int32_t mndProcessSubscribeReq(SNodeMsg *pMsg) {
   code = TSDB_CODE_MND_ACTION_IN_PROGRESS;
 
 SUBSCRIBE_OVER:
+  mndTransDrop(pTrans);
+
   if (pConsumerOld) {
     /*taosRUnLockLatch(&pConsumerOld->lock);*/
     mndReleaseConsumer(pMnode, pConsumerOld);

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

@@ -50,6 +50,14 @@ int32_t mndInitOffset(SMnode *pMnode) {
 
 void mndCleanupOffset(SMnode *pMnode) {}
 
+bool mndOffsetFromTopic(SMqOffsetObj *pOffset, const char *topic) {
+  int32_t i = 0;
+  while (pOffset->key[i] != ':') i++;
+  while (pOffset->key[i] != ':') i++;
+  if (strcmp(&pOffset->key[i + 1], topic) == 0) return true;
+  return false;
+}
+
 SSdbRaw *mndOffsetActionEncode(SMqOffsetObj *pOffset) {
   terrno = TSDB_CODE_OUT_OF_MEMORY;
   void   *buf = NULL;
@@ -134,10 +142,11 @@ int32_t mndCreateOffsets(STrans *pTrans, const char *cgroup, const char *topicNa
   int32_t sz = taosArrayGetSize(vgs);
   for (int32_t i = 0; i < sz; i++) {
     int32_t      vgId = *(int32_t *)taosArrayGet(vgs, i);
-    SMqOffsetObj offsetObj;
+    SMqOffsetObj offsetObj = {0};
     if (mndMakePartitionKey(offsetObj.key, cgroup, topicName, vgId) < 0) {
       return -1;
     }
+    // TODO assign db
     offsetObj.offset = -1;
     SSdbRaw *pOffsetRaw = mndOffsetActionEncode(&offsetObj);
     if (pOffsetRaw == NULL) {
@@ -240,6 +249,14 @@ static int32_t mndSetDropOffsetCommitLogs(SMnode *pMnode, STrans *pTrans, SMqOff
   return 0;
 }
 
+static int32_t mndSetDropOffsetRedoLogs(SMnode *pMnode, STrans *pTrans, SMqOffsetObj *pOffset) {
+  SSdbRaw *pRedoRaw = mndOffsetActionEncode(pOffset);
+  if (pRedoRaw == NULL) return -1;
+  if (mndTransAppendRedolog(pTrans, pRedoRaw) != 0) return -1;
+  if (sdbSetRawStatus(pRedoRaw, SDB_STATUS_DROPPED) != 0) return -1;
+  return 0;
+}
+
 int32_t mndDropOffsetByDB(SMnode *pMnode, STrans *pTrans, SDbObj *pDb) {
   int32_t code = -1;
   SSdb   *pSdb = pMnode->pSdb;
@@ -247,7 +264,7 @@ int32_t mndDropOffsetByDB(SMnode *pMnode, STrans *pTrans, SDbObj *pDb) {
   void         *pIter = NULL;
   SMqOffsetObj *pOffset = NULL;
   while (1) {
-    pIter = sdbFetch(pSdb, SDB_SUBSCRIBE, pIter, (void **)&pOffset);
+    pIter = sdbFetch(pSdb, SDB_OFFSET, pIter, (void **)&pOffset);
     if (pIter == NULL) break;
 
     if (pOffset->dbUid != pDb->uid) {
@@ -256,8 +273,39 @@ int32_t mndDropOffsetByDB(SMnode *pMnode, STrans *pTrans, SDbObj *pDb) {
     }
 
     if (mndSetDropOffsetCommitLogs(pMnode, pTrans, pOffset) < 0) {
+      sdbRelease(pSdb, pOffset);
       goto END;
     }
+
+    sdbRelease(pSdb, pOffset);
+  }
+
+  code = 0;
+END:
+  return code;
+}
+
+int32_t mndDropOffsetByTopic(SMnode *pMnode, STrans *pTrans, const char *topic) {
+  int32_t code = -1;
+  SSdb   *pSdb = pMnode->pSdb;
+
+  void         *pIter = NULL;
+  SMqOffsetObj *pOffset = NULL;
+  while (1) {
+    pIter = sdbFetch(pSdb, SDB_OFFSET, pIter, (void **)&pOffset);
+    if (pIter == NULL) break;
+
+    if (!mndOffsetFromTopic(pOffset, topic)) {
+      sdbRelease(pSdb, pOffset);
+      continue;
+    }
+
+    if (mndSetDropOffsetRedoLogs(pMnode, pTrans, pOffset) < 0) {
+      sdbRelease(pSdb, pOffset);
+      goto END;
+    }
+
+    sdbRelease(pSdb, pOffset);
   }
 
   code = 0;

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

@@ -530,6 +530,7 @@ static int32_t mndProcessMCreateSmaReq(SNodeMsg *pReq) {
   pStream = mndAcquireStream(pMnode, createReq.name);
   if (pStream != NULL) {
     mError("sma:%s, failed to create since stream:%s already exist", createReq.name, createReq.name);
+    terrno = TSDB_CODE_MND_STREAM_ALREADY_EXIST;
     goto _OVER;
   }
 
@@ -565,7 +566,7 @@ static int32_t mndProcessMCreateSmaReq(SNodeMsg *pReq) {
 
 _OVER:
   if (code != 0 && code != TSDB_CODE_MND_ACTION_IN_PROGRESS) {
-    mError("sma:%s, failed to create since %s", createReq.name, terrstr());
+    mError("sma:%s, failed to create since %s", createReq.name, terrstr(terrno));
   }
 
   mndReleaseStb(pMnode, pStb);

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

@@ -386,7 +386,7 @@ static void *mndBuildVCreateStbReq(SMnode *pMnode, SVgObj *pVgroup, SStbObj *pSt
   req.schema.sver = pStb->version;
   req.schema.pSchema = pStb->pColumns;
   req.schemaTag.nCols = pStb->numOfTags;
-  req.schemaTag.nCols = 0;
+  req.schemaTag.sver = 1;
   req.schemaTag.pSchema = pStb->pTags;
 
   if (req.rollup) {

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

@@ -73,6 +73,7 @@ int32_t mndInitSubscribe(SMnode *pMnode) {
                      .deleteFp = (SdbDeleteFp)mndSubActionDelete};
 
   mndSetMsgHandle(pMnode, TDMT_VND_MQ_VG_CHANGE_RSP, mndProcessSubscribeInternalRsp);
+  mndSetMsgHandle(pMnode, TDMT_VND_MQ_VG_DELETE_RSP, mndProcessSubscribeInternalRsp);
   mndSetMsgHandle(pMnode, TDMT_MND_MQ_DO_REBALANCE, mndProcessRebalanceReq);
 
   mndAddShowRetrieveHandle(pMnode, TSDB_MGMT_TABLE_SUBSCRIPTIONS, mndRetrieveSubscribe);
@@ -389,7 +390,7 @@ static int32_t mndDoRebalance(SMnode *pMnode, const SMqRebInputObj *pInput, SMqR
 }
 
 static int32_t mndPersistRebResult(SMnode *pMnode, SNodeMsg *pMsg, const SMqRebOutputObj *pOutput) {
-  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_RETRY, TRN_TYPE_REBALANCE, &pMsg->rpcMsg);
+  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_ROLLBACK, TRN_TYPE_REBALANCE, &pMsg->rpcMsg);
   if (pTrans == NULL) {
     return -1;
   }
@@ -458,6 +459,20 @@ static int32_t mndPersistRebResult(SMnode *pMnode, SNodeMsg *pMsg, const SMqRebO
       goto REB_FAIL;
     }
   }
+  if (consumerNum) {
+    char topic[TSDB_TOPIC_FNAME_LEN];
+    char cgroup[TSDB_CGROUP_LEN];
+    mndSplitSubscribeKey(pOutput->pSub->key, topic, cgroup, true);
+    SMqTopicObj *pTopic = mndAcquireTopic(pMnode, topic);
+    if (pTopic) {
+      // TODO make topic complete
+      SMqTopicObj topicObj = {0};
+      memcpy(&topicObj, pTopic, sizeof(SMqTopicObj));
+      topicObj.refConsumerCnt = pTopic->refConsumerCnt - consumerNum;
+      if (mndSetTopicRedoLogs(pMnode, pTrans, &topicObj) != 0) goto REB_FAIL;
+    }
+  }
+
   // 4. TODO commit log: modification log
 
   // 5. set cb
@@ -688,6 +703,14 @@ static int32_t mndProcessSubscribeInternalRsp(SNodeMsg *pRsp) {
   return 0;
 }
 
+static int32_t mndSetDropSubRedoLogs(SMnode *pMnode, STrans *pTrans, SMqSubscribeObj *pSub) {
+  SSdbRaw *pRedoRaw = mndSubActionEncode(pSub);
+  if (pRedoRaw == NULL) return -1;
+  if (mndTransAppendRedolog(pTrans, pRedoRaw) != 0) return -1;
+  if (sdbSetRawStatus(pRedoRaw, SDB_STATUS_DROPPED) != 0) return -1;
+  return 0;
+}
+
 static int32_t mndSetDropSubCommitLogs(SMnode *pMnode, STrans *pTrans, SMqSubscribeObj *pSub) {
   SSdbRaw *pCommitRaw = mndSubActionEncode(pSub);
   if (pCommitRaw == NULL) return -1;
@@ -712,6 +735,57 @@ int32_t mndDropSubByDB(SMnode *pMnode, STrans *pTrans, SDbObj *pDb) {
     }
 
     if (mndSetDropSubCommitLogs(pMnode, pTrans, pSub) < 0) {
+      sdbRelease(pSdb, pSub);
+      goto END;
+    }
+  }
+
+  code = 0;
+END:
+  return code;
+}
+
+int32_t mndDropSubByTopic(SMnode *pMnode, STrans *pTrans, const char *topicName) {
+  int32_t code = -1;
+  SSdb   *pSdb = pMnode->pSdb;
+
+  void            *pIter = NULL;
+  SMqSubscribeObj *pSub = NULL;
+  while (1) {
+    pIter = sdbFetch(pSdb, SDB_SUBSCRIBE, pIter, (void **)&pSub);
+    if (pIter == NULL) break;
+
+    char topic[TSDB_TOPIC_FNAME_LEN];
+    char cgroup[TSDB_CGROUP_LEN];
+    mndSplitSubscribeKey(pSub->key, topic, cgroup, true);
+    if (strcmp(topic, topicName) != 0) {
+      sdbRelease(pSdb, pSub);
+      continue;
+    }
+
+    // iter all vnode to delete handle
+    ASSERT(taosHashGetSize(pSub->consumerHash) == 0);
+    int32_t sz = taosArrayGetSize(pSub->unassignedVgs);
+    for (int32_t i = 0; i < sz; i++) {
+      SMqVgEp       *pVgEp = taosArrayGetP(pSub->unassignedVgs, i);
+      SMqVDeleteReq *pReq = taosMemoryCalloc(1, sizeof(SMqVDeleteReq));
+      pReq->head.vgId = htonl(pVgEp->vgId);
+      pReq->vgId = pVgEp->vgId;
+      pReq->consumerId = -1;
+      memcpy(pReq->subKey, pSub->key, TSDB_SUBSCRIBE_KEY_LEN);
+      STransAction action = {0};
+      action.epSet = pVgEp->epSet;
+      action.pCont = pReq;
+      action.contLen = sizeof(SMqVDeleteReq);
+      action.msgType = TDMT_VND_MQ_VG_DELETE;
+      if (mndTransAppendRedoAction(pTrans, &action) != 0) {
+        taosMemoryFree(pReq);
+        return -1;
+      }
+    }
+
+    if (mndSetDropSubRedoLogs(pMnode, pTrans, pSub) < 0) {
+      sdbRelease(pSdb, pSub);
       goto END;
     }
   }

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

@@ -18,8 +18,10 @@
 #include "mndDb.h"
 #include "mndDnode.h"
 #include "mndMnode.h"
+#include "mndOffset.h"
 #include "mndShow.h"
 #include "mndStb.h"
+#include "mndSubscribe.h"
 #include "mndTrans.h"
 #include "mndUser.h"
 #include "mndVgroup.h"
@@ -106,6 +108,7 @@ SSdbRaw *mndTopicActionEncode(SMqTopicObj *pTopic) {
   taosEncodeSSchemaWrapper(&aswBuf, &pTopic->schema);
   SDB_SET_INT32(pRaw, dataPos, schemaLen, TOPIC_ENCODE_OVER);
   SDB_SET_BINARY(pRaw, dataPos, swBuf, schemaLen, TOPIC_ENCODE_OVER);
+  SDB_SET_INT32(pRaw, dataPos, pTopic->refConsumerCnt, TOPIC_ENCODE_OVER);
 
   SDB_SET_RESERVE(pRaw, dataPos, MND_TOPIC_RESERVE_SIZE, TOPIC_ENCODE_OVER);
   SDB_SET_DATALEN(pRaw, dataPos, TOPIC_ENCODE_OVER);
@@ -190,6 +193,8 @@ SSdbRow *mndTopicActionDecode(SSdbRaw *pRaw) {
     goto TOPIC_DECODE_OVER;
   }
 
+  SDB_GET_INT32(pRaw, dataPos, &pTopic->refConsumerCnt, TOPIC_DECODE_OVER);
+
   SDB_GET_RESERVE(pRaw, dataPos, MND_TOPIC_RESERVE_SIZE, TOPIC_DECODE_OVER);
 
   terrno = TSDB_CODE_SUCCESS;
@@ -220,11 +225,13 @@ static int32_t mndTopicActionUpdate(SSdb *pSdb, SMqTopicObj *pOldTopic, SMqTopic
   atomic_exchange_64(&pOldTopic->updateTime, pNewTopic->updateTime);
   atomic_exchange_32(&pOldTopic->version, pNewTopic->version);
 
-  taosWLockLatch(&pOldTopic->lock);
+  atomic_store_32(&pOldTopic->refConsumerCnt, pNewTopic->refConsumerCnt);
+
+  /*taosWLockLatch(&pOldTopic->lock);*/
 
   // TODO handle update
 
-  taosWUnLockLatch(&pOldTopic->lock);
+  /*taosWUnLockLatch(&pOldTopic->lock);*/
   return 0;
 }
 
@@ -292,6 +299,7 @@ static int32_t mndCreateTopic(SMnode *pMnode, SNodeMsg *pReq, SCMCreateTopicReq
   topicObj.version = 1;
   topicObj.sql = strdup(pCreate->sql);
   topicObj.sqlLen = strlen(pCreate->sql) + 1;
+  topicObj.refConsumerCnt = 0;
 
   if (pCreate->ast && pCreate->ast[0]) {
     topicObj.ast = strdup(pCreate->ast);
@@ -436,15 +444,7 @@ CREATE_TOPIC_OVER:
   return code;
 }
 
-static int32_t mndDropTopic(SMnode *pMnode, SNodeMsg *pReq, SMqTopicObj *pTopic) {
+static int32_t mndDropTopic(SMnode *pMnode, STrans *pTrans, SNodeMsg *pReq, SMqTopicObj *pTopic) {
-  // TODO: cannot drop when subscribed
-  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_ROLLBACK, TRN_TYPE_DROP_TOPIC, &pReq->rpcMsg);
-  if (pTrans == NULL) {
-    mError("topic:%s, failed to drop since %s", pTopic->name, terrstr());
-    return -1;
-  }
-  mDebug("trans:%d, used to drop topic:%s", pTrans->id, pTopic->name);
-
   SSdbRaw *pRedoRaw = mndTopicActionEncode(pTopic);
   if (pRedoRaw == NULL || mndTransAppendRedolog(pTrans, pRedoRaw) != 0) {
     mError("trans:%d, failed to append redo log since %s", pTrans->id, terrstr());
@@ -465,6 +465,7 @@ static int32_t mndDropTopic(SMnode *pMnode, SNodeMsg *pReq, SMqTopicObj *pTopic)
 
 static int32_t mndProcessDropTopicReq(SNodeMsg *pReq) {
   SMnode        *pMnode = pReq->pNode;
+  SSdb          *pSdb = pMnode->pSdb;
   SMDropTopicReq dropReq = {0};
 
   if (tDeserializeSMDropTopicReq(pReq->rpcMsg.pCont, pReq->rpcMsg.contLen, &dropReq) != 0) {
@@ -485,10 +486,35 @@ static int32_t mndProcessDropTopicReq(SNodeMsg *pReq) {
       return -1;
     }
   }
-  // TODO: check ref
 
-  int32_t code = mndDropTopic(pMnode, pReq, pTopic);
+  // check ref
-  // TODO: iterate and drop related subscriptions and offsets
+  if (pTopic->refConsumerCnt != 0) {
+    terrno = TSDB_CODE_MND_TOPIC_SUBSCRIBED;
+    mError("topic:%s, failed to drop since %s", dropReq.name, terrstr());
+    return -1;
+  }
+
+  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_ROLLBACK, TRN_TYPE_DROP_TOPIC, &pReq->rpcMsg);
+  if (pTrans == NULL) {
+    mError("topic:%s, failed to drop since %s", pTopic->name, terrstr());
+    return -1;
+  }
+
+  mDebug("trans:%d, used to drop topic:%s", pTrans->id, pTopic->name);
+
+#if 1
+  if (mndDropOffsetByTopic(pMnode, pTrans, dropReq.name) < 0) {
+    ASSERT(0);
+    return -1;
+  }
+#endif
+
+  if (mndDropSubByTopic(pMnode, pTrans, dropReq.name) < 0) {
+    ASSERT(0);
+    return -1;
+  }
+
+  int32_t code = mndDropTopic(pMnode, pTrans, pReq, pTopic);
   mndReleaseTopic(pMnode, pTopic);
 
   if (code != 0) {
@@ -577,6 +603,15 @@ static int32_t mndRetrieveTopic(SNodeMsg *pReq, SShowObj *pShow, SSDataBlock *pB
   return numOfRows;
 }
 
+int32_t mndSetTopicRedoLogs(SMnode *pMnode, STrans *pTrans, SMqTopicObj *pTopic) {
+  SSdbRaw *pRedoRaw = mndTopicActionEncode(pTopic);
+  if (pRedoRaw == NULL) return -1;
+  if (mndTransAppendCommitlog(pTrans, pRedoRaw) != 0) return -1;
+  if (sdbSetRawStatus(pRedoRaw, SDB_STATUS_READY) != 0) return -1;
+
+  return 0;
+}
+
 static int32_t mndSetDropTopicCommitLogs(SMnode *pMnode, STrans *pTrans, SMqTopicObj *pTopic) {
   SSdbRaw *pCommitRaw = mndTopicActionEncode(pTopic);
   if (pCommitRaw == NULL) return -1;

source/dnode/vnode/inc/vnode.h

@@ -196,6 +196,8 @@ struct SMetaEntry {
       STSma *tsma;
     } smaEntry;
   };
+
+  uint8_t *pBuf;
 };
 
 struct SMetaReader {

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

@@ -24,7 +24,6 @@ extern "C" {
 
 typedef struct SMetaIdx    SMetaIdx;
 typedef struct SMetaDB     SMetaDB;
-typedef struct SMSmaCursor SMSmaCursor;
 
 // metaDebug ==================
 // clang-format off
@@ -114,22 +113,12 @@ typedef struct {
   int64_t  smaUid;
 } SSmaIdxKey;
 
-#if 1
-
-SMSmaCursor* metaOpenSmaCursor(SMeta* pMeta, tb_uid_t uid);
-void         metaCloseSmaCursor(SMSmaCursor* pSmaCur);
-int64_t      metaSmaCursorNext(SMSmaCursor* pSmaCur);
-
 #ifndef META_REFACT
 // SMetaDB
 int  metaOpenDB(SMeta* pMeta);
 void metaCloseDB(SMeta* pMeta);
 int  metaSaveTableToDB(SMeta* pMeta, STbCfg* pTbCfg, STbDdlH* pHandle);
 int metaRemoveTableFromDb(SMeta* pMeta, tb_uid_t uid);
-int metaSaveSmaToDB(SMeta* pMeta, STSma* pTbCfg);
-int metaRemoveSmaFromDb(SMeta* pMeta, int64_t indexUid);
-#endif
-
 #endif
 
 #ifdef __cplusplus

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

@@ -218,6 +218,11 @@ static int32_t  tdInitSmaEnv(SSma *pSma, int8_t smaType, const char *path, SDisk
 
 void *tdFreeRSmaInfo(SRSmaInfo *pInfo);
 
+int32_t tdProcessTSmaCreateImpl(SSma *pSma, int64_t version, const char *pMsg);
+int32_t tdUpdateExpiredWindowImpl(SSma *pSma, SSubmitReq *pMsg, int64_t version);
+// TODO: This is the basic params, and should wrap the params to a queryHandle.
+int32_t tdGetTSmaDataImpl(SSma *pSma, char *pData, int64_t indexUid, TSKEY querySKey, int32_t nMaxResult);
+
 #ifdef __cplusplus
 }
 #endif

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

@@ -89,11 +89,13 @@ STSchema*       metaGetTbTSchema(SMeta* pMeta, tb_uid_t uid, int32_t sver);
 int             metaGetTableEntryByName(SMetaReader* pReader, const char* name);
 int             metaGetTbNum(SMeta* pMeta);
 SMCtbCursor*    metaOpenCtbCursor(SMeta* pMeta, tb_uid_t uid);
-void            metaCloseCtbCurosr(SMCtbCursor* pCtbCur);
+void            metaCloseCtbCursor(SMCtbCursor* pCtbCur);
 tb_uid_t        metaCtbCursorNext(SMCtbCursor* pCtbCur);
-SArray*         metaGetSmaTbUids(SMeta* pMeta, bool isDup);
+STSma*          metaGetSmaInfoByIndex(SMeta* pMeta, int64_t indexUid);
-void*           metaGetSmaInfoByIndex(SMeta* pMeta, int64_t indexUid, bool isDecode);
+STSmaWrapper*   metaGetSmaInfoByTable(SMeta* pMeta, tb_uid_t uid, bool deepCopy);
-STSmaWrapper*   metaGetSmaInfoByTable(SMeta* pMeta, tb_uid_t uid);
+SArray*         metaGetSmaIdsByTable(SMeta* pMeta, tb_uid_t uid);
+SArray*         metaGetSmaTbUids(SMeta* pMeta);
+
 int32_t         metaCreateTSma(SMeta* pMeta, int64_t version, SSmaCfg* pCfg);
 int32_t         metaDropTSma(SMeta* pMeta, int64_t indexUid);
 
@@ -116,6 +118,7 @@ void    tqClose(STQ*);
 int     tqPushMsg(STQ*, void* msg, int32_t msgLen, tmsg_t msgType, int64_t ver);
 int     tqCommit(STQ*);
 int32_t tqProcessVgChangeReq(STQ* pTq, char* msg, int32_t msgLen);
+int32_t tqProcessVgDeleteReq(STQ* pTq, char* msg, int32_t msgLen);
 int32_t tqProcessTaskExec(STQ* pTq, char* msg, int32_t msgLen, int32_t workerId);
 int32_t tqProcessTaskDeploy(STQ* pTq, char* msg, int32_t msgLen);
 int32_t tqProcessStreamTrigger(STQ* pTq, void* data, int32_t dataLen, int32_t workerId);
@@ -126,7 +129,7 @@ int32_t smaOpen(SVnode* pVnode);
 int32_t smaClose(SSma* pSma);
 
 int32_t tdUpdateExpireWindow(SSma* pSma, SSubmitReq* pMsg, int64_t version);
-int32_t tdProcessTSmaCreate(SSma* pSma, char* pMsg);
+int32_t tdProcessTSmaCreate(SSma* pSma, int64_t version, const char* msg);
 int32_t tdProcessTSmaInsert(SSma* pSma, int64_t indexUid, const char* msg);
 
 int32_t tdProcessRSmaCreate(SSma* pSma, SMeta* pMeta, SVCreateStbReq* pReq, SMsgCb* pMsgCb);

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

@@ -69,6 +69,11 @@ int metaDecodeEntry(SDecoder *pCoder, SMetaEntry *pME) {
     if (tDecodeI32v(pCoder, &pME->ntbEntry.ncid) < 0) return -1;
     if (tDecodeSSchemaWrapper(pCoder, &pME->ntbEntry.schema) < 0) return -1;
   } else if (pME->type == TSDB_TSMA_TABLE) {
+    pME->smaEntry.tsma = taosMemoryCalloc(1, sizeof(STSma));
+    if(!pME->smaEntry.tsma) {
+      terrno = TSDB_CODE_OUT_OF_MEMORY;
+      return -1;
+    }
     if (tDecodeTSma(pCoder, pME->smaEntry.tsma) < 0) return -1;
   } else {
     ASSERT(0);

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

@@ -225,7 +225,7 @@ SMCtbCursor *metaOpenCtbCursor(SMeta *pMeta, tb_uid_t uid) {
   return pCtbCur;
 }
 
-void metaCloseCtbCurosr(SMCtbCursor *pCtbCur) {
+void metaCloseCtbCursor(SMCtbCursor *pCtbCur) {
   if (pCtbCur) {
     if (pCtbCur->pMeta) metaULock(pCtbCur->pMeta);
     if (pCtbCur->pCur) {
@@ -291,178 +291,268 @@ STSchema *metaGetTbTSchema(SMeta *pMeta, tb_uid_t uid, int32_t sver) {
   return pTSchema;
 }
 
-STSmaWrapper *metaGetSmaInfoByTable(SMeta *pMeta, tb_uid_t uid) {
-#if 0
-#ifdef META_TDB_SMA_TEST
-  STSmaWrapper *pSW = NULL;
-
-  pSW = taosMemoryCalloc(1, sizeof(*pSW));
-  if (pSW == NULL) {
-    return NULL;
-  }
-
-  SMSmaCursor *pCur = metaOpenSmaCursor(pMeta, uid);
-  if (pCur == NULL) {
-    taosMemoryFree(pSW);
-    return NULL;
-  }
-
-  void       *pBuf = NULL;
-  SSmaIdxKey *pSmaIdxKey = NULL;
-
-  while (true) {
-    // TODO: lock during iterate?
-    if (tdbDbcNext(pCur->pCur, &pCur->pKey, &pCur->kLen, NULL, &pCur->vLen) == 0) {
-      pSmaIdxKey = pCur->pKey;
-      ASSERT(pSmaIdxKey != NULL);
-
-      void *pSmaVal = metaGetSmaInfoByIndex(pMeta, pSmaIdxKey->smaUid, false);
-
-      if (pSmaVal == NULL) {
-        tsdbWarn("no tsma exists for indexUid: %" PRIi64, pSmaIdxKey->smaUid);
-        continue;
-      }
-
-      ++pSW->number;
-      STSma *tptr = (STSma *)taosMemoryRealloc(pSW->tSma, pSW->number * sizeof(STSma));
-      if (tptr == NULL) {
-        tdbFree(pSmaVal);
-        metaCloseSmaCursor(pCur);
-        tdDestroyTSmaWrapper(pSW);
-        taosMemoryFreeClear(pSW);
-        return NULL;
-      }
-      pSW->tSma = tptr;
-      pBuf = pSmaVal;
-      if (tDecodeTSma(pBuf, pSW->tSma + pSW->number - 1) == NULL) {
-        tdbFree(pSmaVal);
-        metaCloseSmaCursor(pCur);
-        tdDestroyTSmaWrapper(pSW);
-        taosMemoryFreeClear(pSW);
-        return NULL;
-      }
-      tdbFree(pSmaVal);
-      continue;
-    }
-    break;
-  }
-
-  metaCloseSmaCursor(pCur);
-
-  return pSW;
-
-#endif
-#endif
-  return NULL;
-}
-
 int metaGetTbNum(SMeta *pMeta) {
   // TODO
   // ASSERT(0);
   return 0;
 }
 
-SArray *metaGetSmaTbUids(SMeta *pMeta, bool isDup) {
+typedef struct {
-#if 0
+  SMeta   *pMeta;
-  // TODO
+  TDBC    *pCur;
-  // ASSERT(0); // comment this line to pass CI
+  tb_uid_t uid;
-  // return NULL:
-#ifdef META_TDB_SMA_TEST
-  SArray  *pUids = NULL;
-  SMetaDB *pDB = pMeta->pDB;
   void    *pKey;
+  void    *pVal;
+  int      kLen;
+  int      vLen;
+} SMSmaCursor;
 
-  // TODO: lock?
+SMSmaCursor *metaOpenSmaCursor(SMeta *pMeta, tb_uid_t uid) {
-  SMSmaCursor *pCur = metaOpenSmaCursor(pMeta, 0);
+  SMSmaCursor *pSmaCur = NULL;
-  if (pCur == NULL) {
+  SSmaIdxKey   smaIdxKey;
+  int          ret;
+  int          c;
+
+  pSmaCur = (SMSmaCursor *)taosMemoryCalloc(1, sizeof(*pSmaCur));
+  if (pSmaCur == NULL) {
+    terrno = TSDB_CODE_OUT_OF_MEMORY;
     return NULL;
   }
-  // TODO: lock?
 
-  SSmaIdxKey *pSmaIdxKey = NULL;
+  pSmaCur->pMeta = pMeta;
-  tb_uid_t    uid = 0;
+  pSmaCur->uid = uid;
-  while (true) {
+  metaRLock(pMeta);
-    // TODO: lock during iterate?
-    if (tdbDbcNext(pCur->pCur, &pCur->pKey, &pCur->kLen, NULL, &pCur->vLen) == 0) {
-      ASSERT(pSmaIdxKey != NULL);
-      pSmaIdxKey = pCur->pKey;
 
-      if (pSmaIdxKey->uid == 0 || pSmaIdxKey->uid == uid) {
+  ret = tdbDbcOpen(pMeta->pSmaIdx, &pSmaCur->pCur, NULL);
+  if (ret < 0) {
+    metaULock(pMeta);
+    taosMemoryFree(pSmaCur);
+    return NULL;
+  }
+
+  // move to the suid
+  smaIdxKey.uid = uid;
+  smaIdxKey.smaUid = INT64_MIN;
+  tdbDbcMoveTo(pSmaCur->pCur, &smaIdxKey, sizeof(smaIdxKey), &c);
+  if (c > 0) {
+    tdbDbcMoveToNext(pSmaCur->pCur);
+  }
+
+  return pSmaCur;
+}
+
+void metaCloseSmaCursor(SMSmaCursor *pSmaCur) {
+  if (pSmaCur) {
+    if (pSmaCur->pMeta) metaULock(pSmaCur->pMeta);
+    if (pSmaCur->pCur) {
+      tdbDbcClose(pSmaCur->pCur);
+
+      tdbFree(pSmaCur->pKey);
+      tdbFree(pSmaCur->pVal);
+    }
+
+    taosMemoryFree(pSmaCur);
+  }
+}
+
+tb_uid_t metaSmaCursorNext(SMSmaCursor *pSmaCur) {
+  int         ret;
+  SSmaIdxKey *pSmaIdxKey;
+
+  ret = tdbDbcNext(pSmaCur->pCur, &pSmaCur->pKey, &pSmaCur->kLen, &pSmaCur->pVal, &pSmaCur->vLen);
+  if (ret < 0) {
+    return 0;
+  }
+
+  pSmaIdxKey = pSmaCur->pKey;
+  if (pSmaIdxKey->uid > pSmaCur->uid) {
+    return 0;
+  }
+
+  return pSmaIdxKey->uid;
+}
+
+STSmaWrapper *metaGetSmaInfoByTable(SMeta *pMeta, tb_uid_t uid, bool deepCopy) {
+  STSmaWrapper *pSW = NULL;
+  SArray       *pSmaIds = NULL;
+
+  if (!(pSmaIds = metaGetSmaIdsByTable(pMeta, uid))) {
+    return NULL;
+  }
+
+  pSW = taosMemoryCalloc(1, sizeof(*pSW));
+  if (!pSW) {
+    terrno = TSDB_CODE_OUT_OF_MEMORY;
+    goto _err;
+  }
+
+  pSW->number = taosArrayGetSize(pSmaIds);
+  pSW->tSma = taosMemoryCalloc(pSW->number, sizeof(STSma));
+
+  if (!pSW->tSma) {
+    terrno = TSDB_CODE_OUT_OF_MEMORY;
+    goto _err;
+  }
+
+  SMSmaCursor *pCur = metaOpenSmaCursor(pMeta, uid);
+  if (pCur == NULL) {
+    goto _err;
+  }
+
+  SMetaReader mr = {0};
+  metaReaderInit(&mr, pMeta, 0);
+  int64_t smaId;
+  int     smaIdx = 0;
+  STSma  *pTSma = NULL;
+  for (int i = 0; i < pSW->number; ++i) {
+    smaId = *(tb_uid_t *)taosArrayGet(pSmaIds, i);
+    if (metaGetTableEntryByUid(&mr, smaId) < 0) {
+      metaWarn("vgId:%d no entry for tbId: %" PRIi64 ", smaId: %" PRIi64, TD_VID(pMeta->pVnode), uid, smaId);
       continue;
     }
-      uid = pSmaIdxKey->uid;
+    pTSma = pSW->tSma + smaIdx;
+    memcpy(pTSma, mr.me.smaEntry.tsma, sizeof(STSma));
+    if (deepCopy) {
+      if (pTSma->exprLen > 0) {
+        if (!(pTSma->expr = taosMemoryCalloc(1, pTSma->exprLen))) {
+          terrno = TSDB_CODE_OUT_OF_MEMORY;
+          goto _err;
+        }
+        memcpy((void*)pTSma->expr, mr.me.smaEntry.tsma->expr, pTSma->exprLen);
+      }
+      if (pTSma->tagsFilterLen > 0) {
+        if (!(pTSma->tagsFilter = taosMemoryCalloc(1, pTSma->tagsFilterLen))) {
+          terrno = TSDB_CODE_OUT_OF_MEMORY;
+          goto _err;
+        }
+      }
+      memcpy((void*)pTSma->tagsFilter, mr.me.smaEntry.tsma->tagsFilter, pTSma->tagsFilterLen);
+    } else {
+      pTSma->exprLen = 0;
+      pTSma->expr = NULL;
+      pTSma->tagsFilterLen = 0;
+      pTSma->tagsFilter = NULL;
+    }
+    
+    ++smaIdx;
+  }
+
+  if (smaIdx <= 0) goto _err;
+  pSW->number = smaIdx;
+
+  metaReaderClear(&mr);
+  taosArrayDestroy(pSmaIds);
+  metaCloseSmaCursor(pCur);
+  return pSW;
+_err:
+  metaReaderClear(&mr);
+  taosArrayDestroy(pSmaIds);
+  metaCloseSmaCursor(pCur);
+  tdFreeTSmaWrapper(pSW, deepCopy);
+  return NULL;
+}
+
+STSma *metaGetSmaInfoByIndex(SMeta *pMeta, int64_t indexUid) {
+  STSma      *pTSma = NULL;
+  SMetaReader mr = {0};
+  metaReaderInit(&mr, pMeta, 0);
+  if (metaGetTableEntryByUid(&mr, indexUid) < 0) {
+    metaWarn("vgId:%d failed to get table entry for smaId: %" PRIi64, TD_VID(pMeta->pVnode), indexUid);
+    metaReaderClear(&mr);
+    return NULL;
+  }
+  pTSma = (STSma *)taosMemoryMalloc(sizeof(STSma));
+  if (!pTSma) {
+    terrno = TSDB_CODE_OUT_OF_MEMORY;
+    metaReaderClear(&mr);
+    return NULL;
+  }
+
+  memcpy(pTSma, mr.me.smaEntry.tsma, sizeof(STSma));
+
+  metaReaderClear(&mr);
+  return pTSma;
+}
+
+SArray *metaGetSmaIdsByTable(SMeta *pMeta, tb_uid_t uid) {
+  SArray     *pUids = NULL;
+  SSmaIdxKey *pSmaIdxKey = NULL;
+
+  SMSmaCursor *pCur = metaOpenSmaCursor(pMeta, uid);
+  if (!pCur) {
+    return NULL;
+  }
+
+  while (1) {
+    tb_uid_t id = metaSmaCursorNext(pCur);
+    if (id == 0) {
+      break;
+    }
 
     if (!pUids) {
       pUids = taosArrayInit(16, sizeof(tb_uid_t));
       if (!pUids) {
+        terrno = TSDB_CODE_OUT_OF_MEMORY;
         metaCloseSmaCursor(pCur);
         return NULL;
       }
     }
 
-      taosArrayPush(pUids, &uid);
+    pSmaIdxKey = (SSmaIdxKey *)pCur->pKey;
 
-      continue;
+    if (taosArrayPush(pUids, &pSmaIdxKey->smaUid) < 0) {
+      terrno = TSDB_CODE_OUT_OF_MEMORY;
+      metaCloseSmaCursor(pCur);
+      taosArrayDestroy(pUids);
+      return NULL;
     }
+  }
+
+  metaCloseSmaCursor(pCur);
+  return pUids;
+}
+
+SArray *metaGetSmaTbUids(SMeta *pMeta) {
+  SArray     *pUids = NULL;
+  SSmaIdxKey *pSmaIdxKey = NULL;
+  tb_uid_t    lastUid = 0;
+
+  SMSmaCursor *pCur = metaOpenSmaCursor(pMeta, 0);
+  if (!pCur) {
+    return NULL;
+  }
+
+  while (1) {
+    tb_uid_t uid = metaSmaCursorNext(pCur);
+    if (uid == 0) {
       break;
     }
 
+    if (lastUid == uid) {
+      continue;
+    }
+
+    lastUid = uid;
+
+    if (!pUids) {
+      pUids = taosArrayInit(16, sizeof(tb_uid_t));
+      if (!pUids) {
+        terrno = TSDB_CODE_OUT_OF_MEMORY;
         metaCloseSmaCursor(pCur);
+        return NULL;
+      }
+    }
 
+    if (taosArrayPush(pUids, &uid) < 0) {
+      terrno = TSDB_CODE_OUT_OF_MEMORY;
+      metaCloseSmaCursor(pCur);
+      taosArrayDestroy(pUids);
+      return NULL;
+    }
+  }
+
+  metaCloseSmaCursor(pCur);
   return pUids;
-#endif
-#endif
-  return NULL;
-}
-
-void *metaGetSmaInfoByIndex(SMeta *pMeta, int64_t indexUid, bool isDecode) {
-#if 0
-  // TODO
-  // ASSERT(0);
-  // return NULL;
-#ifdef META_TDB_SMA_TEST
-  SMetaDB *pDB = pMeta->pDB;
-  void    *pKey = NULL;
-  void    *pVal = NULL;
-  int      kLen = 0;
-  int      vLen = 0;
-  int      ret = -1;
-
-  // Set key
-  pKey = (void *)&indexUid;
-  kLen = sizeof(indexUid);
-
-  // Query
-  ret = tdbDbGet(pDB->pSmaDB, pKey, kLen, &pVal, &vLen);
-  if (ret != 0 || !pVal) {
-    return NULL;
-  }
-
-  if (!isDecode) {
-    // return raw value
-    return pVal;
-  }
-
-  // Decode
-  STSma *pCfg = (STSma *)taosMemoryCalloc(1, sizeof(STSma));
-  if (pCfg == NULL) {
-    taosMemoryFree(pVal);
-    return NULL;
-  }
-
-  void *pBuf = pVal;
-  if (tDecodeTSma(pBuf, pCfg) == NULL) {
-    tdDestroyTSma(pCfg);
-    taosMemoryFree(pCfg);
-    tdbFree(pVal);
-    return NULL;
-  }
-
-  tdbFree(pVal);
-  return pCfg;
-#endif
-#endif
-  return NULL;
 }
 
 #endif

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

@@ -81,55 +81,6 @@ int32_t metaDropTSma(SMeta *pMeta, int64_t indexUid) {
   return TSDB_CODE_SUCCESS;
 }
 
-// static int metaSaveSmaToDB(SMeta *pMeta, STSma *pSmaCfg) {
-//   int32_t ret = 0;
-//   void   *pBuf = NULL, *qBuf = NULL;
-//   void   *key = {0}, *val = {0};
-
-//   // save sma info
-//   int32_t len = tEncodeTSma(NULL, pSmaCfg);
-//   pBuf = taosMemoryCalloc(1, len);
-//   if (pBuf == NULL) {
-//     terrno = TSDB_CODE_OUT_OF_MEMORY;
-//     return -1;
-//   }
-
-//   key = (void *)&pSmaCfg->indexUid;
-//   qBuf = pBuf;
-//   tEncodeTSma(&qBuf, pSmaCfg);
-//   val = pBuf;
-
-//   int32_t kLen = sizeof(pSmaCfg->indexUid);
-//   int32_t vLen = POINTER_DISTANCE(qBuf, pBuf);
-
-//   ret = tdbDbInsert(pMeta->pTbDb, key, kLen, val, vLen, &pMeta->txn);
-//   if (ret < 0) {
-//     taosMemoryFreeClear(pBuf);
-//     return -1;
-//   }
-
-//   // add sma idx
-//   SSmaIdxKey smaIdxKey;
-//   smaIdxKey.uid = pSmaCfg->tableUid;
-//   smaIdxKey.smaUid = pSmaCfg->indexUid;
-//   key = &smaIdxKey;
-//   kLen = sizeof(smaIdxKey);
-//   val = NULL;
-//   vLen = 0;
-
-//   ret = tdbDbInsert(pMeta->pSmaIdx, key, kLen, val, vLen, &pMeta->txn);
-//   if (ret < 0) {
-//     taosMemoryFreeClear(pBuf);
-//     return -1;
-//   }
-
-//   // release
-//   taosMemoryFreeClear(pBuf);
-
-//   return 0;
-// }
-
-
 static int metaSaveSmaToDB(SMeta *pMeta, const SMetaEntry *pME) {
   STbDbKey tbDbKey;
   void    *pKey = NULL;
@@ -182,6 +133,10 @@ static int metaUpdateUidIdx(SMeta *pMeta, const SMetaEntry *pME) {
   return tdbDbInsert(pMeta->pUidIdx, &pME->uid, sizeof(tb_uid_t), &pME->version, sizeof(int64_t), &pMeta->txn);
 }
 
+static int metaUpdateNameIdx(SMeta *pMeta, const SMetaEntry *pME) {
+  return tdbDbInsert(pMeta->pNameIdx, pME->name, strlen(pME->name) + 1, &pME->uid, sizeof(tb_uid_t), &pMeta->txn);
+}
+
 static int metaUpdateSmaIdx(SMeta *pMeta, const SMetaEntry *pME) {
   SSmaIdxKey smaIdxKey = {.uid = pME->smaEntry.tsma->tableUid, .smaUid = pME->smaEntry.tsma->indexUid};
 
@@ -194,9 +149,13 @@ static int metaHandleSmaEntry(SMeta *pMeta, const SMetaEntry *pME) {
   // save to table.db
   if (metaSaveSmaToDB(pMeta, pME) < 0) goto _err;
 
-  // // update uid.idx
+  // update uid.idx
   if (metaUpdateUidIdx(pMeta, pME) < 0) goto _err;
 
+  // update name.idx
+  if (metaUpdateNameIdx(pMeta, pME) < 0) goto _err;
+
+  // update sma.idx
   if (metaUpdateSmaIdx(pMeta, pME) < 0) goto _err;
 
   metaULock(pMeta);

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

@@ -420,7 +420,8 @@ static int metaAlterTableColumn(SMeta *pMeta, int64_t version, SVAlterTbReq *pAl
   // get table entry
   SDecoder dc = {0};
   tDecoderInit(&dc, pData, nData);
-  metaDecodeEntry(&dc, &entry);
+  ret = metaDecodeEntry(&dc, &entry);
+  ASSERT(ret == 0);
 
   if (entry.type != TSDB_NORMAL_TABLE) {
     terrno = TSDB_CODE_VND_INVALID_TABLE_ACTION;
@@ -468,11 +469,11 @@ static int metaAlterTableColumn(SMeta *pMeta, int64_t version, SVAlterTbReq *pAl
         goto _err;
       }
       pSchema->sver++;
-      pSchema->nCols--;
       tlen = (pSchema->nCols - iCol - 1) * sizeof(SSchema);
       if (tlen) {
         memmove(pColumn, pColumn + 1, tlen);
       }
+      pSchema->nCols--;
       break;
     case TSDB_ALTER_TABLE_UPDATE_COLUMN_BYTES:
       if (pColumn == NULL) {
@@ -498,6 +499,18 @@ static int metaAlterTableColumn(SMeta *pMeta, int64_t version, SVAlterTbReq *pAl
 
   entry.version = version;
 
+  // do actual write
+  metaWLock(pMeta);
+
+  // save to table db
+  metaSaveToTbDb(pMeta, &entry);
+
+  tdbDbcUpsert(pUidIdxc, &entry.uid, sizeof(tb_uid_t), &version, sizeof(version), 0);
+
+  metaSaveToSkmDb(pMeta, &entry);
+
+  metaULock(pMeta);
+
   tDecoderClear(&dc);
   tdbDbcClose(pTbDbc);
   tdbDbcClose(pUidIdxc);
@@ -511,8 +524,128 @@ _err:
 }
 
 static int metaUpdateTableTagVal(SMeta *pMeta, int64_t version, SVAlterTbReq *pAlterTbReq) {
-  // TODO
+  SMetaEntry  ctbEntry = {0};
+  SMetaEntry  stbEntry = {0};
+  void       *pVal = NULL;
+  int         nVal = 0;
+  int         ret;
+  int         c;
+  tb_uid_t    uid;
+  int64_t     oversion;
+  const void *pData = NULL;
+  int         nData = 0;
+
+  // search name index
+  ret = tdbDbGet(pMeta->pNameIdx, pAlterTbReq->tbName, strlen(pAlterTbReq->tbName) + 1, &pVal, &nVal);
+  if (ret < 0) {
+    terrno = TSDB_CODE_VND_TABLE_NOT_EXIST;
+    return -1;
+  }
+
+  uid = *(tb_uid_t *)pVal;
+  tdbFree(pVal);
+  pVal = NULL;
+
+  // search uid index
+  TDBC *pUidIdxc = NULL;
+
+  tdbDbcOpen(pMeta->pUidIdx, &pUidIdxc, &pMeta->txn);
+  tdbDbcMoveTo(pUidIdxc, &uid, sizeof(uid), &c);
+  ASSERT(c == 0);
+
+  tdbDbcGet(pUidIdxc, NULL, NULL, &pData, &nData);
+  oversion = *(int64_t *)pData;
+
+  // search table.db
+  TDBC    *pTbDbc = NULL;
+  SDecoder dc = {0};
+
+  /* get ctbEntry */
+  tdbDbcOpen(pMeta->pTbDb, &pTbDbc, &pMeta->txn);
+  tdbDbcMoveTo(pTbDbc, &((STbDbKey){.uid = uid, .version = oversion}), sizeof(STbDbKey), &c);
+  ASSERT(c == 0);
+  tdbDbcGet(pTbDbc, NULL, NULL, &pData, &nData);
+
+  ctbEntry.pBuf = taosMemoryMalloc(nData);
+  memcpy(ctbEntry.pBuf, pData, nData);
+  tDecoderInit(&dc, ctbEntry.pBuf, nData);
+  metaDecodeEntry(&dc, &ctbEntry);
+  tDecoderClear(&dc);
+
+  /* get stbEntry*/
+  tdbDbGet(pMeta->pUidIdx, &ctbEntry.ctbEntry.suid, sizeof(tb_uid_t), &pVal, &nVal);
+  tdbDbGet(pMeta->pTbDb, &((STbDbKey){.uid = ctbEntry.ctbEntry.suid, .version = *(int64_t *)pVal}), sizeof(STbDbKey),
+           (void **)&stbEntry.pBuf, &nVal);
+  tdbFree(pVal);
+  tDecoderInit(&dc, stbEntry.pBuf, nVal);
+  metaDecodeEntry(&dc, &stbEntry);
+  tDecoderClear(&dc);
+
+  SSchemaWrapper *pTagSchema = &stbEntry.stbEntry.schemaTag;
+  SSchema        *pColumn = NULL;
+  int32_t         iCol = 0;
+  for (;;) {
+    pColumn = NULL;
+
+    if (iCol >= pTagSchema->nCols) break;
+    pColumn = &pTagSchema->pSchema[iCol];
+
+    if (strcmp(pColumn->name, pAlterTbReq->tagName) == 0) break;
+    iCol++;
+  }
+
+  if (pColumn == NULL) {
+    terrno = TSDB_CODE_VND_TABLE_COL_NOT_EXISTS;
+    goto _err;
+  }
+
+  if (iCol == 0) {
+    // TODO : need to update tag index
+  }
+
+  ctbEntry.version = version;
+  SKVRowBuilder kvrb = {0};
+  const SKVRow  pOldTag = (const SKVRow)ctbEntry.ctbEntry.pTags;
+  SKVRow        pNewTag = NULL;
+
+  tdInitKVRowBuilder(&kvrb);
+  for (int32_t i = 0; i < pTagSchema->nCols; i++) {
+    SSchema *pCol = &pTagSchema->pSchema[i];
+    if (iCol == i) {
+      tdAddColToKVRow(&kvrb, pCol->colId, pAlterTbReq->pTagVal, pAlterTbReq->nTagVal);
+    } else {
+      void *p = tdGetKVRowValOfCol(pOldTag, pCol->colId);
+      if (p) {
+        if (IS_VAR_DATA_TYPE(pCol->type)) {
+          tdAddColToKVRow(&kvrb, pCol->colId, p, varDataTLen(p));
+        } else {
+          tdAddColToKVRow(&kvrb, pCol->colId, p, pCol->bytes);
+        }
+      }
+    }
+  }
+
+  ctbEntry.ctbEntry.pTags = tdGetKVRowFromBuilder(&kvrb);
+  tdDestroyKVRowBuilder(&kvrb);
+
+  // save to table.db
+  metaSaveToTbDb(pMeta, &ctbEntry);
+
+  // save to uid.idx
+  tdbDbUpsert(pMeta->pUidIdx, &ctbEntry.uid, sizeof(tb_uid_t), &version, sizeof(version), &pMeta->txn);
+
+  if (ctbEntry.pBuf) taosMemoryFree(ctbEntry.pBuf);
+  if (stbEntry.pBuf) tdbFree(stbEntry.pBuf);
+  tdbDbcClose(pTbDbc);
+  tdbDbcClose(pUidIdxc);
   return 0;
+
+_err:
+  if (ctbEntry.pBuf) taosMemoryFree(ctbEntry.pBuf);
+  if (stbEntry.pBuf) tdbFree(stbEntry.pBuf);
+  tdbDbcClose(pTbDbc);
+  tdbDbcClose(pUidIdxc);
+  return -1;
 }
 
 static int metaUpdateTableOptions(SMeta *pMeta, int64_t version, SVAlterTbReq *pAlterTbReq) {

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

@@ -27,4 +27,28 @@ int32_t tdProcessTSmaInsert(SSma* pSma, int64_t indexUid, const char* msg) {
   return code;
 }
 
+int32_t tdProcessTSmaCreate(SSma* pSma, int64_t version, const char* msg) {
+  int32_t code = TSDB_CODE_SUCCESS;
 
+  if ((code = tdProcessTSmaCreateImpl(pSma, version, msg)) < 0) {
+    smaWarn("vgId:%d create tsma failed since %s", SMA_VID(pSma), tstrerror(terrno));
+  }
+  // TODO: destroy SSDataBlocks(msg)
+  return code;
+}
+
+int32_t tdUpdateExpireWindow(SSma* pSma, SSubmitReq* pMsg, int64_t version) {
+  int32_t code = TSDB_CODE_SUCCESS;
+  if ((code = tdUpdateExpiredWindowImpl(pSma, pMsg, version)) < 0) {
+    smaWarn("vgId:%d update expired sma window failed since %s", SMA_VID(pSma), tstrerror(terrno));
+  }
+  return code;
+}
+
+int32_t tdGetTSmaData(SSma* pSma, char* pData, int64_t indexUid, TSKEY querySKey, int32_t nMaxResult) {
+  int32_t code = TSDB_CODE_SUCCESS;
+  if ((code = tdGetTSmaDataImpl(pSma, pData, indexUid, querySKey, nMaxResult)) < 0) {
+    smaWarn("vgId:%d get tSma data failed since %s", SMA_VID(pSma), tstrerror(terrno));
+  }
+  return code;
+}

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

@@ -122,12 +122,10 @@ static void poolFree(void *arg, void *ptr) {
 }
 
 int32_t tdInitSma(SSma *pSma) {
-  // tSma
+  int32_t numOfTSma = taosArrayGetSize(metaGetSmaTbUids(SMA_META(pSma)));
-  int32_t numOfTSma = taosArrayGetSize(metaGetSmaTbUids(SMA_META(pSma), false));
   if (numOfTSma > 0) {
     atomic_store_16(&SMA_TSMA_NUM(pSma), (int16_t)numOfTSma);
   }
-  // TODO: rSma
   return TSDB_CODE_SUCCESS;
 }
 

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

@@ -443,7 +443,7 @@ static int32_t tdExecuteRSma(SSma *pSma, const void *pMsg, int32_t inputType, tb
 
   if (inputType == STREAM_DATA_TYPE_SUBMIT_BLOCK) {
     // TODO: use the proper schema instead of 0, and cache STSchema in cache
-    STSchema *pTSchema = metaGetTbTSchema(SMA_META(pSma), suid, 0);
+    STSchema *pTSchema = metaGetTbTSchema(SMA_META(pSma), suid, 1);
     if (!pTSchema) {
       terrno = TSDB_CODE_TDB_IVD_TB_SCHEMA_VERSION;
       return TSDB_CODE_FAILED;

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

@@ -70,15 +70,15 @@ static bool    tdSetAndOpenTSmaFile(STSmaReadH *pReadH, TSKEY *queryKey);
 static int32_t tdInsertTSmaBlocks(STSmaWriteH *pSmaH, void *smaKey, int32_t keyLen, void *pData, int32_t dataLen,
                                     TXN *txn);
 // expired window
-static int32_t  tdUpdateExpiredWindowImpl(SSma *pSma, SSubmitReq *pMsg, int64_t version);
+
+
 static int32_t  tdSetExpiredWindow(SSma *pSma, SHashObj *pItemsHash, int64_t indexUid, int64_t winSKey,
                                      int64_t version);
 static int32_t tdResetExpiredWindow(SSma *pSma, SSmaStat *pStat, int64_t indexUid, TSKEY skey);
 static int32_t tdDropTSmaDataImpl(SSma *pSma, int64_t indexUid);
 
 // read data
-// TODO: This is the basic params, and should wrap the params to a queryHandle.
+
-static int32_t tdGetTSmaDataImpl(SSma *pSma, char *pData, int64_t indexUid, TSKEY querySKey, int32_t nMaxResult);
 
 // implementation
 
@@ -713,7 +713,7 @@ static int32_t tdDropTSmaDataImpl(SSma *pSma, int64_t indexUid) {
  * @param nMaxResult The query invoker should control the nMaxResult need to return to avoid OOM.
  * @return int32_t
  */
-static int32_t tdGetTSmaDataImpl(SSma *pSma, char *pData, int64_t indexUid, TSKEY querySKey, int32_t nMaxResult) {
+int32_t tdGetTSmaDataImpl(SSma *pSma, char *pData, int64_t indexUid, TSKEY querySKey, int32_t nMaxResult) {
   SSmaEnv  *pEnv = atomic_load_ptr(&SMA_TSMA_ENV(pSma));
   SSmaStat *pStat = NULL;
 
@@ -834,35 +834,15 @@ static int32_t tdGetTSmaDataImpl(SSma *pSma, char *pData, int64_t indexUid, TSKE
   return TSDB_CODE_SUCCESS;
 }
 
-int32_t tdProcessTSmaCreate(SSma *pSma, char *pMsg) {
+int32_t tdProcessTSmaCreateImpl(SSma *pSma, int64_t version, const char *pMsg) {
-  #if 0
+  SSmaCfg *pCfg = (SSmaCfg *)pMsg;
-  SSmaCfg vCreateSmaReq = {0};
-  if (!tDeserializeSVCreateTSmaReq(pMsg, &vCreateSmaReq)) {
-    terrno = TSDB_CODE_OUT_OF_MEMORY;
-    smaWarn("vgId:%d tsma create msg received but deserialize failed since %s", SMA_VID(pSma), terrstr(terrno));
-    return -1;
-  }
 
-  smaDebug("vgId:%d tsma create msg %s:%" PRIi64 " for table %" PRIi64 " received", SMA_VID(pSma),
+  if (metaCreateTSma(SMA_META(pSma), version, pCfg) < 0) {
-            vCreateSmaReq.tSma.indexName, vCreateSmaReq.tSma.indexUid, vCreateSmaReq.tSma.tableUid);
-
-  // record current timezone of server side
-  vCreateSmaReq.tSma.timezoneInt = tsTimezone;
-
-  if (metaCreateTSma(SMA_META(pSma), &vCreateSmaReq) < 0) {
-    // TODO: handle error
-    smaWarn("vgId:%d tsma %s:%" PRIi64 " create failed for table %" PRIi64 " since %s", SMA_VID(pSma),
-             vCreateSmaReq.tSma.indexName, vCreateSmaReq.tSma.indexUid, vCreateSmaReq.tSma.tableUid, terrstr(terrno));
-    tdDestroyTSma(&vCreateSmaReq.tSma);
     return -1;
   }
 
   tdTSmaAdd(pSma, 1);
-
+  return 0;
-  tdDestroyTSma(&vCreateSmaReq.tSma);
-  // TODO: return directly or go on follow steps?
-#endif
-  return TSDB_CODE_SUCCESS;
 }
 
 int32_t tdDropTSma(SSma *pSma, char *pMsg) {
@@ -930,7 +910,7 @@ static int32_t tdSetExpiredWindow(SSma *pSma, SHashObj *pItemsHash, int64_t inde
     }
 
     // cache smaMeta
-    STSma *pTSma = metaGetSmaInfoByIndex(SMA_META(pSma), indexUid, true);
+    STSma *pTSma = metaGetSmaInfoByIndex(SMA_META(pSma), indexUid);
     if (!pTSma) {
       terrno = TSDB_CODE_TDB_NO_SMA_INDEX_IN_META;
       taosHashCleanup(pItem->expiredWindows);
@@ -1031,25 +1011,25 @@ int32_t tdUpdateExpiredWindowImpl(SSma *pSma, SSubmitReq *pMsg, int64_t version)
 
     SSubmitBlkIter blkIter = {0};
     if (tInitSubmitBlkIter(&msgIter, pBlock, &blkIter) < 0) {
-      pSW = tdFreeTSmaWrapper(pSW);
+      pSW = tdFreeTSmaWrapper(pSW, false);
       break;
     }
 
     while (true) {
       STSRow *row = tGetSubmitBlkNext(&blkIter);
       if (!row) {
-        tdFreeTSmaWrapper(pSW);
+        pSW = tdFreeTSmaWrapper(pSW, false);
         break;
       }
-      if (!pSW || (pTSma->tableUid != pBlock->suid)) {
+      if (!pSW || (pTSma->tableUid != msgIter.suid)) {
         if (pSW) {
-          pSW = tdFreeTSmaWrapper(pSW);
+          pSW = tdFreeTSmaWrapper(pSW, false);
         }
-        if (!(pSW = metaGetSmaInfoByTable(SMA_META(pSma), pBlock->suid))) {
+        if (!(pSW = metaGetSmaInfoByTable(SMA_META(pSma), msgIter.suid, false))) {
           break;
         }
         if ((pSW->number) <= 0 || !pSW->tSma) {
-          pSW = tdFreeTSmaWrapper(pSW);
+          pSW = tdFreeTSmaWrapper(pSW, false);
           break;
         }
 
@@ -1068,6 +1048,7 @@ int32_t tdUpdateExpiredWindowImpl(SSma *pSma, SSubmitReq *pMsg, int64_t version)
       if (lastWinSKey != winSKey) {
         lastWinSKey = winSKey;
         if (tdSetExpiredWindow(pSma, pItemsHash, pTSma->indexUid, winSKey, version) < 0) {
+          pSW = tdFreeTSmaWrapper(pSW, false);
           tdUnRefSmaStat(pSma, pStat);
           return TSDB_CODE_FAILED;
         }
@@ -1083,21 +1064,3 @@ int32_t tdUpdateExpiredWindowImpl(SSma *pSma, SSubmitReq *pMsg, int64_t version)
   return TSDB_CODE_SUCCESS;
 }
 
-
-int32_t tdUpdateExpireWindow(SSma *pSma, SSubmitReq *pMsg, int64_t version) {
-  int32_t code = TSDB_CODE_SUCCESS;
-  if ((code = tdUpdateExpiredWindowImpl(pSma, pMsg, version)) < 0) {
-    smaWarn("vgId:%d update expired sma window failed since %s", SMA_VID(pSma), tstrerror(terrno));
-  }
-  return code;
-}
-
-int32_t tdGetTSmaData(SSma *pSma, char *pData, int64_t indexUid, TSKEY querySKey, int32_t nMaxResult) {
-  int32_t code = TSDB_CODE_SUCCESS;
-  if ((code = tdGetTSmaDataImpl(pSma, pData, indexUid, querySKey, nMaxResult)) < 0) {
-    smaWarn("vgId:%d get tSma data failed since %s", SMA_VID(pSma), tstrerror(terrno));
-  }
-  return code;
-}
-
-

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

@@ -861,6 +861,14 @@ int32_t tqProcessPollReq(STQ* pTq, SRpcMsg* pMsg, int32_t workerId) {
 }
 #endif
 
+int32_t tqProcessVgDeleteReq(STQ* pTq, char* msg, int32_t msgLen) {
+  SMqVDeleteReq* pReq = (SMqVDeleteReq*)msg;
+
+  int32_t code = taosHashRemove(pTq->execs, pReq->subKey, strlen(pReq->subKey));
+  ASSERT(code == 0);
+  return 0;
+}
+
 // TODO: persist meta into tdb
 int32_t tqProcessVgChangeReq(STQ* pTq, char* msg, int32_t msgLen) {
   SMqRebVgReq req = {0};
@@ -1087,35 +1095,6 @@ int32_t tqProcessStreamTrigger2(STQ* pTq, SSubmitReq* pReq, int64_t ver) {
 }
 
 int32_t tqProcessTaskExec2(STQ* pTq, char* msg, int32_t msgLen) {
-  SStreamTaskExecReq req = {0};
+  //
-  tDecodeSStreamTaskExecReq(msg, &req);
-  int32_t taskId = req.taskId;
-
-  SStreamTask* pTask = taosHashGet(pTq->pStreamTasks, &taskId, sizeof(int32_t));
-  ASSERT(pTask);
-  ASSERT(pTask->inputType == TASK_INPUT_TYPE__DATA_BLOCK);
-
-  // enqueue
-  int32_t inputStatus = streamEnqueueDataBlk(pTask, (SStreamDataBlock*)req.data);
-  if (inputStatus == TASK_INPUT_STATUS__BLOCKED) {
-    // TODO rsp blocked
   return 0;
-  }
-
-  // try exec
-  int8_t execStatus = atomic_val_compare_exchange_8(&pTask->status, TASK_STATUS__IDLE, TASK_STATUS__EXECUTING);
-  if (execStatus == TASK_STATUS__IDLE) {
-    if (streamTaskRun(pTask) < 0) {
-      atomic_store_8(&pTask->status, TASK_STATUS__CLOSING);
-
-      goto FAIL;
-    }
-  } else if (execStatus == TASK_STATUS__EXECUTING) {
-    return 0;
-  }
-
-  // TODO rsp success
-  return 0;
-FAIL:
-  return -1;
 }

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

@@ -2760,7 +2760,7 @@ static int32_t getAllTableList(SMeta* pMeta, uint64_t uid, SArray* list) {
     taosArrayPush(list, &info);
   }
 
-  metaCloseCtbCurosr(pCur);
+  metaCloseCtbCursor(pCur);
   return TSDB_CODE_SUCCESS;
 }
 
@@ -3055,7 +3055,8 @@ static bool loadDataBlockFromTableSeq(STsdbReadHandle* pTsdbReadHandle) {
 bool tsdbNextDataBlock(tsdbReaderT pHandle) {
   STsdbReadHandle* pTsdbReadHandle = (STsdbReadHandle*)pHandle;
 
-  for (int32_t i = 0; i < taosArrayGetSize(pTsdbReadHandle->pColumns); ++i) {
+  size_t numOfCols = taosArrayGetSize(pTsdbReadHandle->pColumns);
+  for (int32_t i = 0; i < numOfCols; ++i) {
     SColumnInfoData* pColInfo = taosArrayGet(pTsdbReadHandle->pColumns, i);
     colInfoDataCleanup(pColInfo, pTsdbReadHandle->outputCapacity);
   }

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

@@ -101,6 +101,11 @@ int vnodeProcessWriteReq(SVnode *pVnode, SRpcMsg *pMsg, int64_t version, SRpcMsg
         // TODO: handle error
       }
       break;
+    case TDMT_VND_MQ_VG_DELETE:
+      if (tqProcessVgDeleteReq(pVnode->pTq, pMsg->pCont, pMsg->contLen) < 0) {
+        // TODO: handle error
+      }
+      break;
     case TDMT_VND_TASK_DEPLOY: {
       if (tqProcessTaskDeploy(pVnode->pTq, POINTER_SHIFT(pMsg->pCont, sizeof(SMsgHead)),
                               pMsg->contLen - sizeof(SMsgHead)) < 0) {
@@ -462,7 +467,11 @@ _exit:
 
 static int vnodeProcessAlterTbReq(SVnode *pVnode, int64_t version, void *pReq, int32_t len, SRpcMsg *pRsp) {
   SVAlterTbReq vAlterTbReq = {0};
+  SVAlterTbRsp vAlterTbRsp = {0};
   SDecoder     dc = {0};
+  int          rcode = 0;
+  int          ret;
+  SEncoder     ec = {0};
 
   pRsp->msgType = TDMT_VND_ALTER_TABLE_RSP;
   pRsp->pCont = NULL;
@@ -473,19 +482,27 @@ static int vnodeProcessAlterTbReq(SVnode *pVnode, int64_t version, void *pReq, i
 
   // decode
   if (tDecodeSVAlterTbReq(&dc, &vAlterTbReq) < 0) {
-    pRsp->code = TSDB_CODE_INVALID_MSG;
+    vAlterTbRsp.code = TSDB_CODE_INVALID_MSG;
     tDecoderClear(&dc);
-    return -1;
+    rcode = -1;
+    goto _exit;
   }
 
   // process
   if (metaAlterTable(pVnode->pMeta, version, &vAlterTbReq) < 0) {
-    pRsp->code = terrno;
+    vAlterTbRsp.code = TSDB_CODE_INVALID_MSG;
     tDecoderClear(&dc);
-    return -1;
+    rcode = -1;
+    goto _exit;
   }
-
   tDecoderClear(&dc);
+
+_exit:
+  tEncodeSize(tEncodeSVAlterTbRsp, &vAlterTbRsp, pRsp->contLen, ret);
+  pRsp->pCont = rpcMallocCont(pRsp->contLen);
+  tEncoderInit(&ec, pRsp->pCont, pRsp->contLen);
+  tEncodeSVAlterTbRsp(&ec, &vAlterTbRsp);
+  tEncoderClear(&ec);
   return 0;
 }
 
@@ -711,15 +728,22 @@ static int vnodeProcessCreateTSmaReq(SVnode *pVnode, int64_t version, void *pReq
     goto _err;
   }
 
-  if (metaCreateTSma(pVnode->pMeta, version, &req) < 0) {
+  // record current timezone of server side
+  req.timezoneInt = tsTimezone;
+
+  if (tdProcessTSmaCreate(pVnode->pSma, version, (const char *)&req) < 0) {
     pRsp->code = terrno;
     goto _err;
   }
 
   tDecoderClear(&coder);
+  vDebug("vgId:%d success to create tsma %s:%" PRIi64 " for table %" PRIi64, TD_VID(pVnode), req.indexName,
+         req.indexUid, req.tableUid);
   return 0;
 
 _err:
   tDecoderClear(&coder);
+  vError("vgId:%d failed to create tsma %s:%" PRIi64 " for table %" PRIi64 " since %s", TD_VID(pVnode), req.indexName,
+         req.indexUid, req.tableUid, terrstr(terrno));
   return -1;
 }

source/libs/executor/inc/executil.h

@@ -139,7 +139,7 @@ typedef struct {
   int32_t colId;
 } SStddevInterResult;
 
-void    initGroupedResultInfo(SGroupResInfo* pGroupResInfo, SHashObj* pHashmap, bool sortGroupResult);
+void    initGroupedResultInfo(SGroupResInfo* pGroupResInfo, SHashObj* pHashmap, int32_t order);
 void    initMultiResInfoFromArrayList(SGroupResInfo* pGroupResInfo, SArray* pArrayList);
 
 void    cleanupGroupResInfo(SGroupResInfo* pGroupResInfo);

source/libs/executor/src/executil.c

@@ -184,7 +184,7 @@ void cleanupGroupResInfo(SGroupResInfo* pGroupResInfo) {
   pGroupResInfo->index     = 0;
 }
 
-static int32_t resultrowCompar1(const void* p1, const void* p2) {
+static int32_t resultrowComparAsc(const void* p1, const void* p2) {
   SResKeyPos* pp1 = *(SResKeyPos**) p1;
   SResKeyPos* pp2 = *(SResKeyPos**) p2;
 
@@ -202,7 +202,11 @@ static int32_t resultrowCompar1(const void* p1, const void* p2) {
   }
 }
 
-void initGroupedResultInfo(SGroupResInfo* pGroupResInfo, SHashObj* pHashmap, bool sortGroupResult) {
+static int32_t resultrowComparDesc(const void* p1, const void* p2) {
+  return resultrowComparAsc(p2, p1);
+}
+
+void initGroupedResultInfo(SGroupResInfo* pGroupResInfo, SHashObj* pHashmap, int32_t order) {
   if (pGroupResInfo->pRows != NULL) {
     taosArrayDestroy(pGroupResInfo->pRows);
   }
@@ -224,8 +228,9 @@ void initGroupedResultInfo(SGroupResInfo* pGroupResInfo, SHashObj* pHashmap, boo
     taosArrayPush(pGroupResInfo->pRows, &p);
   }
 
-  if (sortGroupResult) {
+  if (order == TSDB_ORDER_ASC || order == TSDB_ORDER_DESC) {
-    qsort(pGroupResInfo->pRows->pData, taosArrayGetSize(pGroupResInfo->pRows), POINTER_BYTES, resultrowCompar1);
+    __compar_fn_t fn = (order == TSDB_ORDER_ASC)? resultrowComparAsc:resultrowComparDesc;
+    qsort(pGroupResInfo->pRows->pData, taosArrayGetSize(pGroupResInfo->pRows), POINTER_BYTES, fn);
   }
 
   pGroupResInfo->index = 0;

source/libs/executor/src/executorimpl.c

@@ -3707,7 +3707,7 @@ static int32_t doOpenAggregateOptr(SOperatorInfo* pOperator) {
   finalizeMultiTupleQueryResult(pOperator->numOfExprs, pAggInfo->aggSup.pResultBuf, &pAggInfo->binfo.resultRowInfo,
                                 pAggInfo->binfo.rowCellInfoOffset);
 
-  initGroupedResultInfo(&pAggInfo->groupResInfo, pAggInfo->aggSup.pResultRowHashTable, false);
+  initGroupedResultInfo(&pAggInfo->groupResInfo, pAggInfo->aggSup.pResultRowHashTable, 0);
   OPTR_SET_OPENED(pOperator);
   return TSDB_CODE_SUCCESS;
 }

source/libs/executor/src/groupoperator.c

@@ -314,7 +314,7 @@ static SSDataBlock* hashGroupbyAggregate(SOperatorInfo* pOperator) {
   //  }
 
   blockDataEnsureCapacity(pRes, pOperator->resultInfo.capacity);
-  initGroupedResultInfo(&pInfo->groupResInfo, pInfo->aggSup.pResultRowHashTable, false);
+  initGroupedResultInfo(&pInfo->groupResInfo, pInfo->aggSup.pResultRowHashTable, 0);
 
   while(1) {
     doBuildResultDatablock(pOperator, &pInfo->binfo, &pInfo->groupResInfo, pInfo->aggSup.pResultBuf);

source/libs/executor/src/scanoperator.c

@@ -102,7 +102,7 @@ static void getNextTimeWindow(SInterval* pInterval, STimeWindow* tw, int32_t ord
   tw->ekey -= 1;
 }
 
-static bool overlapWithTimeWindow(SInterval* pInterval, SDataBlockInfo* pBlockInfo) {
+static bool overlapWithTimeWindow(SInterval* pInterval, SDataBlockInfo* pBlockInfo, int32_t order) {
   STimeWindow w = {0};
 
   // 0 by default, which means it is not a interval operator of the upstream operator.
@@ -110,13 +110,7 @@ static bool overlapWithTimeWindow(SInterval* pInterval, SDataBlockInfo* pBlockIn
     return false;
   }
 
-  // todo handle the time range case
+  if (order == TSDB_ORDER_ASC) {
-  TSKEY sk = INT64_MIN;
-  TSKEY ek = INT64_MAX;
-  //  TSKEY sk = MIN(pQueryAttr->window.skey, pQueryAttr->window.ekey);
-  //  TSKEY ek = MAX(pQueryAttr->window.skey, pQueryAttr->window.ekey);
-
-  if (true) {
     getAlignQueryTimeWindow(pInterval, pInterval->precision, pBlockInfo->window.skey, &w);
     assert(w.ekey >= pBlockInfo->window.skey);
 
@@ -124,8 +118,8 @@ static bool overlapWithTimeWindow(SInterval* pInterval, SDataBlockInfo* pBlockIn
       return true;
     }
 
-    while (1) {  // todo handle the desc order scan case
+    while (1) {
-      getNextTimeWindow(pInterval, &w, TSDB_ORDER_ASC);
+      getNextTimeWindow(pInterval, &w, order);
       if (w.skey > pBlockInfo->window.ekey) {
         break;
       }
@@ -136,24 +130,24 @@ static bool overlapWithTimeWindow(SInterval* pInterval, SDataBlockInfo* pBlockIn
       }
     }
   } else {
-    //    getAlignQueryTimeWindow(pQueryAttr, pBlockInfo->window.ekey, sk, ek, &w);
+    getAlignQueryTimeWindow(pInterval, pInterval->precision, pBlockInfo->window.ekey, &w);
-    //    assert(w.skey <= pBlockInfo->window.ekey);
+    assert(w.skey <= pBlockInfo->window.ekey);
-    //
+
-    //    if (w.skey > pBlockInfo->window.skey) {
+    if (w.skey > pBlockInfo->window.skey) {
-    //      return true;
+      return true;
-    //    }
+    }
-    //
+
-    //    while(1) {
+    while(1) {
-    //      getNextTimeWindow(pQueryAttr, &w);
+      getNextTimeWindow(pInterval, &w, order);
-    //      if (w.ekey < pBlockInfo->window.skey) {
+      if (w.ekey < pBlockInfo->window.skey) {
-    //        break;
+        break;
-    //      }
+      }
-    //
+
-    //      assert(w.skey < pBlockInfo->window.skey);
+      assert(w.skey < pBlockInfo->window.skey);
-    //      if (w.ekey < pBlockInfo->window.ekey && w.ekey >= pBlockInfo->window.skey) {
+      if (w.ekey < pBlockInfo->window.ekey && w.ekey >= pBlockInfo->window.skey) {
-    //        return true;
+        return true;
-    //      }
+      }
-    //    }
+    }
   }
 
   return false;
@@ -172,7 +166,8 @@ static int32_t loadDataBlock(SOperatorInfo* pOperator, STableScanInfo* pTableSca
   pCost->totalRows += pBlock->info.rows;
 
   *status = pInfo->dataBlockLoadFlag;
-  if (pTableScanInfo->pFilterNode != NULL || overlapWithTimeWindow(&pTableScanInfo->interval, &pBlock->info)) {
+  if (pTableScanInfo->pFilterNode != NULL ||
+      overlapWithTimeWindow(&pTableScanInfo->interval, &pBlock->info, pTableScanInfo->cond.order)) {
     (*status) = FUNC_DATA_REQUIRED_DATA_LOAD;
   }
 
@@ -188,6 +183,13 @@ static int32_t loadDataBlock(SOperatorInfo* pOperator, STableScanInfo* pTableSca
     qDebug("%s data block skipped, brange:%" PRId64 "-%" PRId64 ", rows:%d", GET_TASKID(pTaskInfo),
            pBlockInfo->window.skey, pBlockInfo->window.ekey, pBlockInfo->rows);
     pCost->skipBlocks += 1;
+
+    // clear all data in pBlock that are set when handing the previous block
+    for(int32_t i = 0; i < pBlockInfo->numOfCols; ++i) {
+      SColumnInfoData* pcol = taosArrayGet(pBlock->pDataBlock, i);
+      pcol->pData = NULL;
+    }
+
     return TSDB_CODE_SUCCESS;
   } else if (*status == FUNC_DATA_REQUIRED_STATIS_LOAD) {
     pCost->loadBlockStatis += 1;
@@ -466,6 +468,7 @@ SOperatorInfo* createTableScanOperatorInfo(STableScanPhysiNode* pTableScanNode,
   }
 
   pInfo->scanInfo = (SScanInfo){.numOfAsc = pTableScanNode->scanSeq[0], .numOfDesc = pTableScanNode->scanSeq[1]};
+//  pInfo->scanInfo = (SScanInfo){.numOfAsc = 0, .numOfDesc = 1}; // for debug purpose
 
   pInfo->readHandle        = *readHandle;
   pInfo->interval          = extractIntervalInfo(pTableScanNode);

source/libs/executor/src/timewindowoperator.c

@@ -54,8 +54,8 @@ static TSKEY getStartTsKey(STimeWindow* win, const TSKEY* tsCols, int32_t rows,
   if (tsCols == NULL) {
     ts = ascQuery ? win->skey : win->ekey;
   } else {
-    int32_t offset = ascQuery ? 0 : rows - 1;
+//    int32_t offset = ascQuery ? 0 : rows - 1;
-    ts = tsCols[offset];
+    ts = tsCols[0];
   }
 
   return ts;
@@ -172,14 +172,22 @@ static FORCE_INLINE int32_t getForwardStepsInBlock(int32_t numOfRows, __block_se
       }
     }
   } else {
-    int32_t end = searchFn((char*)pData, pos + 1, ekey, order);
+    int32_t end = searchFn((char*)&pData[pos], numOfRows - pos, ekey, order);
     if (end >= 0) {
-      forwardStep = pos - end;
+      forwardStep = end;
 
-      if (pData[end] == ekey) {
+      if (pData[end + pos] == ekey) {
         forwardStep += 1;
       }
     }
+//    int32_t end = searchFn((char*)pData, pos + 1, ekey, order);
+//    if (end >= 0) {
+//      forwardStep = pos - end;
+//
+//      if (pData[end] == ekey) {
+//        forwardStep += 1;
+//      }
+//    }
   }
 
   assert(forwardStep >= 0);
@@ -203,17 +211,25 @@ int32_t binarySearchForKey(char* pValue, int num, TSKEY key, int order) {
   if (order == TSDB_ORDER_DESC) {
     // find the first position which is smaller than the key
     while (1) {
-      if (key >= keyList[lastPos]) return lastPos;
+      if (key >= keyList[firstPos]) return firstPos;
-      if (key == keyList[firstPos]) return firstPos;
+      if (key == keyList[lastPos]) return lastPos;
-      if (key < keyList[firstPos]) return firstPos - 1;
+
+      if (key < keyList[lastPos]) {
+        lastPos += 1;
+        if (lastPos >= num) {
+          return -1;
+        } else {
+          return lastPos;
+        }
+      }
 
       numOfRows = lastPos - firstPos + 1;
       midPos = (numOfRows >> 1) + firstPos;
 
       if (key < keyList[midPos]) {
-        lastPos = midPos - 1;
-      } else if (key > keyList[midPos]) {
         firstPos = midPos + 1;
+      } else if (key > keyList[midPos]) {
+        lastPos = midPos - 1;
       } else {
         break;
       }
@@ -273,12 +289,12 @@ int32_t getNumOfRowsInTimeWindow(SDataBlockInfo* pDataBlockInfo, TSKEY* pPrimary
     if (ekey > pDataBlockInfo->window.skey && pPrimaryColumn) {
       num = getForwardStepsInBlock(pDataBlockInfo->rows, searchFn, ekey, startPos, order, pPrimaryColumn);
       if (item != NULL) {
-        item->lastKey = pPrimaryColumn[startPos - (num - 1)] + step;
+        item->lastKey = pPrimaryColumn[startPos + (num - 1)] + step;
       }
     } else {
-      num = startPos + 1;
+      num = pDataBlockInfo->rows - startPos;
       if (item != NULL) {
-        item->lastKey = pDataBlockInfo->window.skey + step;
+        item->lastKey = pDataBlockInfo->window.ekey + step;
       }
     }
   }
@@ -470,20 +486,17 @@ static int32_t getNextQualifiedWindow(SInterval* pInterval, STimeWindow* pNext,
     return -1;
   }
 
-  TSKEY   startKey = ascQuery ? pNext->skey : pNext->ekey;
+  TSKEY   skey = ascQuery ? pNext->skey : pNext->ekey;
   int32_t startPos = 0;
 
   // tumbling time window query, a special case of sliding time window query
   if (pInterval->sliding == pInterval->interval && prevPosition != -1) {
-    int32_t factor = GET_FORWARD_DIRECTION_FACTOR(order);
+    startPos = prevPosition + 1;
-    startPos = prevPosition + factor;
   } else {
-    if (startKey <= pDataBlockInfo->window.skey && ascQuery) {
+    if ((skey <= pDataBlockInfo->window.skey && ascQuery) || (skey >= pDataBlockInfo->window.ekey && !ascQuery)) {
       startPos = 0;
-    } else if (startKey >= pDataBlockInfo->window.ekey && !ascQuery) {
-      startPos = pDataBlockInfo->rows - 1;
     } else {
-      startPos = binarySearchForKey((char*)primaryKeys, pDataBlockInfo->rows, startKey, order);
+      startPos = binarySearchForKey((char*)primaryKeys, pDataBlockInfo->rows, skey, order);
     }
   }
 
@@ -608,7 +621,7 @@ static void saveDataBlockLastRow(char** pRow, SArray* pDataBlock, int32_t rowInd
 }
 
 static SArray* hashIntervalAgg(SOperatorInfo* pOperatorInfo, SResultRowInfo* pResultRowInfo, SSDataBlock* pSDataBlock,
-                               int32_t tableGroupId) {
+                               uint64_t tableGroupId) {
   SIntervalAggOperatorInfo* pInfo = (SIntervalAggOperatorInfo*)pOperatorInfo->info;
 
   SExecTaskInfo* pTaskInfo = pOperatorInfo->pTaskInfo;
@@ -620,7 +633,7 @@ static SArray* hashIntervalAgg(SOperatorInfo* pOperatorInfo, SResultRowInfo* pRe
   }
 
   int32_t step = 1;
-  bool    ascScan = true;
+  bool    ascScan = (pInfo->order == TSDB_ORDER_ASC);
 
   //  int32_t prevIndex = pResultRowInfo->curPos;
 
@@ -630,7 +643,7 @@ static SArray* hashIntervalAgg(SOperatorInfo* pOperatorInfo, SResultRowInfo* pRe
     tsCols = (int64_t*)pColDataInfo->pData;
   }
 
-  int32_t startPos = ascScan ? 0 : (pSDataBlock->info.rows - 1);
+  int32_t startPos = 0;
   TSKEY   ts = getStartTsKey(&pSDataBlock->info.window, tsCols, pSDataBlock->info.rows, ascScan);
 
   STimeWindow win = getActiveTimeWindow(pInfo->aggSup.pResultBuf, pResultRowInfo, ts, &pInfo->interval,
@@ -654,9 +667,10 @@ static SArray* hashIntervalAgg(SOperatorInfo* pOperatorInfo, SResultRowInfo* pRe
   }
 
   int32_t forwardStep = 0;
-  TSKEY   ekey = win.ekey;
+  TSKEY   ekey = ascScan? win.ekey:win.skey;
   forwardStep =
-      getNumOfRowsInTimeWindow(&pSDataBlock->info, tsCols, startPos, ekey, binarySearchForKey, NULL, TSDB_ORDER_ASC);
+      getNumOfRowsInTimeWindow(&pSDataBlock->info, tsCols, startPos, ekey, binarySearchForKey, NULL, pInfo->order);
+  ASSERT(forwardStep > 0);
 
   // prev time window not interpolation yet.
   //  int32_t curIndex = pResultRowInfo->curPos;
@@ -731,9 +745,9 @@ static SArray* hashIntervalAgg(SOperatorInfo* pOperatorInfo, SResultRowInfo* pRe
       taosArrayPush(pUpdated, &pos);
     }
 
-    ekey = nextWin.ekey;  // reviseWindowEkey(pQueryAttr, &nextWin);
+    ekey = ascScan? nextWin.ekey:nextWin.skey;
     forwardStep =
-        getNumOfRowsInTimeWindow(&pSDataBlock->info, tsCols, startPos, ekey, binarySearchForKey, NULL, TSDB_ORDER_ASC);
+        getNumOfRowsInTimeWindow(&pSDataBlock->info, tsCols, startPos, ekey, binarySearchForKey, NULL, pInfo->order);
 
     // window start(end) key interpolation
     doWindowBorderInterpolation(pOperatorInfo, pSDataBlock, pInfo->binfo.pCtx, pResult, &nextWin, startPos, forwardStep,
@@ -761,7 +775,8 @@ static int32_t doOpenIntervalAgg(SOperatorInfo* pOperator) {
   SExecTaskInfo*            pTaskInfo = pOperator->pTaskInfo;
   SIntervalAggOperatorInfo* pInfo = pOperator->info;
 
-  int32_t        order = TSDB_ORDER_ASC;
+  int32_t scanFlag = MAIN_SCAN;
+
   SOperatorInfo* downstream = pOperator->pDownstream[0];
 
   while (1) {
@@ -773,8 +788,10 @@ static int32_t doOpenIntervalAgg(SOperatorInfo* pOperator) {
       break;
     }
 
+    getTableScanInfo(pOperator, &pInfo->order, &scanFlag);
+
     // the pDataBlock are always the same one, no need to call this again
-    setInputDataBlock(pOperator, pInfo->binfo.pCtx, pBlock, order, MAIN_SCAN, true);
+    setInputDataBlock(pOperator, pInfo->binfo.pCtx, pBlock, pInfo->order, scanFlag, true);
     STableQueryInfo* pTableQueryInfo = pInfo->pCurrent;
 
     setIntervalQueryRange(pTableQueryInfo, pBlock->info.window.skey, &pTaskInfo->window);
@@ -800,7 +817,7 @@ static int32_t doOpenIntervalAgg(SOperatorInfo* pOperator) {
   finalizeMultiTupleQueryResult(pOperator->numOfExprs, pInfo->aggSup.pResultBuf, &pInfo->binfo.resultRowInfo,
                                 pInfo->binfo.rowCellInfoOffset);
 
-  initGroupedResultInfo(&pInfo->groupResInfo, pInfo->aggSup.pResultRowHashTable, true);
+  initGroupedResultInfo(&pInfo->groupResInfo, pInfo->aggSup.pResultRowHashTable, pInfo->order);
   OPTR_SET_OPENED(pOperator);
   return TSDB_CODE_SUCCESS;
 }
@@ -945,7 +962,7 @@ static SSDataBlock* doStateWindowAgg(SOperatorInfo* pOperator) {
   finalizeMultiTupleQueryResult(pOperator->numOfExprs, pInfo->aggSup.pResultBuf, &pBInfo->resultRowInfo,
                                 pBInfo->rowCellInfoOffset);
 
-  initGroupedResultInfo(&pInfo->groupResInfo, pInfo->aggSup.pResultRowHashTable, true);
+  initGroupedResultInfo(&pInfo->groupResInfo, pInfo->aggSup.pResultRowHashTable, TSDB_ORDER_ASC);
   blockDataEnsureCapacity(pBInfo->pRes, pOperator->resultInfo.capacity);
   doBuildResultDatablock(pOperator, pBInfo, &pInfo->groupResInfo, pInfo->aggSup.pResultBuf);
   if (pBInfo->pRes->info.rows == 0 || !hasRemainDataInCurrentGroup(&pInfo->groupResInfo)) {
@@ -1070,6 +1087,7 @@ static SSDataBlock* doStreamIntervalAgg(SOperatorInfo* pOperator) {
       doClearWindows(pInfo, pOperator->numOfExprs, pBlock);
       continue;
     }
+    pInfo->order = TSDB_ORDER_ASC;
     pUpdated = hashIntervalAgg(pOperator, &pInfo->binfo.resultRowInfo, pBlock, 0);
   }
 
@@ -1119,7 +1137,6 @@ SOperatorInfo* createIntervalOperatorInfo(SOperatorInfo* downstream, SExprInfo*
 
   pInfo->order = TSDB_ORDER_ASC;
   pInfo->interval = *pInterval;
-  //  pInfo->execModel      = OPTR_EXEC_MODEL_STREAM;
   pInfo->execModel = pTaskInfo->execModel;
   pInfo->win = pTaskInfo->window;
   pInfo->twAggSup = *pTwAggSupp;
@@ -1338,7 +1355,7 @@ static SSDataBlock* doSessionWindowAgg(SOperatorInfo* pOperator) {
   finalizeMultiTupleQueryResult(pOperator->numOfExprs, pInfo->aggSup.pResultBuf, &pBInfo->resultRowInfo,
                                 pBInfo->rowCellInfoOffset);
 
-  initGroupedResultInfo(&pInfo->groupResInfo, pInfo->aggSup.pResultRowHashTable, true);
+  initGroupedResultInfo(&pInfo->groupResInfo, pInfo->aggSup.pResultRowHashTable, TSDB_ORDER_ASC);
   blockDataEnsureCapacity(pBInfo->pRes, pOperator->resultInfo.capacity);
   doBuildResultDatablock(pOperator, pBInfo, &pInfo->groupResInfo, pInfo->aggSup.pResultBuf);
   if (pBInfo->pRes->info.rows == 0 || !hasRemainDataInCurrentGroup(&pInfo->groupResInfo)) {

source/libs/function/inc/builtinsimpl.h

@@ -90,6 +90,10 @@ bool histogramFunctionSetup(SqlFunctionCtx *pCtx, SResultRowEntryInfo* pResultIn
 int32_t histogramFunction(SqlFunctionCtx* pCtx);
 int32_t histogramFinalize(SqlFunctionCtx* pCtx, SSDataBlock* pBlock);
 
+bool getHLLFuncEnv(struct SFunctionNode* pFunc, SFuncExecEnv* pEnv);
+int32_t hllFunction(SqlFunctionCtx* pCtx);
+int32_t hllFinalize(SqlFunctionCtx* pCtx, SSDataBlock* pBlock);
+
 bool getStateFuncEnv(struct SFunctionNode* pFunc, SFuncExecEnv* pEnv);
 bool stateFunctionSetup(SqlFunctionCtx *pCtx, SResultRowEntryInfo* pResultInfo);
 int32_t stateCountFunction(SqlFunctionCtx* pCtx);

source/libs/function/src/builtins.c

@@ -263,6 +263,21 @@ static int32_t translateHistogram(SFunctionNode* pFunc, char* pErrBuf, int32_t l
   return TSDB_CODE_SUCCESS;
 }
 
+static int32_t translateHLL(SFunctionNode* pFunc, char* pErrBuf, int32_t len) {
+  if (1 != LIST_LENGTH(pFunc->pParameterList)) {
+    return invaildFuncParaNumErrMsg(pErrBuf, len, pFunc->functionName);
+  }
+
+  SNode* pPara = nodesListGetNode(pFunc->pParameterList, 0);
+  if (QUERY_NODE_COLUMN != nodeType(pPara)) {
+    return buildFuncErrMsg(pErrBuf, len, TSDB_CODE_FUNC_FUNTION_ERROR,
+                           "The input parameter of HYPERLOGLOG function can only be column");
+  }
+
+  pFunc->node.resType = (SDataType){.bytes = tDataTypes[TSDB_DATA_TYPE_UBIGINT].bytes, .type = TSDB_DATA_TYPE_UBIGINT};
+  return TSDB_CODE_SUCCESS;
+}
+
 static int32_t translateStateCount(SFunctionNode* pFunc, char* pErrBuf, int32_t len) {
   if (3 != LIST_LENGTH(pFunc->pParameterList)) {
     return invaildFuncParaNumErrMsg(pErrBuf, len, pFunc->functionName);
@@ -829,6 +844,16 @@ const SBuiltinFuncDefinition funcMgtBuiltins[] = {
     .processFunc  = histogramFunction,
     .finalizeFunc = histogramFinalize
   },
+  {
+    .name = "hyperloglog",
+    .type = FUNCTION_TYPE_HYPERLOGLOG,
+    .classification = FUNC_MGT_AGG_FUNC,
+    .translateFunc = translateHLL,
+    .getEnvFunc   = getHLLFuncEnv,
+    .initFunc     = functionSetup,
+    .processFunc  = hllFunction,
+    .finalizeFunc = hllFinalize
+  },
   {
     .name = "state_count",
     .type = FUNCTION_TYPE_STATE_COUNT,

source/libs/function/src/builtinsimpl.c

@@ -28,6 +28,12 @@
 #define TAIL_MAX_POINTS_NUM      100
 #define TAIL_MAX_OFFSET          100
 
+#define HLL_BUCKET_BITS 14 // The bits of the bucket
+#define HLL_DATA_BITS (64-HLL_BUCKET_BITS)
+#define HLL_BUCKETS (1<<HLL_BUCKET_BITS)
+#define HLL_BUCKET_MASK (HLL_BUCKETS-1)
+#define HLL_ALPHA_INF 0.721347520444481703680 // constant for 0.5/ln(2)
+
 typedef struct SSumRes {
   union {
     int64_t  isum;
@@ -129,6 +135,11 @@ typedef enum {
   LOG_BIN
 } EHistoBinType;
 
+typedef struct SHLLFuncInfo {
+  uint64_t result;
+  uint8_t buckets[HLL_BUCKETS];
+} SHLLInfo;
+
 typedef struct SStateInfo {
   union {
     int64_t count;
@@ -2729,6 +2740,140 @@ int32_t histogramFinalize(SqlFunctionCtx* pCtx, SSDataBlock* pBlock) {
   return pResInfo->numOfRes;
 }
 
+bool getHLLFuncEnv(SFunctionNode* UNUSED_PARAM(pFunc), SFuncExecEnv* pEnv) {
+  pEnv->calcMemSize = sizeof(SHLLInfo);
+  return true;
+}
+
+static uint8_t hllCountNum(void* data, int32_t bytes, int32_t *buk) {
+  uint64_t hash = MurmurHash3_64(data, bytes);
+  int32_t index = hash & HLL_BUCKET_MASK;
+  hash >>= HLL_BUCKET_BITS;
+  hash |= ((uint64_t)1 << HLL_DATA_BITS);
+  uint64_t bit = 1;
+  uint8_t count = 1;
+  while((hash & bit) == 0) {
+    count++;
+    bit <<= 1;
+  }
+  *buk = index;
+  return count;
+}
+
+static void hllBucketHisto(uint8_t *buckets, int32_t* bucketHisto) {
+  uint64_t *word = (uint64_t*) buckets;
+  uint8_t *bytes;
+
+  for (int32_t j = 0; j < HLL_BUCKETS>>3; j++) {
+    if (*word == 0) {
+      bucketHisto[0] += 8;
+    } else {
+      bytes = (uint8_t*) word;
+      bucketHisto[bytes[0]]++;
+      bucketHisto[bytes[1]]++;
+      bucketHisto[bytes[2]]++;
+      bucketHisto[bytes[3]]++;
+      bucketHisto[bytes[4]]++;
+      bucketHisto[bytes[5]]++;
+      bucketHisto[bytes[6]]++;
+      bucketHisto[bytes[7]]++;
+    }
+    word++;
+  }
+}
+static double hllTau(double x) {
+  if (x == 0. || x == 1.) return 0.;
+  double zPrime;
+  double y = 1.0;
+  double z = 1 - x;
+  do {
+    x = sqrt(x);
+    zPrime = z;
+    y *= 0.5;
+    z -= pow(1 - x, 2)*y;
+  } while(zPrime != z);
+  return z / 3;
+}
+
+static double hllSigma(double x) {
+  if (x == 1.0) return INFINITY;
+  double zPrime;
+  double y = 1;
+  double z = x;
+  do {
+    x *= x;
+    zPrime = z;
+    z += x * y;
+    y += y;
+  } while(zPrime != z);
+  return z;
+}
+
+// estimate the cardinality, the algorithm refer this paper: "New cardinality estimation algorithms for HyperLogLog sketches"
+static uint64_t hllCountCnt(uint8_t *buckets) {
+  double m = HLL_BUCKETS;
+  int32_t buckethisto[64] = {0};
+  hllBucketHisto(buckets,buckethisto);
+
+  double z = m * hllTau((m-buckethisto[HLL_DATA_BITS+1])/(double)m);
+  for (int j = HLL_DATA_BITS; j >= 1; --j) {
+    z += buckethisto[j];
+    z *= 0.5;
+  }
+  z += m * hllSigma(buckethisto[0]/(double)m);
+  double E = (double)llroundl(HLL_ALPHA_INF*m*m/z);
+
+  return (uint64_t) E;
+}
+
+
+int32_t hllFunction(SqlFunctionCtx *pCtx) {
+  SHLLInfo* pInfo = GET_ROWCELL_INTERBUF(GET_RES_INFO(pCtx));
+
+  SInputColumnInfoData* pInput = &pCtx->input;
+  SColumnInfoData*      pCol = pInput->pData[0];
+
+  int32_t type  = pCol->info.type;
+  int32_t bytes = pCol->info.bytes;
+
+  int32_t start = pInput->startRowIndex;
+  int32_t numOfRows = pInput->numOfRows;
+
+  int32_t numOfElems = 0;
+  for (int32_t i = start; i < numOfRows + start; ++i) {
+    if (pCol->hasNull && colDataIsNull_s(pCol, i)) {
+      continue;
+    }
+
+    numOfElems++;
+
+    char* data = colDataGetData(pCol, i);
+    if (IS_VAR_DATA_TYPE(type)) {
+      bytes = varDataLen(data);
+      data = varDataVal(data);
+    }
+
+    int32_t index = 0;
+    uint8_t count = hllCountNum(data, bytes, &index);
+    uint8_t oldcount = pInfo->buckets[index];
+    if (count > oldcount) {
+      pInfo->buckets[index] = count;
+    }
+
+  }
+
+  SET_VAL(GET_RES_INFO(pCtx), numOfElems, 1);
+  return TSDB_CODE_SUCCESS;
+}
+
+int32_t hllFinalize(SqlFunctionCtx* pCtx, SSDataBlock* pBlock) {
+  SHLLInfo* pInfo = GET_ROWCELL_INTERBUF(GET_RES_INFO(pCtx));
+
+  pInfo->result = hllCountCnt(pInfo->buckets);
+
+  return functionFinalize(pCtx, pBlock);
+}
+
 bool getStateFuncEnv(SFunctionNode* UNUSED_PARAM(pFunc), SFuncExecEnv* pEnv) {
   pEnv->calcMemSize = sizeof(SStateInfo);
   return true;

source/libs/function/src/tudf.c

@@ -795,7 +795,6 @@ int32_t convertScalarParamToDataBlock(SScalarParam *input, int32_t numOfCols, SS
   }
   output->info.hasVarCol = hasVarCol;
 
-  //TODO: free the array output->pDataBlock
   output->pDataBlock = taosArrayInit(numOfCols, sizeof(SColumnInfoData));
   for (int32_t i = 0; i < numOfCols; ++i) {
     taosArrayPush(output->pDataBlock, (input + i)->columnData);
@@ -809,8 +808,12 @@ int32_t convertDataBlockToScalarParm(SSDataBlock *input, SScalarParam *output) {
     return -1;
   }
   output->numOfRows = input->info.rows;
-  //TODO: memory
+
-  output->columnData = taosArrayGet(input->pDataBlock, 0);
+  output->columnData = taosMemoryMalloc(sizeof(SColumnInfoData));
+  memcpy(output->columnData,
+         taosArrayGet(input->pDataBlock, 0),
+         sizeof(SColumnInfoData));
+
   return 0;
 }
 
@@ -833,7 +836,7 @@ int32_t udfcGetUdfTaskResultFromUvTask(SClientUdfTask *task, SClientUvTaskNode *
   fnDebug("udfc get uv task result. task: %p, uvTask: %p", task, uvTask);
   if (uvTask->type == UV_TASK_REQ_RSP) {
     if (uvTask->rspBuf.base != NULL) {
-      SUdfResponse rsp;
+      SUdfResponse rsp = {0};
       void* buf = decodeUdfResponse(uvTask->rspBuf.base, &rsp);
       assert(uvTask->rspBuf.len == POINTER_DISTANCE(buf, uvTask->rspBuf.base));
       task->errCode = rsp.code;
@@ -1427,7 +1430,10 @@ int32_t doCallUdfScalarFunc(UdfcFuncHandle handle, SScalarParam *input, int32_t
   int32_t err = callUdf(handle, callType, &inputBlock, NULL, NULL, &resultBlock, NULL);
   if (err == 0) {
     convertDataBlockToScalarParm(&resultBlock, output);
+    taosArrayDestroy(resultBlock.pDataBlock);
   }
+
+  taosArrayDestroy(inputBlock.pDataBlock);
   return err;
 }
 
@@ -1508,16 +1514,15 @@ int32_t doTeardownUdf(UdfcFuncHandle handle) {
 
   udfcRunUdfUvTask(task, UV_TASK_REQ_RSP);
 
-  SUdfTeardownResponse *rsp = &task->_teardown.rsp;
   int32_t err = task->errCode;
 
   udfcRunUdfUvTask(task, UV_TASK_DISCONNECT);
 
+  fnInfo("tear down udf. udf name: %s, udf func handle: %p", session->udfName, handle);
+
   taosMemoryFree(task->session);
   taosMemoryFree(task);
 
-  fnInfo("tear down udf. udf name: %s, udf func handle: %p", session->udfName, handle);
-
   return err;
 }
 
@@ -1564,6 +1569,7 @@ bool udfAggInit(struct SqlFunctionCtx *pCtx, struct SResultRowEntryInfo* pResult
   }
   udfRes->interResNum = buf.numOfResult;
   memcpy(udfRes->interResBuf, buf.buf, buf.bufLen);
+  freeUdfInterBuf(&buf);
   return true;
 }
 
@@ -1621,7 +1627,7 @@ int32_t udfAggProcess(struct SqlFunctionCtx *pCtx) {
   blockDataDestroy(inputBlock);
   taosArrayDestroy(tempBlock.pDataBlock);
 
-  taosMemoryFree(newState.buf);
+  freeUdfInterBuf(&newState);
   return udfCode;
 }
 
@@ -1650,6 +1656,8 @@ int32_t udfAggFinalize(struct SqlFunctionCtx *pCtx, SSDataBlock* pBlock) {
     GET_RES_INFO(pCtx)->numOfRes = udfRes->finalResNum;
   }
 
+  freeUdfInterBuf(&resultBuf);
+
   int32_t numOfResults = functionFinalizeWithResultBuf(pCtx, pBlock, udfRes->finalResBuf);
   releaseUdfFuncHandle(pCtx->udfName);
   return udfCallCode == 0 ? numOfResults : udfCallCode;

source/libs/function/src/udfd.c

@@ -96,10 +96,14 @@ int32_t udfdFillUdfInfoFromMNode(void *clientRpc, char *udfName, SUdf *udf);
 
 int32_t udfdLoadUdf(char *udfName, SUdf *udf) {
   strcpy(udf->name, udfName);
+  int32_t err = 0;
+  err = udfdFillUdfInfoFromMNode(global.clientRpc, udf->name, udf);
+  if (err != 0) {
+    fnError("can not retrieve udf from mnode. udf name %s", udfName);
+    return TSDB_CODE_UDF_LOAD_UDF_FAILURE;
+  }
 
-  udfdFillUdfInfoFromMNode(global.clientRpc, udf->name, udf);
+  err = uv_dlopen(udf->path, &udf->lib);
-  //strcpy(udf->path, "/home/slzhou/TDengine/debug/build/lib/libudf1.so");
-  int err = uv_dlopen(udf->path, &udf->lib);
   if (err != 0) {
     fnError("can not load library %s. error: %s", udf->path, uv_strerror(err));
     return TSDB_CODE_UDF_LOAD_UDF_FAILURE;
@@ -142,7 +146,7 @@ int32_t udfdLoadUdf(char *udfName, SUdf *udf) {
 
 void udfdProcessSetupRequest(SUvUdfWork* uvUdf, SUdfRequest* request) {
     // TODO: tracable id from client. connect, setup, call, teardown
-    fnInfo("%" PRId64 " setup request. udf name: %s", request->seqNum, request->setup.udfName);
+    fnInfo( "setup request. seq num: %" PRId64 ", udf name: %s", request->seqNum, request->setup.udfName);
     SUdfSetupRequest *setup = &request->setup;
     int32_t code = TSDB_CODE_SUCCESS;
     SUdf *udf = NULL;
@@ -276,7 +280,7 @@ void udfdProcessCallRequest(SUvUdfWork *uvUdf, SUdfRequest *request) {
 
 void udfdProcessTeardownRequest(SUvUdfWork* uvUdf, SUdfRequest* request) {
   SUdfTeardownRequest *teardown = &request->teardown;
-  fnInfo("teardown. %" PRId64 "handle:%" PRIx64, request->seqNum, teardown->udfHandle);
+  fnInfo("teardown. seq number: %" PRId64 ", handle:%" PRIx64, request->seqNum, teardown->udfHandle);
   SUdfcFuncHandle *handle = (SUdfcFuncHandle *)(teardown->udfHandle);
   SUdf *udf = handle->udf;
   bool  unloadUdf = false;
@@ -800,17 +804,11 @@ static int32_t udfdRun() {
   global.udfsHash = taosHashInit(64, taosGetDefaultHashFunction(TSDB_DATA_TYPE_BINARY), true, HASH_NO_LOCK);
   uv_mutex_init(&global.udfsMutex);
 
-  if (udfdUvInit() != 0) {
-    fnError("uv init failure");
-    return -2;
-  }
-
   fnInfo("start the udfd");
   int code = uv_run(global.loop, UV_RUN_DEFAULT);
   fnInfo("udfd stopped. result: %s, code: %d", uv_err_name(code), code);
   int codeClose = uv_loop_close(global.loop);
   fnDebug("uv loop close. result: %s", uv_err_name(codeClose));
-  removeListeningPipe();
   uv_mutex_destroy(&global.udfsMutex);
   taosHashCleanup(global.udfsHash);
   return 0;
@@ -853,8 +851,14 @@ int main(int argc, char *argv[]) {
     return -4;
   }
 
+  if (udfdUvInit() != 0) {
+    fnError("uv init failure");
+    return -5;
+  }
+
   udfdRun();
 
-  udfdCloseClientRpc();
+  removeListeningPipe();
 
+  udfdCloseClientRpc();
 }

source/libs/function/test/runUdf.c

@@ -34,20 +34,13 @@ static int32_t initLog() {
   return taosCreateLog(logName, 1, configDir, NULL, NULL, NULL, NULL, 0);
 }
 
-int main(int argc, char *argv[]) {
+int scalarFuncTest() {
-  parseArgs(argc, argv);
-  initLog();
-  if (taosInitCfg(configDir, NULL, NULL, NULL, NULL, 0) != 0) {
-    fnError("failed to start since read config error");
-    return -1;
-  }
-
-  udfcOpen();
-  uv_sleep(1000);
-
   UdfcFuncHandle handle;
 
-  doSetupUdf("udf1", &handle);
+  if (doSetupUdf("udf1", &handle) != 0) {
+    fnError("setup udf failure");
+    return -1;
+  }
 
   SSDataBlock  block = {0};
   SSDataBlock *pBlock = █
@@ -74,11 +67,78 @@ int main(int argc, char *argv[]) {
   input.columnData = taosArrayGet(pBlock->pDataBlock, 0);
   SScalarParam output = {0};
   doCallUdfScalarFunc(handle, &input, 1, &output);
-
+  taosArrayDestroy(pBlock->pDataBlock);
   SColumnInfoData *col = output.columnData;
   for (int32_t i = 0; i < output.numOfRows; ++i) {
     fprintf(stderr, "%d\t%d\n", i, *(int32_t *)(col->pData + i * sizeof(int32_t)));
   }
+
+  colDataDestroy(output.columnData);
+  taosMemoryFree(output.columnData);
+
   doTeardownUdf(handle);
+
+  return 0;
+}
+
+int aggregateFuncTest() {
+  UdfcFuncHandle handle;
+
+  if (doSetupUdf("udf2", &handle) != 0) {
+    fnError("setup udf failure");
+    return -1;
+  }
+
+  SSDataBlock  block = {0};
+  SSDataBlock *pBlock = &block;
+  pBlock->pDataBlock = taosArrayInit(1, sizeof(SColumnInfoData));
+  pBlock->info.numOfCols = 1;
+  pBlock->info.rows = 4;
+  char data[16] = {0};
+  char bitmap[4] = {0};
+  for (int32_t i = 0; i < pBlock->info.numOfCols; ++i) {
+    SColumnInfoData colInfo = {0};
+    colInfo.info.type = TSDB_DATA_TYPE_INT;
+    colInfo.info.bytes = sizeof(int32_t);
+    colInfo.info.colId = 1;
+    colInfo.pData = data;
+    colInfo.nullbitmap = bitmap;
+    for (int32_t j = 0; j < pBlock->info.rows; ++j) {
+      colDataAppendInt32(&colInfo, j, &j);
+    }
+    taosArrayPush(pBlock->pDataBlock, &colInfo);
+  }
+
+  SUdfInterBuf buf = {0};
+  SUdfInterBuf newBuf = {0};
+  SUdfInterBuf resultBuf = {0};
+  doCallUdfAggInit(handle, &buf);
+  doCallUdfAggProcess(handle, pBlock, &buf, &newBuf);
+  taosArrayDestroy(pBlock->pDataBlock);
+
+  doCallUdfAggFinalize(handle, &newBuf, &resultBuf);
+  fprintf(stderr, "agg result: %f\n", *(double*)resultBuf.buf);
+
+  freeUdfInterBuf(&buf);
+  freeUdfInterBuf(&newBuf);
+  freeUdfInterBuf(&resultBuf);
+  doTeardownUdf(handle);
+
+  return 0;
+}
+
+int main(int argc, char *argv[]) {
+  parseArgs(argc, argv);
+  initLog();
+  if (taosInitCfg(configDir, NULL, NULL, NULL, NULL, 0) != 0) {
+    fnError("failed to start since read config error");
+    return -1;
+  }
+
+  udfcOpen();
+  uv_sleep(1000);
+
+  scalarFuncTest();
+  aggregateFuncTest();
   udfcClose();
 }

source/libs/index/inc/indexComm.h

@@ -33,11 +33,17 @@ typedef enum { MATCH, CONTINUE, BREAK } TExeCond;
 
 typedef TExeCond (*_cache_range_compare)(void* a, void* b, int8_t type);
 
-TExeCond tDoCommpare(__compar_fn_t func, int8_t comType, void* a, void* b);
+TExeCond tCompare(__compar_fn_t func, int8_t cmpType, void* a, void* b, int8_t dType);
+TExeCond tDoCompare(__compar_fn_t func, int8_t cmpType, void* a, void* b);
 
 _cache_range_compare indexGetCompare(RangeType ty);
 
 int32_t indexConvertData(void* src, int8_t type, void** dst);
+int32_t indexConvertDataToStr(void* src, int8_t type, void** dst);
+
+int32_t indexGetDataByteLen(int8_t type);
+
+char* indexInt2str(int64_t val, char* dst, int radix);
 
 #ifdef __cplusplus
 }

source/libs/index/src/index.c

@@ -109,17 +109,15 @@ int indexOpen(SIndexOpts* opts, const char* path, SIndex** index) {
   taosThreadMutexInit(&sIdx->mtx, NULL);
 
   sIdx->refId = indexAddRef(sIdx);
-  taosAcquireRef(indexRefMgt, sIdx->refId);
+  indexAcquireRef(sIdx->refId);
 
   *index = sIdx;
-
   return 0;
 
 END:
   if (sIdx != NULL) {
     indexClose(sIdx);
   }
-
   *index = NULL;
   return -1;
 }
@@ -273,7 +271,7 @@ SIndexTerm* indexTermCreate(int64_t suid, SIndexOperOnColumn oper, uint8_t colTy
   tm->nColName = nColName;
 
   char*   buf = NULL;
-  int32_t len = indexConvertData((void*)colVal, INDEX_TYPE_GET_TYPE(colType), (void**)&buf);
+  int32_t len = indexConvertDataToStr((void*)colVal, INDEX_TYPE_GET_TYPE(colType), (void**)&buf);
   assert(len != -1);
 
   tm->colVal = buf;

source/libs/index/src/indexCache.c

@@ -282,8 +282,10 @@ static int32_t cacheSearchCompareFunc_JSON(void* cache, SIndexTerm* term, SIdxTe
     if (0 != strncmp(c->colVal, pCt->colVal, skip)) {
       break;
     }
+    char* p = taosMemoryCalloc(1, strlen(c->colVal) + 1);
+    memcpy(p, c->colVal, strlen(c->colVal));
 
-    TExeCond cond = cmpFn(c->colVal + skip, term->colVal, dType);
+    TExeCond cond = cmpFn(p + skip, term->colVal, dType);
     if (cond == MATCH) {
       if (c->operaType == ADD_VALUE) {
         INDEX_MERGE_ADD_DEL(tr->deled, tr->added, c->uid)
@@ -297,6 +299,7 @@ static int32_t cacheSearchCompareFunc_JSON(void* cache, SIndexTerm* term, SIdxTe
     } else if (cond == BREAK) {
       break;
     }
+    taosMemoryFree(p);
   }
 
   taosMemoryFree(pCt);
@@ -463,7 +466,6 @@ int indexCacheSchedToMerge(IndexCache* pCache) {
   // schedMsg.thandle = taosMemoryCalloc(1, sizeof(int64_t));
   // memcpy((char*)(schedMsg.thandle), (char*)&(pCache->index->refId), sizeof(int64_t));
   schedMsg.msg = NULL;
-
   indexAcquireRef(pCache->index->refId);
   taosScheduleTask(indexQhandle, &schedMsg);
 

source/libs/index/src/indexComm.c

@@ -19,10 +19,38 @@
 #include "tcoding.h"
 #include "tcompare.h"
 #include "tdataformat.h"
+#include "ttypes.h"
 
 char JSON_COLUMN[] = "JSON";
 char JSON_VALUE_DELIM = '&';
 
+char* indexInt2str(int64_t val, char* dst, int radix) {
+  char     buffer[65];
+  char*    p;
+  int64_t  new_val;
+  uint64_t uval = (uint64_t)val;
+
+  if (radix < 0) {
+    if (val < 0) {
+      *dst++ = '-';
+      uval = (uint64_t)0 - uval; /* Avoid integer overflow in (-val) for LLONG_MIN (BUG#31799). */
+    }
+  }
+  p = &buffer[sizeof(buffer) - 1];
+  *p = '\0';
+  new_val = (int64_t)(uval / 10);
+  *--p = '0' + (char)(uval - (uint64_t)new_val * 10);
+  val = new_val;
+
+  while (val != 0) {
+    new_val = val / 10;
+    *--p = '0' + (char)(val - new_val * 10);
+    val = new_val;
+  }
+  while ((*dst++ = *p++) != 0)
+    ;
+  return dst - 1;
+}
 static __compar_fn_t indexGetCompar(int8_t type) {
   if (type == TSDB_DATA_TYPE_BINARY || type == TSDB_DATA_TYPE_NCHAR) {
     return (__compar_fn_t)strcmp;
@@ -31,25 +59,49 @@ static __compar_fn_t indexGetCompar(int8_t type) {
 }
 static TExeCond tCompareLessThan(void* a, void* b, int8_t type) {
   __compar_fn_t func = indexGetCompar(type);
-  return tDoCommpare(func, QUERY_LESS_THAN, a, b);
+  return tCompare(func, QUERY_LESS_THAN, a, b, type);
 }
 static TExeCond tCompareLessEqual(void* a, void* b, int8_t type) {
   __compar_fn_t func = indexGetCompar(type);
-  return tDoCommpare(func, QUERY_LESS_EQUAL, a, b);
+  return tCompare(func, QUERY_LESS_EQUAL, a, b, type);
 }
 static TExeCond tCompareGreaterThan(void* a, void* b, int8_t type) {
   __compar_fn_t func = indexGetCompar(type);
-  return tDoCommpare(func, QUERY_GREATER_THAN, a, b);
+  return tCompare(func, QUERY_GREATER_THAN, a, b, type);
 }
 static TExeCond tCompareGreaterEqual(void* a, void* b, int8_t type) {
   __compar_fn_t func = indexGetCompar(type);
-  return tDoCommpare(func, QUERY_GREATER_EQUAL, a, b);
+  return tCompare(func, QUERY_GREATER_EQUAL, a, b, type);
 }
-
+TExeCond tCompare(__compar_fn_t func, int8_t cmptype, void* a, void* b, int8_t dtype) {
-TExeCond tDoCommpare(__compar_fn_t func, int8_t comType, void* a, void* b) {
+  if (dtype == TSDB_DATA_TYPE_BINARY || dtype == TSDB_DATA_TYPE_NCHAR) {
+    return tDoCompare(func, cmptype, a, b);
+  }
+#if 1
+  int8_t bytes = tDataTypes[dtype].bytes;
+  if (bytes == 1) {
+    int8_t va = taosStr2int64(a);
+    int8_t vb = taosStr2int64(b);
+    return tDoCompare(func, cmptype, &va, &vb);
+  } else if (bytes == 2) {
+    int16_t va = taosStr2int64(a);
+    int16_t vb = taosStr2int64(b);
+    return tDoCompare(func, cmptype, &va, &vb);
+  } else if (bytes == 4) {
+    int32_t va = taosStr2int64(a);
+    int32_t vb = taosStr2int64(b);
+    return tDoCompare(func, cmptype, &va, &vb);
+  } else {
+    int64_t va = taosStr2int64(a);
+    int64_t vb = taosStr2int64(b);
+    return tDoCompare(func, cmptype, &va, &vb);
+  }
+#endif
+}
+TExeCond tDoCompare(__compar_fn_t func, int8_t comparType, void* a, void* b) {
   // optime later
   int32_t ret = func(a, b);
-  switch (comType) {
+  switch (comparType) {
     case QUERY_LESS_THAN: {
       if (ret < 0) return MATCH;
     } break;
@@ -174,9 +226,9 @@ int32_t indexConvertData(void* src, int8_t type, void** dst) {
       tlen = taosEncodeFixedU32(dst, *(uint32_t*)src);
       break;
     case TSDB_DATA_TYPE_BIGINT:
-      tlen = taosEncodeFixedI64(NULL, *(uint32_t*)src);
+      tlen = taosEncodeFixedI64(NULL, *(int64_t*)src);
       *dst = taosMemoryCalloc(1, tlen + 1);
-      tlen = taosEncodeFixedI64(dst, *(uint32_t*)src);
+      tlen = taosEncodeFixedI64(dst, *(int64_t*)src);
       break;
     case TSDB_DATA_TYPE_DOUBLE:
       tlen = taosEncodeBinary(NULL, src, sizeof(double));
@@ -184,9 +236,9 @@ int32_t indexConvertData(void* src, int8_t type, void** dst) {
       tlen = taosEncodeBinary(dst, src, sizeof(double));
       break;
     case TSDB_DATA_TYPE_UBIGINT:
-      tlen = taosEncodeFixedU64(NULL, *(uint32_t*)src);
+      tlen = taosEncodeFixedU64(NULL, *(uint64_t*)src);
       *dst = taosMemoryCalloc(1, tlen + 1);
-      tlen = taosEncodeFixedU64(dst, *(uint32_t*)src);
+      tlen = taosEncodeFixedU64(dst, *(uint64_t*)src);
       break;
     case TSDB_DATA_TYPE_NCHAR: {
       tlen = taosEncodeBinary(NULL, varDataVal(src), varDataLen(src));
@@ -215,14 +267,94 @@ int32_t indexConvertData(void* src, int8_t type, void** dst) {
       break;
   }
   *dst = *dst - tlen;
-  if (type != TSDB_DATA_TYPE_BINARY && type != TSDB_DATA_TYPE_NCHAR && type != TSDB_DATA_TYPE_VARBINARY &&
+  // indexMayFillNumbericData(*dst, tlen);
-      type == TSDB_DATA_TYPE_VARCHAR) {
+  return tlen;
-    uint8_t* p = *dst;
+}
-    for (int i = 0; i < tlen; i++) {
+int32_t indexConvertDataToStr(void* src, int8_t type, void** dst) {
-      if (p[i] == 0) {
+  int tlen = tDataTypes[type].bytes;
-        p[i] = (uint8_t)'0';
+
+  switch (type) {
+    case TSDB_DATA_TYPE_TIMESTAMP:
+      *dst = taosMemoryCalloc(1, sizeof(int64_t) + 1);
+      indexInt2str(*(int64_t*)src, *dst, -1);
+      break;
+    case TSDB_DATA_TYPE_BOOL:
+    case TSDB_DATA_TYPE_UTINYINT:
+      // tlen = taosEncodeFixedU8(NULL, *(uint8_t*)src);
+      //*dst = taosMemoryCalloc(1, tlen + 1);
+      // tlen = taosEncodeFixedU8(dst, *(uint8_t*)src);
+      *dst = taosMemoryCalloc(1, sizeof(int64_t) + 1);
+      indexInt2str(*(uint8_t*)src, *dst, 1);
+      break;
+    case TSDB_DATA_TYPE_TINYINT:
+      *dst = taosMemoryCalloc(1, sizeof(int64_t) + 1);
+      indexInt2str(*(int8_t*)src, *dst, 1);
+      break;
+    case TSDB_DATA_TYPE_SMALLINT:
+      *dst = taosMemoryCalloc(1, sizeof(int64_t) + 1);
+      indexInt2str(*(int16_t*)src, *dst, -1);
+      break;
+    case TSDB_DATA_TYPE_USMALLINT:
+      *dst = taosMemoryCalloc(1, sizeof(int64_t) + 1);
+      indexInt2str(*(uint16_t*)src, *dst, -1);
+      break;
+    case TSDB_DATA_TYPE_INT:
+      *dst = taosMemoryCalloc(1, sizeof(int64_t) + 1);
+      indexInt2str(*(int32_t*)src, *dst, -1);
+      break;
+    case TSDB_DATA_TYPE_FLOAT:
+      tlen = taosEncodeBinary(NULL, src, sizeof(float));
+      *dst = taosMemoryCalloc(1, tlen + 1);
+      tlen = taosEncodeBinary(dst, src, sizeof(float));
+      *dst = *dst - tlen;
+      break;
+    case TSDB_DATA_TYPE_UINT:
+      *dst = taosMemoryCalloc(1, sizeof(int64_t) + 1);
+      indexInt2str(*(uint32_t*)src, *dst, 1);
+      break;
+    case TSDB_DATA_TYPE_BIGINT:
+      *dst = taosMemoryCalloc(1, sizeof(int64_t) + 1);
+      indexInt2str(*(int64_t*)src, *dst, 1);
+      break;
+    case TSDB_DATA_TYPE_DOUBLE:
+      tlen = taosEncodeBinary(NULL, src, sizeof(double));
+      *dst = taosMemoryCalloc(1, tlen + 1);
+      tlen = taosEncodeBinary(dst, src, sizeof(double));
+      *dst = *dst - tlen;
+      break;
+    case TSDB_DATA_TYPE_UBIGINT:
+      assert(0);
+      *dst = taosMemoryCalloc(1, sizeof(int64_t) + 1);
+      indexInt2str(*(uint64_t*)src, *dst, 1);
+      break;
+    case TSDB_DATA_TYPE_NCHAR: {
+      tlen = taosEncodeBinary(NULL, varDataVal(src), varDataLen(src));
+      *dst = taosMemoryCalloc(1, tlen + 1);
+      tlen = taosEncodeBinary(dst, varDataVal(src), varDataLen(src));
+      *dst = *dst - tlen;
+
+      break;
     }
+    case TSDB_DATA_TYPE_VARCHAR: {  // TSDB_DATA_TYPE_BINARY
+#if 1
+      tlen = taosEncodeBinary(NULL, src, strlen(src));
+      *dst = taosMemoryCalloc(1, tlen + 1);
+      tlen = taosEncodeBinary(dst, src, strlen(src));
+      *dst = *dst - tlen;
+      break;
+#endif
     }
+    case TSDB_DATA_TYPE_VARBINARY:
+#if 1
+      tlen = taosEncodeBinary(NULL, src, strlen(src));
+      *dst = taosMemoryCalloc(1, tlen + 1);
+      tlen = taosEncodeBinary(dst, src, strlen(src));
+      *dst = *dst - tlen;
+      break;
+#endif
+    default:
+      TASSERT(0);
+      break;
   }
   return tlen;
 }

source/libs/index/src/indexFst.c

@@ -1324,7 +1324,7 @@ StreamWithStateResult* streamWithStateNextWith(StreamWithState* sws, StreamCallb
       if (FST_NODE_ADDR(p->node) != fstGetRootAddr(sws->fst)) {
         taosArrayPop(sws->inp);
       }
-      streamStateDestroy(p);
+      // streamStateDestroy(p);
       continue;
     }
     FstTransition trn;

source/libs/index/src/indexTfile.c

@@ -410,8 +410,9 @@ static int32_t tfSearchTerm_JSON(void* reader, SIndexTerm* tem, SIdxTempResult*
 
     ret = tfileReaderLoadTableIds((TFileReader*)reader, offset, tr->total);
     cost = taosGetTimestampUs() - et;
-    indexInfo("index: %" PRIu64 ", col: %s, colVal: %s, load all table info, time cost: %" PRIu64 "us", tem->suid,
+    indexInfo("index: %" PRIu64 ", col: %s, colVal: %s, load all table info, offset: %" PRIu64
-              tem->colName, tem->colVal, cost);
+              ", size: %d, time cost: %" PRIu64 "us",
+              tem->suid, tem->colName, tem->colVal, offset, (int)taosArrayGetSize(tr->total), cost);
   }
   fstSliceDestroy(&key);
   return 0;
@@ -941,7 +942,7 @@ static int tfileReaderLoadTableIds(TFileReader* reader, int32_t offset, SArray*
   // TODO(yihao): opt later
   WriterCtx* ctx = reader->ctx;
   // add block cache
-  char    block[1024] = {0};
+  char    block[4096] = {0};
   int32_t nread = ctx->readFrom(ctx, block, sizeof(block), offset);
   assert(nread >= sizeof(uint32_t));
 

source/libs/index/test/jsonUT.cc

@@ -56,6 +56,29 @@ class JsonEnv : public ::testing::Test {
   SIndexJson*     index;
 };
 
+static void WriteData(SIndexJson* index, const std::string& colName, int8_t dtype, void* data, int dlen, int tableId,
+                      int8_t operType = ADD_VALUE) {
+  SIndexTerm* term =
+      indexTermCreate(1, (SIndexOperOnColumn)operType, dtype, colName.c_str(), colName.size(), (const char*)data, dlen);
+  SIndexMultiTerm* terms = indexMultiTermCreate();
+  indexMultiTermAdd(terms, term);
+  tIndexJsonPut(index, terms, (int64_t)tableId);
+
+  indexMultiTermDestroy(terms);
+}
+static void Search(SIndexJson* index, const std::string& colNam, int8_t dtype, void* data, int dlen, int8_t filterType,
+                   SArray** result) {
+  std::string colName(colNam);
+
+  SIndexMultiTermQuery* mq = indexMultiTermQueryCreate(MUST);
+  SIndexTerm* q = indexTermCreate(1, ADD_VALUE, dtype, colName.c_str(), colName.size(), (const char*)data, dlen);
+
+  SArray* res = taosArrayInit(1, sizeof(uint64_t));
+  indexMultiTermQueryAdd(mq, q, (EIndexQueryType)filterType);
+  tIndexJsonSearch(index, mq, res);
+  indexMultiTermQueryDestroy(mq);
+  *result = res;
+}
 TEST_F(JsonEnv, testWrite) {
   {
     std::string colName("test");
@@ -204,9 +227,10 @@ TEST_F(JsonEnv, testWriteMillonData) {
 TEST_F(JsonEnv, testWriteJsonNumberData) {
   {
     std::string colName("test");
-    std::string colVal("10");
+    // std::string colVal("10");
+    int         val = 10;
     SIndexTerm* term = indexTermCreate(1, ADD_VALUE, TSDB_DATA_TYPE_INT, colName.c_str(), colName.size(),
-                                       colVal.c_str(), colVal.size());
+                                       (const char*)&val, sizeof(val));
 
     SIndexMultiTerm* terms = indexMultiTermCreate();
     indexMultiTermAdd(terms, term);
@@ -217,35 +241,9 @@ TEST_F(JsonEnv, testWriteJsonNumberData) {
   }
   {
     std::string colName("test2");
-    std::string colVal("20");
+    int         val = 20;
     SIndexTerm* term = indexTermCreate(1, ADD_VALUE, TSDB_DATA_TYPE_INT, colName.c_str(), colName.size(),
-                                       colVal.c_str(), colVal.size());
+                                       (const char*)&val, sizeof(val));
-
-    SIndexMultiTerm* terms = indexMultiTermCreate();
-    indexMultiTermAdd(terms, term);
-    for (size_t i = 0; i < 1000; i++) {
-      tIndexJsonPut(index, terms, i);
-    }
-    indexMultiTermDestroy(terms);
-  }
-  {
-    std::string colName("test2");
-    std::string colVal("15");
-    SIndexTerm* term = indexTermCreate(1, ADD_VALUE, TSDB_DATA_TYPE_INT, colName.c_str(), colName.size(),
-                                       colVal.c_str(), colVal.size());
-
-    SIndexMultiTerm* terms = indexMultiTermCreate();
-    indexMultiTermAdd(terms, term);
-    for (size_t i = 0; i < 1000; i++) {
-      tIndexJsonPut(index, terms, i);
-    }
-    indexMultiTermDestroy(terms);
-  }
-  {
-    std::string colName("test2");
-    std::string colVal("15");
-    SIndexTerm* term = indexTermCreate(1, ADD_VALUE, TSDB_DATA_TYPE_BINARY, colName.c_str(), colName.size(),
-                                       colVal.c_str(), colVal.size());
 
     SIndexMultiTerm* terms = indexMultiTermCreate();
     indexMultiTermAdd(terms, term);
@@ -256,11 +254,36 @@ TEST_F(JsonEnv, testWriteJsonNumberData) {
   }
   {
     std::string colName("test");
-    std::string colVal("10");
+    int         val = 15;
+    SIndexTerm* term = indexTermCreate(1, ADD_VALUE, TSDB_DATA_TYPE_INT, colName.c_str(), colName.size(),
+                                       (const char*)&val, sizeof(val));
 
+    SIndexMultiTerm* terms = indexMultiTermCreate();
+    indexMultiTermAdd(terms, term);
+    for (size_t i = 0; i < 1000; i++) {
+      tIndexJsonPut(index, terms, i);
+    }
+    indexMultiTermDestroy(terms);
+  }
+  {
+    std::string colName("test2");
+    const char* val = "test";
+    SIndexTerm* term = indexTermCreate(1, ADD_VALUE, TSDB_DATA_TYPE_BINARY, colName.c_str(), colName.size(),
+                                       (const char*)val, strlen(val));
+
+    SIndexMultiTerm* terms = indexMultiTermCreate();
+    indexMultiTermAdd(terms, term);
+    for (size_t i = 0; i < 1000; i++) {
+      tIndexJsonPut(index, terms, i);
+    }
+    indexMultiTermDestroy(terms);
+  }
+  {
+    std::string           colName("test");
+    int                   val = 15;
     SIndexMultiTermQuery* mq = indexMultiTermQueryCreate(MUST);
-    SIndexTerm* q = indexTermCreate(1, ADD_VALUE, TSDB_DATA_TYPE_INT, colName.c_str(), colName.size(), colVal.c_str(),
+    SIndexTerm*           q = indexTermCreate(1, ADD_VALUE, TSDB_DATA_TYPE_INT, colName.c_str(), colName.size(),
-                                    colVal.size());
+                                    (const char*)&val, sizeof(val));
 
     SArray* result = taosArrayInit(1, sizeof(uint64_t));
     indexMultiTermQueryAdd(mq, q, QUERY_TERM);
@@ -270,11 +293,11 @@ TEST_F(JsonEnv, testWriteJsonNumberData) {
   }
   {
     std::string colName("test");
-    std::string colVal("10");
+    int         val = 15;
 
     SIndexMultiTermQuery* mq = indexMultiTermQueryCreate(MUST);
-    SIndexTerm* q = indexTermCreate(1, ADD_VALUE, TSDB_DATA_TYPE_INT, colName.c_str(), colName.size(), colVal.c_str(),
+    SIndexTerm*           q = indexTermCreate(1, ADD_VALUE, TSDB_DATA_TYPE_INT, colName.c_str(), colName.size(),
-                                    colVal.size());
+                                    (const char*)&val, sizeof(val));
 
     SArray* result = taosArrayInit(1, sizeof(uint64_t));
     indexMultiTermQueryAdd(mq, q, QUERY_GREATER_THAN);
@@ -284,11 +307,12 @@ TEST_F(JsonEnv, testWriteJsonNumberData) {
   }
   {
     std::string colName("test");
-    std::string colVal("10");
+    int         val = 10;
+    ;
 
     SIndexMultiTermQuery* mq = indexMultiTermQueryCreate(MUST);
-    SIndexTerm* q = indexTermCreate(1, ADD_VALUE, TSDB_DATA_TYPE_INT, colName.c_str(), colName.size(), colVal.c_str(),
+    SIndexTerm*           q = indexTermCreate(1, ADD_VALUE, TSDB_DATA_TYPE_INT, colName.c_str(), colName.size(),
-                                    colVal.size());
+                                    (const char*)&val, sizeof(int));
 
     SArray* result = taosArrayInit(1, sizeof(uint64_t));
     indexMultiTermQueryAdd(mq, q, QUERY_GREATER_EQUAL);
@@ -298,11 +322,12 @@ TEST_F(JsonEnv, testWriteJsonNumberData) {
   }
   {
     std::string colName("test");
-    std::string colVal("10");
+    int         val = 10;
+    // std::string colVal("10");
 
     SIndexMultiTermQuery* mq = indexMultiTermQueryCreate(MUST);
-    SIndexTerm* q = indexTermCreate(1, ADD_VALUE, TSDB_DATA_TYPE_INT, colName.c_str(), colName.size(), colVal.c_str(),
+    SIndexTerm*           q = indexTermCreate(1, ADD_VALUE, TSDB_DATA_TYPE_INT, colName.c_str(), colName.size(),
-                                    colVal.size());
+                                    (const char*)&val, sizeof(val));
 
     SArray* result = taosArrayInit(1, sizeof(uint64_t));
     indexMultiTermQueryAdd(mq, q, QUERY_LESS_THAN);
@@ -312,11 +337,12 @@ TEST_F(JsonEnv, testWriteJsonNumberData) {
   }
   {
     std::string colName("test");
-    std::string colVal("10");
+    int         val = 10;
+    // std::string colVal("10");
 
     SIndexMultiTermQuery* mq = indexMultiTermQueryCreate(MUST);
-    SIndexTerm* q = indexTermCreate(1, ADD_VALUE, TSDB_DATA_TYPE_INT, colName.c_str(), colName.size(), colVal.c_str(),
+    SIndexTerm*           q = indexTermCreate(1, ADD_VALUE, TSDB_DATA_TYPE_INT, colName.c_str(), colName.size(),
-                                    colVal.size());
+                                    (const char*)&val, sizeof(val));
 
     SArray* result = taosArrayInit(1, sizeof(uint64_t));
     indexMultiTermQueryAdd(mq, q, QUERY_LESS_EQUAL);
@@ -326,12 +352,12 @@ TEST_F(JsonEnv, testWriteJsonNumberData) {
   }
 }
 
-TEST_F(JsonEnv, testWriteJsonTfileAndCache) {
+TEST_F(JsonEnv, testWriteJsonTfileAndCache_INT) {
   {
     std::string colName("test1");
-    std::string colVal("10");
+    int         val = 10;
     SIndexTerm* term = indexTermCreate(1, ADD_VALUE, TSDB_DATA_TYPE_INT, colName.c_str(), colName.size(),
-                                       colVal.c_str(), colVal.size());
+                                       (const char*)&val, sizeof(val));
 
     SIndexMultiTerm* terms = indexMultiTermCreate();
     indexMultiTermAdd(terms, term);
@@ -355,11 +381,11 @@ TEST_F(JsonEnv, testWriteJsonTfileAndCache) {
   }
   {
     std::string colName("test1");
-    std::string colVal("10");
+    int         val = 10;
 
     SIndexMultiTermQuery* mq = indexMultiTermQueryCreate(MUST);
-    SIndexTerm* q = indexTermCreate(1, ADD_VALUE, TSDB_DATA_TYPE_INT, colName.c_str(), colName.size(), colVal.c_str(),
+    SIndexTerm*           q = indexTermCreate(1, ADD_VALUE, TSDB_DATA_TYPE_INT, colName.c_str(), colName.size(),
-                                    colVal.size());
+                                    (const char*)&val, sizeof(val));
 
     SArray* result = taosArrayInit(1, sizeof(uint64_t));
     indexMultiTermQueryAdd(mq, q, QUERY_TERM);
@@ -369,11 +395,11 @@ TEST_F(JsonEnv, testWriteJsonTfileAndCache) {
   }
   {
     std::string colName("test1");
-    std::string colVal("10");
+    int         val = 10;
 
     SIndexMultiTermQuery* mq = indexMultiTermQueryCreate(MUST);
-    SIndexTerm* q = indexTermCreate(1, ADD_VALUE, TSDB_DATA_TYPE_INT, colName.c_str(), colName.size(), colVal.c_str(),
+    SIndexTerm*           q = indexTermCreate(1, ADD_VALUE, TSDB_DATA_TYPE_INT, colName.c_str(), colName.size(),
-                                    colVal.size());
+                                    (const char*)&val, sizeof(int));
 
     SArray* result = taosArrayInit(1, sizeof(uint64_t));
     indexMultiTermQueryAdd(mq, q, QUERY_GREATER_THAN);
@@ -383,11 +409,12 @@ TEST_F(JsonEnv, testWriteJsonTfileAndCache) {
   }
   {
     std::string colName("test1");
-    std::string colVal("10");
+    // std::string colVal("10");
+    int val = 10;
 
     SIndexMultiTermQuery* mq = indexMultiTermQueryCreate(MUST);
-    SIndexTerm* q = indexTermCreate(1, ADD_VALUE, TSDB_DATA_TYPE_INT, colName.c_str(), colName.size(), colVal.c_str(),
+    SIndexTerm*           q = indexTermCreate(1, ADD_VALUE, TSDB_DATA_TYPE_INT, colName.c_str(), colName.size(),
-                                    colVal.size());
+                                    (const char*)&val, sizeof(val));
 
     SArray* result = taosArrayInit(1, sizeof(uint64_t));
     indexMultiTermQueryAdd(mq, q, QUERY_GREATER_EQUAL);
@@ -397,11 +424,11 @@ TEST_F(JsonEnv, testWriteJsonTfileAndCache) {
   }
   {
     std::string colName("test1");
-    std::string colVal("10");
+    int         val = 10;
 
     SIndexMultiTermQuery* mq = indexMultiTermQueryCreate(MUST);
-    SIndexTerm* q = indexTermCreate(1, ADD_VALUE, TSDB_DATA_TYPE_INT, colName.c_str(), colName.size(), colVal.c_str(),
+    SIndexTerm*           q = indexTermCreate(1, ADD_VALUE, TSDB_DATA_TYPE_INT, colName.c_str(), colName.size(),
-                                    colVal.size());
+                                    (const char*)&val, sizeof(val));
 
     SArray* result = taosArrayInit(1, sizeof(uint64_t));
     indexMultiTermQueryAdd(mq, q, QUERY_GREATER_THAN);
@@ -411,11 +438,11 @@ TEST_F(JsonEnv, testWriteJsonTfileAndCache) {
   }
   {
     std::string colName("test1");
-    std::string colVal("10");
+    int         val = 10;
 
     SIndexMultiTermQuery* mq = indexMultiTermQueryCreate(MUST);
-    SIndexTerm* q = indexTermCreate(1, ADD_VALUE, TSDB_DATA_TYPE_INT, colName.c_str(), colName.size(), colVal.c_str(),
+    SIndexTerm*           q = indexTermCreate(1, ADD_VALUE, TSDB_DATA_TYPE_INT, colName.c_str(), colName.size(),
-                                    colVal.size());
+                                    (const char*)&val, sizeof(val));
 
     SArray* result = taosArrayInit(1, sizeof(uint64_t));
     indexMultiTermQueryAdd(mq, q, QUERY_LESS_EQUAL);
@@ -425,9 +452,10 @@ TEST_F(JsonEnv, testWriteJsonTfileAndCache) {
   }
   {
     std::string colName("other_column");
-    std::string colVal("100");
+    int         val = 100;
+
     SIndexTerm* term = indexTermCreate(1, ADD_VALUE, TSDB_DATA_TYPE_INT, colName.c_str(), colName.size(),
-                                       colVal.c_str(), colVal.size());
+                                       (const char*)&val, sizeof(val));
 
     SIndexMultiTerm* terms = indexMultiTermCreate();
     indexMultiTermAdd(terms, term);
@@ -438,11 +466,12 @@ TEST_F(JsonEnv, testWriteJsonTfileAndCache) {
   }
   {
     std::string colName("test1");
-    std::string colVal("10");
+    int         val = 10;
+    // std::string colVal("10");
 
     SIndexMultiTermQuery* mq = indexMultiTermQueryCreate(MUST);
-    SIndexTerm* q = indexTermCreate(1, ADD_VALUE, TSDB_DATA_TYPE_INT, colName.c_str(), colName.size(), colVal.c_str(),
+    SIndexTerm*           q = indexTermCreate(1, ADD_VALUE, TSDB_DATA_TYPE_INT, colName.c_str(), colName.size(),
-                                    colVal.size());
+                                    (const char*)&val, sizeof(val));
 
     SArray* result = taosArrayInit(1, sizeof(uint64_t));
     indexMultiTermQueryAdd(mq, q, QUERY_LESS_THAN);
@@ -450,4 +479,102 @@ TEST_F(JsonEnv, testWriteJsonTfileAndCache) {
     EXPECT_EQ(0, taosArrayGetSize(result));
     indexMultiTermQueryDestroy(mq);
   }
+  {
+    std::string colName("test1");
+    int         val = 15;
+    SIndexTerm* term = indexTermCreate(1, ADD_VALUE, TSDB_DATA_TYPE_INT, colName.c_str(), colName.size(),
+                                       (const char*)&val, sizeof(val));
+
+    SIndexMultiTerm* terms = indexMultiTermCreate();
+    indexMultiTermAdd(terms, term);
+    for (size_t i = 0; i < 1000; i++) {
+      tIndexJsonPut(index, terms, i + 1000);
+    }
+    indexMultiTermDestroy(terms);
+  }
+  {
+    std::string colName("test1");
+    int         val = 8;
+    // std::string colVal("10");
+
+    SIndexMultiTermQuery* mq = indexMultiTermQueryCreate(MUST);
+    SIndexTerm*           q = indexTermCreate(1, ADD_VALUE, TSDB_DATA_TYPE_INT, colName.c_str(), colName.size(),
+                                    (const char*)&val, sizeof(val));
+
+    SArray* result = taosArrayInit(1, sizeof(uint64_t));
+    indexMultiTermQueryAdd(mq, q, QUERY_GREATER_EQUAL);
+    tIndexJsonSearch(index, mq, result);
+    EXPECT_EQ(2000, taosArrayGetSize(result));
+    indexMultiTermQueryDestroy(mq);
+  }
+}
+TEST_F(JsonEnv, testWriteJsonTfileAndCache_INT2) {
+  {
+    int         val = 10;
+    std::string colName("test1");
+    for (int i = 0; i < 10000; i++) {
+      val += 1;
+      WriteData(index, colName, TSDB_DATA_TYPE_INT, &val, sizeof(val), i);
+    }
+  }
+  {
+    int         val = 10;
+    std::string colName("test2xxx");
+    std::string colVal("xxxxxxxxxxxxxxx");
+    for (int i = 0; i < 100000; i++) {
+      val += 1;
+      WriteData(index, colName, TSDB_DATA_TYPE_BINARY, (void*)(colVal.c_str()), colVal.size(), i);
+    }
+  }
+  {
+    SArray*     res = NULL;
+    std::string colName("test1");
+    int         val = 9;
+    Search(index, colName, TSDB_DATA_TYPE_INT, &val, sizeof(val), QUERY_GREATER_EQUAL, &res);
+    EXPECT_EQ(10000, taosArrayGetSize(res));
+  }
+  {
+    SArray*     res = NULL;
+    std::string colName("test2xxx");
+    std::string colVal("xxxxxxxxxxxxxxx");
+    Search(index, colName, TSDB_DATA_TYPE_BINARY, (void*)(colVal.c_str()), colVal.size(), QUERY_TERM, &res);
+    EXPECT_EQ(100000, taosArrayGetSize(res));
+  }
+}
+TEST_F(JsonEnv, testWriteJsonTfileAndCache_FLOAT) {
+  {
+    float       val = 10.0;
+    std::string colName("test1");
+    for (int i = 0; i < 1000; i++) {
+      WriteData(index, colName, TSDB_DATA_TYPE_FLOAT, &val, sizeof(val), i);
+    }
+  }
+  {
+    float       val = 2.0;
+    std::string colName("test1");
+    for (int i = 0; i < 1000; i++) {
+      WriteData(index, colName, TSDB_DATA_TYPE_FLOAT, &val, sizeof(val), i);
+    }
+  }
+  {
+    SArray*     res = NULL;
+    std::string colName("test1");
+    float       val = 1.9;
+    Search(index, colName, TSDB_DATA_TYPE_FLOAT, &val, sizeof(val), QUERY_GREATER_EQUAL, &res);
+    EXPECT_EQ(2000, taosArrayGetSize(res));
+  }
+  {
+    SArray*     res = NULL;
+    std::string colName("test1");
+    float       val = 2.1;
+    Search(index, colName, TSDB_DATA_TYPE_FLOAT, &val, sizeof(val), QUERY_GREATER_EQUAL, &res);
+    EXPECT_EQ(1000, taosArrayGetSize(res));
+  }
+  {
+    std::string colName("test1");
+    SArray*     res = NULL;
+    float       val = 2.1;
+    Search(index, colName, TSDB_DATA_TYPE_FLOAT, &val, sizeof(val), QUERY_GREATER_EQUAL, &res);
+    EXPECT_EQ(1000, taosArrayGetSize(res));
+  }
 }

source/libs/index/test/utilUT.cc

@@ -6,12 +6,14 @@
 #include <vector>
 #include "index.h"
 #include "indexCache.h"
+#include "indexComm.h"
 #include "indexFst.h"
 #include "indexFstCountingWriter.h"
 #include "indexFstUtil.h"
 #include "indexInt.h"
 #include "indexTfile.h"
 #include "indexUtil.h"
+#include "tcoding.h"
 #include "tglobal.h"
 #include "tskiplist.h"
 #include "tutil.h"
@@ -305,3 +307,17 @@ TEST_F(UtilEnv, 01Except) {
   ASSERT_EQ(*(uint64_t *)taosArrayGet(total, 0), 1);
   ASSERT_EQ(*(uint64_t *)taosArrayGet(total, 1), 100);
 }
+TEST_F(UtilEnv, testFill) {
+  for (int i = 0; i < 10000000; i++) {
+    int64_t val = i;
+    char    buf[65] = {0};
+    indexInt2str(val, buf, 1);
+    EXPECT_EQ(val, taosStr2int64(buf));
+  }
+  for (int i = 0; i < 10000000; i++) {
+    int64_t val = 0 - i;
+    char    buf[65] = {0};
+    indexInt2str(val, buf, -1);
+    EXPECT_EQ(val, taosStr2int64(buf));
+  }
+}

source/libs/stream/src/tstream.c

@@ -35,7 +35,7 @@ void* streamDataBlockDecode(const void* buf, SStreamDataBlock* pInput) {
   return (void*)buf;
 }
 
-static int32_t streamBuildDispatchMsg(SStreamTask* pTask, SArray* data, SRpcMsg* pMsg, SEpSet** ppEpSet) {
+static int32_t streamBuildExecMsg(SStreamTask* pTask, SArray* data, SRpcMsg* pMsg, SEpSet** ppEpSet) {
   SStreamTaskExecReq req = {
       .streamId = pTask->streamId,
       .data = data,
@@ -107,7 +107,7 @@ static int32_t streamShuffleDispatch(SStreamTask* pTask, SMsgCb* pMsgCb, SHashOb
     SArray* pData = *(SArray**)pIter;
     SRpcMsg dispatchMsg = {0};
     SEpSet* pEpSet;
-    if (streamBuildDispatchMsg(pTask, pData, &dispatchMsg, &pEpSet) < 0) {
+    if (streamBuildExecMsg(pTask, pData, &dispatchMsg, &pEpSet) < 0) {
       ASSERT(0);
       return -1;
     }
@@ -133,7 +133,7 @@ int32_t streamEnqueueDataBlk(SStreamTask* pTask, SStreamDataBlock* input) {
   return inputStatus;
 }
 
-int32_t streamTaskExecImpl(SStreamTask* pTask, void* data, SArray* pRes) {
+static int32_t streamTaskExecImpl(SStreamTask* pTask, void* data, SArray* pRes) {
   void* exec = pTask->exec.runners[0].executor;
 
   // set input
@@ -265,87 +265,42 @@ FAIL:
   return -1;
 }
 
-int32_t streamTaskDispatchDown(SStreamTask* pTask, SMsgCb* pMsgCb) {
+int32_t streamTaskSink(SStreamTask* pTask, SMsgCb* pMsgCb) {
-  //
+  bool firstRun = 1;
-  return 0;
-}
-
-int32_t streamTaskSink(SStreamTask* pTask) {
-  //
-  return 0;
-}
-
-int32_t streamTaskProcessInputReq(SStreamTask* pTask, SMsgCb* pMsgCb, SStreamDataBlock* pBlock, SRpcMsg* pRsp) {
-  // 1. handle input
-  // 1.1 enqueue
-  taosWriteQitem(pTask->inputQ, pBlock);
-  // 1.2 calc back pressure
-  // 1.3 rsp by input status
-  int8_t              inputStatus = atomic_load_8(&pTask->inputStatus);
-  SStreamDispatchRsp* pCont = rpcMallocCont(sizeof(SStreamDispatchRsp));
-  pCont->status = inputStatus;
-  pRsp->pCont = pCont;
-  pRsp->contLen = sizeof(SStreamDispatchRsp);
-  tmsgSendRsp(pRsp);
-  // 2. try exec
-  // 2.1. idle: exec
-  // 2.2. executing: return
-  // 2.3. closing: keep trying
   while (1) {
-    int8_t execStatus = atomic_val_compare_exchange_8(&pTask->status, TASK_STATUS__IDLE, TASK_STATUS__EXECUTING);
+    SStreamDataBlock* pBlock = NULL;
-    if (execStatus == TASK_STATUS__IDLE) {
+    if (!firstRun) {
-      void*         exec = pTask->exec.runners[0].executor;
+      taosReadAllQitems(pTask->outputQ, pTask->outputQAll);
-      SArray*       pRes = taosArrayInit(0, sizeof(void*));
-      const SArray* blocks = pBlock->blocks;
-      qSetMultiStreamInput(exec, blocks->pData, blocks->size, STREAM_DATA_TYPE_SSDATA_BLOCK);
-      while (1) {
-        SSDataBlock* output;
-        uint64_t     ts = 0;
-        if (qExecTask(exec, &output, &ts) < 0) {
-          ASSERT(false);
     }
-        if (output == NULL) break;
+    taosGetQitem(pTask->outputQAll, (void**)&pBlock);
-        taosArrayPush(pRes, &output);
+    if (pBlock == NULL) {
-      }
+      if (firstRun) {
-      // TODO: wrap destroy block
+        firstRun = 0;
-      taosArrayDestroyP(pBlock->blocks, (FDelete)blockDataDestroy);
-
-      if (taosArrayGetSize(pRes) != 0) {
-        SArray** resQ = taosAllocateQitem(sizeof(void**), DEF_QITEM);
-        *resQ = pRes;
-        taosWriteQitem(pTask->outputQ, resQ);
-      }
-
-    } else if (execStatus == TASK_STATUS__CLOSING) {
         continue;
-    } else if (execStatus == TASK_STATUS__EXECUTING)
+      } else {
         break;
-    else {
-      ASSERT(0);
       }
     }
-  // 3. handle output
-  // 3.1 check and set status
-  // 3.2 dispatch / sink
-  STaosQall* qall = taosAllocateQall();
-  taosReadAllQitems(pTask->outputQ, qall);
-  SArray** ppRes = NULL;
-  while (1) {
-    taosGetQitem(qall, (void**)&ppRes);
-    if (ppRes == NULL) break;
 
-    SArray* pRes = *ppRes;
+    SArray* pRes = pBlock->blocks;
+
+    // sink
     if (pTask->sinkType == TASK_SINK__TABLE) {
-      pTask->tbSink.tbSinkFunc(pTask, pTask->tbSink.vnode, pBlock->sourceVer, pRes);
+      // blockDebugShowData(pRes);
+      pTask->tbSink.tbSinkFunc(pTask, pTask->tbSink.vnode, 0, pRes);
     } else if (pTask->sinkType == TASK_SINK__SMA) {
       pTask->smaSink.smaSink(pTask->ahandle, pTask->smaSink.smaId, pRes);
+      //
+    } else if (pTask->sinkType == TASK_SINK__FETCH) {
+      //
     } else {
+      ASSERT(pTask->sinkType == TASK_SINK__NONE);
     }
 
     // dispatch
     if (pTask->dispatchType == TASK_DISPATCH__INPLACE) {
       SRpcMsg dispatchMsg = {0};
-      if (streamBuildDispatchMsg(pTask, pRes, &dispatchMsg, NULL) < 0) {
+      if (streamBuildExecMsg(pTask, pRes, &dispatchMsg, NULL) < 0) {
         ASSERT(0);
         return -1;
       }
@@ -366,7 +321,7 @@ int32_t streamTaskProcessInputReq(SStreamTask* pTask, SMsgCb* pMsgCb, SStreamDat
     } else if (pTask->dispatchType == TASK_DISPATCH__FIXED) {
       SRpcMsg dispatchMsg = {0};
       SEpSet* pEpSet = NULL;
-      if (streamBuildDispatchMsg(pTask, pRes, &dispatchMsg, &pEpSet) < 0) {
+      if (streamBuildExecMsg(pTask, pRes, &dispatchMsg, &pEpSet) < 0) {
         ASSERT(0);
         return -1;
       }
@@ -401,12 +356,53 @@ int32_t streamTaskProcessInputReq(SStreamTask* pTask, SMsgCb* pMsgCb, SStreamDat
       ASSERT(pTask->dispatchType == TASK_DISPATCH__NONE);
     }
   }
-  //
   return 0;
 }
 
-int32_t streamTaskProcessDispatchRsp(SStreamTask* pTask, char* msg, int32_t msgLen) {
+int32_t streamTaskEnqueue(SStreamTask* pTask, SStreamDispatchReq* pReq, SRpcMsg* pRsp) {
-  //
+  SStreamDataBlock* pBlock = taosAllocateQitem(sizeof(SStreamDataBlock), DEF_QITEM);
+  int8_t            status;
+
+  // 1.1 update status
+  // TODO cal backpressure
+  if (pBlock == NULL) {
+    atomic_store_8(&pTask->inputStatus, TASK_INPUT_STATUS__FAILED);
+    status = TASK_INPUT_STATUS__FAILED;
+  } else {
+    status = atomic_load_8(&pTask->inputStatus);
+  }
+
+  // 1.2 enqueue
+  pBlock->type = STREAM_DATA_TYPE_SSDATA_BLOCK;
+  pBlock->sourceVg = pReq->sourceVg;
+  pBlock->sourceVer = pReq->sourceVer;
+  taosWriteQitem(pTask->inputQ, pBlock);
+
+  // 1.3 rsp by input status
+  SStreamDispatchRsp* pCont = rpcMallocCont(sizeof(SStreamDispatchRsp));
+  pCont->inputStatus = status;
+  pRsp->pCont = pCont;
+  pRsp->contLen = sizeof(SStreamDispatchRsp);
+  tmsgSendRsp(pRsp);
+
+  return 0;
+}
+
+int32_t streamTaskProcessDispatchReq(SStreamTask* pTask, SMsgCb* pMsgCb, SStreamDispatchReq* pReq, SRpcMsg* pRsp) {
+  // 1. handle input
+  streamTaskEnqueue(pTask, pReq, pRsp);
+
+  // 2. try exec
+  // 2.1. idle: exec
+  // 2.2. executing: return
+  // 2.3. closing: keep trying
+  streamTaskExec2(pTask, pMsgCb);
+
+  // 3. handle output
+  // 3.1 check and set status
+  // 3.2 dispatch / sink
+  streamTaskSink(pTask, pMsgCb);
+
   return 0;
 }
 
@@ -415,64 +411,6 @@ int32_t streamTaskProcessRecoverReq(SStreamTask* pTask, char* msg) {
   return 0;
 }
 
-int32_t streamTaskRun(SStreamTask* pTask) {
-  SArray* pRes = NULL;
-  if (pTask->execType == TASK_EXEC__PIPE || pTask->execType == TASK_EXEC__MERGE) {
-    // TODO remove multi runner
-    void* exec = pTask->exec.runners[0].executor;
-
-    int8_t status = atomic_val_compare_exchange_8(&pTask->status, TASK_STATUS__IDLE, TASK_STATUS__EXECUTING);
-    if (status == TASK_STATUS__IDLE) {
-      pRes = taosArrayInit(0, sizeof(void*));
-      if (pRes == NULL) {
-        return -1;
-      }
-
-      void* input = NULL;
-      taosWriteQitem(pTask->inputQ, &input);
-      if (input == NULL) return 0;
-
-      // TODO: fix type
-      if (pTask->sourceType == TASK_SOURCE__SCAN) {
-        SStreamDataSubmit* pSubmit = (SStreamDataSubmit*)input;
-        qSetStreamInput(exec, pSubmit->data, STREAM_DATA_TYPE_SUBMIT_BLOCK);
-        while (1) {
-          SSDataBlock* output;
-          uint64_t     ts = 0;
-          if (qExecTask(exec, &output, &ts) < 0) {
-            ASSERT(false);
-          }
-          if (output == NULL) break;
-          taosArrayPush(pRes, &output);
-        }
-        streamDataSubmitRefDec(pSubmit);
-      } else {
-        SStreamDataBlock* pStreamBlock = (SStreamDataBlock*)input;
-        const SArray*     blocks = pStreamBlock->blocks;
-        qSetMultiStreamInput(exec, blocks->pData, blocks->size, STREAM_DATA_TYPE_SSDATA_BLOCK);
-        while (1) {
-          SSDataBlock* output;
-          uint64_t     ts = 0;
-          if (qExecTask(exec, &output, &ts) < 0) {
-            ASSERT(false);
-          }
-          if (output == NULL) break;
-          taosArrayPush(pRes, &output);
-        }
-        // TODO: wrap destroy block
-        taosArrayDestroyP(pStreamBlock->blocks, (FDelete)blockDataDestroy);
-      }
-
-      if (taosArrayGetSize(pRes) != 0) {
-        SArray** resQ = taosAllocateQitem(sizeof(void**), DEF_QITEM);
-        *resQ = pRes;
-        taosWriteQitem(pTask->outputQ, resQ);
-      }
-    }
-  }
-  return 0;
-}
-
 int32_t streamExecTask(SStreamTask* pTask, SMsgCb* pMsgCb, const void* input, int32_t inputType, int32_t workId) {
   SArray* pRes = NULL;
   // source
@@ -545,7 +483,7 @@ int32_t streamExecTask(SStreamTask* pTask, SMsgCb* pMsgCb, const void* input, in
 
   if (pTask->dispatchType == TASK_DISPATCH__INPLACE) {
     SRpcMsg dispatchMsg = {0};
-    if (streamBuildDispatchMsg(pTask, pRes, &dispatchMsg, NULL) < 0) {
+    if (streamBuildExecMsg(pTask, pRes, &dispatchMsg, NULL) < 0) {
       ASSERT(0);
       return -1;
     }
@@ -566,7 +504,7 @@ int32_t streamExecTask(SStreamTask* pTask, SMsgCb* pMsgCb, const void* input, in
   } else if (pTask->dispatchType == TASK_DISPATCH__FIXED) {
     SRpcMsg dispatchMsg = {0};
     SEpSet* pEpSet = NULL;
-    if (streamBuildDispatchMsg(pTask, pRes, &dispatchMsg, &pEpSet) < 0) {
+    if (streamBuildExecMsg(pTask, pRes, &dispatchMsg, &pEpSet) < 0) {
       ASSERT(0);
       return -1;
     }

source/util/src/terror.c

@@ -285,6 +285,7 @@ TAOS_DEFINE_ERROR(TSDB_CODE_MND_INVALID_TOPIC_QUERY,      "Topic with invalid qu
 TAOS_DEFINE_ERROR(TSDB_CODE_MND_INVALID_TOPIC_OPTION,     "Topic with invalid option")
 TAOS_DEFINE_ERROR(TSDB_CODE_MND_CONSUMER_NOT_EXIST,       "Consumer not exist")
 TAOS_DEFINE_ERROR(TSDB_CODE_MND_CONSUMER_NOT_READY,       "Consumer waiting for rebalance")
+TAOS_DEFINE_ERROR(TSDB_CODE_MND_TOPIC_SUBSCRIBED,         "Topic subscribed cannot be dropped")
 
 // mnode-sma
 TAOS_DEFINE_ERROR(TSDB_CODE_MND_SMA_ALREADY_EXIST,        "SMA already exists")
@@ -358,7 +359,7 @@ TAOS_DEFINE_ERROR(TSDB_CODE_TDB_TABLE_RECREATED,          "Table re-created")
 TAOS_DEFINE_ERROR(TSDB_CODE_TDB_TDB_ENV_OPEN_ERROR,       "TDB env open error")
 TAOS_DEFINE_ERROR(TSDB_CODE_TDB_NO_SMA_INDEX_IN_META,     "No sma index in meta")
 TAOS_DEFINE_ERROR(TSDB_CODE_TDB_INVALID_SMA_STAT,         "Invalid sma state")
-TAOS_DEFINE_ERROR(TSDB_CODE_TDB_TSMA_ALREADY_EXIST,       "Tsma already exists")
+TAOS_DEFINE_ERROR(TSDB_CODE_TDB_TSMA_ALREADY_EXIST,       "TSMA already exists")
 
 
 // query

source/util/src/thashutil.c

@@ -78,6 +78,42 @@ uint32_t MurmurHash3_32(const char *key, uint32_t len) {
   return h1;
 }
 
+uint64_t MurmurHash3_64(const char *key, uint32_t len) {
+  const uint64_t m = 0x87c37b91114253d5;
+  const int r = 47;
+  uint32_t seed = 0x12345678;
+  uint64_t h = seed ^ (len * m);
+  const uint8_t *data = (const uint8_t *)key;
+  const uint8_t *end = data + (len-(len&7));
+
+  while(data != end) {
+    uint64_t k = *((uint64_t*)data);
+
+    k *= m;
+    k ^= k >> r;
+    k *= m;
+    h ^= k;
+    h *= m;
+    data += 8;
+  }
+
+  switch(len & 7) {
+    case 7: h ^= (uint64_t)data[6] << 48; /* fall-thru */
+    case 6: h ^= (uint64_t)data[5] << 40; /* fall-thru */
+    case 5: h ^= (uint64_t)data[4] << 32; /* fall-thru */
+    case 4: h ^= (uint64_t)data[3] << 24; /* fall-thru */
+    case 3: h ^= (uint64_t)data[2] << 16; /* fall-thru */
+    case 2: h ^= (uint64_t)data[1] << 8; /* fall-thru */
+    case 1: h ^= (uint64_t)data[0];
+      h *= m; /* fall-thru */
+  };
+
+  h ^= h >> r;
+  h *= m;
+  h ^= h >> r;
+  return h;
+}
+
 uint32_t taosIntHash_32(const char *key, uint32_t UNUSED_PARAM(len)) { return *(uint32_t *)key; }
 uint32_t taosIntHash_16(const char *key, uint32_t UNUSED_PARAM(len)) { return *(uint16_t *)key; }
 uint32_t taosIntHash_8(const char *key, uint32_t UNUSED_PARAM(len)) { return *(uint8_t *)key; }

tests/script/jenkins/basic.txt

@@ -104,7 +104,8 @@
 ./test.sh -f tsim/mnode/basic1.sim -m
 
 # --- sma
-# ./test.sh -f tsim/sma/tsmaCreateInsertData.sim
+./test.sh -f tsim/sma/tsmaCreateInsertData.sim
+./test.sh -f tsim/sma/rsmaCreateInsertQuery.sim
 
 # --- valgrind
 ./test.sh -f tsim/valgrind/checkError.sim -v

tests/script/tsim/sma/rsmaCreateInsertQuery.sim

@@ -0,0 +1,89 @@
+system sh/stop_dnodes.sh
+system sh/deploy.sh -n dnode1 -i 1
+system sh/exec.sh -n dnode1 -s start
+sleep 50
+sql connect
+
+print =============== create database with retentions
+sql create database d0 retentions 15s:7d,1m:21d,15m:365d;
+sql use d0
+
+print =============== create super table and register rsma
+sql create table if not exists stb (ts timestamp, c1 int) tags (city binary(20),district binary(20)) rollup(min) file_factor 0.1 delay 2;
+
+sql show stables
+if $rows != 1 then 
+  return -1
+endi
+
+print =============== create child table
+sql create table ct1 using stb tags("BeiJing", "ChaoYang");
+
+sql show tables
+if $rows != 1 then 
+  return -1
+endi
+
+print =============== insert data and trigger rollup
+sql insert into ct1 values(now, 10);
+sql insert into ct1 values(now+1s, 1);
+sql insert into ct1 values(now+2s, 100);
+
+
+print =============== select * from retention level 2 from memory
+sql select * from ct1;
+print $data00 $data01
+if $rows > 2 then
+  print retention level 2 file rows $rows > 2
+  return -1
+endi
+print =============== select * from retention level 1 from memory
+sql select * from ct1 where ts > now-8d;
+print $data00 $data01
+if $rows > 2 then
+  print retention level 1 file rows $rows > 2
+  return -1
+endi
+print =============== select * from retention level 0 from memory
+sql select * from ct1 where ts > now-3d;
+print $data00 $data01
+print $data10 $data11
+print $data20 $data21
+if $rows < 1 then
+  print retention level 0 file rows $rows < 1
+  return -1
+endi
+#===================================================================
+
+
+#==================== reboot to trigger commit data to file
+system sh/exec.sh -n dnode1 -s stop -x SIGINT
+system sh/exec.sh -n dnode1 -s start
+
+print =============== select * from retention level 2 from file
+sql select * from ct1;
+print $data00 $data01
+if $rows > 2 then
+  print retention level 2 file rows $rows > 2
+  return -1
+endi
+
+print =============== select * from retention level 1 from file
+sql select * from ct1 where ts > now-8d;
+print $data00 $data01
+if $rows > 2 then
+  print retention level 1 file rows $rows > 2
+  return -1
+endi
+
+print =============== select * from retention level 0 from file
+sql select * from ct1 where ts > now-3d;
+print $data00 $data01 
+print $data10 $data11
+print $data20 $data21
+if $rows < 1 then
+  print retention level 0 file rows $rows < 1
+  return -1
+endi
+
+system sh/exec.sh -n dnode1 -s stop -x SIGINT

tests/system-test/7-tmq/basic5.py

@@ -360,7 +360,7 @@ class TDTestCase:
         # wait db ready
         while 1:
             tdSql.query("show databases")
-            if tdSql.getRows() == 4: 
+            if tdSql.getRows() == 5: 
                 print (tdSql.getData(0,0), tdSql.getData(1,0),tdSql.getData(2,0),)           
                 break
             else:

tests/system-test/7-tmq/subscribeDb.py

@@ -251,7 +251,7 @@ class TDTestCase:
         expectrowcnt = parameterDict["rowsPerTbl"] * parameterDict["ctbNum"]
         topicList    = topicName1
         ifcheckdata  = 0
-        ifManualCommit = 0
+        ifManualCommit = 1
         keyList      = 'group.id:cgrp1,\
                         enable.auto.commit:false,\
                         auto.commit.interval.ms:6000,\
@@ -410,7 +410,7 @@ class TDTestCase:
         expectrowcnt = parameterDict["rowsPerTbl"] * parameterDict["ctbNum"] +  parameterDict2["rowsPerTbl"] * parameterDict2["ctbNum"]
         topicList    = topicName1
         ifcheckdata  = 0
-        ifManualCommit = 0
+        ifManualCommit = 1
         keyList      = 'group.id:cgrp1,\
                         enable.auto.commit:false,\
                         auto.commit.interval.ms:6000,\
@@ -487,7 +487,7 @@ class TDTestCase:
         expectrowcnt = parameterDict["rowsPerTbl"] * parameterDict["ctbNum"] +  parameterDict2["rowsPerTbl"] * parameterDict2["ctbNum"]
         topicList    = topicName1
         ifcheckdata  = 0
-        ifManualCommit = 0
+        ifManualCommit = 1
         keyList      = 'group.id:cgrp1,\
                         enable.auto.commit:false,\
                         auto.commit.interval.ms:6000,\
@@ -659,7 +659,7 @@ class TDTestCase:
         expectrowcnt = parameterDict["rowsPerTbl"] * parameterDict["ctbNum"] +  parameterDict2["rowsPerTbl"] * parameterDict2["ctbNum"]
         topicList    = topicName1 + ',' + topicName2
         ifcheckdata  = 0
-        ifManualCommit = 0
+        ifManualCommit = 1
         keyList      = 'group.id:cgrp1,\
                         enable.auto.commit:false,\
                         auto.commit.interval.ms:6000,\
@@ -708,13 +708,13 @@ class TDTestCase:
         cfgPath = buildPath + "/../sim/psim/cfg"
         tdLog.info("cfgPath: %s" % cfgPath)
 
-        #self.tmqCase1(cfgPath, buildPath)
+        self.tmqCase1(cfgPath, buildPath)
-        #self.tmqCase2(cfgPath, buildPath) 
+        self.tmqCase2(cfgPath, buildPath) 
-        #self.tmqCase3(cfgPath, buildPath)
+        self.tmqCase3(cfgPath, buildPath)
         self.tmqCase4(cfgPath, buildPath)
         self.tmqCase5(cfgPath, buildPath)
         self.tmqCase6(cfgPath, buildPath)
-        #self.tmqCase7(cfgPath, buildPath)
+        self.tmqCase7(cfgPath, buildPath)
 
 
     def stop(self):