diff --git a/include/common/tmsg.h b/include/common/tmsg.h
index 47c26ed4317c49a3436ef18d8750b1c87b139504..23945cd50075db667fad613774dcc716edaeab30 100644
--- a/include/common/tmsg.h
+++ b/include/common/tmsg.h
@@ -933,6 +933,7 @@ typedef struct {
   int64_t numOfProcessedFetch;
   int64_t numOfProcessedDrop;
   int64_t numOfProcessedHb;
+  int64_t numOfProcessedDelete;
   int64_t cacheDataSize;
   int64_t numOfQueryInQueue;
   int64_t numOfFetchInQueue;
@@ -2689,20 +2690,20 @@ int32_t tEncodeSVSubmitReq(SEncoder* pCoder, const SVSubmitReq* pReq);
 int32_t tDecodeSVSubmitReq(SDecoder* pCoder, SVSubmitReq* pReq);
 
 typedef struct {
-  int64_t     delUid;
-  int64_t     tbUid;  // super/child/normal table
-  int8_t      type;   // table type
-  int16_t     nWnds;
-  char*       tbFullName;
-  char*       subPlan;
-  STimeWindow wnds[];
+  SMsgHead header;
+  uint64_t sId;
+  uint64_t queryId;
+  uint64_t taskId;
+  uint32_t sqlLen;
+  uint32_t phyLen;
+  char*    sql;
+  char*    msg;
 } SVDeleteReq;
 
-int32_t tEncodeSVDeleteReq(SEncoder* pCoder, const SVDeleteReq* pReq);
-int32_t tDecodeSVDeleteReq(SDecoder* pCoder, SVDeleteReq* pReq);
+int32_t tSerializeSVDeleteReq(void *buf, int32_t bufLen, SVDeleteReq *pReq);
+int32_t tDeserializeSVDeleteReq(void *buf, int32_t bufLen, SVDeleteReq *pReq);
 
 typedef struct {
-  int32_t code;
   int64_t affectedRows;
 } SVDeleteRsp;
 
diff --git a/include/libs/executor/dataSinkMgt.h b/include/libs/executor/dataSinkMgt.h
index 2cc9caca6fa4d8e4dd4bd6a8d7b490e7baaf2c34..c23cf162aa6e1ca35e3750aa54e91659c45d6b08 100644
--- a/include/libs/executor/dataSinkMgt.h
+++ b/include/libs/executor/dataSinkMgt.h
@@ -32,6 +32,18 @@ extern "C" {
 struct SDataSink;
 struct SSDataBlock;
 
+typedef struct SDeleterRes {
+  uint64_t uid;
+  SArray*  uidList;
+  int64_t  skey;
+  int64_t  ekey;
+  int64_t  affectedRows;
+} SDeleterRes;
+
+typedef struct SDeleterParam {
+  SArray* pUidList;
+} SDeleterParam;
+
 typedef struct SDataSinkStat {
   uint64_t cachedSize;
 } SDataSinkStat;
@@ -64,7 +76,7 @@ typedef struct SOutputData {
  * @param pHandle output
  * @return error code
  */
-int32_t dsCreateDataSinker(const SDataSinkNode* pDataSink, DataSinkHandle* pHandle);
+int32_t dsCreateDataSinker(const SDataSinkNode* pDataSink, DataSinkHandle* pHandle, void* pParam);
 
 int32_t dsDataSinkGetCacheSize(SDataSinkStat *pStat);
 
diff --git a/include/libs/qworker/qworker.h b/include/libs/qworker/qworker.h
index aa20082fe02c75c74306adcbdbddc250384ac116..f3f147955a03cb99f1ea806441eddd63ba8f96fe 100644
--- a/include/libs/qworker/qworker.h
+++ b/include/libs/qworker/qworker.h
@@ -31,7 +31,12 @@ enum {
   NODE_TYPE_MNODE,
 };
 
-
+typedef struct SDeleteRes {
+  uint64_t uid;
+  SArray*  uidList;
+  int64_t  skey;
+  int64_t  ekey;
+} SDeleteRes;
 
 typedef struct SQWorkerCfg {
   uint32_t maxSchedulerNum;
@@ -47,6 +52,7 @@ typedef struct {
   uint64_t fetchProcessed;
   uint64_t dropProcessed;
   uint64_t hbProcessed;
+  uint64_t deleteProcessed;
   
   uint64_t numOfQueryInQueue;
   uint64_t numOfFetchInQueue;
@@ -74,6 +80,8 @@ int32_t qWorkerProcessDropMsg(void *node, void *qWorkerMgmt, SRpcMsg *pMsg, int6
 
 int32_t qWorkerProcessHbMsg(void *node, void *qWorkerMgmt, SRpcMsg *pMsg, int64_t ts);
 
+int32_t qWorkerProcessDeleteMsg(void *node, void *qWorkerMgmt, SRpcMsg *pMsg, SRpcMsg *pRsp, SDeleteRes *pRes);
+
 void qWorkerDestroy(void **qWorkerMgmt);
 
 int32_t qWorkerGetStat(SReadHandle *handle, void *qWorkerMgmt, SQWorkerStat *pStat);
diff --git a/source/client/src/clientImpl.c b/source/client/src/clientImpl.c
index c71c7d3487078f79ceee63ba9079670a8a7d4daa..62026dd577a608d6a5595cf508ba62de6b7aa177 100644
--- a/source/client/src/clientImpl.c
+++ b/source/client/src/clientImpl.c
@@ -457,7 +457,7 @@ int32_t scheduleQuery(SRequestObj* pRequest, SQueryPlan* pDag, SArray* pNodeList
     return pRequest->code;
   }
 
-  if (TDMT_VND_SUBMIT == pRequest->type || TDMT_VND_CREATE_TABLE == pRequest->type) {
+  if (TDMT_VND_SUBMIT == pRequest->type || TDMT_VND_DELETE == pRequest->type || TDMT_VND_CREATE_TABLE == pRequest->type) {
     pRequest->body.resInfo.numOfRows = res.numOfRows;
 
     if (pRequest->body.queryJob != 0) {
diff --git a/source/common/src/tmsg.c b/source/common/src/tmsg.c
index 7041c9478e27c3d8421040da225aafeeef177ba3..773e430b8365eb8456c8c90ddcfe4194c921922b 100644
--- a/source/common/src/tmsg.c
+++ b/source/common/src/tmsg.c
@@ -932,6 +932,7 @@ int32_t tSerializeSStatusReq(void *buf, int32_t bufLen, SStatusReq *pReq) {
   if (tEncodeI64(&encoder, pReq->qload.numOfProcessedFetch) < 0) return -1;
   if (tEncodeI64(&encoder, pReq->qload.numOfProcessedDrop) < 0) return -1;
   if (tEncodeI64(&encoder, pReq->qload.numOfProcessedHb) < 0) return -1;
+  if (tEncodeI64(&encoder, pReq->qload.numOfProcessedDelete) < 0) return -1;
   if (tEncodeI64(&encoder, pReq->qload.cacheDataSize) < 0) return -1;
   if (tEncodeI64(&encoder, pReq->qload.numOfQueryInQueue) < 0) return -1;
   if (tEncodeI64(&encoder, pReq->qload.numOfFetchInQueue) < 0) return -1;
@@ -1001,6 +1002,7 @@ int32_t tDeserializeSStatusReq(void *buf, int32_t bufLen, SStatusReq *pReq) {
   if (tDecodeI64(&decoder, &pReq->qload.numOfProcessedFetch) < 0) return -1;
   if (tDecodeI64(&decoder, &pReq->qload.numOfProcessedDrop) < 0) return -1;
   if (tDecodeI64(&decoder, &pReq->qload.numOfProcessedHb) < 0) return -1;
+  if (tDecodeI64(&decoder, &pReq->qload.numOfProcessedDelete) < 0) return -1;
   if (tDecodeI64(&decoder, &pReq->qload.cacheDataSize) < 0) return -1;
   if (tDecodeI64(&decoder, &pReq->qload.numOfQueryInQueue) < 0) return -1;
   if (tDecodeI64(&decoder, &pReq->qload.numOfFetchInQueue) < 0) return -1;
@@ -3823,46 +3825,70 @@ int32_t tDecodeSVGetTsmaExpWndsRsp(SDecoder *pCoder, SVGetTsmaExpWndsRsp *pReq)
   return 0;
 }
 
-int32_t tEncodeSVDeleteReq(SEncoder *pCoder, const SVDeleteReq *pReq) {
-  if (tStartEncode(pCoder) < 0) return -1;
+int32_t tSerializeSVDeleteReq(void *buf, int32_t bufLen, SVDeleteReq *pReq) {
+  int32_t headLen = sizeof(SMsgHead);
+  if (buf != NULL) {
+    buf = (char *)buf + headLen;
+    bufLen -= headLen;
+  }
 
-  if (tEncodeI64(pCoder, pReq->delUid) < 0) return -1;
-  if (tEncodeI64(pCoder, pReq->tbUid) < 0) return -1;
-  if (tEncodeI8(pCoder, pReq->type) < 0) return -1;
-  if (tEncodeI16v(pCoder, pReq->nWnds) < 0) return -1;
-  if (tEncodeCStr(pCoder, pReq->tbFullName) < 0) return -1;
-  if (tEncodeCStr(pCoder, pReq->subPlan) < 0) return -1;
-  for (int16_t i = 0; i < pReq->nWnds; ++i) {
-    if (tEncodeI64(pCoder, pReq->wnds[i].skey) < 0) return -1;
-    if (tEncodeI64(pCoder, pReq->wnds[i].ekey) < 0) return -1;
+  SEncoder encoder = {0};
+  tEncoderInit(&encoder, buf, bufLen);
+
+  if (tStartEncode(&encoder) < 0) return -1;
+  if (tEncodeU64(&encoder, pReq->sId) < 0) return -1;
+  if (tEncodeU64(&encoder, pReq->queryId) < 0) return -1;
+  if (tEncodeU64(&encoder, pReq->taskId) < 0) return -1;
+  if (tEncodeU32(&encoder, pReq->sqlLen) < 0) return -1;
+  if (tEncodeU32(&encoder, pReq->phyLen) < 0) return -1;
+  if (tEncodeCStr(&encoder, pReq->sql) < 0) return -1;
+  if (tEncodeCStr(&encoder, pReq->msg) < 0) return -1;
+  tEndEncode(&encoder);
+
+  int32_t tlen = encoder.pos;
+  tEncoderClear(&encoder);
+
+  if (buf != NULL) {
+    SMsgHead *pHead = (SMsgHead *)((char *)buf - headLen);
+    pHead->vgId = htonl(pReq->header.vgId);
+    pHead->contLen = htonl(tlen + headLen);
   }
 
-  tEndEncode(pCoder);
-  return 0;
+  return tlen + headLen;
 }
 
-int32_t tDecodeSVDeleteReq(SDecoder *pCoder, SVDeleteReq *pReq) {
-  if (tStartDecode(pCoder) < 0) return -1;
+int32_t tDeserializeSVDeleteReq(void *buf, int32_t bufLen, SVDeleteReq *pReq) {
+  int32_t headLen = sizeof(SMsgHead);
 
-  if (tDecodeI64(pCoder, &pReq->delUid) < 0) return -1;
-  if (tDecodeI64(pCoder, &pReq->tbUid) < 0) return -1;
-  if (tDecodeI8(pCoder, &pReq->type) < 0) return -1;
-  if (tDecodeI16v(pCoder, &pReq->nWnds) < 0) return -1;
-  if (tDecodeCStr(pCoder, &pReq->tbFullName) < 0) return -1;
-  if (tDecodeCStr(pCoder, &pReq->subPlan) < 0) return -1;
-  for (int16_t i = 0; i < pReq->nWnds; ++i) {
-    if (tDecodeI64(pCoder, &pReq->wnds[i].skey) < 0) return -1;
-    if (tDecodeI64(pCoder, &pReq->wnds[i].ekey) < 0) return -1;
-  }
+  SMsgHead *pHead = buf;
+  pHead->vgId = pReq->header.vgId;
+  pHead->contLen = pReq->header.contLen;
 
-  tEndDecode(pCoder);
+  SDecoder decoder = {0};
+  tDecoderInit(&decoder, (char *)buf + headLen, bufLen - headLen);
+
+  if (tStartDecode(&decoder) < 0) return -1;
+  if (tDecodeU64(&decoder, &pReq->sId) < 0) return -1;
+  if (tDecodeU64(&decoder, &pReq->queryId) < 0) return -1;
+  if (tDecodeU64(&decoder, &pReq->taskId) < 0) return -1;
+  if (tDecodeU32(&decoder, &pReq->sqlLen) < 0) return -1;
+  if (tDecodeU32(&decoder, &pReq->phyLen) < 0) return -1;
+  pReq->sql = taosMemoryCalloc(1, pReq->sqlLen + 1);
+  if (NULL == pReq->sql) return -1;
+  pReq->msg = taosMemoryCalloc(1, pReq->phyLen + 1);
+  if (NULL == pReq->msg) return -1;
+  if (tDecodeCStrTo(&decoder, pReq->sql) < 0) return -1;
+  if (tDecodeCStrTo(&decoder, pReq->msg) < 0) return -1;
+
+  tEndDecode(&decoder);
+
+  tDecoderClear(&decoder);
   return 0;
 }
 
 int32_t tEncodeSVDeleteRsp(SEncoder *pCoder, const SVDeleteRsp *pReq) {
   if (tStartEncode(pCoder) < 0) return -1;
 
-  if (tEncodeI32(pCoder, pReq->code) < 0) return -1;
   if (tEncodeI64(pCoder, pReq->affectedRows) < 0) return -1;
 
   tEndEncode(pCoder);
@@ -3872,7 +3898,6 @@ int32_t tEncodeSVDeleteRsp(SEncoder *pCoder, const SVDeleteRsp *pReq) {
 int32_t tDecodeSVDeleteRsp(SDecoder *pCoder, SVDeleteRsp *pReq) {
   if (tStartDecode(pCoder) < 0) return -1;
 
-  if (tDecodeI32(pCoder, &pReq->code) < 0) return -1;
   if (tDecodeI64(pCoder, &pReq->affectedRows) < 0) return -1;
 
   tEndDecode(pCoder);
diff --git a/source/dnode/mgmt/mgmt_vnode/src/vmHandle.c b/source/dnode/mgmt/mgmt_vnode/src/vmHandle.c
index 77109ab52f1f9e6067ff48ac6b70a40a221b37ab..ccc399605bd22132265b77a43dd44c4c5e506459 100644
--- a/source/dnode/mgmt/mgmt_vnode/src/vmHandle.c
+++ b/source/dnode/mgmt/mgmt_vnode/src/vmHandle.c
@@ -358,6 +358,7 @@ SArray *vmGetMsgHandles() {
   if (dmSetMgmtHandle(pArray, TDMT_VND_TASK_RUN, vmPutMsgToFetchQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_VND_TASK_DISPATCH, vmPutMsgToFetchQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_VND_TASK_RECOVER, vmPutMsgToFetchQueue, 0) == NULL) goto _OVER;
+  if (dmSetMgmtHandle(pArray, TDMT_VND_DELETE, vmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
 
   if (dmSetMgmtHandle(pArray, TDMT_VND_ALTER_REPLICA, vmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
   if (dmSetMgmtHandle(pArray, TDMT_VND_ALTER_CONFIG, vmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
diff --git a/source/dnode/qnode/src/qnode.c b/source/dnode/qnode/src/qnode.c
index 45b88318c474dfd3f053bb8803e92be32defcecf..ebaf73a952283d0feb4f276b33912dd419becbf7 100644
--- a/source/dnode/qnode/src/qnode.c
+++ b/source/dnode/qnode/src/qnode.c
@@ -59,6 +59,7 @@ int32_t qndGetLoad(SQnode *pQnode, SQnodeLoad *pLoad) {
   pLoad->numOfProcessedFetch = stat.fetchProcessed;
   pLoad->numOfProcessedDrop = stat.dropProcessed;
   pLoad->numOfProcessedHb = stat.hbProcessed;
+  pLoad->numOfProcessedDelete = stat.deleteProcessed;
   
   return 0; 
 }
diff --git a/source/dnode/vnode/src/vnd/vnodeSvr.c b/source/dnode/vnode/src/vnd/vnodeSvr.c
index 4bcd5fc095a5e71f5e87e7ee374993f557bf3000..3eee6713d86c471f89db6a3d9d870faa5563741c 100644
--- a/source/dnode/vnode/src/vnd/vnodeSvr.c
+++ b/source/dnode/vnode/src/vnd/vnodeSvr.c
@@ -24,6 +24,7 @@ static int32_t vnodeProcessDropTbReq(SVnode *pVnode, int64_t version, void *pReq
 static int32_t vnodeProcessSubmitReq(SVnode *pVnode, int64_t version, void *pReq, int32_t len, SRpcMsg *pRsp);
 static int32_t vnodeProcessCreateTSmaReq(SVnode *pVnode, int64_t version, void *pReq, int32_t len, SRpcMsg *pRsp);
 static int32_t vnodeProcessAlterConfirmReq(SVnode *pVnode, int64_t version, void *pReq, int32_t len, SRpcMsg *pRsp);
+static int32_t vnodeProcessWriteMsg(SVnode *pVnode, int64_t version, SRpcMsg *pMsg, SRpcMsg *pRsp);
 
 int32_t vnodePreprocessReq(SVnode *pVnode, SRpcMsg *pMsg) {
   int32_t  code = 0;
@@ -142,6 +143,9 @@ int32_t vnodeProcessWriteReq(SVnode *pVnode, SRpcMsg *pMsg, int64_t version, SRp
     case TDMT_VND_SUBMIT:
       if (vnodeProcessSubmitReq(pVnode, version, pMsg->pCont, pMsg->contLen, pRsp) < 0) goto _err;
       break;
+    case TDMT_VND_DELETE:
+      if (vnodeProcessWriteMsg(pVnode, version, pMsg, pRsp) < 0) goto _err;
+      break;
     /* TQ */
     case TDMT_VND_MQ_VG_CHANGE:
       if (tqProcessVgChangeReq(pVnode->pTq, POINTER_SHIFT(pMsg->pCont, sizeof(SMsgHead)),
@@ -256,6 +260,22 @@ int32_t vnodeProcessFetchMsg(SVnode *pVnode, SRpcMsg *pMsg, SQueueInfo *pInfo) {
   }
 }
 
+int32_t vnodeProcessWriteMsg(SVnode *pVnode, int64_t version, SRpcMsg *pMsg, SRpcMsg *pRsp) {
+  vTrace("message in write queue is processing");
+  char   *msgstr = POINTER_SHIFT(pMsg->pCont, sizeof(SMsgHead));
+  int32_t msgLen = pMsg->contLen - sizeof(SMsgHead);
+  SDeleteRes res = {0};
+  SReadHandle handle = {.meta = pVnode->pMeta, .config = &pVnode->config, .vnode = pVnode, .pMsgCb = &pVnode->msgCb};
+  
+  switch (pMsg->msgType) {
+    case TDMT_VND_DELETE:
+      return qWorkerProcessDeleteMsg(&handle, pVnode->pQuery, pMsg, pRsp, &res);
+    default:
+      vError("unknown msg type:%d in write queue", pMsg->msgType);
+      return TSDB_CODE_VND_APP_ERROR;
+  }
+}
+
 // TODO: remove the function
 void smaHandleRes(void *pVnode, int64_t smaId, const SArray *data) {
   // TODO
@@ -873,4 +893,4 @@ static int32_t vnodeProcessAlterConfirmReq(SVnode *pVnode, int64_t version, void
   pRsp->contLen = 0;
 
   return 0;
-}
\ No newline at end of file
+}
diff --git a/source/libs/executor/inc/dataSinkInt.h b/source/libs/executor/inc/dataSinkInt.h
index 8f49440105c813b512835717e861d3da1b2065df..dead1aff7383a0f6da2b8d83d290bdd7a1be3a31 100644
--- a/source/libs/executor/inc/dataSinkInt.h
+++ b/source/libs/executor/inc/dataSinkInt.h
@@ -49,6 +49,7 @@ typedef struct SDataSinkHandle {
 } SDataSinkHandle;
 
 int32_t createDataDispatcher(SDataSinkManager* pManager, const SDataSinkNode* pDataSink, DataSinkHandle* pHandle);
+int32_t createDataDeleter(SDataSinkManager* pManager, const SDataSinkNode* pDataSink, DataSinkHandle* pHandle, void *pParam);
 
 #ifdef __cplusplus
 }
diff --git a/source/libs/executor/inc/executorimpl.h b/source/libs/executor/inc/executorimpl.h
index 1e0739e06632fac7603854ec6db549adcd6710e9..4804984397bc67df0133fd514ae562d426707082 100644
--- a/source/libs/executor/inc/executorimpl.h
+++ b/source/libs/executor/inc/executorimpl.h
@@ -869,6 +869,7 @@ int32_t decodeOperator(SOperatorInfo* ops, char* data, int32_t length);
 void    setTaskStatus(SExecTaskInfo* pTaskInfo, int8_t status);
 int32_t createExecTaskInfoImpl(SSubplan* pPlan, SExecTaskInfo** pTaskInfo, SReadHandle* pHandle, uint64_t taskId,
                                EOPTR_EXEC_MODEL model);
+int32_t createDataSinkParam(SDataSinkNode *pNode, void **pParam, qTaskInfo_t* pTaskInfo);                               
 int32_t getOperatorExplainExecInfo(SOperatorInfo* operatorInfo, SExplainExecInfo** pRes, int32_t* capacity,
                                    int32_t* resNum);
 
diff --git a/source/libs/executor/src/dataDeleter.c b/source/libs/executor/src/dataDeleter.c
new file mode 100644
index 0000000000000000000000000000000000000000..33b7811e6c0c26032d1bca8cc8643dce2b0ea984
--- /dev/null
+++ b/source/libs/executor/src/dataDeleter.c
@@ -0,0 +1,254 @@
+/*
+ * 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 "dataSinkInt.h"
+#include "dataSinkMgt.h"
+#include "executorimpl.h"
+#include "planner.h"
+#include "tcompression.h"
+#include "tdatablock.h"
+#include "tglobal.h"
+#include "tqueue.h"
+
+extern SDataSinkStat gDataSinkStat;
+
+typedef struct SDataDeleterBuf {
+  int32_t useSize;
+  int32_t allocSize;
+  char*   pData;
+} SDataDeleterBuf;
+
+typedef struct SDataCacheEntry {
+  int32_t dataLen;
+  int32_t numOfRows;
+  int32_t numOfCols;
+  int8_t  compressed;
+  char    data[];
+} SDataCacheEntry;
+
+typedef struct SDataDeleterHandle {
+  SDataSinkHandle     sink;
+  SDataSinkManager*   pManager;
+  SDataBlockDescNode* pSchema;
+  SDataDeleterNode*   pDeleter;
+  SDeleterParam*      pParam;
+  STaosQueue*         pDataBlocks;
+  SDataDeleterBuf     nextOutput;
+  int32_t             status;
+  bool                queryEnd;
+  uint64_t            useconds;
+  uint64_t            cachedSize;
+  TdThreadMutex       mutex;
+} SDataDeleterHandle;
+
+static bool needCompress(const SSDataBlock* pData, int32_t numOfCols) {
+  if (tsCompressColData < 0 || 0 == pData->info.rows) {
+    return false;
+  }
+
+  for (int32_t col = 0; col < numOfCols; ++col) {
+    SColumnInfoData* pColRes = taosArrayGet(pData->pDataBlock, col);
+    int32_t          colSize = pColRes->info.bytes * pData->info.rows;
+    if (NEEDTO_COMPRESS_QUERY(colSize)) {
+      return true;
+    }
+  }
+
+  return false;
+}
+
+static void toDataCacheEntry(SDataDeleterHandle* pHandle, const SInputData* pInput, SDataDeleterBuf* pBuf) {
+  int32_t numOfCols = LIST_LENGTH(pHandle->pSchema->pSlots);
+
+  SDataCacheEntry* pEntry = (SDataCacheEntry*)pBuf->pData;
+  pEntry->compressed = 0;
+  pEntry->numOfRows = pInput->pData->info.rows;
+  pEntry->numOfCols = pInput->pData->info.numOfCols;
+  pEntry->dataLen = sizeof(SDeleterRes);
+
+  ASSERT(1 == pEntry->numOfRows);
+  ASSERT(1 == pEntry->numOfCols);
+
+  pBuf->useSize = sizeof(SDataCacheEntry);
+
+  SColumnInfoData* pColRes = (SColumnInfoData*)taosArrayGet(pInput->pData->pDataBlock, 0);
+
+  SDeleterRes* pRes = (SDeleterRes*)pEntry->data;
+  pRes->uid = pHandle->pDeleter->tableId;
+  pRes->uidList = pHandle->pParam->pUidList;
+  pRes->skey = pHandle->pDeleter->deleteTimeRange.skey;
+  pRes->ekey = pHandle->pDeleter->deleteTimeRange.ekey;
+  pRes->affectedRows = *(int64_t*)pColRes->pData;
+
+  pBuf->useSize += pEntry->dataLen;
+  
+  atomic_add_fetch_64(&pHandle->cachedSize, pEntry->dataLen); 
+  atomic_add_fetch_64(&gDataSinkStat.cachedSize, pEntry->dataLen); 
+}
+
+static bool allocBuf(SDataDeleterHandle* pDeleter, const SInputData* pInput, SDataDeleterBuf* pBuf) {
+  uint32_t capacity = pDeleter->pManager->cfg.maxDataBlockNumPerQuery;
+  if (taosQueueItemSize(pDeleter->pDataBlocks) > capacity) {
+    qError("SinkNode queue is full, no capacity, max:%d, current:%d, no capacity", capacity,
+           taosQueueItemSize(pDeleter->pDataBlocks));
+    return false;
+  }
+
+  pBuf->allocSize = sizeof(SDataCacheEntry) + sizeof(SDeleterRes);
+
+  pBuf->pData = taosMemoryMalloc(pBuf->allocSize);
+  if (pBuf->pData == NULL) {
+    qError("SinkNode failed to malloc memory, size:%d, code:%d", pBuf->allocSize, TAOS_SYSTEM_ERROR(errno));
+  }
+
+  return NULL != pBuf->pData;
+}
+
+static int32_t updateStatus(SDataDeleterHandle* pDeleter) {
+  taosThreadMutexLock(&pDeleter->mutex);
+  int32_t blockNums = taosQueueItemSize(pDeleter->pDataBlocks);
+  int32_t status =
+      (0 == blockNums ? DS_BUF_EMPTY
+                      : (blockNums < pDeleter->pManager->cfg.maxDataBlockNumPerQuery ? DS_BUF_LOW : DS_BUF_FULL));
+  pDeleter->status = status;
+  taosThreadMutexUnlock(&pDeleter->mutex);
+  return status;
+}
+
+static int32_t getStatus(SDataDeleterHandle* pDeleter) {
+  taosThreadMutexLock(&pDeleter->mutex);
+  int32_t status = pDeleter->status;
+  taosThreadMutexUnlock(&pDeleter->mutex);
+  return status;
+}
+
+static int32_t putDataBlock(SDataSinkHandle* pHandle, const SInputData* pInput, bool* pContinue) {
+  SDataDeleterHandle* pDeleter = (SDataDeleterHandle*)pHandle;
+  SDataDeleterBuf*    pBuf = taosAllocateQitem(sizeof(SDataDeleterBuf), DEF_QITEM);
+  if (NULL == pBuf || !allocBuf(pDeleter, pInput, pBuf)) {
+    return TSDB_CODE_QRY_OUT_OF_MEMORY;
+  }
+  toDataCacheEntry(pDeleter, pInput, pBuf);
+  taosWriteQitem(pDeleter->pDataBlocks, pBuf);
+  *pContinue = (DS_BUF_LOW == updateStatus(pDeleter) ? true : false);
+  return TSDB_CODE_SUCCESS;
+}
+
+static void endPut(struct SDataSinkHandle* pHandle, uint64_t useconds) {
+  SDataDeleterHandle* pDeleter = (SDataDeleterHandle*)pHandle;
+  taosThreadMutexLock(&pDeleter->mutex);
+  pDeleter->queryEnd = true;
+  pDeleter->useconds = useconds;
+  taosThreadMutexUnlock(&pDeleter->mutex);
+}
+
+static void getDataLength(SDataSinkHandle* pHandle, int32_t* pLen, bool* pQueryEnd) {
+  SDataDeleterHandle* pDeleter = (SDataDeleterHandle*)pHandle;
+  if (taosQueueEmpty(pDeleter->pDataBlocks)) {
+    *pQueryEnd = pDeleter->queryEnd;
+    *pLen = 0;
+    return;
+  }
+
+  SDataDeleterBuf* pBuf = NULL;
+  taosReadQitem(pDeleter->pDataBlocks, (void**)&pBuf);
+  memcpy(&pDeleter->nextOutput, pBuf, sizeof(SDataDeleterBuf));
+  taosFreeQitem(pBuf);
+  *pLen = ((SDataCacheEntry*)(pDeleter->nextOutput.pData))->dataLen;
+  *pQueryEnd = pDeleter->queryEnd;
+  qDebug("got data len %d, row num %d in sink", *pLen, ((SDataCacheEntry*)(pDeleter->nextOutput.pData))->numOfRows);
+}
+
+static int32_t getDataBlock(SDataSinkHandle* pHandle, SOutputData* pOutput) {
+  SDataDeleterHandle* pDeleter = (SDataDeleterHandle*)pHandle;
+  if (NULL == pDeleter->nextOutput.pData) {
+    assert(pDeleter->queryEnd);
+    pOutput->useconds = pDeleter->useconds;
+    pOutput->precision = pDeleter->pSchema->precision;
+    pOutput->bufStatus = DS_BUF_EMPTY;
+    pOutput->queryEnd = pDeleter->queryEnd;
+    return TSDB_CODE_SUCCESS;
+  }
+  SDataCacheEntry* pEntry = (SDataCacheEntry*)(pDeleter->nextOutput.pData);
+  memcpy(pOutput->pData, pEntry->data, pEntry->dataLen);
+  pOutput->numOfRows = pEntry->numOfRows;
+  pOutput->numOfCols = pEntry->numOfCols;
+  pOutput->compressed = pEntry->compressed;
+
+  atomic_sub_fetch_64(&pDeleter->cachedSize, pEntry->dataLen);  
+  atomic_sub_fetch_64(&gDataSinkStat.cachedSize, pEntry->dataLen); 
+
+  taosMemoryFreeClear(pDeleter->nextOutput.pData);  // todo persistent
+  pOutput->bufStatus = updateStatus(pDeleter);
+  taosThreadMutexLock(&pDeleter->mutex);
+  pOutput->queryEnd = pDeleter->queryEnd;
+  pOutput->useconds = pDeleter->useconds;
+  pOutput->precision = pDeleter->pSchema->precision;
+  taosThreadMutexUnlock(&pDeleter->mutex);
+  
+  return TSDB_CODE_SUCCESS;
+}
+
+static int32_t destroyDataSinker(SDataSinkHandle* pHandle) {
+  SDataDeleterHandle* pDeleter = (SDataDeleterHandle*)pHandle;
+  atomic_sub_fetch_64(&gDataSinkStat.cachedSize, pDeleter->cachedSize);
+  taosMemoryFreeClear(pDeleter->nextOutput.pData);
+  while (!taosQueueEmpty(pDeleter->pDataBlocks)) {
+    SDataDeleterBuf* pBuf = NULL;
+    taosReadQitem(pDeleter->pDataBlocks, (void**)&pBuf);
+    taosMemoryFreeClear(pBuf->pData);
+    taosFreeQitem(pBuf);
+  }
+  taosCloseQueue(pDeleter->pDataBlocks);
+  taosThreadMutexDestroy(&pDeleter->mutex);
+  return TSDB_CODE_SUCCESS;
+}
+
+static int32_t getCacheSize(struct SDataSinkHandle* pHandle, uint64_t* size) {
+  SDataDeleterHandle* pDispatcher = (SDataDeleterHandle*)pHandle;
+
+  *size = atomic_load_64(&pDispatcher->cachedSize);
+  return TSDB_CODE_SUCCESS;
+}
+
+int32_t createDataDeleter(SDataSinkManager* pManager, const SDataSinkNode* pDataSink, DataSinkHandle* pHandle, void *pParam) {
+  SDataDeleterHandle* deleter = taosMemoryCalloc(1, sizeof(SDataDeleterHandle));
+  if (NULL == deleter) {
+    terrno = TSDB_CODE_QRY_OUT_OF_MEMORY;
+    return TSDB_CODE_QRY_OUT_OF_MEMORY;
+  }
+
+  SDataDeleterNode* pDeleterNode = (SDataDeleterNode *)pDataSink;
+  deleter->sink.fPut = putDataBlock;
+  deleter->sink.fEndPut = endPut;
+  deleter->sink.fGetLen = getDataLength;
+  deleter->sink.fGetData = getDataBlock;
+  deleter->sink.fDestroy = destroyDataSinker;
+  deleter->sink.fGetCacheSize = getCacheSize;
+  deleter->pManager = pManager;
+  deleter->pDeleter = pDeleterNode;
+  deleter->pSchema = pDataSink->pInputDataBlockDesc;
+  deleter->pParam = pParam;
+  deleter->status = DS_BUF_EMPTY;
+  deleter->queryEnd = false;
+  deleter->pDataBlocks = taosOpenQueue();
+  taosThreadMutexInit(&deleter->mutex, NULL);
+  if (NULL == deleter->pDataBlocks) {
+    terrno = TSDB_CODE_QRY_OUT_OF_MEMORY;
+    return TSDB_CODE_QRY_OUT_OF_MEMORY;
+  }
+  *pHandle = deleter;
+  return TSDB_CODE_SUCCESS;
+}
diff --git a/source/libs/executor/src/dataDispatcher.c b/source/libs/executor/src/dataDispatcher.c
index 080cf5c2ad44f31f11f0fce0e2350fe121c2c1fb..eaa366b7e75ea75d9cd32f1f99b71685feb6baa5 100644
--- a/source/libs/executor/src/dataDispatcher.c
+++ b/source/libs/executor/src/dataDispatcher.c
@@ -83,7 +83,7 @@ static void toDataCacheEntry(SDataDispatchHandle* pHandle, const SInputData* pIn
   pEntry->numOfCols = pInput->pData->info.numOfCols;
   pEntry->dataLen = 0;
 
-  pBuf->useSize = sizeof(SRetrieveTableRsp);
+  pBuf->useSize = sizeof(SDataCacheEntry);
   blockCompressEncode(pInput->pData, pEntry->data, &pEntry->dataLen, numOfCols, pEntry->compressed);
 
   pBuf->useSize += pEntry->dataLen;
@@ -100,7 +100,7 @@ static bool allocBuf(SDataDispatchHandle* pDispatcher, const SInputData* pInput,
     return false;
   }
 
-  pBuf->allocSize = sizeof(SRetrieveTableRsp) + blockGetEncodeSize(pInput->pData);
+  pBuf->allocSize = sizeof(SDataCacheEntry) + blockGetEncodeSize(pInput->pData);
 
   pBuf->pData = taosMemoryMalloc(pBuf->allocSize);
   if (pBuf->pData == NULL) {
@@ -211,7 +211,7 @@ static int32_t destroyDataSinker(SDataSinkHandle* pHandle) {
   return TSDB_CODE_SUCCESS;
 }
 
-int32_t getCacheSize(struct SDataSinkHandle* pHandle, uint64_t* size) {
+static int32_t getCacheSize(struct SDataSinkHandle* pHandle, uint64_t* size) {
   SDataDispatchHandle* pDispatcher = (SDataDispatchHandle*)pHandle;
 
   *size = atomic_load_64(&pDispatcher->cachedSize);
diff --git a/source/libs/executor/src/dataSinkMgt.c b/source/libs/executor/src/dataSinkMgt.c
index 9016ca274a3567d8cbc45d522d5e1cb93b176e68..ffa9822e927fc10e42a43064e428ccbe45acd00f 100644
--- a/source/libs/executor/src/dataSinkMgt.c
+++ b/source/libs/executor/src/dataSinkMgt.c
@@ -34,9 +34,12 @@ int32_t dsDataSinkGetCacheSize(SDataSinkStat *pStat) {
 }
 
 
-int32_t dsCreateDataSinker(const SDataSinkNode *pDataSink, DataSinkHandle* pHandle) {
-  if (QUERY_NODE_PHYSICAL_PLAN_DISPATCH == nodeType(pDataSink)) {
-    return createDataDispatcher(&gDataSinkManager, pDataSink, pHandle);
+int32_t dsCreateDataSinker(const SDataSinkNode *pDataSink, DataSinkHandle* pHandle, void* pParam) {
+  switch (nodeType(pDataSink)) {
+    case QUERY_NODE_PHYSICAL_PLAN_DISPATCH:
+      return createDataDispatcher(&gDataSinkManager, pDataSink, pHandle);
+    case QUERY_NODE_PHYSICAL_PLAN_DELETE:
+      return createDataDeleter(&gDataSinkManager, pDataSink, pHandle, pParam);
   }
   return TSDB_CODE_FAILED;
 }
diff --git a/source/libs/executor/src/executorMain.c b/source/libs/executor/src/executorMain.c
index 7757825733153741d6e83404051578f7f4e2aef8..c014b2395306c06184ab43875d04e93813ab427e 100644
--- a/source/libs/executor/src/executorMain.c
+++ b/source/libs/executor/src/executorMain.c
@@ -45,8 +45,15 @@ int32_t qCreateExecTask(SReadHandle* readHandle, int32_t vgId, uint64_t taskId,
   if (code != TSDB_CODE_SUCCESS) {
     goto _error;
   }
+  
   if (handle) {
-    code = dsCreateDataSinker(pSubplan->pDataSink, handle);
+    void* pSinkParam = NULL;
+    code = createDataSinkParam(pSubplan->pDataSink, &pSinkParam, pTaskInfo);
+    if (code != TSDB_CODE_SUCCESS) {
+      goto _error;
+    }
+    
+    code = dsCreateDataSinker(pSubplan->pDataSink, handle, pSinkParam);
   }
 
   _error:
diff --git a/source/libs/executor/src/executorimpl.c b/source/libs/executor/src/executorimpl.c
index f08c2f95ba3fc3e8056792cd192892afb806454b..9d7f939bb015d6b074286b0b9a681027419e3f3e 100644
--- a/source/libs/executor/src/executorimpl.c
+++ b/source/libs/executor/src/executorimpl.c
@@ -5099,6 +5099,37 @@ int32_t decodeOperator(SOperatorInfo* ops, char* result, int32_t length) {
   return TDB_CODE_SUCCESS;
 }
 
+int32_t createDataSinkParam(SDataSinkNode *pNode, void **pParam, qTaskInfo_t* pTaskInfo) {
+  SExecTaskInfo* pTask = *(SExecTaskInfo**)pTaskInfo;
+  
+  switch (pNode->type) {
+    case QUERY_NODE_PHYSICAL_PLAN_DELETE: {
+      SDeleterParam *pDeleterParam = taosMemoryCalloc(1, sizeof(SDeleterParam));
+      if (NULL == pDeleterParam) {
+        return TSDB_CODE_OUT_OF_MEMORY;
+      }
+      int32_t tbNum = taosArrayGetSize(pTask->tableqinfoList.pTableList);
+      pDeleterParam->pUidList = taosArrayInit(tbNum, sizeof(uint64_t));
+      if (NULL == pDeleterParam->pUidList) {
+        taosMemoryFree(pDeleterParam);
+        return TSDB_CODE_OUT_OF_MEMORY;
+      }
+      for (int32_t i = 0; i < tbNum; ++i) {
+        STableKeyInfo *pTable = taosArrayGet(pTask->tableqinfoList.pTableList, i);
+        taosArrayPush(pDeleterParam->pUidList, &pTable->uid);
+      }
+
+      *pParam = pDeleterParam;
+      break;
+    }
+    default:
+      break;
+  }
+
+  return TSDB_CODE_SUCCESS;
+}
+
+
 int32_t createExecTaskInfoImpl(SSubplan* pPlan, SExecTaskInfo** pTaskInfo, SReadHandle* pHandle, uint64_t taskId,
                                EOPTR_EXEC_MODEL model) {
   uint64_t queryId = pPlan->id.queryId;
diff --git a/source/libs/monitor/src/monMsg.c b/source/libs/monitor/src/monMsg.c
index 944a7b54750c9e8850d0fe124f36561c54a6630e..a041b582a9eaafc0e8553262150c1a15e7edc469 100644
--- a/source/libs/monitor/src/monMsg.c
+++ b/source/libs/monitor/src/monMsg.c
@@ -569,6 +569,7 @@ int32_t tSerializeSQnodeLoad(void *buf, int32_t bufLen, SQnodeLoad *pInfo) {
   if (tEncodeI64(&encoder, pInfo->numOfProcessedFetch) < 0) return -1;
   if (tEncodeI64(&encoder, pInfo->numOfProcessedDrop) < 0) return -1;
   if (tEncodeI64(&encoder, pInfo->numOfProcessedHb) < 0) return -1;
+  if (tEncodeI64(&encoder, pInfo->numOfProcessedDelete) < 0) return -1;
   if (tEncodeI64(&encoder, pInfo->cacheDataSize) < 0) return -1;
   if (tEncodeI64(&encoder, pInfo->numOfQueryInQueue) < 0) return -1;
   if (tEncodeI64(&encoder, pInfo->numOfFetchInQueue) < 0) return -1;
@@ -591,6 +592,7 @@ int32_t tDeserializeSQnodeLoad(void *buf, int32_t bufLen, SQnodeLoad *pInfo) {
   if (tDecodeI64(&decoder, &pInfo->numOfProcessedFetch) < 0) return -1;
   if (tDecodeI64(&decoder, &pInfo->numOfProcessedDrop) < 0) return -1;
   if (tDecodeI64(&decoder, &pInfo->numOfProcessedHb) < 0) return -1;
+  if (tDecodeI64(&decoder, &pInfo->numOfProcessedDelete) < 0) return -1;
   if (tDecodeI64(&decoder, &pInfo->cacheDataSize) < 0) return -1;
   if (tDecodeI64(&decoder, &pInfo->numOfQueryInQueue) < 0) return -1;
   if (tDecodeI64(&decoder, &pInfo->numOfFetchInQueue) < 0) return -1;
diff --git a/source/libs/parser/src/parTranslater.c b/source/libs/parser/src/parTranslater.c
index 1ba47d0d15d1d5d35829727df6d8e0fb6b230c01..d81c6760e8d0b351614e099d867568b2f448248f 100644
--- a/source/libs/parser/src/parTranslater.c
+++ b/source/libs/parser/src/parTranslater.c
@@ -5015,6 +5015,10 @@ static int32_t setQuery(STranslateContext* pCxt, SQuery* pQuery) {
       pQuery->haveResultSet = true;
       pQuery->msgType = TDMT_VND_QUERY;
       break;
+    case QUERY_NODE_DELETE_STMT:
+      pQuery->execMode = QUERY_EXEC_MODE_SCHEDULE;
+      pQuery->msgType = TDMT_VND_DELETE;
+      break;
     case QUERY_NODE_VNODE_MODIF_STMT:
       pQuery->execMode = QUERY_EXEC_MODE_SCHEDULE;
       pQuery->msgType = toMsgType(((SVnodeModifOpStmt*)pQuery->pRoot)->sqlNodeType);
diff --git a/source/libs/planner/src/planPhysiCreater.c b/source/libs/planner/src/planPhysiCreater.c
index 785f112003ef9684af9162d1a43bb12505f3b72d..6b3d82b64493c3221161a4fcb74849cf518e65f8 100644
--- a/source/libs/planner/src/planPhysiCreater.c
+++ b/source/libs/planner/src/planPhysiCreater.c
@@ -1395,6 +1395,7 @@ static int32_t buildDeleteSubplan(SPhysiPlanContext* pCxt, SVnodeModifyLogicNode
   if (TSDB_CODE_SUCCESS == code) {
     code = createDataDeleter(pCxt, pModify, pSubplan->pNode, &pSubplan->pDataSink);
   }
+  pSubplan->msgType = TDMT_VND_DELETE;
   return code;
 }
 
diff --git a/source/libs/planner/src/planner.c b/source/libs/planner/src/planner.c
index f8d240c7b2d2162800cbc32ee7af2eeb62645d89..1921b1638875c218a417bb18b8da1dfb7ffd48a5 100644
--- a/source/libs/planner/src/planner.c
+++ b/source/libs/planner/src/planner.c
@@ -88,7 +88,7 @@ int32_t qSetSubplanExecutionNode(SSubplan* subplan, int32_t groupId, SDownstream
 }
 
 int32_t qSubPlanToString(const SSubplan* pSubplan, char** pStr, int32_t* pLen) {
-  if (SUBPLAN_TYPE_MODIFY == pSubplan->subplanType) {
+  if (SUBPLAN_TYPE_MODIFY == pSubplan->subplanType && NULL == pSubplan->pNode) {
     SDataInserterNode* insert = (SDataInserterNode*)pSubplan->pDataSink;
     *pLen = insert->size;
     *pStr = insert->pData;
diff --git a/source/libs/qworker/inc/qwInt.h b/source/libs/qworker/inc/qwInt.h
index 082db6428f79575cf29a635a1965f214893d22a5..1d31c86308e749fab5271ccbdb81090c801c9277 100644
--- a/source/libs/qworker/inc/qwInt.h
+++ b/source/libs/qworker/inc/qwInt.h
@@ -160,6 +160,7 @@ typedef struct SQWMsgStat {
   uint64_t cancelProcessed;
   uint64_t dropProcessed;
   uint64_t hbProcessed;
+  uint64_t deleteProcessed;
 } SQWMsgStat;
 
 typedef struct SQWRTStat {
@@ -357,6 +358,7 @@ int32_t qwUpdateTimeInQueue(SQWorker *mgmt, int64_t ts, EQueueType type);
 int64_t qwGetTimeInQueue(SQWorker *mgmt, EQueueType type);
 void qwClearExpiredSch(SArray* pExpiredSch);
 int32_t qwAcquireScheduler(SQWorker *mgmt, uint64_t sId, int32_t rwType, SQWSchStatus **sch);
+void qwFreeTaskCtx(QW_FPARAMS_DEF, SQWTaskCtx *ctx);
 
 void qwDbgDumpMgmtInfo(SQWorker *mgmt);
 int32_t qwDbgValidateStatus(QW_FPARAMS_DEF, int8_t oriStatus, int8_t newStatus, bool *ignore);
diff --git a/source/libs/qworker/inc/qwMsg.h b/source/libs/qworker/inc/qwMsg.h
index 5a8a45cd5150bc5e2aa009f2f3690e8f98a14c05..29861d87ac8957a8ad3e593e796b57a66b5b8eb6 100644
--- a/source/libs/qworker/inc/qwMsg.h
+++ b/source/libs/qworker/inc/qwMsg.h
@@ -30,6 +30,7 @@ int32_t qwProcessReady(QW_FPARAMS_DEF, SQWMsg *qwMsg);
 int32_t qwProcessFetch(QW_FPARAMS_DEF, SQWMsg *qwMsg);
 int32_t qwProcessDrop(QW_FPARAMS_DEF, SQWMsg *qwMsg);
 int32_t qwProcessHb(SQWorker *mgmt, SQWMsg *qwMsg, SSchedulerHbReq *req);
+int32_t qwProcessDelete(QW_FPARAMS_DEF, SQWMsg *qwMsg, SRpcMsg *pRsp, SDeleteRes *pRes);
 
 int32_t qwBuildAndSendDropRsp(SRpcHandleInfo *pConn, int32_t code);
 int32_t qwBuildAndSendCancelRsp(SRpcHandleInfo *pConn, int32_t code);
diff --git a/source/libs/qworker/src/qwMsg.c b/source/libs/qworker/src/qwMsg.c
index 2f1ecc51723a4799071292f5d7eb8d3b3d2f65ab..848a0420cadc01df68ab051949d2fadc9e7d7990 100644
--- a/source/libs/qworker/src/qwMsg.c
+++ b/source/libs/qworker/src/qwMsg.c
@@ -300,13 +300,6 @@ int32_t qWorkerProcessQueryMsg(void *node, void *qWorkerMgmt, SRpcMsg *pMsg, int
     QW_ERR_RET(TSDB_CODE_QRY_INVALID_INPUT);
   }
 
-  msg->sId = msg->sId;
-  msg->queryId = msg->queryId;
-  msg->taskId = msg->taskId;
-  msg->refId = msg->refId;
-  msg->phyLen = msg->phyLen;
-  msg->sqlLen = msg->sqlLen;
-
   uint64_t sId = msg->sId;
   uint64_t qId = msg->queryId;
   uint64_t tId = msg->taskId;
@@ -523,3 +516,37 @@ int32_t qWorkerProcessHbMsg(void *node, void *qWorkerMgmt, SRpcMsg *pMsg, int64_
 
   return TSDB_CODE_SUCCESS;
 }
+
+
+int32_t qWorkerProcessDeleteMsg(void *node, void *qWorkerMgmt, SRpcMsg *pMsg, SRpcMsg *pRsp, SDeleteRes *pRes) {
+  if (NULL == node || NULL == qWorkerMgmt || NULL == pMsg || NULL == pRsp) {
+    QW_ERR_RET(TSDB_CODE_QRY_INVALID_INPUT);
+  }
+
+  int32_t       code = 0;
+  SVDeleteReq req = {0};
+  SQWorker *    mgmt = (SQWorker *)qWorkerMgmt;
+
+  QW_STAT_INC(mgmt->stat.msgStat.deleteProcessed, 1);
+
+  tDeserializeSVDeleteReq(pMsg->pCont, pMsg->contLen, &req);
+  
+  uint64_t sId = req.sId;
+  uint64_t qId = req.queryId;
+  uint64_t tId = req.taskId;
+  int64_t  rId = 0;
+
+  SQWMsg qwMsg = {.node = node, .msg = req.msg, .msgLen = req.phyLen, .connInfo = pMsg->info};
+  QW_SCH_TASK_DLOG("processDelete start, node:%p, handle:%p, sql:%s", node, pMsg->info.handle, req.sql);
+  taosMemoryFreeClear(req.sql);
+
+  QW_ERR_JRET(qwProcessDelete(QW_FPARAMS(), &qwMsg, pRsp, pRes));
+
+  QW_SCH_TASK_DLOG("processDelete end, node:%p", node);
+
+_return:
+
+  QW_RET(code);
+}
+
+
diff --git a/source/libs/qworker/src/qwUtil.c b/source/libs/qworker/src/qwUtil.c
index 8bfb80f0616e223f96eafcd34e8d89768ffb98ca..667008e68e70451f8a4394ab2a792d179f09497f 100644
--- a/source/libs/qworker/src/qwUtil.c
+++ b/source/libs/qworker/src/qwUtil.c
@@ -290,8 +290,11 @@ int32_t qwKillTaskHandle(QW_FPARAMS_DEF, SQWTaskCtx *ctx) {
   QW_RET(code);
 }
 
-void qwFreeTask(QW_FPARAMS_DEF, SQWTaskCtx *ctx) {
-  tmsgReleaseHandle(&ctx->ctrlConnInfo, TAOS_CONN_SERVER);
+void qwFreeTaskCtx(QW_FPARAMS_DEF, SQWTaskCtx *ctx) {
+  if (ctx->ctrlConnInfo.handle) {
+    tmsgReleaseHandle(&ctx->ctrlConnInfo, TAOS_CONN_SERVER);
+  }
+  
   ctx->ctrlConnInfo.handle = NULL;
   ctx->ctrlConnInfo.refId = -1;
 
@@ -333,7 +336,7 @@ int32_t qwDropTaskCtx(QW_FPARAMS_DEF) {
     QW_ERR_RET(TSDB_CODE_QRY_TASK_CTX_NOT_EXIST);
   }
 
-  qwFreeTask(QW_FPARAMS(), &octx);
+  qwFreeTaskCtx(QW_FPARAMS(), &octx);
 
   QW_TASK_DLOG_E("task ctx dropped");
 
diff --git a/source/libs/qworker/src/qworker.c b/source/libs/qworker/src/qworker.c
index 44a8fdf7f4e29c6ad8ee0e8dc5b26612bc1e665f..333884f883d8ee15e1432a0c2bee179d6c3c5f72 100644
--- a/source/libs/qworker/src/qworker.c
+++ b/source/libs/qworker/src/qworker.c
@@ -183,7 +183,7 @@ int32_t qwGenerateSchHbRsp(SQWorker *mgmt, SQWSchStatus *sch, SQWHbInfo *hbInfo)
   return TSDB_CODE_SUCCESS;
 }
 
-int32_t qwGetResFromSink(QW_FPARAMS_DEF, SQWTaskCtx *ctx, int32_t *dataLen, void **rspMsg, SOutputData *pOutput) {
+int32_t qwGetQueryResFromSink(QW_FPARAMS_DEF, SQWTaskCtx *ctx, int32_t *dataLen, void **rspMsg, SOutputData *pOutput) {
   int32_t            len = 0;
   SRetrieveTableRsp *rsp = NULL;
   bool               queryEnd = false;
@@ -242,6 +242,53 @@ int32_t qwGetResFromSink(QW_FPARAMS_DEF, SQWTaskCtx *ctx, int32_t *dataLen, void
   return TSDB_CODE_SUCCESS;
 }
 
+int32_t qwGetDeleteResFromSink(QW_FPARAMS_DEF, SQWTaskCtx *ctx, int32_t *dataLen, void **rspMsg, SDeleteRes *pRes) {
+  int32_t            len = 0;
+  SVDeleteRsp        rsp = {0};
+  bool               queryEnd = false;
+  int32_t            code = 0;
+  SOutputData        output = {0};
+
+  dsGetDataLength(ctx->sinkHandle, &len, &queryEnd);
+
+  if (len <= 0 || len != sizeof(SDeleterRes)) {
+    QW_TASK_ELOG("invalid length from dsGetDataLength, length:%d", len);
+    QW_ERR_RET(TSDB_CODE_QRY_INVALID_INPUT);
+  }
+
+  output.pData = taosMemoryCalloc(1, len);
+  if (NULL == output.pData) {
+    QW_ERR_RET(TSDB_CODE_OUT_OF_MEMORY);
+  }
+  
+  code = dsGetDataBlock(ctx->sinkHandle, &output);
+  if (code) {
+    QW_TASK_ELOG("dsGetDataBlock failed, code:%x - %s", code, tstrerror(code));
+    taosMemoryFree(output.pData);
+    QW_ERR_RET(code);
+  }
+
+  SDeleterRes* pDelRes = (SDeleterRes*)output.pData;
+  
+  rsp.affectedRows = pDelRes->affectedRows;
+  pRes->uid = pDelRes->uid;
+  pRes->uidList = pDelRes->uidList;
+  pRes->skey = pDelRes->skey;
+  pRes->ekey = pDelRes->ekey;
+
+  SEncoder coder = {0};  
+  tEncodeSize(tEncodeSVDeleteRsp, &rsp, len, code);
+  void *msg = rpcMallocCont(len);
+  tEncoderInit(&coder, msg, len);
+  tEncodeSVDeleteRsp(&coder, &rsp);
+  tEncoderClear(&coder);
+
+  *rspMsg = msg;
+  *dataLen = len;
+  
+  return TSDB_CODE_SUCCESS;
+}
+
 
 int32_t qwHandlePrePhaseEvents(QW_FPARAMS_DEF, int8_t phase, SQWPhaseInput *input, SQWPhaseOutput *output) {
   int32_t         code = 0;
@@ -547,7 +594,7 @@ int32_t qwProcessCQuery(QW_FPARAMS_DEF, SQWMsg *qwMsg) {
 
     if (QW_IS_EVENT_RECEIVED(ctx, QW_EVENT_FETCH)) {
       SOutputData sOutput = {0};
-      QW_ERR_JRET(qwGetResFromSink(QW_FPARAMS(), ctx, &dataLen, &rsp, &sOutput));
+      QW_ERR_JRET(qwGetQueryResFromSink(QW_FPARAMS(), ctx, &dataLen, &rsp, &sOutput));
 
       if ((!sOutput.queryEnd) && (DS_BUF_LOW == sOutput.bufStatus || DS_BUF_EMPTY == sOutput.bufStatus)) {
         QW_TASK_DLOG("task not end and buf is %s, need to continue query", qwBufStatusStr(sOutput.bufStatus));
@@ -620,7 +667,7 @@ int32_t qwProcessFetch(QW_FPARAMS_DEF, SQWMsg *qwMsg) {
   QW_ERR_JRET(qwGetTaskCtx(QW_FPARAMS(), &ctx));
 
   SOutputData sOutput = {0};
-  QW_ERR_JRET(qwGetResFromSink(QW_FPARAMS(), ctx, &dataLen, &rsp, &sOutput));
+  QW_ERR_JRET(qwGetQueryResFromSink(QW_FPARAMS(), ctx, &dataLen, &rsp, &sOutput));
 
   if (NULL == rsp) {
     ctx->dataConnInfo = qwMsg->connInfo;
@@ -875,6 +922,47 @@ _return:
   qwRelease(refId);
 }
 
+int32_t qwProcessDelete(QW_FPARAMS_DEF, SQWMsg *qwMsg, SRpcMsg *pRsp, SDeleteRes *pRes) {
+  int32_t        code = 0;
+  SSubplan      *plan = NULL;
+  qTaskInfo_t    pTaskInfo = NULL;
+  DataSinkHandle sinkHandle = NULL;
+  SQWTaskCtx     ctx = {0};
+
+  code = qStringToSubplan(qwMsg->msg, &plan);
+  if (TSDB_CODE_SUCCESS != code) {
+    code = TSDB_CODE_INVALID_MSG;
+    QW_TASK_ELOG("task physical plan to subplan failed, code:%x - %s", code, tstrerror(code));
+    QW_ERR_JRET(code);
+  }
+
+  ctx.plan = plan;
+
+  code = qCreateExecTask(qwMsg->node, mgmt->nodeId, tId, plan, &pTaskInfo, &sinkHandle, OPTR_EXEC_MODEL_BATCH);
+  if (code) {
+    QW_TASK_ELOG("qCreateExecTask failed, code:%x - %s", code, tstrerror(code));
+    QW_ERR_JRET(code);
+  }
+
+  if (NULL == sinkHandle || NULL == pTaskInfo) {
+    QW_TASK_ELOG("create task result error, taskHandle:%p, sinkHandle:%p", pTaskInfo, sinkHandle);
+    QW_ERR_JRET(TSDB_CODE_QRY_APP_ERROR);
+  }
+
+  ctx.taskHandle = pTaskInfo;
+  ctx.sinkHandle = sinkHandle;
+
+  QW_ERR_JRET(qwExecTask(QW_FPARAMS(), &ctx, NULL));
+
+  QW_ERR_JRET(qwGetDeleteResFromSink(QW_FPARAMS(), &ctx, &pRsp->contLen, &pRsp->pCont, pRes));
+
+_return:
+
+  qwFreeTaskCtx(QW_FPARAMS(), &ctx);
+
+  QW_RET(TSDB_CODE_SUCCESS);
+}
+
 
 int32_t qWorkerInit(int8_t nodeType, int32_t nodeId, SQWorkerCfg *cfg, void **qWorkerMgmt, const SMsgCb *pMsgCb) {
   if (NULL == qWorkerMgmt || pMsgCb->mgmt == NULL) {
@@ -1007,6 +1095,7 @@ int32_t qWorkerGetStat(SReadHandle *handle, void *qWorkerMgmt, SQWorkerStat *pSt
   pStat->fetchProcessed = QW_STAT_GET(mgmt->stat.msgStat.fetchProcessed);
   pStat->dropProcessed = QW_STAT_GET(mgmt->stat.msgStat.dropProcessed);
   pStat->hbProcessed = QW_STAT_GET(mgmt->stat.msgStat.hbProcessed);
+  pStat->deleteProcessed = QW_STAT_GET(mgmt->stat.msgStat.deleteProcessed);
 
   pStat->numOfQueryInQueue = handle->pMsgCb->qsizeFp(handle->pMsgCb->mgmt, mgmt->nodeId, QUERY_QUEUE);
   pStat->numOfFetchInQueue = handle->pMsgCb->qsizeFp(handle->pMsgCb->mgmt, mgmt->nodeId, FETCH_QUEUE);
diff --git a/source/libs/scheduler/src/schJob.c b/source/libs/scheduler/src/schJob.c
index 7a0272a4d0009148662e8401e3fa544f4b46b5ee..09f2efc7e2bbe51b2389823f86f0c7391236e396 100644
--- a/source/libs/scheduler/src/schJob.c
+++ b/source/libs/scheduler/src/schJob.c
@@ -1476,7 +1476,7 @@ int32_t schExecJobImpl(void *pTrans, SArray *pNodeList, SQueryPlan *pDag, int64_
   SSchJob *pJob = NULL;
   SCH_ERR_RET(schInitJob(&pJob, pDag, pTrans, pNodeList, sql, pRes, startTs, sync));
 
-  qDebug("QID:0x%" PRIx64 " jobId:0x%"PRIx64 " started", pDag->queryId, pJob->refId);
+  qDebug("QID:0x%" PRIx64 " job refId 0x%"PRIx64 " started", pDag->queryId, pJob->refId);
   *job = pJob->refId;
 
   SCH_ERR_JRET(schLaunchJob(pJob));
diff --git a/source/libs/scheduler/src/schRemote.c b/source/libs/scheduler/src/schRemote.c
index bf51d8d631bfff7501ba40fea396b570e14c5c90..e40db48401426d8bf9cef8f53c3e5ecb9bc4b0c0 100644
--- a/source/libs/scheduler/src/schRemote.c
+++ b/source/libs/scheduler/src/schRemote.c
@@ -62,10 +62,11 @@ int32_t schValidateReceivedMsgType(SSchJob *pJob, SSchTask *pTask, int32_t msgTy
     case TDMT_VND_DROP_TABLE_RSP:
     case TDMT_VND_ALTER_TABLE_RSP:
     case TDMT_VND_SUBMIT_RSP:
+    case TDMT_VND_DELETE_RSP:
       break;
     default:
       SCH_TASK_ELOG("unknown rsp msg, type:%s, status:%s", TMSG_INFO(msgType), jobTaskStatusStr(taskStatus));
-      SCH_ERR_RET(TSDB_CODE_QRY_INVALID_INPUT);
+      SCH_ERR_RET(TSDB_CODE_INVALID_MSG);
   }
 
   if (lastMsgType != reqMsgType) {
@@ -227,6 +228,25 @@ int32_t schHandleResponseMsg(SSchJob *pJob, SSchTask *pTask, int32_t msgType, ch
 
       break;
     }
+    case TDMT_VND_DELETE_RSP: {
+      SCH_ERR_JRET(rspCode);
+
+      if (msg) {
+        SDecoder coder = {0};
+        SVDeleteRsp rsp = {0};
+        tDecoderInit(&coder, msg, msgSize);
+        tDecodeSVDeleteRsp(&coder, &rsp);
+
+        atomic_add_fetch_32(&pJob->resNumOfRows, rsp.affectedRows);
+        SCH_TASK_DLOG("delete succeed, affectedRows:%" PRId64, rsp.affectedRows);
+      }
+
+      taosMemoryFreeClear(msg);
+
+      SCH_ERR_RET(schProcessOnTaskSuccess(pJob, pTask));
+
+      break;
+    }    
     case TDMT_VND_QUERY_RSP: {
       SQueryTableRsp *rsp = (SQueryTableRsp *)msg;
 
@@ -411,6 +431,10 @@ int32_t schHandleQueryCallback(void *param, const SDataBuf *pMsg, int32_t code)
   return schHandleCallback(param, pMsg, TDMT_VND_QUERY_RSP, code);
 }
 
+int32_t schHandleDeleteCallback(void *param, const SDataBuf *pMsg, int32_t code) {
+  return schHandleCallback(param, pMsg, TDMT_VND_DELETE_RSP, code);
+}
+
 int32_t schHandleFetchCallback(void *param, const SDataBuf *pMsg, int32_t code) {
   return schHandleCallback(param, pMsg, TDMT_VND_FETCH_RSP, code);
 }
@@ -501,6 +525,9 @@ int32_t schGetCallbackFp(int32_t msgType, __async_send_cb_fn_t *fp) {
     case TDMT_VND_QUERY:
       *fp = schHandleQueryCallback;
       break;
+    case TDMT_VND_DELETE:
+      *fp = schHandleDeleteCallback;
+      break;
     case TDMT_VND_EXPLAIN:
       *fp = schHandleExplainCallback;
       break;
@@ -982,6 +1009,26 @@ int32_t schBuildAndSendMsg(SSchJob *pJob, SSchTask *pTask, SQueryNodeAddr *addr,
       break;
     }
 
+    case TDMT_VND_DELETE: {
+      SVDeleteReq req = {0};
+      req.header.vgId = addr->nodeId;
+      req.sId = schMgmt.sId;
+      req.queryId = pJob->queryId;
+      req.taskId = pTask->taskId;
+      req.phyLen = pTask->msgLen;
+      req.sqlLen = strlen(pJob->sql);
+      req.sql = (char*)pJob->sql;
+      req.msg = pTask->msg;
+      msgSize = tSerializeSVDeleteReq(NULL, 0, &req);
+      msg = taosMemoryCalloc(1, msgSize);
+      if (NULL == msg) {
+        SCH_TASK_ELOG("calloc %d failed", msgSize);
+        SCH_ERR_RET(TSDB_CODE_QRY_OUT_OF_MEMORY);
+      }
+
+      tSerializeSVDeleteReq(msg, msgSize, &req);
+      break;
+    }
     case TDMT_VND_QUERY: {
       SCH_ERR_RET(schMakeQueryRpcCtx(pJob, pTask, &rpcCtx));
 
diff --git a/source/libs/scheduler/src/schUtil.c b/source/libs/scheduler/src/schUtil.c
index 18398802dbd2ed0c251bd93a1fcafc56363d797e..38c03c74d9617f037591d52574a6b6bf04831d0c 100644
--- a/source/libs/scheduler/src/schUtil.c
+++ b/source/libs/scheduler/src/schUtil.c
@@ -257,6 +257,8 @@ void schFreeRpcCtx(SRpcCtx *pCtx) {
 
   taosHashCleanup(pCtx->args);
 
-  (*pCtx->freeFunc)(pCtx->brokenVal.val);
+  if (pCtx->freeFunc) {
+    (*pCtx->freeFunc)(pCtx->brokenVal.val);
+  }
 }