Merge pull request #1206 from taosdata/feature/liaohj

Feature/liaohj

src/client/inc/tscSecondaryMerge.h

@@ -68,7 +68,7 @@ typedef struct SLocalReducer {
   bool                   hasPrevRow;  // cannot be released
   bool                   hasUnprocessedRow;
   tOrderDescriptor *     pDesc;
-  tColModel *            resColModel;
+  SColumnModel *            resColModel;
   tExtMemBuffer **       pExtMemBuffer;      // disk-based buffer
   SInterpolationInfo     interpolationInfo;  // interpolation support structure
   char *                 pFinalRes;          // result data after interpo
@@ -92,7 +92,7 @@ typedef struct SSubqueryState {
 typedef struct SRetrieveSupport {
   tExtMemBuffer **  pExtMemBuffer;     // for build loser tree
   tOrderDescriptor *pOrderDescriptor;
-  tColModel *       pFinalColModel;   // colModel for final result
+  SColumnModel *    pFinalColModel;    // colModel for final result
   SSubqueryState *  pState;
   int32_t           subqueryIndex;     // index of current vnode in vnode list
   SSqlObj *         pParentSqlObj;
@@ -102,9 +102,9 @@ typedef struct SRetrieveSupport {
 } SRetrieveSupport;
 
 int32_t tscLocalReducerEnvCreate(SSqlObj *pSql, tExtMemBuffer ***pMemBuffer, tOrderDescriptor **pDesc,
-                                 tColModel **pFinalModel, uint32_t nBufferSize);
+                                 SColumnModel **pFinalModel, uint32_t nBufferSize);
 
-void tscLocalReducerEnvDestroy(tExtMemBuffer **pMemBuffer, tOrderDescriptor *pDesc, tColModel *pFinalModel,
+void tscLocalReducerEnvDestroy(tExtMemBuffer **pMemBuffer, tOrderDescriptor *pDesc, SColumnModel *pFinalModel,
                                int32_t numOfVnodes);
 
 int32_t saveToBuffer(tExtMemBuffer *pMemoryBuf, tOrderDescriptor *pDesc, tFilePage *pPage, void *data,
@@ -116,7 +116,7 @@ int32_t tscFlushTmpBuffer(tExtMemBuffer *pMemoryBuf, tOrderDescriptor *pDesc, tF
  * create local reducer to launch the second-stage reduce process at client site
  */
 void tscCreateLocalReducer(tExtMemBuffer **pMemBuffer, int32_t numOfBuffer, tOrderDescriptor *pDesc,
-                           tColModel *finalModel, SSqlCmd *pSqlCmd, SSqlRes *pRes);
+                           SColumnModel *finalModel, SSqlCmd *pSqlCmd, SSqlRes *pRes);
 
 void tscDestroyLocalReducer(SSqlObj *pSql);
 

src/client/src/tscFunctionImpl.c

@@ -27,6 +27,7 @@
 #include "ttime.h"
 #include "ttypes.h"
 #include "tutil.h"
+#include "tpercentile.h"
 
 #define GET_INPUT_CHAR(x) (((char *)((x)->aInputElemBuf)) + ((x)->startOffset) * ((x)->inputBytes))
 #define GET_INPUT_CHAR_INDEX(x, y) (GET_INPUT_CHAR(x) + (y) * (x)->inputBytes)
@@ -220,7 +221,7 @@ int32_t getResultDataInfo(int32_t dataType, int32_t dataBytes, int32_t functionI
     } else if (functionId == TSDB_FUNC_TWA) {
       *type = TSDB_DATA_TYPE_DOUBLE;
       *bytes = sizeof(STwaInfo);
-      *intermediateResBytes = sizeof(STwaInfo);
+      *intermediateResBytes = *bytes;
       return TSDB_CODE_SUCCESS;
     }
   }
@@ -271,7 +272,6 @@ int32_t getResultDataInfo(int32_t dataType, int32_t dataBytes, int32_t functionI
   } else if (functionId == TSDB_FUNC_PERCT) {
     *type = (int16_t)TSDB_DATA_TYPE_DOUBLE;
     *bytes = (int16_t)sizeof(double);
-    //*intermediateResBytes = POINTER_BYTES;
     *intermediateResBytes = (int16_t)sizeof(double);
   } else if (functionId == TSDB_FUNC_LEASTSQR) {
     *type = TSDB_DATA_TYPE_BINARY;
@@ -1878,6 +1878,7 @@ static void do_top_function_add(STopBotInfo *pInfo, int32_t maxLen, void *pData,
   tVariantCreateFromBinary(&val, pData, tDataTypeDesc[type].nSize, type);
 
   tValuePair **pList = pInfo->res;
+  assert(pList != NULL);
   
   if (pInfo->num < maxLen) {
     if (pInfo->num == 0 ||
@@ -2416,7 +2417,7 @@ static bool percentile_function_setup(SQLFunctionCtx *pCtx) {
   SResultInfo *pResInfo = GET_RES_INFO(pCtx);
   SSchema      field[1] = {{pCtx->inputType, "dummyCol", 0, pCtx->inputBytes}};
 
-  tColModel *pModel = tColModelCreate(field, 1, 1000);
+  SColumnModel *pModel = createColumnModel(field, 1, 1000);
   int32_t    orderIdx = 0;
 
   // tOrderDesc object

src/client/src/tscParseInsert.c

@@ -321,7 +321,7 @@ int32_t tsParseOneColumnData(SSchema *pSchema, SSQLToken *pToken, char *payload,
       if (pToken->type == TK_NULL) {
         *(uint32_t *)payload = TSDB_DATA_NCHAR_NULL;
       } else {
-        // if the converted output len is over than pSchema->bytes, return error: 'Argument list too long'
+        // if the converted output len is over than pColumnModel->bytes, return error: 'Argument list too long'
         if (!taosMbsToUcs4(pToken->z, pToken->n, payload, pSchema->bytes)) {
           char buf[512] = {0};
           snprintf(buf, 512, "%s", strerror(errno));

src/client/src/tscSQLParser.c

@@ -72,7 +72,7 @@ static void setColumnOffsetValueInResultset(SQueryInfo* pQueryInfo);
 static int32_t parseGroupbyClause(SQueryInfo* pQueryInfo, tVariantList* pList, SSqlCmd* pCmd);
 
 static int32_t parseIntervalClause(SQueryInfo* pQueryInfo, SQuerySQL* pQuerySql);
-static int32_t setSlidingClause(SQueryInfo* pQueryInfo, SQuerySQL* pQuerySql);
+static int32_t parseSlidingClause(SQueryInfo* pQueryInfo, SQuerySQL* pQuerySql);
 
 static int32_t addProjectionExprAndResultField(SQueryInfo* pQueryInfo, tSQLExprItem* pItem);
 
@@ -657,14 +657,14 @@ int32_t parseIntervalClause(SQueryInfo* pQueryInfo, SQuerySQL* pQuerySql) {
     return ret;
   }
   
-  if (setSlidingClause(pQueryInfo, pQuerySql) != TSDB_CODE_SUCCESS) {
+  if (parseSlidingClause(pQueryInfo, pQuerySql) != TSDB_CODE_SUCCESS) {
     return TSDB_CODE_INVALID_SQL;
   }
   
   return TSDB_CODE_SUCCESS;
 }
 
-int32_t setSlidingClause(SQueryInfo* pQueryInfo, SQuerySQL* pQuerySql) {
+int32_t parseSlidingClause(SQueryInfo* pQueryInfo, SQuerySQL* pQuerySql) {
   const char* msg0 = "sliding value too small";
   const char* msg1 = "sliding value no larger than the interval value";
 
@@ -685,7 +685,7 @@ int32_t setSlidingClause(SQueryInfo* pQueryInfo, SQuerySQL* pQuerySql) {
       return invalidSqlErrMsg(pQueryInfo->msg, msg1);
     }
   } else {
-    pSliding->n = pQueryInfo->nAggTimeInterval;
+    pQueryInfo->nSlidingTime = -1;
   }
 
   return TSDB_CODE_SUCCESS;
@@ -5453,8 +5453,6 @@ int32_t doCheckForQuery(SSqlObj* pSql, SQuerySQL* pQuerySql, int32_t index) {
   const char* msg0 = "invalid table name";
   const char* msg1 = "table name too long";
   const char* msg2 = "point interpolation query needs timestamp";
-  const char* msg3 = "sliding value too small";
-  const char* msg4 = "sliding value no larger than the interval value";
   const char* msg5 = "fill only available for interval query";
   const char* msg6 = "start(end) time of query range required or time range too large";
   const char* msg7 = "illegal number of tables in from clause";
@@ -5588,30 +5586,6 @@ int32_t doCheckForQuery(SSqlObj* pSql, SQuerySQL* pQuerySql, int32_t index) {
     return invalidSqlErrMsg(pQueryInfo->msg, msg2);
   }
 
-//  // set sliding value, the query time range needs to be decide in the first place
-//  SSQLToken* pSliding = &pQuerySql->sliding;
-//  if (pSliding->n != 0) {
-//    if (!tscEmbedded && pCmd->inStream == 0 && hasDefaultQueryTimeRange(pQueryInfo)) {  // sliding only allowed in stream
-//      const char* msg = "time range expected for sliding window query";
-//      return invalidSqlErrMsg(tscGetErrorMsgPayload(pCmd), msg);
-//    }
-//
-//    getTimestampInUsFromStr(pSliding->z, pSliding->n, &pQueryInfo->nSlidingTime);
-//    if (pMeterMetaInfo->pMeterMeta->precision == TSDB_TIME_PRECISION_MILLI) {
-//      pQueryInfo->nSlidingTime /= 1000;
-//    }
-//
-//    if (pQueryInfo->nSlidingTime < tsMinSlidingTime) {
-//      return invalidSqlErrMsg(pQueryInfo->msg, msg3);
-//    }
-//
-//    if (pQueryInfo->nSlidingTime > pQueryInfo->nAggTimeInterval) {
-//      return invalidSqlErrMsg(pQueryInfo->msg, msg4);
-//    }
-//  } else {
-//    pQueryInfo->nSlidingTime = -1;
-//  }
-
   // in case of join query, time range is required.
   if (QUERY_IS_JOIN_QUERY(pQueryInfo->type)) {
     int64_t timeRange = labs(pQueryInfo->stime - pQueryInfo->etime);

src/client/src/tscSecondaryMerge.c

@@ -62,16 +62,19 @@ static void tscInitSqlContext(SSqlCmd *pCmd, SSqlRes *pRes, SLocalReducer *pRedu
   for (int32_t i = 0; i < pQueryInfo->fieldsInfo.numOfOutputCols; ++i) {
     SQLFunctionCtx *pCtx = &pReducer->pCtx[i];
 
-    pCtx->aOutputBuf = pReducer->pResultBuf->data + tscFieldInfoGetOffset(pQueryInfo, i) * pReducer->resColModel->maxCapacity;
+    pCtx->aOutputBuf = pReducer->pResultBuf->data + tscFieldInfoGetOffset(pQueryInfo, i) * pReducer->resColModel->capacity;
     pCtx->order = pQueryInfo->order.order;
     pCtx->functionId = pQueryInfo->exprsInfo.pExprs[i].functionId;
 
     // input buffer hold only one point data
-    pCtx->aInputElemBuf = pReducer->pTempBuffer->data + pDesc->pSchema->colOffset[i];
+    int16_t offset = getColumnModelOffset(pDesc->pColumnModel, i);
+    SSchema* pSchema = getColumnModelSchema(pDesc->pColumnModel, i);
+    
+    pCtx->aInputElemBuf = pReducer->pTempBuffer->data + offset;
 
     // input data format comes from pModel
-    pCtx->inputType = pDesc->pSchema->pFields[i].type;
-    pCtx->inputBytes = pDesc->pSchema->pFields[i].bytes;
+    pCtx->inputType = pSchema->type;
+    pCtx->inputBytes = pSchema->bytes;
 
     TAOS_FIELD *pField = tscFieldInfoGetField(pQueryInfo, i);
     // output data format yet comes from pCmd.
@@ -132,11 +135,11 @@ static void tscInitSqlContext(SSqlCmd *pCmd, SSqlRes *pRes, SLocalReducer *pRedu
  * todo release allocated memory process with async process
  */
 void tscCreateLocalReducer(tExtMemBuffer **pMemBuffer, int32_t numOfBuffer, tOrderDescriptor *pDesc,
-                           tColModel *finalmodel, SSqlCmd *pCmd, SSqlRes *pRes) {
+                           SColumnModel *finalmodel, SSqlCmd *pCmd, SSqlRes *pRes) {
   // offset of cmd in SSqlObj structure
   char *pSqlObjAddr = (char *)pCmd - offsetof(SSqlObj, cmd);
 
-  if (pMemBuffer == NULL || pDesc->pSchema == NULL) {
+  if (pMemBuffer == NULL || pDesc->pColumnModel == NULL) {
     tscLocalReducerEnvDestroy(pMemBuffer, pDesc, finalmodel, numOfBuffer);
 
     tscError("%p no local buffer or intermediate result format model", pSqlObjAddr);
@@ -162,9 +165,9 @@ void tscCreateLocalReducer(tExtMemBuffer **pMemBuffer, int32_t numOfBuffer, tOrd
     return;
   }
 
-  if (pDesc->pSchema->maxCapacity >= pMemBuffer[0]->nPageSize) {
-    tscError("%p Invalid value of buffer capacity %d and page size %d ", pSqlObjAddr, pDesc->pSchema->maxCapacity,
-             pMemBuffer[0]->nPageSize);
+  if (pDesc->pColumnModel->capacity >= pMemBuffer[0]->pageSize) {
+    tscError("%p Invalid value of buffer capacity %d and page size %d ", pSqlObjAddr, pDesc->pColumnModel->capacity,
+             pMemBuffer[0]->pageSize);
 
     tscLocalReducerEnvDestroy(pMemBuffer, pDesc, finalmodel, numOfBuffer);
     pRes->code = TSDB_CODE_APP_ERROR;
@@ -196,7 +199,7 @@ void tscCreateLocalReducer(tExtMemBuffer **pMemBuffer, int32_t numOfBuffer, tOrd
     int32_t numOfFlushoutInFile = pMemBuffer[i]->fileMeta.flushoutData.nLength;
 
     for (int32_t j = 0; j < numOfFlushoutInFile; ++j) {
-      SLocalDataSource *pDS = (SLocalDataSource *)malloc(sizeof(SLocalDataSource) + pMemBuffer[0]->nPageSize);
+      SLocalDataSource *pDS = (SLocalDataSource *)malloc(sizeof(SLocalDataSource) + pMemBuffer[0]->pageSize);
       if (pDS == NULL) {
         tscError("%p failed to create merge structure", pSqlObjAddr);
         pRes->code = TSDB_CODE_CLI_OUT_OF_MEMORY;
@@ -219,7 +222,7 @@ void tscCreateLocalReducer(tExtMemBuffer **pMemBuffer, int32_t numOfBuffer, tOrd
 
       tscGetSrcColumnInfo(colInfo, pQueryInfo);
 
-      tColModelDisplayEx(pDesc->pSchema, pDS->filePage.data, pDS->filePage.numOfElems, pMemBuffer[0]->numOfElemsPerPage,
+      tColModelDisplayEx(pDesc->pColumnModel, pDS->filePage.data, pDS->filePage.numOfElems, pMemBuffer[0]->numOfElemsPerPage,
                          colInfo);
 #endif
       if (pDS->filePage.numOfElems == 0) {  // no data in this flush
@@ -259,7 +262,7 @@ void tscCreateLocalReducer(tExtMemBuffer **pMemBuffer, int32_t numOfBuffer, tOrd
   tscRestoreSQLFunctionForMetricQuery(pQueryInfo);
   tscFieldInfoCalOffset(pQueryInfo);
 
-  if (pReducer->rowSize > pMemBuffer[0]->nPageSize) {
+  if (pReducer->rowSize > pMemBuffer[0]->pageSize) {
     assert(false);  // todo fixed row size is larger than the minimum page size;
   }
 
@@ -274,15 +277,15 @@ void tscCreateLocalReducer(tExtMemBuffer **pMemBuffer, int32_t numOfBuffer, tOrd
   pReducer->discardData = (tFilePage *)calloc(1, pReducer->rowSize + sizeof(tFilePage));
   pReducer->discard = false;
 
-  pReducer->nResultBufSize = pMemBuffer[0]->nPageSize * 16;
+  pReducer->nResultBufSize = pMemBuffer[0]->pageSize * 16;
   pReducer->pResultBuf = (tFilePage *)calloc(1, pReducer->nResultBufSize + sizeof(tFilePage));
   
   int32_t finalRowLength = tscGetResRowLength(pQueryInfo);
   pReducer->resColModel = finalmodel;
-  pReducer->resColModel->maxCapacity = pReducer->nResultBufSize / finalRowLength;
+  pReducer->resColModel->capacity = pReducer->nResultBufSize / finalRowLength;
   assert(finalRowLength <= pReducer->rowSize);
 
-  pReducer->pFinalRes = calloc(1, pReducer->rowSize * pReducer->resColModel->maxCapacity);
+  pReducer->pFinalRes = calloc(1, pReducer->rowSize * pReducer->resColModel->capacity);
   pReducer->pBufForInterpo = calloc(1, pReducer->nResultBufSize);
 
   if (pReducer->pTempBuffer == NULL|| pReducer->discardData == NULL  || pReducer->pResultBuf == NULL ||
@@ -304,8 +307,8 @@ void tscCreateLocalReducer(tExtMemBuffer **pMemBuffer, int32_t numOfBuffer, tOrd
   tscCreateResPointerInfo(pRes, pQueryInfo);
   tscInitSqlContext(pCmd, pRes, pReducer, pDesc);
 
-  // we change the maxCapacity of schema to denote that there is only one row in temp buffer
-  pReducer->pDesc->pSchema->maxCapacity = 1;
+  // we change the capacity of schema to denote that there is only one row in temp buffer
+  pReducer->pDesc->pColumnModel->capacity = 1;
   
   //restore the limitation value at the last stage
   if (tscOrderedProjectionQueryOnSTable(pQueryInfo, 0)) {
@@ -333,7 +336,8 @@ void tscCreateLocalReducer(tExtMemBuffer **pMemBuffer, int32_t numOfBuffer, tOrd
   if (pQueryInfo->groupbyExpr.numOfGroupCols > 0) {
     pInterpoInfo->pTags[0] = (char *)pInterpoInfo->pTags + POINTER_BYTES * pQueryInfo->groupbyExpr.numOfGroupCols;
     for (int32_t i = 1; i < pQueryInfo->groupbyExpr.numOfGroupCols; ++i) {
-      pInterpoInfo->pTags[i] = pReducer->resColModel->pFields[startIndex + i - 1].bytes + pInterpoInfo->pTags[i - 1];
+      SSchema* pSchema = getColumnModelSchema(pReducer->resColModel, startIndex + i - 1);
+      pInterpoInfo->pTags[i] = pSchema->bytes + pInterpoInfo->pTags[i - 1];
     }
   } else {
     assert(pInterpoInfo->pTags == NULL);
@@ -346,16 +350,16 @@ static int32_t tscFlushTmpBufferImpl(tExtMemBuffer *pMemoryBuf, tOrderDescriptor
     return 0;
   }
 
-  assert(pPage->numOfElems <= pDesc->pSchema->maxCapacity);
+  assert(pPage->numOfElems <= pDesc->pColumnModel->capacity);
 
   // sort before flush to disk, the data must be consecutively put on tFilePage.
-  if (pDesc->orderIdx.numOfOrderedCols > 0) {
+  if (pDesc->orderIdx.numOfCols > 0) {
     tColDataQSort(pDesc, pPage->numOfElems, 0, pPage->numOfElems - 1, pPage->data, orderType);
   }
 
 #ifdef _DEBUG_VIEW
   printf("%" PRIu64 " rows data flushed to disk after been sorted:\n", pPage->numOfElems);
-  tColModelDisplay(pDesc->pSchema, pPage->data, pPage->numOfElems, pPage->numOfElems);
+  tColModelDisplay(pDesc->pColumnModel, pPage->data, pPage->numOfElems, pPage->numOfElems);
 #endif
 
   // write to cache after being sorted
@@ -383,18 +387,19 @@ int32_t tscFlushTmpBuffer(tExtMemBuffer *pMemoryBuf, tOrderDescriptor *pDesc, tF
 
 int32_t saveToBuffer(tExtMemBuffer *pMemoryBuf, tOrderDescriptor *pDesc, tFilePage *pPage, void *data,
                      int32_t numOfRows, int32_t orderType) {
-  if (pPage->numOfElems + numOfRows <= pDesc->pSchema->maxCapacity) {
-    tColModelAppend(pDesc->pSchema, pPage, data, 0, numOfRows, numOfRows);
+  SColumnModel *pModel = pDesc->pColumnModel;
+  
+  if (pPage->numOfElems + numOfRows <= pModel->capacity) {
+    tColModelAppend(pModel, pPage, data, 0, numOfRows, numOfRows);
     return 0;
   }
 
-  tColModel *pModel = pDesc->pSchema;
-
-  int32_t numOfRemainEntries = pDesc->pSchema->maxCapacity - pPage->numOfElems;
+  // current buffer is overflow, flush data to extensive buffer
+  int32_t numOfRemainEntries = pModel->capacity - pPage->numOfElems;
   tColModelAppend(pModel, pPage, data, 0, numOfRemainEntries, numOfRows);
 
-  /* current buffer is full, need to flushed to disk */
-  assert(pPage->numOfElems == pDesc->pSchema->maxCapacity);
+  // current buffer is full, need to flushed to disk
+  assert(pPage->numOfElems == pModel->capacity);
   int32_t ret = tscFlushTmpBuffer(pMemoryBuf, pDesc, pPage, orderType);
   if (ret != 0) {
     return -1;
@@ -404,15 +409,15 @@ int32_t saveToBuffer(tExtMemBuffer *pMemoryBuf, tOrderDescriptor *pDesc, tFilePa
 
   while (remain > 0) {
     int32_t numOfWriteElems = 0;
-    if (remain > pModel->maxCapacity) {
-      numOfWriteElems = pModel->maxCapacity;
+    if (remain > pModel->capacity) {
+      numOfWriteElems = pModel->capacity;
     } else {
       numOfWriteElems = remain;
     }
 
     tColModelAppend(pModel, pPage, data, numOfRows - remain, numOfWriteElems, numOfRows);
 
-    if (pPage->numOfElems == pModel->maxCapacity) {
+    if (pPage->numOfElems == pModel->capacity) {
       int32_t ret = tscFlushTmpBuffer(pMemoryBuf, pDesc, pPage, orderType);
       if (ret != 0) {
         return -1;
@@ -508,7 +513,7 @@ void tscDestroyLocalReducer(SSqlObj *pSql) {
   tscTrace("%p free local reducer finished", pSql);
 }
 
-static int32_t createOrderDescriptor(tOrderDescriptor **pOrderDesc, SSqlCmd *pCmd, tColModel *pModel) {
+static int32_t createOrderDescriptor(tOrderDescriptor **pOrderDesc, SSqlCmd *pCmd, SColumnModel *pModel) {
   int32_t numOfGroupByCols = 0;
   SQueryInfo* pQueryInfo = tscGetQueryInfoDetail(pCmd, pCmd->clauseIndex);
   
@@ -567,7 +572,7 @@ bool isSameGroup(SSqlCmd *pCmd, SLocalReducer *pReducer, char *pPrev, tFilePage
   }
 
   tOrderDescriptor *pOrderDesc = pReducer->pDesc;
-  int32_t           numOfCols = pOrderDesc->orderIdx.numOfOrderedCols;
+  int32_t           numOfCols = pOrderDesc->orderIdx.numOfCols;
 
   // no group by columns, all data belongs to one group
   if (numOfCols <= 0) {
@@ -577,25 +582,25 @@ bool isSameGroup(SSqlCmd *pCmd, SLocalReducer *pReducer, char *pPrev, tFilePage
   if (pOrderDesc->orderIdx.pData[numOfCols - 1] == PRIMARYKEY_TIMESTAMP_COL_INDEX) {  //<= 0
     // super table interval query
     assert(pQueryInfo->nAggTimeInterval > 0);
-    pOrderDesc->orderIdx.numOfOrderedCols -= 1;
+    pOrderDesc->orderIdx.numOfCols -= 1;
   } else {  // simple group by query
     assert(pQueryInfo->nAggTimeInterval == 0);
   }
 
   // only one row exists
   int32_t ret = compare_a(pOrderDesc, 1, 0, pPrev, 1, 0, tmpBuffer->data);
-  pOrderDesc->orderIdx.numOfOrderedCols = numOfCols;
+  pOrderDesc->orderIdx.numOfCols = numOfCols;
 
   return (ret == 0);
 }
 
 int32_t tscLocalReducerEnvCreate(SSqlObj *pSql, tExtMemBuffer ***pMemBuffer, tOrderDescriptor **pOrderDesc,
-                                 tColModel **pFinalModel, uint32_t nBufferSizes) {
+                                 SColumnModel **pFinalModel, uint32_t nBufferSizes) {
   SSqlCmd *pCmd = &pSql->cmd;
   SSqlRes *pRes = &pSql->res;
 
   SSchema *  pSchema = NULL;
-  tColModel *pModel = NULL;
+  SColumnModel *pModel = NULL;
   *pFinalModel = NULL;
 
   SQueryInfo* pQueryInfo = tscGetQueryInfoDetail(pCmd, pCmd->clauseIndex);
@@ -630,14 +635,10 @@ int32_t tscLocalReducerEnvCreate(SSqlObj *pSql, tExtMemBuffer ***pMemBuffer, tOr
     capacity = nBufferSizes / rlen;
   }
   
-  pModel = tColModelCreate(pSchema, pQueryInfo->fieldsInfo.numOfOutputCols, capacity);
+  pModel = createColumnModel(pSchema, pQueryInfo->fieldsInfo.numOfOutputCols, capacity);
 
   for (int32_t i = 0; i < pMeterMetaInfo->pMetricMeta->numOfVnodes; ++i) {
-    char tmpPath[512] = {0};
-    getTmpfilePath("tv_bf_db", tmpPath);
-    tscTrace("%p create [%d](%d) tmp file for subquery:%s", pSql, pMeterMetaInfo->pMetricMeta->numOfVnodes, i, tmpPath);
-
-    tExtMemBufferCreate(&(*pMemBuffer)[i], nBufferSizes, rlen, tmpPath, pModel);
+    (*pMemBuffer)[i] = createExtMemBuffer(nBufferSizes, rlen, pModel);
     (*pMemBuffer)[i]->flushModel = MULTIPLE_APPEND_MODEL;
   }
 
@@ -655,7 +656,7 @@ int32_t tscLocalReducerEnvCreate(SSqlObj *pSql, tExtMemBuffer ***pMemBuffer, tOr
     strcpy(pSchema[i].name, pField->name);
   }
 
-  *pFinalModel = tColModelCreate(pSchema, pQueryInfo->fieldsInfo.numOfOutputCols, capacity);
+  *pFinalModel = createColumnModel(pSchema, pQueryInfo->fieldsInfo.numOfOutputCols, capacity);
   tfree(pSchema);
 
   return TSDB_CODE_SUCCESS;
@@ -667,12 +668,12 @@ int32_t tscLocalReducerEnvCreate(SSqlObj *pSql, tExtMemBuffer ***pMemBuffer, tOr
  * @param pFinalModel
  * @param numOfVnodes
  */
-void tscLocalReducerEnvDestroy(tExtMemBuffer **pMemBuffer, tOrderDescriptor *pDesc, tColModel *pFinalModel,
+void tscLocalReducerEnvDestroy(tExtMemBuffer **pMemBuffer, tOrderDescriptor *pDesc, SColumnModel *pFinalModel,
                                int32_t numOfVnodes) {
-  tColModelDestroy(pFinalModel);
+  destroyColumnModel(pFinalModel);
   tOrderDescDestroy(pDesc);
   for (int32_t i = 0; i < numOfVnodes; ++i) {
-    tExtMemBufferDestroy(&pMemBuffer[i]);
+    pMemBuffer[i] = destoryExtMemBuffer(pMemBuffer[i]);
   }
 
   tfree(pMemBuffer);
@@ -697,8 +698,8 @@ int32_t loadNewDataFromDiskFor(SLocalReducer *pLocalReducer, SLocalDataSource *p
 
 #if defined(_DEBUG_VIEW)
     printf("new page load to buffer\n");
-    tColModelDisplay(pOneInterDataSrc->pMemBuffer->pColModel, pOneInterDataSrc->filePage.data,
-                     pOneInterDataSrc->filePage.numOfElems, pOneInterDataSrc->pMemBuffer->pColModel->maxCapacity);
+    tColModelDisplay(pOneInterDataSrc->pMemBuffer->pColumnModel, pOneInterDataSrc->filePage.data,
+                     pOneInterDataSrc->filePage.numOfElems, pOneInterDataSrc->pMemBuffer->pColumnModel->capacity);
 #endif
     *needAdjustLoserTree = true;
   } else {
@@ -759,7 +760,7 @@ void savePrevRecordAndSetupInterpoInfo(SLocalReducer *pLocalReducer, SQueryInfo*
   pLocalReducer->discard = true;
   pLocalReducer->discardData->numOfElems = 0;
 
-  tColModel *pModel = pLocalReducer->pDesc->pSchema;
+  SColumnModel *pModel = pLocalReducer->pDesc->pColumnModel;
   tColModelAppend(pModel, pLocalReducer->discardData, pLocalReducer->prevRowOfInput, 0, 1, 1);
 }
 
@@ -782,11 +783,12 @@ static void reversedCopyResultToDstBuf(SQueryInfo* pQueryInfo, SSqlRes *pRes, tF
 }
 
 static void reversedCopyFromInterpolationToDstBuf(SQueryInfo* pQueryInfo, SSqlRes *pRes, tFilePage **pResPages, SLocalReducer *pLocalReducer) {
+  assert(0);
   for (int32_t i = 0; i < pQueryInfo->exprsInfo.numOfExprs; ++i) {
     TAOS_FIELD *pField = tscFieldInfoGetField(pQueryInfo, i);
 
     int32_t offset = tscFieldInfoGetOffset(pQueryInfo, i);
-    assert(offset == pLocalReducer->resColModel->colOffset[i]);
+    assert(offset == getColumnModelOffset(pLocalReducer->resColModel, i));
 
     char *src = pResPages[i]->data + (pRes->numOfRows - 1) * pField->bytes;
     char *dst = pRes->data + pRes->numOfRows * offset;
@@ -880,7 +882,7 @@ static void doInterpolateResult(SSqlObj *pSql, SLocalReducer *pLocalReducer, boo
   tFilePage **pResPages = malloc(POINTER_BYTES * pQueryInfo->fieldsInfo.numOfOutputCols);
   for (int32_t i = 0; i < pQueryInfo->fieldsInfo.numOfOutputCols; ++i) {
     TAOS_FIELD *pField = tscFieldInfoGetField(pQueryInfo, i);
-    pResPages[i] = calloc(1, sizeof(tFilePage) + pField->bytes * pLocalReducer->resColModel->maxCapacity);
+    pResPages[i] = calloc(1, sizeof(tFilePage) + pField->bytes * pLocalReducer->resColModel->capacity);
   }
 
   char **  srcData = (char **)malloc((POINTER_BYTES + sizeof(int32_t)) * pQueryInfo->fieldsInfo.numOfOutputCols);
@@ -899,11 +901,11 @@ static void doInterpolateResult(SSqlObj *pSql, SLocalReducer *pLocalReducer, boo
     TSKEY etime = taosGetRevisedEndKey(actualETime, pQueryInfo->order.order, pQueryInfo->nAggTimeInterval, pQueryInfo->intervalTimeUnit,
                                        precision);
     int32_t nrows = taosGetNumOfResultWithInterpo(pInterpoInfo, pPrimaryKeys, remains, pQueryInfo->nAggTimeInterval, etime,
-                                                  pLocalReducer->resColModel->maxCapacity);
+                                                  pLocalReducer->resColModel->capacity);
 
     int32_t newRows = taosDoInterpoResult(pInterpoInfo, pQueryInfo->interpoType, pResPages, remains, nrows,
                                           pQueryInfo->nAggTimeInterval, pPrimaryKeys, pLocalReducer->resColModel, srcData,
-                                          pQueryInfo->defaultVal, functions, pLocalReducer->resColModel->maxCapacity);
+                                          pQueryInfo->defaultVal, functions, pLocalReducer->resColModel->capacity);
     assert(newRows <= nrows);
 
     if (pQueryInfo->limit.offset < newRows) {
@@ -960,11 +962,10 @@ static void doInterpolateResult(SSqlObj *pSql, SLocalReducer *pLocalReducer, boo
     if (pQueryInfo->order.order == TSQL_SO_ASC) {
       for (int32_t i = 0; i < pQueryInfo->fieldsInfo.numOfOutputCols; ++i) {
         TAOS_FIELD *pField = tscFieldInfoGetField(pQueryInfo, i);
-
-        memcpy(pRes->data + pLocalReducer->resColModel->colOffset[i] * pRes->numOfRows, pResPages[i]->data,
-               pField->bytes * pRes->numOfRows);
+        int16_t offset = getColumnModelOffset(pLocalReducer->resColModel, i);
+        memcpy(pRes->data + offset * pRes->numOfRows, pResPages[i]->data, pField->bytes * pRes->numOfRows);
       }
-    } else {
+    } else {//todo bug??
       reversedCopyFromInterpolationToDstBuf(pQueryInfo, pRes, pResPages, pLocalReducer);
     }
   }
@@ -979,13 +980,15 @@ static void doInterpolateResult(SSqlObj *pSql, SLocalReducer *pLocalReducer, boo
 }
 
 static void savePreviousRow(SLocalReducer *pLocalReducer, tFilePage *tmpBuffer) {
-  tColModel *pColModel = pLocalReducer->pDesc->pSchema;
-  assert(pColModel->maxCapacity == 1 && tmpBuffer->numOfElems == 1);
+  SColumnModel *pColumnModel = pLocalReducer->pDesc->pColumnModel;
+  assert(pColumnModel->capacity == 1 && tmpBuffer->numOfElems == 1);
 
   // copy to previous temp buffer
-  for (int32_t i = 0; i < pLocalReducer->pDesc->pSchema->numOfCols; ++i) {
-    memcpy(pLocalReducer->prevRowOfInput + pColModel->colOffset[i], tmpBuffer->data + pColModel->colOffset[i],
-           pColModel->pFields[i].bytes);
+  for (int32_t i = 0; i < pColumnModel->numOfCols; ++i) {
+    SSchema* pSchema = getColumnModelSchema(pColumnModel, i);
+    int16_t offset = getColumnModelOffset(pColumnModel, i);
+    
+    memcpy(pLocalReducer->prevRowOfInput + offset, tmpBuffer->data + offset, pSchema->bytes);
   }
 
   tmpBuffer->numOfElems = 0;
@@ -1127,7 +1130,7 @@ bool needToMerge(SQueryInfo* pQueryInfo, SLocalReducer *pLocalReducer, tFilePage
     ret = 1;                                                            // disable merge procedure
   } else {
     tOrderDescriptor *pDesc = pLocalReducer->pDesc;
-    if (pDesc->orderIdx.numOfOrderedCols > 0) {
+    if (pDesc->orderIdx.numOfCols > 0) {
       if (pDesc->tsOrder == TSQL_SO_ASC) {  // asc
         // todo refactor comparator
         ret = compare_a(pLocalReducer->pDesc, 1, 0, pLocalReducer->prevRowOfInput, 1, 0, tmpBuffer->data);
@@ -1177,7 +1180,7 @@ bool doGenerateFinalResults(SSqlObj *pSql, SLocalReducer *pLocalReducer, bool no
   
   SQueryInfo* pQueryInfo = tscGetQueryInfoDetail(pCmd, pCmd->clauseIndex);
   tFilePage *pResBuf = pLocalReducer->pResultBuf;
-  tColModel *pModel = pLocalReducer->resColModel;
+  SColumnModel *pModel = pLocalReducer->resColModel;
 
   pRes->code = TSDB_CODE_SUCCESS;
 
@@ -1192,7 +1195,7 @@ bool doGenerateFinalResults(SSqlObj *pSql, SLocalReducer *pLocalReducer, bool no
     return false;
   }
 
-  tColModelCompact(pModel, pResBuf, pModel->maxCapacity);
+  tColModelCompact(pModel, pResBuf, pModel->capacity);
   memcpy(pLocalReducer->pBufForInterpo, pResBuf->data, pLocalReducer->nResultBufSize);
 
 #ifdef _DEBUG_VIEW
@@ -1204,9 +1207,11 @@ bool doGenerateFinalResults(SSqlObj *pSql, SLocalReducer *pLocalReducer, bool no
   int32_t             startIndex = pQueryInfo->fieldsInfo.numOfOutputCols - pQueryInfo->groupbyExpr.numOfGroupCols;
 
   for (int32_t i = 0; i < pQueryInfo->groupbyExpr.numOfGroupCols; ++i) {
+    int16_t offset = getColumnModelOffset(pModel, startIndex + i);
+    SSchema* pSchema = getColumnModelSchema(pModel, startIndex + i);
+    
     memcpy(pInterpoInfo->pTags[i],
-           pLocalReducer->pBufForInterpo + pModel->colOffset[startIndex + i] * pResBuf->numOfElems,
-           pModel->pFields[startIndex + i].bytes);
+           pLocalReducer->pBufForInterpo + offset * pResBuf->numOfElems, pSchema->bytes);
   }
 
   taosInterpoSetStartInfo(&pLocalReducer->interpolationInfo, pResBuf->numOfElems, pQueryInfo->interpoType);
@@ -1218,7 +1223,7 @@ bool doGenerateFinalResults(SSqlObj *pSql, SLocalReducer *pLocalReducer, bool no
 void resetOutputBuf(SQueryInfo* pQueryInfo, SLocalReducer *pLocalReducer) {  // reset output buffer to the beginning
   for (int32_t i = 0; i < pQueryInfo->fieldsInfo.numOfOutputCols; ++i) {
     pLocalReducer->pCtx[i].aOutputBuf =
-        pLocalReducer->pResultBuf->data + tscFieldInfoGetOffset(pQueryInfo, i) * pLocalReducer->resColModel->maxCapacity;
+        pLocalReducer->pResultBuf->data + tscFieldInfoGetOffset(pQueryInfo, i) * pLocalReducer->resColModel->capacity;
   }
 
   memset(pLocalReducer->pResultBuf, 0, pLocalReducer->nResultBufSize + sizeof(tFilePage));
@@ -1270,7 +1275,7 @@ static bool doInterpolationForCurrentGroup(SSqlObj *pSql) {
     int32_t remain = taosNumOfRemainPoints(pInterpoInfo);
     TSKEY   ekey = taosGetRevisedEndKey(etime, pQueryInfo->order.order, pQueryInfo->nAggTimeInterval, pQueryInfo->intervalTimeUnit, p);
     int32_t rows = taosGetNumOfResultWithInterpo(pInterpoInfo, (TSKEY *)pLocalReducer->pBufForInterpo, remain,
-                                                 pQueryInfo->nAggTimeInterval, ekey, pLocalReducer->resColModel->maxCapacity);
+                                                 pQueryInfo->nAggTimeInterval, ekey, pLocalReducer->resColModel->capacity);
     if (rows > 0) {  // do interpo
       doInterpolateResult(pSql, pLocalReducer, false);
     }
@@ -1302,7 +1307,7 @@ static bool doHandleLastRemainData(SSqlObj *pSql) {
 
       etime = taosGetRevisedEndKey(etime, pQueryInfo->order.order, pQueryInfo->nAggTimeInterval, pQueryInfo->intervalTimeUnit, precision);
       int32_t rows = taosGetNumOfResultWithInterpo(pInterpoInfo, NULL, 0, pQueryInfo->nAggTimeInterval, etime,
-                                                   pLocalReducer->resColModel->maxCapacity);
+                                                   pLocalReducer->resColModel->capacity);
       if (rows > 0) {  // do interpo
         doInterpolateResult(pSql, pLocalReducer, true);
       }
@@ -1391,7 +1396,7 @@ int32_t tscDoLocalreduce(SSqlObj *pSql) {
 
   // clear buffer
   handleUnprocessedRow(pCmd, pLocalReducer, tmpBuffer);
-  tColModel *pModel = pLocalReducer->pDesc->pSchema;
+  SColumnModel *pModel = pLocalReducer->pDesc->pColumnModel;
 
   while (1) {
     if (isAllSourcesCompleted(pLocalReducer)) {
@@ -1408,14 +1413,14 @@ int32_t tscDoLocalreduce(SSqlObj *pSql) {
     SLocalDataSource *pOneDataSrc = pLocalReducer->pLocalDataSrc[pTree->pNode[0].index];
 
     tColModelAppend(pModel, tmpBuffer, pOneDataSrc->filePage.data, pOneDataSrc->rowIdx, 1,
-                    pOneDataSrc->pMemBuffer->pColModel->maxCapacity);
+                    pOneDataSrc->pMemBuffer->pColumnModel->capacity);
 
 #if defined(_DEBUG_VIEW)
     printf("chosen row:\t");
     SSrcColumnInfo colInfo[256] = {0};
     tscGetSrcColumnInfo(colInfo, pQueryInfo);
 
-    tColModelDisplayEx(pModel, tmpBuffer->data, tmpBuffer->numOfElems, pModel->maxCapacity, colInfo);
+    tColModelDisplayEx(pModel, tmpBuffer->data, tmpBuffer->numOfElems, pModel->capacity, colInfo);
 #endif
 
     if (pLocalReducer->discard) {
@@ -1470,7 +1475,7 @@ int32_t tscDoLocalreduce(SSqlObj *pSql) {
          * continue to process results instead of return results.
          */
         if ((!sameGroup && pResBuf->numOfElems > 0) ||
-            (pResBuf->numOfElems == pLocalReducer->resColModel->maxCapacity)) {
+            (pResBuf->numOfElems == pLocalReducer->resColModel->capacity)) {
           // does not belong to the same group
           bool notSkipped = doGenerateFinalResults(pSql, pLocalReducer, !sameGroup);
 

src/client/src/tscServer.c

@@ -901,7 +901,7 @@ int tscLaunchSTableSubqueries(SSqlObj *pSql) {
 
   tExtMemBuffer **  pMemoryBuf = NULL;
   tOrderDescriptor *pDesc = NULL;
-  tColModel *       pModel = NULL;
+  SColumnModel *       pModel = NULL;
 
   pRes->qhandle = 1;  // hack the qhandle check
 
@@ -1181,7 +1181,7 @@ void tscRetrieveFromVnodeCallBack(void *param, TAOS_RES *tres, int numOfRows) {
     tscTrace("%p sub:%p retrieve numOfRows:%d totalNumOfRows:%d from ip:%u,vid:%d,orderOfSub:%d", pPObj, pSql,
              pRes->numOfRows, pState->numOfRetrievedRows, pSvd->ip, pSvd->vnode, idx);
     
-    if (num > tsMaxNumOfOrderedResults) {
+    if (num > tsMaxNumOfOrderedResults && tscIsProjectionQueryOnSTable(pQueryInfo, 0)) {
       tscError("%p sub:%p num of OrderedRes is too many, max allowed:%" PRId64 " , current:%" PRId64,
           pPObj, pSql, tsMaxNumOfOrderedResults, num);
       tscAbortFurtherRetryRetrieval(trsupport, tres, TSDB_CODE_SORTED_RES_TOO_MANY);
@@ -1194,7 +1194,7 @@ void tscRetrieveFromVnodeCallBack(void *param, TAOS_RES *tres, int numOfRows) {
     SSrcColumnInfo colInfo[256] = {0};
 
     tscGetSrcColumnInfo(colInfo, pQueryInfo);
-    tColModelDisplayEx(pDesc->pSchema, pRes->data, pRes->numOfRows, pRes->numOfRows, colInfo);
+    tColModelDisplayEx(pDesc->pColumnModel, pRes->data, pRes->numOfRows, pRes->numOfRows, colInfo);
 #endif
     if (tsTotalTmpDirGB != 0 && tsAvailTmpDirGB < tsMinimalTmpDirGB) {
       tscError("%p sub:%p client disk space remain %.3f GB, need at least %.3f GB, stop query", pPObj, pSql,
@@ -1202,6 +1202,7 @@ void tscRetrieveFromVnodeCallBack(void *param, TAOS_RES *tres, int numOfRows) {
       tscAbortFurtherRetryRetrieval(trsupport, tres, TSDB_CODE_CLI_NO_DISKSPACE);
       return;
     }
+    
     int32_t ret = saveToBuffer(trsupport->pExtMemBuffer[idx], pDesc, trsupport->localBuffer, pRes->data,
                                pRes->numOfRows, pQueryInfo->groupbyExpr.orderType);
     if (ret < 0) {
@@ -1214,17 +1215,17 @@ void tscRetrieveFromVnodeCallBack(void *param, TAOS_RES *tres, int numOfRows) {
 
   } else {  // all data has been retrieved to client
     /* data in from current vnode is stored in cache and disk */
-    uint32_t numOfRowsFromVnode = trsupport->pExtMemBuffer[idx]->numOfAllElems + trsupport->localBuffer->numOfElems;
+    uint32_t numOfRowsFromVnode = trsupport->pExtMemBuffer[idx]->numOfTotalElems + trsupport->localBuffer->numOfElems;
     tscTrace("%p sub:%p all data retrieved from ip:%u,vid:%d, numOfRows:%d, orderOfSub:%d", pPObj, pSql, pSvd->ip,
              pSvd->vnode, numOfRowsFromVnode, idx);
 
-    tColModelCompact(pDesc->pSchema, trsupport->localBuffer, pDesc->pSchema->maxCapacity);
+    tColModelCompact(pDesc->pColumnModel, trsupport->localBuffer, pDesc->pColumnModel->capacity);
 
 #ifdef _DEBUG_VIEW
     printf("%" PRIu64 " rows data flushed to disk:\n", trsupport->localBuffer->numOfElems);
     SSrcColumnInfo colInfo[256] = {0};
     tscGetSrcColumnInfo(colInfo, pQueryInfo);
-    tColModelDisplayEx(pDesc->pSchema, trsupport->localBuffer->data, trsupport->localBuffer->numOfElems,
+    tColModelDisplayEx(pDesc->pColumnModel, trsupport->localBuffer->data, trsupport->localBuffer->numOfElems,
                        trsupport->localBuffer->numOfElems, colInfo);
 #endif
     
@@ -1256,7 +1257,7 @@ void tscRetrieveFromVnodeCallBack(void *param, TAOS_RES *tres, int numOfRows) {
     }
 
     // all sub-queries are returned, start to local merge process
-    pDesc->pSchema->maxCapacity = trsupport->pExtMemBuffer[idx]->numOfElemsPerPage;
+    pDesc->pColumnModel->capacity = trsupport->pExtMemBuffer[idx]->numOfElemsPerPage;
 
     tscTrace("%p retrieve from %d vnodes completed.final NumOfRows:%d,start to build loser tree", pPObj,
              pState->numOfTotal, pState->numOfRetrievedRows);
@@ -1516,7 +1517,7 @@ void tscUpdateVnodeInQueryMsg(SSqlObj *pSql, char *buf) {
   char *          pStart = buf + tsRpcHeadSize;
   SQueryMeterMsg *pQueryMsg = (SQueryMeterMsg *)pStart;
 
-  if (UTIL_METER_IS_NOMRAL_METER(pMeterMetaInfo)) {  // pSchema == NULL, query on meter
+  if (UTIL_METER_IS_NOMRAL_METER(pMeterMetaInfo)) {  // pColumnModel == NULL, query on meter
     SMeterMeta *pMeterMeta = pMeterMetaInfo->pMeterMeta;
     pQueryMsg->vnode = htons(pMeterMeta->vpeerDesc[pSql->index].vnode);
   } else {  // query on metric

src/client/src/tscStream.c

@@ -396,7 +396,9 @@ static void tscSetSlidingWindowInfo(SSqlObj *pSql, SSqlStream *pStream) {
   int64_t minSlidingTime =
       (pStream->precision == TSDB_TIME_PRECISION_MICRO) ? tsMinSlidingTime * 1000L : tsMinSlidingTime;
 
-  if (pQueryInfo->nSlidingTime < minSlidingTime) {
+  if (pQueryInfo->nSlidingTime == -1) {
+    pQueryInfo->nSlidingTime = pQueryInfo->nAggTimeInterval;
+  } else if (pQueryInfo->nSlidingTime < minSlidingTime) {
     tscWarn("%p stream:%p, original sliding value:%" PRId64 " too small, reset to:%" PRId64 "", pSql, pStream,
         pQueryInfo->nSlidingTime, minSlidingTime);
 

src/client/src/tscUtil.c

@@ -676,7 +676,7 @@ int32_t tscGetDataBlockFromList(void* pHashList, SDataBlockList* pDataBlockList,
                                 STableDataBlocks** dataBlocks) {
   *dataBlocks = NULL;
 
-  STableDataBlocks** t1 = (STableDataBlocks**)taosGetDataFromHash(pHashList, (const char*)&id, sizeof(id));
+  STableDataBlocks** t1 = (STableDataBlocks**)taosGetDataFromHashTable(pHashList, (const char*)&id, sizeof(id));
   if (t1 != NULL) {
     *dataBlocks = *t1;
   }

src/inc/hash.h

@@ -43,10 +43,10 @@ typedef struct SHashEntry {
 
 typedef struct HashObj {
   SHashEntry **hashList;
-  uint32_t     capacity;
-  int          size;
-  _hash_fn_t   hashFp;
-  bool         multithreadSafe;  // enable lock
+  uint32_t     capacity;         // number of slots
+  int          size;             // number of elements in hash table
+  _hash_fn_t   hashFp;           // hash function
+  bool         multithreadSafe;  // enable lock or not
 
 #if defined LINUX
   pthread_rwlock_t lock;
@@ -57,11 +57,13 @@ typedef struct HashObj {
 } HashObj;
 
 void *taosInitHashTable(uint32_t capacity, _hash_fn_t fn, bool multithreadSafe);
+void  taosDeleteFromHashTable(HashObj *pObj, const char *key, uint32_t keyLen);
 
 int32_t taosAddToHashTable(HashObj *pObj, const char *key, uint32_t keyLen, void *data, uint32_t size);
-void    taosDeleteFromHashTable(HashObj *pObj, const char *key, uint32_t keyLen);
+int32_t taosNumElemsInHashTable(HashObj *pObj);
+
+char *taosGetDataFromHashTable(HashObj *pObj, const char *key, uint32_t keyLen);
 
-char *taosGetDataFromHash(HashObj *pObj, const char *key, uint32_t keyLen);
 
 void taosCleanUpHashTable(void *handle);
 

src/inc/textbuffer.h

@@ -19,21 +19,15 @@
 extern "C" {
 #endif
 
-#include <stdio.h>
-#include <stdlib.h>
-#include <string.h>
-
-#include "tutil.h"
+#include "os.h"
 #include "taosmsg.h"
+#include "tutil.h"
 
 #define DEFAULT_PAGE_SIZE 16384  // 16k larger than the SHistoInfo
 #define MIN_BUFFER_SIZE (1 << 19)
 #define MAX_TMPFILE_PATH_LENGTH PATH_MAX
 #define INITIAL_ALLOCATION_BUFFER_SIZE 64
 
-// forward declare
-struct tTagSchema;
-
 typedef enum EXT_BUFFER_FLUSH_MODEL {
   /*
    * all data that have been flushed to disk is belonged to the same group
@@ -61,12 +55,12 @@ typedef struct tFlushoutData {
   tFlushoutInfo *pFlushoutInfo;
 } tFlushoutData;
 
-typedef struct tFileMeta {
+typedef struct SFileInfo {
   uint32_t      nFileSize;  // in pages
-  uint32_t      nPageSize;
+  uint32_t      pageSize;
   uint32_t      numOfElemsInFile;
   tFlushoutData flushoutData;
-} tFileMeta;
+} SFileInfo;
 
 typedef struct tFilePage {
   uint64_t numOfElems;
@@ -78,65 +72,73 @@ typedef struct tFilePagesItem {
   tFilePage              item;
 } tFilePagesItem;
 
-typedef struct tColModel {
-  int32_t         maxCapacity;
+typedef struct SSchemaEx {
+  struct SSchema field;
+  int16_t        offset;
+} SSchemaEx;
+
+typedef struct SColumnModel {
+  int32_t    capacity;
   int32_t    numOfCols;
-  int16_t *       colOffset;
-  struct SSchema *pFields;
-} tColModel;
+  int16_t    rowSize;
+  SSchemaEx *pFields;
+} SColumnModel;
 
-typedef struct tOrderIdx {
-  int32_t numOfOrderedCols;
+typedef struct SColumnOrderInfo {
+  int32_t numOfCols;
   int16_t pData[];
-} tOrderIdx;
+} SColumnOrderInfo;
 
 typedef struct tOrderDescriptor {
-  union {
-    struct tTagSchema *pTagSchema;
-    tColModel *        pSchema;
-  };
+  SColumnModel *   pColumnModel;
   int32_t          tsOrder;  // timestamp order type if exists
-  tOrderIdx orderIdx;
+  SColumnOrderInfo orderIdx;
 } tOrderDescriptor;
 
 typedef struct tExtMemBuffer {
-  int32_t nMaxSizeInPages;
-
+  int32_t inMemCapacity;
   int32_t nElemSize;
-  int32_t nPageSize;
-
-  int32_t numOfAllElems;
+  int32_t pageSize;
+  int32_t numOfTotalElems;
   int32_t numOfElemsInBuffer;
   int32_t numOfElemsPerPage;
+  int16_t numOfInMemPages;
 
-  int16_t         numOfPagesInMem;
   tFilePagesItem *pHead;
   tFilePagesItem *pTail;
 
-  tFileMeta fileMeta;
-
-  char  dataFilePath[MAX_TMPFILE_PATH_LENGTH];
-  FILE *dataFile;
-
-  tColModel *pColModel;
+  char *    path;
+  FILE *    file;
+  SFileInfo fileMeta;
 
+  SColumnModel *         pColumnModel;
   EXT_BUFFER_FLUSH_MODEL flushModel;
 } tExtMemBuffer;
 
+/**
+ *
+ * @param fileNamePattern
+ * @param dstPath
+ */
 void getTmpfilePath(const char *fileNamePattern, char *dstPath);
 
-/*
- * create ext-memory buffer
+/**
+ *
+ * @param inMemSize
+ * @param elemSize
+ * @param pModel
+ * @return
  */
-void tExtMemBufferCreate(tExtMemBuffer **pMemBuffer, int32_t numOfBufferSize, int32_t elemSize,
-                         const char *tmpDataFilePath, tColModel *pModel);
+tExtMemBuffer *createExtMemBuffer(int32_t inMemSize, int32_t elemSize, SColumnModel *pModel);
 
-/*
- * destroy ext-memory buffer
+/**
+ *
+ * @param pMemBuffer
+ * @return
  */
-void tExtMemBufferDestroy(tExtMemBuffer **pMemBuffer);
+void *destoryExtMemBuffer(tExtMemBuffer *pMemBuffer);
 
-/*
+/**
  * @param pMemBuffer
  * @param data       input data pointer
  * @param numOfRows  number of rows in data
@@ -145,12 +147,15 @@ void tExtMemBufferDestroy(tExtMemBuffer **pMemBuffer);
  */
 int16_t tExtMemBufferPut(tExtMemBuffer *pMemBuffer, void *data, int32_t numOfRows);
 
-/*
- * flush all data into disk and release all in-memory buffer
+/**
+ *
+ * @param pMemBuffer
+ * @return
  */
 bool tExtMemBufferFlush(tExtMemBuffer *pMemBuffer);
 
-/*
+/**
+ *
  * remove all data that has been put into buffer, including in buffer or
  * ext-buffer(disk)
  */
@@ -163,11 +168,44 @@ void tExtMemBufferClear(tExtMemBuffer *pMemBuffer);
  */
 bool tExtMemBufferLoadData(tExtMemBuffer *pMemBuffer, tFilePage *pFilePage, int32_t flushIdx, int32_t pageIdx);
 
+/**
+ *
+ * @param pMemBuffer
+ * @return
+ */
 bool tExtMemBufferIsAllDataInMem(tExtMemBuffer *pMemBuffer);
 
-tColModel *tColModelCreate(SSchema *field, int32_t numOfCols, int32_t maxCapacity);
+/**
+ *
+ * @param fields
+ * @param numOfCols
+ * @param blockCapacity
+ * @return
+ */
+SColumnModel *createColumnModel(SSchema *fields, int32_t numOfCols, int32_t blockCapacity);
+
+/**
+ *
+ * @param pSrc
+ * @return
+ */
+SColumnModel *cloneColumnModel(SColumnModel *pSrc);
+
+/**
+ *
+ * @param pModel
+ */
+void destroyColumnModel(SColumnModel *pModel);
+
+/*
+ * compress data into consecutive block without hole in data
+ */
+void tColModelCompact(SColumnModel *pModel, tFilePage *inputBuffer, int32_t maxElemsCapacity);
+
+void     tColModelErase(SColumnModel *pModel, tFilePage *inputBuffer, int32_t maxCapacity, int32_t s, int32_t e);
+SSchema *getColumnModelSchema(SColumnModel *pColumnModel, int32_t index);
 
-void tColModelDestroy(tColModel *pModel);
+int16_t getColumnModelOffset(SColumnModel *pColumnModel, int32_t index);
 
 typedef struct SSrcColumnInfo {
   int32_t functionId;
@@ -177,68 +215,18 @@ typedef struct SSrcColumnInfo {
 /*
  * display data in column format model for debug purpose only
  */
-void tColModelDisplay(tColModel *pModel, void *pData, int32_t numOfRows, int32_t maxCount);
+void tColModelDisplay(SColumnModel *pModel, void *pData, int32_t numOfRows, int32_t maxCount);
 
-void tColModelDisplayEx(tColModel *pModel, void *pData, int32_t numOfRows, int32_t maxCount, SSrcColumnInfo *pInfo);
+void tColModelDisplayEx(SColumnModel *pModel, void *pData, int32_t numOfRows, int32_t maxCount, SSrcColumnInfo *pInfo);
 
-/*
- * compress data into consecutive block without hole in data
- */
-void tColModelCompact(tColModel *pModel, tFilePage *inputBuffer, int32_t maxElemsCapacity);
-
-void tColModelErase(tColModel *pModel, tFilePage *inputBuffer, int32_t maxCapacity, int32_t s, int32_t e);
-
-tOrderDescriptor *tOrderDesCreate(int32_t *orderColIdx, int32_t numOfOrderCols, tColModel *pModel, int32_t tsOrderType);
+tOrderDescriptor *tOrderDesCreate(const int32_t *orderColIdx, int32_t numOfOrderCols, SColumnModel *pModel,
+                                  int32_t tsOrderType);
 
 void tOrderDescDestroy(tOrderDescriptor *pDesc);
 
-void tColModelAppend(tColModel *dstModel, tFilePage *dstPage, void *srcData, int32_t srcStartRows,
+void tColModelAppend(SColumnModel *dstModel, tFilePage *dstPage, void *srcData, int32_t srcStartRows,
                      int32_t numOfRowsToWrite, int32_t srcCapacity);
 
-///////////////////////////////////////////////////////////////////////////////////////////////////////
-typedef struct MinMaxEntry {
-  union {
-    double  dMinVal;
-    int32_t iMinVal;
-    int64_t i64MinVal;
-  };
-  union {
-    double  dMaxVal;
-    int32_t iMaxVal;
-    int64_t i64MaxVal;
-  };
-} MinMaxEntry;
-
-typedef struct tMemBucketSegment {
-  int32_t         numOfSlots;
-  MinMaxEntry *   pBoundingEntries;
-  tExtMemBuffer **pBuffer;
-} tMemBucketSegment;
-
-typedef struct tMemBucket {
-  int16_t numOfSegs;
-  int16_t nTotalSlots;
-  int16_t nSlotsOfSeg;
-  int16_t dataType;
-
-  int16_t nElemSize;
-  int32_t numOfElems;
-
-  int32_t nTotalBufferSize;
-  int32_t maxElemsCapacity;
-
-  int16_t nPageSize;
-  int16_t numOfTotalPages;
-  int16_t numOfAvailPages; /* remain available buffer pages */
-
-  tMemBucketSegment *pSegs;
-  tOrderDescriptor * pOrderDesc;
-
-  MinMaxEntry nRange;
-
-  void (*HashFunc)(struct tMemBucket *pBucket, void *value, int16_t *segIdx, int16_t *slotIdx);
-} tMemBucket;
-
 typedef int (*__col_compar_fn_t)(tOrderDescriptor *, int32_t numOfRows, int32_t idx1, int32_t idx2, char *data);
 
 void tColDataQSort(tOrderDescriptor *, int32_t numOfRows, int32_t start, int32_t end, char *data, int32_t orderType);
@@ -253,19 +241,6 @@ int32_t compare_a(tOrderDescriptor *, int32_t numOfRow1, int32_t s1, char *data1
 int32_t compare_d(tOrderDescriptor *, int32_t numOfRow1, int32_t s1, char *data1, int32_t numOfRow2, int32_t s2,
                   char *data2);
 
-tMemBucket* tMemBucketCreate(int32_t totalSlots, int32_t nBufferSize, int16_t nElemSize,
-                      int16_t dataType, tOrderDescriptor *pDesc);
-
-void tMemBucketDestroy(tMemBucket *pBucket);
-
-void tMemBucketPut(tMemBucket *pBucket, void *data, int32_t numOfRows);
-
-double getPercentile(tMemBucket *pMemBucket, double percent);
-
-void tBucketIntHash(tMemBucket *pBucket, void *value, int16_t *segIdx, int16_t *slotIdx);
-
-void tBucketDoubleHash(tMemBucket *pBucket, void *value, int16_t *segIdx, int16_t *slotIdx);
-
 #ifdef __cplusplus
 }
 #endif

src/inc/thistogram.h

@@ -20,8 +20,6 @@
 extern "C" {
 #endif
 
-#include "tskiplist.h"
-
 #define USE_ARRAYLIST
 
 #define MAX_HISTOGRAM_BIN 500

src/inc/tinterpolation.h

@@ -78,7 +78,7 @@ int32_t taosNumOfRemainPoints(SInterpolationInfo *pInterpoInfo);
  */
 int32_t taosDoInterpoResult(SInterpolationInfo *pInterpoInfo, int16_t interpoType, tFilePage **data,
                             int32_t numOfRawDataInRows, int32_t outputRows, int64_t nInterval,
-                            const int64_t *pPrimaryKeyArray, tColModel *pModel, char **srcData, int64_t *defaultVal,
+                            const int64_t *pPrimaryKeyArray, SColumnModel *pModel, char **srcData, int64_t *defaultVal,
                             const int32_t *functionIDs, int32_t bufSize);
 
 int taosDoLinearInterpolation(int32_t type, SPoint *point1, SPoint *point2, SPoint *point);

src/inc/tpercentile.h

+/*
+ * 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/>.
+ */
+
+#ifndef TDENGINE_TPERCENTILE_H
+#define TDENGINE_TPERCENTILE_H
+
+#include "textbuffer.h"
+
+typedef struct MinMaxEntry {
+  union {
+    double  dMinVal;
+    int32_t iMinVal;
+    int64_t i64MinVal;
+  };
+  union {
+    double  dMaxVal;
+    int32_t iMaxVal;
+    int64_t i64MaxVal;
+  };
+} MinMaxEntry;
+
+typedef struct tMemBucketSegment {
+  int32_t         numOfSlots;
+  MinMaxEntry *   pBoundingEntries;
+  tExtMemBuffer **pBuffer;
+} tMemBucketSegment;
+
+typedef struct tMemBucket {
+  int16_t numOfSegs;
+  int16_t nTotalSlots;
+  int16_t nSlotsOfSeg;
+  int16_t dataType;
+  
+  int16_t nElemSize;
+  int32_t numOfElems;
+  
+  int32_t nTotalBufferSize;
+  int32_t maxElemsCapacity;
+  
+  int16_t pageSize;
+  int16_t numOfTotalPages;
+  int16_t numOfAvailPages; /* remain available buffer pages */
+  
+  tMemBucketSegment *pSegs;
+  tOrderDescriptor * pOrderDesc;
+  
+  MinMaxEntry nRange;
+  
+  void (*HashFunc)(struct tMemBucket *pBucket, void *value, int16_t *segIdx, int16_t *slotIdx);
+} tMemBucket;
+
+tMemBucket *tMemBucketCreate(int32_t totalSlots, int32_t nBufferSize, int16_t nElemSize, int16_t dataType,
+                             tOrderDescriptor *pDesc);
+
+void tMemBucketDestroy(tMemBucket *pBucket);
+
+void tMemBucketPut(tMemBucket *pBucket, void *data, int32_t numOfRows);
+
+double getPercentile(tMemBucket *pMemBucket, double percent);
+
+void tBucketIntHash(tMemBucket *pBucket, void *value, int16_t *segIdx, int16_t *slotIdx);
+
+void tBucketDoubleHash(tMemBucket *pBucket, void *value, int16_t *segIdx, int16_t *slotIdx);
+
+#endif  // TDENGINE_TPERCENTILE_H

src/inc/tresultBuf.h

+#ifndef TDENGINE_VNODEQUERYUTIL_H
+#define TDENGINE_VNODEQUERYUTIL_H
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#include "os.h"
+#include "textbuffer.h"
+
+typedef struct SIDList {
+  uint32_t alloc;
+  int32_t  size;
+  int32_t* pData;
+} SIDList;
+
+typedef struct SQueryResultBuf {
+  int32_t  numOfRowsPerPage;
+  int32_t  numOfPages;
+  int64_t  totalBufSize;
+  int32_t  fd;                  // data file fd
+  int32_t  allocateId;          // allocated page id
+  int32_t  incStep;             // minimum allocated pages
+  char*    pBuf;                // mmap buffer pointer
+  char*    path;                // file path
+  
+  uint32_t numOfAllocGroupIds;  // number of allocated id list
+  void*    idsTable;            // id hash table
+  SIDList* list;                // for each id, there is a page id list
+} SQueryResultBuf;
+
+/**
+ * create disk-based result buffer
+ * @param pResultBuf
+ * @param size
+ * @param rowSize
+ * @return
+ */
+int32_t createResultBuf(SQueryResultBuf** pResultBuf, int32_t size, int32_t rowSize);
+
+/**
+ *
+ * @param pResultBuf
+ * @param groupId
+ * @param pageId
+ * @return
+ */
+tFilePage* getNewDataBuf(SQueryResultBuf* pResultBuf, int32_t groupId, int32_t* pageId);
+
+/**
+ *
+ * @param pResultBuf
+ * @return
+ */
+int32_t getNumOfRowsPerPage(SQueryResultBuf* pResultBuf);
+
+/**
+ *
+ * @param pResultBuf
+ * @param groupId
+ * @return
+ */
+SIDList getDataBufPagesIdList(SQueryResultBuf* pResultBuf, int32_t groupId);
+
+/**
+ * get the specified buffer page by id
+ * @param pResultBuf
+ * @param id
+ * @return
+ */
+tFilePage* getResultBufferPageById(SQueryResultBuf* pResultBuf, int32_t id);
+
+/**
+ * get the total buffer size in the format of disk file
+ * @param pResultBuf
+ * @return
+ */
+int32_t getResBufSize(SQueryResultBuf* pResultBuf);
+
+/**
+ * get the number of groups in the result buffer
+ * @param pResultBuf
+ * @return
+ */
+int32_t getNumOfResultBufGroupId(SQueryResultBuf* pResultBuf);
+
+/**
+ * destroy result buffer
+ * @param pResultBuf
+ */
+void destroyResultBuf(SQueryResultBuf* pResultBuf);
+
+/**
+ *
+ * @param pList
+ * @return
+ */
+int32_t getLastPageId(SIDList *pList);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif  // TDENGINE_VNODEQUERYUTIL_H

src/system/detail/inc/vnodeQueryImpl.h

@@ -13,8 +13,8 @@
  * along with this program. If not, see <http://www.gnu.org/licenses/>.
  */
 
-#ifndef TDENGINE_VNODEQUERYUTIL_H
-#define TDENGINE_VNODEQUERYUTIL_H
+#ifndef TDENGINE_VNODEQUERYIMPL_H
+#define TDENGINE_VNODEQUERYIMPL_H
 
 #ifdef __cplusplus
 extern "C" {
@@ -120,7 +120,7 @@ typedef enum {
 typedef int (*__block_search_fn_t)(char* data, int num, int64_t key, int order);
 
 static FORCE_INLINE SMeterObj* getMeterObj(void* hashHandle, int32_t sid) {
-  return *(SMeterObj**)taosGetDataFromHash(hashHandle, (const char*) &sid, sizeof(sid));
+  return *(SMeterObj**)taosGetDataFromHashTable(hashHandle, (const char*) &sid, sizeof(sid));
 }
 
 bool isQueryKilled(SQuery* pQuery);
@@ -209,7 +209,7 @@ int32_t vnodeGetHeaderFile(SQueryRuntimeEnv *pRuntimeEnv, int32_t fileIndex);
  * @param ekey
  * @return
  */
-SMeterQueryInfo* createMeterQueryInfo(SQuery* pQuery, TSKEY skey, TSKEY ekey);
+SMeterQueryInfo* createMeterQueryInfo(SQuery* pQuery, int32_t sid, TSKEY skey, TSKEY ekey);
 
 /**
  * Destroy meter query info
@@ -224,7 +224,7 @@ void destroyMeterQueryInfo(SMeterQueryInfo *pMeterQueryInfo, int32_t numOfCols);
  * @param skey
  * @param ekey
  */
-void changeMeterQueryInfoForSuppleQuery(SMeterQueryInfo *pMeterQueryInfo, TSKEY skey, TSKEY ekey);
+void changeMeterQueryInfoForSuppleQuery(SQueryResultBuf* pResultBuf, SMeterQueryInfo *pMeterQueryInfo, TSKEY skey, TSKEY ekey);
 
 /**
  * add the new allocated disk page to meter query info
@@ -276,11 +276,11 @@ void displayInterResult(SData** pdata, SQuery* pQuery, int32_t numOfRows);
 
 void vnodePrintQueryStatistics(SMeterQuerySupportObj* pSupporter);
 
-void clearGroupResultBuf(SOutputRes* pOneOutputRes, int32_t nOutputCols);
-void copyGroupResultBuf(SOutputRes* dst, const SOutputRes* src, int32_t nOutputCols);
+void clearGroupResultBuf(SQueryRuntimeEnv *pRuntimeEnv, SOutputRes *pOneOutputRes);
+void copyGroupResultBuf(SQueryRuntimeEnv *pRuntimeEnv, SOutputRes* dst, const SOutputRes* src);
 
-void resetSlidingWindowInfo(SSlidingWindowInfo* pSlidingWindowInfo, int32_t numOfCols);
-void clearCompletedSlidingWindows(SSlidingWindowInfo* pSlidingWindowInfo, int32_t numOfCols);
+void resetSlidingWindowInfo(SQueryRuntimeEnv *pRuntimeEnv, SSlidingWindowInfo* pSlidingWindowInfo);
+void clearCompletedSlidingWindows(SQueryRuntimeEnv* pRuntimeEnv);
 int32_t numOfClosedSlidingWindow(SSlidingWindowInfo* pSlidingWindowInfo);
 void closeSlidingWindow(SSlidingWindowInfo* pSlidingWindowInfo, int32_t slot);
 void closeAllSlidingWindow(SSlidingWindowInfo* pSlidingWindowInfo);
@@ -289,4 +289,4 @@ void closeAllSlidingWindow(SSlidingWindowInfo* pSlidingWindowInfo);
 }
 #endif
 
-#endif  // TDENGINE_VNODEQUERYUTIL_H
+#endif  // TDENGINE_VNODEQUERYIMPL_H

src/system/detail/inc/vnodeRead.h

@@ -21,6 +21,7 @@ extern "C" {
 #endif
 
 #include "os.h"
+#include "tresultBuf.h"
 
 #include "tinterpolation.h"
 #include "vnodeTagMgmt.h"
@@ -84,10 +85,15 @@ typedef struct SQueryCostSummary {
   int64_t tmpBufferInDisk;  // size of buffer for intermediate result
 } SQueryCostSummary;
 
+typedef struct SPosInfo {
+  int64_t pageId;
+  int32_t rowId;
+} SPosInfo;
+
 typedef struct SOutputRes {
   uint16_t     numOfRows;
   int32_t      nAlloc;
-  tFilePage**  result;
+  SPosInfo     pos;
   SResultInfo* resultInfo;
 } SOutputRes;
 
@@ -174,16 +180,17 @@ typedef struct SQueryRuntimeEnv {
    * So we keep a copy of the support structure as well as the cache block data itself.
    */
   SCacheBlock         cacheBlock;
+  
+  SQueryResultBuf*    pResultBuf;
+  bool                stableQuery;  // is super table query or not
 } SQueryRuntimeEnv;
 
-/* intermediate result during multimeter query involves interval */
+/* intermediate pos during multimeter query involves interval */
 typedef struct SMeterQueryInfo {
   int64_t      lastKey;
   int64_t      skey;
   int64_t      ekey;
   int32_t      numOfRes;
-  uint32_t     numOfPages;
-  uint32_t     numOfAlloc;
   int32_t      reverseIndex;    // reversed output indicator, start from (numOfRes-1)
   int16_t      reverseFillRes;  // denote if reverse fill the results in supplementary scan required or not
   int16_t      queryRangeSet;   // denote if the query range is set, only available for interval query
@@ -191,7 +198,7 @@ typedef struct SMeterQueryInfo {
   int64_t      tag;
   STSCursor    cur;
   SResultInfo* resultInfo;
-  uint32_t*    pageList;
+  int32_t      sid; // for retrieve the page id list
 } SMeterQueryInfo;
 
 typedef struct SMeterDataInfo {
@@ -235,16 +242,8 @@ typedef struct SMeterQuerySupportObj {
    */
   int32_t meterIdx;
 
-  int32_t meterOutputFd;
-  int32_t lastPageId;
-  int32_t numOfPages;
   int32_t numOfGroupResultPages;
   int32_t groupResultSize;
-
-  char*   meterOutputMMapBuf;
-  int64_t bufSize;
-  char    extBufFile[256];  // external file name
-
   SMeterDataInfo* pMeterDataInfo;
 
   TSKEY* tsList;

src/system/detail/inc/vnodeTagMgmt.h

@@ -32,21 +32,14 @@ extern "C" {
  * Note:
  * 1. we implement a quick sort algorithm, may remove it later.
  */
-
-typedef struct tTagSchema {
-  struct SSchema *pSchema;
-  int32_t         numOfCols;
-  int32_t         colOffset[];
-} tTagSchema;
-
 typedef struct tSidSet {
   int32_t            numOfSids;
   int32_t            numOfSubSet;
   SMeterSidExtInfo **pSids;
   int32_t *          starterPos;  // position of each subgroup, generated according to
 
-  tTagSchema *pTagSchema;
-  tOrderIdx   orderIdx;
+  SColumnModel      *pColumnModel;
+  SColumnOrderInfo   orderIdx;
 } tSidSet;
 
 typedef int32_t (*__ext_compar_fn_t)(const void *p1, const void *p2, void *param);
@@ -54,8 +47,6 @@ typedef int32_t (*__ext_compar_fn_t)(const void *p1, const void *p2, void *param
 tSidSet *tSidSetCreate(struct SMeterSidExtInfo **pMeterSidExtInfo, int32_t numOfMeters, SSchema *pSchema,
                        int32_t numOfTags, SColIndexEx *colList, int32_t numOfOrderCols);
 
-tTagSchema *tCreateTagSchema(SSchema *pSchema, int32_t numOfTagCols);
-
 int32_t *calculateSubGroup(void **pSids, int32_t numOfMeters, int32_t *numOfSubset, tOrderDescriptor *pOrderDesc,
                            __ext_compar_fn_t compareFn);
 

src/system/detail/src/mgmtDnodeInt.c

@@ -210,7 +210,7 @@ char *mgmtBuildCreateMeterIe(STabObj *pMeter, char *pMsg, int vnode) {
 
   for (int i = 0; i < pMeter->numOfColumns; ++i) {
     pCreateMeter->schema[i].type = pSchema[i].type;
-    /* strcpy(pCreateMeter->schema[i].name, pSchema[i].name); */
+    /* strcpy(pCreateMeter->schema[i].name, pColumnModel[i].name); */
     pCreateMeter->schema[i].bytes = htons(pSchema[i].bytes);
     pCreateMeter->schema[i].colId = htons(pSchema[i].colId);
   }

src/system/detail/src/mgmtSupertableQuery.c

@@ -70,7 +70,7 @@ static int32_t tabObjResultComparator(const void* p1, const void* p2, void* para
   STabObj* pNode1 = (STabObj*)p1;
   STabObj* pNode2 = (STabObj*)p2;
 
-  for (int32_t i = 0; i < pOrderDesc->orderIdx.numOfOrderedCols; ++i) {
+  for (int32_t i = 0; i < pOrderDesc->orderIdx.numOfCols; ++i) {
     int32_t colIdx = pOrderDesc->orderIdx.pData[i];
 
     char* f1 = NULL;
@@ -86,7 +86,9 @@ static int32_t tabObjResultComparator(const void* p1, const void* p2, void* para
     } else {
       f1 = mgmtMeterGetTag(pNode1, colIdx, NULL);
       f2 = mgmtMeterGetTag(pNode2, colIdx, &schema);
-      assert(schema.type == pOrderDesc->pTagSchema->pSchema[colIdx].type);
+      
+      SSchema* pSchema = getColumnModelSchema(pOrderDesc->pColumnModel, colIdx);
+      assert(schema.type == pSchema->type);
     }
 
     int32_t ret = doCompare(f1, f2, schema.type, schema.bytes);
@@ -109,7 +111,7 @@ static int32_t tabObjResultComparator(const void* p1, const void* p2, void* para
  * @param pOrderIndexInfo
  * @param numOfTags
  */
-static void mgmtUpdateOrderTagColIndex(SMetricMetaMsg* pMetricMetaMsg, int32_t tableIndex, tOrderIdx* pOrderIndexInfo,
+static void mgmtUpdateOrderTagColIndex(SMetricMetaMsg* pMetricMetaMsg, int32_t tableIndex, SColumnOrderInfo* pOrderIndexInfo,
      int32_t numOfTags) {
   SMetricMetaElemMsg* pElem = (SMetricMetaElemMsg*)((char*)pMetricMetaMsg + pMetricMetaMsg->metaElem[tableIndex]);
   SColIndexEx* groupColumnList = (SColIndexEx*)((char*)pMetricMetaMsg + pElem->groupbyTagColumnList);
@@ -123,7 +125,7 @@ static void mgmtUpdateOrderTagColIndex(SMetricMetaMsg* pMetricMetaMsg, int32_t t
     }
   }
   
-  pOrderIndexInfo->numOfOrderedCols = numOfGroupbyTags;
+  pOrderIndexInfo->numOfCols = numOfGroupbyTags;
 }
 
 // todo merge sort function with losertree used
@@ -143,14 +145,14 @@ void mgmtReorganizeMetersInMetricMeta(SMetricMetaMsg* pMetricMetaMsg, int32_t ta
    */
   tOrderDescriptor* descriptor =
       (tOrderDescriptor*)calloc(1, sizeof(tOrderDescriptor) + sizeof(int32_t) * pElem->numOfGroupCols);
-  descriptor->pTagSchema = tCreateTagSchema(pTagSchema, pMetric->numOfTags);
-  descriptor->orderIdx.numOfOrderedCols = pElem->numOfGroupCols;
+  descriptor->pColumnModel = createColumnModel(pTagSchema, pMetric->numOfTags, 1);
+  descriptor->orderIdx.numOfCols = pElem->numOfGroupCols;
 
   int32_t* startPos = NULL;
   int32_t  numOfSubset = 1;
   
   mgmtUpdateOrderTagColIndex(pMetricMetaMsg, tableIndex, &descriptor->orderIdx, pMetric->numOfTags);
-  if (descriptor->orderIdx.numOfOrderedCols > 0) {
+  if (descriptor->orderIdx.numOfCols > 0) {
     tQSortEx(pRes->pRes, POINTER_BYTES, 0, pRes->num - 1, descriptor, tabObjResultComparator);
     startPos = calculateSubGroup(pRes->pRes, pRes->num, &numOfSubset, descriptor, tabObjResultComparator);
   } else {
@@ -166,7 +168,7 @@ void mgmtReorganizeMetersInMetricMeta(SMetricMetaMsg* pMetricMetaMsg, int32_t ta
    */
   qsort(pRes->pRes, (size_t)pRes->num, POINTER_BYTES, tabObjVGIDComparator);
 
-  free(descriptor->pTagSchema);
+  free(descriptor->pColumnModel);
   free(descriptor);
   free(startPos);
 }
@@ -291,15 +293,15 @@ static void orderResult(SMetricMetaMsg* pMetricMetaMsg, tQueryResultset* pRes, i
   STabObj* pMetric = mgmtGetMeter(pElem->meterId);
   SSchema* pTagSchema = (SSchema*)(pMetric->schema + pMetric->numOfColumns * sizeof(SSchema));
 
-  descriptor->pTagSchema = tCreateTagSchema(pTagSchema, pMetric->numOfTags);
+  descriptor->pColumnModel = createColumnModel(pTagSchema, pMetric->numOfTags, 1);
 
   descriptor->orderIdx.pData[0] = colIndex;
-  descriptor->orderIdx.numOfOrderedCols = 1;
+  descriptor->orderIdx.numOfCols = 1;
 
   // sort results list
   tQSortEx(pRes->pRes, POINTER_BYTES, 0, pRes->num - 1, descriptor, tabObjResultComparator);
 
-  free(descriptor->pTagSchema);
+  free(descriptor->pColumnModel);
   free(descriptor);
 }
 

src/system/detail/src/vnodeFile.c

@@ -127,7 +127,7 @@ int vnodeCreateHeadDataFile(int vnode, int fileId, char *headName, char *dataNam
   if (symlink(dDataName, dataName) != 0) return -1;
   if (symlink(dLastName, lastName) != 0) return -1;
 
-  dPrint("vid:%d, fileId:%d, empty header file:%s dataFile:%s lastFile:%s on disk:%s is created ",
+  dPrint("vid:%d, fileId:%d, empty header file:%s file:%s lastFile:%s on disk:%s is created ",
           vnode, fileId, headName, dataName, lastName, path);
 
   return 0;

src/system/detail/src/vnodeQueryImpl.c

@@ -56,7 +56,7 @@ static void    vnodeInitLoadCompBlockInfo(SLoadCompBlockInfo *pCompBlockLoadInfo
 static int32_t moveToNextBlock(SQueryRuntimeEnv *pRuntimeEnv, int32_t step, __block_search_fn_t searchFn,
                                bool loadData);
 static int32_t doMergeMetersResultsToGroupRes(SMeterQuerySupportObj *pSupporter, SQuery *pQuery,
-                                              SQueryRuntimeEnv *pRuntimeEnv, SMeterDataInfo *pMeterHeadDataInfo,
+                                              SQueryRuntimeEnv *pRuntimeEnv, SMeterDataInfo *pMeterDataInfo,
                                               int32_t start, int32_t end);
 
 static TSKEY getTimestampInCacheBlock(SQueryRuntimeEnv *pRuntimeEnv, SCacheBlock *pBlock, int32_t index);
@@ -68,9 +68,6 @@ static int32_t getNextDataFileCompInfo(SQueryRuntimeEnv *pRuntimeEnv, SMeterObj
 static void setGroupOutputBuffer(SQueryRuntimeEnv *pRuntimeEnv, SOutputRes *pResult);
 
 static void getAlignedIntervalQueryRange(SQueryRuntimeEnv *pRuntimeEnv, TSKEY keyInData, TSKEY skey, TSKEY ekey);
-static void doApplyIntervalQueryOnBlock(SMeterQuerySupportObj *pSupporter, SMeterQueryInfo *pMeterQueryInfo,
-                                        SBlockInfo *pBlockInfo, int64_t *pPrimaryCol, SField *pFields,
-                                        __block_search_fn_t searchFn);
 
 static int32_t saveResult(SMeterQuerySupportObj *pSupporter, SMeterQueryInfo *pMeterQueryInfo, int32_t numOfResult);
 static void applyIntervalQueryOnBlock(SMeterQuerySupportObj *pSupporter, SMeterDataInfo *pMeterDataInfo,
@@ -589,7 +586,8 @@ static void setExecParams(SQuery *pQuery, SQLFunctionCtx *pCtx, int64_t StartQue
                           char *primaryColumnData, int32_t size, int32_t functionId, SField *pField, bool hasNull,
                           int32_t blockStatus, void *param, int32_t scanFlag);
 
-void        createGroupResultBuf(SQuery *pQuery, SOutputRes *pOneResult, bool isMetricQuery);
+void createQueryResultBuf(SQueryRuntimeEnv *pRuntimeEnv, SOutputRes *pResultRow, bool isSTableQuery, SPosInfo *posInfo);
+
 static void destroyGroupResultBuf(SOutputRes *pOneOutputRes, int32_t nOutputCols);
 
 static int32_t binarySearchForBlockImpl(SCompBlock *pBlock, int32_t numOfBlocks, TSKEY skey, int32_t order) {
@@ -1576,7 +1574,7 @@ static SOutputRes *doSetSlidingWindowFromKey(SSlidingWindowInfo *pSlidingWindowI
                                 SWindowStatus **pStatus) {
   int32_t p = -1;
 
-  int32_t *p1 = (int32_t *)taosGetDataFromHash(pSlidingWindowInfo->hashList, pData, bytes);
+  int32_t *p1 = (int32_t *)taosGetDataFromHashTable(pSlidingWindowInfo->hashList, pData, bytes);
   if (p1 != NULL) {
     p = *p1;
 
@@ -1614,7 +1612,7 @@ static SOutputRes *doSetSlidingWindowFromKey(SSlidingWindowInfo *pSlidingWindowI
   return &pSlidingWindowInfo->pResult[p];
 }
 
-static int32_t initSlidingWindowInfo(SSlidingWindowInfo *pSlidingWindowInfo, int32_t threshold, int16_t type,
+static int32_t initSlidingWindowInfo(SSlidingWindowInfo *pSlidingWindowInfo, int32_t threshold, int16_t type, int32_t rowSizes,
                                  SOutputRes *pRes) {
   pSlidingWindowInfo->capacity = threshold;
   pSlidingWindowInfo->threshold = threshold;
@@ -1627,7 +1625,17 @@ static int32_t initSlidingWindowInfo(SSlidingWindowInfo *pSlidingWindowInfo, int
   pSlidingWindowInfo->curIndex = -1;
   pSlidingWindowInfo->size = 0;
   pSlidingWindowInfo->pResult = pRes;
+
+//  createResultBuf(&pSlidingWindowInfo->pResultBuf, 10, rowSizes);
+  
   pSlidingWindowInfo->pStatus = calloc(threshold, sizeof(SWindowStatus));
+//  pSlidingWindowInfo->pResultInfo = calloc(threshold, POINTER_BYTES);
+  
+//  for(int32_t i = 0; i < threshold; ++i) {
+//    pSlidingWindowInfo->pResultInfo[i] = calloc((size_t)numOfOutput, sizeof(SResultInfo));
+    
+    
+//  }
   
   if (pSlidingWindowInfo->pStatus == NULL || pSlidingWindowInfo->hashList == NULL) {
     return -1;
@@ -1643,17 +1651,19 @@ static void destroySlidingWindowInfo(SSlidingWindowInfo *pSlidingWindowInfo) {
   }
 
   taosCleanUpHashTable(pSlidingWindowInfo->hashList);
+//  destroyResultBuf(pSlidingWindowInfo->pResultBuf);
+  
   tfree(pSlidingWindowInfo->pStatus);
 }
 
-void resetSlidingWindowInfo(SSlidingWindowInfo *pSlidingWindowInfo, int32_t numOfCols) {
+void resetSlidingWindowInfo(SQueryRuntimeEnv *pRuntimeEnv, SSlidingWindowInfo *pSlidingWindowInfo) {
   if (pSlidingWindowInfo == NULL || pSlidingWindowInfo->capacity == 0) {
     return;
   }
 
   for (int32_t i = 0; i < pSlidingWindowInfo->size; ++i) {
     SOutputRes *pOneRes = &pSlidingWindowInfo->pResult[i];
-    clearGroupResultBuf(pOneRes, numOfCols);
+    clearGroupResultBuf(pRuntimeEnv, pOneRes);
   }
 
   memset(pSlidingWindowInfo->pStatus, 0, sizeof(SWindowStatus) * pSlidingWindowInfo->capacity);
@@ -1669,7 +1679,8 @@ void resetSlidingWindowInfo(SSlidingWindowInfo *pSlidingWindowInfo, int32_t numO
   pSlidingWindowInfo->prevSKey = 0;
 }
 
-void clearCompletedSlidingWindows(SSlidingWindowInfo *pSlidingWindowInfo, int32_t numOfCols) {
+void clearCompletedSlidingWindows(SQueryRuntimeEnv* pRuntimeEnv) {
+  SSlidingWindowInfo* pSlidingWindowInfo = &pRuntimeEnv->swindowResInfo;
   if (pSlidingWindowInfo == NULL || pSlidingWindowInfo->capacity == 0 || pSlidingWindowInfo->size == 0) {
     return;
   }
@@ -1689,24 +1700,25 @@ void clearCompletedSlidingWindows(SSlidingWindowInfo *pSlidingWindowInfo, int32_
   }
 
   int32_t remain = pSlidingWindowInfo->size - i;
+  
   //clear remain list
   memmove(pSlidingWindowInfo->pStatus, &pSlidingWindowInfo->pStatus[i], remain * sizeof(SWindowStatus));
   memset(&pSlidingWindowInfo->pStatus[remain], 0, (pSlidingWindowInfo->capacity - remain) * sizeof(SWindowStatus));
   
   for(int32_t k = 0; k < remain; ++k) {
-    copyGroupResultBuf(&pSlidingWindowInfo->pResult[k], &pSlidingWindowInfo->pResult[i + k], numOfCols);
+    copyGroupResultBuf(pRuntimeEnv, &pSlidingWindowInfo->pResult[k], &pSlidingWindowInfo->pResult[i + k]);
   }
   
   for(int32_t k = remain; k < pSlidingWindowInfo->size; ++k) {
     SOutputRes *pOneRes = &pSlidingWindowInfo->pResult[k];
-    clearGroupResultBuf(pOneRes, numOfCols);
+    clearGroupResultBuf(pRuntimeEnv, pOneRes);
   }
 
   pSlidingWindowInfo->size = remain;
 
   for(int32_t k = 0; k < pSlidingWindowInfo->size; ++k) {
     SWindowStatus* pStatus = &pSlidingWindowInfo->pStatus[k];
-    int32_t *p = (int32_t*) taosGetDataFromHash(pSlidingWindowInfo->hashList, (const char*)&pStatus->window.skey, TSDB_KEYSIZE);
+    int32_t *p = (int32_t*) taosGetDataFromHashTable(pSlidingWindowInfo->hashList, (const char*)&pStatus->window.skey, TSDB_KEYSIZE);
     int32_t v = *p;
     v = (v - i);
   
@@ -1720,14 +1732,12 @@ void clearCompletedSlidingWindows(SSlidingWindowInfo *pSlidingWindowInfo, int32_
 }
 
 int32_t numOfClosedSlidingWindow(SSlidingWindowInfo *pSlidingWindowInfo) {
-  for (int32_t i = 0; i < pSlidingWindowInfo->size; ++i) {
-    SWindowStatus *pStatus = &pSlidingWindowInfo->pStatus[i];
-    if (pStatus->closed == false) {
-      return i;
-    }
+  int32_t i = 0;
+  while(i < pSlidingWindowInfo->size && pSlidingWindowInfo->pStatus[i].closed) {
+    ++i;
   }
   
-  return 0;
+  return i;
 }
 
 void closeSlidingWindow(SSlidingWindowInfo* pSlidingWindowInfo, int32_t slot) {
@@ -2469,20 +2479,21 @@ void setExecParams(SQuery *pQuery, SQLFunctionCtx *pCtx, int64_t startQueryTimes
 }
 
 // set the output buffer for the selectivity + tag query
-static void setCtxTagColumnInfo(SQuery *pQuery, SQueryRuntimeEnv *pRuntimeEnv) {
+static void setCtxTagColumnInfo(SQuery *pQuery, SQLFunctionCtx *pCtx) {
   if (isSelectivityWithTagsQuery(pQuery)) {
     int32_t         num = 0;
-    SQLFunctionCtx *pCtx = NULL;
+    SQLFunctionCtx *p = NULL;
+    
     int16_t         tagLen = 0;
 
     SQLFunctionCtx **pTagCtx = calloc(pQuery->numOfOutputCols, POINTER_BYTES);
     for (int32_t i = 0; i < pQuery->numOfOutputCols; ++i) {
       SSqlFuncExprMsg *pSqlFuncMsg = &pQuery->pSelectExpr[i].pBase;
       if (pSqlFuncMsg->functionId == TSDB_FUNC_TAG_DUMMY || pSqlFuncMsg->functionId == TSDB_FUNC_TS_DUMMY) {
-        tagLen += pRuntimeEnv->pCtx[i].outputBytes;
-        pTagCtx[num++] = &pRuntimeEnv->pCtx[i];
+        tagLen += pCtx[i].outputBytes;
+        pTagCtx[num++] = &pCtx[i];
       } else if ((aAggs[pSqlFuncMsg->functionId].nStatus & TSDB_FUNCSTATE_SELECTIVITY) != 0) {
-        pCtx = &pRuntimeEnv->pCtx[i];
+        p = &pCtx[i];
       } else if (pSqlFuncMsg->functionId == TSDB_FUNC_TS || pSqlFuncMsg->functionId == TSDB_FUNC_TAG) {
         // tag function may be the group by tag column
         // ts may be the required primary timestamp column
@@ -2492,14 +2503,14 @@ static void setCtxTagColumnInfo(SQuery *pQuery, SQueryRuntimeEnv *pRuntimeEnv) {
       }
     }
 
-    pCtx->tagInfo.pTagCtxList = pTagCtx;
-    pCtx->tagInfo.numOfTagCols = num;
-    pCtx->tagInfo.tagsLen = tagLen;
+    p->tagInfo.pTagCtxList = pTagCtx;
+    p->tagInfo.numOfTagCols = num;
+    p->tagInfo.tagsLen = tagLen;
   }
 }
 
 static int32_t setupQueryRuntimeEnv(SMeterObj *pMeterObj, SQuery *pQuery, SQueryRuntimeEnv *pRuntimeEnv,
-                                    SSchema *pTagsSchema, int16_t order, bool isMetricQuery) {
+                                    SColumnModel *pTagsSchema, int16_t order, bool isSTableQuery) {
   dTrace("QInfo:%p setup runtime env", GET_QINFO_ADDR(pQuery));
 
   pRuntimeEnv->pMeterObj = pMeterObj;
@@ -2520,8 +2531,10 @@ static int32_t setupQueryRuntimeEnv(SMeterObj *pMeterObj, SQuery *pQuery, SQuery
     SQLFunctionCtx *pCtx = &pRuntimeEnv->pCtx[i];
 
     if (TSDB_COL_IS_TAG(pSqlFuncMsg->colInfo.flag)) {  // process tag column info
-      pCtx->inputType = pTagsSchema[pColIndexEx->colIdx].type;
-      pCtx->inputBytes = pTagsSchema[pColIndexEx->colIdx].bytes;
+      SSchema* pSchema = getColumnModelSchema(pTagsSchema, pColIndexEx->colIdx);
+      
+      pCtx->inputType = pSchema->type;
+      pCtx->inputBytes = pSchema->bytes;
     } else {
       pCtx->inputType = GET_COLUMN_TYPE(pQuery, i);
       pCtx->inputBytes = GET_COLUMN_BYTES(pQuery, i);
@@ -2567,15 +2580,15 @@ static int32_t setupQueryRuntimeEnv(SMeterObj *pMeterObj, SQuery *pQuery, SQuery
 
     // set the intermediate result output buffer
     SResultInfo *pResInfo = &pRuntimeEnv->resultInfo[i];
-    setResultInfoBuf(pResInfo, pQuery->pSelectExpr[i].interResBytes, isMetricQuery);
+    setResultInfoBuf(pResInfo, pQuery->pSelectExpr[i].interResBytes, isSTableQuery);
   }
 
   // if it is group by normal column, do not set output buffer, the output buffer is pResult
-  if (!isGroupbyNormalCol(pQuery->pGroupbyExpr) && !isMetricQuery) {
+  if (!isGroupbyNormalCol(pQuery->pGroupbyExpr) && !isSTableQuery) {
     resetCtxOutputBuf(pRuntimeEnv);
   }
 
-  setCtxTagColumnInfo(pQuery, pRuntimeEnv);
+  setCtxTagColumnInfo(pQuery, pRuntimeEnv->pCtx);
 
   // for loading block data in memory
   assert(vnodeList[pMeterObj->vnode].cfg.rowsInFileBlock == pMeterObj->pointsPerFileBlock);
@@ -2651,6 +2664,7 @@ static int64_t getOldestKey(int32_t numOfFiles, int64_t fileId, SVnodeCfg *pCfg)
 }
 
 bool isQueryKilled(SQuery *pQuery) {
+  return false;
   SQInfo *pQInfo = (SQInfo *)GET_QINFO_ADDR(pQuery);
 
   /*
@@ -4120,21 +4134,29 @@ static void allocMemForInterpo(SMeterQuerySupportObj *pSupporter, SQuery *pQuery
   }
 }
 
-static int32_t allocateOutputBufForGroup(SMeterQuerySupportObj *pSupporter, SQuery *pQuery, bool isMetricQuery) {
-  int32_t slot = 0;
+static int32_t createQueryOutputBuffer(SMeterQuerySupportObj *pSupporter, SQuery *pQuery, bool isSTableQuery) {
+  SQueryRuntimeEnv* pRuntimeEnv = &pSupporter->runtimeEnv;
+  
+  int32_t numOfRows = 0;
 
   if (isGroupbyNormalCol(pQuery->pGroupbyExpr) || (pQuery->nAggTimeInterval > 0 && pQuery->slidingTime > 0)) {
-    slot = 10000;
+    numOfRows = 10000;
   } else {
-    slot = pSupporter->pSidSet->numOfSubSet;
+    numOfRows = pSupporter->pSidSet->numOfSubSet;
   }
 
-  pSupporter->pResult = calloc(1, sizeof(SOutputRes) * slot);
+  createResultBuf(&pRuntimeEnv->pResultBuf, 100, pQuery->rowSize);
+  
+  // total number of initial results
+  pSupporter->pResult = calloc(numOfRows, sizeof(SOutputRes));
   if (pSupporter->pResult == NULL) {
     return TSDB_CODE_SERV_OUT_OF_MEMORY;
   }
 
-  for (int32_t k = 0; k < slot; ++k) {
+  int32_t pageId = -1;
+  tFilePage* page = NULL;
+  
+  for (int32_t k = 0; k < numOfRows; ++k) {
     SOutputRes *pOneRes = &pSupporter->pResult[k];
     pOneRes->nAlloc = 1;
 
@@ -4142,14 +4164,23 @@ static int32_t allocateOutputBufForGroup(SMeterQuerySupportObj *pSupporter, SQue
      * for single table top/bottom query, the output for group by normal column, the output rows is
      * equals to the maximum rows, instead of 1.
      */
-    if (!isMetricQuery && isTopBottomQuery(pQuery)) {
+    if (!isSTableQuery && isTopBottomQuery(pQuery)) {
       assert(pQuery->numOfOutputCols > 1);
 
       SSqlFunctionExpr *pExpr = &pQuery->pSelectExpr[1];
       pOneRes->nAlloc = pExpr->pBase.arg[0].argValue.i64;
     }
 
-    createGroupResultBuf(pQuery, pOneRes, isMetricQuery);
+    if (page == NULL || page->numOfElems >= pRuntimeEnv->numOfRowsPerPage) {
+      page = getNewDataBuf(pRuntimeEnv->pResultBuf, 0, &pageId);
+    }
+    
+    assert(pageId >= 0);
+    
+    SPosInfo posInfo = {.pageId = pageId, .rowId = page->numOfElems};
+    
+    createQueryResultBuf(pRuntimeEnv, pOneRes, isSTableQuery, &posInfo);
+    page->numOfElems += 1; // next row is available
   }
 
   return TSDB_CODE_SUCCESS;
@@ -4211,6 +4242,32 @@ _error_clean:
   return TSDB_CODE_SERV_OUT_OF_MEMORY;
 }
 
+static int32_t getRowParamForMultiRowsOutput(SQuery* pQuery, bool isSTableQuery) {
+  int32_t rowparam = 1;
+  
+  if (isTopBottomQuery(pQuery) && (!isSTableQuery)) {
+    rowparam = pQuery->pSelectExpr[1].pBase.arg->argValue.i64;
+  }
+  
+  return rowparam;
+}
+
+static int32_t getNumOfRowsInResultPage(SQuery* pQuery, bool isSTableQuery) {
+  int32_t rowSize = pQuery->rowSize * getRowParamForMultiRowsOutput(pQuery, isSTableQuery);
+  return (DEFAULT_INTERN_BUF_SIZE - sizeof(tFilePage)) / rowSize;
+}
+
+static char* getPosInResultPage(SQueryRuntimeEnv* pRuntimeEnv, int32_t columnIndex, SOutputRes* pResult) {
+  SQuery* pQuery = pRuntimeEnv->pQuery;
+  tFilePage* page = getResultBufferPageById(pRuntimeEnv->pResultBuf, pResult->pos.pageId);
+  
+  int32_t numOfRows = getNumOfRowsInResultPage(pQuery, pRuntimeEnv->stableQuery);
+  int32_t realRowId = pResult->pos.rowId * getRowParamForMultiRowsOutput(pQuery, pRuntimeEnv->stableQuery);
+  
+  return ((char*)page->data) + pRuntimeEnv->offset[columnIndex] * numOfRows +
+  pQuery->pSelectExpr[columnIndex].resBytes * realRowId;
+}
+
 int32_t vnodeQuerySingleMeterPrepare(SQInfo *pQInfo, SMeterObj *pMeterObj, SMeterQuerySupportObj *pSupporter,
                                      void *param) {
   SQuery *pQuery = &pQInfo->query;
@@ -4281,8 +4338,9 @@ int32_t vnodeQuerySingleMeterPrepare(SQInfo *pQInfo, SMeterObj *pMeterObj, SMete
 
   vnodeRecordAllFiles(pQInfo, pMeterObj->vnode);
   
+  pRuntimeEnv->numOfRowsPerPage = getNumOfRowsInResultPage(pQuery, false);
   if (isGroupbyNormalCol(pQuery->pGroupbyExpr) || (pQuery->nAggTimeInterval > 0 && pQuery->slidingTime > 0)) {
-    if ((code = allocateOutputBufForGroup(pSupporter, pQuery, false)) != TSDB_CODE_SUCCESS) {
+    if ((code = createQueryOutputBuffer(pSupporter, pQuery, false)) != TSDB_CODE_SUCCESS) {
       return code;
     }
 
@@ -4293,8 +4351,7 @@ int32_t vnodeQuerySingleMeterPrepare(SQInfo *pQInfo, SMeterObj *pMeterObj, SMete
       type = TSDB_DATA_TYPE_TIMESTAMP;
     }
 
-    // todo bug!
-    initSlidingWindowInfo(&pRuntimeEnv->swindowResInfo, 3, type, pSupporter->pResult);
+    initSlidingWindowInfo(&pRuntimeEnv->swindowResInfo, 3, type, pQuery->rowSize, pSupporter->pResult);
   }
 
   pSupporter->rawSKey = pQuery->skey;
@@ -4404,6 +4461,8 @@ int32_t vnodeQuerySingleMeterPrepare(SQInfo *pQInfo, SMeterObj *pMeterObj, SMete
 
   // the pQuery->skey is changed during normalizedFirstQueryRange, so set the newest lastkey value
   pQuery->lastKey = pQuery->skey;
+  pRuntimeEnv->stableQuery = false;
+  
   return TSDB_CODE_SUCCESS;
 }
 
@@ -4437,15 +4496,6 @@ void vnodeQueryFreeQInfoEx(SQInfo *pQInfo) {
     }
   }
 
-  if (FD_VALID(pSupporter->meterOutputFd)) {
-    assert(pSupporter->meterOutputMMapBuf != NULL);
-    dTrace("QInfo:%p disk-based output buffer during query:%" PRId64 " bytes", pQInfo, pSupporter->bufSize);
-    munmap(pSupporter->meterOutputMMapBuf, pSupporter->bufSize);
-    tclose(pSupporter->meterOutputFd);
-
-    unlink(pSupporter->extBufFile);
-  }
-
   tSidSetDestroy(&pSupporter->pSidSet);
 
   if (pSupporter->pMeterDataInfo != NULL) {
@@ -4498,12 +4548,7 @@ int32_t vnodeMultiMeterQueryPrepare(SQInfo *pQInfo, SQuery *pQuery, void *param)
   pQuery->lastKey = pQuery->skey;
 
   // create runtime environment
-  SSchema *pTagSchema = NULL;
-
-  tTagSchema *pTagSchemaInfo = pSupporter->pSidSet->pTagSchema;
-  if (pTagSchemaInfo != NULL) {
-    pTagSchema = pTagSchemaInfo->pSchema;
-  }
+  SColumnModel *pTagSchemaInfo = pSupporter->pSidSet->pColumnModel;
 
   // get one queried meter
   SMeterObj *pMeter = getMeterObj(pSupporter->pMetersHashTable, pSupporter->pSidSet->pSids[0]->sid);
@@ -4517,7 +4562,7 @@ int32_t vnodeMultiMeterQueryPrepare(SQInfo *pQInfo, SQuery *pQuery, void *param)
     tsBufSetTraverseOrder(pRuntimeEnv->pTSBuf, order);
   }
 
-  int32_t ret = setupQueryRuntimeEnv(pMeter, pQuery, &pSupporter->runtimeEnv, pTagSchema, TSQL_SO_ASC, true);
+  int32_t ret = setupQueryRuntimeEnv(pMeter, pQuery, &pSupporter->runtimeEnv, pTagSchemaInfo, TSQL_SO_ASC, true);
   if (ret != TSDB_CODE_SUCCESS) {
     return ret;
   }
@@ -4530,45 +4575,25 @@ int32_t vnodeMultiMeterQueryPrepare(SQInfo *pQInfo, SQuery *pQuery, void *param)
   tSidSetSort(pSupporter->pSidSet);
   vnodeRecordAllFiles(pQInfo, pMeter->vnode);
 
-  if ((ret = allocateOutputBufForGroup(pSupporter, pQuery, true)) != TSDB_CODE_SUCCESS) {
+  if ((ret = createQueryOutputBuffer(pSupporter, pQuery, true)) != TSDB_CODE_SUCCESS) {
     return ret;
   }
 
   if (isGroupbyNormalCol(pQuery->pGroupbyExpr)) {  // group by columns not tags;
     int16_t type = getGroupbyColumnType(pQuery, pQuery->pGroupbyExpr);
-    initSlidingWindowInfo(&pRuntimeEnv->swindowResInfo, 10039, type, pSupporter->pResult);
+    initSlidingWindowInfo(&pRuntimeEnv->swindowResInfo, 4096, type, pQuery->rowSize, pSupporter->pResult);
   }
 
   if (pQuery->nAggTimeInterval != 0) {
-    getTmpfilePath("tb_metric_mmap", pSupporter->extBufFile);
-    pSupporter->meterOutputFd = open(pSupporter->extBufFile, O_CREAT | O_RDWR, 0666);
-
-    if (!FD_VALID(pSupporter->meterOutputFd)) {
-      dError("QInfo:%p failed to create file: %s on disk. %s", pQInfo, pSupporter->extBufFile, strerror(errno));
-      return TSDB_CODE_SERV_OUT_OF_MEMORY;
-    }
-
-    pSupporter->numOfPages = pSupporter->numOfMeters;
-
-    ret = ftruncate(pSupporter->meterOutputFd, pSupporter->numOfPages * DEFAULT_INTERN_BUF_SIZE);
+    // one page for each table at least
+    ret = createResultBuf(&pRuntimeEnv->pResultBuf, pSupporter->numOfMeters, pQuery->rowSize);
     if (ret != TSDB_CODE_SUCCESS) {
-      dError("QInfo:%p failed to create intermediate result output file:%s. %s", pQInfo, pSupporter->extBufFile,
-             strerror(errno));
-      return TSDB_CODE_SERV_NO_DISKSPACE;
-    }
-
-    pRuntimeEnv->numOfRowsPerPage = (DEFAULT_INTERN_BUF_SIZE - sizeof(tFilePage)) / pQuery->rowSize;
-    pSupporter->lastPageId = -1;
-    pSupporter->bufSize = pSupporter->numOfPages * DEFAULT_INTERN_BUF_SIZE;
-
-    pSupporter->meterOutputMMapBuf =
-        mmap(NULL, pSupporter->bufSize, PROT_READ | PROT_WRITE, MAP_SHARED, pSupporter->meterOutputFd, 0);
-    if (pSupporter->meterOutputMMapBuf == MAP_FAILED) {
-      dError("QInfo:%p failed to map temp file: %s. %s", pQInfo, pSupporter->extBufFile, strerror(errno));
-      return TSDB_CODE_SERV_OUT_OF_MEMORY;
+      return ret;
     }
   }
   
+  pRuntimeEnv->numOfRowsPerPage = getNumOfRowsInResultPage(pQuery, true);
+  
   // metric query do not invoke interpolation, it will be done at the second-stage merge
   if (!isPointInterpoQuery(pQuery)) {
     pQuery->interpoType = TSDB_INTERPO_NONE;
@@ -4577,6 +4602,7 @@ int32_t vnodeMultiMeterQueryPrepare(SQInfo *pQInfo, SQuery *pQuery, void *param)
   TSKEY revisedStime = taosGetIntervalStartTimestamp(pSupporter->rawSKey, pQuery->nAggTimeInterval,
                                                      pQuery->intervalTimeUnit, pQuery->precision);
   taosInitInterpoInfo(&pRuntimeEnv->interpoInfo, pQuery->order.order, revisedStime, 0, 0);
+  pRuntimeEnv->stableQuery = true;
   
   return TSDB_CODE_SUCCESS;
 }
@@ -5061,14 +5087,14 @@ void queryOnBlock(SMeterQuerySupportObj *pSupporter, int64_t *primaryKeys, int32
  * set tag value in SQLFunctionCtx
  * e.g.,tag information into input buffer
  */
-static void doSetTagValueInParam(tTagSchema *pTagSchema, int32_t tagColIdx, SMeterSidExtInfo *pMeterSidInfo,
+static void doSetTagValueInParam(SColumnModel *pTagSchema, int32_t tagColIdx, SMeterSidExtInfo *pMeterSidInfo,
                                  tVariant *param) {
   assert(tagColIdx >= 0);
 
-  int32_t *fieldValueOffset = pTagSchema->colOffset;
+  int16_t offset = getColumnModelOffset(pTagSchema, tagColIdx);
   
-  void *   pStr = (char *)pMeterSidInfo->tags + fieldValueOffset[tagColIdx];
-  SSchema *pCol = &pTagSchema->pSchema[tagColIdx];
+  void *   pStr = (char *)pMeterSidInfo->tags + offset;
+  SSchema *pCol = getColumnModelSchema(pTagSchema, tagColIdx);
 
   tVariantDestroy(param);
 
@@ -5081,7 +5107,7 @@ static void doSetTagValueInParam(tTagSchema *pTagSchema, int32_t tagColIdx, SMet
 
 void vnodeSetTagValueInParam(tSidSet *pSidSet, SQueryRuntimeEnv *pRuntimeEnv, SMeterSidExtInfo *pMeterSidInfo) {
   SQuery *    pQuery = pRuntimeEnv->pQuery;
-  tTagSchema *pTagSchema = pSidSet->pTagSchema;
+  SColumnModel *pTagSchema = pSidSet->pColumnModel;
 
   SSqlFuncExprMsg *pFuncMsg = &pQuery->pSelectExpr[0].pBase;
   if (pQuery->numOfOutputCols == 1 && pFuncMsg->functionId == TSDB_FUNC_TS_COMP) {
@@ -5259,19 +5285,9 @@ void UNUSED_FUNC displayInterResult(SData **pdata, SQuery *pQuery, int32_t numOf
   }
 }
 
-static tFilePage *getFilePage(SMeterQuerySupportObj *pSupporter, int32_t pageId) {
-  assert(pageId <= pSupporter->lastPageId && pageId >= 0);
-  return (tFilePage *)(pSupporter->meterOutputMMapBuf + DEFAULT_INTERN_BUF_SIZE * pageId);
-}
-
-static tFilePage *getMeterDataPage(SMeterQuerySupportObj *pSupporter, SMeterDataInfo *pMeterDataInfo, int32_t pageId) {
-  SMeterQueryInfo *pMeterQueryInfo = pMeterDataInfo->pMeterQInfo;
-  if (pageId >= pMeterQueryInfo->numOfPages) {
-    return NULL;
-  }
-
-  int32_t realId = pMeterQueryInfo->pageList[pageId];
-  return getFilePage(pSupporter, realId);
+static tFilePage *getMeterDataPage(SQueryResultBuf* pResultBuf, SMeterQueryInfo *pMeterQueryInfo, int32_t index) {
+  SIDList pList = getDataBufPagesIdList(pResultBuf, pMeterQueryInfo->sid);
+  return getResultBufferPageById(pResultBuf, pList.pData[index]);
 }
 
 typedef struct Position {
@@ -5287,7 +5303,9 @@ typedef struct SCompSupporter {
 
 int64_t getCurrentTimestamp(SCompSupporter *pSupportor, int32_t meterIdx) {
   Position * pPos = &pSupportor->pPosition[meterIdx];
-  tFilePage *pPage = getMeterDataPage(pSupportor->pSupporter, pSupportor->pMeterDataInfo[meterIdx], pPos->pageIdx);
+  tFilePage *pPage = getMeterDataPage(pSupportor->pSupporter->runtimeEnv.pResultBuf,
+      pSupportor->pMeterDataInfo[meterIdx]->pMeterQInfo, pPos->pageIdx);
+  
   return *(int64_t *)(pPage->data + TSDB_KEYSIZE * pPos->rowIdx);
 }
 
@@ -5295,10 +5313,11 @@ int32_t meterResultComparator(const void *pLeft, const void *pRight, void *param
   int32_t left = *(int32_t *)pLeft;
   int32_t right = *(int32_t *)pRight;
 
-  SCompSupporter *supportor = (SCompSupporter *)param;
+  SCompSupporter *supporter = (SCompSupporter *)param;
+  SQueryResultBuf* pResultBuf = supporter->pSupporter->runtimeEnv.pResultBuf;
   
-  Position leftPos = supportor->pPosition[left];
-  Position rightPos = supportor->pPosition[right];
+  Position leftPos = supporter->pPosition[left];
+  Position rightPos = supporter->pPosition[right];
 
   /* left source is exhausted */
   if (leftPos.pageIdx == -1 && leftPos.rowIdx == -1) {
@@ -5310,10 +5329,10 @@ int32_t meterResultComparator(const void *pLeft, const void *pRight, void *param
     return -1;
   }
 
-  tFilePage *pPageLeft = getMeterDataPage(supportor->pSupporter, supportor->pMeterDataInfo[left], leftPos.pageIdx);
+  tFilePage *pPageLeft = getMeterDataPage(pResultBuf, supporter->pMeterDataInfo[left]->pMeterQInfo, leftPos.pageIdx);
   int64_t    leftTimestamp = *(int64_t *)(pPageLeft->data + TSDB_KEYSIZE * leftPos.rowIdx);
 
-  tFilePage *pPageRight = getMeterDataPage(supportor->pSupporter, supportor->pMeterDataInfo[right], rightPos.pageIdx);
+  tFilePage *pPageRight = getMeterDataPage(pResultBuf, supporter->pMeterDataInfo[right]->pMeterQInfo, rightPos.pageIdx);
   int64_t    rightTimestamp = *(int64_t *)(pPageRight->data + TSDB_KEYSIZE * rightPos.rowIdx);
 
   if (leftTimestamp == rightTimestamp) {
@@ -5373,23 +5392,40 @@ void copyResToQueryResultBuf(SMeterQuerySupportObj *pSupporter, SQuery *pQuery)
   }
 
   SQueryRuntimeEnv *pRuntimeEnv = &pSupporter->runtimeEnv;
-  char *            pStart = pSupporter->meterOutputMMapBuf + DEFAULT_INTERN_BUF_SIZE * (pSupporter->lastPageId + 1) +
-                 pSupporter->groupResultSize * pSupporter->offset;
+  SQueryResultBuf* pResultBuf = pRuntimeEnv->pResultBuf;
+  
+  SIDList list = getDataBufPagesIdList(pResultBuf, 200000 + pSupporter->offset + (pSupporter->subgroupIdx - 1)* 10000);
   
-  uint64_t numOfElem = ((tFilePage *)pStart)->numOfElems;
-  assert(numOfElem <= pQuery->pointsToRead);
+  int32_t total = 0;
+  for(int32_t i = 0; i < list.size; ++i) {
+    tFilePage* pData = getResultBufferPageById(pResultBuf, list.pData[i]);
+    total += pData->numOfElems;
+  }
+  
+  pQuery->sdata[0]->len = total;
+  
+  int32_t offset = 0;
+  for(int32_t num = 0; num < list.size; ++num) {
+    tFilePage* pData = getResultBufferPageById(pResultBuf, list.pData[num]);
     
     for (int32_t i = 0; i < pQuery->numOfOutputCols; ++i) {
-    memcpy(pQuery->sdata[i], pStart, pRuntimeEnv->pCtx[i].outputBytes * numOfElem + sizeof(tFilePage));
-    pStart += pRuntimeEnv->pCtx[i].outputBytes * pQuery->pointsToRead + sizeof(tFilePage);
+      int32_t bytes = pRuntimeEnv->pCtx[i].outputBytes;
+      char* pDest = pQuery->sdata[i]->data;
+      
+      memcpy(pDest + offset*bytes, pData->data + pRuntimeEnv->offset[i] * pData->numOfElems, bytes * pData->numOfElems);
+    }
+    
+    offset += pData->numOfElems;
   }
 
-  pQuery->pointsRead += numOfElem;
+  assert(pQuery->pointsRead == 0);
+  
+  pQuery->pointsRead += pQuery->sdata[0]->len;
   pSupporter->offset += 1;
 }
 
 int32_t doMergeMetersResultsToGroupRes(SMeterQuerySupportObj *pSupporter, SQuery *pQuery, SQueryRuntimeEnv *pRuntimeEnv,
-                                       SMeterDataInfo *pMeterHeadDataInfo, int32_t start, int32_t end) {
+                                       SMeterDataInfo *pMeterDataInfo, int32_t start, int32_t end) {
   // calculate the maximum required space
   if (pSupporter->groupResultSize == 0) {
     for (int32_t i = 0; i < pQuery->numOfOutputCols; ++i) {
@@ -5403,8 +5439,11 @@ int32_t doMergeMetersResultsToGroupRes(SMeterQuerySupportObj *pSupporter, SQuery
 
   int32_t numOfMeters = 0;
   for (int32_t i = start; i < end; ++i) {
-    if (pMeterHeadDataInfo[i].pMeterQInfo->numOfPages > 0 && pMeterHeadDataInfo[i].pMeterQInfo->numOfRes > 0) {
-      pValidMeter[numOfMeters] = &pMeterHeadDataInfo[i];
+    int32_t sid = pMeterDataInfo[i].pMeterQInfo->sid;
+    SIDList list = getDataBufPagesIdList(pRuntimeEnv->pResultBuf, sid);
+    
+    if (list.size > 0 && pMeterDataInfo[i].pMeterQInfo->numOfRes > 0) {
+      pValidMeter[numOfMeters] = &pMeterDataInfo[i];
       // set the merge start position: page:0, index:0
       posArray[numOfMeters].pageIdx = 0;
       posArray[numOfMeters++].rowIdx = 0;
@@ -5433,10 +5472,11 @@ int32_t doMergeMetersResultsToGroupRes(SMeterQuerySupportObj *pSupporter, SQuery
   while (1) {
     int32_t    pos = pTree->pNode[0].index;
     Position * position = &cs.pPosition[pos];
-    tFilePage *pPage = getMeterDataPage(cs.pSupporter, pValidMeter[pos], position->pageIdx);
+    SQueryResultBuf* pResultBuf = cs.pSupporter->runtimeEnv.pResultBuf;
+    tFilePage *pPage = getMeterDataPage(pResultBuf, pValidMeter[pos]->pMeterQInfo, position->pageIdx);
 
     int64_t ts = getCurrentTimestamp(&cs, pos);
-    if (ts == lastTimestamp) {  // merge with the last one
+    if (ts == lastTimestamp) {// merge with the last one
       doMerge(pRuntimeEnv, ts, pPage, position->rowIdx, true);
     } else {
       // copy data to disk buffer
@@ -5448,7 +5488,7 @@ int32_t doMergeMetersResultsToGroupRes(SMeterQuerySupportObj *pSupporter, SQuery
         resetMergeResultBuf(pQuery, pCtx);
       }
 
-      pPage = getMeterDataPage(cs.pSupporter, pValidMeter[pos], position->pageIdx);
+      pPage = getMeterDataPage(pResultBuf, pValidMeter[pos]->pMeterQInfo, position->pageIdx);
       if (pPage->numOfElems <= 0) {  // current source data page is empty
         // do nothing
       } else {
@@ -5464,17 +5504,19 @@ int32_t doMergeMetersResultsToGroupRes(SMeterQuerySupportObj *pSupporter, SQuery
       cs.pPosition[pos].pageIdx += 1;  // try next page
 
       // check if current page is empty or not. if it is empty, ignore it and try next
-      if (cs.pPosition[pos].pageIdx <= cs.pMeterDataInfo[pos]->pMeterQInfo->numOfPages - 1) {
-        tFilePage *newPage = getMeterDataPage(cs.pSupporter, pValidMeter[pos], position->pageIdx);
-        if (newPage->numOfElems <= 0) {
+      SIDList list = getDataBufPagesIdList(pRuntimeEnv->pResultBuf, cs.pMeterDataInfo[pos]->pMeterQInfo->sid);
+      if (cs.pPosition[pos].pageIdx <= list.size - 1) {
+        tFilePage *newPage = getMeterDataPage(pResultBuf, pValidMeter[pos]->pMeterQInfo, position->pageIdx);
+        
         // if current source data page is null, it must be the last page of source output page
+        if (newPage->numOfElems <= 0) {
           cs.pPosition[pos].pageIdx += 1;
-          assert(cs.pPosition[pos].pageIdx >= cs.pMeterDataInfo[pos]->pMeterQInfo->numOfPages - 1);
+          assert(cs.pPosition[pos].pageIdx >= list.size - 1);
         }
       }
 
       // the following code must be executed if current source pages are exhausted
-      if (cs.pPosition[pos].pageIdx >= cs.pMeterDataInfo[pos]->pMeterQInfo->numOfPages) {
+      if (cs.pPosition[pos].pageIdx >= list.size) {
         cs.pPosition[pos].pageIdx = -1;
         cs.pPosition[pos].rowIdx = -1;
 
@@ -5492,8 +5534,8 @@ int32_t doMergeMetersResultsToGroupRes(SMeterQuerySupportObj *pSupporter, SQuery
 
   if (buffer[0]->numOfElems != 0) {  // there are data in buffer
     if (flushFromResultBuf(pSupporter, pQuery, pRuntimeEnv) != TSDB_CODE_SUCCESS) {
-      dError("QInfo:%p failed to flush data into temp file, abort query", GET_QINFO_ADDR(pQuery),
-             pSupporter->extBufFile);
+//      dError("QInfo:%p failed to flush data into temp file, abort query", GET_QINFO_ADDR(pQuery),
+//             pSupporter->extBufFile);
       tfree(pTree);
       tfree(pValidMeter);
       tfree(posArray);
@@ -5518,68 +5560,37 @@ int32_t doMergeMetersResultsToGroupRes(SMeterQuerySupportObj *pSupporter, SQuery
   return pSupporter->numOfGroupResultPages;
 }
 
-static int32_t extendDiskBuf(const SQuery *pQuery, SMeterQuerySupportObj *pSupporter, int32_t numOfPages) {
-  assert(pSupporter->numOfPages * DEFAULT_INTERN_BUF_SIZE == pSupporter->bufSize);
-
-  SQInfo *pQInfo = (SQInfo *)GET_QINFO_ADDR(pQuery);
-
-  int32_t ret = munmap(pSupporter->meterOutputMMapBuf, pSupporter->bufSize);
-  pSupporter->numOfPages = numOfPages;
-
-  /*
-   * disk-based output buffer is exhausted, try to extend the disk-based buffer, the available disk space may
-   * be insufficient
-   */
-  ret = ftruncate(pSupporter->meterOutputFd, pSupporter->numOfPages * DEFAULT_INTERN_BUF_SIZE);
-  if (ret != 0) {
-    dError("QInfo:%p failed to create intermediate result output file:%s. %s", pQInfo, pSupporter->extBufFile,
-           strerror(errno));
-    pQInfo->code = -TSDB_CODE_SERV_NO_DISKSPACE;
-    pQInfo->killed = 1;
-
-    return pQInfo->code;
-  }
-
-  pSupporter->bufSize = pSupporter->numOfPages * DEFAULT_INTERN_BUF_SIZE;
-  pSupporter->meterOutputMMapBuf =
-      mmap(NULL, pSupporter->bufSize, PROT_READ | PROT_WRITE, MAP_SHARED, pSupporter->meterOutputFd, 0);
-
-  if (pSupporter->meterOutputMMapBuf == MAP_FAILED) {
-    dError("QInfo:%p failed to map temp file: %s. %s", pQInfo, pSupporter->extBufFile, strerror(errno));
-    pQInfo->code = -TSDB_CODE_SERV_OUT_OF_MEMORY;
-    pQInfo->killed = 1;
-
-    return pQInfo->code;
-  }
-
-  return TSDB_CODE_SUCCESS;
-}
-
 int32_t flushFromResultBuf(SMeterQuerySupportObj *pSupporter, const SQuery *pQuery,
                            const SQueryRuntimeEnv *pRuntimeEnv) {
-  int32_t numOfMeterResultBufPages = pSupporter->lastPageId + 1;
-  int64_t dstSize = numOfMeterResultBufPages * DEFAULT_INTERN_BUF_SIZE +
-                    pSupporter->groupResultSize * (pSupporter->numOfGroupResultPages + 1);
+  SQueryResultBuf* pResultBuf = pRuntimeEnv->pResultBuf;
+  int32_t capacity = (DEFAULT_INTERN_BUF_SIZE - sizeof(tFilePage))/ pQuery->rowSize;
   
-  int32_t requiredPages = pSupporter->numOfPages;
-  if (requiredPages * DEFAULT_INTERN_BUF_SIZE < dstSize) {
-    while (requiredPages * DEFAULT_INTERN_BUF_SIZE < dstSize) {
-      requiredPages += pSupporter->numOfMeters;
-    }
+  // the base value for group result, since the maximum number of table for each vnode will not exceed 100,000.
+  int32_t base = 200000;
+  int32_t pageId = -1;
   
-    if (extendDiskBuf(pQuery, pSupporter, requiredPages) != TSDB_CODE_SUCCESS) {
-      return -1;
-    }
+  int32_t remain = pQuery->sdata[0]->len;
+  int32_t offset = 0;
+  
+  while(remain > 0) {
+    int32_t r = remain;
+    if (r > capacity) {
+      r = capacity;
     }
   
-  char *lastPosition = pSupporter->meterOutputMMapBuf + DEFAULT_INTERN_BUF_SIZE * numOfMeterResultBufPages +
-                       pSupporter->groupResultSize * pSupporter->numOfGroupResultPages;
+    tFilePage* buf = getNewDataBuf(pResultBuf, base + pSupporter->subgroupIdx*10000 + pSupporter->numOfGroupResultPages, &pageId);
     
+    //pagewise copy to dest buffer
     for (int32_t i = 0; i < pQuery->numOfOutputCols; ++i) {
-    int32_t size = pRuntimeEnv->pCtx[i].outputBytes * pQuery->sdata[0]->len + sizeof(tFilePage);
-    memcpy(lastPosition, pQuery->sdata[i], size);
+      int32_t bytes = pRuntimeEnv->pCtx[i].outputBytes;
+      buf->numOfElems = r;
+      
+      memcpy(buf->data + pRuntimeEnv->offset[i] * buf->numOfElems, ((char*)pQuery->sdata[i]->data) + offset * bytes,
+          buf->numOfElems * bytes);
+    }
     
-    lastPosition += pRuntimeEnv->pCtx[i].outputBytes * pQuery->pointsToRead + sizeof(tFilePage);
+    offset += r;
+    remain -= r;
   }
   
   pSupporter->numOfGroupResultPages += 1;
@@ -5691,49 +5702,62 @@ void enableFunctForMasterScan(SQueryRuntimeEnv *pRuntimeEnv, int32_t order) {
   pQuery->order.order = (pQuery->order.order ^ 1);
 }
 
-void createGroupResultBuf(SQuery *pQuery, SOutputRes *pOneResult, bool isMetricQuery) {
-  int32_t numOfOutput = pQuery->numOfOutputCols;
+void createQueryResultBuf(SQueryRuntimeEnv *pRuntimeEnv, SOutputRes *pResultRow, bool isSTableQuery, SPosInfo *posInfo) {
+  SQuery* pQuery = pRuntimeEnv->pQuery;
   
-  pOneResult->resultInfo = calloc((size_t)numOfOutput, sizeof(SResultInfo));
-
-  pOneResult->result = malloc(POINTER_BYTES * numOfOutput);
-  for (int32_t i = 0; i < numOfOutput; ++i) {
-    size_t       size = pQuery->pSelectExpr[i].interResBytes;
-    SResultInfo *pResInfo = &pOneResult->resultInfo[i];
+  int32_t numOfCols = pQuery->numOfOutputCols;
 
-    pOneResult->result[i] = malloc(sizeof(tFilePage) + size * pOneResult->nAlloc);
-    pOneResult->result[i]->numOfElems = 0;
+  pResultRow->resultInfo = calloc((size_t)numOfCols, sizeof(SResultInfo));
+  pResultRow->pos = *posInfo;//page->data + (pRuntimeEnv->offset[i] * pRuntimeEnv->numOfRowsPerPage) + page->numOfElems*s1;
   
-    setResultInfoBuf(pResInfo, (int32_t)size, isMetricQuery);
+  for (int32_t i = 0; i < numOfCols; ++i) {
+    SResultInfo *pResultInfo = &pResultRow->resultInfo[i];
+    size_t size = pQuery->pSelectExpr[i].interResBytes;
+    setResultInfoBuf(pResultInfo, (int32_t)size, isSTableQuery);
   }
 }
 
-void clearGroupResultBuf(SOutputRes *pOneOutputRes, int32_t nOutputCols) {
+void clearGroupResultBuf(SQueryRuntimeEnv *pRuntimeEnv, SOutputRes *pOneOutputRes) {
   if (pOneOutputRes == NULL) {
     return;
   }
 
-  for (int32_t i = 0; i < nOutputCols; ++i) {
-    SResultInfo *pResInfo = &pOneOutputRes->resultInfo[i];
-    int32_t      size = sizeof(tFilePage) + pResInfo->bufLen * pOneOutputRes->nAlloc;
+  for (int32_t i = 0; i < pRuntimeEnv->pQuery->numOfOutputCols; ++i) {
+    SResultInfo *pResultInfo = &pOneOutputRes->resultInfo[i];
+//    int32_t      size = sizeof(tFilePage) + pResultInfo->bufLen * pOneOutputRes->nAlloc;
 
-    memset(pOneOutputRes->result[i], 0, (size_t)size);
-    resetResultInfo(pResInfo);
+//    memset(pOneOutputRes->pos[i], 0, (size_t)size);
+    char* s = getPosInResultPage(pRuntimeEnv, i, pOneOutputRes);
+    size_t size = pRuntimeEnv->pQuery->pSelectExpr[i].resBytes;
+    memset(s, 0, size);
+    
+    resetResultInfo(pResultInfo);
   }
 }
 
-void copyGroupResultBuf(SOutputRes* dst, const SOutputRes* src, int32_t nOutputCols) {
+void copyGroupResultBuf(SQueryRuntimeEnv *pRuntimeEnv, SOutputRes* dst, const SOutputRes* src) {
+  dst->numOfRows = src->numOfRows;
+  dst->nAlloc = src->nAlloc;
+  
+  int32_t nOutputCols = pRuntimeEnv->pQuery->numOfOutputCols;
+  
   for(int32_t i = 0; i < nOutputCols; ++i) {
     SResultInfo *pDst = &dst->resultInfo[i];
     SResultInfo *pSrc = &src->resultInfo[i];
     
     char* buf = pDst->interResultBuf;
     memcpy(pDst, pSrc, sizeof(SResultInfo));
-    pDst->interResultBuf = buf;
+    pDst->interResultBuf = buf;  // restore the allocated buffer
+    
+    // copy the result info struct
     memcpy(pDst->interResultBuf, pSrc->interResultBuf, pDst->bufLen);
 
-    int32_t size = sizeof(tFilePage) + pSrc->bufLen * src->nAlloc;
-    memcpy(dst->result[i], src->result[i], size);
+    // copy the output buffer data from src to dst, the position info keep unchanged
+    char* dstBuf = getPosInResultPage(pRuntimeEnv, i, dst);
+    char* srcBuf = getPosInResultPage(pRuntimeEnv, i, src);
+    size_t s = pRuntimeEnv->pQuery->pSelectExpr[i].resBytes;
+    
+    memcpy(dstBuf, srcBuf, s);
   }
 }
 
@@ -5743,12 +5767,12 @@ void destroyGroupResultBuf(SOutputRes *pOneOutputRes, int32_t nOutputCols) {
   }
 
   for (int32_t i = 0; i < nOutputCols; ++i) {
-    free(pOneOutputRes->result[i]);
+//    free(pOneOutputRes->pos[i]);
     free(pOneOutputRes->resultInfo[i].interResultBuf);
   }
 
   free(pOneOutputRes->resultInfo);
-  free(pOneOutputRes->result);
+//  free(pOneOutputRes->result);
 }
 
 void resetCtxOutputBuf(SQueryRuntimeEnv *pRuntimeEnv) {
@@ -5812,6 +5836,7 @@ void initCtxOutputBuf(SQueryRuntimeEnv *pRuntimeEnv) {
   for (int32_t j = 0; j < pQuery->numOfOutputCols; ++j) {
     int32_t functionId = pQuery->pSelectExpr[j].pBase.functionId;
     pRuntimeEnv->pCtx[j].currentStage = 0;
+    
     aAggs[functionId].init(&pRuntimeEnv->pCtx[j]);
   }
 }
@@ -6153,7 +6178,7 @@ void forwardIntervalQueryRange(SMeterQuerySupportObj *pSupporter, SQueryRuntimeE
         (!QUERY_IS_ASC_QUERY(pQuery) && pQuery->lastKey <= pQuery->ekey)) {
       setQueryStatus(pQuery, QUERY_COMPLETED);
     } else {
-      TSKEY nextTimestamp = loadRequiredBlockIntoMem(pRuntimeEnv, &pRuntimeEnv->nextPos);
+      /*TSKEY nextTimestamp =*/ loadRequiredBlockIntoMem(pRuntimeEnv, &pRuntimeEnv->nextPos);
     }
 
     return;
@@ -6306,7 +6331,7 @@ int32_t vnodeFilterQualifiedMeters(SQInfo *pQInfo, int32_t vid, tSidSet *pSidSet
     pOneMeterDataInfo->offsetInHeaderFile = (uint64_t)compHeader->compInfoOffset;
 
     if (pOneMeterDataInfo->pMeterQInfo == NULL) {
-      pOneMeterDataInfo->pMeterQInfo = createMeterQueryInfo(pQuery, pSupporter->rawSKey, pSupporter->rawEKey);
+      pOneMeterDataInfo->pMeterQInfo = createMeterQueryInfo(pQuery, pMeterObj->sid, pSupporter->rawSKey, pSupporter->rawEKey);
     }
 
     (*pReqMeterDataInfo)[*numOfMeters] = pOneMeterDataInfo;
@@ -6325,18 +6350,18 @@ int32_t vnodeFilterQualifiedMeters(SQInfo *pQInfo, int32_t vid, tSidSet *pSidSet
   return TSDB_CODE_SUCCESS;
 }
 
-SMeterQueryInfo *createMeterQueryInfo(SQuery *pQuery, TSKEY skey, TSKEY ekey) {
+SMeterQueryInfo *createMeterQueryInfo(SQuery *pQuery, int32_t sid, TSKEY skey, TSKEY ekey) {
   SMeterQueryInfo *pMeterQueryInfo = calloc(1, sizeof(SMeterQueryInfo));
 
   pMeterQueryInfo->skey = skey;
   pMeterQueryInfo->ekey = ekey;
   pMeterQueryInfo->lastKey = skey;
 
-  pMeterQueryInfo->numOfPages = 0;
-  pMeterQueryInfo->numOfAlloc = INIT_ALLOCATE_DISK_PAGES;
-  pMeterQueryInfo->pageList = calloc(pMeterQueryInfo->numOfAlloc, sizeof(uint32_t));
+//  pMeterQueryInfo->numOfPages = 0;
+//  pMeterQueryInfo->numOfAlloc = INIT_ALLOCATE_DISK_PAGES;
+//  pMeterQueryInfo->pageList = calloc(pMeterQueryInfo->numOfAlloc, sizeof(uint32_t));
   pMeterQueryInfo->lastResRows = 0;
-
+  pMeterQueryInfo->sid = sid;
   pMeterQueryInfo->cur.vnodeIndex = -1;
 
   pMeterQueryInfo->resultInfo = calloc((size_t)pQuery->numOfOutputCols, sizeof(SResultInfo));
@@ -6353,7 +6378,7 @@ void destroyMeterQueryInfo(SMeterQueryInfo *pMeterQueryInfo, int32_t numOfCols)
     return;
   }
 
-  free(pMeterQueryInfo->pageList);
+//  free(pMeterQueryInfo->pageList);
   for (int32_t i = 0; i < numOfCols; ++i) {
     tfree(pMeterQueryInfo->resultInfo[i].interResultBuf);
   }
@@ -6362,7 +6387,8 @@ void destroyMeterQueryInfo(SMeterQueryInfo *pMeterQueryInfo, int32_t numOfCols)
   free(pMeterQueryInfo);
 }
 
-void changeMeterQueryInfoForSuppleQuery(SMeterQueryInfo *pMeterQueryInfo, TSKEY skey, TSKEY ekey) {
+void changeMeterQueryInfoForSuppleQuery(SQueryResultBuf* pResultBuf, SMeterQueryInfo *pMeterQueryInfo, TSKEY skey,
+    TSKEY ekey) {
   if (pMeterQueryInfo == NULL) {
     return;
   }
@@ -6376,7 +6402,9 @@ void changeMeterQueryInfoForSuppleQuery(SMeterQueryInfo *pMeterQueryInfo, TSKEY
   pMeterQueryInfo->cur.vnodeIndex = -1;
 
   // previous does not generate any results
-  if (pMeterQueryInfo->numOfPages == 0) {
+  SIDList list = getDataBufPagesIdList(pResultBuf, pMeterQueryInfo->sid);
+  
+  if (list.size == 0) {
     pMeterQueryInfo->reverseFillRes = 0;
   } else {
     pMeterQueryInfo->reverseIndex = pMeterQueryInfo->numOfRes;
@@ -6384,35 +6412,6 @@ void changeMeterQueryInfoForSuppleQuery(SMeterQueryInfo *pMeterQueryInfo, TSKEY
   }
 }
 
-static tFilePage *allocNewPage(SQuery *pQuery, SMeterQuerySupportObj *pSupporter, uint32_t *pageId) {
-  if (pSupporter->lastPageId == pSupporter->numOfPages - 1) {
-    if (extendDiskBuf(pQuery, pSupporter, pSupporter->numOfPages + pSupporter->numOfMeters) != TSDB_CODE_SUCCESS) {
-      return NULL;
-    }
-  }
-
-  *pageId = (++pSupporter->lastPageId);
-  return getFilePage(pSupporter, *pageId);
-}
-
-tFilePage *addDataPageForMeterQueryInfo(SQuery *pQuery, SMeterQueryInfo *pMeterQueryInfo,
-                                        SMeterQuerySupportObj *pSupporter) {
-  uint32_t pageId = 0;
-
-  tFilePage *pPage = allocNewPage(pQuery, pSupporter, &pageId);
-  if (pPage == NULL) {  // failed to allocate disk-based buffer for intermediate results
-    return NULL;
-  }
-
-  if (pMeterQueryInfo->numOfPages >= pMeterQueryInfo->numOfAlloc) {
-    pMeterQueryInfo->numOfAlloc = pMeterQueryInfo->numOfAlloc << 1;
-    pMeterQueryInfo->pageList = realloc(pMeterQueryInfo->pageList, sizeof(uint32_t) * pMeterQueryInfo->numOfAlloc);
-  }
-
-  pMeterQueryInfo->pageList[pMeterQueryInfo->numOfPages++] = pageId;
-  return pPage;
-}
-
 void saveIntervalQueryRange(SQueryRuntimeEnv *pRuntimeEnv, SMeterQueryInfo *pMeterQueryInfo) {
   SQuery *pQuery = pRuntimeEnv->pQuery;
 
@@ -6825,12 +6824,10 @@ void setExecutionContext(SMeterQuerySupportObj *pSupporter, SOutputRes *outputRe
 static void setGroupOutputBuffer(SQueryRuntimeEnv *pRuntimeEnv, SOutputRes *pResult) {
   SQuery *pQuery = pRuntimeEnv->pQuery;
   
-  // Note: pResult->result[i]->numOfElems == 0, there is only fixed number of results for each group
+  // Note: pResult->pos[i]->numOfElems == 0, there is only fixed number of results for each group
   for (int32_t i = 0; i < pQuery->numOfOutputCols; ++i) {
-    assert(pResult->result[i]->numOfElems == 0 || pResult->result[i]->numOfElems == 1);
-
     SQLFunctionCtx *pCtx = &pRuntimeEnv->pCtx[i];
-    pCtx->aOutputBuf = pResult->result[i]->data + pCtx->outputBytes * pResult->result[i]->numOfElems;
+    pCtx->aOutputBuf = getPosInResultPage(pRuntimeEnv, i, pResult);
     
     int32_t functionId = pQuery->pSelectExpr[i].pBase.functionId;
     if (functionId == TSDB_FUNC_TOP || functionId == TSDB_FUNC_BOTTOM || functionId == TSDB_FUNC_DIFF) {
@@ -6867,12 +6864,16 @@ void setCtxOutputPointerForSupplementScan(SMeterQuerySupportObj *pSupporter, SMe
   tFilePage *pData = NULL;
   int32_t    i = 0;
 
+  SQueryResultBuf* pResultBuf = pRuntimeEnv->pResultBuf;
+  
   // find the position for this output result
-  for (; i < pMeterQueryInfo->numOfPages; ++i) {
-    pData = getFilePage(pSupporter, pMeterQueryInfo->pageList[i]);
+  SIDList list = getDataBufPagesIdList(pResultBuf, pMeterQueryInfo->sid);
+  for (; i < list.size; ++i) {
+    pData = getResultBufferPageById(pResultBuf, list.pData[i]);
     if (index <= pData->numOfElems) {
       break;
     }
+    
     index -= pData->numOfElems;
   }
 
@@ -6934,17 +6935,18 @@ int32_t setOutputBufferForIntervalQuery(SMeterQuerySupportObj *pSupporter, SMete
   SQueryRuntimeEnv *pRuntimeEnv = &pSupporter->runtimeEnv;
   tFilePage *       pData = NULL;
 
-  SQuery *pQuery = pRuntimeEnv->pQuery;
+  SQueryResultBuf* pResultBuf = pRuntimeEnv->pResultBuf;
 
   // in the first scan, new space needed for results
-  if (pMeterQueryInfo->numOfPages == 0) {
-    pData = addDataPageForMeterQueryInfo(pQuery, pMeterQueryInfo, pSupporter);
+  SIDList list = getDataBufPagesIdList(pResultBuf, pMeterQueryInfo->sid);
+  int32_t pageId = -1;
+  if (list.size == 0) {
+    pData = getNewDataBuf(pResultBuf, pMeterQueryInfo->sid, &pageId);
   } else {
-    int32_t lastPageId = pMeterQueryInfo->pageList[pMeterQueryInfo->numOfPages - 1];
-    pData = getFilePage(pSupporter, lastPageId);
+    pData = getResultBufferPageById(pResultBuf, getLastPageId(&list));
 
     if (pData->numOfElems >= pRuntimeEnv->numOfRowsPerPage) {
-      pData = addDataPageForMeterQueryInfo(pRuntimeEnv->pQuery, pMeterQueryInfo, pSupporter);
+      pData = getNewDataBuf(pResultBuf, pMeterQueryInfo->sid, &pageId);
       if (pData != NULL) {
         assert(pData->numOfElems == 0);  // number of elements must be 0 for new allocated buffer
       }
@@ -7012,171 +7014,171 @@ int32_t setIntervalQueryExecutionContext(SMeterQuerySupportObj *pSupporter, int3
   return 0;
 }
 
-static void doApplyIntervalQueryOnBlock(SMeterQuerySupportObj *pSupporter, SMeterQueryInfo *pMeterQueryInfo,
-                                        SBlockInfo *pBlockInfo, int64_t *pPrimaryCol, SField *pFields,
-                                        __block_search_fn_t searchFn) {
-  SQueryRuntimeEnv *pRuntimeEnv = &pSupporter->runtimeEnv;
-  SQuery *          pQuery = pRuntimeEnv->pQuery;
-  int32_t           factor = GET_FORWARD_DIRECTION_FACTOR(pQuery->order.order);
-
-  int64_t nextKey = -1;
-  bool    queryCompleted = false;
-
-  while (1) {
-    int32_t numOfRes = 0;
-    int32_t steps = applyFunctionsOnBlock(pRuntimeEnv, pBlockInfo, pPrimaryCol, pFields, searchFn, &numOfRes);
-    assert(steps > 0);
-
-    // NOTE: in case of stable query, only ONE(or ZERO) row of result generated for each query range
-    if (pMeterQueryInfo->lastResRows == 0) {
-      pMeterQueryInfo->lastResRows = numOfRes;
-    } else {
-      assert(pMeterQueryInfo->lastResRows == 1);
-    }
-
-    int32_t pos = pQuery->pos + steps * factor;
-
-    // query does not reach the end of current block
-    if ((pos < pBlockInfo->size && QUERY_IS_ASC_QUERY(pQuery)) || (pos >= 0 && !QUERY_IS_ASC_QUERY(pQuery))) {
-      nextKey = pPrimaryCol[pos];
-    } else {
-      assert((pQuery->lastKey > pBlockInfo->keyLast && QUERY_IS_ASC_QUERY(pQuery)) ||
-             (pQuery->lastKey < pBlockInfo->keyFirst && !QUERY_IS_ASC_QUERY(pQuery)));
-    }
-
-    // all data satisfy current query are checked, query completed
-    if (QUERY_IS_ASC_QUERY(pQuery)) {
-      queryCompleted = (nextKey > pQuery->ekey || pQuery->ekey <= pBlockInfo->keyLast);
-    } else {
-      queryCompleted = (nextKey < pQuery->ekey || pQuery->ekey >= pBlockInfo->keyFirst);
-    }
-
-    /*
-     * 1. there may be more date that satisfy current query interval, other than
-     *    current block, we need to try next data blocks
-     * 2. query completed, since reaches the upper bound of the main query range
-     */
-    if (QUERY_IS_ASC_QUERY(pQuery)) {
-      if (pQuery->lastKey > pBlockInfo->keyLast || pQuery->lastKey > pSupporter->rawEKey ||
-          nextKey > pSupporter->rawEKey) {
-        /*
-         * current interval query is completed, set query result flag closed and
-         * try next data block if pQuery->ekey == pSupporter->rawEKey, whole query is completed
-         */
-        if (pQuery->lastKey > pBlockInfo->keyLast) {
-          assert(pQuery->ekey >= pBlockInfo->keyLast);
-        }
-
-        if (pQuery->lastKey > pSupporter->rawEKey || nextKey > pSupporter->rawEKey) {
-          /* whole query completed, save result and abort */
-          assert(queryCompleted);
-          saveResult(pSupporter, pMeterQueryInfo, pMeterQueryInfo->lastResRows);
-
-          // save the pQuery->lastKey for retrieve data in cache, actually, there will be no qualified data in cache.
-          saveIntervalQueryRange(pRuntimeEnv, pMeterQueryInfo);
-        } else if (pQuery->ekey == pBlockInfo->keyLast) {
-          /* current interval query is completed, set the next query range on other data blocks if exist */
-          int64_t prevEKey = pQuery->ekey;
-
-          getAlignedIntervalQueryRange(pRuntimeEnv, pQuery->lastKey, pSupporter->rawSKey, pSupporter->rawEKey);
-          saveIntervalQueryRange(pRuntimeEnv, pMeterQueryInfo);
-
-          assert(queryCompleted && prevEKey < pQuery->skey);
-          if (pMeterQueryInfo->lastResRows > 0) {
-            saveResult(pSupporter, pMeterQueryInfo, pMeterQueryInfo->lastResRows);
-          }
-        } else {
-          /*
-           * Data that satisfy current query range may locate in current block and blocks that are directly right
-           * next to current block. Therefore, we need to keep the query range(interval) unchanged until reaching
-           * the direct next data block, while only forwards the pQuery->lastKey.
-           *
-           * With the information of the directly next data block, whether locates in cache or disk,
-           * current interval query being completed or not can be decided.
-           */
-          saveIntervalQueryRange(pRuntimeEnv, pMeterQueryInfo);
-          assert(pQuery->lastKey > pBlockInfo->keyLast && pQuery->lastKey <= pQuery->ekey);
-
-          /*
-           * if current block is the last block of current file, we still close the result flag, and
-           * merge with other meters in the same group
-           */
-          if (queryCompleted) {
-            saveResult(pSupporter, pMeterQueryInfo, pMeterQueryInfo->lastResRows);
-          }
-        }
-
-        break;
-      }
-    } else {
-      if (pQuery->lastKey < pBlockInfo->keyFirst || pQuery->lastKey < pSupporter->rawEKey ||
-          nextKey < pSupporter->rawEKey) {
-        if (pQuery->lastKey < pBlockInfo->keyFirst) {
-          assert(pQuery->ekey <= pBlockInfo->keyFirst);
-        }
-
-        if (pQuery->lastKey < pSupporter->rawEKey || (nextKey < pSupporter->rawEKey && nextKey != -1)) {
-          /* whole query completed, save result and abort */
-          assert(queryCompleted);
-          saveResult(pSupporter, pMeterQueryInfo, pMeterQueryInfo->lastResRows);
-
-          /*
-           * save the pQuery->lastKey for retrieve data in cache, actually,
-           * there will be no qualified data in cache.
-           */
-          saveIntervalQueryRange(pRuntimeEnv, pMeterQueryInfo);
-        } else if (pQuery->ekey == pBlockInfo->keyFirst) {
-          // current interval query is completed, set the next query range on other data blocks if exist
-          int64_t prevEKey = pQuery->ekey;
-
-          getAlignedIntervalQueryRange(pRuntimeEnv, pQuery->lastKey, pSupporter->rawSKey, pSupporter->rawEKey);
-          saveIntervalQueryRange(pRuntimeEnv, pMeterQueryInfo);
-
-          assert(queryCompleted && prevEKey > pQuery->skey);
-          if (pMeterQueryInfo->lastResRows > 0) {
-            saveResult(pSupporter, pMeterQueryInfo, pMeterQueryInfo->lastResRows);
-          }
-        } else {
-          /*
-           * Data that satisfy current query range may locate in current block and blocks that are
-           * directly right next to current block. Therefore, we need to keep the query range(interval)
-           * unchanged until reaching the direct next data block, while only forwards the pQuery->lastKey.
-           *
-           * With the information of the directly next data block, whether locates in cache or disk,
-           * current interval query being completed or not can be decided.
-           */
-          saveIntervalQueryRange(pRuntimeEnv, pMeterQueryInfo);
-          assert(pQuery->lastKey < pBlockInfo->keyFirst && pQuery->lastKey >= pQuery->ekey);
-
-          /*
-           * if current block is the last block of current file, we still close the result
-           * flag, and merge with other meters in the same group
-           */
-          if (queryCompleted) {
-            saveResult(pSupporter, pMeterQueryInfo, pMeterQueryInfo->lastResRows);
-          }
-        }
-
-        break;
-      }
-    }
-
-    assert(queryCompleted);
-    saveResult(pSupporter, pMeterQueryInfo, pMeterQueryInfo->lastResRows);
-
-    assert((nextKey >= pQuery->lastKey && QUERY_IS_ASC_QUERY(pQuery)) ||
-           (nextKey <= pQuery->lastKey && !QUERY_IS_ASC_QUERY(pQuery)));
-
-    /* still in the same block to query */
-    getAlignedIntervalQueryRange(pRuntimeEnv, nextKey, pSupporter->rawSKey, pSupporter->rawEKey);
-    saveIntervalQueryRange(pRuntimeEnv, pMeterQueryInfo);
-
-    int32_t newPos = searchFn((char *)pPrimaryCol, pBlockInfo->size, pQuery->skey, pQuery->order.order);
-    assert(newPos == pQuery->pos + steps * factor);
-
-    pQuery->pos = newPos;
-  }
-}
+//static void doApplyIntervalQueryOnBlock(SMeterQuerySupportObj *pSupporter, SMeterQueryInfo *pMeterQueryInfo,
+//                                        SBlockInfo *pBlockInfo, int64_t *pPrimaryCol, SField *pFields,
+//                                        __block_search_fn_t searchFn) {
+//  SQueryRuntimeEnv *pRuntimeEnv = &pSupporter->runtimeEnv;
+//  SQuery *          pQuery = pRuntimeEnv->pQuery;
+//  int32_t           factor = GET_FORWARD_DIRECTION_FACTOR(pQuery->order.order);
+//
+//  int64_t nextKey = -1;
+//  bool    queryCompleted = false;
+//
+//  while (1) {
+//    int32_t numOfRes = 0;
+//    int32_t steps = applyFunctionsOnBlock(pRuntimeEnv, pBlockInfo, pPrimaryCol, pFields, searchFn, &numOfRes);
+//    assert(steps > 0);
+//
+//    // NOTE: in case of stable query, only ONE(or ZERO) row of pos generated for each query range
+//    if (pMeterQueryInfo->lastResRows == 0) {
+//      pMeterQueryInfo->lastResRows = numOfRes;
+//    } else {
+//      assert(pMeterQueryInfo->lastResRows == 1);
+//    }
+//
+//    int32_t pos = pQuery->pos + steps * factor;
+//
+//    // query does not reach the end of current block
+//    if ((pos < pBlockInfo->size && QUERY_IS_ASC_QUERY(pQuery)) || (pos >= 0 && !QUERY_IS_ASC_QUERY(pQuery))) {
+//      nextKey = pPrimaryCol[pos];
+//    } else {
+//      assert((pQuery->lastKey > pBlockInfo->keyLast && QUERY_IS_ASC_QUERY(pQuery)) ||
+//             (pQuery->lastKey < pBlockInfo->keyFirst && !QUERY_IS_ASC_QUERY(pQuery)));
+//    }
+//
+//    // all data satisfy current query are checked, query completed
+//    if (QUERY_IS_ASC_QUERY(pQuery)) {
+//      queryCompleted = (nextKey > pQuery->ekey || pQuery->ekey <= pBlockInfo->keyLast);
+//    } else {
+//      queryCompleted = (nextKey < pQuery->ekey || pQuery->ekey >= pBlockInfo->keyFirst);
+//    }
+//
+//    /*
+//     * 1. there may be more date that satisfy current query interval, other than
+//     *    current block, we need to try next data blocks
+//     * 2. query completed, since reaches the upper bound of the main query range
+//     */
+//    if (QUERY_IS_ASC_QUERY(pQuery)) {
+//      if (pQuery->lastKey > pBlockInfo->keyLast || pQuery->lastKey > pSupporter->rawEKey ||
+//          nextKey > pSupporter->rawEKey) {
+//        /*
+//         * current interval query is completed, set query pos flag closed and
+//         * try next data block if pQuery->ekey == pSupporter->rawEKey, whole query is completed
+//         */
+//        if (pQuery->lastKey > pBlockInfo->keyLast) {
+//          assert(pQuery->ekey >= pBlockInfo->keyLast);
+//        }
+//
+//        if (pQuery->lastKey > pSupporter->rawEKey || nextKey > pSupporter->rawEKey) {
+//          /* whole query completed, save pos and abort */
+//          assert(queryCompleted);
+//          saveResult(pSupporter, pMeterQueryInfo, pMeterQueryInfo->lastResRows);
+//
+//          // save the pQuery->lastKey for retrieve data in cache, actually, there will be no qualified data in cache.
+//          saveIntervalQueryRange(pRuntimeEnv, pMeterQueryInfo);
+//        } else if (pQuery->ekey == pBlockInfo->keyLast) {
+//          /* current interval query is completed, set the next query range on other data blocks if exist */
+//          int64_t prevEKey = pQuery->ekey;
+//
+//          getAlignedIntervalQueryRange(pRuntimeEnv, pQuery->lastKey, pSupporter->rawSKey, pSupporter->rawEKey);
+//          saveIntervalQueryRange(pRuntimeEnv, pMeterQueryInfo);
+//
+//          assert(queryCompleted && prevEKey < pQuery->skey);
+//          if (pMeterQueryInfo->lastResRows > 0) {
+//            saveResult(pSupporter, pMeterQueryInfo, pMeterQueryInfo->lastResRows);
+//          }
+//        } else {
+//          /*
+//           * Data that satisfy current query range may locate in current block and blocks that are directly right
+//           * next to current block. Therefore, we need to keep the query range(interval) unchanged until reaching
+//           * the direct next data block, while only forwards the pQuery->lastKey.
+//           *
+//           * With the information of the directly next data block, whether locates in cache or disk,
+//           * current interval query being completed or not can be decided.
+//           */
+//          saveIntervalQueryRange(pRuntimeEnv, pMeterQueryInfo);
+//          assert(pQuery->lastKey > pBlockInfo->keyLast && pQuery->lastKey <= pQuery->ekey);
+//
+//          /*
+//           * if current block is the last block of current file, we still close the pos flag, and
+//           * merge with other meters in the same group
+//           */
+//          if (queryCompleted) {
+//            saveResult(pSupporter, pMeterQueryInfo, pMeterQueryInfo->lastResRows);
+//          }
+//        }
+//
+//        break;
+//      }
+//    } else {
+//      if (pQuery->lastKey < pBlockInfo->keyFirst || pQuery->lastKey < pSupporter->rawEKey ||
+//          nextKey < pSupporter->rawEKey) {
+//        if (pQuery->lastKey < pBlockInfo->keyFirst) {
+//          assert(pQuery->ekey <= pBlockInfo->keyFirst);
+//        }
+//
+//        if (pQuery->lastKey < pSupporter->rawEKey || (nextKey < pSupporter->rawEKey && nextKey != -1)) {
+//          /* whole query completed, save pos and abort */
+//          assert(queryCompleted);
+//          saveResult(pSupporter, pMeterQueryInfo, pMeterQueryInfo->lastResRows);
+//
+//          /*
+//           * save the pQuery->lastKey for retrieve data in cache, actually,
+//           * there will be no qualified data in cache.
+//           */
+//          saveIntervalQueryRange(pRuntimeEnv, pMeterQueryInfo);
+//        } else if (pQuery->ekey == pBlockInfo->keyFirst) {
+//          // current interval query is completed, set the next query range on other data blocks if exist
+//          int64_t prevEKey = pQuery->ekey;
+//
+//          getAlignedIntervalQueryRange(pRuntimeEnv, pQuery->lastKey, pSupporter->rawSKey, pSupporter->rawEKey);
+//          saveIntervalQueryRange(pRuntimeEnv, pMeterQueryInfo);
+//
+//          assert(queryCompleted && prevEKey > pQuery->skey);
+//          if (pMeterQueryInfo->lastResRows > 0) {
+//            saveResult(pSupporter, pMeterQueryInfo, pMeterQueryInfo->lastResRows);
+//          }
+//        } else {
+//          /*
+//           * Data that satisfy current query range may locate in current block and blocks that are
+//           * directly right next to current block. Therefore, we need to keep the query range(interval)
+//           * unchanged until reaching the direct next data block, while only forwards the pQuery->lastKey.
+//           *
+//           * With the information of the directly next data block, whether locates in cache or disk,
+//           * current interval query being completed or not can be decided.
+//           */
+//          saveIntervalQueryRange(pRuntimeEnv, pMeterQueryInfo);
+//          assert(pQuery->lastKey < pBlockInfo->keyFirst && pQuery->lastKey >= pQuery->ekey);
+//
+//          /*
+//           * if current block is the last block of current file, we still close the pos
+//           * flag, and merge with other meters in the same group
+//           */
+//          if (queryCompleted) {
+//            saveResult(pSupporter, pMeterQueryInfo, pMeterQueryInfo->lastResRows);
+//          }
+//        }
+//
+//        break;
+//      }
+//    }
+//
+//    assert(queryCompleted);
+//    saveResult(pSupporter, pMeterQueryInfo, pMeterQueryInfo->lastResRows);
+//
+//    assert((nextKey >= pQuery->lastKey && QUERY_IS_ASC_QUERY(pQuery)) ||
+//           (nextKey <= pQuery->lastKey && !QUERY_IS_ASC_QUERY(pQuery)));
+//
+//    /* still in the same block to query */
+//    getAlignedIntervalQueryRange(pRuntimeEnv, nextKey, pSupporter->rawSKey, pSupporter->rawEKey);
+//    saveIntervalQueryRange(pRuntimeEnv, pMeterQueryInfo);
+//
+//    int32_t newPos = searchFn((char *)pPrimaryCol, pBlockInfo->size, pQuery->skey, pQuery->order.order);
+//    assert(newPos == pQuery->pos + steps * factor);
+//
+//    pQuery->pos = newPos;
+//  }
+//}
 
 static void doApplyIntervalQueryOnBlock_rv(SMeterQuerySupportObj *pSupporter, SMeterQueryInfo *pMeterQueryInfo,
                                         SBlockInfo *pBlockInfo, int64_t *pPrimaryCol, SField *pFields,
@@ -7244,7 +7246,10 @@ static void doApplyIntervalQueryOnBlock_rv(SMeterQuerySupportObj *pSupporter, SM
   
     assert(completed);
     
-    if (pQuery->ekey == pSupporter->rawEKey) {
+    // while the interval time window is less than the time range gap between two points, nextKey may be greater than
+    // pSupporter->rawEKey
+    if (pQuery->ekey == pSupporter->rawEKey || (nextKey > pSupporter->rawEKey && QUERY_IS_ASC_QUERY(pQuery)) ||
+        (nextKey < pSupporter->rawEKey && !QUERY_IS_ASC_QUERY(pQuery))) {
       /* whole query completed, save result and abort */
       saveResult(pSupporter, pMeterQueryInfo, pMeterQueryInfo->lastResRows);
     
@@ -7493,8 +7498,12 @@ bool onDemandLoadDatablock(SQuery *pQuery, int16_t queryRangeSet) {
 static void validateResultBuf(SMeterQuerySupportObj *pSupporter, SMeterQueryInfo *pMeterQueryInfo) {
   SQueryRuntimeEnv *pRuntimeEnv = &pSupporter->runtimeEnv;
   SQuery *          pQuery = pSupporter->runtimeEnv.pQuery;
+  SQueryResultBuf* pResultBuf = pRuntimeEnv->pResultBuf;
 
-  tFilePage *newOutput = getFilePage(pSupporter, pMeterQueryInfo->pageList[pMeterQueryInfo->numOfPages - 1]);
+  SIDList list = getDataBufPagesIdList(pResultBuf, pMeterQueryInfo->sid);
+  int32_t id = getLastPageId(&list);
+  
+  tFilePage* newOutput = getResultBufferPageById(pResultBuf, id);
   for (int32_t i = 0; i < pQuery->numOfOutputCols; ++i) {
     assert(pRuntimeEnv->pCtx[i].aOutputBuf - newOutput->data < DEFAULT_INTERN_BUF_SIZE);
   }
@@ -7547,8 +7556,10 @@ int32_t saveResult(SMeterQuerySupportObj *pSupporter, SMeterQueryInfo *pMeterQue
     pMeterQueryInfo->reverseIndex -= 1;
     setCtxOutputPointerForSupplementScan(pSupporter, pMeterQueryInfo);
   } else {
-    int32_t    pageId = pMeterQueryInfo->pageList[pMeterQueryInfo->numOfPages - 1];
-    tFilePage *pData = getFilePage(pSupporter, pageId);
+    SIDList list = getDataBufPagesIdList(pRuntimeEnv->pResultBuf, pMeterQueryInfo->sid);
+    
+    int32_t pageId = getLastPageId(&list);
+    tFilePage* pData = getResultBufferPageById(pRuntimeEnv->pResultBuf, pageId);
     
     // in handling records occuring around '1970-01-01', the aligned start timestamp may be 0.
     TSKEY ts = *(TSKEY *)getOutputResPos(pRuntimeEnv, pData, pData->numOfElems, 0);
@@ -7580,7 +7591,7 @@ int32_t saveResult(SMeterQuerySupportObj *pSupporter, SMeterQueryInfo *pMeterQue
     sc[1].bytes = 8;
 
     UNUSED(sc);
-    tColModel *cm = tColModelCreate(sc, pQuery->numOfOutputCols, pRuntimeEnv->numOfRowsPerPage);
+    SColumnModel *cm = createColumnModel(sc, pQuery->numOfOutputCols, pRuntimeEnv->numOfRowsPerPage);
 
 //    if (outputPage->numOfElems + numOfResult >= pRuntimeEnv->numOfRowsPerPage)
         tColModelDisplay(cm, outputPage->data, outputPage->numOfElems, pRuntimeEnv->numOfRowsPerPage);
@@ -7590,7 +7601,7 @@ int32_t saveResult(SMeterQuerySupportObj *pSupporter, SMeterQueryInfo *pMeterQue
   return TSDB_CODE_SUCCESS;
 }
 
-static int32_t getSubsetNumber(SMeterQuerySupportObj *pSupporter) {
+static int32_t getNumOfSubset(SMeterQuerySupportObj *pSupporter) {
   SQuery *pQuery = pSupporter->runtimeEnv.pQuery;
 
   int32_t totalSubset = 0;
@@ -7613,7 +7624,7 @@ static int32_t doCopyFromGroupBuf(SMeterQuerySupportObj *pSupporter, SOutputRes
 
   dTrace("QInfo:%p start to copy data to dest buf", GET_QINFO_ADDR(pSupporter->runtimeEnv.pQuery));
 
-  int32_t totalSubset = getSubsetNumber(pSupporter);
+  int32_t totalSubset = getNumOfSubset(pSupporter);
 
   if (orderType == TSQL_SO_ASC) {
     startIdx = pSupporter->subgroupIdx;
@@ -7631,8 +7642,6 @@ static int32_t doCopyFromGroupBuf(SMeterQuerySupportObj *pSupporter, SOutputRes
 
     assert(result[i].numOfRows >= 0 && pSupporter->offset <= 1);
 
-    tFilePage **srcBuf = result[i].result;
-
     int32_t numOfRowsToCopy = result[i].numOfRows - pSupporter->offset;
     int32_t oldOffset = pSupporter->offset;
 
@@ -7648,8 +7657,8 @@ static int32_t doCopyFromGroupBuf(SMeterQuerySupportObj *pSupporter, SOutputRes
     for (int32_t j = 0; j < pQuery->numOfOutputCols; ++j) {
       int32_t elemSize = pRuntimeEnv->pCtx[j].outputBytes;
       char *  outputBuf = pQuery->sdata[j]->data + numOfResult * elemSize;
-
-      memcpy(outputBuf, srcBuf[j]->data + oldOffset * elemSize, elemSize * numOfRowsToCopy);
+      char* p = getPosInResultPage(pRuntimeEnv, j, &result[i]);
+      memcpy(outputBuf, p + oldOffset * elemSize, elemSize * numOfRowsToCopy);
     }
 
     numOfResult += numOfRowsToCopy;
@@ -7717,7 +7726,7 @@ static void applyIntervalQueryOnBlock(SMeterQuerySupportObj *pSupporter, SMeterD
 
     saveIntervalQueryRange(pRuntimeEnv, pMeterQueryInfo);
   } else {
-    doApplyIntervalQueryOnBlock(pSupporter, pMeterQueryInfo, pBlockInfo, pPrimaryKey, pFields, searchFn);
+    doApplyIntervalQueryOnBlock_rv(pSupporter, pMeterQueryInfo, pBlockInfo, pPrimaryKey, pFields, searchFn);
   }
 }
 
@@ -7802,7 +7811,7 @@ static int32_t resultInterpolate(SQInfo *pQInfo, tFilePage **data, tFilePage **p
     pSchema[i].type = pQuery->pSelectExpr[i].resType;
   }
 
-  tColModel *pModel = tColModelCreate(pSchema, pQuery->numOfOutputCols, pQuery->pointsToRead);
+  SColumnModel *pModel = createColumnModel(pSchema, pQuery->numOfOutputCols, pQuery->pointsToRead);
 
   char *  srcData[TSDB_MAX_COLUMNS] = {0};
   int32_t functions[TSDB_MAX_COLUMNS] = {0};
@@ -7816,7 +7825,7 @@ static int32_t resultInterpolate(SQInfo *pQInfo, tFilePage **data, tFilePage **p
                                          pQuery->nAggTimeInterval, (int64_t *)pDataSrc[0]->data, pModel, srcData,
                                          pQuery->defaultVal, functions, pRuntimeEnv->pMeterObj->pointsPerFileBlock);
 
-  tColModelDestroy(pModel);
+  destroyColumnModel(pModel);
   free(pSchema);
 
   return numOfRes;
@@ -7906,18 +7915,10 @@ int32_t vnodeQueryResultInterpolate(SQInfo *pQInfo, tFilePage **pDst, tFilePage
       ret -= pQuery->limit.offset;
       // todo !!!!there exactly number of interpo is not valid.
       // todo refactor move to the beginning of buffer
-      //      if (QUERY_IS_ASC_QUERY(pQuery)) {
       for (int32_t i = 0; i < pQuery->numOfOutputCols; ++i) {
         memmove(pDst[i]->data, pDst[i]->data + pQuery->pSelectExpr[i].resBytes * pQuery->limit.offset,
                 ret * pQuery->pSelectExpr[i].resBytes);
       }
-      //      } else {
-      //        for (int32_t i = 0; i < pQuery->numOfOutputCols; ++i) {
-      //          memmove(pDst[i]->data + (pQuery->pointsToRead - ret) * pQuery->pSelectExpr[i].resBytes,
-      //                  pDst[i]->data + (pQuery->pointsToRead - ret - pQuery->limit.offset) *
-      //                  pQuery->pSelectExpr[i].resBytes, ret * pQuery->pSelectExpr[i].resBytes);
-      //        }
-      //      }
       pQuery->limit.offset = 0;
       return ret;
     } else {
@@ -7938,7 +7939,11 @@ void vnodePrintQueryStatistics(SMeterQuerySupportObj *pSupporter) {
   SQInfo *pQInfo = (SQInfo *)GET_QINFO_ADDR(pQuery);
 
   SQueryCostSummary *pSummary = &pRuntimeEnv->summary;
-  pSummary->tmpBufferInDisk = pSupporter->bufSize;
+  if (pRuntimeEnv->pResultBuf == NULL) {
+    pSummary->tmpBufferInDisk = 0;
+  } else {
+    pSummary->tmpBufferInDisk = getResBufSize(pRuntimeEnv->pResultBuf);
+  }
 
   dTrace("QInfo:%p statis: comp blocks:%d, size:%d Bytes, elapsed time:%.2f ms", pQInfo, pSummary->readCompInfo,
          pSummary->totalCompInfoSize, pSummary->loadCompInfoUs / 1000.0);

src/system/detail/src/vnodeQueryProcess.c

@@ -132,7 +132,7 @@ static void queryOnMultiDataCache(SQInfo *pQInfo, SMeterDataInfo *pMeterInfo) {
       pRuntimeEnv->pMeterObj = pMeterObj;
 
       if (pMeterInfo[k].pMeterQInfo == NULL) {
-        pMeterInfo[k].pMeterQInfo = createMeterQueryInfo(pQuery, pSupporter->rawSKey, pSupporter->rawEKey);
+        pMeterInfo[k].pMeterQInfo = createMeterQueryInfo(pQuery, pMeterObj->sid, pSupporter->rawSKey, pSupporter->rawEKey);
       }
 
       if (pMeterInfo[k].pMeterObj == NULL) {  // no data in disk for this meter, set its pointer
@@ -683,7 +683,7 @@ static void vnodeSTableSeqProcessor(SQInfo *pQInfo) {
     }
   
     resetCtxOutputBuf(pRuntimeEnv);
-    resetSlidingWindowInfo(&pRuntimeEnv->swindowResInfo, pQuery->numOfOutputCols);
+    resetSlidingWindowInfo(pRuntimeEnv, &pRuntimeEnv->swindowResInfo);
     
     while (pSupporter->meterIdx < pSupporter->numOfMeters) {
       int32_t k = pSupporter->meterIdx;
@@ -858,7 +858,9 @@ static void doOrderedScan(SQInfo *pQInfo) {
 static void setupMeterQueryInfoForSupplementQuery(SMeterQuerySupportObj *pSupporter) {
   for (int32_t i = 0; i < pSupporter->numOfMeters; ++i) {
     SMeterQueryInfo *pMeterQueryInfo = pSupporter->pMeterDataInfo[i].pMeterQInfo;
-    changeMeterQueryInfoForSuppleQuery(pMeterQueryInfo, pSupporter->rawSKey, pSupporter->rawEKey);
+    SQueryResultBuf* pResultBuf = pSupporter->runtimeEnv.pResultBuf;
+    
+    changeMeterQueryInfoForSuppleQuery(pResultBuf, pMeterQueryInfo, pSupporter->rawSKey, pSupporter->rawEKey);
   }
 }
 
@@ -1086,7 +1088,7 @@ static void vnodeSingleMeterIntervalMainLooper(SMeterQuerySupportObj *pSupporter
            (pQuery->skey >= pQuery->ekey && !QUERY_IS_ASC_QUERY(pQuery)));
 
     initCtxOutputBuf(pRuntimeEnv);
-    clearCompletedSlidingWindows(&pRuntimeEnv->swindowResInfo, pQuery->numOfOutputCols);
+    clearCompletedSlidingWindows(pRuntimeEnv);
     
     vnodeScanAllData(pRuntimeEnv);
     if (isQueryKilled(pQuery)) {

src/system/detail/src/vnodeShell.c

@@ -442,7 +442,6 @@ void vnodeExecuteRetrieveReq(SSchedMsg *pSched) {
 
   if (code == TSDB_CODE_SUCCESS) {
     size = vnodeGetResultSize((void *)(pRetrieve->qhandle), &numOfRows);
-  }
   
     // buffer size for progress information, including meter count,
     // and for each meter, including 'uid' and 'TSKEY'.
@@ -457,6 +456,7 @@ void vnodeExecuteRetrieveReq(SSchedMsg *pSched) {
       taosSendSimpleRsp(pObj->thandle, TSDB_MSG_TYPE_RETRIEVE_RSP, TSDB_CODE_SERV_OUT_OF_MEMORY);
       goto _exit;
     }
+  }
 
   pMsg = pStart;
 
@@ -485,27 +485,29 @@ void vnodeExecuteRetrieveReq(SSchedMsg *pSched) {
 
   // write the progress information of each meter to response
   // this is required by subscriptions
+  if (numOfRows > 0 && code == TSDB_CODE_SUCCESS) {
     if (pQInfo->pMeterQuerySupporter != NULL && pQInfo->pMeterQuerySupporter->pMeterSidExtInfo != NULL) {
-    *((int32_t*)pMsg) = htonl(pQInfo->pMeterQuerySupporter->numOfMeters);
+      *((int32_t *)pMsg) = htonl(pQInfo->pMeterQuerySupporter->numOfMeters);
       pMsg += sizeof(int32_t);
       for (int32_t i = 0; i < pQInfo->pMeterQuerySupporter->numOfMeters; i++) {
-      *((int64_t*)pMsg) = htobe64(pQInfo->pMeterQuerySupporter->pMeterSidExtInfo[i]->uid);
+        *((int64_t *)pMsg) = htobe64(pQInfo->pMeterQuerySupporter->pMeterSidExtInfo[i]->uid);
         pMsg += sizeof(int64_t);
-      *((TSKEY*)pMsg) = htobe64(pQInfo->pMeterQuerySupporter->pMeterSidExtInfo[i]->key);
+        *((TSKEY *)pMsg) = htobe64(pQInfo->pMeterQuerySupporter->pMeterSidExtInfo[i]->key);
         pMsg += sizeof(TSKEY);
       }
     } else if (pQInfo->pObj != NULL) {
-    *((int32_t*)pMsg) = htonl(1);
+      *((int32_t *)pMsg) = htonl(1);
       pMsg += sizeof(int32_t);
-    *((int64_t*)pMsg) = htobe64(pQInfo->pObj->uid);
+      *((int64_t *)pMsg) = htobe64(pQInfo->pObj->uid);
       pMsg += sizeof(int64_t);
       if (pQInfo->pointsRead > 0) {
-      *((TSKEY*)pMsg) = htobe64(pQInfo->query.lastKey + 1);
+        *((TSKEY *)pMsg) = htobe64(pQInfo->query.lastKey + 1);
       } else {
-      *((TSKEY*)pMsg) = htobe64(pQInfo->query.lastKey);
+        *((TSKEY *)pMsg) = htobe64(pQInfo->query.lastKey);
       }
       pMsg += sizeof(TSKEY);
     }
+  }
 
   msgLen = pMsg - pStart;
 

src/system/detail/src/vnodeTagMgmt.c

@@ -24,10 +24,10 @@
 #include "tast.h"
 #include "vnodeTagMgmt.h"
 
-#define GET_TAG_VAL_POINTER(s, col, sc, t) ((t *)(&((s)->tags[(sc)->colOffset[(col)]])))
+#define GET_TAG_VAL_POINTER(s, col, sc, t) ((t *)(&((s)->tags[getColumnModelOffset(sc, col)])))
 #define GET_TAG_VAL(s, col, sc, t) (*GET_TAG_VAL_POINTER(s, col, sc, t))
 
-static void tTagsPrints(SMeterSidExtInfo *pMeterInfo, tTagSchema *pSchema, tOrderIdx *pOrder);
+static void tTagsPrints(SMeterSidExtInfo *pMeterInfo, SColumnModel *pSchema, SColumnOrderInfo *pOrder);
 
 static void tSidSetDisplay(tSidSet *pSets);
 
@@ -65,7 +65,7 @@ int32_t meterSidComparator(const void *p1, const void *p2, void *param) {
   SMeterSidExtInfo *s1 = (SMeterSidExtInfo *)p1;
   SMeterSidExtInfo *s2 = (SMeterSidExtInfo *)p2;
 
-  for (int32_t i = 0; i < pOrderDesc->orderIdx.numOfOrderedCols; ++i) {
+  for (int32_t i = 0; i < pOrderDesc->orderIdx.numOfCols; ++i) {
     int32_t colIdx = pOrderDesc->orderIdx.pData[i];
 
     char *  f1 = NULL;
@@ -79,9 +79,9 @@ int32_t meterSidComparator(const void *p1, const void *p2, void *param) {
       type = TSDB_DATA_TYPE_BINARY;
       bytes = TSDB_METER_NAME_LEN;
     } else {
-      f1 = GET_TAG_VAL_POINTER(s1, colIdx, pOrderDesc->pTagSchema, char);
-      f2 = GET_TAG_VAL_POINTER(s2, colIdx, pOrderDesc->pTagSchema, char);
-      SSchema *pSchema = &pOrderDesc->pTagSchema->pSchema[colIdx];
+      f1 = GET_TAG_VAL_POINTER(s1, colIdx, pOrderDesc->pColumnModel, char);
+      f2 = GET_TAG_VAL_POINTER(s2, colIdx, pOrderDesc->pColumnModel, char);
+      SSchema *pSchema = getColumnModelSchema(pOrderDesc->pColumnModel, colIdx);
       type = pSchema->type;
       bytes = pSchema->bytes;
     }
@@ -116,9 +116,9 @@ static void median(void **pMeterSids, size_t size, int32_t s1, int32_t s2, tOrde
          compareFn(pMeterSids[s1], pMeterSids[s2], pOrderDesc) <= 0);
 
 #ifdef _DEBUG_VIEW
-  tTagsPrints(pMeterSids[s1], pOrderDesc->pTagSchema, &pOrderDesc->orderIdx);
-  tTagsPrints(pMeterSids[midIdx], pOrderDesc->pTagSchema, &pOrderDesc->orderIdx);
-  tTagsPrints(pMeterSids[s2], pOrderDesc->pTagSchema, &pOrderDesc->orderIdx);
+  tTagsPrints(pMeterSids[s1], pOrderDesc->pColumnModel, &pOrderDesc->orderIdx);
+  tTagsPrints(pMeterSids[midIdx], pOrderDesc->pColumnModel, &pOrderDesc->orderIdx);
+  tTagsPrints(pMeterSids[s2], pOrderDesc->pColumnModel, &pOrderDesc->orderIdx);
 #endif
 }
 
@@ -241,25 +241,6 @@ int32_t *calculateSubGroup(void **pSids, int32_t numOfMeters, int32_t *numOfSubs
   return starterPos;
 }
 
-tTagSchema *tCreateTagSchema(SSchema *pSchema, int32_t numOfTagCols) {
-  if (numOfTagCols == 0 || pSchema == NULL) {
-    return NULL;
-  }
-
-  tTagSchema *pTagSchema =
-      (tTagSchema *)calloc(1, sizeof(tTagSchema) + numOfTagCols * sizeof(int32_t) + sizeof(SSchema) * numOfTagCols);
-
-  pTagSchema->colOffset[0] = 0;
-  pTagSchema->numOfCols = numOfTagCols;
-  for (int32_t i = 1; i < numOfTagCols; ++i) {
-    pTagSchema->colOffset[i] = (pTagSchema->colOffset[i - 1] + pSchema[i - 1].bytes);
-  }
-
-  pTagSchema->pSchema = (SSchema *)&(pTagSchema->colOffset[numOfTagCols]);
-  memcpy(pTagSchema->pSchema, pSchema, sizeof(SSchema) * numOfTagCols);
-  return pTagSchema;
-}
-
 tSidSet *tSidSetCreate(struct SMeterSidExtInfo **pMeterSidExtInfo, int32_t numOfMeters, SSchema *pSchema,
                        int32_t numOfTags, SColIndexEx *colList, int32_t numOfCols) {
   tSidSet *pSidSet = (tSidSet *)calloc(1, sizeof(tSidSet) + numOfCols * sizeof(int16_t));
@@ -269,8 +250,8 @@ tSidSet *tSidSetCreate(struct SMeterSidExtInfo **pMeterSidExtInfo, int32_t numOf
 
   pSidSet->numOfSids = numOfMeters;
   pSidSet->pSids = pMeterSidExtInfo;
-  pSidSet->pTagSchema = tCreateTagSchema(pSchema, numOfTags);
-  pSidSet->orderIdx.numOfOrderedCols = numOfCols;
+  pSidSet->pColumnModel = createColumnModel(pSchema, numOfTags, 1);
+  pSidSet->orderIdx.numOfCols = numOfCols;
 
   /*
    * in case of "group by tbname,normal_col", the normal_col is ignored
@@ -282,7 +263,7 @@ tSidSet *tSidSetCreate(struct SMeterSidExtInfo **pMeterSidExtInfo, int32_t numOf
     }
   }
 
-  pSidSet->orderIdx.numOfOrderedCols = numOfTagCols;
+  pSidSet->orderIdx.numOfCols = numOfTagCols;
 
   pSidSet->starterPos = NULL;
   return pSidSet;
@@ -291,19 +272,19 @@ tSidSet *tSidSetCreate(struct SMeterSidExtInfo **pMeterSidExtInfo, int32_t numOf
 void tSidSetDestroy(tSidSet **pSets) {
   if ((*pSets) != NULL) {
     tfree((*pSets)->starterPos);
-    tfree((*pSets)->pTagSchema)(*pSets)->pSids = NULL;
+    tfree((*pSets)->pColumnModel)(*pSets)->pSids = NULL;
     tfree(*pSets);
   }
 }
 
-void tTagsPrints(SMeterSidExtInfo *pMeterInfo, tTagSchema *pSchema, tOrderIdx *pOrder) {
+void tTagsPrints(SMeterSidExtInfo *pMeterInfo, SColumnModel *pSchema, SColumnOrderInfo *pOrder) {
   if (pSchema == NULL) {
     return;
   }
 
   printf("sid: %-5d tags(", pMeterInfo->sid);
 
-  for (int32_t i = 0; i < pOrder->numOfOrderedCols; ++i) {
+  for (int32_t i = 0; i < pOrder->numOfCols; ++i) {
     int32_t colIndex = pOrder->pData[i];
 
     // it is the tbname column
@@ -312,7 +293,9 @@ void tTagsPrints(SMeterSidExtInfo *pMeterInfo, tTagSchema *pSchema, tOrderIdx *p
       continue;
     }
 
-    switch (pSchema->pSchema[colIndex].type) {
+    SSchema* s = getColumnModelSchema(pSchema, colIndex);
+    
+    switch (s->type) {
       case TSDB_DATA_TYPE_INT:
         printf("%d, ", GET_TAG_VAL(pMeterInfo, colIndex, pSchema, int32_t));
         break;
@@ -336,9 +319,9 @@ void tTagsPrints(SMeterSidExtInfo *pMeterInfo, tTagSchema *pSchema, tOrderIdx *p
         break;
       case TSDB_DATA_TYPE_NCHAR: {
         char *data = GET_TAG_VAL_POINTER(pMeterInfo, colIndex, pSchema, char);
-        char  buffer[512] = {0};
         
-        taosUcs4ToMbs(data, pSchema->pSchema[colIndex].bytes, buffer);
+        char  buffer[512] = {0};
+        taosUcs4ToMbs(data, s->bytes, buffer);
         printf("%s, ", buffer);
         break;
       }
@@ -370,16 +353,16 @@ static void UNUSED_FUNC tSidSetDisplay(tSidSet *pSets) {
 
     printf("the %d-th subgroup: \n", i + 1);
     for (int32_t j = s; j < e; ++j) {
-      tTagsPrints(pSets->pSids[j], pSets->pTagSchema, &pSets->orderIdx);
+      tTagsPrints(pSets->pSids[j], pSets->pColumnModel, &pSets->orderIdx);
     }
   }
 }
 
 void tSidSetSort(tSidSet *pSets) {
   pTrace("number of meters in sort: %d", pSets->numOfSids);
-  tOrderIdx *pOrderIdx = &pSets->orderIdx;
+  SColumnOrderInfo *pOrderIdx = &pSets->orderIdx;
 
-  if (pOrderIdx->numOfOrderedCols == 0 || pSets->numOfSids <= 1 || pSets->pTagSchema == NULL) { // no group by tags clause
+  if (pOrderIdx->numOfCols == 0 || pSets->numOfSids <= 1 || pSets->pColumnModel == NULL) { // no group by tags clause
     pSets->numOfSubSet = 1;
     pSets->starterPos = (int32_t *)malloc(sizeof(int32_t) * (pSets->numOfSubSet + 1));
     pSets->starterPos[0] = 0;
@@ -390,11 +373,11 @@ void tSidSetSort(tSidSet *pSets) {
 #endif
   } else {
     tOrderDescriptor *descriptor =
-        (tOrderDescriptor *)calloc(1, sizeof(tOrderDescriptor) + sizeof(int16_t) * pSets->orderIdx.numOfOrderedCols);
-    descriptor->pTagSchema = pSets->pTagSchema;
+        (tOrderDescriptor *)calloc(1, sizeof(tOrderDescriptor) + sizeof(int16_t) * pSets->orderIdx.numOfCols);
+    descriptor->pColumnModel = pSets->pColumnModel;
     descriptor->orderIdx = pSets->orderIdx;
 
-    memcpy(descriptor->orderIdx.pData, pOrderIdx->pData, sizeof(int16_t) * pSets->orderIdx.numOfOrderedCols);
+    memcpy(descriptor->orderIdx.pData, pOrderIdx->pData, sizeof(int16_t) * pSets->orderIdx.numOfCols);
 
     tQSortEx((void **)pSets->pSids, POINTER_BYTES, 0, pSets->numOfSids - 1, descriptor, meterSidComparator);
     pSets->starterPos =

src/system/detail/src/vnodeUtil.c

@@ -247,7 +247,7 @@ SSqlFunctionExpr* vnodeCreateSqlFunctionExpr(SQueryMeterMsg* pQueryMsg, int32_t*
 
     SColIndexEx* pColumnIndexExInfo = &pExprs[i].pBase.colInfo;
 
-    // tag column schema is kept in pQueryMsg->pTagSchema
+    // tag column schema is kept in pQueryMsg->pColumnModel
     if (TSDB_COL_IS_TAG(pColumnIndexExInfo->flag)) {
       if (pColumnIndexExInfo->colIdx >= pQueryMsg->numOfTagsCols) {
         *code = TSDB_CODE_INVALID_QUERY_MSG;

src/util/src/hash.c

@@ -346,6 +346,14 @@ static void doAddToHashTable(HashObj *pObj, SHashNode *pNode) {
 //  pTrace("key:%s %p add to hash table", key, pNode);
 }
 
+int32_t taosNumElemsInHashTable(HashObj *pObj) {
+  if (pObj == NULL) {
+    return 0;
+  }
+  
+  return pObj->size;
+}
+
 /**
  * add data node into hash table
  * @param pObj    hash object
@@ -392,7 +400,7 @@ int32_t taosAddToHashTable(HashObj *pObj, const char *key, uint32_t keyLen, void
   return 0;
 }
 
-char *taosGetDataFromHash(HashObj *pObj, const char *key, uint32_t keyLen) {
+char *taosGetDataFromHashTable(HashObj *pObj, const char *key, uint32_t keyLen) {
   if (pObj->multithreadSafe) {
     __rd_lock(&pObj->lock);
   }

src/util/src/textbuffer.c

@@ -23,7 +23,7 @@
 #include "tutil.h"
 
 #define COLMODEL_GET_VAL(data, schema, allrow, rowId, colId) \
-  (data + (schema)->colOffset[colId] * (allrow) + (rowId) * (schema)->pFields[colId].bytes)
+  (data + (schema)->pFields[colId].offset * (allrow) + (rowId) * (schema)->pFields[colId].field.bytes)
 
 int32_t tmpFileSerialNum = 0;
 
@@ -49,46 +49,50 @@ void getTmpfilePath(const char *fileNamePrefix, char *dstPath) {
 }
 
 /*
- * tColModel is deeply copy
+ * SColumnModel is deeply copy
  */
-void tExtMemBufferCreate(tExtMemBuffer **pMemBuffer, int32_t nBufferSize, int32_t elemSize, const char *tmpDataFilePath,
-                         tColModel *pModel) {
-  (*pMemBuffer) = (tExtMemBuffer *)calloc(1, sizeof(tExtMemBuffer));
+tExtMemBuffer* createExtMemBuffer(int32_t inMemSize, int32_t elemSize, SColumnModel *pModel) {
+  tExtMemBuffer* pMemBuffer = (tExtMemBuffer *)calloc(1, sizeof(tExtMemBuffer));
 
-  (*pMemBuffer)->nPageSize = DEFAULT_PAGE_SIZE;
-  (*pMemBuffer)->nMaxSizeInPages = ALIGN8(nBufferSize) / (*pMemBuffer)->nPageSize;
-  (*pMemBuffer)->nElemSize = elemSize;
+  pMemBuffer->pageSize = DEFAULT_PAGE_SIZE;
+  pMemBuffer->inMemCapacity = ALIGN8(inMemSize) / pMemBuffer->pageSize;
+  pMemBuffer->nElemSize = elemSize;
 
-  (*pMemBuffer)->numOfElemsPerPage = ((*pMemBuffer)->nPageSize - sizeof(tFilePage)) / (*pMemBuffer)->nElemSize;
+  pMemBuffer->numOfElemsPerPage = (pMemBuffer->pageSize - sizeof(tFilePage)) / pMemBuffer->nElemSize;
   
-  strcpy((*pMemBuffer)->dataFilePath, tmpDataFilePath);
+  char name[MAX_TMPFILE_PATH_LENGTH] = {0};
+  getTmpfilePath("extbuf", name);
+  
+  pMemBuffer->path = strdup(name);
+  pTrace("create tmp file:%s", pMemBuffer->path);
   
-  tFileMeta *pFMeta = &(*pMemBuffer)->fileMeta;
+  SFileInfo *pFMeta = &pMemBuffer->fileMeta;
 
-  pFMeta->numOfElemsInFile = 0;
-  pFMeta->nFileSize = 0;
-  pFMeta->nPageSize = DEFAULT_PAGE_SIZE;
+  pFMeta->pageSize = DEFAULT_PAGE_SIZE;
 
   pFMeta->flushoutData.nAllocSize = 4;
   pFMeta->flushoutData.nLength = 0;
   pFMeta->flushoutData.pFlushoutInfo = (tFlushoutInfo *)calloc(4, sizeof(tFlushoutInfo));
 
-  (*pMemBuffer)->pColModel = tColModelCreate(pModel->pFields, pModel->numOfCols, (*pMemBuffer)->numOfElemsPerPage);
+  pMemBuffer->pColumnModel = cloneColumnModel(pModel);
+  pMemBuffer->pColumnModel->capacity = pMemBuffer->numOfElemsPerPage;
+  
+  return pMemBuffer;
 }
 
-void tExtMemBufferDestroy(tExtMemBuffer **pMemBuffer) {
-  if ((*pMemBuffer) == NULL) {
-    return;
+void* destoryExtMemBuffer(tExtMemBuffer *pMemBuffer) {
+  if (pMemBuffer == NULL) {
+    return NULL;
   }
 
   // release flush out info link
-  tFileMeta *pFileMeta = &(*pMemBuffer)->fileMeta;
+  SFileInfo *pFileMeta = &pMemBuffer->fileMeta;
   if (pFileMeta->flushoutData.nAllocSize != 0 && pFileMeta->flushoutData.pFlushoutInfo != NULL) {
     tfree(pFileMeta->flushoutData.pFlushoutInfo);
   }
 
   // release all in-memory buffer pages
-  tFilePagesItem *pFilePages = (*pMemBuffer)->pHead;
+  tFilePagesItem *pFilePages = pMemBuffer->pHead;
   while (pFilePages != NULL) {
     tFilePagesItem *pTmp = pFilePages;
     pFilePages = pFilePages->pNext;
@@ -96,24 +100,27 @@ void tExtMemBufferDestroy(tExtMemBuffer **pMemBuffer) {
   }
 
   // close temp file
-  if ((*pMemBuffer)->dataFile != 0) {
-    if (fclose((*pMemBuffer)->dataFile) != 0) {
-      pError("failed to close file:%s, reason:%s", (*pMemBuffer)->dataFilePath, strerror(errno));
+  if (pMemBuffer->file != 0) {
+    if (fclose(pMemBuffer->file) != 0) {
+      pError("failed to close file:%s, reason:%s", pMemBuffer->path, strerror(errno));
     }
     
-    pTrace("remove temp file:%s for external buffer", (*pMemBuffer)->dataFilePath);
-    unlink((*pMemBuffer)->dataFilePath);
+    pTrace("remove temp file:%s for external buffer", pMemBuffer->path);
+    unlink(pMemBuffer->path);
   }
 
-  tColModelDestroy((*pMemBuffer)->pColModel);
+  destroyColumnModel(pMemBuffer->pColumnModel);
+
+  tfree(pMemBuffer->path);
+  tfree(pMemBuffer);
   
-  tfree(*pMemBuffer);
+  return NULL;
 }
 
 /*
  * alloc more memory for flush out info entries.
  */
-static bool allocFlushoutInfoEntries(tFileMeta *pFileMeta) {
+static bool allocFlushoutInfoEntries(SFileInfo *pFileMeta) {
   pFileMeta->flushoutData.nAllocSize = pFileMeta->flushoutData.nAllocSize << 1;
 
   tFlushoutInfo *tmp = (tFlushoutInfo *)realloc(pFileMeta->flushoutData.pFlushoutInfo,
@@ -127,12 +134,12 @@ static bool allocFlushoutInfoEntries(tFileMeta *pFileMeta) {
   return true;
 }
 
-bool tExtMemBufferAlloc(tExtMemBuffer *pMemBuffer) {
-  if (pMemBuffer->numOfPagesInMem > 0 && pMemBuffer->numOfPagesInMem == pMemBuffer->nMaxSizeInPages) {
+static bool tExtMemBufferAlloc(tExtMemBuffer *pMemBuffer) {
   /*
    * the in-mem buffer is full.
    * To flush data to disk to accommodate more data
    */
+  if (pMemBuffer->numOfInMemPages > 0 && pMemBuffer->numOfInMemPages == pMemBuffer->inMemCapacity) {
     if (!tExtMemBufferFlush(pMemBuffer)) {
       return false;
     }
@@ -140,12 +147,12 @@ bool tExtMemBufferAlloc(tExtMemBuffer *pMemBuffer) {
 
   /*
    * We do not recycle the file page structure. And in flush data operations, all
-   * filepage that are full of data are destroyed after data being flushed to disk.
+   * file page that are full of data are destroyed after data being flushed to disk.
    *
    * The memory buffer pages may be recycle in order to avoid unnecessary memory
    * allocation later.
    */
-  tFilePagesItem *item = (tFilePagesItem *)calloc(1, pMemBuffer->nPageSize + sizeof(tFilePagesItem));
+  tFilePagesItem *item = (tFilePagesItem *)calloc(1, pMemBuffer->pageSize + sizeof(tFilePagesItem));
   if (item == NULL) {
     return false;
   }
@@ -161,8 +168,7 @@ bool tExtMemBufferAlloc(tExtMemBuffer *pMemBuffer) {
     pMemBuffer->pHead = item;
   }
 
-  pMemBuffer->numOfPagesInMem += 1;
-
+  pMemBuffer->numOfInMemPages += 1;
   return true;
 }
 
@@ -171,7 +177,7 @@ bool tExtMemBufferAlloc(tExtMemBuffer *pMemBuffer) {
  */
 int16_t tExtMemBufferPut(tExtMemBuffer *pMemBuffer, void *data, int32_t numOfRows) {
   if (numOfRows == 0) {
-    return pMemBuffer->numOfPagesInMem;
+    return pMemBuffer->numOfInMemPages;
   }
 
   tFilePagesItem *pLast = pMemBuffer->pTail;
@@ -183,24 +189,23 @@ int16_t tExtMemBufferPut(tExtMemBuffer *pMemBuffer, void *data, int32_t numOfRow
     pLast = pMemBuffer->pTail;
   }
 
-  if (pLast->item.numOfElems + numOfRows <= pMemBuffer->numOfElemsPerPage) {
-    // enough space for records
-    tColModelAppend(pMemBuffer->pColModel, &pLast->item, data, 0, numOfRows, numOfRows);
+  if (pLast->item.numOfElems + numOfRows <= pMemBuffer->numOfElemsPerPage) { // enough space for records
+    tColModelAppend(pMemBuffer->pColumnModel, &pLast->item, data, 0, numOfRows, numOfRows);
+    
     pMemBuffer->numOfElemsInBuffer += numOfRows;
-    pMemBuffer->numOfAllElems += numOfRows;
+    pMemBuffer->numOfTotalElems += numOfRows;
   } else {
     int32_t numOfRemainEntries = pMemBuffer->numOfElemsPerPage - pLast->item.numOfElems;
-    tColModelAppend(pMemBuffer->pColModel, &pLast->item, data, 0, numOfRemainEntries, numOfRows);
+    tColModelAppend(pMemBuffer->pColumnModel, &pLast->item, data, 0, numOfRemainEntries, numOfRows);
 
     pMemBuffer->numOfElemsInBuffer += numOfRemainEntries;
-    pMemBuffer->numOfAllElems += numOfRemainEntries;
+    pMemBuffer->numOfTotalElems += numOfRemainEntries;
 
     int32_t hasWritten = numOfRemainEntries;
     int32_t remain = numOfRows - numOfRemainEntries;
 
     while (remain > 0) {
-      if (!tExtMemBufferAlloc(pMemBuffer)) {
-        // failed to allocate memory buffer
+      if (!tExtMemBufferAlloc(pMemBuffer)) { // failed to allocate memory buffer
         return -1;
       }
 
@@ -211,10 +216,10 @@ int16_t tExtMemBufferPut(tExtMemBuffer *pMemBuffer, void *data, int32_t numOfRow
         numOfWriteElems = remain;
       }
 
-      pMemBuffer->numOfAllElems += numOfWriteElems;
+      pMemBuffer->numOfTotalElems += numOfWriteElems;
 
       pLast = pMemBuffer->pTail;
-      tColModelAppend(pMemBuffer->pColModel, &pLast->item, data, hasWritten, numOfWriteElems, numOfRows);
+      tColModelAppend(pMemBuffer->pColumnModel, &pLast->item, data, hasWritten, numOfWriteElems, numOfRows);
 
       remain -= numOfWriteElems;
       pMemBuffer->numOfElemsInBuffer += numOfWriteElems;
@@ -222,11 +227,11 @@ int16_t tExtMemBufferPut(tExtMemBuffer *pMemBuffer, void *data, int32_t numOfRow
     }
   }
 
-  return pMemBuffer->numOfPagesInMem;
+  return pMemBuffer->numOfInMemPages;
 }
 
 static bool tExtMemBufferUpdateFlushoutInfo(tExtMemBuffer *pMemBuffer) {
-  tFileMeta *pFileMeta = &pMemBuffer->fileMeta;
+  SFileInfo *pFileMeta = &pMemBuffer->fileMeta;
 
   if (pMemBuffer->flushModel == MULTIPLE_APPEND_MODEL) {
     if (pFileMeta->flushoutData.nLength == pFileMeta->flushoutData.nAllocSize && !allocFlushoutInfoEntries(pFileMeta)) {
@@ -243,46 +248,47 @@ static bool tExtMemBufferUpdateFlushoutInfo(tExtMemBuffer *pMemBuffer) {
     }
 
     // only the page still in buffer is flushed out to disk
-    pFlushoutInfo->numOfPages = pMemBuffer->numOfPagesInMem;
+    pFlushoutInfo->numOfPages = pMemBuffer->numOfInMemPages;
     pFileMeta->flushoutData.nLength += 1;
   } else {
-    // always update the first flushout array in single_flush_model
+    // always update the first flush out array in single_flush_model
     pFileMeta->flushoutData.nLength = 1;
     tFlushoutInfo *pFlushoutInfo = &pFileMeta->flushoutData.pFlushoutInfo[0];
-    pFlushoutInfo->numOfPages += pMemBuffer->numOfPagesInMem;
+    pFlushoutInfo->numOfPages += pMemBuffer->numOfInMemPages;
   }
 
   return true;
 }
 
 static void tExtMemBufferClearFlushoutInfo(tExtMemBuffer *pMemBuffer) {
-  tFileMeta *pFileMeta = &pMemBuffer->fileMeta;
+  SFileInfo *pFileMeta = &pMemBuffer->fileMeta;
 
   pFileMeta->flushoutData.nLength = 0;
   memset(pFileMeta->flushoutData.pFlushoutInfo, 0, sizeof(tFlushoutInfo) * pFileMeta->flushoutData.nAllocSize);
 }
 
 bool tExtMemBufferFlush(tExtMemBuffer *pMemBuffer) {
-  if (pMemBuffer->numOfAllElems == 0) {
+  if (pMemBuffer->numOfTotalElems == 0) {
     return true;
   }
 
-  if (pMemBuffer->dataFile == NULL) {
-    if ((pMemBuffer->dataFile = fopen(pMemBuffer->dataFilePath, "wb+")) == NULL) {
+  if (pMemBuffer->file == NULL) {
+    if ((pMemBuffer->file = fopen(pMemBuffer->path, "wb+")) == NULL) {
       return false;
     }
   }
 
-  if (pMemBuffer->numOfElemsInBuffer == 0) {
   /* all data has been flushed to disk, ignore flush operation */
+  if (pMemBuffer->numOfElemsInBuffer == 0) {
     return true;
   }
 
   bool ret = true;
+  
   tFilePagesItem *first = pMemBuffer->pHead;
 
   while (first != NULL) {
-    size_t retVal = fwrite((char *)&(first->item), pMemBuffer->nPageSize, 1, pMemBuffer->dataFile);
+    size_t retVal = fwrite((char *)&(first->item), pMemBuffer->pageSize, 1, pMemBuffer->file);
     if (retVal <= 0) {  // failed to write to buffer, may be not enough space
       ret = false;
     }
@@ -296,12 +302,12 @@ bool tExtMemBufferFlush(tExtMemBuffer *pMemBuffer) {
     tfree(ptmp);  // release all data in memory buffer
   }
 
-  fflush(pMemBuffer->dataFile);  // flush to disk
+  fflush(pMemBuffer->file);  // flush to disk
 
   tExtMemBufferUpdateFlushoutInfo(pMemBuffer);
 
   pMemBuffer->numOfElemsInBuffer = 0;
-  pMemBuffer->numOfPagesInMem = 0;
+  pMemBuffer->numOfInMemPages = 0;
   pMemBuffer->pHead = NULL;
   pMemBuffer->pTail = NULL;
 
@@ -309,11 +315,11 @@ bool tExtMemBufferFlush(tExtMemBuffer *pMemBuffer) {
 }
 
 void tExtMemBufferClear(tExtMemBuffer *pMemBuffer) {
-  if (pMemBuffer == NULL || pMemBuffer->numOfAllElems == 0) return;
+  if (pMemBuffer == NULL || pMemBuffer->numOfTotalElems == 0) {
+    return;
+  }
 
-  /*
-   * release all data in memory buffer
-   */
+  //release all data in memory buffer
   tFilePagesItem *first = pMemBuffer->pHead;
   while (first != NULL) {
     tFilePagesItem *ptmp = first;
@@ -325,15 +331,16 @@ void tExtMemBufferClear(tExtMemBuffer *pMemBuffer) {
   pMemBuffer->fileMeta.nFileSize = 0;
 
   pMemBuffer->numOfElemsInBuffer = 0;
-  pMemBuffer->numOfPagesInMem = 0;
+  pMemBuffer->numOfInMemPages = 0;
+  
   pMemBuffer->pHead = NULL;
   pMemBuffer->pTail = NULL;
 
   tExtMemBufferClearFlushoutInfo(pMemBuffer);
 
-  if (pMemBuffer->dataFile != NULL) {
   // reset the write pointer to the header
-    fseek(pMemBuffer->dataFile, 0, SEEK_SET);
+  if (pMemBuffer->file != NULL) {
+    fseek(pMemBuffer->file, 0, SEEK_SET);
   }
 }
 
@@ -347,8 +354,8 @@ bool tExtMemBufferLoadData(tExtMemBuffer *pMemBuffer, tFilePage *pFilePage, int3
     return false;
   }
 
-  size_t ret = fseek(pMemBuffer->dataFile, (pInfo->startPageId + pageIdx) * pMemBuffer->nPageSize, SEEK_SET);
-  ret = fread(pFilePage, pMemBuffer->nPageSize, 1, pMemBuffer->dataFile);
+  size_t ret = fseek(pMemBuffer->file, (pInfo->startPageId + pageIdx) * pMemBuffer->pageSize, SEEK_SET);
+  ret = fread(pFilePage, pMemBuffer->pageSize, 1, pMemBuffer->file);
 
   return (ret > 0);
 }
@@ -356,469 +363,6 @@ bool tExtMemBufferLoadData(tExtMemBuffer *pMemBuffer, tFilePage *pFilePage, int3
 bool tExtMemBufferIsAllDataInMem(tExtMemBuffer *pMemBuffer) { return (pMemBuffer->fileMeta.nFileSize == 0); }
 
 //////////////////////////////////////////////////////////////////////////////////////////////////////////////
-// TODO safty check in result
-void tBucketBigIntHash(tMemBucket *pBucket, void *value, int16_t *segIdx, int16_t *slotIdx) {
-  int64_t v = *(int64_t *)value;
-
-  if (pBucket->nRange.i64MaxVal == INT64_MIN) {
-    if (v >= 0) {
-      *segIdx = ((v >> (64 - 9)) >> 6) + 8;
-      *slotIdx = (v >> (64 - 9)) & 0x3F;
-    } else {  // v<0
-      *segIdx = ((-v) >> (64 - 9)) >> 6;
-      *slotIdx = ((-v) >> (64 - 9)) & 0x3F;
-      *segIdx = 7 - (*segIdx);
-    }
-  } else {
-    // todo hash for bigint and float and double
-    int64_t span = pBucket->nRange.i64MaxVal - pBucket->nRange.i64MinVal;
-    if (span < pBucket->nTotalSlots) {
-      int32_t delta = (int32_t)(v - pBucket->nRange.i64MinVal);
-      *segIdx = delta / pBucket->nSlotsOfSeg;
-      *slotIdx = delta % pBucket->nSlotsOfSeg;
-    } else {
-      double x = (double)span / pBucket->nTotalSlots;
-      double posx = (v - pBucket->nRange.i64MinVal) / x;
-      if (v == pBucket->nRange.i64MaxVal) {
-        posx -= 1;
-      }
-
-      *segIdx = ((int32_t)posx) / pBucket->nSlotsOfSeg;
-      *slotIdx = ((int32_t)posx) % pBucket->nSlotsOfSeg;
-    }
-  }
-}
-
-// todo refactor to more generic
-void tBucketIntHash(tMemBucket *pBucket, void *value, int16_t *segIdx, int16_t *slotIdx) {
-  int32_t v = *(int32_t *)value;
-
-  if (pBucket->nRange.iMaxVal == INT32_MIN) {
-    /*
-     * taking negative integer into consideration,
-     * there is only half of pBucket->segs available for non-negative integer
-     */
-    //        int32_t numOfSlots = pBucket->nTotalSlots>>1;
-    //        int32_t bits = bitsOfNumber(numOfSlots)-1;
-
-    if (v >= 0) {
-      *segIdx = ((v >> (32 - 9)) >> 6) + 8;
-      *slotIdx = (v >> (32 - 9)) & 0x3F;
-    } else {  // v<0
-      *segIdx = ((-v) >> (32 - 9)) >> 6;
-      *slotIdx = ((-v) >> (32 - 9)) & 0x3F;
-      *segIdx = 7 - (*segIdx);
-    }
-  } else {
-    // divide a range of [iMinVal, iMaxVal] into 1024 buckets
-    int32_t span = pBucket->nRange.iMaxVal - pBucket->nRange.iMinVal;
-    if (span < pBucket->nTotalSlots) {
-      int32_t delta = v - pBucket->nRange.iMinVal;
-      *segIdx = delta / pBucket->nSlotsOfSeg;
-      *slotIdx = delta % pBucket->nSlotsOfSeg;
-    } else {
-      double x = (double)span / pBucket->nTotalSlots;
-      double posx = (v - pBucket->nRange.iMinVal) / x;
-      if (v == pBucket->nRange.iMaxVal) {
-        posx -= 1;
-      }
-      *segIdx = ((int32_t)posx) / pBucket->nSlotsOfSeg;
-      *slotIdx = ((int32_t)posx) % pBucket->nSlotsOfSeg;
-    }
-  }
-}
-
-void tBucketDoubleHash(tMemBucket *pBucket, void *value, int16_t *segIdx, int16_t *slotIdx) {
-  //double v = *(double *)value;
-  double v = GET_DOUBLE_VAL(value);
-
-  if (pBucket->nRange.dMinVal == DBL_MAX) {
-    /*
-     * taking negative integer into consideration,
-     * there is only half of pBucket->segs available for non-negative integer
-     */
-    double x = DBL_MAX / (pBucket->nTotalSlots >> 1);
-    double posx = (v + DBL_MAX) / x;
-    *segIdx = ((int32_t)posx) / pBucket->nSlotsOfSeg;
-    *slotIdx = ((int32_t)posx) % pBucket->nSlotsOfSeg;
-  } else {
-    // divide a range of [dMinVal, dMaxVal] into 1024 buckets
-    double span = pBucket->nRange.dMaxVal - pBucket->nRange.dMinVal;
-    if (span < pBucket->nTotalSlots) {
-      int32_t delta = (int32_t)(v - pBucket->nRange.dMinVal);
-      *segIdx = delta / pBucket->nSlotsOfSeg;
-      *slotIdx = delta % pBucket->nSlotsOfSeg;
-    } else {
-      double x = span / pBucket->nTotalSlots;
-      double posx = (v - pBucket->nRange.dMinVal) / x;
-      if (v == pBucket->nRange.dMaxVal) {
-        posx -= 1;
-      }
-      *segIdx = ((int32_t)posx) / pBucket->nSlotsOfSeg;
-      *slotIdx = ((int32_t)posx) % pBucket->nSlotsOfSeg;
-    }
-
-    if (*segIdx < 0 || *segIdx > 16 || *slotIdx < 0 || *slotIdx > 64) {
-      pError("error in hash process. segment is: %d, slot id is: %d\n", *segIdx, *slotIdx);
-    }
-  }
-}
-
-tMemBucket* tMemBucketCreate(int32_t totalSlots, int32_t nBufferSize, int16_t nElemSize, int16_t dataType, tOrderDescriptor *pDesc) {
-  tMemBucket* pBucket = (tMemBucket *)malloc(sizeof(tMemBucket));
-
-  pBucket->nTotalSlots = totalSlots;
-  pBucket->nSlotsOfSeg = 1 << 6;  // 64 Segments, 16 slots each seg.
-  pBucket->dataType = dataType;
-  pBucket->nElemSize = nElemSize;
-  pBucket->nPageSize = DEFAULT_PAGE_SIZE;
-
-  pBucket->numOfElems = 0;
-  pBucket->numOfSegs = pBucket->nTotalSlots / pBucket->nSlotsOfSeg;
-
-  pBucket->nTotalBufferSize = nBufferSize;
-
-  pBucket->maxElemsCapacity = pBucket->nTotalBufferSize / pBucket->nElemSize;
-
-  pBucket->numOfTotalPages = pBucket->nTotalBufferSize / pBucket->nPageSize;
-  pBucket->numOfAvailPages = pBucket->numOfTotalPages;
-
-  pBucket->pOrderDesc = pDesc;
-
-  switch (pBucket->dataType) {
-    case TSDB_DATA_TYPE_INT:
-    case TSDB_DATA_TYPE_SMALLINT:
-    case TSDB_DATA_TYPE_TINYINT: {
-      pBucket->nRange.iMinVal = INT32_MAX;
-      pBucket->nRange.iMaxVal = INT32_MIN;
-      pBucket->HashFunc = tBucketIntHash;
-      break;
-    };
-    case TSDB_DATA_TYPE_DOUBLE:
-    case TSDB_DATA_TYPE_FLOAT: {
-      pBucket->nRange.dMinVal = DBL_MAX;
-      pBucket->nRange.dMaxVal = -DBL_MAX;
-      pBucket->HashFunc = tBucketDoubleHash;
-      break;
-    };
-    case TSDB_DATA_TYPE_BIGINT: {
-      pBucket->nRange.i64MinVal = INT64_MAX;
-      pBucket->nRange.i64MaxVal = INT64_MIN;
-      pBucket->HashFunc = tBucketBigIntHash;
-      break;
-    };
-    default: {
-      pError("MemBucket:%p,not support data type %d,failed", *pBucket, pBucket->dataType);
-      tfree(pBucket);
-      return NULL;
-    }
-  }
-
-  if (pDesc->pSchema->numOfCols != 1 || pDesc->pSchema->colOffset[0] != 0) {
-    pError("MemBucket:%p,only consecutive data is allowed,invalid numOfCols:%d or offset:%d",
-           pBucket, pDesc->pSchema->numOfCols, pDesc->pSchema->colOffset[0]);
-    tfree(pBucket);
-    return NULL;
-  }
-
-  if (pDesc->pSchema->pFields[0].type != dataType) {
-    pError("MemBucket:%p,data type is not consistent,%d in schema, %d in param", pBucket,
-           pDesc->pSchema->pFields[0].type, dataType);
-    tfree(pBucket);
-    return NULL;
-  }
-
-  if (pBucket->numOfTotalPages < pBucket->nTotalSlots) {
-    pWarn("MemBucket:%p,total buffer pages %d are not enough for all slots", pBucket, pBucket->numOfTotalPages);
-  }
-
-  pBucket->pSegs = (tMemBucketSegment *)malloc(pBucket->numOfSegs * sizeof(tMemBucketSegment));
-
-  for (int32_t i = 0; i < pBucket->numOfSegs; ++i) {
-    pBucket->pSegs[i].numOfSlots = pBucket->nSlotsOfSeg;
-    pBucket->pSegs[i].pBuffer = NULL;
-    pBucket->pSegs[i].pBoundingEntries = NULL;
-  }
-
-  pTrace("MemBucket:%p,created,buffer size:%d,elem size:%d", pBucket, pBucket->numOfTotalPages * DEFAULT_PAGE_SIZE,
-         pBucket->nElemSize);
-
-  return pBucket;
-}
-
-void tMemBucketDestroy(tMemBucket *pBucket) {
-  if (pBucket == NULL) {
-    return;
-  }
-
-  if (pBucket->pSegs) {
-    for (int32_t i = 0; i < pBucket->numOfSegs; ++i) {
-      tMemBucketSegment *pSeg = &(pBucket->pSegs[i]);
-      tfree(pSeg->pBoundingEntries);
-
-      if (pSeg->pBuffer == NULL || pSeg->numOfSlots == 0) {
-        continue;
-      }
-
-      for (int32_t j = 0; j < pSeg->numOfSlots; ++j) {
-        if (pSeg->pBuffer[j] != NULL) {
-          tExtMemBufferDestroy(&pSeg->pBuffer[j]);
-        }
-      }
-      tfree(pSeg->pBuffer);
-    }
-  }
-
-  tfree(pBucket->pSegs);
-  tfree(pBucket);
-}
-
-/*
- * find the slots which accounts for largest proportion of total in-memory buffer
- */
-static void tBucketGetMaxMemSlot(tMemBucket *pBucket, int16_t *segIdx, int16_t *slotIdx) {
-  *segIdx = -1;
-  *slotIdx = -1;
-
-  int32_t val = 0;
-  for (int32_t k = 0; k < pBucket->numOfSegs; ++k) {
-    tMemBucketSegment *pSeg = &pBucket->pSegs[k];
-    for (int32_t i = 0; i < pSeg->numOfSlots; ++i) {
-      if (pSeg->pBuffer == NULL || pSeg->pBuffer[i] == NULL) {
-        continue;
-      }
-
-      if (val < pSeg->pBuffer[i]->numOfPagesInMem) {
-        val = pSeg->pBuffer[i]->numOfPagesInMem;
-        *segIdx = k;
-        *slotIdx = i;
-      }
-    }
-  }
-}
-
-static void resetBoundingBox(tMemBucketSegment *pSeg, int32_t type) {
-  switch (type) {
-    case TSDB_DATA_TYPE_BIGINT: {
-      for (int32_t i = 0; i < pSeg->numOfSlots; ++i) {
-        pSeg->pBoundingEntries[i].i64MaxVal = INT64_MIN;
-        pSeg->pBoundingEntries[i].i64MinVal = INT64_MAX;
-      }
-      break;
-    };
-    case TSDB_DATA_TYPE_INT:
-    case TSDB_DATA_TYPE_SMALLINT:
-    case TSDB_DATA_TYPE_TINYINT: {
-      for (int32_t i = 0; i < pSeg->numOfSlots; ++i) {
-        pSeg->pBoundingEntries[i].iMaxVal = INT32_MIN;
-        pSeg->pBoundingEntries[i].iMinVal = INT32_MAX;
-      }
-      break;
-    };
-    case TSDB_DATA_TYPE_DOUBLE:
-    case TSDB_DATA_TYPE_FLOAT: {
-      for (int32_t i = 0; i < pSeg->numOfSlots; ++i) {
-        pSeg->pBoundingEntries[i].dMaxVal = -DBL_MAX;
-        pSeg->pBoundingEntries[i].dMinVal = DBL_MAX;
-      }
-      break;
-    }
-  }
-}
-
-void tMemBucketUpdateBoundingBox(MinMaxEntry *r, char *data, int32_t dataType) {
-  switch (dataType) {
-    case TSDB_DATA_TYPE_INT: {
-      int32_t val = *(int32_t *)data;
-      if (r->iMinVal > val) {
-        r->iMinVal = val;
-      }
-
-      if (r->iMaxVal < val) {
-        r->iMaxVal = val;
-      }
-      break;
-    };
-    case TSDB_DATA_TYPE_BIGINT: {
-      int64_t val = *(int64_t *)data;
-      if (r->i64MinVal > val) {
-        r->i64MinVal = val;
-      }
-
-      if (r->i64MaxVal < val) {
-        r->i64MaxVal = val;
-      }
-      break;
-    };
-    case TSDB_DATA_TYPE_SMALLINT: {
-      int32_t val = *(int16_t *)data;
-      if (r->iMinVal > val) {
-        r->iMinVal = val;
-      }
-
-      if (r->iMaxVal < val) {
-        r->iMaxVal = val;
-      }
-      break;
-    };
-    case TSDB_DATA_TYPE_TINYINT: {
-      int32_t val = *(int8_t *)data;
-      if (r->iMinVal > val) {
-        r->iMinVal = val;
-      }
-
-      if (r->iMaxVal < val) {
-        r->iMaxVal = val;
-      }
-
-      break;
-    };
-    case TSDB_DATA_TYPE_DOUBLE: {
-      //double val = *(double *)data;
-      double val = GET_DOUBLE_VAL(data);
-      if (r->dMinVal > val) {
-        r->dMinVal = val;
-      }
-
-      if (r->dMaxVal < val) {
-        r->dMaxVal = val;
-      }
-      break;
-    };
-    case TSDB_DATA_TYPE_FLOAT: {
-      //double val = *(float *)data;
-      double val = GET_FLOAT_VAL(data);
-
-      if (r->dMinVal > val) {
-        r->dMinVal = val;
-      }
-
-      if (r->dMaxVal < val) {
-        r->dMaxVal = val;
-      }
-      break;
-    };
-    default: { assert(false); }
-  }
-}
-
-/*
- * in memory bucket, we only accept the simple data consecutive put in a row/column
- * no column-model in this case.
- */
-void tMemBucketPut(tMemBucket *pBucket, void *data, int32_t numOfRows) {
-  pBucket->numOfElems += numOfRows;
-  int16_t segIdx = 0, slotIdx = 0;
-
-  for (int32_t i = 0; i < numOfRows; ++i) {
-    char *d = (char *)data + i * tDataTypeDesc[pBucket->dataType].nSize;
-
-    switch (pBucket->dataType) {
-      case TSDB_DATA_TYPE_SMALLINT: {
-        int32_t val = *(int16_t *)d;
-        (pBucket->HashFunc)(pBucket, &val, &segIdx, &slotIdx);
-        break;
-      }
-      case TSDB_DATA_TYPE_TINYINT: {
-        int32_t val = *(int8_t *)d;
-        (pBucket->HashFunc)(pBucket, &val, &segIdx, &slotIdx);
-        break;
-      }
-      case TSDB_DATA_TYPE_INT: {
-        int32_t val = *(int32_t *)d;
-        (pBucket->HashFunc)(pBucket, &val, &segIdx, &slotIdx);
-        break;
-      }
-      case TSDB_DATA_TYPE_BIGINT: {
-        int64_t val = *(int64_t *)d;
-        (pBucket->HashFunc)(pBucket, &val, &segIdx, &slotIdx);
-        break;
-      }
-      case TSDB_DATA_TYPE_DOUBLE: {
-        //double val = *(double *)d;
-        double val = GET_DOUBLE_VAL(d);
-        (pBucket->HashFunc)(pBucket, &val, &segIdx, &slotIdx);
-        break;
-      }
-      case TSDB_DATA_TYPE_FLOAT: {
-        //double val = *(float *)d;
-        double val = GET_FLOAT_VAL(d);
-        (pBucket->HashFunc)(pBucket, &val, &segIdx, &slotIdx);
-        break;
-      }
-    }
-
-    tMemBucketSegment *pSeg = &pBucket->pSegs[segIdx];
-    if (pSeg->pBoundingEntries == NULL) {
-      pSeg->pBoundingEntries = (MinMaxEntry *)malloc(sizeof(MinMaxEntry) * pBucket->nSlotsOfSeg);
-      resetBoundingBox(pSeg, pBucket->dataType);
-    }
-
-    if (pSeg->pBuffer == NULL) {
-      pSeg->pBuffer = (tExtMemBuffer **)calloc(pBucket->nSlotsOfSeg, sizeof(void *));
-    }
-
-    if (pSeg->pBuffer[slotIdx] == NULL) {
-      char    name[MAX_TMPFILE_PATH_LENGTH] = {0};
-      getTmpfilePath("tb_ex_bk_%lld_%lld_%d_%d", name);
-
-      tExtMemBufferCreate(&pSeg->pBuffer[slotIdx], pBucket->numOfTotalPages * pBucket->nPageSize, pBucket->nElemSize,
-                          name, pBucket->pOrderDesc->pSchema);
-      pSeg->pBuffer[slotIdx]->flushModel = SINGLE_APPEND_MODEL;
-      pBucket->pOrderDesc->pSchema->maxCapacity = pSeg->pBuffer[slotIdx]->numOfElemsPerPage;
-    }
-
-    tMemBucketUpdateBoundingBox(&pSeg->pBoundingEntries[slotIdx], d, pBucket->dataType);
-
-    // ensure available memory pages to allocate
-    int16_t cseg = 0, cslot = 0;
-    if (pBucket->numOfAvailPages == 0) {
-      pTrace("MemBucket:%p,max avail size:%d, no avail memory pages,", pBucket, pBucket->numOfTotalPages);
-
-      tBucketGetMaxMemSlot(pBucket, &cseg, &cslot);
-      if (cseg == -1 || cslot == -1) {
-        pError("MemBucket:%p,failed to find appropriated avail buffer", pBucket);
-        return;
-      }
-
-      if (cseg != segIdx || cslot != slotIdx) {
-        pBucket->numOfAvailPages += pBucket->pSegs[cseg].pBuffer[cslot]->numOfPagesInMem;
-
-        int32_t avail = pBucket->pSegs[cseg].pBuffer[cslot]->numOfPagesInMem;
-        UNUSED(avail);
-        tExtMemBufferFlush(pBucket->pSegs[cseg].pBuffer[cslot]);
-
-        pTrace("MemBucket:%p,seg:%d,slot:%d flushed to disk,new avail pages:%d", pBucket, cseg, cslot,
-               pBucket->numOfAvailPages);
-      } else {
-        pTrace("MemBucket:%p,failed to choose slot to flush to disk seg:%d,slot:%d",
-               pBucket, cseg, cslot);
-      }
-    }
-    int16_t consumedPgs = pSeg->pBuffer[slotIdx]->numOfPagesInMem;
-
-    int16_t newPgs = tExtMemBufferPut(pSeg->pBuffer[slotIdx], d, 1);
-    /*
-     * trigger 1. page re-allocation, to reduce the available pages
-     *         2. page flushout, to increase the available pages
-     */
-    pBucket->numOfAvailPages += (consumedPgs - newPgs);
-  }
-}
-
-void releaseBucket(tMemBucket *pMemBucket, int32_t segIdx, int32_t slotIdx) {
-  if (segIdx < 0 || segIdx > pMemBucket->numOfSegs || slotIdx < 0) {
-    return;
-  }
-
-  tMemBucketSegment *pSeg = &pMemBucket->pSegs[segIdx];
-  if (slotIdx < 0 || slotIdx >= pSeg->numOfSlots || pSeg->pBuffer[slotIdx] == NULL) {
-    return;
-  }
-
-  tExtMemBufferDestroy(&pSeg->pBuffer[slotIdx]);
-}
-
 static FORCE_INLINE int32_t primaryKeyComparator(int64_t f1, int64_t f2, int32_t colIdx, int32_t tsOrder) {
   if (f1 == f2) {
     return 0;
@@ -831,7 +375,6 @@ static FORCE_INLINE int32_t primaryKeyComparator(int64_t f1, int64_t f2, int32_t
   }
 }
 
-// todo refactor
 static FORCE_INLINE int32_t columnValueAscendingComparator(char *f1, char *f2, int32_t type, int32_t bytes) {
   switch (type) {
     case TSDB_DATA_TYPE_INT: {
@@ -910,14 +453,14 @@ int32_t compare_a(tOrderDescriptor *pDescriptor, int32_t numOfRows1, int32_t s1,
                   int32_t s2, char *data2) {
   assert(numOfRows1 == numOfRows2);
 
-  int32_t cmpCnt = pDescriptor->orderIdx.numOfOrderedCols;
+  int32_t cmpCnt = pDescriptor->orderIdx.numOfCols;
   for (int32_t i = 0; i < cmpCnt; ++i) {
     int32_t colIdx = pDescriptor->orderIdx.pData[i];
 
-    char *f1 = COLMODEL_GET_VAL(data1, pDescriptor->pSchema, numOfRows1, s1, colIdx);
-    char *f2 = COLMODEL_GET_VAL(data2, pDescriptor->pSchema, numOfRows2, s2, colIdx);
+    char *f1 = COLMODEL_GET_VAL(data1, pDescriptor->pColumnModel, numOfRows1, s1, colIdx);
+    char *f2 = COLMODEL_GET_VAL(data2, pDescriptor->pColumnModel, numOfRows2, s2, colIdx);
 
-    if (pDescriptor->pSchema->pFields[colIdx].type == TSDB_DATA_TYPE_TIMESTAMP) {
+    if (pDescriptor->pColumnModel->pFields[colIdx].field.type == TSDB_DATA_TYPE_TIMESTAMP) {
       int32_t ret = primaryKeyComparator(*(int64_t *)f1, *(int64_t *)f2, colIdx, pDescriptor->tsOrder);
       if (ret == 0) {
         continue;
@@ -925,7 +468,7 @@ int32_t compare_a(tOrderDescriptor *pDescriptor, int32_t numOfRows1, int32_t s1,
         return ret;
       }
     } else {
-      SSchema *pSchema = &pDescriptor->pSchema->pFields[colIdx];
+      SSchema *pSchema = &pDescriptor->pColumnModel->pFields[colIdx];
       int32_t  ret = columnValueAscendingComparator(f1, f2, pSchema->type, pSchema->bytes);
       if (ret == 0) {
         continue;
@@ -942,14 +485,14 @@ int32_t compare_d(tOrderDescriptor *pDescriptor, int32_t numOfRows1, int32_t s1,
                   int32_t s2, char *data2) {
   assert(numOfRows1 == numOfRows2);
 
-  int32_t cmpCnt = pDescriptor->orderIdx.numOfOrderedCols;
+  int32_t cmpCnt = pDescriptor->orderIdx.numOfCols;
   for (int32_t i = 0; i < cmpCnt; ++i) {
     int32_t colIdx = pDescriptor->orderIdx.pData[i];
 
-    char *f1 = COLMODEL_GET_VAL(data1, pDescriptor->pSchema, numOfRows1, s1, colIdx);
-    char *f2 = COLMODEL_GET_VAL(data2, pDescriptor->pSchema, numOfRows2, s2, colIdx);
+    char *f1 = COLMODEL_GET_VAL(data1, pDescriptor->pColumnModel, numOfRows1, s1, colIdx);
+    char *f2 = COLMODEL_GET_VAL(data2, pDescriptor->pColumnModel, numOfRows2, s2, colIdx);
 
-    if (pDescriptor->pSchema->pFields[colIdx].type == TSDB_DATA_TYPE_TIMESTAMP) {
+    if (pDescriptor->pColumnModel->pFields[colIdx].field.type == TSDB_DATA_TYPE_TIMESTAMP) {
       int32_t ret = primaryKeyComparator(*(int64_t *)f1, *(int64_t *)f2, colIdx, pDescriptor->tsOrder);
       if (ret == 0) {
         continue;
@@ -957,7 +500,7 @@ int32_t compare_d(tOrderDescriptor *pDescriptor, int32_t numOfRows1, int32_t s1,
         return ret;
       }
     } else {
-      SSchema *pSchema = &pDescriptor->pSchema->pFields[colIdx];
+      SSchema *pSchema = &pDescriptor->pColumnModel->pFields[colIdx];
       int32_t  ret = columnValueAscendingComparator(f1, f2, pSchema->type, pSchema->bytes);
       if (ret == 0) {
         continue;
@@ -979,12 +522,13 @@ FORCE_INLINE int32_t compare_sd(tOrderDescriptor *pDescriptor, int32_t numOfRows
   return compare_d(pDescriptor, numOfRows, idx1, data, numOfRows, idx2, data);
 }
 
-static void swap(tOrderDescriptor *pDescriptor, int32_t count, int32_t s1, char *data1, int32_t s2) {
-  for (int32_t i = 0; i < pDescriptor->pSchema->numOfCols; ++i) {
-    void *first = COLMODEL_GET_VAL(data1, pDescriptor->pSchema, count, s1, i);
-    void *second = COLMODEL_GET_VAL(data1, pDescriptor->pSchema, count, s2, i);
+static void swap(SColumnModel *pColumnModel, int32_t count, int32_t s1, char *data1, int32_t s2) {
+  for (int32_t i = 0; i < pColumnModel->numOfCols; ++i) {
+    void *first = COLMODEL_GET_VAL(data1, pColumnModel, count, s1, i);
+    void *second = COLMODEL_GET_VAL(data1, pColumnModel, count, s2, i);
 
-    tsDataSwap(first, second, pDescriptor->pSchema->pFields[i].type, pDescriptor->pSchema->pFields[i].bytes);
+    SSchema* pSchema = &pColumnModel->pFields[i].field;
+    tsDataSwap(first, second, pSchema->type, pSchema->bytes);
   }
 }
 
@@ -993,7 +537,7 @@ static void tColDataInsertSort(tOrderDescriptor *pDescriptor, int32_t numOfRows,
   for (int32_t i = start + 1; i <= end; ++i) {
     for (int32_t j = i; j > start; --j) {
       if (compareFn(pDescriptor, numOfRows, j, j - 1, data) == -1) {
-        swap(pDescriptor, numOfRows, j - 1, data, j);
+        swap(pDescriptor->pColumnModel, numOfRows, j - 1, data, j);
       } else {
         break;
       }
@@ -1035,33 +579,33 @@ static void median(tOrderDescriptor *pDescriptor, int32_t numOfRows, int32_t sta
 #if defined(_DEBUG_VIEW)
   int32_t f = pDescriptor->orderIdx.pData[0];
 
-  char *midx = COLMODEL_GET_VAL(data, pDescriptor->pSchema, numOfRows, midIdx, f);
-  char *startx = COLMODEL_GET_VAL(data, pDescriptor->pSchema, numOfRows, start, f);
-  char *endx = COLMODEL_GET_VAL(data, pDescriptor->pSchema, numOfRows, end, f);
+  char *midx = COLMODEL_GET_VAL(data, pDescriptor->pColumnModel, numOfRows, midIdx, f);
+  char *startx = COLMODEL_GET_VAL(data, pDescriptor->pColumnModel, numOfRows, start, f);
+  char *endx = COLMODEL_GET_VAL(data, pDescriptor->pColumnModel, numOfRows, end, f);
 
   int32_t colIdx = pDescriptor->orderIdx.pData[0];
-  tSortDataPrint(pDescriptor->pSchema->pFields[colIdx].type, "before", startx, midx, endx);
+  tSortDataPrint(pDescriptor->pColumnModel->pFields[colIdx].field.type, "before", startx, midx, endx);
 #endif
 
   if (compareFn(pDescriptor, numOfRows, midIdx, start, data) == 1) {
-    swap(pDescriptor, numOfRows, start, data, midIdx);
+    swap(pDescriptor->pColumnModel, numOfRows, start, data, midIdx);
   }
 
   if (compareFn(pDescriptor, numOfRows, midIdx, end, data) == 1) {
-    swap(pDescriptor, numOfRows, midIdx, data, start);
-    swap(pDescriptor, numOfRows, midIdx, data, end);
+    swap(pDescriptor->pColumnModel, numOfRows, midIdx, data, start);
+    swap(pDescriptor->pColumnModel, numOfRows, midIdx, data, end);
   } else if (compareFn(pDescriptor, numOfRows, start, end, data) == 1) {
-    swap(pDescriptor, numOfRows, start, data, end);
+    swap(pDescriptor->pColumnModel, numOfRows, start, data, end);
   }
 
   assert(compareFn(pDescriptor, numOfRows, midIdx, start, data) <= 0 &&
          compareFn(pDescriptor, numOfRows, start, end, data) <= 0);
 
 #if defined(_DEBUG_VIEW)
-  midx = COLMODEL_GET_VAL(data, pDescriptor->pSchema, numOfRows, midIdx, f);
-  startx = COLMODEL_GET_VAL(data, pDescriptor->pSchema, numOfRows, start, f);
-  endx = COLMODEL_GET_VAL(data, pDescriptor->pSchema, numOfRows, end, f);
-  tSortDataPrint(pDescriptor->pSchema->pFields[colIdx].type, "after", startx, midx, endx);
+  midx = COLMODEL_GET_VAL(data, pDescriptor->pColumnModel, numOfRows, midIdx, f);
+  startx = COLMODEL_GET_VAL(data, pDescriptor->pColumnModel, numOfRows, start, f);
+  endx = COLMODEL_GET_VAL(data, pDescriptor->pColumnModel, numOfRows, end, f);
+  tSortDataPrint(pDescriptor->pColumnModel->pFields[colIdx].field.type, "after", startx, midx, endx);
 #endif
 }
 
@@ -1069,9 +613,9 @@ static UNUSED_FUNC void tRowModelDisplay(tOrderDescriptor *pDescriptor, int32_t
   int32_t colIdx = pDescriptor->orderIdx.pData[0];
 
   for (int32_t i = 0; i < len; ++i) {
-    char *startx = COLMODEL_GET_VAL(d, pDescriptor->pSchema, numOfRows, i, colIdx);
+    char *startx = COLMODEL_GET_VAL(d, pDescriptor->pColumnModel, numOfRows, i, colIdx);
 
-    switch (pDescriptor->pSchema->pFields[colIdx].type) {
+    switch (pDescriptor->pColumnModel->pFields[colIdx].field.type) {
       case TSDB_DATA_TYPE_DOUBLE:
         printf("%lf\t", *(double *)startx);
         break;
@@ -1115,15 +659,15 @@ void tColDataQSort(tOrderDescriptor *pDescriptor, int32_t numOfRows, int32_t sta
   }
 
 #ifdef _DEBUG_VIEW
-  printf("before sort:\n");
-  tRowModelDisplay(pDescriptor, numOfRows, data, end - start + 1);
+//  printf("before sort:\n");
+//  tRowModelDisplay(pDescriptor, numOfRows, data, end - start + 1);
 #endif
 
   int32_t s = start, e = end;
   median(pDescriptor, numOfRows, start, end, data, compareFn);
 
 #ifdef _DEBUG_VIEW
-  printf("%s called: %d\n", __FUNCTION__, qsort_call++);
+//  printf("%s called: %d\n", __FUNCTION__, qsort_call++);
 #endif
 
   UNUSED(qsort_call);
@@ -1139,17 +683,17 @@ void tColDataQSort(tOrderDescriptor *pDescriptor, int32_t numOfRows, int32_t sta
       }
 
       if (ret == 0 && e != end_same) {
-        swap(pDescriptor, numOfRows, e, data, end_same--);
+        swap(pDescriptor->pColumnModel, numOfRows, e, data, end_same--);
       }
       e--;
     }
 
     if (e != s) {
-      swap(pDescriptor, numOfRows, s, data, e);
+      swap(pDescriptor->pColumnModel, numOfRows, s, data, e);
     }
 
 #ifdef _DEBUG_VIEW
-    tRowModelDisplay(pDescriptor, numOfRows, data, end - start + 1);
+//    tRowModelDisplay(pDescriptor, numOfRows, data, end - start + 1);
 #endif
 
     while (s < e) {
@@ -1159,16 +703,16 @@ void tColDataQSort(tOrderDescriptor *pDescriptor, int32_t numOfRows, int32_t sta
       }
 
       if (ret == 0 && s != start_same) {
-        swap(pDescriptor, numOfRows, s, data, start_same++);
+        swap(pDescriptor->pColumnModel, numOfRows, s, data, start_same++);
       }
       s++;
     }
 
     if (s != e) {
-      swap(pDescriptor, numOfRows, s, data, e);
+      swap(pDescriptor->pColumnModel, numOfRows, s, data, e);
     }
 #ifdef _DEBUG_VIEW
-    tRowModelDisplay(pDescriptor, numOfRows, data, end - start + 1);
+//    tRowModelDisplay(pDescriptor, numOfRows, data, end - start + 1);
 #endif
   }
 
@@ -1178,14 +722,14 @@ void tColDataQSort(tOrderDescriptor *pDescriptor, int32_t numOfRows, int32_t sta
     int32_t right = end;
 
     while (right > end_same && left <= end_same) {
-      swap(pDescriptor, numOfRows, left++, data, right--);
+      swap(pDescriptor->pColumnModel, numOfRows, left++, data, right--);
     }
 
     // (pivotal+1) + steps of number that are identical pivotal
     rightx += (end - end_same);
 
 #ifdef _DEBUG_VIEW
-    tRowModelDisplay(pDescriptor, numOfRows, data, end - start + 1);
+//    tRowModelDisplay(pDescriptor, numOfRows, data, end - start + 1);
 #endif
   }
 
@@ -1195,14 +739,14 @@ void tColDataQSort(tOrderDescriptor *pDescriptor, int32_t numOfRows, int32_t sta
     int32_t right = e - 1;
 
     while (left < start_same && right >= start_same) {
-      swap(pDescriptor, numOfRows, left++, data, right--);
+      swap(pDescriptor->pColumnModel, numOfRows, left++, data, right--);
     }
 
     // (pivotal-1) - steps of number that are identical pivotal
     leftx -= (start_same - start);
 
 #ifdef _DEBUG_VIEW
-    tRowModelDisplay(pDescriptor, numOfRows, data, end - start + 1);
+//    tRowModelDisplay(pDescriptor, numOfRows, data, end - start + 1);
 #endif
   }
 
@@ -1215,143 +759,50 @@ void tColDataQSort(tOrderDescriptor *pDescriptor, int32_t numOfRows, int32_t sta
   }
 }
 
-tExtMemBuffer *releaseBucketsExceptFor(tMemBucket *pMemBucket, int16_t segIdx, int16_t slotIdx) {
-  tExtMemBuffer *pBuffer = NULL;
-
-  for (int32_t i = 0; i < pMemBucket->numOfSegs; ++i) {
-    tMemBucketSegment *pSeg = &pMemBucket->pSegs[i];
-
-    for (int32_t j = 0; j < pSeg->numOfSlots; ++j) {
-      if (i == segIdx && j == slotIdx) {
-        pBuffer = pSeg->pBuffer[j];
-      } else {
-        if (pSeg->pBuffer && pSeg->pBuffer[j]) {
-          tExtMemBufferDestroy(&pSeg->pBuffer[j]);
-        }
-      }
-    }
+/*
+ * deep copy of sschema
+ */
+SColumnModel *createColumnModel(SSchema *fields, int32_t numOfCols, int32_t blockCapacity) {
+  SColumnModel *pColumnModel = (SColumnModel *)calloc(1, sizeof(SColumnModel) + numOfCols * sizeof(SSchemaEx));
+  if (pColumnModel == NULL) {
+    return NULL;
   }
 
-  return pBuffer;
-}
-
-static tFilePage *loadIntoBucketFromDisk(tMemBucket *pMemBucket, int32_t segIdx, int32_t slotIdx,
-                                         tOrderDescriptor *pDesc) {
-  // release all data in other slots
-  tExtMemBuffer *pMemBuffer = pMemBucket->pSegs[segIdx].pBuffer[slotIdx];
-  tFilePage *    buffer = (tFilePage *)calloc(1, pMemBuffer->nElemSize * pMemBuffer->numOfAllElems + sizeof(tFilePage));
-  int32_t        oldCapacity = pDesc->pSchema->maxCapacity;
-  pDesc->pSchema->maxCapacity = pMemBuffer->numOfAllElems;
-
-  if (!tExtMemBufferIsAllDataInMem(pMemBuffer)) {
-    pMemBuffer = releaseBucketsExceptFor(pMemBucket, segIdx, slotIdx);
-    assert(pMemBuffer->numOfAllElems > 0);
-
-    // load data in disk to memory
-    tFilePage *pPage = (tFilePage *)calloc(1, pMemBuffer->nPageSize);
+  pColumnModel->pFields = (SSchemaEx *)(&pColumnModel[1]);
   
-    for (int32_t i = 0; i < pMemBuffer->fileMeta.flushoutData.nLength; ++i) {
-      tFlushoutInfo *pFlushInfo = &pMemBuffer->fileMeta.flushoutData.pFlushoutInfo[i];
+  for(int32_t i = 0; i < numOfCols; ++i) {
+    SSchemaEx* pSchemaEx = &pColumnModel->pFields[i];
+    pSchemaEx->field = fields[i];
+    pSchemaEx->offset = pColumnModel->rowSize;
     
-      int32_t ret = fseek(pMemBuffer->dataFile, pFlushInfo->startPageId * pMemBuffer->nPageSize, SEEK_SET);
-      UNUSED(ret);
-
-      for (uint32_t j = 0; j < pFlushInfo->numOfPages; ++j) {
-        ret = fread(pPage, pMemBuffer->nPageSize, 1, pMemBuffer->dataFile);
-        UNUSED(ret);
-        assert(pPage->numOfElems > 0);
-
-        tColModelAppend(pDesc->pSchema, buffer, pPage->data, 0, pPage->numOfElems, pPage->numOfElems);
-        printf("id: %d  count: %" PRIu64 "\n", j, buffer->numOfElems);
-      }
+    pColumnModel->rowSize += pSchemaEx->field.bytes;
   }
-    tfree(pPage);
 
-    assert(buffer->numOfElems == pMemBuffer->fileMeta.numOfElemsInFile);
-  }
-
-  // load data in pMemBuffer to buffer
-  tFilePagesItem *pListItem = pMemBuffer->pHead;
-  while (pListItem != NULL) {
-    tColModelAppend(pDesc->pSchema, buffer, pListItem->item.data, 0, pListItem->item.numOfElems,
-                    pListItem->item.numOfElems);
-    pListItem = pListItem->pNext;
-  }
-
-  tColDataQSort(pDesc, buffer->numOfElems, 0, buffer->numOfElems - 1, buffer->data, TSQL_SO_ASC);
+  pColumnModel->numOfCols = numOfCols;
+  pColumnModel->capacity = blockCapacity;
 
-  pDesc->pSchema->maxCapacity = oldCapacity;  // restore value
-  return buffer;
+  return pColumnModel;
 }
 
-double findOnlyResult(tMemBucket *pMemBucket) {
-  assert(pMemBucket->numOfElems == 1);
-
-  for (int32_t i = 0; i < pMemBucket->numOfSegs; ++i) {
-    tMemBucketSegment *pSeg = &pMemBucket->pSegs[i];
-    if (pSeg->pBuffer) {
-      for (int32_t j = 0; j < pSeg->numOfSlots; ++j) {
-        tExtMemBuffer *pBuffer = pSeg->pBuffer[j];
-        if (pBuffer) {
-          assert(pBuffer->numOfAllElems == 1);
-          tFilePage *pPage = &pBuffer->pHead->item;
-          if (pBuffer->numOfElemsInBuffer == 1) {
-            switch (pMemBucket->dataType) {
-              case TSDB_DATA_TYPE_INT:
-                return *(int32_t *)pPage->data;
-              case TSDB_DATA_TYPE_SMALLINT:
-                return *(int16_t *)pPage->data;
-              case TSDB_DATA_TYPE_TINYINT:
-                return *(int8_t *)pPage->data;
-              case TSDB_DATA_TYPE_BIGINT:
-                return (double)(*(int64_t *)pPage->data);
-              case TSDB_DATA_TYPE_DOUBLE: {
-		double dv = GET_DOUBLE_VAL(pPage->data);				  
-                //return *(double *)pPage->data;
-		return dv;
-	      }
-              case TSDB_DATA_TYPE_FLOAT: {
-		float fv = GET_FLOAT_VAL(pPage->data);
-                //return *(float *)pPage->data;
-		return fv;
-	      }
-              default:
-                return 0;
-            }
-          }
-        }
-      }
-    }
+SColumnModel *cloneColumnModel(SColumnModel *pSrc) {
+  if (pSrc == NULL) {
+    return NULL;
   }
-  return 0;
-}
   
-/*
- * deep copy of sschema
- */
-tColModel *tColModelCreate(SSchema *field, int32_t numOfCols, int32_t maxCapacity) {
-  tColModel *pSchema =
-      (tColModel *)calloc(1, sizeof(tColModel) + numOfCols * sizeof(SSchema) + numOfCols * sizeof(int16_t));
-  if (pSchema == NULL) {
+  SColumnModel *pColumnModel = (SColumnModel *)calloc(1, sizeof(SColumnModel) + pSrc->numOfCols * sizeof(SSchemaEx));
+  if (pColumnModel == NULL) {
     return NULL;
   }
   
-  pSchema->pFields = (SSchema *)(&pSchema[1]);
-  memcpy(pSchema->pFields, field, sizeof(SSchema) * numOfCols);
+  *pColumnModel = *pSrc;
   
-  pSchema->colOffset = (int16_t *)(&pSchema->pFields[numOfCols]);
-  pSchema->colOffset[0] = 0;
-  for (int32_t i = 1; i < numOfCols; ++i) {
-    pSchema->colOffset[i] = pSchema->colOffset[i - 1] + pSchema->pFields[i - 1].bytes;
-  }
+  pColumnModel->pFields = (SSchemaEx*) (&pColumnModel[1]);
+  memcpy(pColumnModel->pFields, pSrc->pFields, pSrc->numOfCols * sizeof(SSchemaEx));
   
-  pSchema->numOfCols = numOfCols;
-  pSchema->maxCapacity = maxCapacity;
-
-  return pSchema;
+  return pColumnModel;
 }
 
-void tColModelDestroy(tColModel *pModel) {
+void destroyColumnModel(SColumnModel *pModel) {
   if (pModel == NULL) {
     return;
   }
@@ -1444,12 +895,12 @@ static void printBinaryDataEx(char *data, int32_t len, SSrcColumnInfo *param) {
   }
 }
 
-void tColModelDisplay(tColModel *pModel, void *pData, int32_t numOfRows, int32_t totalCapacity) {
+void tColModelDisplay(SColumnModel *pModel, void *pData, int32_t numOfRows, int32_t totalCapacity) {
   for (int32_t i = 0; i < numOfRows; ++i) {
     for (int32_t j = 0; j < pModel->numOfCols; ++j) {
       char *val = COLMODEL_GET_VAL((char *)pData, pModel, totalCapacity, i, j);
 
-      int type = pModel->pFields[j].type;
+      int type = pModel->pFields[j].field.type;
       printf("type:%d ", type);
 
       switch (type) {
@@ -1461,11 +912,11 @@ void tColModelDisplay(tColModel *pModel, void *pData, int32_t numOfRows, int32_t
           break;
         case TSDB_DATA_TYPE_NCHAR: {
           char buf[4096] = {0};
-          taosUcs4ToMbs(val, pModel->pFields[j].bytes, buf);
+          taosUcs4ToMbs(val, pModel->pFields[j].field.bytes, buf);
           printf("%s\t", buf);
         }
         case TSDB_DATA_TYPE_BINARY: {
-          printBinaryData(val, pModel->pFields[j].bytes);
+          printBinaryData(val, pModel->pFields[j].field.bytes);
           break;
         }
         case TSDB_DATA_TYPE_DOUBLE:
@@ -1495,15 +946,15 @@ void tColModelDisplay(tColModel *pModel, void *pData, int32_t numOfRows, int32_t
   printf("\n");
 }
 
-void tColModelDisplayEx(tColModel *pModel, void *pData, int32_t numOfRows, int32_t totalCapacity,
+void tColModelDisplayEx(SColumnModel *pModel, void *pData, int32_t numOfRows, int32_t totalCapacity,
                         SSrcColumnInfo *param) {
   for (int32_t i = 0; i < numOfRows; ++i) {
     for (int32_t j = 0; j < pModel->numOfCols; ++j) {
       char *val = COLMODEL_GET_VAL((char *)pData, pModel, totalCapacity, i, j);
 
-      printf("type:%d\t", pModel->pFields[j].type);
+      printf("type:%d\t", pModel->pFields[j].field.type);
 
-      switch (pModel->pFields[j].type) {
+      switch (pModel->pFields[j].field.type) {
         case TSDB_DATA_TYPE_BIGINT:
           printf("%" PRId64 "\t", *(int64_t *)val);
           break;
@@ -1512,11 +963,11 @@ void tColModelDisplayEx(tColModel *pModel, void *pData, int32_t numOfRows, int32
           break;
         case TSDB_DATA_TYPE_NCHAR: {
           char buf[128] = {0};
-          taosUcs4ToMbs(val, pModel->pFields[j].bytes, buf);
+          taosUcs4ToMbs(val, pModel->pFields[j].field.bytes, buf);
           printf("%s\t", buf);
         }
         case TSDB_DATA_TYPE_BINARY: {
-          printBinaryDataEx(val, pModel->pFields[j].bytes, &param[j]);
+          printBinaryDataEx(val, pModel->pFields[j].field.bytes, &param[j]);
           break;
         }
         case TSDB_DATA_TYPE_DOUBLE:
@@ -1547,20 +998,31 @@ void tColModelDisplayEx(tColModel *pModel, void *pData, int32_t numOfRows, int32
 }
 
 ////////////////////////////////////////////////////////////////////////////////////////////
-void tColModelCompact(tColModel *pModel, tFilePage *inputBuffer, int32_t maxElemsCapacity) {
+void tColModelCompact(SColumnModel *pModel, tFilePage *inputBuffer, int32_t maxElemsCapacity) {
   if (inputBuffer->numOfElems == 0 || maxElemsCapacity == inputBuffer->numOfElems) {
     return;
   }
 
   /* start from the second column */
   for (int32_t i = 1; i < pModel->numOfCols; ++i) {
-    memmove(inputBuffer->data + pModel->colOffset[i] * inputBuffer->numOfElems,
-            inputBuffer->data + pModel->colOffset[i] * maxElemsCapacity,
-            pModel->pFields[i].bytes * inputBuffer->numOfElems);
+    SSchemaEx* pSchemaEx = &pModel->pFields[i];
+    memmove(inputBuffer->data + pSchemaEx->offset * inputBuffer->numOfElems,
+            inputBuffer->data + pSchemaEx->offset * maxElemsCapacity,
+            pSchemaEx->field.bytes * inputBuffer->numOfElems);
   }
 }
 
-void tColModelErase(tColModel *pModel, tFilePage *inputBuffer, int32_t maxCapacity, int32_t s, int32_t e) {
+SSchema* getColumnModelSchema(SColumnModel *pColumnModel, int32_t index) {
+  assert(pColumnModel != NULL && index >= 0 && index < pColumnModel->numOfCols);
+  return &pColumnModel->pFields[index].field;
+}
+
+int16_t getColumnModelOffset(SColumnModel *pColumnModel, int32_t index) {
+  assert(pColumnModel != NULL && index >= 0 && index < pColumnModel->numOfCols);
+  return pColumnModel->pFields[index].offset;
+}
+
+void tColModelErase(SColumnModel *pModel, tFilePage *inputBuffer, int32_t blockCapacity, int32_t s, int32_t e) {
   if (inputBuffer->numOfElems == 0 || (e - s + 1) <= 0) {
     return;
   }
@@ -1571,10 +1033,13 @@ void tColModelErase(tColModel *pModel, tFilePage *inputBuffer, int32_t maxCapaci
 
   /* start from the second column */
   for (int32_t i = 0; i < pModel->numOfCols; ++i) {
-    char *startPos = inputBuffer->data + pModel->colOffset[i] * maxCapacity + s * pModel->pFields[i].bytes;
-    char *endPos = startPos + pModel->pFields[i].bytes * removed;
+    int16_t offset = getColumnModelOffset(pModel, i);
+    SSchema* pSchema = getColumnModelSchema(pModel, i);
     
-    memmove(startPos, endPos, pModel->pFields[i].bytes * secPart);
+    char *startPos = inputBuffer->data + offset * blockCapacity + s * pSchema->bytes;
+    char *endPos = startPos + pSchema->bytes * removed;
+
+    memmove(startPos, endPos, pSchema->bytes * secPart);
   }
 
   inputBuffer->numOfElems = remain;
@@ -1587,31 +1052,31 @@ void tColModelErase(tColModel *pModel, tFilePage *inputBuffer, int32_t maxCapaci
  * data in srcData must has the same schema as data in dstPage, that can be
  * described by dstModel
  */
-void tColModelAppend(tColModel *dstModel, tFilePage *dstPage, void *srcData, int32_t start, int32_t numOfRows,
+void tColModelAppend(SColumnModel *dstModel, tFilePage *dstPage, void *srcData, int32_t start, int32_t numOfRows,
                      int32_t srcCapacity) {
-  assert(dstPage->numOfElems + numOfRows <= dstModel->maxCapacity);
+  assert(dstPage->numOfElems + numOfRows <= dstModel->capacity);
 
   for (int32_t col = 0; col < dstModel->numOfCols; ++col) {
-    char *dst = COLMODEL_GET_VAL(dstPage->data, dstModel, dstModel->maxCapacity, dstPage->numOfElems, col);
+    char *dst = COLMODEL_GET_VAL(dstPage->data, dstModel, dstModel->capacity, dstPage->numOfElems, col);
     char *src = COLMODEL_GET_VAL((char *)srcData, dstModel, srcCapacity, start, col);
 
-    memmove(dst, src, dstModel->pFields[col].bytes * numOfRows);
+    memmove(dst, src, dstModel->pFields[col].field.bytes * numOfRows);
   }
 
   dstPage->numOfElems += numOfRows;
 }
 
-tOrderDescriptor *tOrderDesCreate(int32_t *orderColIdx, int32_t numOfOrderCols, tColModel *pModel,
+tOrderDescriptor *tOrderDesCreate(const int32_t *orderColIdx, int32_t numOfOrderCols, SColumnModel *pModel,
                                   int32_t tsOrderType) {
   tOrderDescriptor *desc = (tOrderDescriptor *)calloc(1, sizeof(tOrderDescriptor) + sizeof(int32_t) * numOfOrderCols);
   if (desc == NULL) {
     return NULL;
   }
 
-  desc->pSchema = pModel;
+  desc->pColumnModel = pModel;
   desc->tsOrder = tsOrderType;
 
-  desc->orderIdx.numOfOrderedCols = numOfOrderCols;
+  desc->orderIdx.numOfCols = numOfOrderCols;
   for (int32_t i = 0; i < numOfOrderCols; ++i) {
     desc->orderIdx.pData[i] = orderColIdx[i];
   }
@@ -1624,390 +1089,6 @@ void tOrderDescDestroy(tOrderDescriptor *pDesc) {
     return;
   }
 
-  tColModelDestroy(pDesc->pSchema);
+  destroyColumnModel(pDesc->pColumnModel);
   tfree(pDesc);
 }
-
-////////////////////////////////////////////////////////////////////////////////////////////
-static void findMaxMinValue(tMemBucket *pMemBucket, double *maxVal, double *minVal) {
-  *minVal = DBL_MAX;
-  *maxVal = -DBL_MAX;
-
-  for (int32_t i = 0; i < pMemBucket->numOfSegs; ++i) {
-    tMemBucketSegment *pSeg = &pMemBucket->pSegs[i];
-    if (pSeg->pBuffer == NULL) {
-      continue;
-    }
-    switch (pMemBucket->dataType) {
-      case TSDB_DATA_TYPE_INT:
-      case TSDB_DATA_TYPE_SMALLINT:
-      case TSDB_DATA_TYPE_TINYINT: {
-        for (int32_t j = 0; j < pSeg->numOfSlots; ++j) {
-          double minv = pSeg->pBoundingEntries[j].iMinVal;
-          double maxv = pSeg->pBoundingEntries[j].iMaxVal;
-
-          if (*minVal > minv) {
-            *minVal = minv;
-          }
-          if (*maxVal < maxv) {
-            *maxVal = maxv;
-          }
-        }
-        break;
-      }
-      case TSDB_DATA_TYPE_DOUBLE:
-      case TSDB_DATA_TYPE_FLOAT: {
-        for (int32_t j = 0; j < pSeg->numOfSlots; ++j) {
-          double minv = pSeg->pBoundingEntries[j].dMinVal;
-          double maxv = pSeg->pBoundingEntries[j].dMaxVal;
-
-          if (*minVal > minv) {
-            *minVal = minv;
-          }
-          if (*maxVal < maxv) {
-            *maxVal = maxv;
-          }
-        }
-        break;
-      }
-      case TSDB_DATA_TYPE_BIGINT: {
-        for (int32_t j = 0; j < pSeg->numOfSlots; ++j) {
-          double minv = (double)pSeg->pBoundingEntries[j].i64MinVal;
-          double maxv = (double)pSeg->pBoundingEntries[j].i64MaxVal;
-
-          if (*minVal > minv) {
-            *minVal = minv;
-          }
-          if (*maxVal < maxv) {
-            *maxVal = maxv;
-          }
-        }
-        break;
-      }
-    }
-  }
-}
-
-static MinMaxEntry getMinMaxEntryOfNearestSlotInNextSegment(tMemBucket *pMemBucket, int32_t segIdx) {
-  int32_t i = segIdx + 1;
-  while (i < pMemBucket->numOfSegs && pMemBucket->pSegs[i].numOfSlots == 0) ++i;
-
-  tMemBucketSegment *pSeg = &pMemBucket->pSegs[i];
-  assert(pMemBucket->numOfSegs > i && pMemBucket->pSegs[i].pBuffer != NULL);
-
-  i = 0;
-  while (i < pMemBucket->nSlotsOfSeg && pSeg->pBuffer[i] == NULL) ++i;
-
-  assert(i < pMemBucket->nSlotsOfSeg);
-  return pSeg->pBoundingEntries[i];
-}
-
-/*
- *
- * now, we need to find the minimum value of the next slot for
- * interpolating the percentile value
- * j is the last slot of current segment, we need to get the first
- * slot of the next segment.
- */
-static MinMaxEntry getMinMaxEntryOfNextSlotWithData(tMemBucket *pMemBucket, int32_t segIdx, int32_t slotIdx) {
-  tMemBucketSegment *pSeg = &pMemBucket->pSegs[segIdx];
-
-  MinMaxEntry next;
-  if (slotIdx == pSeg->numOfSlots - 1) {  // find next segment with data
-    return getMinMaxEntryOfNearestSlotInNextSegment(pMemBucket, segIdx);
-  } else {
-    int32_t j = slotIdx + 1;
-    for (; j < pMemBucket->nSlotsOfSeg && pMemBucket->pSegs[segIdx].pBuffer[j] == 0; ++j) {
-    };
-
-    if (j == pMemBucket->nSlotsOfSeg) {  // current slot has no available
-                                         // slot,try next segment
-      return getMinMaxEntryOfNearestSlotInNextSegment(pMemBucket, segIdx);
-    } else {
-      next = pSeg->pBoundingEntries[slotIdx + 1];
-      assert(pSeg->pBuffer[slotIdx + 1] != NULL);
-    }
-  }
-
-  return next;
-}
-
-bool isIdenticalData(tMemBucket *pMemBucket, int32_t segIdx, int32_t slotIdx);
-char *getFirstElemOfMemBuffer(tMemBucketSegment *pSeg, int32_t slotIdx, tFilePage *pPage);
-
-double getPercentileImpl(tMemBucket *pMemBucket, int32_t count, double fraction) {
-  int32_t num = 0;
-
-  for (int32_t i = 0; i < pMemBucket->numOfSegs; ++i) {
-    tMemBucketSegment *pSeg = &pMemBucket->pSegs[i];
-    for (int32_t j = 0; j < pSeg->numOfSlots; ++j) {
-      if (pSeg->pBuffer == NULL || pSeg->pBuffer[j] == NULL) {
-        continue;
-      }
-      // required value in current slot
-      if (num < (count + 1) && num + pSeg->pBuffer[j]->numOfAllElems >= (count + 1)) {
-        if (pSeg->pBuffer[j]->numOfAllElems + num == (count + 1)) {
-          /*
-           * now, we need to find the minimum value of the next slot for interpolating the percentile value
-           * j is the last slot of current segment, we need to get the first slot of the next segment.
-           *
-           */
-          MinMaxEntry next = getMinMaxEntryOfNextSlotWithData(pMemBucket, i, j);
-
-          double maxOfThisSlot = 0;
-          double minOfNextSlot = 0;
-          switch (pMemBucket->dataType) {
-            case TSDB_DATA_TYPE_INT:
-            case TSDB_DATA_TYPE_SMALLINT:
-            case TSDB_DATA_TYPE_TINYINT: {
-              maxOfThisSlot = pSeg->pBoundingEntries[j].iMaxVal;
-              minOfNextSlot = next.iMinVal;
-              break;
-            };
-            case TSDB_DATA_TYPE_FLOAT:
-            case TSDB_DATA_TYPE_DOUBLE: {
-              maxOfThisSlot = pSeg->pBoundingEntries[j].dMaxVal;
-              minOfNextSlot = next.dMinVal;
-              break;
-            };
-            case TSDB_DATA_TYPE_BIGINT: {
-              maxOfThisSlot = (double)pSeg->pBoundingEntries[j].i64MaxVal;
-              minOfNextSlot = (double)next.i64MinVal;
-              break;
-            }
-          };
-
-          assert(minOfNextSlot > maxOfThisSlot);
-
-          double val = (1 - fraction) * maxOfThisSlot + fraction * minOfNextSlot;
-          return val;
-        }
-        if (pSeg->pBuffer[j]->numOfAllElems <= pMemBucket->maxElemsCapacity) {
-          // data in buffer and file are merged together to be processed.
-          tFilePage *buffer = loadIntoBucketFromDisk(pMemBucket, i, j, pMemBucket->pOrderDesc);
-          int32_t    currentIdx = count - num;
-
-          char * thisVal = buffer->data + pMemBucket->nElemSize * currentIdx;
-          char * nextVal = thisVal + pMemBucket->nElemSize;
-          double td, nd;
-          switch (pMemBucket->dataType) {
-            case TSDB_DATA_TYPE_SMALLINT: {
-              td = *(int16_t *)thisVal;
-              nd = *(int16_t *)nextVal;
-              break;
-            }
-            case TSDB_DATA_TYPE_TINYINT: {
-              td = *(int8_t *)thisVal;
-              nd = *(int8_t *)nextVal;
-              break;
-            }
-            case TSDB_DATA_TYPE_INT: {
-              td = *(int32_t *)thisVal;
-              nd = *(int32_t *)nextVal;
-              break;
-            };
-            case TSDB_DATA_TYPE_FLOAT: {
-              //td = *(float *)thisVal;
-              //nd = *(float *)nextVal;
-              td = GET_FLOAT_VAL(thisVal);
-              nd = GET_FLOAT_VAL(nextVal);
-              break;
-            }
-            case TSDB_DATA_TYPE_DOUBLE: {
-              //td = *(double *)thisVal;
-              td = GET_DOUBLE_VAL(thisVal);
-              //nd = *(double *)nextVal;
-              nd = GET_DOUBLE_VAL(nextVal);
-              break;
-            }
-            case TSDB_DATA_TYPE_BIGINT: {
-              td = (double)*(int64_t *)thisVal;
-              nd = (double)*(int64_t *)nextVal;
-              break;
-            }
-          }
-          double val = (1 - fraction) * td + fraction * nd;
-          tfree(buffer);
-
-          return val;
-        } else {  // incur a second round bucket split
-          if (isIdenticalData(pMemBucket, i, j)) {
-            tExtMemBuffer *pMemBuffer = pSeg->pBuffer[j];
-
-            tFilePage *pPage = (tFilePage *)malloc(pMemBuffer->nPageSize);
-
-            char *thisVal = getFirstElemOfMemBuffer(pSeg, j, pPage);
-
-            double finalResult = 0.0;
-
-            switch (pMemBucket->dataType) {
-              case TSDB_DATA_TYPE_SMALLINT: {
-                finalResult = *(int16_t *)thisVal;
-                break;
-              }
-              case TSDB_DATA_TYPE_TINYINT: {
-                finalResult = *(int8_t *)thisVal;
-                break;
-              }
-              case TSDB_DATA_TYPE_INT: {
-                finalResult = *(int32_t *)thisVal;
-                break;
-              };
-              case TSDB_DATA_TYPE_FLOAT: {
-                //finalResult = *(float *)thisVal;
-                finalResult = GET_FLOAT_VAL(thisVal);
-                break;
-              }
-              case TSDB_DATA_TYPE_DOUBLE: {
-                //finalResult = *(double *)thisVal;
-                finalResult = GET_DOUBLE_VAL(thisVal);
-                break;
-              }
-              case TSDB_DATA_TYPE_BIGINT: {
-                finalResult = (double)(*(int64_t *)thisVal);
-                break;
-              }
-            }
-
-            free(pPage);
-            return finalResult;
-          }
-
-          pTrace("MemBucket:%p,start second round bucketing", pMemBucket);
-
-          if (pSeg->pBuffer[j]->numOfElemsInBuffer != 0) {
-            pTrace("MemBucket:%p,flush %d pages to disk, clear status", pMemBucket, pSeg->pBuffer[j]->numOfPagesInMem);
-
-            pMemBucket->numOfAvailPages += pSeg->pBuffer[j]->numOfPagesInMem;
-            tExtMemBufferFlush(pSeg->pBuffer[j]);
-          }
-
-          tExtMemBuffer *pMemBuffer = pSeg->pBuffer[j];
-          pSeg->pBuffer[j] = NULL;
-
-          // release all
-          for (int32_t tt = 0; tt < pMemBucket->numOfSegs; ++tt) {
-            tMemBucketSegment *pSeg = &pMemBucket->pSegs[tt];
-            for (int32_t ttx = 0; ttx < pSeg->numOfSlots; ++ttx) {
-              if (pSeg->pBuffer && pSeg->pBuffer[ttx]) {
-                tExtMemBufferDestroy(&pSeg->pBuffer[ttx]);
-              }
-            }
-          }
-
-          pMemBucket->nRange.i64MaxVal = pSeg->pBoundingEntries->i64MaxVal;
-          pMemBucket->nRange.i64MinVal = pSeg->pBoundingEntries->i64MinVal;
-          pMemBucket->numOfElems = 0;
-
-          for (int32_t tt = 0; tt < pMemBucket->numOfSegs; ++tt) {
-            tMemBucketSegment *pSeg = &pMemBucket->pSegs[tt];
-            for (int32_t ttx = 0; ttx < pSeg->numOfSlots; ++ttx) {
-              if (pSeg->pBoundingEntries) {
-                resetBoundingBox(pSeg, pMemBucket->dataType);
-              }
-            }
-          }
-
-          tFilePage *pPage = (tFilePage *)malloc(pMemBuffer->nPageSize);
-
-          tFlushoutInfo *pFlushInfo = &pMemBuffer->fileMeta.flushoutData.pFlushoutInfo[0];
-          assert(pFlushInfo->numOfPages == pMemBuffer->fileMeta.nFileSize);
-
-          int32_t ret = fseek(pMemBuffer->dataFile, pFlushInfo->startPageId * pMemBuffer->nPageSize, SEEK_SET);
-          UNUSED(ret);
-
-          for (uint32_t jx = 0; jx < pFlushInfo->numOfPages; ++jx) {
-            ret = fread(pPage, pMemBuffer->nPageSize, 1, pMemBuffer->dataFile);
-            UNUSED(ret);
-            tMemBucketPut(pMemBucket, pPage->data, pPage->numOfElems);
-          }
-
-          fclose(pMemBuffer->dataFile);
-          if (unlink(pMemBuffer->dataFilePath) != 0) {
-            pError("MemBucket:%p,remove tmp file %s failed", pMemBucket, pMemBuffer->dataFilePath);
-          }
-          tfree(pMemBuffer);
-          tfree(pPage);
-
-          return getPercentileImpl(pMemBucket, count - num, fraction);
-        }
-      } else {
-        num += pSeg->pBuffer[j]->numOfAllElems;
-      }
-    }
-  }
-  return 0;
-}
-
-double getPercentile(tMemBucket *pMemBucket, double percent) {
-  if (pMemBucket->numOfElems == 0) {
-    return 0.0;
-  }
-
-  if (pMemBucket->numOfElems == 1) {  // return the only element
-    return findOnlyResult(pMemBucket);
-  }
-
-  percent = fabs(percent);
-
-  // validate the parameters
-  if (fabs(percent - 100.0) < DBL_EPSILON || (percent < DBL_EPSILON)) {
-    double minx = 0, maxx = 0;
-    /*
-     * find the min/max value, no need to scan all data in bucket
-     */
-    findMaxMinValue(pMemBucket, &maxx, &minx);
-
-    return fabs(percent - 100) < DBL_EPSILON ? maxx : minx;
-  }
-
-  double  percentVal = (percent * (pMemBucket->numOfElems - 1)) / ((double)100.0);
-  int32_t orderIdx = (int32_t)percentVal;
-
-  // do put data by using buckets
-  return getPercentileImpl(pMemBucket, orderIdx, percentVal - orderIdx);
-}
-
-/*
- * check if data in one slot are all identical
- * only need to compare with the bounding box
- */
-bool isIdenticalData(tMemBucket *pMemBucket, int32_t segIdx, int32_t slotIdx) {
-  tMemBucketSegment *pSeg = &pMemBucket->pSegs[segIdx];
-
-  if (pMemBucket->dataType == TSDB_DATA_TYPE_INT || pMemBucket->dataType == TSDB_DATA_TYPE_BIGINT ||
-      pMemBucket->dataType == TSDB_DATA_TYPE_SMALLINT || pMemBucket->dataType == TSDB_DATA_TYPE_TINYINT) {
-    return pSeg->pBoundingEntries[slotIdx].i64MinVal == pSeg->pBoundingEntries[slotIdx].i64MaxVal;
-  }
-
-  if (pMemBucket->dataType == TSDB_DATA_TYPE_FLOAT || pMemBucket->dataType == TSDB_DATA_TYPE_DOUBLE) {
-    return fabs(pSeg->pBoundingEntries[slotIdx].dMaxVal - pSeg->pBoundingEntries[slotIdx].dMinVal) < DBL_EPSILON;
-  }
-
-  return false;
-}
-
-/*
- * get the first element of one slot into memory.
- * if no data of current slot in memory, load it from disk
- */
-char *getFirstElemOfMemBuffer(tMemBucketSegment *pSeg, int32_t slotIdx, tFilePage *pPage) {
-  tExtMemBuffer *pMemBuffer = pSeg->pBuffer[slotIdx];
-  char *         thisVal = NULL;
-
-  if (pSeg->pBuffer[slotIdx]->numOfElemsInBuffer != 0) {
-    thisVal = pSeg->pBuffer[slotIdx]->pHead->item.data;
-  } else {
-    /*
-     * no data in memory, load one page into memory
-     */
-    tFlushoutInfo *pFlushInfo = &pMemBuffer->fileMeta.flushoutData.pFlushoutInfo[0];
-    assert(pFlushInfo->numOfPages == pMemBuffer->fileMeta.nFileSize);
-
-    fseek(pMemBuffer->dataFile, pFlushInfo->startPageId * pMemBuffer->nPageSize, SEEK_SET);
-    size_t ret = fread(pPage, pMemBuffer->nPageSize, 1, pMemBuffer->dataFile);
-    UNUSED(ret);
-    thisVal = pPage->data;
-  }
-  return thisVal;
-}

src/util/src/tinterpolation.c

@@ -205,16 +205,18 @@ static char* getPos(char* data, int32_t bytes, int32_t order, int32_t capacity,
 //  }
 }
 
-static void setTagsValueInInterpolation(tFilePage** data, char** pTags, tColModel* pModel, int32_t order, int32_t start,
+static void setTagsValueInInterpolation(tFilePage** data, char** pTags, SColumnModel* pModel, int32_t order, int32_t start,
                                         int32_t capacity, int32_t num) {
   for (int32_t j = 0, i = start; i < pModel->numOfCols; ++i, ++j) {
-    char* val1 = getPos(data[i]->data, pModel->pFields[i].bytes, order, capacity, num);
-    assignVal(val1, pTags[j], pModel->pFields[i].bytes, pModel->pFields[i].type);
+    SSchema* pSchema = getColumnModelSchema(pModel, i);
+    
+    char* val1 = getPos(data[i]->data, pSchema->bytes, order, capacity, num);
+    assignVal(val1, pTags[j], pSchema->bytes, pSchema->type);
   }
 }
 
 static void doInterpoResultImpl(SInterpolationInfo* pInterpoInfo, int16_t interpoType, tFilePage** data,
-                                tColModel* pModel, int32_t* num, char** srcData, int64_t nInterval, int64_t* defaultVal,
+                                SColumnModel* pModel, int32_t* num, char** srcData, int64_t nInterval, int64_t* defaultVal,
                                 int64_t currentTimestamp, int32_t capacity, int32_t numOfTags, char** pTags,
                                 bool outOfBound) {
   char** prevValues = &pInterpoInfo->prevValues;
@@ -234,18 +236,23 @@ static void doInterpoResultImpl(SInterpolationInfo* pInterpoInfo, int16_t interp
     char* pInterpolationData = INTERPOL_IS_ASC_INTERPOL(pInterpoInfo) ? *prevValues : *nextValues;
     if (pInterpolationData != NULL) {
       for (int32_t i = 1; i < numOfValCols; ++i) {
-        char* val1 = getPos(data[i]->data, pModel->pFields[i].bytes, pInterpoInfo->order, capacity, *num);
+        SSchema* pSchema = getColumnModelSchema(pModel, i);
+        int16_t offset = getColumnModelOffset(pModel, i);
+        
+        char* val1 = getPos(data[i]->data, pSchema->bytes, pInterpoInfo->order, capacity, *num);
 
-        if (isNull(pInterpolationData + pModel->colOffset[i], pModel->pFields[i].type)) {
-          setNull(val1, pModel->pFields[i].type, pModel->pFields[i].bytes);
+        if (isNull(pInterpolationData + offset, pSchema->type)) {
+          setNull(val1, pSchema->type, pSchema->bytes);
         } else {
-          assignVal(val1, pInterpolationData + pModel->colOffset[i], pModel->pFields[i].bytes, pModel->pFields[i].type);
+          assignVal(val1, pInterpolationData + offset, pSchema->bytes, pSchema->type);
         }
       }
     } else { /* no prev value yet, set the value for null */
       for (int32_t i = 1; i < numOfValCols; ++i) {
-        char* val1 = getPos(data[i]->data, pModel->pFields[i].bytes, pInterpoInfo->order, capacity, *num);
-        setNull(val1, pModel->pFields[i].type, pModel->pFields[i].bytes);
+        SSchema* pSchema = getColumnModelSchema(pModel, i);
+  
+        char* val1 = getPos(data[i]->data, pSchema->bytes, pInterpoInfo->order, capacity, *num);
+        setNull(val1, pSchema->type, pSchema->bytes);
       }
     }
 
@@ -254,34 +261,41 @@ static void doInterpoResultImpl(SInterpolationInfo* pInterpoInfo, int16_t interp
     // TODO : linear interpolation supports NULL value
     if (*prevValues != NULL && !outOfBound) {
       for (int32_t i = 1; i < numOfValCols; ++i) {
-        int32_t type = pModel->pFields[i].type;
-        char*   val1 = getPos(data[i]->data, pModel->pFields[i].bytes, pInterpoInfo->order, capacity, *num);
+        SSchema* pSchema = getColumnModelSchema(pModel, i);
+        int16_t offset = getColumnModelOffset(pModel, i);
+        
+        int16_t type = pSchema->type;
+        char*   val1 = getPos(data[i]->data, pSchema->bytes, pInterpoInfo->order, capacity, *num);
 
         if (type == TSDB_DATA_TYPE_BINARY || type == TSDB_DATA_TYPE_NCHAR || type == TSDB_DATA_TYPE_BOOL) {
-          setNull(val1, pModel->pFields[i].type, pModel->pFields[i].bytes);
+          setNull(val1, type, pSchema->bytes);
           continue;
         }
 
-        point1 = (SPoint){.key = *(TSKEY*)(*prevValues), .val = *prevValues + pModel->colOffset[i]};
-        point2 = (SPoint){.key = currentTimestamp, .val = srcData[i] + pInterpoInfo->rowIdx * pModel->pFields[i].bytes};
+        point1 = (SPoint){.key = *(TSKEY*)(*prevValues), .val = *prevValues + offset};
+        point2 = (SPoint){.key = currentTimestamp, .val = srcData[i] + pInterpoInfo->rowIdx * pSchema->bytes};
         point = (SPoint){.key = pInterpoInfo->startTimestamp, .val = val1};
-        taosDoLinearInterpolation(pModel->pFields[i].type, &point1, &point2, &point);
+        taosDoLinearInterpolation(type, &point1, &point2, &point);
       }
 
       setTagsValueInInterpolation(data, pTags, pModel, pInterpoInfo->order, numOfValCols, capacity, *num);
 
     } else {
       for (int32_t i = 1; i < numOfValCols; ++i) {
-        char* val1 = getPos(data[i]->data, pModel->pFields[i].bytes, pInterpoInfo->order, capacity, *num);
-        setNull(val1, pModel->pFields[i].type, pModel->pFields[i].bytes);
+        SSchema* pSchema = getColumnModelSchema(pModel, i);
+        
+        char* val1 = getPos(data[i]->data, pSchema->bytes, pInterpoInfo->order, capacity, *num);
+        setNull(val1, pSchema->type, pSchema->bytes);
       }
 
       setTagsValueInInterpolation(data, pTags, pModel, pInterpoInfo->order, numOfValCols, capacity, *num);
     }
   } else { /* default value interpolation */
     for (int32_t i = 1; i < numOfValCols; ++i) {
-      char* val1 = getPos(data[i]->data, pModel->pFields[i].bytes, pInterpoInfo->order, capacity, *num);
-      assignVal(val1, (char*)&defaultVal[i], pModel->pFields[i].bytes, pModel->pFields[i].type);
+      SSchema* pSchema = getColumnModelSchema(pModel, i);
+  
+      char* val1 = getPos(data[i]->data, pSchema->bytes, pInterpoInfo->order, capacity, *num);
+      assignVal(val1, (char*)&defaultVal[i], pSchema->bytes, pSchema->type);
     }
 
     setTagsValueInInterpolation(data, pTags, pModel, pInterpoInfo->order, numOfValCols, capacity, *num);
@@ -295,7 +309,7 @@ static void doInterpoResultImpl(SInterpolationInfo* pInterpoInfo, int16_t interp
 
 int32_t taosDoInterpoResult(SInterpolationInfo* pInterpoInfo, int16_t interpoType, tFilePage** data,
                             int32_t numOfRawDataInRows, int32_t outputRows, int64_t nInterval,
-                            const int64_t* pPrimaryKeyArray, tColModel* pModel, char** srcData, int64_t* defaultVal,
+                            const int64_t* pPrimaryKeyArray, SColumnModel* pModel, char** srcData, int64_t* defaultVal,
                             const int32_t* functionIDs, int32_t bufSize) {
   int32_t num = 0;
   pInterpoInfo->numOfCurrentInterpo = 0;
@@ -328,17 +342,21 @@ int32_t taosDoInterpoResult(SInterpolationInfo* pInterpoInfo, int16_t interpoTyp
           (pInterpoInfo->startTimestamp > currentTimestamp && !INTERPOL_IS_ASC_INTERPOL(pInterpoInfo))) {
         /* set the next value for interpolation */
         if (*nextValues == NULL) {
-          *nextValues =
-              calloc(1, pModel->colOffset[pModel->numOfCols - 1] + pModel->pFields[pModel->numOfCols - 1].bytes);
+          *nextValues = calloc(1, pModel->rowSize);
           for (int i = 1; i < pModel->numOfCols; i++) {
-            setNull(*nextValues + pModel->colOffset[i], pModel->pFields[i].type, pModel->pFields[i].bytes);
+            int16_t offset = getColumnModelOffset(pModel, i);
+            SSchema* pSchema = getColumnModelSchema(pModel, i);
+            
+            setNull(*nextValues + offset, pSchema->type, pSchema->bytes);
           }
         }
 
         int32_t offset = pInterpoInfo->rowIdx;
         for (int32_t tlen = 0, i = 0; i < pModel->numOfCols - numOfTags; ++i) {
-          memcpy(*nextValues + tlen, srcData[i] + offset * pModel->pFields[i].bytes, pModel->pFields[i].bytes);
-          tlen += pModel->pFields[i].bytes;
+          SSchema* pSchema = getColumnModelSchema(pModel, i);
+          
+          memcpy(*nextValues + tlen, srcData[i] + offset * pSchema->bytes, pSchema->bytes);
+          tlen += pSchema->bytes;
         }
       }
 
@@ -358,37 +376,41 @@ int32_t taosDoInterpoResult(SInterpolationInfo* pInterpoInfo, int16_t interpoTyp
 
       if (pInterpoInfo->startTimestamp == currentTimestamp) {
         if (*prevValues == NULL) {
-          *prevValues =
-              calloc(1, pModel->colOffset[pModel->numOfCols - 1] + pModel->pFields[pModel->numOfCols - 1].bytes);
+          *prevValues = calloc(1, pModel->rowSize);
           for (int i = 1; i < pModel->numOfCols; i++) {
-            setNull(*prevValues + pModel->colOffset[i], pModel->pFields[i].type, pModel->pFields[i].bytes);
+            int16_t offset = getColumnModelOffset(pModel, i);
+            SSchema* pSchema = getColumnModelSchema(pModel, i);
+            
+            setNull(*prevValues + offset, pSchema->type, pSchema->bytes);
           }
         }
 
         // assign rows to dst buffer
         int32_t i = 0;
         for (int32_t tlen = 0; i < pModel->numOfCols - numOfTags; ++i) {
-          char* val1 = getPos(data[i]->data, pModel->pFields[i].bytes, pInterpoInfo->order, bufSize, num);
+          int16_t offset = getColumnModelOffset(pModel, i);
+          SSchema* pSchema = getColumnModelSchema(pModel, i);
+          
+          char* val1 = getPos(data[i]->data, pSchema->bytes, pInterpoInfo->order, bufSize, num);
 
           if (i == 0 ||
               (functionIDs[i] != TSDB_FUNC_COUNT &&
-               !isNull(srcData[i] + pInterpoInfo->rowIdx * pModel->pFields[i].bytes, pModel->pFields[i].type)) ||
+               !isNull(srcData[i] + pInterpoInfo->rowIdx * pSchema->bytes, pSchema->type)) ||
               (functionIDs[i] == TSDB_FUNC_COUNT &&
-               *(int64_t*)(srcData[i] + pInterpoInfo->rowIdx * pModel->pFields[i].bytes) != 0)) {
-            assignVal(val1, srcData[i] + pInterpoInfo->rowIdx * pModel->pFields[i].bytes, pModel->pFields[i].bytes,
-                      pModel->pFields[i].type);
-            memcpy(*prevValues + tlen, srcData[i] + pInterpoInfo->rowIdx * pModel->pFields[i].bytes,
-                   pModel->pFields[i].bytes);
+               *(int64_t*)(srcData[i] + pInterpoInfo->rowIdx * pSchema->bytes) != 0)) {
+            
+            assignVal(val1, srcData[i] + pInterpoInfo->rowIdx * pSchema->bytes, pSchema->bytes, pSchema->type);
+            memcpy(*prevValues + tlen, srcData[i] + pInterpoInfo->rowIdx * pSchema->bytes, pSchema->bytes);
           } else {  // i > 0 and isNULL, do interpolation
             if (interpoType == TSDB_INTERPO_PREV) {
-              assignVal(val1, *prevValues + pModel->colOffset[i], pModel->pFields[i].bytes, pModel->pFields[i].type);
+              assignVal(val1, *prevValues + offset, pSchema->bytes, pSchema->type);
             } else if (interpoType == TSDB_INTERPO_LINEAR) {
               // TODO:
             } else {
-              assignVal(val1, (char*)&defaultVal[i], pModel->pFields[i].bytes, pModel->pFields[i].type);
+              assignVal(val1, (char*)&defaultVal[i], pSchema->bytes, pSchema->type);
             }
           }
-          tlen += pModel->pFields[i].bytes;
+          tlen += pSchema->bytes;
         }
 
         /* set the tag value for final result */

src/util/src/tpercentile.c

+/*
+ * 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 "os.h"
+
+#include "taosmsg.h"
+#include "tsdb.h"
+#include "tlog.h"
+#include "ttypes.h"
+#include "tpercentile.h"
+
+tExtMemBuffer *releaseBucketsExceptFor(tMemBucket *pMemBucket, int16_t segIdx, int16_t slotIdx) {
+  tExtMemBuffer *pBuffer = NULL;
+  
+  for (int32_t i = 0; i < pMemBucket->numOfSegs; ++i) {
+    tMemBucketSegment *pSeg = &pMemBucket->pSegs[i];
+    
+    for (int32_t j = 0; j < pSeg->numOfSlots; ++j) {
+      if (i == segIdx && j == slotIdx) {
+        pBuffer = pSeg->pBuffer[j];
+      } else {
+        if (pSeg->pBuffer && pSeg->pBuffer[j]) {
+          pSeg->pBuffer[j] = destoryExtMemBuffer(pSeg->pBuffer[j]);
+        }
+      }
+    }
+  }
+  
+  return pBuffer;
+}
+
+static tFilePage *loadIntoBucketFromDisk(tMemBucket *pMemBucket, int32_t segIdx, int32_t slotIdx,
+                                         tOrderDescriptor *pDesc) {
+  // release all data in other slots
+  tExtMemBuffer *pMemBuffer = pMemBucket->pSegs[segIdx].pBuffer[slotIdx];
+  tFilePage *    buffer = (tFilePage *)calloc(1, pMemBuffer->nElemSize * pMemBuffer->numOfTotalElems + sizeof(tFilePage));
+  int32_t        oldCapacity = pDesc->pColumnModel->capacity;
+  pDesc->pColumnModel->capacity = pMemBuffer->numOfTotalElems;
+  
+  if (!tExtMemBufferIsAllDataInMem(pMemBuffer)) {
+    pMemBuffer = releaseBucketsExceptFor(pMemBucket, segIdx, slotIdx);
+    assert(pMemBuffer->numOfTotalElems > 0);
+    
+    // load data in disk to memory
+    tFilePage *pPage = (tFilePage *)calloc(1, pMemBuffer->pageSize);
+    
+    for (int32_t i = 0; i < pMemBuffer->fileMeta.flushoutData.nLength; ++i) {
+      tFlushoutInfo *pFlushInfo = &pMemBuffer->fileMeta.flushoutData.pFlushoutInfo[i];
+      
+      int32_t ret = fseek(pMemBuffer->file, pFlushInfo->startPageId * pMemBuffer->pageSize, SEEK_SET);
+      UNUSED(ret);
+      
+      for (uint32_t j = 0; j < pFlushInfo->numOfPages; ++j) {
+        ret = fread(pPage, pMemBuffer->pageSize, 1, pMemBuffer->file);
+        UNUSED(ret);
+        assert(pPage->numOfElems > 0);
+        
+        tColModelAppend(pDesc->pColumnModel, buffer, pPage->data, 0, pPage->numOfElems, pPage->numOfElems);
+        printf("id: %d  count: %" PRIu64 "\n", j, buffer->numOfElems);
+      }
+    }
+    tfree(pPage);
+    
+    assert(buffer->numOfElems == pMemBuffer->fileMeta.numOfElemsInFile);
+  }
+  
+  // load data in pMemBuffer to buffer
+  tFilePagesItem *pListItem = pMemBuffer->pHead;
+  while (pListItem != NULL) {
+    tColModelAppend(pDesc->pColumnModel, buffer, pListItem->item.data, 0, pListItem->item.numOfElems,
+                    pListItem->item.numOfElems);
+    pListItem = pListItem->pNext;
+  }
+  
+  tColDataQSort(pDesc, buffer->numOfElems, 0, buffer->numOfElems - 1, buffer->data, TSQL_SO_ASC);
+  
+  pDesc->pColumnModel->capacity = oldCapacity;  // restore value
+  return buffer;
+}
+
+double findOnlyResult(tMemBucket *pMemBucket) {
+  assert(pMemBucket->numOfElems == 1);
+  
+  for (int32_t i = 0; i < pMemBucket->numOfSegs; ++i) {
+    tMemBucketSegment *pSeg = &pMemBucket->pSegs[i];
+    if (pSeg->pBuffer) {
+      for (int32_t j = 0; j < pSeg->numOfSlots; ++j) {
+        tExtMemBuffer *pBuffer = pSeg->pBuffer[j];
+        if (pBuffer) {
+          assert(pBuffer->numOfTotalElems == 1);
+          tFilePage *pPage = &pBuffer->pHead->item;
+          if (pBuffer->numOfElemsInBuffer == 1) {
+            switch (pMemBucket->dataType) {
+              case TSDB_DATA_TYPE_INT:
+                return *(int32_t *)pPage->data;
+              case TSDB_DATA_TYPE_SMALLINT:
+                return *(int16_t *)pPage->data;
+              case TSDB_DATA_TYPE_TINYINT:
+                return *(int8_t *)pPage->data;
+              case TSDB_DATA_TYPE_BIGINT:
+                return (double)(*(int64_t *)pPage->data);
+              case TSDB_DATA_TYPE_DOUBLE: {
+                double dv = GET_DOUBLE_VAL(pPage->data);
+                //return *(double *)pPage->data;
+                return dv;
+              }
+              case TSDB_DATA_TYPE_FLOAT: {
+                float fv = GET_FLOAT_VAL(pPage->data);
+                //return *(float *)pPage->data;
+                return fv;
+              }
+              default:
+                return 0;
+            }
+          }
+        }
+      }
+    }
+  }
+  return 0;
+}
+
+void tBucketBigIntHash(tMemBucket *pBucket, void *value, int16_t *segIdx, int16_t *slotIdx) {
+  int64_t v = *(int64_t *)value;
+
+  if (pBucket->nRange.i64MaxVal == INT64_MIN) {
+    if (v >= 0) {
+      *segIdx = ((v >> (64 - 9)) >> 6) + 8;
+      *slotIdx = (v >> (64 - 9)) & 0x3F;
+    } else {  // v<0
+      *segIdx = ((-v) >> (64 - 9)) >> 6;
+      *slotIdx = ((-v) >> (64 - 9)) & 0x3F;
+      *segIdx = 7 - (*segIdx);
+    }
+  } else {
+    // todo hash for bigint and float and double
+    int64_t span = pBucket->nRange.i64MaxVal - pBucket->nRange.i64MinVal;
+    if (span < pBucket->nTotalSlots) {
+      int32_t delta = (int32_t)(v - pBucket->nRange.i64MinVal);
+      *segIdx = delta / pBucket->nSlotsOfSeg;
+      *slotIdx = delta % pBucket->nSlotsOfSeg;
+    } else {
+      double x = (double)span / pBucket->nTotalSlots;
+      double posx = (v - pBucket->nRange.i64MinVal) / x;
+      if (v == pBucket->nRange.i64MaxVal) {
+        posx -= 1;
+      }
+
+      *segIdx = ((int32_t)posx) / pBucket->nSlotsOfSeg;
+      *slotIdx = ((int32_t)posx) % pBucket->nSlotsOfSeg;
+    }
+  }
+}
+
+// todo refactor to more generic
+void tBucketIntHash(tMemBucket *pBucket, void *value, int16_t *segIdx, int16_t *slotIdx) {
+  int32_t v = *(int32_t *)value;
+
+  if (pBucket->nRange.iMaxVal == INT32_MIN) {
+    /*
+     * taking negative integer into consideration,
+     * there is only half of pBucket->segs available for non-negative integer
+     */
+    //        int32_t numOfSlots = pBucket->nTotalSlots>>1;
+    //        int32_t bits = bitsOfNumber(numOfSlots)-1;
+
+    if (v >= 0) {
+      *segIdx = ((v >> (32 - 9)) >> 6) + 8;
+      *slotIdx = (v >> (32 - 9)) & 0x3F;
+    } else {  // v<0
+      *segIdx = ((-v) >> (32 - 9)) >> 6;
+      *slotIdx = ((-v) >> (32 - 9)) & 0x3F;
+      *segIdx = 7 - (*segIdx);
+    }
+  } else {
+    // divide a range of [iMinVal, iMaxVal] into 1024 buckets
+    int32_t span = pBucket->nRange.iMaxVal - pBucket->nRange.iMinVal;
+    if (span < pBucket->nTotalSlots) {
+      int32_t delta = v - pBucket->nRange.iMinVal;
+      *segIdx = delta / pBucket->nSlotsOfSeg;
+      *slotIdx = delta % pBucket->nSlotsOfSeg;
+    } else {
+      double x = (double)span / pBucket->nTotalSlots;
+      double posx = (v - pBucket->nRange.iMinVal) / x;
+      if (v == pBucket->nRange.iMaxVal) {
+        posx -= 1;
+      }
+      *segIdx = ((int32_t)posx) / pBucket->nSlotsOfSeg;
+      *slotIdx = ((int32_t)posx) % pBucket->nSlotsOfSeg;
+    }
+  }
+}
+
+void tBucketDoubleHash(tMemBucket *pBucket, void *value, int16_t *segIdx, int16_t *slotIdx) {
+  // double v = *(double *)value;
+  double v = GET_DOUBLE_VAL(value);
+
+  if (pBucket->nRange.dMinVal == DBL_MAX) {
+    /*
+     * taking negative integer into consideration,
+     * there is only half of pBucket->segs available for non-negative integer
+     */
+    double x = DBL_MAX / (pBucket->nTotalSlots >> 1);
+    double posx = (v + DBL_MAX) / x;
+    *segIdx = ((int32_t)posx) / pBucket->nSlotsOfSeg;
+    *slotIdx = ((int32_t)posx) % pBucket->nSlotsOfSeg;
+  } else {
+    // divide a range of [dMinVal, dMaxVal] into 1024 buckets
+    double span = pBucket->nRange.dMaxVal - pBucket->nRange.dMinVal;
+    if (span < pBucket->nTotalSlots) {
+      int32_t delta = (int32_t)(v - pBucket->nRange.dMinVal);
+      *segIdx = delta / pBucket->nSlotsOfSeg;
+      *slotIdx = delta % pBucket->nSlotsOfSeg;
+    } else {
+      double x = span / pBucket->nTotalSlots;
+      double posx = (v - pBucket->nRange.dMinVal) / x;
+      if (v == pBucket->nRange.dMaxVal) {
+        posx -= 1;
+      }
+      *segIdx = ((int32_t)posx) / pBucket->nSlotsOfSeg;
+      *slotIdx = ((int32_t)posx) % pBucket->nSlotsOfSeg;
+    }
+
+    if (*segIdx < 0 || *segIdx > 16 || *slotIdx < 0 || *slotIdx > 64) {
+      pError("error in hash process. segment is: %d, slot id is: %d\n", *segIdx, *slotIdx);
+    }
+  }
+}
+
+tMemBucket *tMemBucketCreate(int32_t totalSlots, int32_t nBufferSize, int16_t nElemSize, int16_t dataType,
+                             tOrderDescriptor *pDesc) {
+  tMemBucket *pBucket = (tMemBucket *)malloc(sizeof(tMemBucket));
+
+  pBucket->nTotalSlots = totalSlots;
+  pBucket->nSlotsOfSeg = 1 << 6;  // 64 Segments, 16 slots each seg.
+  pBucket->dataType = dataType;
+  pBucket->nElemSize = nElemSize;
+  pBucket->pageSize = DEFAULT_PAGE_SIZE;
+
+  pBucket->numOfElems = 0;
+  pBucket->numOfSegs = pBucket->nTotalSlots / pBucket->nSlotsOfSeg;
+
+  pBucket->nTotalBufferSize = nBufferSize;
+
+  pBucket->maxElemsCapacity = pBucket->nTotalBufferSize / pBucket->nElemSize;
+
+  pBucket->numOfTotalPages = pBucket->nTotalBufferSize / pBucket->pageSize;
+  pBucket->numOfAvailPages = pBucket->numOfTotalPages;
+
+  pBucket->pOrderDesc = pDesc;
+
+  switch (pBucket->dataType) {
+    case TSDB_DATA_TYPE_INT:
+    case TSDB_DATA_TYPE_SMALLINT:
+    case TSDB_DATA_TYPE_TINYINT: {
+      pBucket->nRange.iMinVal = INT32_MAX;
+      pBucket->nRange.iMaxVal = INT32_MIN;
+      pBucket->HashFunc = tBucketIntHash;
+      break;
+    };
+    case TSDB_DATA_TYPE_DOUBLE:
+    case TSDB_DATA_TYPE_FLOAT: {
+      pBucket->nRange.dMinVal = DBL_MAX;
+      pBucket->nRange.dMaxVal = -DBL_MAX;
+      pBucket->HashFunc = tBucketDoubleHash;
+      break;
+    };
+    case TSDB_DATA_TYPE_BIGINT: {
+      pBucket->nRange.i64MinVal = INT64_MAX;
+      pBucket->nRange.i64MaxVal = INT64_MIN;
+      pBucket->HashFunc = tBucketBigIntHash;
+      break;
+    };
+    default: {
+      pError("MemBucket:%p,not support data type %d,failed", *pBucket, pBucket->dataType);
+      tfree(pBucket);
+      return NULL;
+    }
+  }
+
+  int32_t numOfCols = pDesc->pColumnModel->numOfCols;
+  if (numOfCols != 1) {
+    pError("MemBucket:%p,only consecutive data is allowed,invalid numOfCols:%d", pBucket, numOfCols);
+    tfree(pBucket);
+    return NULL;
+  }
+
+  SSchema* pSchema = getColumnModelSchema(pDesc->pColumnModel, 0);
+  if (pSchema->type != dataType) {
+    pError("MemBucket:%p,data type is not consistent,%d in schema, %d in param", pBucket, pSchema->type, dataType);
+    tfree(pBucket);
+    return NULL;
+  }
+
+  if (pBucket->numOfTotalPages < pBucket->nTotalSlots) {
+    pWarn("MemBucket:%p,total buffer pages %d are not enough for all slots", pBucket, pBucket->numOfTotalPages);
+  }
+
+  pBucket->pSegs = (tMemBucketSegment *)malloc(pBucket->numOfSegs * sizeof(tMemBucketSegment));
+
+  for (int32_t i = 0; i < pBucket->numOfSegs; ++i) {
+    pBucket->pSegs[i].numOfSlots = pBucket->nSlotsOfSeg;
+    pBucket->pSegs[i].pBuffer = NULL;
+    pBucket->pSegs[i].pBoundingEntries = NULL;
+  }
+
+  pTrace("MemBucket:%p,created,buffer size:%d,elem size:%d", pBucket, pBucket->numOfTotalPages * DEFAULT_PAGE_SIZE,
+         pBucket->nElemSize);
+
+  return pBucket;
+}
+
+void tMemBucketDestroy(tMemBucket *pBucket) {
+  if (pBucket == NULL) {
+    return;
+  }
+
+  if (pBucket->pSegs) {
+    for (int32_t i = 0; i < pBucket->numOfSegs; ++i) {
+      tMemBucketSegment *pSeg = &(pBucket->pSegs[i]);
+      tfree(pSeg->pBoundingEntries);
+
+      if (pSeg->pBuffer == NULL || pSeg->numOfSlots == 0) {
+        continue;
+      }
+
+      for (int32_t j = 0; j < pSeg->numOfSlots; ++j) {
+        if (pSeg->pBuffer[j] != NULL) {
+          pSeg->pBuffer[j] = destoryExtMemBuffer(pSeg->pBuffer[j]);
+        }
+      }
+      tfree(pSeg->pBuffer);
+    }
+  }
+
+  tfree(pBucket->pSegs);
+  tfree(pBucket);
+}
+
+/*
+ * find the slots which accounts for largest proportion of total in-memory buffer
+ */
+static void tBucketGetMaxMemSlot(tMemBucket *pBucket, int16_t *segIdx, int16_t *slotIdx) {
+  *segIdx = -1;
+  *slotIdx = -1;
+
+  int32_t val = 0;
+  for (int32_t k = 0; k < pBucket->numOfSegs; ++k) {
+    tMemBucketSegment *pSeg = &pBucket->pSegs[k];
+    for (int32_t i = 0; i < pSeg->numOfSlots; ++i) {
+      if (pSeg->pBuffer == NULL || pSeg->pBuffer[i] == NULL) {
+        continue;
+      }
+
+      if (val < pSeg->pBuffer[i]->numOfInMemPages) {
+        val = pSeg->pBuffer[i]->numOfInMemPages;
+        *segIdx = k;
+        *slotIdx = i;
+      }
+    }
+  }
+}
+
+static void resetBoundingBox(tMemBucketSegment *pSeg, int32_t type) {
+  switch (type) {
+    case TSDB_DATA_TYPE_BIGINT: {
+      for (int32_t i = 0; i < pSeg->numOfSlots; ++i) {
+        pSeg->pBoundingEntries[i].i64MaxVal = INT64_MIN;
+        pSeg->pBoundingEntries[i].i64MinVal = INT64_MAX;
+      }
+      break;
+    };
+    case TSDB_DATA_TYPE_INT:
+    case TSDB_DATA_TYPE_SMALLINT:
+    case TSDB_DATA_TYPE_TINYINT: {
+      for (int32_t i = 0; i < pSeg->numOfSlots; ++i) {
+        pSeg->pBoundingEntries[i].iMaxVal = INT32_MIN;
+        pSeg->pBoundingEntries[i].iMinVal = INT32_MAX;
+      }
+      break;
+    };
+    case TSDB_DATA_TYPE_DOUBLE:
+    case TSDB_DATA_TYPE_FLOAT: {
+      for (int32_t i = 0; i < pSeg->numOfSlots; ++i) {
+        pSeg->pBoundingEntries[i].dMaxVal = -DBL_MAX;
+        pSeg->pBoundingEntries[i].dMinVal = DBL_MAX;
+      }
+      break;
+    }
+  }
+}
+
+void tMemBucketUpdateBoundingBox(MinMaxEntry *r, char *data, int32_t dataType) {
+  switch (dataType) {
+    case TSDB_DATA_TYPE_INT: {
+      int32_t val = *(int32_t *)data;
+      if (r->iMinVal > val) {
+        r->iMinVal = val;
+      }
+
+      if (r->iMaxVal < val) {
+        r->iMaxVal = val;
+      }
+      break;
+    };
+    case TSDB_DATA_TYPE_BIGINT: {
+      int64_t val = *(int64_t *)data;
+      if (r->i64MinVal > val) {
+        r->i64MinVal = val;
+      }
+
+      if (r->i64MaxVal < val) {
+        r->i64MaxVal = val;
+      }
+      break;
+    };
+    case TSDB_DATA_TYPE_SMALLINT: {
+      int32_t val = *(int16_t *)data;
+      if (r->iMinVal > val) {
+        r->iMinVal = val;
+      }
+
+      if (r->iMaxVal < val) {
+        r->iMaxVal = val;
+      }
+      break;
+    };
+    case TSDB_DATA_TYPE_TINYINT: {
+      int32_t val = *(int8_t *)data;
+      if (r->iMinVal > val) {
+        r->iMinVal = val;
+      }
+
+      if (r->iMaxVal < val) {
+        r->iMaxVal = val;
+      }
+
+      break;
+    };
+    case TSDB_DATA_TYPE_DOUBLE: {
+      // double val = *(double *)data;
+      double val = GET_DOUBLE_VAL(data);
+      if (r->dMinVal > val) {
+        r->dMinVal = val;
+      }
+
+      if (r->dMaxVal < val) {
+        r->dMaxVal = val;
+      }
+      break;
+    };
+    case TSDB_DATA_TYPE_FLOAT: {
+      // double val = *(float *)data;
+      double val = GET_FLOAT_VAL(data);
+
+      if (r->dMinVal > val) {
+        r->dMinVal = val;
+      }
+
+      if (r->dMaxVal < val) {
+        r->dMaxVal = val;
+      }
+      break;
+    };
+    default: { assert(false); }
+  }
+}
+
+/*
+ * in memory bucket, we only accept the simple data consecutive put in a row/column
+ * no column-model in this case.
+ */
+void tMemBucketPut(tMemBucket *pBucket, void *data, int32_t numOfRows) {
+  pBucket->numOfElems += numOfRows;
+  int16_t segIdx = 0, slotIdx = 0;
+
+  for (int32_t i = 0; i < numOfRows; ++i) {
+    char *d = (char *)data + i * tDataTypeDesc[pBucket->dataType].nSize;
+
+    switch (pBucket->dataType) {
+      case TSDB_DATA_TYPE_SMALLINT: {
+        int32_t val = *(int16_t *)d;
+        (pBucket->HashFunc)(pBucket, &val, &segIdx, &slotIdx);
+        break;
+      }
+      case TSDB_DATA_TYPE_TINYINT: {
+        int32_t val = *(int8_t *)d;
+        (pBucket->HashFunc)(pBucket, &val, &segIdx, &slotIdx);
+        break;
+      }
+      case TSDB_DATA_TYPE_INT: {
+        int32_t val = *(int32_t *)d;
+        (pBucket->HashFunc)(pBucket, &val, &segIdx, &slotIdx);
+        break;
+      }
+      case TSDB_DATA_TYPE_BIGINT: {
+        int64_t val = *(int64_t *)d;
+        (pBucket->HashFunc)(pBucket, &val, &segIdx, &slotIdx);
+        break;
+      }
+      case TSDB_DATA_TYPE_DOUBLE: {
+        // double val = *(double *)d;
+        double val = GET_DOUBLE_VAL(d);
+        (pBucket->HashFunc)(pBucket, &val, &segIdx, &slotIdx);
+        break;
+      }
+      case TSDB_DATA_TYPE_FLOAT: {
+        // double val = *(float *)d;
+        double val = GET_FLOAT_VAL(d);
+        (pBucket->HashFunc)(pBucket, &val, &segIdx, &slotIdx);
+        break;
+      }
+    }
+
+    tMemBucketSegment *pSeg = &pBucket->pSegs[segIdx];
+    if (pSeg->pBoundingEntries == NULL) {
+      pSeg->pBoundingEntries = (MinMaxEntry *)malloc(sizeof(MinMaxEntry) * pBucket->nSlotsOfSeg);
+      resetBoundingBox(pSeg, pBucket->dataType);
+    }
+
+    if (pSeg->pBuffer == NULL) {
+      pSeg->pBuffer = (tExtMemBuffer **)calloc(pBucket->nSlotsOfSeg, sizeof(void *));
+    }
+
+    if (pSeg->pBuffer[slotIdx] == NULL) {
+      pSeg->pBuffer[slotIdx] = createExtMemBuffer(pBucket->numOfTotalPages * pBucket->pageSize, pBucket->nElemSize,
+                                                  pBucket->pOrderDesc->pColumnModel);
+      pSeg->pBuffer[slotIdx]->flushModel = SINGLE_APPEND_MODEL;
+      pBucket->pOrderDesc->pColumnModel->capacity = pSeg->pBuffer[slotIdx]->numOfElemsPerPage;
+    }
+
+    tMemBucketUpdateBoundingBox(&pSeg->pBoundingEntries[slotIdx], d, pBucket->dataType);
+
+    // ensure available memory pages to allocate
+    int16_t cseg = 0, cslot = 0;
+    if (pBucket->numOfAvailPages == 0) {
+      pTrace("MemBucket:%p,max avail size:%d, no avail memory pages,", pBucket, pBucket->numOfTotalPages);
+
+      tBucketGetMaxMemSlot(pBucket, &cseg, &cslot);
+      if (cseg == -1 || cslot == -1) {
+        pError("MemBucket:%p,failed to find appropriated avail buffer", pBucket);
+        return;
+      }
+
+      if (cseg != segIdx || cslot != slotIdx) {
+        pBucket->numOfAvailPages += pBucket->pSegs[cseg].pBuffer[cslot]->numOfInMemPages;
+
+        int32_t avail = pBucket->pSegs[cseg].pBuffer[cslot]->numOfInMemPages;
+        UNUSED(avail);
+        tExtMemBufferFlush(pBucket->pSegs[cseg].pBuffer[cslot]);
+
+        pTrace("MemBucket:%p,seg:%d,slot:%d flushed to disk,new avail pages:%d", pBucket, cseg, cslot,
+               pBucket->numOfAvailPages);
+      } else {
+        pTrace("MemBucket:%p,failed to choose slot to flush to disk seg:%d,slot:%d", pBucket, cseg, cslot);
+      }
+    }
+    int16_t consumedPgs = pSeg->pBuffer[slotIdx]->numOfInMemPages;
+
+    int16_t newPgs = tExtMemBufferPut(pSeg->pBuffer[slotIdx], d, 1);
+    /*
+     * trigger 1. page re-allocation, to reduce the available pages
+     *         2. page flushout, to increase the available pages
+     */
+    pBucket->numOfAvailPages += (consumedPgs - newPgs);
+  }
+}
+
+void releaseBucket(tMemBucket *pMemBucket, int32_t segIdx, int32_t slotIdx) {
+  if (segIdx < 0 || segIdx > pMemBucket->numOfSegs || slotIdx < 0) {
+    return;
+  }
+
+  tMemBucketSegment *pSeg = &pMemBucket->pSegs[segIdx];
+  if (slotIdx < 0 || slotIdx >= pSeg->numOfSlots || pSeg->pBuffer[slotIdx] == NULL) {
+    return;
+  }
+
+  pSeg->pBuffer[slotIdx] = destoryExtMemBuffer(pSeg->pBuffer[slotIdx]);
+}
+
+////////////////////////////////////////////////////////////////////////////////////////////
+static void findMaxMinValue(tMemBucket *pMemBucket, double *maxVal, double *minVal) {
+  *minVal = DBL_MAX;
+  *maxVal = -DBL_MAX;
+
+  for (int32_t i = 0; i < pMemBucket->numOfSegs; ++i) {
+    tMemBucketSegment *pSeg = &pMemBucket->pSegs[i];
+    if (pSeg->pBuffer == NULL) {
+      continue;
+    }
+    switch (pMemBucket->dataType) {
+      case TSDB_DATA_TYPE_INT:
+      case TSDB_DATA_TYPE_SMALLINT:
+      case TSDB_DATA_TYPE_TINYINT: {
+        for (int32_t j = 0; j < pSeg->numOfSlots; ++j) {
+          double minv = pSeg->pBoundingEntries[j].iMinVal;
+          double maxv = pSeg->pBoundingEntries[j].iMaxVal;
+
+          if (*minVal > minv) {
+            *minVal = minv;
+          }
+          if (*maxVal < maxv) {
+            *maxVal = maxv;
+          }
+        }
+        break;
+      }
+      case TSDB_DATA_TYPE_DOUBLE:
+      case TSDB_DATA_TYPE_FLOAT: {
+        for (int32_t j = 0; j < pSeg->numOfSlots; ++j) {
+          double minv = pSeg->pBoundingEntries[j].dMinVal;
+          double maxv = pSeg->pBoundingEntries[j].dMaxVal;
+
+          if (*minVal > minv) {
+            *minVal = minv;
+          }
+          if (*maxVal < maxv) {
+            *maxVal = maxv;
+          }
+        }
+        break;
+      }
+      case TSDB_DATA_TYPE_BIGINT: {
+        for (int32_t j = 0; j < pSeg->numOfSlots; ++j) {
+          double minv = (double)pSeg->pBoundingEntries[j].i64MinVal;
+          double maxv = (double)pSeg->pBoundingEntries[j].i64MaxVal;
+
+          if (*minVal > minv) {
+            *minVal = minv;
+          }
+          if (*maxVal < maxv) {
+            *maxVal = maxv;
+          }
+        }
+        break;
+      }
+    }
+  }
+}
+
+static MinMaxEntry getMinMaxEntryOfNearestSlotInNextSegment(tMemBucket *pMemBucket, int32_t segIdx) {
+  int32_t i = segIdx + 1;
+  while (i < pMemBucket->numOfSegs && pMemBucket->pSegs[i].numOfSlots == 0) ++i;
+
+  tMemBucketSegment *pSeg = &pMemBucket->pSegs[i];
+  assert(pMemBucket->numOfSegs > i && pMemBucket->pSegs[i].pBuffer != NULL);
+
+  i = 0;
+  while (i < pMemBucket->nSlotsOfSeg && pSeg->pBuffer[i] == NULL) ++i;
+
+  assert(i < pMemBucket->nSlotsOfSeg);
+  return pSeg->pBoundingEntries[i];
+}
+
+/*
+ *
+ * now, we need to find the minimum value of the next slot for
+ * interpolating the percentile value
+ * j is the last slot of current segment, we need to get the first
+ * slot of the next segment.
+ */
+static MinMaxEntry getMinMaxEntryOfNextSlotWithData(tMemBucket *pMemBucket, int32_t segIdx, int32_t slotIdx) {
+  tMemBucketSegment *pSeg = &pMemBucket->pSegs[segIdx];
+
+  MinMaxEntry next;
+  if (slotIdx == pSeg->numOfSlots - 1) {  // find next segment with data
+    return getMinMaxEntryOfNearestSlotInNextSegment(pMemBucket, segIdx);
+  } else {
+    int32_t j = slotIdx + 1;
+    for (; j < pMemBucket->nSlotsOfSeg && pMemBucket->pSegs[segIdx].pBuffer[j] == 0; ++j) {
+    };
+
+    if (j == pMemBucket->nSlotsOfSeg) {  // current slot has no available
+      // slot,try next segment
+      return getMinMaxEntryOfNearestSlotInNextSegment(pMemBucket, segIdx);
+    } else {
+      next = pSeg->pBoundingEntries[slotIdx + 1];
+      assert(pSeg->pBuffer[slotIdx + 1] != NULL);
+    }
+  }
+
+  return next;
+}
+
+bool  isIdenticalData(tMemBucket *pMemBucket, int32_t segIdx, int32_t slotIdx);
+char *getFirstElemOfMemBuffer(tMemBucketSegment *pSeg, int32_t slotIdx, tFilePage *pPage);
+
+double getPercentileImpl(tMemBucket *pMemBucket, int32_t count, double fraction) {
+  int32_t num = 0;
+
+  for (int32_t i = 0; i < pMemBucket->numOfSegs; ++i) {
+    tMemBucketSegment *pSeg = &pMemBucket->pSegs[i];
+    for (int32_t j = 0; j < pSeg->numOfSlots; ++j) {
+      if (pSeg->pBuffer == NULL || pSeg->pBuffer[j] == NULL) {
+        continue;
+      }
+      // required value in current slot
+      if (num < (count + 1) && num + pSeg->pBuffer[j]->numOfTotalElems >= (count + 1)) {
+        if (pSeg->pBuffer[j]->numOfTotalElems + num == (count + 1)) {
+          /*
+           * now, we need to find the minimum value of the next slot for interpolating the percentile value
+           * j is the last slot of current segment, we need to get the first slot of the next segment.
+           *
+           */
+          MinMaxEntry next = getMinMaxEntryOfNextSlotWithData(pMemBucket, i, j);
+
+          double maxOfThisSlot = 0;
+          double minOfNextSlot = 0;
+          switch (pMemBucket->dataType) {
+            case TSDB_DATA_TYPE_INT:
+            case TSDB_DATA_TYPE_SMALLINT:
+            case TSDB_DATA_TYPE_TINYINT: {
+              maxOfThisSlot = pSeg->pBoundingEntries[j].iMaxVal;
+              minOfNextSlot = next.iMinVal;
+              break;
+            };
+            case TSDB_DATA_TYPE_FLOAT:
+            case TSDB_DATA_TYPE_DOUBLE: {
+              maxOfThisSlot = pSeg->pBoundingEntries[j].dMaxVal;
+              minOfNextSlot = next.dMinVal;
+              break;
+            };
+            case TSDB_DATA_TYPE_BIGINT: {
+              maxOfThisSlot = (double)pSeg->pBoundingEntries[j].i64MaxVal;
+              minOfNextSlot = (double)next.i64MinVal;
+              break;
+            }
+          };
+
+          assert(minOfNextSlot > maxOfThisSlot);
+
+          double val = (1 - fraction) * maxOfThisSlot + fraction * minOfNextSlot;
+          return val;
+        }
+        if (pSeg->pBuffer[j]->numOfTotalElems <= pMemBucket->maxElemsCapacity) {
+          // data in buffer and file are merged together to be processed.
+          tFilePage *buffer = loadIntoBucketFromDisk(pMemBucket, i, j, pMemBucket->pOrderDesc);
+          int32_t    currentIdx = count - num;
+
+          char * thisVal = buffer->data + pMemBucket->nElemSize * currentIdx;
+          char * nextVal = thisVal + pMemBucket->nElemSize;
+          double td, nd;
+          switch (pMemBucket->dataType) {
+            case TSDB_DATA_TYPE_SMALLINT: {
+              td = *(int16_t *)thisVal;
+              nd = *(int16_t *)nextVal;
+              break;
+            }
+            case TSDB_DATA_TYPE_TINYINT: {
+              td = *(int8_t *)thisVal;
+              nd = *(int8_t *)nextVal;
+              break;
+            }
+            case TSDB_DATA_TYPE_INT: {
+              td = *(int32_t *)thisVal;
+              nd = *(int32_t *)nextVal;
+              break;
+            };
+            case TSDB_DATA_TYPE_FLOAT: {
+              // td = *(float *)thisVal;
+              // nd = *(float *)nextVal;
+              td = GET_FLOAT_VAL(thisVal);
+              nd = GET_FLOAT_VAL(nextVal);
+              break;
+            }
+            case TSDB_DATA_TYPE_DOUBLE: {
+              // td = *(double *)thisVal;
+              td = GET_DOUBLE_VAL(thisVal);
+              // nd = *(double *)nextVal;
+              nd = GET_DOUBLE_VAL(nextVal);
+              break;
+            }
+            case TSDB_DATA_TYPE_BIGINT: {
+              td = (double)*(int64_t *)thisVal;
+              nd = (double)*(int64_t *)nextVal;
+              break;
+            }
+          }
+          double val = (1 - fraction) * td + fraction * nd;
+          tfree(buffer);
+
+          return val;
+        } else {  // incur a second round bucket split
+          if (isIdenticalData(pMemBucket, i, j)) {
+            tExtMemBuffer *pMemBuffer = pSeg->pBuffer[j];
+
+            tFilePage *pPage = (tFilePage *)malloc(pMemBuffer->pageSize);
+
+            char *thisVal = getFirstElemOfMemBuffer(pSeg, j, pPage);
+
+            double finalResult = 0.0;
+
+            switch (pMemBucket->dataType) {
+              case TSDB_DATA_TYPE_SMALLINT: {
+                finalResult = *(int16_t *)thisVal;
+                break;
+              }
+              case TSDB_DATA_TYPE_TINYINT: {
+                finalResult = *(int8_t *)thisVal;
+                break;
+              }
+              case TSDB_DATA_TYPE_INT: {
+                finalResult = *(int32_t *)thisVal;
+                break;
+              };
+              case TSDB_DATA_TYPE_FLOAT: {
+                // finalResult = *(float *)thisVal;
+                finalResult = GET_FLOAT_VAL(thisVal);
+                break;
+              }
+              case TSDB_DATA_TYPE_DOUBLE: {
+                // finalResult = *(double *)thisVal;
+                finalResult = GET_DOUBLE_VAL(thisVal);
+                break;
+              }
+              case TSDB_DATA_TYPE_BIGINT: {
+                finalResult = (double)(*(int64_t *)thisVal);
+                break;
+              }
+            }
+
+            free(pPage);
+            return finalResult;
+          }
+
+          pTrace("MemBucket:%p,start second round bucketing", pMemBucket);
+
+          if (pSeg->pBuffer[j]->numOfElemsInBuffer != 0) {
+            pTrace("MemBucket:%p,flush %d pages to disk, clear status", pMemBucket, pSeg->pBuffer[j]->numOfInMemPages);
+
+            pMemBucket->numOfAvailPages += pSeg->pBuffer[j]->numOfInMemPages;
+            tExtMemBufferFlush(pSeg->pBuffer[j]);
+          }
+
+          tExtMemBuffer *pMemBuffer = pSeg->pBuffer[j];
+          pSeg->pBuffer[j] = NULL;
+
+          // release all
+          for (int32_t tt = 0; tt < pMemBucket->numOfSegs; ++tt) {
+            tMemBucketSegment *pSeg = &pMemBucket->pSegs[tt];
+            for (int32_t ttx = 0; ttx < pSeg->numOfSlots; ++ttx) {
+              if (pSeg->pBuffer && pSeg->pBuffer[ttx]) {
+                pSeg->pBuffer[ttx] = destoryExtMemBuffer(pSeg->pBuffer[ttx]);
+              }
+            }
+          }
+
+          pMemBucket->nRange.i64MaxVal = pSeg->pBoundingEntries->i64MaxVal;
+          pMemBucket->nRange.i64MinVal = pSeg->pBoundingEntries->i64MinVal;
+          pMemBucket->numOfElems = 0;
+
+          for (int32_t tt = 0; tt < pMemBucket->numOfSegs; ++tt) {
+            tMemBucketSegment *pSeg = &pMemBucket->pSegs[tt];
+            for (int32_t ttx = 0; ttx < pSeg->numOfSlots; ++ttx) {
+              if (pSeg->pBoundingEntries) {
+                resetBoundingBox(pSeg, pMemBucket->dataType);
+              }
+            }
+          }
+
+          tFilePage *pPage = (tFilePage *)malloc(pMemBuffer->pageSize);
+
+          tFlushoutInfo *pFlushInfo = &pMemBuffer->fileMeta.flushoutData.pFlushoutInfo[0];
+          assert(pFlushInfo->numOfPages == pMemBuffer->fileMeta.nFileSize);
+
+          int32_t ret = fseek(pMemBuffer->file, pFlushInfo->startPageId * pMemBuffer->pageSize, SEEK_SET);
+          UNUSED(ret);
+
+          for (uint32_t jx = 0; jx < pFlushInfo->numOfPages; ++jx) {
+            ret = fread(pPage, pMemBuffer->pageSize, 1, pMemBuffer->file);
+            UNUSED(ret);
+            tMemBucketPut(pMemBucket, pPage->data, pPage->numOfElems);
+          }
+
+          fclose(pMemBuffer->file);
+          if (unlink(pMemBuffer->path) != 0) {
+            pError("MemBucket:%p, remove tmp file %s failed", pMemBucket, pMemBuffer->path);
+          }
+          tfree(pMemBuffer);
+          tfree(pPage);
+
+          return getPercentileImpl(pMemBucket, count - num, fraction);
+        }
+      } else {
+        num += pSeg->pBuffer[j]->numOfTotalElems;
+      }
+    }
+  }
+  return 0;
+}
+
+double getPercentile(tMemBucket *pMemBucket, double percent) {
+  if (pMemBucket->numOfElems == 0) {
+    return 0.0;
+  }
+
+  if (pMemBucket->numOfElems == 1) {  // return the only element
+    return findOnlyResult(pMemBucket);
+  }
+
+  percent = fabs(percent);
+
+  // validate the parameters
+  if (fabs(percent - 100.0) < DBL_EPSILON || (percent < DBL_EPSILON)) {
+    double minx = 0, maxx = 0;
+    /*
+     * find the min/max value, no need to scan all data in bucket
+     */
+    findMaxMinValue(pMemBucket, &maxx, &minx);
+
+    return fabs(percent - 100) < DBL_EPSILON ? maxx : minx;
+  }
+
+  double  percentVal = (percent * (pMemBucket->numOfElems - 1)) / ((double)100.0);
+  int32_t orderIdx = (int32_t)percentVal;
+
+  // do put data by using buckets
+  return getPercentileImpl(pMemBucket, orderIdx, percentVal - orderIdx);
+}
+
+/*
+ * check if data in one slot are all identical
+ * only need to compare with the bounding box
+ */
+bool isIdenticalData(tMemBucket *pMemBucket, int32_t segIdx, int32_t slotIdx) {
+  tMemBucketSegment *pSeg = &pMemBucket->pSegs[segIdx];
+
+  if (pMemBucket->dataType == TSDB_DATA_TYPE_INT || pMemBucket->dataType == TSDB_DATA_TYPE_BIGINT ||
+      pMemBucket->dataType == TSDB_DATA_TYPE_SMALLINT || pMemBucket->dataType == TSDB_DATA_TYPE_TINYINT) {
+    return pSeg->pBoundingEntries[slotIdx].i64MinVal == pSeg->pBoundingEntries[slotIdx].i64MaxVal;
+  }
+
+  if (pMemBucket->dataType == TSDB_DATA_TYPE_FLOAT || pMemBucket->dataType == TSDB_DATA_TYPE_DOUBLE) {
+    return fabs(pSeg->pBoundingEntries[slotIdx].dMaxVal - pSeg->pBoundingEntries[slotIdx].dMinVal) < DBL_EPSILON;
+  }
+
+  return false;
+}
+
+/*
+ * get the first element of one slot into memory.
+ * if no data of current slot in memory, load it from disk
+ */
+char *getFirstElemOfMemBuffer(tMemBucketSegment *pSeg, int32_t slotIdx, tFilePage *pPage) {
+  tExtMemBuffer *pMemBuffer = pSeg->pBuffer[slotIdx];
+  char *         thisVal = NULL;
+
+  if (pSeg->pBuffer[slotIdx]->numOfElemsInBuffer != 0) {
+    thisVal = pSeg->pBuffer[slotIdx]->pHead->item.data;
+  } else {
+    /*
+     * no data in memory, load one page into memory
+     */
+    tFlushoutInfo *pFlushInfo = &pMemBuffer->fileMeta.flushoutData.pFlushoutInfo[0];
+    assert(pFlushInfo->numOfPages == pMemBuffer->fileMeta.nFileSize);
+
+    fseek(pMemBuffer->file, pFlushInfo->startPageId * pMemBuffer->pageSize, SEEK_SET);
+    size_t ret = fread(pPage, pMemBuffer->pageSize, 1, pMemBuffer->file);
+    UNUSED(ret);
+    thisVal = pPage->data;
+  }
+  return thisVal;
+}

src/util/src/tresultBuf.c

+#include "hash.h"
+#include "taoserror.h"
+#include "textbuffer.h"
+#include "tlog.h"
+#include "tsqlfunction.h"
+#include "tresultBuf.h"
+
+#define DEFAULT_INTERN_BUF_SIZE 16384L
+
+int32_t createResultBuf(SQueryResultBuf** pResultBuf, int32_t size, int32_t rowSize) {
+  SQueryResultBuf* pResBuf = calloc(1, sizeof(SQueryResultBuf));
+  pResBuf->numOfRowsPerPage = (DEFAULT_INTERN_BUF_SIZE - sizeof(tFilePage)) / rowSize;
+  pResBuf->numOfPages = size;
+
+  pResBuf->totalBufSize = pResBuf->numOfPages * DEFAULT_INTERN_BUF_SIZE;
+  pResBuf->incStep = 4;
+
+  // init id hash table
+  pResBuf->idsTable = taosInitHashTable(size, taosGetDefaultHashFunction(TSDB_DATA_TYPE_INT), false);
+  pResBuf->list = calloc(size, sizeof(SIDList));
+  pResBuf->numOfAllocGroupIds = size;
+
+  char path[4096] = {0};
+  getTmpfilePath("tsdb_q_buf", path);
+  pResBuf->path = strdup(path);
+
+  pResBuf->fd = open(pResBuf->path, O_CREAT | O_RDWR, 0666);
+  
+  memset(path, 0, tListLen(path));
+
+  if (!FD_VALID(pResBuf->fd)) {
+    pError("failed to create tmp file: %s on disk. %s", pResBuf->path, strerror(errno));
+    return TSDB_CODE_CLI_NO_DISKSPACE;
+  }
+
+  int32_t ret = ftruncate(pResBuf->fd, pResBuf->numOfPages * DEFAULT_INTERN_BUF_SIZE);
+  if (ret != TSDB_CODE_SUCCESS) {
+    pError("failed to create tmp file: %s on disk. %s", pResBuf->path, strerror(errno));
+    return TSDB_CODE_CLI_NO_DISKSPACE;
+  }
+
+  pResBuf->pBuf = mmap(NULL, pResBuf->totalBufSize, PROT_READ | PROT_WRITE, MAP_SHARED, pResBuf->fd, 0);
+  if (pResBuf->pBuf == MAP_FAILED) {
+    pError("QInfo:%p failed to map temp file: %s. %s", pResBuf->path, strerror(errno));
+    return TSDB_CODE_CLI_OUT_OF_MEMORY; // todo change error code
+  }
+
+  pTrace("create tmp file for output result, %s, " PRId64 "bytes", pResBuf->path, pResBuf->totalBufSize);
+  *pResultBuf = pResBuf;
+  return TSDB_CODE_SUCCESS;
+}
+
+tFilePage* getResultBufferPageById(SQueryResultBuf* pResultBuf, int32_t id) {
+  assert(id < pResultBuf->numOfPages && id >= 0);
+
+  return (tFilePage*)(pResultBuf->pBuf + DEFAULT_INTERN_BUF_SIZE * id);
+}
+
+int32_t getNumOfResultBufGroupId(SQueryResultBuf* pResultBuf) { return taosNumElemsInHashTable(pResultBuf->idsTable); }
+
+int32_t getResBufSize(SQueryResultBuf* pResultBuf) { return pResultBuf->totalBufSize; }
+
+static int32_t extendDiskFileSize(SQueryResultBuf* pResultBuf, int32_t numOfPages) {
+  assert(pResultBuf->numOfPages * DEFAULT_INTERN_BUF_SIZE == pResultBuf->totalBufSize);
+
+  int32_t ret = munmap(pResultBuf->pBuf, pResultBuf->totalBufSize);
+  pResultBuf->numOfPages += numOfPages;
+
+  /*
+   * disk-based output buffer is exhausted, try to extend the disk-based buffer, the available disk space may
+   * be insufficient
+   */
+  ret = ftruncate(pResultBuf->fd, pResultBuf->numOfPages * DEFAULT_INTERN_BUF_SIZE);
+  if (ret != 0) {
+    //    dError("QInfo:%p failed to create intermediate result output file:%s. %s", pQInfo, pSupporter->extBufFile,
+    //           strerror(errno));
+    return -TSDB_CODE_SERV_NO_DISKSPACE;
+  }
+
+  pResultBuf->totalBufSize = pResultBuf->numOfPages * DEFAULT_INTERN_BUF_SIZE;
+  pResultBuf->pBuf = mmap(NULL, pResultBuf->totalBufSize, PROT_READ | PROT_WRITE, MAP_SHARED, pResultBuf->fd, 0);
+
+  if (pResultBuf->pBuf == MAP_FAILED) {
+    //    dError("QInfo:%p failed to map temp file: %s. %s", pQInfo, pSupporter->extBufFile, strerror(errno));
+    return -TSDB_CODE_SERV_OUT_OF_MEMORY;
+  }
+
+  return TSDB_CODE_SUCCESS;
+}
+
+static bool noMoreAvailablePages(SQueryResultBuf* pResultBuf) {
+  return (pResultBuf->allocateId == pResultBuf->numOfPages - 1);
+}
+
+static int32_t getGroupIndex(SQueryResultBuf* pResultBuf, int32_t groupId) {
+  assert(pResultBuf != NULL);
+
+  char* p = taosGetDataFromHashTable(pResultBuf->idsTable, (const char*)&groupId, sizeof(int32_t));
+  if (p == NULL) {  // it is a new group id
+    return -1;
+  }
+
+  int32_t slot = GET_INT32_VAL(p);
+  assert(slot >= 0 && slot < pResultBuf->numOfAllocGroupIds);
+
+  return slot;
+}
+
+static int32_t addNewGroupId(SQueryResultBuf* pResultBuf, int32_t groupId) {
+  int32_t num = getNumOfResultBufGroupId(pResultBuf); // the num is the newest allocated group id slot
+
+  if (pResultBuf->numOfAllocGroupIds <= num) {
+    size_t n = pResultBuf->numOfAllocGroupIds << 1u;
+
+    SIDList* p = (SIDList*)realloc(pResultBuf->list, sizeof(SIDList) * n);
+    assert(p != NULL);
+
+    memset(&p[pResultBuf->numOfAllocGroupIds], 0, sizeof(SIDList) * pResultBuf->numOfAllocGroupIds);
+    
+    pResultBuf->list = p;
+    pResultBuf->numOfAllocGroupIds = n;
+  }
+
+  taosAddToHashTable(pResultBuf->idsTable, (const char*)&groupId, sizeof(int32_t), &num, sizeof(int32_t));
+  return num;
+}
+
+static int32_t doRegisterId(SIDList* pList, int32_t id) {
+  if (pList->size >= pList->alloc) {
+    int32_t s = 0;
+    if (pList->alloc == 0) {
+      s = 4;
+      assert(pList->pData == NULL);
+    } else {
+      s = pList->alloc << 1u;
+    }
+    
+    int32_t* c = realloc(pList->pData, s * sizeof(int32_t));
+    assert(c);
+    
+    memset(&c[pList->alloc], 0, sizeof(int32_t) * pList->alloc);
+
+    pList->pData = c;
+    pList->alloc = s;
+  }
+
+  pList->pData[pList->size++] = id;
+  return 0;
+}
+
+static void registerPageId(SQueryResultBuf* pResultBuf, int32_t groupId, int32_t pageId) {
+  int32_t slot = getGroupIndex(pResultBuf, groupId);
+  if (slot < 0) {
+    slot = addNewGroupId(pResultBuf, groupId);
+  }
+
+  SIDList* pList = &pResultBuf->list[slot];
+  doRegisterId(pList, pageId);
+}
+
+tFilePage* getNewDataBuf(SQueryResultBuf* pResultBuf, int32_t groupId, int32_t* pageId) {
+  if (noMoreAvailablePages(pResultBuf)) {
+    if (extendDiskFileSize(pResultBuf, pResultBuf->incStep) != TSDB_CODE_SUCCESS) {
+      return NULL;
+    }
+  }
+
+  // register new id in this group
+  *pageId = (pResultBuf->allocateId++);
+  registerPageId(pResultBuf, groupId, *pageId);
+
+  tFilePage* page = getResultBufferPageById(pResultBuf, *pageId);
+  
+  // clear memory for the new page
+  memset(page, 0, DEFAULT_INTERN_BUF_SIZE);
+  
+  return page;
+}
+
+int32_t getNumOfRowsPerPage(SQueryResultBuf* pResultBuf) { return pResultBuf->numOfRowsPerPage; }
+
+SIDList getDataBufPagesIdList(SQueryResultBuf* pResultBuf, int32_t groupId) {
+  SIDList list = {0};
+  int32_t slot = getGroupIndex(pResultBuf, groupId);
+  if (slot < 0) {
+    return list;
+  } else {
+    return pResultBuf->list[slot];
+  }
+}
+
+void destroyResultBuf(SQueryResultBuf* pResultBuf) {
+  if (pResultBuf == NULL) {
+    return;
+  }
+
+  if (FD_VALID(pResultBuf->fd)) {
+    close(pResultBuf->fd);
+  }
+
+  pTrace("disk-based output buffer closed, %" PRId64 " bytes, file:%s", pResultBuf->totalBufSize, pResultBuf->path);
+  munmap(pResultBuf->pBuf, pResultBuf->totalBufSize);
+  unlink(pResultBuf->path);
+  
+  tfree(pResultBuf->path);
+
+  for (int32_t i = 0; i < pResultBuf->numOfAllocGroupIds; ++i) {
+    SIDList* pList = &pResultBuf->list[i];
+    tfree(pList->pData);
+  }
+
+  tfree(pResultBuf->list);
+  taosCleanUpHashTable(pResultBuf->idsTable);
+  
+  tfree(pResultBuf);
+}
+
+int32_t getLastPageId(SIDList *pList) {
+  if (pList == NULL && pList->size <= 0) {
+    return -1;
+  }
+  
+  return pList->pData[pList->size - 1];
+}
+

src/util/src/ttokenizer.c

@@ -274,7 +274,7 @@ int tSQLKeywordCode(const char* z, int n) {
     }
   }
 
-  SKeyword** pKey = (SKeyword**)taosGetDataFromHash(KeywordHashTable, key, n);
+  SKeyword** pKey = (SKeyword**)taosGetDataFromHashTable(KeywordHashTable, key, n);
   if (pKey != NULL) {
     return (*pKey)->type;
   } else {