From 3e5ab6b54cecbb658f0028fcb11af85969183e8a Mon Sep 17 00:00:00 2001
From: Xiaoyu Wang <xiaoyuwang@taosdata.com>
Date: Tue, 31 May 2022 11:29:10 +0800
Subject: [PATCH 01/34] feat: add physical plannode of indefinite rows func

---
 include/libs/nodes/nodes.h                   |   2 +
 include/libs/nodes/plannodes.h               |  11 ++
 source/libs/function/src/builtins.c          | 109 ++-----------------
 source/libs/nodes/src/nodesCloneFuncs.c      |   8 ++
 source/libs/nodes/src/nodesCodeFuncs.c       |  67 ++++++++++++
 source/libs/nodes/src/nodesTraverseFuncs.c   |   2 +
 source/libs/nodes/src/nodesUtilFuncs.c       |   4 +
 source/libs/planner/src/planLogicCreater.c   |  34 ++++++
 source/libs/planner/src/planPhysiCreater.c   |  42 ++++++-
 source/libs/planner/test/planProjectTest.cpp |  34 ++++++
 10 files changed, 214 insertions(+), 99 deletions(-)
 create mode 100644 source/libs/planner/test/planProjectTest.cpp

diff --git a/include/libs/nodes/nodes.h b/include/libs/nodes/nodes.h
index 3860266725..639a1173cd 100644
--- a/include/libs/nodes/nodes.h
+++ b/include/libs/nodes/nodes.h
@@ -194,6 +194,7 @@ typedef enum ENodeType {
   QUERY_NODE_LOGIC_PLAN_FILL,
   QUERY_NODE_LOGIC_PLAN_SORT,
   QUERY_NODE_LOGIC_PLAN_PARTITION,
+  QUERY_NODE_LOGIC_PLAN_INDEF_ROWS_FUNC,
   QUERY_NODE_LOGIC_SUBPLAN,
   QUERY_NODE_LOGIC_PLAN,
 
@@ -217,6 +218,7 @@ typedef enum ENodeType {
   QUERY_NODE_PHYSICAL_PLAN_STREAM_FINAL_SESSION_WINDOW,
   QUERY_NODE_PHYSICAL_PLAN_STATE_WINDOW,
   QUERY_NODE_PHYSICAL_PLAN_PARTITION,
+  QUERY_NODE_PHYSICAL_PLAN_INDEF_ROWS_FUNC,
   QUERY_NODE_PHYSICAL_PLAN_DISPATCH,
   QUERY_NODE_PHYSICAL_PLAN_INSERT,
   QUERY_NODE_PHYSICAL_SUBPLAN,
diff --git a/include/libs/nodes/plannodes.h b/include/libs/nodes/plannodes.h
index 2648a468dd..56a1343dcb 100644
--- a/include/libs/nodes/plannodes.h
+++ b/include/libs/nodes/plannodes.h
@@ -84,6 +84,11 @@ typedef struct SProjectLogicNode {
   int64_t    soffset;
 } SProjectLogicNode;
 
+typedef struct SIndefRowsFuncLogicNode {
+  SLogicNode node;
+  SNodeList* pVectorFuncs;
+} SIndefRowsFuncLogicNode;
+
 typedef struct SVnodeModifLogicNode {
   SLogicNode     node;
   int32_t        msgType;
@@ -236,6 +241,12 @@ typedef struct SProjectPhysiNode {
   int64_t    soffset;
 } SProjectPhysiNode;
 
+typedef struct SIndefRowsFuncPhysiNode {
+  SPhysiNode node;
+  SNodeList* pExprs;
+  SNodeList* pVectorFuncs;
+} SIndefRowsFuncPhysiNode;
+
 typedef struct SJoinPhysiNode {
   SPhysiNode node;
   EJoinType  joinType;
diff --git a/source/libs/function/src/builtins.c b/source/libs/function/src/builtins.c
index c9d556c86b..831af33add 100644
--- a/source/libs/function/src/builtins.c
+++ b/source/libs/function/src/builtins.c
@@ -156,14 +156,7 @@ static int32_t translatePercentile(SFunctionNode* pFunc, char* pErrBuf, int32_t
     return invaildFuncParaNumErrMsg(pErrBuf, len, pFunc->functionName);
   }
 
-  //param0
-  SNode* pParamNode0 = nodesListGetNode(pFunc->pParameterList, 0);
-  if (nodeType(pParamNode0) != QUERY_NODE_COLUMN) {
-    return buildFuncErrMsg(pErrBuf, len, TSDB_CODE_FUNC_FUNTION_ERROR,
-                           "The first parameter of PERCENTILE function can only be column");
-  }
-
-  //param1
+  // param1
   SValueNode* pValue = (SValueNode*)nodesListGetNode(pFunc->pParameterList, 1);
 
   if (pValue->datum.i < 0 || pValue->datum.i > 100) {
@@ -178,7 +171,7 @@ static int32_t translatePercentile(SFunctionNode* pFunc, char* pErrBuf, int32_t
     return invaildFuncParaTypeErrMsg(pErrBuf, len, pFunc->functionName);
   }
 
-  //set result type
+  // set result type
   pFunc->node.resType = (SDataType){.bytes = tDataTypes[TSDB_DATA_TYPE_DOUBLE].bytes, .type = TSDB_DATA_TYPE_DOUBLE};
   return TSDB_CODE_SUCCESS;
 }
@@ -197,14 +190,7 @@ static int32_t translateApercentile(SFunctionNode* pFunc, char* pErrBuf, int32_t
     return invaildFuncParaNumErrMsg(pErrBuf, len, pFunc->functionName);
   }
 
-  //param0
-  SNode* pParamNode0 = nodesListGetNode(pFunc->pParameterList, 0);
-  if (nodeType(pParamNode0) != QUERY_NODE_COLUMN) {
-    return buildFuncErrMsg(pErrBuf, len, TSDB_CODE_FUNC_FUNTION_ERROR,
-                           "The first parameter of APERCENTILE function can only be column");
-  }
-
-  //param1
+  // param1
   SNode* pParamNode1 = nodesListGetNode(pFunc->pParameterList, 1);
   if (nodeType(pParamNode1) != QUERY_NODE_VALUE) {
     return invaildFuncParaTypeErrMsg(pErrBuf, len, pFunc->functionName);
@@ -223,7 +209,7 @@ static int32_t translateApercentile(SFunctionNode* pFunc, char* pErrBuf, int32_t
     return invaildFuncParaTypeErrMsg(pErrBuf, len, pFunc->functionName);
   }
 
-  //param2
+  // param2
   if (3 == numOfParams) {
     uint8_t para3Type = ((SExprNode*)nodesListGetNode(pFunc->pParameterList, 2))->resType.type;
     if (!IS_VAR_DATA_TYPE(para3Type)) {
@@ -263,14 +249,7 @@ static int32_t translateTop(SFunctionNode* pFunc, char* pErrBuf, int32_t len) {
     return invaildFuncParaTypeErrMsg(pErrBuf, len, pFunc->functionName);
   }
 
-  //param0
-  SNode* pParamNode0 = nodesListGetNode(pFunc->pParameterList, 0);
-  if (nodeType(pParamNode0) != QUERY_NODE_COLUMN) {
-    return buildFuncErrMsg(pErrBuf, len, TSDB_CODE_FUNC_FUNTION_ERROR,
-                           "The first parameter of TOP/BOTTOM function can only be column");
-  }
-
-  //param1
+  // param1
   SNode* pParamNode1 = nodesListGetNode(pFunc->pParameterList, 1);
   if (nodeType(pParamNode1) != QUERY_NODE_VALUE) {
     return invaildFuncParaTypeErrMsg(pErrBuf, len, pFunc->functionName);
@@ -287,7 +266,7 @@ static int32_t translateTop(SFunctionNode* pFunc, char* pErrBuf, int32_t len) {
 
   pValue->notReserved = true;
 
-  //set result type
+  // set result type
   SDataType* pType = &((SExprNode*)nodesListGetNode(pFunc->pParameterList, 0))->resType;
   pFunc->node.resType = (SDataType){.bytes = pType->bytes, .type = pType->type};
   return TSDB_CODE_SUCCESS;
@@ -317,13 +296,6 @@ static int32_t translateElapsed(SFunctionNode* pFunc, char* pErrBuf, int32_t len
     return invaildFuncParaNumErrMsg(pErrBuf, len, pFunc->functionName);
   }
 
-  // param0
-  SNode* pParaNode0 = nodesListGetNode(pFunc->pParameterList, 0);
-  if (QUERY_NODE_COLUMN != nodeType(pParaNode0)) {
-    return buildFuncErrMsg(pErrBuf, len, TSDB_CODE_FUNC_FUNTION_ERROR,
-                           "The first parameter of ELAPSED function can only be column");
-  }
-
   uint8_t paraType = ((SExprNode*)nodesListGetNode(pFunc->pParameterList, 0))->resType.type;
   if (TSDB_DATA_TYPE_TIMESTAMP != paraType) {
     return invaildFuncParaTypeErrMsg(pErrBuf, len, pFunc->functionName);
@@ -389,13 +361,6 @@ static int32_t translateHistogram(SFunctionNode* pFunc, char* pErrBuf, int32_t l
     return invaildFuncParaNumErrMsg(pErrBuf, len, pFunc->functionName);
   }
 
-  // param0
-  SNode* pParaNode0 = nodesListGetNode(pFunc->pParameterList, 0);
-  if (QUERY_NODE_COLUMN != nodeType(pParaNode0)) {
-    return buildFuncErrMsg(pErrBuf, len, TSDB_CODE_FUNC_FUNTION_ERROR,
-                           "The first parameter of HISTOGRAM function can only be column");
-  }
-
   uint8_t colType = ((SExprNode*)nodesListGetNode(pFunc->pParameterList, 0))->resType.type;
   if (!IS_NUMERIC_TYPE(colType)) {
     return invaildFuncParaTypeErrMsg(pErrBuf, len, pFunc->functionName);
@@ -428,12 +393,6 @@ static int32_t translateHLL(SFunctionNode* pFunc, char* pErrBuf, int32_t len) {
     return invaildFuncParaNumErrMsg(pErrBuf, len, pFunc->functionName);
   }
 
-  SNode* pPara = nodesListGetNode(pFunc->pParameterList, 0);
-  if (QUERY_NODE_COLUMN != nodeType(pPara)) {
-    return buildFuncErrMsg(pErrBuf, len, TSDB_CODE_FUNC_FUNTION_ERROR,
-                           "The input parameter of HYPERLOGLOG function can only be column");
-  }
-
   pFunc->node.resType = (SDataType){.bytes = tDataTypes[TSDB_DATA_TYPE_UBIGINT].bytes, .type = TSDB_DATA_TYPE_UBIGINT};
   return TSDB_CODE_SUCCESS;
 }
@@ -444,12 +403,6 @@ static int32_t translateStateCount(SFunctionNode* pFunc, char* pErrBuf, int32_t
     return invaildFuncParaNumErrMsg(pErrBuf, len, pFunc->functionName);
   }
 
-  // param0
-  SNode* pParaNode0 = nodesListGetNode(pFunc->pParameterList, 0);
-  if (QUERY_NODE_COLUMN != nodeType(pParaNode0)) {
-    return buildFuncErrMsg(pErrBuf, len, TSDB_CODE_FUNC_FUNTION_ERROR,
-                           "The input parameter of STATECOUNT function can only be column");
-  }
   uint8_t colType = ((SExprNode*)nodesListGetNode(pFunc->pParameterList, 0))->resType.type;
   if (!IS_NUMERIC_TYPE(colType)) {
     return invaildFuncParaTypeErrMsg(pErrBuf, len, pFunc->functionName);
@@ -484,12 +437,6 @@ static int32_t translateStateDuration(SFunctionNode* pFunc, char* pErrBuf, int32
     return invaildFuncParaNumErrMsg(pErrBuf, len, pFunc->functionName);
   }
 
-  // param0
-  SNode* pParaNode0 = nodesListGetNode(pFunc->pParameterList, 0);
-  if (QUERY_NODE_COLUMN != nodeType(pParaNode0)) {
-    return buildFuncErrMsg(pErrBuf, len, TSDB_CODE_FUNC_FUNTION_ERROR,
-                           "The input parameter of STATEDURATION function can only be column");
-  }
   uint8_t colType = ((SExprNode*)nodesListGetNode(pFunc->pParameterList, 0))->resType.type;
   if (!IS_NUMERIC_TYPE(colType)) {
     return invaildFuncParaTypeErrMsg(pErrBuf, len, pFunc->functionName);
@@ -528,12 +475,6 @@ static int32_t translateCsum(SFunctionNode* pFunc, char* pErrBuf, int32_t len) {
     return invaildFuncParaNumErrMsg(pErrBuf, len, pFunc->functionName);
   }
 
-  SNode* pPara = nodesListGetNode(pFunc->pParameterList, 0);
-  if (QUERY_NODE_COLUMN != nodeType(pPara)) {
-    return buildFuncErrMsg(pErrBuf, len, TSDB_CODE_FUNC_FUNTION_ERROR,
-                           "The input parameter of CSUM function can only be column");
-  }
-
   uint8_t colType = ((SExprNode*)nodesListGetNode(pFunc->pParameterList, 0))->resType.type;
   uint8_t resType;
   if (!IS_NUMERIC_TYPE(colType)) {
@@ -559,13 +500,6 @@ static int32_t translateMavg(SFunctionNode* pFunc, char* pErrBuf, int32_t len) {
     return invaildFuncParaNumErrMsg(pErrBuf, len, pFunc->functionName);
   }
 
-  // param0
-  SNode* pParaNode0 = nodesListGetNode(pFunc->pParameterList, 0);
-  if (QUERY_NODE_COLUMN != nodeType(pParaNode0)) {
-    return buildFuncErrMsg(pErrBuf, len, TSDB_CODE_FUNC_FUNTION_ERROR,
-                           "The first parameter of MAVG function can only be column");
-  }
-
   uint8_t colType = ((SExprNode*)nodesListGetNode(pFunc->pParameterList, 0))->resType.type;
 
   // param1
@@ -595,13 +529,6 @@ static int32_t translateSample(SFunctionNode* pFunc, char* pErrBuf, int32_t len)
     return invaildFuncParaNumErrMsg(pErrBuf, len, pFunc->functionName);
   }
 
-  // param0
-  SNode* pParamNode0 = nodesListGetNode(pFunc->pParameterList, 0);
-  if (QUERY_NODE_COLUMN != nodeType(pParamNode0)) {
-    return buildFuncErrMsg(pErrBuf, len, TSDB_CODE_FUNC_FUNTION_ERROR,
-                           "The first parameter of SAMPLE function can only be column");
-  }
-
   SExprNode* pCol = (SExprNode*)nodesListGetNode(pFunc->pParameterList, 0);
   uint8_t    colType = pCol->resType.type;
 
@@ -639,12 +566,6 @@ static int32_t translateTail(SFunctionNode* pFunc, char* pErrBuf, int32_t len) {
     return invaildFuncParaNumErrMsg(pErrBuf, len, pFunc->functionName);
   }
 
-  // param0
-  SNode* pPara = nodesListGetNode(pFunc->pParameterList, 0);
-  if (QUERY_NODE_COLUMN != nodeType(pPara)) {
-    return buildFuncErrMsg(pErrBuf, len, TSDB_CODE_FUNC_FUNTION_ERROR,
-                           "The first parameter of TAIL function can only be column");
-  }
   SExprNode* pCol = (SExprNode*)nodesListGetNode(pFunc->pParameterList, 0);
   uint8_t    colType = pCol->resType.type;
 
@@ -659,8 +580,8 @@ static int32_t translateTail(SFunctionNode* pFunc, char* pErrBuf, int32_t len) {
 
     if (pValue->datum.i < ((i > 1) ? 0 : 1) || pValue->datum.i > 100) {
       return buildFuncErrMsg(pErrBuf, len, TSDB_CODE_FUNC_FUNTION_ERROR,
-                            "TAIL function second parameter should be in range [1, 100], "
-                            "third parameter should be in range [0, 100]");
+                             "TAIL function second parameter should be in range [1, 100], "
+                             "third parameter should be in range [0, 100]");
     }
 
     pValue->notReserved = true;
@@ -721,20 +642,12 @@ static int32_t translateDiff(SFunctionNode* pFunc, char* pErrBuf, int32_t len) {
     return invaildFuncParaNumErrMsg(pErrBuf, len, pFunc->functionName);
   }
 
-  //param0
-  SNode* pParamNode0 = nodesListGetNode(pFunc->pParameterList, 0);
-  if (nodeType(pParamNode0) != QUERY_NODE_COLUMN) {
-    return buildFuncErrMsg(pErrBuf, len, TSDB_CODE_FUNC_FUNTION_ERROR,
-                           "The first parameter of DIFF function can only be column");
-  }
-
   uint8_t colType = ((SExprNode*)nodesListGetNode(pFunc->pParameterList, 0))->resType.type;
-  if (!IS_SIGNED_NUMERIC_TYPE(colType) && !IS_FLOAT_TYPE(colType) &&
-      TSDB_DATA_TYPE_BOOL != colType) {
+  if (!IS_SIGNED_NUMERIC_TYPE(colType) && !IS_FLOAT_TYPE(colType) && TSDB_DATA_TYPE_BOOL != colType) {
     return invaildFuncParaTypeErrMsg(pErrBuf, len, pFunc->functionName);
   }
 
-  //param1
+  // param1
   if (numOfParams == 2) {
     uint8_t paraType = ((SExprNode*)nodesListGetNode(pFunc->pParameterList, 1))->resType.type;
     if (!IS_INTEGER_TYPE(paraType)) {
@@ -852,7 +765,7 @@ static int32_t translateSubstr(SFunctionNode* pFunc, char* pErrBuf, int32_t len)
 
   if (3 == numOfParams) {
     SExprNode* p2 = (SExprNode*)nodesListGetNode(pFunc->pParameterList, 2);
-    uint8_t para2Type = p2->resType.type;
+    uint8_t    para2Type = p2->resType.type;
     if (!IS_INTEGER_TYPE(para2Type)) {
       return invaildFuncParaTypeErrMsg(pErrBuf, len, pFunc->functionName);
     }
diff --git a/source/libs/nodes/src/nodesCloneFuncs.c b/source/libs/nodes/src/nodesCloneFuncs.c
index 68d3741b48..40c9e68620 100644
--- a/source/libs/nodes/src/nodesCloneFuncs.c
+++ b/source/libs/nodes/src/nodesCloneFuncs.c
@@ -408,6 +408,12 @@ static SNode* logicPartitionCopy(const SPartitionLogicNode* pSrc, SPartitionLogi
   return (SNode*)pDst;
 }
 
+static SNode* logicIndefRowsFuncCopy(const SIndefRowsFuncLogicNode* pSrc, SIndefRowsFuncLogicNode* pDst) {
+  COPY_BASE_OBJECT_FIELD(node, logicNodeCopy);
+  CLONE_NODE_LIST_FIELD(pVectorFuncs);
+  return (SNode*)pDst;
+}
+
 static SNode* logicSubplanCopy(const SLogicSubplan* pSrc, SLogicSubplan* pDst) {
   COPY_OBJECT_FIELD(id, sizeof(SSubplanId));
   CLONE_NODE_FIELD(pNode);
@@ -537,6 +543,8 @@ SNodeptr nodesCloneNode(const SNodeptr pNode) {
       return logicSortCopy((const SSortLogicNode*)pNode, (SSortLogicNode*)pDst);
     case QUERY_NODE_LOGIC_PLAN_PARTITION:
       return logicPartitionCopy((const SPartitionLogicNode*)pNode, (SPartitionLogicNode*)pDst);
+    case QUERY_NODE_LOGIC_PLAN_INDEF_ROWS_FUNC:
+      return logicIndefRowsFuncCopy((const SIndefRowsFuncLogicNode*)pNode, (SIndefRowsFuncLogicNode*)pDst);
     case QUERY_NODE_LOGIC_SUBPLAN:
       return logicSubplanCopy((const SLogicSubplan*)pNode, (SLogicSubplan*)pDst);
     default:
diff --git a/source/libs/nodes/src/nodesCodeFuncs.c b/source/libs/nodes/src/nodesCodeFuncs.c
index 78710569cb..e47913a91b 100644
--- a/source/libs/nodes/src/nodesCodeFuncs.c
+++ b/source/libs/nodes/src/nodesCodeFuncs.c
@@ -198,6 +198,8 @@ const char* nodesNodeName(ENodeType type) {
       return "LogicSort";
     case QUERY_NODE_LOGIC_PLAN_PARTITION:
       return "LogicPartition";
+    case QUERY_NODE_LOGIC_PLAN_INDEF_ROWS_FUNC:
+      return "LogicIndefRowsFunc";
     case QUERY_NODE_LOGIC_SUBPLAN:
       return "LogicSubplan";
     case QUERY_NODE_LOGIC_PLAN:
@@ -236,6 +238,8 @@ const char* nodesNodeName(ENodeType type) {
       return "PhysiStateWindow";
     case QUERY_NODE_PHYSICAL_PLAN_PARTITION:
       return "PhysiPartition";
+    case QUERY_NODE_PHYSICAL_PLAN_INDEF_ROWS_FUNC:
+      return "PhysiIndefRowsFunc";
     case QUERY_NODE_PHYSICAL_PLAN_DISPATCH:
       return "PhysiDispatch";
     case QUERY_NODE_PHYSICAL_PLAN_INSERT:
@@ -727,6 +731,30 @@ static int32_t jsonToLogicPartitionNode(const SJson* pJson, void* pObj) {
   return code;
 }
 
+static const char* jkIndefRowsFuncLogicPlanVectorFuncs = "VectorFuncs";
+
+static int32_t logicIndefRowsFuncNodeToJson(const void* pObj, SJson* pJson) {
+  const SIndefRowsFuncLogicNode* pNode = (const SIndefRowsFuncLogicNode*)pObj;
+
+  int32_t code = logicPlanNodeToJson(pObj, pJson);
+  if (TSDB_CODE_SUCCESS == code) {
+    code = nodeListToJson(pJson, jkIndefRowsFuncLogicPlanVectorFuncs, pNode->pVectorFuncs);
+  }
+
+  return code;
+}
+
+static int32_t jsonToLogicIndefRowsFuncNode(const SJson* pJson, void* pObj) {
+  SIndefRowsFuncLogicNode* pNode = (SIndefRowsFuncLogicNode*)pObj;
+
+  int32_t code = jsonToLogicPlanNode(pJson, pObj);
+  if (TSDB_CODE_SUCCESS == code) {
+    code = jsonToNodeList(pJson, jkIndefRowsFuncLogicPlanVectorFuncs, &pNode->pVectorFuncs);
+  }
+
+  return code;
+}
+
 static const char* jkSubplanIdQueryId = "QueryId";
 static const char* jkSubplanIdGroupId = "GroupId";
 static const char* jkSubplanIdSubplanId = "SubplanId";
@@ -1744,6 +1772,37 @@ static int32_t jsonToPhysiPartitionNode(const SJson* pJson, void* pObj) {
   return code;
 }
 
+static const char* jkIndefRowsFuncPhysiPlanExprs = "Exprs";
+static const char* jkIndefRowsFuncPhysiPlanVectorFuncs = "VectorFuncs";
+
+static int32_t physiIndefRowsFuncNodeToJson(const void* pObj, SJson* pJson) {
+  const SIndefRowsFuncPhysiNode* pNode = (const SIndefRowsFuncPhysiNode*)pObj;
+
+  int32_t code = physicPlanNodeToJson(pObj, pJson);
+  if (TSDB_CODE_SUCCESS == code) {
+    code = nodeListToJson(pJson, jkIndefRowsFuncPhysiPlanExprs, pNode->pExprs);
+  }
+  if (TSDB_CODE_SUCCESS == code) {
+    code = nodeListToJson(pJson, jkIndefRowsFuncPhysiPlanVectorFuncs, pNode->pVectorFuncs);
+  }
+
+  return code;
+}
+
+static int32_t jsonToPhysiIndefRowsFuncNode(const SJson* pJson, void* pObj) {
+  SIndefRowsFuncPhysiNode* pNode = (SIndefRowsFuncPhysiNode*)pObj;
+
+  int32_t code = jsonToPhysicPlanNode(pJson, pObj);
+  if (TSDB_CODE_SUCCESS == code) {
+    code = jsonToNodeList(pJson, jkIndefRowsFuncPhysiPlanExprs, &pNode->pExprs);
+  }
+  if (TSDB_CODE_SUCCESS == code) {
+    code = jsonToNodeList(pJson, jkIndefRowsFuncPhysiPlanVectorFuncs, &pNode->pVectorFuncs);
+  }
+
+  return code;
+}
+
 static const char* jkDataSinkInputDataBlockDesc = "InputDataBlockDesc";
 
 static int32_t physicDataSinkNodeToJson(const void* pObj, SJson* pJson) {
@@ -3394,6 +3453,8 @@ static int32_t specificNodeToJson(const void* pObj, SJson* pJson) {
       return logicSortNodeToJson(pObj, pJson);
     case QUERY_NODE_LOGIC_PLAN_PARTITION:
       return logicPartitionNodeToJson(pObj, pJson);
+    case QUERY_NODE_LOGIC_PLAN_INDEF_ROWS_FUNC:
+      return logicIndefRowsFuncNodeToJson(pObj, pJson);
     case QUERY_NODE_LOGIC_SUBPLAN:
       return logicSubplanToJson(pObj, pJson);
     case QUERY_NODE_LOGIC_PLAN:
@@ -3428,6 +3489,8 @@ static int32_t specificNodeToJson(const void* pObj, SJson* pJson) {
       return physiStateWindowNodeToJson(pObj, pJson);
     case QUERY_NODE_PHYSICAL_PLAN_PARTITION:
       return physiPartitionNodeToJson(pObj, pJson);
+    case QUERY_NODE_PHYSICAL_PLAN_INDEF_ROWS_FUNC:
+      return physiIndefRowsFuncNodeToJson(pObj, pJson);
     case QUERY_NODE_PHYSICAL_PLAN_DISPATCH:
       return physiDispatchNodeToJson(pObj, pJson);
     case QUERY_NODE_PHYSICAL_PLAN_INSERT:
@@ -3505,6 +3568,8 @@ static int32_t jsonToSpecificNode(const SJson* pJson, void* pObj) {
       return jsonToLogicSortNode(pJson, pObj);
     case QUERY_NODE_LOGIC_PLAN_PARTITION:
       return jsonToLogicPartitionNode(pJson, pObj);
+    case QUERY_NODE_LOGIC_PLAN_INDEF_ROWS_FUNC:
+      return jsonToLogicIndefRowsFuncNode(pJson, pObj);
     case QUERY_NODE_LOGIC_SUBPLAN:
       return jsonToLogicSubplan(pJson, pObj);
     case QUERY_NODE_LOGIC_PLAN:
@@ -3539,6 +3604,8 @@ static int32_t jsonToSpecificNode(const SJson* pJson, void* pObj) {
       return jsonToPhysiStateWindowNode(pJson, pObj);
     case QUERY_NODE_PHYSICAL_PLAN_PARTITION:
       return jsonToPhysiPartitionNode(pJson, pObj);
+    case QUERY_NODE_PHYSICAL_PLAN_INDEF_ROWS_FUNC:
+      return jsonToPhysiIndefRowsFuncNode(pJson, pObj);
     case QUERY_NODE_PHYSICAL_PLAN_DISPATCH:
       return jsonToPhysiDispatchNode(pJson, pObj);
     case QUERY_NODE_PHYSICAL_SUBPLAN:
diff --git a/source/libs/nodes/src/nodesTraverseFuncs.c b/source/libs/nodes/src/nodesTraverseFuncs.c
index ae1ff5744b..9c433aa678 100644
--- a/source/libs/nodes/src/nodesTraverseFuncs.c
+++ b/source/libs/nodes/src/nodesTraverseFuncs.c
@@ -350,6 +350,7 @@ void nodesWalkSelectStmt(SSelectStmt* pSelect, ESqlClause clause, FNodeWalker wa
     case SQL_CLAUSE_GROUP_BY:
       nodesWalkExpr(pSelect->pHaving, walker, pContext);
     case SQL_CLAUSE_HAVING:
+    case SQL_CLAUSE_SELECT:
     case SQL_CLAUSE_DISTINCT:
       nodesWalkExprs(pSelect->pOrderByList, walker, pContext);
     case SQL_CLAUSE_ORDER_BY:
@@ -382,6 +383,7 @@ void nodesRewriteSelectStmt(SSelectStmt* pSelect, ESqlClause clause, FNodeRewrit
     case SQL_CLAUSE_GROUP_BY:
       nodesRewriteExpr(&(pSelect->pHaving), rewriter, pContext);
     case SQL_CLAUSE_HAVING:
+    case SQL_CLAUSE_SELECT:
     case SQL_CLAUSE_DISTINCT:
       nodesRewriteExprs(pSelect->pOrderByList, rewriter, pContext);
     case SQL_CLAUSE_ORDER_BY:
diff --git a/source/libs/nodes/src/nodesUtilFuncs.c b/source/libs/nodes/src/nodesUtilFuncs.c
index e28844f2e1..5232c96140 100644
--- a/source/libs/nodes/src/nodesUtilFuncs.c
+++ b/source/libs/nodes/src/nodesUtilFuncs.c
@@ -230,6 +230,8 @@ SNodeptr nodesMakeNode(ENodeType type) {
       return makeNode(type, sizeof(SSortLogicNode));
     case QUERY_NODE_LOGIC_PLAN_PARTITION:
       return makeNode(type, sizeof(SPartitionLogicNode));
+    case QUERY_NODE_LOGIC_PLAN_INDEF_ROWS_FUNC:
+      return makeNode(type, sizeof(SIndefRowsFuncLogicNode));
     case QUERY_NODE_LOGIC_SUBPLAN:
       return makeNode(type, sizeof(SLogicSubplan));
     case QUERY_NODE_LOGIC_PLAN:
@@ -268,6 +270,8 @@ SNodeptr nodesMakeNode(ENodeType type) {
       return makeNode(type, sizeof(SStateWinodwPhysiNode));
     case QUERY_NODE_PHYSICAL_PLAN_PARTITION:
       return makeNode(type, sizeof(SPartitionPhysiNode));
+    case QUERY_NODE_PHYSICAL_PLAN_INDEF_ROWS_FUNC:
+      return makeNode(type, sizeof(SIndefRowsFuncPhysiNode));
     case QUERY_NODE_PHYSICAL_PLAN_DISPATCH:
       return makeNode(type, sizeof(SDataDispatcherNode));
     case QUERY_NODE_PHYSICAL_PLAN_INSERT:
diff --git a/source/libs/planner/src/planLogicCreater.c b/source/libs/planner/src/planLogicCreater.c
index 467b26b7c4..273f46ac17 100644
--- a/source/libs/planner/src/planLogicCreater.c
+++ b/source/libs/planner/src/planLogicCreater.c
@@ -476,6 +476,37 @@ static int32_t createAggLogicNode(SLogicPlanContext* pCxt, SSelectStmt* pSelect,
   return code;
 }
 
+static int32_t createIndefRowsFuncLogicNode(SLogicPlanContext* pCxt, SSelectStmt* pSelect, SLogicNode** pLogicNode) {
+  // top/bottom are both an aggregate function and a indefinite rows function
+  if (!pSelect->hasIndefiniteRowsFunc || pSelect->hasAggFuncs) {
+    return TSDB_CODE_SUCCESS;
+  }
+
+  SIndefRowsFuncLogicNode* pIdfRowsFunc =
+      (SIndefRowsFuncLogicNode*)nodesMakeNode(QUERY_NODE_LOGIC_PLAN_INDEF_ROWS_FUNC);
+  if (NULL == pIdfRowsFunc) {
+    return TSDB_CODE_OUT_OF_MEMORY;
+  }
+
+  int32_t code = nodesCollectFuncs(pSelect, SQL_CLAUSE_SELECT, fmIsVectorFunc, &pIdfRowsFunc->pVectorFuncs);
+  if (TSDB_CODE_SUCCESS == code) {
+    code = rewriteExprForSelect(pIdfRowsFunc->pVectorFuncs, pSelect, SQL_CLAUSE_SELECT);
+  }
+
+  // set the output
+  if (TSDB_CODE_SUCCESS == code) {
+    code = createColumnByRewriteExps(pCxt, pIdfRowsFunc->pVectorFuncs, &pIdfRowsFunc->node.pTargets);
+  }
+
+  if (TSDB_CODE_SUCCESS == code) {
+    *pLogicNode = (SLogicNode*)pIdfRowsFunc;
+  } else {
+    nodesDestroyNode(pIdfRowsFunc);
+  }
+
+  return code;
+}
+
 static int32_t createWindowLogicNodeFinalize(SLogicPlanContext* pCxt, SSelectStmt* pSelect, SWindowLogicNode* pWindow,
                                              SLogicNode** pLogicNode) {
   int32_t code = nodesCollectFuncs(pSelect, SQL_CLAUSE_WINDOW, fmIsWindowClauseFunc, &pWindow->pFuncs);
@@ -787,6 +818,9 @@ static int32_t createSelectLogicNode(SLogicPlanContext* pCxt, SSelectStmt* pSele
   if (TSDB_CODE_SUCCESS == code) {
     code = createChildLogicNode(pCxt, pSelect, createAggLogicNode, &pRoot);
   }
+  if (TSDB_CODE_SUCCESS == code) {
+    code = createChildLogicNode(pCxt, pSelect, createIndefRowsFuncLogicNode, &pRoot);
+  }
   if (TSDB_CODE_SUCCESS == code) {
     code = createChildLogicNode(pCxt, pSelect, createDistinctLogicNode, &pRoot);
   }
diff --git a/source/libs/planner/src/planPhysiCreater.c b/source/libs/planner/src/planPhysiCreater.c
index a45eabefb9..69fa5499c4 100644
--- a/source/libs/planner/src/planPhysiCreater.c
+++ b/source/libs/planner/src/planPhysiCreater.c
@@ -800,6 +800,43 @@ static int32_t createAggPhysiNode(SPhysiPlanContext* pCxt, SNodeList* pChildren,
   return code;
 }
 
+static int32_t createIndefRowsFuncPhysiNode(SPhysiPlanContext* pCxt, SNodeList* pChildren,
+                                            SIndefRowsFuncLogicNode* pFuncLogicNode, SPhysiNode** pPhyNode) {
+  SIndefRowsFuncPhysiNode* pIdfRowsFunc = (SIndefRowsFuncPhysiNode*)makePhysiNode(
+      pCxt, getPrecision(pChildren), (SLogicNode*)pFuncLogicNode, QUERY_NODE_PHYSICAL_PLAN_INDEF_ROWS_FUNC);
+  if (NULL == pIdfRowsFunc) {
+    return TSDB_CODE_OUT_OF_MEMORY;
+  }
+
+  SNodeList* pPrecalcExprs = NULL;
+  SNodeList* pVectorFuncs = NULL;
+  int32_t    code = rewritePrecalcExprs(pCxt, pFuncLogicNode->pVectorFuncs, &pPrecalcExprs, &pVectorFuncs);
+
+  SDataBlockDescNode* pChildTupe = (((SPhysiNode*)nodesListGetNode(pChildren, 0))->pOutputDataBlockDesc);
+  // push down expression to pOutputDataBlockDesc of child node
+  if (TSDB_CODE_SUCCESS == code && NULL != pPrecalcExprs) {
+    code = setListSlotId(pCxt, pChildTupe->dataBlockId, -1, pPrecalcExprs, &pIdfRowsFunc->pExprs);
+    if (TSDB_CODE_SUCCESS == code) {
+      code = pushdownDataBlockSlots(pCxt, pIdfRowsFunc->pExprs, pChildTupe);
+    }
+  }
+
+  if (TSDB_CODE_SUCCESS == code && NULL != pVectorFuncs) {
+    code = setListSlotId(pCxt, pChildTupe->dataBlockId, -1, pVectorFuncs, &pIdfRowsFunc->pVectorFuncs);
+    if (TSDB_CODE_SUCCESS == code) {
+      code = addDataBlockSlots(pCxt, pIdfRowsFunc->pVectorFuncs, pIdfRowsFunc->node.pOutputDataBlockDesc);
+    }
+  }
+
+  if (TSDB_CODE_SUCCESS == code) {
+    *pPhyNode = (SPhysiNode*)pIdfRowsFunc;
+  } else {
+    nodesDestroyNode(pIdfRowsFunc);
+  }
+
+  return code;
+}
+
 static int32_t createProjectPhysiNode(SPhysiPlanContext* pCxt, SNodeList* pChildren,
                                       SProjectLogicNode* pProjectLogicNode, SPhysiNode** pPhyNode) {
   SProjectPhysiNode* pProject = (SProjectPhysiNode*)makePhysiNode(
@@ -949,7 +986,8 @@ static int32_t createSessionWindowPhysiNode(SPhysiPlanContext* pCxt, SNodeList*
                                             SWindowLogicNode* pWindowLogicNode, SPhysiNode** pPhyNode) {
   SSessionWinodwPhysiNode* pSession = (SSessionWinodwPhysiNode*)makePhysiNode(
       pCxt, getPrecision(pChildren), (SLogicNode*)pWindowLogicNode,
-      (pCxt->pPlanCxt->streamQuery ? QUERY_NODE_PHYSICAL_PLAN_STREAM_SESSION_WINDOW : QUERY_NODE_PHYSICAL_PLAN_SESSION_WINDOW));
+      (pCxt->pPlanCxt->streamQuery ? QUERY_NODE_PHYSICAL_PLAN_STREAM_SESSION_WINDOW
+                                   : QUERY_NODE_PHYSICAL_PLAN_SESSION_WINDOW));
   if (NULL == pSession) {
     return TSDB_CODE_OUT_OF_MEMORY;
   }
@@ -1153,6 +1191,8 @@ static int32_t doCreatePhysiNode(SPhysiPlanContext* pCxt, SLogicNode* pLogicNode
       return createPartitionPhysiNode(pCxt, pChildren, (SPartitionLogicNode*)pLogicNode, pPhyNode);
     case QUERY_NODE_LOGIC_PLAN_FILL:
       return createFillPhysiNode(pCxt, pChildren, (SFillLogicNode*)pLogicNode, pPhyNode);
+    case QUERY_NODE_LOGIC_PLAN_INDEF_ROWS_FUNC:
+      return createIndefRowsFuncPhysiNode(pCxt, pChildren, (SIndefRowsFuncLogicNode*)pLogicNode, pPhyNode);
     default:
       break;
   }
diff --git a/source/libs/planner/test/planProjectTest.cpp b/source/libs/planner/test/planProjectTest.cpp
new file mode 100644
index 0000000000..3ef0038ae0
--- /dev/null
+++ b/source/libs/planner/test/planProjectTest.cpp
@@ -0,0 +1,34 @@
+/*
+ * Copyright (c) 2019 TAOS Data, Inc. <jhtao@taosdata.com>
+ *
+ * This program is free software: you can use, redistribute, and/or modify
+ * it under the terms of the GNU Affero General Public License, version 3
+ * or later ("AGPL"), as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.
+ *
+ * You should have received a copy of the GNU Affero General Public License
+ * along with this program. If not, see <http://www.gnu.org/licenses/>.
+ */
+
+#include "planTestUtil.h"
+
+using namespace std;
+
+class PlanProjectTest : public PlannerTestBase {};
+
+TEST_F(PlanProjectTest, basic) {
+  useDb("root", "test");
+
+  run("SELECT CEIL(c1) FROM t1");
+}
+
+TEST_F(PlanProjectTest, indefiniteRowsFunc) {
+  useDb("root", "test");
+
+  run("SELECT MAVG(c1, 10) FROM t1");
+
+  run("SELECT MAVG(CEIL(c1), 20) + 2 FROM t1");
+}

From 0244c4ff9435217bdfb2f6c9090dba3acb23c6ba Mon Sep 17 00:00:00 2001
From: Haojun Liao <hjliao@taosdata.com>
Date: Sat, 4 Jun 2022 19:19:49 +0800
Subject: [PATCH 02/34] enh(query): add a new operator.

---
 source/libs/executor/inc/executorimpl.h |  21 ++-
 source/libs/executor/src/executorimpl.c | 174 +++++++++++++++++++++---
 2 files changed, 171 insertions(+), 24 deletions(-)

diff --git a/source/libs/executor/inc/executorimpl.h b/source/libs/executor/inc/executorimpl.h
index cb45fa9730..16d758c0e3 100644
--- a/source/libs/executor/inc/executorimpl.h
+++ b/source/libs/executor/inc/executorimpl.h
@@ -493,11 +493,11 @@ typedef struct SAggOperatorInfo {
 
 typedef struct SProjectOperatorInfo {
   // SOptrBasicInfo should be first, SAggSupporter should be second for stream encode
-  SOptrBasicInfo     binfo;
-  SAggSupporter      aggSup;
+  SOptrBasicInfo     binfo;   // todo remove it
+  SAggSupporter      aggSup;  // todo remove it
 
-  SSDataBlock*       existDataBlock;
-  SArray*            pPseudoColInfo;
+  SSDataBlock*       existDataBlock;  // todo remove it
+  SArray*            pPseudoColInfo;  // todo remove it
   SLimit             limit;
   SLimit             slimit;
 
@@ -509,6 +509,17 @@ typedef struct SProjectOperatorInfo {
   int64_t            curOutput;
 } SProjectOperatorInfo;
 
+typedef struct SIndefOperatorInfo {
+  SOptrBasicInfo     binfo;
+  SAggSupporter      aggSup;
+  SArray*            pPseudoColInfo;
+
+  SExprInfo*         pScalarExpr;
+  int32_t            numOfScalarExpr;
+  SqlFunctionCtx*    pScalarCtx;
+  int32_t*           rowCellInfoOffset;
+} SIndefOperatorInfo;
+
 typedef struct SFillOperatorInfo {
   struct SFillInfo* pFillInfo;
   SSDataBlock*      pRes;
@@ -731,6 +742,8 @@ SOperatorInfo* createAggregateOperatorInfo(SOperatorInfo* downstream, SExprInfo*
                                            int32_t numOfScalarExpr, SExecTaskInfo* pTaskInfo);
 
 SOperatorInfo* createProjectOperatorInfo(SOperatorInfo* downstream, SExprInfo* pExprInfo, int32_t num, SSDataBlock* pResBlock, SLimit* pLimit, SLimit* pSlimit, SExecTaskInfo* pTaskInfo);
+SOperatorInfo* createIndefinitOutputOperatorInfo(SOperatorInfo* downstream, SPhysiNode *pNode, SExecTaskInfo* pTaskInfo);
+
 SOperatorInfo *createSortOperatorInfo(SOperatorInfo* downstream, SSDataBlock* pResBlock, SArray* pSortInfo, SExprInfo* pExprInfo, int32_t numOfCols,
                                       SArray* pIndexMap, SExecTaskInfo* pTaskInfo);
 
diff --git a/source/libs/executor/src/executorimpl.c b/source/libs/executor/src/executorimpl.c
index e7de886b03..acc38735f8 100644
--- a/source/libs/executor/src/executorimpl.c
+++ b/source/libs/executor/src/executorimpl.c
@@ -3750,18 +3750,6 @@ static SSDataBlock* doProjectOperation(SOperatorInfo* pOperator) {
       break;
     }
 
-#if 0
-    // Return result of the previous group in the firstly.
-    if (false) {
-      if (pRes->info.rows > 0) {
-        pProjectInfo->existDataBlock = pBlock;
-        break;
-      } else {  // init output buffer for a new group data
-        initCtxOutputBuffer(pInfo->pCtx, pOperator->numOfExprs);
-      }
-    }
-#endif
-
     // the pDataBlock are always the same one, no need to call this again
     int32_t code = getTableScanInfo(pOperator->pDownstream[0], &order, &scanFlag);
     if (code != TSDB_CODE_SUCCESS) {
@@ -3917,6 +3905,17 @@ static int32_t getNumOfScanTimes(STaskAttr* pQueryAttr) {
   return 1;
 }
 
+static void destroyExprInfo(SExprInfo* pExpr, int32_t numOfExprs) {
+  for (int32_t i = 0; i < numOfExprs; ++i) {
+    SExprInfo* pExprInfo = &pExpr[i];
+    if (pExprInfo->pExpr->nodeType == QUERY_NODE_COLUMN) {
+      taosMemoryFree(pExprInfo->base.pParam[0].pCol);
+    }
+    taosMemoryFree(pExprInfo->base.pParam);
+    taosMemoryFree(pExprInfo->pExpr);
+  }
+}
+
 static void destroyOperatorInfo(SOperatorInfo* pOperator) {
   if (pOperator == NULL) {
     return;
@@ -3936,14 +3935,7 @@ static void destroyOperatorInfo(SOperatorInfo* pOperator) {
   }
 
   if (pOperator->pExpr != NULL) {
-    for (int32_t i = 0; i < pOperator->numOfExprs; ++i) {
-      SExprInfo* pExprInfo = &pOperator->pExpr[i];
-      if (pExprInfo->pExpr->nodeType == QUERY_NODE_COLUMN) {
-        taosMemoryFree(pExprInfo->base.pParam[0].pCol);
-      }
-      taosMemoryFree(pExprInfo->base.pParam);
-      taosMemoryFree(pExprInfo->pExpr);
-    }
+    destroyExprInfo(pOperator->pExpr, pOperator->numOfExprs);
   }
 
   taosMemoryFreeClear(pOperator->pExpr);
@@ -4132,6 +4124,19 @@ static void destroyProjectOperatorInfo(void* param, int32_t numOfOutput) {
   taosArrayDestroy(pInfo->pPseudoColInfo);
 }
 
+static void destroyIndefinitOperatorInfo(void* param, int32_t numOfOutput) {
+  SIndefOperatorInfo* pInfo = (SIndefOperatorInfo*) param;
+  doDestroyBasicInfo(&pInfo->binfo, numOfOutput);
+
+  taosArrayDestroy(pInfo->pPseudoColInfo);
+  cleanupAggSup(&pInfo->aggSup);
+
+  destroySqlFunctionCtx(pInfo->pScalarCtx, numOfOutput);
+  destroyExprInfo(pInfo->pScalarExpr, pInfo->numOfScalarExpr);
+
+  taosMemoryFree(pInfo->rowCellInfoOffset);
+}
+
 void destroyExchangeOperatorInfo(void* param, int32_t numOfOutput) {
   SExchangeInfo* pExInfo = (SExchangeInfo*)param;
   taosArrayDestroy(pExInfo->pSources);
@@ -4209,6 +4214,133 @@ _error:
   return NULL;
 }
 
+static SSDataBlock* doApplyIndefinitFunction(SOperatorInfo* pOperator) {
+  SIndefOperatorInfo* pIndefInfo = pOperator->info;
+  SOptrBasicInfo*       pInfo = &pIndefInfo->binfo;
+
+  SSDataBlock* pRes = pInfo->pRes;
+  blockDataCleanup(pRes);
+
+  SExecTaskInfo* pTaskInfo = pOperator->pTaskInfo;
+  if (pOperator->status == OP_EXEC_DONE) {
+    return NULL;
+  }
+
+  int64_t st = 0;
+  int32_t order = 0;
+  int32_t scanFlag = 0;
+
+  if (pOperator->cost.openCost == 0) {
+    st = taosGetTimestampUs();
+  }
+
+  SOperatorInfo* downstream = pOperator->pDownstream[0];
+
+  while (1) {
+    // The downstream exec may change the value of the newgroup, so use a local variable instead.
+    SSDataBlock* pBlock = downstream->fpSet.getNextFn(downstream);
+    if (pBlock == NULL) {
+      doSetOperatorCompleted(pOperator);
+      break;
+    }
+
+    // the pDataBlock are always the same one, no need to call this again
+    int32_t code = getTableScanInfo(pOperator->pDownstream[0], &order, &scanFlag);
+    if (code != TSDB_CODE_SUCCESS) {
+      longjmp(pTaskInfo->env, code);
+    }
+
+    // there is an scalar expression that needs to be calculated before apply the group aggregation.
+    if (pIndefInfo->pScalarExpr != NULL) {
+      code = projectApplyFunctions(pIndefInfo->pScalarExpr, pBlock, pBlock, pIndefInfo->pScalarCtx,
+                                   pIndefInfo->numOfScalarExpr, pIndefInfo->pPseudoColInfo);
+      if (code != TSDB_CODE_SUCCESS) {
+        longjmp(pTaskInfo->env, code);
+      }
+    }
+
+    setInputDataBlock(pOperator, pInfo->pCtx, pBlock, order, scanFlag, false);
+    blockDataEnsureCapacity(pInfo->pRes, pInfo->pRes->info.rows + pBlock->info.rows);
+
+    code = projectApplyFunctions(pOperator->pExpr, pInfo->pRes, pBlock, pInfo->pCtx, pOperator->numOfExprs, pIndefInfo->pPseudoColInfo);
+    if (code != TSDB_CODE_SUCCESS) {
+      longjmp(pTaskInfo->env, code);
+    }
+  }
+
+  size_t rows = pInfo->pRes->info.rows;
+  pOperator->resultInfo.totalRows += rows;
+
+  if (pOperator->cost.openCost == 0) {
+    pOperator->cost.openCost = (taosGetTimestampUs() - st) / 1000.0;
+  }
+
+  return (rows > 0) ? pInfo->pRes : NULL;
+}
+
+SOperatorInfo* createIndefinitOutputOperatorInfo(SOperatorInfo* downstream, SPhysiNode *pNode, SExecTaskInfo* pTaskInfo) {
+  SIndefOperatorInfo* pInfo = taosMemoryCalloc(1, sizeof(SIndefOperatorInfo));
+  SOperatorInfo*        pOperator = taosMemoryCalloc(1, sizeof(SOperatorInfo));
+  if (pInfo == NULL || pOperator == NULL) {
+    goto _error;
+  }
+
+  SIndefRowsFuncPhysiNode* pPhyNode = (SIndefRowsFuncPhysiNode*)pNode;
+
+  int32_t    numOfExpr = 0;
+  SExprInfo* pExprInfo = createExprInfo(pPhyNode->pVectorFuncs, NULL, &numOfExpr);
+
+  int32_t numOfScalarExpr = 0;
+  if (pPhyNode->pExprs != NULL) {
+    pInfo->pScalarExpr = createExprInfo(pPhyNode->pExprs, NULL, &numOfScalarExpr);
+    pInfo->pScalarCtx =  createSqlFunctionCtx(pInfo->pScalarExpr, numOfScalarExpr, &pInfo->rowCellInfoOffset);
+  }
+
+  SSDataBlock* pResBlock = createResDataBlock(pPhyNode->node.pOutputDataBlockDesc);;
+
+  int32_t numOfRows = 4096;
+  size_t  keyBufSize = sizeof(int64_t) + sizeof(int64_t) + POINTER_BYTES;
+
+  // Make sure the size of SSDataBlock will never exceed the size of 2MB.
+  int32_t TWOMB = 2 * 1024 * 1024;
+  if (numOfRows * pResBlock->info.rowSize > TWOMB) {
+    numOfRows = TWOMB / pResBlock->info.rowSize;
+  }
+  initResultSizeInfo(pOperator, numOfRows);
+
+  initAggInfo(&pInfo->binfo, &pInfo->aggSup, pExprInfo, numOfExpr, pResBlock, keyBufSize, pTaskInfo->id.str);
+  setFunctionResultOutput(&pInfo->binfo, &pInfo->aggSup, MAIN_SCAN, numOfExpr, pTaskInfo);
+
+  pInfo->binfo.pRes       = pResBlock;
+  pInfo->numOfScalarExpr  = numOfScalarExpr;
+  pInfo->pPseudoColInfo   = setRowTsColumnOutputInfo(pInfo->binfo.pCtx, numOfExpr);
+
+  pOperator->name         = "IndefinitOperator";
+  pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_PROJECT;
+  pOperator->blocking     = false;
+  pOperator->status       = OP_NOT_OPENED;
+  pOperator->info         = pInfo;
+  pOperator->pExpr        = pExprInfo;
+  pOperator->numOfExprs   = numOfExpr;
+  pOperator->pTaskInfo    = pTaskInfo;
+
+  pOperator->fpSet = createOperatorFpSet(operatorDummyOpenFn, doApplyIndefinitFunction, NULL, NULL,
+                                         destroyIndefinitOperatorInfo, NULL, NULL, NULL);
+
+  int32_t code = appendDownstream(pOperator, &downstream, 1);
+  if (code != TSDB_CODE_SUCCESS) {
+    goto _error;
+  }
+
+  return pOperator;
+
+  _error:
+  taosMemoryFree(pInfo);
+  taosMemoryFree(pOperator);
+  pTaskInfo->code = TSDB_CODE_OUT_OF_MEMORY;
+  return NULL;
+}
+
 static int32_t initFillInfo(SFillOperatorInfo* pInfo, SExprInfo* pExpr, int32_t numOfCols, SNodeListNode* pValNode,
                             STimeWindow win, int32_t capacity, const char* id, SInterval* pInterval, int32_t fillType) {
   SFillColInfo* pColInfo = createFillColInfo(pExpr, numOfCols, pValNode);
@@ -4710,6 +4842,8 @@ SOperatorInfo* createOperatorTree(SPhysiNode* pPhyNode, SExecTaskInfo* pTaskInfo
     SInterval* pInterval = &((SIntervalAggOperatorInfo*)ops[0]->info)->interval;
     pOptr = createFillOperatorInfo(ops[0], pExprInfo, num, pInterval, &pFillNode->timeRange, pResBlock, pFillNode->mode,
                                    (SNodeListNode*)pFillNode->pValues, false, pTaskInfo);
+  } else if (QUERY_NODE_PHYSICAL_PLAN_INDEF_ROWS_FUNC == type) {
+    pOptr = createIndefinitOutputOperatorInfo(ops[0], pPhyNode, pTaskInfo);
   } else {
     ASSERT(0);
   }

From 8e8f5fbab2701e9a713753c7574a99c1078b1a45 Mon Sep 17 00:00:00 2001
From: "wenzhouwww@live.cn" <wenzhouwww@live.cn>
Date: Mon, 6 Jun 2022 11:30:11 +0800
Subject: [PATCH 03/34] update csum case

---
 tests/system-test/2-query/csum.py | 10 +++++++++-
 1 file changed, 9 insertions(+), 1 deletion(-)

diff --git a/tests/system-test/2-query/csum.py b/tests/system-test/2-query/csum.py
index a331311fd2..5c2de5ea14 100644
--- a/tests/system-test/2-query/csum.py
+++ b/tests/system-test/2-query/csum.py
@@ -240,7 +240,7 @@ class TDTestCase:
         tdSql.error("select csum(c1)   t1")         # no from
         tdSql.error("select csum( c1 )  from ")     # no table_expr
         # tdSql.error(self.csum_query_form(col="st1"))    # tag col
-        tdSql.error(self.csum_query_form(col=1))        # col is a value
+        # tdSql.error(self.csum_query_form(col=1))        # col is a value
         tdSql.error(self.csum_query_form(col="'c1'"))   # col is a string
         tdSql.error(self.csum_query_form(col=None))     # col is NULL 1
         tdSql.error(self.csum_query_form(col="NULL"))   # col is NULL 2
@@ -407,6 +407,14 @@ class TDTestCase:
         tdDnodes.start(index)
         self.csum_current_query()
         self.csum_error_query()
+        tdSql.query("select csum(1) from t1 ")
+        tdSql.checkRows(7)
+        tdSql.checkData(0,0,1)
+        tdSql.checkData(1,0,2)
+        tdSql.checkData(2,0,3)
+        tdSql.checkData(3,0,4)
+        tdSql.query("select csum(abs(c1))+2 from t1 ")
+        tdSql.checkRows(4)
 
     def run(self):
         import traceback

From 7fc1e3a40fd9fe61707afcd5e926f650ccea1d35 Mon Sep 17 00:00:00 2001
From: "wenzhouwww@live.cn" <wenzhouwww@live.cn>
Date: Mon, 6 Jun 2022 11:35:28 +0800
Subject: [PATCH 04/34] update case for mavg

---
 tests/system-test/2-query/mavg.py | 12 ++++++++++--
 1 file changed, 10 insertions(+), 2 deletions(-)

diff --git a/tests/system-test/2-query/mavg.py b/tests/system-test/2-query/mavg.py
index 1d92964615..13d2b4d420 100644
--- a/tests/system-test/2-query/mavg.py
+++ b/tests/system-test/2-query/mavg.py
@@ -417,8 +417,8 @@ class TDTestCase:
 
         # err9 = {"col": "st1"}
         # self.checkmavg(**err9)          # col: tag
-        err10 = {"col": 1}
-        self.checkmavg(**err10)         # col: value
+        # err10 = {"col": 1}
+        # self.checkmavg(**err10)         # col: value
         err11 = {"col": "NULL"}
         self.checkmavg(**err11)         # col: NULL
         err12 = {"col": "%_"}
@@ -660,6 +660,14 @@ class TDTestCase:
         tdDnodes.start(index)
         self.mavg_current_query()
         self.mavg_error_query()
+        tdSql.query("select mavg(1,1) from t1")
+        tdSql.checkRows(7)
+        tdSql.checkData(0,0,1.000000000)
+        tdSql.checkData(1,0,1.000000000)
+        tdSql.checkData(5,0,1.000000000)
+        
+        tdSql.query("select mavg(abs(c1),1) from t1")
+        tdSql.checkRows(4)
 
     def run(self):
         import traceback

From f35ba2a85e80a7fc5ddf6796b97f6c2e554e96e0 Mon Sep 17 00:00:00 2001
From: yihaoDeng <luomoxyz@126.com>
Date: Mon, 6 Jun 2022 11:51:00 +0800
Subject: [PATCH 05/34] fix: failed to update json idx

---
 source/dnode/vnode/src/meta/metaTable.c | 2 +-
 source/libs/index/src/indexFilter.c     | 3 ++-
 2 files changed, 3 insertions(+), 2 deletions(-)

diff --git a/source/dnode/vnode/src/meta/metaTable.c b/source/dnode/vnode/src/meta/metaTable.c
index 4f4a940bc7..c6a6dd8147 100644
--- a/source/dnode/vnode/src/meta/metaTable.c
+++ b/source/dnode/vnode/src/meta/metaTable.c
@@ -103,7 +103,7 @@ static int metaSaveJsonVarToIdx(SMeta *pMeta, const SMetaEntry *pCtbEntry, const
   tIndexJsonPut(pMeta->pTagIvtIdx, terms, tuid);
   indexMultiTermDestroy(terms);
 #endif
-  return -1;
+  return 0;
 }
 
 int metaCreateSTable(SMeta *pMeta, int64_t version, SVCreateStbReq *pReq) {
diff --git a/source/libs/index/src/indexFilter.c b/source/libs/index/src/indexFilter.c
index 34251ab918..2e30ca704a 100644
--- a/source/libs/index/src/indexFilter.c
+++ b/source/libs/index/src/indexFilter.c
@@ -38,7 +38,7 @@ typedef struct SIFParam {
   col_id_t      colId;
   int64_t       suid;  // add later
   char          dbName[TSDB_DB_NAME_LEN];
-  char          colName[TSDB_COL_NAME_LEN];
+  char          colName[TSDB_COL_NAME_LEN * 2 + 4];
 
   SIndexMetaArg arg;
 } SIFParam;
@@ -171,6 +171,7 @@ static int32_t sifInitJsonParam(SNode *node, SIFParam *param, SIFCtx *ctx) {
   param->colId = l->colId;
   param->colValType = l->node.resType.type;
   memcpy(param->dbName, l->dbName, sizeof(l->dbName));
+
   sprintf(param->colName, "%s_%s", l->colName, r->literal);
   param->colValType = r->typeData;
   return 0;

From 79769dd2ad7642bc0c4ba902d96fb4c897f9136e Mon Sep 17 00:00:00 2001
From: Cary Xu <klxu@taosdata.com>
Date: Mon, 6 Jun 2022 13:47:32 +0800
Subject: [PATCH 06/34] feat: add vgroup epset for tsma

---
 include/common/tmsg.h                | 44 +++++++++----------
 source/common/src/tmsg.c             | 62 ++++++++++++++-------------
 source/dnode/mnode/impl/inc/mndDef.h | 46 ++++++++++----------
 source/dnode/mnode/impl/src/mndSma.c | 63 +++++++++++++++++++++++++++-
 4 files changed, 142 insertions(+), 73 deletions(-)

diff --git a/include/common/tmsg.h b/include/common/tmsg.h
index 39dc5361e6..2afab8b2f8 100644
--- a/include/common/tmsg.h
+++ b/include/common/tmsg.h
@@ -1122,13 +1122,13 @@ typedef struct {
   SSchema* pSchemas;
 } STableMetaRsp;
 
-typedef struct {  
+typedef struct {
   STableMetaRsp* pMeta;
 } SMAlterStbRsp;
 
-int32_t tEncodeSMAlterStbRsp(SEncoder *pEncoder, const SMAlterStbRsp *pRsp);
-int32_t tDecodeSMAlterStbRsp(SDecoder *pDecoder, SMAlterStbRsp *pRsp);
-void tFreeSMAlterStbRsp(SMAlterStbRsp* pRsp);
+int32_t tEncodeSMAlterStbRsp(SEncoder* pEncoder, const SMAlterStbRsp* pRsp);
+int32_t tDecodeSMAlterStbRsp(SDecoder* pDecoder, SMAlterStbRsp* pRsp);
+void    tFreeSMAlterStbRsp(SMAlterStbRsp* pRsp);
 
 int32_t tSerializeSTableMetaRsp(void* buf, int32_t bufLen, STableMetaRsp* pRsp);
 int32_t tDeserializeSTableMetaRsp(void* buf, int32_t bufLen, STableMetaRsp* pRsp);
@@ -2303,23 +2303,23 @@ typedef struct {
 } SVgEpSet;
 
 typedef struct {
-  int8_t   version;       // for compatibility(default 0)
-  int8_t   intervalUnit;  // MACRO: TIME_UNIT_XXX
-  int8_t   slidingUnit;   // MACRO: TIME_UNIT_XXX
-  int8_t   timezoneInt;   // sma data expired if timezone changes.
-  int32_t  dstVgId;
-  char     indexName[TSDB_INDEX_NAME_LEN];
-  int32_t  exprLen;
-  int32_t  tagsFilterLen;
-  int32_t  numOfVgroups;
-  int64_t  indexUid;
-  tb_uid_t tableUid;  // super/child/common table uid
-  int64_t  interval;
-  int64_t  offset;  // use unit by precision of DB
-  int64_t  sliding;
-  char*    expr;  // sma expression
-  char*    tagsFilter;
-  SVgEpSet vgEpSet[];
+  int8_t    version;       // for compatibility(default 0)
+  int8_t    intervalUnit;  // MACRO: TIME_UNIT_XXX
+  int8_t    slidingUnit;   // MACRO: TIME_UNIT_XXX
+  int8_t    timezoneInt;   // sma data expired if timezone changes.
+  int32_t   dstVgId;
+  char      indexName[TSDB_INDEX_NAME_LEN];
+  int32_t   exprLen;
+  int32_t   tagsFilterLen;
+  int32_t   numOfVgroups;
+  int64_t   indexUid;
+  tb_uid_t  tableUid;  // super/child/common table uid
+  int64_t   interval;
+  int64_t   offset;  // use unit by precision of DB
+  int64_t   sliding;
+  char*     expr;  // sma expression
+  char*     tagsFilter;
+  SVgEpSet* pVgEpSet;
 } STSma;  // Time-range-wise SMA
 
 typedef STSma SVCreateTSmaReq;
@@ -2405,7 +2405,7 @@ static int32_t tDecodeTSmaWrapper(SDecoder* pDecoder, STSmaWrapper* pReq) {
 }
 
 typedef struct {
-  int64_t indexUid;
+  int64_t     indexUid;
   STimeWindow queryWindow;
 } SVGetTsmaExpWndsReq;
 
diff --git a/source/common/src/tmsg.c b/source/common/src/tmsg.c
index 9c9f33ac96..7041c9478e 100644
--- a/source/common/src/tmsg.c
+++ b/source/common/src/tmsg.c
@@ -694,7 +694,6 @@ void tFreeSMAltertbReq(SMAlterStbReq *pReq) {
   pReq->pFields = NULL;
 }
 
-
 int32_t tSerializeSEpSet(void *buf, int32_t bufLen, const SEpSet *pEpset) {
   SEncoder encoder = {0};
   tEncoderInit(&encoder, buf, bufLen);
@@ -3674,12 +3673,12 @@ int32_t tEncodeTSma(SEncoder *pCoder, const STSma *pSma) {
     if (tEncodeCStr(pCoder, pSma->tagsFilter) < 0) return -1;
   }
   for (int32_t v = 0; v < pSma->numOfVgroups; ++v) {
-    if (tEncodeI32(pCoder, pSma->vgEpSet[v].vgId) < 0) return -1;
-    if (tEncodeI8(pCoder, pSma->vgEpSet[v].epSet.inUse) < 0) return -1;
-    int8_t numOfEps = pSma->vgEpSet[v].epSet.numOfEps;
+    if (tEncodeI32(pCoder, pSma->pVgEpSet[v].vgId) < 0) return -1;
+    if (tEncodeI8(pCoder, pSma->pVgEpSet[v].epSet.inUse) < 0) return -1;
+    int8_t numOfEps = pSma->pVgEpSet[v].epSet.numOfEps;
     if (tEncodeI8(pCoder, numOfEps) < 0) return -1;
     for (int32_t n = 0; n < numOfEps; ++n) {
-      const SEp *pEp = &pSma->vgEpSet[v].epSet.eps[n];
+      const SEp *pEp = &pSma->pVgEpSet[v].epSet.eps[n];
       if (tEncodeCStr(pCoder, pEp->fqdn) < 0) return -1;
       if (tEncodeU16(pCoder, pEp->port) < 0) return -1;
     }
@@ -3712,15 +3711,25 @@ int32_t tDecodeTSma(SDecoder *pCoder, STSma *pSma) {
   } else {
     pSma->tagsFilter = NULL;
   }
-  for (int32_t v = 0; v < pSma->numOfVgroups; ++v) {
-    if (tDecodeI32(pCoder, &pSma->vgEpSet[v].vgId) < 0) return -1;
-    if (tDecodeI8(pCoder, &pSma->vgEpSet[v].epSet.inUse) < 0) return -1;
-    if (tDecodeI8(pCoder, &pSma->vgEpSet[v].epSet.numOfEps) < 0) return -1;
-    int8_t numOfEps = pSma->vgEpSet[v].epSet.numOfEps;
-    for (int32_t n = 0; n < numOfEps; ++n) {
-      SEp *pEp = &pSma->vgEpSet[v].epSet.eps[n];
-      if (tDecodeCStrTo(pCoder, pEp->fqdn) < 0) return -1;
-      if (tDecodeU16(pCoder, &pEp->port) < 0) return -1;
+  if (pSma->numOfVgroups > 0) {
+    pSma->pVgEpSet = (SVgEpSet *)tDecoderMalloc(pCoder, pSma->numOfVgroups * sizeof(SVgEpSet));
+    if (!pSma->pVgEpSet) {
+      terrno = TSDB_CODE_OUT_OF_MEMORY;
+      return -1;
+    }
+
+    memset(pSma->pVgEpSet, 0, pSma->numOfVgroups * sizeof(SVgEpSet));
+
+    for (int32_t v = 0; v < pSma->numOfVgroups; ++v) {
+      if (tDecodeI32(pCoder, &pSma->pVgEpSet[v].vgId) < 0) return -1;
+      if (tDecodeI8(pCoder, &pSma->pVgEpSet[v].epSet.inUse) < 0) return -1;
+      if (tDecodeI8(pCoder, &pSma->pVgEpSet[v].epSet.numOfEps) < 0) return -1;
+      int8_t numOfEps = pSma->pVgEpSet[v].epSet.numOfEps;
+      for (int32_t n = 0; n < numOfEps; ++n) {
+        SEp *pEp = &pSma->pVgEpSet[v].epSet.eps[n];
+        if (tDecodeCStrTo(pCoder, pEp->fqdn) < 0) return -1;
+        if (tDecodeU16(pCoder, &pEp->port) < 0) return -1;
+      }
     }
   }
 
@@ -3765,7 +3774,7 @@ int32_t tDecodeSVDropTSmaReq(SDecoder *pCoder, SVDropTSmaReq *pReq) {
   return 0;
 }
 
-int32_t tEncodeSVGetTSmaExpWndsReq(SEncoder* pCoder, const SVGetTsmaExpWndsReq* pReq) {
+int32_t tEncodeSVGetTSmaExpWndsReq(SEncoder *pCoder, const SVGetTsmaExpWndsReq *pReq) {
   if (tStartEncode(pCoder) < 0) return -1;
 
   if (tEncodeI64(pCoder, pReq->indexUid) < 0) return -1;
@@ -3773,10 +3782,10 @@ int32_t tEncodeSVGetTSmaExpWndsReq(SEncoder* pCoder, const SVGetTsmaExpWndsReq*
   if (tEncodeI64(pCoder, pReq->queryWindow.ekey) < 0) return -1;
 
   tEndEncode(pCoder);
-  return 0;  
+  return 0;
 }
 
-int32_t tDecodeSVGetTsmaExpWndsReq(SDecoder* pCoder, SVGetTsmaExpWndsReq* pReq) {
+int32_t tDecodeSVGetTsmaExpWndsReq(SDecoder *pCoder, SVGetTsmaExpWndsReq *pReq) {
   if (tStartDecode(pCoder) < 0) return -1;
 
   if (tDecodeI64(pCoder, &pReq->indexUid) < 0) return -1;
@@ -3787,7 +3796,7 @@ int32_t tDecodeSVGetTsmaExpWndsReq(SDecoder* pCoder, SVGetTsmaExpWndsReq* pReq)
   return 0;
 }
 
-int32_t tEncodeSVGetTSmaExpWndsRsp(SEncoder* pCoder, const SVGetTsmaExpWndsRsp* pReq) {
+int32_t tEncodeSVGetTSmaExpWndsRsp(SEncoder *pCoder, const SVGetTsmaExpWndsRsp *pReq) {
   if (tStartEncode(pCoder) < 0) return -1;
 
   if (tEncodeI64(pCoder, pReq->indexUid) < 0) return -1;
@@ -3814,7 +3823,7 @@ int32_t tDecodeSVGetTsmaExpWndsRsp(SDecoder *pCoder, SVGetTsmaExpWndsRsp *pReq)
   return 0;
 }
 
-int32_t tEncodeSVDeleteReq(SEncoder* pCoder, const SVDeleteReq* pReq) {
+int32_t tEncodeSVDeleteReq(SEncoder *pCoder, const SVDeleteReq *pReq) {
   if (tStartEncode(pCoder) < 0) return -1;
 
   if (tEncodeI64(pCoder, pReq->delUid) < 0) return -1;
@@ -3832,7 +3841,7 @@ int32_t tEncodeSVDeleteReq(SEncoder* pCoder, const SVDeleteReq* pReq) {
   return 0;
 }
 
-int32_t tDecodeSVDeleteReq(SDecoder* pCoder, SVDeleteReq* pReq) {
+int32_t tDecodeSVDeleteReq(SDecoder *pCoder, SVDeleteReq *pReq) {
   if (tStartDecode(pCoder) < 0) return -1;
 
   if (tDecodeI64(pCoder, &pReq->delUid) < 0) return -1;
@@ -3850,7 +3859,7 @@ int32_t tDecodeSVDeleteReq(SDecoder* pCoder, SVDeleteReq* pReq) {
   return 0;
 }
 
-int32_t tEncodeSVDeleteRsp(SEncoder* pCoder, const SVDeleteRsp* pReq) {
+int32_t tEncodeSVDeleteRsp(SEncoder *pCoder, const SVDeleteRsp *pReq) {
   if (tStartEncode(pCoder) < 0) return -1;
 
   if (tEncodeI32(pCoder, pReq->code) < 0) return -1;
@@ -3860,7 +3869,7 @@ int32_t tEncodeSVDeleteRsp(SEncoder* pCoder, const SVDeleteRsp* pReq) {
   return 0;
 }
 
-int32_t tDecodeSVDeleteRsp(SDecoder* pCoder, SVDeleteRsp* pReq) {
+int32_t tDecodeSVDeleteRsp(SDecoder *pCoder, SVDeleteRsp *pReq) {
   if (tStartDecode(pCoder) < 0) return -1;
 
   if (tDecodeI32(pCoder, &pReq->code) < 0) return -1;
@@ -4502,7 +4511,7 @@ int32_t tDecodeSVAlterTbRsp(SDecoder *pDecoder, SVAlterTbRsp *pRsp) {
 }
 
 int32_t tDeserializeSVAlterTbRsp(void *buf, int32_t bufLen, SVAlterTbRsp *pRsp) {
-  int32_t meta = 0;
+  int32_t  meta = 0;
   SDecoder decoder = {0};
   tDecoderInit(&decoder, buf, bufLen);
 
@@ -4543,7 +4552,7 @@ int32_t tDecodeSMAlterStbRsp(SDecoder *pDecoder, SMAlterStbRsp *pRsp) {
 }
 
 int32_t tDeserializeSMAlterStbRsp(void *buf, int32_t bufLen, SMAlterStbRsp *pRsp) {
-  int32_t meta = 0;
+  int32_t  meta = 0;
   SDecoder decoder = {0};
   tDecoderInit(&decoder, buf, bufLen);
 
@@ -4559,7 +4568,7 @@ int32_t tDeserializeSMAlterStbRsp(void *buf, int32_t bufLen, SMAlterStbRsp *pRsp
   return 0;
 }
 
-void tFreeSMAlterStbRsp(SMAlterStbRsp* pRsp) {
+void tFreeSMAlterStbRsp(SMAlterStbRsp *pRsp) {
   if (NULL == pRsp) {
     return;
   }
@@ -4569,6 +4578,3 @@ void tFreeSMAlterStbRsp(SMAlterStbRsp* pRsp) {
     taosMemoryFree(pRsp->pMeta);
   }
 }
-
-
-
diff --git a/source/dnode/mnode/impl/inc/mndDef.h b/source/dnode/mnode/impl/inc/mndDef.h
index 83a36f4b0d..8ea172ef94 100644
--- a/source/dnode/mnode/impl/inc/mndDef.h
+++ b/source/dnode/mnode/impl/inc/mndDef.h
@@ -298,28 +298,30 @@ typedef struct {
 } SVgObj;
 
 typedef struct {
-  char    name[TSDB_TABLE_FNAME_LEN];
-  char    stb[TSDB_TABLE_FNAME_LEN];
-  char    db[TSDB_DB_FNAME_LEN];
-  int64_t createdTime;
-  int64_t uid;
-  int64_t stbUid;
-  int64_t dbUid;
-  int8_t  intervalUnit;
-  int8_t  slidingUnit;
-  int8_t  timezone;
-  int32_t dstVgId;  // for stream
-  int64_t interval;
-  int64_t offset;
-  int64_t sliding;
-  int32_t exprLen;  // strlen + 1
-  int32_t tagsFilterLen;
-  int32_t sqlLen;
-  int32_t astLen;
-  char*   expr;
-  char*   tagsFilter;
-  char*   sql;
-  char*   ast;
+  char      name[TSDB_TABLE_FNAME_LEN];
+  char      stb[TSDB_TABLE_FNAME_LEN];
+  char      db[TSDB_DB_FNAME_LEN];
+  int64_t   createdTime;
+  int64_t   uid;
+  int64_t   stbUid;
+  int64_t   dbUid;
+  int8_t    intervalUnit;
+  int8_t    slidingUnit;
+  int8_t    timezone;
+  int32_t   dstVgId;  // for stream
+  int64_t   interval;
+  int64_t   offset;
+  int64_t   sliding;
+  int32_t   exprLen;  // strlen + 1
+  int32_t   tagsFilterLen;
+  int32_t   sqlLen;
+  int32_t   astLen;
+  int32_t   numOfVgroups;
+  char*     expr;
+  char*     tagsFilter;
+  char*     sql;
+  char*     ast;
+  SVgEpSet* pVgEpSet;
 } SSmaObj;
 
 typedef struct {
diff --git a/source/dnode/mnode/impl/src/mndSma.c b/source/dnode/mnode/impl/src/mndSma.c
index e8e23ccc2a..f9cb63dc74 100644
--- a/source/dnode/mnode/impl/src/mndSma.c
+++ b/source/dnode/mnode/impl/src/mndSma.c
@@ -36,6 +36,7 @@ static SSdbRow *mndSmaActionDecode(SSdbRaw *pRaw);
 static int32_t  mndSmaActionInsert(SSdb *pSdb, SSmaObj *pSma);
 static int32_t  mndSmaActionDelete(SSdb *pSdb, SSmaObj *pSpSmatb);
 static int32_t  mndSmaActionUpdate(SSdb *pSdb, SSmaObj *pOld, SSmaObj *pNew);
+static int32_t  mndSmaGetVgEpSet(SMnode *pMnode, SDbObj *pDb, SVgEpSet **ppVgEpSet, int32_t *numOfVgroups);
 static int32_t  mndProcessMCreateSmaReq(SRpcMsg *pReq);
 static int32_t  mndProcessMDropSmaReq(SRpcMsg *pReq);
 static int32_t  mndProcessGetSmaReq(SRpcMsg *pReq);
@@ -262,7 +263,9 @@ static void *mndBuildVCreateSmaReq(SMnode *pMnode, SVgObj *pVgroup, SSmaObj *pSm
   req.sliding = pSma->sliding;
   req.expr = pSma->expr;
   req.tagsFilter = pSma->tagsFilter;
-
+  req.numOfVgroups = pSma->numOfVgroups;
+  req.pVgEpSet = pSma->pVgEpSet;
+  
   // get length
   int32_t ret = 0;
   tEncodeSize(tEncodeSVCreateTSmaReq, &req, contLen, ret);
@@ -420,6 +423,15 @@ static int32_t mndSetCreateSmaVgroupRedoActions(SMnode *pMnode, STrans *pTrans,
   mndReleaseDnode(pMnode, pDnode);
 
   // todo add sma info here
+  SVgEpSet *pVgEpSet = NULL;
+  int32_t   numOfVgroups = 0;
+  if (mndSmaGetVgEpSet(pMnode, pDb, &pVgEpSet, &numOfVgroups) != 0) {
+    return -1;
+  }
+
+  pSma->pVgEpSet = pVgEpSet;
+  pSma->numOfVgroups = numOfVgroups;
+
   int32_t smaContLen = 0;
   void   *pSmaReq = mndBuildVCreateSmaReq(pMnode, pVgroup, pSma, &smaContLen);
   if (pSmaReq == NULL) return -1;
@@ -964,3 +976,52 @@ static void mndCancelGetNextSma(SMnode *pMnode, void *pIter) {
   SSdb *pSdb = pMnode->pSdb;
   sdbCancelFetch(pSdb, pIter);
 }
+
+static int32_t mndSmaGetVgEpSet(SMnode *pMnode, SDbObj *pDb, SVgEpSet **ppVgEpSet, int32_t *numOfVgroups) {
+  SSdb     *pSdb = pMnode->pSdb;
+  SVgObj   *pVgroup = NULL;
+  void     *pIter = NULL;
+  SVgEpSet *pVgEpSet = NULL;
+  int32_t   nAllocVgs = 16;
+  int32_t   nVgs = 0;
+
+  pVgEpSet = taosMemoryCalloc(nAllocVgs, sizeof(SVgEpSet));
+  if (!pVgEpSet) {
+    terrno = TSDB_CODE_OUT_OF_MEMORY;
+    return -1;
+  }
+
+  while (1) {
+    pIter = sdbFetch(pSdb, SDB_VGROUP, pIter, (void **)&pVgroup);
+    if (pIter == NULL) break;
+    if (pVgroup->dbUid != pDb->uid) {
+      sdbRelease(pSdb, pVgroup);
+      continue;
+    }
+
+    if (nVgs >= nAllocVgs) {
+      void *p = taosMemoryRealloc(pVgEpSet, nAllocVgs * 2 * sizeof(SVgEpSet));
+      if (!p) {
+        taosMemoryFree(pVgEpSet);
+        sdbCancelFetch(pSdb, pIter);
+        sdbRelease(pSdb, pVgroup);
+        terrno = TSDB_CODE_OUT_OF_MEMORY;
+        return -1;
+      }
+      pVgEpSet = (SVgEpSet *)p;
+      nAllocVgs *= 2;
+    }
+
+    (pVgEpSet + nVgs)->vgId = pVgroup->vgId;
+    (pVgEpSet + nVgs)->epSet = mndGetVgroupEpset(pMnode, pVgroup);
+
+    ++nVgs;
+
+    sdbRelease(pSdb, pVgroup);
+  }
+
+  *ppVgEpSet = pVgEpSet;
+  *numOfVgroups = nVgs;
+
+  return 0;
+}

From 5033cf0c8419bd7493ad15b687e27f9155214841 Mon Sep 17 00:00:00 2001
From: "wenzhouwww@live.cn" <wenzhouwww@live.cn>
Date: Mon, 6 Jun 2022 14:04:51 +0800
Subject: [PATCH 07/34] update case for sample

---
 tests/system-test/2-query/sample.py | 26 +++++++++++++-------------
 1 file changed, 13 insertions(+), 13 deletions(-)

diff --git a/tests/system-test/2-query/sample.py b/tests/system-test/2-query/sample.py
index 94e06347d2..4b651b56e7 100644
--- a/tests/system-test/2-query/sample.py
+++ b/tests/system-test/2-query/sample.py
@@ -427,10 +427,10 @@ class TDTestCase:
         # err9 = {"col": "st1"}
         # self.checksample(**err9)          # col: tag
         tdSql.query(" select sample(st1 ,1) from t1 ")
-        err10 = {"col": 1}
-        self.checksample(**err10)         # col: value
-        err11 = {"col": "NULL"}
-        self.checksample(**err11)         # col: NULL
+        # err10 = {"col": 1}
+        # self.checksample(**err10)         # col: value
+        # err11 = {"col": "NULL"}
+        # self.checksample(**err11)         # col: NULL
         err12 = {"col": "%_"}
         self.checksample(**err12)         # col: %_
         err13 = {"col": "c3"}
@@ -445,12 +445,12 @@ class TDTestCase:
         self.checksample(**err17)         # nchar col
         err18 = {"col": "c6"}
         self.checksample(**err18)         # bool col
-        err19 = {"col": "'c1'"}
-        self.checksample(**err19)         # col: string
+        # err19 = {"col": "'c1'"}
+        # self.checksample(**err19)         # col: string
         err20 = {"col": None}
         self.checksample(**err20)         # col: None
-        err21 = {"col": "''"}
-        self.checksample(**err21)         # col: ''
+        # err21 = {"col": "''"}
+        # self.checksample(**err21)         # col: ''
         err22 = {"col": "tt1.c1"}
         self.checksample(**err22)         # not table_expr col
         err23 = {"col": "t1"}
@@ -459,10 +459,10 @@ class TDTestCase:
         self.checksample(**err24)         # stbname
         err25 = {"col": "db"}
         self.checksample(**err25)         # datbasename
-        err26 = {"col": "True"}
-        self.checksample(**err26)         # col: BOOL 1
-        err27 = {"col": True}
-        self.checksample(**err27)         # col: BOOL 2
+        # err26 = {"col": "True"}
+        # self.checksample(**err26)         # col: BOOL 1
+        # err27 = {"col": True}
+        # self.checksample(**err27)         # col: BOOL 2
         err28 = {"col": "*"}
         self.checksample(**err28)         # col: all col
         err29 = {"func": "sample[", "r_comm": "]"}
@@ -678,7 +678,7 @@ class TDTestCase:
         tdSql.error(" select sample(c1,tbname) from t1 ")
         tdSql.error(" select sample(c1,ts) from t1 ")
         tdSql.error(" select sample(c1,false) from t1 ")
-        tdSql.error(" select sample(123,1) from t1 ")
+        tdSql.query(" select sample(123,1) from t1 ")
         
         tdSql.query(" select sample(c1,2) from t1 ")
         tdSql.checkRows(2)

From bd3e0a76fb3c23cf3d1ce955b7d4cd78598341ce Mon Sep 17 00:00:00 2001
From: "wenzhouwww@live.cn" <wenzhouwww@live.cn>
Date: Mon, 6 Jun 2022 14:13:56 +0800
Subject: [PATCH 08/34] update case for statedurcation

---
 tests/system-test/2-query/function_stateduration.py | 9 +++++++--
 1 file changed, 7 insertions(+), 2 deletions(-)

diff --git a/tests/system-test/2-query/function_stateduration.py b/tests/system-test/2-query/function_stateduration.py
index b25a658469..9aa6fdaefa 100644
--- a/tests/system-test/2-query/function_stateduration.py
+++ b/tests/system-test/2-query/function_stateduration.py
@@ -85,8 +85,8 @@ class TDTestCase:
             "select stateduration(c1 ,'GT','*',1s) from t1",
             "select stateduration(c1 ,'GT',ts,1s) from t1",
             "select stateduration(c1 ,'GT',max(c1),1s) from t1",
-            "select stateduration(abs(c1) ,'GT',1,1s) from t1",
-            "select stateduration(c1+2 ,'GT',1,1s) from t1",
+            # "select stateduration(abs(c1) ,'GT',1,1s) from t1",
+            # "select stateduration(c1+2 ,'GT',1,1s) from t1",
             "select stateduration(c1 ,'GT',1,1u) from t1",
             "select stateduration(c1 ,'GT',1,now) from t1",
             "select stateduration(c1 ,'GT','1',1s) from t1",
@@ -323,6 +323,11 @@ class TDTestCase:
         tdSql.checkData(0, 0, None)
         tdSql.checkData(1, 0, 0.000000000)
         tdSql.checkData(3, 0, -86404.000000000)
+
+        tdSql.query("select stateduration(abs(c1) ,'GT',1,1s) from t1")
+        tdSql.checkRows(12)
+        tdSql.query("select stateduration(c1+2 ,'GT',1,1s) from t1")
+        tdSql.checkRows(12)
         
 
         # bug for stable

From 4f33ecbf23129cf576d67e93e80dd36dbff06fdb Mon Sep 17 00:00:00 2001
From: "wenzhouwww@live.cn" <wenzhouwww@live.cn>
Date: Mon, 6 Jun 2022 14:15:20 +0800
Subject: [PATCH 09/34] update case

---
 tests/system-test/2-query/statecount.py | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/tests/system-test/2-query/statecount.py b/tests/system-test/2-query/statecount.py
index 2634d9a9ab..ccda55f7ba 100644
--- a/tests/system-test/2-query/statecount.py
+++ b/tests/system-test/2-query/statecount.py
@@ -85,8 +85,8 @@ class TDTestCase:
             "select statecount(c1 ,'GT','*') from t1",
             "select statecount(c1 ,'GT',ts) from t1",
             "select statecount(c1 ,'GT',max(c1)) from t1",
-            "select statecount(abs(c1) ,'GT',1) from t1",
-            "select statecount(c1+2 ,'GT',1) from t1",
+            # "select statecount(abs(c1) ,'GT',1) from t1",
+            # "select statecount(c1+2 ,'GT',1) from t1",
             "select statecount(c1 ,'GT',1,1u) from t1",
             "select statecount(c1 ,'GT',1,now) from t1",
             "select statecount(c1 ,'GT','1') from t1",

From 37b7f06b4a663b4b1864671c400e410527d028f4 Mon Sep 17 00:00:00 2001
From: "wenzhouwww@live.cn" <wenzhouwww@live.cn>
Date: Mon, 6 Jun 2022 14:18:36 +0800
Subject: [PATCH 10/34] update case for elapsed function

---
 tests/system-test/2-query/elapsed.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/tests/system-test/2-query/elapsed.py b/tests/system-test/2-query/elapsed.py
index 017090128d..1553e06914 100644
--- a/tests/system-test/2-query/elapsed.py
+++ b/tests/system-test/2-query/elapsed.py
@@ -141,7 +141,7 @@ class TDTestCase:
         tablenames = ["sub_table1_1","sub_table1_2","sub_table1_3","sub_table2_1","sub_table2_2","sub_table2_3","regular_table_1","regular_table_2","regular_table_3"]
 
         abnormal_list = ["()","(NULL)","(*)","(abc)","( , )","(NULL,*)","( ,NULL)","(%)","(+)","(*,)","(*, /)","(ts,*)" "(ts,tbname*10)","(ts,tagname)",
-        "(ts,2d+3m-2s,NULL)","(ts+1d,10s)","(ts+10d,NULL)" ,"(ts,now -1m%1d)","(ts+10d)","(ts+10d,_c0)","(ts+10d,)","(ts,%)","(ts, , m)","(ts,abc)","(ts,/)","(ts,*)","(ts,1s,100)",
+        "(ts,2d+3m-2s,NULL)","(ts+10d,NULL)" ,"(ts,now -1m%1d)","(ts+10d,_c0)","(ts+10d,)","(ts,%)","(ts, , m)","(ts,abc)","(ts,/)","(ts,*)","(ts,1s,100)",
         "(ts,1s,abc)","(ts,1s,_c0)","(ts,1s,*)","(ts,1s,NULL)","(ts,,_c0)","(ts,tbname,ts)","(ts,0,tbname)","('2021-11-18 00:00:10')","('2021-11-18 00:00:10', 1s)",
         "('2021-11-18T00:00:10+0800', '1s')","('2021-11-18T00:00:10Z', '1s')","('2021-11-18T00:00:10+0800', 10000000d,)","('ts', ,2021-11-18T00:00:10+0800, )"]
         

From a8be204254e05a7b2b604bf380d0e14f80379ac0 Mon Sep 17 00:00:00 2001
From: shenglian zhou <slzhou@taosdata.com>
Date: Mon, 6 Jun 2022 14:37:00 +0800
Subject: [PATCH 11/34] feat: add multiwaysort merge operator

---
 source/libs/executor/inc/executorimpl.h |   3 +-
 source/libs/executor/src/executorimpl.c |   9 ++
 source/libs/executor/src/sortoperator.c | 156 ++++++++++++++++++++++++
 source/libs/index/src/indexFilter.c     |   3 +
 4 files changed, 170 insertions(+), 1 deletion(-)

diff --git a/source/libs/executor/inc/executorimpl.h b/source/libs/executor/inc/executorimpl.h
index 7d597217ee..bd33d7f3c7 100644
--- a/source/libs/executor/inc/executorimpl.h
+++ b/source/libs/executor/inc/executorimpl.h
@@ -773,7 +773,8 @@ SOperatorInfo* createAggregateOperatorInfo(SOperatorInfo* downstream, SExprInfo*
 SOperatorInfo* createProjectOperatorInfo(SOperatorInfo* downstream, SExprInfo* pExprInfo, int32_t num, SSDataBlock* pResBlock, SLimit* pLimit, SLimit* pSlimit, SNode* pCondition, SExecTaskInfo* pTaskInfo);
 SOperatorInfo *createSortOperatorInfo(SOperatorInfo* downstream, SSDataBlock* pResBlock, SArray* pSortInfo, SExprInfo* pExprInfo, int32_t numOfCols,
                                       SArray* pIndexMap, SExecTaskInfo* pTaskInfo);
-
+SOperatorInfo* createMultiwaySortMergeOperatorInfo(SOperatorInfo** downStreams, int32_t numStreams,
+                                                   SSDataBlock* pResBlock, SArray* pSortInfo, SExecTaskInfo* pTaskInfo);
 SOperatorInfo* createSortedMergeOperatorInfo(SOperatorInfo** downstream, int32_t numOfDownstream, SExprInfo* pExprInfo, int32_t num, SArray* pSortInfo, SArray* pGroupInfo, SExecTaskInfo* pTaskInfo);
 
 SOperatorInfo* createIntervalOperatorInfo(SOperatorInfo* downstream, SExprInfo* pExprInfo, int32_t numOfCols,
diff --git a/source/libs/executor/src/executorimpl.c b/source/libs/executor/src/executorimpl.c
index 92b3195f36..d094dc0203 100644
--- a/source/libs/executor/src/executorimpl.c
+++ b/source/libs/executor/src/executorimpl.c
@@ -4517,6 +4517,15 @@ SOperatorInfo* createOperatorTree(SPhysiNode* pPhyNode, SExecTaskInfo* pTaskInfo
         extractColMatchInfo(pSortPhyNode->pTargets, pDescNode, &numOfOutputCols, pTaskInfo, COL_MATCH_FROM_SLOT_ID);
 
     pOptr = createSortOperatorInfo(ops[0], pResBlock, info, pExprInfo, numOfCols, pColList, pTaskInfo);
+  }  else if (QUERY_NODE_PHYSICAL_PLAN_MERGE == type) {
+    SMergePhysiNode* pMergePhyNode = (SMergePhysiNode*)pPhyNode;
+
+    SDataBlockDescNode* pDescNode = pPhyNode->pOutputDataBlockDesc;
+    SSDataBlock* pResBlock = createResDataBlock(pDescNode);
+
+    SArray*      sortInfo = createSortInfo(pMergePhyNode->pMergeKeys);
+
+    pOptr = createMultiwaySortMergeOperatorInfo(ops, size, pResBlock, sortInfo, pTaskInfo);
   } else if (QUERY_NODE_PHYSICAL_PLAN_SESSION_WINDOW == type) {
     SSessionWinodwPhysiNode* pSessionNode = (SSessionWinodwPhysiNode*)pPhyNode;
 
diff --git a/source/libs/executor/src/sortoperator.c b/source/libs/executor/src/sortoperator.c
index 5e1b47cc91..0618c80682 100644
--- a/source/libs/executor/src/sortoperator.c
+++ b/source/libs/executor/src/sortoperator.c
@@ -215,3 +215,159 @@ int32_t getExplainExecInfo(SOperatorInfo* pOptr, void** pOptrExplain, uint32_t*
   *len = sizeof(SSortExecInfo);
   return TSDB_CODE_SUCCESS;
 }
+
+typedef struct SMultiwaySortMergeOperatorInfo {
+  SOptrBasicInfo binfo;
+
+  int32_t      bufPageSize;
+  uint32_t     sortBufSize;    // max buffer size for in-memory sort
+
+  SArray*      pSortInfo;
+  SSortHandle* pSortHandle;
+
+  int64_t      startTs;       // sort start time
+} SMultiwaySortMergeOperatorInfo;
+
+SSDataBlock* getMultiwaySortMergedBlockData(SSortHandle* pHandle, SSDataBlock* pDataBlock, int32_t capacity) {
+  blockDataCleanup(pDataBlock);
+
+  blockDataEnsureCapacity(pDataBlock, capacity);
+
+  while (1) {
+    STupleHandle* pTupleHandle = tsortNextTuple(pHandle);
+    if (pTupleHandle == NULL) {
+      break;
+    }
+
+    appendOneRowToDataBlock(pDataBlock, pTupleHandle);
+    if (pDataBlock->info.rows >= capacity) {
+      return pDataBlock;
+    }
+  }
+
+  return (pDataBlock->info.rows > 0) ? pDataBlock : NULL;
+}
+
+
+int32_t doOpenMultiwaySortMergeOperator(SOperatorInfo* pOperator) {
+  SMultiwaySortMergeOperatorInfo * pInfo = pOperator->info;
+  SExecTaskInfo* pTaskInfo = pOperator->pTaskInfo;
+
+  if (OPTR_IS_OPENED(pOperator)) {
+    return TSDB_CODE_SUCCESS;
+  }
+
+  pInfo->startTs = taosGetTimestampUs();
+
+  int32_t numOfBufPage = pInfo->sortBufSize / pInfo->bufPageSize;
+
+  pInfo->pSortHandle = tsortCreateSortHandle(pInfo->pSortInfo, NULL, SORT_MULTISOURCE_MERGE,
+                                             pInfo->bufPageSize, numOfBufPage, NULL, pTaskInfo->id.str);
+
+  tsortSetFetchRawDataFp(pInfo->pSortHandle, loadNextDataBlock, NULL, NULL);
+
+  for (int32_t i = 0; i < pOperator->numOfDownstream; ++i) {
+    SSortSource ps = {0};
+    ps.param = pOperator->pDownstream[i];
+    tsortAddSource(pInfo->pSortHandle, &ps);
+  }
+
+  int32_t code = tsortOpen(pInfo->pSortHandle);
+
+  if (code != TSDB_CODE_SUCCESS) {
+    longjmp(pTaskInfo->env, terrno);
+  }
+
+  pOperator->cost.openCost = (taosGetTimestampUs() - pInfo->startTs)/1000.0;
+  pOperator->status = OP_RES_TO_RETURN;
+
+  OPTR_SET_OPENED(pOperator);
+  return TSDB_CODE_SUCCESS;
+}
+
+SSDataBlock* doMultiwaySortMerge(SOperatorInfo* pOperator) {
+  if (pOperator->status == OP_EXEC_DONE) {
+    return NULL;
+  }
+
+  SExecTaskInfo*     pTaskInfo = pOperator->pTaskInfo;
+  SSortOperatorInfo* pInfo = pOperator->info;
+
+  int32_t code = pOperator->fpSet._openFn(pOperator);
+  if (code != TSDB_CODE_SUCCESS) {
+    longjmp(pTaskInfo->env, code);
+  }
+
+  SSDataBlock* pBlock = getMultiwaySortMergedBlockData(pInfo->pSortHandle, pInfo->binfo.pRes, pOperator->resultInfo.capacity);
+
+  if (pBlock != NULL) {
+    pOperator->resultInfo.totalRows += pBlock->info.rows;
+  } else {
+    doSetOperatorCompleted(pOperator);
+  }
+  return pBlock;
+}
+
+void destroyMultiwaySortMergeOperatorInfo(void* param, int32_t numOfOutput) {
+  SSortOperatorInfo* pInfo = (SSortOperatorInfo*)param;
+  pInfo->binfo.pRes = blockDataDestroy(pInfo->binfo.pRes);
+
+  taosArrayDestroy(pInfo->pSortInfo);
+}
+
+int32_t getMultiwaySortMergeExplainExecInfo(SOperatorInfo* pOptr, void** pOptrExplain, uint32_t* len) {
+  ASSERT(pOptr != NULL);
+  SSortExecInfo* pInfo = taosMemoryCalloc(1, sizeof(SSortExecInfo));
+
+  SMultiwaySortMergeOperatorInfo *pOperatorInfo = (SMultiwaySortMergeOperatorInfo*)pOptr->info;
+
+  *pInfo = tsortGetSortExecInfo(pOperatorInfo->pSortHandle);
+  *pOptrExplain = pInfo;
+  *len = sizeof(SSortExecInfo);
+  return TSDB_CODE_SUCCESS;
+}
+
+SOperatorInfo* createMultiwaySortMergeOperatorInfo(SOperatorInfo** downStreams, int32_t numStreams,
+                                                   SSDataBlock* pResBlock, SArray* pSortInfo, SExecTaskInfo* pTaskInfo) {
+  SMultiwaySortMergeOperatorInfo* pInfo = taosMemoryCalloc(1, sizeof(SMultiwaySortMergeOperatorInfo));
+  SOperatorInfo*     pOperator = taosMemoryCalloc(1, sizeof(SOperatorInfo));
+  int32_t            rowSize = pResBlock->info.rowSize;
+
+  if (pInfo == NULL || pOperator == NULL || rowSize > 100 * 1024 * 1024) {
+    goto _error;
+  }
+
+  pInfo->binfo.pRes   = pResBlock;
+
+  initResultSizeInfo(pOperator, 1024);
+
+  pInfo->pSortInfo    = pSortInfo;
+  pOperator->name     = "MultiwaySortMerge";
+  pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_MERGE;
+  pOperator->blocking = true;
+  pOperator->status   = OP_NOT_OPENED;
+  pOperator->info     = pInfo;
+
+  pInfo->bufPageSize = rowSize < 1024 ? 1024 : rowSize * 2;
+  pInfo->sortBufSize = pInfo->bufPageSize * 16;
+
+  pOperator->pTaskInfo = pTaskInfo;
+  pOperator->fpSet =
+      createOperatorFpSet(doOpenMultiwaySortMergeOperator, doMultiwaySortMerge,
+                          NULL,
+                          NULL, destroyMultiwaySortMergeOperatorInfo,
+                          NULL, NULL,
+                          getMultiwaySortMergeExplainExecInfo);
+
+  int32_t code = appendDownstream(pOperator, downStreams, numStreams);
+  if (code != TSDB_CODE_SUCCESS) {
+    goto _error;
+  }
+  return pOperator;
+
+_error:
+  pTaskInfo->code = TSDB_CODE_OUT_OF_MEMORY;
+  taosMemoryFree(pInfo);
+  taosMemoryFree(pOperator);
+  return NULL;
+}
\ No newline at end of file
diff --git a/source/libs/index/src/indexFilter.c b/source/libs/index/src/indexFilter.c
index 34251ab918..72d29469db 100644
--- a/source/libs/index/src/indexFilter.c
+++ b/source/libs/index/src/indexFilter.c
@@ -171,7 +171,10 @@ static int32_t sifInitJsonParam(SNode *node, SIFParam *param, SIFCtx *ctx) {
   param->colId = l->colId;
   param->colValType = l->node.resType.type;
   memcpy(param->dbName, l->dbName, sizeof(l->dbName));
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wformat-overflow"
   sprintf(param->colName, "%s_%s", l->colName, r->literal);
+#pragma GCC diagnostic pop
   param->colValType = r->typeData;
   return 0;
   // memcpy(param->colName, l->colName, sizeof(l->colName));

From 58389568319199119b461292107367e79ea88743 Mon Sep 17 00:00:00 2001
From: Shengliang Guan <slguan@taosdata.com>
Date: Mon, 6 Jun 2022 11:49:30 +0800
Subject: [PATCH 12/34] refactor: dnode online

---
 source/dnode/mnode/impl/inc/mndDnode.h  |   2 +-
 source/dnode/mnode/impl/inc/mndMnode.h  |   1 +
 source/dnode/mnode/impl/inc/mndVgroup.h |  22 +-
 source/dnode/mnode/impl/src/mndDb.c     | 108 +-------
 source/dnode/mnode/impl/src/mndDnode.c  |  80 +++---
 source/dnode/mnode/impl/src/mndMain.c   |   2 +-
 source/dnode/mnode/impl/src/mndMnode.c  |  22 +-
 source/dnode/mnode/impl/src/mndVgroup.c | 345 ++++++++++++++++++------
 source/dnode/mnode/sdb/src/sdbRaw.c     |   6 +-
 9 files changed, 344 insertions(+), 244 deletions(-)

diff --git a/source/dnode/mnode/impl/inc/mndDnode.h b/source/dnode/mnode/impl/inc/mndDnode.h
index c76186c0a2..cf1e7422be 100644
--- a/source/dnode/mnode/impl/inc/mndDnode.h
+++ b/source/dnode/mnode/impl/inc/mndDnode.h
@@ -28,7 +28,7 @@ SDnodeObj *mndAcquireDnode(SMnode *pMnode, int32_t dnodeId);
 void       mndReleaseDnode(SMnode *pMnode, SDnodeObj *pDnode);
 SEpSet     mndGetDnodeEpset(SDnodeObj *pDnode);
 int32_t    mndGetDnodeSize(SMnode *pMnode);
-bool       mndIsDnodeOnline(SMnode *pMnode, SDnodeObj *pDnode, int64_t curMs);
+bool       mndIsDnodeOnline(SDnodeObj *pDnode, int64_t curMs);
 
 #ifdef __cplusplus
 }
diff --git a/source/dnode/mnode/impl/inc/mndMnode.h b/source/dnode/mnode/impl/inc/mndMnode.h
index fd62b3ce75..a433af9947 100644
--- a/source/dnode/mnode/impl/inc/mndMnode.h
+++ b/source/dnode/mnode/impl/inc/mndMnode.h
@@ -28,6 +28,7 @@ SMnodeObj *mndAcquireMnode(SMnode *pMnode, int32_t mnodeId);
 void       mndReleaseMnode(SMnode *pMnode, SMnodeObj *pObj);
 bool       mndIsMnode(SMnode *pMnode, int32_t dnodeId);
 void       mndGetMnodeEpSet(SMnode *pMnode, SEpSet *pEpSet);
+int32_t    mndSetDropMnodeInfoToTrans(SMnode *pMnode, STrans *pTrans, SMnodeObj *pObj);
 
 #ifdef __cplusplus
 }
diff --git a/source/dnode/mnode/impl/inc/mndVgroup.h b/source/dnode/mnode/impl/inc/mndVgroup.h
index 2119320de5..89f93d30b5 100644
--- a/source/dnode/mnode/impl/inc/mndVgroup.h
+++ b/source/dnode/mnode/impl/inc/mndVgroup.h
@@ -30,15 +30,21 @@ SSdbRaw *mndVgroupActionEncode(SVgObj *pVgroup);
 SEpSet   mndGetVgroupEpset(SMnode *pMnode, const SVgObj *pVgroup);
 int32_t  mndGetVnodesNum(SMnode *pMnode, int32_t dnodeId);
 
-int32_t mndAllocSmaVgroup(SMnode *pMnode, SDbObj *pDb, SVgObj *pVgroup);
-int32_t mndAllocVgroup(SMnode *pMnode, SDbObj *pDb, SVgObj **ppVgroups);
-SArray *mndBuildDnodesArray(SMnode *pMnode);
-int32_t mndAddVnodeToVgroup(SMnode *pMnode, SVgObj *pVgroup, SArray *pArray);
-int32_t mndRemoveVnodeFromVgroup(SMnode *pMnode, SVgObj *pVgroup, SArray *pArray, SVnodeGid *pVgId);
+SArray *mndBuildDnodesArray(SMnode *, int32_t exceptDnodeId);
+int32_t mndAllocSmaVgroup(SMnode *, SDbObj *pDb, SVgObj *pVgroup);
+int32_t mndAllocVgroup(SMnode *, SDbObj *pDb, SVgObj **ppVgroups);
+int32_t mndAddVnodeToVgroup(SMnode *, SVgObj *pVgroup, SArray *pArray);
+int32_t mndRemoveVnodeFromVgroup(SMnode *, SVgObj *pVgroup, SArray *pArray, SVnodeGid *pDelVgid);
+int32_t mndAddCreateVnodeAction(SMnode *, STrans *pTrans, SDbObj *pDb, SVgObj *pVgroup, SVnodeGid *pVgid, bool standby);
+int32_t mndAddAlterVnodeConfirmAction(SMnode *, STrans *pTrans, SDbObj *pDb, SVgObj *pVgroup);
+int32_t mndAddAlterVnodeAction(SMnode *, STrans *pTrans, SDbObj *pDb, SVgObj *pVgroup, tmsg_t msgType);
+int32_t mndAddDropVnodeAction(SMnode *, STrans *pTrans, SDbObj *pDb, SVgObj *pVgroup, SVnodeGid *pVgid, bool isRedo);
+int32_t mndSetMoveVgroupInfoToTrans(SMnode *, STrans *pTrans, SDbObj *pDb, SVgObj *pVgroup, int32_t vn, SArray *pArray);
+int32_t mndSetMoveVgroupsInfoToTrans(SMnode *, STrans *pTrans, int32_t dropDnodeId);
 
-void *mndBuildCreateVnodeReq(SMnode *pMnode, SDnodeObj *pDnode, SDbObj *pDb, SVgObj *pVgroup, int32_t *pContLen, bool standby);
-void *mndBuildDropVnodeReq(SMnode *pMnode, SDnodeObj *pDnode, SDbObj *pDb, SVgObj *pVgroup, int32_t *pContLen);
-void *mndBuildAlterVnodeReq(SMnode *pMnode, SDbObj *pDb, SVgObj *pVgroup, int32_t *pContLen);
+void *mndBuildCreateVnodeReq(SMnode *, SDnodeObj *pDnode, SDbObj *pDb, SVgObj *pVgroup, int32_t *cntlen, bool standby);
+void *mndBuildDropVnodeReq(SMnode *, SDnodeObj *pDnode, SDbObj *pDb, SVgObj *pVgroup, int32_t *pContLen);
+void *mndBuildAlterVnodeReq(SMnode *, SDbObj *pDb, SVgObj *pVgroup, int32_t *pContLen);
 
 #ifdef __cplusplus
 }
diff --git a/source/dnode/mnode/impl/src/mndDb.c b/source/dnode/mnode/impl/src/mndDb.c
index 1ce1a2590a..d2b468d588 100644
--- a/source/dnode/mnode/impl/src/mndDb.c
+++ b/source/dnode/mnode/impl/src/mndDb.c
@@ -263,111 +263,6 @@ void mndReleaseDb(SMnode *pMnode, SDbObj *pDb) {
   sdbRelease(pSdb, pDb);
 }
 
-static int32_t mndAddCreateVnodeAction(SMnode *pMnode, STrans *pTrans, SDbObj *pDb, SVgObj *pVgroup, SVnodeGid *pVgid,
-                                       bool standby) {
-  STransAction action = {0};
-
-  SDnodeObj *pDnode = mndAcquireDnode(pMnode, pVgid->dnodeId);
-  if (pDnode == NULL) return -1;
-  action.epSet = mndGetDnodeEpset(pDnode);
-  mndReleaseDnode(pMnode, pDnode);
-
-  int32_t contLen = 0;
-  void   *pReq = mndBuildCreateVnodeReq(pMnode, pDnode, pDb, pVgroup, &contLen, standby);
-  if (pReq == NULL) return -1;
-
-  action.pCont = pReq;
-  action.contLen = contLen;
-  action.msgType = TDMT_DND_CREATE_VNODE;
-  action.acceptableCode = TSDB_CODE_NODE_ALREADY_DEPLOYED;
-
-  if (mndTransAppendRedoAction(pTrans, &action) != 0) {
-    taosMemoryFree(pReq);
-    return -1;
-  }
-
-  return 0;
-}
-
-static int32_t mndAddAlterVnodeConfirmAction(SMnode *pMnode, STrans *pTrans, SDbObj *pDb, SVgObj *pVgroup) {
-  STransAction action = {0};
-  action.epSet = mndGetVgroupEpset(pMnode, pVgroup);
-
-  int32_t   contLen = sizeof(SMsgHead);
-  SMsgHead *pHead = taosMemoryMalloc(contLen);
-  if (pHead == NULL) {
-    terrno = TSDB_CODE_OUT_OF_MEMORY;
-    return -1;
-  }
-
-  pHead->contLen = htonl(contLen);
-  pHead->vgId = htonl(pVgroup->vgId);
-
-  action.pCont = pHead;
-  action.contLen = contLen;
-  action.msgType = TDMT_VND_ALTER_CONFIRM;
-
-  if (mndTransAppendRedoAction(pTrans, &action) != 0) {
-    taosMemoryFree(pHead);
-    return -1;
-  }
-
-  return 0;
-}
-
-static int32_t mndAddAlterVnodeAction(SMnode *pMnode, STrans *pTrans, SDbObj *pDb, SVgObj *pVgroup, tmsg_t msgType) {
-  STransAction action = {0};
-  action.epSet = mndGetVgroupEpset(pMnode, pVgroup);
-
-  int32_t contLen = 0;
-  void   *pReq = mndBuildAlterVnodeReq(pMnode, pDb, pVgroup, &contLen);
-  if (pReq == NULL) return -1;
-
-  action.pCont = pReq;
-  action.contLen = contLen;
-  action.msgType = msgType;
-
-  if (mndTransAppendRedoAction(pTrans, &action) != 0) {
-    taosMemoryFree(pReq);
-    return -1;
-  }
-
-  return 0;
-}
-
-static int32_t mndAddDropVnodeAction(SMnode *pMnode, STrans *pTrans, SDbObj *pDb, SVgObj *pVgroup, SVnodeGid *pVgid,
-                                     bool isRedo) {
-  STransAction action = {0};
-
-  SDnodeObj *pDnode = mndAcquireDnode(pMnode, pVgid->dnodeId);
-  if (pDnode == NULL) return -1;
-  action.epSet = mndGetDnodeEpset(pDnode);
-  mndReleaseDnode(pMnode, pDnode);
-
-  int32_t contLen = 0;
-  void   *pReq = mndBuildDropVnodeReq(pMnode, pDnode, pDb, pVgroup, &contLen);
-  if (pReq == NULL) return -1;
-
-  action.pCont = pReq;
-  action.contLen = contLen;
-  action.msgType = TDMT_DND_DROP_VNODE;
-  action.acceptableCode = TSDB_CODE_NODE_NOT_DEPLOYED;
-
-  if (isRedo) {
-    if (mndTransAppendRedoAction(pTrans, &action) != 0) {
-      taosMemoryFree(pReq);
-      return -1;
-    }
-  } else {
-    if (mndTransAppendUndoAction(pTrans, &action) != 0) {
-      taosMemoryFree(pReq);
-      return -1;
-    }
-  }
-
-  return 0;
-}
-
 static int32_t mndCheckDbName(const char *dbName, SUserObj *pUser) {
   char *pos = strstr(dbName, TS_PATH_DELIMITER);
   if (pos == NULL) {
@@ -795,7 +690,7 @@ static int32_t mndBuildAlterVgroupAction(SMnode *pMnode, STrans *pTrans, SDbObj
 static int32_t mndSetAlterDbRedoActions(SMnode *pMnode, STrans *pTrans, SDbObj *pOld, SDbObj *pNew) {
   SSdb   *pSdb = pMnode->pSdb;
   void   *pIter = NULL;
-  SArray *pArray = mndBuildDnodesArray(pMnode);
+  SArray *pArray = mndBuildDnodesArray(pMnode, 0);
 
   while (1) {
     SVgObj *pVgroup = NULL;
@@ -1742,3 +1637,4 @@ static void mndCancelGetNextDb(SMnode *pMnode, void *pIter) {
   SSdb *pSdb = pMnode->pSdb;
   sdbCancelFetch(pSdb, pIter);
 }
+
diff --git a/source/dnode/mnode/impl/src/mndDnode.c b/source/dnode/mnode/impl/src/mndDnode.c
index aeff018aa8..9bdee92113 100644
--- a/source/dnode/mnode/impl/src/mndDnode.c
+++ b/source/dnode/mnode/impl/src/mndDnode.c
@@ -254,7 +254,7 @@ int32_t mndGetDnodeSize(SMnode *pMnode) {
   return sdbGetSize(pSdb, SDB_DNODE);
 }
 
-bool mndIsDnodeOnline(SMnode *pMnode, SDnodeObj *pDnode, int64_t curMs) {
+bool mndIsDnodeOnline(SDnodeObj *pDnode, int64_t curMs) {
   int64_t interval = TABS(pDnode->lastAccessTime - curMs);
   if (interval > 5000 * tsStatusInterval) {
     if (pDnode->rebootTime > 0) {
@@ -393,7 +393,7 @@ static int32_t mndProcessStatusReq(SRpcMsg *pReq) {
 
   int64_t dnodeVer = sdbGetTableVer(pMnode->pSdb, SDB_DNODE) + sdbGetTableVer(pMnode->pSdb, SDB_MNODE);
   int64_t curMs = taosGetTimestampMs();
-  bool    online = mndIsDnodeOnline(pMnode, pDnode, curMs);
+  bool    online = mndIsDnodeOnline(pDnode, curMs);
   bool    dnodeChanged = (statusReq.dnodeVer != dnodeVer);
   bool    reboot = (pDnode->rebootTime != statusReq.rebootTime);
   bool    needCheck = !online || dnodeChanged || reboot;
@@ -559,30 +559,36 @@ CREATE_DNODE_OVER:
   return code;
 }
 
-static int32_t mndDropDnode(SMnode *pMnode, SRpcMsg *pReq, SDnodeObj *pDnode) {
-  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_ROLLBACK, TRN_CONFLICT_GLOBAL, pReq);
-  if (pTrans == NULL) {
-    mError("dnode:%d, failed to drop since %s", pDnode->id, terrstr());
-    return -1;
-  }
+static int32_t mndDropDnode(SMnode *pMnode, SRpcMsg *pReq, SDnodeObj *pDnode, SMnodeObj *pMObj) {
+  int32_t  code = -1;
+  SSdbRaw *pRaw = NULL;
+  STrans  *pTrans = NULL;
+
+  pTrans = mndTransCreate(pMnode, TRN_POLICY_RETRY, TRN_CONFLICT_GLOBAL, pReq);
+  if (pTrans == NULL) goto _OVER;
   mDebug("trans:%d, used to drop dnode:%d", pTrans->id, pDnode->id);
 
-  SSdbRaw *pCommitRaw = mndDnodeActionEncode(pDnode);
-  if (pCommitRaw == NULL || mndTransAppendCommitlog(pTrans, pCommitRaw) != 0) {
-    mError("trans:%d, failed to append commit log since %s", pTrans->id, terrstr());
-    mndTransDrop(pTrans);
-    return -1;
-  }
-  sdbSetRawStatus(pCommitRaw, SDB_STATUS_DROPPED);
+  pRaw = mndDnodeActionEncode(pDnode);
+  if (pRaw == NULL || mndTransAppendRedolog(pTrans, pRaw) != 0) goto _OVER;
+  sdbSetRawStatus(pRaw, SDB_STATUS_DROPPING);
+  pRaw = NULL;
 
-  if (mndTransPrepare(pMnode, pTrans) != 0) {
-    mError("trans:%d, failed to prepare since %s", pTrans->id, terrstr());
-    mndTransDrop(pTrans);
-    return -1;
-  }
+  pRaw = mndDnodeActionEncode(pDnode);
+  if (pRaw == NULL || mndTransAppendCommitlog(pTrans, pRaw) != 0) goto _OVER;
+  sdbSetRawStatus(pRaw, SDB_STATUS_DROPPED);
+  pRaw = NULL;
 
+  if (mndSetDropMnodeInfoToTrans(pMnode, pTrans, pMObj) != 0) goto _OVER;
+  if (mndSetMoveVgroupsInfoToTrans(pMnode, pTrans, pDnode->id) != 0) goto _OVER;
+  if (mndTransPrepare(pMnode, pTrans) != 0) goto _OVER;
+
+  mndTransSetSerial(pTrans);
+  code = 0;
+
+_OVER:
   mndTransDrop(pTrans);
-  return 0;
+  sdbFreeRaw(pRaw);
+  return code;
 }
 
 static int32_t mndProcessDropDnodeReq(SRpcMsg *pReq) {
@@ -595,42 +601,53 @@ static int32_t mndProcessDropDnodeReq(SRpcMsg *pReq) {
 
   if (tDeserializeSCreateDropMQSBNodeReq(pReq->pCont, pReq->contLen, &dropReq) != 0) {
     terrno = TSDB_CODE_INVALID_MSG;
-    goto DROP_DNODE_OVER;
+    goto _OVER;
   }
 
   mDebug("dnode:%d, start to drop", dropReq.dnodeId);
 
   if (dropReq.dnodeId <= 0) {
     terrno = TSDB_CODE_MND_INVALID_DNODE_ID;
-    goto DROP_DNODE_OVER;
+    goto _OVER;
   }
 
   pDnode = mndAcquireDnode(pMnode, dropReq.dnodeId);
   if (pDnode == NULL) {
     terrno = TSDB_CODE_MND_DNODE_NOT_EXIST;
-    goto DROP_DNODE_OVER;
+    goto _OVER;
+  }
+
+  if (!mndIsDnodeOnline(pDnode, taosGetTimestampMs())) {
+    terrno = TSDB_CODE_NODE_OFFLINE;
+    goto _OVER;
   }
 
   pMObj = mndAcquireMnode(pMnode, dropReq.dnodeId);
   if (pMObj != NULL) {
-    terrno = TSDB_CODE_MND_MNODE_NOT_EXIST;
-    goto DROP_DNODE_OVER;
+    if (sdbGetSize(pMnode->pSdb, SDB_MNODE) <= 1) {
+      terrno = TSDB_CODE_MND_TOO_FEW_MNODES;
+      goto _OVER;
+    }
+    if (pMnode->selfDnodeId == dropReq.dnodeId) {
+      terrno = TSDB_CODE_MND_CANT_DROP_MASTER;
+      goto _OVER;
+    }
   }
 
   pUser = mndAcquireUser(pMnode, pReq->conn.user);
   if (pUser == NULL) {
     terrno = TSDB_CODE_MND_NO_USER_FROM_CONN;
-    goto DROP_DNODE_OVER;
+    goto _OVER;
   }
 
   if (mndCheckNodeAuth(pUser)) {
-    goto DROP_DNODE_OVER;
+    goto _OVER;
   }
 
-  code = mndDropDnode(pMnode, pReq, pDnode);
+  code = mndDropDnode(pMnode, pReq, pDnode, pMObj);
   if (code == 0) code = TSDB_CODE_ACTION_IN_PROGRESS;
 
-DROP_DNODE_OVER:
+_OVER:
   if (code != 0 && code != TSDB_CODE_ACTION_IN_PROGRESS) {
     mError("dnode:%d, failed to drop since %s", dropReq.dnodeId, terrstr());
   }
@@ -638,7 +655,6 @@ DROP_DNODE_OVER:
   mndReleaseDnode(pMnode, pDnode);
   mndReleaseUser(pMnode, pUser);
   mndReleaseMnode(pMnode, pMObj);
-
   return code;
 }
 
@@ -736,7 +752,7 @@ static int32_t mndRetrieveDnodes(SRpcMsg *pReq, SShowObj *pShow, SSDataBlock *pB
   while (numOfRows < rows) {
     pShow->pIter = sdbFetch(pSdb, SDB_DNODE, pShow->pIter, (void **)&pDnode);
     if (pShow->pIter == NULL) break;
-    bool online = mndIsDnodeOnline(pMnode, pDnode, curMs);
+    bool online = mndIsDnodeOnline(pDnode, curMs);
 
     cols = 0;
 
diff --git a/source/dnode/mnode/impl/src/mndMain.c b/source/dnode/mnode/impl/src/mndMain.c
index d989ba6ec3..69300c0889 100644
--- a/source/dnode/mnode/impl/src/mndMain.c
+++ b/source/dnode/mnode/impl/src/mndMain.c
@@ -529,7 +529,7 @@ int32_t mndGetMonitorInfo(SMnode *pMnode, SMonClusterInfo *pClusterInfo, SMonVgr
     SMonDnodeDesc desc = {0};
     desc.dnode_id = pObj->id;
     tstrncpy(desc.dnode_ep, pObj->ep, sizeof(desc.dnode_ep));
-    if (mndIsDnodeOnline(pMnode, pObj, ms)) {
+    if (mndIsDnodeOnline(pObj, ms)) {
       tstrncpy(desc.status, "ready", sizeof(desc.status));
     } else {
       tstrncpy(desc.status, "offline", sizeof(desc.status));
diff --git a/source/dnode/mnode/impl/src/mndMnode.c b/source/dnode/mnode/impl/src/mndMnode.c
index 1cc832b3af..abd65c1724 100644
--- a/source/dnode/mnode/impl/src/mndMnode.c
+++ b/source/dnode/mnode/impl/src/mndMnode.c
@@ -408,7 +408,7 @@ static int32_t mndProcessCreateMnodeReq(SRpcMsg *pReq) {
     goto _OVER;
   }
 
-  if (!mndIsDnodeOnline(pMnode, pDnode, taosGetTimestampMs())) {
+  if (!mndIsDnodeOnline(pDnode, taosGetTimestampMs())) {
     terrno = TSDB_CODE_NODE_OFFLINE;
     goto _OVER;
   }
@@ -535,20 +535,26 @@ static int32_t mndSetDropMnodeRedoActions(SMnode *pMnode, STrans *pTrans, SDnode
   return 0;
 }
 
+int32_t mndSetDropMnodeInfoToTrans(SMnode *pMnode, STrans *pTrans, SMnodeObj *pObj) {
+  if (mndSetDropMnodeRedoLogs(pMnode, pTrans, pObj) != 0) return -1;
+  if (mndSetDropMnodeCommitLogs(pMnode, pTrans, pObj) != 0) return -1;
+  if (mndSetDropMnodeRedoActions(pMnode, pTrans, pObj->pDnode, pObj) != 0) return -1;
+  if (mndTransAppendNullLog(pTrans) != 0) return -1;
+  return 0;
+}
+
 static int32_t mndDropMnode(SMnode *pMnode, SRpcMsg *pReq, SMnodeObj *pObj) {
   int32_t code = -1;
+  STrans *pTrans = NULL;
 
-  STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_RETRY, TRN_CONFLICT_GLOBAL, pReq);
+  pTrans = mndTransCreate(pMnode, TRN_POLICY_RETRY, TRN_CONFLICT_GLOBAL, pReq);
   if (pTrans == NULL) goto _OVER;
 
   mDebug("trans:%d, used to drop mnode:%d", pTrans->id, pObj->id);
-  mndTransSetSerial(pTrans);
-  if (mndSetDropMnodeRedoLogs(pMnode, pTrans, pObj) != 0) goto _OVER;
-  if (mndSetDropMnodeCommitLogs(pMnode, pTrans, pObj) != 0) goto _OVER;
-  if (mndSetDropMnodeRedoActions(pMnode, pTrans, pObj->pDnode, pObj) != 0) goto _OVER;
-  if (mndTransAppendNullLog(pTrans) != 0) goto _OVER;
+  if (mndSetDropMnodeInfoToTrans(pMnode, pTrans, pObj) != 0) goto _OVER;
   if (mndTransPrepare(pMnode, pTrans) != 0) goto _OVER;
 
+  mndTransSetSerial(pTrans);
   code = 0;
 
 _OVER:
@@ -642,7 +648,7 @@ static int32_t mndRetrieveMnodes(SRpcMsg *pReq, SShowObj *pShow, SSDataBlock *pB
     if (pObj->id == pMnode->selfDnodeId) {
       roles = syncStr(TAOS_SYNC_STATE_LEADER);
     }
-    if (pObj->pDnode && mndIsDnodeOnline(pMnode, pObj->pDnode, curMs)) {
+    if (pObj->pDnode && mndIsDnodeOnline(pObj->pDnode, curMs)) {
       roles = syncStr(pObj->state);
     }
     char b2[12 + VARSTR_HEADER_SIZE] = {0};
diff --git a/source/dnode/mnode/impl/src/mndVgroup.c b/source/dnode/mnode/impl/src/mndVgroup.c
index 4f09eda733..afca8b14f0 100644
--- a/source/dnode/mnode/impl/src/mndVgroup.c
+++ b/source/dnode/mnode/impl/src/mndVgroup.c
@@ -344,9 +344,14 @@ static bool mndResetDnodesArrayFp(SMnode *pMnode, void *pObj, void *p1, void *p2
 static bool mndBuildDnodesArrayFp(SMnode *pMnode, void *pObj, void *p1, void *p2, void *p3) {
   SDnodeObj *pDnode = pObj;
   SArray    *pArray = p1;
+  int32_t    exceptDnodeId = *(int32_t *)p2;
+
+  if (exceptDnodeId == pDnode->id) {
+    return true;
+  }
 
   int64_t curMs = taosGetTimestampMs();
-  bool    online = mndIsDnodeOnline(pMnode, pDnode, curMs);
+  bool    online = mndIsDnodeOnline(pDnode, curMs);
   bool    isMnode = mndIsMnode(pMnode, pDnode->id);
   pDnode->numOfVnodes = mndGetVnodesNum(pMnode, pDnode->id);
 
@@ -363,7 +368,7 @@ static bool mndBuildDnodesArrayFp(SMnode *pMnode, void *pObj, void *p1, void *p2
   return true;
 }
 
-SArray *mndBuildDnodesArray(SMnode *pMnode) {
+SArray *mndBuildDnodesArray(SMnode *pMnode, int32_t exceptDnodeId) {
   SSdb   *pSdb = pMnode->pSdb;
   int32_t numOfDnodes = mndGetDnodeSize(pMnode);
 
@@ -374,7 +379,7 @@ SArray *mndBuildDnodesArray(SMnode *pMnode) {
   }
 
   sdbTraverse(pSdb, SDB_DNODE, mndResetDnodesArrayFp, NULL, NULL, NULL);
-  sdbTraverse(pSdb, SDB_DNODE, mndBuildDnodesArrayFp, pArray, NULL, NULL);
+  sdbTraverse(pSdb, SDB_DNODE, mndBuildDnodesArrayFp, pArray, &exceptDnodeId, NULL);
   return pArray;
 }
 
@@ -422,7 +427,7 @@ static int32_t mndGetAvailableDnode(SMnode *pMnode, SVgObj *pVgroup, SArray *pAr
 }
 
 int32_t mndAllocSmaVgroup(SMnode *pMnode, SDbObj *pDb, SVgObj *pVgroup) {
-  SArray *pArray = mndBuildDnodesArray(pMnode);
+  SArray *pArray = mndBuildDnodesArray(pMnode, 0);
   if (pArray == NULL) return -1;
 
   pVgroup->vgId = sdbGetMaxId(pMnode->pSdb, SDB_VGROUP);
@@ -451,7 +456,7 @@ int32_t mndAllocVgroup(SMnode *pMnode, SDbObj *pDb, SVgObj **ppVgroups) {
     goto _OVER;
   }
 
-  pArray = mndBuildDnodesArray(pMnode);
+  pArray = mndBuildDnodesArray(pMnode, 0);
   if (pArray == NULL) goto _OVER;
 
   mInfo("db:%s, total %d dnodes used to create %d vgroups (%d vnodes)", pDb->name, (int32_t)taosArrayGetSize(pArray),
@@ -501,86 +506,6 @@ _OVER:
   return code;
 }
 
-int32_t mndAddVnodeToVgroup(SMnode *pMnode, SVgObj *pVgroup, SArray *pArray) {
-  taosArraySort(pArray, (__compar_fn_t)mndCompareDnodeVnodes);
-  for (int32_t i = 0; i < taosArrayGetSize(pArray); ++i) {
-    SDnodeObj *pDnode = taosArrayGet(pArray, i);
-    mDebug("dnode:%d, equivalent vnodes:%d", pDnode->id, pDnode->numOfVnodes);
-  }
-
-  SVnodeGid *pVgid = &pVgroup->vnodeGid[pVgroup->replica];
-  for (int32_t d = 0; d < taosArrayGetSize(pArray); ++d) {
-    SDnodeObj *pDnode = taosArrayGet(pArray, d);
-
-    bool used = false;
-    for (int32_t vn = 0; vn < pVgroup->replica; ++vn) {
-      if (pDnode->id == pVgroup->vnodeGid[vn].dnodeId) {
-        used = true;
-        break;
-      }
-    }
-    if (used) continue;
-
-    if (pDnode == NULL || pDnode->numOfVnodes > pDnode->numOfSupportVnodes) {
-      terrno = TSDB_CODE_MND_NO_ENOUGH_DNODES;
-      return -1;
-    }
-
-    pVgid->dnodeId = pDnode->id;
-    pVgid->role = TAOS_SYNC_STATE_ERROR;
-    mInfo("db:%s, vgId:%d, vn:%d dnode:%d, is added", pVgroup->dbName, pVgroup->vgId, pVgroup->replica,
-          pVgid->dnodeId);
-
-    pVgroup->replica++;
-    pDnode->numOfVnodes++;
-    return 0;
-  }
-
-  terrno = TSDB_CODE_MND_NO_ENOUGH_DNODES;
-  mError("db:%s, failed to add vnode to vgId:%d since %s", pVgroup->dbName, pVgroup->vgId, terrstr());
-  return -1;
-}
-
-int32_t mndRemoveVnodeFromVgroup(SMnode *pMnode, SVgObj *pVgroup, SArray *pArray, SVnodeGid *pDelVgid) {
-  taosArraySort(pArray, (__compar_fn_t)mndCompareDnodeVnodes);
-  for (int32_t i = 0; i < taosArrayGetSize(pArray); ++i) {
-    SDnodeObj *pDnode = taosArrayGet(pArray, i);
-    mDebug("dnode:%d, equivalent vnodes:%d", pDnode->id, pDnode->numOfVnodes);
-  }
-
-  int32_t code = -1;
-  for (int32_t d = taosArrayGetSize(pArray) - 1; d >= 0; --d) {
-    SDnodeObj *pDnode = taosArrayGet(pArray, d);
-
-    for (int32_t vn = 0; vn < pVgroup->replica; ++vn) {
-      SVnodeGid *pVgid = &pVgroup->vnodeGid[vn];
-      if (pVgid->dnodeId == pDnode->id) {
-        mInfo("db:%s, vgId:%d, vn:%d dnode:%d, is removed", pVgroup->dbName, pVgroup->vgId, vn, pVgid->dnodeId);
-        pDnode->numOfVnodes--;
-        pVgroup->replica--;
-        *pDelVgid = *pVgid;
-        *pVgid = pVgroup->vnodeGid[pVgroup->replica];
-        memset(&pVgroup->vnodeGid[pVgroup->replica], 0, sizeof(SVnodeGid));
-        code = 0;
-        goto _OVER;
-      }
-    }
-  }
-
-_OVER:
-  if (code != 0) {
-    terrno = TSDB_CODE_APP_ERROR;
-    mError("db:%s, failed to remove vnode from vgId:%d since %s", pVgroup->dbName, pVgroup->vgId, terrstr());
-    return -1;
-  }
-
-  for (int32_t vn = 0; vn < pVgroup->replica; ++vn) {
-    SVnodeGid *pVgid = &pVgroup->vnodeGid[vn];
-    mInfo("db:%s, vgId:%d, vn:%d dnode:%d is reserved", pVgroup->dbName, pVgroup->vgId, vn, pVgid->dnodeId);
-  }
-  return 0;
-}
-
 SEpSet mndGetVgroupEpset(SMnode *pMnode, const SVgObj *pVgroup) {
   SEpSet epset = {0};
 
@@ -678,7 +603,7 @@ static int32_t mndRetrieveVgroups(SRpcMsg *pReq, SShowObj *pShow, SSDataBlock *p
         bool       online = false;
         SDnodeObj *pDnode = mndAcquireDnode(pMnode, pVgroup->vnodeGid[i].dnodeId);
         if (pDnode != NULL) {
-          online = mndIsDnodeOnline(pMnode, pDnode, curMs);
+          online = mndIsDnodeOnline(pDnode, curMs);
           mndReleaseDnode(pMnode, pDnode);
         }
 
@@ -797,3 +722,251 @@ static void mndCancelGetNextVnode(SMnode *pMnode, void *pIter) {
   SSdb *pSdb = pMnode->pSdb;
   sdbCancelFetch(pSdb, pIter);
 }
+
+int32_t mndAddVnodeToVgroup(SMnode *pMnode, SVgObj *pVgroup, SArray *pArray) {
+  taosArraySort(pArray, (__compar_fn_t)mndCompareDnodeVnodes);
+  for (int32_t i = 0; i < taosArrayGetSize(pArray); ++i) {
+    SDnodeObj *pDnode = taosArrayGet(pArray, i);
+    mDebug("dnode:%d, equivalent vnodes:%d", pDnode->id, pDnode->numOfVnodes);
+  }
+
+  SVnodeGid *pVgid = &pVgroup->vnodeGid[pVgroup->replica];
+  for (int32_t d = 0; d < taosArrayGetSize(pArray); ++d) {
+    SDnodeObj *pDnode = taosArrayGet(pArray, d);
+
+    bool used = false;
+    for (int32_t vn = 0; vn < pVgroup->replica; ++vn) {
+      if (pDnode->id == pVgroup->vnodeGid[vn].dnodeId) {
+        used = true;
+        break;
+      }
+    }
+    if (used) continue;
+
+    if (pDnode == NULL || pDnode->numOfVnodes > pDnode->numOfSupportVnodes) {
+      terrno = TSDB_CODE_MND_NO_ENOUGH_DNODES;
+      return -1;
+    }
+
+    pVgid->dnodeId = pDnode->id;
+    pVgid->role = TAOS_SYNC_STATE_ERROR;
+    mInfo("db:%s, vgId:%d, vn:%d dnode:%d, is added", pVgroup->dbName, pVgroup->vgId, pVgroup->replica, pVgid->dnodeId);
+
+    pVgroup->replica++;
+    pDnode->numOfVnodes++;
+    return 0;
+  }
+
+  terrno = TSDB_CODE_MND_NO_ENOUGH_DNODES;
+  mError("db:%s, failed to add vnode to vgId:%d since %s", pVgroup->dbName, pVgroup->vgId, terrstr());
+  return -1;
+}
+
+int32_t mndRemoveVnodeFromVgroup(SMnode *pMnode, SVgObj *pVgroup, SArray *pArray, SVnodeGid *pDelVgid) {
+  taosArraySort(pArray, (__compar_fn_t)mndCompareDnodeVnodes);
+  for (int32_t i = 0; i < taosArrayGetSize(pArray); ++i) {
+    SDnodeObj *pDnode = taosArrayGet(pArray, i);
+    mDebug("dnode:%d, equivalent vnodes:%d", pDnode->id, pDnode->numOfVnodes);
+  }
+
+  int32_t code = -1;
+  for (int32_t d = taosArrayGetSize(pArray) - 1; d >= 0; --d) {
+    SDnodeObj *pDnode = taosArrayGet(pArray, d);
+
+    for (int32_t vn = 0; vn < pVgroup->replica; ++vn) {
+      SVnodeGid *pVgid = &pVgroup->vnodeGid[vn];
+      if (pVgid->dnodeId == pDnode->id) {
+        mInfo("db:%s, vgId:%d, vn:%d dnode:%d, is removed", pVgroup->dbName, pVgroup->vgId, vn, pVgid->dnodeId);
+        pDnode->numOfVnodes--;
+        pVgroup->replica--;
+        *pDelVgid = *pVgid;
+        *pVgid = pVgroup->vnodeGid[pVgroup->replica];
+        memset(&pVgroup->vnodeGid[pVgroup->replica], 0, sizeof(SVnodeGid));
+        code = 0;
+        goto _OVER;
+      }
+    }
+  }
+
+_OVER:
+  if (code != 0) {
+    terrno = TSDB_CODE_APP_ERROR;
+    mError("db:%s, failed to remove vnode from vgId:%d since %s", pVgroup->dbName, pVgroup->vgId, terrstr());
+    return -1;
+  }
+
+  for (int32_t vn = 0; vn < pVgroup->replica; ++vn) {
+    SVnodeGid *pVgid = &pVgroup->vnodeGid[vn];
+    mInfo("db:%s, vgId:%d, vn:%d dnode:%d is reserved", pVgroup->dbName, pVgroup->vgId, vn, pVgid->dnodeId);
+  }
+  return 0;
+}
+
+int32_t mndAddCreateVnodeAction(SMnode *pMnode, STrans *pTrans, SDbObj *pDb, SVgObj *pVgroup, SVnodeGid *pVgid,
+                                bool standby) {
+  STransAction action = {0};
+
+  SDnodeObj *pDnode = mndAcquireDnode(pMnode, pVgid->dnodeId);
+  if (pDnode == NULL) return -1;
+  action.epSet = mndGetDnodeEpset(pDnode);
+  mndReleaseDnode(pMnode, pDnode);
+
+  int32_t contLen = 0;
+  void   *pReq = mndBuildCreateVnodeReq(pMnode, pDnode, pDb, pVgroup, &contLen, standby);
+  if (pReq == NULL) return -1;
+
+  action.pCont = pReq;
+  action.contLen = contLen;
+  action.msgType = TDMT_DND_CREATE_VNODE;
+  action.acceptableCode = TSDB_CODE_NODE_ALREADY_DEPLOYED;
+
+  if (mndTransAppendRedoAction(pTrans, &action) != 0) {
+    taosMemoryFree(pReq);
+    return -1;
+  }
+
+  return 0;
+}
+
+int32_t mndAddAlterVnodeConfirmAction(SMnode *pMnode, STrans *pTrans, SDbObj *pDb, SVgObj *pVgroup) {
+  STransAction action = {0};
+  action.epSet = mndGetVgroupEpset(pMnode, pVgroup);
+
+  int32_t   contLen = sizeof(SMsgHead);
+  SMsgHead *pHead = taosMemoryMalloc(contLen);
+  if (pHead == NULL) {
+    terrno = TSDB_CODE_OUT_OF_MEMORY;
+    return -1;
+  }
+
+  pHead->contLen = htonl(contLen);
+  pHead->vgId = htonl(pVgroup->vgId);
+
+  action.pCont = pHead;
+  action.contLen = contLen;
+  action.msgType = TDMT_VND_ALTER_CONFIRM;
+
+  if (mndTransAppendRedoAction(pTrans, &action) != 0) {
+    taosMemoryFree(pHead);
+    return -1;
+  }
+
+  return 0;
+}
+
+int32_t mndAddAlterVnodeAction(SMnode *pMnode, STrans *pTrans, SDbObj *pDb, SVgObj *pVgroup, tmsg_t msgType) {
+  STransAction action = {0};
+  action.epSet = mndGetVgroupEpset(pMnode, pVgroup);
+
+  int32_t contLen = 0;
+  void   *pReq = mndBuildAlterVnodeReq(pMnode, pDb, pVgroup, &contLen);
+  if (pReq == NULL) return -1;
+
+  action.pCont = pReq;
+  action.contLen = contLen;
+  action.msgType = msgType;
+
+  if (mndTransAppendRedoAction(pTrans, &action) != 0) {
+    taosMemoryFree(pReq);
+    return -1;
+  }
+
+  return 0;
+}
+
+int32_t mndAddDropVnodeAction(SMnode *pMnode, STrans *pTrans, SDbObj *pDb, SVgObj *pVgroup, SVnodeGid *pVgid,
+                              bool isRedo) {
+  STransAction action = {0};
+
+  SDnodeObj *pDnode = mndAcquireDnode(pMnode, pVgid->dnodeId);
+  if (pDnode == NULL) return -1;
+  action.epSet = mndGetDnodeEpset(pDnode);
+  mndReleaseDnode(pMnode, pDnode);
+
+  int32_t contLen = 0;
+  void   *pReq = mndBuildDropVnodeReq(pMnode, pDnode, pDb, pVgroup, &contLen);
+  if (pReq == NULL) return -1;
+
+  action.pCont = pReq;
+  action.contLen = contLen;
+  action.msgType = TDMT_DND_DROP_VNODE;
+  action.acceptableCode = TSDB_CODE_NODE_NOT_DEPLOYED;
+
+  if (isRedo) {
+    if (mndTransAppendRedoAction(pTrans, &action) != 0) {
+      taosMemoryFree(pReq);
+      return -1;
+    }
+  } else {
+    if (mndTransAppendUndoAction(pTrans, &action) != 0) {
+      taosMemoryFree(pReq);
+      return -1;
+    }
+  }
+
+  return 0;
+}
+
+int32_t mndSetMoveVgroupInfoToTrans(SMnode *pMnode, STrans *pTrans, SDbObj *pDb, SVgObj *pVgroup, int32_t vnIndex,
+                                    SArray *pArray) {
+  SVgObj newVgroup = {0};
+  memcpy(&newVgroup, pVgroup, sizeof(SVgObj));
+
+  mInfo("vgId:%d, vgroup info before move, replica:%d", newVgroup.vgId, newVgroup.replica);
+  for (int32_t i = 0; i < newVgroup.replica; ++i) {
+    mInfo("vgId:%d, vnode:%d dnode:%d", newVgroup.vgId, i, newVgroup.vnodeGid[i].dnodeId);
+  }
+
+  mInfo("vgId:%d, will add 1 vnodes", pVgroup->vgId);
+  if (mndAddVnodeToVgroup(pMnode, &newVgroup, pArray) != 0) return -1;
+  if (mndAddCreateVnodeAction(pMnode, pTrans, pDb, &newVgroup, &newVgroup.vnodeGid[1], true) != 0) return -1;
+  if (mndAddAlterVnodeAction(pMnode, pTrans, pDb, &newVgroup, TDMT_VND_ALTER_REPLICA) != 0) return -1;
+  if (mndAddAlterVnodeConfirmAction(pMnode, pTrans, pDb, &newVgroup) != 0) return -1;
+
+  mInfo("vgId:%d, will remove 1 vnodes", pVgroup->vgId);
+  newVgroup.replica--;
+  SVnodeGid del = newVgroup.vnodeGid[vnIndex];
+  newVgroup.vnodeGid[vnIndex] = newVgroup.vnodeGid[newVgroup.replica];
+  memset(&newVgroup.vnodeGid[newVgroup.replica], 0, sizeof(SVnodeGid));
+  if (mndAddAlterVnodeAction(pMnode, pTrans, pDb, &newVgroup, TDMT_VND_ALTER_REPLICA) != 0) return -1;
+  if (mndAddDropVnodeAction(pMnode, pTrans, pDb, &newVgroup, &del, true) != 0) return -1;
+  if (mndAddAlterVnodeConfirmAction(pMnode, pTrans, pDb, &newVgroup) != 0) return -1;
+
+  mInfo("vgId:%d, vgroup info after move, replica:%d", newVgroup.vgId, newVgroup.replica);
+  for (int32_t i = 0; i < newVgroup.replica; ++i) {
+    mInfo("vgId:%d, vnode:%d dnode:%d", newVgroup.vgId, i, newVgroup.vnodeGid[i].dnodeId);
+  }
+  return 0;
+}
+
+int32_t mndSetMoveVgroupsInfoToTrans(SMnode *pMnode, STrans *pTrans, int32_t dropDnodeId) {
+  SArray *pArray = mndBuildDnodesArray(pMnode, dropDnodeId);
+  if (pArray == NULL) return -1;
+
+  void *pIter = NULL;
+  while (1) {
+    SVgObj *pVgroup = NULL;
+    pIter = sdbFetch(pMnode->pSdb, SDB_VGROUP, pIter, (void **)&pVgroup);
+    if (pIter == NULL) break;
+
+    int32_t vnIndex = -1;
+    for (int32_t i = 0; i < pVgroup->replica; ++i) {
+      if (pVgroup->vnodeGid[i].dnodeId == dropDnodeId) {
+        vnIndex = i;
+        break;
+      }
+    }
+
+    if (vnIndex != -1) {
+      mInfo("vgId:%d, vnode:%d will be removed from dnode:%d", pVgroup->vgId, vnIndex, dropDnodeId);
+      SDbObj *pDb = mndAcquireDb(pMnode, pVgroup->dbName);
+      mndSetMoveVgroupInfoToTrans(pMnode, pTrans, pDb, pVgroup, vnIndex, pArray);
+      mndReleaseDb(pMnode, pDb);
+    }
+
+    sdbRelease(pMnode->pSdb, pVgroup);
+  }
+
+  taosArrayDestroy(pArray);
+  return 0;
+}
\ No newline at end of file
diff --git a/source/dnode/mnode/sdb/src/sdbRaw.c b/source/dnode/mnode/sdb/src/sdbRaw.c
index 90643a54a9..7720a8e88a 100644
--- a/source/dnode/mnode/sdb/src/sdbRaw.c
+++ b/source/dnode/mnode/sdb/src/sdbRaw.c
@@ -42,8 +42,10 @@ SSdbRaw *sdbAllocRaw(ESdbType type, int8_t sver, int32_t dataLen) {
 }
 
 void sdbFreeRaw(SSdbRaw *pRaw) {
-  mTrace("raw:%p, is freed", pRaw);
-  taosMemoryFree(pRaw);
+  if (pRaw != NULL) {
+    mTrace("raw:%p, is freed", pRaw);
+    taosMemoryFree(pRaw);
+  }
 }
 
 int32_t sdbSetRawInt8(SSdbRaw *pRaw, int32_t dataPos, int8_t val) {

From 187750ec335d7a844bd81a63b22267c03a8c7b49 Mon Sep 17 00:00:00 2001
From: plum-lihui <huili@taosdata.com>
Date: Mon, 6 Jun 2022 14:53:46 +0800
Subject: [PATCH 13/34] test: modify test case

---
 tests/system-test/7-tmq/schema.py | 20 ++++++++++----------
 1 file changed, 10 insertions(+), 10 deletions(-)

diff --git a/tests/system-test/7-tmq/schema.py b/tests/system-test/7-tmq/schema.py
index 633a097db6..f837e334f6 100644
--- a/tests/system-test/7-tmq/schema.py
+++ b/tests/system-test/7-tmq/schema.py
@@ -29,8 +29,8 @@ class TDTestCase:
 
     def init(self, conn, logSql):
         tdLog.debug(f"start to excute {__file__}")
-        tdSql.init(conn.cursor())
-        #tdSql.init(conn.cursor(), logSql)  # output sql.txt file
+        #tdSql.init(conn.cursor())
+        tdSql.init(conn.cursor(), logSql)  # output sql.txt file
 
     def getBuildPath(self):
         selfPath = os.path.dirname(os.path.realpath(__file__))
@@ -282,7 +282,7 @@ class TDTestCase:
         tdSql.execute("create topic %s as select ts, c1, c2, t1, t2 from %s.%s" %(columnTopicFromStb, parameterDict['dbName'], parameterDict['stbName']))
         tdSql.execute("create topic %s as select ts, c1, c2 from %s.%s" %(columnTopicFromNtb, parameterDict['dbName'], ntbName))
         
-        tsLog.info("======== super table test:")
+        tdLog.info("======== super table test:")
         # alter actions prohibited: drop column/tag, modify column/tag type, rename column/tag included in topic
         tdSql.error("alter table %s.%s drop column c1"%(parameterDict['dbName'], parameterDict['stbName']))
         tdSql.error("alter table %s.%s drop column c2"%(parameterDict['dbName'], parameterDict['stbName']))
@@ -316,7 +316,7 @@ class TDTestCase:
         tdSql.query("alter table %s.%s add tag t3 int"%(parameterDict['dbName'], parameterDict['stbName']))
         tdSql.query("alter table %s.%s add tag t4 float"%(parameterDict['dbName'], parameterDict['stbName']))
 
-        tsLog.info("======== normal table test:")
+        tdLog.info("======== normal table test:")
         # alter actions prohibited: drop column/tag, modify column/tag type, rename column/tag included in topic
         tdSql.error("alter table %s.%s drop column c1"%(parameterDict['dbName'], ntbName))
         tdSql.error("alter table %s.%s drop column c2"%(parameterDict['dbName'], ntbName))
@@ -419,7 +419,7 @@ class TDTestCase:
         tdSql.execute("create topic %s as select ts, c1, c2, t1, t2 from %s.%s where c3 > 3 and c4 like 'abc' and t3 = 5 and t4 = 'beijing'" %(columnTopicFromStb, parameterDict['dbName'], parameterDict['stbName']))
         tdSql.execute("create topic %s as select ts, c1, c2 from %s.%s where c3 > 3 and c4 like 'abc'" %(columnTopicFromNtb, parameterDict['dbName'], ntbName))
         
-        tsLog.info("======== super table test:")
+        tdLog.info("======== super table test:")
         # alter actions prohibited: drop column/tag, modify column/tag type, rename column/tag included in topic
         tdSql.error("alter table %s.%s drop column c1"%(parameterDict['dbName'], parameterDict['stbName']))
         tdSql.error("alter table %s.%s drop column c2"%(parameterDict['dbName'], parameterDict['stbName']))
@@ -457,7 +457,7 @@ class TDTestCase:
         tdSql.query("alter table %s.%s add column c5 int"%(parameterDict['dbName'], parameterDict['stbName']))
         tdSql.query("alter table %s.%s add tag t5 float"%(parameterDict['dbName'], parameterDict['stbName']))
 
-        tsLog.info("======== normal table test:")
+        tdLog.info("======== normal table test:")
         # alter actions prohibited: drop column/tag, modify column/tag type, rename column/tag included in topic
         tdSql.error("alter table %s.%s drop column c1"%(parameterDict['dbName'], ntbName))
         tdSql.error("alter table %s.%s drop column c2"%(parameterDict['dbName'], ntbName))
@@ -566,7 +566,7 @@ class TDTestCase:
         tdSql.execute("create topic %s as select * from %s.%s" %(columnTopicFromStb, parameterDict['dbName'], parameterDict['stbName']))
         tdSql.execute("create topic %s as select * from %s.%s " %(columnTopicFromNtb, parameterDict['dbName'], ntbName))
         
-        tsLog.info("======== super table test:")
+        tdLog.info("======== super table test:")
         # alter actions prohibited: drop column/tag, modify column/tag type, rename column/tag included in topic
         tdSql.error("alter table %s.%s drop column c1"%(parameterDict['dbName'], parameterDict['stbName']))
         tdSql.error("alter table %s.%s drop column c2"%(parameterDict['dbName'], parameterDict['stbName']))
@@ -601,7 +601,7 @@ class TDTestCase:
         tdSql.query("alter table %s.%s add column c6 int"%(parameterDict['dbName'], parameterDict['stbName']))
         tdSql.query("alter table %s.%s add tag t6 float"%(parameterDict['dbName'], parameterDict['stbName']))
 
-        tsLog.info("======== normal table test:")
+        tdLog.info("======== normal table test:")
         # alter actions prohibited: drop column/tag, modify column/tag type, rename column/tag included in topic
         tdSql.error("alter table %s.%s drop column c1"%(parameterDict['dbName'], ntbName))
         tdSql.error("alter table %s.%s drop column c2"%(parameterDict['dbName'], ntbName))
@@ -687,8 +687,8 @@ class TDTestCase:
         tdLog.info("cfgPath: %s" % cfgPath)
 
         self.tmqCase1(cfgPath, buildPath)
-        self.tmqCase2(cfgPath, buildPath) 
-        self.tmqCase3(cfgPath, buildPath)
+        # self.tmqCase2(cfgPath, buildPath) 
+        # self.tmqCase3(cfgPath, buildPath)
 
     def stop(self):
         tdSql.close()

From 6e6a492ab3b5f1dfbf1d5278c808a537d9882dfb Mon Sep 17 00:00:00 2001
From: plum-lihui <huili@taosdata.com>
Date: Mon, 6 Jun 2022 15:00:45 +0800
Subject: [PATCH 14/34] test: modify test case

---
 tests/system-test/7-tmq/schema.py | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/tests/system-test/7-tmq/schema.py b/tests/system-test/7-tmq/schema.py
index f837e334f6..196b9a5323 100644
--- a/tests/system-test/7-tmq/schema.py
+++ b/tests/system-test/7-tmq/schema.py
@@ -29,8 +29,8 @@ class TDTestCase:
 
     def init(self, conn, logSql):
         tdLog.debug(f"start to excute {__file__}")
-        #tdSql.init(conn.cursor())
-        tdSql.init(conn.cursor(), logSql)  # output sql.txt file
+        tdSql.init(conn.cursor())
+        #tdSql.init(conn.cursor(), logSql)  # output sql.txt file
 
     def getBuildPath(self):
         selfPath = os.path.dirname(os.path.realpath(__file__))

From 62caafa35687466d0554a630ff283eb55f1f30d3 Mon Sep 17 00:00:00 2001
From: "wenzhouwww@live.cn" <wenzhouwww@live.cn>
Date: Mon, 6 Jun 2022 15:08:43 +0800
Subject: [PATCH 15/34] update case for hyperlog function

---
 tests/system-test/2-query/hyperloglog.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/tests/system-test/2-query/hyperloglog.py b/tests/system-test/2-query/hyperloglog.py
index 35703e441d..b20ec35f07 100644
--- a/tests/system-test/2-query/hyperloglog.py
+++ b/tests/system-test/2-query/hyperloglog.py
@@ -223,7 +223,7 @@ class TDTestCase:
         tdLog.printNoPrefix("===step 0: err case, must return err")
         tdSql.error( "select hyperloglog() from ct1" )
         tdSql.error( "select hyperloglog(c1, c2) from ct2" )
-        tdSql.error( "select hyperloglog(1) from ct2" )
+        # tdSql.error( "select hyperloglog(1) from ct2" )
         tdSql.error( f"select hyperloglog({NUM_COL[0]}, {NUM_COL[1]}) from ct4" )
         tdSql.error( ''' select hyperloglog(['c1 + c1', 'c1 + c2', 'c1 + c3', 'c1 + c4', 'c1 + c5', 'c1 + c6', 'c1 + c7', 'c1 + c8', 'c1 + c9', 'c1 + c10'])
                     from ct1

From c438b699c937b49e60989516edc429063a3158b7 Mon Sep 17 00:00:00 2001
From: shenglian zhou <slzhou@taosdata.com>
Date: Mon, 6 Jun 2022 15:12:07 +0800
Subject: [PATCH 16/34] feat: support multiway sort merge

---
 source/libs/executor/inc/executorimpl.h |  3 ++-
 source/libs/executor/src/executorimpl.c |  6 ++++-
 source/libs/executor/src/sortoperator.c | 32 +++++++------------------
 3 files changed, 16 insertions(+), 25 deletions(-)

diff --git a/source/libs/executor/inc/executorimpl.h b/source/libs/executor/inc/executorimpl.h
index bd33d7f3c7..17a06a40c1 100644
--- a/source/libs/executor/inc/executorimpl.h
+++ b/source/libs/executor/inc/executorimpl.h
@@ -774,7 +774,8 @@ SOperatorInfo* createProjectOperatorInfo(SOperatorInfo* downstream, SExprInfo* p
 SOperatorInfo *createSortOperatorInfo(SOperatorInfo* downstream, SSDataBlock* pResBlock, SArray* pSortInfo, SExprInfo* pExprInfo, int32_t numOfCols,
                                       SArray* pIndexMap, SExecTaskInfo* pTaskInfo);
 SOperatorInfo* createMultiwaySortMergeOperatorInfo(SOperatorInfo** downStreams, int32_t numStreams,
-                                                   SSDataBlock* pResBlock, SArray* pSortInfo, SExecTaskInfo* pTaskInfo);
+                                                   SSDataBlock* pResBlock, SArray* pSortInfo, SArray* pColMatchColInfo,
+                                                   SExecTaskInfo* pTaskInfo);
 SOperatorInfo* createSortedMergeOperatorInfo(SOperatorInfo** downstream, int32_t numOfDownstream, SExprInfo* pExprInfo, int32_t num, SArray* pSortInfo, SArray* pGroupInfo, SExecTaskInfo* pTaskInfo);
 
 SOperatorInfo* createIntervalOperatorInfo(SOperatorInfo* downstream, SExprInfo* pExprInfo, int32_t numOfCols,
diff --git a/source/libs/executor/src/executorimpl.c b/source/libs/executor/src/executorimpl.c
index d094dc0203..beb5e1152f 100644
--- a/source/libs/executor/src/executorimpl.c
+++ b/source/libs/executor/src/executorimpl.c
@@ -4524,8 +4524,12 @@ SOperatorInfo* createOperatorTree(SPhysiNode* pPhyNode, SExecTaskInfo* pTaskInfo
     SSDataBlock* pResBlock = createResDataBlock(pDescNode);
 
     SArray*      sortInfo = createSortInfo(pMergePhyNode->pMergeKeys);
+    int32_t numOfOutputCols = 0;
+    SArray* pColList = NULL;
+        //extractColMatchInfo(pMergePhyNode->pTargets, pDescNode, &numOfOutputCols, pTaskInfo, COL_MATCH_FROM_SLOT_ID);
 
-    pOptr = createMultiwaySortMergeOperatorInfo(ops, size, pResBlock, sortInfo, pTaskInfo);
+
+    pOptr = createMultiwaySortMergeOperatorInfo(ops, size, pResBlock, sortInfo, pColList, pTaskInfo);
   } else if (QUERY_NODE_PHYSICAL_PLAN_SESSION_WINDOW == type) {
     SSessionWinodwPhysiNode* pSessionNode = (SSessionWinodwPhysiNode*)pPhyNode;
 
diff --git a/source/libs/executor/src/sortoperator.c b/source/libs/executor/src/sortoperator.c
index 0618c80682..c219acb687 100644
--- a/source/libs/executor/src/sortoperator.c
+++ b/source/libs/executor/src/sortoperator.c
@@ -224,30 +224,11 @@ typedef struct SMultiwaySortMergeOperatorInfo {
 
   SArray*      pSortInfo;
   SSortHandle* pSortHandle;
+  SArray*      pColMatchInfo;  // for index map from table scan output
 
   int64_t      startTs;       // sort start time
 } SMultiwaySortMergeOperatorInfo;
 
-SSDataBlock* getMultiwaySortMergedBlockData(SSortHandle* pHandle, SSDataBlock* pDataBlock, int32_t capacity) {
-  blockDataCleanup(pDataBlock);
-
-  blockDataEnsureCapacity(pDataBlock, capacity);
-
-  while (1) {
-    STupleHandle* pTupleHandle = tsortNextTuple(pHandle);
-    if (pTupleHandle == NULL) {
-      break;
-    }
-
-    appendOneRowToDataBlock(pDataBlock, pTupleHandle);
-    if (pDataBlock->info.rows >= capacity) {
-      return pDataBlock;
-    }
-  }
-
-  return (pDataBlock->info.rows > 0) ? pDataBlock : NULL;
-}
-
 
 int32_t doOpenMultiwaySortMergeOperator(SOperatorInfo* pOperator) {
   SMultiwaySortMergeOperatorInfo * pInfo = pOperator->info;
@@ -261,7 +242,7 @@ int32_t doOpenMultiwaySortMergeOperator(SOperatorInfo* pOperator) {
 
   int32_t numOfBufPage = pInfo->sortBufSize / pInfo->bufPageSize;
 
-  pInfo->pSortHandle = tsortCreateSortHandle(pInfo->pSortInfo, NULL, SORT_MULTISOURCE_MERGE,
+  pInfo->pSortHandle = tsortCreateSortHandle(pInfo->pSortInfo, pInfo->pColMatchInfo, SORT_MULTISOURCE_MERGE,
                                              pInfo->bufPageSize, numOfBufPage, NULL, pTaskInfo->id.str);
 
   tsortSetFetchRawDataFp(pInfo->pSortHandle, loadNextDataBlock, NULL, NULL);
@@ -298,7 +279,10 @@ SSDataBlock* doMultiwaySortMerge(SOperatorInfo* pOperator) {
     longjmp(pTaskInfo->env, code);
   }
 
-  SSDataBlock* pBlock = getMultiwaySortMergedBlockData(pInfo->pSortHandle, pInfo->binfo.pRes, pOperator->resultInfo.capacity);
+  SSDataBlock* pBlock = getSortedBlockData(pInfo->pSortHandle,
+                                           pInfo->binfo.pRes,
+                                           pOperator->resultInfo.capacity,
+                                           pInfo->pColMatchInfo);
 
   if (pBlock != NULL) {
     pOperator->resultInfo.totalRows += pBlock->info.rows;
@@ -328,7 +312,8 @@ int32_t getMultiwaySortMergeExplainExecInfo(SOperatorInfo* pOptr, void** pOptrEx
 }
 
 SOperatorInfo* createMultiwaySortMergeOperatorInfo(SOperatorInfo** downStreams, int32_t numStreams,
-                                                   SSDataBlock* pResBlock, SArray* pSortInfo, SExecTaskInfo* pTaskInfo) {
+                                                   SSDataBlock* pResBlock, SArray* pSortInfo, SArray* pColMatchColInfo,
+                                                   SExecTaskInfo* pTaskInfo) {
   SMultiwaySortMergeOperatorInfo* pInfo = taosMemoryCalloc(1, sizeof(SMultiwaySortMergeOperatorInfo));
   SOperatorInfo*     pOperator = taosMemoryCalloc(1, sizeof(SOperatorInfo));
   int32_t            rowSize = pResBlock->info.rowSize;
@@ -342,6 +327,7 @@ SOperatorInfo* createMultiwaySortMergeOperatorInfo(SOperatorInfo** downStreams,
   initResultSizeInfo(pOperator, 1024);
 
   pInfo->pSortInfo    = pSortInfo;
+  pInfo->pColMatchInfo= pColMatchColInfo;
   pOperator->name     = "MultiwaySortMerge";
   pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_MERGE;
   pOperator->blocking = true;

From 72c71141b2bfeabe07fc7bab64a4d2ca13234dfc Mon Sep 17 00:00:00 2001
From: shenglian zhou <slzhou@taosdata.com>
Date: Mon, 6 Jun 2022 15:16:20 +0800
Subject: [PATCH 17/34] feat: support multiway sort merge

---
 source/libs/executor/src/sortoperator.c | 111 ++++++++++++------------
 1 file changed, 55 insertions(+), 56 deletions(-)

diff --git a/source/libs/executor/src/sortoperator.c b/source/libs/executor/src/sortoperator.c
index c219acb687..c84d4491af 100644
--- a/source/libs/executor/src/sortoperator.c
+++ b/source/libs/executor/src/sortoperator.c
@@ -13,17 +13,18 @@
  * along with this program. If not, see <http://www.gnu.org/licenses/>.
  */
 
-#include "tdatablock.h"
 #include "executorimpl.h"
+#include "tdatablock.h"
 
 static SSDataBlock* doSort(SOperatorInfo* pOperator);
-static int32_t doOpenSortOperator(SOperatorInfo* pOperator);
-static int32_t getExplainExecInfo(SOperatorInfo* pOptr, void** pOptrExplain, uint32_t* len);
+static int32_t      doOpenSortOperator(SOperatorInfo* pOperator);
+static int32_t      getExplainExecInfo(SOperatorInfo* pOptr, void** pOptrExplain, uint32_t* len);
 
 static void destroyOrderOperatorInfo(void* param, int32_t numOfOutput);
 
-SOperatorInfo* createSortOperatorInfo(SOperatorInfo* downstream, SSDataBlock* pResBlock, SArray* pSortInfo, SExprInfo* pExprInfo, int32_t numOfCols,
-                                      SArray* pColMatchColInfo, SExecTaskInfo* pTaskInfo) {
+SOperatorInfo* createSortOperatorInfo(SOperatorInfo* downstream, SSDataBlock* pResBlock, SArray* pSortInfo,
+                                      SExprInfo* pExprInfo, int32_t numOfCols, SArray* pColMatchColInfo,
+                                      SExecTaskInfo* pTaskInfo) {
   SSortOperatorInfo* pInfo = taosMemoryCalloc(1, sizeof(SSortOperatorInfo));
   SOperatorInfo*     pOperator = taosMemoryCalloc(1, sizeof(SOperatorInfo));
   int32_t            rowSize = pResBlock->info.rowSize;
@@ -32,33 +33,33 @@ SOperatorInfo* createSortOperatorInfo(SOperatorInfo* downstream, SSDataBlock* pR
     goto _error;
   }
 
-  pOperator->pExpr    = pExprInfo;
+  pOperator->pExpr = pExprInfo;
   pOperator->numOfExprs = numOfCols;
-  pInfo->binfo.pCtx   = createSqlFunctionCtx(pExprInfo, numOfCols, &pInfo->binfo.rowCellInfoOffset);
-  pInfo->binfo.pRes   = pResBlock;
+  pInfo->binfo.pCtx = createSqlFunctionCtx(pExprInfo, numOfCols, &pInfo->binfo.rowCellInfoOffset);
+  pInfo->binfo.pRes = pResBlock;
 
   initResultSizeInfo(pOperator, 1024);
 
-  pInfo->pSortInfo    = pSortInfo;
-  pInfo->pColMatchInfo= pColMatchColInfo;
-  pOperator->name     = "SortOperator";
+  pInfo->pSortInfo = pSortInfo;
+  pInfo->pColMatchInfo = pColMatchColInfo;
+  pOperator->name = "SortOperator";
   pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_SORT;
   pOperator->blocking = true;
-  pOperator->status   = OP_NOT_OPENED;
-  pOperator->info     = pInfo;
+  pOperator->status = OP_NOT_OPENED;
+  pOperator->info = pInfo;
 
   // lazy evaluation for the following parameter since the input datablock is not known till now.
-//  pInfo->bufPageSize  = rowSize < 1024 ? 1024 * 2 : rowSize * 2;  // there are headers, so pageSize = rowSize + header
-//  pInfo->sortBufSize  = pInfo->bufPageSize * 16;  // TODO dynamic set the available sort buffer
+  //  pInfo->bufPageSize  = rowSize < 1024 ? 1024 * 2 : rowSize * 2;  // there are headers, so pageSize = rowSize +
+  //  header pInfo->sortBufSize  = pInfo->bufPageSize * 16;  // TODO dynamic set the available sort buffer
 
   pOperator->pTaskInfo = pTaskInfo;
-  pOperator->fpSet =
-      createOperatorFpSet(doOpenSortOperator, doSort, NULL, NULL, destroyOrderOperatorInfo, NULL, NULL, getExplainExecInfo);
+  pOperator->fpSet = createOperatorFpSet(doOpenSortOperator, doSort, NULL, NULL, destroyOrderOperatorInfo, NULL, NULL,
+                                         getExplainExecInfo);
 
   int32_t code = appendDownstream(pOperator, &downstream, 1);
   return pOperator;
 
-  _error:
+_error:
   pTaskInfo->code = TSDB_CODE_OUT_OF_MEMORY;
   taosMemoryFree(pInfo);
   taosMemoryFree(pOperator);
@@ -68,7 +69,7 @@ SOperatorInfo* createSortOperatorInfo(SOperatorInfo* downstream, SSDataBlock* pR
 void appendOneRowToDataBlock(SSDataBlock* pBlock, STupleHandle* pTupleHandle) {
   for (int32_t i = 0; i < pBlock->info.numOfCols; ++i) {
     SColumnInfoData* pColInfo = taosArrayGet(pBlock->pDataBlock, i);
-    bool isNull = tsortIsNullVal(pTupleHandle, i);
+    bool             isNull = tsortIsNullVal(pTupleHandle, i);
     if (isNull) {
       colDataAppendNULL(pColInfo, pBlock->info.rows);
     } else {
@@ -80,7 +81,8 @@ void appendOneRowToDataBlock(SSDataBlock* pBlock, STupleHandle* pTupleHandle) {
   pBlock->info.rows += 1;
 }
 
-SSDataBlock* getSortedBlockData(SSortHandle* pHandle, SSDataBlock* pDataBlock, int32_t capacity, SArray* pColMatchInfo) {
+SSDataBlock* getSortedBlockData(SSortHandle* pHandle, SSDataBlock* pDataBlock, int32_t capacity,
+                                SArray* pColMatchInfo) {
   blockDataCleanup(pDataBlock);
   ASSERT(taosArrayGetSize(pColMatchInfo) == pDataBlock->info.numOfCols);
 
@@ -129,10 +131,11 @@ SSDataBlock* loadNextDataBlock(void* param) {
 
 // todo refactor: merged with fetch fp
 void applyScalarFunction(SSDataBlock* pBlock, void* param) {
-  SOperatorInfo* pOperator = param;
+  SOperatorInfo*     pOperator = param;
   SSortOperatorInfo* pSort = pOperator->info;
   if (pOperator->pExpr != NULL) {
-    int32_t code = projectApplyFunctions(pOperator->pExpr, pBlock, pBlock, pSort->binfo.pCtx, pOperator->numOfExprs, NULL);
+    int32_t code =
+        projectApplyFunctions(pOperator->pExpr, pBlock, pBlock, pSort->binfo.pCtx, pOperator->numOfExprs, NULL);
     if (code != TSDB_CODE_SUCCESS) {
       longjmp(pOperator->pTaskInfo->env, code);
     }
@@ -141,7 +144,7 @@ void applyScalarFunction(SSDataBlock* pBlock, void* param) {
 
 int32_t doOpenSortOperator(SOperatorInfo* pOperator) {
   SSortOperatorInfo* pInfo = pOperator->info;
-  SExecTaskInfo* pTaskInfo = pOperator->pTaskInfo;
+  SExecTaskInfo*     pTaskInfo = pOperator->pTaskInfo;
 
   if (OPTR_IS_OPENED(pOperator)) {
     return TSDB_CODE_SUCCESS;
@@ -150,8 +153,8 @@ int32_t doOpenSortOperator(SOperatorInfo* pOperator) {
   pInfo->startTs = taosGetTimestampUs();
 
   //  pInfo->binfo.pRes is not equalled to the input datablock.
-  pInfo->pSortHandle = tsortCreateSortHandle(pInfo->pSortInfo, pInfo->pColMatchInfo, SORT_SINGLESOURCE_SORT,
-                                             -1, -1, NULL, pTaskInfo->id.str);
+  pInfo->pSortHandle = tsortCreateSortHandle(pInfo->pSortInfo, pInfo->pColMatchInfo, SORT_SINGLESOURCE_SORT, -1, -1,
+                                             NULL, pTaskInfo->id.str);
 
   tsortSetFetchRawDataFp(pInfo->pSortHandle, loadNextDataBlock, applyScalarFunction, pOperator);
 
@@ -166,7 +169,7 @@ int32_t doOpenSortOperator(SOperatorInfo* pOperator) {
     longjmp(pTaskInfo->env, terrno);
   }
 
-  pOperator->cost.openCost = (taosGetTimestampUs() - pInfo->startTs)/1000.0;
+  pOperator->cost.openCost = (taosGetTimestampUs() - pInfo->startTs) / 1000.0;
   pOperator->status = OP_RES_TO_RETURN;
 
   OPTR_SET_OPENED(pOperator);
@@ -186,7 +189,8 @@ SSDataBlock* doSort(SOperatorInfo* pOperator) {
     longjmp(pTaskInfo->env, code);
   }
 
-  SSDataBlock* pBlock = getSortedBlockData(pInfo->pSortHandle, pInfo->binfo.pRes, pOperator->resultInfo.capacity, pInfo->pColMatchInfo);
+  SSDataBlock* pBlock =
+      getSortedBlockData(pInfo->pSortHandle, pInfo->binfo.pRes, pOperator->resultInfo.capacity, pInfo->pColMatchInfo);
 
   if (pBlock != NULL) {
     pOperator->resultInfo.totalRows += pBlock->info.rows;
@@ -208,7 +212,7 @@ int32_t getExplainExecInfo(SOperatorInfo* pOptr, void** pOptrExplain, uint32_t*
   ASSERT(pOptr != NULL);
   SSortExecInfo* pInfo = taosMemoryCalloc(1, sizeof(SSortExecInfo));
 
-  SSortOperatorInfo *pOperatorInfo = (SSortOperatorInfo*)pOptr->info;
+  SSortOperatorInfo* pOperatorInfo = (SSortOperatorInfo*)pOptr->info;
 
   *pInfo = tsortGetSortExecInfo(pOperatorInfo->pSortHandle);
   *pOptrExplain = pInfo;
@@ -219,20 +223,19 @@ int32_t getExplainExecInfo(SOperatorInfo* pOptr, void** pOptrExplain, uint32_t*
 typedef struct SMultiwaySortMergeOperatorInfo {
   SOptrBasicInfo binfo;
 
-  int32_t      bufPageSize;
-  uint32_t     sortBufSize;    // max buffer size for in-memory sort
+  int32_t  bufPageSize;
+  uint32_t sortBufSize;  // max buffer size for in-memory sort
 
   SArray*      pSortInfo;
   SSortHandle* pSortHandle;
   SArray*      pColMatchInfo;  // for index map from table scan output
 
-  int64_t      startTs;       // sort start time
+  int64_t startTs;  // sort start time
 } SMultiwaySortMergeOperatorInfo;
 
-
 int32_t doOpenMultiwaySortMergeOperator(SOperatorInfo* pOperator) {
-  SMultiwaySortMergeOperatorInfo * pInfo = pOperator->info;
-  SExecTaskInfo* pTaskInfo = pOperator->pTaskInfo;
+  SMultiwaySortMergeOperatorInfo* pInfo = pOperator->info;
+  SExecTaskInfo*                  pTaskInfo = pOperator->pTaskInfo;
 
   if (OPTR_IS_OPENED(pOperator)) {
     return TSDB_CODE_SUCCESS;
@@ -259,7 +262,7 @@ int32_t doOpenMultiwaySortMergeOperator(SOperatorInfo* pOperator) {
     longjmp(pTaskInfo->env, terrno);
   }
 
-  pOperator->cost.openCost = (taosGetTimestampUs() - pInfo->startTs)/1000.0;
+  pOperator->cost.openCost = (taosGetTimestampUs() - pInfo->startTs) / 1000.0;
   pOperator->status = OP_RES_TO_RETURN;
 
   OPTR_SET_OPENED(pOperator);
@@ -271,18 +274,16 @@ SSDataBlock* doMultiwaySortMerge(SOperatorInfo* pOperator) {
     return NULL;
   }
 
-  SExecTaskInfo*     pTaskInfo = pOperator->pTaskInfo;
-  SSortOperatorInfo* pInfo = pOperator->info;
+  SExecTaskInfo*                  pTaskInfo = pOperator->pTaskInfo;
+  SMultiwaySortMergeOperatorInfo* pInfo = pOperator->info;
 
   int32_t code = pOperator->fpSet._openFn(pOperator);
   if (code != TSDB_CODE_SUCCESS) {
     longjmp(pTaskInfo->env, code);
   }
 
-  SSDataBlock* pBlock = getSortedBlockData(pInfo->pSortHandle,
-                                           pInfo->binfo.pRes,
-                                           pOperator->resultInfo.capacity,
-                                           pInfo->pColMatchInfo);
+  SSDataBlock* pBlock =
+      getSortedBlockData(pInfo->pSortHandle, pInfo->binfo.pRes, pOperator->resultInfo.capacity, pInfo->pColMatchInfo);
 
   if (pBlock != NULL) {
     pOperator->resultInfo.totalRows += pBlock->info.rows;
@@ -293,17 +294,18 @@ SSDataBlock* doMultiwaySortMerge(SOperatorInfo* pOperator) {
 }
 
 void destroyMultiwaySortMergeOperatorInfo(void* param, int32_t numOfOutput) {
-  SSortOperatorInfo* pInfo = (SSortOperatorInfo*)param;
+  SMultiwaySortMergeOperatorInfo * pInfo = (SMultiwaySortMergeOperatorInfo*)param;
   pInfo->binfo.pRes = blockDataDestroy(pInfo->binfo.pRes);
 
   taosArrayDestroy(pInfo->pSortInfo);
+  taosArrayDestroy(pInfo->pColMatchInfo);
 }
 
 int32_t getMultiwaySortMergeExplainExecInfo(SOperatorInfo* pOptr, void** pOptrExplain, uint32_t* len) {
   ASSERT(pOptr != NULL);
   SSortExecInfo* pInfo = taosMemoryCalloc(1, sizeof(SSortExecInfo));
 
-  SMultiwaySortMergeOperatorInfo *pOperatorInfo = (SMultiwaySortMergeOperatorInfo*)pOptr->info;
+  SMultiwaySortMergeOperatorInfo* pOperatorInfo = (SMultiwaySortMergeOperatorInfo*)pOptr->info;
 
   *pInfo = tsortGetSortExecInfo(pOperatorInfo->pSortHandle);
   *pOptrExplain = pInfo;
@@ -315,35 +317,32 @@ SOperatorInfo* createMultiwaySortMergeOperatorInfo(SOperatorInfo** downStreams,
                                                    SSDataBlock* pResBlock, SArray* pSortInfo, SArray* pColMatchColInfo,
                                                    SExecTaskInfo* pTaskInfo) {
   SMultiwaySortMergeOperatorInfo* pInfo = taosMemoryCalloc(1, sizeof(SMultiwaySortMergeOperatorInfo));
-  SOperatorInfo*     pOperator = taosMemoryCalloc(1, sizeof(SOperatorInfo));
-  int32_t            rowSize = pResBlock->info.rowSize;
+  SOperatorInfo*                  pOperator = taosMemoryCalloc(1, sizeof(SOperatorInfo));
+  int32_t                         rowSize = pResBlock->info.rowSize;
 
   if (pInfo == NULL || pOperator == NULL || rowSize > 100 * 1024 * 1024) {
     goto _error;
   }
 
-  pInfo->binfo.pRes   = pResBlock;
+  pInfo->binfo.pRes = pResBlock;
 
   initResultSizeInfo(pOperator, 1024);
 
-  pInfo->pSortInfo    = pSortInfo;
-  pInfo->pColMatchInfo= pColMatchColInfo;
-  pOperator->name     = "MultiwaySortMerge";
+  pInfo->pSortInfo = pSortInfo;
+  pInfo->pColMatchInfo = pColMatchColInfo;
+  pOperator->name = "MultiwaySortMerge";
   pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_MERGE;
   pOperator->blocking = true;
-  pOperator->status   = OP_NOT_OPENED;
-  pOperator->info     = pInfo;
+  pOperator->status = OP_NOT_OPENED;
+  pOperator->info = pInfo;
 
   pInfo->bufPageSize = rowSize < 1024 ? 1024 : rowSize * 2;
   pInfo->sortBufSize = pInfo->bufPageSize * 16;
 
   pOperator->pTaskInfo = pTaskInfo;
   pOperator->fpSet =
-      createOperatorFpSet(doOpenMultiwaySortMergeOperator, doMultiwaySortMerge,
-                          NULL,
-                          NULL, destroyMultiwaySortMergeOperatorInfo,
-                          NULL, NULL,
-                          getMultiwaySortMergeExplainExecInfo);
+      createOperatorFpSet(doOpenMultiwaySortMergeOperator, doMultiwaySortMerge, NULL, NULL,
+                          destroyMultiwaySortMergeOperatorInfo, NULL, NULL, getMultiwaySortMergeExplainExecInfo);
 
   int32_t code = appendDownstream(pOperator, downStreams, numStreams);
   if (code != TSDB_CODE_SUCCESS) {

From 4700a83f099520ffe21a6b94a79191860eca9c0b Mon Sep 17 00:00:00 2001
From: jiacy-jcy <cyjia@taosdata.com>
Date: Mon, 6 Jun 2022 15:34:19 +0800
Subject: [PATCH 18/34] add case for create_table_comment

---
 .../1-insert/create_table_comment.py          | 113 ++++++++++++++++++
 1 file changed, 113 insertions(+)
 create mode 100644 tests/system-test/1-insert/create_table_comment.py

diff --git a/tests/system-test/1-insert/create_table_comment.py b/tests/system-test/1-insert/create_table_comment.py
new file mode 100644
index 0000000000..92ea083c5a
--- /dev/null
+++ b/tests/system-test/1-insert/create_table_comment.py
@@ -0,0 +1,113 @@
+###################################################################
+#           Copyright (c) 2016 by TAOS Technologies, Inc.
+#                     All rights reserved.
+#
+#  This file is proprietary and confidential to TAOS Technologies.
+#  No part of this file may be reproduced, stored, transmitted,
+#  disclosed or used in any form or by any means other than as
+#  expressly provided by the written permission from Jianhui Tao
+#
+###################################################################
+
+# -*- coding: utf-8 -*-
+
+import random
+import string
+from util.log import *
+from util.cases import *
+from util.sql import *
+
+class TDTestCase:
+    def init(self, conn, logSql):
+        tdLog.debug("start to execute %s" % __file__)
+        tdSql.init(conn.cursor())
+
+    def get_long_name(self, length, mode="mixed"):
+        """
+        generate long name
+        mode could be numbers/letters/letters_mixed/mixed
+        """
+        if mode == "numbers":
+            population = string.digits
+        elif mode == "letters":
+            population = string.ascii_letters.lower()
+        elif mode == "letters_mixed":
+            population = string.ascii_letters.upper() + string.ascii_letters.lower()
+        else:
+            population = string.ascii_letters.lower() + string.digits
+        return "".join(random.choices(population, k=length))
+
+    def __create_tb(self,dbname,stbname,tbname,comment):
+        tdSql.execute(f'create database if not exists {dbname}')
+        tdSql.execute(f'use {dbname}')
+        tdSql.execute(
+            f'create table {stbname} (ts timestamp,c0 int) tags(t0 int) ')
+        tdSql.execute(
+            f'create table {tbname} using {stbname} tags(1) comment "{comment}"')
+    def __create_normaltb(self,dbname,tbname,comment):
+        tdSql.execute(f'create database if not exists {dbname}')
+        tdSql.execute(f'use {dbname}')
+        tdSql.execute(
+            f'create table {tbname} (ts timestamp,c0 int) comment "{comment}"')
+
+    def check_comment(self):
+        dbname = self.get_long_name(length=10, mode="letters")
+        ntbname = self.get_long_name(length=5, mode="letters")
+
+        # create normal table with comment
+        comment = self.get_long_name(length=10, mode="letters")
+        self.__create_normaltb(dbname,ntbname,comment)
+        ntb_kv_list = tdSql.getResult("show tables")
+        print(ntb_kv_list)
+        tdSql.checkEqual(ntb_kv_list[0][8], comment)
+        tdSql.error('alter table {ntbname} comment "test1"')
+        tdSql.execute(f'drop database {dbname}')
+
+        # max length(1024)
+        comment = self.get_long_name(length=1024, mode="letters")
+        self.__create_normaltb(dbname,ntbname,comment)
+        ntb_kv_list = tdSql.getResult("show tables")
+        tdSql.checkEqual(ntb_kv_list[0][8], comment)
+        tdSql.execute(f'drop database {dbname}')
+
+        # error overlength
+        comment = self.get_long_name(length=1025, mode="letters")
+        tdSql.execute(f'create database if not exists {dbname}')
+        tdSql.execute(f'use {dbname}')
+        tdSql.error(f"create table ntb (ts timestamp,c0 int) comment '{comment}'")
+        tdSql.execute(f'drop database {dbname}')
+
+        # create child table with comment
+        comment = self.get_long_name(length=10, mode="letters")
+        stbname = self.get_long_name(length=5, mode="letters")
+        tbname = self.get_long_name(length=3, mode="letters")
+        self.__create_tb(dbname,stbname,tbname,comment)
+        ntb_kv_list = tdSql.getResult("show tables")
+        tdSql.checkEqual(ntb_kv_list[0][8], comment)
+        tdSql.error(f'alter table {tbname} comment "test1"')
+        tdSql.execute(f'drop database {dbname}')
+
+        # max length 1024
+        comment = self.get_long_name(length=1024, mode="letters")
+        self.__create_tb(dbname,ntbname,comment)
+        ntb_kv_list = tdSql.getResult("show tables")
+        tdSql.checkEqual(ntb_kv_list[0][8], comment)
+        tdSql.execute(f'drop database {dbname}')
+
+        # error overlength
+        comment = self.get_long_name(length=1025, mode="letters")
+        tdSql.execute(f'create database if not exists {dbname}')
+        tdSql.execute(f'use {dbname}')
+        tdSql.execute(f"create table stb (ts timestamp,c0 int) tags(t0 int)")
+        tdSql.error(f'create table stb_1 us stb tags(1) comment "{comment}"')
+        tdSql.execute(f'drop database {dbname}')
+
+    def run(self):
+        self.check_comment()
+
+    def stop(self):
+        tdSql.close()
+        tdLog.success("%s successfully executed" % __file__)
+
+tdCases.addWindows(__file__, TDTestCase())
+tdCases.addLinux(__file__, TDTestCase())
\ No newline at end of file

From 8015c71f20a7d3300a676b985fa8fbd2ee991185 Mon Sep 17 00:00:00 2001
From: jiacy-jcy <cyjia@taosdata.com>
Date: Mon, 6 Jun 2022 15:34:56 +0800
Subject: [PATCH 19/34] add case into ci

---
 tests/system-test/fulltest.sh | 1 +
 1 file changed, 1 insertion(+)

diff --git a/tests/system-test/fulltest.sh b/tests/system-test/fulltest.sh
index 3991654350..64e1fef77d 100755
--- a/tests/system-test/fulltest.sh
+++ b/tests/system-test/fulltest.sh
@@ -21,6 +21,7 @@ python3 ./test.py -f 1-insert/opentsdb_json_taosc_insert.py
 #python3 ./test.py -f 1-insert/test_stmt_muti_insert_query.py
 python3 ./test.py -f 1-insert/alter_stable.py
 python3 ./test.py -f 1-insert/alter_table.py
+python3 ./test.py -f 1-inerst/create_table_comment.py
 python3 ./test.py -f 2-query/between.py
 python3 ./test.py -f 2-query/distinct.py
 python3 ./test.py -f 2-query/varchar.py

From 1d1f392a01b4ccaea239a00603b5343044a87cbd Mon Sep 17 00:00:00 2001
From: jiacy-jcy <cyjia@taosdata.com>
Date: Mon, 6 Jun 2022 15:35:11 +0800
Subject: [PATCH 20/34] update

---
 tests/system-test/fulltest.sh | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/tests/system-test/fulltest.sh b/tests/system-test/fulltest.sh
index 64e1fef77d..792ba6ee4a 100755
--- a/tests/system-test/fulltest.sh
+++ b/tests/system-test/fulltest.sh
@@ -21,7 +21,7 @@ python3 ./test.py -f 1-insert/opentsdb_json_taosc_insert.py
 #python3 ./test.py -f 1-insert/test_stmt_muti_insert_query.py
 python3 ./test.py -f 1-insert/alter_stable.py
 python3 ./test.py -f 1-insert/alter_table.py
-python3 ./test.py -f 1-inerst/create_table_comment.py
+# python3 ./test.py -f 1-inerst/create_table_comment.py
 python3 ./test.py -f 2-query/between.py
 python3 ./test.py -f 2-query/distinct.py
 python3 ./test.py -f 2-query/varchar.py

From bae49baa9868194b3c1572ab8b482e24bdacfb94 Mon Sep 17 00:00:00 2001
From: wangmm0220 <wangmm0220@gmail.com>
Date: Mon, 6 Jun 2022 16:09:36 +0800
Subject: [PATCH 21/34] fix: fix error in json and add test cases for json

---
 source/libs/parser/src/parInsert.c    |  8 ++-
 tests/system-test/2-query/json_tag.py | 85 ++++++++++++++-------------
 tests/system-test/fulltest.sh         |  1 +
 3 files changed, 50 insertions(+), 44 deletions(-)

diff --git a/source/libs/parser/src/parInsert.c b/source/libs/parser/src/parInsert.c
index efe0bd75b2..8f17d500ab 100644
--- a/source/libs/parser/src/parInsert.c
+++ b/source/libs/parser/src/parInsert.c
@@ -1006,7 +1006,7 @@ static int32_t parseTagsClause(SInsertParseContext* pCxt, SSchema* pSchema, uint
     }
 
     SSchema* pTagSchema = &pSchema[pCxt->tags.boundColumns[i]];
-    char*    tmpTokenBuf = taosMemoryCalloc(1, sToken.n);  // this can be optimize with parse column
+    char*    tmpTokenBuf = taosMemoryCalloc(1, sToken.n);  //todo this can be optimize with parse column
     code = checkAndTrimValue(&sToken, tmpTokenBuf, &pCxt->msg);
     if (code != TSDB_CODE_SUCCESS) {
       taosMemoryFree(tmpTokenBuf);
@@ -1018,7 +1018,11 @@ static int32_t parseTagsClause(SInsertParseContext* pCxt, SSchema* pSchema, uint
         taosMemoryFree(tmpTokenBuf);
         goto end;
       }
-      code = parseJsontoTagData(sToken.z, pTagVals, &pTag, &pCxt->msg);
+      if(isNullStr(&sToken)) {
+        code = tTagNew(pTagVals, 1, true, &pTag);
+      } else {
+        code = parseJsontoTagData(sToken.z, pTagVals, &pTag, &pCxt->msg);
+      }
       taosMemoryFree(tmpTokenBuf);
       if (code != TSDB_CODE_SUCCESS) {
         goto end;
diff --git a/tests/system-test/2-query/json_tag.py b/tests/system-test/2-query/json_tag.py
index 6a855ebd4b..f6f09d189d 100644
--- a/tests/system-test/2-query/json_tag.py
+++ b/tests/system-test/2-query/json_tag.py
@@ -61,6 +61,8 @@ class TDTestCase:
         # test invalidate json
         tdSql.error("CREATE TABLE if not exists jsons1_14 using jsons1 tags('\"efwewf\"')")
         tdSql.error("CREATE TABLE if not exists jsons1_14 using jsons1 tags('3333')")
+        tdSql.error("CREATE TABLE if not exists jsons1_14 using jsons1 tags(76)")
+        tdSql.error("CREATE TABLE if not exists jsons1_14 using jsons1 tags(hell)")
         tdSql.error("CREATE TABLE if not exists jsons1_14 using jsons1 tags('33.33')")
         tdSql.error("CREATE TABLE if not exists jsons1_14 using jsons1 tags('false')")
         tdSql.error("CREATE TABLE if not exists jsons1_14 using jsons1 tags('[1,true]')")
@@ -128,7 +130,7 @@ class TDTestCase:
         tdSql.error("select ceil(jtag) from jsons1")
         #
         # # test select normal column
-        tdSql.query("select dataint from jsons1")
+        tdSql.query("select dataint from jsons1 order by dataint")
         tdSql.checkRows(9)
         tdSql.checkData(1, 0, 1)
 
@@ -137,9 +139,9 @@ class TDTestCase:
         tdSql.checkRows(9)
         tdSql.query("select jtag from jsons1")
         tdSql.checkRows(13)
-        tdSql.query("select jtag from jsons1 where jtag is null")
+        # tdSql.query("select jtag from jsons1 where jtag is null")
         # tdSql.checkRows(5)
-        tdSql.query("select jtag from jsons1 where jtag is not null")
+        # tdSql.query("select jtag from jsons1 where jtag is not null")
         # tdSql.checkRows(8)
 
         # test jtag is NULL
@@ -259,12 +261,6 @@ class TDTestCase:
         # tdSql.query("select * from jsons1 where jtag->'tag1'=null")     # only json suport =null. This synatx will change later.
         # tdSql.checkRows(1)
         #
-        # # where json is null
-        tdSql.query("select * from jsons1 where jtag is null")
-        # tdSql.checkRows(1)
-        tdSql.query("select * from jsons1 where jtag is not null")
-        # tdSql.checkRows(8)
-        #
         # # where json key is null
         # tdSql.query("select * from jsons1 where jtag->'tag_no_exist'=3")
         # tdSql.checkRows(0)
@@ -358,25 +354,28 @@ class TDTestCase:
         # tdSql.checkRows(0)
         #
         # # test join
-        # tdSql.execute("create table if not exists jsons2(ts timestamp, dataInt int, dataBool bool, dataStr nchar(50), dataStrBin binary(150)) tags(jtag json)")
-        # tdSql.execute("insert into jsons2_1 using jsons2 tags('{\"tag1\":\"fff\",\"tag2\":5, \"tag3\":true}') values(1591060618000, 2, false, 'json2', '你是2')")
-        # tdSql.execute("insert into jsons2_2 using jsons2 tags('{\"tag1\":5,\"tag2\":null}') values (1591060628000, 2, true, 'json2', 'sss')")
-        #
-        # tdSql.execute("create table if not exists jsons3(ts timestamp, dataInt int, dataBool bool, dataStr nchar(50), dataStrBin binary(150)) tags(jtag json)")
-        # tdSql.execute("insert into jsons3_1 using jsons3 tags('{\"tag1\":\"fff\",\"tag2\":5, \"tag3\":true}') values(1591060618000, 3, false, 'json3', '你是3')")
-        # tdSql.execute("insert into jsons3_2 using jsons3 tags('{\"tag1\":5,\"tag2\":\"beijing\"}') values (1591060638000, 2, true, 'json3', 'sss')")
-        # tdSql.query("select 'sss',33,a.jtag->'tag3' from jsons2 a,jsons3 b where a.ts=b.ts and a.jtag->'tag1'=b.jtag->'tag1'")
-        # tdSql.checkData(0, 0, "sss")
-        # tdSql.checkData(0, 2, "true")
-        #
-        # res = tdSql.getColNameList("select 'sss',33,a.jtag->'tag3' from jsons2 a,jsons3 b where a.ts=b.ts and a.jtag->'tag1'=b.jtag->'tag1'")
-        # cname_list = []
-        # cname_list.append("sss")
-        # cname_list.append("33")
-        # cname_list.append("a.jtag->'tag3'")
-        # tdSql.checkColNameList(res, cname_list)
+        tdSql.execute("create table if not exists jsons2(ts timestamp, dataInt int, dataBool bool, dataStr nchar(50), dataStrBin binary(150)) tags(jtag json)")
+        tdSql.execute("insert into jsons2_1 using jsons2 tags('{\"tag1\":\"fff\",\"tag2\":5, \"tag3\":true}') values(1591060618000, 2, false, 'json2', '你是2')")
+        tdSql.execute("insert into jsons2_2 using jsons2 tags('{\"tag1\":5,\"tag2\":null}') values (1591060628000, 2, true, 'json2', 'sss')")
+
+        tdSql.execute("create table if not exists jsons3(ts timestamp, dataInt int, dataBool bool, dataStr nchar(50), dataStrBin binary(150)) tags(jtag json)")
+        tdSql.execute("insert into jsons3_1 using jsons3 tags('{\"tag1\":\"fff\",\"tag2\":5, \"tag3\":true}') values(1591060618000, 3, false, 'json3', '你是3')")
+        tdSql.execute("insert into jsons3_2 using jsons3 tags('{\"tag1\":5,\"tag2\":\"beijing\"}') values (1591060638000, 2, true, 'json3', 'sss')")
+        tdSql.query("select 'sss',33,a.jtag->'tag3' from jsons2 a,jsons3 b where a.ts=b.ts and a.jtag->'tag1'=b.jtag->'tag1'")
+        tdSql.checkData(0, 0, "sss")
+        tdSql.checkData(0, 2, "true")
+
+        res = tdSql.getColNameList("select 'sss',33,a.jtag->'tag3' from jsons2 a,jsons3 b where a.ts=b.ts and a.jtag->'tag1'=b.jtag->'tag1'")
+        cname_list = []
+        cname_list.append("'sss'")
+        cname_list.append("33")
+        cname_list.append("a.jtag->'tag3'")
+        tdSql.checkColNameList(res, cname_list)
         #
         # test group by & order by  json tag
+        # tdSql.error("select count(*) from jsons1 group by jtag")
+        # tdSql.error("select count(*) from jsons1 partition by jtag")
+        # tdSql.error("select count(*) from jsons1 group by jtag order by jtag")
         tdSql.error("select count(*) from jsons1 group by jtag->'tag1' order by jtag->'tag2'")
         tdSql.error("select count(*) from jsons1 group by jtag->'tag1' order by jtag")
         tdSql.query("select count(*),jtag->'tag1' from jsons1 group by jtag->'tag1' order by jtag->'tag1' desc")
@@ -432,14 +431,14 @@ class TDTestCase:
         tdSql.checkData(10, 1, '"femail"')
 
         # test having
-        # tdSql.query("select stddev(dataint) from jsons1 group by jtag->'tag1' having stddev(dataint) > 0")
-        # tdSql.checkRows(2)
+        tdSql.query("select count(*),jtag->'tag1' from jsons1 group by jtag->'tag1' having count(*) > 1")
+        tdSql.checkRows(3)
 
         # subquery with json tag
         tdSql.query("select * from (select jtag, dataint from jsons1) order by dataint")
         tdSql.checkRows(11)
         tdSql.checkData(1, 1, 1)
-        tdSql.checkData(2, 0, '{"tag1":5,"tag2":"beijing"}')
+        tdSql.checkData(5, 0, '{"tag1":false,"tag2":"beijing"}')
 
         # tdSql.query("select jtag->'tag1' from (select jtag->'tag1', dataint from jsons1)")
         # tdSql.checkRows(11)
@@ -457,16 +456,18 @@ class TDTestCase:
         # tdSql.checkData(1, 2, '"femail"')
         #
         # # union all
-        # tdSql.error("select jtag->'tag1' from jsons1 union all select jtag->'tag2' from jsons2")
-        # tdSql.error("select jtag->'tag1' from jsons1_1 union all select jtag->'tag2' from jsons2_1")
-        #
-        # tdSql.query("select jtag->'tag1' from jsons1_1 union all select jtag->'tag1' from jsons2_1")
-        # tdSql.checkRows(2)
-        # tdSql.query("select dataint,jtag->'tag1',tbname from jsons1 union all select dataint,jtag->'tag1',tbname from jsons2")
-        # tdSql.checkRows(13)
-        # tdSql.query("select dataint,jtag,tbname from jsons1 union all select dataint,jtag,tbname from jsons2")
-        # tdSql.checkRows(13)
-        #
+        tdSql.query("select jtag->'tag1' from jsons1 union all select jtag->'tag2' from jsons2")
+        tdSql.checkRows(17)
+        tdSql.query("select jtag->'tag1' from jsons1_1 union all select jtag->'tag2' from jsons2_1")
+        tdSql.checkRows(2)
+
+        tdSql.query("select jtag->'tag1' from jsons1_1 union all select jtag->'tag1' from jsons2_1")
+        tdSql.checkRows(2)
+        tdSql.query("select dataint,jtag->'tag1',tbname from jsons1 union all select dataint,jtag->'tag1',tbname from jsons2")
+        tdSql.checkRows(13)
+        tdSql.query("select dataint,jtag,tbname from jsons1 union all select dataint,jtag,tbname from jsons2")
+        tdSql.checkRows(13)
+
         # #show create table
         # tdSql.query("show create table jsons1")
         # tdSql.checkData(0, 1, 'CREATE TABLE `jsons1` (`ts` TIMESTAMP,`dataint` INT,`databool` BOOL,`datastr` NCHAR(50),`datastrbin` BINARY(150)) TAGS (`jtag` JSON)')
@@ -528,9 +529,9 @@ class TDTestCase:
         tdSql.execute("CREATE TABLE if not exists jsons1_20 using jsons1 tags(NULL)")
         tdSql.query("select jtag from jsons1_20")
         tdSql.checkData(0, 0, None)
-        # tdSql.execute("insert into jsons1_21 using jsons1 tags(NULL) values(1591061628000, 11, false, '你就会','')")
-        # tdSql.query("select jtag from jsons1_21")
-        # tdSql.checkData(0, 0, None)
+        tdSql.execute("insert into jsons1_21 using jsons1 tags(NULL) values(1591061628000, 11, false, '你就会','')")
+        tdSql.query("select jtag from jsons1_21")
+        tdSql.checkData(0, 0, None)
         #
         # #test TD-12389
         tdSql.query("describe jsons1")
diff --git a/tests/system-test/fulltest.sh b/tests/system-test/fulltest.sh
index 3991654350..1fbbc9f37e 100755
--- a/tests/system-test/fulltest.sh
+++ b/tests/system-test/fulltest.sh
@@ -57,6 +57,7 @@ python3 ./test.py -f 2-query/To_unixtimestamp.py
 python3 ./test.py -f 2-query/timetruncate.py
 python3 ./test.py -f 2-query/diff.py
 python3 ./test.py -f 2-query/Timediff.py
+python3 ./test.py -f 2-query/json_tag.py
 
 python3 ./test.py -f 2-query/top.py
 python3 ./test.py -f 2-query/bottom.py

From ea8239c8e5716909aedd4f7142a0d9d01e77ae90 Mon Sep 17 00:00:00 2001
From: Cary Xu <klxu@taosdata.com>
Date: Mon, 6 Jun 2022 16:40:37 +0800
Subject: [PATCH 22/34] fix: fix type convert of sum/avg for rsma

---
 source/common/src/tdatablock.c | 24 +++++++++++++++++++++---
 1 file changed, 21 insertions(+), 3 deletions(-)

diff --git a/source/common/src/tdatablock.c b/source/common/src/tdatablock.c
index c615af705a..d9293433ea 100644
--- a/source/common/src/tdatablock.c
+++ b/source/common/src/tdatablock.c
@@ -1590,7 +1590,6 @@ int32_t buildSubmitReqFromDataBlock(SSubmitReq** pReq, const SArray* pDataBlocks
       for (int32_t k = 0; k < colNum; ++k) {  // iterate by column
         SColumnInfoData* pColInfoData = taosArrayGet(pDataBlock->pDataBlock, k);
         STColumn*         pCol = &pTSchema->columns[k];
-        ASSERT(pCol->type == pColInfoData->info.type);
         void* var = POINTER_SHIFT(pColInfoData->pData, j * pColInfoData->info.bytes);
         switch (pColInfoData->info.type) {
           case TSDB_DATA_TYPE_TIMESTAMP:
@@ -1614,20 +1613,39 @@ int32_t buildSubmitReqFromDataBlock(SSubmitReq** pReq, const SArray* pDataBlocks
           case TSDB_DATA_TYPE_VARBINARY:
           case TSDB_DATA_TYPE_DECIMAL:
           case TSDB_DATA_TYPE_BLOB:
+          case TSDB_DATA_TYPE_JSON:
           case TSDB_DATA_TYPE_MEDIUMBLOB:
             uError("the column type %" PRIi16 " is defined but not implemented yet", pColInfoData->info.type);
             TASSERT(0);
             break;
           default:
             if (pColInfoData->info.type < TSDB_DATA_TYPE_MAX && pColInfoData->info.type > TSDB_DATA_TYPE_NULL) {
-              tdAppendColValToRow(&rb, PRIMARYKEY_TIMESTAMP_COL_ID + k, pColInfoData->info.type, TD_VTYPE_NORM, var, true, offset, k);
+              char tv[8] = {0};
+              if (pColInfoData->info.type == TSDB_DATA_TYPE_FLOAT) {
+                float v = 0;
+                GET_TYPED_DATA(v, float, pColInfoData->info.type, var);
+                SET_TYPED_DATA(&tv, pCol->type, v);
+              } else if (pColInfoData->info.type == TSDB_DATA_TYPE_DOUBLE) {
+                double v = 0;
+                GET_TYPED_DATA(v, double, pColInfoData->info.type, var);
+                SET_TYPED_DATA(&tv, pCol->type, v);
+              } else if (IS_SIGNED_NUMERIC_TYPE(pColInfoData->info.type)) {
+                int64_t v = 0;
+                GET_TYPED_DATA(v, int64_t, pColInfoData->info.type, var);
+                SET_TYPED_DATA(&tv, pCol->type, v);
+              } else {
+                uint64_t v = 0;
+                GET_TYPED_DATA(v, uint64_t, pColInfoData->info.type, var);
+                SET_TYPED_DATA(&tv, pCol->type, v);
+              }
+              tdAppendColValToRow(&rb, PRIMARYKEY_TIMESTAMP_COL_ID + k, pCol->type, TD_VTYPE_NORM, tv, true, offset, k);
             } else {
               uError("the column type %" PRIi16 " is undefined\n", pColInfoData->info.type);
               TASSERT(0);
             }
             break;
         }
-        offset += TYPE_BYTES[pColInfoData->info.type];
+        offset += TYPE_BYTES[pCol->type]; // sum/avg would convert to int64_t/uint64_t/double during aggregation
       }
       dataLen += TD_ROW_LEN(rb.pBuf);
 #ifdef TD_DEBUG_PRINT_ROW

From 362f37b5ed8ad42de2783e1eefb4fb8baf9802a7 Mon Sep 17 00:00:00 2001
From: Shengliang Guan <slguan@taosdata.com>
Date: Mon, 6 Jun 2022 16:21:56 +0800
Subject: [PATCH 23/34] feat: redistribute vgroup

---
 include/util/taoserror.h                |   1 +
 source/dnode/mnode/impl/src/mndBnode.c  |   4 +-
 source/dnode/mnode/impl/src/mndDnode.c  |   6 +-
 source/dnode/mnode/impl/src/mndMnode.c  |  12 +-
 source/dnode/mnode/impl/src/mndQnode.c  |   4 +-
 source/dnode/mnode/impl/src/mndSma.c    |   3 +-
 source/dnode/mnode/impl/src/mndSnode.c  |   4 +-
 source/dnode/mnode/impl/src/mndVgroup.c | 244 +++++++++++++++++++++---
 8 files changed, 237 insertions(+), 41 deletions(-)

diff --git a/include/util/taoserror.h b/include/util/taoserror.h
index 0b331abf8a..17851bc4a8 100644
--- a/include/util/taoserror.h
+++ b/include/util/taoserror.h
@@ -218,6 +218,7 @@ int32_t* taosGetErrno();
 #define TSDB_CODE_MND_VGROUP_NOT_EXIST          TAOS_DEF_ERROR_CODE(0, 0x0390)
 #define TSDB_CODE_MND_VGROUP_NOT_IN_DNODE       TAOS_DEF_ERROR_CODE(0, 0x0391)
 #define TSDB_CODE_MND_VGROUP_ALREADY_IN_DNODE   TAOS_DEF_ERROR_CODE(0, 0x0392)
+#define TSDB_CODE_MND_VGROUP_UN_CHANGED         TAOS_DEF_ERROR_CODE(0, 0x0393)
 
 // mnode-stable
 #define TSDB_CODE_MND_STB_ALREADY_EXIST         TAOS_DEF_ERROR_CODE(0, 0x03A0)
diff --git a/source/dnode/mnode/impl/src/mndBnode.c b/source/dnode/mnode/impl/src/mndBnode.c
index 38a166900d..ed07e15c63 100644
--- a/source/dnode/mnode/impl/src/mndBnode.c
+++ b/source/dnode/mnode/impl/src/mndBnode.c
@@ -299,7 +299,7 @@ static int32_t mndProcessCreateBnodeReq(SRpcMsg *pReq) {
     goto _OVER;
   }
 
-  if (mndCheckNodeAuth(pUser)) {
+  if (mndCheckNodeAuth(pUser) != 0) {
     goto _OVER;
   }
 
@@ -409,7 +409,7 @@ static int32_t mndProcessDropBnodeReq(SRpcMsg *pReq) {
     goto _OVER;
   }
 
-  if (mndCheckNodeAuth(pUser)) {
+  if (mndCheckNodeAuth(pUser) != 0) {
     goto _OVER;
   }
 
diff --git a/source/dnode/mnode/impl/src/mndDnode.c b/source/dnode/mnode/impl/src/mndDnode.c
index 9bdee92113..c7e347fb8b 100644
--- a/source/dnode/mnode/impl/src/mndDnode.c
+++ b/source/dnode/mnode/impl/src/mndDnode.c
@@ -542,7 +542,7 @@ static int32_t mndProcessCreateDnodeReq(SRpcMsg *pReq) {
     goto CREATE_DNODE_OVER;
   }
 
-  if (mndCheckNodeAuth(pUser)) {
+  if (mndCheckNodeAuth(pUser) != 0) {
     goto CREATE_DNODE_OVER;
   }
 
@@ -566,6 +566,7 @@ static int32_t mndDropDnode(SMnode *pMnode, SRpcMsg *pReq, SDnodeObj *pDnode, SM
 
   pTrans = mndTransCreate(pMnode, TRN_POLICY_RETRY, TRN_CONFLICT_GLOBAL, pReq);
   if (pTrans == NULL) goto _OVER;
+  mndTransSetSerial(pTrans);
   mDebug("trans:%d, used to drop dnode:%d", pTrans->id, pDnode->id);
 
   pRaw = mndDnodeActionEncode(pDnode);
@@ -582,7 +583,6 @@ static int32_t mndDropDnode(SMnode *pMnode, SRpcMsg *pReq, SDnodeObj *pDnode, SM
   if (mndSetMoveVgroupsInfoToTrans(pMnode, pTrans, pDnode->id) != 0) goto _OVER;
   if (mndTransPrepare(pMnode, pTrans) != 0) goto _OVER;
 
-  mndTransSetSerial(pTrans);
   code = 0;
 
 _OVER:
@@ -640,7 +640,7 @@ static int32_t mndProcessDropDnodeReq(SRpcMsg *pReq) {
     goto _OVER;
   }
 
-  if (mndCheckNodeAuth(pUser)) {
+  if (mndCheckNodeAuth(pUser) != 0) {
     goto _OVER;
   }
 
diff --git a/source/dnode/mnode/impl/src/mndMnode.c b/source/dnode/mnode/impl/src/mndMnode.c
index abd65c1724..d315bb12ed 100644
--- a/source/dnode/mnode/impl/src/mndMnode.c
+++ b/source/dnode/mnode/impl/src/mndMnode.c
@@ -358,9 +358,9 @@ static int32_t mndCreateMnode(SMnode *pMnode, SRpcMsg *pReq, SDnodeObj *pDnode,
 
   STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_RETRY, TRN_CONFLICT_GLOBAL, pReq);
   if (pTrans == NULL) goto _OVER;
-
-  mDebug("trans:%d, used to create mnode:%d", pTrans->id, pCreate->dnodeId);
   mndTransSetSerial(pTrans);
+  mDebug("trans:%d, used to create mnode:%d", pTrans->id, pCreate->dnodeId);
+
   if (mndSetCreateMnodeRedoLogs(pMnode, pTrans, &mnodeObj) != 0) goto _OVER;
   if (mndSetCreateMnodeCommitLogs(pMnode, pTrans, &mnodeObj) != 0) goto _OVER;
   if (mndSetCreateMnodeRedoActions(pMnode, pTrans, pDnode, &mnodeObj) != 0) goto _OVER;
@@ -419,7 +419,7 @@ static int32_t mndProcessCreateMnodeReq(SRpcMsg *pReq) {
     goto _OVER;
   }
 
-  if (mndCheckNodeAuth(pUser)) {
+  if (mndCheckNodeAuth(pUser) != 0) {
     goto _OVER;
   }
 
@@ -549,12 +549,12 @@ static int32_t mndDropMnode(SMnode *pMnode, SRpcMsg *pReq, SMnodeObj *pObj) {
 
   pTrans = mndTransCreate(pMnode, TRN_POLICY_RETRY, TRN_CONFLICT_GLOBAL, pReq);
   if (pTrans == NULL) goto _OVER;
-
+  mndTransSetSerial(pTrans);
   mDebug("trans:%d, used to drop mnode:%d", pTrans->id, pObj->id);
+
   if (mndSetDropMnodeInfoToTrans(pMnode, pTrans, pObj) != 0) goto _OVER;
   if (mndTransPrepare(pMnode, pTrans) != 0) goto _OVER;
 
-  mndTransSetSerial(pTrans);
   code = 0;
 
 _OVER:
@@ -602,7 +602,7 @@ static int32_t mndProcessDropMnodeReq(SRpcMsg *pReq) {
     goto _OVER;
   }
 
-  if (mndCheckNodeAuth(pUser)) {
+  if (mndCheckNodeAuth(pUser) != 0) {
     goto _OVER;
   }
 
diff --git a/source/dnode/mnode/impl/src/mndQnode.c b/source/dnode/mnode/impl/src/mndQnode.c
index 4fb9226144..aac6eaba47 100644
--- a/source/dnode/mnode/impl/src/mndQnode.c
+++ b/source/dnode/mnode/impl/src/mndQnode.c
@@ -301,7 +301,7 @@ static int32_t mndProcessCreateQnodeReq(SRpcMsg *pReq) {
     goto _OVER;
   }
 
-  if (mndCheckNodeAuth(pUser)) {
+  if (mndCheckNodeAuth(pUser) != 0) {
     goto _OVER;
   }
 
@@ -411,7 +411,7 @@ static int32_t mndProcessDropQnodeReq(SRpcMsg *pReq) {
     goto _OVER;
   }
 
-  if (mndCheckNodeAuth(pUser)) {
+  if (mndCheckNodeAuth(pUser) != 0) {
     goto _OVER;
   }
 
diff --git a/source/dnode/mnode/impl/src/mndSma.c b/source/dnode/mnode/impl/src/mndSma.c
index e8e23ccc2a..f7fa74f11a 100644
--- a/source/dnode/mnode/impl/src/mndSma.c
+++ b/source/dnode/mnode/impl/src/mndSma.c
@@ -510,10 +510,9 @@ static int32_t mndCreateSma(SMnode *pMnode, SRpcMsg *pReq, SMCreateSmaReq *pCrea
   int32_t code = -1;
   STrans *pTrans = mndTransCreate(pMnode, TRN_POLICY_RETRY, TRN_CONFLICT_DB, pReq);
   if (pTrans == NULL) goto _OVER;
-
-  mDebug("trans:%d, used to create sma:%s", pTrans->id, pCreate->name);
   mndTransSetDbName(pTrans, pDb->name);
   mndTransSetSerial(pTrans);
+  mDebug("trans:%d, used to create sma:%s", pTrans->id, pCreate->name);
 
   if (mndSetCreateSmaRedoLogs(pMnode, pTrans, &smaObj) != 0) goto _OVER;
   if (mndSetCreateSmaVgroupRedoLogs(pMnode, pTrans, &streamObj.fixedSinkVg) != 0) goto _OVER;
diff --git a/source/dnode/mnode/impl/src/mndSnode.c b/source/dnode/mnode/impl/src/mndSnode.c
index f5a1e76723..7d21528260 100644
--- a/source/dnode/mnode/impl/src/mndSnode.c
+++ b/source/dnode/mnode/impl/src/mndSnode.c
@@ -307,7 +307,7 @@ static int32_t mndProcessCreateSnodeReq(SRpcMsg *pReq) {
     goto _OVER;
   }
 
-  if (mndCheckNodeAuth(pUser)) {
+  if (mndCheckNodeAuth(pUser) != 0) {
     goto _OVER;
   }
 
@@ -419,7 +419,7 @@ static int32_t mndProcessDropSnodeReq(SRpcMsg *pReq) {
     goto _OVER;
   }
 
-  if (mndCheckNodeAuth(pUser)) {
+  if (mndCheckNodeAuth(pUser) != 0) {
     goto _OVER;
   }
 
diff --git a/source/dnode/mnode/impl/src/mndVgroup.c b/source/dnode/mnode/impl/src/mndVgroup.c
index afca8b14f0..d8f2dae234 100644
--- a/source/dnode/mnode/impl/src/mndVgroup.c
+++ b/source/dnode/mnode/impl/src/mndVgroup.c
@@ -15,11 +15,13 @@
 
 #define _DEFAULT_SOURCE
 #include "mndVgroup.h"
+#include "mndAuth.h"
 #include "mndDb.h"
 #include "mndDnode.h"
 #include "mndMnode.h"
 #include "mndShow.h"
 #include "mndTrans.h"
+#include "mndUser.h"
 
 #define VGROUP_VER_NUMBER   1
 #define VGROUP_RESERVE_SIZE 64
@@ -34,6 +36,9 @@ static void    mndCancelGetNextVgroup(SMnode *pMnode, void *pIter);
 static int32_t mndRetrieveVnodes(SRpcMsg *pReq, SShowObj *pShow, SSDataBlock *pBlock, int32_t rows);
 static void    mndCancelGetNextVnode(SMnode *pMnode, void *pIter);
 
+static int32_t mndProcessRedistributeVgroupMsg(SRpcMsg *pReq);
+static int32_t mndProcessSplitVgroupMsg(SRpcMsg *pReq);
+
 int32_t mndInitVgroup(SMnode *pMnode) {
   SSdbTable table = {
       .sdbType = SDB_VGROUP,
@@ -909,38 +914,38 @@ int32_t mndAddDropVnodeAction(SMnode *pMnode, STrans *pTrans, SDbObj *pDb, SVgOb
 
 int32_t mndSetMoveVgroupInfoToTrans(SMnode *pMnode, STrans *pTrans, SDbObj *pDb, SVgObj *pVgroup, int32_t vnIndex,
                                     SArray *pArray) {
-  SVgObj newVgroup = {0};
-  memcpy(&newVgroup, pVgroup, sizeof(SVgObj));
+  SVgObj newVg = {0};
+  memcpy(&newVg, pVgroup, sizeof(SVgObj));
 
-  mInfo("vgId:%d, vgroup info before move, replica:%d", newVgroup.vgId, newVgroup.replica);
-  for (int32_t i = 0; i < newVgroup.replica; ++i) {
-    mInfo("vgId:%d, vnode:%d dnode:%d", newVgroup.vgId, i, newVgroup.vnodeGid[i].dnodeId);
+  mInfo("vgId:%d, vgroup info before move, replica:%d", newVg.vgId, newVg.replica);
+  for (int32_t i = 0; i < newVg.replica; ++i) {
+    mInfo("vgId:%d, vnode:%d dnode:%d", newVg.vgId, i, newVg.vnodeGid[i].dnodeId);
   }
 
   mInfo("vgId:%d, will add 1 vnodes", pVgroup->vgId);
-  if (mndAddVnodeToVgroup(pMnode, &newVgroup, pArray) != 0) return -1;
-  if (mndAddCreateVnodeAction(pMnode, pTrans, pDb, &newVgroup, &newVgroup.vnodeGid[1], true) != 0) return -1;
-  if (mndAddAlterVnodeAction(pMnode, pTrans, pDb, &newVgroup, TDMT_VND_ALTER_REPLICA) != 0) return -1;
-  if (mndAddAlterVnodeConfirmAction(pMnode, pTrans, pDb, &newVgroup) != 0) return -1;
+  if (mndAddVnodeToVgroup(pMnode, &newVg, pArray) != 0) return -1;
+  if (mndAddCreateVnodeAction(pMnode, pTrans, pDb, &newVg, &newVg.vnodeGid[1], true) != 0) return -1;
+  if (mndAddAlterVnodeAction(pMnode, pTrans, pDb, &newVg, TDMT_VND_ALTER_REPLICA) != 0) return -1;
+  if (mndAddAlterVnodeConfirmAction(pMnode, pTrans, pDb, &newVg) != 0) return -1;
 
   mInfo("vgId:%d, will remove 1 vnodes", pVgroup->vgId);
-  newVgroup.replica--;
-  SVnodeGid del = newVgroup.vnodeGid[vnIndex];
-  newVgroup.vnodeGid[vnIndex] = newVgroup.vnodeGid[newVgroup.replica];
-  memset(&newVgroup.vnodeGid[newVgroup.replica], 0, sizeof(SVnodeGid));
-  if (mndAddAlterVnodeAction(pMnode, pTrans, pDb, &newVgroup, TDMT_VND_ALTER_REPLICA) != 0) return -1;
-  if (mndAddDropVnodeAction(pMnode, pTrans, pDb, &newVgroup, &del, true) != 0) return -1;
-  if (mndAddAlterVnodeConfirmAction(pMnode, pTrans, pDb, &newVgroup) != 0) return -1;
+  newVg.replica--;
+  SVnodeGid del = newVg.vnodeGid[vnIndex];
+  newVg.vnodeGid[vnIndex] = newVg.vnodeGid[newVg.replica];
+  memset(&newVg.vnodeGid[newVg.replica], 0, sizeof(SVnodeGid));
+  if (mndAddAlterVnodeAction(pMnode, pTrans, pDb, &newVg, TDMT_VND_ALTER_REPLICA) != 0) return -1;
+  if (mndAddDropVnodeAction(pMnode, pTrans, pDb, &newVg, &del, true) != 0) return -1;
+  if (mndAddAlterVnodeConfirmAction(pMnode, pTrans, pDb, &newVg) != 0) return -1;
 
-  mInfo("vgId:%d, vgroup info after move, replica:%d", newVgroup.vgId, newVgroup.replica);
-  for (int32_t i = 0; i < newVgroup.replica; ++i) {
-    mInfo("vgId:%d, vnode:%d dnode:%d", newVgroup.vgId, i, newVgroup.vnodeGid[i].dnodeId);
+  mInfo("vgId:%d, vgroup info after move, replica:%d", newVg.vgId, newVg.replica);
+  for (int32_t i = 0; i < newVg.replica; ++i) {
+    mInfo("vgId:%d, vnode:%d dnode:%d", newVg.vgId, i, newVg.vnodeGid[i].dnodeId);
   }
   return 0;
 }
 
-int32_t mndSetMoveVgroupsInfoToTrans(SMnode *pMnode, STrans *pTrans, int32_t dropDnodeId) {
-  SArray *pArray = mndBuildDnodesArray(pMnode, dropDnodeId);
+int32_t mndSetMoveVgroupsInfoToTrans(SMnode *pMnode, STrans *pTrans, int32_t delDnodeId) {
+  SArray *pArray = mndBuildDnodesArray(pMnode, delDnodeId);
   if (pArray == NULL) return -1;
 
   void *pIter = NULL;
@@ -951,14 +956,14 @@ int32_t mndSetMoveVgroupsInfoToTrans(SMnode *pMnode, STrans *pTrans, int32_t dro
 
     int32_t vnIndex = -1;
     for (int32_t i = 0; i < pVgroup->replica; ++i) {
-      if (pVgroup->vnodeGid[i].dnodeId == dropDnodeId) {
+      if (pVgroup->vnodeGid[i].dnodeId == delDnodeId) {
         vnIndex = i;
         break;
       }
     }
 
     if (vnIndex != -1) {
-      mInfo("vgId:%d, vnode:%d will be removed from dnode:%d", pVgroup->vgId, vnIndex, dropDnodeId);
+      mInfo("vgId:%d, vnode:%d will be removed from dnode:%d", pVgroup->vgId, vnIndex, delDnodeId);
       SDbObj *pDb = mndAcquireDb(pMnode, pVgroup->dbName);
       mndSetMoveVgroupInfoToTrans(pMnode, pTrans, pDb, pVgroup, vnIndex, pArray);
       mndReleaseDb(pMnode, pDb);
@@ -969,4 +974,195 @@ int32_t mndSetMoveVgroupsInfoToTrans(SMnode *pMnode, STrans *pTrans, int32_t dro
 
   taosArrayDestroy(pArray);
   return 0;
-}
\ No newline at end of file
+}
+
+static int32_t mndAddIncVgroupReplicaToTrans(SMnode *pMnode, STrans *pTrans, SDbObj *pDb, SVgObj *pVgroup,
+                                             int32_t newDnodeId) {
+  mDebug("vgId:%d, will add 1 vnode, replica:%d, dnode:%d", pVgroup->vgId, pVgroup->replica, newDnodeId);
+
+  SVnodeGid *pGid = &pVgroup->vnodeGid[pVgroup->replica];
+  pVgroup->replica++;
+  pGid->dnodeId = newDnodeId;
+  pGid->role = TAOS_SYNC_STATE_ERROR;
+
+  if (mndAddCreateVnodeAction(pMnode, pTrans, pDb, pVgroup, pGid, true) != 0) return -1;
+  if (mndAddAlterVnodeAction(pMnode, pTrans, pDb, pVgroup, TDMT_VND_ALTER_REPLICA) != 0) return -1;
+  if (mndAddAlterVnodeConfirmAction(pMnode, pTrans, pDb, pVgroup) != 0) return -1;
+
+  return 0;
+}
+
+static int32_t mndAddDecVgroupReplicaFromTrans(SMnode *pMnode, STrans *pTrans, SDbObj *pDb, SVgObj *pVgroup,
+                                               int32_t delDnodeId) {
+  mDebug("vgId:%d, will remove 1 vnode, replica:%d, dnode:%d", pVgroup->vgId, pVgroup->replica, delDnodeId);
+
+  SVnodeGid *pGid = NULL;
+  SVnodeGid  delGid = {0};
+  for (int32_t i = 0; i < pVgroup->replica; ++i) {
+    if (pVgroup->vnodeGid[i].dnodeId == delDnodeId) {
+      pGid = &pVgroup->vnodeGid[i];
+      break;
+    }
+  }
+
+  if (pGid == NULL) return 0;
+
+  memcpy(&delGid, pGid, sizeof(SVnodeGid));
+  memcpy(pGid, &pVgroup->vnodeGid[pVgroup->replica], sizeof(SVnodeGid));
+  memset(&pVgroup->vnodeGid[pVgroup->replica], 0, sizeof(SVnodeGid));
+  pVgroup->replica--;
+
+  if (mndAddAlterVnodeAction(pMnode, pTrans, pDb, pVgroup, TDMT_VND_ALTER_REPLICA) != 0) return -1;
+  if (mndAddDropVnodeAction(pMnode, pTrans, pDb, pVgroup, &delGid, true) != 0) return -1;
+  if (mndAddAlterVnodeConfirmAction(pMnode, pTrans, pDb, pVgroup) != 0) return -1;
+
+  return 0;
+}
+
+static int32_t mndRedistributeVgroup(SMnode *pMnode, SRpcMsg *pReq, SDbObj *pDb, SVgObj *pVgroup, SDnodeObj *pNew1,
+                                     SDnodeObj *pOld1, SDnodeObj *pNew2, SDnodeObj *pOld2, SDnodeObj *pNew3,
+                                     SDnodeObj *pOld3) {
+  int32_t  code = -1;
+  SSdbRaw *pRaw = NULL;
+  STrans  *pTrans = NULL;
+
+  pTrans = mndTransCreate(pMnode, TRN_POLICY_RETRY, TRN_CONFLICT_GLOBAL, pReq);
+  if (pTrans == NULL) goto _OVER;
+  mndTransSetSerial(pTrans);
+  mDebug("trans:%d, used to drop redistribute vgId:%d", pTrans->id, pVgroup->vgId);
+
+  SVgObj newVg = {0};
+  memcpy(&newVg, pVgroup, sizeof(SVgObj));
+  mInfo("vgId:%d, vgroup info before redistribute, replica:%d", newVg.vgId, newVg.replica);
+  for (int32_t i = 0; i < newVg.replica; ++i) {
+    mInfo("vgId:%d, vnode:%d dnode:%d", newVg.vgId, i, newVg.vnodeGid[i].dnodeId);
+  }
+
+  if (mndAddIncVgroupReplicaToTrans(pMnode, pTrans, pDb, &newVg, pNew1->id) != 0) goto _OVER;
+  if (mndAddDecVgroupReplicaFromTrans(pMnode, pTrans, pDb, &newVg, pOld1->id) != 0) goto _OVER;
+  if (pNew2 != NULL) {
+    if (mndAddIncVgroupReplicaToTrans(pMnode, pTrans, pDb, &newVg, pNew2->id) != 0) goto _OVER;
+    if (mndAddDecVgroupReplicaFromTrans(pMnode, pTrans, pDb, &newVg, pOld2->id) != 0) goto _OVER;
+    if (mndAddIncVgroupReplicaToTrans(pMnode, pTrans, pDb, &newVg, pNew3->id) != 0) goto _OVER;
+    if (mndAddDecVgroupReplicaFromTrans(pMnode, pTrans, pDb, &newVg, pOld3->id) != 0) goto _OVER;
+  }
+
+  pRaw = mndVgroupActionEncode(&newVg);
+  if (pRaw == NULL || mndTransAppendCommitlog(pTrans, pRaw) != 0) goto _OVER;
+  sdbSetRawStatus(pRaw, SDB_STATUS_READY);
+  pRaw = NULL;
+
+  mInfo("vgId:%d, vgroup info after redistribute, replica:%d", newVg.vgId, newVg.replica);
+  for (int32_t i = 0; i < newVg.replica; ++i) {
+    mInfo("vgId:%d, vnode:%d dnode:%d", newVg.vgId, i, newVg.vnodeGid[i].dnodeId);
+  }
+
+  if (mndTransPrepare(pMnode, pTrans) != 0) goto _OVER;
+  code = 0;
+
+_OVER:
+  mndTransDrop(pTrans);
+  sdbFreeRaw(pRaw);
+  mndReleaseDb(pMnode, pDb);
+  return code;
+}
+
+static int32_t mndProcessRedistributeVgroupMsg(SRpcMsg *pReq) {
+  SMnode        *pMnode = pReq->info.node;
+  SUserObj      *pUser = NULL;
+  SDnodeObj     *pNew1 = NULL;
+  SDnodeObj     *pNew2 = NULL;
+  SDnodeObj     *pNew3 = NULL;
+  SDnodeObj     *pOld1 = NULL;
+  SDnodeObj     *pOld2 = NULL;
+  SDnodeObj     *pOld3 = NULL;
+  SVgObj        *pVgroup = NULL;
+  SDbObj        *pDb = NULL;
+  int32_t        code = -1;
+  int64_t        curMs = taosGetTimestampMs();
+  SMDropMnodeReq redReq = {0};
+
+#if 0
+  if (tDeserializeSCreateDropMQSBNodeReq(pReq->pCont, pReq->contLen, &dropReq) != 0) {
+    terrno = TSDB_CODE_INVALID_MSG;
+    goto _OVER;
+  }
+#endif
+
+  mDebug("vgId:%d, start to redistribute", 2);
+  pUser = mndAcquireUser(pMnode, pReq->conn.user);
+  if (pUser == NULL) {
+    terrno = TSDB_CODE_MND_NO_USER_FROM_CONN;
+    goto _OVER;
+  }
+
+  if (mndCheckNodeAuth(pUser) != 0) {
+    goto _OVER;
+  }
+
+  pVgroup = mndAcquireVgroup(pMnode, 2);
+  if (pVgroup == NULL) goto _OVER;
+
+  pDb = mndAcquireDb(pMnode, pVgroup->dbName);
+  if (pDb == NULL) goto _OVER;
+
+  if (pVgroup->replica == 1) {
+    pNew1 = mndAcquireDnode(pMnode, 1);
+    pOld1 = mndAcquireDnode(pMnode, pVgroup->vnodeGid[0].dnodeId);
+    if (pNew1 == NULL || pOld1 == NULL) goto _OVER;
+    if (!mndIsDnodeOnline(pNew1, curMs) || !mndIsDnodeOnline(pOld1, curMs)) {
+      terrno = TSDB_CODE_NODE_OFFLINE;
+      goto _OVER;
+    }
+    if (pNew1 == pOld1) {
+      terrno = TSDB_CODE_MND_VGROUP_UN_CHANGED;
+      goto _OVER;
+    }
+    if (mndRedistributeVgroup(pMnode, pReq, pDb, pVgroup, pNew1, pOld1, NULL, NULL, NULL, NULL) != 0) goto _OVER;
+  }
+
+  if (pVgroup->replica == 3) {
+    pNew1 = mndAcquireDnode(pMnode, 1);
+    pNew2 = mndAcquireDnode(pMnode, 2);
+    pNew3 = mndAcquireDnode(pMnode, 3);
+    pOld1 = mndAcquireDnode(pMnode, pVgroup->vnodeGid[0].dnodeId);
+    pOld2 = mndAcquireDnode(pMnode, pVgroup->vnodeGid[1].dnodeId);
+    pOld3 = mndAcquireDnode(pMnode, pVgroup->vnodeGid[2].dnodeId);
+    if (pNew1 == NULL || pOld1 == NULL || pNew2 == NULL || pOld2 == NULL || pNew3 == NULL || pOld3 == NULL) goto _OVER;
+    if (!mndIsDnodeOnline(pNew1, curMs) || !mndIsDnodeOnline(pOld1, curMs) || !mndIsDnodeOnline(pNew2, curMs) ||
+        !mndIsDnodeOnline(pOld2, curMs) || !mndIsDnodeOnline(pNew3, curMs) || !mndIsDnodeOnline(pOld3, curMs)) {
+      terrno = TSDB_CODE_NODE_OFFLINE;
+      goto _OVER;
+    }
+    bool changed = true;
+    if (pNew1 != pOld1 || pNew1 != pOld2 || pNew1 != pOld3) changed = true;
+    if (pNew2 != pOld1 || pNew2 != pOld2 || pNew2 != pOld3) changed = true;
+    if (pNew3 != pOld1 || pNew3 != pOld2 || pNew3 != pOld3) changed = true;
+    if (!changed) {
+      terrno = TSDB_CODE_MND_VGROUP_UN_CHANGED;
+      goto _OVER;
+    }
+    if (mndRedistributeVgroup(pMnode, pReq, pDb, pVgroup, pNew1, pOld1, pNew2, pOld2, pNew3, pOld3) != 0) goto _OVER;
+  }
+
+  if (code == 0) code = TSDB_CODE_ACTION_IN_PROGRESS;
+
+_OVER:
+  if (code != 0 && code != TSDB_CODE_ACTION_IN_PROGRESS) {
+    mDebug("vgId:%d, failed to redistribute since %s", 1, terrstr());
+  }
+
+  mndReleaseDnode(pMnode, pNew1);
+  mndReleaseDnode(pMnode, pNew2);
+  mndReleaseDnode(pMnode, pNew3);
+  mndReleaseDnode(pMnode, pOld1);
+  mndReleaseDnode(pMnode, pOld2);
+  mndReleaseDnode(pMnode, pOld3);
+  mndReleaseUser(pMnode, pUser);
+  mndReleaseVgroup(pMnode, pVgroup);
+  mndReleaseDb(pMnode, pDb);
+
+  return code;
+}
+
+static int32_t mndProcessSplitVgroupMsg(SRpcMsg *pReq) { return 0; }
\ No newline at end of file

From 268236d804ffa2d1e8ce4a873f6dea3ac1903a7f Mon Sep 17 00:00:00 2001
From: afwerar <1296468573@qq.com>
Date: Mon, 6 Jun 2022 17:28:45 +0800
Subject: [PATCH 24/34] fix(os): do not kill taosd when case init dnodes

---
 tests/pytest/util/dnodes.py | 36 +-----------------------------------
 1 file changed, 1 insertion(+), 35 deletions(-)

diff --git a/tests/pytest/util/dnodes.py b/tests/pytest/util/dnodes.py
index b6ffda5c86..fb0c1881c9 100644
--- a/tests/pytest/util/dnodes.py
+++ b/tests/pytest/util/dnodes.py
@@ -388,7 +388,7 @@ class TDDnode:
             if os.system(cmd) != 0:
                 tdLog.exit(cmd)
         else:
-            self.remoteExec(self.cfgDict, "tdDnodes.deploy(%d,updateCfgDict)\ntdDnodes.startWithoutSleep(%d)"%(self.index, self.index))
+            self.remoteExec(self.cfgDict, "tdDnodes.dnodes[%d].deployed=1\ntdDnodes.dnodes[%d].logDir=\"%%s/sim/dnode%%d/log\"%%(tdDnodes.dnodes[%d].path,%d)\ntdDnodes.dnodes[%d].cfgDir=\"%%s/sim/dnode%%d/cfg\"%%(tdDnodes.dnodes[%d].path,%d)\ntdDnodes.startWithoutSleep(%d)"%(self.index-1,self.index-1,self.index-1,self.index,self.index-1,self.index-1,self.index,self.index))
 
         self.running = 1
         tdLog.debug("dnode:%d is running with %s " % (self.index, cmd))
@@ -497,46 +497,12 @@ class TDDnodes:
         self.killValgrind = 1
 
     def init(self, path, remoteIP = ""):
-        psCmd = "ps -ef|grep -w taosd| grep -v grep| grep -v defunct | awk '{print $2}'"
-        processID = subprocess.check_output(psCmd, shell=True).decode("utf-8")
-        while(processID):
-            killCmd = "kill -9 %s > /dev/null 2>&1" % processID
-            os.system(killCmd)
-            time.sleep(1)
-            processID = subprocess.check_output(
-                psCmd, shell=True).decode("utf-8")
-
-        if self.killValgrind == 1:
-            psCmd = "ps -ef|grep -w valgrind.bin| grep -v grep | awk '{print $2}'"
-            processID = subprocess.check_output(psCmd, shell=True).decode("utf-8")
-            while(processID):
-                killCmd = "kill -9 %s > /dev/null 2>&1" % processID
-                os.system(killCmd)
-                time.sleep(1)
-                processID = subprocess.check_output(
-                    psCmd, shell=True).decode("utf-8")
-
         binPath = self.dnodes[0].getPath() + "/../../../"
         # tdLog.debug("binPath %s" % (binPath))
         binPath = os.path.realpath(binPath)
         # tdLog.debug("binPath real path %s" % (binPath))
 
-        # cmd = "sudo cp %s/build/lib/libtaos.so /usr/local/lib/taos/" % (binPath)
-        # tdLog.debug(cmd)
-        # os.system(cmd)
-
-        # cmd = "sudo cp %s/build/bin/taos /usr/local/bin/taos/" % (binPath)
-        # if os.system(cmd) != 0 :
-        #  tdLog.exit(cmd)
-        # tdLog.debug("execute %s" % (cmd))
-
-        # cmd = "sudo cp %s/build/bin/taosd /usr/local/bin/taos/" % (binPath)
-        # if os.system(cmd) != 0 :
-        # tdLog.exit(cmd)
-        # tdLog.debug("execute %s" % (cmd))
-
         if path == "":
-            # self.path = os.path.expanduser('~')
             self.path = os.path.abspath(binPath + "../../")
         else:
             self.path = os.path.realpath(path)

From 41d9c89b8e019dec3aeeb3a09bc48c5207ba79ba Mon Sep 17 00:00:00 2001
From: Shengliang Guan <slguan@taosdata.com>
Date: Mon, 6 Jun 2022 17:50:18 +0800
Subject: [PATCH 25/34] fix: cannot modify columns used by topic

---
 source/dnode/mnode/impl/src/mndStb.c   | 4 ++--
 source/dnode/mnode/impl/src/mndTopic.c | 3 ++-
 2 files changed, 4 insertions(+), 3 deletions(-)

diff --git a/source/dnode/mnode/impl/src/mndStb.c b/source/dnode/mnode/impl/src/mndStb.c
index 388172f6ae..9981dc8530 100644
--- a/source/dnode/mnode/impl/src/mndStb.c
+++ b/source/dnode/mnode/impl/src/mndStb.c
@@ -1108,7 +1108,7 @@ static int32_t mndDropSuperTableColumn(SMnode *pMnode, const SStbObj *pOld, SStb
     return -1;
   }
 
-  col_id_t colId = pOld->pTags[col].colId;
+  col_id_t colId = pOld->pColumns[col].colId;
   if (mndCheckColAndTagModifiable(pMnode, pOld->uid, colId) != 0) {
     return -1;
   }
@@ -1142,7 +1142,7 @@ static int32_t mndAlterStbColumnBytes(SMnode *pMnode, const SStbObj *pOld, SStbO
     return -1;
   }
 
-  col_id_t colId = pOld->pTags[col].colId;
+  col_id_t colId = pOld->pColumns[col].colId;
   if (mndCheckColAndTagModifiable(pMnode, pOld->uid, colId) != 0) {
     return -1;
   }
diff --git a/source/dnode/mnode/impl/src/mndTopic.c b/source/dnode/mnode/impl/src/mndTopic.c
index a810a6a487..605e446e2e 100644
--- a/source/dnode/mnode/impl/src/mndTopic.c
+++ b/source/dnode/mnode/impl/src/mndTopic.c
@@ -91,7 +91,7 @@ int32_t mndCheckColAndTagModifiable(SMnode *pMnode, int64_t suid, col_id_t colId
     }
 
     SHashObj  *pColHash = NULL;
-    SNodeList *pNodeList;
+    SNodeList *pNodeList = NULL;
     nodesCollectColumns((SSelectStmt *)pAst, SQL_CLAUSE_FROM, NULL, COLLECT_COL_TYPE_ALL, &pNodeList);
     SNode *pNode = NULL;
     FOREACH(pNode, pNodeList) {
@@ -103,6 +103,7 @@ int32_t mndCheckColAndTagModifiable(SMnode *pMnode, int64_t suid, col_id_t colId
       if (pCol->colId > 0) {
         taosHashPut(pColHash, &pCol->colId, sizeof(int16_t), NULL, 0);
       }
+      mTrace("topic:%s, colId:%d is used", pTopic->name, pCol->colId);
     }
 
     if (taosHashGet(pColHash, &colId, sizeof(int16_t)) != NULL) {

From f77c72c25a5500250830f24ff81a688ba654ee67 Mon Sep 17 00:00:00 2001
From: Xiaoyu Wang <xiaoyuwang@taosdata.com>
Date: Mon, 6 Jun 2022 17:51:25 +0800
Subject: [PATCH 26/34] feat: order by distributed split

---
 include/libs/nodes/nodes.h                   |   2 +-
 include/libs/nodes/plannodes.h               |   8 +-
 source/dnode/mnode/impl/src/mndTopic.c       |   2 +-
 source/libs/nodes/src/nodesCloneFuncs.c      |   2 +
 source/libs/nodes/src/nodesCodeFuncs.c       |  18 ++
 source/libs/nodes/src/nodesUtilFuncs.c       |   4 +-
 source/libs/parser/test/parInitialDTest.cpp  |   2 +-
 source/libs/planner/src/planLogicCreater.c   |   2 +-
 source/libs/planner/src/planPhysiCreater.c   |  24 ++-
 source/libs/planner/src/planSpliter.c        | 190 +++++++++++++------
 source/libs/planner/test/planOrderByTest.cpp |   3 +
 11 files changed, 186 insertions(+), 71 deletions(-)

diff --git a/include/libs/nodes/nodes.h b/include/libs/nodes/nodes.h
index 29fabb5418..6f52f8dab0 100644
--- a/include/libs/nodes/nodes.h
+++ b/include/libs/nodes/nodes.h
@@ -211,7 +211,7 @@ typedef enum ENodeType {
   QUERY_NODE_PHYSICAL_PLAN_MERGE,
   QUERY_NODE_PHYSICAL_PLAN_SORT,
   QUERY_NODE_PHYSICAL_PLAN_HASH_INTERVAL,
-  QUERY_NODE_PHYSICAL_PLAN_SORT_MERGE_INTERVAL,
+  QUERY_NODE_PHYSICAL_PLAN_MERGE_INTERVAL,
   QUERY_NODE_PHYSICAL_PLAN_STREAM_INTERVAL,
   QUERY_NODE_PHYSICAL_PLAN_STREAM_FINAL_INTERVAL,
   QUERY_NODE_PHYSICAL_PLAN_STREAM_SEMI_INTERVAL,
diff --git a/include/libs/nodes/plannodes.h b/include/libs/nodes/plannodes.h
index 38c3f150ef..d646effa58 100644
--- a/include/libs/nodes/plannodes.h
+++ b/include/libs/nodes/plannodes.h
@@ -42,6 +42,7 @@ typedef struct SScanLogicNode {
   SNodeList*    pScanPseudoCols;
   int8_t        tableType;
   uint64_t      tableId;
+  uint64_t      stableId;
   SVgroupsInfo* pVgroupList;
   EScanType     scanType;
   uint8_t       scanSeq[2];  // first is scan count, and second is reverse scan count
@@ -109,6 +110,7 @@ typedef struct SExchangeLogicNode {
 typedef struct SMergeLogicNode {
   SLogicNode node;
   SNodeList* pMergeKeys;
+  SNodeList* pInputs;
   int32_t    numOfChannels;
   int32_t    srcGroupId;
 } SMergeLogicNode;
@@ -117,7 +119,7 @@ typedef enum EWindowType { WINDOW_TYPE_INTERVAL = 1, WINDOW_TYPE_SESSION, WINDOW
 
 typedef enum EIntervalAlgorithm {
   INTERVAL_ALGO_HASH = 1,
-  INTERVAL_ALGO_SORT_MERGE,
+  INTERVAL_ALGO_MERGE,
   INTERVAL_ALGO_STREAM_FINAL,
   INTERVAL_ALGO_STREAM_SEMI,
   INTERVAL_ALGO_STREAM_SINGLE,
@@ -220,6 +222,7 @@ typedef struct SScanPhysiNode {
   SNodeList* pScanCols;
   SNodeList* pScanPseudoCols;
   uint64_t   uid;  // unique id of the table
+  uint64_t   suid;
   int8_t     tableType;
   SName      tableName;
 } SScanPhysiNode;
@@ -296,6 +299,7 @@ typedef struct SExchangePhysiNode {
 typedef struct SMergePhysiNode {
   SPhysiNode node;
   SNodeList* pMergeKeys;
+  SNodeList* pTargets;
   int32_t    numOfChannels;
   int32_t    srcGroupId;
 } SMergePhysiNode;
@@ -319,7 +323,7 @@ typedef struct SIntervalPhysiNode {
   int8_t           slidingUnit;
 } SIntervalPhysiNode;
 
-typedef SIntervalPhysiNode SSortMergeIntervalPhysiNode;
+typedef SIntervalPhysiNode SMergeIntervalPhysiNode;
 typedef SIntervalPhysiNode SStreamIntervalPhysiNode;
 typedef SIntervalPhysiNode SStreamFinalIntervalPhysiNode;
 typedef SIntervalPhysiNode SStreamSemiIntervalPhysiNode;
diff --git a/source/dnode/mnode/impl/src/mndTopic.c b/source/dnode/mnode/impl/src/mndTopic.c
index a810a6a487..060bd94889 100644
--- a/source/dnode/mnode/impl/src/mndTopic.c
+++ b/source/dnode/mnode/impl/src/mndTopic.c
@@ -91,7 +91,7 @@ int32_t mndCheckColAndTagModifiable(SMnode *pMnode, int64_t suid, col_id_t colId
     }
 
     SHashObj  *pColHash = NULL;
-    SNodeList *pNodeList;
+    SNodeList *pNodeList = NULL;
     nodesCollectColumns((SSelectStmt *)pAst, SQL_CLAUSE_FROM, NULL, COLLECT_COL_TYPE_ALL, &pNodeList);
     SNode *pNode = NULL;
     FOREACH(pNode, pNodeList) {
diff --git a/source/libs/nodes/src/nodesCloneFuncs.c b/source/libs/nodes/src/nodesCloneFuncs.c
index f277d30eb1..657ffd32a5 100644
--- a/source/libs/nodes/src/nodesCloneFuncs.c
+++ b/source/libs/nodes/src/nodesCloneFuncs.c
@@ -318,6 +318,7 @@ static SNode* logicScanCopy(const SScanLogicNode* pSrc, SScanLogicNode* pDst) {
   CLONE_NODE_LIST_FIELD(pScanPseudoCols);
   COPY_SCALAR_FIELD(tableType);
   COPY_SCALAR_FIELD(tableId);
+  COPY_SCALAR_FIELD(stableId);
   CLONE_OBJECT_FIELD(pVgroupList, vgroupsInfoClone);
   COPY_SCALAR_FIELD(scanType);
   COPY_OBJECT_FIELD(scanSeq[0], sizeof(uint8_t) * 2);
@@ -387,6 +388,7 @@ static SNode* logicExchangeCopy(const SExchangeLogicNode* pSrc, SExchangeLogicNo
 static SNode* logicMergeCopy(const SMergeLogicNode* pSrc, SMergeLogicNode* pDst) {
   COPY_BASE_OBJECT_FIELD(node, logicNodeCopy);
   CLONE_NODE_LIST_FIELD(pMergeKeys);
+  CLONE_NODE_LIST_FIELD(pInputs);
   COPY_SCALAR_FIELD(numOfChannels);
   COPY_SCALAR_FIELD(srcGroupId);
   return (SNode*)pDst;
diff --git a/source/libs/nodes/src/nodesCodeFuncs.c b/source/libs/nodes/src/nodesCodeFuncs.c
index c54289bc56..246ee2bfed 100644
--- a/source/libs/nodes/src/nodesCodeFuncs.c
+++ b/source/libs/nodes/src/nodesCodeFuncs.c
@@ -230,6 +230,8 @@ const char* nodesNodeName(ENodeType type) {
       return "PhysiSort";
     case QUERY_NODE_PHYSICAL_PLAN_HASH_INTERVAL:
       return "PhysiHashInterval";
+    case QUERY_NODE_PHYSICAL_PLAN_MERGE_INTERVAL:
+      return "PhysiMergeInterval";
     case QUERY_NODE_PHYSICAL_PLAN_STREAM_INTERVAL:
       return "PhysiStreamInterval";
     case QUERY_NODE_PHYSICAL_PLAN_STREAM_FINAL_INTERVAL:
@@ -1249,6 +1251,7 @@ static int32_t jsonToName(const SJson* pJson, void* pObj) {
 static const char* jkScanPhysiPlanScanCols = "ScanCols";
 static const char* jkScanPhysiPlanScanPseudoCols = "ScanPseudoCols";
 static const char* jkScanPhysiPlanTableId = "TableId";
+static const char* jkScanPhysiPlanSTableId = "STableId";
 static const char* jkScanPhysiPlanTableType = "TableType";
 static const char* jkScanPhysiPlanTableName = "TableName";
 
@@ -1265,6 +1268,9 @@ static int32_t physiScanNodeToJson(const void* pObj, SJson* pJson) {
   if (TSDB_CODE_SUCCESS == code) {
     code = tjsonAddIntegerToObject(pJson, jkScanPhysiPlanTableId, pNode->uid);
   }
+  if (TSDB_CODE_SUCCESS == code) {
+    code = tjsonAddIntegerToObject(pJson, jkScanPhysiPlanSTableId, pNode->suid);
+  }
   if (TSDB_CODE_SUCCESS == code) {
     code = tjsonAddIntegerToObject(pJson, jkScanPhysiPlanTableType, pNode->tableType);
   }
@@ -1288,6 +1294,9 @@ static int32_t jsonToPhysiScanNode(const SJson* pJson, void* pObj) {
   if (TSDB_CODE_SUCCESS == code) {
     code = tjsonGetUBigIntValue(pJson, jkScanPhysiPlanTableId, &pNode->uid);
   }
+  if (TSDB_CODE_SUCCESS == code) {
+    code = tjsonGetUBigIntValue(pJson, jkScanPhysiPlanSTableId, &pNode->suid);
+  }
   if (TSDB_CODE_SUCCESS == code) {
     code = tjsonGetTinyIntValue(pJson, jkScanPhysiPlanTableType, &pNode->tableType);
   }
@@ -1644,6 +1653,7 @@ static int32_t jsonToPhysiExchangeNode(const SJson* pJson, void* pObj) {
 }
 
 static const char* jkMergePhysiPlanMergeKeys = "MergeKeys";
+static const char* jkMergePhysiPlanTargets = "Targets";
 static const char* jkMergePhysiPlanNumOfChannels = "NumOfChannels";
 static const char* jkMergePhysiPlanSrcGroupId = "SrcGroupId";
 
@@ -1654,6 +1664,9 @@ static int32_t physiMergeNodeToJson(const void* pObj, SJson* pJson) {
   if (TSDB_CODE_SUCCESS == code) {
     code = nodeListToJson(pJson, jkMergePhysiPlanMergeKeys, pNode->pMergeKeys);
   }
+  if (TSDB_CODE_SUCCESS == code) {
+    code = nodeListToJson(pJson, jkMergePhysiPlanTargets, pNode->pTargets);
+  }
   if (TSDB_CODE_SUCCESS == code) {
     code = tjsonAddIntegerToObject(pJson, jkMergePhysiPlanNumOfChannels, pNode->numOfChannels);
   }
@@ -1671,6 +1684,9 @@ static int32_t jsonToPhysiMergeNode(const SJson* pJson, void* pObj) {
   if (TSDB_CODE_SUCCESS == code) {
     code = jsonToNodeList(pJson, jkMergePhysiPlanMergeKeys, &pNode->pMergeKeys);
   }
+  if (TSDB_CODE_SUCCESS == code) {
+    code = jsonToNodeList(pJson, jkMergePhysiPlanTargets, &pNode->pTargets);
+  }
   if (TSDB_CODE_SUCCESS == code) {
     code = tjsonGetIntValue(pJson, jkMergePhysiPlanNumOfChannels, &pNode->numOfChannels);
   }
@@ -3802,6 +3818,7 @@ static int32_t specificNodeToJson(const void* pObj, SJson* pJson) {
     case QUERY_NODE_PHYSICAL_PLAN_SORT:
       return physiSortNodeToJson(pObj, pJson);
     case QUERY_NODE_PHYSICAL_PLAN_HASH_INTERVAL:
+    case QUERY_NODE_PHYSICAL_PLAN_MERGE_INTERVAL:
     case QUERY_NODE_PHYSICAL_PLAN_STREAM_INTERVAL:
     case QUERY_NODE_PHYSICAL_PLAN_STREAM_FINAL_INTERVAL:
     case QUERY_NODE_PHYSICAL_PLAN_STREAM_SEMI_INTERVAL:
@@ -3930,6 +3947,7 @@ static int32_t jsonToSpecificNode(const SJson* pJson, void* pObj) {
     case QUERY_NODE_PHYSICAL_PLAN_SORT:
       return jsonToPhysiSortNode(pJson, pObj);
     case QUERY_NODE_PHYSICAL_PLAN_HASH_INTERVAL:
+    case QUERY_NODE_PHYSICAL_PLAN_MERGE_INTERVAL:
     case QUERY_NODE_PHYSICAL_PLAN_STREAM_INTERVAL:
     case QUERY_NODE_PHYSICAL_PLAN_STREAM_FINAL_INTERVAL:
     case QUERY_NODE_PHYSICAL_PLAN_STREAM_SEMI_INTERVAL:
diff --git a/source/libs/nodes/src/nodesUtilFuncs.c b/source/libs/nodes/src/nodesUtilFuncs.c
index 16ce4d7a48..0aec211e4c 100644
--- a/source/libs/nodes/src/nodesUtilFuncs.c
+++ b/source/libs/nodes/src/nodesUtilFuncs.c
@@ -260,8 +260,8 @@ SNodeptr nodesMakeNode(ENodeType type) {
       return makeNode(type, sizeof(SSortPhysiNode));
     case QUERY_NODE_PHYSICAL_PLAN_HASH_INTERVAL:
       return makeNode(type, sizeof(SIntervalPhysiNode));
-    case QUERY_NODE_PHYSICAL_PLAN_SORT_MERGE_INTERVAL:
-      return makeNode(type, sizeof(SSortMergeIntervalPhysiNode));
+    case QUERY_NODE_PHYSICAL_PLAN_MERGE_INTERVAL:
+      return makeNode(type, sizeof(SMergeIntervalPhysiNode));
     case QUERY_NODE_PHYSICAL_PLAN_STREAM_INTERVAL:
       return makeNode(type, sizeof(SStreamIntervalPhysiNode));
     case QUERY_NODE_PHYSICAL_PLAN_STREAM_FINAL_INTERVAL:
diff --git a/source/libs/parser/test/parInitialDTest.cpp b/source/libs/parser/test/parInitialDTest.cpp
index d4f9a540ae..528d465dbd 100644
--- a/source/libs/parser/test/parInitialDTest.cpp
+++ b/source/libs/parser/test/parInitialDTest.cpp
@@ -51,7 +51,7 @@ TEST_F(ParserInitialDTest, dropBnode) {
 }
 
 // DROP CONSUMER GROUP [ IF EXISTS ] cgroup_name ON topic_name
-TEST_F(ParserInitialDTest, dropCGroup) {
+TEST_F(ParserInitialDTest, dropConsumerGroup) {
   useDb("root", "test");
 
   SMDropCgroupReq expect = {0};
diff --git a/source/libs/planner/src/planLogicCreater.c b/source/libs/planner/src/planLogicCreater.c
index c3aad632c7..d37e22b316 100644
--- a/source/libs/planner/src/planLogicCreater.c
+++ b/source/libs/planner/src/planLogicCreater.c
@@ -219,9 +219,9 @@ static int32_t makeScanLogicNode(SLogicPlanContext* pCxt, SRealTableNode* pRealT
     return TSDB_CODE_OUT_OF_MEMORY;
   }
 
-  // TSWAP(pScan->pMeta, pRealTable->pMeta);
   TSWAP(pScan->pVgroupList, pRealTable->pVgroupList);
   pScan->tableId = pRealTable->pMeta->uid;
+  pScan->stableId = pRealTable->pMeta->suid;
   pScan->tableType = pRealTable->pMeta->tableType;
   pScan->scanSeq[0] = hasRepeatScanFuncs ? 2 : 1;
   pScan->scanSeq[1] = 0;
diff --git a/source/libs/planner/src/planPhysiCreater.c b/source/libs/planner/src/planPhysiCreater.c
index 3c000e3ff7..b2a675dbb1 100644
--- a/source/libs/planner/src/planPhysiCreater.c
+++ b/source/libs/planner/src/planPhysiCreater.c
@@ -425,6 +425,7 @@ static int32_t createScanPhysiNodeFinalize(SPhysiPlanContext* pCxt, SSubplan* pS
 
   if (TSDB_CODE_SUCCESS == code) {
     pScanPhysiNode->uid = pScanLogicNode->tableId;
+    pScanPhysiNode->suid = pScanLogicNode->stableId;
     pScanPhysiNode->tableType = pScanLogicNode->tableType;
     memcpy(&pScanPhysiNode->tableName, &pScanLogicNode->tableName, sizeof(SName));
     if (NULL != pScanLogicNode->pTagCond) {
@@ -923,8 +924,8 @@ static ENodeType getIntervalOperatorType(EIntervalAlgorithm intervalAlgo) {
   switch (intervalAlgo) {
     case INTERVAL_ALGO_HASH:
       return QUERY_NODE_PHYSICAL_PLAN_HASH_INTERVAL;
-    case INTERVAL_ALGO_SORT_MERGE:
-      return QUERY_NODE_PHYSICAL_PLAN_SORT_MERGE_INTERVAL;
+    case INTERVAL_ALGO_MERGE:
+      return QUERY_NODE_PHYSICAL_PLAN_MERGE_INTERVAL;
     case INTERVAL_ALGO_STREAM_FINAL:
       return QUERY_NODE_PHYSICAL_PLAN_STREAM_FINAL_INTERVAL;
     case INTERVAL_ALGO_STREAM_SEMI:
@@ -1155,6 +1156,8 @@ static int32_t createExchangePhysiNodeByMerge(SMergePhysiNode* pMerge) {
     nodesDestroyNode(pExchange);
     return TSDB_CODE_OUT_OF_MEMORY;
   }
+  SNode* pSlot = NULL;
+  FOREACH(pSlot, pExchange->node.pOutputDataBlockDesc->pSlots) { ((SSlotDescNode*)pSlot)->output = true; }
   return nodesListMakeStrictAppend(&pMerge->node.pChildren, pExchange);
 }
 
@@ -1168,12 +1171,14 @@ static int32_t createMergePhysiNode(SPhysiPlanContext* pCxt, SMergeLogicNode* pM
   pMerge->numOfChannels = pMergeLogicNode->numOfChannels;
   pMerge->srcGroupId = pMergeLogicNode->srcGroupId;
 
-  int32_t code = TSDB_CODE_SUCCESS;
+  int32_t code = addDataBlockSlots(pCxt, pMergeLogicNode->pInputs, pMerge->node.pOutputDataBlockDesc);
 
-  for (int32_t i = 0; i < pMerge->numOfChannels; ++i) {
-    code = createExchangePhysiNodeByMerge(pMerge);
-    if (TSDB_CODE_SUCCESS != code) {
-      break;
+  if (TSDB_CODE_SUCCESS == code) {
+    for (int32_t i = 0; i < pMerge->numOfChannels; ++i) {
+      code = createExchangePhysiNodeByMerge(pMerge);
+      if (TSDB_CODE_SUCCESS != code) {
+        break;
+      }
     }
   }
 
@@ -1182,6 +1187,11 @@ static int32_t createMergePhysiNode(SPhysiPlanContext* pCxt, SMergeLogicNode* pM
                          &pMerge->pMergeKeys);
   }
 
+  if (TSDB_CODE_SUCCESS == code) {
+    code = setListSlotId(pCxt, pMerge->node.pOutputDataBlockDesc->dataBlockId, -1, pMergeLogicNode->node.pTargets,
+                         &pMerge->pTargets);
+  }
+
   if (TSDB_CODE_SUCCESS == code) {
     *pPhyNode = (SPhysiNode*)pMerge;
   } else {
diff --git a/source/libs/planner/src/planSpliter.c b/source/libs/planner/src/planSpliter.c
index 9e6c25ce78..7be33d54e3 100644
--- a/source/libs/planner/src/planSpliter.c
+++ b/source/libs/planner/src/planSpliter.c
@@ -80,30 +80,36 @@ static int32_t splCreateExchangeNode(SSplitContext* pCxt, SLogicNode* pChild, SE
   return TSDB_CODE_SUCCESS;
 }
 
-static int32_t splCreateExchangeNodeForSubplan(SSplitContext* pCxt, SLogicSubplan* pSubplan, SLogicNode* pSplitNode,
-                                               ESubplanType subplanType) {
-  SExchangeLogicNode* pExchange = NULL;
-  if (TSDB_CODE_SUCCESS != splCreateExchangeNode(pCxt, pSplitNode, &pExchange)) {
-    return TSDB_CODE_OUT_OF_MEMORY;
-  }
-
-  pSubplan->subplanType = subplanType;
-
-  if (NULL == pSplitNode->pParent) {
-    pSubplan->pNode = (SLogicNode*)pExchange;
+static int32_t splReplaceLogicNode(SLogicSubplan* pSubplan, SLogicNode* pOld, SLogicNode* pNew) {
+  if (NULL == pOld->pParent) {
+    pSubplan->pNode = (SLogicNode*)pNew;
     return TSDB_CODE_SUCCESS;
   }
 
   SNode* pNode;
-  FOREACH(pNode, pSplitNode->pParent->pChildren) {
-    if (nodesEqualNode(pNode, pSplitNode)) {
-      REPLACE_NODE(pExchange);
-      pExchange->node.pParent = pSplitNode->pParent;
+  FOREACH(pNode, pOld->pParent->pChildren) {
+    if (nodesEqualNode(pNode, pOld)) {
+      REPLACE_NODE(pNew);
+      pNew->pParent = pOld->pParent;
       return TSDB_CODE_SUCCESS;
     }
   }
-  nodesDestroyNode(pExchange);
-  return TSDB_CODE_FAILED;
+  return TSDB_CODE_PLAN_INTERNAL_ERROR;
+}
+
+static int32_t splCreateExchangeNodeForSubplan(SSplitContext* pCxt, SLogicSubplan* pSubplan, SLogicNode* pSplitNode,
+                                               ESubplanType subplanType) {
+  SExchangeLogicNode* pExchange = NULL;
+  int32_t             code = splCreateExchangeNode(pCxt, pSplitNode, &pExchange);
+  if (TSDB_CODE_SUCCESS == code) {
+    code = splReplaceLogicNode(pSubplan, pSplitNode, (SLogicNode*)pExchange);
+  }
+  if (TSDB_CODE_SUCCESS == code) {
+    pSubplan->subplanType = subplanType;
+  } else {
+    nodesDestroyNode(pExchange);
+  }
+  return code;
 }
 
 static bool splMatch(SSplitContext* pCxt, SLogicSubplan* pSubplan, int32_t flag, FSplFindSplitNode func, void* pInfo) {
@@ -295,24 +301,34 @@ static int32_t stbSplCreatePartWindowNode(SWindowLogicNode* pMergeWindow, SLogic
   return code;
 }
 
-static int32_t stbSplCreateMergeNode(SSplitContext* pCxt, SLogicNode* pParent, SNodeList* pMergeKeys,
-                                     SLogicNode* pPartChild) {
+static int32_t stbSplCreateMergeNode(SSplitContext* pCxt, SLogicSubplan* pSubplan, SLogicNode* pSplitNode,
+                                     SNodeList* pMergeKeys, SLogicNode* pPartChild) {
   SMergeLogicNode* pMerge = nodesMakeNode(QUERY_NODE_LOGIC_PLAN_MERGE);
   if (NULL == pMerge) {
     return TSDB_CODE_OUT_OF_MEMORY;
   }
   pMerge->numOfChannels = ((SScanLogicNode*)nodesListGetNode(pPartChild->pChildren, 0))->pVgroupList->numOfVgroups;
   pMerge->srcGroupId = pCxt->groupId;
-  pMerge->node.pParent = pParent;
   pMerge->node.precision = pPartChild->precision;
   pMerge->pMergeKeys = pMergeKeys;
-  pMerge->node.pTargets = nodesCloneList(pPartChild->pTargets);
-  if (NULL == pMerge->node.pTargets) {
-    nodesDestroyNode(pMerge);
-    return TSDB_CODE_OUT_OF_MEMORY;
-  }
 
-  return nodesListMakeAppend(&pParent->pChildren, pMerge);
+  int32_t code = TSDB_CODE_SUCCESS;
+  pMerge->pInputs = nodesCloneList(pPartChild->pTargets);
+  pMerge->node.pTargets = nodesCloneList(pSplitNode->pTargets);
+  if (NULL == pMerge->node.pTargets || NULL == pMerge->pInputs) {
+    code = TSDB_CODE_OUT_OF_MEMORY;
+  }
+  if (TSDB_CODE_SUCCESS == code) {
+    if (NULL == pSubplan) {
+      code = nodesListMakeAppend(&pSplitNode->pChildren, pMerge);
+    } else {
+      code = splReplaceLogicNode(pSubplan, pSplitNode, (SLogicNode*)pMerge);
+    }
+  }
+  if (TSDB_CODE_SUCCESS != code) {
+    nodesDestroyNode(pMerge);
+  }
+  return code;
 }
 
 static int32_t stbSplCreateExchangeNode(SSplitContext* pCxt, SLogicNode* pParent, SLogicNode* pPartChild) {
@@ -329,8 +345,15 @@ static int32_t stbSplSplitWindowNodeForBatch(SSplitContext* pCxt, SStableSplitIn
   int32_t     code = stbSplCreatePartWindowNode((SWindowLogicNode*)pInfo->pSplitNode, &pPartWindow);
   if (TSDB_CODE_SUCCESS == code) {
     ((SWindowLogicNode*)pPartWindow)->intervalAlgo = INTERVAL_ALGO_HASH;
-    ((SWindowLogicNode*)pInfo->pSplitNode)->intervalAlgo = INTERVAL_ALGO_SORT_MERGE;
-    code = stbSplCreateExchangeNode(pCxt, pInfo->pSplitNode, pPartWindow);
+    ((SWindowLogicNode*)pInfo->pSplitNode)->intervalAlgo = INTERVAL_ALGO_MERGE;
+    SNodeList* pMergeKeys = NULL;
+    code = nodesListMakeStrictAppend(&pMergeKeys, nodesCloneNode(((SWindowLogicNode*)pInfo->pSplitNode)->pTspk));
+    if (TSDB_CODE_SUCCESS == code) {
+      code = stbSplCreateMergeNode(pCxt, NULL, pInfo->pSplitNode, pMergeKeys, pPartWindow);
+    }
+    if (TSDB_CODE_SUCCESS != code) {
+      nodesDestroyList(pMergeKeys);
+    }
   }
   if (TSDB_CODE_SUCCESS == code) {
     code = nodesListMakeStrictAppend(&pInfo->pSubplan->pChildren,
@@ -424,37 +447,99 @@ static int32_t stbSplSplitAggNode(SSplitContext* pCxt, SStableSplitInfo* pInfo)
   return code;
 }
 
-static int32_t stbSplCreatePartSortNode(SSortLogicNode* pMergeSort, SLogicNode** pOutput) {
-  SNodeList* pSortKeys = pMergeSort->pSortKeys;
-  pMergeSort->pSortKeys = NULL;
-  SNodeList* pTargets = pMergeSort->node.pTargets;
-  pMergeSort->node.pTargets = NULL;
-  SNodeList* pChildren = pMergeSort->node.pChildren;
-  pMergeSort->node.pChildren = NULL;
+static SNode* stbSplCreateColumnNode(SExprNode* pExpr) {
+  SColumnNode* pCol = nodesMakeNode(QUERY_NODE_COLUMN);
+  if (NULL == pCol) {
+    return NULL;
+  }
+  if (QUERY_NODE_COLUMN == nodeType(pExpr)) {
+    strcpy(pCol->tableAlias, ((SColumnNode*)pExpr)->tableAlias);
+  }
+  strcpy(pCol->colName, pExpr->aliasName);
+  strcpy(pCol->node.aliasName, pExpr->aliasName);
+  pCol->node.resType = pExpr->resType;
+  return (SNode*)pCol;
+}
+
+static SNode* stbSplCreateOrderByExpr(SOrderByExprNode* pSortKey, SNode* pCol) {
+  SOrderByExprNode* pOutput = nodesMakeNode(QUERY_NODE_ORDER_BY_EXPR);
+  if (NULL == pOutput) {
+    return NULL;
+  }
+  pOutput->pExpr = nodesCloneNode(pCol);
+  if (NULL == pOutput->pExpr) {
+    nodesDestroyNode(pOutput);
+    return NULL;
+  }
+  pOutput->order = pSortKey->order;
+  pOutput->nullOrder = pSortKey->nullOrder;
+  return (SNode*)pOutput;
+}
+
+static int32_t stbSplCreateMergeKeys(SNodeList* pSortKeys, SNodeList* pTargets, SNodeList** pOutput) {
+  int32_t    code = TSDB_CODE_SUCCESS;
+  SNodeList* pMergeKeys = NULL;
+  SNode*     pNode = NULL;
+  FOREACH(pNode, pSortKeys) {
+    SOrderByExprNode* pSortKey = (SOrderByExprNode*)pNode;
+    SNode*            pTarget = NULL;
+    bool              found = false;
+    FOREACH(pTarget, pTargets) {
+      if (0 == strcmp(((SExprNode*)pSortKey->pExpr)->aliasName, ((SColumnNode*)pTarget)->colName)) {
+        code = nodesListMakeStrictAppend(&pMergeKeys, stbSplCreateOrderByExpr(pSortKey, pTarget));
+        if (TSDB_CODE_SUCCESS != code) {
+          break;
+        }
+        found = true;
+      }
+    }
+    if (TSDB_CODE_SUCCESS == code && !found) {
+      SNode* pCol = stbSplCreateColumnNode((SExprNode*)pSortKey->pExpr);
+      code = nodesListMakeStrictAppend(&pMergeKeys, stbSplCreateOrderByExpr(pSortKey, pCol));
+      if (TSDB_CODE_SUCCESS == code) {
+        code = nodesListStrictAppend(pTargets, pCol);
+      } else {
+        nodesDestroyNode(pCol);
+      }
+    }
+    if (TSDB_CODE_SUCCESS != code) {
+      break;
+    }
+  }
+  if (TSDB_CODE_SUCCESS == code) {
+    *pOutput = pMergeKeys;
+  } else {
+    nodesDestroyList(pMergeKeys);
+  }
+  return code;
+}
+
+static int32_t stbSplCreatePartSortNode(SSortLogicNode* pSort, SLogicNode** pOutputPartSort,
+                                        SNodeList** pOutputMergeKeys) {
+  SNodeList* pSortKeys = pSort->pSortKeys;
+  pSort->pSortKeys = NULL;
+  SNodeList* pChildren = pSort->node.pChildren;
+  pSort->node.pChildren = NULL;
 
   int32_t         code = TSDB_CODE_SUCCESS;
-  SSortLogicNode* pPartSort = nodesCloneNode(pMergeSort);
+  SSortLogicNode* pPartSort = nodesCloneNode(pSort);
   if (NULL == pPartSort) {
     code = TSDB_CODE_OUT_OF_MEMORY;
   }
 
-  pMergeSort->node.pTargets = pTargets;
-  pPartSort->node.pChildren = pChildren;
+  SNodeList* pMergeKeys = NULL;
   if (TSDB_CODE_SUCCESS == code) {
+    pPartSort->node.pChildren = pChildren;
     pPartSort->pSortKeys = pSortKeys;
-    code = createColumnByRewriteExps(pPartSort->pSortKeys, &pPartSort->node.pTargets);
-  }
-  if (TSDB_CODE_SUCCESS == code) {
-    pMergeSort->pSortKeys = nodesCloneList(pPartSort->node.pTargets);
-    if (NULL == pMergeSort->pSortKeys) {
-      code = TSDB_CODE_OUT_OF_MEMORY;
-    }
+    code = stbSplCreateMergeKeys(pPartSort->pSortKeys, pPartSort->node.pTargets, &pMergeKeys);
   }
 
   if (TSDB_CODE_SUCCESS == code) {
-    *pOutput = (SLogicNode*)pPartSort;
+    *pOutputPartSort = (SLogicNode*)pPartSort;
+    *pOutputMergeKeys = pMergeKeys;
   } else {
     nodesDestroyNode(pPartSort);
+    nodesDestroyList(pMergeKeys);
   }
 
   return code;
@@ -462,17 +547,10 @@ static int32_t stbSplCreatePartSortNode(SSortLogicNode* pMergeSort, SLogicNode**
 
 static int32_t stbSplSplitSortNode(SSplitContext* pCxt, SStableSplitInfo* pInfo) {
   SLogicNode* pPartSort = NULL;
-  int32_t     code = stbSplCreatePartSortNode((SSortLogicNode*)pInfo->pSplitNode, &pPartSort);
+  SNodeList*  pMergeKeys = NULL;
+  int32_t     code = stbSplCreatePartSortNode((SSortLogicNode*)pInfo->pSplitNode, &pPartSort, &pMergeKeys);
   if (TSDB_CODE_SUCCESS == code) {
-    SNodeList* pMergeKeys = nodesCloneList(((SSortLogicNode*)pInfo->pSplitNode)->pSortKeys);
-    if (NULL != pMergeKeys) {
-      code = stbSplCreateMergeNode(pCxt, pInfo->pSplitNode, pMergeKeys, pPartSort);
-      if (TSDB_CODE_SUCCESS != code) {
-        nodesDestroyList(pMergeKeys);
-      }
-    } else {
-      code = TSDB_CODE_OUT_OF_MEMORY;
-    }
+    code = stbSplCreateMergeNode(pCxt, pInfo->pSubplan, pInfo->pSplitNode, pMergeKeys, pPartSort);
   }
   if (TSDB_CODE_SUCCESS == code) {
     code = nodesListMakeStrictAppend(&pInfo->pSubplan->pChildren,
diff --git a/source/libs/planner/test/planOrderByTest.cpp b/source/libs/planner/test/planOrderByTest.cpp
index ef6f438b55..851eda81b5 100644
--- a/source/libs/planner/test/planOrderByTest.cpp
+++ b/source/libs/planner/test/planOrderByTest.cpp
@@ -46,4 +46,7 @@ TEST_F(PlanOrderByTest, stable) {
 
   // ORDER BY key is in the projection list
   run("SELECT c1 FROM st1 ORDER BY c1");
+
+  // ORDER BY key is not in the projection list
+  run("SELECT c2 FROM st1 ORDER BY c1");
 }

From f9cb481413b339d8ca39eb7316918bd5df3ff06a Mon Sep 17 00:00:00 2001
From: "wenzhouwww@live.cn" <wenzhouwww@live.cn>
Date: Mon, 6 Jun 2022 18:01:20 +0800
Subject: [PATCH 27/34] log case update for support e as base

---
 tests/system-test/2-query/log.py | 67 +++++++++++++++++++++++++++++++-
 1 file changed, 66 insertions(+), 1 deletion(-)

diff --git a/tests/system-test/2-query/log.py b/tests/system-test/2-query/log.py
index 6e4c292183..907ba329ee 100644
--- a/tests/system-test/2-query/log.py
+++ b/tests/system-test/2-query/log.py
@@ -66,6 +66,37 @@ class TDTestCase:
             ( '2023-02-21 01:01:01.000', NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL )
             '''
         )
+
+
+    def check_result_auto_log(self ,origin_query , log_query):
+    
+        log_result = tdSql.getResult(log_query)
+        origin_result = tdSql.getResult(origin_query)
+
+        auto_result =[]
+        
+        for row in origin_result:
+            row_check = []
+            for elem in row:
+                if elem == None:
+                    elem = None
+                elif elem >0:
+                    elem = math.log(elem)
+                elif elem <=0:
+                    elem = None
+                row_check.append(elem)
+            auto_result.append(row_check)
+
+        check_status = True
+        for row_index , row in enumerate(log_result):
+            for col_index , elem in enumerate(row):
+                if auto_result[row_index][col_index] != elem:
+                    check_status = False 
+        if not check_status:
+            tdLog.notice("log function value has not as expected , sql is \"%s\" "%log_query )
+            sys.exit(1)
+        else:
+            tdLog.info("log value check pass , it work as expected ,sql is \"%s\"   "%log_query )
     
     def check_result_auto_log2(self ,origin_query , log_query):
 
@@ -270,6 +301,7 @@ class TDTestCase:
         tdSql.checkRows(25)
 
         # used for empty table  , ct3 is empty
+
         tdSql.query("select log(c1 ,2) from ct3")
         tdSql.checkRows(0)
         tdSql.query("select log(c2 ,2) from ct3")
@@ -291,6 +323,13 @@ class TDTestCase:
         tdSql.checkData(3 , 0, 1.584962501)
         tdSql.checkData(5 , 0, None)
 
+        tdSql.query("select log(c1) from t1")
+        tdSql.checkData(0, 0, None)
+        tdSql.checkData(1 , 0, 0.000000000)
+        tdSql.checkData(2 , 0, 0.693147181)
+        tdSql.checkData(3 , 0, 1.098612289)
+        tdSql.checkData(4 , 0, 1.386294361)
+
         tdSql.query("select c1, c2, c3 , c4, c5 from t1")
         tdSql.checkData(1, 4, 1.11000)
         tdSql.checkData(3, 3, 33)
@@ -301,6 +340,7 @@ class TDTestCase:
         tdSql.checkData(3, 4, 33)
         tdSql.checkData(5, 5, None)
 
+        self.check_result_auto_log( "select c1, c2, c3 , c4, c5 from t1", "select log(c1), log(c2) ,log(c3), log(c4), log(c5) from t1")
         self.check_result_auto_log2( "select c1, c2, c3 , c4, c5 from t1", "select log(c1 ,2), log(c2 ,2) ,log(c3, 2), log(c4 ,2), log(c5 ,2) from t1")
         self.check_result_auto_log1( "select c1, c2, c3 , c4, c5 from t1", "select log(c1 ,1), log(c2 ,1) ,log(c3, 1), log(c4 ,1), log(c5 ,1) from t1")
         self.check_result_auto_log__10( "select c1, c2, c3 , c4, c5 from t1", "select log(c1 ,-10), log(c2 ,-10) ,log(c3, -10), log(c4 ,-10), log(c5 ,-10) from t1")
@@ -321,7 +361,7 @@ class TDTestCase:
         tdSql.checkData(3 , 2, 0.147315235)
         tdSql.checkData(4 , 2, None)
 
-
+        self.check_result_auto_log( "select c1, c2, c3 , c4, c5 from ct1", "select log(c1), log(c2) ,log(c3), log(c4), log(c5) from ct1")
         self.check_result_auto_log2( "select c1, c2, c3 , c4, c5 from ct1", "select log(c1,2), log(c2,2) ,log(c3,2), log(c4,2), log(c5,2) from ct1")
         self.check_result_auto_log__10( "select c1, c2, c3 , c4, c5 from ct1", "select log(c1,-10), log(c2,-10) ,log(c3,-10), log(c4,-10), log(c5,-10) from ct1")
 
@@ -569,10 +609,12 @@ class TDTestCase:
         tdSql.error(
                 f"insert into sub1_bound values ( now()+1s, 2147483648, 9223372036854775808, 32768, 128, 3.40E+38, 1.7e+308, True, 'binary_tb1', 'nchar_tb1', now() )"
             )
+        self.check_result_auto_log( "select c1, c2, c3 , c4, c5 ,c6 from sub1_bound ", "select log(c1), log(c2) ,log(c3), log(c4), log(c5) ,log(c6) from sub1_bound")
         self.check_result_auto_log2( "select c1, c2, c3 , c4, c5 ,c6 from sub1_bound ", "select log(c1,2), log(c2,2) ,log(c3,2), log(c4,2), log(c5,2) ,log(c6,2) from sub1_bound")
         self.check_result_auto_log__10( "select c1, c2, c3 , c4, c5 ,c6 from sub1_bound ", "select log(c1,-10), log(c2,-10) ,log(c3,-10), log(c4,-10), log(c5,-10) ,log(c6,-10) from sub1_bound")
         
         self.check_result_auto_log2( "select c1, c2, c3 , c3, c2 ,c1 from sub1_bound ", "select log(c1,2), log(c2,2) ,log(c3,2), log(c3,2), log(c2,2) ,log(c1,2) from sub1_bound")
+        self.check_result_auto_log( "select c1, c2, c3 , c3, c2 ,c1 from sub1_bound ", "select log(c1), log(c2) ,log(c3), log(c3), log(c2) ,log(c1) from sub1_bound")
 
 
         self.check_result_auto_log2("select abs(abs(abs(abs(abs(abs(abs(abs(abs(c1)))))))))  nest_col_func from sub1_bound" , "select log(abs(c1) ,2) from sub1_bound" )
@@ -598,6 +640,29 @@ class TDTestCase:
         tdSql.checkData(3,4,math.log(339999995214436424907732413799364296704.00000,2))
         tdSql.checkData(3,5,math.log(169999999999999993883079578865998174333346074304075874502773119193537729178160565864330091787584707988572262467983188919169916105593357174268369962062473635296474636515660464935663040684957844303524367815028553272712298986386310828644513212353921123253311675499856875650512437415429217994623324794855339589632.000000000,2))
 
+        # check basic elem for table per row
+        tdSql.query("select log(abs(c1)) ,log(abs(c2)) , log(abs(c3)) , log(abs(c4)), log(abs(c5)), log(abs(c6)) from sub1_bound ")
+        tdSql.checkData(0,0,math.log(2147483647))
+        tdSql.checkData(0,1,math.log(9223372036854775807))
+        tdSql.checkData(0,2,math.log(32767))
+        tdSql.checkData(0,3,math.log(127))
+        tdSql.checkData(0,4,math.log(339999995214436424907732413799364296704.00000))
+        tdSql.checkData(0,5,math.log(169999999999999993883079578865998174333346074304075874502773119193537729178160565864330091787584707988572262467983188919169916105593357174268369962062473635296474636515660464935663040684957844303524367815028553272712298986386310828644513212353921123253311675499856875650512437415429217994623324794855339589632.000000000 ))
+        tdSql.checkData(1,0,math.log(2147483647))
+        tdSql.checkData(1,1,math.log(9223372036854775807))
+        tdSql.checkData(1,2,math.log(32767))
+        tdSql.checkData(1,3,math.log(127))
+        tdSql.checkData(1,4,math.log(339999995214436424907732413799364296704.00000 ))
+        tdSql.checkData(1,5,math.log(169999999999999993883079578865998174333346074304075874502773119193537729178160565864330091787584707988572262467983188919169916105593357174268369962062473635296474636515660464935663040684957844303524367815028553272712298986386310828644513212353921123253311675499856875650512437415429217994623324794855339589632.000000000))
+        tdSql.checkData(3,0,math.log(2147483646))
+        tdSql.checkData(3,1,math.log(9223372036854775806))
+        tdSql.checkData(3,2,math.log(32766))
+        tdSql.checkData(3,3,math.log(126))
+        tdSql.checkData(3,4,math.log(339999995214436424907732413799364296704.00000))
+        tdSql.checkData(3,5,math.log(169999999999999993883079578865998174333346074304075874502773119193537729178160565864330091787584707988572262467983188919169916105593357174268369962062473635296474636515660464935663040684957844303524367815028553272712298986386310828644513212353921123253311675499856875650512437415429217994623324794855339589632.000000000))
+
+
+
         # check  + - * / in functions
         tdSql.query("select log(abs(c1+1) ,2) ,log(abs(c2),2) , log(abs(c3*1),2) , log(abs(c4/2),2), log(abs(c5) ,2)/2, log(abs(c6) ,2) from sub1_bound ")
         tdSql.checkData(0,0,math.log(2147483648.000000000,2))

From 5f71c3a6e273bcdcc44f54c3e728eb3484ed7825 Mon Sep 17 00:00:00 2001
From: Shengliang Guan <slguan@taosdata.com>
Date: Mon, 6 Jun 2022 18:03:23 +0800
Subject: [PATCH 28/34] refactor: disable drop dnode script

---
 source/dnode/mnode/impl/src/mndMnode.c |  1 +
 tests/script/tsim/dnode/basic1.sim     | 16 ++++++++--------
 2 files changed, 9 insertions(+), 8 deletions(-)

diff --git a/source/dnode/mnode/impl/src/mndMnode.c b/source/dnode/mnode/impl/src/mndMnode.c
index d315bb12ed..072ad05f5a 100644
--- a/source/dnode/mnode/impl/src/mndMnode.c
+++ b/source/dnode/mnode/impl/src/mndMnode.c
@@ -536,6 +536,7 @@ static int32_t mndSetDropMnodeRedoActions(SMnode *pMnode, STrans *pTrans, SDnode
 }
 
 int32_t mndSetDropMnodeInfoToTrans(SMnode *pMnode, STrans *pTrans, SMnodeObj *pObj) {
+  if (pObj == NULL) return 0;
   if (mndSetDropMnodeRedoLogs(pMnode, pTrans, pObj) != 0) return -1;
   if (mndSetDropMnodeCommitLogs(pMnode, pTrans, pObj) != 0) return -1;
   if (mndSetDropMnodeRedoActions(pMnode, pTrans, pObj->pDnode, pObj) != 0) return -1;
diff --git a/tests/script/tsim/dnode/basic1.sim b/tests/script/tsim/dnode/basic1.sim
index a5b5427e03..8a28897d5a 100644
--- a/tests/script/tsim/dnode/basic1.sim
+++ b/tests/script/tsim/dnode/basic1.sim
@@ -183,15 +183,15 @@ if $rows != 15 then
 endi
 
 print =============== drop dnode
-sql drop dnode 2;
-sql show dnodes;
-if $rows != 1 then 
-  return -1
-endi
+#sql drop dnode 2;
+#sql show dnodes;
+#if $rows != 1 then 
+#  return -1
+#endi
 
-if $data00 != 1 then 
-  return -1
-endi
+#if $data00 != 1 then 
+#  return -1
+#endi
 
 system sh/exec.sh -n dnode1 -s stop -x SIGINT
 system sh/exec.sh -n dnode2 -s stop -x SIGINT

From ce3aacf4d76bc62c5dd891fe2c46603cd3841fbc Mon Sep 17 00:00:00 2001
From: Xiaoyu Wang <xiaoyuwang@taosdata.com>
Date: Mon, 6 Jun 2022 18:07:34 +0800
Subject: [PATCH 29/34] feat: order by distributed split

---
 source/libs/executor/src/executorimpl.c    | 39 +++++++++++-----------
 source/libs/planner/src/planPhysiCreater.c |  3 ++
 2 files changed, 22 insertions(+), 20 deletions(-)

diff --git a/source/libs/executor/src/executorimpl.c b/source/libs/executor/src/executorimpl.c
index beb5e1152f..10381f72fd 100644
--- a/source/libs/executor/src/executorimpl.c
+++ b/source/libs/executor/src/executorimpl.c
@@ -2522,8 +2522,8 @@ void relocateColumnData(SSDataBlock* pBlock, const SArray* pColMatchInfo, SArray
 }
 
 int32_t setDataBlockFromFetchRsp(SSDataBlock* pRes, SLoadRemoteDataInfo* pLoadInfo, int32_t numOfRows, char* pData,
-                                  int32_t compLen, int32_t numOfOutput, int64_t startTs, uint64_t* total,
-                                  SArray* pColList) {
+                                 int32_t compLen, int32_t numOfOutput, int64_t startTs, uint64_t* total,
+                                 SArray* pColList) {
   if (pColList == NULL) {  // data from other sources
     blockCompressDecode(pRes, numOfOutput, numOfRows, pData);
   } else {  // extract data according to pColList
@@ -2677,7 +2677,7 @@ static SSDataBlock* concurrentlyLoadRemoteDataImpl(SOperatorInfo* pOperator, SEx
 
       SRetrieveTableRsp* pTableRsp = pDataInfo->pRsp;
       code = setDataBlockFromFetchRsp(pExchangeInfo->pResult, pLoadInfo, pTableRsp->numOfRows, pTableRsp->data,
-                                       pTableRsp->compLen, pTableRsp->numOfCols, startTs, &pDataInfo->totalRows, NULL);
+                                      pTableRsp->compLen, pTableRsp->numOfCols, startTs, &pDataInfo->totalRows, NULL);
       if (code != 0) {
         taosMemoryFreeClear(pDataInfo->pRsp);
         goto _error;
@@ -2790,7 +2790,7 @@ static SSDataBlock* seqLoadRemoteData(SOperatorInfo* pOperator) {
     SRetrieveTableRsp* pTableRsp = pDataInfo->pRsp;
     int32_t            code =
         setDataBlockFromFetchRsp(pExchangeInfo->pResult, pLoadInfo, pTableRsp->numOfRows, pTableRsp->data,
-                                  pTableRsp->compLen, pTableRsp->numOfCols, startTs, &pDataInfo->totalRows, NULL);
+                                 pTableRsp->compLen, pTableRsp->numOfCols, startTs, &pDataInfo->totalRows, NULL);
 
     if (pRsp->completed == 1) {
       qDebug("%s fetch msg rsp from vgId:%d, taskId:0x%" PRIx64 " numOfRows:%d, rowsOfSource:%" PRIu64
@@ -2856,7 +2856,7 @@ static SSDataBlock* doLoadRemoteData(SOperatorInfo* pOperator) {
     return seqLoadRemoteData(pOperator);
   } else {
     return concurrentlyLoadRemoteDataImpl(pOperator, pExchangeInfo, pTaskInfo);
-//    return concurrentlyLoadRemoteData(pOperator);
+    //    return concurrentlyLoadRemoteData(pOperator);
   }
 }
 
@@ -2911,18 +2911,18 @@ SOperatorInfo* createExchangeOperatorInfo(void* pTransporter, SExchangePhysiNode
     goto _error;
   }
 
-  pInfo->seqLoadData  = false;
+  pInfo->seqLoadData = false;
   pInfo->pTransporter = pTransporter;
-  pInfo->pResult      = createResDataBlock(pExNode->node.pOutputDataBlockDesc);
+  pInfo->pResult = createResDataBlock(pExNode->node.pOutputDataBlockDesc);
   tsem_init(&pInfo->ready, 0, 0);
 
-  pOperator->name         = "ExchangeOperator";
+  pOperator->name = "ExchangeOperator";
   pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_EXCHANGE;
-  pOperator->blocking     = false;
-  pOperator->status       = OP_NOT_OPENED;
-  pOperator->info         = pInfo;
-  pOperator->numOfExprs   = pInfo->pResult->info.numOfCols;
-  pOperator->pTaskInfo    = pTaskInfo;
+  pOperator->blocking = false;
+  pOperator->status = OP_NOT_OPENED;
+  pOperator->info = pInfo;
+  pOperator->numOfExprs = pInfo->pResult->info.numOfCols;
+  pOperator->pTaskInfo = pTaskInfo;
 
   pOperator->fpSet = createOperatorFpSet(prepareLoadRemoteData, doLoadRemoteData, NULL, NULL,
                                          destroyExchangeOperatorInfo, NULL, NULL, NULL);
@@ -4389,7 +4389,7 @@ SOperatorInfo* createOperatorTree(SPhysiNode* pPhyNode, SExecTaskInfo* pTaskInfo
       SScanPhysiNode*      pScanPhyNode = (SScanPhysiNode*)pPhyNode;  // simple child table.
       STableScanPhysiNode* pTableScanNode = (STableScanPhysiNode*)pPhyNode;
       STimeWindowAggSupp   twSup = {
-          .waterMark = pTableScanNode->watermark, .calTrigger = pTableScanNode->triggerType, .maxTs = INT64_MIN};
+            .waterMark = pTableScanNode->watermark, .calTrigger = pTableScanNode->triggerType, .maxTs = INT64_MIN};
       tsdbReaderT pDataReader = NULL;
       if (pHandle->vnode) {
         pDataReader = doCreateDataReader(pTableScanNode, pHandle, pTableListInfo, (uint64_t)queryId, taskId, pTagCond);
@@ -4517,17 +4517,16 @@ SOperatorInfo* createOperatorTree(SPhysiNode* pPhyNode, SExecTaskInfo* pTaskInfo
         extractColMatchInfo(pSortPhyNode->pTargets, pDescNode, &numOfOutputCols, pTaskInfo, COL_MATCH_FROM_SLOT_ID);
 
     pOptr = createSortOperatorInfo(ops[0], pResBlock, info, pExprInfo, numOfCols, pColList, pTaskInfo);
-  }  else if (QUERY_NODE_PHYSICAL_PLAN_MERGE == type) {
+  } else if (QUERY_NODE_PHYSICAL_PLAN_MERGE == type) {
     SMergePhysiNode* pMergePhyNode = (SMergePhysiNode*)pPhyNode;
 
     SDataBlockDescNode* pDescNode = pPhyNode->pOutputDataBlockDesc;
-    SSDataBlock* pResBlock = createResDataBlock(pDescNode);
+    SSDataBlock*        pResBlock = createResDataBlock(pDescNode);
 
-    SArray*      sortInfo = createSortInfo(pMergePhyNode->pMergeKeys);
+    SArray* sortInfo = createSortInfo(pMergePhyNode->pMergeKeys);
     int32_t numOfOutputCols = 0;
-    SArray* pColList = NULL;
-        //extractColMatchInfo(pMergePhyNode->pTargets, pDescNode, &numOfOutputCols, pTaskInfo, COL_MATCH_FROM_SLOT_ID);
-
+    SArray* pColList =
+        extractColMatchInfo(pMergePhyNode->pTargets, pDescNode, &numOfOutputCols, pTaskInfo, COL_MATCH_FROM_SLOT_ID);
 
     pOptr = createMultiwaySortMergeOperatorInfo(ops, size, pResBlock, sortInfo, pColList, pTaskInfo);
   } else if (QUERY_NODE_PHYSICAL_PLAN_SESSION_WINDOW == type) {
diff --git a/source/libs/planner/src/planPhysiCreater.c b/source/libs/planner/src/planPhysiCreater.c
index b2a675dbb1..991d6d66b8 100644
--- a/source/libs/planner/src/planPhysiCreater.c
+++ b/source/libs/planner/src/planPhysiCreater.c
@@ -1191,6 +1191,9 @@ static int32_t createMergePhysiNode(SPhysiPlanContext* pCxt, SMergeLogicNode* pM
     code = setListSlotId(pCxt, pMerge->node.pOutputDataBlockDesc->dataBlockId, -1, pMergeLogicNode->node.pTargets,
                          &pMerge->pTargets);
   }
+  if (TSDB_CODE_SUCCESS == code) {
+    code = addDataBlockSlots(pCxt, pMerge->pTargets, pMerge->node.pOutputDataBlockDesc);
+  }
 
   if (TSDB_CODE_SUCCESS == code) {
     *pPhyNode = (SPhysiNode*)pMerge;

From 2a1ec3c304d889f1629b4bf4e7f2fee8b0a53380 Mon Sep 17 00:00:00 2001
From: Xiaoyu Wang <xiaoyuwang@taosdata.com>
Date: Mon, 6 Jun 2022 18:39:23 +0800
Subject: [PATCH 30/34] feat: order by distributed split

---
 include/libs/nodes/plannodes.h             |  3 ++-
 source/libs/nodes/src/nodesCloneFuncs.c    |  2 +-
 source/libs/nodes/src/nodesCodeFuncs.c     | 13 ++++++++++---
 source/libs/planner/src/planLogicCreater.c |  4 ++--
 source/libs/planner/src/planPhysiCreater.c | 18 +++++++++++++-----
 5 files changed, 28 insertions(+), 12 deletions(-)

diff --git a/include/libs/nodes/plannodes.h b/include/libs/nodes/plannodes.h
index d646effa58..5811b76588 100644
--- a/include/libs/nodes/plannodes.h
+++ b/include/libs/nodes/plannodes.h
@@ -95,7 +95,7 @@ typedef struct SVnodeModifyLogicNode {
   int32_t          msgType;
   SArray*          pDataBlocks;
   SVgDataBlocks*   pVgDataBlocks;
-  SNode*           pModifyRows;  // SColumnNode
+  SNode*           pAffectedRows;  // SColumnNode
   uint64_t         tableId;
   int8_t           tableType;  // table type
   char             tableFName[TSDB_TABLE_FNAME_LEN];
@@ -392,6 +392,7 @@ typedef struct SDataDeleterNode {
   int8_t        tableType;  // table type
   char          tableFName[TSDB_TABLE_FNAME_LEN];
   STimeWindow   deleteTimeRange;
+  SNode*        pAffectedRows;
 } SDataDeleterNode;
 
 typedef struct SSubplan {
diff --git a/source/libs/nodes/src/nodesCloneFuncs.c b/source/libs/nodes/src/nodesCloneFuncs.c
index 657ffd32a5..7d58b999dd 100644
--- a/source/libs/nodes/src/nodesCloneFuncs.c
+++ b/source/libs/nodes/src/nodesCloneFuncs.c
@@ -371,7 +371,7 @@ static SNode* logicVnodeModifCopy(const SVnodeModifyLogicNode* pSrc, SVnodeModif
   COPY_BASE_OBJECT_FIELD(node, logicNodeCopy);
   COPY_SCALAR_FIELD(modifyType);
   COPY_SCALAR_FIELD(msgType);
-  CLONE_NODE_FIELD(pModifyRows);
+  CLONE_NODE_FIELD(pAffectedRows);
   COPY_SCALAR_FIELD(tableId);
   COPY_SCALAR_FIELD(tableType);
   COPY_CHAR_ARRAY_FIELD(tableFName);
diff --git a/source/libs/nodes/src/nodesCodeFuncs.c b/source/libs/nodes/src/nodesCodeFuncs.c
index 246ee2bfed..8583f72c26 100644
--- a/source/libs/nodes/src/nodesCodeFuncs.c
+++ b/source/libs/nodes/src/nodesCodeFuncs.c
@@ -613,7 +613,7 @@ static int32_t jsonToLogicProjectNode(const SJson* pJson, void* pObj) {
 
 static const char* jkVnodeModifyLogicPlanModifyType = "ModifyType";
 static const char* jkVnodeModifyLogicPlanMsgType = "MsgType";
-static const char* jkVnodeModifyLogicPlanModifyRows = "ModifyRows";
+static const char* jkVnodeModifyLogicPlanAffectedRows = "AffectedRows";
 
 static int32_t logicVnodeModifyNodeToJson(const void* pObj, SJson* pJson) {
   const SVnodeModifyLogicNode* pNode = (const SVnodeModifyLogicNode*)pObj;
@@ -626,7 +626,7 @@ static int32_t logicVnodeModifyNodeToJson(const void* pObj, SJson* pJson) {
     code = tjsonAddIntegerToObject(pJson, jkVnodeModifyLogicPlanMsgType, pNode->msgType);
   }
   if (TSDB_CODE_SUCCESS == code) {
-    code = tjsonAddObject(pJson, jkVnodeModifyLogicPlanModifyRows, nodeToJson, pNode->pModifyRows);
+    code = tjsonAddObject(pJson, jkVnodeModifyLogicPlanAffectedRows, nodeToJson, pNode->pAffectedRows);
   }
 
   return code;
@@ -643,7 +643,7 @@ static int32_t jsonToLogicVnodeModifyNode(const SJson* pJson, void* pObj) {
     code = tjsonGetIntValue(pJson, jkVnodeModifyLogicPlanMsgType, &pNode->msgType);
   }
   if (TSDB_CODE_SUCCESS == code) {
-    code = jsonToNodeObject(pJson, jkVnodeModifyLogicPlanModifyRows, &pNode->pModifyRows);
+    code = jsonToNodeObject(pJson, jkVnodeModifyLogicPlanAffectedRows, &pNode->pAffectedRows);
   }
 
   return code;
@@ -2016,6 +2016,7 @@ static const char* jkDeletePhysiPlanTableType = "TableType";
 static const char* jkDeletePhysiPlanTableFName = "TableFName";
 static const char* jkDeletePhysiPlanDeleteTimeRangeStartKey = "DeleteTimeRangeStartKey";
 static const char* jkDeletePhysiPlanDeleteTimeRangeEndKey = "DeleteTimeRangeEndKey";
+static const char* jkDeletePhysiPlanAffectedRows = "AffectedRows";
 
 static int32_t physiDeleteNodeToJson(const void* pObj, SJson* pJson) {
   const SDataDeleterNode* pNode = (const SDataDeleterNode*)pObj;
@@ -2036,6 +2037,9 @@ static int32_t physiDeleteNodeToJson(const void* pObj, SJson* pJson) {
   if (TSDB_CODE_SUCCESS == code) {
     code = tjsonAddIntegerToObject(pJson, jkDeletePhysiPlanDeleteTimeRangeEndKey, pNode->deleteTimeRange.ekey);
   }
+  if (TSDB_CODE_SUCCESS == code) {
+    code = tjsonAddObject(pJson, jkDeletePhysiPlanAffectedRows, nodeToJson, pNode->pAffectedRows);
+  }
 
   return code;
 }
@@ -2059,6 +2063,9 @@ static int32_t jsonToPhysiDeleteNode(const SJson* pJson, void* pObj) {
   if (TSDB_CODE_SUCCESS == code) {
     code = tjsonGetBigIntValue(pJson, jkDeletePhysiPlanDeleteTimeRangeEndKey, &pNode->deleteTimeRange.ekey);
   }
+  if (TSDB_CODE_SUCCESS == code) {
+    code = jsonToNodeObject(pJson, jkDeletePhysiPlanAffectedRows, &pNode->pAffectedRows);
+  }
 
   return code;
 }
diff --git a/source/libs/planner/src/planLogicCreater.c b/source/libs/planner/src/planLogicCreater.c
index d37e22b316..ac46531fad 100644
--- a/source/libs/planner/src/planLogicCreater.c
+++ b/source/libs/planner/src/planLogicCreater.c
@@ -1057,8 +1057,8 @@ static int32_t createVnodeModifLogicNodeByDelete(SLogicPlanContext* pCxt, SDelet
   snprintf(pModify->tableFName, sizeof(pModify->tableFName), "%d.%s.%s", pCxt->pPlanCxt->acctId,
            pRealTable->table.dbName, pRealTable->table.tableName);
   pModify->deleteTimeRange = pDelete->timeRange;
-  pModify->pModifyRows = nodesCloneNode(pDelete->pCountFunc);
-  if (NULL == pModify->pModifyRows) {
+  pModify->pAffectedRows = nodesCloneNode(pDelete->pCountFunc);
+  if (NULL == pModify->pAffectedRows) {
     nodesDestroyNode(pModify);
     return TSDB_CODE_OUT_OF_MEMORY;
   }
diff --git a/source/libs/planner/src/planPhysiCreater.c b/source/libs/planner/src/planPhysiCreater.c
index 991d6d66b8..bf5ff57949 100644
--- a/source/libs/planner/src/planPhysiCreater.c
+++ b/source/libs/planner/src/planPhysiCreater.c
@@ -1332,13 +1332,21 @@ static int32_t createDataDeleter(SPhysiPlanContext* pCxt, SVnodeModifyLogicNode*
   strcpy(pDeleter->tableFName, pModify->tableFName);
   pDeleter->deleteTimeRange = pModify->deleteTimeRange;
 
-  pDeleter->sink.pInputDataBlockDesc = nodesCloneNode(pRoot->pOutputDataBlockDesc);
-  if (NULL == pDeleter->sink.pInputDataBlockDesc) {
-    nodesDestroyNode(pDeleter);
-    return TSDB_CODE_OUT_OF_MEMORY;
+  int32_t code = setNodeSlotId(pCxt, pRoot->pOutputDataBlockDesc->dataBlockId, -1, pModify->pAffectedRows,
+                               &pDeleter->pAffectedRows);
+  if (TSDB_CODE_SUCCESS == code) {
+    pDeleter->sink.pInputDataBlockDesc = nodesCloneNode(pRoot->pOutputDataBlockDesc);
+    if (NULL == pDeleter->sink.pInputDataBlockDesc) {
+      code = TSDB_CODE_OUT_OF_MEMORY;
+    }
+  }
+
+  if (TSDB_CODE_SUCCESS == code) {
+    *pSink = (SDataSinkNode*)pDeleter;
+  } else {
+    nodesDestroyNode(pDeleter);
   }
 
-  *pSink = (SDataSinkNode*)pDeleter;
   return TSDB_CODE_SUCCESS;
 }
 

From da03a3b6fbf688fd322382106097816408058630 Mon Sep 17 00:00:00 2001
From: afwerar <1296468573@qq.com>
Date: Mon, 6 Jun 2022 19:40:13 +0800
Subject: [PATCH 31/34] fix(os): do not kill taosd when case init dnodes

---
 tests/pytest/util/dnodes.py                   |   4 +-
 tests/system-test/0-others/taosShellNetChk.py |  85 +++++++--------
 tests/system-test/0-others/telemetry.py       |  24 +++--
 tests/system-test/0-others/udf_create.py      |  11 ++
 tests/system-test/fulltest.bat                | 100 +++++++++++++++++-
 tests/system-test/test-all.bat                |  10 +-
 6 files changed, 177 insertions(+), 57 deletions(-)

diff --git a/tests/pytest/util/dnodes.py b/tests/pytest/util/dnodes.py
index fb0c1881c9..63b841cfc5 100644
--- a/tests/pytest/util/dnodes.py
+++ b/tests/pytest/util/dnodes.py
@@ -395,7 +395,7 @@ class TDDnode:
 
     def stop(self):
         if (not self.remoteIP == ""):
-            self.remoteExec(self.cfgDict, "tdDnodes.stop(%d)"%self.index)
+            self.remoteExec(self.cfgDict, "tdDnodes.dnodes[%d].running=1\ntdDnodes.dnodes[%d].stop()"%(self.index-1,self.index-1))
             tdLog.info("stop dnode%d"%self.index)
             return
         if self.valgrind == 0:
@@ -426,7 +426,7 @@ class TDDnode:
 
     def forcestop(self):
         if (not self.remoteIP == ""):
-            self.remoteExec(self.cfgDict, "tdDnodes.forcestop(%d)"%self.index)
+            self.remoteExec(self.cfgDict, "tdDnodes.dnodes[%d].running=1\ntdDnodes.dnodes[%d].forcestop()"%(self.index-1,self.index-1))
             return
         if self.valgrind == 0:
             toBeKilled = "taosd"
diff --git a/tests/system-test/0-others/taosShellNetChk.py b/tests/system-test/0-others/taosShellNetChk.py
index 3c99ddb8d6..22c9c8c0c5 100644
--- a/tests/system-test/0-others/taosShellNetChk.py
+++ b/tests/system-test/0-others/taosShellNetChk.py
@@ -187,50 +187,51 @@ class TDTestCase:
         # stop taosd
         tdDnodes.stop(1)        
 
-        role   = 'server'
-        if platform.system().lower() == 'windows':
-            taosCmd = 'mintty -h never -w hide ' + buildPath + '\\build\\bin\\taos.exe -c ' + keyDict['c']
-            taosCmd = taosCmd.replace('\\','\\\\')
-            taosCmd = taosCmd + ' -n ' + role
-        else:
-            taosCmd = 'nohup ' + buildPath + '/build/bin/taos -c ' + keyDict['c']
-            taosCmd = taosCmd + ' -n ' + role + ' > /dev/null 2>&1 &'
-        print (taosCmd)
-        os.system(taosCmd)
+        try:
+            role   = 'server'
+            if platform.system().lower() == 'windows':
+                taosCmd = 'mintty -h never -w hide ' + buildPath + '\\build\\bin\\taos.exe -c ' + keyDict['c']
+                taosCmd = taosCmd.replace('\\','\\\\')
+                taosCmd = taosCmd + ' -n ' + role
+            else:
+                taosCmd = 'nohup ' + buildPath + '/build/bin/taos -c ' + keyDict['c']
+                taosCmd = taosCmd + ' -n ' + role + ' > /dev/null 2>&1 &'
+            print (taosCmd)
+            os.system(taosCmd)
 
-        pktLen = '2000'
-        pktNum = '10'
-        role   = 'client'
-        if platform.system().lower() == 'windows':
-            taosCmd = buildPath + '\\build\\bin\\taos.exe -h 127.0.0.1 -c ' + keyDict['c']
-            taosCmd = taosCmd.replace('\\','\\\\')
-        else:
-            taosCmd = buildPath + '/build/bin/taos -c ' + keyDict['c']
-        taosCmd = taosCmd + ' -n ' + role + ' -l ' + pktLen + ' -N ' +  pktNum
-        print (taosCmd)
-        child = taosExpect.spawn(taosCmd, timeout=3)
-        i = child.expect([taosExpect.TIMEOUT, taosExpect.EOF], timeout=6)
+            pktLen = '2000'
+            pktNum = '10'
+            role   = 'client'
+            if platform.system().lower() == 'windows':
+                taosCmd = buildPath + '\\build\\bin\\taos.exe -h 127.0.0.1 -c ' + keyDict['c']
+                taosCmd = taosCmd.replace('\\','\\\\')
+            else:
+                taosCmd = buildPath + '/build/bin/taos -c ' + keyDict['c']
+            taosCmd = taosCmd + ' -n ' + role + ' -l ' + pktLen + ' -N ' +  pktNum
+            print (taosCmd)
+            child = taosExpect.spawn(taosCmd, timeout=3)
+            i = child.expect([taosExpect.TIMEOUT, taosExpect.EOF], timeout=6)
 
-        if platform.system().lower() == 'windows':
-            retResult = child.before
-        else:
-            retResult = child.before.decode()
-        print("expect() return code: %d, content:\n %s\n"%(i, retResult))
-        #print(child.after.decode())
-        if i == 0:
-            tdLog.exit('taos -n server fail!')
-        
-        expectString1 = 'response is received, size:' + pktLen
-        expectSTring2 = pktNum + '/' + pktNum
-        if expectString1 in retResult and expectSTring2 in retResult:
-            tdLog.info("taos -n client success")
-        else:
-            tdLog.exit('taos -n client fail!')
-
-        if platform.system().lower() == 'windows':
-            os.system('ps -a | grep taos | awk \'{print $2}\' | xargs kill -9')
-        else:
-            os.system('pkill taos')
+            if platform.system().lower() == 'windows':
+                retResult = child.before
+            else:
+                retResult = child.before.decode()
+            print("expect() return code: %d, content:\n %s\n"%(i, retResult))
+            #print(child.after.decode())
+            if i == 0:
+                tdLog.exit('taos -n server fail!')
+            
+            expectString1 = 'response is received, size:' + pktLen
+            expectSTring2 = pktNum + '/' + pktNum
+            if expectString1 in retResult and expectSTring2 in retResult:
+                tdLog.info("taos -n client success")
+            else:
+                tdLog.exit('taos -n client fail!')
+        finally:
+            if platform.system().lower() == 'windows':
+                os.system('ps -a | grep taos | awk \'{print $2}\' | xargs kill -9')
+            else:
+                os.system('pkill taos')
 
     def stop(self):
         tdSql.close()
diff --git a/tests/system-test/0-others/telemetry.py b/tests/system-test/0-others/telemetry.py
index 203f87c085..4483e113bf 100644
--- a/tests/system-test/0-others/telemetry.py
+++ b/tests/system-test/0-others/telemetry.py
@@ -8,6 +8,7 @@ import http.server
 import gzip
 import threading
 import json
+import pickle
 
 from util.log import *
 from util.sql import *
@@ -136,13 +137,19 @@ class RequestHandlerImpl(http.server.BaseHTTPRequestHandler):
         telemetryInfoCheck(infoDict)
 
         # 4. shutdown the server and exit case 
-        assassin = threading.Thread(target=httpServer.shutdown)
+        assassin = threading.Thread(target=self.server.shutdown)
         assassin.daemon = True
         assassin.start()
         print ("==== shutdown http server ====")
 
 class TDTestCase:
     hostname = socket.gethostname()
+    if (platform.system().lower() == 'windows' and not tdDnodes.dnodes[0].remoteIP == ""):
+        try:
+            config = eval(tdDnodes.dnodes[0].remoteIP)
+            hostname = config["host"]
+        except Exception:
+            hostname = tdDnodes.dnodes[0].remoteIP
     serverPort = '7080'
     rpcDebugFlagVal = '143'
     clientCfgDict = {'serverPort': '', 'firstEp': '', 'secondEp':'', 'rpcDebugFlag':'135', 'fqdn':''}
@@ -177,17 +184,20 @@ class TDTestCase:
         sql = "create database db3 vgroups " + vgroups
         tdSql.query(sql)
 
-        # loop to wait request
-        httpServer.serve_forever()
+        # create http server: bing ip/port , and  request processor
+        if (platform.system().lower() == 'windows' and not tdDnodes.dnodes[0].remoteIP == ""):
+            RequestHandlerImplStr = base64.b64encode(pickle.dumps(RequestHandlerImpl)).decode()
+            telemetryInfoCheckStr = base64.b64encode(pickle.dumps(telemetryInfoCheck)).decode()
+            cmdStr = "import pickle\nimport http\ntelemetryInfoCheck=pickle.loads(base64.b64decode(\"%s\".encode()))\nRequestHandlerImpl=pickle.loads(base64.b64decode(\"%s\".encode()))\nhttp.server.HTTPServer((\"\", %d), RequestHandlerImpl).serve_forever()"%(telemetryInfoCheckStr,RequestHandlerImplStr,int(telemetryPort))
+            tdDnodes.dnodes[0].remoteExec({}, cmdStr)
+        else:
+            serverAddress = ("", int(telemetryPort))
+            http.server.HTTPServer(serverAddress, RequestHandlerImpl).serve_forever()
 
     def stop(self):
         tdSql.close()
         tdLog.success(f"{__file__} successfully executed")
 
-# create http server: bing ip/port , and  request processor
-serverAddress = ("", int(telemetryPort))
-httpServer = http.server.HTTPServer(serverAddress, RequestHandlerImpl)
-
 tdCases.addLinux(__file__, TDTestCase())
 tdCases.addWindows(__file__, TDTestCase())
 
diff --git a/tests/system-test/0-others/udf_create.py b/tests/system-test/0-others/udf_create.py
index 88ed667161..0d24b09616 100644
--- a/tests/system-test/0-others/udf_create.py
+++ b/tests/system-test/0-others/udf_create.py
@@ -9,6 +9,8 @@ from util.sql import *
 from util.cases import *
 from util.dnodes import *
 import subprocess
+# import win32gui
+# import threading
 
 class TDTestCase:
 
@@ -533,8 +535,17 @@ class TDTestCase:
         return udf1_sqls ,udf2_sqls
 
 
+    # def checkRunTimeError(self):
+    #     while 1:
+    #         time.sleep(1)
+    #         hwnd = win32gui.FindWindow(None, "Microsoft Visual C++ Runtime Library")
+    #         if hwnd:
+    #             os.system("TASKKILL /F /IM udfd.exe")
 
     def unexpected_create(self):
+        # if (platform.system().lower() == 'windows' and tdDnodes.dnodes[0].remoteIP == ""):
+        #     checkErrorThread = threading.Thread(target=self.checkRunTimeError,daemon=True)
+        #     checkErrorThread.start()
 
         tdLog.info(" create function with out bufsize ")
         tdSql.query("drop function udf1 ")
diff --git a/tests/system-test/fulltest.bat b/tests/system-test/fulltest.bat
index ea57bc7427..c4eaeed2ad 100644
--- a/tests/system-test/fulltest.bat
+++ b/tests/system-test/fulltest.bat
@@ -1,7 +1,103 @@
 
-python3 .\test.py -f 0-others\taosShell.py
+@REM python3 .\test.py -f 0-others\taosShell.py
 python3 .\test.py -f 0-others\taosShellError.py
 python3 .\test.py -f 0-others\taosShellNetChk.py
+python3 .\test.py -f 0-others\telemetry.py
+@REM python3 .\test.py -f 0-others\taosdMonitor.py
 python3 .\test.py -f 0-others\udfTest.py
 python3 .\test.py -f 0-others\udf_create.py
-python3 .\test.py -f 0-others\udf_restart_taosd.py
\ No newline at end of file
+python3 .\test.py -f 0-others\udf_restart_taosd.py
+@REM python3 .\test.py -f 0-others\cachelast.py
+
+@REM python3 .\test.py -f 0-others\user_control.py
+@REM python3 .\test.py -f 0-others\fsync.py
+
+@REM python3 .\test.py -f 1-insert\influxdb_line_taosc_insert.py
+@REM python3 .\test.py -f 1-insert\opentsdb_telnet_line_taosc_insert.py
+@REM python3 .\test.py -f 1-insert\opentsdb_json_taosc_insert.py
+@REM #python3 .\test.py -f 1-insert\test_stmt_muti_insert_query.py
+@REM python3 .\test.py -f 1-insert\alter_stable.py
+@REM python3 .\test.py -f 1-insert\alter_table.py
+@REM python3 .\test.py -f 2-query\between.py
+@REM python3 .\test.py -f 2-query\distinct.py
+@REM python3 .\test.py -f 2-query\varchar.py
+@REM python3 .\test.py -f 2-query\ltrim.py
+@REM python3 .\test.py -f 2-query\rtrim.py
+@REM python3 .\test.py -f 2-query\length.py
+@REM python3 .\test.py -f 2-query\char_length.py
+@REM python3 .\test.py -f 2-query\upper.py
+@REM python3 .\test.py -f 2-query\lower.py
+@REM python3 .\test.py -f 2-query\join.py
+@REM python3 .\test.py -f 2-query\join2.py
+@REM python3 .\test.py -f 2-query\cast.py
+@REM python3 .\test.py -f 2-query\union.py
+@REM python3 .\test.py -f 2-query\union1.py
+@REM python3 .\test.py -f 2-query\concat.py
+@REM python3 .\test.py -f 2-query\concat2.py
+@REM python3 .\test.py -f 2-query\concat_ws.py
+@REM python3 .\test.py -f 2-query\concat_ws2.py
+@REM python3 .\test.py -f 2-query\check_tsdb.py
+@REM python3 .\test.py -f 2-query\spread.py
+@REM python3 .\test.py -f 2-query\hyperloglog.py
+
+
+@REM python3 .\test.py -f 2-query\timezone.py
+@REM python3 .\test.py -f 2-query\Now.py
+@REM python3 .\test.py -f 2-query\Today.py
+@REM python3 .\test.py -f 2-query\max.py
+@REM python3 .\test.py -f 2-query\min.py
+@REM python3 .\test.py -f 2-query\count.py
+@REM python3 .\test.py -f 2-query\last.py
+@REM python3 .\test.py -f 2-query\first.py
+@REM python3 .\test.py -f 2-query\To_iso8601.py
+@REM python3 .\test.py -f 2-query\To_unixtimestamp.py
+@REM python3 .\test.py -f 2-query\timetruncate.py
+@REM python3 .\test.py -f 2-query\diff.py
+@REM python3 .\test.py -f 2-query\Timediff.py
+
+@REM python3 .\test.py -f 2-query\top.py
+@REM python3 .\test.py -f 2-query\bottom.py
+@REM python3 .\test.py -f 2-query\percentile.py
+@REM python3 .\test.py -f 2-query\apercentile.py
+@REM python3 .\test.py -f 2-query\abs.py
+@REM python3 .\test.py -f 2-query\ceil.py
+@REM python3 .\test.py -f 2-query\floor.py
+@REM python3 .\test.py -f 2-query\round.py
+@REM python3 .\test.py -f 2-query\log.py
+@REM python3 .\test.py -f 2-query\pow.py
+@REM python3 .\test.py -f 2-query\sqrt.py
+@REM python3 .\test.py -f 2-query\sin.py
+@REM python3 .\test.py -f 2-query\cos.py
+@REM python3 .\test.py -f 2-query\tan.py
+@REM python3 .\test.py -f 2-query\arcsin.py
+@REM python3 .\test.py -f 2-query\arccos.py
+@REM python3 .\test.py -f 2-query\arctan.py
+@REM python3 .\test.py -f 2-query\query_cols_tags_and_or.py
+@REM # python3 .\test.py -f 2-query\nestedQuery.py
+@REM # TD-15983 subquery output duplicate name column. 
+@REM # Please Xiangyang Guo modify the following script
+@REM # python3 .\test.py -f 2-query\nestedQuery_str.py
+
+@REM python3 .\test.py -f 2-query\avg.py
+@REM python3 .\test.py -f 2-query\elapsed.py
+@REM python3 .\test.py -f 2-query\csum.py
+@REM python3 .\test.py -f 2-query\mavg.py
+@REM python3 .\test.py -f 2-query\diff.py
+@REM python3 .\test.py -f 2-query\sample.py
+@REM python3 .\test.py -f 2-query\function_diff.py
+@REM python3 .\test.py -f 2-query\unique.py
+@REM python3 .\test.py -f 2-query\stateduration.py
+@REM python3 .\test.py -f 2-query\function_stateduration.py
+@REM python3 .\test.py -f 2-query\statecount.py
+
+@REM python3 .\test.py -f 7-tmq\basic5.py
+@REM python3 .\test.py -f 7-tmq\subscribeDb.py
+@REM python3 .\test.py -f 7-tmq\subscribeDb0.py
+@REM python3 .\test.py -f 7-tmq\subscribeDb1.py
+@REM python3 .\test.py -f 7-tmq\subscribeStb.py
+@REM python3 .\test.py -f 7-tmq\subscribeStb0.py
+@REM python3 .\test.py -f 7-tmq\subscribeStb1.py
+@REM python3 .\test.py -f 7-tmq\subscribeStb2.py
+@REM python3 .\test.py -f 7-tmq\subscribeStb3.py
+@REM python3 .\test.py -f 7-tmq\subscribeStb4.py
+@REM python3 .\test.py -f 7-tmq\db.py
\ No newline at end of file
diff --git a/tests/system-test/test-all.bat b/tests/system-test/test-all.bat
index ae6c98b06f..baafbc9e08 100644
--- a/tests/system-test/test-all.bat
+++ b/tests/system-test/test-all.bat
@@ -11,10 +11,12 @@ set /a a=0
 @REM )
 echo Linux Taosd Test
 for /F "usebackq tokens=*" %%i in (fulltest.bat) do (
-    echo Processing %%i
-    set /a a+=1
-    call %%i ARG1 -m %1 > result_!a!.txt 2>error_!a!.txt
-    if errorlevel 1 ( call :colorEcho 0c "failed" &echo. && exit 8 ) else ( call :colorEcho 0a "Success" &echo. ) 
+    for /f "tokens=1* delims= " %%a in ("%%i") do if not "%%a" == "@REM" (
+        echo Processing %%i
+        set /a a+=1
+        call %%i ARG1 -m %1 > result_!a!.txt 2>error_!a!.txt
+        if errorlevel 1 ( call :colorEcho 0c "failed" &echo. && echo result: && cat result_!a!.txt && echo error: && cat error_!a!.txt && exit 8 ) else ( call :colorEcho 0a "Success" &echo. ) 
+    )
 )
 exit
 

From 98a9789c7e40456aef8f549d06022d7a9032e97e Mon Sep 17 00:00:00 2001
From: Xiaoyu Wang <xiaoyuwang@taosdata.com>
Date: Mon, 6 Jun 2022 19:55:28 +0800
Subject: [PATCH 32/34] fix: top function with order by clause

---
 source/libs/planner/src/planLogicCreater.c    | 2 +-
 source/libs/planner/test/planIntervalTest.cpp | 2 ++
 source/libs/planner/test/planTestUtil.cpp     | 6 +++++-
 3 files changed, 8 insertions(+), 2 deletions(-)

diff --git a/source/libs/planner/src/planLogicCreater.c b/source/libs/planner/src/planLogicCreater.c
index cda2ace5ba..329712bc28 100644
--- a/source/libs/planner/src/planLogicCreater.c
+++ b/source/libs/planner/src/planLogicCreater.c
@@ -458,7 +458,7 @@ static int32_t createAggLogicNode(SLogicPlanContext* pCxt, SSelectStmt* pSelect,
 
 static int32_t createIndefRowsFuncLogicNode(SLogicPlanContext* pCxt, SSelectStmt* pSelect, SLogicNode** pLogicNode) {
   // top/bottom are both an aggregate function and a indefinite rows function
-  if (!pSelect->hasIndefiniteRowsFunc || pSelect->hasAggFuncs) {
+  if (!pSelect->hasIndefiniteRowsFunc || pSelect->hasAggFuncs || NULL != pSelect->pWindow) {
     return TSDB_CODE_SUCCESS;
   }
 
diff --git a/source/libs/planner/test/planIntervalTest.cpp b/source/libs/planner/test/planIntervalTest.cpp
index a04f47741e..edd735922b 100644
--- a/source/libs/planner/test/planIntervalTest.cpp
+++ b/source/libs/planner/test/planIntervalTest.cpp
@@ -50,6 +50,8 @@ TEST_F(PlanIntervalTest, selectFunc) {
   run("SELECT MAX(c1), MIN(c1) FROM t1 INTERVAL(10s)");
   // select function along with the columns of select row, and with INTERVAL clause
   run("SELECT MAX(c1), c2 FROM t1 INTERVAL(10s)");
+
+  run("SELECT TOP(c1, 1) FROM t1 INTERVAL(10s) ORDER BY c1");
 }
 
 TEST_F(PlanIntervalTest, stable) {
diff --git a/source/libs/planner/test/planTestUtil.cpp b/source/libs/planner/test/planTestUtil.cpp
index e2082d4936..84c46cb9bb 100644
--- a/source/libs/planner/test/planTestUtil.cpp
+++ b/source/libs/planner/test/planTestUtil.cpp
@@ -81,6 +81,8 @@ int32_t getLogLevel() { return g_logLevel; }
 
 class PlannerTestBaseImpl {
  public:
+  PlannerTestBaseImpl() : sqlNo_(0) {}
+
   void useDb(const string& acctId, const string& db) {
     caseEnv_.acctId_ = acctId;
     caseEnv_.db_ = db;
@@ -88,6 +90,7 @@ class PlannerTestBaseImpl {
   }
 
   void run(const string& sql) {
+    ++sqlNo_;
     if (caseEnv_.nsql_ > 0) {
       --(caseEnv_.nsql_);
       return;
@@ -229,7 +232,7 @@ class PlannerTestBaseImpl {
       return;
     }
 
-    cout << "==========================================sql : [" << stmtEnv_.sql_ << "]" << endl;
+    cout << "========================================== " << sqlNo_ << " sql : [" << stmtEnv_.sql_ << "]" << endl;
 
     if (DUMP_MODULE_ALL == module || DUMP_MODULE_PARSER == module) {
       if (res_.prepareAst_.empty()) {
@@ -382,6 +385,7 @@ class PlannerTestBaseImpl {
   caseEnv caseEnv_;
   stmtEnv stmtEnv_;
   stmtRes res_;
+  int32_t sqlNo_;
 };
 
 PlannerTestBase::PlannerTestBase() : impl_(new PlannerTestBaseImpl()) {}

From b105716aac8f2aa56994812ee7c888c45010ff01 Mon Sep 17 00:00:00 2001
From: plum-lihui <huili@taosdata.com>
Date: Mon, 6 Jun 2022 20:09:51 +0800
Subject: [PATCH 33/34] test: modify test case of tmq

---
 tests/system-test/7-tmq/schema.py | 11 ++++++-----
 1 file changed, 6 insertions(+), 5 deletions(-)

diff --git a/tests/system-test/7-tmq/schema.py b/tests/system-test/7-tmq/schema.py
index 196b9a5323..51c8da2413 100644
--- a/tests/system-test/7-tmq/schema.py
+++ b/tests/system-test/7-tmq/schema.py
@@ -301,13 +301,13 @@ class TDTestCase:
         tdSql.query("alter table %s.%s modify column c4 binary(60)"%(parameterDict['dbName'], parameterDict['stbName']))
         tdSql.query("alter table %s.%s modify tag t4 binary(60)"%(parameterDict['dbName'], parameterDict['stbName']))
 
-        tdSql.query("alter table %s.%s rename column c3 c3new"%(parameterDict['dbName'], parameterDict['stbName']))
-        tdSql.query("alter table %s.%s rename column c4 c4new"%(parameterDict['dbName'], parameterDict['stbName']))
+        # tdSql.query("alter table %s.%s rename column c3 c3new"%(parameterDict['dbName'], parameterDict['stbName']))
+        # tdSql.query("alter table %s.%s rename column c4 c4new"%(parameterDict['dbName'], parameterDict['stbName']))
         tdSql.query("alter table %s.%s rename tag t3 t3new"%(parameterDict['dbName'], parameterDict['stbName']))
         tdSql.query("alter table %s.%s rename tag t4 t4new"%(parameterDict['dbName'], parameterDict['stbName']))
 
-        tdSql.query("alter table %s.%s drop column c3new"%(parameterDict['dbName'], parameterDict['stbName']))
-        tdSql.query("alter table %s.%s drop column c4new"%(parameterDict['dbName'], parameterDict['stbName']))
+        tdSql.query("alter table %s.%s drop column c3"%(parameterDict['dbName'], parameterDict['stbName']))
+        tdSql.query("alter table %s.%s drop column c4"%(parameterDict['dbName'], parameterDict['stbName']))
         tdSql.query("alter table %s.%s drop tag t3new"%(parameterDict['dbName'], parameterDict['stbName']))
         tdSql.query("alter table %s.%s drop tag t4new"%(parameterDict['dbName'], parameterDict['stbName']))
 
@@ -408,6 +408,7 @@ class TDTestCase:
         parameterDict['cfg'] = cfgPath
         
         # tdLog.info("create database, super table, child table, normal table")
+        # self.create_database(tdSql, parameterDict["dbName"])
         ntbName = 'ntb2'
         tdSql.query("create table %s.%s (ts timestamp, c1 int, c2 binary(32), c3 double, c4 binary(32), c5 nchar(10)) tags (t1 int, t2 binary(32), t3  double, t4 binary(32), t5 nchar(10))"%(parameterDict["dbName"],parameterDict["stbName"]))
         tdSql.query("create table %s.%s (ts timestamp, c1 int, c2 binary(32), c3 double, c4 binary(32), c5 nchar(10))"%(parameterDict["dbName"],ntbName))
@@ -687,7 +688,7 @@ class TDTestCase:
         tdLog.info("cfgPath: %s" % cfgPath)
 
         self.tmqCase1(cfgPath, buildPath)
-        # self.tmqCase2(cfgPath, buildPath) 
+        self.tmqCase2(cfgPath, buildPath) 
         # self.tmqCase3(cfgPath, buildPath)
 
     def stop(self):

From 26d2c9e63d105334d4bb4acc435cb3b770a17731 Mon Sep 17 00:00:00 2001
From: Xiaoyu Wang <xiaoyuwang@taosdata.com>
Date: Mon, 6 Jun 2022 20:19:04 +0800
Subject: [PATCH 34/34] fix: top function with order by clause

---
 source/libs/planner/test/planTestUtil.cpp | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/source/libs/planner/test/planTestUtil.cpp b/source/libs/planner/test/planTestUtil.cpp
index 84c46cb9bb..f5c8b58e43 100644
--- a/source/libs/planner/test/planTestUtil.cpp
+++ b/source/libs/planner/test/planTestUtil.cpp
@@ -190,6 +190,8 @@ class PlannerTestBaseImpl {
     string  acctId_;
     string  db_;
     int32_t nsql_;
+
+    caseEnv() : nsql_(0) {}
   };
 
   struct stmtEnv {
@@ -197,6 +199,7 @@ class PlannerTestBaseImpl {
     array<char, 1024> msgBuf_;
     SQuery*           pQuery_;
 
+    stmtEnv() : pQuery_(nullptr) {}
     ~stmtEnv() { qDestroyQuery(pQuery_); }
   };