Merge pull request #10317 from taosdata/feature/TD-11208-new

Feature/td-11208-new  add unique function

src/client/inc/tscUtil.h

@@ -242,7 +242,7 @@ SExprInfo* tscExprAppend(SQueryInfo* pQueryInfo, int16_t functionId, SColumnInde
                            int16_t size, int16_t resColId, int16_t interSize, bool isTagCol);
 
 SExprInfo* tscExprUpdate(SQueryInfo* pQueryInfo, int32_t index, int16_t functionId, int16_t srcColumnIndex, int16_t type,
-                           int16_t size);
+                           int32_t size);
 
 size_t     tscNumOfExprs(SQueryInfo* pQueryInfo);
 int32_t     tscExprTopBottomIndex(SQueryInfo* pQueryInfo);

src/client/src/tscGlobalmerge.c

@@ -440,6 +440,15 @@ int32_t tscCreateGlobalMergerEnv(SQueryInfo *pQueryInfo, tExtMemBuffer ***pMemBu
     rlen += pExpr->base.resBytes;
   }
 
+  int32_t pg = DEFAULT_PAGE_SIZE;
+  int32_t overhead = sizeof(tFilePage);
+  while((pg - overhead) < rlen * 2) {
+    pg *= 2;
+  }
+
+  if (*nBufferSizes < pg){
+    *nBufferSizes = 2 * pg;
+  }
   int32_t capacity = 0;
   if (rlen != 0) {
     if ((*nBufferSizes) < rlen) {
@@ -447,19 +456,13 @@ int32_t tscCreateGlobalMergerEnv(SQueryInfo *pQueryInfo, tExtMemBuffer ***pMemBu
     }
     capacity = (*nBufferSizes) / rlen;
   }
-  
+
   pModel = createColumnModel(pSchema, (int32_t)size, capacity);
   tfree(pSchema);
   if (pModel == NULL){
     return TSDB_CODE_TSC_OUT_OF_MEMORY;
   }
 
-  int32_t pg = DEFAULT_PAGE_SIZE;
-  int32_t overhead = sizeof(tFilePage);
-  while((pg - overhead) < pModel->rowSize * 2) {
-    pg *= 2;
-  }
-
   assert(numOfSub <= pTableMetaInfo->vgroupList->numOfVgroups);
   for (int32_t i = 0; i < numOfSub; ++i) {
     (*pMemBuffer)[i] = createExtMemBuffer(*nBufferSizes, rlen, pg, pModel);
@@ -593,7 +596,7 @@ static void setTagValueForMultipleRows(SQLFunctionCtx* pCtx, int32_t numOfOutput
   }
 }
 
-static void doMergeResultImpl(SMultiwayMergeInfo* pInfo, SQLFunctionCtx *pCtx, int32_t numOfExpr, int32_t rowIndex, char** pDataPtr) {
+static void doMergeResultImpl(SOperatorInfo* pInfo, SQLFunctionCtx *pCtx, int32_t numOfExpr, int32_t rowIndex, char** pDataPtr) {
   for (int32_t j = 0; j < numOfExpr; ++j) {
     pCtx[j].pInput = pDataPtr[j] + pCtx[j].inputBytes * rowIndex;
   }
@@ -605,12 +608,19 @@ static void doMergeResultImpl(SMultiwayMergeInfo* pInfo, SQLFunctionCtx *pCtx, i
     }
 
     if (functionId < 0) {
-      SUdfInfo* pUdfInfo = taosArrayGet(pInfo->udfInfo, -1 * functionId - 1);
+      SUdfInfo* pUdfInfo = taosArrayGet(((SMultiwayMergeInfo*)(pInfo->info))->udfInfo, -1 * functionId - 1);
       doInvokeUdf(pUdfInfo, &pCtx[j], 0, TSDB_UDF_FUNC_MERGE);
     } else {
       assert(!TSDB_FUNC_IS_SCALAR(functionId));
       aAggs[functionId].mergeFunc(&pCtx[j]);
     }
+
+    if (functionId == TSDB_FUNC_UNIQUE &&
+        (GET_RES_INFO(&(pCtx[j]))->numOfRes > MAX_UNIQUE_RESULT_ROWS || GET_RES_INFO(&(pCtx[j]))->numOfRes == -1)){
+      tscError("Unique result num is too large. num: %d, limit: %d",
+             GET_RES_INFO(&(pCtx[j]))->numOfRes, MAX_UNIQUE_RESULT_ROWS);
+      longjmp(pInfo->pRuntimeEnv->env, TSDB_CODE_QRY_UNIQUE_RESULT_TOO_LARGE);
+    }
   }
 }
 
@@ -644,7 +654,7 @@ static void doExecuteFinalMerge(SOperatorInfo* pOperator, int32_t numOfExpr, SSD
   for(int32_t i = 0; i < pBlock->info.rows; ++i) {
     if (pInfo->hasPrev) {
       if (needToMerge(pBlock, pInfo->orderColumnList, i, pInfo->prevRow)) {
-        doMergeResultImpl(pInfo, pCtx, numOfExpr, i, addrPtr);
+        doMergeResultImpl(pOperator, pCtx, numOfExpr, i, addrPtr);
       } else {
         doFinalizeResultImpl(pInfo, pCtx, numOfExpr);
 
@@ -656,7 +666,7 @@ static void doExecuteFinalMerge(SOperatorInfo* pOperator, int32_t numOfExpr, SSD
         for(int32_t j = 0; j < numOfExpr; ++j) {
           pCtx[j].pOutput += (pCtx[j].outputBytes * numOfRows);
           if (pCtx[j].functionId == TSDB_FUNC_TOP || pCtx[j].functionId == TSDB_FUNC_BOTTOM ||
-              pCtx[j].functionId == TSDB_FUNC_SAMPLE) {
+              pCtx[j].functionId == TSDB_FUNC_SAMPLE || pCtx[j].functionId == TSDB_FUNC_UNIQUE) {
             if(j > 0) pCtx[j].ptsOutputBuf = pCtx[j - 1].pOutput;
           }
         }
@@ -671,10 +681,10 @@ static void doExecuteFinalMerge(SOperatorInfo* pOperator, int32_t numOfExpr, SSD
           }
         }
 
-        doMergeResultImpl(pInfo, pCtx, numOfExpr, i, addrPtr);
+        doMergeResultImpl(pOperator, pCtx, numOfExpr, i, addrPtr);
       }
     } else {
-      doMergeResultImpl(pInfo, pCtx, numOfExpr, i, addrPtr);
+      doMergeResultImpl(pOperator, pCtx, numOfExpr, i, addrPtr);
     }
 
     savePrevOrderColumns(pInfo->prevRow, pInfo->orderColumnList, pBlock, i, &pInfo->hasPrev);

src/client/src/tscServer.c

@@ -1045,8 +1045,8 @@ int tscBuildQueryMsg(SSqlObj *pSql, SSqlInfo *pInfo) {
   
   SGroupbyExpr *pGroupbyExpr = query.pGroupbyExpr;
   if (pGroupbyExpr != NULL && pGroupbyExpr->numOfGroupCols > 0) {
-    pQueryMsg->orderByIdx = htons(pGroupbyExpr->orderIndex);
-    pQueryMsg->orderType = htons(pGroupbyExpr->orderType);
+    //pQueryMsg->orderByIdx = htons(pGroupbyExpr->orderIndex);
+    pQueryMsg->groupOrderType = htons(pGroupbyExpr->orderType);
 
     for (int32_t j = 0; j < pGroupbyExpr->numOfGroupCols; ++j) {
       SColIndex* pCol = taosArrayGet(pGroupbyExpr->columnInfo, j);
@@ -1947,7 +1947,6 @@ int tscProcessRetrieveGlobalMergeRsp(SSqlObj *pSql) {
   SQueryInfo *pQueryInfo = tscGetQueryInfo(pCmd);
   if (pQueryInfo->pQInfo == NULL) {
     STableGroupInfo tableGroupInfo = {.numOfTables = 1, .pGroupList = taosArrayInit(1, POINTER_BYTES),};
-    tableGroupInfo.map = taosHashInit(1, taosGetDefaultHashFunction(TSDB_DATA_TYPE_INT), true, HASH_NO_LOCK);
 
     STableKeyInfo tableKeyInfo = {.pTable = NULL, .lastKey = INT64_MIN};
 
@@ -1958,8 +1957,6 @@ int tscProcessRetrieveGlobalMergeRsp(SSqlObj *pSql) {
     tscDebug("0x%"PRIx64" create QInfo 0x%"PRIx64" to execute query processing", pSql->self, pSql->self);
     pQueryInfo->pQInfo = createQInfoFromQueryNode(pQueryInfo, &tableGroupInfo, NULL, NULL, pRes->pMerger, MERGE_STAGE, pSql->self);
     if (pQueryInfo->pQInfo == NULL) {
-      taosHashCleanup(tableGroupInfo.map);
-      taosArrayDestroy(&group);
       tscAsyncResultOnError(pSql);
       pRes->code = TSDB_CODE_QRY_OUT_OF_MEMORY;
       return pRes->code;

src/client/src/tscSubquery.c

@@ -3805,6 +3805,7 @@ void* createQInfoFromQueryNode(SQueryInfo* pQueryInfo, STableGroupInfo* pTableGr
   assert(pQueryInfo != NULL);
   SQInfo *pQInfo = (SQInfo *)calloc(1, sizeof(SQInfo));
   if (pQInfo == NULL) {
+    tsdbDestroyTableGroup(pTableGroupInfo);
     goto _cleanup;
   }
 
@@ -3913,6 +3914,7 @@ void* createQInfoFromQueryNode(SQueryInfo* pQueryInfo, STableGroupInfo* pTableGr
   int32_t code = initQInfo(&bufInfo, NULL, pSourceOperator, pQInfo, &param, NULL, 0, merger);
   taosArrayDestroy(&pa);
   if (code != TSDB_CODE_SUCCESS) {
+    pQInfo = NULL;
     goto _cleanup;
   }
 

src/client/src/tscUtil.c

@@ -74,11 +74,11 @@ int32_t converToStr(char *str, int type, void *buf, int32_t bufSize, int32_t *le
       break;
 
     case TSDB_DATA_TYPE_UINT:
-      n = sprintf(str, "%d", *(uint32_t*)buf);
+      n = sprintf(str, "%u", *(uint32_t*)buf);
       break;
 
     case TSDB_DATA_TYPE_UBIGINT:
-      n = sprintf(str, "%" PRId64, *(uint64_t*)buf);
+      n = sprintf(str, "%" PRIu64, *(uint64_t*)buf);
       break;
 
     case TSDB_DATA_TYPE_FLOAT:
@@ -304,7 +304,7 @@ bool tscNonOrderedProjectionQueryOnSTable(SQueryInfo* pQueryInfo, int32_t tableI
     return false;
   }
 
-  // order by columnIndex exists, not a non-ordered projection query
+  // order by columnIndex not exists, not a ordered projection query
   return pQueryInfo->order.orderColId < 0;
 }
 
@@ -313,7 +313,7 @@ bool tscOrderedProjectionQueryOnSTable(SQueryInfo* pQueryInfo, int32_t tableInde
     return false;
   }
 
-  // order by columnIndex exists, a non-ordered projection query
+  // order by columnIndex exists, a ordered projection query
   return pQueryInfo->order.orderColId >= 0;
 }
 
@@ -689,7 +689,8 @@ bool isSimpleAggregateRv(SQueryInfo* pQueryInfo) {
         (functionId == TSDB_FUNC_TOP || functionId == TSDB_FUNC_BOTTOM ||
          functionId == TSDB_FUNC_TS_COMP ||
          functionId == TSDB_FUNC_SAMPLE ||
-         functionId == TSDB_FUNC_HISTOGRAM)) {
+         functionId == TSDB_FUNC_HISTOGRAM ||
+         functionId == TSDB_FUNC_UNIQUE)) {
       return true;
     }
   }
@@ -1404,8 +1405,6 @@ void handleDownstreamOperator(SSqlObj** pSqlObjList, int32_t numOfUpstream, SQue
       }
     }
 
-    tableGroupInfo.map = taosHashInit(1, taosGetDefaultHashFunction(TSDB_DATA_TYPE_INT), true, HASH_NO_LOCK);
-
     STableKeyInfo tableKeyInfo = {.pTable = NULL, .lastKey = INT64_MIN};
 
     SArray* group = taosArrayInit(1, sizeof(STableKeyInfo));
@@ -2614,7 +2613,7 @@ SExprInfo* tscExprAppend(SQueryInfo* pQueryInfo, int16_t functionId, SColumnInde
 }
 
 SExprInfo* tscExprUpdate(SQueryInfo* pQueryInfo, int32_t index, int16_t functionId, int16_t srcColumnIndex,
-                           int16_t type, int16_t size) {
+                           int16_t type, int32_t size) {
   STableMetaInfo* pTableMetaInfo = tscGetMetaInfo(pQueryInfo, 0);
   SExprInfo* pExpr = tscExprGet(pQueryInfo, index);
   if (pExpr == NULL) {
@@ -2659,7 +2658,8 @@ int32_t tscExprTopBottomIndex(SQueryInfo* pQueryInfo){
     SExprInfo* pExpr = tscExprGet(pQueryInfo, i);
     if (pExpr == NULL)
       continue;
-    if (pExpr->base.functionId == TSDB_FUNC_TOP || pExpr->base.functionId == TSDB_FUNC_BOTTOM) {
+    if (pExpr->base.functionId == TSDB_FUNC_TOP || pExpr->base.functionId == TSDB_FUNC_BOTTOM
+        || pExpr->base.functionId == TSDB_FUNC_UNIQUE) {
       return i;
     }
   }
@@ -4937,7 +4937,11 @@ static int32_t createGlobalAggregateExpr(SQueryAttr* pQueryAttr, SQueryInfo* pQu
     pse->colInfo.colIndex = i;
 
     pse->colType = pExpr->base.resType;
-    pse->colBytes = pExpr->base.resBytes;
+    if(pExpr->base.resBytes > INT16_MAX && pExpr->base.functionId == TSDB_FUNC_UNIQUE){
+      pQueryAttr->interBytesForGlobal = pExpr->base.resBytes;
+    }else{
+      pse->colBytes = pExpr->base.resBytes;
+    }
   }
 
   {
@@ -5081,6 +5085,7 @@ int32_t tscCreateQueryFromQueryInfo(SQueryInfo* pQueryInfo, SQueryAttr* pQueryAt
   pQueryAttr->pUdfInfo          = pQueryInfo->pUdfInfo;
   pQueryAttr->range             = pQueryInfo->range;
 
+
   if (pQueryInfo->order.order == TSDB_ORDER_ASC) {   // TODO refactor
     pQueryAttr->window = pQueryInfo->window;
   } else {
@@ -5112,6 +5117,8 @@ int32_t tscCreateQueryFromQueryInfo(SQueryInfo* pQueryInfo, SQueryAttr* pQueryAt
     }
   }
 
+  pQueryAttr->uniqueQuery       = isUniqueQuery(numOfOutput, pQueryAttr->pExpr1);
+
   pQueryAttr->tableCols = calloc(numOfCols, sizeof(SColumnInfo));
   for(int32_t i = 0; i < numOfCols; ++i) {
     SColumn* pCol = taosArrayGetP(pQueryInfo->colList, i);
@@ -5403,7 +5410,7 @@ int parseJsontoTagData(char* json, SKVRowBuilder* kvRowBuilder, char* errMsg, in
   // set json real data
   cJSON *root = cJSON_Parse(json);
   if (root == NULL){
-    tscError("json parse error");
+    tscError("json parse error:%s", json);
     return tscSQLSyntaxErrMsg(errMsg, "json parse error", NULL);
   }
 

src/common/inc/tname.h

@@ -54,7 +54,7 @@ typedef struct SSqlExpr {
   int32_t   resBytes;       // length of return value
   int32_t   interBytes;     // inter result buffer size
 
-  int16_t   colType;        // table column type
+  int16_t   colType;        // table column type, this should be int32_t, because it is too small for globale merge stage, pQueryAttr->interBytesForGlobal
   int16_t   colBytes;       // table column bytes
 
   int16_t   numOfParams;    // argument value of each function

src/inc/taoserror.h

@@ -293,6 +293,7 @@ int32_t* taosGetErrno();
 #define TSDB_CODE_QRY_SYS_ERROR                 TAOS_DEF_ERROR_CODE(0, 0x070D)  //"System error")
 #define TSDB_CODE_QRY_INVALID_TIME_CONDITION    TAOS_DEF_ERROR_CODE(0, 0x070E)  //"invalid time condition")
 #define TSDB_CODE_QRY_INVALID_SCHEMA_VERSION    TAOS_DEF_ERROR_CODE(0, 0x0710)  //"invalid schema version")
+#define TSDB_CODE_QRY_UNIQUE_RESULT_TOO_LARGE   TAOS_DEF_ERROR_CODE(0, 0x0711)  //"unique result num is too large")
 
 // grant
 #define TSDB_CODE_GRANT_EXPIRED                 TAOS_DEF_ERROR_CODE(0, 0x0800)  //"License expired"

src/inc/taosmsg.h

@@ -503,8 +503,8 @@ typedef struct {
   uint32_t    tagCondLen;       // tag length in current query
   int32_t    colCondLen;       // column length in current query
   int16_t     numOfGroupCols;   // num of group by columns
-  int16_t     orderByIdx;
-  int16_t     orderType;        // used in group by xx order by xxx
+  int16_t     orderByIdx;       // useless
+  int16_t     groupOrderType;   // used for group order
   int64_t     vgroupLimit;      // limit the number of rows for each table, used in order by + limit in stable projection query.
   int16_t     prjOrder;         // global order in super table projection query.
   int64_t     limit;

src/query/inc/qAggMain.h

@@ -78,8 +78,9 @@ extern "C" {
 
 #define TSDB_FUNC_ELAPSED      37
 #define TSDB_FUNC_HISTOGRAM    38
+#define TSDB_FUNC_UNIQUE       39
 
-#define TSDB_FUNC_MAX_NUM    39
+#define TSDB_FUNC_MAX_NUM    40
 
 #define TSDB_FUNCSTATE_SO           0x1u    // single output
 #define TSDB_FUNCSTATE_MO           0x2u    // dynamic number of output, not multinumber of output e.g., TOP/BOTTOM
@@ -174,7 +175,7 @@ typedef struct SQLFunctionCtx {
   void *       pInput;    // input data buffer
   uint32_t     order;     // asc|desc
   int16_t      inputType;
-  int16_t      inputBytes;
+  int32_t      inputBytes;
   
   int16_t      outputType;
   int32_t      outputBytes;   // size of results, determined by function and input column data type
@@ -200,6 +201,8 @@ typedef struct SQLFunctionCtx {
   SExtTagsInfo tagInfo;
   SPoint1      start;
   SPoint1      end;
+
+  SHashObj     **pUniqueSet;   // for unique function
 } SQLFunctionCtx;
 
 typedef struct SAggFunctionInfo {
@@ -249,7 +252,7 @@ void blockDistInfoToBinary(STableBlockDist* pDist, struct SBufferWriter* bw);
 void blockDistInfoFromBinary(const char* data, int32_t len, STableBlockDist* pDist);
 
 /* global sql function array */
-extern struct SAggFunctionInfo aAggs[40];
+extern struct SAggFunctionInfo aAggs[TSDB_FUNC_MAX_NUM];
 
 extern int32_t functionCompatList[]; // compatible check array list
 

src/query/inc/qExecutor.h

@@ -90,6 +90,7 @@ typedef struct SResultRow {
   SResultRowCellInfo*  pCellInfo;  // For each result column, there is a resultInfo
   STimeWindow   win;
   char         *key;               // start key of current result row
+  SHashObj     *uniqueHash;  // for unique function
 } SResultRow;
 
 typedef struct SResultRowCell {
@@ -221,6 +222,7 @@ typedef struct SQueryAttr {
 
   bool             stableQuery;      // super table query or not
   bool             topBotQuery;      // TODO used bitwise flag
+  bool             uniqueQuery;
   bool             groupbyColumn;    // denote if this is a groupby normal column query
   bool             hasTagResults;    // if there are tag values in final result or not
   bool             timeWindowInterpo;// if the time window start/end required interpolation
@@ -281,6 +283,7 @@ typedef struct SQueryAttr {
   STableGroupInfo  tableGroupInfo;       // table <tid, last_key> list  SArray<STableKeyInfo>
   int32_t          vgId;
   SArray          *pUdfInfo;             // no need to free
+  int32_t          interBytesForGlobal;
 } SQueryAttr;
 
 typedef SSDataBlock* (*__operator_fn_t)(void* param, bool* newgroup);
@@ -730,4 +733,5 @@ void addTableReadRows(SQueryRuntimeEnv* pEnv, int32_t tid, int32_t rows);
 // tsdb scan table callback table or query is over. param is SQueryRuntimeEnv*
 bool qReadOverCB(void* param, int8_t type, int32_t tid);
 
+bool isUniqueQuery(int32_t numOfOutput, SExprInfo* pExprs);
 #endif  // TDENGINE_QEXECUTOR_H

src/query/inc/qExtbuffer.h

@@ -53,14 +53,14 @@ typedef struct tFlushoutInfo {
 } tFlushoutInfo;
 
 typedef struct tFlushoutData {
-  uint32_t       nAllocSize;
-  uint32_t       nLength;
-  tFlushoutInfo *pFlushoutInfo;
+  uint32_t       nAllocSize;    // capacity
+  uint32_t       nLength;       // size
+  tFlushoutInfo *pFlushoutInfo; // dynamic allocate
 } tFlushoutData;
 
 typedef struct SExtFileInfo {
-  uint32_t      nFileSize;  // in pages
-  uint32_t      pageSize;
+  uint32_t      nFileSize;  // how many pages in file
+  //uint32_t      pageSize; // useless
   uint32_t      numOfElemsInFile;
   tFlushoutData flushoutData;
 } SExtFileInfo;

src/query/inc/qResultbuf.h

@@ -78,7 +78,8 @@ typedef struct SDiskbasedResultBuf {
 
 #define DEFAULT_INTERN_BUF_PAGE_SIZE  (1024L)                          // in bytes
 #define PAGE_INFO_INITIALIZER         (SPageDiskInfo){-1, -1}
-
+#define MAX_UNIQUE_RESULT_ROWS (1000)
+#define MAX_UNIQUE_RESULT_SIZE (1024*1024*1)
 /**
  * create disk-based result buffer
  * @param pResultBuf

src/query/inc/qTableMeta.h

@@ -50,7 +50,7 @@ typedef struct SGroupbyExpr {
   int16_t tableIndex;
   SArray* columnInfo;  // SArray<SColIndex>, group by columns information
   int16_t numOfGroupCols;  // todo remove it
-  int16_t orderIndex;  // order by column index
+  //int16_t orderIndex; // order by column index, rm useless orderIndex
   int16_t orderType;   // order by type: asc/desc
 } SGroupbyExpr;
 

src/query/inc/qUtil.h

@@ -70,8 +70,12 @@ static FORCE_INLINE char* getPosInResultPage(SQueryAttr* pQueryAttr, tFilePage*
                                              int32_t offset) {
   assert(rowOffset >= 0 && pQueryAttr != NULL);
 
-  int32_t numOfRows = (int32_t)getRowNumForMultioutput(pQueryAttr, pQueryAttr->topBotQuery, pQueryAttr->stableQuery);
-  return ((char *)page->data) + rowOffset + offset * numOfRows;
+  int64_t numOfRows = (int64_t)getRowNumForMultioutput(pQueryAttr, pQueryAttr->topBotQuery, pQueryAttr->stableQuery);
+  numOfRows *= offset;
+  if(numOfRows >= INT32_MAX){
+    assert(0);
+  }
+  return ((char *)page->data) + rowOffset + numOfRows;
 }
 
 bool isNullOperator(SColumnFilterElem *pFilter, const char* minval, const char* maxval, int16_t type);

src/query/src/qAggMain.c

@@ -223,6 +223,16 @@ typedef struct{
   SHistogramFuncBin* orderedBins;
 } SHistogramFuncInfo;
 
+typedef struct {
+  int64_t  timestamp;
+  char     data[];
+} UniqueUnit;
+
+typedef struct {
+  int32_t      num;
+  char         res[];
+} SUniqueFuncInfo;
+
 int32_t getResultDataInfo(int32_t dataType, int32_t dataBytes, int32_t functionId, int32_t param, int16_t *type,
                           int32_t *bytes, int32_t *interBytes, int16_t extLength, bool isSuperTable, SUdfInfo* pUdfInfo) {
   if (!isValidDataType(dataType)) {
@@ -353,6 +363,18 @@ int32_t getResultDataInfo(int32_t dataType, int32_t dataBytes, int32_t functionI
       *bytes = (sizeof(STopBotInfo) + (sizeof(tValuePair) + POINTER_BYTES + extLength) * param);
       *interBytes = *bytes;
 
+      return TSDB_CODE_SUCCESS;
+    } else if (functionId == TSDB_FUNC_UNIQUE) {
+      *type = TSDB_DATA_TYPE_BINARY;
+      int64_t size = sizeof(UniqueUnit) + dataBytes + extLength;
+      size *= param;
+      size += sizeof(SUniqueFuncInfo);
+      if (size > MAX_UNIQUE_RESULT_SIZE){
+        size = MAX_UNIQUE_RESULT_SIZE;
+      }
+      *bytes = size;
+      *interBytes = *bytes;
+
       return TSDB_CODE_SUCCESS;
     } else if (functionId == TSDB_FUNC_SAMPLE) {
       *type = TSDB_DATA_TYPE_BINARY;
@@ -477,10 +499,20 @@ int32_t getResultDataInfo(int32_t dataType, int32_t dataBytes, int32_t functionI
     *type = (int16_t)dataType;
     *bytes = dataBytes;
 
-    size_t size = sizeof(STopBotInfo) + (sizeof(tValuePair) + POINTER_BYTES + extLength) * param;
+    size_t size = sizeof(STopBotInfo) + (sizeof(tValuePair) + extLength) * param;
 
     // the output column may be larger than sizeof(STopBotInfo)
     *interBytes = (int32_t)size;
+  } else if (functionId == TSDB_FUNC_UNIQUE) {
+    *type = (int16_t)dataType;
+    *bytes = dataBytes;
+    int64_t size = sizeof(UniqueUnit) + dataBytes + extLength;
+    size *= param;
+    size += sizeof(SUniqueFuncInfo);
+    if (size > MAX_UNIQUE_RESULT_SIZE){
+      size = MAX_UNIQUE_RESULT_SIZE;
+    }
+    *interBytes = (int32_t)size;
   } else if (functionId == TSDB_FUNC_SAMPLE) {
       *type = (int16_t)dataType;
       *bytes = dataBytes;
@@ -2130,7 +2162,7 @@ static void copyTopBotRes(SQLFunctionCtx *pCtx, int32_t type) {
   
   int32_t step = QUERY_ASC_FORWARD_STEP;
   int32_t len = (int32_t)(GET_RES_INFO(pCtx)->numOfRes);
-  
+
   switch (type) {
     case TSDB_DATA_TYPE_UINT:
     case TSDB_DATA_TYPE_INT: {
@@ -5097,6 +5129,194 @@ static void histogram_func_finalizer(SQLFunctionCtx *pCtx) {
   doFinalizer(pCtx);
 }
 
+// unique use the intermediate result buffer to keep the intermediate result
+static SUniqueFuncInfo *getUniqueOutputInfo(SQLFunctionCtx *pCtx) {
+  SResultRowCellInfo *pResInfo = GET_RES_INFO(pCtx);
+
+  // only the first_stage_merge is directly written data into final output buffer
+  if (pCtx->stableQuery && pCtx->currentStage != MERGE_STAGE) {
+    return (SUniqueFuncInfo*) pCtx->pOutput;
+  } else { // during normal table query and super table at the secondary_stage, result is written to intermediate buffer
+    return GET_ROWCELL_INTERBUF(pResInfo);
+  }
+}
+
+// unique
+static void copyUniqueRes(SQLFunctionCtx *pCtx, int32_t bytes) {
+  SResultRowCellInfo *pResInfo = GET_RES_INFO(pCtx);
+  SUniqueFuncInfo *pRes = GET_ROWCELL_INTERBUF(pResInfo);
+
+  size_t size = sizeof(UniqueUnit) + bytes + pCtx->tagInfo.tagsLen;
+  int32_t len = (int32_t)(GET_RES_INFO(pCtx)->numOfRes);
+
+  char *tsOutput = pCtx->ptsOutputBuf;
+  char *output = pCtx->pOutput;
+  int32_t step = GET_FORWARD_DIRECTION_FACTOR(pCtx->param[2].i64);
+  char *tvp = pRes->res + (size * ((pCtx->param[2].i64 == TSDB_ORDER_ASC) ? 0 : len -1));
+  for (int32_t i = 0; i < len; ++i) {
+    memcpy(tsOutput, tvp, sizeof(int64_t));
+    memcpy(output, tvp + sizeof(UniqueUnit), bytes);
+    tvp += (step * size);
+    tsOutput += sizeof(int64_t);
+    output += bytes;
+  }
+
+  // set the corresponding tag data for each record
+  // todo check malloc failure
+  if (pCtx->tagInfo.numOfTagCols == 0) {
+    return ;
+  }
+
+  char **pData = calloc(pCtx->tagInfo.numOfTagCols, POINTER_BYTES);
+  for (int32_t i = 0; i < pCtx->tagInfo.numOfTagCols; ++i) {
+    pData[i] = pCtx->tagInfo.pTagCtxList[i]->pOutput;
+  }
+
+  tvp = pRes->res + (size * ((pCtx->param[2].i64 == TSDB_ORDER_ASC) ? 0 : len -1));
+  for (int32_t i = 0; i < len; ++i) {
+    int16_t offset = sizeof(UniqueUnit) + bytes;
+    for (int32_t j = 0; j < pCtx->tagInfo.numOfTagCols; ++j) {
+      memcpy(pData[j], tvp + offset, (size_t)pCtx->tagInfo.pTagCtxList[j]->outputBytes);
+      offset += pCtx->tagInfo.pTagCtxList[j]->outputBytes;
+      pData[j] += pCtx->tagInfo.pTagCtxList[j]->outputBytes;
+    }
+    tvp += (step * size);
+  }
+
+  tfree(pData);
+}
+
+static bool unique_function_setup(SQLFunctionCtx *pCtx, SResultRowCellInfo* pResInfo) {
+  if (!function_setup(pCtx, pResInfo)) {
+    return false;
+  }
+  if(*pCtx->pUniqueSet != NULL){
+    taosHashClear(*pCtx->pUniqueSet);
+  }else{
+    *pCtx->pUniqueSet = taosHashInit(64, taosGetDefaultHashFunction(TSDB_DATA_TYPE_BINARY), true, HASH_NO_LOCK);
+  }
+
+  return true;
+}
+
+static void do_unique_function(SQLFunctionCtx *pCtx, SUniqueFuncInfo *pInfo, TSKEY timestamp, char *pData, char *tag, int32_t bytes, int16_t type){
+  int32_t hashKeyBytes = bytes;
+  if(IS_VAR_DATA_TYPE(type)){     // for var data, we can not use bytes, because there are dirty data in the back of var data
+    hashKeyBytes = varDataTLen(pData);
+  }
+  UniqueUnit **unique = taosHashGet(*pCtx->pUniqueSet, pData, hashKeyBytes);
+  if (unique == NULL) {
+    size_t size = sizeof(UniqueUnit) + bytes + pCtx->tagInfo.tagsLen;
+    char *tmp = pInfo->res + pInfo->num * size;
+    ((UniqueUnit*)tmp)->timestamp = timestamp;
+    char *data = tmp + sizeof(UniqueUnit);
+    char *tags = tmp + sizeof(UniqueUnit) + bytes;
+    memcpy(data, pData, bytes);
+
+    if (pCtx->currentStage == MERGE_STAGE && tag != NULL) {
+      memcpy(tags, tag, (size_t)pCtx->tagInfo.tagsLen);
+    }else{
+      int32_t offset = 0;
+      for (int32_t j = 0; j < pCtx->tagInfo.numOfTagCols; ++j) {
+        SQLFunctionCtx *tagCtx = pCtx->tagInfo.pTagCtxList[j];
+        if (tagCtx->functionId == TSDB_FUNC_TS_DUMMY) {
+          tagCtx->tag.nType = TSDB_DATA_TYPE_BIGINT;
+          tagCtx->tag.i64 = timestamp;
+        }
+
+        tVariantDump(&tagCtx->tag, tagCtx->pOutput, tagCtx->tag.nType, true);
+        memcpy(tags + offset, tagCtx->pOutput, tagCtx->outputBytes);
+        offset += tagCtx->outputBytes;
+      }
+    }
+
+    taosHashPut(*pCtx->pUniqueSet, pData, hashKeyBytes, &tmp, sizeof(UniqueUnit*));
+    pInfo->num++;
+  }else if((*unique)->timestamp > timestamp){
+    (*unique)->timestamp = timestamp;
+  }
+}
+
+static void unique_function(SQLFunctionCtx *pCtx) {
+  SUniqueFuncInfo *pInfo = getUniqueOutputInfo(pCtx);
+
+  for (int32_t i = 0; i < pCtx->size; i++) {
+    char *pData = GET_INPUT_DATA(pCtx, i);
+    TSKEY k = 0;
+    if (pCtx->ptsList != NULL) {
+      k = GET_TS_DATA(pCtx, i);
+    }
+    do_unique_function(pCtx, pInfo, k, pData, NULL, pCtx->inputBytes, pCtx->inputType);
+
+    if (sizeof(SUniqueFuncInfo) + pInfo->num * (sizeof(UniqueUnit) + pCtx->inputBytes + pCtx->tagInfo.tagsLen) >= MAX_UNIQUE_RESULT_SIZE){
+      GET_RES_INFO(pCtx)->numOfRes = -1;    // mark out of memory
+      return;
+    }
+  }
+
+  GET_RES_INFO(pCtx)->numOfRes = 1;
+}
+
+static void unique_function_merge(SQLFunctionCtx *pCtx) {
+  SUniqueFuncInfo *pInput = (SUniqueFuncInfo *)GET_INPUT_DATA_LIST(pCtx);
+  SUniqueFuncInfo *pOutput = getUniqueOutputInfo(pCtx);
+  size_t size = sizeof(UniqueUnit) + pCtx->outputBytes + pCtx->tagInfo.tagsLen;
+  for (int32_t i = 0; i < pInput->num; ++i) {
+    char *tmp = pInput->res + i* size;
+    TSKEY timestamp = ((UniqueUnit*)tmp)->timestamp;
+    char *data = tmp + sizeof(UniqueUnit);
+    char *tags = tmp + sizeof(UniqueUnit) + pCtx->outputBytes;
+    do_unique_function(pCtx, pOutput, timestamp, data, tags, pCtx->outputBytes, pCtx->outputType);
+
+    if (sizeof(SUniqueFuncInfo) + pOutput->num * (sizeof(UniqueUnit) + pCtx->outputBytes + pCtx->tagInfo.tagsLen) >= MAX_UNIQUE_RESULT_SIZE){
+      GET_RES_INFO(pCtx)->numOfRes = -1;    // mark out of memory
+      return;
+    }
+  }
+
+  GET_RES_INFO(pCtx)->numOfRes = pOutput->num;
+}
+
+typedef struct{
+  int32_t dataOffset;
+  __compar_fn_t comparFn;
+} UiqueSupporter;
+
+static int32_t uniqueCompareFn(const void *p1, const void *p2, const void *param) {
+  UiqueSupporter *support = (UiqueSupporter *)param;
+  return support->comparFn(p1 + support->dataOffset, p2 + support->dataOffset);
+}
+
+static void unique_func_finalizer(SQLFunctionCtx *pCtx) {
+  SUniqueFuncInfo *pInfo = getUniqueOutputInfo(pCtx);
+
+  GET_RES_INFO(pCtx)->numOfRes = pInfo->num;
+  int32_t bytes = 0;
+  int32_t type = 0;
+  if (pCtx->currentStage == MERGE_STAGE) {
+    bytes = pCtx->outputBytes;
+    type = pCtx->outputType;
+    assert(pCtx->inputType == TSDB_DATA_TYPE_BINARY);
+  } else {
+    bytes = pCtx->inputBytes;
+    type = pCtx->inputType;
+  }
+  UiqueSupporter support = {0};
+  // user specify the order of output by sort the result according to timestamp
+  if (pCtx->param[1].i64 == PRIMARYKEY_TIMESTAMP_COL_INDEX) {
+    support.dataOffset = 0;
+    support.comparFn = compareInt64Val;
+  } else{
+    support.dataOffset = sizeof(UniqueUnit);
+    support.comparFn = getComparFunc(type, 0);
+  }
+
+  size_t size = sizeof(UniqueUnit) + bytes + pCtx->tagInfo.tagsLen;
+  taosqsort(pInfo->res, (size_t)GET_RES_INFO(pCtx)->numOfRes, size, &support, uniqueCompareFn);
+  copyUniqueRes(pCtx, bytes);
+  doFinalizer(pCtx);
+}
+
 /////////////////////////////////////////////////////////////////////////////////////////////
 /*
  * function compatible list.
@@ -5117,11 +5337,11 @@ int32_t functionCompatList[] = {
     1,          1,        1,         1,       -1,      1,          1,           1,          5,          1,      1,
     // tid_tag, deriv,    csum,       mavg,        sample,
     6,          8,        -1,         -1,          -1,
-    // block_info,elapsed,histogram
-    7,          1,        -1
+    // block_info,elapsed,histogram,unique
+    7,          1,        -1,        -1
 };
 
-SAggFunctionInfo aAggs[40] = {{
+SAggFunctionInfo aAggs[TSDB_FUNC_MAX_NUM] = {{
                               // 0, count function does not invoke the finalize function
                               "count",
                               TSDB_FUNC_COUNT,
@@ -5591,5 +5811,17 @@ SAggFunctionInfo aAggs[40] = {{
                               histogram_func_finalizer,
                               histogram_func_merge,
                               dataBlockRequired,
+                          },
+                          {
+                              // 39
+                              "unique",
+                              TSDB_FUNC_UNIQUE,
+                              TSDB_FUNC_UNIQUE,
+                              TSDB_FUNCSTATE_MO | TSDB_FUNCSTATE_STABLE | TSDB_FUNCSTATE_SELECTIVITY,
+                              unique_function_setup,
+                              unique_function,
+                              unique_func_finalizer,
+                              unique_function_merge,
+                              dataBlockRequired,
                           }
 };

src/query/src/qExecutor.c

@@ -281,7 +281,7 @@ static int compareRowData(const void *a, const void *b, const void *userData) {
   tFilePage *page1 = getResBufPage(pRuntimeEnv->pResultBuf, pRow1->pageId);
   tFilePage *page2 = getResBufPage(pRuntimeEnv->pResultBuf, pRow2->pageId);
 
-  int16_t offset = supporter->dataOffset;
+  int32_t offset = supporter->dataOffset;
   char *in1  = getPosInResultPage(pRuntimeEnv->pQueryAttr, page1, pRow1->offset, offset);
   char *in2  = getPosInResultPage(pRuntimeEnv->pQueryAttr, page2, pRow2->offset, offset);
 
@@ -289,9 +289,8 @@ static int compareRowData(const void *a, const void *b, const void *userData) {
 }
 
 static void sortGroupResByOrderList(SGroupResInfo *pGroupResInfo, SQueryRuntimeEnv *pRuntimeEnv, SSDataBlock* pDataBlock, SQLFunctionCtx *pCtx) {
-  SArray *columnOrderList = getOrderCheckColumns(pRuntimeEnv->pQueryAttr);
-  size_t size = taosArrayGetSize(columnOrderList);
-  taosArrayDestroy(&columnOrderList);
+  int32_t size = pRuntimeEnv->pQueryAttr->pGroupbyExpr == NULL? 0: pRuntimeEnv->pQueryAttr->pGroupbyExpr->numOfGroupCols;
+  if (pRuntimeEnv->pQueryAttr->interval.interval > 0) size++;
 
   if (size <= 0) {
     return;
@@ -357,7 +356,13 @@ SSDataBlock* createOutputBuf(SExprInfo* pExpr, int32_t numOfOutput, int32_t numO
     idata.info.bytes = pExpr[i].base.resBytes;
     idata.info.colId = pExpr[i].base.resColId;
 
-    int32_t size = MAX(idata.info.bytes * numOfRows, minSize);
+    int64_t tmp = idata.info.bytes;
+    tmp *= numOfRows;
+    if (tmp >= 1024*1024*1024) {   // 1G
+      qError("size is too large, failed to allocate column buffer for output buffer");
+      tmp = 128*1024*1024;
+    }
+    int32_t size = MAX(tmp, minSize);
     idata.pData = calloc(1, size);  // at least to hold a pointer on x64 platform
     if (idata.pData == NULL) {
       qError("failed to allocate column buffer for output buffer");
@@ -1004,6 +1009,13 @@ static void doApplyFunctions(SQueryRuntimeEnv* pRuntimeEnv, SQLFunctionCtx* pCtx
       }
     }
 
+    if (functionId == TSDB_FUNC_UNIQUE &&
+        (GET_RES_INFO(&(pCtx[k]))->numOfRes > MAX_UNIQUE_RESULT_ROWS || GET_RES_INFO(&(pCtx[k]))->numOfRes == -1)){
+      qError("Unique result num is too large. num: %d, limit: %d",
+             GET_RES_INFO(&(pCtx[k]))->numOfRes, MAX_UNIQUE_RESULT_ROWS);
+      longjmp(pRuntimeEnv->env, TSDB_CODE_QRY_UNIQUE_RESULT_TOO_LARGE);
+    }
+
     // restore it
     pCtx[k].preAggVals.isSet = hasAggregates;
     pCtx[k].pInput = start;
@@ -1263,6 +1275,13 @@ static void doAggregateImpl(SOperatorInfo* pOperator, TSKEY startTs, SQLFunction
       } else {
         assert(0);
       }
+
+      if (functionId == TSDB_FUNC_UNIQUE &&
+          (GET_RES_INFO(&(pCtx[k]))->numOfRes > MAX_UNIQUE_RESULT_ROWS || GET_RES_INFO(&(pCtx[k]))->numOfRes == -1)){
+        qError("Unique result num is too large. num: %d, limit: %d",
+               GET_RES_INFO(&(pCtx[k]))->numOfRes, MAX_UNIQUE_RESULT_ROWS);
+        longjmp(pRuntimeEnv->env, TSDB_CODE_QRY_UNIQUE_RESULT_TOO_LARGE);
+      }
     }
   }
 }
@@ -1893,7 +1912,7 @@ static int32_t setCtxTagColumnInfo(SQLFunctionCtx *pCtx, int32_t numOfOutput) {
       continue;
     }
 
-    if (functionId == TSDB_FUNC_TAG_DUMMY || functionId == TSDB_FUNC_TS_DUMMY) {
+    if (functionId == TSDB_FUNC_TAG_DUMMY || functionId == TSDB_FUNC_TS_DUMMY) { //ts_select ts,top(col,2)
       tagLen += pCtx[i].outputBytes;
       pTagCtx[num++] = &pCtx[i];
     } else if ((aAggs[functionId].status & TSDB_FUNCSTATE_SELECTIVITY) != 0) {
@@ -1945,8 +1964,12 @@ static SQLFunctionCtx* createSQLFunctionCtx(SQueryRuntimeEnv* pRuntimeEnv, SExpr
       pCtx->requireNull = false;
     }
 
-    pCtx->inputBytes = pSqlExpr->colBytes;
     pCtx->inputType  = pSqlExpr->colType;
+    if (pRuntimeEnv->pQueryAttr->interBytesForGlobal > INT16_MAX && pSqlExpr->functionId == TSDB_FUNC_UNIQUE){
+      pCtx->inputBytes  = pRuntimeEnv->pQueryAttr->interBytesForGlobal;
+    }else{
+      pCtx->inputBytes = pSqlExpr->colBytes;
+    }
 
     pCtx->ptsOutputBuf = NULL;
 
@@ -1980,7 +2003,8 @@ static SQLFunctionCtx* createSQLFunctionCtx(SQueryRuntimeEnv* pRuntimeEnv, SExpr
     // set the order information for top/bottom query
     int32_t functionId = pCtx->functionId;
 
-    if (functionId == TSDB_FUNC_TOP || functionId == TSDB_FUNC_BOTTOM || functionId == TSDB_FUNC_DIFF) {
+    if (functionId == TSDB_FUNC_TOP || functionId == TSDB_FUNC_BOTTOM
+        || functionId == TSDB_FUNC_DIFF || functionId == TSDB_FUNC_UNIQUE) {
       int32_t f = pExpr[i-1].base.functionId;
       assert(f == TSDB_FUNC_TS || f == TSDB_FUNC_TS_DUMMY);
 
@@ -3165,7 +3189,7 @@ int32_t loadDataBlockOnDemand(SQueryRuntimeEnv* pRuntimeEnv, STableScanInfo* pTa
   if ((*status) != BLK_DATA_ALL_NEEDED) {
     // the pCtx[i] result is belonged to previous time window since the outputBuf has not been set yet,
     // the filter result may be incorrect. So in case of interval query, we need to set the correct time output buffer
-    if (QUERY_IS_INTERVAL_QUERY(pQueryAttr) && (!pQueryAttr->pointInterpQuery)) {
+    if (QUERY_IS_INTERVAL_QUERY(pQueryAttr) && (!pQueryAttr->pointInterpQuery) && (!pQueryAttr->uniqueQuery)) {
       SResultRow* pResult = NULL;
 
       bool  masterScan = IS_MASTER_SCAN(pRuntimeEnv);
@@ -3177,7 +3201,7 @@ int32_t loadDataBlockOnDemand(SQueryRuntimeEnv* pRuntimeEnv, STableScanInfo* pTa
                                   pTableScanInfo->rowCellInfoOffset) != TSDB_CODE_SUCCESS) {
         longjmp(pRuntimeEnv->env, TSDB_CODE_QRY_OUT_OF_MEMORY);
       }
-    } else if (pQueryAttr->stableQuery && (!pQueryAttr->tsCompQuery) && (!pQueryAttr->diffQuery) && (!pQueryAttr->pointInterpQuery)) { // stable aggregate, not interval aggregate or normal column aggregate
+    } else if (pQueryAttr->stableQuery && (!pQueryAttr->tsCompQuery) && (!pQueryAttr->diffQuery) && (!pQueryAttr->pointInterpQuery) && (!pQueryAttr->uniqueQuery)) { // stable aggregate, not interval aggregate or normal column aggregate
       doSetTableGroupOutputBuf(pRuntimeEnv, pTableScanInfo->pResultRowInfo, pTableScanInfo->pCtx,
                                pTableScanInfo->rowCellInfoOffset, pTableScanInfo->numOfOutput,
                                pRuntimeEnv->current->groupIndex);
@@ -3664,6 +3688,9 @@ void setDefaultOutputBuf(SQueryRuntimeEnv *pRuntimeEnv, SOptrBasicInfo *pInfo, i
     RESET_RESULT_INFO(pCellInfo);
 
     pCtx[i].resultInfo   = pCellInfo;
+    if (pCtx[i].functionId == TSDB_FUNC_UNIQUE) {
+      pCtx[i].pUniqueSet = &pRow->uniqueHash;
+    }
     pCtx[i].pOutput      = pData->pData;
     pCtx[i].currentStage = stage;
     assert(pCtx[i].pOutput != NULL);
@@ -3671,7 +3698,7 @@ void setDefaultOutputBuf(SQueryRuntimeEnv *pRuntimeEnv, SOptrBasicInfo *pInfo, i
     // set the timestamp output buffer for top/bottom/diff query
     int32_t fid = pCtx[i].functionId;
     if (fid == TSDB_FUNC_TOP || fid == TSDB_FUNC_BOTTOM || fid == TSDB_FUNC_DIFF || fid == TSDB_FUNC_DERIVATIVE ||
-        fid == TSDB_FUNC_SAMPLE || fid == TSDB_FUNC_MAVG || fid == TSDB_FUNC_CSUM) {
+        fid == TSDB_FUNC_SAMPLE || fid == TSDB_FUNC_MAVG || fid == TSDB_FUNC_CSUM || fid == TSDB_FUNC_UNIQUE) {
       if (i > 0) pCtx[i].ptsOutputBuf = pCtx[i-1].pOutput;
     } else if (fid == TSDB_FUNC_INTERP) {
       assert(pCtx[0].functionId == TSDB_FUNC_TS_DUMMY || pCtx[0].functionId == TSDB_FUNC_TS);
@@ -3742,7 +3769,7 @@ void updateOutputBuf(SOptrBasicInfo* pBInfo, int32_t *bufCapacity, int32_t numOf
     if (functionId == TSDB_FUNC_TOP || functionId == TSDB_FUNC_BOTTOM ||
         functionId == TSDB_FUNC_DIFF || functionId == TSDB_FUNC_DERIVATIVE ||
         functionId == TSDB_FUNC_CSUM || functionId == TSDB_FUNC_MAVG ||
-        functionId == TSDB_FUNC_SAMPLE ) {
+        functionId == TSDB_FUNC_SAMPLE || functionId == TSDB_FUNC_UNIQUE) {
       if (i > 0) pBInfo->pCtx[i].ptsOutputBuf = pBInfo->pCtx[i-1].pOutput;
     } else if (functionId == TSDB_FUNC_INTERP) {
       assert(pBInfo->pCtx[0].functionId == TSDB_FUNC_TS_DUMMY || pBInfo->pCtx[0].functionId == TSDB_FUNC_TS);
@@ -3918,6 +3945,15 @@ void finalizeQueryResult(SOperatorInfo* pOperator, SQLFunctionCtx* pCtx, SResult
   }
 }
 
+bool isUniqueQuery(int32_t numOfOutput, SExprInfo* pExprs) {
+  for (int32_t i = 0; i < numOfOutput; ++i) {
+    if (pExprs[i].base.functionId == TSDB_FUNC_UNIQUE) {
+      return true;
+    }
+  }
+  return false;
+}
+
 static bool hasMainOutput(SQueryAttr *pQueryAttr) {
   for (int32_t i = 0; i < pQueryAttr->numOfOutput; ++i) {
     int32_t functionId = pQueryAttr->pExpr1[i].base.functionId;
@@ -3989,6 +4025,9 @@ void setResultRowOutputBufInitCtx(SQueryRuntimeEnv *pRuntimeEnv, SResultRow *pRe
   int32_t offset = 0;
   for (int32_t i = 0; i < numOfOutput; ++i) {
     pCtx[i].resultInfo = getResultCell(pResult, i, rowCellInfoOffset);
+    if (pCtx[i].functionId == TSDB_FUNC_UNIQUE){
+      pCtx[i].pUniqueSet = &pResult->uniqueHash;
+    }
 
     SResultRowCellInfo* pResInfo = pCtx[i].resultInfo;
     if (pResInfo->initialized && pResInfo->complete) {
@@ -4002,7 +4041,8 @@ void setResultRowOutputBufInitCtx(SQueryRuntimeEnv *pRuntimeEnv, SResultRow *pRe
     int32_t functionId = pCtx[i].functionId;
 
     if (functionId == TSDB_FUNC_TOP || functionId == TSDB_FUNC_BOTTOM || functionId == TSDB_FUNC_DIFF ||
-        functionId == TSDB_FUNC_CSUM || functionId == TSDB_FUNC_MAVG || functionId == TSDB_FUNC_SAMPLE) {
+        functionId == TSDB_FUNC_CSUM || functionId == TSDB_FUNC_MAVG || functionId == TSDB_FUNC_SAMPLE ||
+        functionId == TSDB_FUNC_UNIQUE) {
       if(i > 0) pCtx[i].ptsOutputBuf = pCtx[i-1].pOutput;
     }
 
@@ -4063,7 +4103,7 @@ void setResultOutputBuf(SQueryRuntimeEnv *pRuntimeEnv, SResultRow *pResult, SQLF
   // Note: pResult->pos[i]->num == 0, there is only fixed number of results for each group
   tFilePage *page = getResBufPage(pRuntimeEnv->pResultBuf, pResult->pageId);
 
-  int16_t offset = 0;
+  int32_t offset = 0;
   for (int32_t i = 0; i < numOfCols; ++i) {
     pCtx[i].pOutput = getPosInResultPage(pRuntimeEnv->pQueryAttr, page, pResult->offset, offset);
     offset += pCtx[i].outputBytes;
@@ -4071,7 +4111,8 @@ void setResultOutputBuf(SQueryRuntimeEnv *pRuntimeEnv, SResultRow *pResult, SQLF
     int32_t functionId = pCtx[i].functionId;
     if (functionId == TSDB_FUNC_TOP || functionId == TSDB_FUNC_BOTTOM ||
         functionId == TSDB_FUNC_DIFF || functionId == TSDB_FUNC_DERIVATIVE ||
-        functionId == TSDB_FUNC_SAMPLE || functionId == TSDB_FUNC_MAVG || functionId == TSDB_FUNC_CSUM) {
+        functionId == TSDB_FUNC_SAMPLE || functionId == TSDB_FUNC_MAVG ||
+        functionId == TSDB_FUNC_CSUM || functionId == TSDB_FUNC_UNIQUE) {
       if(i > 0) pCtx[i].ptsOutputBuf = pCtx[i-1].pOutput;
     }
 
@@ -4080,6 +4121,9 @@ void setResultOutputBuf(SQueryRuntimeEnv *pRuntimeEnv, SResultRow *pResult, SQLF
      * not all queries require the interResultBuf, such as COUNT
      */
     pCtx[i].resultInfo = getResultCell(pResult, i, rowCellInfoOffset);
+    if (pCtx[i].functionId == TSDB_FUNC_UNIQUE) {
+      pCtx[i].pUniqueSet = &pResult->uniqueHash;
+    }
   }
 }
 
@@ -5511,10 +5555,6 @@ SArray* getOrderCheckColumns(SQueryAttr* pQuery) {
   }
 
   if (pQuery->interval.interval > 0) {
-    if (pOrderColumns == NULL) {
-      pOrderColumns = taosArrayInit(1, sizeof(SColIndex));
-    }
-
     SColIndex colIndex = {.colIndex = 0, .colId = 0, .flag = TSDB_COL_NORMAL};
     taosArrayPush(pOrderColumns, &colIndex);
   }
@@ -8753,8 +8793,8 @@ int32_t convertQueryMsg(SQueryTableMsg *pQueryMsg, SQueryParam* param) {
       pMsg += tListLen(param->pGroupColIndex[i].name);
     }
 
-    pQueryMsg->orderByIdx = htons(pQueryMsg->orderByIdx);
-    pQueryMsg->orderType = htons(pQueryMsg->orderType);
+    //pQueryMsg->orderByIdx = htons(pQueryMsg->orderByIdx);
+    pQueryMsg->groupOrderType = htons(pQueryMsg->groupOrderType);
   }
 
   pQueryMsg->fillType = htons(pQueryMsg->fillType);
@@ -8962,7 +9002,7 @@ static int32_t updateOutputBufForTopBotQuery(SQueriedTableInfo* pTableInfo, SCol
   for (int32_t i = 0; i < numOfOutput; ++i) {
     int16_t functId = pExprs[i].base.functionId;
 
-    if (functId == TSDB_FUNC_TOP || functId == TSDB_FUNC_BOTTOM || functId == TSDB_FUNC_SAMPLE) {
+    if (functId == TSDB_FUNC_TOP || functId == TSDB_FUNC_BOTTOM || functId == TSDB_FUNC_SAMPLE || functId == TSDB_FUNC_UNIQUE) {
       int32_t j = getColumnIndexInSource(pTableInfo, &pExprs[i].base, pTagCols);
       if (j < 0 || j >= pTableInfo->numOfCols) {
         return TSDB_CODE_QRY_INVALID_MSG;
@@ -9331,8 +9371,8 @@ SGroupbyExpr *createGroupbyExprFromMsg(SQueryTableMsg *pQueryMsg, SColIndex *pCo
   }
 
   pGroupbyExpr->numOfGroupCols = pQueryMsg->numOfGroupCols;
-  pGroupbyExpr->orderType = pQueryMsg->orderType;
-  pGroupbyExpr->orderIndex = pQueryMsg->orderByIdx;
+  pGroupbyExpr->orderType = pQueryMsg->groupOrderType;
+  //pGroupbyExpr->orderIndex = pQueryMsg->orderByIdx;
 
   pGroupbyExpr->columnInfo = taosArrayInit(pQueryMsg->numOfGroupCols, sizeof(SColIndex));
   for(int32_t i = 0; i < pQueryMsg->numOfGroupCols; ++i) {
@@ -9547,7 +9587,8 @@ SQInfo* createQInfoImpl(SQueryTableMsg* pQueryMsg, SGroupbyExpr* pGroupbyExpr, S
   pQueryAttr->vgId            = vgId;
   pQueryAttr->pFilters        = pFilters;
   pQueryAttr->range           = pQueryMsg->range;
-  
+  pQueryAttr->uniqueQuery     = isUniqueQuery(numOfOutput, pExprs);
+
   pQueryAttr->tableCols = calloc(numOfCols, sizeof(SSingleColumnFilterInfo));
   if (pQueryAttr->tableCols == NULL) {
     goto _cleanup;

src/query/src/qExtbuffer.c

@@ -46,7 +46,7 @@ tExtMemBuffer* createExtMemBuffer(int32_t inMemSize, int32_t elemSize, int32_t p
   
   SExtFileInfo *pFMeta = &pMemBuffer->fileMeta;
 
-  pFMeta->pageSize = DEFAULT_PAGE_SIZE;
+  //pFMeta->pageSize = DEFAULT_PAGE_SIZE;
 
   pFMeta->flushoutData.nAllocSize = 4;
   pFMeta->flushoutData.nLength = 0;

src/query/src/qPlan.c

@@ -77,7 +77,7 @@ static SQueryNode* createQueryNode(int32_t type, const char* name, SQueryNode**
 
       pGroupbyExpr->tableIndex = p->tableIndex;
       pGroupbyExpr->orderType  = p->orderType;
-      pGroupbyExpr->orderIndex = p->orderIndex;
+      //pGroupbyExpr->orderIndex = p->orderIndex;
       pGroupbyExpr->numOfGroupCols = p->numOfGroupCols;
       pGroupbyExpr->columnInfo = taosArrayDup(p->columnInfo);
       pNode->pExtInfo = pGroupbyExpr;

src/query/src/qResultbuf.c

@@ -20,7 +20,7 @@ int32_t createDiskbasedResultBuffer(SDiskbasedResultBuf** pResultBuf, int32_t pa
   pResBuf->pageSize     = pagesize;
   pResBuf->numOfPages   = 0;                        // all pages are in buffer in the first place
   pResBuf->totalBufSize = 0;
-  pResBuf->inMemPages   = inMemBufSize/pagesize;    // maximum allowed pages, it is a soft limit.
+  pResBuf->inMemPages   = inMemBufSize/pagesize + 1;    // maximum allowed pages, it is a soft limit.
   pResBuf->allocateId   = -1;
   pResBuf->comp         = true;
   pResBuf->file         = NULL;
@@ -28,7 +28,7 @@ int32_t createDiskbasedResultBuffer(SDiskbasedResultBuf** pResultBuf, int32_t pa
   pResBuf->fileSize     = 0;
 
   // at least more than 2 pages must be in memory
-  assert(inMemBufSize >= pagesize * 2);
+  // assert(inMemBufSize >= pagesize * 2);
 
   pResBuf->lruList = tdListNew(POINTER_BYTES);
 
@@ -257,7 +257,7 @@ static char* evicOneDataPage(SDiskbasedResultBuf* pResultBuf) {
     int32_t prev = pResultBuf->inMemPages;
 
     // increase by 50% of previous mem pages
-    pResultBuf->inMemPages = (int32_t)(pResultBuf->inMemPages * 1.5f);
+    pResultBuf->inMemPages = (int32_t)(pResultBuf->inMemPages * 1.5f) + 1;  // if pResultBuf->inMemPages == 1, *1.5 always == 1
 
     qWarn("%p in memory buf page not sufficient, expand from %d to %d, page size:%d", pResultBuf, prev,
           pResultBuf->inMemPages, pResultBuf->pageSize);

src/query/src/qUtil.c

@@ -33,7 +33,7 @@ typedef struct SCompSupporter {
 } SCompSupporter;
 
 int32_t getRowNumForMultioutput(SQueryAttr* pQueryAttr, bool topBottomQuery, bool stable) {
-  if (pQueryAttr && (!stable)) {
+  if (pQueryAttr && (!stable)) {  // if table is stable, no need return more than 1 no in merge stage
     for (int16_t i = 0; i < pQueryAttr->numOfOutput; ++i) {
       if (pQueryAttr->pExpr1[i].base.functionId == TSDB_FUNC_TOP ||
           pQueryAttr->pExpr1[i].base.functionId == TSDB_FUNC_BOTTOM ||
@@ -42,6 +42,9 @@ int32_t getRowNumForMultioutput(SQueryAttr* pQueryAttr, bool topBottomQuery, boo
         return (int32_t)pQueryAttr->pExpr1[i].base.param[0].i64;
       }
     }
+    if (pQueryAttr->uniqueQuery){
+      return MAX_UNIQUE_RESULT_ROWS;
+    }
   }
 
   return 1;
@@ -85,6 +88,10 @@ void cleanupResultRowInfo(SResultRowInfo *pResultRowInfo) {
   for(int32_t i = 0; i < pResultRowInfo->size; ++i) {
     if (pResultRowInfo->pResult[i]) {
       tfree(pResultRowInfo->pResult[i]->key);
+      if (pResultRowInfo->pResult[i]->uniqueHash){
+        taosHashCleanup(pResultRowInfo->pResult[i]->uniqueHash);
+        pResultRowInfo->pResult[i]->uniqueHash = NULL;
+      }
     }
   }
   
@@ -150,11 +157,11 @@ void clearResultRow(SQueryRuntimeEnv *pRuntimeEnv, SResultRow *pResultRow, int16
   if (pResultRow->pageId >= 0) {
     tFilePage *page = getResBufPage(pRuntimeEnv->pResultBuf, pResultRow->pageId);
 
-    int16_t offset = 0;
+    int32_t offset = 0;
     for (int32_t i = 0; i < pRuntimeEnv->pQueryAttr->numOfOutput; ++i) {
       SResultRowCellInfo *pResultInfo = &pResultRow->pCellInfo[i];
 
-      int16_t size = pRuntimeEnv->pQueryAttr->pExpr1[i].base.resType;
+      int32_t size = pRuntimeEnv->pQueryAttr->pExpr1[i].base.resBytes;
       char * s = getPosInResultPage(pRuntimeEnv->pQueryAttr, page, pResultRow->offset, offset);
       memset(s, 0, size);
 
@@ -192,7 +199,13 @@ SResultRowPool* initResultRowPool(size_t size) {
   p->numOfElemPerBlock = 128;
 
   p->elemSize = (int32_t) size;
-  p->blockSize = p->numOfElemPerBlock * p->elemSize;
+  int64_t tmp = p->elemSize;
+  tmp *= p->numOfElemPerBlock;
+  if (tmp > 1024*1024*1024){
+    qError("ResultRow blockSize is too large:%" PRId64, tmp);
+    tmp = 128*1024*1024;
+  }
+  p->blockSize = tmp;
   p->position.pos = 0;
 
   p->pData = taosArrayInit(8, POINTER_BYTES);
@@ -217,7 +230,6 @@ SResultRow* getNewResultRow(SResultRowPool* p) {
   }
 
   p->position.pos = (p->position.pos + 1)%p->numOfElemPerBlock;
-  initResultRow(ptr);
 
   return ptr;
 }
@@ -451,9 +463,7 @@ int32_t tsDescOrder(const void* p1, const void* p2) {
   }
 }
 
-void
-
-orderTheResultRows(SQueryRuntimeEnv* pRuntimeEnv) {
+void orderTheResultRows(SQueryRuntimeEnv* pRuntimeEnv) {
   __compar_fn_t  fn = NULL;
   if (pRuntimeEnv->pQueryAttr->order.order == TSDB_ORDER_ASC) {
     fn = tsAscOrder;

src/tsdb/src/tsdbRead.c

@@ -3150,8 +3150,7 @@ static bool loadCachedLast(STsdbQueryHandle* pQueryHandle) {
       }
     
       pData = (char*)pColInfo->pData + numOfRows * pColInfo->info.bytes;
-    
-      if (pTable->lastCols[j].bytes > 0) {        
+      if (pTable->lastCols[j].bytes > 0) {
         void* value = pTable->lastCols[j].pData;
         switch (pColInfo->info.type) {
           case TSDB_DATA_TYPE_BINARY:
@@ -3205,7 +3204,6 @@ static bool loadCachedLast(STsdbQueryHandle* pQueryHandle) {
 
           pColInfo = taosArrayGet(pQueryHandle->pColumns, n);
           pData = (char*)pColInfo->pData + numOfRows * pColInfo->info.bytes;;
-
           if (pColInfo->info.colId == PRIMARYKEY_TIMESTAMP_COL_INDEX) {
             *(TSKEY *)pData = pTable->lastCols[j].ts;
             continue;
@@ -3231,7 +3229,7 @@ static bool loadCachedLast(STsdbQueryHandle* pQueryHandle) {
     if (priKey != TSKEY_INITIAL_VAL) {
       pColInfo = taosArrayGet(pQueryHandle->pColumns, priIdx);
       pData = (char*)pColInfo->pData + numOfRows * pColInfo->info.bytes;
-    
+
       *(TSKEY *)pData = priKey;
 
       for (int32_t n = 0; n < tgNumOfCols; ++n) {
@@ -3241,7 +3239,7 @@ static bool loadCachedLast(STsdbQueryHandle* pQueryHandle) {
       
         pColInfo = taosArrayGet(pQueryHandle->pColumns, n);
         pData = (char*)pColInfo->pData + numOfRows * pColInfo->info.bytes;;
-      
+
         assert (pColInfo->info.colId != PRIMARYKEY_TIMESTAMP_COL_INDEX);
         
         if (pColInfo->info.type == TSDB_DATA_TYPE_BINARY || pColInfo->info.type == TSDB_DATA_TYPE_NCHAR) {
@@ -4288,6 +4286,7 @@ void tsdbDestroyTableGroup(STableGroupInfo *pGroupList) {
   }
 
   taosHashCleanup(pGroupList->map);
+  pGroupList->map = NULL;
   taosArrayDestroy(&pGroupList->pGroupList);
   pGroupList->numOfTables = 0;
 }
@@ -4662,4 +4661,4 @@ void tsdbAddScanCallback(TsdbQueryHandleT* queryHandle, readover_callback callba
   pQueryHandle->readover_cb = callback;
   pQueryHandle->param       = param;
   return ;
-}
+}
\ No newline at end of file

src/util/src/terror.c

@@ -299,6 +299,7 @@ TAOS_DEFINE_ERROR(TSDB_CODE_QRY_NOT_ENOUGH_BUFFER,        "Query buffer limit ha
 TAOS_DEFINE_ERROR(TSDB_CODE_QRY_INCONSISTAN,              "File inconsistance in replica")
 TAOS_DEFINE_ERROR(TSDB_CODE_QRY_INVALID_TIME_CONDITION,   "One valid time range condition expected")
 TAOS_DEFINE_ERROR(TSDB_CODE_QRY_SYS_ERROR,                "System error")
+TAOS_DEFINE_ERROR(TSDB_CODE_QRY_UNIQUE_RESULT_TOO_LARGE,  "Unique result num is too large")
 
 // grant
 TAOS_DEFINE_ERROR(TSDB_CODE_GRANT_EXPIRED,                "License expired")