diff --git a/.gitignore b/.gitignore
index d5c7f763cf..d7fcb019ae 100644
--- a/.gitignore
+++ b/.gitignore
@@ -22,6 +22,7 @@ mac/
 .mypy_cache
 *.tmp
 *.swp
+*.swo
 *.orig
 src/connector/nodejs/node_modules/
 src/connector/nodejs/out/
diff --git a/include/common/tcommon.h b/include/common/tcommon.h
index 9e7aea03ea..e1aadd4486 100644
--- a/include/common/tcommon.h
+++ b/include/common/tcommon.h
@@ -153,11 +153,10 @@ typedef struct SQueryTableDataCond {
   int32_t      order;  // desc|asc order to iterate the data block
   int32_t      numOfCols;
   SColumnInfo* colList;
-  int32_t      type;  // data block load type:
-                      //  int32_t      numOfTWindows;
-  STimeWindow twindows;
-  int64_t     startVersion;
-  int64_t     endVersion;
+  int32_t      type;   // data block load type:
+  STimeWindow  twindows;
+  int64_t      startVersion;
+  int64_t      endVersion;
 } SQueryTableDataCond;
 
 int32_t tEncodeDataBlock(void** buf, const SSDataBlock* pBlock);
diff --git a/include/libs/nodes/plannodes.h b/include/libs/nodes/plannodes.h
index e382fa4efd..ba16acf7b0 100644
--- a/include/libs/nodes/plannodes.h
+++ b/include/libs/nodes/plannodes.h
@@ -104,6 +104,7 @@ typedef struct SJoinLogicNode {
   SNode*     pMergeCondition;
   SNode*     pOnConditions;
   bool       isSingleTableJoin;
+  EOrder     inputTsOrder;
 } SJoinLogicNode;
 
 typedef struct SAggLogicNode {
@@ -201,6 +202,7 @@ typedef struct SWindowLogicNode {
   int64_t          watermark;
   int8_t           igExpired;
   EWindowAlgorithm windowAlgo;
+  EOrder           inputTsOrder;
 } SWindowLogicNode;
 
 typedef struct SFillLogicNode {
@@ -356,15 +358,14 @@ typedef struct SInterpFuncPhysiNode {
   SNode*      pTimeSeries;  // SColumnNode
 } SInterpFuncPhysiNode;
 
-typedef struct SJoinPhysiNode {
+typedef struct SSortMergeJoinPhysiNode {
   SPhysiNode node;
   EJoinType  joinType;
   SNode*     pMergeCondition;
   SNode*     pOnConditions;
   SNodeList* pTargets;
-} SJoinPhysiNode;
-
-typedef SJoinPhysiNode SSortMergeJoinPhysiNode;
+  EOrder     inputTsOrder;
+} SSortMergeJoinPhysiNode;
 
 typedef struct SAggPhysiNode {
   SPhysiNode node;
diff --git a/include/libs/nodes/querynodes.h b/include/libs/nodes/querynodes.h
index f8c7024591..81ed5b5ecd 100644
--- a/include/libs/nodes/querynodes.h
+++ b/include/libs/nodes/querynodes.h
@@ -255,6 +255,7 @@ typedef struct SSelectStmt {
   int32_t     selectFuncNum;
   bool        isEmptyResult;
   bool        isTimeLineResult;
+  bool        isSubquery;
   bool        hasAggFuncs;
   bool        hasRepeatScanFuncs;
   bool        hasIndefiniteRowsFunc;
diff --git a/include/libs/wal/wal.h b/include/libs/wal/wal.h
index 907b3be560..220c4f73e0 100644
--- a/include/libs/wal/wal.h
+++ b/include/libs/wal/wal.h
@@ -114,21 +114,30 @@ typedef struct SWal {
   int64_t       refId;
   TdThreadMutex mutex;
   // ref
-  SHashObj *pRefHash;  // ref -> SWalRef
+  SHashObj *pRefHash;  // refId -> SWalRef
   // path
   char path[WAL_PATH_LEN];
   // reusable write head
   SWalCkHead writeHead;
-} SWal;  // WAL HANDLE
+} SWal;
+
+typedef struct {
+  int64_t refId;
+  int64_t refVer;
+  int64_t refFile;
+  SWal   *pWal;
+} SWalRef;
 
 typedef struct {
   int8_t scanUncommited;
+  int8_t scanNotApplied;
   int8_t scanMeta;
   int8_t enableRef;
 } SWalFilterCond;
 
 typedef struct {
   SWal          *pWal;
+  int64_t        readerId;
   TdFilePtr      pLogFile;
   TdFilePtr      pIdxFile;
   int64_t        curFileFirstVer;
@@ -138,7 +147,8 @@ typedef struct {
   int8_t         curStopped;
   TdThreadMutex  mutex;
   SWalFilterCond cond;
-  SWalCkHead    *pHead;
+  // TODO remove it
+  SWalCkHead *pHead;
 } SWalReader;
 
 // module initialization
@@ -157,11 +167,7 @@ int32_t walWrite(SWal *, int64_t index, tmsg_t msgType, const void *body, int32_
 int32_t walWriteWithSyncInfo(SWal *, int64_t index, tmsg_t msgType, SWalSyncInfo syncMeta, const void *body,
                              int32_t bodyLen);
 
-// This interface assign version automatically and return to caller.
-// When using this interface with concurrent writes,
-// wal will write all logs atomically,
-// but not sure which one will be actually write first,
-// and then the unique index of successful writen is returned.
+// Assign version automatically and return to caller,
 // -1 will be returned for failed writes
 int64_t walAppendLog(SWal *, tmsg_t msgType, SWalSyncInfo syncMeta, const void *body, int32_t bodyLen);
 
@@ -191,17 +197,15 @@ void    walSetReaderCapacity(SWalReader *pRead, int32_t capacity);
 int32_t walFetchHead(SWalReader *pRead, int64_t ver, SWalCkHead *pHead);
 int32_t walFetchBody(SWalReader *pRead, SWalCkHead **ppHead);
 int32_t walSkipFetchBody(SWalReader *pRead, const SWalCkHead *pHead);
-typedef struct {
-  int64_t refId;
-  int64_t ver;
-} SWalRef;
+
+SWalRef *walRefCommittedVer(SWal *);
 
 SWalRef *walOpenRef(SWal *);
-void     walCloseRef(SWalRef *);
+void     walCloseRef(SWal *pWal, int64_t refId);
 int32_t  walRefVer(SWalRef *, int64_t ver);
-int32_t  walUnrefVer(SWal *);
+void     walUnrefVer(SWalRef *);
 
-// help function for raft
+// helper function for raft
 bool walLogExist(SWal *, int64_t ver);
 bool walIsEmpty(SWal *);
 
diff --git a/source/common/src/tglobal.c b/source/common/src/tglobal.c
index a73808f2ed..36900e3dfa 100644
--- a/source/common/src/tglobal.c
+++ b/source/common/src/tglobal.c
@@ -213,10 +213,6 @@ static int32_t taosSetTfsCfg(SConfig *pCfg) {
       memcpy(&tsDiskCfg[index], pCfg, sizeof(SDiskCfg));
       if (pCfg->level == 0 && pCfg->primary == 1) {
         tstrncpy(tsDataDir, pCfg->dir, PATH_MAX);
-        if (taosMulMkDir(tsDataDir) != 0) {
-          uError("failed to create dataDir:%s since %s", tsDataDir, terrstr());
-          return -1;
-        }
       }
       if (taosMulMkDir(pCfg->dir) != 0) {
         uError("failed to create tfsDir:%s since %s", tsDataDir, terrstr());
@@ -227,12 +223,13 @@ static int32_t taosSetTfsCfg(SConfig *pCfg) {
 
   if (tsDataDir[0] == 0) {
     if (pItem->str != NULL) {
-      taosAddDataDir(0, pItem->str, 0, 1);
+      taosAddDataDir(tsDiskCfgNum, pItem->str, 0, 1);
       tstrncpy(tsDataDir, pItem->str, PATH_MAX);
       if (taosMulMkDir(tsDataDir) != 0) {
-        uError("failed to create dataDir:%s since %s", tsDataDir, terrstr());
+        uError("failed to create tfsDir:%s since %s", tsDataDir, terrstr());
         return -1;
       }
+      tsDiskCfgNum++;
     } else {
       uError("datadir not set");
       return -1;
diff --git a/source/dnode/mnode/impl/src/mndProfile.c b/source/dnode/mnode/impl/src/mndProfile.c
index 5871d56a8f..f6ecd4493d 100644
--- a/source/dnode/mnode/impl/src/mndProfile.c
+++ b/source/dnode/mnode/impl/src/mndProfile.c
@@ -312,7 +312,7 @@ static int32_t mndSaveQueryList(SConnObj *pConn, SQueryHbReqBasic *pBasic) {
   pConn->numOfQueries = pBasic->queryDesc ? taosArrayGetSize(pBasic->queryDesc) : 0;
   pBasic->queryDesc = NULL;
 
-  mDebug("queries updated in conn %d, num:%d", pConn->id, pConn->numOfQueries);
+  mDebug("queries updated in conn %u, num:%d", pConn->id, pConn->numOfQueries);
 
   taosWUnLockLatch(&pConn->queryLock);
 
diff --git a/source/dnode/vnode/inc/vnode.h b/source/dnode/vnode/inc/vnode.h
index 6d976083d2..4d95a9d7a5 100644
--- a/source/dnode/vnode/inc/vnode.h
+++ b/source/dnode/vnode/inc/vnode.h
@@ -118,9 +118,8 @@ int32_t     metaTbCursorNext(SMTbCursor *pTbCur);
 // typedef struct STsdb STsdb;
 typedef struct STsdbReader STsdbReader;
 
-#define BLOCK_LOAD_OFFSET_ORDER   1
-#define BLOCK_LOAD_TABLESEQ_ORDER 2
-#define BLOCK_LOAD_EXTERN_ORDER   3
+#define TIMEWINDOW_RANGE_CONTAINED 1
+#define TIMEWINDOW_RANGE_EXTERNAL  2
 
 #define LASTROW_RETRIEVE_TYPE_ALL    0x1
 #define LASTROW_RETRIEVE_TYPE_SINGLE 0x2
diff --git a/source/dnode/vnode/src/inc/tq.h b/source/dnode/vnode/src/inc/tq.h
index b063e552f6..262300a3e7 100644
--- a/source/dnode/vnode/src/inc/tq.h
+++ b/source/dnode/vnode/src/inc/tq.h
@@ -104,6 +104,8 @@ typedef struct {
   // TODO remove
   SWalReader* pWalReader;
 
+  SWalRef* pRef;
+
   // push
   STqPushHandle pushHandle;
 
diff --git a/source/dnode/vnode/src/tq/tq.c b/source/dnode/vnode/src/tq/tq.c
index 8ea4b8d98b..118e3a5d43 100644
--- a/source/dnode/vnode/src/tq/tq.c
+++ b/source/dnode/vnode/src/tq/tq.c
@@ -212,6 +212,15 @@ int32_t tqProcessOffsetCommitReq(STQ* pTq, char* msg, int32_t msgLen) {
     ASSERT(0);
     return -1;
   }
+
+  if (offset.val.type == TMQ_OFFSET__LOG) {
+    STqHandle* pHandle = taosHashGet(pTq->handles, offset.subKey, strlen(offset.subKey));
+    if (walRefVer(pHandle->pRef, offset.val.version) < 0) {
+      ASSERT(0);
+      return -1;
+    }
+  }
+
   /*}*/
   /*}*/
 
@@ -376,8 +385,8 @@ int32_t tqProcessPollReq(STQ* pTq, SRpcMsg* pMsg) {
   }
 
   if (pHandle->execHandle.subType != TOPIC_SUB_TYPE__COLUMN) {
-    int64_t     fetchVer = fetchOffsetNew.version + 1;
-    SWalCkHead* pCkHead = taosMemoryMalloc(sizeof(SWalCkHead) + 2048);
+    int64_t fetchVer = fetchOffsetNew.version + 1;
+    pCkHead = taosMemoryMalloc(sizeof(SWalCkHead) + 2048);
     if (pCkHead == NULL) {
       code = -1;
       goto OVER;
@@ -534,11 +543,14 @@ int32_t tqProcessVgChangeReq(STQ* pTq, char* msg, int32_t msgLen) {
 
     pHandle->execHandle.subType = req.subType;
     pHandle->fetchMeta = req.withMeta;
+    // TODO version should be assigned and refed during preprocess
+    SWalRef* pRef = walRefCommittedVer(pTq->pVnode->pWal);
+    if (pRef == NULL) {
+      ASSERT(0);
+    }
+    int64_t ver = pRef->refVer;
+    pHandle->pRef = pRef;
 
-    pHandle->pWalReader = walOpenReader(pTq->pVnode->pWal, NULL);
-
-    // TODO version should be assigned in preprocess
-    int64_t ver = walGetCommittedVer(pTq->pVnode->pWal);
     if (pHandle->execHandle.subType == TOPIC_SUB_TYPE__COLUMN) {
       pHandle->execHandle.execCol.qmsg = req.qmsg;
       pHandle->snapshotVer = ver;
@@ -560,10 +572,14 @@ int32_t tqProcessVgChangeReq(STQ* pTq, char* msg, int32_t msgLen) {
       pHandle->execHandle.pExecReader = qExtractReaderFromStreamScanner(scanner);
       ASSERT(pHandle->execHandle.pExecReader);
     } else if (pHandle->execHandle.subType == TOPIC_SUB_TYPE__DB) {
+      pHandle->pWalReader = walOpenReader(pTq->pVnode->pWal, NULL);
+
       pHandle->execHandle.pExecReader = tqOpenReader(pTq->pVnode);
       pHandle->execHandle.execDb.pFilterOutTbUid =
           taosHashInit(64, taosGetDefaultHashFunction(TSDB_DATA_TYPE_BIGINT), false, HASH_NO_LOCK);
     } else if (pHandle->execHandle.subType == TOPIC_SUB_TYPE__TABLE) {
+      pHandle->pWalReader = walOpenReader(pTq->pVnode->pWal, NULL);
+
       pHandle->execHandle.execTb.suid = req.suid;
       SArray* tbUidList = taosArrayInit(0, sizeof(int64_t));
       vnodeGetCtbIdList(pTq->pVnode, req.suid, tbUidList);
diff --git a/source/dnode/vnode/src/tq/tqMeta.c b/source/dnode/vnode/src/tq/tqMeta.c
index 620417016f..290ffe5c8d 100644
--- a/source/dnode/vnode/src/tq/tqMeta.c
+++ b/source/dnode/vnode/src/tq/tqMeta.c
@@ -52,7 +52,7 @@ int32_t tqMetaOpen(STQ* pTq) {
     ASSERT(0);
   }
 
-  TXN txn;
+  TXN txn = {0};
 
   if (tdbTxnOpen(&txn, 0, tdbDefaultMalloc, tdbDefaultFree, NULL, 0) < 0) {
     ASSERT(0);
@@ -75,7 +75,13 @@ int32_t tqMetaOpen(STQ* pTq) {
     STqHandle handle;
     tDecoderInit(&decoder, (uint8_t*)pVal, vLen);
     tDecodeSTqHandle(&decoder, &handle);
-    handle.pWalReader = walOpenReader(pTq->pVnode->pWal, NULL);
+
+    handle.pRef = walOpenRef(pTq->pVnode->pWal);
+    if (handle.pRef == NULL) {
+      ASSERT(0);
+    }
+    walRefVer(handle.pRef, handle.snapshotVer);
+
     if (handle.execHandle.subType == TOPIC_SUB_TYPE__COLUMN) {
       SReadHandle reader = {
           .meta = pTq->pVnode->pMeta,
@@ -94,6 +100,7 @@ int32_t tqMetaOpen(STQ* pTq) {
       handle.execHandle.pExecReader = qExtractReaderFromStreamScanner(scanner);
       ASSERT(handle.execHandle.pExecReader);
     } else {
+      handle.pWalReader = walOpenReader(pTq->pVnode->pWal, NULL);
       handle.execHandle.execDb.pFilterOutTbUid =
           taosHashInit(64, taosGetDefaultHashFunction(TSDB_DATA_TYPE_BIGINT), false, HASH_NO_LOCK);
     }
diff --git a/source/dnode/vnode/src/tsdb/tsdbRead.c b/source/dnode/vnode/src/tsdb/tsdbRead.c
index cec714e0ee..ea8ac09429 100644
--- a/source/dnode/vnode/src/tsdb/tsdbRead.c
+++ b/source/dnode/vnode/src/tsdb/tsdbRead.c
@@ -16,6 +16,12 @@
 #include "tsdb.h"
 #define ASCENDING_TRAVERSE(o) (o == TSDB_ORDER_ASC)
 
+typedef enum {
+  EXTERNAL_ROWS_PREV = 0x1,
+  EXTERNAL_ROWS_MAIN = 0x2,
+  EXTERNAL_ROWS_NEXT = 0x3,
+} EContentData;
+
 typedef struct {
   STbDataIter* iter;
   int32_t      index;
@@ -70,9 +76,9 @@ typedef struct SFilesetIter {
 } SFilesetIter;
 
 typedef struct SFileDataBlockInfo {
-  int32_t
-      tbBlockIdx;  // index position in STableBlockScanInfo in order to check whether neighbor block overlaps with it
+  // index position in STableBlockScanInfo in order to check whether neighbor block overlaps with it
   uint64_t uid;
+  int32_t tbBlockIdx;
 } SFileDataBlockInfo;
 
 typedef struct SDataBlockIter {
@@ -99,12 +105,11 @@ typedef struct SReaderStatus {
   SHashObj*            pTableMap;     // SHash<STableBlockScanInfo>
   STableBlockScanInfo* pTableIter;    // table iterator used in building in-memory buffer data blocks.
   SFileBlockDumpInfo   fBlockDumpInfo;
-
-  SDFileSet*     pCurrentFileset;  // current opened file set
-  SBlockData     fileBlockData;
-  SFilesetIter   fileIter;
-  SDataBlockIter blockIter;
-  bool           composedDataBlock;  // the returned data block is a composed block or not
+  SDFileSet*           pCurrentFileset;  // current opened file set
+  SBlockData           fileBlockData;
+  SFilesetIter         fileIter;
+  SDataBlockIter       blockIter;
+  bool                 composedDataBlock;  // the returned data block is a composed block or not
 } SReaderStatus;
 
 struct STsdbReader {
@@ -115,15 +120,17 @@ struct STsdbReader {
   SSDataBlock*       pResBlock;
   int32_t            capacity;
   SReaderStatus      status;
-  char*              idStr;  // query info handle, for debug purpose
-  int32_t            type;   // query type: 1. retrieve all data blocks, 2. retrieve direct prev|next rows
+  char*              idStr;   // query info handle, for debug purpose
+  int32_t            type;    // query type: 1. retrieve all data blocks, 2. retrieve direct prev|next rows
   SBlockLoadSuppInfo suppInfo;
   STsdbReadSnap*     pReadSnap;
+  SIOCostSummary     cost;
+  STSchema*          pSchema;
+  SDataFReader*      pFileReader;
+  SVersionRange      verRange;
 
-  SIOCostSummary cost;
-  STSchema*      pSchema;
-  SDataFReader*  pFileReader;
-  SVersionRange  verRange;
+  int32_t            step;
+  STsdbReader*       innerReader[2];
 };
 
 static SFileDataBlockInfo* getCurrentBlockInfo(SDataBlockIter* pBlockIter);
@@ -200,6 +207,9 @@ static SHashObj* createDataBlockScanInfo(STsdbReader* pTsdbReader, const STableK
               pTsdbReader->idStr);
   }
 
+  tsdbDebug("%p create %d tables scan-info, size:%.2f Kb, %s", pTsdbReader, numOfTables, (sizeof(STableBlockScanInfo)*numOfTables)/1024.0,
+      pTsdbReader->idStr);
+
   return pTableMap;
 }
 
@@ -328,7 +338,7 @@ static bool filesetIteratorNext(SFilesetIter* pIter, STsdbReader* pReader) {
       continue;
     }
 
-    tsdbDebug("%p file found fid:%d for qrange:%" PRId64 "-%" PRId64 ", ignore, %s", pReader, fid, pReader->window.skey,
+    tsdbDebug("%p file found fid:%d for qrange:%" PRId64 "-%" PRId64 ", %s", pReader, fid, pReader->window.skey,
               pReader->window.ekey, pReader->idStr);
     return true;
   }
@@ -378,7 +388,7 @@ static SSDataBlock* createResBlock(SQueryTableDataCond* pCond, int32_t capacity)
   return pResBlock;
 }
 
-static int32_t tsdbReaderCreate(SVnode* pVnode, SQueryTableDataCond* pCond, STsdbReader** ppReader, const char* idstr) {
+static int32_t tsdbReaderCreate(SVnode* pVnode, SQueryTableDataCond* pCond, STsdbReader** ppReader, int32_t capacity, const char* idstr) {
   int32_t      code = 0;
   int8_t       level = 0;
   STsdbReader* pReader = (STsdbReader*)taosMemoryCalloc(1, sizeof(*pReader));
@@ -392,7 +402,7 @@ static int32_t tsdbReaderCreate(SVnode* pVnode, SQueryTableDataCond* pCond, STsd
   pReader->pTsdb = getTsdbByRetentions(pVnode, pCond->twindows.skey, pVnode->config.tsdbCfg.retentions, idstr, &level);
   pReader->suid = pCond->suid;
   pReader->order = pCond->order;
-  pReader->capacity = 4096;
+  pReader->capacity = capacity;
   pReader->idStr = (idstr != NULL) ? strdup(idstr) : NULL;
   pReader->verRange = getQueryVerRange(pVnode, pCond, level);
   pReader->type = pCond->type;
@@ -483,95 +493,6 @@ _end:
 //   return res;
 // }
 
-// static TSKEY extractFirstTraverseKey(STableBlockScanInfo* pCheckInfo, int32_t order, int32_t update, TDRowVerT
-// maxVer) {
-//   TSDBROW row = {0};
-//   STSRow *rmem = NULL, *rimem = NULL;
-
-//   if (pCheckInfo->iter) {
-//     if (tsdbTbDataIterGet(pCheckInfo->iter, &row)) {
-//       rmem = row.pTSRow;
-//     }
-//   }
-
-//   if (pCheckInfo->iiter) {
-//     if (tsdbTbDataIterGet(pCheckInfo->iiter, &row)) {
-//       rimem = row.pTSRow;
-//     }
-//   }
-
-//   if (rmem == NULL && rimem == NULL) {
-//     return TSKEY_INITIAL_VAL;
-//   }
-
-//   if (rmem != NULL && rimem == NULL) {
-//     pCheckInfo->chosen = CHECKINFO_CHOSEN_MEM;
-//     return TD_ROW_KEY(rmem);
-//   }
-
-//   if (rmem == NULL && rimem != NULL) {
-//     pCheckInfo->chosen = CHECKINFO_CHOSEN_IMEM;
-//     return TD_ROW_KEY(rimem);
-//   }
-
-//   TSKEY r1 = TD_ROW_KEY(rmem);
-//   TSKEY r2 = TD_ROW_KEY(rimem);
-
-//   if (r1 == r2) {
-//     if (TD_SUPPORT_UPDATE(update)) {
-//       pCheckInfo->chosen = CHECKINFO_CHOSEN_BOTH;
-//     } else {
-//       pCheckInfo->chosen = CHECKINFO_CHOSEN_IMEM;
-//       tsdbTbDataIterNext(pCheckInfo->iter);
-//     }
-//     return r1;
-//   } else if (r1 < r2 && ASCENDING_TRAVERSE(order)) {
-//     pCheckInfo->chosen = CHECKINFO_CHOSEN_MEM;
-//     return r1;
-//   } else {
-//     pCheckInfo->chosen = CHECKINFO_CHOSEN_IMEM;
-//     return r2;
-//   }
-// }
-
-// static bool moveToNextRowInMem(STableBlockScanInfo* pCheckInfo) {
-//   bool hasNext = false;
-//   if (pCheckInfo->chosen == CHECKINFO_CHOSEN_MEM) {
-//     if (pCheckInfo->iter != NULL) {
-//       hasNext = tsdbTbDataIterNext(pCheckInfo->iter);
-//     }
-
-//     if (hasNext) {
-//       return hasNext;
-//     }
-
-//     if (pCheckInfo->iiter != NULL) {
-//       return tsdbTbDataIterGet(pCheckInfo->iiter, NULL);
-//     }
-//   } else if (pCheckInfo->chosen == CHECKINFO_CHOSEN_IMEM) {
-//     if (pCheckInfo->iiter != NULL) {
-//       hasNext = tsdbTbDataIterNext(pCheckInfo->iiter);
-//     }
-
-//     if (hasNext) {
-//       return hasNext;
-//     }
-
-//     if (pCheckInfo->iter != NULL) {
-//       return tsdbTbDataIterGet(pCheckInfo->iter, NULL);
-//     }
-//   } else {
-//     if (pCheckInfo->iter != NULL) {
-//       hasNext = tsdbTbDataIterNext(pCheckInfo->iter);
-//     }
-//     if (pCheckInfo->iiter != NULL) {
-//       hasNext = tsdbTbDataIterNext(pCheckInfo->iiter) || hasNext;
-//     }
-//   }
-
-//   return hasNext;
-// }
-
 // static int32_t binarySearchForBlock(SBlock* pBlock, int32_t numOfBlocks, TSKEY skey, int32_t order) {
 //   int32_t firstSlot = 0;
 //   int32_t lastSlot = numOfBlocks - 1;
@@ -602,18 +523,22 @@ _end:
 static int32_t doLoadBlockIndex(STsdbReader* pReader, SDataFReader* pFileReader, SArray* pIndexList) {
   SArray* aBlockIdx = taosArrayInit(0, sizeof(SBlockIdx));
 
+  int64_t st = taosGetTimestampUs();
   int32_t code = tsdbReadBlockIdx(pFileReader, aBlockIdx, NULL);
   if (code != TSDB_CODE_SUCCESS) {
     goto _end;
   }
 
-  if (taosArrayGetSize(aBlockIdx) == 0) {
+  size_t num = taosArrayGetSize(aBlockIdx);
+  if (num == 0) {
     taosArrayClear(aBlockIdx);
     return TSDB_CODE_SUCCESS;
   }
 
-  SBlockIdx* pBlockIdx;
-  for (int32_t i = 0; i < taosArrayGetSize(aBlockIdx); ++i) {
+  int64_t et1 = taosGetTimestampUs();
+
+  SBlockIdx* pBlockIdx = NULL;
+  for (int32_t i = 0; i < num; ++i) {
     pBlockIdx = (SBlockIdx*)taosArrayGet(aBlockIdx, i);
 
     // uid check
@@ -627,17 +552,6 @@ static int32_t doLoadBlockIndex(STsdbReader* pReader, SDataFReader* pFileReader,
       continue;
     }
 
-    // todo: not valid info in bockIndex
-    // time range check
-    //    if (pBlockIdx->minKey > pReader->window.ekey || pBlockIdx->maxKey < pReader->window.skey) {
-    //      continue;
-    //    }
-
-    // version check
-    //    if (pBlockIdx->minVersion > pReader->verRange.maxVer || pBlockIdx->maxVersion < pReader->verRange.minVer) {
-    //      continue;
-    //    }
-
     STableBlockScanInfo* pScanInfo = p;
     if (pScanInfo->pBlockList == NULL) {
       pScanInfo->pBlockList = taosArrayInit(16, sizeof(SBlock));
@@ -647,6 +561,9 @@ static int32_t doLoadBlockIndex(STsdbReader* pReader, SDataFReader* pFileReader,
     taosArrayPush(pIndexList, pBlockIdx);
   }
 
+  int64_t et2 = taosGetTimestampUs();
+  tsdbDebug("load block index for %d tables completed, elapsed time:%.2f ms, set blockIdx:%.2f ms, size:%d bytes %s",
+            (int32_t)num, (et1 - st)/1000.0, (et2-et1)/1000.0, num * sizeof(SBlockIdx), pReader->idStr);
 _end:
   taosArrayDestroy(aBlockIdx);
   return code;
@@ -655,9 +572,11 @@ _end:
 static int32_t doLoadFileBlock(STsdbReader* pReader, SArray* pIndexList, uint32_t* numOfValidTables,
                                int32_t* numOfBlocks) {
   size_t numOfTables = taosArrayGetSize(pIndexList);
-
   *numOfValidTables = 0;
 
+  int64_t st = taosGetTimestampUs();
+  size_t size = 0;
+
   STableBlockScanInfo* px = NULL;
   while (1) {
     px = taosHashIterate(pReader->status.pTableMap, px);
@@ -675,6 +594,8 @@ static int32_t doLoadFileBlock(STsdbReader* pReader, SArray* pIndexList, uint32_
     tMapDataReset(&mapData);
     tsdbReadBlock(pReader->pFileReader, pBlockIdx, &mapData, NULL);
 
+    size += mapData.nData;
+
     STableBlockScanInfo* pScanInfo = taosHashGet(pReader->status.pTableMap, &pBlockIdx->uid, sizeof(int64_t));
     for (int32_t j = 0; j < mapData.nItem; ++j) {
       SBlock block = {0};
@@ -706,6 +627,10 @@ static int32_t doLoadFileBlock(STsdbReader* pReader, SArray* pIndexList, uint32_
     }
   }
 
+  int64_t et = taosGetTimestampUs();
+  tsdbDebug("load block of %d tables completed, blocks:%d in %d tables, size:%.2f Kb, elapsed time:%.2f ms %s",
+      numOfTables, *numOfBlocks, *numOfValidTables, size/1000.0, (et-st)/1000.0, pReader->idStr);
+
   return TSDB_CODE_SUCCESS;
 }
 
@@ -816,7 +741,6 @@ static int32_t copyBlockDataToSDataBlock(STsdbReader* pReader, STableBlockScanIn
   return TSDB_CODE_SUCCESS;
 }
 
-// todo consider the output buffer size
 static int32_t doLoadFileBlockData(STsdbReader* pReader, SDataBlockIter* pBlockIter,
                                    STableBlockScanInfo* pBlockScanInfo, SBlockData* pBlockData) {
   int64_t st = taosGetTimestampUs();
@@ -853,346 +777,6 @@ _error:
   return code;
 }
 
-// static int doBinarySearchKey(char* pValue, int num, TSKEY key, int order) {
-//   int    firstPos, lastPos, midPos = -1;
-//   int    numOfRows;
-//   TSKEY* keyList;
-
-//   assert(order == TSDB_ORDER_ASC || order == TSDB_ORDER_DESC);
-
-//   if (num <= 0) return -1;
-
-//   keyList = (TSKEY*)pValue;
-//   firstPos = 0;
-//   lastPos = num - 1;
-
-//   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 - 1;
-
-//       numOfRows = lastPos - firstPos + 1;
-//       midPos = (numOfRows >> 1) + firstPos;
-
-//       if (key < keyList[midPos]) {
-//         lastPos = midPos - 1;
-//       } else if (key > keyList[midPos]) {
-//         firstPos = midPos + 1;
-//       } else {
-//         break;
-//       }
-//     }
-
-//   } else {
-//     // find the first position which is bigger than the key
-//     while (1) {
-//       if (key <= keyList[firstPos]) return firstPos;
-//       if (key == keyList[lastPos]) return lastPos;
-
-//       if (key > keyList[lastPos]) {
-//         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 {
-//         break;
-//       }
-//     }
-//   }
-
-//   return midPos;
-// }
-
-// static void doCheckGeneratedBlockRange(STsdbReader* pTsdbReadHandle) {
-//   SQueryFilePos* cur = &pTsdbReadHandle->cur;
-
-//   if (cur->rows > 0) {
-//     if (ASCENDING_TRAVERSE(pTsdbReadHandle->order)) {
-//       assert(cur->win.skey >= pTsdbReadHandle->window.skey && cur->win.ekey <= pTsdbReadHandle->window.ekey);
-//     } else {
-//       assert(cur->win.skey >= pTsdbReadHandle->window.ekey && cur->win.ekey <= pTsdbReadHandle->window.skey);
-//     }
-
-//     SColumnInfoData* pColInfoData = taosArrayGet(pTsdbReadHandle->pColumns, 0);
-//     assert(cur->win.skey == ((TSKEY*)pColInfoData->pData)[0] &&
-//            cur->win.ekey == ((TSKEY*)pColInfoData->pData)[cur->rows - 1]);
-//   } else {
-//     cur->win = pTsdbReadHandle->window;
-
-//     int32_t step = ASCENDING_TRAVERSE(pTsdbReadHandle->order) ? 1 : -1;
-//     cur->lastKey = pTsdbReadHandle->window.ekey + step;
-//   }
-// }
-
-// static void copyAllRemainRowsFromFileBlock(STsdbReader* pTsdbReadHandle, STableBlockScanInfo* pCheckInfo,
-//                                            SDataBlockInfo* pBlockInfo, int32_t endPos) {
-//   SQueryFilePos* cur = &pTsdbReadHandle->cur;
-
-//   SDataCols* pCols = pTsdbReadHandle->rhelper.pDCols[0];
-//   TSKEY*     tsArray = pCols->cols[0].pData;
-
-//   bool ascScan = ASCENDING_TRAVERSE(pTsdbReadHandle->order);
-
-//   int32_t step = ascScan ? 1 : -1;
-
-//   int32_t start = cur->pos;
-//   int32_t end = endPos;
-
-//   if (!ascScan) {
-//     TSWAP(start, end);
-//   }
-
-//   assert(pTsdbReadHandle->outputCapacity >= (end - start + 1));
-//   int32_t numOfRows = doCopyRowsFromFileBlock(pTsdbReadHandle, pTsdbReadHandle->outputCapacity, 0, start, end);
-
-//   // the time window should always be ascending order: skey <= ekey
-//   cur->win = (STimeWindow){.skey = tsArray[start], .ekey = tsArray[end]};
-//   cur->mixBlock = (numOfRows != pBlockInfo->rows);
-//   cur->lastKey = tsArray[endPos] + step;
-//   cur->blockCompleted = (ascScan ? (endPos == pBlockInfo->rows - 1) : (endPos == 0));
-
-//   // The value of pos may be -1 or pBlockInfo->rows, and it is invalid in both cases.
-//   int32_t pos = endPos + step;
-//   updateInfoAfterMerge(pTsdbReadHandle, pCheckInfo, numOfRows, pos);
-//   doCheckGeneratedBlockRange(pTsdbReadHandle);
-
-//   tsdbDebug("%p uid:%" PRIu64 ", data block created, mixblock:%d, brange:%" PRIu64 "-%" PRIu64 " rows:%d, %s",
-//             pTsdbReadHandle, pCheckInfo->tableId, cur->mixBlock, cur->win.skey, cur->win.ekey, cur->rows,
-//             pTsdbReadHandle->idStr);
-// }
-
-// // only return the qualified data to client in terms of query time window, data rows in the same block but do not
-// // be included in the query time window will be discarded
-// static void doMergeTwoLevelData(STsdbReader* pTsdbReadHandle, STableBlockScanInfo* pCheckInfo, SBlock* pBlock) {
-//   SQueryFilePos* cur = &pTsdbReadHandle->cur;
-//   SDataBlockInfo blockInfo = GET_FILE_DATA_BLOCK_INFO(pCheckInfo, pBlock);
-//   STsdbCfg*      pCfg = REPO_CFG(pTsdbReadHandle->pTsdb);
-
-//   initTableMemIterator(pTsdbReadHandle, pCheckInfo);
-
-//   SDataCols* pCols = pTsdbReadHandle->rhelper.pDCols[0];
-//   assert(pCols->cols[0].type == TSDB_DATA_TYPE_TIMESTAMP && pCols->cols[0].colId == PRIMARYKEY_TIMESTAMP_COL_ID &&
-//          cur->pos >= 0 && cur->pos < pBlock->numOfRows);
-//   // Even Multi-Version supported, the records with duplicated TSKEY would be merged inside of tsdbLoadData
-//   interface. TSKEY* tsArray = pCols->cols[0].pData; assert(pCols->numOfRows == pBlock->numOfRows && tsArray[0] ==
-//   pBlock->minKey.ts &&
-//          tsArray[pBlock->numOfRows - 1] == pBlock->maxKey.ts);
-
-//   bool    ascScan = ASCENDING_TRAVERSE(pTsdbReadHandle->order);
-//   int32_t step = ascScan ? 1 : -1;
-
-//   // for search the endPos, so the order needs to reverse
-//   int32_t order = ascScan ? TSDB_ORDER_DESC : TSDB_ORDER_ASC;
-
-//   int32_t numOfCols = (int32_t)(QH_GET_NUM_OF_COLS(pTsdbReadHandle));
-//   int32_t endPos = getEndPosInDataBlock(pTsdbReadHandle, &blockInfo);
-
-//   STimeWindow* pWin = &blockInfo.window;
-//   tsdbDebug("%p uid:%" PRIu64 " start merge data block, file block range:%" PRIu64 "-%" PRIu64
-//             " rows:%d, start:%d, end:%d, %s",
-//             pTsdbReadHandle, pCheckInfo->tableId, pWin->skey, pWin->ekey, blockInfo.rows, cur->pos, endPos,
-//             pTsdbReadHandle->idStr);
-
-//   // compared with the data from in-memory buffer, to generate the correct timestamp array list
-//   int32_t numOfRows = 0;
-//   int32_t curRow = 0;
-
-//   int16_t   rv1 = -1;
-//   int16_t   rv2 = -1;
-//   STSchema* pSchema1 = NULL;
-//   STSchema* pSchema2 = NULL;
-
-//   int32_t pos = cur->pos;
-//   cur->win = TSWINDOW_INITIALIZER;
-//   bool adjustPos = false;
-
-//   // no data in buffer, load data from file directly
-//   if (pCheckInfo->iiter == NULL && pCheckInfo->iter == NULL) {
-//     copyAllRemainRowsFromFileBlock(pTsdbReadHandle, pCheckInfo, &blockInfo, endPos);
-//     return;
-//   } else if (pCheckInfo->iter != NULL || pCheckInfo->iiter != NULL) {
-//     SSkipListNode* node = NULL;
-//     TSKEY          lastKeyAppend = TSKEY_INITIAL_VAL;
-
-//     do {
-//       STSRow* row2 = NULL;
-//       STSRow* row1 = getSRowInTableMem(pCheckInfo, pTsdbReadHandle->order, pCfg->update, &row2, TD_VER_MAX);
-//       if (row1 == NULL) {
-//         break;
-//       }
-
-//       TSKEY key = TD_ROW_KEY(row1);
-//       if ((key > pTsdbReadHandle->window.ekey && ascScan) || (key < pTsdbReadHandle->window.ekey && !ascScan)) {
-//         break;
-//       }
-
-//       if (adjustPos) {
-//         if (key == lastKeyAppend) {
-//           pos -= step;
-//         }
-//         adjustPos = false;
-//       }
-
-//       if (((pos > endPos || tsArray[pos] > pTsdbReadHandle->window.ekey) && ascScan) ||
-//           ((pos < endPos || tsArray[pos] < pTsdbReadHandle->window.ekey) && !ascScan)) {
-//         break;
-//       }
-
-//       if ((key < tsArray[pos] && ascScan) || (key > tsArray[pos] && !ascScan)) {
-//         if (rv1 != TD_ROW_SVER(row1)) {
-//           //          pSchema1 = tsdbGetTableSchemaByVersion(pTable, memRowVersion(row1));
-//           rv1 = TD_ROW_SVER(row1);
-//         }
-//         if (row2 && rv2 != TD_ROW_SVER(row2)) {
-//           //          pSchema2 = tsdbGetTableSchemaByVersion(pTable, memRowVersion(row2));
-//           rv2 = TD_ROW_SVER(row2);
-//         }
-
-//         numOfRows +=
-//             mergeTwoRowFromMem(pTsdbReadHandle, pTsdbReadHandle->outputCapacity, &curRow, row1, row2, numOfCols,
-//                                pCheckInfo->tableId, pSchema1, pSchema2, pCfg->update, &lastKeyAppend);
-//         if (cur->win.skey == TSKEY_INITIAL_VAL) {
-//           cur->win.skey = key;
-//         }
-
-//         cur->win.ekey = key;
-//         cur->lastKey = key + step;
-//         cur->mixBlock = true;
-//         moveToNextRowInMem(pCheckInfo);
-//       } else if (key == tsArray[pos]) {  // data in buffer has the same timestamp of data in file block, ignore it
-//         if (TD_SUPPORT_UPDATE(pCfg->update)) {
-//           if (lastKeyAppend != key) {
-//             if (lastKeyAppend != TSKEY_INITIAL_VAL) {
-//               ++curRow;
-//             }
-//             lastKeyAppend = key;
-//           }
-//           // load data from file firstly
-//           numOfRows = doCopyRowsFromFileBlock(pTsdbReadHandle, pTsdbReadHandle->outputCapacity, curRow, pos, pos);
-
-//           if (rv1 != TD_ROW_SVER(row1)) {
-//             rv1 = TD_ROW_SVER(row1);
-//           }
-//           if (row2 && rv2 != TD_ROW_SVER(row2)) {
-//             rv2 = TD_ROW_SVER(row2);
-//           }
-
-//           // still assign data into current row
-//           numOfRows +=
-//               mergeTwoRowFromMem(pTsdbReadHandle, pTsdbReadHandle->outputCapacity, &curRow, row1, row2, numOfCols,
-//                                  pCheckInfo->tableId, pSchema1, pSchema2, pCfg->update, &lastKeyAppend);
-
-//           if (cur->win.skey == TSKEY_INITIAL_VAL) {
-//             cur->win.skey = key;
-//           }
-
-//           cur->win.ekey = key;
-//           cur->lastKey = key + step;
-//           cur->mixBlock = true;
-
-//           moveToNextRowInMem(pCheckInfo);
-
-//           pos += step;
-//           adjustPos = true;
-//         } else {
-//           // discard the memory record
-//           moveToNextRowInMem(pCheckInfo);
-//         }
-//       } else if ((key > tsArray[pos] && ascScan) || (key < tsArray[pos] && !ascScan)) {
-//         if (cur->win.skey == TSKEY_INITIAL_VAL) {
-//           cur->win.skey = tsArray[pos];
-//         }
-
-//         int32_t end = doBinarySearchKey(pCols->cols[0].pData, pCols->numOfRows, key, order);
-//         assert(end != -1);
-
-//         if (tsArray[end] == key) {  // the value of key in cache equals to the end timestamp value, ignore it
-// #if 0
-//           if (pCfg->update == TD_ROW_DISCARD_UPDATE) {
-//             moveToNextRowInMem(pCheckInfo);
-//           } else {
-//             end -= step;
-//           }
-// #endif
-//           if (!TD_SUPPORT_UPDATE(pCfg->update)) {
-//             moveToNextRowInMem(pCheckInfo);
-//           } else {
-//             end -= step;
-//           }
-//         }
-
-//         int32_t qstart = 0, qend = 0;
-//         getQualifiedRowsPos(pTsdbReadHandle, pos, end, numOfRows, &qstart, &qend);
-
-//         if ((lastKeyAppend != TSKEY_INITIAL_VAL) && (lastKeyAppend != (ascScan ? tsArray[qstart] : tsArray[qend]))) {
-//           ++curRow;
-//         }
-
-//         numOfRows = doCopyRowsFromFileBlock(pTsdbReadHandle, pTsdbReadHandle->outputCapacity, curRow, qstart, qend);
-//         pos += (qend - qstart + 1) * step;
-//         if (numOfRows > 0) {
-//           curRow = numOfRows - 1;
-//         }
-
-//         cur->win.ekey = ascScan ? tsArray[qend] : tsArray[qstart];
-//         cur->lastKey = cur->win.ekey + step;
-//         lastKeyAppend = cur->win.ekey;
-//       }
-//     } while (numOfRows < pTsdbReadHandle->outputCapacity);
-
-//     if (numOfRows < pTsdbReadHandle->outputCapacity) {
-//       /**
-//        * if cache is empty, load remain file block data. In contrast, if there are remain data in cache, do NOT
-//        * copy them all to result buffer, since it may be overlapped with file data block.
-//        */
-//       if (node == NULL || ((TD_ROW_KEY((STSRow*)SL_GET_NODE_DATA(node)) > pTsdbReadHandle->window.ekey) && ascScan)
-//       ||
-//           ((TD_ROW_KEY((STSRow*)SL_GET_NODE_DATA(node)) < pTsdbReadHandle->window.ekey) && !ascScan)) {
-//         // no data in cache or data in cache is greater than the ekey of time window, load data from file block
-//         if (cur->win.skey == TSKEY_INITIAL_VAL) {
-//           cur->win.skey = tsArray[pos];
-//         }
-
-//         int32_t start = -1, end = -1;
-//         getQualifiedRowsPos(pTsdbReadHandle, pos, endPos, numOfRows, &start, &end);
-
-//         numOfRows = doCopyRowsFromFileBlock(pTsdbReadHandle, pTsdbReadHandle->outputCapacity, numOfRows, start, end);
-//         pos += (end - start + 1) * step;
-
-//         cur->win.ekey = ascScan ? tsArray[end] : tsArray[start];
-//         cur->lastKey = cur->win.ekey + step;
-//         cur->mixBlock = true;
-//       }
-//     }
-//   }
-
-//   cur->blockCompleted = (((pos > endPos || cur->lastKey > pTsdbReadHandle->window.ekey) && ascScan) ||
-//                          ((pos < endPos || cur->lastKey < pTsdbReadHandle->window.ekey) && !ascScan));
-
-//   if (!ascScan) {
-//     TSWAP(cur->win.skey, cur->win.ekey);
-//   }
-
-//   updateInfoAfterMerge(pTsdbReadHandle, pCheckInfo, numOfRows, pos);
-//   doCheckGeneratedBlockRange(pTsdbReadHandle);
-
-//   tsdbDebug("%p uid:%" PRIu64 ", data block created, mixblock:%d, brange:%" PRIu64 "-%" PRIu64 " rows:%d, %s",
-//             pTsdbReadHandle, pCheckInfo->tableId, cur->mixBlock, cur->win.skey, cur->win.ekey, cur->rows,
-//             pTsdbReadHandle->idStr);
-// }
-
 static void cleanupBlockOrderSupporter(SBlockOrderSupporter* pSup) {
   taosMemoryFreeClear(pSup->numOfBlocksPerTable);
   taosMemoryFreeClear(pSup->indexPerTable);
@@ -1252,8 +836,9 @@ static int32_t initBlockIterator(STsdbReader* pReader, SDataBlockIter* pBlockIte
   // access data blocks according to the offset of each block in asc/desc order.
   int32_t numOfTables = (int32_t)taosHashGetSize(pReader->status.pTableMap);
 
-  SBlockOrderSupporter sup = {0};
+  int64_t st = taosGetTimestampUs();
 
+  SBlockOrderSupporter sup = {0};
   int32_t code = initBlockOrderSupporter(&sup, numOfTables);
   if (code != TSDB_CODE_SUCCESS) {
     return code;
@@ -1302,11 +887,12 @@ static int32_t initBlockIterator(STsdbReader* pReader, SDataBlockIter* pBlockIte
       SFileDataBlockInfo blockInfo = {.uid = sup.pDataBlockInfo[0][i].uid, .tbBlockIdx = i};
       taosArrayPush(pBlockIter->blockList, &blockInfo);
     }
-    tsdbDebug("%p create blocks info struct completed for one table, %d blocks not sorted %s", pReader, cnt,
-              pReader->idStr);
+
+    int64_t et = taosGetTimestampUs();
+    tsdbDebug("%p create blocks info struct completed for one table, %d blocks not sorted, elapsed time:%.2f ms %s", pReader, cnt,
+              (et - st)/1000.0, pReader->idStr);
 
     pBlockIter->index = asc ? 0 : (numOfBlocks - 1);
-
     cleanupBlockOrderSupporter(&sup);
     return TSDB_CODE_SUCCESS;
   }
@@ -1340,7 +926,8 @@ static int32_t initBlockIterator(STsdbReader* pReader, SDataBlockIter* pBlockIte
     tMergeTreeAdjust(pTree, tMergeTreeGetAdjustIndex(pTree));
   }
 
-  tsdbDebug("%p %d data blocks sort completed, %s", pReader, cnt, pReader->idStr);
+  int64_t et = taosGetTimestampUs();
+  tsdbDebug("%p %d data blocks access order completed, elapsed time:%.2f ms %s", pReader, cnt, (et-st)/1000.0, pReader->idStr);
   cleanupBlockOrderSupporter(&sup);
   taosMemoryFree(pTree);
 
@@ -1813,6 +1400,8 @@ static int32_t buildComposedDataBlock(STsdbReader* pReader, STableBlockScanInfo*
   SBlockData*         pBlockData = &pReader->status.fileBlockData;
   int32_t             step = ASCENDING_TRAVERSE(pReader->order) ? 1 : -1;
 
+  int64_t st = taosGetTimestampUs();
+
   while (1) {
     // todo check the validate of row in file block
     {
@@ -1851,10 +1440,11 @@ static int32_t buildComposedDataBlock(STsdbReader* pReader, STableBlockScanInfo*
   blockDataUpdateTsWindow(pResBlock, 0);
 
   setComposedBlockFlag(pReader, true);
+  int64_t et = taosGetTimestampUs();
 
-  tsdbDebug("%p uid:%" PRIu64 ", composed data block created, brange:%" PRIu64 "-%" PRIu64 " rows:%d, %s", pReader,
+  tsdbDebug("%p uid:%" PRIu64 ", composed data block created, brange:%" PRIu64 "-%" PRIu64 " rows:%d, elapsed time:%.2f ms %s", pReader,
             pBlockScanInfo->uid, pResBlock->info.window.skey, pResBlock->info.window.ekey, pResBlock->info.rows,
-            pReader->idStr);
+            (et - st)/1000.0, pReader->idStr);
 
   return TSDB_CODE_SUCCESS;
 }
@@ -2031,7 +1621,9 @@ static TSDBKEY getCurrentKeyInBuf(SDataBlockIter* pBlockIter, STsdbReader* pRead
 
 static int32_t moveToNextFile(STsdbReader* pReader, int32_t* numOfBlocks) {
   SReaderStatus* pStatus = &pReader->status;
-  SArray*        pIndexList = taosArrayInit(4, sizeof(SBlockIdx));
+
+  size_t numOfTables = taosHashGetSize(pReader->status.pTableMap);
+  SArray* pIndexList = taosArrayInit(numOfTables, sizeof(SBlockIdx));
 
   while (1) {
     bool hasNext = filesetIteratorNext(&pStatus->fileIter, pReader);
@@ -2799,24 +2391,57 @@ int32_t tsdbGetStbIdList(SMeta* pMeta, int64_t suid, SArray* list) {
 // ====================================== EXPOSED APIs ======================================
 int32_t tsdbReaderOpen(SVnode* pVnode, SQueryTableDataCond* pCond, SArray* pTableList, STsdbReader** ppReader,
                        const char* idstr) {
-  int32_t code = tsdbReaderCreate(pVnode, pCond, ppReader, idstr);
-  if (code) {
+  int32_t code = tsdbReaderCreate(pVnode, pCond, ppReader, 4096, idstr);
+  if (code != TSDB_CODE_SUCCESS) {
     goto _err;
   }
 
-  if (pCond->suid != 0) {
-    (*ppReader)->pSchema = metaGetTbTSchema((*ppReader)->pTsdb->pVnode->pMeta, (*ppReader)->suid, -1);
-  } else if (taosArrayGetSize(pTableList) > 0) {
-    STableKeyInfo* pKey = taosArrayGet(pTableList, 0);
-    (*ppReader)->pSchema = metaGetTbTSchema((*ppReader)->pTsdb->pVnode->pMeta, pKey->uid, -1);
-  }
-
+  // check for query time window
   STsdbReader* pReader = *ppReader;
   if (isEmptyQueryTimeWindow(&pReader->window)) {
     tsdbDebug("%p query window not overlaps with the data set, no result returned, %s", pReader, pReader->idStr);
     return TSDB_CODE_SUCCESS;
   }
 
+  if (pCond->type == TIMEWINDOW_RANGE_EXTERNAL) {
+    // update the SQueryTableDataCond to create inner reader
+    STimeWindow w = pCond->twindows;
+    int32_t order = pCond->order;
+    if (order == TSDB_ORDER_ASC) {
+      pCond->twindows.ekey = pCond->twindows.skey;
+      pCond->twindows.skey = INT64_MIN;
+      pCond->order = TSDB_ORDER_DESC;
+    } else {
+      pCond->twindows.skey = pCond->twindows.ekey;
+      pCond->twindows.ekey = INT64_MAX;
+      pCond->order = TSDB_ORDER_ASC;
+    }
+
+    code = tsdbReaderCreate(pVnode, pCond, &pReader->innerReader[0], 1, idstr);
+    if (code != TSDB_CODE_SUCCESS) {
+      goto _err;
+    }
+
+    if (order == TSDB_ORDER_ASC) {
+      pCond->twindows.skey = w.ekey;
+      pCond->twindows.ekey = INT64_MAX;
+    } else  {
+      pCond->twindows.skey = INT64_MIN;
+      pCond->twindows.ekey = w.ekey;
+    }
+    code = tsdbReaderCreate(pVnode, pCond, &pReader->innerReader[1], 1, idstr);
+    if (code != TSDB_CODE_SUCCESS) {
+      goto _err;
+    }
+  }
+
+  if (pCond->suid != 0) {
+    pReader->pSchema = metaGetTbTSchema(pReader->pTsdb->pVnode->pMeta, pReader->suid, -1);
+  } else if (taosArrayGetSize(pTableList) > 0) {
+    STableKeyInfo* pKey = taosArrayGet(pTableList, 0);
+    pReader->pSchema = metaGetTbTSchema(pReader->pTsdb->pVnode->pMeta, pKey->uid, -1);
+  }
+
   int32_t numOfTables = taosArrayGetSize(pTableList);
   pReader->status.pTableMap = createDataBlockScanInfo(pReader, pTableList->pData, numOfTables);
   if (pReader->status.pTableMap == NULL) {
@@ -2827,21 +2452,41 @@ int32_t tsdbReaderOpen(SVnode* pVnode, SQueryTableDataCond* pCond, SArray* pTabl
     goto _err;
   }
 
-  SDataBlockIter* pBlockIter = &pReader->status.blockIter;
-
   code = tsdbTakeReadSnap(pReader->pTsdb, &pReader->pReadSnap);
-  if (code) goto _err;
+  if (code != TSDB_CODE_SUCCESS) {
+    goto _err;
+  }
 
-  initFilesetIterator(&pReader->status.fileIter, (*ppReader)->pReadSnap->fs.aDFileSet, pReader->order, pReader->idStr);
-  resetDataBlockIterator(&pReader->status.blockIter, pReader->order);
+  if (pReader->type == TIMEWINDOW_RANGE_CONTAINED) {
+    SDataBlockIter* pBlockIter = &pReader->status.blockIter;
 
-  // no data in files, let's try buffer in memory
-  if (pReader->status.fileIter.numOfFiles == 0) {
-    pReader->status.loadFromFile = false;
+    initFilesetIterator(&pReader->status.fileIter, pReader->pReadSnap->fs.aDFileSet, pReader->order, pReader->idStr);
+    resetDataBlockIterator(&pReader->status.blockIter, pReader->order);
+
+    // no data in files, let's try buffer in memory
+    if (pReader->status.fileIter.numOfFiles == 0) {
+      pReader->status.loadFromFile = false;
+    } else {
+      code = initForFirstBlockInFile(pReader, pBlockIter);
+      if (code != TSDB_CODE_SUCCESS) {
+        return code;
+      }
+    }
   } else {
-    code = initForFirstBlockInFile(pReader, pBlockIter);
-    if (code != TSDB_CODE_SUCCESS) {
-      return code;
+    STsdbReader* pPrevReader = pReader->innerReader[0];
+    SDataBlockIter* pBlockIter = &pPrevReader->status.blockIter;
+
+    initFilesetIterator(&pPrevReader->status.fileIter, pPrevReader->pReadSnap->fs.aDFileSet, pPrevReader->order, pPrevReader->idStr);
+    resetDataBlockIterator(&pPrevReader->status.blockIter, pPrevReader->order);
+
+    // no data in files, let's try buffer in memory
+    if (pPrevReader->status.fileIter.numOfFiles == 0) {
+      pPrevReader->status.loadFromFile = false;
+    } else {
+      code = initForFirstBlockInFile(pPrevReader, pBlockIter);
+      if (code != TSDB_CODE_SUCCESS) {
+        return code;
+      }
     }
   }
 
@@ -2881,20 +2526,6 @@ void tsdbReaderClose(STsdbReader* pReader) {
     tsdbDataFReaderClose(&pReader->pFileReader);
   }
 
-#if 0
-//   if (pReader->status.pTableScanInfo != NULL) {
-//     pReader->status.pTableScanInfo = destroyTableCheckInfo(pReader->status.pTableScanInfo);
-//   }
-
-//   tsdbDestroyReadH(&pReader->rhelper);
-
-//   tdFreeDataCols(pReader->pDataCols);
-//   pReader->pDataCols = NULL;
-//
-//   pReader->prev = doFreeColumnInfoData(pReader->prev);
-//   pReader->next = doFreeColumnInfoData(pReader->next);
-#endif
-
   SIOCostSummary* pCost = &pReader->cost;
 
   tsdbDebug("%p :io-cost summary: head-file read cnt:%" PRIu64 ", head-file time:%" PRIu64 " us, statis-info:%" PRId64
@@ -2907,55 +2538,100 @@ void tsdbReaderClose(STsdbReader* pReader) {
   taosMemoryFreeClear(pReader);
 }
 
-bool tsdbNextDataBlock(STsdbReader* pReader) {
-  if (isEmptyQueryTimeWindow(&pReader->window)) {
-    return false;
-  }
-
+static bool doTsdbNextDataBlock(STsdbReader* pReader) {
   // cleanup the data that belongs to the previous data block
   SSDataBlock* pBlock = pReader->pResBlock;
   blockDataCleanup(pBlock);
 
   int64_t        stime = taosGetTimestampUs();
-  int64_t        elapsedTime = stime;
   SReaderStatus* pStatus = &pReader->status;
 
-  if (pReader->type == BLOCK_LOAD_OFFSET_ORDER) {
-    if (pStatus->loadFromFile) {
-      int32_t code = buildBlockFromFiles(pReader);
-      if (code != TSDB_CODE_SUCCESS) {
-        return false;
-      }
+  if (pStatus->loadFromFile) {
+    int32_t code = buildBlockFromFiles(pReader);
+    if (code != TSDB_CODE_SUCCESS) {
+      return false;
+    }
 
-      if (pBlock->info.rows > 0) {
-        return true;
-      } else {
-        buildBlockFromBufferSequentially(pReader);
-        return pBlock->info.rows > 0;
-      }
-    } else {  // no data in files, let's try the buffer
+    if (pBlock->info.rows > 0) {
+      return true;
+    } else {
       buildBlockFromBufferSequentially(pReader);
       return pBlock->info.rows > 0;
     }
-  } else if (pReader->type == BLOCK_LOAD_TABLESEQ_ORDER) {
-  } else if (pReader->type == BLOCK_LOAD_EXTERN_ORDER) {
-  } else {
-    ASSERT(0);
+  } else {  // no data in files, let's try the buffer
+    buildBlockFromBufferSequentially(pReader);
+    return pBlock->info.rows > 0;
   }
+
   return false;
 }
 
-void tsdbRetrieveDataBlockInfo(STsdbReader* pReader, SDataBlockInfo* pDataBlockInfo) {
+bool tsdbNextDataBlock(STsdbReader* pReader) {
+  if (isEmptyQueryTimeWindow(&pReader->window)) {
+    return false;
+  }
+
+  if (pReader->innerReader[0] != NULL) {
+    bool ret = doTsdbNextDataBlock(pReader->innerReader[0]);
+    if (ret) {
+      pReader->step = EXTERNAL_ROWS_PREV;
+      return ret;
+    }
+
+    tsdbReaderClose(pReader->innerReader[0]);
+    pReader->innerReader[0] = NULL;
+  }
+
+  pReader->step = EXTERNAL_ROWS_MAIN;
+  bool ret = doTsdbNextDataBlock(pReader);
+  if (ret) {
+    return ret;
+  }
+
+  if (pReader->innerReader[1] != NULL) {
+    bool ret1 = doTsdbNextDataBlock(pReader->innerReader[1]);
+    if (ret1) {
+      pReader->step = EXTERNAL_ROWS_NEXT;
+      return ret1;
+    }
+
+    tsdbReaderClose(pReader->innerReader[1]);
+    pReader->innerReader[1] = NULL;
+  }
+
+  return false;
+}
+
+static void setBlockInfo(STsdbReader* pReader, SDataBlockInfo* pDataBlockInfo) {
   ASSERT(pDataBlockInfo != NULL && pReader != NULL);
   pDataBlockInfo->rows = pReader->pResBlock->info.rows;
   pDataBlockInfo->uid = pReader->pResBlock->info.uid;
   pDataBlockInfo->window = pReader->pResBlock->info.window;
 }
 
+void tsdbRetrieveDataBlockInfo(STsdbReader* pReader, SDataBlockInfo* pDataBlockInfo) {
+  if (pReader->type == TIMEWINDOW_RANGE_EXTERNAL) {
+    if (pReader->step ==  EXTERNAL_ROWS_MAIN) {
+      setBlockInfo(pReader, pDataBlockInfo);
+    } else if (pReader->step ==  EXTERNAL_ROWS_PREV) {
+      setBlockInfo(pReader->innerReader[0], pDataBlockInfo);
+    } else {
+      setBlockInfo(pReader->innerReader[1], pDataBlockInfo);
+    }
+  } else {
+    setBlockInfo(pReader, pDataBlockInfo);
+  }
+}
+
 int32_t tsdbRetrieveDatablockSMA(STsdbReader* pReader, SColumnDataAgg*** pBlockStatis, bool* allHave) {
   int32_t code = 0;
   *allHave = false;
 
+  if(pReader->type == TIMEWINDOW_RANGE_EXTERNAL) {
+    *pBlockStatis = NULL;
+    return TSDB_CODE_SUCCESS;
+  }
+
   // there is no statistics data for composed block
   if (pReader->status.composedDataBlock) {
     *pBlockStatis = NULL;
@@ -3025,7 +2701,7 @@ int32_t tsdbRetrieveDatablockSMA(STsdbReader* pReader, SColumnDataAgg*** pBlockS
   return code;
 }
 
-SArray* tsdbRetrieveDataBlock(STsdbReader* pReader, SArray* pIdList) {
+static SArray* doRetrieveDataBlock(STsdbReader* pReader) {
   SReaderStatus* pStatus = &pReader->status;
 
   if (pStatus->composedDataBlock) {
@@ -3054,16 +2730,27 @@ SArray* tsdbRetrieveDataBlock(STsdbReader* pReader, SArray* pIdList) {
   return pReader->pResBlock->pDataBlock;
 }
 
+SArray* tsdbRetrieveDataBlock(STsdbReader* pReader, SArray* pIdList) {
+  if (pReader->type == TIMEWINDOW_RANGE_EXTERNAL) {
+    if (pReader->step == EXTERNAL_ROWS_PREV) {
+      return doRetrieveDataBlock(pReader->innerReader[0]);
+    } else if (pReader->step == EXTERNAL_ROWS_NEXT) {
+      return doRetrieveDataBlock(pReader->innerReader[1]);
+    }
+  }
+
+  return doRetrieveDataBlock(pReader);
+}
+
 int32_t tsdbReaderReset(STsdbReader* pReader, SQueryTableDataCond* pCond) {
   if (isEmptyQueryTimeWindow(&pReader->window)) {
     return TSDB_CODE_SUCCESS;
   }
 
   pReader->order = pCond->order;
-  pReader->type = BLOCK_LOAD_OFFSET_ORDER;
+  pReader->type  = TIMEWINDOW_RANGE_CONTAINED;
   pReader->status.loadFromFile = true;
   pReader->status.pTableIter = NULL;
-
   pReader->window = updateQueryTimeWindow(pReader->pTsdb, &pCond->twindows);
 
   // allocate buffer in order to load data blocks from file
@@ -3073,10 +2760,7 @@ int32_t tsdbReaderReset(STsdbReader* pReader, SQueryTableDataCond* pCond) {
   pReader->suppInfo.tsColAgg.colId = PRIMARYKEY_TIMESTAMP_COL_ID;
   tsdbDataFReaderClose(&pReader->pFileReader);
 
-  // todo set the correct numOfTables
-  int32_t         numOfTables = 1;
-  SDataBlockIter* pBlockIter = &pReader->status.blockIter;
-
+  int32_t numOfTables = taosHashGetSize(pReader->status.pTableMap);
   tsdbDataFReaderClose(&pReader->pFileReader);
 
   initFilesetIterator(&pReader->status.fileIter, pReader->pReadSnap->fs.aDFileSet, pReader->order, pReader->idStr);
@@ -3084,18 +2768,23 @@ int32_t tsdbReaderReset(STsdbReader* pReader, SQueryTableDataCond* pCond) {
   resetDataBlockScanInfo(pReader->status.pTableMap);
 
   int32_t code = 0;
+  SDataBlockIter* pBlockIter = &pReader->status.blockIter;
+
   // no data in files, let's try buffer in memory
   if (pReader->status.fileIter.numOfFiles == 0) {
     pReader->status.loadFromFile = false;
   } else {
     code = initForFirstBlockInFile(pReader, pBlockIter);
     if (code != TSDB_CODE_SUCCESS) {
+      tsdbError("%p reset reader failed, numOfTables:%d, query range:%" PRId64 " - %" PRId64 " in query %s",
+                pReader, numOfTables, pReader->window.skey, pReader->window.ekey, pReader->idStr);
       return code;
     }
   }
 
   tsdbDebug("%p reset reader, suid:%" PRIu64 ", numOfTables:%d, query range:%" PRId64 " - %" PRId64 " in query %s",
             pReader, pReader->suid, numOfTables, pReader->window.skey, pReader->window.ekey, pReader->idStr);
+
   return code;
 }
 
@@ -3186,7 +2875,7 @@ int64_t tsdbGetNumOfRowsInMemTable(STsdbReader* pReader) {
 
     STbData* d = NULL;
     if (pReader->pTsdb->mem != NULL) {
-      tsdbGetTbDataFromMemTable(pReader->pTsdb->mem, pReader->suid, pBlockScanInfo->uid, &d);
+      tsdbGetTbDataFromMemTable(pReader->pReadSnap->pMem, pReader->suid, pBlockScanInfo->uid, &d);
       if (d != NULL) {
         rows += tsdbGetNRowsInTbData(d);
       }
@@ -3194,7 +2883,7 @@ int64_t tsdbGetNumOfRowsInMemTable(STsdbReader* pReader) {
 
     STbData* di = NULL;
     if (pReader->pTsdb->imem != NULL) {
-      tsdbGetTbDataFromMemTable(pReader->pTsdb->imem, pReader->suid, pBlockScanInfo->uid, &di);
+      tsdbGetTbDataFromMemTable(pReader->pReadSnap->pIMem, pReader->suid, pBlockScanInfo->uid, &di);
       if (di != NULL) {
         rows += tsdbGetNRowsInTbData(di);
       }
diff --git a/source/libs/command/src/explain.c b/source/libs/command/src/explain.c
index 9ffdfc2289..266f96b41e 100644
--- a/source/libs/command/src/explain.c
+++ b/source/libs/command/src/explain.c
@@ -135,7 +135,7 @@ int32_t qExplainGenerateResChildren(SPhysiNode *pNode, SExplainGroup *group, SNo
       break;
     }
     case QUERY_NODE_PHYSICAL_PLAN_MERGE_JOIN: {
-      SJoinPhysiNode *pJoinNode = (SJoinPhysiNode *)pNode;
+      SSortMergeJoinPhysiNode *pJoinNode = (SSortMergeJoinPhysiNode *)pNode;
       pPhysiChildren = pJoinNode->node.pChildren;
       break;
     }
@@ -434,7 +434,8 @@ int32_t qExplainResNodeToRowsImpl(SExplainResNode *pResNode, SExplainCtx *ctx, i
     case QUERY_NODE_PHYSICAL_PLAN_TABLE_SCAN: {
       STableScanPhysiNode *pTblScanNode = (STableScanPhysiNode *)pNode;
       EXPLAIN_ROW_NEW(level,
-                      QUERY_NODE_PHYSICAL_PLAN_TABLE_MERGE_SCAN == pNode->type ? EXPLAIN_TBL_MERGE_SCAN_FORMAT : EXPLAIN_TBL_SCAN_FORMAT,
+                      QUERY_NODE_PHYSICAL_PLAN_TABLE_MERGE_SCAN == pNode->type ? EXPLAIN_TBL_MERGE_SCAN_FORMAT
+                                                                               : EXPLAIN_TBL_SCAN_FORMAT,
                       pTblScanNode->scan.tableName.tname);
       EXPLAIN_ROW_APPEND(EXPLAIN_LEFT_PARENTHESIS_FORMAT);
       if (pResNode->pExecInfo) {
@@ -551,7 +552,7 @@ int32_t qExplainResNodeToRowsImpl(SExplainResNode *pResNode, SExplainCtx *ctx, i
       EXPLAIN_ROW_APPEND(EXPLAIN_BLANK_FORMAT);
       if (pSTblScanNode->scan.pScanPseudoCols) {
         EXPLAIN_ROW_APPEND(EXPLAIN_PSEUDO_COLUMNS_FORMAT, pSTblScanNode->scan.pScanPseudoCols->length);
-        EXPLAIN_ROW_APPEND(EXPLAIN_BLANK_FORMAT); 
+        EXPLAIN_ROW_APPEND(EXPLAIN_BLANK_FORMAT);
       }
       EXPLAIN_ROW_APPEND(EXPLAIN_WIDTH_FORMAT, pSTblScanNode->scan.node.pOutputDataBlockDesc->totalRowSize);
       EXPLAIN_ROW_APPEND(EXPLAIN_BLANK_FORMAT);
@@ -613,7 +614,7 @@ int32_t qExplainResNodeToRowsImpl(SExplainResNode *pResNode, SExplainCtx *ctx, i
       break;
     }
     case QUERY_NODE_PHYSICAL_PLAN_MERGE_JOIN: {
-      SJoinPhysiNode *pJoinNode = (SJoinPhysiNode *)pNode;
+      SSortMergeJoinPhysiNode *pJoinNode = (SSortMergeJoinPhysiNode *)pNode;
       EXPLAIN_ROW_NEW(level, EXPLAIN_JOIN_FORMAT, EXPLAIN_JOIN_STRING(pJoinNode->joinType));
       EXPLAIN_ROW_APPEND(EXPLAIN_LEFT_PARENTHESIS_FORMAT);
       if (pResNode->pExecInfo) {
@@ -1180,7 +1181,7 @@ int32_t qExplainResNodeToRowsImpl(SExplainResNode *pResNode, SExplainCtx *ctx, i
       EXPLAIN_ROW_APPEND(EXPLAIN_BLANK_FORMAT);
       if (pDistScanNode->pScanPseudoCols) {
         EXPLAIN_ROW_APPEND(EXPLAIN_PSEUDO_COLUMNS_FORMAT, pDistScanNode->pScanPseudoCols->length);
-        EXPLAIN_ROW_APPEND(EXPLAIN_BLANK_FORMAT);   
+        EXPLAIN_ROW_APPEND(EXPLAIN_BLANK_FORMAT);
       }
       EXPLAIN_ROW_APPEND(EXPLAIN_WIDTH_FORMAT, pDistScanNode->node.pOutputDataBlockDesc->totalRowSize);
       EXPLAIN_ROW_APPEND(EXPLAIN_BLANK_FORMAT);
@@ -1367,7 +1368,7 @@ int32_t qExplainResNodeToRowsImpl(SExplainResNode *pResNode, SExplainCtx *ctx, i
         EXPLAIN_ROW_APPEND(EXPLAIN_FUNCTIONS_FORMAT, pInterpNode->pFuncs->length);
         EXPLAIN_ROW_APPEND(EXPLAIN_BLANK_FORMAT);
       }
-      
+
       EXPLAIN_ROW_APPEND(EXPLAIN_MODE_FORMAT, nodesGetFillModeString(pInterpNode->fillMode));
       EXPLAIN_ROW_APPEND(EXPLAIN_BLANK_FORMAT);
 
@@ -1419,7 +1420,7 @@ int32_t qExplainResNodeToRowsImpl(SExplainResNode *pResNode, SExplainCtx *ctx, i
         }
       }
       break;
-    }    
+    }
     default:
       qError("not supported physical node type %d", pNode->type);
       return TSDB_CODE_QRY_APP_ERROR;
diff --git a/source/libs/executor/inc/executil.h b/source/libs/executor/inc/executil.h
index 23732a6f9a..be97b20455 100644
--- a/source/libs/executor/inc/executil.h
+++ b/source/libs/executor/inc/executil.h
@@ -82,8 +82,6 @@ size_t getResultRowSize(struct SqlFunctionCtx* pCtx, int32_t numOfOutput);
 void   initResultRowInfo(SResultRowInfo* pResultRowInfo);
 void   cleanupResultRowInfo(SResultRowInfo* pResultRowInfo);
 
-void closeAllResultRows(SResultRowInfo* pResultRowInfo);
-
 void initResultRow(SResultRow* pResultRow);
 void closeResultRow(SResultRow* pResultRow);
 bool isResultRowClosed(SResultRow* pResultRow);
@@ -96,6 +94,11 @@ static FORCE_INLINE SResultRow* getResultRowByPos(SDiskbasedBuf* pBuf, SResultRo
   return pRow;
 }
 
+static FORCE_INLINE void setResultBufPageDirty(SDiskbasedBuf* pBuf, SResultRowPosition* pos) {
+  void* pPage = getBufPage(pBuf, pos->pageId);
+  setBufPageDirty(pPage, true);
+}
+
 void initGroupedResultInfo(SGroupResInfo* pGroupResInfo, SHashObj* pHashmap, int32_t order);
 void cleanupGroupResInfo(SGroupResInfo* pGroupResInfo);
 
diff --git a/source/libs/executor/inc/executorimpl.h b/source/libs/executor/inc/executorimpl.h
index 1ad17bbc76..0beb6f1784 100644
--- a/source/libs/executor/inc/executorimpl.h
+++ b/source/libs/executor/inc/executorimpl.h
@@ -108,7 +108,6 @@ typedef struct STaskCostInfo {
   SFileBlockLoadRecorder* pRecoder;
   uint64_t                elapsedTime;
 
-  uint64_t firstStageMergeTime;
   uint64_t winInfoSize;
   uint64_t tableInfoSize;
   uint64_t hashSize;
@@ -321,6 +320,47 @@ typedef struct STableScanInfo {
   int8_t          noTable;
 } STableScanInfo;
 
+typedef struct STableMergeScanInfo {
+  STableListInfo* tableListInfo;
+  int32_t         tableStartIndex;
+  int32_t         tableEndIndex;
+  bool            hasGroupId;
+  uint64_t        groupId;
+  SArray*         dataReaders;  // array of tsdbReaderT*
+  SReadHandle     readHandle;
+  int32_t         bufPageSize;
+  uint32_t        sortBufSize;  // max buffer size for in-memory sort
+  SArray*         pSortInfo;
+  SSortHandle*    pSortHandle;
+
+  SSDataBlock* pSortInputBlock;
+  int64_t      startTs;  // sort start time
+  SArray*      sortSourceParams;
+
+  SFileBlockLoadRecorder readRecorder;
+  int64_t                numOfRows;
+  SScanInfo              scanInfo;
+  int32_t                scanTimes;
+  SNode*                 pFilterNode;  // filter info, which is push down by optimizer
+  SqlFunctionCtx*        pCtx;         // which belongs to the direct upstream operator operator query context
+  SResultRowInfo*        pResultRowInfo;
+  int32_t*               rowEntryInfoOffset;
+  SExprInfo*             pExpr;
+  SSDataBlock*           pResBlock;
+  SArray*                pColMatchInfo;
+  int32_t                numOfOutput;
+
+  SExprSupp       pseudoSup;
+
+  SQueryTableDataCond cond;
+  int32_t             scanFlag;  // table scan flag to denote if it is a repeat/reverse/main scan
+  int32_t             dataBlockLoadFlag;
+  // if the upstream is an interval operator, the interval info is also kept here to get the time
+  // window to check if current data block needs to be loaded.
+  SInterval       interval;
+  SSampleExecInfo sample;  // sample execution info
+} STableMergeScanInfo;
+
 typedef struct STagScanInfo {
   SColumnInfo     *pCols;
   SSDataBlock     *pRes;
@@ -352,6 +392,11 @@ typedef enum EStreamScanMode {
   STREAM_SCAN_FROM_DATAREADER_RANGE,
 } EStreamScanMode;
 
+enum {
+  PROJECT_RETRIEVE_CONTINUE = 0x1,
+  PROJECT_RETRIEVE_DONE = 0x2,
+};
+
 typedef struct SCatchSupporter {
   SHashObj*      pWindowHashTable;  // quick locate the window object for each window
   SDiskbasedBuf* pDataBuf;          // buffer based on blocked-wised disk file
@@ -549,6 +594,7 @@ typedef struct SProjectOperatorInfo {
   SLimitInfo         limitInfo;
   bool               mergeDataBlocks;
   SSDataBlock*       pFinalRes;
+  SNode*             pCondition;
 } SProjectOperatorInfo;
 
 typedef struct SIndefOperatorInfo {
@@ -881,7 +927,7 @@ SOperatorInfo* createPartitionOperatorInfo(SOperatorInfo* downstream, SPartition
 
 SOperatorInfo* createTimeSliceOperatorInfo(SOperatorInfo* downstream, SPhysiNode* pNode, SExecTaskInfo* pTaskInfo);
 
-SOperatorInfo* createMergeJoinOperatorInfo(SOperatorInfo** pDownstream, int32_t numOfDownstream, SJoinPhysiNode* pJoinNode,
+SOperatorInfo* createMergeJoinOperatorInfo(SOperatorInfo** pDownstream, int32_t numOfDownstream, SSortMergeJoinPhysiNode* pJoinNode,
                                            SExecTaskInfo* pTaskInfo);
 
 SOperatorInfo* createStreamSessionAggOperatorInfo(SOperatorInfo* downstream,
diff --git a/source/libs/executor/src/executil.c b/source/libs/executor/src/executil.c
index a76253ab20..ec8e3c4abb 100644
--- a/source/libs/executor/src/executil.c
+++ b/source/libs/executor/src/executil.c
@@ -43,10 +43,6 @@ void cleanupResultRowInfo(SResultRowInfo* pResultRowInfo) {
   }
 }
 
-void closeAllResultRows(SResultRowInfo* pResultRowInfo) {
-  // do nothing
-}
-
 bool isResultRowClosed(SResultRow* pRow) { return (pRow->closed == true); }
 
 void closeResultRow(SResultRow* pResultRow) { pResultRow->closed = true; }
@@ -160,11 +156,13 @@ int32_t getNumOfTotalRes(SGroupResInfo* pGroupResInfo) {
 
 SArray* createSortInfo(SNodeList* pNodeList) {
   size_t numOfCols = 0;
+
   if (pNodeList != NULL) {
     numOfCols = LIST_LENGTH(pNodeList);
   } else {
     numOfCols = 0;
   }
+
   SArray* pList = taosArrayInit(numOfCols, sizeof(SBlockOrderInfo));
   if (pList == NULL) {
     terrno = TSDB_CODE_OUT_OF_MEMORY;
@@ -196,10 +194,6 @@ SSDataBlock* createResDataBlock(SDataBlockDescNode* pNode) {
 
   for (int32_t i = 0; i < numOfCols; ++i) {
     SSlotDescNode* pDescNode = (SSlotDescNode*)nodesListGetNode(pNode->pSlots, i);
-    /*if (!pDescNode->output) {  // todo disable it temporarily*/
-    /*continue;*/
-    /*}*/
-
     SColumnInfoData idata =
         createColumnInfoData(pDescNode->dataType.type, pDescNode->dataType.bytes, pDescNode->slotId);
     idata.info.scale = pDescNode->dataType.scale;
@@ -701,9 +695,6 @@ static int32_t setSelectValueColumnInfo(SqlFunctionCtx* pCtx, int32_t numOfOutpu
     }
   }
 
-#ifdef BUF_PAGE_DEBUG
-  qDebug("page_setSelect num:%d", num);
-#endif
   if (p != NULL) {
     p->subsidiaries.pCtx = pValCtx;
     p->subsidiaries.num = num;
@@ -852,7 +843,7 @@ int32_t initQueryTableDataCond(SQueryTableDataCond* pCond, const STableScanPhysi
   // TODO: get it from stable scan node
   pCond->twindows = pTableScanNode->scanRange;
   pCond->suid = pTableScanNode->scan.suid;
-  pCond->type = BLOCK_LOAD_OFFSET_ORDER;
+  pCond->type = TIMEWINDOW_RANGE_CONTAINED;
   pCond->startVersion = -1;
   pCond->endVersion = -1;
   //  pCond->type = pTableScanNode->scanFlag;
@@ -947,6 +938,7 @@ STimeWindow getFirstQualifiedTimeWindow(int64_t ts, STimeWindow* pWindow, SInter
 }
 
 // get the correct time window according to the handled timestamp
+// todo refactor
 STimeWindow getActiveTimeWindow(SDiskbasedBuf* pBuf, SResultRowInfo* pResultRowInfo, int64_t ts, SInterval* pInterval,
                                 int32_t order) {
   STimeWindow w = {0};
diff --git a/source/libs/executor/src/executorimpl.c b/source/libs/executor/src/executorimpl.c
index 7bac828a53..36ae1d19ec 100644
--- a/source/libs/executor/src/executorimpl.c
+++ b/source/libs/executor/src/executorimpl.c
@@ -42,11 +42,6 @@
 
 #define GET_FORWARD_DIRECTION_FACTOR(ord) (((ord) == TSDB_ORDER_ASC) ? QUERY_ASC_FORWARD_STEP : QUERY_DESC_FORWARD_STEP)
 
-enum {
-  PROJECT_RETRIEVE_CONTINUE = 0x1,
-  PROJECT_RETRIEVE_DONE = 0x2,
-};
-
 #if 0
 static UNUSED_FUNC void *u_malloc (size_t __size) {
   uint32_t v = taosRand();
@@ -575,6 +570,26 @@ static void setPseudoOutputColInfo(SSDataBlock* pResult, SqlFunctionCtx* pCtx, S
 int32_t projectApplyFunctions(SExprInfo* pExpr, SSDataBlock* pResult, SSDataBlock* pSrcBlock, SqlFunctionCtx* pCtx,
                               int32_t numOfOutput, SArray* pPseudoList) {
   setPseudoOutputColInfo(pResult, pCtx, pPseudoList);
+
+  if (pSrcBlock == NULL) {
+    for (int32_t k = 0; k < numOfOutput; ++k) {
+      int32_t outputSlotId = pExpr[k].base.resSchema.slotId;
+
+      ASSERT(pExpr[k].pExpr->nodeType == QUERY_NODE_VALUE);
+      SColumnInfoData* pColInfoData = taosArrayGet(pResult->pDataBlock, outputSlotId);
+
+      int32_t type = pExpr[k].base.pParam[0].param.nType;
+      if (TSDB_DATA_TYPE_NULL == type) {
+        colDataAppendNNULL(pColInfoData, 0, 1);
+      } else {
+        colDataAppend(pColInfoData, 0, taosVariantGet(&pExpr[k].base.pParam[0].param, type), false);
+      }
+    }
+
+    pResult->info.rows = 1;
+    return TSDB_CODE_SUCCESS;
+  }
+
   pResult->info.groupId = pSrcBlock->info.groupId;
 
   // if the source equals to the destination, it is to create a new column as the result of scalar
@@ -1243,52 +1258,6 @@ void initResultRow(SResultRow* pResultRow) {
   //  pResultRow->pEntryInfo = (struct SResultRowEntryInfo*)((char*)pResultRow + sizeof(SResultRow));
 }
 
-/*
- * The start of each column SResultRowEntryInfo is denote by RowCellInfoOffset.
- * Note that in case of top/bottom query, the whole multiple rows of result is treated as only one row of results.
- * +------------+-----------------result column 1------------+------------------result column 2-----------+
- * | SResultRow | SResultRowEntryInfo | intermediate buffer1 | SResultRowEntryInfo | intermediate buffer 2|
- * +------------+--------------------------------------------+--------------------------------------------+
- *           offset[0]                                  offset[1]                                   offset[2]
- */
-// TODO refactor: some function move away
-void setFunctionResultOutput(SOperatorInfo* pOperator, SOptrBasicInfo* pInfo, SAggSupporter* pSup, int32_t stage,
-                             int32_t numOfExprs) {
-  SExecTaskInfo*  pTaskInfo = pOperator->pTaskInfo;
-  SqlFunctionCtx* pCtx = pOperator->exprSupp.pCtx;
-  int32_t*        rowEntryInfoOffset = pOperator->exprSupp.rowEntryInfoOffset;
-
-  SResultRowInfo* pResultRowInfo = &pInfo->resultRowInfo;
-  initResultRowInfo(pResultRowInfo);
-
-  int64_t     tid = 0;
-  int64_t     groupId = 0;
-  SResultRow* pRow = doSetResultOutBufByKey(pSup->pResultBuf, pResultRowInfo, (char*)&tid, sizeof(tid), true, groupId,
-                                            pTaskInfo, false, pSup);
-
-  for (int32_t i = 0; i < numOfExprs; ++i) {
-    struct SResultRowEntryInfo* pEntry = getResultEntryInfo(pRow, i, rowEntryInfoOffset);
-    cleanupResultRowEntry(pEntry);
-
-    pCtx[i].resultInfo = pEntry;
-    pCtx[i].scanFlag = stage;
-  }
-
-  initCtxOutputBuffer(pCtx, numOfExprs);
-}
-
-void initCtxOutputBuffer(SqlFunctionCtx* pCtx, int32_t size) {
-  for (int32_t j = 0; j < size; ++j) {
-    struct SResultRowEntryInfo* pResInfo = GET_RES_INFO(&pCtx[j]);
-    if (isRowEntryInitialized(pResInfo) || fmIsPseudoColumnFunc(pCtx[j].functionId) || pCtx[j].functionId == -1 ||
-        fmIsScalarFunc(pCtx[j].functionId)) {
-      continue;
-    }
-
-    pCtx[j].fpSet.init(&pCtx[j], pCtx[j].resultInfo);
-  }
-}
-
 void setTaskStatus(SExecTaskInfo* pTaskInfo, int8_t status) {
   if (status == TASK_NOT_COMPLETED) {
     pTaskInfo->status = status;
@@ -1356,7 +1325,7 @@ void doFilter(const SNode* pFilterNode, SSDataBlock* pBlock, const SArray* pColM
   extractQualifiedTupleByFilterResult(pBlock, rowRes, keep);
 
   if (pColMatchInfo != NULL) {
-    for(int32_t i = 0; i < taosArrayGetSize(pColMatchInfo); ++i) {
+    for (int32_t i = 0; i < taosArrayGetSize(pColMatchInfo); ++i) {
       SColMatchInfo* pInfo = taosArrayGet(pColMatchInfo, i);
       if (pInfo->colId == PRIMARYKEY_TIMESTAMP_COL_ID) {
         SColumnInfoData* pColData = taosArrayGet(pBlock->pDataBlock, pInfo->targetSlotId);
@@ -1665,9 +1634,6 @@ void queryCostStatis(SExecTaskInfo* pTaskInfo) {
   //  hashSize += taosHashGetMemSize(pRuntimeEnv->tableqinfoGroupInfo.map);
   //  pSummary->hashSize = hashSize;
 
-  // add the merge time
-  pSummary->elapsedTime += pSummary->firstStageMergeTime;
-
   //  SResultRowPool* p = pTaskInfo->pool;
   //  if (p != NULL) {
   //    pSummary->winInfoSize = getResultRowPoolMemSize(p);
@@ -1676,17 +1642,16 @@ void queryCostStatis(SExecTaskInfo* pTaskInfo) {
   //    pSummary->winInfoSize = 0;
   //    pSummary->numOfTimeWindows = 0;
   //  }
-  //
-  //  calculateOperatorProfResults(pQInfo);
 
   SFileBlockLoadRecorder* pRecorder = pSummary->pRecoder;
   if (pSummary->pRecoder != NULL) {
-    qDebug("%s :cost summary: elapsed time:%" PRId64 " us, first merge:%" PRId64
-           " us, total blocks:%d, "
-           "load block statis:%d, load data block:%d, total rows:%" PRId64 ", check rows:%" PRId64,
-           GET_TASKID(pTaskInfo), pSummary->elapsedTime, pSummary->firstStageMergeTime, pRecorder->totalBlocks,
-           pRecorder->loadBlockStatis, pRecorder->loadBlocks, pRecorder->totalRows, pRecorder->totalCheckedRows);
+    qDebug(
+        "%s :cost summary: elapsed time:%.2f ms, total blocks:%d, load block SMA:%d, load data block:%d, total "
+        "rows:%" PRId64 ", check rows:%" PRId64,
+        GET_TASKID(pTaskInfo), pSummary->elapsedTime / 1000.0, pRecorder->totalBlocks, pRecorder->loadBlockStatis,
+        pRecorder->loadBlocks, pRecorder->totalRows, pRecorder->totalCheckedRows);
   }
+
   // qDebug("QInfo:0x%"PRIx64" :cost summary: winResPool size:%.2f Kb, numOfWin:%"PRId64", tableInfoSize:%.2f Kb,
   // hashTable:%.2f Kb", pQInfo->qId, pSummary->winInfoSize/1024.0,
   //      pSummary->numOfTimeWindows, pSummary->tableInfoSize/1024.0, pSummary->hashSize/1024.0);
@@ -2809,73 +2774,6 @@ static int32_t initGroupCol(SExprInfo* pExprInfo, int32_t numOfCols, SArray* pGr
   return TSDB_CODE_SUCCESS;
 }
 
-SOperatorInfo* createSortedMergeOperatorInfo(SOperatorInfo** downstream, int32_t numOfDownstream, SExprInfo* pExprInfo,
-                                             int32_t num, SArray* pSortInfo, SArray* pGroupInfo,
-                                             SExecTaskInfo* pTaskInfo) {
-  SSortedMergeOperatorInfo* pInfo = taosMemoryCalloc(1, sizeof(SSortedMergeOperatorInfo));
-  SOperatorInfo*            pOperator = taosMemoryCalloc(1, sizeof(SOperatorInfo));
-  if (pInfo == NULL || pOperator == NULL) {
-    goto _error;
-  }
-
-  int32_t code = initExprSupp(&pOperator->exprSupp, pExprInfo, num);
-  if (code != TSDB_CODE_SUCCESS) {
-    goto _error;
-  }
-
-  initResultRowInfo(&pInfo->binfo.resultRowInfo);
-
-  if (pOperator->exprSupp.pCtx == NULL || pInfo->binfo.pRes == NULL) {
-    goto _error;
-  }
-
-  size_t keyBufSize = sizeof(int64_t) + sizeof(int64_t) + POINTER_BYTES;
-  code = doInitAggInfoSup(&pInfo->aggSup, pOperator->exprSupp.pCtx, num, keyBufSize, pTaskInfo->id.str);
-  if (code != TSDB_CODE_SUCCESS) {
-    goto _error;
-  }
-
-  setFunctionResultOutput(pOperator, &pInfo->binfo, &pInfo->aggSup, MAIN_SCAN, num);
-  code = initGroupCol(pExprInfo, num, pGroupInfo, pInfo);
-  if (code != TSDB_CODE_SUCCESS) {
-    goto _error;
-  }
-
-  //  pInfo->resultRowFactor = (int32_t)(getRowNumForMultioutput(pRuntimeEnv->pQueryAttr,
-  //      pRuntimeEnv->pQueryAttr->topBotQuery, false));
-  pInfo->sortBufSize = 1024 * 16;  // 1MB
-  pInfo->bufPageSize = 1024;
-  pInfo->pSortInfo = pSortInfo;
-
-  pOperator->resultInfo.capacity = blockDataGetCapacityInRow(pInfo->binfo.pRes, pInfo->bufPageSize);
-
-  pOperator->name = "SortedMerge";
-  // pOperator->operatorType = OP_SortedMerge;
-  pOperator->blocking = true;
-  pOperator->status = OP_NOT_OPENED;
-  pOperator->info = pInfo;
-  pOperator->pTaskInfo = pTaskInfo;
-
-  pOperator->fpSet = createOperatorFpSet(operatorDummyOpenFn, doSortedMerge, NULL, NULL, destroySortedMergeOperatorInfo,
-                                         NULL, NULL, NULL);
-  code = appendDownstream(pOperator, downstream, numOfDownstream);
-  if (code != TSDB_CODE_SUCCESS) {
-    goto _error;
-  }
-
-  return pOperator;
-
-_error:
-  if (pInfo != NULL) {
-    destroySortedMergeOperatorInfo(pInfo, num);
-  }
-
-  taosMemoryFreeClear(pInfo);
-  taosMemoryFreeClear(pOperator);
-  terrno = TSDB_CODE_QRY_OUT_OF_MEMORY;
-  return NULL;
-}
-
 int32_t getTableScanInfo(SOperatorInfo* pOperator, int32_t* order, int32_t* scanFlag) {
   // todo add more information about exchange operation
   int32_t type = pOperator->operatorType;
@@ -2885,11 +2783,16 @@ int32_t getTableScanInfo(SOperatorInfo* pOperator, int32_t* order, int32_t* scan
     *order = TSDB_ORDER_ASC;
     *scanFlag = MAIN_SCAN;
     return TSDB_CODE_SUCCESS;
-  } else if (type == QUERY_NODE_PHYSICAL_PLAN_TABLE_SCAN || type == QUERY_NODE_PHYSICAL_PLAN_TABLE_MERGE_SCAN) {
+  } else if (type == QUERY_NODE_PHYSICAL_PLAN_TABLE_SCAN) {
     STableScanInfo* pTableScanInfo = pOperator->info;
     *order = pTableScanInfo->cond.order;
     *scanFlag = pTableScanInfo->scanFlag;
     return TSDB_CODE_SUCCESS;
+  } else if (type == QUERY_NODE_PHYSICAL_PLAN_TABLE_MERGE_SCAN) {
+    STableMergeScanInfo* pTableScanInfo = pOperator->info;
+    *order = pTableScanInfo->cond.order;
+    *scanFlag = pTableScanInfo->scanFlag;
+    return TSDB_CODE_SUCCESS;
   } else {
     if (pOperator->pDownstream == NULL || pOperator->pDownstream[0] == NULL) {
       return TSDB_CODE_INVALID_PARA;
@@ -3031,7 +2934,6 @@ static int32_t doOpenAggregateOptr(SOperatorInfo* pOperator) {
     }
   }
 
-  closeAllResultRows(&pAggInfo->binfo.resultRowInfo);
   initGroupedResultInfo(&pAggInfo->groupResInfo, pAggInfo->aggSup.pResultRowHashTable, 0);
   OPTR_SET_OPENED(pOperator);
 
@@ -3279,162 +3181,6 @@ int32_t handleLimitOffset(SOperatorInfo* pOperator, SLimitInfo* pLimitInfo, SSDa
   }
 }
 
-static SSDataBlock* doProjectOperation(SOperatorInfo* pOperator) {
-  SProjectOperatorInfo* pProjectInfo = pOperator->info;
-  SOptrBasicInfo*       pInfo = &pProjectInfo->binfo;
-
-  SExprSupp*   pSup = &pOperator->exprSupp;
-  SSDataBlock* pRes = pInfo->pRes;
-  SSDataBlock* pFinalRes = pProjectInfo->pFinalRes;
-
-  blockDataCleanup(pFinalRes);
-
-  SExecTaskInfo* pTaskInfo = pOperator->pTaskInfo;
-  if (pOperator->status == OP_EXEC_DONE) {
-    if (pTaskInfo->execModel == OPTR_EXEC_MODEL_QUEUE) {
-      pOperator->status = OP_OPENED;
-      return NULL;
-    }
-    return NULL;
-  }
-
-  int64_t st = 0;
-  int32_t order = 0;
-  int32_t scanFlag = 0;
-
-  if (pOperator->cost.openCost == 0) {
-    st = taosGetTimestampUs();
-  }
-
-  SOperatorInfo* downstream = pOperator->pDownstream[0];
-  SLimitInfo* pLimitInfo = &pProjectInfo->limitInfo;
-
-  while(1) {
-    while (1) {
-      blockDataCleanup(pRes);
-
-      // The downstream exec may change the value of the newgroup, so use a local variable instead.
-      SSDataBlock* pBlock = downstream->fpSet.getNextFn(downstream);
-      if (pBlock == NULL) {
-        doSetOperatorCompleted(pOperator);
-        break;
-      }
-
-      if (pBlock->info.type == STREAM_RETRIEVE) {
-        // for stream interval
-        return pBlock;
-      }
-
-      if (pLimitInfo->remainGroupOffset > 0) {
-        if (pLimitInfo->currentGroupId == 0 || pLimitInfo->currentGroupId == pBlock->info.groupId) {  // it is the first group
-          pLimitInfo->currentGroupId = pBlock->info.groupId;
-          continue;
-        } else if (pLimitInfo->currentGroupId != pBlock->info.groupId) {
-          // now it is the data from a new group
-          pLimitInfo->remainGroupOffset -= 1;
-          pLimitInfo->currentGroupId = pBlock->info.groupId;
-
-          // ignore data block in current group
-          if (pLimitInfo->remainGroupOffset > 0) {
-            continue;
-          }
-        }
-
-        // set current group id of the project operator
-        pLimitInfo->currentGroupId = pBlock->info.groupId;
-      }
-
-      // remainGroupOffset == 0
-      // here check for a new group data, we need to handle the data of the previous group.
-      if (pLimitInfo->currentGroupId != 0 && pLimitInfo->currentGroupId != pBlock->info.groupId) {
-        pLimitInfo->numOfOutputGroups += 1;
-        if ((pLimitInfo->slimit.limit > 0) && (pLimitInfo->slimit.limit <= pLimitInfo->numOfOutputGroups)) {
-          doSetOperatorCompleted(pOperator);
-          break;
-        }
-
-        // reset the value for a new group data
-        // existing rows that belongs to previous group.
-        pLimitInfo->numOfOutputRows = 0;
-        pLimitInfo->remainOffset = pLimitInfo->limit.offset;
-      }
-
-      // the pDataBlock are always the same one, no need to call this again
-      int32_t code = getTableScanInfo(pOperator->pDownstream[0], &order, &scanFlag);
-      if (code != TSDB_CODE_SUCCESS) {
-        longjmp(pTaskInfo->env, code);
-      }
-
-      setInputDataBlock(pOperator, pSup->pCtx, pBlock, order, scanFlag, false);
-      blockDataEnsureCapacity(pInfo->pRes, pInfo->pRes->info.rows + pBlock->info.rows);
-
-      code = projectApplyFunctions(pSup->pExprInfo, pInfo->pRes, pBlock, pSup->pCtx, pSup->numOfExprs,
-                                   pProjectInfo->pPseudoColInfo);
-      if (code != TSDB_CODE_SUCCESS) {
-        longjmp(pTaskInfo->env, code);
-      }
-
-      // set current group id
-      pLimitInfo->currentGroupId = pBlock->info.groupId;
-
-      if (pLimitInfo->remainOffset >= pInfo->pRes->info.rows) {
-        pLimitInfo->remainOffset -= pInfo->pRes->info.rows;
-        blockDataCleanup(pInfo->pRes);
-        continue;
-      } else if (pLimitInfo->remainOffset < pInfo->pRes->info.rows && pLimitInfo->remainOffset > 0) {
-        blockDataTrimFirstNRows(pInfo->pRes, pLimitInfo->remainOffset);
-        pLimitInfo->remainOffset = 0;
-      }
-
-      // check for the limitation in each group
-      if (pLimitInfo->limit.limit >= 0 &&
-          pLimitInfo->numOfOutputRows + pInfo->pRes->info.rows >= pLimitInfo->limit.limit) {
-        int32_t keepRows = (int32_t)(pLimitInfo->limit.limit - pLimitInfo->numOfOutputRows);
-        blockDataKeepFirstNRows(pInfo->pRes, keepRows);
-        if (pLimitInfo->slimit.limit > 0 && pLimitInfo->slimit.limit <= pLimitInfo->numOfOutputGroups) {
-          pOperator->status = OP_EXEC_DONE;
-        }
-      }
-
-      pLimitInfo->numOfOutputRows += pInfo->pRes->info.rows;
-      break;
-    }
-
-    // no results generated
-    if (pInfo->pRes->info.rows == 0 || (!pProjectInfo->mergeDataBlocks)) {
-      break;
-    }
-
-    if (pProjectInfo->mergeDataBlocks) {
-      pFinalRes->info.groupId = pInfo->pRes->info.groupId;
-      pFinalRes->info.version = pInfo->pRes->info.version;
-
-      // continue merge data, ignore the group id
-      blockDataMerge(pFinalRes, pInfo->pRes);
-
-      if (pFinalRes->info.rows + pInfo->pRes->info.rows <= pOperator->resultInfo.threshold) {
-        continue;
-      }
-    }
-
-    // do apply filter
-    SSDataBlock* p = pProjectInfo->mergeDataBlocks ? pFinalRes : pRes;
-    doFilter(pProjectInfo->pFilterNode, p, NULL);
-    if (p->info.rows > 0) {
-      break;
-    }
-  }
-
-  SSDataBlock* p = pProjectInfo->mergeDataBlocks ? pFinalRes : pRes;
-  pOperator->resultInfo.totalRows += p->info.rows;
-
-  if (pOperator->cost.openCost == 0) {
-    pOperator->cost.openCost = (taosGetTimestampUs() - st) / 1000.0;
-  }
-
-  return (p->info.rows > 0) ? p : NULL;
-}
-
 static void doHandleRemainBlockForNewGroupImpl(SFillOperatorInfo* pInfo, SResultInfo* pResultInfo,
                                                SExecTaskInfo* pTaskInfo) {
   pInfo->totalInputRows = pInfo->existNewGroupBlock->info.rows;
@@ -3815,30 +3561,6 @@ void destroySFillOperatorInfo(void* param, int32_t numOfOutput) {
   taosMemoryFreeClear(param);
 }
 
-static void destroyProjectOperatorInfo(void* param, int32_t numOfOutput) {
-  if (NULL == param) {
-    return;
-  }
-  SProjectOperatorInfo* pInfo = (SProjectOperatorInfo*)param;
-  cleanupBasicInfo(&pInfo->binfo);
-  cleanupAggSup(&pInfo->aggSup);
-  taosArrayDestroy(pInfo->pPseudoColInfo);
-
-  blockDataDestroy(pInfo->pFinalRes);
-  taosMemoryFreeClear(param);
-}
-
-static void destroyIndefinitOperatorInfo(void* param, int32_t numOfOutput) {
-  SIndefOperatorInfo* pInfo = (SIndefOperatorInfo*)param;
-  cleanupBasicInfo(&pInfo->binfo);
-
-  taosArrayDestroy(pInfo->pPseudoColInfo);
-  cleanupAggSup(&pInfo->aggSup);
-  cleanupExprSupp(&pInfo->scalarSup);
-
-  taosMemoryFreeClear(param);
-}
-
 void destroyExchangeOperatorInfo(void* param, int32_t numOfOutput) {
   SExchangeInfo* pExInfo = (SExchangeInfo*)param;
   taosRemoveRef(exchangeObjRefPool, pExInfo->self);
@@ -3858,260 +3580,6 @@ void doDestroyExchangeOperatorInfo(void* param) {
   taosMemoryFreeClear(param);
 }
 
-static SArray* setRowTsColumnOutputInfo(SqlFunctionCtx* pCtx, int32_t numOfCols) {
-  SArray* pList = taosArrayInit(4, sizeof(int32_t));
-  for (int32_t i = 0; i < numOfCols; ++i) {
-    if (fmIsPseudoColumnFunc(pCtx[i].functionId)) {
-      taosArrayPush(pList, &i);
-    }
-  }
-
-  return pList;
-}
-
-SOperatorInfo* createProjectOperatorInfo(SOperatorInfo* downstream, SProjectPhysiNode* pProjPhyNode,
-                                         SExecTaskInfo* pTaskInfo) {
-  SProjectOperatorInfo* pInfo = taosMemoryCalloc(1, sizeof(SProjectOperatorInfo));
-  SOperatorInfo*        pOperator = taosMemoryCalloc(1, sizeof(SOperatorInfo));
-  if (pInfo == NULL || pOperator == NULL) {
-    goto _error;
-  }
-
-  int32_t    numOfCols = 0;
-  SExprInfo* pExprInfo = createExprInfo(pProjPhyNode->pProjections, NULL, &numOfCols);
-
-  SSDataBlock* pResBlock = createResDataBlock(pProjPhyNode->node.pOutputDataBlockDesc);
-  initLimitInfo(pProjPhyNode->node.pLimit, pProjPhyNode->node.pSlimit, &pInfo->limitInfo);
-
-  pInfo->binfo.pRes = pResBlock;
-  pInfo->pFinalRes = createOneDataBlock(pResBlock, false);
-
-  pInfo->pFilterNode = pProjPhyNode->node.pConditions;
-  pInfo->mergeDataBlocks = pProjPhyNode->mergeDataBlock;
-
-  int32_t numOfRows = 4096;
-  size_t  keyBufSize = sizeof(int64_t) + sizeof(int64_t) + POINTER_BYTES;
-
-  // Make sure the size of SSDataBlock will never exceed the size of 2MB.
-  int32_t TWOMB = 2 * 1024 * 1024;
-  if (numOfRows * pResBlock->info.rowSize > TWOMB) {
-    numOfRows = TWOMB / pResBlock->info.rowSize;
-  }
-  initResultSizeInfo(&pOperator->resultInfo, numOfRows);
-
-  initAggInfo(&pOperator->exprSupp, &pInfo->aggSup, pExprInfo, numOfCols, keyBufSize, pTaskInfo->id.str);
-  initBasicInfo(&pInfo->binfo, pResBlock);
-  setFunctionResultOutput(pOperator, &pInfo->binfo, &pInfo->aggSup, MAIN_SCAN, numOfCols);
-
-  pInfo->pPseudoColInfo = setRowTsColumnOutputInfo(pOperator->exprSupp.pCtx, numOfCols);
-  pOperator->name = "ProjectOperator";
-  pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_PROJECT;
-  pOperator->blocking = false;
-  pOperator->status = OP_NOT_OPENED;
-  pOperator->info = pInfo;
-  pOperator->pTaskInfo = pTaskInfo;
-
-  pOperator->fpSet = createOperatorFpSet(operatorDummyOpenFn, doProjectOperation, NULL, NULL,
-                                         destroyProjectOperatorInfo, NULL, NULL, NULL);
-
-  int32_t code = appendDownstream(pOperator, &downstream, 1);
-  if (code != TSDB_CODE_SUCCESS) {
-    goto _error;
-  }
-
-  return pOperator;
-
-_error:
-  pTaskInfo->code = TSDB_CODE_OUT_OF_MEMORY;
-  return NULL;
-}
-
-static void doHandleDataBlock(SOperatorInfo* pOperator, SSDataBlock* pBlock, SOperatorInfo* downstream,
-                              SExecTaskInfo* pTaskInfo) {
-  int32_t order = 0;
-  int32_t scanFlag = 0;
-
-  SIndefOperatorInfo* pIndefInfo = pOperator->info;
-  SOptrBasicInfo*     pInfo = &pIndefInfo->binfo;
-  SExprSupp*          pSup = &pOperator->exprSupp;
-
-  // the pDataBlock are always the same one, no need to call this again
-  int32_t code = getTableScanInfo(downstream, &order, &scanFlag);
-  if (code != TSDB_CODE_SUCCESS) {
-    longjmp(pTaskInfo->env, code);
-  }
-
-  // there is an scalar expression that needs to be calculated before apply the group aggregation.
-  SExprSupp* pScalarSup = &pIndefInfo->scalarSup;
-  if (pScalarSup->pExprInfo != NULL) {
-    code = projectApplyFunctions(pScalarSup->pExprInfo, pBlock, pBlock, pScalarSup->pCtx, pScalarSup->numOfExprs,
-                                 pIndefInfo->pPseudoColInfo);
-    if (code != TSDB_CODE_SUCCESS) {
-      longjmp(pTaskInfo->env, code);
-    }
-  }
-
-  setInputDataBlock(pOperator, pSup->pCtx, pBlock, order, scanFlag, false);
-  blockDataEnsureCapacity(pInfo->pRes, pInfo->pRes->info.rows + pBlock->info.rows);
-
-  code = projectApplyFunctions(pSup->pExprInfo, pInfo->pRes, pBlock, pSup->pCtx, pSup->numOfExprs,
-                               pIndefInfo->pPseudoColInfo);
-  if (code != TSDB_CODE_SUCCESS) {
-    longjmp(pTaskInfo->env, code);
-  }
-}
-
-static SSDataBlock* doApplyIndefinitFunction(SOperatorInfo* pOperator) {
-  SIndefOperatorInfo* pIndefInfo = pOperator->info;
-  SOptrBasicInfo*     pInfo = &pIndefInfo->binfo;
-  SExprSupp*          pSup = &pOperator->exprSupp;
-
-  SSDataBlock* pRes = pInfo->pRes;
-  blockDataCleanup(pRes);
-
-  SExecTaskInfo* pTaskInfo = pOperator->pTaskInfo;
-  if (pOperator->status == OP_EXEC_DONE) {
-    return NULL;
-  }
-
-  int64_t st = 0;
-
-  if (pOperator->cost.openCost == 0) {
-    st = taosGetTimestampUs();
-  }
-
-  SOperatorInfo* downstream = pOperator->pDownstream[0];
-
-  while (1) {
-    // here we need to handle the existsed group results
-    if (pIndefInfo->pNextGroupRes != NULL) {  // todo extract method
-      for (int32_t k = 0; k < pSup->numOfExprs; ++k) {
-        SqlFunctionCtx* pCtx = &pSup->pCtx[k];
-
-        SResultRowEntryInfo* pResInfo = GET_RES_INFO(pCtx);
-        pResInfo->initialized = false;
-        pCtx->pOutput = NULL;
-      }
-
-      doHandleDataBlock(pOperator, pIndefInfo->pNextGroupRes, downstream, pTaskInfo);
-      pIndefInfo->pNextGroupRes = NULL;
-    }
-
-    if (pInfo->pRes->info.rows < pOperator->resultInfo.threshold) {
-      while (1) {
-        // The downstream exec may change the value of the newgroup, so use a local variable instead.
-        SSDataBlock* pBlock = downstream->fpSet.getNextFn(downstream);
-        if (pBlock == NULL) {
-          doSetOperatorCompleted(pOperator);
-          break;
-        }
-
-        if (pIndefInfo->groupId == 0 && pBlock->info.groupId != 0) {
-          pIndefInfo->groupId = pBlock->info.groupId;  // this is the initial group result
-        } else {
-          if (pIndefInfo->groupId != pBlock->info.groupId) {  // reset output buffer and computing status
-            pIndefInfo->groupId = pBlock->info.groupId;
-            pIndefInfo->pNextGroupRes = pBlock;
-            break;
-          }
-        }
-
-        doHandleDataBlock(pOperator, pBlock, downstream, pTaskInfo);
-        if (pInfo->pRes->info.rows >= pOperator->resultInfo.threshold) {
-          break;
-        }
-      }
-    }
-
-    doFilter(pIndefInfo->pCondition, pInfo->pRes, NULL);
-    size_t rows = pInfo->pRes->info.rows;
-    if (rows > 0 || pOperator->status == OP_EXEC_DONE) {
-      break;
-    } else {
-      blockDataCleanup(pInfo->pRes);
-    }
-  }
-
-  size_t rows = pInfo->pRes->info.rows;
-  pOperator->resultInfo.totalRows += rows;
-
-  if (pOperator->cost.openCost == 0) {
-    pOperator->cost.openCost = (taosGetTimestampUs() - st) / 1000.0;
-  }
-
-  return (rows > 0) ? pInfo->pRes : NULL;
-}
-
-SOperatorInfo* createIndefinitOutputOperatorInfo(SOperatorInfo* downstream, SPhysiNode* pNode,
-                                                 SExecTaskInfo* pTaskInfo) {
-  SIndefOperatorInfo* pInfo = taosMemoryCalloc(1, sizeof(SIndefOperatorInfo));
-  SOperatorInfo*      pOperator = taosMemoryCalloc(1, sizeof(SOperatorInfo));
-  if (pInfo == NULL || pOperator == NULL) {
-    goto _error;
-  }
-
-  SExprSupp* pSup = &pOperator->exprSupp;
-
-  SIndefRowsFuncPhysiNode* pPhyNode = (SIndefRowsFuncPhysiNode*)pNode;
-
-  int32_t    numOfExpr = 0;
-  SExprInfo* pExprInfo = createExprInfo(pPhyNode->pFuncs, NULL, &numOfExpr);
-
-  if (pPhyNode->pExprs != NULL) {
-    int32_t    num = 0;
-    SExprInfo* pSExpr = createExprInfo(pPhyNode->pExprs, NULL, &num);
-    int32_t    code = initExprSupp(&pInfo->scalarSup, pSExpr, num);
-    if (code != TSDB_CODE_SUCCESS) {
-      goto _error;
-    }
-  }
-
-  SSDataBlock* pResBlock = createResDataBlock(pPhyNode->node.pOutputDataBlockDesc);
-
-  int32_t numOfRows = 4096;
-  size_t  keyBufSize = sizeof(int64_t) + sizeof(int64_t) + POINTER_BYTES;
-
-  // Make sure the size of SSDataBlock will never exceed the size of 2MB.
-  int32_t TWOMB = 2 * 1024 * 1024;
-  if (numOfRows * pResBlock->info.rowSize > TWOMB) {
-    numOfRows = TWOMB / pResBlock->info.rowSize;
-  }
-
-  initResultSizeInfo(&pOperator->resultInfo, numOfRows);
-
-  initAggInfo(pSup, &pInfo->aggSup, pExprInfo, numOfExpr, keyBufSize, pTaskInfo->id.str);
-  initBasicInfo(&pInfo->binfo, pResBlock);
-
-  setFunctionResultOutput(pOperator, &pInfo->binfo, &pInfo->aggSup, MAIN_SCAN, numOfExpr);
-
-  pInfo->binfo.pRes = pResBlock;
-  pInfo->pCondition = pPhyNode->node.pConditions;
-  pInfo->pPseudoColInfo = setRowTsColumnOutputInfo(pSup->pCtx, numOfExpr);
-
-  pOperator->name = "IndefinitOperator";
-  pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_INDEF_ROWS_FUNC;
-  pOperator->blocking = false;
-  pOperator->status = OP_NOT_OPENED;
-  pOperator->info = pInfo;
-  pOperator->pTaskInfo = pTaskInfo;
-
-  pOperator->fpSet = createOperatorFpSet(operatorDummyOpenFn, doApplyIndefinitFunction, NULL, NULL,
-                                         destroyIndefinitOperatorInfo, NULL, NULL, NULL);
-
-  int32_t code = appendDownstream(pOperator, &downstream, 1);
-  if (code != TSDB_CODE_SUCCESS) {
-    goto _error;
-  }
-
-  return pOperator;
-
-_error:
-  taosMemoryFree(pInfo);
-  taosMemoryFree(pOperator);
-  pTaskInfo->code = TSDB_CODE_OUT_OF_MEMORY;
-  return NULL;
-}
-
 static int32_t initFillInfo(SFillOperatorInfo* pInfo, SExprInfo* pExpr, int32_t numOfCols, SNodeListNode* pValNode,
                             STimeWindow win, int32_t capacity, const char* id, SInterval* pInterval, int32_t fillType) {
   SFillColInfo* pColInfo = createFillColInfo(pExpr, numOfCols, pValNode);
@@ -4265,7 +3733,7 @@ SSchemaWrapper* extractQueriedColumnSchema(SScanPhysiNode* pScanNode) {
   }
 
   // this the tags and pseudo function columns, we only keep the tag columns
-  for(int32_t i = 0; i < numOfTags; ++i) {
+  for (int32_t i = 0; i < numOfTags; ++i) {
     STargetNode* pNode = (STargetNode*)nodesListGetNode(pScanNode->pScanPseudoCols, i);
 
     int32_t type = nodeType(pNode->pExpr);
@@ -4381,7 +3849,7 @@ int32_t generateGroupIdMap(STableListInfo* pTableListInfo, SReadHandle* pHandle,
   int32_t groupNum = 0;
   for (int32_t i = 0; i < taosArrayGetSize(pTableListInfo->pTableList); i++) {
     STableKeyInfo* info = taosArrayGet(pTableListInfo->pTableList, i);
-    int32_t code = getGroupIdFromTagsVal(pHandle->meta, info->uid, group, keyBuf, &info->groupId);
+    int32_t        code = getGroupIdFromTagsVal(pHandle->meta, info->uid, group, keyBuf, &info->groupId);
     if (code != TSDB_CODE_SUCCESS) {
       return code;
     }
@@ -4416,7 +3884,7 @@ static int32_t initTableblockDistQueryCond(uint64_t uid, SQueryTableDataCond* pC
 
   pCond->twindows = (STimeWindow){.skey = INT64_MIN, .ekey = INT64_MAX};
   pCond->suid = uid;
-  pCond->type = BLOCK_LOAD_OFFSET_ORDER;
+  pCond->type = TIMEWINDOW_RANGE_CONTAINED;
   pCond->startVersion = -1;
   pCond->endVersion = -1;
 
@@ -4504,7 +3972,6 @@ SOperatorInfo* createOperatorTree(SPhysiNode* pPhyNode, SExecTaskInfo* pTaskInfo
       return createSysTableScanOperatorInfo(pHandle, pSysScanPhyNode, pUser, pTaskInfo);
     } else if (QUERY_NODE_PHYSICAL_PLAN_TAG_SCAN == type) {
       STagScanPhysiNode* pScanPhyNode = (STagScanPhysiNode*)pPhyNode;
-
       int32_t code = getTableList(pHandle->meta, pHandle->vnode, pScanPhyNode, pTagCond, pTagIndexCond, pTableListInfo);
       if (code != TSDB_CODE_SUCCESS) {
         pTaskInfo->code = terrno;
@@ -4555,6 +4022,8 @@ SOperatorInfo* createOperatorTree(SPhysiNode* pPhyNode, SExecTaskInfo* pTaskInfo
       }
 
       return createLastrowScanOperator(pScanNode, pHandle, pTaskInfo);
+    } else if (QUERY_NODE_PHYSICAL_PLAN_PROJECT == type) {
+      return createProjectOperatorInfo(NULL, (SProjectPhysiNode*)pPhyNode, pTaskInfo);
     } else {
       ASSERT(0);
     }
@@ -4701,7 +4170,7 @@ SOperatorInfo* createOperatorTree(SPhysiNode* pPhyNode, SExecTaskInfo* pTaskInfo
   } else if (QUERY_NODE_PHYSICAL_PLAN_STREAM_STATE == type) {
     pOptr = createStreamStateAggOperatorInfo(ops[0], pPhyNode, pTaskInfo);
   } else if (QUERY_NODE_PHYSICAL_PLAN_MERGE_JOIN == type) {
-    pOptr = createMergeJoinOperatorInfo(ops, size, (SJoinPhysiNode*)pPhyNode, pTaskInfo);
+    pOptr = createMergeJoinOperatorInfo(ops, size, (SSortMergeJoinPhysiNode*)pPhyNode, pTaskInfo);
   } else if (QUERY_NODE_PHYSICAL_PLAN_FILL == type) {
     pOptr = createFillOperatorInfo(ops[0], (SFillPhysiNode*)pPhyNode, pTaskInfo);
   } else if (QUERY_NODE_PHYSICAL_PLAN_INDEF_ROWS_FUNC == type) {
diff --git a/source/libs/executor/src/joinoperator.c b/source/libs/executor/src/joinoperator.c
index 2e6c9bd351..f26b2f4f0a 100644
--- a/source/libs/executor/src/joinoperator.c
+++ b/source/libs/executor/src/joinoperator.c
@@ -28,30 +28,30 @@ static SSDataBlock* doMergeJoin(struct SOperatorInfo* pOperator);
 static void         destroyMergeJoinOperator(void* param, int32_t numOfOutput);
 static void         extractTimeCondition(SJoinOperatorInfo* Info, SLogicConditionNode* pLogicConditionNode);
 
-SOperatorInfo* createMergeJoinOperatorInfo(SOperatorInfo** pDownstream, int32_t numOfDownstream, SJoinPhysiNode* pJoinNode,
-    SExecTaskInfo* pTaskInfo) {
+SOperatorInfo* createMergeJoinOperatorInfo(SOperatorInfo** pDownstream, int32_t numOfDownstream,
+                                           SSortMergeJoinPhysiNode* pJoinNode, SExecTaskInfo* pTaskInfo) {
   SJoinOperatorInfo* pInfo = taosMemoryCalloc(1, sizeof(SJoinOperatorInfo));
   SOperatorInfo*     pOperator = taosMemoryCalloc(1, sizeof(SOperatorInfo));
   if (pOperator == NULL || pInfo == NULL) {
     goto _error;
   }
 
-  SSDataBlock*    pResBlock = createResDataBlock(pJoinNode->node.pOutputDataBlockDesc);
+  SSDataBlock* pResBlock = createResDataBlock(pJoinNode->node.pOutputDataBlockDesc);
 
-  int32_t numOfCols = 0;
+  int32_t    numOfCols = 0;
   SExprInfo* pExprInfo = createExprInfo(pJoinNode->pTargets, NULL, &numOfCols);
 
   initResultSizeInfo(&pOperator->resultInfo, 4096);
 
-  pInfo->pRes             = pResBlock;
-  pOperator->name         = "MergeJoinOperator";
+  pInfo->pRes = pResBlock;
+  pOperator->name = "MergeJoinOperator";
   pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_MERGE_JOIN;
-  pOperator->blocking     = false;
-  pOperator->status       = OP_NOT_OPENED;
-  pOperator->exprSupp.pExprInfo   = pExprInfo;
-  pOperator->exprSupp.numOfExprs  = numOfCols;
-  pOperator->info         = pInfo;
-  pOperator->pTaskInfo    = pTaskInfo;
+  pOperator->blocking = false;
+  pOperator->status = OP_NOT_OPENED;
+  pOperator->exprSupp.pExprInfo = pExprInfo;
+  pOperator->exprSupp.numOfExprs = numOfCols;
+  pOperator->info = pInfo;
+  pOperator->pTaskInfo = pTaskInfo;
 
   SNode* pMergeCondition = pJoinNode->pMergeCondition;
   if (nodeType(pMergeCondition) == QUERY_NODE_OPERATOR) {
@@ -104,7 +104,7 @@ void setJoinColumnInfo(SColumnInfo* pColumn, const SColumnNode* pColumnNode) {
 void destroyMergeJoinOperator(void* param, int32_t numOfOutput) {
   SJoinOperatorInfo* pJoinOperator = (SJoinOperatorInfo*)param;
   nodesDestroyNode(pJoinOperator->pCondAfterMerge);
-  
+
   taosMemoryFreeClear(param);
 }
 
diff --git a/source/libs/executor/src/projectoperator.c b/source/libs/executor/src/projectoperator.c
new file mode 100644
index 0000000000..34149d7499
--- /dev/null
+++ b/source/libs/executor/src/projectoperator.c
@@ -0,0 +1,590 @@
+/*
+ * Copyright (c) 2019 TAOS Data, Inc. <jhtao@taosdata.com>
+ *
+ * This program is free software: you can use, redistribute, and/or modify
+ * it under the terms of the GNU Affero General Public License, version 3
+ * or later ("AGPL"), as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.
+ *
+ * You should have received a copy of the GNU Affero General Public License
+ * along with this program. If not, see <http://www.gnu.org/licenses/>.
+ */
+
+#include "executorimpl.h"
+#include "functionMgt.h"
+
+static SSDataBlock* doGenerateSourceData(SOperatorInfo* pOperator);
+static SSDataBlock* doProjectOperation(SOperatorInfo* pOperator);
+static SSDataBlock* doApplyIndefinitFunction(SOperatorInfo* pOperator);
+static SArray*      setRowTsColumnOutputInfo(SqlFunctionCtx* pCtx, int32_t numOfCols);
+static void setFunctionResultOutput(SOperatorInfo* pOperator, SOptrBasicInfo* pInfo, SAggSupporter* pSup, int32_t stage,
+                                    int32_t numOfExprs);
+
+static void destroyProjectOperatorInfo(void* param, int32_t numOfOutput) {
+  if (NULL == param) {
+    return;
+  }
+
+  SProjectOperatorInfo* pInfo = (SProjectOperatorInfo*)param;
+  cleanupBasicInfo(&pInfo->binfo);
+  cleanupAggSup(&pInfo->aggSup);
+  taosArrayDestroy(pInfo->pPseudoColInfo);
+
+  blockDataDestroy(pInfo->pFinalRes);
+  taosMemoryFreeClear(param);
+}
+
+static void destroyIndefinitOperatorInfo(void* param, int32_t numOfOutput) {
+  SIndefOperatorInfo* pInfo = (SIndefOperatorInfo*)param;
+  cleanupBasicInfo(&pInfo->binfo);
+
+  taosArrayDestroy(pInfo->pPseudoColInfo);
+  cleanupAggSup(&pInfo->aggSup);
+  cleanupExprSupp(&pInfo->scalarSup);
+
+  taosMemoryFreeClear(param);
+}
+
+SOperatorInfo* createProjectOperatorInfo(SOperatorInfo* downstream, SProjectPhysiNode* pProjPhyNode,
+                                         SExecTaskInfo* pTaskInfo) {
+  SProjectOperatorInfo* pInfo = taosMemoryCalloc(1, sizeof(SProjectOperatorInfo));
+  SOperatorInfo*        pOperator = taosMemoryCalloc(1, sizeof(SOperatorInfo));
+  if (pInfo == NULL || pOperator == NULL) {
+    goto _error;
+  }
+
+  int32_t    numOfCols = 0;
+  SExprInfo* pExprInfo = createExprInfo(pProjPhyNode->pProjections, NULL, &numOfCols);
+
+  SSDataBlock* pResBlock = createResDataBlock(pProjPhyNode->node.pOutputDataBlockDesc);
+  initLimitInfo(pProjPhyNode->node.pLimit, pProjPhyNode->node.pSlimit, &pInfo->limitInfo);
+
+  pInfo->binfo.pRes = pResBlock;
+  pInfo->pFinalRes  = createOneDataBlock(pResBlock, false);
+  pInfo->pFilterNode = pProjPhyNode->node.pConditions;
+  pInfo->mergeDataBlocks = pProjPhyNode->mergeDataBlock;
+
+  // todo remove it soon
+  if (pTaskInfo->execModel == OPTR_EXEC_MODEL_STREAM) {
+    pInfo->mergeDataBlocks = true;
+  }
+
+  int32_t numOfRows = 4096;
+  size_t  keyBufSize = sizeof(int64_t) + sizeof(int64_t) + POINTER_BYTES;
+
+  // Make sure the size of SSDataBlock will never exceed the size of 2MB.
+  int32_t TWOMB = 2 * 1024 * 1024;
+  if (numOfRows * pResBlock->info.rowSize > TWOMB) {
+    numOfRows = TWOMB / pResBlock->info.rowSize;
+  }
+  initResultSizeInfo(&pOperator->resultInfo, numOfRows);
+
+  initAggInfo(&pOperator->exprSupp, &pInfo->aggSup, pExprInfo, numOfCols, keyBufSize, pTaskInfo->id.str);
+  initBasicInfo(&pInfo->binfo, pResBlock);
+  setFunctionResultOutput(pOperator, &pInfo->binfo, &pInfo->aggSup, MAIN_SCAN, numOfCols);
+
+  pInfo->pPseudoColInfo = setRowTsColumnOutputInfo(pOperator->exprSupp.pCtx, numOfCols);
+  pOperator->name         = "ProjectOperator";
+  pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_PROJECT;
+  pOperator->blocking     = false;
+  pOperator->status       = OP_NOT_OPENED;
+  pOperator->info         = pInfo;
+  pOperator->pTaskInfo    = pTaskInfo;
+
+  pOperator->fpSet = createOperatorFpSet(operatorDummyOpenFn, doProjectOperation, NULL, NULL,
+                                         destroyProjectOperatorInfo, NULL, NULL, NULL);
+
+  int32_t code = appendDownstream(pOperator, &downstream, 1);
+  if (code != TSDB_CODE_SUCCESS) {
+    goto _error;
+  }
+
+  return pOperator;
+
+  _error:
+  pTaskInfo->code = TSDB_CODE_OUT_OF_MEMORY;
+  return NULL;
+}
+
+static int32_t discardGroupDataBlock(SSDataBlock* pBlock, SLimitInfo* pLimitInfo) {
+  if (pLimitInfo->remainGroupOffset > 0) {
+    // it is the first group
+    if (pLimitInfo->currentGroupId == 0 || pLimitInfo->currentGroupId == pBlock->info.groupId) {
+      pLimitInfo->currentGroupId = pBlock->info.groupId;
+      return PROJECT_RETRIEVE_CONTINUE;
+    } else if (pLimitInfo->currentGroupId != pBlock->info.groupId) {
+      // now it is the data from a new group
+      pLimitInfo->remainGroupOffset -= 1;
+      pLimitInfo->currentGroupId = pBlock->info.groupId;
+
+      // ignore data block in current group
+      if (pLimitInfo->remainGroupOffset > 0) {
+        return PROJECT_RETRIEVE_CONTINUE;
+      }
+    }
+
+    // set current group id of the project operator
+    pLimitInfo->currentGroupId = pBlock->info.groupId;
+  }
+
+  return PROJECT_RETRIEVE_DONE;
+}
+
+static int32_t setInfoForNewGroup(SSDataBlock* pBlock, SLimitInfo* pLimitInfo, SOperatorInfo* pOperator) {
+  // remainGroupOffset == 0
+  // here check for a new group data, we need to handle the data of the previous group.
+  ASSERT(pLimitInfo->remainGroupOffset == 0 || pLimitInfo->remainGroupOffset == -1);
+
+  if (pLimitInfo->currentGroupId != 0 && pLimitInfo->currentGroupId != pBlock->info.groupId) {
+    pLimitInfo->numOfOutputGroups += 1;
+    if ((pLimitInfo->slimit.limit > 0) && (pLimitInfo->slimit.limit <= pLimitInfo->numOfOutputGroups)) {
+      doSetOperatorCompleted(pOperator);
+      return PROJECT_RETRIEVE_DONE;
+    }
+
+    // reset the value for a new group data
+    // existing rows that belongs to previous group.
+    pLimitInfo->numOfOutputRows = 0;
+    pLimitInfo->remainOffset = pLimitInfo->limit.offset;
+  }
+
+  return PROJECT_RETRIEVE_DONE;
+}
+
+static int32_t doIngroupLimitOffset(SLimitInfo* pLimitInfo, uint64_t groupId, SSDataBlock* pBlock, SOperatorInfo* pOperator) {
+  // set current group id
+  pLimitInfo->currentGroupId = groupId;
+
+  if (pLimitInfo->remainOffset >= pBlock->info.rows) {
+    pLimitInfo->remainOffset -= pBlock->info.rows;
+    blockDataCleanup(pBlock);
+    return PROJECT_RETRIEVE_CONTINUE;
+  } else if (pLimitInfo->remainOffset < pBlock->info.rows && pLimitInfo->remainOffset > 0) {
+    blockDataTrimFirstNRows(pBlock, pLimitInfo->remainOffset);
+    pLimitInfo->remainOffset = 0;
+  }
+
+  // check for the limitation in each group
+  if (pLimitInfo->limit.limit >= 0 &&
+      pLimitInfo->numOfOutputRows + pBlock->info.rows >= pLimitInfo->limit.limit) {
+    int32_t keepRows = (int32_t)(pLimitInfo->limit.limit - pLimitInfo->numOfOutputRows);
+    blockDataKeepFirstNRows(pBlock, keepRows);
+    if (pLimitInfo->slimit.limit > 0 && pLimitInfo->slimit.limit <= pLimitInfo->numOfOutputGroups) {
+      doSetOperatorCompleted(pOperator);
+    }
+  }
+
+  pLimitInfo->numOfOutputRows += pBlock->info.rows;
+  return PROJECT_RETRIEVE_DONE;
+}
+
+SSDataBlock* doProjectOperation(SOperatorInfo* pOperator) {
+  SProjectOperatorInfo* pProjectInfo = pOperator->info;
+  SOptrBasicInfo*       pInfo = &pProjectInfo->binfo;
+
+  SExprSupp*   pSup = &pOperator->exprSupp;
+  SSDataBlock* pRes = pInfo->pRes;
+  SSDataBlock* pFinalRes = pProjectInfo->pFinalRes;
+
+  blockDataCleanup(pFinalRes);
+
+  SExecTaskInfo* pTaskInfo = pOperator->pTaskInfo;
+  if (pOperator->status == OP_EXEC_DONE) {
+    if (pTaskInfo->execModel == OPTR_EXEC_MODEL_QUEUE) {
+      pOperator->status = OP_OPENED;
+      return NULL;
+    }
+
+    return NULL;
+  }
+
+  int64_t st = 0;
+  int32_t order = 0;
+  int32_t scanFlag = 0;
+
+  if (pOperator->cost.openCost == 0) {
+    st = taosGetTimestampUs();
+  }
+
+  SOperatorInfo* downstream = pOperator->pDownstream[0];
+  SLimitInfo* pLimitInfo = &pProjectInfo->limitInfo;
+
+  if (downstream == NULL) {
+    return doGenerateSourceData(pOperator);
+  }
+
+  while (1) {
+    while (1) {
+      blockDataCleanup(pRes);
+
+      // The downstream exec may change the value of the newgroup, so use a local variable instead.
+      SSDataBlock* pBlock = downstream->fpSet.getNextFn(downstream);
+      if (pBlock == NULL) {
+        doSetOperatorCompleted(pOperator);
+        break;
+      }
+
+      // for stream interval
+      if (pBlock->info.type == STREAM_RETRIEVE) {
+        return pBlock;
+      }
+
+      int32_t status = discardGroupDataBlock(pBlock, pLimitInfo);
+      if (status == PROJECT_RETRIEVE_CONTINUE) {
+        continue;
+      }
+
+      setInfoForNewGroup(pBlock, pLimitInfo, pOperator);
+      if (pOperator->status == OP_EXEC_DONE) {
+        break;
+      }
+
+      // the pDataBlock are always the same one, no need to call this again
+      int32_t code = getTableScanInfo(downstream, &order, &scanFlag);
+      if (code != TSDB_CODE_SUCCESS) {
+        longjmp(pTaskInfo->env, code);
+      }
+
+      setInputDataBlock(pOperator, pSup->pCtx, pBlock, order, scanFlag, false);
+      blockDataEnsureCapacity(pInfo->pRes, pInfo->pRes->info.rows + pBlock->info.rows);
+
+      code = projectApplyFunctions(pSup->pExprInfo, pInfo->pRes, pBlock, pSup->pCtx, pSup->numOfExprs,
+                                   pProjectInfo->pPseudoColInfo);
+      if (code != TSDB_CODE_SUCCESS) {
+        longjmp(pTaskInfo->env, code);
+      }
+
+      status = doIngroupLimitOffset(pLimitInfo, pBlock->info.groupId, pInfo->pRes, pOperator);
+      if (status == PROJECT_RETRIEVE_CONTINUE) {
+        continue;
+      }
+
+      break;
+    }
+
+    if (pProjectInfo->mergeDataBlocks) {
+      if (pRes->info.rows > 0) {
+        pFinalRes->info.groupId = pRes->info.groupId;
+        pFinalRes->info.version = pRes->info.version;
+
+        // continue merge data, ignore the group id
+        blockDataMerge(pFinalRes, pRes);
+        if (pFinalRes->info.rows + pRes->info.rows <= pOperator->resultInfo.threshold) {
+          continue;
+        }
+      }
+
+      // do apply filter
+      doFilter(pProjectInfo->pFilterNode, pFinalRes, NULL);
+      if (pFinalRes->info.rows > 0 || pRes->info.rows == 0) {
+        break;
+      }
+    } else {
+      // do apply filter
+      if (pRes->info.rows > 0) {
+        doFilter(pProjectInfo->pFilterNode, pRes, NULL);
+        if (pRes->info.rows == 0) {
+          continue;
+        }
+      }
+
+      // no results generated
+      break;
+    }
+  }
+
+  SSDataBlock* p = pProjectInfo->mergeDataBlocks ? pFinalRes : pRes;
+  pOperator->resultInfo.totalRows += p->info.rows;
+
+  if (pOperator->cost.openCost == 0) {
+    pOperator->cost.openCost = (taosGetTimestampUs() - st) / 1000.0;
+  }
+
+  return (p->info.rows > 0) ? p : NULL;
+}
+
+SOperatorInfo* createIndefinitOutputOperatorInfo(SOperatorInfo* downstream, SPhysiNode* pNode,
+                                                 SExecTaskInfo* pTaskInfo) {
+  SIndefOperatorInfo* pInfo = taosMemoryCalloc(1, sizeof(SIndefOperatorInfo));
+  SOperatorInfo*      pOperator = taosMemoryCalloc(1, sizeof(SOperatorInfo));
+  if (pInfo == NULL || pOperator == NULL) {
+    goto _error;
+  }
+
+  SExprSupp* pSup = &pOperator->exprSupp;
+
+  SIndefRowsFuncPhysiNode* pPhyNode = (SIndefRowsFuncPhysiNode*)pNode;
+
+  int32_t    numOfExpr = 0;
+  SExprInfo* pExprInfo = createExprInfo(pPhyNode->pFuncs, NULL, &numOfExpr);
+
+  if (pPhyNode->pExprs != NULL) {
+    int32_t    num = 0;
+    SExprInfo* pSExpr = createExprInfo(pPhyNode->pExprs, NULL, &num);
+    int32_t    code = initExprSupp(&pInfo->scalarSup, pSExpr, num);
+    if (code != TSDB_CODE_SUCCESS) {
+      goto _error;
+    }
+  }
+
+  SSDataBlock* pResBlock = createResDataBlock(pPhyNode->node.pOutputDataBlockDesc);
+
+  int32_t numOfRows = 4096;
+  size_t  keyBufSize = sizeof(int64_t) + sizeof(int64_t) + POINTER_BYTES;
+
+  // Make sure the size of SSDataBlock will never exceed the size of 2MB.
+  int32_t TWOMB = 2 * 1024 * 1024;
+  if (numOfRows * pResBlock->info.rowSize > TWOMB) {
+    numOfRows = TWOMB / pResBlock->info.rowSize;
+  }
+
+  initResultSizeInfo(&pOperator->resultInfo, numOfRows);
+
+  initAggInfo(pSup, &pInfo->aggSup, pExprInfo, numOfExpr, keyBufSize, pTaskInfo->id.str);
+  initBasicInfo(&pInfo->binfo, pResBlock);
+
+  setFunctionResultOutput(pOperator, &pInfo->binfo, &pInfo->aggSup, MAIN_SCAN, numOfExpr);
+
+  pInfo->binfo.pRes = pResBlock;
+  pInfo->pCondition = pPhyNode->node.pConditions;
+  pInfo->pPseudoColInfo = setRowTsColumnOutputInfo(pSup->pCtx, numOfExpr);
+
+  pOperator->name = "IndefinitOperator";
+  pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_INDEF_ROWS_FUNC;
+  pOperator->blocking = false;
+  pOperator->status = OP_NOT_OPENED;
+  pOperator->info = pInfo;
+  pOperator->pTaskInfo = pTaskInfo;
+
+  pOperator->fpSet = createOperatorFpSet(operatorDummyOpenFn, doApplyIndefinitFunction, NULL, NULL,
+                                         destroyIndefinitOperatorInfo, NULL, NULL, NULL);
+
+  int32_t code = appendDownstream(pOperator, &downstream, 1);
+  if (code != TSDB_CODE_SUCCESS) {
+    goto _error;
+  }
+
+  return pOperator;
+
+  _error:
+  taosMemoryFree(pInfo);
+  taosMemoryFree(pOperator);
+  pTaskInfo->code = TSDB_CODE_OUT_OF_MEMORY;
+  return NULL;
+}
+
+static void doHandleDataBlock(SOperatorInfo* pOperator, SSDataBlock* pBlock, SOperatorInfo* downstream,
+                              SExecTaskInfo* pTaskInfo) {
+  int32_t order = 0;
+  int32_t scanFlag = 0;
+
+  SIndefOperatorInfo* pIndefInfo = pOperator->info;
+  SOptrBasicInfo*     pInfo = &pIndefInfo->binfo;
+  SExprSupp*          pSup = &pOperator->exprSupp;
+
+  // the pDataBlock are always the same one, no need to call this again
+  int32_t code = getTableScanInfo(downstream, &order, &scanFlag);
+  if (code != TSDB_CODE_SUCCESS) {
+    longjmp(pTaskInfo->env, code);
+  }
+
+  // there is an scalar expression that needs to be calculated before apply the group aggregation.
+  SExprSupp* pScalarSup = &pIndefInfo->scalarSup;
+  if (pScalarSup->pExprInfo != NULL) {
+    code = projectApplyFunctions(pScalarSup->pExprInfo, pBlock, pBlock, pScalarSup->pCtx, pScalarSup->numOfExprs,
+                                 pIndefInfo->pPseudoColInfo);
+    if (code != TSDB_CODE_SUCCESS) {
+      longjmp(pTaskInfo->env, code);
+    }
+  }
+
+  setInputDataBlock(pOperator, pSup->pCtx, pBlock, order, scanFlag, false);
+  blockDataEnsureCapacity(pInfo->pRes, pInfo->pRes->info.rows + pBlock->info.rows);
+
+  code = projectApplyFunctions(pSup->pExprInfo, pInfo->pRes, pBlock, pSup->pCtx, pSup->numOfExprs,
+                               pIndefInfo->pPseudoColInfo);
+  if (code != TSDB_CODE_SUCCESS) {
+    longjmp(pTaskInfo->env, code);
+  }
+}
+
+SSDataBlock* doApplyIndefinitFunction(SOperatorInfo* pOperator) {
+  SIndefOperatorInfo* pIndefInfo = pOperator->info;
+  SOptrBasicInfo*     pInfo = &pIndefInfo->binfo;
+  SExprSupp*          pSup = &pOperator->exprSupp;
+
+  SSDataBlock* pRes = pInfo->pRes;
+  blockDataCleanup(pRes);
+
+  SExecTaskInfo* pTaskInfo = pOperator->pTaskInfo;
+  if (pOperator->status == OP_EXEC_DONE) {
+    return NULL;
+  }
+
+  int64_t st = 0;
+
+  if (pOperator->cost.openCost == 0) {
+    st = taosGetTimestampUs();
+  }
+
+  SOperatorInfo* downstream = pOperator->pDownstream[0];
+
+  while (1) {
+    // here we need to handle the existsed group results
+    if (pIndefInfo->pNextGroupRes != NULL) {  // todo extract method
+      for (int32_t k = 0; k < pSup->numOfExprs; ++k) {
+        SqlFunctionCtx* pCtx = &pSup->pCtx[k];
+
+        SResultRowEntryInfo* pResInfo = GET_RES_INFO(pCtx);
+        pResInfo->initialized = false;
+        pCtx->pOutput = NULL;
+      }
+
+      doHandleDataBlock(pOperator, pIndefInfo->pNextGroupRes, downstream, pTaskInfo);
+      pIndefInfo->pNextGroupRes = NULL;
+    }
+
+    if (pInfo->pRes->info.rows < pOperator->resultInfo.threshold) {
+      while (1) {
+        // The downstream exec may change the value of the newgroup, so use a local variable instead.
+        SSDataBlock* pBlock = downstream->fpSet.getNextFn(downstream);
+        if (pBlock == NULL) {
+          doSetOperatorCompleted(pOperator);
+          break;
+        }
+
+        if (pIndefInfo->groupId == 0 && pBlock->info.groupId != 0) {
+          pIndefInfo->groupId = pBlock->info.groupId;  // this is the initial group result
+        } else {
+          if (pIndefInfo->groupId != pBlock->info.groupId) {  // reset output buffer and computing status
+            pIndefInfo->groupId = pBlock->info.groupId;
+            pIndefInfo->pNextGroupRes = pBlock;
+            break;
+          }
+        }
+
+        doHandleDataBlock(pOperator, pBlock, downstream, pTaskInfo);
+        if (pInfo->pRes->info.rows >= pOperator->resultInfo.threshold) {
+          break;
+        }
+      }
+    }
+
+    doFilter(pIndefInfo->pCondition, pInfo->pRes, NULL);
+    size_t rows = pInfo->pRes->info.rows;
+    if (rows > 0 || pOperator->status == OP_EXEC_DONE) {
+      break;
+    } else {
+      blockDataCleanup(pInfo->pRes);
+    }
+  }
+
+  size_t rows = pInfo->pRes->info.rows;
+  pOperator->resultInfo.totalRows += rows;
+
+  if (pOperator->cost.openCost == 0) {
+    pOperator->cost.openCost = (taosGetTimestampUs() - st) / 1000.0;
+  }
+
+  return (rows > 0) ? pInfo->pRes : NULL;
+}
+
+void initCtxOutputBuffer(SqlFunctionCtx* pCtx, int32_t size) {
+  for (int32_t j = 0; j < size; ++j) {
+    struct SResultRowEntryInfo* pResInfo = GET_RES_INFO(&pCtx[j]);
+    if (isRowEntryInitialized(pResInfo) || fmIsPseudoColumnFunc(pCtx[j].functionId) || pCtx[j].functionId == -1 ||
+        fmIsScalarFunc(pCtx[j].functionId)) {
+      continue;
+    }
+
+    pCtx[j].fpSet.init(&pCtx[j], pCtx[j].resultInfo);
+  }
+}
+
+/*
+ * The start of each column SResultRowEntryInfo is denote by RowCellInfoOffset.
+ * Note that in case of top/bottom query, the whole multiple rows of result is treated as only one row of results.
+ * +------------+-----------------result column 1------------+------------------result column 2-----------+
+ * | SResultRow | SResultRowEntryInfo | intermediate buffer1 | SResultRowEntryInfo | intermediate buffer 2|
+ * +------------+--------------------------------------------+--------------------------------------------+
+ *           offset[0]                                  offset[1]                                   offset[2]
+ */
+// TODO refactor: some function move away
+void setFunctionResultOutput(SOperatorInfo* pOperator, SOptrBasicInfo* pInfo, SAggSupporter* pSup, int32_t stage,
+                             int32_t numOfExprs) {
+  SExecTaskInfo*  pTaskInfo = pOperator->pTaskInfo;
+  SqlFunctionCtx* pCtx = pOperator->exprSupp.pCtx;
+  int32_t*        rowEntryInfoOffset = pOperator->exprSupp.rowEntryInfoOffset;
+
+  SResultRowInfo* pResultRowInfo = &pInfo->resultRowInfo;
+  initResultRowInfo(pResultRowInfo);
+
+  int64_t     tid = 0;
+  int64_t     groupId = 0;
+  SResultRow* pRow = doSetResultOutBufByKey(pSup->pResultBuf, pResultRowInfo, (char*)&tid, sizeof(tid), true, groupId,
+                                            pTaskInfo, false, pSup);
+
+  for (int32_t i = 0; i < numOfExprs; ++i) {
+    struct SResultRowEntryInfo* pEntry = getResultEntryInfo(pRow, i, rowEntryInfoOffset);
+    cleanupResultRowEntry(pEntry);
+
+    pCtx[i].resultInfo = pEntry;
+    pCtx[i].scanFlag = stage;
+  }
+
+  initCtxOutputBuffer(pCtx, numOfExprs);
+}
+
+SArray* setRowTsColumnOutputInfo(SqlFunctionCtx* pCtx, int32_t numOfCols) {
+  SArray* pList = taosArrayInit(4, sizeof(int32_t));
+  for (int32_t i = 0; i < numOfCols; ++i) {
+    if (fmIsPseudoColumnFunc(pCtx[i].functionId)) {
+      taosArrayPush(pList, &i);
+    }
+  }
+
+  return pList;
+}
+
+SSDataBlock* doGenerateSourceData(SOperatorInfo* pOperator) {
+  SProjectOperatorInfo* pProjectInfo = pOperator->info;
+
+  SExprSupp*   pSup = &pOperator->exprSupp;
+  SSDataBlock* pRes = pProjectInfo->binfo.pRes;
+
+  blockDataEnsureCapacity(pRes, pOperator->resultInfo.capacity);
+  SExprInfo* pExpr = pSup->pExprInfo;
+
+  int64_t st = taosGetTimestampUs();
+
+  for (int32_t k = 0; k < pSup->numOfExprs; ++k) {
+    int32_t outputSlotId = pExpr[k].base.resSchema.slotId;
+
+    ASSERT(pExpr[k].pExpr->nodeType == QUERY_NODE_VALUE);
+    SColumnInfoData* pColInfoData = taosArrayGet(pRes->pDataBlock, outputSlotId);
+
+    int32_t type = pExpr[k].base.pParam[0].param.nType;
+    if (TSDB_DATA_TYPE_NULL == type) {
+      colDataAppendNNULL(pColInfoData, 0, 1);
+    } else {
+      colDataAppend(pColInfoData, 0, taosVariantGet(&pExpr[k].base.pParam[0].param, type), false);
+    }
+  }
+
+  pRes->info.rows = 1;
+  doFilter(pProjectInfo->pFilterNode, pRes, NULL);
+
+  /*int32_t status = */doIngroupLimitOffset(&pProjectInfo->limitInfo, 0, pRes, pOperator);
+
+  pOperator->resultInfo.totalRows += pRes->info.rows;
+
+  doSetOperatorCompleted(pOperator);
+  if (pOperator->cost.openCost == 0) {
+    pOperator->cost.openCost = (taosGetTimestampUs() - st) / 1000.0;
+  }
+
+  return (pRes->info.rows > 0) ? pRes : NULL;
+}
\ No newline at end of file
diff --git a/source/libs/executor/src/scanoperator.c b/source/libs/executor/src/scanoperator.c
index a9d03aebbe..2dcb555834 100644
--- a/source/libs/executor/src/scanoperator.c
+++ b/source/libs/executor/src/scanoperator.c
@@ -274,7 +274,7 @@ static int32_t loadDataBlock(SOperatorInfo* pOperator, STableScanInfo* pTableSca
     qDebug("%s data block filter out, brange:%" PRId64 "-%" PRId64 ", rows:%d", GET_TASKID(pTaskInfo),
            pBlockInfo->window.skey, pBlockInfo->window.ekey, pBlockInfo->rows);
   } else {
-    qDebug("%s data block filter out, elapsed time:%"PRId64, GET_TASKID(pTaskInfo), (et - st));
+    qDebug("%s data block filter out, elapsed time:%" PRId64, GET_TASKID(pTaskInfo), (et - st));
   }
 
   return TSDB_CODE_SUCCESS;
@@ -1838,11 +1838,14 @@ static SSDataBlock* sysTableScanUserTags(SOperatorInfo* pOperator) {
       int8_t tagType = smr.me.stbEntry.schemaTag.pSchema[i].type;
       pColInfoData = taosArrayGet(p->pDataBlock, 4);
       char tagTypeStr[VARSTR_HEADER_SIZE + 32];
-      int tagTypeLen = sprintf(varDataVal(tagTypeStr), "%s", tDataTypes[tagType].name);
+      int  tagTypeLen = sprintf(varDataVal(tagTypeStr), "%s", tDataTypes[tagType].name);
       if (tagType == TSDB_DATA_TYPE_VARCHAR) {
-        tagTypeLen += sprintf(varDataVal(tagTypeStr) + tagTypeLen, "(%d)", (int32_t)(smr.me.stbEntry.schemaTag.pSchema[i].bytes - VARSTR_HEADER_SIZE));
+        tagTypeLen += sprintf(varDataVal(tagTypeStr) + tagTypeLen, "(%d)",
+                              (int32_t)(smr.me.stbEntry.schemaTag.pSchema[i].bytes - VARSTR_HEADER_SIZE));
       } else if (tagType == TSDB_DATA_TYPE_NCHAR) {
-        tagTypeLen += sprintf(varDataVal(tagTypeStr) + tagTypeLen, "(%d)", (int32_t)((smr.me.stbEntry.schemaTag.pSchema[i].bytes - VARSTR_HEADER_SIZE) / TSDB_NCHAR_SIZE));
+        tagTypeLen +=
+            sprintf(varDataVal(tagTypeStr) + tagTypeLen, "(%d)",
+                    (int32_t)((smr.me.stbEntry.schemaTag.pSchema[i].bytes - VARSTR_HEADER_SIZE) / TSDB_NCHAR_SIZE));
       }
       varDataSetLen(tagTypeStr, tagTypeLen);
       colDataAppend(pColInfoData, numOfRows, (char*)tagTypeStr, false);
@@ -2527,49 +2530,6 @@ _error:
   return NULL;
 }
 
-typedef struct STableMergeScanInfo {
-  STableListInfo* tableListInfo;
-  int32_t         tableStartIndex;
-  int32_t         tableEndIndex;
-  bool            hasGroupId;
-  uint64_t        groupId;
-  SArray*         dataReaders;  // array of tsdbReaderT*
-  SReadHandle     readHandle;
-  int32_t         bufPageSize;
-  uint32_t        sortBufSize;  // max buffer size for in-memory sort
-  SArray*         pSortInfo;
-  SSortHandle*    pSortHandle;
-
-  SSDataBlock* pSortInputBlock;
-  int64_t      startTs;  // sort start time
-  SArray*      sortSourceParams;
-
-  SFileBlockLoadRecorder readRecorder;
-  int64_t                numOfRows;
-  SScanInfo              scanInfo;
-  int32_t                scanTimes;
-  SNode*                 pFilterNode;  // filter info, which is push down by optimizer
-  SqlFunctionCtx*        pCtx;         // which belongs to the direct upstream operator operator query context
-  SResultRowInfo*        pResultRowInfo;
-  int32_t*               rowEntryInfoOffset;
-  SExprInfo*             pExpr;
-  SSDataBlock*           pResBlock;
-  SArray*                pColMatchInfo;
-  int32_t                numOfOutput;
-
-  SExprInfo*      pPseudoExpr;
-  int32_t         numOfPseudoExpr;
-  SqlFunctionCtx* pPseudoCtx;
-
-  SQueryTableDataCond cond;
-  int32_t             scanFlag;  // table scan flag to denote if it is a repeat/reverse/main scan
-  int32_t             dataBlockLoadFlag;
-  // if the upstream is an interval operator, the interval info is also kept here to get the time
-  // window to check if current data block needs to be loaded.
-  SInterval       interval;
-  SSampleExecInfo sample;  // sample execution info
-} STableMergeScanInfo;
-
 int32_t createScanTableListInfo(SScanPhysiNode* pScanNode, SNodeList* pGroupTags, bool groupSort, SReadHandle* pHandle,
                                 STableListInfo* pTableListInfo, SNode* pTagCond, SNode* pTagIndexCond,
                                 const char* idStr) {
@@ -2700,9 +2660,9 @@ static int32_t loadDataBlockFromOneTable(SOperatorInfo* pOperator, STableMergeSc
   relocateColumnData(pBlock, pTableScanInfo->pColMatchInfo, pCols, true);
 
   // currently only the tbname pseudo column
-  if (pTableScanInfo->numOfPseudoExpr > 0) {
-    int32_t code = addTagPseudoColumnData(&pTableScanInfo->readHandle, pTableScanInfo->pPseudoExpr,
-                                          pTableScanInfo->numOfPseudoExpr, pBlock, GET_TASKID(pTaskInfo));
+  if (pTableScanInfo->pseudoSup.numOfExprs > 0) {
+    int32_t code = addTagPseudoColumnData(&pTableScanInfo->readHandle, pTableScanInfo->pseudoSup.pExprInfo,
+                                          pTableScanInfo->pseudoSup.numOfExprs, pBlock, GET_TASKID(pTaskInfo));
     if (code != TSDB_CODE_SUCCESS) {
       longjmp(pTaskInfo->env, code);
     }
@@ -2869,29 +2829,31 @@ int32_t stopGroupTableMergeScan(SOperatorInfo* pOperator) {
   STableMergeScanInfo* pInfo = pOperator->info;
   SExecTaskInfo*       pTaskInfo = pOperator->pTaskInfo;
 
-  tsortDestroySortHandle(pInfo->pSortHandle);
+  size_t numReaders = taosArrayGetSize(pInfo->dataReaders);
+
+  for (int32_t i = 0; i < numReaders; ++i) {
+    STableMergeScanSortSourceParam* param = taosArrayGet(pInfo->sortSourceParams, i);
+    blockDataDestroy(param->inputBlock);
+  }
   taosArrayClear(pInfo->sortSourceParams);
 
-  for (int32_t i = 0; i < taosArrayGetSize(pInfo->dataReaders); ++i) {
+  tsortDestroySortHandle(pInfo->pSortHandle);
+
+  for (int32_t i = 0; i < numReaders; ++i) {
     STsdbReader* reader = taosArrayGetP(pInfo->dataReaders, i);
     tsdbReaderClose(reader);
   }
-
   taosArrayDestroy(pInfo->dataReaders);
   pInfo->dataReaders = NULL;
   return TSDB_CODE_SUCCESS;
 }
 
-SSDataBlock* getSortedTableMergeScanBlockData(SSortHandle* pHandle, int32_t capacity, SOperatorInfo* pOperator) {
+SSDataBlock* getSortedTableMergeScanBlockData(SSortHandle* pHandle, SSDataBlock* pResBlock, int32_t capacity, SOperatorInfo* pOperator) {
   STableMergeScanInfo* pInfo = pOperator->info;
   SExecTaskInfo*       pTaskInfo = pOperator->pTaskInfo;
 
-  SSDataBlock* p = tsortGetSortedDataBlock(pHandle);
-  if (p == NULL) {
-    return NULL;
-  }
-
-  blockDataEnsureCapacity(p, capacity);
+  blockDataCleanup(pResBlock);
+  blockDataEnsureCapacity(pResBlock, capacity);
 
   while (1) {
     STupleHandle* pTupleHandle = tsortNextTuple(pHandle);
@@ -2899,14 +2861,15 @@ SSDataBlock* getSortedTableMergeScanBlockData(SSortHandle* pHandle, int32_t capa
       break;
     }
 
-    appendOneRowToDataBlock(p, pTupleHandle);
-    if (p->info.rows >= capacity) {
+    appendOneRowToDataBlock(pResBlock, pTupleHandle);
+    if (pResBlock->info.rows >= capacity) {
       break;
     }
   }
 
-  qDebug("%s get sorted row blocks, rows:%d", GET_TASKID(pTaskInfo), p->info.rows);
-  return (p->info.rows > 0) ? p : NULL;
+
+  qDebug("%s get sorted row blocks, rows:%d", GET_TASKID(pTaskInfo), pResBlock->info.rows);
+  return (pResBlock->info.rows > 0) ? pResBlock : NULL;
 }
 
 SSDataBlock* doTableMergeScan(SOperatorInfo* pOperator) {
@@ -2935,7 +2898,7 @@ SSDataBlock* doTableMergeScan(SOperatorInfo* pOperator) {
   }
   SSDataBlock* pBlock = NULL;
   while (pInfo->tableStartIndex < tableListSize) {
-    pBlock = getSortedTableMergeScanBlockData(pInfo->pSortHandle, pOperator->resultInfo.capacity, pOperator);
+    pBlock = getSortedTableMergeScanBlockData(pInfo->pSortHandle, pInfo->pResBlock, pOperator->resultInfo.capacity, pOperator);
     if (pBlock != NULL) {
       pBlock->info.groupId = pInfo->groupId;
       pOperator->resultInfo.totalRows += pBlock->info.rows;
@@ -2959,6 +2922,7 @@ SSDataBlock* doTableMergeScan(SOperatorInfo* pOperator) {
 void destroyTableMergeScanOperatorInfo(void* param, int32_t numOfOutput) {
   STableMergeScanInfo* pTableScanInfo = (STableMergeScanInfo*)param;
   cleanupQueryTableDataCond(&pTableScanInfo->cond);
+  taosArrayDestroy(pTableScanInfo->sortSourceParams);
 
   for (int32_t i = 0; i < taosArrayGetSize(pTableScanInfo->dataReaders); ++i) {
     STsdbReader* reader = taosArrayGetP(pTableScanInfo->dataReaders, i);
@@ -2974,7 +2938,9 @@ void destroyTableMergeScanOperatorInfo(void* param, int32_t numOfOutput) {
   pTableScanInfo->pSortInputBlock = blockDataDestroy(pTableScanInfo->pSortInputBlock);
 
   taosArrayDestroy(pTableScanInfo->pSortInfo);
+  cleanupExprSupp(&pTableScanInfo->pseudoSup);
 
+  taosMemoryFreeClear(pTableScanInfo->rowEntryInfoOffset);
   taosMemoryFreeClear(param);
 }
 
@@ -3031,8 +2997,9 @@ SOperatorInfo* createTableMergeScanOperatorInfo(STableScanPhysiNode* pTableScanN
   }
 
   if (pTableScanNode->scan.pScanPseudoCols != NULL) {
-    pInfo->pPseudoExpr = createExprInfo(pTableScanNode->scan.pScanPseudoCols, NULL, &pInfo->numOfPseudoExpr);
-    pInfo->pPseudoCtx = createSqlFunctionCtx(pInfo->pPseudoExpr, pInfo->numOfPseudoExpr, &pInfo->rowEntryInfoOffset);
+    SExprSupp* pSup = &pInfo->pseudoSup;
+    pSup->pExprInfo = createExprInfo(pTableScanNode->scan.pScanPseudoCols, NULL, &pSup->numOfExprs);
+    pSup->pCtx = createSqlFunctionCtx(pSup->pExprInfo, pSup->numOfExprs, &pSup->rowEntryInfoOffset);
   }
 
   pInfo->scanInfo = (SScanInfo){.numOfAsc = pTableScanNode->scanSeq[0], .numOfDesc = pTableScanNode->scanSeq[1]};
diff --git a/source/libs/executor/src/timewindowoperator.c b/source/libs/executor/src/timewindowoperator.c
index 0f1272c964..1e001a29a0 100644
--- a/source/libs/executor/src/timewindowoperator.c
+++ b/source/libs/executor/src/timewindowoperator.c
@@ -940,6 +940,7 @@ static void hashIntervalAgg(SOperatorInfo* pOperatorInfo, SResultRowInfo* pResul
 
     if (pInfo->execModel == OPTR_EXEC_MODEL_STREAM && pInfo->twAggSup.calTrigger == STREAM_TRIGGER_AT_ONCE) {
       saveResultRow(pResult, tableGroupId, pUpdated);
+      setResultBufPageDirty(pInfo->aggSup.pResultBuf, &pResultRowInfo->cur);
     }
   }
 
@@ -996,6 +997,7 @@ static void hashIntervalAgg(SOperatorInfo* pOperatorInfo, SResultRowInfo* pResul
 
     if (pInfo->execModel == OPTR_EXEC_MODEL_STREAM && pInfo->twAggSup.calTrigger == STREAM_TRIGGER_AT_ONCE) {
       saveResultRow(pResult, tableGroupId, pUpdated);
+      setResultBufPageDirty(pInfo->aggSup.pResultBuf, &pResultRowInfo->cur);
     }
 
     ekey = ascScan ? nextWin.ekey : nextWin.skey;
@@ -1092,7 +1094,6 @@ static int32_t doOpenIntervalAgg(SOperatorInfo* pOperator) {
     hashIntervalAgg(pOperator, &pInfo->binfo.resultRowInfo, pBlock, scanFlag, NULL);
   }
 
-  closeAllResultRows(&pInfo->binfo.resultRowInfo);
   initGroupedResultInfo(&pInfo->groupResInfo, pInfo->aggSup.pResultRowHashTable, pInfo->order);
   OPTR_SET_OPENED(pOperator);
 
@@ -1248,7 +1249,6 @@ static SSDataBlock* doStateWindowAgg(SOperatorInfo* pOperator) {
   pOperator->cost.openCost = (taosGetTimestampUs() - st) / 1000.0;
 
   pOperator->status = OP_RES_TO_RETURN;
-  closeAllResultRows(&pBInfo->resultRowInfo);
 
   initGroupedResultInfo(&pInfo->groupResInfo, pInfo->aggSup.pResultRowHashTable, TSDB_ORDER_ASC);
   blockDataEnsureCapacity(pBInfo->pRes, pOperator->resultInfo.capacity);
@@ -2043,7 +2043,6 @@ static SSDataBlock* doSessionWindowAgg(SOperatorInfo* pOperator) {
 
   // restore the value
   pOperator->status = OP_RES_TO_RETURN;
-  closeAllResultRows(&pBInfo->resultRowInfo);
 
   initGroupedResultInfo(&pInfo->groupResInfo, pInfo->aggSup.pResultRowHashTable, TSDB_ORDER_ASC);
   blockDataEnsureCapacity(pBInfo->pRes, pOperator->resultInfo.capacity);
@@ -2207,8 +2206,6 @@ static SSDataBlock* doTimeslice(SOperatorInfo* pOperator) {
   SSDataBlock*            pResBlock = pSliceInfo->pRes;
   SExprSupp*              pSup = &pOperator->exprSupp;
 
-  blockDataEnsureCapacity(pResBlock, pOperator->resultInfo.capacity);
-
   //  if (pOperator->status == OP_RES_TO_RETURN) {
   //    //    doBuildResultDatablock(&pRuntimeEnv->groupResInfo, pRuntimeEnv, pIntervalInfo->pRes);
   //    if (pResBlock->info.rows == 0 || !hasDataInGroupInfo(&pSliceInfo->groupResInfo)) {
@@ -2348,10 +2345,10 @@ SOperatorInfo* createTimeSliceOperatorInfo(SOperatorInfo* downstream, SPhysiNode
   initResultSizeInfo(&pOperator->resultInfo, 4096);
 
   pInfo->pFillColInfo = createFillColInfo(pExprInfo, numOfExprs, (SNodeListNode*)pInterpPhyNode->pFillValues);
-  pInfo->pRes = createResDataBlock(pPhyNode->pOutputDataBlockDesc);
-  pInfo->win = pInterpPhyNode->timeRange;
+  pInfo->pRes         = createResDataBlock(pPhyNode->pOutputDataBlockDesc);
+  pInfo->win          = pInterpPhyNode->timeRange;
   pInfo->interval.interval = pInterpPhyNode->interval;
-  pInfo->current = pInfo->win.skey;
+  pInfo->current      = pInfo->win.skey;
 
   pOperator->name = "TimeSliceOperator";
   pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_INTERP_FUNC;
@@ -2542,6 +2539,7 @@ static void rebuildIntervalWindow(SStreamFinalIntervalOperatorInfo* pInfo, SExpr
     }
     if (find && pUpdated) {
       saveResultRow(pCurResult, pWinRes->groupId, pUpdated);
+      setResultBufPageDirty(pInfo->aggSup.pResultBuf, &pInfo->binfo.resultRowInfo.cur);
     }
   }
 }
@@ -2662,6 +2660,7 @@ static void doHashInterval(SOperatorInfo* pOperatorInfo, SSDataBlock* pSDataBloc
     }
     if (pInfo->twAggSup.calTrigger == STREAM_TRIGGER_AT_ONCE && pUpdated) {
       saveResultRow(pResult, tableGroupId, pUpdated);
+      setResultBufPageDirty(pInfo->aggSup.pResultBuf, &pResultRowInfo->cur);
     }
     updateTimeWindowInfo(&pInfo->twAggSup.timeWindowData, &nextWin, true);
     doApplyFunctions(pTaskInfo, pSup->pCtx, &nextWin, &pInfo->twAggSup.timeWindowData, startPos, forwardRows, tsCols,
diff --git a/source/libs/nodes/src/nodesCloneFuncs.c b/source/libs/nodes/src/nodesCloneFuncs.c
index 121e697630..5279d015b4 100644
--- a/source/libs/nodes/src/nodesCloneFuncs.c
+++ b/source/libs/nodes/src/nodesCloneFuncs.c
@@ -375,6 +375,7 @@ static int32_t logicJoinCopy(const SJoinLogicNode* pSrc, SJoinLogicNode* pDst) {
   CLONE_NODE_FIELD(pMergeCondition);
   CLONE_NODE_FIELD(pOnConditions);
   COPY_SCALAR_FIELD(isSingleTableJoin);
+  COPY_SCALAR_FIELD(inputTsOrder);
   return TSDB_CODE_SUCCESS;
 }
 
@@ -440,6 +441,7 @@ static int32_t logicWindowCopy(const SWindowLogicNode* pSrc, SWindowLogicNode* p
   COPY_SCALAR_FIELD(watermark);
   COPY_SCALAR_FIELD(igExpired);
   COPY_SCALAR_FIELD(windowAlgo);
+  COPY_SCALAR_FIELD(inputTsOrder);
   return TSDB_CODE_SUCCESS;
 }
 
diff --git a/source/libs/nodes/src/nodesCodeFuncs.c b/source/libs/nodes/src/nodesCodeFuncs.c
index eec4780293..186a51f000 100644
--- a/source/libs/nodes/src/nodesCodeFuncs.c
+++ b/source/libs/nodes/src/nodesCodeFuncs.c
@@ -1717,7 +1717,7 @@ static const char* jkJoinPhysiPlanOnConditions = "OnConditions";
 static const char* jkJoinPhysiPlanTargets = "Targets";
 
 static int32_t physiJoinNodeToJson(const void* pObj, SJson* pJson) {
-  const SJoinPhysiNode* pNode = (const SJoinPhysiNode*)pObj;
+  const SSortMergeJoinPhysiNode* pNode = (const SSortMergeJoinPhysiNode*)pObj;
 
   int32_t code = physicPlanNodeToJson(pObj, pJson);
   if (TSDB_CODE_SUCCESS == code) {
@@ -1737,7 +1737,7 @@ static int32_t physiJoinNodeToJson(const void* pObj, SJson* pJson) {
 }
 
 static int32_t jsonToPhysiJoinNode(const SJson* pJson, void* pObj) {
-  SJoinPhysiNode* pNode = (SJoinPhysiNode*)pObj;
+  SSortMergeJoinPhysiNode* pNode = (SSortMergeJoinPhysiNode*)pObj;
 
   int32_t code = jsonToPhysicPlanNode(pJson, pObj);
   if (TSDB_CODE_SUCCESS == code) {
diff --git a/source/libs/nodes/src/nodesTraverseFuncs.c b/source/libs/nodes/src/nodesTraverseFuncs.c
index b12e3b14c7..77681af1bc 100644
--- a/source/libs/nodes/src/nodesTraverseFuncs.c
+++ b/source/libs/nodes/src/nodesTraverseFuncs.c
@@ -468,7 +468,7 @@ static EDealRes dispatchPhysiPlan(SNode* pNode, ETraversalOrder order, FNodeWalk
       break;
     }
     case QUERY_NODE_PHYSICAL_PLAN_MERGE_JOIN: {
-      SJoinPhysiNode* pJoin = (SJoinPhysiNode*)pNode;
+      SSortMergeJoinPhysiNode* pJoin = (SSortMergeJoinPhysiNode*)pNode;
       res = walkPhysiNode((SPhysiNode*)pNode, order, walker, pContext);
       if (DEAL_RES_ERROR != res && DEAL_RES_END != res) {
         res = walkPhysiPlan(pJoin->pMergeCondition, order, walker, pContext);
diff --git a/source/libs/nodes/src/nodesUtilFuncs.c b/source/libs/nodes/src/nodesUtilFuncs.c
index 23f0bb088d..3c6fbe409c 100644
--- a/source/libs/nodes/src/nodesUtilFuncs.c
+++ b/source/libs/nodes/src/nodesUtilFuncs.c
@@ -287,7 +287,7 @@ SNode* nodesMakeNode(ENodeType type) {
     case QUERY_NODE_PHYSICAL_PLAN_PROJECT:
       return makeNode(type, sizeof(SProjectPhysiNode));
     case QUERY_NODE_PHYSICAL_PLAN_MERGE_JOIN:
-      return makeNode(type, sizeof(SJoinPhysiNode));
+      return makeNode(type, sizeof(SSortMergeJoinPhysiNode));
     case QUERY_NODE_PHYSICAL_PLAN_HASH_AGG:
       return makeNode(type, sizeof(SAggPhysiNode));
     case QUERY_NODE_PHYSICAL_PLAN_EXCHANGE:
@@ -883,7 +883,7 @@ void nodesDestroyNode(SNode* pNode) {
       break;
     }
     case QUERY_NODE_PHYSICAL_PLAN_MERGE_JOIN: {
-      SJoinPhysiNode* pPhyNode = (SJoinPhysiNode*)pNode;
+      SSortMergeJoinPhysiNode* pPhyNode = (SSortMergeJoinPhysiNode*)pNode;
       destroyPhysiNode((SPhysiNode*)pPhyNode);
       nodesDestroyNode(pPhyNode->pMergeCondition);
       nodesDestroyNode(pPhyNode->pOnConditions);
diff --git a/source/libs/parser/inc/parAst.h b/source/libs/parser/inc/parAst.h
index f3ea332fe2..12c18733b1 100644
--- a/source/libs/parser/inc/parAst.h
+++ b/source/libs/parser/inc/parAst.h
@@ -90,7 +90,7 @@ SNode* createValueNode(SAstCreateContext* pCxt, int32_t dataType, const SToken*
 SNode* createDurationValueNode(SAstCreateContext* pCxt, const SToken* pLiteral);
 SNode* createDefaultDatabaseCondValue(SAstCreateContext* pCxt);
 SNode* createPlaceholderValueNode(SAstCreateContext* pCxt, const SToken* pLiteral);
-SNode* setProjectionAlias(SAstCreateContext* pCxt, SNode* pNode, const SToken* pAlias);
+SNode* setProjectionAlias(SAstCreateContext* pCxt, SNode* pNode, SToken* pAlias);
 SNode* createLogicConditionNode(SAstCreateContext* pCxt, ELogicConditionType type, SNode* pParam1, SNode* pParam2);
 SNode* createOperatorNode(SAstCreateContext* pCxt, EOperatorType type, SNode* pLeft, SNode* pRight);
 SNode* createBetweenAnd(SAstCreateContext* pCxt, SNode* pExpr, SNode* pLeft, SNode* pRight);
diff --git a/source/libs/parser/src/parAstCreater.c b/source/libs/parser/src/parAstCreater.c
index b237fd2c6e..a54dae1ee9 100644
--- a/source/libs/parser/src/parAstCreater.c
+++ b/source/libs/parser/src/parAstCreater.c
@@ -527,6 +527,7 @@ SNode* createTempTableNode(SAstCreateContext* pCxt, SNode* pSubquery, const STok
   }
   if (QUERY_NODE_SELECT_STMT == nodeType(pSubquery)) {
     strcpy(((SSelectStmt*)pSubquery)->stmtName, tempTable->table.tableAlias);
+    ((SSelectStmt*)pSubquery)->isSubquery = true;
   } else if (QUERY_NODE_SET_OPERATOR == nodeType(pSubquery)) {
     strcpy(((SSetOperator*)pSubquery)->stmtName, tempTable->table.tableAlias);
   }
@@ -637,8 +638,9 @@ SNode* createInterpTimeRange(SAstCreateContext* pCxt, SNode* pStart, SNode* pEnd
   return createBetweenAnd(pCxt, createPrimaryKeyCol(pCxt), pStart, pEnd);
 }
 
-SNode* setProjectionAlias(SAstCreateContext* pCxt, SNode* pNode, const SToken* pAlias) {
+SNode* setProjectionAlias(SAstCreateContext* pCxt, SNode* pNode, SToken* pAlias) {
   CHECK_PARSER_STATUS(pCxt);
+  trimEscape(pAlias);
   int32_t len = TMIN(sizeof(((SExprNode*)pNode)->aliasName) - 1, pAlias->n);
   strncpy(((SExprNode*)pNode)->aliasName, pAlias->z, len);
   ((SExprNode*)pNode)->aliasName[len] = '\0';
diff --git a/source/libs/parser/src/parInsert.c b/source/libs/parser/src/parInsert.c
index 702422e022..d564d53633 100644
--- a/source/libs/parser/src/parInsert.c
+++ b/source/libs/parser/src/parInsert.c
@@ -739,12 +739,13 @@ static int32_t parseBoundColumns(SInsertParseContext* pCxt, SParsedDataColInfo*
   return TSDB_CODE_SUCCESS;
 }
 
-static void buildCreateTbReq(SVCreateTbReq* pTbReq, const char* tname, STag* pTag, int64_t suid, const char* sname, SArray* tagName, uint8_t tagNum) {
+static void buildCreateTbReq(SVCreateTbReq* pTbReq, const char* tname, STag* pTag, int64_t suid, const char* sname,
+                             SArray* tagName, uint8_t tagNum) {
   pTbReq->type = TD_CHILD_TABLE;
   pTbReq->name = strdup(tname);
   pTbReq->ctb.suid = suid;
   pTbReq->ctb.tagNum = tagNum;
-  if(sname) pTbReq->ctb.name = strdup(sname);
+  if (sname) pTbReq->ctb.name = strdup(sname);
   pTbReq->ctb.pTag = (uint8_t*)pTag;
   pTbReq->ctb.tagName = taosArrayDup(tagName);
   pTbReq->commentLen = -1;
@@ -969,7 +970,7 @@ static int32_t parseTagsClause(SInsertParseContext* pCxt, SSchema* pSchema, uint
     }
 
     SSchema* pTagSchema = &pSchema[pCxt->tags.boundColumns[i]];
-    char     tmpTokenBuf[TSDB_MAX_BYTES_PER_ROW] = {0}; // todo this can be optimize with parse column
+    char     tmpTokenBuf[TSDB_MAX_BYTES_PER_ROW] = {0};  // todo this can be optimize with parse column
     code = checkAndTrimValue(&sToken, tmpTokenBuf, &pCxt->msg);
     if (code != TSDB_CODE_SUCCESS) {
       goto end;
@@ -1012,7 +1013,8 @@ static int32_t parseTagsClause(SInsertParseContext* pCxt, SSchema* pSchema, uint
     goto end;
   }
 
-  buildCreateTbReq(&pCxt->createTblReq, tName, pTag, pCxt->pTableMeta->suid, pCxt->sTableName, tagName, pCxt->pTableMeta->tableInfo.numOfTags);
+  buildCreateTbReq(&pCxt->createTblReq, tName, pTag, pCxt->pTableMeta->suid, pCxt->sTableName, tagName,
+                   pCxt->pTableMeta->tableInfo.numOfTags);
 
 end:
   for (int i = 0; i < taosArrayGetSize(pTagVals); ++i) {
@@ -1650,7 +1652,6 @@ static int32_t skipUsingClause(SInsertParseSyntaxCxt* pCxt) {
 static int32_t collectTableMetaKey(SInsertParseSyntaxCxt* pCxt, SToken* pTbToken) {
   SName name;
   CHECK_CODE(createSName(&name, pTbToken, pCxt->pComCxt->acctId, pCxt->pComCxt->db, &pCxt->msg));
-  CHECK_CODE(reserveDbCfgInCache(pCxt->pComCxt->acctId, name.dbname, pCxt->pMetaCache));
   CHECK_CODE(reserveUserAuthInCacheExt(pCxt->pComCxt->pUser, &name, AUTH_TYPE_WRITE, pCxt->pMetaCache));
   CHECK_CODE(reserveTableMetaInCacheExt(&name, pCxt->pMetaCache));
   CHECK_CODE(reserveTableVgroupInCacheExt(&name, pCxt->pMetaCache));
@@ -2332,7 +2333,8 @@ int32_t smlBindData(void* handle, SArray* tags, SArray* colsSchema, SArray* cols
     return ret;
   }
 
-  buildCreateTbReq(&smlHandle->tableExecHandle.createTblReq, tableName, pTag, pTableMeta->suid, NULL, tagName, pTableMeta->tableInfo.numOfTags);
+  buildCreateTbReq(&smlHandle->tableExecHandle.createTblReq, tableName, pTag, pTableMeta->suid, NULL, tagName,
+                   pTableMeta->tableInfo.numOfTags);
   taosArrayDestroy(tagName);
 
   smlHandle->tableExecHandle.createTblReq.ctb.name = taosMemoryMalloc(sTableNameLen + 1);
diff --git a/source/libs/parser/src/parUtil.c b/source/libs/parser/src/parUtil.c
index 74d5f03dc1..7c9a8b10dd 100644
--- a/source/libs/parser/src/parUtil.c
+++ b/source/libs/parser/src/parUtil.c
@@ -92,7 +92,7 @@ static char* getSyntaxErrFormat(int32_t errCode) {
     case TSDB_CODE_PAR_INTER_SLIDING_TOO_BIG:
       return "sliding value no larger than the interval value";
     case TSDB_CODE_PAR_INTER_SLIDING_TOO_SMALL:
-      return "sliding value can not less than 1% of interval value";
+      return "sliding value can not less than 1%% of interval value";
     case TSDB_CODE_PAR_ONLY_ONE_JSON_TAG:
       return "Only one tag if there is a json tag";
     case TSDB_CODE_PAR_INCORRECT_NUM_OF_COL:
diff --git a/source/libs/planner/src/planLogicCreater.c b/source/libs/planner/src/planLogicCreater.c
index 84e712b466..30e3b676df 100644
--- a/source/libs/planner/src/planLogicCreater.c
+++ b/source/libs/planner/src/planLogicCreater.c
@@ -339,6 +339,7 @@ static int32_t createJoinLogicNode(SLogicPlanContext* pCxt, SSelectStmt* pSelect
 
   pJoin->joinType = pJoinTable->joinType;
   pJoin->isSingleTableJoin = pJoinTable->table.singleTable;
+  pJoin->inputTsOrder = ORDER_ASC;
   pJoin->node.groupAction = GROUP_ACTION_CLEAR;
   pJoin->node.requireDataOrder = DATA_ORDER_LEVEL_GLOBAL;
   pJoin->node.requireDataOrder = DATA_ORDER_LEVEL_GLOBAL;
@@ -625,14 +626,14 @@ static int32_t createInterpFuncLogicNode(SLogicPlanContext* pCxt, SSelectStmt* p
 
 static int32_t createWindowLogicNodeFinalize(SLogicPlanContext* pCxt, SSelectStmt* pSelect, SWindowLogicNode* pWindow,
                                              SLogicNode** pLogicNode) {
-  int32_t code = nodesCollectFuncs(pSelect, SQL_CLAUSE_WINDOW, fmIsWindowClauseFunc, &pWindow->pFuncs);
-
   if (pCxt->pPlanCxt->streamQuery) {
     pWindow->triggerType = pCxt->pPlanCxt->triggerType;
     pWindow->watermark = pCxt->pPlanCxt->watermark;
     pWindow->igExpired = pCxt->pPlanCxt->igExpired;
   }
+  pWindow->inputTsOrder = ORDER_ASC;
 
+  int32_t code = nodesCollectFuncs(pSelect, SQL_CLAUSE_WINDOW, fmIsWindowClauseFunc, &pWindow->pFuncs);
   if (TSDB_CODE_SUCCESS == code) {
     code = rewriteExprsForSelect(pWindow->pFuncs, pSelect, SQL_CLAUSE_WINDOW);
   }
@@ -861,7 +862,8 @@ static int32_t createProjectLogicNode(SLogicPlanContext* pCxt, SSelectStmt* pSel
 
   TSWAP(pProject->node.pLimit, pSelect->pLimit);
   TSWAP(pProject->node.pSlimit, pSelect->pSlimit);
-  pProject->node.groupAction = GROUP_ACTION_CLEAR;
+  pProject->node.groupAction =
+      (!pSelect->isSubquery && pCxt->pPlanCxt->streamQuery) ? GROUP_ACTION_KEEP : GROUP_ACTION_CLEAR;
   pProject->node.requireDataOrder = DATA_ORDER_LEVEL_NONE;
   pProject->node.resultDataOrder = DATA_ORDER_LEVEL_NONE;
 
diff --git a/source/libs/planner/src/planOptimizer.c b/source/libs/planner/src/planOptimizer.c
index 7b60710c7d..fcc395af62 100644
--- a/source/libs/planner/src/planOptimizer.c
+++ b/source/libs/planner/src/planOptimizer.c
@@ -993,25 +993,28 @@ static bool sortPriKeyOptMayBeOptimized(SLogicNode* pNode) {
 }
 
 static int32_t sortPriKeyOptGetScanNodesImpl(SLogicNode* pNode, bool* pNotOptimize, SNodeList** pScanNodes) {
-  int32_t code = TSDB_CODE_SUCCESS;
-
   switch (nodeType(pNode)) {
-    case QUERY_NODE_LOGIC_PLAN_SCAN:
-      if (TSDB_SUPER_TABLE != ((SScanLogicNode*)pNode)->tableType) {
-        return nodesListMakeAppend(pScanNodes, (SNode*)pNode);
+    case QUERY_NODE_LOGIC_PLAN_SCAN: {
+      SScanLogicNode* pScan = (SScanLogicNode*)pNode;
+      if (NULL != pScan->pGroupTags) {
+        *pNotOptimize = true;
+        return TSDB_CODE_SUCCESS;
       }
-      break;
-    case QUERY_NODE_LOGIC_PLAN_JOIN:
-      code =
+      return nodesListMakeAppend(pScanNodes, (SNode*)pNode);
+    }
+    case QUERY_NODE_LOGIC_PLAN_JOIN: {
+      int32_t code =
           sortPriKeyOptGetScanNodesImpl((SLogicNode*)nodesListGetNode(pNode->pChildren, 0), pNotOptimize, pScanNodes);
       if (TSDB_CODE_SUCCESS == code) {
         code =
             sortPriKeyOptGetScanNodesImpl((SLogicNode*)nodesListGetNode(pNode->pChildren, 1), pNotOptimize, pScanNodes);
       }
       return code;
+    }
     case QUERY_NODE_LOGIC_PLAN_AGG:
+    case QUERY_NODE_LOGIC_PLAN_PARTITION:
       *pNotOptimize = true;
-      return code;
+      return TSDB_CODE_SUCCESS;
     default:
       break;
   }
@@ -1037,17 +1040,33 @@ static EOrder sortPriKeyOptGetPriKeyOrder(SSortLogicNode* pSort) {
   return ((SOrderByExprNode*)nodesListGetNode(pSort->pSortKeys, 0))->order;
 }
 
+static void sortPriKeyOptSetParentOrder(SLogicNode* pNode, EOrder order) {
+  if (NULL == pNode) {
+    return;
+  }
+  if (QUERY_NODE_LOGIC_PLAN_WINDOW == nodeType(pNode)) {
+    ((SWindowLogicNode*)pNode)->inputTsOrder = order;
+  } else if (QUERY_NODE_LOGIC_PLAN_JOIN == nodeType(pNode)) {
+    ((SJoinLogicNode*)pNode)->inputTsOrder = order;
+  }
+  sortPriKeyOptSetParentOrder(pNode->pParent, order);
+}
+
 static int32_t sortPriKeyOptApply(SOptimizeContext* pCxt, SLogicSubplan* pLogicSubplan, SSortLogicNode* pSort,
                                   SNodeList* pScanNodes) {
   EOrder order = sortPriKeyOptGetPriKeyOrder(pSort);
-  if (ORDER_DESC == order) {
-    SNode* pScanNode = NULL;
-    FOREACH(pScanNode, pScanNodes) {
-      SScanLogicNode* pScan = (SScanLogicNode*)pScanNode;
-      if (pScan->scanSeq[0] > 0) {
-        TSWAP(pScan->scanSeq[0], pScan->scanSeq[1]);
-      }
+  SNode* pScanNode = NULL;
+  FOREACH(pScanNode, pScanNodes) {
+    SScanLogicNode* pScan = (SScanLogicNode*)pScanNode;
+    if (ORDER_DESC == order && pScan->scanSeq[0] > 0) {
+      TSWAP(pScan->scanSeq[0], pScan->scanSeq[1]);
     }
+    if (TSDB_SUPER_TABLE == pScan->tableType) {
+      pScan->scanType = SCAN_TYPE_TABLE_MERGE;
+      pScan->node.resultDataOrder = DATA_ORDER_LEVEL_GLOBAL;
+      pScan->node.requireDataOrder = DATA_ORDER_LEVEL_GLOBAL;
+    }
+    sortPriKeyOptSetParentOrder(pScan->node.pParent, order);
   }
 
   SLogicNode* pChild = (SLogicNode*)nodesListGetNode(pSort->node.pChildren, 0);
@@ -1613,10 +1632,10 @@ static void alignProjectionWithTarget(SLogicNode* pNode) {
   }
 
   SProjectLogicNode* pProjectNode = (SProjectLogicNode*)pNode;
-  SNode* pProjection = NULL;
+  SNode*             pProjection = NULL;
   FOREACH(pProjection, pProjectNode->pProjections) {
     SNode* pTarget = NULL;
-    bool keep = false;
+    bool   keep = false;
     FOREACH(pTarget, pNode->pTargets) {
       if (0 == strcmp(((SColumnNode*)pProjection)->node.aliasName, ((SColumnNode*)pTarget)->colName)) {
         keep = true;
@@ -2214,7 +2233,7 @@ static bool tagScanMayBeOptimized(SLogicNode* pNode) {
       !planOptNodeListHasTbname(pAgg->pGroupKeys)) {
     return false;
   }
-  
+
   SNode* pGroupKey = NULL;
   FOREACH(pGroupKey, pAgg->pGroupKeys) {
     SNode* pGroup = NULL;
diff --git a/source/libs/planner/src/planPhysiCreater.c b/source/libs/planner/src/planPhysiCreater.c
index 0a1f8bbd0b..587e566939 100644
--- a/source/libs/planner/src/planPhysiCreater.c
+++ b/source/libs/planner/src/planPhysiCreater.c
@@ -415,7 +415,6 @@ static int32_t createScanPhysiNodeFinalize(SPhysiPlanContext* pCxt, SSubplan* pS
                                            SScanPhysiNode* pScanPhysiNode, SPhysiNode** pPhyNode) {
   int32_t code = createScanCols(pCxt, pScanPhysiNode, pScanLogicNode->pScanCols);
   if (TSDB_CODE_SUCCESS == code) {
-    // Data block describe also needs to be set without scanning column, such as SELECT COUNT(*) FROM t
     code = addDataBlockSlots(pCxt, pScanPhysiNode->pScanCols, pScanPhysiNode->node.pOutputDataBlockDesc);
   }
 
@@ -622,8 +621,8 @@ static int32_t createScanPhysiNode(SPhysiPlanContext* pCxt, SSubplan* pSubplan,
 
 static int32_t createJoinPhysiNode(SPhysiPlanContext* pCxt, SNodeList* pChildren, SJoinLogicNode* pJoinLogicNode,
                                    SPhysiNode** pPhyNode) {
-  SJoinPhysiNode* pJoin =
-      (SJoinPhysiNode*)makePhysiNode(pCxt, (SLogicNode*)pJoinLogicNode, QUERY_NODE_PHYSICAL_PLAN_MERGE_JOIN);
+  SSortMergeJoinPhysiNode* pJoin =
+      (SSortMergeJoinPhysiNode*)makePhysiNode(pCxt, (SLogicNode*)pJoinLogicNode, QUERY_NODE_PHYSICAL_PLAN_MERGE_JOIN);
   if (NULL == pJoin) {
     return TSDB_CODE_OUT_OF_MEMORY;
   }
@@ -975,6 +974,9 @@ static int32_t createInterpFuncPhysiNode(SPhysiPlanContext* pCxt, SNodeList* pCh
 }
 
 static bool projectCanMergeDataBlock(SProjectLogicNode* pProject) {
+  if (GROUP_ACTION_KEEP == pProject->node.groupAction) {
+    return false;
+  }
   if (DATA_ORDER_LEVEL_NONE == pProject->node.resultDataOrder) {
     return true;
   }
diff --git a/source/libs/planner/src/planSpliter.c b/source/libs/planner/src/planSpliter.c
index 8586234b7e..81e2bff179 100644
--- a/source/libs/planner/src/planSpliter.c
+++ b/source/libs/planner/src/planSpliter.c
@@ -469,7 +469,7 @@ static int32_t stbSplCreateExchangeNode(SSplitContext* pCxt, SLogicNode* pParent
   return code;
 }
 
-static int32_t stbSplCreateMergeKeysByPrimaryKey(SNode* pPrimaryKey, SNodeList** pMergeKeys) {
+static int32_t stbSplCreateMergeKeysByPrimaryKey(SNode* pPrimaryKey, EOrder order, SNodeList** pMergeKeys) {
   SOrderByExprNode* pMergeKey = (SOrderByExprNode*)nodesMakeNode(QUERY_NODE_ORDER_BY_EXPR);
   if (NULL == pMergeKey) {
     return TSDB_CODE_OUT_OF_MEMORY;
@@ -479,7 +479,7 @@ static int32_t stbSplCreateMergeKeysByPrimaryKey(SNode* pPrimaryKey, SNodeList**
     nodesDestroyNode((SNode*)pMergeKey);
     return TSDB_CODE_OUT_OF_MEMORY;
   }
-  pMergeKey->order = ORDER_ASC;
+  pMergeKey->order = order;
   pMergeKey->nullOrder = NULL_ORDER_FIRST;
   return nodesListMakeStrictAppend(pMergeKeys, (SNode*)pMergeKey);
 }
@@ -491,7 +491,8 @@ static int32_t stbSplSplitIntervalForBatch(SSplitContext* pCxt, SStableSplitInfo
     ((SWindowLogicNode*)pPartWindow)->windowAlgo = INTERVAL_ALGO_HASH;
     ((SWindowLogicNode*)pInfo->pSplitNode)->windowAlgo = INTERVAL_ALGO_MERGE;
     SNodeList* pMergeKeys = NULL;
-    code = stbSplCreateMergeKeysByPrimaryKey(((SWindowLogicNode*)pInfo->pSplitNode)->pTspk, &pMergeKeys);
+    code = stbSplCreateMergeKeysByPrimaryKey(((SWindowLogicNode*)pInfo->pSplitNode)->pTspk,
+                                             ((SWindowLogicNode*)pInfo->pSplitNode)->inputTsOrder, &pMergeKeys);
     if (TSDB_CODE_SUCCESS == code) {
       code = stbSplCreateMergeNode(pCxt, NULL, pInfo->pSplitNode, pMergeKeys, pPartWindow, true);
     }
@@ -579,7 +580,8 @@ static int32_t stbSplSplitSessionOrStateForBatch(SSplitContext* pCxt, SStableSpl
   SLogicNode* pChild = (SLogicNode*)nodesListGetNode(pWindow->pChildren, 0);
 
   SNodeList* pMergeKeys = NULL;
-  int32_t    code = stbSplCreateMergeKeysByPrimaryKey(((SWindowLogicNode*)pWindow)->pTspk, &pMergeKeys);
+  int32_t    code = stbSplCreateMergeKeysByPrimaryKey(((SWindowLogicNode*)pWindow)->pTspk,
+                                                      ((SWindowLogicNode*)pWindow)->inputTsOrder, &pMergeKeys);
 
   if (TSDB_CODE_SUCCESS == code) {
     code = stbSplCreateMergeNode(pCxt, pInfo->pSubplan, pChild, pMergeKeys, (SLogicNode*)pChild, true);
@@ -913,27 +915,70 @@ static int32_t stbSplSplitScanNodeWithPartTags(SSplitContext* pCxt, SStableSplit
 }
 
 static SNode* stbSplFindPrimaryKeyFromScan(SScanLogicNode* pScan) {
+  bool   find = false;
   SNode* pCol = NULL;
   FOREACH(pCol, pScan->pScanCols) {
     if (PRIMARYKEY_TIMESTAMP_COL_ID == ((SColumnNode*)pCol)->colId) {
+      find = true;
+      break;
+    }
+  }
+  if (!find) {
+    return NULL;
+  }
+  SNode* pTarget = NULL;
+  FOREACH(pTarget, pScan->node.pTargets) {
+    if (nodesEqualNode(pTarget, pCol)) {
       return pCol;
     }
   }
-  return NULL;
+  nodesListStrictAppend(pScan->node.pTargets, nodesCloneNode(pCol));
+  return pCol;
+}
+
+static int32_t stbSplCreateMergeScanNode(SScanLogicNode* pScan, SLogicNode** pOutputMergeScan,
+                                         SNodeList** pOutputMergeKeys) {
+  SNodeList* pChildren = pScan->node.pChildren;
+  pScan->node.pChildren = NULL;
+
+  int32_t         code = TSDB_CODE_SUCCESS;
+  SScanLogicNode* pMergeScan = (SScanLogicNode*)nodesCloneNode((SNode*)pScan);
+  if (NULL == pMergeScan) {
+    code = TSDB_CODE_OUT_OF_MEMORY;
+  }
+
+  SNodeList* pMergeKeys = NULL;
+  if (TSDB_CODE_SUCCESS == code) {
+    pMergeScan->scanType = SCAN_TYPE_TABLE_MERGE;
+    pMergeScan->node.pChildren = pChildren;
+    splSetParent((SLogicNode*)pMergeScan);
+    code = stbSplCreateMergeKeysByPrimaryKey(stbSplFindPrimaryKeyFromScan(pMergeScan),
+                                             pMergeScan->scanSeq[0] > 0 ? ORDER_ASC : ORDER_DESC, &pMergeKeys);
+  }
+
+  if (TSDB_CODE_SUCCESS == code) {
+    *pOutputMergeScan = (SLogicNode*)pMergeScan;
+    *pOutputMergeKeys = pMergeKeys;
+  } else {
+    nodesDestroyNode((SNode*)pMergeScan);
+    nodesDestroyList(pMergeKeys);
+  }
+
+  return code;
 }
 
 static int32_t stbSplSplitMergeScanNode(SSplitContext* pCxt, SLogicSubplan* pSubplan, SScanLogicNode* pScan,
                                         bool groupSort) {
-  SNodeList* pMergeKeys = NULL;
-  int32_t    code = stbSplCreateMergeKeysByPrimaryKey(stbSplFindPrimaryKeyFromScan(pScan), &pMergeKeys);
+  SLogicNode* pMergeScan = NULL;
+  SNodeList*  pMergeKeys = NULL;
+  int32_t     code = stbSplCreateMergeScanNode(pScan, &pMergeScan, &pMergeKeys);
   if (TSDB_CODE_SUCCESS == code) {
-    code = stbSplCreateMergeNode(pCxt, pSubplan, (SLogicNode*)pScan, pMergeKeys, (SLogicNode*)pScan, groupSort);
+    code = stbSplCreateMergeNode(pCxt, pSubplan, (SLogicNode*)pScan, pMergeKeys, pMergeScan, groupSort);
   }
   if (TSDB_CODE_SUCCESS == code) {
     code = nodesListMakeStrictAppend(&pSubplan->pChildren,
-                                     (SNode*)splCreateScanSubplan(pCxt, (SLogicNode*)pScan, SPLIT_FLAG_STABLE_SPLIT));
+                                     (SNode*)splCreateScanSubplan(pCxt, pMergeScan, SPLIT_FLAG_STABLE_SPLIT));
   }
-  pScan->scanType = SCAN_TYPE_TABLE_MERGE;
   ++(pCxt->groupId);
   return code;
 }
@@ -978,14 +1023,14 @@ static int32_t stbSplSplitJoinNode(SSplitContext* pCxt, SStableSplitInfo* pInfo)
 }
 
 static int32_t stbSplCreateMergeKeysForPartitionNode(SLogicNode* pPart, SNodeList** pMergeKeys) {
-  SNode* pPrimaryKey =
-      nodesCloneNode(stbSplFindPrimaryKeyFromScan((SScanLogicNode*)nodesListGetNode(pPart->pChildren, 0)));
+  SScanLogicNode* pScan = (SScanLogicNode*)nodesListGetNode(pPart->pChildren, 0);
+  SNode*          pPrimaryKey = nodesCloneNode(stbSplFindPrimaryKeyFromScan(pScan));
   if (NULL == pPrimaryKey) {
     return TSDB_CODE_OUT_OF_MEMORY;
   }
   int32_t code = nodesListAppend(pPart->pTargets, pPrimaryKey);
   if (TSDB_CODE_SUCCESS == code) {
-    code = stbSplCreateMergeKeysByPrimaryKey(pPrimaryKey, pMergeKeys);
+    code = stbSplCreateMergeKeysByPrimaryKey(pPrimaryKey, pScan->scanSeq[0] > 0 ? ORDER_ASC : ORDER_DESC, pMergeKeys);
   }
   return code;
 }
diff --git a/source/libs/planner/src/planUtil.c b/source/libs/planner/src/planUtil.c
index bfa6079cb1..7aab8a7ca3 100644
--- a/source/libs/planner/src/planUtil.c
+++ b/source/libs/planner/src/planUtil.c
@@ -124,7 +124,8 @@ int32_t replaceLogicNode(SLogicSubplan* pSubplan, SLogicNode* pOld, SLogicNode*
 }
 
 static int32_t adjustScanDataRequirement(SScanLogicNode* pScan, EDataOrderLevel requirement) {
-  if (SCAN_TYPE_TABLE != pScan->scanType && SCAN_TYPE_TABLE_MERGE != pScan->scanType) {
+  if ((SCAN_TYPE_TABLE != pScan->scanType && SCAN_TYPE_TABLE_MERGE != pScan->scanType) ||
+      DATA_ORDER_LEVEL_GLOBAL == pScan->node.requireDataOrder) {
     return TSDB_CODE_SUCCESS;
   }
   // The lowest sort level of scan output data is DATA_ORDER_LEVEL_IN_BLOCK
diff --git a/source/libs/planner/test/planBasicTest.cpp b/source/libs/planner/test/planBasicTest.cpp
index 8f9cd94c19..9cfae68d34 100644
--- a/source/libs/planner/test/planBasicTest.cpp
+++ b/source/libs/planner/test/planBasicTest.cpp
@@ -24,9 +24,10 @@ TEST_F(PlanBasicTest, selectClause) {
   useDb("root", "test");
 
   run("SELECT * FROM t1");
-  run("SELECT 1 FROM t1");
-  run("SELECT * FROM st1");
-  run("SELECT 1 FROM st1");
+
+  run("SELECT MAX(c1) c2, c2 FROM t1");
+
+  run("SELECT MAX(c1) c2, c2 FROM st1");
 }
 
 TEST_F(PlanBasicTest, whereClause) {
diff --git a/source/libs/planner/test/planOptimizeTest.cpp b/source/libs/planner/test/planOptimizeTest.cpp
index 770ac94e5b..058705403b 100644
--- a/source/libs/planner/test/planOptimizeTest.cpp
+++ b/source/libs/planner/test/planOptimizeTest.cpp
@@ -53,6 +53,8 @@ TEST_F(PlanOptimizeTest, sortPrimaryKey) {
 
   run("SELECT c1 FROM t1 ORDER BY ts");
 
+  run("SELECT c1 FROM st1 ORDER BY ts");
+
   run("SELECT c1 FROM t1 ORDER BY ts DESC");
 
   run("SELECT COUNT(*) FROM t1 INTERVAL(10S) ORDER BY _WSTART DESC");
diff --git a/source/libs/qworker/src/qworker.c b/source/libs/qworker/src/qworker.c
index ebccb7950c..d77e42388b 100644
--- a/source/libs/qworker/src/qworker.c
+++ b/source/libs/qworker/src/qworker.c
@@ -66,7 +66,7 @@ int32_t qwHandleTaskComplete(QW_FPARAMS_DEF, SQWTaskCtx *ctx) {
   return TSDB_CODE_SUCCESS;
 }
 
-int32_t qwExecTask(QW_FPARAMS_DEF, SQWTaskCtx *ctx, bool *queryEnd) {
+int32_t qwExecTask(QW_FPARAMS_DEF, SQWTaskCtx *ctx, bool *queryStop) {
   int32_t        code = 0;
   bool           qcontinue = true;
   SSDataBlock   *pRes = NULL;
@@ -104,8 +104,8 @@ int32_t qwExecTask(QW_FPARAMS_DEF, SQWTaskCtx *ctx, bool *queryEnd) {
 
       QW_ERR_RET(qwHandleTaskComplete(QW_FPARAMS(), ctx));
 
-      if (queryEnd) {
-        *queryEnd = true;
+      if (queryStop) {
+        *queryStop = true;
       }
 
       break;
@@ -125,6 +125,10 @@ int32_t qwExecTask(QW_FPARAMS_DEF, SQWTaskCtx *ctx, bool *queryEnd) {
     QW_TASK_DLOG("data put into sink, rows:%d, continueExecTask:%d", rows, qcontinue);
 
     if (!qcontinue) {
+      if (queryStop) {
+        *queryStop = true;
+      }
+      
       break;
     }
 
@@ -566,7 +570,7 @@ int32_t qwProcessCQuery(QW_FPARAMS_DEF, SQWMsg *qwMsg) {
   SQWPhaseInput input = {0};
   void         *rsp = NULL;
   int32_t       dataLen = 0;
-  bool          queryEnd = false;
+  bool          queryStop = false;
 
   do {
     QW_ERR_JRET(qwHandlePrePhaseEvents(QW_FPARAMS(), QW_PHASE_PRE_CQUERY, &input, NULL));
@@ -576,7 +580,7 @@ int32_t qwProcessCQuery(QW_FPARAMS_DEF, SQWMsg *qwMsg) {
     atomic_store_8((int8_t *)&ctx->queryInQueue, 0);
     atomic_store_8((int8_t *)&ctx->queryContinue, 0);
 
-    QW_ERR_JRET(qwExecTask(QW_FPARAMS(), ctx, &queryEnd));
+    QW_ERR_JRET(qwExecTask(QW_FPARAMS(), ctx, &queryStop));
 
     if (QW_EVENT_RECEIVED(ctx, QW_EVENT_FETCH)) {
       SOutputData sOutput = {0};
@@ -627,7 +631,7 @@ int32_t qwProcessCQuery(QW_FPARAMS_DEF, SQWMsg *qwMsg) {
     }
 
     QW_LOCK(QW_WRITE, &ctx->lock);
-    if (queryEnd || code || 0 == atomic_load_8((int8_t *)&ctx->queryContinue)) {
+    if (queryStop || code || 0 == atomic_load_8((int8_t *)&ctx->queryContinue)) {
       // Note: query is not running anymore
       QW_SET_PHASE(ctx, 0);
       QW_UNLOCK(QW_WRITE, &ctx->lock);
diff --git a/source/libs/wal/src/walMgmt.c b/source/libs/wal/src/walMgmt.c
index 491b982968..85238e87b9 100644
--- a/source/libs/wal/src/walMgmt.c
+++ b/source/libs/wal/src/walMgmt.c
@@ -93,7 +93,7 @@ SWal *walOpen(const char *path, SWalCfg *pCfg) {
   }
 
   // init ref
-  pWal->pRefHash = taosHashInit(64, taosGetDefaultHashFunction(TSDB_DATA_TYPE_UBIGINT), true, HASH_ENTRY_LOCK);
+  pWal->pRefHash = taosHashInit(64, taosGetDefaultHashFunction(TSDB_DATA_TYPE_BIGINT), true, HASH_ENTRY_LOCK);
   if (pWal->pRefHash == NULL) {
     taosMemoryFree(pWal);
     return NULL;
diff --git a/source/libs/wal/src/walRead.c b/source/libs/wal/src/walRead.c
index 6d0e844e8e..ac62b7d98d 100644
--- a/source/libs/wal/src/walRead.c
+++ b/source/libs/wal/src/walRead.c
@@ -21,107 +21,112 @@ static int32_t walFetchBodyNew(SWalReader *pRead);
 static int32_t walSkipFetchBodyNew(SWalReader *pRead);
 
 SWalReader *walOpenReader(SWal *pWal, SWalFilterCond *cond) {
-  SWalReader *pRead = taosMemoryCalloc(1, sizeof(SWalReader));
-  if (pRead == NULL) {
+  SWalReader *pReader = taosMemoryCalloc(1, sizeof(SWalReader));
+  if (pReader == NULL) {
     terrno = TSDB_CODE_OUT_OF_MEMORY;
     return NULL;
   }
 
-  pRead->pWal = pWal;
-  pRead->pIdxFile = NULL;
-  pRead->pLogFile = NULL;
-  pRead->curVersion = -1;
-  pRead->curFileFirstVer = -1;
-  pRead->curInvalid = 1;
-  pRead->capacity = 0;
+  pReader->pWal = pWal;
+  pReader->readerId = tGenIdPI64();
+  pReader->pIdxFile = NULL;
+  pReader->pLogFile = NULL;
+  pReader->curVersion = -1;
+  pReader->curFileFirstVer = -1;
+  pReader->curInvalid = 1;
+  pReader->capacity = 0;
   if (cond) {
-    pRead->cond = *cond;
+    pReader->cond = *cond;
   } else {
-    pRead->cond.scanMeta = 0;
-    pRead->cond.scanUncommited = 0;
-    pRead->cond.enableRef = 0;
+    pReader->cond.scanUncommited = 0;
+    pReader->cond.scanNotApplied = 0;
+    pReader->cond.scanMeta = 0;
+    pReader->cond.enableRef = 0;
   }
 
-  taosThreadMutexInit(&pRead->mutex, NULL);
+  taosThreadMutexInit(&pReader->mutex, NULL);
 
-  /*if (pRead->cond.enableRef) {*/
-  /*walOpenRef(pWal);*/
-  /*}*/
-
-  pRead->pHead = taosMemoryMalloc(sizeof(SWalCkHead));
-  if (pRead->pHead == NULL) {
+  pReader->pHead = taosMemoryMalloc(sizeof(SWalCkHead));
+  if (pReader->pHead == NULL) {
     terrno = TSDB_CODE_WAL_OUT_OF_MEMORY;
-    taosMemoryFree(pRead);
+    taosMemoryFree(pReader);
     return NULL;
   }
 
-  return pRead;
+  /*if (pReader->cond.enableRef) {*/
+  /* taosHashPut(pWal->pRefHash, &pReader->readerId, sizeof(int64_t), &pReader, sizeof(void *));*/
+  /*}*/
+
+  return pReader;
 }
 
-void walCloseReader(SWalReader *pRead) {
-  taosCloseFile(&pRead->pIdxFile);
-  taosCloseFile(&pRead->pLogFile);
-  taosMemoryFreeClear(pRead->pHead);
-  taosMemoryFree(pRead);
+void walCloseReader(SWalReader *pReader) {
+  taosCloseFile(&pReader->pIdxFile);
+  taosCloseFile(&pReader->pLogFile);
+  /*if (pReader->cond.enableRef) {*/
+  /*taosHashRemove(pReader->pWal->pRefHash, &pReader->readerId, sizeof(int64_t));*/
+  /*}*/
+  taosMemoryFreeClear(pReader->pHead);
+  taosMemoryFree(pReader);
 }
 
-int32_t walNextValidMsg(SWalReader *pRead) {
-  int64_t fetchVer = pRead->curVersion;
-  int64_t lastVer = walGetLastVer(pRead->pWal);
-  int64_t committedVer = walGetCommittedVer(pRead->pWal);
-  int64_t appliedVer = walGetAppliedVer(pRead->pWal);
-  int64_t endVer = pRead->cond.scanUncommited ? lastVer : committedVer;
+int32_t walNextValidMsg(SWalReader *pReader) {
+  int64_t fetchVer = pReader->curVersion;
+  int64_t lastVer = walGetLastVer(pReader->pWal);
+  int64_t committedVer = walGetCommittedVer(pReader->pWal);
+  int64_t appliedVer = walGetAppliedVer(pReader->pWal);
+  int64_t endVer = pReader->cond.scanUncommited ? lastVer : committedVer;
   endVer = TMIN(appliedVer, endVer);
 
   wDebug("vgId:%d wal start to fetch, ver %ld, last ver %ld commit ver %ld, applied ver %ld, end ver %ld",
-         pRead->pWal->cfg.vgId, fetchVer, lastVer, committedVer, appliedVer, endVer);
-  pRead->curStopped = 0;
+         pReader->pWal->cfg.vgId, fetchVer, lastVer, committedVer, appliedVer, endVer);
+  pReader->curStopped = 0;
   while (fetchVer <= endVer) {
-    if (walFetchHeadNew(pRead, fetchVer) < 0) {
+    if (walFetchHeadNew(pReader, fetchVer) < 0) {
       return -1;
     }
-    if (pRead->pHead->head.msgType == TDMT_VND_SUBMIT ||
-        (IS_META_MSG(pRead->pHead->head.msgType) && pRead->cond.scanMeta)) {
-      if (walFetchBodyNew(pRead) < 0) {
+    if (pReader->pHead->head.msgType == TDMT_VND_SUBMIT ||
+        (IS_META_MSG(pReader->pHead->head.msgType) && pReader->cond.scanMeta)) {
+      if (walFetchBodyNew(pReader) < 0) {
         return -1;
       }
       return 0;
     } else {
-      if (walSkipFetchBodyNew(pRead) < 0) {
+      if (walSkipFetchBodyNew(pReader) < 0) {
         return -1;
       }
       fetchVer++;
-      ASSERT(fetchVer == pRead->curVersion);
+      ASSERT(fetchVer == pReader->curVersion);
     }
   }
-  pRead->curStopped = 1;
+  pReader->curStopped = 1;
   return -1;
 }
 
-static int64_t walReadSeekFilePos(SWalReader *pRead, int64_t fileFirstVer, int64_t ver) {
+static int64_t walReadSeekFilePos(SWalReader *pReader, int64_t fileFirstVer, int64_t ver) {
   int64_t ret = 0;
 
-  TdFilePtr pIdxTFile = pRead->pIdxFile;
-  TdFilePtr pLogTFile = pRead->pLogFile;
+  TdFilePtr pIdxTFile = pReader->pIdxFile;
+  TdFilePtr pLogTFile = pReader->pLogFile;
 
   // seek position
   int64_t offset = (ver - fileFirstVer) * sizeof(SWalIdxEntry);
   ret = taosLSeekFile(pIdxTFile, offset, SEEK_SET);
   if (ret < 0) {
     terrno = TAOS_SYSTEM_ERROR(errno);
-    wError("vgId:%d, failed to seek idx file, index:%" PRId64 ", pos:%" PRId64 ", since %s", pRead->pWal->cfg.vgId, ver,
-           offset, terrstr());
+    wError("vgId:%d, failed to seek idx file, index:%" PRId64 ", pos:%" PRId64 ", since %s", pReader->pWal->cfg.vgId,
+           ver, offset, terrstr());
     return -1;
   }
   SWalIdxEntry entry = {0};
   if ((ret = taosReadFile(pIdxTFile, &entry, sizeof(SWalIdxEntry))) != sizeof(SWalIdxEntry)) {
     if (ret < 0) {
       terrno = TAOS_SYSTEM_ERROR(errno);
-      wError("vgId:%d, failed to read idx file, since %s", pRead->pWal->cfg.vgId, terrstr());
+      wError("vgId:%d, failed to read idx file, since %s", pReader->pWal->cfg.vgId, terrstr());
     } else {
       terrno = TSDB_CODE_WAL_FILE_CORRUPTED;
       wError("vgId:%d, read idx file incompletely, read bytes %" PRId64 ", bytes should be %" PRIu64,
-             pRead->pWal->cfg.vgId, ret, sizeof(SWalIdxEntry));
+             pReader->pWal->cfg.vgId, ret, sizeof(SWalIdxEntry));
     }
     return -1;
   }
@@ -130,79 +135,79 @@ static int64_t walReadSeekFilePos(SWalReader *pRead, int64_t fileFirstVer, int64
   ret = taosLSeekFile(pLogTFile, entry.offset, SEEK_SET);
   if (ret < 0) {
     terrno = TAOS_SYSTEM_ERROR(errno);
-    wError("vgId:%d, failed to seek log file, index:%" PRId64 ", pos:%" PRId64 ", since %s", pRead->pWal->cfg.vgId, ver,
-           entry.offset, terrstr());
+    wError("vgId:%d, failed to seek log file, index:%" PRId64 ", pos:%" PRId64 ", since %s", pReader->pWal->cfg.vgId,
+           ver, entry.offset, terrstr());
     return -1;
   }
   return ret;
 }
 
-static int32_t walReadChangeFile(SWalReader *pRead, int64_t fileFirstVer) {
+static int32_t walReadChangeFile(SWalReader *pReader, int64_t fileFirstVer) {
   char fnameStr[WAL_FILE_LEN];
 
-  taosCloseFile(&pRead->pIdxFile);
-  taosCloseFile(&pRead->pLogFile);
+  taosCloseFile(&pReader->pIdxFile);
+  taosCloseFile(&pReader->pLogFile);
 
-  walBuildLogName(pRead->pWal, fileFirstVer, fnameStr);
-  TdFilePtr pLogTFile = taosOpenFile(fnameStr, TD_FILE_READ);
-  if (pLogTFile == NULL) {
+  walBuildLogName(pReader->pWal, fileFirstVer, fnameStr);
+  TdFilePtr pLogFile = taosOpenFile(fnameStr, TD_FILE_READ);
+  if (pLogFile == NULL) {
     terrno = TAOS_SYSTEM_ERROR(errno);
-    wError("vgId:%d, cannot open file %s, since %s", pRead->pWal->cfg.vgId, fnameStr, terrstr());
+    wError("vgId:%d, cannot open file %s, since %s", pReader->pWal->cfg.vgId, fnameStr, terrstr());
     return -1;
   }
 
-  pRead->pLogFile = pLogTFile;
+  pReader->pLogFile = pLogFile;
 
-  walBuildIdxName(pRead->pWal, fileFirstVer, fnameStr);
-  TdFilePtr pIdxTFile = taosOpenFile(fnameStr, TD_FILE_READ);
-  if (pIdxTFile == NULL) {
+  walBuildIdxName(pReader->pWal, fileFirstVer, fnameStr);
+  TdFilePtr pIdxFile = taosOpenFile(fnameStr, TD_FILE_READ);
+  if (pIdxFile == NULL) {
     terrno = TAOS_SYSTEM_ERROR(errno);
-    wError("vgId:%d, cannot open file %s, since %s", pRead->pWal->cfg.vgId, fnameStr, terrstr());
+    wError("vgId:%d, cannot open file %s, since %s", pReader->pWal->cfg.vgId, fnameStr, terrstr());
     return -1;
   }
 
-  pRead->pIdxFile = pIdxTFile;
+  pReader->pIdxFile = pIdxFile;
   return 0;
 }
 
-int32_t walReadSeekVerImpl(SWalReader *pRead, int64_t ver) {
-  SWal *pWal = pRead->pWal;
+int32_t walReadSeekVerImpl(SWalReader *pReader, int64_t ver) {
+  SWal *pWal = pReader->pWal;
 
+  // bsearch in fileSet
   SWalFileInfo tmpInfo;
   tmpInfo.firstVer = ver;
-  // bsearch in fileSet
   SWalFileInfo *pRet = taosArraySearch(pWal->fileInfoSet, &tmpInfo, compareWalFileInfo, TD_LE);
   ASSERT(pRet != NULL);
-  if (pRead->curFileFirstVer != pRet->firstVer) {
+  if (pReader->curFileFirstVer != pRet->firstVer) {
     // error code was set inner
-    if (walReadChangeFile(pRead, pRet->firstVer) < 0) {
+    if (walReadChangeFile(pReader, pRet->firstVer) < 0) {
       return -1;
     }
   }
 
   // error code was set inner
-  if (walReadSeekFilePos(pRead, pRet->firstVer, ver) < 0) {
+  if (walReadSeekFilePos(pReader, pRet->firstVer, ver) < 0) {
     return -1;
   }
 
-  wDebug("wal version reset from %ld(invalid: %d) to %ld", pRead->curVersion, pRead->curInvalid, ver);
+  wDebug("wal version reset from %ld(invalid: %d) to %ld", pReader->curVersion, pReader->curInvalid, ver);
 
-  pRead->curVersion = ver;
+  pReader->curVersion = ver;
   return 0;
 }
 
-int32_t walReadSeekVer(SWalReader *pRead, int64_t ver) {
-  SWal *pWal = pRead->pWal;
-  if (!pRead->curInvalid && ver == pRead->curVersion) {
+int32_t walReadSeekVer(SWalReader *pReader, int64_t ver) {
+  SWal *pWal = pReader->pWal;
+  if (!pReader->curInvalid && ver == pReader->curVersion) {
     wDebug("wal version %ld match, no need to reset", ver);
     return 0;
   }
 
-  pRead->curInvalid = 1;
-  pRead->curVersion = ver;
+  pReader->curInvalid = 1;
+  pReader->curVersion = ver;
 
   if (ver > pWal->vers.lastVer || ver < pWal->vers.firstVer) {
-    wDebug("vgId:%d, invalid index:%" PRId64 ", first index:%" PRId64 ", last index:%" PRId64, pRead->pWal->cfg.vgId,
+    wDebug("vgId:%d, invalid index:%" PRId64 ", first index:%" PRId64 ", last index:%" PRId64, pReader->pWal->cfg.vgId,
            ver, pWal->vers.firstVer, pWal->vers.lastVer);
     terrno = TSDB_CODE_WAL_LOG_NOT_EXIST;
     return -1;
@@ -210,7 +215,7 @@ int32_t walReadSeekVer(SWalReader *pRead, int64_t ver) {
   if (ver < pWal->vers.snapshotVer) {
   }
 
-  if (walReadSeekVerImpl(pRead, ver) < 0) {
+  if (walReadSeekVerImpl(pReader, ver) < 0) {
     return -1;
   }
 
diff --git a/source/libs/wal/src/walRef.c b/source/libs/wal/src/walRef.c
new file mode 100644
index 0000000000..bd0f6fb1a8
--- /dev/null
+++ b/source/libs/wal/src/walRef.c
@@ -0,0 +1,89 @@
+/*
+ * Copyright (c) 2019 TAOS Data, Inc. <jhtao@taosdata.com>
+ *
+ * This program is free software: you can use, redistribute, and/or modify
+ * it under the terms of the GNU Affero General Public License, version 3
+ * or later ("AGPL"), as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.
+ *
+ * You should have received a copy of the GNU Affero General Public License
+ * along with this program. If not, see <http://www.gnu.org/licenses/>.
+ */
+
+#include "cJSON.h"
+#include "os.h"
+#include "taoserror.h"
+#include "tutil.h"
+#include "walInt.h"
+
+SWalRef *walOpenRef(SWal *pWal) {
+  SWalRef *pRef = taosMemoryCalloc(1, sizeof(SWalRef));
+  if (pRef == NULL) {
+    return NULL;
+  }
+  pRef->refId = tGenIdPI64();
+  pRef->refVer = -1;
+  pRef->refFile = -1;
+  pRef->pWal = pWal;
+  taosHashPut(pWal->pRefHash, &pRef->refId, sizeof(int64_t), &pRef, sizeof(void *));
+  return pRef;
+}
+
+void walCloseRef(SWal *pWal, int64_t refId) {
+  SWalRef *pRef = *(SWalRef **)taosHashGet(pWal->pRefHash, &refId, sizeof(int64_t));
+  taosHashRemove(pWal->pRefHash, &refId, sizeof(int64_t));
+  taosMemoryFree(pRef);
+}
+
+int32_t walRefVer(SWalRef *pRef, int64_t ver) {
+  SWal *pWal = pRef->pWal;
+  if (pRef->refVer != ver) {
+    taosThreadMutexLock(&pWal->mutex);
+    if (ver < pWal->vers.firstVer || ver > pWal->vers.lastVer) {
+      taosThreadMutexUnlock(&pWal->mutex);
+      terrno = TSDB_CODE_WAL_INVALID_VER;
+      return -1;
+    }
+
+    pRef->refVer = ver;
+    // bsearch in fileSet
+    SWalFileInfo tmpInfo;
+    tmpInfo.firstVer = ver;
+    SWalFileInfo *pRet = taosArraySearch(pWal->fileInfoSet, &tmpInfo, compareWalFileInfo, TD_LE);
+    ASSERT(pRet != NULL);
+    pRef->refFile = pRet->firstVer;
+
+    taosThreadMutexUnlock(&pWal->mutex);
+  }
+
+  return 0;
+}
+
+void walUnrefVer(SWalRef *pRef) {
+  pRef->refId = -1;
+  pRef->refFile = -1;
+}
+
+SWalRef *walRefCommittedVer(SWal *pWal) {
+  SWalRef *pRef = walOpenRef(pWal);
+  if (pRef == NULL) {
+    return NULL;
+  }
+  taosThreadMutexLock(&pWal->mutex);
+
+  int64_t ver = walGetCommittedVer(pWal);
+
+  pRef->refVer = ver;
+  // bsearch in fileSet
+  SWalFileInfo tmpInfo;
+  tmpInfo.firstVer = ver;
+  SWalFileInfo *pRet = taosArraySearch(pWal->fileInfoSet, &tmpInfo, compareWalFileInfo, TD_LE);
+  ASSERT(pRet != NULL);
+  pRef->refFile = pRet->firstVer;
+
+  taosThreadMutexUnlock(&pWal->mutex);
+  return pRef;
+}
diff --git a/source/libs/wal/src/walWrite.c b/source/libs/wal/src/walWrite.c
index d6348cc5dd..81500d8088 100644
--- a/source/libs/wal/src/walWrite.c
+++ b/source/libs/wal/src/walWrite.c
@@ -26,7 +26,7 @@ int32_t walRestoreFromSnapshot(SWal *pWal, int64_t ver) {
     pIter = taosHashIterate(pWal->pRefHash, pIter);
     if (pIter == NULL) break;
     SWalRef *pRef = (SWalRef *)pIter;
-    if (pRef->ver != -1) {
+    if (pRef->refVer != -1) {
       taosHashCancelIterate(pWal->pRefHash, pIter);
       return -1;
     }
@@ -215,22 +215,23 @@ int32_t walRollback(SWal *pWal, int64_t ver) {
 
 static FORCE_INLINE int32_t walCheckAndRoll(SWal *pWal) {
   if (taosArrayGetSize(pWal->fileInfoSet) == 0) {
-    /*pWal->vers.firstVer = index;*/
     if (walRollImpl(pWal) < 0) {
       return -1;
     }
-  } else {
-    int64_t passed = walGetSeq() - pWal->lastRollSeq;
-    if (pWal->cfg.rollPeriod != -1 && pWal->cfg.rollPeriod != 0 && passed > pWal->cfg.rollPeriod) {
-      if (walRollImpl(pWal) < 0) {
-        return -1;
-      }
-    } else if (pWal->cfg.segSize != -1 && pWal->cfg.segSize != 0 && walGetLastFileSize(pWal) > pWal->cfg.segSize) {
-      if (walRollImpl(pWal) < 0) {
-        return -1;
-      }
+    return 0;
+  }
+
+  int64_t passed = walGetSeq() - pWal->lastRollSeq;
+  if (pWal->cfg.rollPeriod != -1 && pWal->cfg.rollPeriod != 0 && passed > pWal->cfg.rollPeriod) {
+    if (walRollImpl(pWal) < 0) {
+      return -1;
+    }
+  } else if (pWal->cfg.segSize != -1 && pWal->cfg.segSize != 0 && walGetLastFileSize(pWal) > pWal->cfg.segSize) {
+    if (walRollImpl(pWal) < 0) {
+      return -1;
     }
   }
+
   return 0;
 }
 
@@ -260,6 +261,16 @@ int32_t walEndSnapshot(SWal *pWal) {
   pWal->vers.snapshotVer = ver;
   int ts = taosGetTimestampSec();
 
+  int64_t minVerToDelete = ver;
+  void   *pIter = NULL;
+  while (1) {
+    pIter = taosHashIterate(pWal->pRefHash, pIter);
+    if (pIter == NULL) break;
+    SWalRef *pRef = *(SWalRef **)pIter;
+    if (pRef->refVer == -1) continue;
+    minVerToDelete = TMIN(minVerToDelete, pRef->refVer);
+  }
+
   int          deleteCnt = 0;
   int64_t      newTotSize = pWal->totSize;
   SWalFileInfo tmp;
diff --git a/source/util/src/terror.c b/source/util/src/terror.c
index 80ad480e43..ad6eff3c12 100644
--- a/source/util/src/terror.c
+++ b/source/util/src/terror.c
@@ -512,7 +512,7 @@ TAOS_DEFINE_ERROR(TSDB_CODE_PAR_INTER_OFFSET_UNIT,         "Cannot use 'year' as
 TAOS_DEFINE_ERROR(TSDB_CODE_PAR_INTER_OFFSET_TOO_BIG,      "Interval offset should be shorter than interval")
 TAOS_DEFINE_ERROR(TSDB_CODE_PAR_INTER_SLIDING_UNIT,        "Does not support sliding when interval is natural month/year")
 TAOS_DEFINE_ERROR(TSDB_CODE_PAR_INTER_SLIDING_TOO_BIG,     "sliding value no larger than the interval value")
-TAOS_DEFINE_ERROR(TSDB_CODE_PAR_INTER_SLIDING_TOO_SMALL,   "sliding value can not less than 1% of interval value")
+TAOS_DEFINE_ERROR(TSDB_CODE_PAR_INTER_SLIDING_TOO_SMALL,   "sliding value can not less than 1%% of interval value")
 TAOS_DEFINE_ERROR(TSDB_CODE_PAR_ONLY_ONE_JSON_TAG,         "Only one tag if there is a json tag")
 TAOS_DEFINE_ERROR(TSDB_CODE_PAR_INCORRECT_NUM_OF_COL,      "Query block has incorrect number of result columns")
 TAOS_DEFINE_ERROR(TSDB_CODE_PAR_INCORRECT_TIMESTAMP_VAL,   "Incorrect TIMESTAMP value")
diff --git a/tests/pytest/crash_gen/crash_gen_main.py b/tests/pytest/crash_gen/crash_gen_main.py
index 8990c24305..d5ffc1b7c1 100755
--- a/tests/pytest/crash_gen/crash_gen_main.py
+++ b/tests/pytest/crash_gen/crash_gen_main.py
@@ -1327,6 +1327,8 @@ class Task():
 
                 # TDengine 3.0 Error Codes:
                 0x0333, # Object is creating # TODO: this really is NOT an acceptable error
+                0x0369, # Tag already exists
+                0x0388, # Database not exist
                 0x03A0, # STable already exists
                 0x03A1, # STable [does] not exist
                 0x03AA, # Tag already exists
diff --git a/tests/system-test/7-tmq/TD-17803.py b/tests/system-test/7-tmq/TD-17803.py
new file mode 100644
index 0000000000..771ff83a29
--- /dev/null
+++ b/tests/system-test/7-tmq/TD-17803.py
@@ -0,0 +1,198 @@
+from distutils.log import error
+import taos
+import sys
+import time
+import socket
+import os
+import threading
+import subprocess
+import platform
+
+from util.log import *
+from util.sql import *
+from util.cases import *
+from util.dnodes import *
+from util.common import *
+sys.path.append("./7-tmq")
+from tmqCommon import *
+
+
+
+class TDTestCase:
+    def __init__(self):
+        self.snapshot   = 0
+        self.replica    = 3
+        self.vgroups    = 3
+        self.ctbNum     = 2
+        self.rowsPerTbl = 2
+
+    def init(self, conn, logSql):
+        tdLog.debug(f"start to excute {__file__}")
+        tdSql.init(conn.cursor())
+        #tdSql.init(conn.cursor(), logSql)  # output sql.txt file
+
+    def checkFileContent(self, consumerId, queryString):
+        buildPath = tdCom.getBuildPath()
+        cfgPath = tdCom.getClientCfgPath()
+        dstFile = '%s/../log/dstrows_%d.txt'%(cfgPath, consumerId)
+        cmdStr = '%s/build/bin/taos -c %s -s "%s >> %s"'%(buildPath, cfgPath, queryString, dstFile)
+        tdLog.info(cmdStr)
+        os.system(cmdStr)
+        
+        consumeRowsFile = '%s/../log/consumerid_%d.txt'%(cfgPath, consumerId)
+        tdLog.info("rows file: %s, %s"%(consumeRowsFile, dstFile))
+
+        consumeFile = open(consumeRowsFile, mode='r')
+        queryFile = open(dstFile, mode='r')
+        
+        # skip first line for it is schema
+        queryFile.readline()
+
+        while True:
+            dst = queryFile.readline()
+            src = consumeFile.readline()
+            
+            if dst:
+                if dst != src:
+                    tdLog.exit("consumerId %d consume rows is not match the rows by direct query"%consumerId)
+            else:
+                break
+        return
+
+    def prepareTestEnv(self):
+        tdLog.printNoPrefix("======== prepare test env include database, stable, ctables, and insert data: ")
+        paraDict = {'dbName':     'dbt',
+                    'dropFlag':   1,
+                    'event':      '',
+                    'vgroups':    4,
+                    'stbName':    'stb',
+                    'colPrefix':  'c',
+                    'tagPrefix':  't',
+                    'colSchema':   [{'type': 'INT', 'count':1},{'type': 'BIGINT', 'count':1}],
+                    'tagSchema':   [{'type': 'INT', 'count':1},{'type': 'BIGINT', 'count':1}],
+                    'ctbPrefix':  'ctb',
+                    'ctbStartIdx': 0,
+                    'ctbNum':     2,
+                    'rowsPerTbl': 1000,
+                    'batchNum':   10,
+                    'startTs':    1640966400000,  # 2022-01-01 00:00:00.000
+                    'pollDelay':  3,
+                    'showMsg':    1,
+                    'showRow':    1,
+                    'snapshot':   0}
+
+        paraDict['vgroups'] = self.vgroups
+        paraDict['ctbNum'] = self.ctbNum
+        paraDict['rowsPerTbl'] = self.rowsPerTbl
+        
+        tmqCom.initConsumerTable()
+        tdCom.create_database(tdSql, paraDict["dbName"],paraDict["dropFlag"], vgroups=paraDict["vgroups"],replica=self.replica)
+        tdLog.info("create stb")
+        tmqCom.create_stable(tdSql, dbName=paraDict["dbName"],stbName=paraDict["stbName"])
+        tdLog.info("create ctb")
+        tmqCom.create_ctable(tdSql, dbName=paraDict["dbName"],stbName=paraDict["stbName"],ctbPrefix=paraDict['ctbPrefix'],
+                             ctbNum=paraDict["ctbNum"],ctbStartIdx=paraDict['ctbStartIdx'])
+        tdLog.info("insert data")
+        tmqCom.insert_data_interlaceByMultiTbl(tsql=tdSql,dbName=paraDict["dbName"],ctbPrefix=paraDict["ctbPrefix"],
+                                               ctbNum=paraDict["ctbNum"],rowsPerTbl=paraDict["rowsPerTbl"],batchNum=paraDict["batchNum"],
+                                               startTs=paraDict["startTs"],ctbStartIdx=paraDict['ctbStartIdx'])
+        # tmqCom.insert_data_with_autoCreateTbl(tsql=tdSql,dbName=paraDict["dbName"],stbName=paraDict["stbName"],ctbPrefix="ctbx",
+        #                                       ctbNum=paraDict["ctbNum"],rowsPerTbl=paraDict["rowsPerTbl"],batchNum=paraDict["batchNum"],
+        #                                       startTs=paraDict["startTs"],ctbStartIdx=paraDict['ctbStartIdx'])
+        # tmqCom.asyncInsertDataByInterlace(paraDict)
+        tdLog.printNoPrefix("11111111111111111111111") 
+        tmqCom.create_ntable(tdSql, dbname=paraDict["dbName"], tbname_prefix="ntb", tbname_index_start_num = 1, column_elm_list=paraDict["colSchema"], colPrefix='c', tblNum=1)
+        tdLog.printNoPrefix("222222222222222") 
+        tmqCom.insert_rows_into_ntbl(tdSql, dbname=paraDict["dbName"], tbname_prefix="ntb", tbname_index_start_num = 1, column_ele_list=paraDict["colSchema"], startTs=paraDict["startTs"], tblNum=1, rows=2)        # tdLog.info("restart taosd to ensure that the data falls into the disk")        
+        
+        tdLog.printNoPrefix("333333333333333333333")  
+        tdSql.query("drop database %s"%paraDict["dbName"])
+        tdLog.printNoPrefix("44444444444444444") 
+        return
+
+    def tmqCase1(self):        
+        tdLog.printNoPrefix("======== test case 1: ")       
+        
+        # create and start thread
+        paraDict = {'dbName':     'dbt',
+                    'dropFlag':   1,
+                    'event':      '',
+                    'vgroups':    4,
+                    'stbName':    'stb',
+                    'colPrefix':  'c',
+                    'tagPrefix':  't',
+                    'colSchema':   [{'type': 'INT', 'count':1},{'type': 'BIGINT', 'count':1},{'type': 'DOUBLE', 'count':1},{'type': 'BINARY', 'len':32, 'count':1},{'type': 'NCHAR', 'len':32, 'count':1},{'type': 'TIMESTAMP', 'count':1}],
+                    'tagSchema':   [{'type': 'INT', 'count':1},{'type': 'BIGINT', 'count':1},{'type': 'DOUBLE', 'count':1},{'type': 'BINARY', 'len':32, 'count':1},{'type': 'NCHAR', 'len':32, 'count':1}],
+                    'ctbPrefix':  'ctb',
+                    'ctbStartIdx': 0,
+                    'ctbNum':     100,
+                    'rowsPerTbl': 1000,
+                    'batchNum':   100,
+                    'startTs':    1640966400000,  # 2022-01-01 00:00:00.000
+                    'pollDelay':  3,
+                    'showMsg':    1,
+                    'showRow':    1,
+                    'snapshot':   1}
+
+        paraDict['vgroups'] = self.vgroups
+        paraDict['ctbNum'] = self.ctbNum
+        paraDict['rowsPerTbl'] = self.rowsPerTbl
+        
+        tdLog.info("create topics from stb1")
+        topicFromStb1 = 'topic_stb1'        
+        queryString = "select ts, c1, c2 from %s.%s  where t4 == 'beijing' or t4 == 'changsha' "%(paraDict['dbName'], paraDict['stbName'])
+        sqlString = "create topic %s as %s" %(topicFromStb1, queryString)
+        tdLog.info("create topic sql: %s"%sqlString)
+        tdSql.execute(sqlString)
+        
+        consumerId     = 0
+        expectrowcnt   = paraDict["rowsPerTbl"] * paraDict["ctbNum"]
+        topicList      = topicFromStb1
+        ifcheckdata    = 0
+        ifManualCommit = 0
+        keyList        = 'group.id:cgrp1,\
+                        enable.auto.commit:false,\
+                        auto.commit.interval.ms:6000,\
+                        auto.offset.reset:earliest'
+        tmqCom.insertConsumerInfo(consumerId, expectrowcnt,topicList,keyList,ifcheckdata,ifManualCommit)
+
+        tdLog.info("start consume processor")
+        pollDelay = 100
+        showMsg   = 1
+        showRow   = 1
+        tmqCom.startTmqSimProcess(pollDelay=paraDict['pollDelay'],dbName=paraDict["dbName"],showMsg=paraDict['showMsg'], showRow=paraDict['showRow'],snapshot=paraDict['snapshot'])
+
+        tdLog.info("start to check consume result")
+        expectRows = 1
+        resultList = tmqCom.selectConsumeResult(expectRows)
+        totalConsumeRows = 0
+        for i in range(expectRows):
+            totalConsumeRows += resultList[i]
+
+        tdSql.query(queryString)
+        totalRowsInserted = tdSql.getRows()
+        
+        tdLog.info("act consume rows: %d, act insert rows: %d, expect consume rows: %d, "%(totalConsumeRows, totalRowsInserted, expectrowcnt))
+        
+        if totalConsumeRows != expectrowcnt:
+            tdLog.exit("tmq consume rows error!")
+            
+        # tmqCom.checkFileContent(consumerId, queryString)   
+
+        tmqCom.waitSubscriptionExit(tdSql, topicFromStb1)
+        tdSql.query("drop topic %s"%topicFromStb1)
+
+        tdLog.printNoPrefix("======== test case 1 end ...... ")
+
+    def run(self):
+        self.prepareTestEnv()
+        # self.tmqCase1()
+
+    def stop(self):
+        tdSql.close()
+        tdLog.success(f"{__file__} successfully executed")
+
+event = threading.Event()
+
+tdCases.addLinux(__file__, TDTestCase())
+tdCases.addWindows(__file__, TDTestCase())
diff --git a/tests/system-test/7-tmq/dataFromTsdbNWal-multiCtb.py b/tests/system-test/7-tmq/dataFromTsdbNWal-multiCtb.py
new file mode 100644
index 0000000000..2216000214
--- /dev/null
+++ b/tests/system-test/7-tmq/dataFromTsdbNWal-multiCtb.py
@@ -0,0 +1,250 @@
+
+import taos
+import sys
+import time
+import socket
+import os
+import threading
+import math
+
+from util.log import *
+from util.sql import *
+from util.cases import *
+from util.dnodes import *
+from util.common import *
+sys.path.append("./7-tmq")
+from tmqCommon import *
+
+class TDTestCase:
+    def __init__(self):
+        self.vgroups    = 4
+        self.ctbNum     = 100
+        self.rowsPerTbl = 1000
+        
+    def init(self, conn, logSql):
+        tdLog.debug(f"start to excute {__file__}")
+        tdSql.init(conn.cursor(), False)
+
+    def prepareTestEnv(self):
+        tdLog.printNoPrefix("======== prepare test env include database, stable, ctables, and insert data: ")
+        paraDict = {'dbName':     'dbt',
+                    'dropFlag':   1,
+                    'event':      '',
+                    'vgroups':    1,
+                    'stbName':    'stb',
+                    'colPrefix':  'c',
+                    'tagPrefix':  't',
+                    'colSchema':   [{'type': 'INT', 'count':1},{'type': 'BIGINT', 'count':1},{'type': 'DOUBLE', 'count':1},{'type': 'BINARY', 'len':32, 'count':1},{'type': 'NCHAR', 'len':32, 'count':1},{'type': 'TIMESTAMP', 'count':1}],
+                    'tagSchema':   [{'type': 'INT', 'count':1},{'type': 'BIGINT', 'count':1},{'type': 'DOUBLE', 'count':1},{'type': 'BINARY', 'len':32, 'count':1},{'type': 'NCHAR', 'len':32, 'count':1}],
+                    'ctbPrefix':  'ctb',
+                    'ctbStartIdx': 0,
+                    'ctbNum':     100,
+                    'rowsPerTbl': 1000,
+                    'batchNum':   100,
+                    'startTs':    1640966400000,  # 2022-01-01 00:00:00.000
+                    'pollDelay':  10,
+                    'showMsg':    1,
+                    'showRow':    1,
+                    'snapshot':   1}
+
+        paraDict['vgroups'] = self.vgroups
+        paraDict['ctbNum'] = self.ctbNum
+        paraDict['rowsPerTbl'] = self.rowsPerTbl
+        
+        tmqCom.initConsumerTable()
+        tdCom.create_database(tdSql, paraDict["dbName"],paraDict["dropFlag"], vgroups=paraDict["vgroups"],replica=1)
+        tdLog.info("create stb")
+        tmqCom.create_stable(tdSql, dbName=paraDict["dbName"],stbName=paraDict["stbName"])
+        tdLog.info("create ctb")
+        tmqCom.create_ctable(tdSql, dbName=paraDict["dbName"],stbName=paraDict["stbName"],ctbPrefix=paraDict['ctbPrefix'],
+                             ctbNum=paraDict["ctbNum"],ctbStartIdx=paraDict['ctbStartIdx'])
+        
+        tdLog.info("insert data")
+        tmqCom.insert_data_interlaceByMultiTbl(tsql=tdSql,dbName=paraDict["dbName"],ctbPrefix=paraDict["ctbPrefix"],
+                                               ctbNum=paraDict["ctbNum"],rowsPerTbl=paraDict["rowsPerTbl"],batchNum=paraDict["batchNum"],
+                                               startTs=paraDict["startTs"],ctbStartIdx=paraDict['ctbStartIdx'])
+        
+        tdLog.info("flush db to let data falls into the disk")
+        tdSql.query("flush database %s"%(paraDict['dbName']))
+        return
+
+    def tmqCase1(self):
+        tdLog.printNoPrefix("======== test case 1: ")
+        paraDict = {'dbName':     'dbt',
+                    'dropFlag':   1,
+                    'event':      '',
+                    'vgroups':    1,
+                    'stbName':    'stb',
+                    'colPrefix':  'c',
+                    'tagPrefix':  't',
+                    'colSchema':   [{'type': 'INT', 'count':1},{'type': 'BIGINT', 'count':1},{'type': 'DOUBLE', 'count':1},{'type': 'BINARY', 'len':32, 'count':1},{'type': 'NCHAR', 'len':32, 'count':1},{'type': 'TIMESTAMP', 'count':1}],
+                    'tagSchema':   [{'type': 'INT', 'count':1},{'type': 'BIGINT', 'count':1},{'type': 'DOUBLE', 'count':1},{'type': 'BINARY', 'len':32, 'count':1},{'type': 'NCHAR', 'len':32, 'count':1}],
+                    'ctbPrefix':  'ctb',
+                    'ctbStartIdx': 0,
+                    'ctbNum':     100,
+                    'rowsPerTbl': 1000,
+                    'batchNum':   500,
+                    'startTs':    1640966400000,  # 2022-01-01 00:00:00.000
+                    'pollDelay':  5,
+                    'showMsg':    1,
+                    'showRow':    1,
+                    'snapshot':   1}
+
+        paraDict['vgroups'] = self.vgroups
+        paraDict['ctbNum'] = self.ctbNum
+        paraDict['rowsPerTbl'] = self.rowsPerTbl
+        
+        topicNameList = ['topic1']
+        expectRowsList = []
+        tmqCom.initConsumerTable()
+        
+        tdLog.info("create topics from stb with filter")
+        queryString = "select * from %s.%s"%(paraDict['dbName'], paraDict['stbName'])
+        # sqlString = "create topic %s as stable %s" %(topicNameList[0], paraDict['stbName'])
+        sqlString = "create topic %s as %s" %(topicNameList[0], queryString)
+        tdLog.info("create topic sql: %s"%sqlString)
+        tdSql.execute(sqlString)
+        # tdSql.query(queryString)        
+        # expectRowsList.append(tdSql.getRows())
+
+        # init consume info, and start tmq_sim, then check consume result
+        tdLog.info("insert consume info to consume processor")
+        consumerId   = 0
+        expectrowcnt = paraDict["rowsPerTbl"] * paraDict["ctbNum"] * 2
+        topicList    = topicNameList[0]
+        ifcheckdata  = 1
+        ifManualCommit = 1
+        keyList      = 'group.id:cgrp1, enable.auto.commit:true, auto.commit.interval.ms:1000, auto.offset.reset:earliest'
+        tmqCom.insertConsumerInfo(consumerId, expectrowcnt,topicList,keyList,ifcheckdata,ifManualCommit)
+
+        # after start consume, continue insert some data
+        paraDict['batchNum']   = 100
+        paraDict['startTs'] = paraDict['startTs'] + self.rowsPerTbl
+        pInsertThread = tmqCom.asyncInsertDataByInterlace(paraDict)
+        
+        tdLog.info("start consume processor")
+        tmqCom.startTmqSimProcess(pollDelay=paraDict['pollDelay'],dbName=paraDict["dbName"],showMsg=paraDict['showMsg'], showRow=paraDict['showRow'],snapshot=paraDict['snapshot'])
+
+        pInsertThread.join()
+        
+        tdSql.query(queryString)        
+        expectRowsList.append(tdSql.getRows())
+        
+        tdLog.info("wait the consume result")         
+        expectRows = 1
+        resultList = tmqCom.selectConsumeResult(expectRows)
+                
+        tdLog.info("expect consume rows: %d, act consume rows: %d"%(expectRowsList[0], resultList[0]))
+        if expectRowsList[0] != resultList[0]:
+            tdLog.exit("%d tmq consume rows error!"%consumerId)
+
+        # tmqCom.checkFileContent(consumerId, queryString) 
+
+        tdSql.query("flush database %s"%(paraDict['dbName']))
+        
+        for i in range(len(topicNameList)):
+            tmqCom.waitSubscriptionExit(tdSql,topicNameList[i])    
+            tdSql.query("drop topic %s"%topicNameList[i])
+
+        tdLog.printNoPrefix("======== test case 1 end ...... ")
+
+    def tmqCase2(self):
+        tdLog.printNoPrefix("======== test case 2: ")
+        paraDict = {'dbName':     'dbt',
+                    'dropFlag':   1,
+                    'event':      '',
+                    'vgroups':    1,
+                    'stbName':    'stb',
+                    'colPrefix':  'c',
+                    'tagPrefix':  't',
+                    'colSchema':   [{'type': 'INT', 'count':1},{'type': 'BIGINT', 'count':1},{'type': 'DOUBLE', 'count':1},{'type': 'BINARY', 'len':32, 'count':1},{'type': 'NCHAR', 'len':32, 'count':1},{'type': 'TIMESTAMP', 'count':1}],
+                    'tagSchema':   [{'type': 'INT', 'count':1},{'type': 'BIGINT', 'count':1},{'type': 'DOUBLE', 'count':1},{'type': 'BINARY', 'len':32, 'count':1},{'type': 'NCHAR', 'len':32, 'count':1}],
+                    'ctbPrefix':  'ctb',
+                    'ctbStartIdx': 0,
+                    'ctbNum':     100,
+                    'rowsPerTbl': 1000,
+                    'batchNum':   500,
+                    'startTs':    1640966400000,  # 2022-01-01 00:00:00.000
+                    'pollDelay':  3,
+                    'showMsg':    1,
+                    'showRow':    1,
+                    'snapshot':   1}
+
+        paraDict['vgroups'] = self.vgroups
+        paraDict['ctbNum'] = self.ctbNum
+        paraDict['rowsPerTbl'] = self.rowsPerTbl
+        
+        topicNameList = ['topic1']
+        expectRowsList = []
+        tmqCom.initConsumerTable()
+        
+        tdLog.info("create topics from stb with filter")
+        queryString = "select * from %s.%s"%(paraDict['dbName'], paraDict['stbName'])
+        # sqlString = "create topic %s as stable %s" %(topicNameList[0], paraDict['stbName'])
+        sqlString = "create topic %s as %s" %(topicNameList[0], queryString)
+        tdLog.info("create topic sql: %s"%sqlString)
+        tdSql.execute(sqlString)
+        tdSql.query(queryString)        
+        expectRowsList.append(tdSql.getRows())
+        totalRowsInserted = expectRowsList[0]
+
+        # init consume info, and start tmq_sim, then check consume result
+        tdLog.info("insert consume info to consume processor")
+        consumerId   = 1
+        expectrowcnt = math.ceil(paraDict["rowsPerTbl"] * paraDict["ctbNum"] / 3)
+        topicList    = topicNameList[0]
+        ifcheckdata  = 1
+        ifManualCommit = 1
+        keyList      = 'group.id:cgrp1, enable.auto.commit:true, auto.commit.interval.ms:1000, auto.offset.reset:earliest'
+        tmqCom.insertConsumerInfo(consumerId, expectrowcnt,topicList,keyList,ifcheckdata,ifManualCommit)
+
+        tdLog.info("start consume processor 0")
+        tmqCom.startTmqSimProcess(pollDelay=paraDict['pollDelay'],dbName=paraDict["dbName"],showMsg=paraDict['showMsg'], showRow=paraDict['showRow'],snapshot=paraDict['snapshot'])
+        tdLog.info("wait the consume result") 
+        
+        expectRows = 1
+        resultList = tmqCom.selectConsumeResult(expectRows)
+        actConsumeRows = resultList[0]
+                
+        tdLog.info("act consume rows: %d, expect consume rows between %d and %d"%(actConsumeRows, expectrowcnt, totalRowsInserted))
+        if not (expectrowcnt <= actConsumeRows and totalRowsInserted >= actConsumeRows):
+            tdLog.exit("%d tmq consume rows error!"%consumerId)
+        
+        # reinit consume info, and start tmq_sim, then check consume result
+        tmqCom.initConsumerTable()
+        consumerId   = 2
+        expectrowcnt = math.ceil(paraDict["rowsPerTbl"] * paraDict["ctbNum"] * 2/3)
+        tmqCom.insertConsumerInfo(consumerId, expectrowcnt,topicList,keyList,ifcheckdata,ifManualCommit)
+        
+        tdLog.info("start consume processor 1")
+        tmqCom.startTmqSimProcess(pollDelay=paraDict['pollDelay'],dbName=paraDict["dbName"],showMsg=paraDict['showMsg'], showRow=paraDict['showRow'],snapshot=paraDict['snapshot'])
+        tdLog.info("wait the consume result") 
+        
+        expectRows = 1
+        resultList = tmqCom.selectConsumeResult(expectRows)
+        
+        actConsumeRows = resultList[0]
+        tdLog.info("act consume rows: %d, expect rows: %d, act insert rows: %d"%(actConsumeRows, expectrowcnt, totalRowsInserted))    
+        if not ((actConsumeRows >= expectrowcnt) and (totalRowsInserted > actConsumeRows)):
+            tdLog.exit("%d tmq consume rows error!"%consumerId)
+            
+        for i in range(len(topicNameList)):
+            tmqCom.waitSubscriptionExit(tdSql,topicNameList[i])    
+            tdSql.query("drop topic %s"%topicNameList[i])
+
+        tdLog.printNoPrefix("======== test case 2 end ...... ")
+
+    def run(self):
+        tdSql.prepare()
+        self.prepareTestEnv()
+        self.tmqCase1()
+        self.tmqCase2()
+
+    def stop(self):
+        tdSql.close()
+        tdLog.success(f"{__file__} successfully executed")
+
+event = threading.Event()
+
+tdCases.addLinux(__file__, TDTestCase())
+tdCases.addWindows(__file__, TDTestCase())
diff --git a/tests/system-test/7-tmq/dataFromTsdbNWal.py b/tests/system-test/7-tmq/dataFromTsdbNWal.py
index 227ce9d5a5..faa70f4820 100644
--- a/tests/system-test/7-tmq/dataFromTsdbNWal.py
+++ b/tests/system-test/7-tmq/dataFromTsdbNWal.py
@@ -17,8 +17,8 @@ from tmqCommon import *
 
 class TDTestCase:
     def __init__(self):
-        self.vgroups    = 1
-        self.ctbNum     = 100
+        self.vgroups    = 4
+        self.ctbNum     = 1
         self.rowsPerTbl = 10000
         
     def init(self, conn, logSql):
@@ -38,9 +38,9 @@ class TDTestCase:
                     'tagSchema':   [{'type': 'INT', 'count':1},{'type': 'BIGINT', 'count':1},{'type': 'DOUBLE', 'count':1},{'type': 'BINARY', 'len':32, 'count':1},{'type': 'NCHAR', 'len':32, 'count':1}],
                     'ctbPrefix':  'ctb',
                     'ctbStartIdx': 0,
-                    'ctbNum':     100,
+                    'ctbNum':     1,
                     'rowsPerTbl': 10000,
-                    'batchNum':   3000,
+                    'batchNum':   100,
                     'startTs':    1640966400000,  # 2022-01-01 00:00:00.000
                     'pollDelay':  10,
                     'showMsg':    1,
@@ -85,7 +85,7 @@ class TDTestCase:
                     'rowsPerTbl': 10000,
                     'batchNum':   100,
                     'startTs':    1640966400000,  # 2022-01-01 00:00:00.000
-                    'pollDelay':  3,
+                    'pollDelay':  5,
                     'showMsg':    1,
                     'showRow':    1,
                     'snapshot':   1}
@@ -117,17 +117,16 @@ class TDTestCase:
         keyList      = 'group.id:cgrp1, enable.auto.commit:true, auto.commit.interval.ms:1000, auto.offset.reset:earliest'
         tmqCom.insertConsumerInfo(consumerId, expectrowcnt,topicList,keyList,ifcheckdata,ifManualCommit)
 
-        tdLog.info("start consume processor")
-        tmqCom.startTmqSimProcess(pollDelay=paraDict['pollDelay'],dbName=paraDict["dbName"],showMsg=paraDict['showMsg'], showRow=paraDict['showRow'],snapshot=paraDict['snapshot'])
-
         # after start consume, continue insert some data
         paraDict['batchNum']   = 100
         paraDict['startTs'] = paraDict['startTs'] + self.rowsPerTbl
-        tmqCom.insert_data_interlaceByMultiTbl(tsql=tdSql,dbName=paraDict["dbName"],ctbPrefix=paraDict["ctbPrefix"],
-                                               ctbNum=paraDict["ctbNum"],rowsPerTbl=paraDict["rowsPerTbl"],batchNum=paraDict["batchNum"],
-                                               startTs=paraDict["startTs"],ctbStartIdx=paraDict['ctbStartIdx'])
+        pInsertThread = tmqCom.asyncInsertDataByInterlace(paraDict)
+        
+        tdLog.info("start consume processor")
+        tmqCom.startTmqSimProcess(pollDelay=paraDict['pollDelay'],dbName=paraDict["dbName"],showMsg=paraDict['showMsg'], showRow=paraDict['showRow'],snapshot=paraDict['snapshot'])
 
-        # 
+        pInsertThread.join()
+        
         tdSql.query(queryString)        
         expectRowsList.append(tdSql.getRows())
         
@@ -135,15 +134,16 @@ class TDTestCase:
         expectRows = 1
         resultList = tmqCom.selectConsumeResult(expectRows)
                 
-                
         tdLog.info("expect consume rows: %d, act consume rows: %d"%(expectRowsList[0], resultList[0]))
         if expectRowsList[0] != resultList[0]:
             tdLog.exit("%d tmq consume rows error!"%consumerId)
 
         tmqCom.checkFileContent(consumerId, queryString) 
 
-        time.sleep(10)        
+        tdSql.query("flush database %s"%(paraDict['dbName']))
+
         for i in range(len(topicNameList)):
+            tmqCom.waitSubscriptionExit(tdSql,topicNameList[i])   
             tdSql.query("drop topic %s"%topicNameList[i])
 
         tdLog.printNoPrefix("======== test case 1 end ...... ")
@@ -204,13 +204,12 @@ class TDTestCase:
         
         expectRows = 1
         resultList = tmqCom.selectConsumeResult(expectRows)
-        
-        if not (expectrowcnt <= resultList[0] and totalRowsInserted >= resultList[0]):
-            tdLog.info("act consume rows: %d, expect consume rows between %d and %d"%(resultList[0], expectrowcnt, totalRowsInserted))
+        actConsumeRows = resultList[0]
+                
+        tdLog.info("act consume rows: %d, expect consume rows between %d and %d"%(actConsumeRows, expectrowcnt, totalRowsInserted))
+        if not (expectrowcnt <= actConsumeRows and totalRowsInserted >= actConsumeRows):
             tdLog.exit("%d tmq consume rows error!"%consumerId)
-            
-        firstConsumeRows = resultList[0]
-
+        
         # reinit consume info, and start tmq_sim, then check consume result
         tmqCom.initConsumerTable()
         consumerId   = 2
@@ -224,15 +223,13 @@ class TDTestCase:
         expectRows = 1
         resultList = tmqCom.selectConsumeResult(expectRows)
         
-        actConsumeTotalRows = firstConsumeRows + resultList[0]
-        
-        if not (expectrowcnt >= resultList[0] and totalRowsInserted == actConsumeTotalRows):
-            tdLog.info("act consume rows, first: %d, second: %d "%(firstConsumeRows, resultList[0]))            
-            tdLog.info("and sum of two consume rows: %d should be equal to total inserted rows: %d"%(actConsumeTotalRows, totalRowsInserted))
+        actConsumeRows = resultList[0]
+        tdLog.info("act consume rows: %d, expect rows: %d, act insert rows: %d"%(actConsumeRows, expectrowcnt, totalRowsInserted))    
+        if not ((actConsumeRows >= expectrowcnt) and (totalRowsInserted > actConsumeRows)):
             tdLog.exit("%d tmq consume rows error!"%consumerId)
 
-        time.sleep(10)        
         for i in range(len(topicNameList)):
+            tmqCom.waitSubscriptionExit(tdSql,topicNameList[i]) 
             tdSql.query("drop topic %s"%topicNameList[i])
 
         tdLog.printNoPrefix("======== test case 2 end ...... ")
@@ -241,7 +238,7 @@ class TDTestCase:
         tdSql.prepare()
         self.prepareTestEnv()
         self.tmqCase1()
-        # self.tmqCase2()
+        self.tmqCase2()
 
     def stop(self):
         tdSql.close()
diff --git a/tests/system-test/7-tmq/tmqDnodeRestart.py b/tests/system-test/7-tmq/tmqDnodeRestart.py
index cec6985a4e..5117ee3d24 100644
--- a/tests/system-test/7-tmq/tmqDnodeRestart.py
+++ b/tests/system-test/7-tmq/tmqDnodeRestart.py
@@ -151,41 +151,6 @@ class TDTestCase:
         if not (totalConsumeRows == totalRowsFromQury):
             tdLog.exit("tmq consume rows error!")
 
-
-
-
-        # tdLog.info("****************************************************************************")
-        # tmqCom.initConsumerTable()
-        # consumerId     = 1
-        # expectrowcnt   = paraDict["rowsPerTbl"] * paraDict["ctbNum"] * 2
-        # topicList      = topicFromStb1
-        # ifcheckdata    = 0
-        # ifManualCommit = 0
-        # keyList        = 'group.id:cgrp2,\
-        #                 enable.auto.commit:true,\
-        #                 auto.commit.interval.ms:3000,\
-        #                 auto.offset.reset:earliest'
-        # tmqCom.insertConsumerInfo(consumerId, expectrowcnt,topicList,keyList,ifcheckdata,ifManualCommit)
-
-        # tdLog.info("start consume processor")
-        # tmqCom.startTmqSimProcess(pollDelay=paraDict['pollDelay'],dbName=paraDict["dbName"],showMsg=paraDict['showMsg'], showRow=paraDict['showRow'],snapshot=paraDict['snapshot'])
-        
-        # expectRows = 1
-        # resultList = tmqCom.selectConsumeResult(expectRows)
-        # totalConsumeRows = 0
-        # for i in range(expectRows):
-        #     totalConsumeRows += resultList[i]
-
-        # tdSql.query(queryString)
-        # totalRowsFromQury = tdSql.getRows()
-        
-        # tdLog.info("act consume rows: %d, act query rows: %d"%(totalConsumeRows, totalRowsFromQury))
-        # if not (totalConsumeRows == totalRowsFromQury):
-        #     tdLog.exit("tmq consume rows error!")
-
-        
-        # tdLog.info("****************************************************************************")
-
         tmqCom.waitSubscriptionExit(tdSql, topicFromStb1)
         tdSql.query("drop topic %s"%topicFromStb1)
 
@@ -259,7 +224,7 @@ class TDTestCase:
         tdLog.info("create some new child table and insert data ")
         paraDict["batchNum"] = 100
         paraDict["ctbPrefix"] = 'newCtb'
-        # tmqCom.insert_data_with_autoCreateTbl(tdSql,paraDict["dbName"],paraDict["stbName"],paraDict["ctbPrefix"],paraDict["ctbNum"],paraDict["rowsPerTbl"],paraDict["batchNum"])
+        tmqCom.insert_data_with_autoCreateTbl(tdSql,paraDict["dbName"],paraDict["stbName"],paraDict["ctbPrefix"],paraDict["ctbNum"],paraDict["rowsPerTbl"],paraDict["batchNum"])
         
         tdLog.info("insert process end, and start to check consume result")
         expectRows = 1
diff --git a/tests/system-test/7-tmq/tmqDropNtb-snapshot0.py b/tests/system-test/7-tmq/tmqDropNtb-snapshot0.py
new file mode 100644
index 0000000000..650d918828
--- /dev/null
+++ b/tests/system-test/7-tmq/tmqDropNtb-snapshot0.py
@@ -0,0 +1,225 @@
+
+import taos
+import sys
+import time
+import socket
+import os
+import threading
+from enum import Enum
+
+from util.log import *
+from util.sql import *
+from util.cases import *
+from util.dnodes import *
+sys.path.append("./7-tmq")
+from tmqCommon import *
+
+class TDTestCase:
+    def __init__(self):
+        self.snapshot   = 0
+        self.vgroups    = 4
+        self.ctbNum     = 1000
+        self.rowsPerTbl = 10
+        
+    def init(self, conn, logSql):
+        tdLog.debug(f"start to excute {__file__}")
+        tdSql.init(conn.cursor(), False)
+        
+    # drop some ntbs
+    def tmqCase1(self):
+        tdLog.printNoPrefix("======== test case 1: ")
+        paraDict = {'dbName':     'dbt',
+                    'dropFlag':   1,
+                    'event':      '',
+                    'vgroups':    4,
+                    'stbName':    'stb',
+                    'colPrefix':  'c',
+                    'tagPrefix':  't',
+                    'colSchema':   [{'type': 'INT', 'count':1},{'type': 'BIGINT', 'count':1},{'type': 'DOUBLE', 'count':1},{'type': 'TIMESTAMP', 'count':1}],
+                    'tagSchema':   [{'type': 'INT', 'count':1},{'type': 'BIGINT', 'count':1},{'type': 'DOUBLE', 'count':1},{'type': 'BINARY', 'len':32, 'count':1},{'type': 'NCHAR', 'len':32, 'count':1}],
+                    'ctbPrefix':  'ntb',
+                    'ctbStartIdx': 0,
+                    'ctbNum':     1000,
+                    'rowsPerTbl': 100,
+                    'batchNum':   100,
+                    'startTs':    1640966400000,  # 2022-01-01 00:00:00.000
+                    'endTs': 0,
+                    'pollDelay':  5,
+                    'showMsg':    1,
+                    'showRow':    1,
+                    'snapshot':   0}
+        paraDict['snapshot'] = self.snapshot
+        paraDict['vgroups'] = self.vgroups
+        paraDict['ctbNum'] = self.ctbNum
+        paraDict['rowsPerTbl'] = self.rowsPerTbl        
+        
+        tmqCom.initConsumerTable()
+        tdLog.info("start create database....")
+        tdCom.create_database(tdSql, paraDict["dbName"],paraDict["dropFlag"], vgroups=paraDict["vgroups"],replica=1)
+        tdLog.info("start create normal tables....")
+        tmqCom.create_ntable(tsql=tdSql, dbname=paraDict["dbName"], tbname_prefix=paraDict["ctbPrefix"], tbname_index_start_num = 1, column_elm_list=paraDict["colSchema"], colPrefix='c', tblNum=paraDict["ctbNum"])
+        tdLog.info("start insert data into normal tables....")
+        tmqCom.insert_rows_into_ntbl(tsql=tdSql, dbname=paraDict["dbName"], tbname_prefix=paraDict["ctbPrefix"], tbname_index_start_num = 1, column_ele_list=paraDict["colSchema"],startTs=paraDict["startTs"], tblNum=paraDict["ctbNum"], rows=paraDict["rowsPerTbl"])
+                
+        tdLog.info("create topics from database")
+        topicFromDb = 'topic_dbt'                
+        tdSql.execute("create topic %s as database %s" %(topicFromDb, paraDict['dbName']))
+        
+        if self.snapshot == 0:
+            consumerId     = 0
+        elif self.snapshot == 1:
+            consumerId     = 1
+
+        expectrowcnt   = int(paraDict["rowsPerTbl"] * paraDict["ctbNum"])
+        topicList      = topicFromDb
+        ifcheckdata    = 1
+        ifManualCommit = 1
+        keyList        = 'group.id:cgrp1,\
+                        enable.auto.commit:true,\
+                        auto.commit.interval.ms:1000,\
+                        auto.offset.reset:earliest'
+        tmqCom.insertConsumerInfo(consumerId, expectrowcnt,topicList,keyList,ifcheckdata,ifManualCommit)
+
+        tdLog.info("start consume processor")
+        tmqCom.startTmqSimProcess(pollDelay=paraDict['pollDelay'],dbName=paraDict["dbName"],showMsg=paraDict['showMsg'], showRow=paraDict['showRow'],snapshot=paraDict['snapshot'])
+
+        tmqCom.getStartConsumeNotifyFromTmqsim()  
+        tdLog.info("drop some ntables")
+        # drop 1/4 ctbls from half offset
+        paraDict["ctbStartIdx"] = paraDict["ctbStartIdx"] + int(paraDict["ctbNum"] * 1 / 2)
+        paraDict["ctbNum"] = int(paraDict["ctbNum"] / 4)
+        tmqCom.drop_ctable(tdSql, dbname=paraDict['dbName'], count=paraDict["ctbNum"], default_ctbname_prefix=paraDict["ctbPrefix"], ctbStartIdx=paraDict["ctbStartIdx"])
+        
+        tdLog.info("start to check consume result")
+        expectRows = 1
+        resultList = tmqCom.selectConsumeResult(expectRows)
+        totalConsumeRows = 0
+        for i in range(expectRows):
+            totalConsumeRows += resultList[i]
+
+        tdLog.info("act consume rows: %d, expect consume rows: %d"%(totalConsumeRows, expectrowcnt))
+        
+        if not ((totalConsumeRows >= expectrowcnt * 3/4) and (totalConsumeRows < expectrowcnt)):
+            tdLog.exit("tmq consume rows error with snapshot = 0!")
+        
+        tdLog.info("wait subscriptions exit ....")      
+        tmqCom.waitSubscriptionExit(tdSql, topicFromDb)
+            
+        tdSql.query("drop topic %s"%topicFromDb)
+        tdLog.info("success dorp topic: %s"%topicFromDb)
+        tdLog.printNoPrefix("======== test case 1 end ...... ")
+    
+
+    
+    # drop some ntbs and create some new ntbs
+    def tmqCase2(self):
+        tdLog.printNoPrefix("======== test case 2: ")
+        paraDict = {'dbName':     'dbt',
+                    'dropFlag':   1,
+                    'event':      '',
+                    'vgroups':    4,
+                    'stbName':    'stb',
+                    'colPrefix':  'c',
+                    'tagPrefix':  't',
+                    'colSchema':   [{'type': 'INT', 'count':1},{'type': 'BIGINT', 'count':1},{'type': 'DOUBLE', 'count':1},{'type': 'TIMESTAMP', 'count':1}],
+                    'tagSchema':   [{'type': 'INT', 'count':1},{'type': 'BIGINT', 'count':1},{'type': 'DOUBLE', 'count':1},{'type': 'BINARY', 'len':32, 'count':1},{'type': 'NCHAR', 'len':32, 'count':1}],
+                    'ctbPrefix':  'ntb',
+                    'ctbStartIdx': 0,
+                    'ctbNum':     1000,
+                    'rowsPerTbl': 100,
+                    'batchNum':   100,
+                    'startTs':    1640966400000,  # 2022-01-01 00:00:00.000
+                    'endTs': 0,
+                    'pollDelay':  10,
+                    'showMsg':    1,
+                    'showRow':    1,
+                    'snapshot':   0}
+        paraDict['snapshot'] = self.snapshot
+        paraDict['vgroups'] = self.vgroups
+        paraDict['ctbNum'] = self.ctbNum
+        paraDict['rowsPerTbl'] = self.rowsPerTbl        
+        
+        tmqCom.initConsumerTable()
+        tdLog.info("start create database....")
+        tdCom.create_database(tdSql, paraDict["dbName"],paraDict["dropFlag"], vgroups=paraDict["vgroups"],replica=1)
+        tdLog.info("start create normal tables....")
+        tmqCom.create_ntable(tsql=tdSql, dbname=paraDict["dbName"], tbname_prefix=paraDict["ctbPrefix"], tbname_index_start_num = 1, column_elm_list=paraDict["colSchema"], colPrefix='c', tblNum=paraDict["ctbNum"])
+        tdLog.info("start insert data into normal tables....")
+        tmqCom.insert_rows_into_ntbl(tsql=tdSql, dbname=paraDict["dbName"], tbname_prefix=paraDict["ctbPrefix"], tbname_index_start_num = 1, column_ele_list=paraDict["colSchema"],startTs=paraDict["startTs"], tblNum=paraDict["ctbNum"], rows=paraDict["rowsPerTbl"])
+                
+        tdLog.info("create topics from database")
+        topicFromDb = 'topic_dbt'                
+        tdSql.execute("create topic %s as database %s" %(topicFromDb, paraDict['dbName']))
+        
+        if self.snapshot == 0:
+            consumerId     = 2
+        elif self.snapshot == 1:
+            consumerId     = 3
+
+        expectrowcnt   = int(paraDict["rowsPerTbl"] * paraDict["ctbNum"] * 2)
+        topicList      = topicFromDb
+        ifcheckdata    = 1
+        ifManualCommit = 1
+        keyList        = 'group.id:cgrp1,\
+                        enable.auto.commit:true,\
+                        auto.commit.interval.ms:1000,\
+                        auto.offset.reset:earliest'
+        tmqCom.insertConsumerInfo(consumerId, expectrowcnt,topicList,keyList,ifcheckdata,ifManualCommit)
+
+        tdLog.info("start consume processor")
+        tmqCom.startTmqSimProcess(pollDelay=paraDict['pollDelay'],dbName=paraDict["dbName"],showMsg=paraDict['showMsg'], showRow=paraDict['showRow'],snapshot=paraDict['snapshot'])
+
+        tmqCom.getStartConsumeNotifyFromTmqsim()  
+        tdLog.info("drop some ntables")
+        # drop 1/4 ctbls from half offset
+        paraDict["ctbStartIdx"] = paraDict["ctbStartIdx"] + int(paraDict["ctbNum"] * 1 / 2)
+        paraDict["ctbNum"] = int(paraDict["ctbNum"] / 4)
+        tmqCom.drop_ctable(tdSql, dbname=paraDict['dbName'], count=paraDict["ctbNum"], default_ctbname_prefix=paraDict["ctbPrefix"], ctbStartIdx=paraDict["ctbStartIdx"])
+                
+        tdLog.info("start create some new normal tables....")
+        paraDict["ctbPrefix"] = 'newCtb'
+        paraDict["ctbNum"] = self.ctbNum
+        tmqCom.create_ntable(tsql=tdSql, dbname=paraDict["dbName"], tbname_prefix=paraDict["ctbPrefix"], tbname_index_start_num = 1, column_elm_list=paraDict["colSchema"], colPrefix='c', tblNum=paraDict["ctbNum"])
+        tdLog.info("start insert data into these new normal tables....")
+        tmqCom.insert_rows_into_ntbl(tsql=tdSql, dbname=paraDict["dbName"], tbname_prefix=paraDict["ctbPrefix"], tbname_index_start_num = 1, column_ele_list=paraDict["colSchema"],startTs=paraDict["startTs"], tblNum=paraDict["ctbNum"], rows=paraDict["rowsPerTbl"])
+        
+        tdLog.info("start to check consume result")
+        expectRows = 1
+        resultList = tmqCom.selectConsumeResult(expectRows)
+        totalConsumeRows = 0
+        for i in range(expectRows):
+            totalConsumeRows += resultList[i]
+
+        tdLog.info("act consume rows: %d, expect consume rows: %d"%(totalConsumeRows, expectrowcnt))
+        
+        if not ((totalConsumeRows >= expectrowcnt / 2 * (1 + 3/4)) and (totalConsumeRows < expectrowcnt)):
+            tdLog.exit("tmq consume rows error with snapshot = 0!")
+        
+        tdLog.info("wait subscriptions exit ....")      
+        tmqCom.waitSubscriptionExit(tdSql, topicFromDb)
+            
+        tdSql.query("drop topic %s"%topicFromDb)
+        tdLog.info("success dorp topic: %s"%topicFromDb)
+        tdLog.printNoPrefix("======== test case 2 end ...... ")
+        
+    def run(self):        
+        tdLog.printNoPrefix("=============================================")
+        tdLog.printNoPrefix("======== snapshot is 0: only consume from wal")
+        self.snapshot = 0
+        self.tmqCase1()
+        self.tmqCase2()               
+        
+        # tdLog.printNoPrefix("====================================================================")
+        # tdLog.printNoPrefix("======== snapshot is 1: firstly consume from tsbs, and then from wal")
+        # self.snapshot = 1
+        # self.tmqCase1()
+        # self.tmqCase2()
+
+    def stop(self):
+        tdSql.close()
+        tdLog.success(f"{__file__} successfully executed")
+
+event = threading.Event()
+
+tdCases.addLinux(__file__, TDTestCase())
+tdCases.addWindows(__file__, TDTestCase())
diff --git a/tests/system-test/7-tmq/tmqDropNtb.py b/tests/system-test/7-tmq/tmqDropNtb-snapshot1.py
similarity index 93%
rename from tests/system-test/7-tmq/tmqDropNtb.py
rename to tests/system-test/7-tmq/tmqDropNtb-snapshot1.py
index e1f5794ce2..b23f422585 100644
--- a/tests/system-test/7-tmq/tmqDropNtb.py
+++ b/tests/system-test/7-tmq/tmqDropNtb-snapshot1.py
@@ -18,7 +18,7 @@ class TDTestCase:
     def __init__(self):
         self.snapshot   = 0
         self.vgroups    = 4
-        self.ctbNum     = 100
+        self.ctbNum     = 1000
         self.rowsPerTbl = 10
         
     def init(self, conn, logSql):
@@ -39,9 +39,9 @@ class TDTestCase:
                     'tagSchema':   [{'type': 'INT', 'count':1},{'type': 'BIGINT', 'count':1},{'type': 'DOUBLE', 'count':1},{'type': 'BINARY', 'len':32, 'count':1},{'type': 'NCHAR', 'len':32, 'count':1}],
                     'ctbPrefix':  'ntb',
                     'ctbStartIdx': 0,
-                    'ctbNum':     100,
-                    'rowsPerTbl': 1000,
-                    'batchNum':   1000,
+                    'ctbNum':     1000,
+                    'rowsPerTbl': 100,
+                    'batchNum':   100,
                     'startTs':    1640966400000,  # 2022-01-01 00:00:00.000
                     'endTs': 0,
                     'pollDelay':  5,
@@ -125,9 +125,9 @@ class TDTestCase:
                     'tagSchema':   [{'type': 'INT', 'count':1},{'type': 'BIGINT', 'count':1},{'type': 'DOUBLE', 'count':1},{'type': 'BINARY', 'len':32, 'count':1},{'type': 'NCHAR', 'len':32, 'count':1}],
                     'ctbPrefix':  'ntb',
                     'ctbStartIdx': 0,
-                    'ctbNum':     100,
-                    'rowsPerTbl': 1000,
-                    'batchNum':   1000,
+                    'ctbNum':     1000,
+                    'rowsPerTbl': 100,
+                    'batchNum':   100,
                     'startTs':    1640966400000,  # 2022-01-01 00:00:00.000
                     'endTs': 0,
                     'pollDelay':  10,
@@ -203,16 +203,16 @@ class TDTestCase:
         tdLog.printNoPrefix("======== test case 2 end ...... ")
         
     def run(self):        
-        tdLog.printNoPrefix("=============================================")
-        tdLog.printNoPrefix("======== snapshot is 0: only consume from wal")
-        self.snapshot = 0
+        # tdLog.printNoPrefix("=============================================")
+        # tdLog.printNoPrefix("======== snapshot is 0: only consume from wal")
+        # self.snapshot = 0
         # self.tmqCase1()
-        self.tmqCase2()               
+        # self.tmqCase2()               
         
         tdLog.printNoPrefix("====================================================================")
         tdLog.printNoPrefix("======== snapshot is 1: firstly consume from tsbs, and then from wal")
         self.snapshot = 1
-        # self.tmqCase1()
+        self.tmqCase1()
         self.tmqCase2()
 
     def stop(self):
diff --git a/tests/system-test/fulltest.sh b/tests/system-test/fulltest.sh
index 5cc7aca675..f074bd8850 100755
--- a/tests/system-test/fulltest.sh
+++ b/tests/system-test/fulltest.sh
@@ -210,7 +210,7 @@ python3 ./test.py -f 7-tmq/tmqConsFromTsdb1-1ctb-funcNFilter.py
 python3 ./test.py -f 7-tmq/tmqConsFromTsdb1-mutilVg-mutilCtb-funcNFilter.py
 python3 ./test.py -f 7-tmq/tmqConsFromTsdb1-mutilVg-mutilCtb.py
 python3 ./test.py -f 7-tmq/tmqAutoCreateTbl.py
-#python3 ./test.py -f 7-tmq/tmqDnodeRestart.py
+python3 ./test.py -f 7-tmq/tmqDnodeRestart.py
 python3 ./test.py -f 7-tmq/tmqUpdate-1ctb.py
 python3 ./test.py -f 7-tmq/tmqUpdateWithConsume.py
 python3 ./test.py -f 7-tmq/tmqUpdate-multiCtb-snapshot0.py
@@ -219,12 +219,14 @@ python3 ./test.py -f 7-tmq/tmqDelete-1ctb.py
 python3 ./test.py -f 7-tmq/tmqDelete-multiCtb.py
 python3 ./test.py -f 7-tmq/tmqDropStb.py
 python3 ./test.py -f 7-tmq/tmqDropStbCtb.py
-python3 ./test.py -f 7-tmq/tmqDropNtb.py
+python3 ./test.py -f 7-tmq/tmqDropNtb-snapshot0.py
+python3 ./test.py -f 7-tmq/tmqDropNtb-snapshot1.py
 python3 ./test.py -f 7-tmq/tmqUdf.py
 python3 ./test.py -f 7-tmq/tmqUdf-multCtb-snapshot0.py
 python3 ./test.py -f 7-tmq/tmqUdf-multCtb-snapshot1.py
 python3 ./test.py -f 7-tmq/stbTagFilter-1ctb.py
-
+python3 ./test.py -f 7-tmq/dataFromTsdbNWal.py
+python3 ./test.py -f 7-tmq/dataFromTsdbNWal-multiCtb.py
 # python3 ./test.py -f 7-tmq/stbTagFilter-multiCtb.py
 
 #------------querPolicy  2-----------
diff --git a/tools/taos-tools b/tools/taos-tools
index 9cfa195713..0b8a3373bb 160000
--- a/tools/taos-tools
+++ b/tools/taos-tools
@@ -1 +1 @@
-Subproject commit 9cfa195713d1cae9edf417a8d49bde87dd971016
+Subproject commit 0b8a3373bb7548f8106d13e7d3b0a988d3c4d48a