diff --git a/include/common/tdatablock.h b/include/common/tdatablock.h
index 93ce7ea984..7839859e8b 100644
--- a/include/common/tdatablock.h
+++ b/include/common/tdatablock.h
@@ -184,7 +184,7 @@ static FORCE_INLINE void colDataAppendDouble(SColumnInfoData* pColumnInfoData, u
 int32_t getJsonValueLen(const char* data);
 
 int32_t colDataAppend(SColumnInfoData* pColumnInfoData, uint32_t currentRow, const char* pData, bool isNull);
-int32_t colDataReserve(SColumnInfoData* pColumnInfoData, size_t newSize);
+int32_t colDataAppendNItems(SColumnInfoData* pColumnInfoData, uint32_t currentRow, const char* pData, uint32_t numOfRows);
 int32_t colDataMergeCol(SColumnInfoData* pColumnInfoData, int32_t numOfRow1, int32_t* capacity,
                         const SColumnInfoData* pSource, int32_t numOfRow2);
 int32_t colDataAssign(SColumnInfoData* pColumnInfoData, const SColumnInfoData* pSource, int32_t numOfRows,
diff --git a/source/common/src/tdatablock.c b/source/common/src/tdatablock.c
index 5da7cfcbba..9cd00b4cb2 100644
--- a/source/common/src/tdatablock.c
+++ b/source/common/src/tdatablock.c
@@ -140,6 +140,54 @@ int32_t colDataReserve(SColumnInfoData* pColumnInfoData, size_t newSize) {
   return TSDB_CODE_SUCCESS;
 }
 
+static void doCopyNItems(struct SColumnInfoData* pColumnInfoData, int32_t currentRow, const char* pData, int32_t itemLen, int32_t numOfRows) {
+  ASSERT(pColumnInfoData->info.bytes >= itemLen);
+  size_t start = 1;
+
+  // the first item
+  memcpy(pColumnInfoData->pData, pData, itemLen);
+
+  int32_t t = 0;
+  int32_t count = log(numOfRows)/log(2);
+  while(t < count) {
+    int32_t xlen = 1 << t;
+    memcpy(pColumnInfoData->pData + start * itemLen + pColumnInfoData->varmeta.length, pColumnInfoData->pData, xlen * itemLen);
+    t += 1;
+    start += xlen;
+  }
+
+  // the tail part
+  if (numOfRows > start) {
+    memcpy(pColumnInfoData->pData + start * itemLen + currentRow * itemLen, pColumnInfoData->pData, (numOfRows - start) * itemLen);
+  }
+
+  if (IS_VAR_DATA_TYPE(pColumnInfoData->info.type)) {
+    for(int32_t i = 0; i < numOfRows; ++i) {
+      pColumnInfoData->varmeta.offset[i + currentRow] = pColumnInfoData->varmeta.length + i * itemLen;
+    }
+
+    pColumnInfoData->varmeta.length += numOfRows * itemLen;
+  }
+}
+
+int32_t colDataAppendNItems(SColumnInfoData* pColumnInfoData, uint32_t currentRow, const char* pData, uint32_t numOfRows) {
+  ASSERT(pData != NULL && pColumnInfoData != NULL);
+
+  int32_t len = pColumnInfoData->info.bytes;
+  if (IS_VAR_DATA_TYPE(pColumnInfoData->info.type)) {
+    len = varDataTLen(pData);
+    if (pColumnInfoData->varmeta.allocLen < (numOfRows + currentRow) * len) {
+      int32_t code = colDataReserve(pColumnInfoData, (numOfRows + currentRow) * len);
+      if (code != TSDB_CODE_SUCCESS) {
+        return code;
+      }
+    }
+  }
+
+  doCopyNItems(pColumnInfoData, currentRow, pData, len, numOfRows);
+  return TSDB_CODE_SUCCESS;
+}
+
 static void doBitmapMerge(SColumnInfoData* pColumnInfoData, int32_t numOfRow1, const SColumnInfoData* pSource,
                           int32_t numOfRow2) {
   if (numOfRow2 <= 0) return;
diff --git a/source/libs/executor/src/scanoperator.c b/source/libs/executor/src/scanoperator.c
index 3439259224..b8257a49f9 100644
--- a/source/libs/executor/src/scanoperator.c
+++ b/source/libs/executor/src/scanoperator.c
@@ -399,22 +399,17 @@ int32_t addTagPseudoColumnData(SReadHandle* pHandle, SExprInfo* pPseudoExpr, int
       const char* p = metaGetTableTagVal(&mr.me, pColInfoData->info.type, &tagVal);
 
       char* data = NULL;
-      int32_t len = 0;
       if (pColInfoData->info.type != TSDB_DATA_TYPE_JSON && p != NULL) {
         data = tTagValToData((const STagVal*)p, false);
-        len = varDataTLen(data);
-        code = colDataReserve(pColInfoData, len * pBlock->info.rows);
-        if (code != TSDB_CODE_SUCCESS) {
-          terrno = code;
-          return code;
-        }
       } else {
         data = (char*)p;
       }
 
-      for (int32_t i = 0; i < pBlock->info.rows; ++i) {
-        colDataAppend(pColInfoData, i, data,
-                      (data == NULL) || (pColInfoData->info.type == TSDB_DATA_TYPE_JSON && tTagIsJsonNull(data)));
+      bool isNullVal = (data == NULL) || (pColInfoData->info.type == TSDB_DATA_TYPE_JSON && tTagIsJsonNull(data));
+      if (isNullVal) {
+        colDataAppendNNULL(pColInfoData, 0, pBlock->info.rows);
+      } else {
+        colDataAppendNItems(pColInfoData, 0, data, pBlock->info.rows);
       }
 
       if (data && (pColInfoData->info.type != TSDB_DATA_TYPE_JSON) && p != NULL &&
diff --git a/source/libs/scalar/src/sclfunc.c b/source/libs/scalar/src/sclfunc.c
index c162b13799..2d889dd925 100644
--- a/source/libs/scalar/src/sclfunc.c
+++ b/source/libs/scalar/src/sclfunc.c
@@ -1727,17 +1727,11 @@ int32_t qTbnameFunction(SScalarParam *pInput, int32_t inputNum, SScalarParam *pO
 
   char str[TSDB_TABLE_FNAME_LEN + VARSTR_HEADER_SIZE] = {0};
   metaGetTableNameByUid(pInput->param, uid, str);
-
-  colDataReserve(pOutput->columnData, varDataTLen(str) * (pInput->numOfRows + pOutput->numOfRows));
-  for(int32_t i = 0; i < pInput->numOfRows; ++i) {
-    colDataAppend(pOutput->columnData, pOutput->numOfRows + i, str, false);
-  }
-
+  colDataAppendNItems(pOutput->columnData, pOutput->numOfRows, str, pInput->numOfRows);
   pOutput->numOfRows += pInput->numOfRows;
   return TSDB_CODE_SUCCESS;
 }
 
-
 /** Aggregation functions **/
 int32_t countScalarFunction(SScalarParam *pInput, int32_t inputNum, SScalarParam *pOutput) {
   SColumnInfoData *pInputData  = pInput->columnData;
diff --git a/source/os/src/osMath.c b/source/os/src/osMath.c
index 98cd63a831..cd2acac261 100644
--- a/source/os/src/osMath.c
+++ b/source/os/src/osMath.c
@@ -31,6 +31,7 @@ void swapStr(char* j, char* J, int width) {
 }
 #endif
 
+// todo refactor: 1) move away; 2) use merge sort instead; 3) qsort is not a stable sort actually.
 void taosSort(void* arr, int64_t sz, int64_t width, __compar_fn_t compar) {
 #ifdef WINDOWS
   int64_t i, j;