diff --git a/include/libs/nodes/plannodes.h b/include/libs/nodes/plannodes.h
index 453c5d49142934b79758f3faf5e3c12c80c16a55..821defc7972866510a35a2dd925cb75ff1ebd6ef 100644
--- a/include/libs/nodes/plannodes.h
+++ b/include/libs/nodes/plannodes.h
@@ -55,6 +55,7 @@ typedef struct SLogicNode {
   EGroupAction       groupAction;
   EOrder             inputTsOrder;
   EOrder             outputTsOrder;
+  bool               forceCreateNonBlockingOptr;  // true if the operator can use non-blocking(pipeline) mode
 } SLogicNode;
 
 typedef enum EScanType {
@@ -105,6 +106,7 @@ typedef struct SScanLogicNode {
   bool          hasNormalCols;  // neither tag column nor primary key tag column
   bool          sortPrimaryKey;
   bool          igLastNull;
+  bool          groupOrderScan;
 } SScanLogicNode;
 
 typedef struct SJoinLogicNode {
@@ -316,6 +318,7 @@ typedef struct SPhysiNode {
   struct SPhysiNode*  pParent;
   SNode*              pLimit;
   SNode*              pSlimit;
+  bool                forceCreateNonBlockingOptr;
 } SPhysiNode;
 
 typedef struct SScanPhysiNode {
@@ -326,6 +329,7 @@ typedef struct SScanPhysiNode {
   uint64_t   suid;
   int8_t     tableType;
   SName      tableName;
+  bool       groupOrderScan;
 } SScanPhysiNode;
 
 typedef SScanPhysiNode STagScanPhysiNode;
diff --git a/source/common/src/tdatablock.c b/source/common/src/tdatablock.c
index b2f03fa7ba54606a924214f98559c818bbef3ac9..887a11083153c0daa3bf6f06a36ae49d6c65dab5 100644
--- a/source/common/src/tdatablock.c
+++ b/source/common/src/tdatablock.c
@@ -632,7 +632,10 @@ int32_t blockDataToBuf(char* buf, const SSDataBlock* pBlock) {
         pStart += colSize;
       }
     } else {
-      memcpy(pStart, pCol->pData, dataSize);
+      if (dataSize != 0) {
+        // ubsan reports error if pCol->pData==NULL && dataSize==0
+        memcpy(pStart, pCol->pData, dataSize);
+      }
       pStart += dataSize;
     }
   }
@@ -684,8 +687,10 @@ int32_t blockDataFromBuf(SSDataBlock* pBlock, const char* buf) {
         return TSDB_CODE_FAILED;
       }
     }
-
-    memcpy(pCol->pData, pStart, colLength);
+    if (colLength != 0) {
+      // ubsan reports error if colLength==0 && pCol->pData == 0
+      memcpy(pCol->pData, pStart, colLength);
+    }
     pStart += colLength;
   }
 
diff --git a/source/dnode/mnode/impl/inc/mndStb.h b/source/dnode/mnode/impl/inc/mndStb.h
index 99af413539b850eb4f62808fdf7899ab679933f0..db960d790f708e76ce2921f4ad6b6fefba82e441 100644
--- a/source/dnode/mnode/impl/inc/mndStb.h
+++ b/source/dnode/mnode/impl/inc/mndStb.h
@@ -39,6 +39,7 @@ int32_t mndBuildSMCreateStbRsp(SMnode *pMnode, char *dbFName, char *stbFName, vo
 
 void mndExtractDbNameFromStbFullName(const char *stbFullName, char *dst);
 void mndExtractShortDbNameFromStbFullName(const char *stbFullName, char *dst);
+void mndExtractShortDbNameFromDbFullName(const char *stbFullName, char *dst);
 void mndExtractTbNameFromStbFullName(const char *stbFullName, char *dst, int32_t dstSize);
 
 const char *mndGetStbStr(const char *src);
diff --git a/source/dnode/mnode/impl/src/mndStb.c b/source/dnode/mnode/impl/src/mndStb.c
index d6537ef9926cd01bffd8f2ed985e9d3c7fff7d7f..fa104d06ff5c0393ed3535b2cf0d36570b7b2dc9 100644
--- a/source/dnode/mnode/impl/src/mndStb.c
+++ b/source/dnode/mnode/impl/src/mndStb.c
@@ -2498,12 +2498,14 @@ static int32_t mndProcessTableCfgReq(SRpcMsg *pReq) {
     goto _OVER;
   }
 
-  if (0 == strcmp(cfgReq.dbFName, TSDB_INFORMATION_SCHEMA_DB)) {
+  char dbName[TSDB_DB_NAME_LEN] = {0};
+  mndExtractShortDbNameFromDbFullName(cfgReq.dbFName, dbName);
+  if (0 == strcmp(dbName, TSDB_INFORMATION_SCHEMA_DB)) {
     mInfo("information_schema table:%s.%s, start to retrieve cfg", cfgReq.dbFName, cfgReq.tbName);
     if (mndBuildInsTableCfg(pMnode, cfgReq.dbFName, cfgReq.tbName, &cfgRsp) != 0) {
       goto _OVER;
     }
-  } else if (0 == strcmp(cfgReq.dbFName, TSDB_PERFORMANCE_SCHEMA_DB)) {
+  } else if (0 == strcmp(dbName, TSDB_PERFORMANCE_SCHEMA_DB)) {
     mInfo("performance_schema table:%s.%s, start to retrieve cfg", cfgReq.dbFName, cfgReq.tbName);
     if (mndBuildPerfsTableCfg(pMnode, cfgReq.dbFName, cfgReq.tbName, &cfgRsp) != 0) {
       goto _OVER;
@@ -2672,6 +2674,13 @@ void mndExtractShortDbNameFromStbFullName(const char *stbFullName, char *dst) {
   tNameGetDbName(&name, dst);
 }
 
+void mndExtractShortDbNameFromDbFullName(const char *stbFullName, char *dst) {
+  SName name = {0};
+  tNameFromString(&name, stbFullName, T_NAME_ACCT | T_NAME_DB);
+
+  tNameGetDbName(&name, dst);
+}
+
 void mndExtractTbNameFromStbFullName(const char *stbFullName, char *dst, int32_t dstSize) {
   int32_t pos = -1;
   int32_t num = 0;
diff --git a/source/libs/catalog/src/ctgAsync.c b/source/libs/catalog/src/ctgAsync.c
index 562343c9c70b4b8a38c866069e897535c5fdfddd..784126eee941a4c9cd0af7898a58e1b6d5d4648b 100644
--- a/source/libs/catalog/src/ctgAsync.c
+++ b/source/libs/catalog/src/ctgAsync.c
@@ -2090,7 +2090,7 @@ int32_t ctgLaunchGetTbCfgTask(SCtgTask* pTask) {
   }
 
   CTG_CACHE_NHIT_INC(CTG_CI_TBL_CFG, 1);
-  
+
   if (pCtx->tbType <= 0) {
     CTG_ERR_JRET(ctgReadTbTypeFromCache(pCtg, dbFName, pCtx->pName->tname, &pCtx->tbType));
     if (pCtx->tbType <= 0) {
@@ -2102,7 +2102,7 @@ int32_t ctgLaunchGetTbCfgTask(SCtgTask* pTask) {
     }
   }
 
-  if (TSDB_SUPER_TABLE == pCtx->tbType) {
+  if (TSDB_SUPER_TABLE == pCtx->tbType || TSDB_SYSTEM_TABLE == pCtx->tbType) {
     CTG_ERR_JRET(ctgGetTableCfgFromMnode(pCtg, pConn, pCtx->pName, NULL, pTask));
   } else {
     if (NULL == pCtx->pVgInfo) {
diff --git a/source/libs/executor/inc/executorInt.h b/source/libs/executor/inc/executorInt.h
index 9c41eb2bf08d3b81c0f86ce46b053a83615d586a..5baf0978cd84a79ab77dd935f2e35bcfa70efa9a 100644
--- a/source/libs/executor/inc/executorInt.h
+++ b/source/libs/executor/inc/executorInt.h
@@ -232,19 +232,20 @@ typedef struct STableMergeScanInfo {
   int32_t         tableEndIndex;
   bool            hasGroupId;
   uint64_t        groupId;
-  SArray*         queryConds;  // array of queryTableDataCond
   STableScanBase  base;
   int32_t         bufPageSize;
   uint32_t        sortBufSize;  // max buffer size for in-memory sort
   SArray*         pSortInfo;
   SSortHandle*    pSortHandle;
   SSDataBlock*    pSortInputBlock;
+  SSDataBlock*    pReaderBlock;
   int64_t         startTs;  // sort start time
   SArray*         sortSourceParams;
   SLimitInfo      limitInfo;
   int64_t         numOfRows;
   SScanInfo       scanInfo;
   int32_t         scanTimes;
+  int32_t         readIdx;
   SSDataBlock*    pResBlock;
   SSampleExecInfo sample;  // sample execution info
   SSortExecInfo   sortExecInfo;
diff --git a/source/libs/executor/inc/tsort.h b/source/libs/executor/inc/tsort.h
index 627aa825c671ee0d87700f62b00e6b3c9fdfef4d..538a9f18f6012a110b52fa8e2d899f551dd88597 100644
--- a/source/libs/executor/inc/tsort.h
+++ b/source/libs/executor/inc/tsort.h
@@ -26,6 +26,7 @@ extern "C" {
 enum {
   SORT_MULTISOURCE_MERGE = 0x1,
   SORT_SINGLESOURCE_SORT = 0x2,
+  SORT_BLOCK_TS_MERGE = 0x3
 };
 
 typedef struct SMultiMergeSource {
@@ -53,6 +54,12 @@ typedef struct SMsortComparParam {
   int32_t numOfSources;
   SArray* orderInfo;  // SArray<SBlockOrderInfo>
   bool    cmpGroupId;
+
+  int32_t sortType;
+  // the following field to speed up when sortType == SORT_BLOCK_TS_MERGE
+  int32_t tsSlotId;
+  int32_t order;
+  __compar_fn_t cmpFn;
 } SMsortComparParam;
 
 typedef struct SSortHandle  SSortHandle;
@@ -70,8 +77,8 @@ typedef int32_t (*_sort_merge_compar_fn_t)(const void* p1, const void* p2, void*
  * @return
  */
 SSortHandle* tsortCreateSortHandle(SArray* pOrderInfo, int32_t type, int32_t pageSize, int32_t numOfPages,
-                                   SSDataBlock* pBlock, const char* idstr, uint64_t maxRows, uint32_t maxTupleLength,
-                                   uint32_t sortBufSize);
+                                   SSDataBlock* pBlock, const char* idstr, uint64_t pqMaxRows, uint32_t pqMaxTupleLength,
+                                   uint32_t pqSortBufSize);
 
 void tsortSetForceUsePQSort(SSortHandle* pHandle);
 
@@ -110,6 +117,10 @@ int32_t tsortSetFetchRawDataFp(SSortHandle* pHandle, _sort_fetch_block_fn_t fetc
  */
 int32_t tsortSetComparFp(SSortHandle* pHandle, _sort_merge_compar_fn_t fp);
 
+/**
+ * 
+*/
+void tsortSetMergeLimit(SSortHandle* pHandle, int64_t mergeLimit);
 /**
  *
  */
diff --git a/source/libs/executor/src/aggregateoperator.c b/source/libs/executor/src/aggregateoperator.c
index be0ad1c2399c1b423bf6a09635d6334581c321fe..176c4b53be828a0ad356953793f722cc39d43893 100644
--- a/source/libs/executor/src/aggregateoperator.c
+++ b/source/libs/executor/src/aggregateoperator.c
@@ -45,6 +45,8 @@ typedef struct SAggOperatorInfo {
   SGroupResInfo    groupResInfo;
   SExprSupp        scalarExprSup;
   bool             groupKeyOptimized;
+  bool             hasValidBlock;
+  SSDataBlock*     pNewGroupBlock;
 } SAggOperatorInfo;
 
 static void destroyAggOperatorInfo(void* param);
@@ -53,7 +55,6 @@ static void setExecutionContext(SOperatorInfo* pOperator, int32_t numOfOutput, u
 static int32_t createDataBlockForEmptyInput(SOperatorInfo* pOperator, SSDataBlock** ppBlock);
 static void destroyDataBlockForEmptyInput(bool blockAllocated, SSDataBlock** ppBlock);
 
-static int32_t doOpenAggregateOptr(SOperatorInfo* pOperator);
 static int32_t doAggregateImpl(SOperatorInfo* pOperator, SqlFunctionCtx* pCtx);
 static SSDataBlock* getAggregateResult(SOperatorInfo* pOperator);
 
@@ -111,9 +112,9 @@ SOperatorInfo* createAggregateOperatorInfo(SOperatorInfo* downstream, SAggPhysiN
   pInfo->binfo.inputTsOrder = pAggNode->node.inputTsOrder;
   pInfo->binfo.outputTsOrder = pAggNode->node.outputTsOrder;
 
-  setOperatorInfo(pOperator, "TableAggregate", QUERY_NODE_PHYSICAL_PLAN_HASH_AGG, true, OP_NOT_OPENED, pInfo,
-                  pTaskInfo);
-  pOperator->fpSet = createOperatorFpSet(doOpenAggregateOptr, getAggregateResult, NULL, destroyAggOperatorInfo,
+  setOperatorInfo(pOperator, "TableAggregate", QUERY_NODE_PHYSICAL_PLAN_HASH_AGG,
+                  !pAggNode->node.forceCreateNonBlockingOptr, OP_NOT_OPENED, pInfo, pTaskInfo);
+  pOperator->fpSet = createOperatorFpSet(optrDummyOpenFn, getAggregateResult, NULL, destroyAggOperatorInfo,
                                          optrDefaultBufFn, NULL);
 
   if (downstream->operatorType == QUERY_NODE_PHYSICAL_PLAN_TABLE_SCAN) {
@@ -153,28 +154,42 @@ void destroyAggOperatorInfo(void* param) {
   taosMemoryFreeClear(param);
 }
 
-// this is a blocking operator
-int32_t doOpenAggregateOptr(SOperatorInfo* pOperator) {
-  if (OPTR_IS_OPENED(pOperator)) {
-    return TSDB_CODE_SUCCESS;
-  }
-
+/**
+ * @brief get blocks from downstream and fill results into groupedRes after aggragation
+ * @retval false if no more groups
+ * @retval true if there could have new groups coming
+ * @note if pOperator.blocking is true, scan all blocks from downstream, all groups are handled
+ *       if false, fill results of ONE GROUP
+ * */
+static bool nextGroupedResult(SOperatorInfo* pOperator) {
   SExecTaskInfo*    pTaskInfo = pOperator->pTaskInfo;
   SAggOperatorInfo* pAggInfo = pOperator->info;
 
+  if (pOperator->blocking && pAggInfo->hasValidBlock) return false;
+
   SExprSupp*     pSup = &pOperator->exprSupp;
   SOperatorInfo* downstream = pOperator->pDownstream[0];
 
-  int64_t st = taosGetTimestampUs();
-  int32_t code = TSDB_CODE_SUCCESS;
-  int32_t order = pAggInfo->binfo.inputTsOrder;
-  bool hasValidBlock = false;
+  int64_t      st = taosGetTimestampUs();
+  int32_t      code = TSDB_CODE_SUCCESS;
+  int32_t      order = pAggInfo->binfo.inputTsOrder;
+  SSDataBlock* pBlock = pAggInfo->pNewGroupBlock;
 
+  if (pBlock) {
+    pAggInfo->pNewGroupBlock = NULL;
+    tSimpleHashClear(pAggInfo->aggSup.pResultRowHashTable);
+    setExecutionContext(pOperator, pOperator->exprSupp.numOfExprs, pBlock->info.id.groupId);
+    setInputDataBlock(pSup, pBlock, order, pBlock->info.scanFlag, true);
+    code = doAggregateImpl(pOperator, pSup->pCtx);
+    if (code != TSDB_CODE_SUCCESS) {
+      T_LONG_JMP(pTaskInfo->env, code);
+    }
+  }
   while (1) {
     bool blockAllocated = false;
-    SSDataBlock* pBlock = downstream->fpSet.getNextFn(downstream);
+    pBlock = downstream->fpSet.getNextFn(downstream);
     if (pBlock == NULL) {
-      if (!hasValidBlock) {
+      if (!pAggInfo->hasValidBlock) {
         createDataBlockForEmptyInput(pOperator, &pBlock);
         if (pBlock == NULL) {
           break;
@@ -184,7 +199,7 @@ int32_t doOpenAggregateOptr(SOperatorInfo* pOperator) {
         break;
       }
     }
-    hasValidBlock = true;
+    pAggInfo->hasValidBlock = true;
     pAggInfo->binfo.pRes->info.scanFlag = pBlock->info.scanFlag;
 
     // there is an scalar expression that needs to be calculated before apply the group aggregation.
@@ -196,7 +211,11 @@ int32_t doOpenAggregateOptr(SOperatorInfo* pOperator) {
         T_LONG_JMP(pTaskInfo->env, code);
       }
     }
-
+    // if non-blocking mode and new group arrived, save the block and break
+    if (!pOperator->blocking && pAggInfo->groupId != UINT64_MAX && pBlock->info.id.groupId != pAggInfo->groupId) {
+      pAggInfo->pNewGroupBlock = pBlock;
+      break;
+    }
     // the pDataBlock are always the same one, no need to call this again
     setExecutionContext(pOperator, pOperator->exprSupp.numOfExprs, pBlock->info.id.groupId);
     setInputDataBlock(pSup, pBlock, order, pBlock->info.scanFlag, true);
@@ -215,10 +234,7 @@ int32_t doOpenAggregateOptr(SOperatorInfo* pOperator) {
   }
 
   initGroupedResultInfo(&pAggInfo->groupResInfo, pAggInfo->aggSup.pResultRowHashTable, 0);
-  OPTR_SET_OPENED(pOperator);
-
-  pOperator->cost.openCost = (taosGetTimestampUs() - st) / 1000.0;
-  return pTaskInfo->code;
+  return pBlock != NULL;
 }
 
 SSDataBlock* getAggregateResult(SOperatorInfo* pOperator) {
@@ -230,26 +246,25 @@ SSDataBlock* getAggregateResult(SOperatorInfo* pOperator) {
   }
 
   SExecTaskInfo* pTaskInfo = pOperator->pTaskInfo;
-  pTaskInfo->code = pOperator->fpSet._openFn(pOperator);
-  if (pTaskInfo->code != TSDB_CODE_SUCCESS) {
-    setOperatorCompleted(pOperator);
-    return NULL;
-  }
+  bool hasNewGroups = false;
+  do {
+    hasNewGroups = nextGroupedResult(pOperator);
+    blockDataEnsureCapacity(pInfo->pRes, pOperator->resultInfo.capacity);
 
-  blockDataEnsureCapacity(pInfo->pRes, pOperator->resultInfo.capacity);
-  while (1) {
-    doBuildResultDatablock(pOperator, pInfo, &pAggInfo->groupResInfo, pAggInfo->aggSup.pResultBuf);
-    doFilter(pInfo->pRes, pOperator->exprSupp.pFilterInfo, NULL);
+    while (1) {
+      doBuildResultDatablock(pOperator, pInfo, &pAggInfo->groupResInfo, pAggInfo->aggSup.pResultBuf);
+      doFilter(pInfo->pRes, pOperator->exprSupp.pFilterInfo, NULL);
 
-    if (!hasRemainResults(&pAggInfo->groupResInfo)) {
-      setOperatorCompleted(pOperator);
-      break;
-    }
+      if (!hasRemainResults(&pAggInfo->groupResInfo)) {
+        if (!hasNewGroups) setOperatorCompleted(pOperator);
+        break;
+      }
 
-    if (pInfo->pRes->info.rows > 0) {
-      break;
+      if (pInfo->pRes->info.rows > 0) {
+        break;
+      }
     }
-  }
+  } while (pInfo->pRes->info.rows == 0 && hasNewGroups);
 
   size_t rows = blockDataGetNumOfRows(pInfo->pRes);
   pOperator->resultInfo.totalRows += rows;
diff --git a/source/libs/executor/src/executil.c b/source/libs/executor/src/executil.c
index 832750e967dfd6f996d12281a9ece48b6c6d26c7..e1bf4e7cb0e3b3e2df8cc7931d57c153a952b78b 100644
--- a/source/libs/executor/src/executil.c
+++ b/source/libs/executor/src/executil.c
@@ -127,6 +127,10 @@ void initGroupedResultInfo(SGroupResInfo* pGroupResInfo, SSHashObj* pHashmap, in
   if (pGroupResInfo->pRows != NULL) {
     taosArrayDestroy(pGroupResInfo->pRows);
   }
+  if (pGroupResInfo->pBuf) {
+    taosMemoryFree(pGroupResInfo->pBuf);
+    pGroupResInfo->pBuf = NULL;
+  }
 
   // extract the result rows information from the hash map
   int32_t size = tSimpleHashGetSize(pHashmap);
@@ -2104,6 +2108,8 @@ int32_t buildGroupIdMapForAllTables(STableListInfo* pTableListInfo, SReadHandle*
     if (groupSort && groupByTbname) {
       taosArraySort(pTableListInfo->pTableList, orderbyGroupIdComparFn);
       pTableListInfo->numOfOuputGroups = numOfTables;
+    } else if (groupByTbname && pScanNode->groupOrderScan){
+      pTableListInfo->numOfOuputGroups = numOfTables;
     } else {
       pTableListInfo->numOfOuputGroups = 1;
     }
diff --git a/source/libs/executor/src/scanoperator.c b/source/libs/executor/src/scanoperator.c
index f1de80ec4cf92adb3bb06646c0787816252308b9..810f9709f5887d9bb526eeccfdd07e10dc782e9d 100644
--- a/source/libs/executor/src/scanoperator.c
+++ b/source/libs/executor/src/scanoperator.c
@@ -13,8 +13,6 @@
  * along with this program. If not, see <http://www.gnu.org/licenses/>.
  */
 
-// clang-format off
-
 #include "executorInt.h"
 #include "filter.h"
 #include "function.h"
@@ -55,8 +53,7 @@ typedef struct STableMergeScanSortSourceParam {
   SOperatorInfo* pOperator;
   int32_t        readerIdx;
   uint64_t       uid;
-  SSDataBlock*   inputBlock;
-  STsdbReader*   dataReader;
+  STsdbReader*   reader;
 } STableMergeScanSortSourceParam;
 
 typedef struct STableCountScanOperatorInfo {
@@ -2734,32 +2731,17 @@ static SSDataBlock* getTableDataBlockImpl(void* param) {
   SExecTaskInfo*                  pTaskInfo = pOperator->pTaskInfo;
   SStorageAPI* pAPI = &pTaskInfo->storageAPI;
 
-  int32_t                         readIdx = source->readerIdx;
-  SSDataBlock*                    pBlock = source->inputBlock;
+  SSDataBlock*                    pBlock = pInfo->pReaderBlock;
   int32_t                         code = 0;
 
-  SQueryTableDataCond* pQueryCond = taosArrayGet(pInfo->queryConds, readIdx);
-
   int64_t      st = taosGetTimestampUs();
-  void*        p = tableListGetInfo(pInfo->base.pTableListInfo, readIdx + pInfo->tableStartIndex);
-  SReadHandle* pHandle = &pInfo->base.readHandle;
-  if (NULL == source->dataReader) {
-    code = pAPI->tsdReader.tsdReaderOpen(pHandle->vnode, pQueryCond, p, 1, pBlock, (void**)&source->dataReader, GET_TASKID(pTaskInfo), false, NULL);
-    if (code != 0) {
-      T_LONG_JMP(pTaskInfo->env, code);
-    }
-  }
-
-  pInfo->base.dataReader = source->dataReader;
-  STsdbReader* reader = pInfo->base.dataReader;
   bool         hasNext = false;
-  qTrace("tsdb/read-table-data: %p, enter next reader", reader);
 
+  STsdbReader* reader = pInfo->base.dataReader;
   while (true) {
     code = pAPI->tsdReader.tsdNextDataBlock(reader, &hasNext);
     if (code != 0) {
       pAPI->tsdReader.tsdReaderReleaseDataBlock(reader);
-      pInfo->base.dataReader = NULL;
       T_LONG_JMP(pTaskInfo->env, code);
     }
 
@@ -2769,7 +2751,6 @@ static SSDataBlock* getTableDataBlockImpl(void* param) {
 
     if (isTaskKilled(pTaskInfo)) {
       pAPI->tsdReader.tsdReaderReleaseDataBlock(reader);
-      pInfo->base.dataReader = NULL;
       T_LONG_JMP(pTaskInfo->env, pTaskInfo->code);
     }
 
@@ -2779,12 +2760,6 @@ static SSDataBlock* getTableDataBlockImpl(void* param) {
       continue;
     }
 
-    if (pQueryCond->order == TSDB_ORDER_ASC) {
-      pQueryCond->twindows.skey = pBlock->info.window.ekey + 1;
-    } else {
-      pQueryCond->twindows.ekey = pBlock->info.window.skey - 1;
-    }
-
     uint32_t status = 0;
     code = loadDataBlock(pOperator, &pInfo->base, pBlock, &status);
     //    code = loadDataBlockFromOneTable(pOperator, pTableScanInfo, pBlock, &status);
@@ -2806,16 +2781,9 @@ static SSDataBlock* getTableDataBlockImpl(void* param) {
     pOperator->resultInfo.totalRows += pBlock->info.rows;
     pInfo->base.readRecorder.elapsedTime += (taosGetTimestampUs() - st) / 1000.0;
 
-    qTrace("tsdb/read-table-data: %p, close reader", reader);
-    pInfo->base.dataReader = NULL;
     return pBlock;
   }
 
-  pAPI->tsdReader.tsdReaderClose(source->dataReader);
-  source->dataReader = NULL;
-  pInfo->base.dataReader = NULL;
-  blockDataDestroy(source->inputBlock);
-  source->inputBlock = NULL;
   return NULL;
 }
 
@@ -2851,6 +2819,8 @@ int32_t dumpQueryTableCond(const SQueryTableDataCond* src, SQueryTableDataCond*
 int32_t startGroupTableMergeScan(SOperatorInfo* pOperator) {
   STableMergeScanInfo* pInfo = pOperator->info;
   SExecTaskInfo*       pTaskInfo = pOperator->pTaskInfo;
+  SReadHandle* pHandle = &pInfo->base.readHandle;
+  SStorageAPI* pAPI = &pTaskInfo->storageAPI;
 
   {
     size_t  numOfTables = tableListGetSize(pInfo->base.pTableListInfo);
@@ -2867,53 +2837,29 @@ int32_t startGroupTableMergeScan(SOperatorInfo* pOperator) {
   int32_t tableStartIdx = pInfo->tableStartIndex;
   int32_t tableEndIdx = pInfo->tableEndIndex;
 
-  pInfo->base.dataReader = NULL;
-
-  // todo the total available buffer should be determined by total capacity of buffer of this task.
-  // the additional one is reserved for merge result
-  // pInfo->sortBufSize = pInfo->bufPageSize * (tableEndIdx - tableStartIdx + 1 + 1);
-  int32_t kWay = (TSDB_MAX_BYTES_PER_ROW * 2) / (pInfo->pResBlock->info.rowSize);
-  if (kWay >= 128) {
-    kWay = 128;
-  } else if (kWay <= 2) {
-    kWay = 2;
-  } else {
-    int i = 2;
-    while (i * 2 <= kWay) i = i * 2;
-    kWay = i;
-  }
-
-  pInfo->sortBufSize = pInfo->bufPageSize * (kWay + 1);
+  pInfo->sortBufSize = 2048 * pInfo->bufPageSize;
   int32_t numOfBufPage = pInfo->sortBufSize / pInfo->bufPageSize;
-  pInfo->pSortHandle = tsortCreateSortHandle(pInfo->pSortInfo, SORT_MULTISOURCE_MERGE, pInfo->bufPageSize, numOfBufPage,
+  pInfo->pSortHandle = tsortCreateSortHandle(pInfo->pSortInfo, SORT_BLOCK_TS_MERGE, pInfo->bufPageSize, numOfBufPage,
                                              pInfo->pSortInputBlock, pTaskInfo->id.str, 0, 0, 0);
-
+  int64_t mergeLimit = -1;
+  if (pInfo->limitInfo.limit.limit != -1 || pInfo->limitInfo.limit.offset != -1) {
+    mergeLimit = pInfo->limitInfo.limit.limit + pInfo->limitInfo.limit.offset;
+  }                                             
+  tsortSetMergeLimit(pInfo->pSortHandle, mergeLimit);
   tsortSetFetchRawDataFp(pInfo->pSortHandle, getTableDataBlockImpl, NULL, NULL);
 
   // one table has one data block
   int32_t numOfTable = tableEndIdx - tableStartIdx + 1;
-  pInfo->queryConds = taosArrayInit(numOfTable, sizeof(SQueryTableDataCond));
-
-  for (int32_t i = 0; i < numOfTable; ++i) {
-    STableMergeScanSortSourceParam param = {0};
-    param.readerIdx = i;
-    param.pOperator = pOperator;
-    param.inputBlock = createOneDataBlock(pInfo->pResBlock, false);
 
-    taosArrayPush(pInfo->sortSourceParams, &param);
+  STableMergeScanSortSourceParam param = {0};
+  param.pOperator = pOperator;
+  STableKeyInfo* startKeyInfo = tableListGetInfo(pInfo->base.pTableListInfo, tableStartIdx);
+  pAPI->tsdReader.tsdReaderOpen(pHandle->vnode, &pInfo->base.cond, startKeyInfo, numOfTable, pInfo->pReaderBlock, (void**)&pInfo->base.dataReader, GET_TASKID(pTaskInfo), false, NULL);
 
-    SQueryTableDataCond cond;
-    dumpQueryTableCond(&pInfo->base.cond, &cond);
-    taosArrayPush(pInfo->queryConds, &cond);
-  }
-
-  for (int32_t i = 0; i < numOfTable; ++i) {
-    SSortSource*                    ps = taosMemoryCalloc(1, sizeof(SSortSource));
-    STableMergeScanSortSourceParam* param = taosArrayGet(pInfo->sortSourceParams, i);
-    ps->param = param;
-    ps->onlyRef = true;
-    tsortAddSource(pInfo->pSortHandle, ps);
-  }
+  SSortSource* ps = taosMemoryCalloc(1, sizeof(SSortSource));
+  ps->param = &param;
+  ps->onlyRef = true;
+  tsortAddSource(pInfo->pSortHandle, ps);
 
   int32_t code = tsortOpen(pInfo->pSortHandle);
 
@@ -2929,8 +2875,6 @@ int32_t stopGroupTableMergeScan(SOperatorInfo* pOperator) {
   SExecTaskInfo*       pTaskInfo = pOperator->pTaskInfo;
   SStorageAPI*         pAPI = &pTaskInfo->storageAPI;
 
-  int32_t numOfTable = taosArrayGetSize(pInfo->queryConds);
-
   SSortExecInfo sortExecInfo = tsortGetSortExecInfo(pInfo->pSortHandle);
   pInfo->sortExecInfo.sortMethod = sortExecInfo.sortMethod;
   pInfo->sortExecInfo.sortBuffer = sortExecInfo.sortBuffer;
@@ -2938,24 +2882,14 @@ int32_t stopGroupTableMergeScan(SOperatorInfo* pOperator) {
   pInfo->sortExecInfo.readBytes += sortExecInfo.readBytes;
   pInfo->sortExecInfo.writeBytes += sortExecInfo.writeBytes;
 
-  for (int32_t i = 0; i < numOfTable; ++i) {
-    STableMergeScanSortSourceParam* param = taosArrayGet(pInfo->sortSourceParams, i);
-    blockDataDestroy(param->inputBlock);
-    pAPI->tsdReader.tsdReaderClose(param->dataReader);
-    param->dataReader = NULL;
+  if (pInfo->base.dataReader != NULL) {
+    pAPI->tsdReader.tsdReaderClose(pInfo->base.dataReader);
+    pInfo->base.dataReader = NULL;
   }
-  taosArrayClear(pInfo->sortSourceParams);
 
   tsortDestroySortHandle(pInfo->pSortHandle);
   pInfo->pSortHandle = NULL;
 
-  for (int32_t i = 0; i < taosArrayGetSize(pInfo->queryConds); i++) {
-    SQueryTableDataCond* cond = taosArrayGet(pInfo->queryConds, i);
-    taosMemoryFree(cond->colList);
-  }
-  taosArrayDestroy(pInfo->queryConds);
-  pInfo->queryConds = NULL;
-
   resetLimitInfoForNextGroup(&pInfo->limitInfo);
   return TSDB_CODE_SUCCESS;
 }
@@ -2968,28 +2902,32 @@ SSDataBlock* getSortedTableMergeScanBlockData(SSortHandle* pHandle, SSDataBlock*
   SExecTaskInfo*       pTaskInfo = pOperator->pTaskInfo;
 
   blockDataCleanup(pResBlock);
-
+  STupleHandle* pTupleHandle = NULL;
   while (1) {
-    STupleHandle* pTupleHandle = tsortNextTuple(pHandle);
-    if (pTupleHandle == NULL) {
-      break;
+    while (1) {
+      pTupleHandle = tsortNextTuple(pHandle);
+      if (pTupleHandle == NULL) {
+        break;
+      }
+
+      appendOneRowToDataBlock(pResBlock, pTupleHandle);
+      if (pResBlock->info.rows >= capacity) {
+        break;
+      }
     }
 
-    appendOneRowToDataBlock(pResBlock, pTupleHandle);
-    if (pResBlock->info.rows >= capacity) {
-      break;
+    if (tsortIsClosed(pHandle)) {
+      terrno = TSDB_CODE_TSC_QUERY_CANCELLED;
+      T_LONG_JMP(pOperator->pTaskInfo->env, terrno);
     }
-  }
 
-  if (tsortIsClosed(pHandle)) {
-    terrno = TSDB_CODE_TSC_QUERY_CANCELLED;
-    T_LONG_JMP(pOperator->pTaskInfo->env, terrno);
+    bool limitReached = applyLimitOffset(&pInfo->limitInfo, pResBlock, pTaskInfo);
+    qDebug("%s get sorted row block, rows:%" PRId64 ", limit:%" PRId64, GET_TASKID(pTaskInfo), pResBlock->info.rows,
+          pInfo->limitInfo.numOfOutputRows);
+    if (pTupleHandle == NULL || limitReached || pResBlock->info.rows > 0) {
+      break;
+    }  
   }
-
-  bool limitReached = applyLimitOffset(&pInfo->limitInfo, pResBlock, pTaskInfo);
-  qDebug("%s get sorted row block, rows:%" PRId64 ", limit:%" PRId64, GET_TASKID(pTaskInfo), pResBlock->info.rows,
-         pInfo->limitInfo.numOfOutputRows);
-
   return (pResBlock->info.rows > 0) ? pResBlock : NULL;
 }
 
@@ -3053,14 +2991,7 @@ void destroyTableMergeScanOperatorInfo(void* param) {
   STableMergeScanInfo* pTableScanInfo = (STableMergeScanInfo*)param;
   cleanupQueryTableDataCond(&pTableScanInfo->base.cond);
 
-  int32_t numOfTable = taosArrayGetSize(pTableScanInfo->queryConds);
-
-  for (int32_t i = 0; i < numOfTable; i++) {
-    STableMergeScanSortSourceParam* p = taosArrayGet(pTableScanInfo->sortSourceParams, i);
-    blockDataDestroy(p->inputBlock);
-    pTableScanInfo->base.readerAPI.tsdReaderClose(p->dataReader);
-    p->dataReader = NULL;
-  }
+  int32_t numOfTable = taosArrayGetSize(pTableScanInfo->sortSourceParams);
 
   pTableScanInfo->base.readerAPI.tsdReaderClose(pTableScanInfo->base.dataReader);
   pTableScanInfo->base.dataReader = NULL;
@@ -3069,16 +3000,11 @@ void destroyTableMergeScanOperatorInfo(void* param) {
   tsortDestroySortHandle(pTableScanInfo->pSortHandle);
   pTableScanInfo->pSortHandle = NULL;
 
-  for (int i = 0; i < taosArrayGetSize(pTableScanInfo->queryConds); i++) {
-    SQueryTableDataCond* pCond = taosArrayGet(pTableScanInfo->queryConds, i);
-    taosMemoryFree(pCond->colList);
-  }
-
-  taosArrayDestroy(pTableScanInfo->queryConds);
   destroyTableScanBase(&pTableScanInfo->base, &pTableScanInfo->base.readerAPI);
 
   pTableScanInfo->pResBlock = blockDataDestroy(pTableScanInfo->pResBlock);
   pTableScanInfo->pSortInputBlock = blockDataDestroy(pTableScanInfo->pSortInputBlock);
+  pTableScanInfo->pReaderBlock = blockDataDestroy(pTableScanInfo->pReaderBlock);
 
   taosArrayDestroy(pTableScanInfo->pSortInfo);
   taosMemoryFreeClear(param);
@@ -3140,6 +3066,8 @@ SOperatorInfo* createTableMergeScanOperatorInfo(STableScanPhysiNode* pTableScanN
   pInfo->base.scanFlag = MAIN_SCAN;
   pInfo->base.readHandle = *readHandle;
 
+  pInfo->readIdx = -1;
+
   pInfo->base.limitInfo.limit.limit = -1;
   pInfo->base.limitInfo.slimit.limit = -1;
   pInfo->base.pTableListInfo = pTableListInfo;
@@ -3162,6 +3090,8 @@ SOperatorInfo* createTableMergeScanOperatorInfo(STableScanPhysiNode* pTableScanN
   pInfo->pSortInputBlock = createOneDataBlock(pInfo->pResBlock, false);
   initLimitInfo(pTableScanNode->scan.node.pLimit, pTableScanNode->scan.node.pSlimit, &pInfo->limitInfo);
 
+  pInfo->pReaderBlock = createOneDataBlock(pInfo->pResBlock, false);
+
   int32_t  rowSize = pInfo->pResBlock->info.rowSize;
   uint32_t nCols = taosArrayGetSize(pInfo->pResBlock->pDataBlock);
   pInfo->bufPageSize = getProperSortPageSize(rowSize, nCols);
@@ -3570,6 +3500,4 @@ static void destoryTableCountScanOperator(void* param) {
 
   taosArrayDestroy(pTableCountScanInfo->stbUidList);
   taosMemoryFreeClear(param);
-}
-
-// clang-format on
+}
\ No newline at end of file
diff --git a/source/libs/executor/src/tsort.c b/source/libs/executor/src/tsort.c
index d26db6536facc3d36e9de03afefbf2dc923e32af..30e71487368d719025e4c76c7b1580e816d8e810 100644
--- a/source/libs/executor/src/tsort.c
+++ b/source/libs/executor/src/tsort.c
@@ -24,6 +24,7 @@
 #include "tpagedbuf.h"
 #include "tsort.h"
 #include "tutil.h"
+#include "tsimplehash.h"
 
 struct STupleHandle {
   SSDataBlock* pBlock;
@@ -42,13 +43,15 @@ struct SSortHandle {
   int64_t        startTs;
   uint64_t       totalElapsed;
 
-  uint64_t         maxRows;
-  uint32_t         maxTupleLength;
-  uint32_t         sortBufSize;
+  uint64_t         pqMaxRows;
+  uint32_t         pqMaxTupleLength;
+  uint32_t         pqSortBufSize;
   bool             forceUsePQSort;
   BoundedQueue*    pBoundedQueue;
   uint32_t         tmpRowIdx;
 
+  int64_t          mergeLimit;
+
   int32_t           sourceId;
   SSDataBlock*      pDataBlock;
   SMsortComparParam cmpParam;
@@ -173,8 +176,8 @@ void destroyTuple(void* t) {
  * @return
  */
 SSortHandle* tsortCreateSortHandle(SArray* pSortInfo, int32_t type, int32_t pageSize, int32_t numOfPages,
-                                   SSDataBlock* pBlock, const char* idstr, uint64_t maxRows, uint32_t maxTupleLength,
-                                   uint32_t sortBufSize) {
+                                   SSDataBlock* pBlock, const char* idstr, uint64_t pqMaxRows, uint32_t pqMaxTupleLength,
+                                   uint32_t pqSortBufSize) {
   SSortHandle* pSortHandle = taosMemoryCalloc(1, sizeof(SSortHandle));
 
   pSortHandle->type = type;
@@ -183,10 +186,10 @@ SSortHandle* tsortCreateSortHandle(SArray* pSortInfo, int32_t type, int32_t page
   pSortHandle->pSortInfo = pSortInfo;
   pSortHandle->loops = 0;
 
-  pSortHandle->maxTupleLength = maxTupleLength;
-  if (maxRows != 0) {
-    pSortHandle->sortBufSize = sortBufSize;
-    pSortHandle->maxRows = maxRows;
+  pSortHandle->pqMaxTupleLength = pqMaxTupleLength;
+  if (pqMaxRows != 0) {
+    pSortHandle->pqSortBufSize = pqSortBufSize;
+    pSortHandle->pqMaxRows = pqMaxRows;
   }
   pSortHandle->forceUsePQSort = false;
 
@@ -194,10 +197,18 @@ SSortHandle* tsortCreateSortHandle(SArray* pSortInfo, int32_t type, int32_t page
     pSortHandle->pDataBlock = createOneDataBlock(pBlock, false);
   }
 
+  pSortHandle->mergeLimit = -1;
+
   pSortHandle->pOrderedSource = taosArrayInit(4, POINTER_BYTES);
   pSortHandle->cmpParam.orderInfo = pSortInfo;
   pSortHandle->cmpParam.cmpGroupId = false;
-
+  pSortHandle->cmpParam.sortType = type;
+  if (type == SORT_BLOCK_TS_MERGE) {
+    SBlockOrderInfo* pOrder = TARRAY_GET_ELEM(pSortInfo, 0);
+    pSortHandle->cmpParam.tsSlotId = pOrder->slotId;
+    pSortHandle->cmpParam.order = pOrder->order;
+    pSortHandle->cmpParam.cmpFn = (pOrder->order == TSDB_ORDER_ASC) ? compareInt64Val : compareInt64ValDesc;
+  }
   tsortSetComparFp(pSortHandle, msortComparFn);
 
   if (idstr != NULL) {
@@ -469,11 +480,14 @@ static int32_t adjustMergeTreeForNextTuple(SSortSource* pSource, SMultiwayMergeT
     if (pHandle->type == SORT_SINGLESOURCE_SORT) {
       pSource->pageIndex++;
       if (pSource->pageIndex >= taosArrayGetSize(pSource->pageIdList)) {
+        qDebug("adjust merge tree. %d source completed %d", *numOfCompleted, pSource->pageIndex);
         (*numOfCompleted) += 1;
         pSource->src.rowIndex = -1;
         pSource->pageIndex = -1;
         pSource->src.pBlock = blockDataDestroy(pSource->src.pBlock);
       } else {
+        if (pSource->pageIndex % 512 == 0) qDebug("begin source %p page %d", pSource, pSource->pageIndex);
+
         int32_t* pPgId = taosArrayGet(pSource->pageIdList, pSource->pageIndex);
 
         void*   pPage = getBufPage(pHandle->pBuf, *pPgId);
@@ -486,7 +500,6 @@ static int32_t adjustMergeTreeForNextTuple(SSortSource* pSource, SMultiwayMergeT
         if (code != TSDB_CODE_SUCCESS) {
           return code;
         }
-
         releaseBufPage(pHandle->pBuf, pPage);
       }
     } else {
@@ -497,6 +510,7 @@ static int32_t adjustMergeTreeForNextTuple(SSortSource* pSource, SMultiwayMergeT
       if (pSource->src.pBlock == NULL) {
         (*numOfCompleted) += 1;
         pSource->src.rowIndex = -1;
+        qDebug("adjust merge tree. %d source completed", *numOfCompleted);
       }
     }
   }
@@ -577,53 +591,63 @@ int32_t msortComparFn(const void* pLeft, const void* pRight, void* param) {
     }
   }
 
-  for (int32_t i = 0; i < pInfo->size; ++i) {
-    SBlockOrderInfo* pOrder = TARRAY_GET_ELEM(pInfo, i);
-    SColumnInfoData* pLeftColInfoData = TARRAY_GET_ELEM(pLeftBlock->pDataBlock, pOrder->slotId);
+  if (pParam->sortType == SORT_BLOCK_TS_MERGE) {
+    SColumnInfoData* pLeftColInfoData = TARRAY_GET_ELEM(pLeftBlock->pDataBlock, pParam->tsSlotId);
+    SColumnInfoData* pRightColInfoData = TARRAY_GET_ELEM(pRightBlock->pDataBlock, pParam->tsSlotId);
+    int64_t*            left1 = (int64_t*)(pLeftColInfoData->pData) + pLeftSource->src.rowIndex;
+    int64_t*            right1 =  (int64_t*)(pRightColInfoData->pData) + pRightSource->src.rowIndex;
 
-    bool leftNull = false;
-    if (pLeftColInfoData->hasNull) {
-      if (pLeftBlock->pBlockAgg == NULL) {
-        leftNull = colDataIsNull_s(pLeftColInfoData, pLeftSource->src.rowIndex);
-      } else {
-        leftNull =
-            colDataIsNull(pLeftColInfoData, pLeftBlock->info.rows, pLeftSource->src.rowIndex, pLeftBlock->pBlockAgg[i]);
+    int ret = pParam->cmpFn(left1, right1);
+    return ret;
+  } else {
+    for (int32_t i = 0; i < pInfo->size; ++i) {
+      SBlockOrderInfo* pOrder = TARRAY_GET_ELEM(pInfo, i);
+      SColumnInfoData* pLeftColInfoData = TARRAY_GET_ELEM(pLeftBlock->pDataBlock, pOrder->slotId);
+      SColumnInfoData* pRightColInfoData = TARRAY_GET_ELEM(pRightBlock->pDataBlock, pOrder->slotId);
+
+      bool leftNull = false;
+      if (pLeftColInfoData->hasNull) {
+        if (pLeftBlock->pBlockAgg == NULL) {
+          leftNull = colDataIsNull_s(pLeftColInfoData, pLeftSource->src.rowIndex);
+        } else {
+          leftNull = colDataIsNull(pLeftColInfoData, pLeftBlock->info.rows, pLeftSource->src.rowIndex,
+                                   pLeftBlock->pBlockAgg[i]);
+        }
       }
-    }
 
-    SColumnInfoData* pRightColInfoData = TARRAY_GET_ELEM(pRightBlock->pDataBlock, pOrder->slotId);
-    bool             rightNull = false;
-    if (pRightColInfoData->hasNull) {
-      if (pRightBlock->pBlockAgg == NULL) {
-        rightNull = colDataIsNull_s(pRightColInfoData, pRightSource->src.rowIndex);
-      } else {
-        rightNull = colDataIsNull(pRightColInfoData, pRightBlock->info.rows, pRightSource->src.rowIndex,
-                                  pRightBlock->pBlockAgg[i]);
+      bool rightNull = false;
+      if (pRightColInfoData->hasNull) {
+        if (pRightBlock->pBlockAgg == NULL) {
+          rightNull = colDataIsNull_s(pRightColInfoData, pRightSource->src.rowIndex);
+        } else {
+          rightNull = colDataIsNull(pRightColInfoData, pRightBlock->info.rows, pRightSource->src.rowIndex,
+                                    pRightBlock->pBlockAgg[i]);
+        }
       }
-    }
 
-    if (leftNull && rightNull) {
-      continue;  // continue to next slot
-    }
+      if (leftNull && rightNull) {
+        continue;  // continue to next slot
+      }
 
-    if (rightNull) {
-      return pOrder->nullFirst ? 1 : -1;
-    }
+      if (rightNull) {
+        return pOrder->nullFirst ? 1 : -1;
+      }
 
-    if (leftNull) {
-      return pOrder->nullFirst ? -1 : 1;
-    }
+      if (leftNull) {
+        return pOrder->nullFirst ? -1 : 1;
+      }
 
-    void* left1 = colDataGetData(pLeftColInfoData, pLeftSource->src.rowIndex);
-    void* right1 = colDataGetData(pRightColInfoData, pRightSource->src.rowIndex);
+      void* left1 = colDataGetData(pLeftColInfoData, pLeftSource->src.rowIndex);
+      void* right1 = colDataGetData(pRightColInfoData, pRightSource->src.rowIndex);
 
-    __compar_fn_t fn = getKeyComparFunc(pLeftColInfoData->info.type, pOrder->order);
+      __compar_fn_t fn = getKeyComparFunc(pLeftColInfoData->info.type, pOrder->order);
 
-    int ret = fn(left1, right1);
-    if (ret == 0) {
-      continue;
-    } else {
-      return ret;
+      int ret = fn(left1, right1);
+      if (ret == 0) {
+        continue;
+      } else {
+        return ret;
+      }
     }
   }
   return 0;
@@ -668,6 +692,7 @@ static int32_t doInternalMergeSort(SSortHandle* pHandle) {
 
     // Only *numOfInputSources* can be loaded into buffer to perform the external sort.
     for (int32_t i = 0; i < sortGroup; ++i) {
+      qDebug("internal merge sort pass %d group %d. num input sources %d ", t, i, numOfInputSources);
       pHandle->sourceId += 1;
 
       int32_t end = (i + 1) * numOfInputSources - 1;
@@ -690,13 +715,15 @@ static int32_t doInternalMergeSort(SSortHandle* pHandle) {
         return code;
       }
 
+      int nMergedRows = 0;
+
       SArray* pPageIdList = taosArrayInit(4, sizeof(int32_t));
       while (1) {
         if (tsortIsClosed(pHandle)) {
           code = terrno = TSDB_CODE_TSC_QUERY_CANCELLED;
           return code;
         }
-        
+
         SSDataBlock* pDataBlock = getSortedBlockDataInner(pHandle, &pHandle->cmpParam, numOfRows);
         if (pDataBlock == NULL) {
           break;
@@ -720,8 +747,12 @@ static int32_t doInternalMergeSort(SSortHandle* pHandle) {
 
         setBufPageDirty(pPage, true);
         releaseBufPage(pHandle->pBuf, pPage);
+        nMergedRows += pDataBlock->info.rows;
 
         blockDataCleanup(pDataBlock);
+        if ((pHandle->mergeLimit != -1) && (nMergedRows >= pHandle->mergeLimit)) {
+          break;
+        }        
       }
 
       sortComparCleanup(&pHandle->cmpParam);
@@ -769,39 +800,285 @@ int32_t getProperSortPageSize(size_t rowSize, uint32_t numOfCols) {
   return pgSize;
 }
 
-static int32_t createInitialSources(SSortHandle* pHandle) {
-  size_t sortBufSize = pHandle->numOfPages * pHandle->pageSize;
-  int32_t code = 0;
+static int32_t createPageBuf(SSortHandle* pHandle) {
+  if (pHandle->pBuf == NULL) {
+    if (!osTempSpaceAvailable()) {
+      terrno = TSDB_CODE_NO_DISKSPACE;
+      qError("create page buf failed since %s, tempDir:%s", terrstr(), tsTempDir);
+      return terrno;
+    }
 
-  if (pHandle->type == SORT_SINGLESOURCE_SORT) {
-    SSortSource** pSource = taosArrayGet(pHandle->pOrderedSource, 0);
-    SSortSource*  source = *pSource;
-    *pSource = NULL;
+    int32_t code = createDiskbasedBuf(&pHandle->pBuf, pHandle->pageSize, pHandle->numOfPages * pHandle->pageSize,
+                                      "tableBlocksBuf", tsTempDir);
+    dBufSetPrintInfo(pHandle->pBuf);
+    if (code != TSDB_CODE_SUCCESS) {
+      return code;
+    }
+  }
+  return 0;
+}
 
-    tsortClearOrderdSource(pHandle->pOrderedSource, NULL, NULL);
+typedef struct SBlkMergeSupport {
+  int64_t** aTs;
+  int32_t* aRowIdx;
+  int32_t order;
+} SBlkMergeSupport;
+
+static int32_t blockCompareTsFn(const void* pLeft, const void* pRight, void* param) {
+  int32_t left = *(int32_t*)pLeft;
+  int32_t right = *(int32_t*)pRight;
+
+  SBlkMergeSupport* pSup = (SBlkMergeSupport*)param;
+  if (pSup->aRowIdx[left] == -1) {
+    return 1;
+  } else if (pSup->aRowIdx[right] == -1) {
+    return -1;
+  }
+
+  int64_t leftTs = pSup->aTs[left][pSup->aRowIdx[left]];
+  int64_t rightTs = pSup->aTs[right][pSup->aRowIdx[right]];
+
+  int32_t ret = leftTs>rightTs ? 1 : ((leftTs < rightTs) ? -1 : 0);
+  if (pSup->order == TSDB_ORDER_DESC) {
+    ret = -1 * ret;
+  }
+  return ret;
+}
+
+static int32_t appendDataBlockToPageBuf(SSortHandle* pHandle, SSDataBlock* blk, SArray* aPgId) {
+  int32_t pageId = -1;
+  void*   pPage = getNewBufPage(pHandle->pBuf, &pageId);
+  taosArrayPush(aPgId, &pageId);
+
+  int32_t size = blockDataGetSize(blk) + sizeof(int32_t) + taosArrayGetSize(blk->pDataBlock) * sizeof(int32_t);
+  ASSERT(size <= getBufPageSize(pHandle->pBuf));
+
+  blockDataToBuf(pPage, blk);
+
+  setBufPageDirty(pPage, true);
+  releaseBufPage(pHandle->pBuf, pPage);
+
+  return 0;
+}
+
+static int32_t getPageBufIncForRow(SSDataBlock* blk, int32_t row, int32_t rowIdxInPage) {
+  int sz = 0;
+  int numCols = taosArrayGetSize(blk->pDataBlock);
+  if (!blk->info.hasVarCol) {
+    sz += numCols * ((rowIdxInPage & 0x7) == 0 ? 1: 0);
+    sz += blockDataGetRowSize(blk);
+  } else {
+    for (int32_t i = 0; i < numCols; ++i) {
+      SColumnInfoData* pColInfoData = TARRAY_GET_ELEM(blk->pDataBlock, i);
+      if (IS_VAR_DATA_TYPE(pColInfoData->info.type)) {
+        if (pColInfoData->varmeta.offset[row] != -1) {
+          char* p = colDataGetData(pColInfoData, row);
+          sz += varDataTLen(p);
+        }
+
+        sz += sizeof(pColInfoData->varmeta.offset[0]);
+      } else {
+        sz += pColInfoData->info.bytes;
 
-    while (1) {
-      SSDataBlock* pBlock = pHandle->fetchfp(source->param);
-      if (pBlock == NULL) {
-        break;
+        if (((rowIdxInPage) & 0x07) == 0) {
+          sz += 1; // bitmap
+        }
       }
+    }    
+  }
+  return sz;
+}
+
+static int32_t sortBlocksToExtSource(SSortHandle* pHandle, SArray* aBlk, SBlockOrderInfo* order, SArray* aExtSrc) {
+  int pgHeaderSz = sizeof(int32_t) + sizeof(int32_t) * taosArrayGetSize(pHandle->pDataBlock->pDataBlock);
+  int32_t rowCap = blockDataGetCapacityInRow(pHandle->pDataBlock, pHandle->pageSize, pgHeaderSz);
+  blockDataEnsureCapacity(pHandle->pDataBlock, rowCap);
+  blockDataCleanup(pHandle->pDataBlock);
+  int32_t numBlks = taosArrayGetSize(aBlk);
+
+  SBlkMergeSupport sup;
+  sup.aRowIdx = taosMemoryCalloc(numBlks, sizeof(int32_t));
+  sup.aTs = taosMemoryCalloc(numBlks, sizeof(int64_t*));
+  sup.order = order->order;
+  for (int i = 0; i < numBlks; ++i) {
+    SSDataBlock* blk = taosArrayGetP(aBlk, i);
+    SColumnInfoData* col = taosArrayGet(blk->pDataBlock, order->slotId);
+    sup.aTs[i] = (int64_t*)col->pData;
+    sup.aRowIdx[i] = 0;
+  }
+
+  int32_t totalRows = 0;
+  for (int i = 0; i < numBlks; ++i) {
+    SSDataBlock* blk = taosArrayGetP(aBlk, i);
+    totalRows += blk->info.rows;
+  }
+
+  SArray* aPgId = taosArrayInit(8, sizeof(int32_t));
+
+  SMultiwayMergeTreeInfo* pTree = NULL;        
+  tMergeTreeCreate(&pTree, taosArrayGetSize(aBlk), &sup, blockCompareTsFn);
+  int32_t nRows = 0;
+  int32_t nMergedRows = 0;
+  bool mergeLimitReached = false;
+  size_t blkPgSz = pgHeaderSz;
+
+  while (nRows < totalRows) {
+    int32_t minIdx = tMergeTreeGetChosenIndex(pTree);
+    SSDataBlock* minBlk = taosArrayGetP(aBlk, minIdx);
+    int32_t minRow = sup.aRowIdx[minIdx];
+    int32_t bufInc = getPageBufIncForRow(minBlk, minRow, pHandle->pDataBlock->info.rows);
+
+    if (blkPgSz <= pHandle->pageSize && blkPgSz + bufInc > pHandle->pageSize) {
+        appendDataBlockToPageBuf(pHandle, pHandle->pDataBlock, aPgId);
+        nMergedRows += pHandle->pDataBlock->info.rows;
+
+        blockDataCleanup(pHandle->pDataBlock);
+        blkPgSz = pgHeaderSz;
+        bufInc = getPageBufIncForRow(minBlk, minRow, 0);
+        if ((pHandle->mergeLimit != -1) && (nMergedRows >= pHandle->mergeLimit)) {
+          mergeLimitReached = true;
+          break;
+        }        
+    }
+    blockDataEnsureCapacity(pHandle->pDataBlock, pHandle->pDataBlock->info.rows + 1);
+    appendOneRowToDataBlock(pHandle->pDataBlock, minBlk, &minRow);
+    blkPgSz += bufInc;
 
-      if (pHandle->pDataBlock == NULL) {
-        uint32_t numOfCols = taosArrayGetSize(pBlock->pDataBlock);
-        pHandle->pageSize = getProperSortPageSize(blockDataGetRowSize(pBlock), numOfCols);
+    ++nRows;
 
-        // todo, number of pages are set according to the total available sort buffer
-        pHandle->numOfPages = 1024;
-        sortBufSize = pHandle->numOfPages * pHandle->pageSize;
-        pHandle->pDataBlock = createOneDataBlock(pBlock, false);
+    if (sup.aRowIdx[minIdx] == minBlk->info.rows - 1) {
+      sup.aRowIdx[minIdx] = -1;
+    } else {
+      ++sup.aRowIdx[minIdx];
+    }
+    tMergeTreeAdjust(pTree, tMergeTreeGetAdjustIndex(pTree));
+  }
+  if (pHandle->pDataBlock->info.rows > 0) {
+    if (!mergeLimitReached) {
+      appendDataBlockToPageBuf(pHandle, pHandle->pDataBlock, aPgId);
+      nMergedRows += pHandle->pDataBlock->info.rows;
+    }
+    blockDataCleanup(pHandle->pDataBlock);
+  }
+  SSDataBlock* pMemSrcBlk = createOneDataBlock(pHandle->pDataBlock, false);
+  doAddNewExternalMemSource(pHandle->pBuf, aExtSrc, pMemSrcBlk, &pHandle->sourceId, aPgId);
+
+  taosMemoryFree(sup.aRowIdx);
+  taosMemoryFree(sup.aTs);
+
+  tMergeTreeDestroy(&pTree);
+
+  return 0;
+}
+
+static int32_t createBlocksMergeSortInitialSources(SSortHandle* pHandle) {
+  SBlockOrderInfo* pOrder = taosArrayGet(pHandle->pSortInfo, 0);
+  size_t           nSrc = taosArrayGetSize(pHandle->pOrderedSource);
+  SArray*          aExtSrc = taosArrayInit(nSrc, POINTER_BYTES);
+
+  size_t maxBufSize = pHandle->numOfPages * pHandle->pageSize;
+  createPageBuf(pHandle);
+
+  SSortSource* pSrc = taosArrayGetP(pHandle->pOrderedSource, 0);
+  int32_t      szSort = 0;
+
+  SArray* aBlkSort = taosArrayInit(8, POINTER_BYTES);
+  SSHashObj* mUidBlk = tSimpleHashInit(64, taosGetDefaultHashFunction(TSDB_DATA_TYPE_UBIGINT));
+  while (1) {
+    SSDataBlock* pBlk = pHandle->fetchfp(pSrc->param);
+
+    if (pBlk != NULL) {
+      szSort += blockDataGetSize(pBlk);
+
+      void* ppBlk = tSimpleHashGet(mUidBlk, &pBlk->info.id.uid, sizeof(pBlk->info.id.uid));
+      if (ppBlk != NULL) {
+        SSDataBlock* tBlk = *(SSDataBlock**)(ppBlk);
+        blockDataMerge(tBlk, pBlk);
+      } else {
+        SSDataBlock* tBlk = createOneDataBlock(pBlk, true);
+        tSimpleHashPut(mUidBlk, &pBlk->info.id.uid, sizeof(pBlk->info.id.uid), &tBlk, POINTER_BYTES);
+        taosArrayPush(aBlkSort, &tBlk);
       }
+    }
+
+    if ((pBlk != NULL && szSort > maxBufSize) || (pBlk == NULL && szSort > 0)) {
+      tSimpleHashClear(mUidBlk);
+
+      int64_t p = taosGetTimestampUs();
+      sortBlocksToExtSource(pHandle, aBlkSort, pOrder, aExtSrc);
+      int64_t el = taosGetTimestampUs() - p;
+      pHandle->sortElapsed += el;
 
-      if (pHandle->beforeFp != NULL) {
-        pHandle->beforeFp(pBlock, pHandle->param);
+      for (int i = 0; i < taosArrayGetSize(aBlkSort); ++i) {
+        blockDataDestroy(taosArrayGetP(aBlkSort, i));
       }
+      taosArrayClear(aBlkSort);
+      szSort = 0;
+      qDebug("source %zu created", taosArrayGetSize(aExtSrc));
+    }
+    if (pBlk == NULL) {
+      break;
+    };
+  }
+  tSimpleHashCleanup(mUidBlk);
+  taosArrayDestroy(aBlkSort);
+  tsortClearOrderdSource(pHandle->pOrderedSource, NULL, NULL);
+  taosArrayAddAll(pHandle->pOrderedSource, aExtSrc);
+  taosArrayDestroy(aExtSrc);
 
-      code = blockDataMerge(pHandle->pDataBlock, pBlock);
-      if (code != TSDB_CODE_SUCCESS) {
+  pHandle->type = SORT_SINGLESOURCE_SORT;
+  return 0;
+}
+
+static int32_t createBlocksQuickSortInitialSources(SSortHandle* pHandle) {
+  int32_t code = 0;
+  size_t  sortBufSize = pHandle->numOfPages * pHandle->pageSize;
+
+  SSortSource** pSource = taosArrayGet(pHandle->pOrderedSource, 0);
+  SSortSource*  source = *pSource;
+  *pSource = NULL;
+
+  tsortClearOrderdSource(pHandle->pOrderedSource, NULL, NULL);
+
+  while (1) {
+    SSDataBlock* pBlock = pHandle->fetchfp(source->param);
+    if (pBlock == NULL) {
+      break;
+    }
+
+    if (pHandle->pDataBlock == NULL) {
+      uint32_t numOfCols = taosArrayGetSize(pBlock->pDataBlock);
+      pHandle->pageSize = getProperSortPageSize(blockDataGetRowSize(pBlock), numOfCols);
+
+      // todo, number of pages are set according to the total available sort buffer
+      pHandle->numOfPages = 1024;
+      sortBufSize = pHandle->numOfPages * pHandle->pageSize;
+      pHandle->pDataBlock = createOneDataBlock(pBlock, false);
+    }
+
+    if (pHandle->beforeFp != NULL) {
+      pHandle->beforeFp(pBlock, pHandle->param);
+    }
+
+    code = blockDataMerge(pHandle->pDataBlock, pBlock);
+    if (code != TSDB_CODE_SUCCESS) {
+      if (source->param && !source->onlyRef) {
+        taosMemoryFree(source->param);
+      }
+      if (!source->onlyRef && source->src.pBlock) {
+        blockDataDestroy(source->src.pBlock);
+        source->src.pBlock = NULL;
+      }
+      taosMemoryFree(source);
+      return code;
+    }
+
+    size_t size = blockDataGetSize(pHandle->pDataBlock);
+    if (size > sortBufSize) {
+      // Perform the in-memory sort and then flush data in the buffer into disk.
+      int64_t p = taosGetTimestampUs();
+      code = blockDataSort(pHandle->pDataBlock, pHandle->pSortInfo);
+      if (code != 0) {
         if (source->param && !source->onlyRef) {
           taosMemoryFree(source->param);
         }
@@ -809,74 +1086,67 @@ static int32_t createInitialSources(SSortHandle* pHandle) {
           blockDataDestroy(source->src.pBlock);
           source->src.pBlock = NULL;
         }
+
         taosMemoryFree(source);
         return code;
       }
 
-      size_t size = blockDataGetSize(pHandle->pDataBlock);
-      if (size > sortBufSize) {
-        // Perform the in-memory sort and then flush data in the buffer into disk.
-        int64_t p = taosGetTimestampUs();
-        code = blockDataSort(pHandle->pDataBlock, pHandle->pSortInfo);
-        if (code != 0) {
-          if (source->param && !source->onlyRef) {
-            taosMemoryFree(source->param);
-          }
-          if (!source->onlyRef && source->src.pBlock) {
-            blockDataDestroy(source->src.pBlock);
-            source->src.pBlock = NULL;
-          }
-
-          taosMemoryFree(source);
-          return code;
-        }
-
-        int64_t el = taosGetTimestampUs() - p;
-        pHandle->sortElapsed += el;
-        if (pHandle->maxRows > 0) blockDataKeepFirstNRows(pHandle->pDataBlock, pHandle->maxRows);
-        code = doAddToBuf(pHandle->pDataBlock, pHandle);
-        if (code != TSDB_CODE_SUCCESS) {
-          return code;
-        }
+      int64_t el = taosGetTimestampUs() - p;
+      pHandle->sortElapsed += el;
+      if (pHandle->pqMaxRows > 0) blockDataKeepFirstNRows(pHandle->pDataBlock, pHandle->pqMaxRows);
+      code = doAddToBuf(pHandle->pDataBlock, pHandle);
+      if (code != TSDB_CODE_SUCCESS) {
+        return code;
       }
     }
+  }
 
-    if (source->param && !source->onlyRef) {
-      taosMemoryFree(source->param);
-    }
+  if (source->param && !source->onlyRef) {
+    taosMemoryFree(source->param);
+  }
 
-    taosMemoryFree(source);
+  taosMemoryFree(source);
 
-    if (pHandle->pDataBlock != NULL && pHandle->pDataBlock->info.rows > 0) {
-      size_t size = blockDataGetSize(pHandle->pDataBlock);
+  if (pHandle->pDataBlock != NULL && pHandle->pDataBlock->info.rows > 0) {
+    size_t size = blockDataGetSize(pHandle->pDataBlock);
 
-      // Perform the in-memory sort and then flush data in the buffer into disk.
-      int64_t p = taosGetTimestampUs();
+    // Perform the in-memory sort and then flush data in the buffer into disk.
+    int64_t p = taosGetTimestampUs();
 
-      code = blockDataSort(pHandle->pDataBlock, pHandle->pSortInfo);
-      if (code != 0) {
-        return code;
-      }
+    code = blockDataSort(pHandle->pDataBlock, pHandle->pSortInfo);
+    if (code != 0) {
+      return code;
+    }
 
-      if (pHandle->maxRows > 0) blockDataKeepFirstNRows(pHandle->pDataBlock, pHandle->maxRows);
-      int64_t el = taosGetTimestampUs() - p;
-      pHandle->sortElapsed += el;
+    if (pHandle->pqMaxRows > 0) blockDataKeepFirstNRows(pHandle->pDataBlock, pHandle->pqMaxRows);
+    int64_t el = taosGetTimestampUs() - p;
+    pHandle->sortElapsed += el;
 
-      // All sorted data can fit in memory, external memory sort is not needed. Return to directly
-      if (size <= sortBufSize && pHandle->pBuf == NULL) {
-        pHandle->cmpParam.numOfSources = 1;
-        pHandle->inMemSort = true;
+    // All sorted data can fit in memory, external memory sort is not needed. Return to directly
+    if (size <= sortBufSize && pHandle->pBuf == NULL) {
+      pHandle->cmpParam.numOfSources = 1;
+      pHandle->inMemSort = true;
 
-        pHandle->loops = 1;
-        pHandle->tupleHandle.rowIndex = -1;
-        pHandle->tupleHandle.pBlock = pHandle->pDataBlock;
-        return 0;
-      } else {
-        code = doAddToBuf(pHandle->pDataBlock, pHandle);
-      }
+      pHandle->loops = 1;
+      pHandle->tupleHandle.rowIndex = -1;
+      pHandle->tupleHandle.pBlock = pHandle->pDataBlock;
+      return 0;
+    } else {
+      code = doAddToBuf(pHandle->pDataBlock, pHandle);
     }
   }
+  return code;
+}
+
+static int32_t createInitialSources(SSortHandle* pHandle) {
+  int32_t code = 0;
 
+  if (pHandle->type == SORT_SINGLESOURCE_SORT) {
+    code = createBlocksQuickSortInitialSources(pHandle);
+  } else if (pHandle->type == SORT_BLOCK_TS_MERGE) {
+    code = createBlocksMergeSortInitialSources(pHandle);
+  }
+  qDebug("%zu sources created", taosArrayGetSize(pHandle->pOrderedSource));
   return code;
 }
 
@@ -923,6 +1193,10 @@ void tsortSetClosed(SSortHandle* pHandle) {
   atomic_store_8(&pHandle->closed, 2);
 }
 
+void tsortSetMergeLimit(SSortHandle* pHandle, int64_t mergeLimit) {
+  pHandle->mergeLimit = mergeLimit;
+}
+
 int32_t tsortSetFetchRawDataFp(SSortHandle* pHandle, _sort_fetch_block_fn_t fetchFp, void (*fp)(SSDataBlock*, void*),
                                void* param) {
   pHandle->fetchfp = fetchFp;
@@ -1002,8 +1276,8 @@ void tsortSetForceUsePQSort(SSortHandle* pHandle) {
 static bool tsortIsPQSortApplicable(SSortHandle* pHandle) {
   if (pHandle->type != SORT_SINGLESOURCE_SORT) return false;
   if (tsortIsForceUsePQSort(pHandle)) return true;
-  uint64_t maxRowsFitInMemory = pHandle->sortBufSize / (pHandle->maxTupleLength + sizeof(char*));
-  return maxRowsFitInMemory > pHandle->maxRows;
+  uint64_t maxRowsFitInMemory = pHandle->pqSortBufSize / (pHandle->pqMaxTupleLength + sizeof(char*));
+  return maxRowsFitInMemory > pHandle->pqMaxRows;
 }
 
 static bool tsortPQCompFn(void* a, void* b, void* param) {
@@ -1049,7 +1323,7 @@ static int32_t tupleComparFn(const void* pLeft, const void* pRight, void* param)
 }
 
 static int32_t tsortOpenForPQSort(SSortHandle* pHandle) {
-  pHandle->pBoundedQueue = createBoundedQueue(pHandle->maxRows, tsortPQCompFn, destroyTuple, pHandle);
+  pHandle->pBoundedQueue = createBoundedQueue(pHandle->pqMaxRows, tsortPQCompFn, destroyTuple, pHandle);
   if (NULL == pHandle->pBoundedQueue) return TSDB_CODE_OUT_OF_MEMORY;
   tsortSetComparFp(pHandle, tupleComparFn);
 
diff --git a/source/libs/nodes/src/nodesCloneFuncs.c b/source/libs/nodes/src/nodesCloneFuncs.c
index 6e4dde4ec1752f4bb3349c9c41658ab0140264ae..f5eacf0bd5d1c15bb7c773ee60caea7abc0dc0b5 100644
--- a/source/libs/nodes/src/nodesCloneFuncs.c
+++ b/source/libs/nodes/src/nodesCloneFuncs.c
@@ -361,6 +361,7 @@ static int32_t logicNodeCopy(const SLogicNode* pSrc, SLogicNode* pDst) {
   COPY_SCALAR_FIELD(groupAction);
   COPY_SCALAR_FIELD(inputTsOrder);
   COPY_SCALAR_FIELD(outputTsOrder);
+  COPY_SCALAR_FIELD(forceCreateNonBlockingOptr);
   return TSDB_CODE_SUCCESS;
 }
 
@@ -397,6 +398,7 @@ static int32_t logicScanCopy(const SScanLogicNode* pSrc, SScanLogicNode* pDst) {
   CLONE_NODE_LIST_FIELD(pTags);
   CLONE_NODE_FIELD(pSubtable);
   COPY_SCALAR_FIELD(igLastNull);
+  COPY_SCALAR_FIELD(groupOrderScan);
   return TSDB_CODE_SUCCESS;
 }
 
@@ -545,6 +547,7 @@ static int32_t physiNodeCopy(const SPhysiNode* pSrc, SPhysiNode* pDst) {
   CLONE_NODE_LIST_FIELD(pChildren);
   COPY_SCALAR_FIELD(inputTsOrder);
   COPY_SCALAR_FIELD(outputTsOrder);
+  COPY_SCALAR_FIELD(forceCreateNonBlockingOptr);
   return TSDB_CODE_SUCCESS;
 }
 
@@ -556,6 +559,7 @@ static int32_t physiScanCopy(const SScanPhysiNode* pSrc, SScanPhysiNode* pDst) {
   COPY_SCALAR_FIELD(suid);
   COPY_SCALAR_FIELD(tableType);
   COPY_OBJECT_FIELD(tableName, sizeof(SName));
+  COPY_SCALAR_FIELD(groupOrderScan);
   return TSDB_CODE_SUCCESS;
 }
 
diff --git a/source/libs/nodes/src/nodesCodeFuncs.c b/source/libs/nodes/src/nodesCodeFuncs.c
index 81116a60b06272a8b9be9a7c2438eec4b317f784..f25616065eb064d0160209320ecb744ba6ac23d8 100644
--- a/source/libs/nodes/src/nodesCodeFuncs.c
+++ b/source/libs/nodes/src/nodesCodeFuncs.c
@@ -1559,6 +1559,7 @@ static const char* jkScanPhysiPlanTableId = "TableId";
 static const char* jkScanPhysiPlanSTableId = "STableId";
 static const char* jkScanPhysiPlanTableType = "TableType";
 static const char* jkScanPhysiPlanTableName = "TableName";
+static const char* jkScanPhysiPlanGroupOrderScan = "GroupOrderScan";
 
 static int32_t physiScanNodeToJson(const void* pObj, SJson* pJson) {
   const STagScanPhysiNode* pNode = (const STagScanPhysiNode*)pObj;
@@ -1582,6 +1583,9 @@ static int32_t physiScanNodeToJson(const void* pObj, SJson* pJson) {
   if (TSDB_CODE_SUCCESS == code) {
     code = tjsonAddObject(pJson, jkScanPhysiPlanTableName, nameToJson, &pNode->tableName);
   }
+  if (TSDB_CODE_SUCCESS == code) {
+    code = tjsonAddBoolToObject(pJson, jkScanPhysiPlanGroupOrderScan, pNode->groupOrderScan);
+  }
 
   return code;
 }
@@ -1608,6 +1612,9 @@ static int32_t jsonToPhysiScanNode(const SJson* pJson, void* pObj) {
   if (TSDB_CODE_SUCCESS == code) {
     code = tjsonToObject(pJson, jkScanPhysiPlanTableName, jsonToName, &pNode->tableName);
   }
+  if (TSDB_CODE_SUCCESS == code) {
+    code = tjsonGetBoolValue(pJson, jkScanPhysiPlanGroupOrderScan, &pNode->groupOrderScan);
+  }
 
   return code;
 }
diff --git a/source/libs/nodes/src/nodesMsgFuncs.c b/source/libs/nodes/src/nodesMsgFuncs.c
index 1ca37defa4a76a6b679e85facae10a6cd758fb80..20e829766dafe75aa2372aa1e4147ac9b856ee71 100644
--- a/source/libs/nodes/src/nodesMsgFuncs.c
+++ b/source/libs/nodes/src/nodesMsgFuncs.c
@@ -1853,7 +1853,8 @@ enum {
   PHY_NODE_CODE_LIMIT,
   PHY_NODE_CODE_SLIMIT,
   PHY_NODE_CODE_INPUT_TS_ORDER,
-  PHY_NODE_CODE_OUTPUT_TS_ORDER
+  PHY_NODE_CODE_OUTPUT_TS_ORDER,
+  PHY_NODE_CODE_FORCE_NONBLOCKING_OPTR
 };
 
 static int32_t physiNodeToMsg(const void* pObj, STlvEncoder* pEncoder) {
@@ -1878,6 +1879,9 @@ static int32_t physiNodeToMsg(const void* pObj, STlvEncoder* pEncoder) {
   if (TSDB_CODE_SUCCESS == code) {
     code = tlvEncodeEnum(pEncoder, PHY_NODE_CODE_OUTPUT_TS_ORDER, pNode->outputTsOrder);
   }
+  if (TSDB_CODE_SUCCESS == code) {
+    code = tlvEncodeBool(pEncoder, PHY_NODE_CODE_FORCE_NONBLOCKING_OPTR, pNode->forceCreateNonBlockingOptr);
+  }
 
   return code;
 }
@@ -1910,6 +1914,8 @@ static int32_t msgToPhysiNode(STlvDecoder* pDecoder, void* pObj) {
       case PHY_NODE_CODE_OUTPUT_TS_ORDER:
         code = tlvDecodeEnum(pTlv, &pNode->outputTsOrder, sizeof(pNode->outputTsOrder));
         break;
+      case PHY_NODE_CODE_FORCE_NONBLOCKING_OPTR:
+        code = tlvDecodeBool(pTlv, &pNode->forceCreateNonBlockingOptr);
       default:
         break;
     }
@@ -1925,7 +1931,8 @@ enum {
   PHY_SCAN_CODE_BASE_UID,
   PHY_SCAN_CODE_BASE_SUID,
   PHY_SCAN_CODE_BASE_TABLE_TYPE,
-  PHY_SCAN_CODE_BASE_TABLE_NAME
+  PHY_SCAN_CODE_BASE_TABLE_NAME,
+  PHY_SCAN_CODE_BASE_GROUP_ORDER_SCAN
 };
 
 static int32_t physiScanNodeToMsg(const void* pObj, STlvEncoder* pEncoder) {
@@ -1950,6 +1957,9 @@ static int32_t physiScanNodeToMsg(const void* pObj, STlvEncoder* pEncoder) {
   if (TSDB_CODE_SUCCESS == code) {
     code = tlvEncodeObj(pEncoder, PHY_SCAN_CODE_BASE_TABLE_NAME, nameToMsg, &pNode->tableName);
   }
+  if (TSDB_CODE_SUCCESS == code) {
+    code = tlvEncodeBool(pEncoder, PHY_SCAN_CODE_BASE_GROUP_ORDER_SCAN, pNode->groupOrderScan);
+  }
 
   return code;
 }
@@ -1982,6 +1992,9 @@ static int32_t msgToPhysiScanNode(STlvDecoder* pDecoder, void* pObj) {
       case PHY_SCAN_CODE_BASE_TABLE_NAME:
         code = tlvDecodeObjFromTlv(pTlv, msgToName, &pNode->tableName);
         break;
+      case PHY_SCAN_CODE_BASE_GROUP_ORDER_SCAN:
+        code = tlvDecodeBool(pTlv, &pNode->groupOrderScan);
+        break;
       default:
         break;
     }
diff --git a/source/libs/planner/inc/planInt.h b/source/libs/planner/inc/planInt.h
index 82abc5d1a973dbafa820c41e0614e1b2c90a92c3..092fe1741187dcf3706a64a7be64a3032835b44a 100644
--- a/source/libs/planner/inc/planInt.h
+++ b/source/libs/planner/inc/planInt.h
@@ -43,6 +43,9 @@ int32_t splitLogicPlan(SPlanContext* pCxt, SLogicSubplan* pLogicSubplan);
 int32_t scaleOutLogicPlan(SPlanContext* pCxt, SLogicSubplan* pLogicSubplan, SQueryLogicPlan** pLogicPlan);
 int32_t createPhysiPlan(SPlanContext* pCxt, SQueryLogicPlan* pLogicPlan, SQueryPlan** pPlan, SArray* pExecNodeList);
 
+bool    isPartTableAgg(SAggLogicNode* pAgg);
+bool    isPartTableWinodw(SWindowLogicNode* pWindow);
+
 #ifdef __cplusplus
 }
 #endif
diff --git a/source/libs/planner/src/planOptimizer.c b/source/libs/planner/src/planOptimizer.c
index 05f478b116518ac04d824c9e812440e5976ded87..32721d8060405ae9b21e6cdce06ec78d56bdadeb 100644
--- a/source/libs/planner/src/planOptimizer.c
+++ b/source/libs/planner/src/planOptimizer.c
@@ -363,6 +363,18 @@ static void scanPathOptSetScanOrder(EScanOrder scanOrder, SScanLogicNode* pScan)
   }
 }
 
+static void scanPathOptSetGroupOrderScan(SScanLogicNode* pScan) {
+  if (pScan->tableType != TSDB_SUPER_TABLE) return;
+
+  if (pScan->node.pParent && nodeType(pScan->node.pParent) == QUERY_NODE_LOGIC_PLAN_AGG) {
+    SAggLogicNode* pAgg = (SAggLogicNode*)pScan->node.pParent;
+    bool withSlimit = pAgg->node.pSlimit != NULL || (pAgg->node.pParent && pAgg->node.pParent->pSlimit);
+    if (withSlimit && isPartTableAgg(pAgg)) {
+      pScan->groupOrderScan = pAgg->node.forceCreateNonBlockingOptr = true;
+    }
+  }
+}
+
 static int32_t scanPathOptimize(SOptimizeContext* pCxt, SLogicSubplan* pLogicSubplan) {
   SOsdInfo info = {.scanOrder = SCAN_ORDER_ASC};
   int32_t  code = scanPathOptMatch(pCxt, pLogicSubplan->pNode, &info);
@@ -371,6 +383,7 @@ static int32_t scanPathOptimize(SOptimizeContext* pCxt, SLogicSubplan* pLogicSub
     if (!pCxt->pPlanCxt->streamQuery) {
       scanPathOptSetScanOrder(info.scanOrder, info.pScan);
     }
+    scanPathOptSetGroupOrderScan(info.pScan);
   }
   if (TSDB_CODE_SUCCESS == code && (NULL != info.pDsoFuncs || NULL != info.pSdrFuncs)) {
     info.pScan->dataRequired = scanPathOptGetDataRequired(info.pSdrFuncs);
@@ -1675,6 +1688,7 @@ static int32_t partTagsOptimize(SOptimizeContext* pCxt, SLogicSubplan* pLogicSub
     if (TSDB_CODE_SUCCESS == code) {
       if (QUERY_NODE_LOGIC_PLAN_AGG == pNode->pParent->type) {
         SAggLogicNode* pParent = (SAggLogicNode*)(pNode->pParent);
+        scanPathOptSetGroupOrderScan(pScan);
         pParent->hasGroupKeyOptimized = true;
       }
 
diff --git a/source/libs/planner/src/planPhysiCreater.c b/source/libs/planner/src/planPhysiCreater.c
index b3d94a5e47aef06960954b88c612b9568f0e45d4..1b92dcd2e75acc6915c46ec2e509939c2189bc55 100644
--- a/source/libs/planner/src/planPhysiCreater.c
+++ b/source/libs/planner/src/planPhysiCreater.c
@@ -447,6 +447,7 @@ static int32_t createScanPhysiNodeFinalize(SPhysiPlanContext* pCxt, SSubplan* pS
     pScanPhysiNode->uid = pScanLogicNode->tableId;
     pScanPhysiNode->suid = pScanLogicNode->stableId;
     pScanPhysiNode->tableType = pScanLogicNode->tableType;
+    pScanPhysiNode->groupOrderScan = pScanLogicNode->groupOrderScan;
     memcpy(&pScanPhysiNode->tableName, &pScanLogicNode->tableName, sizeof(SName));
     if (NULL != pScanLogicNode->pTagCond) {
       pSubplan->pTagCond = nodesCloneNode(pScanLogicNode->pTagCond);
@@ -880,6 +881,7 @@ static int32_t createAggPhysiNode(SPhysiPlanContext* pCxt, SNodeList* pChildren,
 
   pAgg->mergeDataBlock = (GROUP_ACTION_KEEP == pAggLogicNode->node.groupAction ? false : true);
   pAgg->groupKeyOptimized = pAggLogicNode->hasGroupKeyOptimized;
+  pAgg->node.forceCreateNonBlockingOptr = pAggLogicNode->node.forceCreateNonBlockingOptr;
 
   SNodeList* pPrecalcExprs = NULL;
   SNodeList* pGroupKeys = NULL;
diff --git a/source/libs/planner/src/planSpliter.c b/source/libs/planner/src/planSpliter.c
index 246ee13fb00aa7d30857e63a03f18262ffb10510..84a486649efe4265794d24adf1bde7a295c779d7 100644
--- a/source/libs/planner/src/planSpliter.c
+++ b/source/libs/planner/src/planSpliter.c
@@ -306,54 +306,6 @@ static bool stbSplIsTableCountQuery(SLogicNode* pNode) {
   return QUERY_NODE_LOGIC_PLAN_SCAN == nodeType(pChild) && SCAN_TYPE_TABLE_COUNT == ((SScanLogicNode*)pChild)->scanType;
 }
 
-static SNodeList* stbSplGetPartKeys(SLogicNode* pNode) {
-  if (QUERY_NODE_LOGIC_PLAN_SCAN == nodeType(pNode)) {
-    return ((SScanLogicNode*)pNode)->pGroupTags;
-  } else if (QUERY_NODE_LOGIC_PLAN_PARTITION == nodeType(pNode)) {
-    return ((SPartitionLogicNode*)pNode)->pPartitionKeys;
-  } else {
-    return NULL;
-  }
-}
-
-static bool stbSplHasPartTbname(SNodeList* pPartKeys) {
-  if (NULL == pPartKeys) {
-    return false;
-  }
-  SNode* pPartKey = NULL;
-  FOREACH(pPartKey, pPartKeys) {
-    if (QUERY_NODE_GROUPING_SET == nodeType(pPartKey)) {
-      pPartKey = nodesListGetNode(((SGroupingSetNode*)pPartKey)->pParameterList, 0);
-    }
-    if ((QUERY_NODE_FUNCTION == nodeType(pPartKey) && FUNCTION_TYPE_TBNAME == ((SFunctionNode*)pPartKey)->funcType) ||
-        (QUERY_NODE_COLUMN == nodeType(pPartKey) && COLUMN_TYPE_TBNAME == ((SColumnNode*)pPartKey)->colType)) {
-      return true;
-    }
-  }
-  return false;
-}
-
-static bool stbSplNotSystemScan(SLogicNode* pNode) {
-  if (QUERY_NODE_LOGIC_PLAN_SCAN == nodeType(pNode)) {
-    return SCAN_TYPE_SYSTEM_TABLE != ((SScanLogicNode*)pNode)->scanType;
-  } else if (QUERY_NODE_LOGIC_PLAN_PARTITION == nodeType(pNode)) {
-    return stbSplNotSystemScan((SLogicNode*)nodesListGetNode(pNode->pChildren, 0));
-  } else {
-    return true;
-  }
-}
-
-static bool stbSplIsPartTableAgg(SAggLogicNode* pAgg) {
-  if (1 != LIST_LENGTH(pAgg->node.pChildren)) {
-    return false;
-  }
-  if (NULL != pAgg->pGroupKeys) {
-    return stbSplHasPartTbname(pAgg->pGroupKeys) &&
-           stbSplNotSystemScan((SLogicNode*)nodesListGetNode(pAgg->node.pChildren, 0));
-  }
-  return stbSplHasPartTbname(stbSplGetPartKeys((SLogicNode*)nodesListGetNode(pAgg->node.pChildren, 0)));
-}
-
 static bool stbSplNeedSplit(bool streamQuery, SLogicNode* pNode) {
   switch (nodeType(pNode)) {
     case QUERY_NODE_LOGIC_PLAN_SCAN:
@@ -364,7 +316,7 @@ static bool stbSplNeedSplit(bool streamQuery, SLogicNode* pNode) {
       return streamQuery ? false : stbSplIsMultiTbScanChild(streamQuery, pNode);
     case QUERY_NODE_LOGIC_PLAN_AGG:
       return (!stbSplHasGatherExecFunc(((SAggLogicNode*)pNode)->pAggFuncs) ||
-              stbSplIsPartTableAgg((SAggLogicNode*)pNode)) &&
+              isPartTableAgg((SAggLogicNode*)pNode)) &&
              stbSplHasMultiTbScan(streamQuery, pNode) && !stbSplIsTableCountQuery(pNode);
     case QUERY_NODE_LOGIC_PLAN_WINDOW:
       return stbSplNeedSplitWindow(streamQuery, pNode);
@@ -778,10 +730,6 @@ static int32_t stbSplSplitEvent(SSplitContext* pCxt, SStableSplitInfo* pInfo) {
   }
 }
 
-static bool stbSplIsPartTableWinodw(SWindowLogicNode* pWindow) {
-  return stbSplHasPartTbname(stbSplGetPartKeys((SLogicNode*)nodesListGetNode(pWindow->node.pChildren, 0)));
-}
-
 static int32_t stbSplSplitWindowForCrossTable(SSplitContext* pCxt, SStableSplitInfo* pInfo) {
   switch (((SWindowLogicNode*)pInfo->pSplitNode)->winType) {
     case WINDOW_TYPE_INTERVAL:
@@ -834,7 +782,7 @@ static int32_t stbSplSplitWindowForPartTable(SSplitContext* pCxt, SStableSplitIn
 }
 
 static int32_t stbSplSplitWindowNode(SSplitContext* pCxt, SStableSplitInfo* pInfo) {
-  if (stbSplIsPartTableWinodw((SWindowLogicNode*)pInfo->pSplitNode)) {
+  if (isPartTableWinodw((SWindowLogicNode*)pInfo->pSplitNode)) {
     return stbSplSplitWindowForPartTable(pCxt, pInfo);
   } else {
     return stbSplSplitWindowForCrossTable(pCxt, pInfo);
@@ -920,7 +868,7 @@ static int32_t stbSplSplitAggNodeForCrossTable(SSplitContext* pCxt, SStableSplit
 }
 
 static int32_t stbSplSplitAggNode(SSplitContext* pCxt, SStableSplitInfo* pInfo) {
-  if (stbSplIsPartTableAgg((SAggLogicNode*)pInfo->pSplitNode)) {
+  if (isPartTableAgg((SAggLogicNode*)pInfo->pSplitNode)) {
     return stbSplSplitAggNodeForPartTable(pCxt, pInfo);
   }
   return stbSplSplitAggNodeForCrossTable(pCxt, pInfo);
diff --git a/source/libs/planner/src/planUtil.c b/source/libs/planner/src/planUtil.c
index 29e87b34ce046d166678ab54ae5095d6e9a859fe..88086cde1d0edb91e2918a26935495be0b3120ce 100644
--- a/source/libs/planner/src/planUtil.c
+++ b/source/libs/planner/src/planUtil.c
@@ -321,3 +321,57 @@ int32_t adjustLogicNodeDataRequirement(SLogicNode* pNode, EDataOrderLevel requir
   }
   return code;
 }
+
+static bool stbNotSystemScan(SLogicNode* pNode) {
+  if (QUERY_NODE_LOGIC_PLAN_SCAN == nodeType(pNode)) {
+    return SCAN_TYPE_SYSTEM_TABLE != ((SScanLogicNode*)pNode)->scanType;
+  } else if (QUERY_NODE_LOGIC_PLAN_PARTITION == nodeType(pNode)) {
+    return stbNotSystemScan((SLogicNode*)nodesListGetNode(pNode->pChildren, 0));
+  } else {
+    return true;
+  }
+}
+
+static bool stbHasPartTbname(SNodeList* pPartKeys) {
+  if (NULL == pPartKeys) {
+    return false;
+  }
+  SNode* pPartKey = NULL;
+  FOREACH(pPartKey, pPartKeys) {
+    if (QUERY_NODE_GROUPING_SET == nodeType(pPartKey)) {
+      pPartKey = nodesListGetNode(((SGroupingSetNode*)pPartKey)->pParameterList, 0);
+    }
+    if ((QUERY_NODE_FUNCTION == nodeType(pPartKey) && FUNCTION_TYPE_TBNAME == ((SFunctionNode*)pPartKey)->funcType) ||
+        (QUERY_NODE_COLUMN == nodeType(pPartKey) && COLUMN_TYPE_TBNAME == ((SColumnNode*)pPartKey)->colType)) {
+      return true;
+    }
+  }
+  return false;
+}
+
+static SNodeList* stbSplGetPartKeys(SLogicNode* pNode) {
+  if (QUERY_NODE_LOGIC_PLAN_SCAN == nodeType(pNode)) {
+    return ((SScanLogicNode*)pNode)->pGroupTags;
+  } else if (QUERY_NODE_LOGIC_PLAN_PARTITION == nodeType(pNode)) {
+    return ((SPartitionLogicNode*)pNode)->pPartitionKeys;
+  } else {
+    return NULL;
+  }
+}
+
+bool isPartTableAgg(SAggLogicNode* pAgg) {
+  if (1 != LIST_LENGTH(pAgg->node.pChildren)) {
+    return false;
+  }
+  if (NULL != pAgg->pGroupKeys) {
+    return stbHasPartTbname(pAgg->pGroupKeys) &&
+           stbNotSystemScan((SLogicNode*)nodesListGetNode(pAgg->node.pChildren, 0));
+  }
+  return stbHasPartTbname(stbSplGetPartKeys((SLogicNode*)nodesListGetNode(pAgg->node.pChildren, 0)));
+}
+
+bool isPartTableWinodw(SWindowLogicNode* pWindow) {
+  return stbHasPartTbname(stbSplGetPartKeys((SLogicNode*)nodesListGetNode(pWindow->node.pChildren, 0)));
+}
+
+
diff --git a/tests/system-test/0-others/show.py b/tests/system-test/0-others/show.py
index 50a1662ba013d4e4dd52ffbf4bdfb9f47f7c92d7..9d26b3a2aeb679195246e0989605b48f12b91a77 100644
--- a/tests/system-test/0-others/show.py
+++ b/tests/system-test/0-others/show.py
@@ -95,6 +95,23 @@ class TDTestCase:
           tdSql.checkEqual(f'{db}',tdSql.queryResult[0][0])
           tdSql.checkEqual(f'CREATE DATABASE `{db}`',tdSql.queryResult[0][1])
 
+    def show_create_systb_sql(self):
+        for param in self.ins_param_list:
+          tdSql.query(f'show create table information_schema.ins_{param}')
+          tdSql.checkEqual(f'ins_{param}',tdSql.queryResult[0][0])
+
+          tdSql.execute(f'use information_schema')
+          tdSql.query(f'show create table ins_{param}')
+          tdSql.checkEqual(f'ins_{param}',tdSql.queryResult[0][0])
+
+        for param in self.perf_param_list:
+          tdSql.query(f'show create table performance_schema.perf_{param}')
+          tdSql.checkEqual(f'perf_{param}',tdSql.queryResult[0][0])
+
+          tdSql.execute(f'use performance_schema')
+          tdSql.query(f'show create table perf_{param}')
+          tdSql.checkEqual(f'perf_{param}',tdSql.queryResult[0][0])
+
     def show_create_sql(self):
         create_db_sql = self.set_create_database_sql(self.db_param)
         print(create_db_sql)
@@ -200,6 +217,7 @@ class TDTestCase:
         self.perf_check()
         self.show_create_sql()
         self.show_create_sysdb_sql()
+        self.show_create_systb_sql()
 
     def stop(self):
         tdSql.close()
diff --git a/tests/system-test/2-query/columnLenUpdated.py b/tests/system-test/2-query/columnLenUpdated.py
index e43b32a716017702382ce0384bb377d692e64684..93d9a492f946db423b78f7891b533af8c9a6da9f 100644
--- a/tests/system-test/2-query/columnLenUpdated.py
+++ b/tests/system-test/2-query/columnLenUpdated.py
@@ -202,7 +202,7 @@ class TDTestCase:
         if retCode != "TAOS_OK":
             tdLog.exit("taos -s fail")
 
-        tdSql.query("select count(*) from stb group by tg1")
+        tdSql.query("select count(*) from stb group by tg1 order by count(*) desc")
         tdSql.checkData(0, 0, 2)
         tdSql.checkData(1, 0, 1)
 
diff --git a/tests/system-test/5-taos-tools/taosbenchmark/insertMix.py b/tests/system-test/5-taos-tools/taosbenchmark/insertMix.py
index 60daa8cdc27f5e683239d6722e4f28ae1b8d90d1..b4046b8c98b311d8f73fc811cf4e70a2d2510e38 100644
--- a/tests/system-test/5-taos-tools/taosbenchmark/insertMix.py
+++ b/tests/system-test/5-taos-tools/taosbenchmark/insertMix.py
@@ -79,6 +79,11 @@ class TDTestCase:
         tdSql.query("select count(*) from (select * from meters order by ts desc)")
         tdSql.checkData(0, 0, allCnt)
 
+        rowCnt = tdSql.query("select tbname, count(*) from meters partition by tbname slimit 11")
+        if rowCnt != 10:
+            tdLog.exit("partition by tbname should return 10 rows of table data which is " + str(rowCnt))
+            return
+
 
     def run(self):
         binPath = self.getPath()