Merge pull request #7556 from taosdata/fix/query

Fix/query

src/client/inc/tscUtil.h

@@ -36,7 +36,7 @@ extern "C" {
   (((metaInfo)->pTableMeta != NULL) && ((metaInfo)->pTableMeta->tableType == TSDB_CHILD_TABLE))
 
 #define UTIL_TABLE_IS_NORMAL_TABLE(metaInfo) \
-  (!(UTIL_TABLE_IS_SUPER_TABLE(metaInfo) || UTIL_TABLE_IS_CHILD_TABLE(metaInfo) || UTIL_TABLE_IS_TMP_TABLE(metaInfo)))
+  (!(UTIL_TABLE_IS_SUPER_TABLE(metaInfo) || UTIL_TABLE_IS_CHILD_TABLE(metaInfo)))
 
 #define UTIL_TABLE_IS_TMP_TABLE(metaInfo) \
   (((metaInfo)->pTableMeta != NULL) && ((metaInfo)->pTableMeta->tableType == TSDB_TEMP_TABLE))

src/client/src/tscGlobalmerge.c

@@ -35,6 +35,7 @@ typedef struct SCompareParam {
 
 static bool needToMerge(SSDataBlock* pBlock, SArray* columnIndexList, int32_t index, char **buf) {
   int32_t ret = 0;
+
   size_t  size = taosArrayGetSize(columnIndexList);
   if (size > 0) {
     ret = compare_aRv(pBlock, columnIndexList, (int32_t) size, index, buf, TSDB_ORDER_ASC);
@@ -564,9 +565,11 @@ static void savePrevOrderColumns(char** prevRow, SArray* pColumnList, SSDataBloc
   (*hasPrev) = true;
 }
 
+// tsdb_func_tag function only produce one row of result. Therefore, we need to copy the
+// output value to multiple rows
 static void setTagValueForMultipleRows(SQLFunctionCtx* pCtx, int32_t numOfOutput, int32_t numOfRows) {
   if (numOfRows <= 1) {
-    return ;
+    return;
   }
 
   for (int32_t k = 0; k < numOfOutput; ++k) {
@@ -574,12 +577,49 @@ static void setTagValueForMultipleRows(SQLFunctionCtx* pCtx, int32_t numOfOutput
       continue;
     }
 
-    int32_t inc = numOfRows - 1;  // tsdb_func_tag function only produce one row of result
-    char* src = pCtx[k].pOutput;
+    char* src  = pCtx[k].pOutput;
+    char* dst = pCtx[k].pOutput + pCtx[k].outputBytes;
 
-    for (int32_t i = 0; i < inc; ++i) {
-      pCtx[k].pOutput += pCtx[k].outputBytes;
-      memcpy(pCtx[k].pOutput, src, (size_t)pCtx[k].outputBytes);
+    // Let's start from the second row, as the first row has result value already.
+    for (int32_t i = 1; i < numOfRows; ++i) {
+      memcpy(dst, src, (size_t)pCtx[k].outputBytes);
+      dst += pCtx[k].outputBytes;
+    }
+  }
+}
+
+static void doMergeResultImpl(SMultiwayMergeInfo* pInfo, SQLFunctionCtx *pCtx, int32_t numOfExpr, int32_t rowIndex, char** pDataPtr) {
+  for (int32_t j = 0; j < numOfExpr; ++j) {
+    pCtx[j].pInput = pDataPtr[j] + pCtx[j].inputBytes * rowIndex;
+  }
+
+  for (int32_t j = 0; j < numOfExpr; ++j) {
+    int32_t functionId = pCtx[j].functionId;
+    if (functionId == TSDB_FUNC_TAG_DUMMY || functionId == TSDB_FUNC_TS_DUMMY) {
+      continue;
+    }
+
+    if (functionId < 0) {
+      SUdfInfo* pUdfInfo = taosArrayGet(pInfo->udfInfo, -1 * functionId - 1);
+      doInvokeUdf(pUdfInfo, &pCtx[j], 0, TSDB_UDF_FUNC_MERGE);
+    } else {
+      aAggs[functionId].mergeFunc(&pCtx[j]);
+    }
+  }
+}
+
+static void doFinalizeResultImpl(SMultiwayMergeInfo* pInfo, SQLFunctionCtx *pCtx, int32_t numOfExpr) {
+  for(int32_t j = 0; j < numOfExpr; ++j) {
+    int32_t functionId = pCtx[j].functionId;
+    if (functionId == TSDB_FUNC_TAG_DUMMY || functionId == TSDB_FUNC_TS_DUMMY) {
+      continue;
+    }
+
+    if (functionId < 0) {
+      SUdfInfo* pUdfInfo = taosArrayGet(pInfo->udfInfo, -1 * functionId - 1);
+      doInvokeUdf(pUdfInfo, &pCtx[j], 0, TSDB_UDF_FUNC_FINALIZE);
+    } else {
+      aAggs[functionId].xFinalize(&pCtx[j]);
     }
   }
 }
@@ -588,52 +628,18 @@ static void doExecuteFinalMerge(SOperatorInfo* pOperator, int32_t numOfExpr, SSD
   SMultiwayMergeInfo* pInfo = pOperator->info;
   SQLFunctionCtx* pCtx = pInfo->binfo.pCtx;
 
-  char** add = calloc(pBlock->info.numOfCols, POINTER_BYTES);
+  char** addrPtr = calloc(pBlock->info.numOfCols, POINTER_BYTES);
   for(int32_t i = 0; i < pBlock->info.numOfCols; ++i) {
-    add[i] = pCtx[i].pInput;
+    addrPtr[i] = pCtx[i].pInput;
     pCtx[i].size = 1;
   }
 
   for(int32_t i = 0; i < pBlock->info.rows; ++i) {
     if (pInfo->hasPrev) {
       if (needToMerge(pBlock, pInfo->orderColumnList, i, pInfo->prevRow)) {
-        for (int32_t j = 0; j < numOfExpr; ++j) {
-          pCtx[j].pInput = add[j] + pCtx[j].inputBytes * i;
-        }
-
-        for (int32_t j = 0; j < numOfExpr; ++j) {
-          int32_t functionId = pCtx[j].functionId;
-          if (functionId == TSDB_FUNC_TAG_DUMMY || functionId == TSDB_FUNC_TS_DUMMY) {
-            continue;
-          }
-
-          if (functionId < 0) {
-            SUdfInfo* pUdfInfo = taosArrayGet(pInfo->udfInfo, -1 * functionId - 1);
-
-            doInvokeUdf(pUdfInfo, &pCtx[j], 0, TSDB_UDF_FUNC_MERGE);
-
-            continue;
-          }
-
-          aAggs[functionId].mergeFunc(&pCtx[j]);
-        }
+        doMergeResultImpl(pInfo, pCtx, numOfExpr, i, addrPtr);
       } else {
-        for(int32_t j = 0; j < numOfExpr; ++j) {  // TODO refactor
-          int32_t functionId = pCtx[j].functionId;
-          if (functionId == TSDB_FUNC_TAG_DUMMY || functionId == TSDB_FUNC_TS_DUMMY) {
-            continue;
-          }
-
-          if (functionId < 0) {
-            SUdfInfo* pUdfInfo = taosArrayGet(pInfo->udfInfo, -1 * functionId - 1);
-
-            doInvokeUdf(pUdfInfo, &pCtx[j], 0, TSDB_UDF_FUNC_FINALIZE);
-
-            continue;
-          }
-
-          aAggs[functionId].xFinalize(&pCtx[j]);
-        }
+        doFinalizeResultImpl(pInfo, pCtx, numOfExpr);
 
         int32_t numOfRows = getNumOfResult(pOperator->pRuntimeEnv, pInfo->binfo.pCtx, pOperator->numOfOutput);
         setTagValueForMultipleRows(pCtx, pOperator->numOfOutput, numOfRows);
@@ -655,48 +661,10 @@ static void doExecuteFinalMerge(SOperatorInfo* pOperator, int32_t numOfExpr, SSD
           aAggs[pCtx[j].functionId].init(&pCtx[j], pCtx[j].resultInfo);
         }
 
-        for (int32_t j = 0; j < numOfExpr; ++j) {
-          pCtx[j].pInput = add[j] + pCtx[j].inputBytes * i;
-        }
-
-        for (int32_t j = 0; j < numOfExpr; ++j) {
-          int32_t functionId = pCtx[j].functionId;
-          if (functionId == TSDB_FUNC_TAG_DUMMY || functionId == TSDB_FUNC_TS_DUMMY) {
-            continue;
-          }
-
-          if (functionId < 0) {
-            SUdfInfo* pUdfInfo = taosArrayGet(pInfo->udfInfo, -1 * functionId - 1);
-
-            doInvokeUdf(pUdfInfo, &pCtx[j], 0, TSDB_UDF_FUNC_MERGE);
-
-            continue;
-          }
-
-          aAggs[functionId].mergeFunc(&pCtx[j]);
-        }
+        doMergeResultImpl(pInfo, pCtx, numOfExpr, i, addrPtr);
       }
     } else {
-      for (int32_t j = 0; j < numOfExpr; ++j) {
-        pCtx[j].pInput = add[j] + pCtx[j].inputBytes * i;
-      }
-
-      for (int32_t j = 0; j < numOfExpr; ++j) {
-        int32_t functionId = pCtx[j].functionId;
-        if (functionId == TSDB_FUNC_TAG_DUMMY || functionId == TSDB_FUNC_TS_DUMMY) {
-          continue;
-        }
-
-        if (functionId < 0) {
-          SUdfInfo* pUdfInfo = taosArrayGet(pInfo->udfInfo, -1 * functionId - 1);
-
-          doInvokeUdf(pUdfInfo, &pCtx[j], 0, TSDB_UDF_FUNC_MERGE);
-
-          continue;
-        }
-
-        aAggs[functionId].mergeFunc(&pCtx[j]);
-      }
+      doMergeResultImpl(pInfo, pCtx, numOfExpr, i, addrPtr);
     }
 
     savePrevOrderColumns(pInfo->prevRow, pInfo->orderColumnList, pBlock, i, &pInfo->hasPrev);
@@ -704,11 +672,11 @@ static void doExecuteFinalMerge(SOperatorInfo* pOperator, int32_t numOfExpr, SSD
 
   {
     for(int32_t i = 0; i < pBlock->info.numOfCols; ++i) {
-      pCtx[i].pInput = add[i];
+      pCtx[i].pInput = addrPtr[i];
     }
   }
 
-  tfree(add);
+  tfree(addrPtr);
 }
 
 static bool isAllSourcesCompleted(SGlobalMerger *pMerger) {
@@ -816,6 +784,8 @@ SSDataBlock* doMultiwayMergeSort(void* param, bool* newgroup) {
     SLocalDataSource *pOneDataSrc = pMerger->pLocalDataSrc[pTree->pNode[0].index];
     bool sameGroup = true;
     if (pInfo->hasPrev) {
+
+      // todo refactor extract method
       int32_t numOfCols = (int32_t)taosArrayGetSize(pInfo->orderColumnList);
 
       // if this row belongs to current result set group
@@ -955,9 +925,10 @@ SSDataBlock* doGlobalAggregate(void* param, bool* newgroup) {
       break;
     }
 
+    bool sameGroup = true;
     if (pAggInfo->hasGroupColData) {
-      bool sameGroup = isSameGroup(pAggInfo->groupColumnList, pBlock, pAggInfo->currentGroupColData);
-      if (!sameGroup) {
+      sameGroup = isSameGroup(pAggInfo->groupColumnList, pBlock, pAggInfo->currentGroupColData);
+      if (!sameGroup && !pAggInfo->multiGroupResults) {
         *newgroup = true;
         pAggInfo->hasDataBlockForNewGroup = true;
         pAggInfo->pExistBlock = pBlock;
@@ -976,26 +947,11 @@ SSDataBlock* doGlobalAggregate(void* param, bool* newgroup) {
   }
 
   if (handleData) { // data in current group is all handled
-    for(int32_t j = 0; j < pOperator->numOfOutput; ++j) {
-      int32_t functionId = pAggInfo->binfo.pCtx[j].functionId;
-      if (functionId == TSDB_FUNC_TAG_DUMMY || functionId == TSDB_FUNC_TS_DUMMY) {
-        continue;
-      }
-
-      if (functionId < 0) {
-        SUdfInfo* pUdfInfo = taosArrayGet(pAggInfo->udfInfo, -1 * functionId - 1);
-
-        doInvokeUdf(pUdfInfo, &pAggInfo->binfo.pCtx[j], 0, TSDB_UDF_FUNC_FINALIZE);
-
-        continue;
-      }
-
-      aAggs[functionId].xFinalize(&pAggInfo->binfo.pCtx[j]);
-    }
+    doFinalizeResultImpl(pAggInfo, pAggInfo->binfo.pCtx, pOperator->numOfOutput);
 
     int32_t numOfRows = getNumOfResult(pOperator->pRuntimeEnv, pAggInfo->binfo.pCtx, pOperator->numOfOutput);
-    pAggInfo->binfo.pRes->info.rows += numOfRows;
 
+    pAggInfo->binfo.pRes->info.rows += numOfRows;
     setTagValueForMultipleRows(pAggInfo->binfo.pCtx, pOperator->numOfOutput, numOfRows);
   }
 
@@ -1019,71 +975,127 @@ SSDataBlock* doGlobalAggregate(void* param, bool* newgroup) {
   return (pRes->info.rows != 0)? pRes:NULL;
 }
 
-static SSDataBlock* skipGroupBlock(SOperatorInfo* pOperator, bool* newgroup) {
-  SSLimitOperatorInfo *pInfo = pOperator->info;
-  assert(pInfo->currentGroupOffset >= 0);
+static void doHandleDataInCurrentGroup(SSLimitOperatorInfo* pInfo, SSDataBlock* pBlock, int32_t rowIndex) {
+  if (pInfo->currentOffset > 0) {
+    pInfo->currentOffset -= 1;
+  } else {
+    // discard the data rows in current group
+    if (pInfo->limit.limit < 0 || (pInfo->limit.limit >= 0 && pInfo->rowsTotal < pInfo->limit.limit)) {
+      size_t num1 = taosArrayGetSize(pInfo->pRes->pDataBlock);
+      for (int32_t i = 0; i < num1; ++i) {
+        SColumnInfoData *pColInfoData = taosArrayGet(pBlock->pDataBlock, i);
+        SColumnInfoData *pDstInfoData = taosArrayGet(pInfo->pRes->pDataBlock, i);
 
-  SSDataBlock* pBlock = NULL;
-  if (pInfo->currentGroupOffset == 0) {
-    publishOperatorProfEvent(pOperator->upstream[0], QUERY_PROF_BEFORE_OPERATOR_EXEC);
-    pBlock = pOperator->upstream[0]->exec(pOperator->upstream[0], newgroup);
-    publishOperatorProfEvent(pOperator->upstream[0], QUERY_PROF_AFTER_OPERATOR_EXEC);
-    if (pBlock == NULL) {
-      setQueryStatus(pOperator->pRuntimeEnv, QUERY_COMPLETED);
-      pOperator->status = OP_EXEC_DONE;
-    }
+        SColumnInfo *pColInfo = &pColInfoData->info;
+        
+        char *pSrc = rowIndex * pColInfo->bytes + (char *)pColInfoData->pData;
+        char *pDst = (char *)pDstInfoData->pData + (pInfo->pRes->info.rows * pColInfo->bytes);
 
-    if (*newgroup == false && pInfo->limit.limit > 0 && pInfo->rowsTotal >= pInfo->limit.limit) {
-      while ((*newgroup) == false) {  // ignore the remain blocks
-        publishOperatorProfEvent(pOperator->upstream[0], QUERY_PROF_BEFORE_OPERATOR_EXEC);
-        pBlock = pOperator->upstream[0]->exec(pOperator->upstream[0], newgroup);
-        publishOperatorProfEvent(pOperator->upstream[0], QUERY_PROF_AFTER_OPERATOR_EXEC);
-        if (pBlock == NULL) {
-          setQueryStatus(pOperator->pRuntimeEnv, QUERY_COMPLETED);
-          pOperator->status = OP_EXEC_DONE;
-          return NULL;
-        }
+        memcpy(pDst, pSrc, pColInfo->bytes);
       }
+
+      pInfo->rowsTotal += 1;
+      pInfo->pRes->info.rows += 1;
     }
+  }
+}
+
+static void ensureOutputBuf(SSLimitOperatorInfo * pInfo, SSDataBlock *pResultBlock, int32_t numOfRows) {
+  if (pInfo->capacity < pResultBlock->info.rows + numOfRows) {
+    int32_t total = pResultBlock->info.rows + numOfRows;
+
+    size_t num = taosArrayGetSize(pResultBlock->pDataBlock);
+    for (int32_t i = 0; i < num; ++i) {
+      SColumnInfoData *pInfoData = taosArrayGet(pResultBlock->pDataBlock, i);
+
+      char *tmp = realloc(pInfoData->pData, total * pInfoData->info.bytes);
+      if (tmp != NULL) {
+        pInfoData->pData = tmp;
+      } else {
+        // todo handle the malloc failure
+      }
 
-    return pBlock;
+      pInfo->capacity  = total;
+      pInfo->threshold = (int64_t) (total * 0.8);
+    }
   }
+}
 
-    publishOperatorProfEvent(pOperator->upstream[0], QUERY_PROF_BEFORE_OPERATOR_EXEC);
-    pBlock = pOperator->upstream[0]->exec(pOperator->upstream[0], newgroup);
-    publishOperatorProfEvent(pOperator->upstream[0], QUERY_PROF_AFTER_OPERATOR_EXEC);
+enum {
+  BLOCK_NEW_GROUP  = 1,
+  BLOCK_NO_GROUP   = 2,
+  BLOCK_SAME_GROUP = 3,
+};
 
-    if (pBlock == NULL) {
-      setQueryStatus(pOperator->pRuntimeEnv, QUERY_COMPLETED);
-      pOperator->status = OP_EXEC_DONE;
-      return NULL;
+static int32_t doSlimitImpl(SOperatorInfo* pOperator, SSLimitOperatorInfo* pInfo, SSDataBlock* pBlock) {
+  int32_t rowIndex = 0;
+
+  while (rowIndex < pBlock->info.rows) {
+    int32_t numOfCols = (int32_t)taosArrayGetSize(pInfo->orderColumnList);
+
+    bool samegroup = true;
+    if (pInfo->hasPrev) {
+      for (int32_t i = 0; i < numOfCols; ++i) {
+        SColIndex       *pIndex = taosArrayGet(pInfo->orderColumnList, i);
+        SColumnInfoData *pColInfoData = taosArrayGet(pBlock->pDataBlock, pIndex->colIndex);
+
+        SColumnInfo *pColInfo = &pColInfoData->info;
+
+        char   *d = rowIndex * pColInfo->bytes + (char *)pColInfoData->pData;
+        int32_t ret = columnValueAscendingComparator(pInfo->prevRow[i], d, pColInfo->type, pColInfo->bytes);
+        if (ret != 0) {  // it is a new group
+          samegroup = false;
+          break;
+        }
+      }
     }
 
-    while(1) {
-      if (*newgroup) {
-        pInfo->currentGroupOffset -= 1;
-        *newgroup = false;
+    if (!samegroup || !pInfo->hasPrev) {
+      pInfo->ignoreCurrentGroup = false;
+      savePrevOrderColumns(pInfo->prevRow, pInfo->orderColumnList, pBlock, rowIndex, &pInfo->hasPrev);
+
+      pInfo->currentOffset = pInfo->limit.offset;  // reset the offset value for a new group
+      pInfo->rowsTotal = 0;
+
+      if (pInfo->currentGroupOffset > 0) {
+        pInfo->ignoreCurrentGroup = true;
+        pInfo->currentGroupOffset -= 1;  // now we are in the next group data
+        rowIndex += 1;
+        continue;
+      }
+
+      // A new group has arrived according to the result rows, and the group limitation has already reached.
+      // Let's jump out of current loop and return immediately.
+      if (pInfo->slimit.limit >= 0 && pInfo->groupTotal >= pInfo->slimit.limit) {
+        setQueryStatus(pOperator->pRuntimeEnv, QUERY_COMPLETED);
+        pOperator->status = OP_EXEC_DONE;
+        return BLOCK_NO_GROUP;
       }
 
-      while ((*newgroup) == false) {
-        publishOperatorProfEvent(pOperator->upstream[0], QUERY_PROF_BEFORE_OPERATOR_EXEC);
-        pBlock = pOperator->upstream[0]->exec(pOperator->upstream[0], newgroup);
-        publishOperatorProfEvent(pOperator->upstream[0], QUERY_PROF_AFTER_OPERATOR_EXEC);
+      pInfo->groupTotal += 1;
 
-        if (pBlock == NULL) {
-          setQueryStatus(pOperator->pRuntimeEnv, QUERY_COMPLETED);
-          pOperator->status = OP_EXEC_DONE;
-          return NULL;
-        }
+      // data in current group not allowed, return if current result does not belong to the previous group.And there
+      // are results exists in current SSDataBlock
+      if (!pInfo->multigroupResult && !samegroup && pInfo->pRes->info.rows > 0) {
+        return BLOCK_NEW_GROUP;
       }
 
-      // now we have got the first data block of the next group.
-      if (pInfo->currentGroupOffset == 0) {
-        return pBlock;
+      doHandleDataInCurrentGroup(pInfo, pBlock, rowIndex);
+
+    } else {  // handle the offset in the same group
+      // All the data in current group needs to be discarded, due to the limit parameter in the SQL statement
+      if (pInfo->ignoreCurrentGroup) {
+        rowIndex += 1;
+        continue;
       }
+
+      doHandleDataInCurrentGroup(pInfo, pBlock, rowIndex);
     }
 
-  return NULL;
+    rowIndex += 1;
+  }
+
+  return BLOCK_SAME_GROUP;
 }
 
 SSDataBlock* doSLimit(void* param, bool* newgroup) {
@@ -1093,63 +1105,41 @@ SSDataBlock* doSLimit(void* param, bool* newgroup) {
   }
 
   SSLimitOperatorInfo *pInfo = pOperator->info;
+  pInfo->pRes->info.rows = 0;
 
-  SSDataBlock *pBlock = NULL;
-  while (1) {
-    pBlock = skipGroupBlock(pOperator, newgroup);
-    if (pBlock == NULL) {
-      setQueryStatus(pOperator->pRuntimeEnv, QUERY_COMPLETED);
-      pOperator->status = OP_EXEC_DONE;
-      return NULL;
-    }
-
-    if (*newgroup) {  // a new group arrives
-      pInfo->groupTotal += 1;
-      pInfo->rowsTotal = 0;
-      pInfo->currentOffset = pInfo->limit.offset;
-    }
+  if (pInfo->pPrevBlock != NULL) {
+    ensureOutputBuf(pInfo, pInfo->pRes, pInfo->pPrevBlock->info.rows);
+    int32_t ret = doSlimitImpl(pOperator, pInfo, pInfo->pPrevBlock);
+    assert(ret != BLOCK_NEW_GROUP);
 
-    assert(pInfo->currentGroupOffset == 0);
-
-    if (pInfo->currentOffset >= pBlock->info.rows) {
-      pInfo->currentOffset -= pBlock->info.rows;
-    } else {
-      if (pInfo->currentOffset == 0) {
-        break;
-      }
-
-      int32_t remain = (int32_t)(pBlock->info.rows - pInfo->currentOffset);
-      pBlock->info.rows = remain;
+    pInfo->pPrevBlock = NULL;
+  }
 
-      // move the remain rows of this data block to the front.
-      for (int32_t i = 0; i < pBlock->info.numOfCols; ++i) {
-        SColumnInfoData *pColInfoData = taosArrayGet(pBlock->pDataBlock, i);
+  assert(pInfo->currentGroupOffset >= 0);
 
-        int16_t bytes = pColInfoData->info.bytes;
-        memmove(pColInfoData->pData, pColInfoData->pData + bytes * pInfo->currentOffset, remain * bytes);
-      }
+  while(1) {
+    publishOperatorProfEvent(pOperator->upstream[0], QUERY_PROF_BEFORE_OPERATOR_EXEC);
+    SSDataBlock *pBlock = pOperator->upstream[0]->exec(pOperator->upstream[0], newgroup);
+    publishOperatorProfEvent(pOperator->upstream[0], QUERY_PROF_AFTER_OPERATOR_EXEC);
 
-      pInfo->currentOffset = 0;
-      break;
+    if (pBlock == NULL) {
+      return pInfo->pRes->info.rows == 0 ? NULL : pInfo->pRes;
     }
-  }
-
-  if (pInfo->slimit.limit > 0 && pInfo->groupTotal > pInfo->slimit.limit) {  // reach the group limit, abort
-    return NULL;
-  }
 
-  if (pInfo->limit.limit > 0 && (pInfo->rowsTotal + pBlock->info.rows >= pInfo->limit.limit)) {
-    pBlock->info.rows = (int32_t)(pInfo->limit.limit - pInfo->rowsTotal);
-    pInfo->rowsTotal = pInfo->limit.limit;
+    ensureOutputBuf(pInfo, pInfo->pRes, pBlock->info.rows);
+    int32_t ret = doSlimitImpl(pOperator, pInfo, pBlock);
+    if (ret == BLOCK_NEW_GROUP) {
+      pInfo->pPrevBlock = pBlock;
+      return pInfo->pRes;
+    }
 
-    if (pInfo->slimit.limit > 0 && pInfo->groupTotal >= pInfo->slimit.limit) {
-      pOperator->status = OP_EXEC_DONE;
+    if (pOperator->status == OP_EXEC_DONE) {
+      return pInfo->pRes->info.rows == 0 ? NULL : pInfo->pRes;
     }
 
-    //    setQueryStatus(pOperator->pRuntimeEnv, QUERY_COMPLETED);
-  } else {
-    pInfo->rowsTotal += pBlock->info.rows;
+    // now the number of rows in current group is enough, let's return to the invoke function
+    if (pInfo->pRes->info.rows > pInfo->threshold) {
+      return pInfo->pRes;
+    }
   }
-
-  return pBlock;
 }

src/client/src/tscSQLParser.c

@@ -925,7 +925,6 @@ int32_t tscValidateSqlInfo(SSqlObj* pSql, struct SSqlInfo* pInfo) {
           pQueryInfo = pCmd->active;
           pQueryInfo->pUdfInfo = pUdfInfo;
           pQueryInfo->udfCopy = true;
-
         }
       }
 
@@ -6916,7 +6915,6 @@ static int32_t doAddGroupbyColumnsOnDemand(SSqlCmd* pCmd, SQueryInfo* pQueryInfo
   const char* msg1 = "interval not allowed in group by normal column";
 
   STableMetaInfo* pTableMetaInfo = tscGetMetaInfo(pQueryInfo, 0);
-
   SSchema* pSchema = tscGetTableSchema(pTableMetaInfo->pTableMeta);
 
   SSchema* tagSchema = NULL;
@@ -8438,6 +8436,7 @@ static int32_t doValidateSubquery(SSqlNode* pSqlNode, int32_t index, SSqlObj* pS
   if (taosArrayGetSize(subInfo->pSubquery) >= 2) {
     return invalidOperationMsg(msgBuf, "not support union in subquery");
   }
+
   SQueryInfo* pSub = calloc(1, sizeof(SQueryInfo));
   tscInitQueryInfo(pSub);
 
@@ -8460,6 +8459,7 @@ static int32_t doValidateSubquery(SSqlNode* pSqlNode, int32_t index, SSqlObj* pS
   if (pTableMetaInfo1 == NULL) {
     return TSDB_CODE_TSC_OUT_OF_MEMORY;
   }
+
   pTableMetaInfo1->pTableMeta        = extractTempTableMetaFromSubquery(pSub);
   pTableMetaInfo1->tableMetaCapacity = tscGetTableMetaSize(pTableMetaInfo1->pTableMeta);
 
@@ -8543,7 +8543,7 @@ int32_t validateSqlNode(SSqlObj* pSql, SSqlNode* pSqlNode, SQueryInfo* pQueryInf
       // check if there is 3 level select
       SRelElementPair* subInfo = taosArrayGet(pSqlNode->from->list, i);
       SSqlNode* p = taosArrayGetP(subInfo->pSubquery, 0);
-      if (p->from->type == SQL_NODE_FROM_SUBQUERY){
+      if (p->from->type == SQL_NODE_FROM_SUBQUERY) {
         return invalidOperationMsg(tscGetErrorMsgPayload(pCmd), msg9);
       }
 
@@ -8635,6 +8635,15 @@ int32_t validateSqlNode(SSqlObj* pSql, SSqlNode* pSqlNode, SQueryInfo* pQueryInf
       }
     }
 
+    // disable group result mixed up if interval/session window query exists.
+    if (isTimeWindowQuery(pQueryInfo)) {
+      size_t num = taosArrayGetSize(pQueryInfo->pUpstream);
+      for(int32_t i = 0; i < num; ++i) {
+        SQueryInfo* pUp = taosArrayGetP(pQueryInfo->pUpstream, i);
+        pUp->multigroupResult = false;
+      }
+    }
+
     // parse the having clause in the first place
     int32_t joinQuery = (pSqlNode->from != NULL && taosArrayGetSize(pSqlNode->from->list) > 1);
     if (validateHavingClause(pQueryInfo, pSqlNode->pHaving, pCmd, pSqlNode->pSelNodeList, joinQuery, timeWindowQuery) !=

src/client/src/tscUtil.c

@@ -3078,6 +3078,7 @@ void tscInitQueryInfo(SQueryInfo* pQueryInfo) {
   pQueryInfo->slimit.offset  = 0;
   pQueryInfo->pUpstream      = taosArrayInit(4, POINTER_BYTES);
   pQueryInfo->window         = TSWINDOW_INITIALIZER;
+  pQueryInfo->multigroupResult = true;
 }
 
 int32_t tscAddQueryInfo(SSqlCmd* pCmd) {
@@ -3089,7 +3090,6 @@ int32_t tscAddQueryInfo(SSqlCmd* pCmd) {
   }
 
   tscInitQueryInfo(pQueryInfo);
-
   pQueryInfo->msg = pCmd->payload;  // pointer to the parent error message buffer
 
   if (pCmd->pQueryInfo == NULL) {
@@ -3171,6 +3171,7 @@ int32_t tscQueryInfoCopy(SQueryInfo* pQueryInfo, const SQueryInfo* pSrc) {
   pQueryInfo->window         = pSrc->window;
   pQueryInfo->sessionWindow  = pSrc->sessionWindow;
   pQueryInfo->pTableMetaInfo = NULL;
+  pQueryInfo->multigroupResult = pSrc->multigroupResult;
 
   pQueryInfo->bufLen         = pSrc->bufLen;
   pQueryInfo->orderProjectQuery = pSrc->orderProjectQuery;
@@ -3570,7 +3571,7 @@ SSqlObj* createSubqueryObj(SSqlObj* pSql, int16_t tableIndex, __async_cb_func_t
   pNewQueryInfo->pTableMetaInfo = NULL;
   pNewQueryInfo->bufLen = pQueryInfo->bufLen;
   pNewQueryInfo->buf = malloc(pQueryInfo->bufLen);
-
+  pNewQueryInfo->multigroupResult = pQueryInfo->multigroupResult;
 
   pNewQueryInfo->distinct =  pQueryInfo->distinct;
   if (pNewQueryInfo->buf == NULL) {
@@ -3795,12 +3796,12 @@ static void tscSubqueryCompleteCallback(void* param, TAOS_RES* tres, int code) {
 
     if (code && !((code == TSDB_CODE_TDB_INVALID_TABLE_ID || code == TSDB_CODE_VND_INVALID_VGROUP_ID) && pParentSql->retry < pParentSql->maxRetry)) {
       pParentSql->res.code = code;
-      
+
       tscAsyncResultOnError(pParentSql);
       return;
     }
 
-    tscFreeSubobj(pParentSql);    
+    tscFreeSubobj(pParentSql);
     tfree(pParentSql->pSubs);
 
     pParentSql->res.code = TSDB_CODE_SUCCESS;
@@ -3809,9 +3810,9 @@ static void tscSubqueryCompleteCallback(void* param, TAOS_RES* tres, int code) {
     tscDebug("0x%"PRIx64" retry parse sql and send query, prev error: %s, retry:%d", pParentSql->self,
              tstrerror(code), pParentSql->retry);
 
-    
+
     tscResetSqlCmd(&pParentSql->cmd, true);
-    
+
     code = tsParseSql(pParentSql, true);
     if (code == TSDB_CODE_TSC_ACTION_IN_PROGRESS) {
       return;
@@ -3880,7 +3881,7 @@ void executeQuery(SSqlObj* pSql, SQueryInfo* pQueryInfo) {
       pNew->maxRetry  = pSql->maxRetry;
 
       pNew->cmd.resColumnId = TSDB_RES_COL_ID;
-      
+
       tsem_init(&pNew->rspSem, 0, 0);
 
       SRetrieveSupport* ps = calloc(1, sizeof(SRetrieveSupport));  // todo use object id
@@ -4416,7 +4417,7 @@ int32_t tscCreateTableMetaFromSTableMeta(STableMeta** ppChild, const char* name,
   size_t      sz   = 0;
   STableMeta* pChild = *ppChild;
   STableMeta* pChild1;
-   
+
   taosHashGetCloneExt(tscTableMetaMap, pChild->sTableName, strnlen(pChild->sTableName, TSDB_TABLE_FNAME_LEN), NULL, (void **)&p, &sz);
 
   // tableMeta exists, build child table meta according to the super table meta
@@ -4427,9 +4428,9 @@ int32_t tscCreateTableMetaFromSTableMeta(STableMeta** ppChild, const char* name,
     int32_t tableMetaSize =  sizeof(STableMeta)  + totalBytes;
     if (*tableMetaCapacity < tableMetaSize) {
       pChild1 = realloc(pChild, tableMetaSize);
-      if(pChild1 == NULL) 
+      if(pChild1 == NULL)
         return -1;
-      pChild = pChild1; 
+      pChild = pChild1;
       *tableMetaCapacity = (size_t)tableMetaSize;
     }
 
@@ -4701,6 +4702,7 @@ int32_t tscCreateQueryFromQueryInfo(SQueryInfo* pQueryInfo, SQueryAttr* pQueryAt
   pQueryAttr->distinct          = pQueryInfo->distinct;
   pQueryAttr->sw                = pQueryInfo->sessionWindow;
   pQueryAttr->stateWindow       = pQueryInfo->stateWindow;
+  pQueryAttr->multigroupResult  = pQueryInfo->multigroupResult;
 
   pQueryAttr->numOfCols         = numOfCols;
   pQueryAttr->numOfOutput       = numOfOutput;

src/query/inc/qExecutor.h

@@ -219,6 +219,7 @@ typedef struct SQueryAttr {
   bool             distinct;         // distinct  query or not
   bool             stateWindow;       // window State on sub/normal table
   bool             createFilterOperator; // if filter operator is needed
+  bool             multigroupResult; // multigroup result can exist in one SSDataBlock
   int32_t          interBufSize;     // intermediate buffer sizse
 
   int32_t          havingNum;        // having expr number
@@ -460,16 +461,23 @@ typedef struct SLimitOperatorInfo {
 } SLimitOperatorInfo;
 
 typedef struct SSLimitOperatorInfo {
-  int64_t   groupTotal;
-  int64_t   currentGroupOffset;
-
-  int64_t   rowsTotal;
-  int64_t   currentOffset;
-  SLimitVal limit;
-  SLimitVal slimit;
-
-  char    **prevRow;
-  SArray   *orderColumnList;
+  int64_t      groupTotal;
+  int64_t      currentGroupOffset;
+
+  int64_t      rowsTotal;
+  int64_t      currentOffset;
+  SLimitVal    limit;
+  SLimitVal    slimit;
+
+  char       **prevRow;
+  SArray      *orderColumnList;
+  bool         hasPrev;
+  bool         ignoreCurrentGroup;
+  bool         multigroupResult;
+  SSDataBlock *pRes;   // result buffer
+  SSDataBlock *pPrevBlock;
+  int64_t      capacity;
+  int64_t      threshold;
 } SSLimitOperatorInfo;
 
 typedef struct SFilterOperatorInfo {
@@ -481,8 +489,9 @@ typedef struct SFillOperatorInfo {
   SFillInfo   *pFillInfo;
   SSDataBlock *pRes;
   int64_t      totalInputRows;
-
+  void       **p;
   SSDataBlock *existNewGroupBlock;
+  bool         multigroupResult;
 } SFillOperatorInfo;
 
 typedef struct SGroupbyOperatorInfo {
@@ -544,9 +553,9 @@ typedef struct SMultiwayMergeInfo {
   bool                 hasDataBlockForNewGroup;
   SSDataBlock         *pExistBlock;
 
-  bool                 hasPrev;
-  bool                 groupMix;
   SArray              *udfInfo;
+  bool                 hasPrev;
+  bool                 multiGroupResults;
 } SMultiwayMergeInfo;
 
 // todo support the disk-based sort
@@ -568,7 +577,7 @@ SOperatorInfo* createLimitOperatorInfo(SQueryRuntimeEnv* pRuntimeEnv, SOperatorI
 SOperatorInfo* createTimeIntervalOperatorInfo(SQueryRuntimeEnv* pRuntimeEnv, SOperatorInfo* upstream, SExprInfo* pExpr, int32_t numOfOutput);
 SOperatorInfo* createAllTimeIntervalOperatorInfo(SQueryRuntimeEnv* pRuntimeEnv, SOperatorInfo* upstream, SExprInfo* pExpr, int32_t numOfOutput);
 SOperatorInfo* createSWindowOperatorInfo(SQueryRuntimeEnv* pRuntimeEnv, SOperatorInfo* upstream, SExprInfo* pExpr, int32_t numOfOutput);
-SOperatorInfo* createFillOperatorInfo(SQueryRuntimeEnv* pRuntimeEnv, SOperatorInfo* upstream, SExprInfo* pExpr, int32_t numOfOutput);
+SOperatorInfo* createFillOperatorInfo(SQueryRuntimeEnv* pRuntimeEnv, SOperatorInfo* upstream, SExprInfo* pExpr, int32_t numOfOutput, bool multigroupResult);
 SOperatorInfo* createGroupbyOperatorInfo(SQueryRuntimeEnv* pRuntimeEnv, SOperatorInfo* upstream, SExprInfo* pExpr, int32_t numOfOutput);
 SOperatorInfo* createMultiTableAggOperatorInfo(SQueryRuntimeEnv* pRuntimeEnv, SOperatorInfo* upstream, SExprInfo* pExpr, int32_t numOfOutput);
 SOperatorInfo* createMultiTableTimeIntervalOperatorInfo(SQueryRuntimeEnv* pRuntimeEnv, SOperatorInfo* upstream, SExprInfo* pExpr, int32_t numOfOutput);
@@ -577,10 +586,10 @@ SOperatorInfo* createTagScanOperatorInfo(SQueryRuntimeEnv* pRuntimeEnv, SExprInf
 SOperatorInfo* createDistinctOperatorInfo(SQueryRuntimeEnv* pRuntimeEnv, SOperatorInfo* upstream, SExprInfo* pExpr, int32_t numOfOutput);
 SOperatorInfo* createTableBlockInfoScanOperator(void* pTsdbQueryHandle, SQueryRuntimeEnv* pRuntimeEnv);
 SOperatorInfo* createMultiwaySortOperatorInfo(SQueryRuntimeEnv* pRuntimeEnv, SExprInfo* pExpr, int32_t numOfOutput,
-                                              int32_t numOfRows, void* merger, bool groupMix);
-SOperatorInfo* createGlobalAggregateOperatorInfo(SQueryRuntimeEnv* pRuntimeEnv, SOperatorInfo* upstream, SExprInfo* pExpr, int32_t numOfOutput, void* param, SArray* pUdfInfo);
+                                              int32_t numOfRows, void* merger);
+SOperatorInfo* createGlobalAggregateOperatorInfo(SQueryRuntimeEnv* pRuntimeEnv, SOperatorInfo* upstream, SExprInfo* pExpr, int32_t numOfOutput, void* param, SArray* pUdfInfo, bool groupResultMixedUp);
 SOperatorInfo* createStatewindowOperatorInfo(SQueryRuntimeEnv* pRuntimeEnv, SOperatorInfo* upstream, SExprInfo* pExpr, int32_t numOfOutput);
-SOperatorInfo* createSLimitOperatorInfo(SQueryRuntimeEnv* pRuntimeEnv, SOperatorInfo* upstream, SExprInfo* pExpr, int32_t numOfOutput, void* merger);
+SOperatorInfo* createSLimitOperatorInfo(SQueryRuntimeEnv* pRuntimeEnv, SOperatorInfo* upstream, SExprInfo* pExpr, int32_t numOfOutput, void* merger, bool multigroupResult);
 SOperatorInfo* createFilterOperatorInfo(SQueryRuntimeEnv* pRuntimeEnv, SOperatorInfo* upstream, SExprInfo* pExpr,
                                         int32_t numOfOutput, SColumnInfo* pCols, int32_t numOfFilter);
 

src/query/inc/qTableMeta.h

@@ -148,6 +148,7 @@ typedef struct SQueryInfo {
   bool               orderProjectQuery;
   bool               stateWindow;
   bool               globalMerge;
+  bool               multigroupResult;
 } SQueryInfo;
 
 /**

src/query/src/qFill.c

@@ -430,7 +430,7 @@ void taosFillSetInputDataBlock(SFillInfo* pFillInfo, const SSDataBlock* pInput)
     SColumnInfoData* pColData = taosArrayGet(pInput->pDataBlock, i);
     pFillInfo->pData[i] = pColData->pData;
 
-    if (TSDB_COL_IS_TAG(pCol->flag)/* || IS_VAR_DATA_TYPE(pCol->col.type)*/) {  // copy the tag value to tag value buffer
+    if (TSDB_COL_IS_TAG(pCol->flag)) {  // copy the tag value to tag value buffer
       SFillTagColInfo* pTag = &pFillInfo->pTags[pCol->tagIndex];
       assert (pTag->col.colId == pCol->col.colId);
       memcpy(pTag->tagVal, pColData->pData, pCol->col.bytes);  // TODO not memcpy??