From 9cfb2c9bacdb0bd146a32c7ba836b93abe4d2fb9 Mon Sep 17 00:00:00 2001
From: Jeff Tao <jhtao@taosdata.com>
Date: Thu, 2 Jul 2020 13:16:00 +0000
Subject: [PATCH 01/19] try to avoid race condition

---
 src/rpc/src/rpcMain.c | 9 +++++++--
 1 file changed, 7 insertions(+), 2 deletions(-)

diff --git a/src/rpc/src/rpcMain.c b/src/rpc/src/rpcMain.c
index e4f364d3d3..86b41747c4 100644
--- a/src/rpc/src/rpcMain.c
+++ b/src/rpc/src/rpcMain.c
@@ -73,6 +73,7 @@ typedef struct {
   SRpcInfo *pRpc;       // associated SRpcInfo
   SRpcIpSet ipSet;      // ip list provided by app
   void     *ahandle;    // handle provided by app
+  void     *signature   // for validation
   struct SRpcConn *pConn; // pConn allocated
   char      msgType;    // message type
   uint8_t  *pCont;      // content provided by app
@@ -361,6 +362,7 @@ void rpcSendRequest(void *shandle, const SRpcIpSet *pIpSet, SRpcMsg *pMsg) {
   int contLen = rpcCompressRpcMsg(pMsg->pCont, pMsg->contLen);
   pContext = (SRpcReqContext *) (pMsg->pCont-sizeof(SRpcHead)-sizeof(SRpcReqContext));
   pContext->ahandle = pMsg->ahandle;
+  pContext->signature = pContext;
   pContext->pRpc = (SRpcInfo *)shandle;
   pContext->ipSet = *pIpSet;
   pContext->contLen = contLen;
@@ -527,11 +529,13 @@ int rpcReportProgress(void *handle, char *pCont, int contLen) {
   return code;
 }
 
-/* todo: cancel process may have race condition, pContext may have been released 
-   just before app calls the rpcCancelRequest */
 void rpcCancelRequest(void *handle) {
   SRpcReqContext *pContext = handle;
 
+  // signature is used to check if pContext is freed. 
+  // pContext may have been released just before app calls the rpcCancelRequest 
+  if (pContext->signature != pContext) return;
+
   if (pContext->pConn) {
     tDebug("%s, app trys to cancel request", pContext->pConn->info);
     rpcCloseConn(pContext->pConn);
@@ -1005,6 +1009,7 @@ static void *rpcProcessMsgFromPeer(SRecvInfo *pRecv) {
 static void rpcNotifyClient(SRpcReqContext *pContext, SRpcMsg *pMsg) {
   SRpcInfo       *pRpc = pContext->pRpc;
 
+  pContext->signature = NULL;
   pContext->pConn = NULL;
   if (pContext->pRsp) { 
     // for synchronous API

From 0c91ffae759e8a6b38b386bed1f63ed61dc898ea Mon Sep 17 00:00:00 2001
From: Shengliang Guan <slguan@taosdata.com>
Date: Sat, 4 Jul 2020 14:44:25 +0800
Subject: [PATCH 02/19] [TD-825] vnodehash may be null while close all vnodes

---
 src/dnode/src/dnodeMgmt.c         | 17 ++++++++++------
 src/inc/vnode.h                   |  4 +++-
 src/plugins/mqtt/src/mqttSystem.c |  2 +-
 src/vnode/src/vnodeMain.c         | 34 +++++++++++++++++++++++++------
 4 files changed, 43 insertions(+), 14 deletions(-)

diff --git a/src/dnode/src/dnodeMgmt.c b/src/dnode/src/dnodeMgmt.c
index 9cf024ba83..10eb77058b 100644
--- a/src/dnode/src/dnodeMgmt.c
+++ b/src/dnode/src/dnodeMgmt.c
@@ -176,6 +176,7 @@ void dnodeCleanupMgmt() {
   tsMgmtQset = NULL;
   tsMgmtQueue = NULL;
 
+  vnodeCleanupResources();
 }
 
 void dnodeDispatchToMgmtQueue(SRpcMsg *pMsg) {
@@ -242,8 +243,14 @@ static int32_t dnodeGetVnodeList(int32_t vnodeList[], int32_t *numOfVnodes) {
       int32_t vnode = atoi(de->d_name + 5);
       if (vnode == 0) continue;
 
-      vnodeList[*numOfVnodes] = vnode;
       (*numOfVnodes)++;
+
+      if (*numOfVnodes >= TSDB_MAX_VNODES) {
+        dError("vgId:%d, too many vnode directory in disk, exist:%d max:%d", vnode, *numOfVnodes, TSDB_MAX_VNODES);
+        continue;
+      } else {
+        vnodeList[*numOfVnodes - 1] = vnode;
+      }
     }
   }
   closedir(dir);
@@ -337,7 +344,7 @@ static int32_t dnodeOpenVnodes() {
 void dnodeStartStream() {
   int32_t vnodeList[TSDB_MAX_VNODES];
   int32_t numOfVnodes = 0;
-  int32_t status = dnodeGetVnodeList(vnodeList, &numOfVnodes);
+  int32_t status = vnodeGetVnodeList(vnodeList, &numOfVnodes);
 
   if (status != TSDB_CODE_SUCCESS) {
     dInfo("get dnode list failed");
@@ -352,15 +359,14 @@ void dnodeStartStream() {
 }
 
 static void dnodeCloseVnodes() {
-  int32_t *vnodeList = (int32_t *)malloc(sizeof(int32_t) * TSDB_MAX_VNODES);
+  int32_t vnodeList[TSDB_MAX_VNODES];
   int32_t numOfVnodes;
   int32_t status;
 
-  status = dnodeGetVnodeList(vnodeList, &numOfVnodes);
+  status = vnodeGetVnodeList(vnodeList, &numOfVnodes);
 
   if (status != TSDB_CODE_SUCCESS) {
     dInfo("get dnode list failed");
-    free(vnodeList);
     return;
   }
 
@@ -368,7 +374,6 @@ static void dnodeCloseVnodes() {
     vnodeClose(vnodeList[i]);
   }
 
-  free(vnodeList);
   dInfo("total vnodes:%d are all closed", numOfVnodes);
 }
 
diff --git a/src/inc/vnode.h b/src/inc/vnode.h
index 9f0c8cc241..971f341258 100644
--- a/src/inc/vnode.h
+++ b/src/inc/vnode.h
@@ -57,8 +57,10 @@ void*   vnodeGetWqueue(int32_t vgId);
 void*   vnodeGetWal(void *pVnode);
 
 int32_t vnodeProcessWrite(void *pVnode, int qtype, void *pHead, void *item);
-void    vnodeBuildStatusMsg(void * param);
+int32_t vnodeGetVnodeList(int32_t vnodeList[], int32_t *numOfVnodes);
+void    vnodeBuildStatusMsg(void *param);
 void    vnodeSetAccess(SDMVgroupAccess *pAccess, int32_t numOfVnodes);
+void    vnodeCleanupResources();
 
 int32_t vnodeProcessRead(void *pVnode, SReadMsg *pReadMsg);
 
diff --git a/src/plugins/mqtt/src/mqttSystem.c b/src/plugins/mqtt/src/mqttSystem.c
index 3df62f8bf4..2687106124 100644
--- a/src/plugins/mqtt/src/mqttSystem.c
+++ b/src/plugins/mqtt/src/mqttSystem.c
@@ -111,7 +111,7 @@ void mqttStopSystem() {
 }
 
 void mqttCleanUpSystem() {
-  mqttInfo("starting to clean  up mqtt");
+  mqttInfo("starting to cleanup mqtt");
   free(recntStatus.user_name);
   free(recntStatus.password);
   free(recntStatus.hostname);
diff --git a/src/vnode/src/vnodeMain.c b/src/vnode/src/vnodeMain.c
index 5eb78fda52..ec97cbc48c 100644
--- a/src/vnode/src/vnodeMain.c
+++ b/src/vnode/src/vnodeMain.c
@@ -69,6 +69,12 @@ static void vnodeInit() {
   }
 }
 
+void vnodeCleanupResources() {
+  taosHashCleanup(tsDnodeVnodesHash);
+  vnodeModuleInit = PTHREAD_ONCE_INIT;
+  tsDnodeVnodesHash = NULL;
+}
+
 int32_t vnodeCreate(SMDCreateVnodeMsg *pVnodeCfg) {
   int32_t code;
   pthread_once(&vnodeModuleInit, vnodeInit);
@@ -362,12 +368,6 @@ void vnodeRelease(void *pVnodeRaw) {
 
   int32_t count = atomic_sub_fetch_32(&tsOpennedVnodes, 1);
   vDebug("vgId:%d, vnode is released, vnodes:%d", vgId, count);
-
-  if (count <= 0) {
-    taosHashCleanup(tsDnodeVnodesHash);
-    vnodeModuleInit = PTHREAD_ONCE_INIT;
-    tsDnodeVnodesHash = NULL;
-  }
 }
 
 void *vnodeGetVnode(int32_t vgId) {
@@ -424,6 +424,28 @@ static void vnodeBuildVloadMsg(SVnodeObj *pVnode, SDMStatusMsg *pStatus) {
   pLoad->replica = pVnode->syncCfg.replica;  
 }
 
+int32_t vnodeGetVnodeList(int32_t vnodeList[], int32_t *numOfVnodes) {
+  if (tsDnodeVnodesHash == NULL) return TSDB_CODE_SUCCESS;
+
+  SHashMutableIterator *pIter = taosHashCreateIter(tsDnodeVnodesHash);
+  while (taosHashIterNext(pIter)) {
+    SVnodeObj **pVnode = taosHashIterGet(pIter);
+    if (pVnode == NULL) continue;
+    if (*pVnode == NULL) continue;
+
+    (*numOfVnodes)++;
+    if (*numOfVnodes >= TSDB_MAX_VNODES) {
+      vError("vgId:%d, too many open vnodes, exist:%d max:%d", (*pVnode)->vgId, *numOfVnodes, TSDB_MAX_VNODES);
+      continue;
+    } else {
+      vnodeList[*numOfVnodes - 1] = (*pVnode)->vgId;
+    }
+  }
+
+  taosHashDestroyIter(pIter);
+  return TSDB_CODE_SUCCESS;
+}
+
 void vnodeBuildStatusMsg(void *param) {
   SDMStatusMsg *pStatus = param;
   SHashMutableIterator *pIter = taosHashCreateIter(tsDnodeVnodesHash);

From a744e419f4d1e0ae237a4602c4a785197bf252e2 Mon Sep 17 00:00:00 2001
From: Hui Li <huili@taosdata.com>
Date: Sat, 4 Jul 2020 15:50:46 +0800
Subject: [PATCH 03/19] [TD-687]

---
 tests/script/jenkins/basic.txt                |   4 +
 tests/script/jenkins/unique.txt               |   4 +
 .../migrate/mn2_vn2_repl2_rmMnodeDir.sim      | 272 +++++++++++++++++
 .../migrate/mn2_vn2_repl2_rmMnodeVnodeDir.sim | 274 ++++++++++++++++++
 ..._repl2_rmMnodeVnodeDir_stopAll_starAll.sim | 210 ++++++++++++++
 .../migrate/mn2_vn2_repl2_rmVnodeDir.sim      | 272 +++++++++++++++++
 6 files changed, 1036 insertions(+)
 create mode 100644 tests/script/unique/migrate/mn2_vn2_repl2_rmMnodeDir.sim
 create mode 100644 tests/script/unique/migrate/mn2_vn2_repl2_rmMnodeVnodeDir.sim
 create mode 100644 tests/script/unique/migrate/mn2_vn2_repl2_rmMnodeVnodeDir_stopAll_starAll.sim
 create mode 100644 tests/script/unique/migrate/mn2_vn2_repl2_rmVnodeDir.sim

diff --git a/tests/script/jenkins/basic.txt b/tests/script/jenkins/basic.txt
index e2ebd9af63..3c4733a25b 100644
--- a/tests/script/jenkins/basic.txt
+++ b/tests/script/jenkins/basic.txt
@@ -365,3 +365,7 @@ cd ../../../debug; make
 ./test.sh -f unique/arbitrator/sync_replica3_dropDb.sim
 ./test.sh -f unique/arbitrator/sync_replica3_dropTable.sim
 
+./test.sh -f unique/migrate/mn2_vn2_repl2_rmMnodeDir.sim
+./test.sh -f unique/migrate/mn2_vn2_repl2_rmMnodeVnodeDir.sim
+./test.sh -f unique/migrate/mn2_vn2_repl2_rmMnodeVnodeDir_stopAll_starAll.sim
+./test.sh -f unique/migrate/mn2_vn2_repl2_rmVnodeDir.sim
diff --git a/tests/script/jenkins/unique.txt b/tests/script/jenkins/unique.txt
index afd0ea55c0..06edb8890a 100644
--- a/tests/script/jenkins/unique.txt
+++ b/tests/script/jenkins/unique.txt
@@ -133,3 +133,7 @@ cd ../../../debug; make
 ./test.sh -f unique/arbitrator/sync_replica3_dropDb.sim
 ./test.sh -f unique/arbitrator/sync_replica3_dropTable.sim
 
+./test.sh -f unique/migrate/mn2_vn2_repl2_rmMnodeDir.sim
+./test.sh -f unique/migrate/mn2_vn2_repl2_rmMnodeVnodeDir.sim
+./test.sh -f unique/migrate/mn2_vn2_repl2_rmMnodeVnodeDir_stopAll_starAll.sim
+./test.sh -f unique/migrate/mn2_vn2_repl2_rmVnodeDir.sim
diff --git a/tests/script/unique/migrate/mn2_vn2_repl2_rmMnodeDir.sim b/tests/script/unique/migrate/mn2_vn2_repl2_rmMnodeDir.sim
new file mode 100644
index 0000000000..e0b5e9b931
--- /dev/null
+++ b/tests/script/unique/migrate/mn2_vn2_repl2_rmMnodeDir.sim
@@ -0,0 +1,272 @@
+# Test case describe: dnode1/dnode2 include mnode and vnode roles
+# step 1: start dnode1/dnode2, and added into cluster
+# step 2: create db(repl = 2), table, insert data, 
+# step 4: stop dnode1, remove its mnode dir, and copy mnode dir of dnode2 to dnode1
+# step 5: restart dnode1, waiting sync end
+# step 6: stop dnode2, reset query cache, and query
+
+system sh/stop_dnodes.sh
+system sh/deploy.sh -n dnode1 -i 1
+system sh/deploy.sh -n dnode2 -i 2
+#system sh/deploy.sh -n dnode3 -i 3
+#system sh/deploy.sh -n dnode4 -i 4
+
+system sh/cfg.sh -n dnode1 -c numOfMnodes -v 2
+system sh/cfg.sh -n dnode2 -c numOfMnodes -v 2
+#system sh/cfg.sh -n dnode3 -c numOfMnodes -v 1
+#system sh/cfg.sh -n dnode4 -c numOfMnodes -v 1
+
+system sh/cfg.sh -n dnode1 -c walLevel -v 2
+system sh/cfg.sh -n dnode2 -c walLevel -v 2
+#system sh/cfg.sh -n dnode3 -c walLevel -v 2
+#system sh/cfg.sh -n dnode4 -c walLevel -v 2
+
+system sh/cfg.sh -n dnode1 -c balanceInterval -v 10
+system sh/cfg.sh -n dnode2 -c balanceInterval -v 10
+#system sh/cfg.sh -n dnode3 -c balanceInterval -v 10
+#system sh/cfg.sh -n dnode4 -c balanceInterval -v 10
+
+system sh/cfg.sh -n dnode1 -c numOfTotalVnodes -v 4
+system sh/cfg.sh -n dnode2 -c numOfTotalVnodes -v 4
+#system sh/cfg.sh -n dnode3 -c numOfTotalVnodes -v 4
+#system sh/cfg.sh -n dnode4 -c numOfTotalVnodes -v 4
+
+system sh/cfg.sh -n dnode1 -c alternativeRole -v 0
+system sh/cfg.sh -n dnode2 -c alternativeRole -v 0
+#system sh/cfg.sh -n dnode3 -c alternativeRole -v 2
+#system sh/cfg.sh -n dnode4 -c alternativeRole -v 2
+
+system sh/cfg.sh -n dnode1 -c maxtablesPerVnode -v 4
+system sh/cfg.sh -n dnode2 -c maxtablesPerVnode -v 4
+#system sh/cfg.sh -n dnode3 -c maxtablesPerVnode -v 4
+#system sh/cfg.sh -n dnode4 -c maxtablesPerVnode -v 4
+#system sh/cfg.sh -n dnode5 -c maxtablesPerVnode -v 4
+
+system sh/cfg.sh -n dnode1 -c arbitrator -v $arbitrator
+system sh/cfg.sh -n dnode2 -c arbitrator -v $arbitrator
+#system sh/cfg.sh -n dnode3 -c arbitrator -v $arbitrator
+
+print ============== step0: start tarbitrator
+system sh/exec_tarbitrator.sh -s start
+
+print ============== step1: start dnode1/dnode2 and add into cluster
+system sh/exec.sh -n dnode1 -s start
+system sh/exec.sh -n dnode2 -s start
+sleep 1000
+sql connect
+sleep 1000
+sql create dnode $hostname2
+sleep 1000
+
+print ============== step2: create database with replica 2, and create table, insert data
+$totalTableNum = 10
+$sleepTimer = 3000
+
+$db = db
+sql create database $db replica 2 cache 1
+sql use $db
+
+# create table , insert data
+$stb = stb
+sql create table $stb (ts timestamp, c1 double) tags(t1 int)
+$rowNum = 1200
+$tblNum = $totalTableNum
+$totalRows = 0
+$tsStart = 1577808000000  # 2020-01-01 00:00:00.000
+
+$i = 0
+while $i < $tblNum
+  $tb = tb . $i
+  sql create table $tb using $stb tags( $i )
+ 
+  $x = 0
+  while $x < $rowNum
+    $ts = $tsStart + $x
+    sql insert into $tb values ( $ts + 0a , $x ) ( $ts + 1a , $x ) ( $ts + 2a , $x ) ( $ts + 3a , $x ) ( $ts + 4a , $x ) ( $ts + 5a , $x ) ( $ts + 6a , $x ) ( $ts + 7a , $x ) ( $ts + 8a , $x ) ( $ts + 9a , $x ) ( $ts + 10a , $x ) ( $ts + 11a , $x ) ( $ts + 12a , $x ) ( $ts + 13a , $x ) ( $ts + 14a , $x ) ( $ts + 15a , $x ) ( $ts + 16a , $x ) ( $ts + 17a , $x ) ( $ts + 18a , $x ) ( $ts + 19a , $x ) ( $ts + 20a , $x ) ( $ts + 21a , $x ) ( $ts + 22a , $x ) ( $ts + 23a , $x ) ( $ts + 24a , $x ) ( $ts + 25a , $x ) ( $ts + 26a , $x ) ( $ts + 27a , $x ) ( $ts + 28a , $x ) ( $ts + 29a , $x ) ( $ts + 30a , $x ) ( $ts + 31a , $x ) ( $ts + 32a , $x ) ( $ts + 33a , $x ) ( $ts + 34a , $x ) ( $ts + 25a , $x ) ( $ts + 26a , $x ) ( $ts + 27a , $x ) ( $ts + 28a , $x ) ( $ts + 29a , $x ) ( $ts + 30a , $x ) ( $ts + 31a , $x ) ( $ts + 32a , $x ) ( $ts + 33a , $x ) ( $ts + 34a , $x ) ( $ts + 35a , $x ) ( $ts + 36a , $x ) ( $ts + 37a , $x ) ( $ts + 38a , $x ) ( $ts + 39a , $x ) ( $ts + 40a , $x ) ( $ts + 41a , $x ) ( $ts + 42a , $x ) ( $ts + 43a , $x ) ( $ts + 44a , $x ) ( $ts + 45a , $x ) ( $ts + 46a , $x ) ( $ts + 47a , $x ) ( $ts + 48a , $x ) ( $ts + 49a , $x ) ( $ts + 50a , $x ) ( $ts + 51a , $x ) ( $ts + 52a , $x ) ( $ts + 53a , $x ) ( $ts + 54a , $x ) ( $ts + 55a , $x ) ( $ts + 56a , $x ) ( $ts + 57a , $x ) ( $ts + 58a , $x ) ( $ts + 59a , $x ) 
+    $x = $x + 60
+  endw 
+  $totalRows = $totalRows + $x
+  print       info: inserted $x rows into $tb and totalRows: $totalRows
+  $i = $i + 1
+endw 
+
+sql select count(*) from $stb
+print rows:$rows    data00:$data00    totalRows:$totalRows
+if $rows != 1 then
+  return -1
+endi
+
+if $data00 != $totalRows then
+  return -1
+endi
+
+
+print ============== step3: insert old data(now-15d) and new data(now+15d), control data rows  in order to save in cache, not falling disc
+sql insert into $tb values ( now - 20d , -20 )
+sql insert into $tb values ( now - 40d , -40 )
+$totalRows = $totalRows + 2
+
+print ============== step4: stop dnode1
+system sh/exec.sh -n dnode1 -s stop -x SIGINT
+
+$loopCnt = 0
+wait_dnode1_offline:
+$loopCnt = $loopCnt + 1
+if $loopCnt == 10 then 
+  return -1
+endi
+
+sql show dnodes
+if $rows != 2 then
+  sleep 2000
+  goto wait_dnode1_offline
+endi
+print $data0_1  $data1_1  $data2_1  $data3_1  $data4_1
+print $data0_2  $data1_2  $data2_2  $data3_2  $data4_2
+$dnode1Status = $data4_1  
+$dnode2Status = $data4_2
+
+if $dnode1Status != offline then
+  sleep 2000
+  goto wait_dnode1_offline
+endi
+if $dnode2Status != ready then
+  sleep 2000
+  goto wait_dnode1_offline
+endi
+
+# check using select
+sql select count(*) from $stb
+print data00 $data00
+if $data00 != $totalRows then
+	return -1
+endi
+
+#sql show vgroups
+#print show vgroups:
+#print $data0_1  $data1_1  $data2_1  $data3_1  $data4_1  $data5_1  $data6_1  $data7_1  $data8_1  $data9_1
+#print $data0_2  $data1_2  $data2_2  $data3_2  $data4_2  $data5_2  $data6_2  $data7_2  $data8_2  $data9_2
+#print $data0_3  $data1_3  $data2_3  $data3_3  $data4_3  $data5_3  $data6_3  $data7_3  $data8_3  $data9_3
+
+print ============== step5: remove the mnode dir of dnode1, then copy the monde dir of dnode2
+system_content rm -rf ../../../sim/dnode1/data/mnode
+system_content cp -rf ../../../sim/dnode2/data/mnode ../../../sim/dnode1/data/
+
+print ============== step6: restart dnode1, waiting sync end
+system sh/exec.sh -n dnode1 -s start
+sleep 1000
+
+$loopCnt = 0
+wait_dnode1_ready:
+$loopCnt = $loopCnt + 1
+if $loopCnt == 20 then 
+  return -1
+endi
+
+sql show dnodes -x wait_dnode1_ready
+if $rows != 2 then
+  sleep 2000
+  goto wait_dnode1_ready
+endi
+print $data0_1  $data1_1  $data2_1  $data3_1  $data4_1
+print $data0_2  $data1_2  $data2_2  $data3_2  $data4_2
+$dnode1Status = $data4_1  
+$dnode2Status = $data4_2
+
+if $dnode1Status != ready then
+  sleep 2000
+  goto wait_dnode1_ready
+endi
+if $dnode2Status != ready then
+  sleep 2000
+  goto wait_dnode1_ready
+endi
+
+$loopCnt = 0
+wait_dnode1_vgroup_slave:
+$loopCnt = $loopCnt + 1
+if $loopCnt == 10 then 
+  return -1
+endi
+
+sql show vgroups
+if $rows != 3 then
+  sleep 2000
+  goto wait_dnode1_vgroup_slave
+endi
+print show vgroups:
+print $data0_1  $data1_1  $data2_1  $data3_1  $data4_1  $data5_1  $data6_1  $data7_1  $data8_1  $data9_1
+print $data0_2  $data1_2  $data2_2  $data3_2  $data4_2  $data5_2  $data6_2  $data7_2  $data8_2  $data9_2
+print $data0_3  $data1_3  $data2_3  $data3_3  $data4_3  $data5_3  $data6_3  $data7_3  $data8_3  $data9_3
+print $data0_4  $data1_4  $data2_4  $data3_4  $data4_4  $data5_4  $data6_4  $data7_4  $data8_4  $data9_4
+$d2v2status = $data4_4
+$d2v3status = $data4_2
+$d2v4status = $data4_3
+
+$d1v2status = $data7_4
+$d1v3status = $data7_2
+$d1v4status = $data7_3
+
+if $d2v2status != master then
+  sleep 2000
+  goto wait_dnode1_vgroup_slave
+endi
+if $d2v3status != master then
+  sleep 2000
+  goto wait_dnode1_vgroup_slave
+endi
+if $d2v4status != master then
+  sleep 2000
+  goto wait_dnode1_vgroup_slave
+endi
+
+if $d1v2status != slave then
+  sleep 2000
+  goto wait_dnode1_vgroup_slave
+endi
+if $d1v3status != slave then
+  sleep 2000
+  goto wait_dnode1_vgroup_slave
+endi
+if $d1v4status != slave then
+  sleep 2000
+  goto wait_dnode1_vgroup_slave
+endi
+
+print ============== step7: stop dnode2
+system sh/exec.sh -n dnode2 -s stop -x SIGINT
+
+$loopCnt = 0
+wait_dnode2_offline:
+$loopCnt = $loopCnt + 1
+if $loopCnt == 10 then 
+  return -1
+endi
+
+sql show dnodes
+if $rows != 2 then
+  sleep 2000
+  goto wait_dnode2_offline
+endi
+print $data0_1  $data1_1  $data2_1  $data3_1  $data4_1
+print $data0_2  $data1_2  $data2_2  $data3_2  $data4_2
+$dnode1Status = $data4_1  
+$dnode2Status = $data4_2
+
+if $dnode1Status != ready then
+  sleep 2000
+  goto wait_dnode2_offline
+endi
+if $dnode2Status != offline then
+  sleep 2000
+  goto wait_dnode2_offline
+endi
+
+sql reset query cache 
+
+# check using select
+sql select count(*) from $stb
+print data00 $data00
+if $data00 != $totalRows then
+	return -1
+endi
\ No newline at end of file
diff --git a/tests/script/unique/migrate/mn2_vn2_repl2_rmMnodeVnodeDir.sim b/tests/script/unique/migrate/mn2_vn2_repl2_rmMnodeVnodeDir.sim
new file mode 100644
index 0000000000..ae7fc6af17
--- /dev/null
+++ b/tests/script/unique/migrate/mn2_vn2_repl2_rmMnodeVnodeDir.sim
@@ -0,0 +1,274 @@
+# Test case describe: dnode1/dnode2 include mnode and vnode roles
+# step 1: start dnode1/dnode2, and added into cluster
+# step 2: create db(repl = 2), table, insert data, 
+# step 4: stop dnode1, remove its mnode and vnode dir, and copy mnode and vnode dir of dnode2 to dnode1
+# step 5: restart dnode1, waiting sync end
+# step 6: stop dnode2, reset query cache, and query
+
+system sh/stop_dnodes.sh
+system sh/deploy.sh -n dnode1 -i 1
+system sh/deploy.sh -n dnode2 -i 2
+#system sh/deploy.sh -n dnode3 -i 3
+#system sh/deploy.sh -n dnode4 -i 4
+
+system sh/cfg.sh -n dnode1 -c numOfMnodes -v 2
+system sh/cfg.sh -n dnode2 -c numOfMnodes -v 2
+#system sh/cfg.sh -n dnode3 -c numOfMnodes -v 1
+#system sh/cfg.sh -n dnode4 -c numOfMnodes -v 1
+
+system sh/cfg.sh -n dnode1 -c walLevel -v 2
+system sh/cfg.sh -n dnode2 -c walLevel -v 2
+#system sh/cfg.sh -n dnode3 -c walLevel -v 2
+#system sh/cfg.sh -n dnode4 -c walLevel -v 2
+
+system sh/cfg.sh -n dnode1 -c balanceInterval -v 10
+system sh/cfg.sh -n dnode2 -c balanceInterval -v 10
+#system sh/cfg.sh -n dnode3 -c balanceInterval -v 10
+#system sh/cfg.sh -n dnode4 -c balanceInterval -v 10
+
+system sh/cfg.sh -n dnode1 -c numOfTotalVnodes -v 4
+system sh/cfg.sh -n dnode2 -c numOfTotalVnodes -v 4
+#system sh/cfg.sh -n dnode3 -c numOfTotalVnodes -v 4
+#system sh/cfg.sh -n dnode4 -c numOfTotalVnodes -v 4
+
+system sh/cfg.sh -n dnode1 -c alternativeRole -v 0
+system sh/cfg.sh -n dnode2 -c alternativeRole -v 0
+#system sh/cfg.sh -n dnode3 -c alternativeRole -v 2
+#system sh/cfg.sh -n dnode4 -c alternativeRole -v 2
+
+system sh/cfg.sh -n dnode1 -c maxtablesPerVnode -v 4
+system sh/cfg.sh -n dnode2 -c maxtablesPerVnode -v 4
+#system sh/cfg.sh -n dnode3 -c maxtablesPerVnode -v 4
+#system sh/cfg.sh -n dnode4 -c maxtablesPerVnode -v 4
+#system sh/cfg.sh -n dnode5 -c maxtablesPerVnode -v 4
+
+system sh/cfg.sh -n dnode1 -c arbitrator -v $arbitrator
+system sh/cfg.sh -n dnode2 -c arbitrator -v $arbitrator
+#system sh/cfg.sh -n dnode3 -c arbitrator -v $arbitrator
+
+print ============== step0: start tarbitrator
+system sh/exec_tarbitrator.sh -s start
+
+print ============== step1: start dnode1/dnode2 and add into cluster
+system sh/exec.sh -n dnode1 -s start
+system sh/exec.sh -n dnode2 -s start
+sleep 1000
+sql connect
+sleep 1000
+sql create dnode $hostname2
+sleep 1000
+
+print ============== step2: create database with replica 2, and create table, insert data
+$totalTableNum = 10
+$sleepTimer = 3000
+
+$db = db
+sql create database $db replica 2 cache 1
+sql use $db
+
+# create table , insert data
+$stb = stb
+sql create table $stb (ts timestamp, c1 double) tags(t1 int)
+$rowNum = 1200
+$tblNum = $totalTableNum
+$totalRows = 0
+$tsStart = 1577808000000  # 2020-01-01 00:00:00.000
+
+$i = 0
+while $i < $tblNum
+  $tb = tb . $i
+  sql create table $tb using $stb tags( $i )
+ 
+  $x = 0
+  while $x < $rowNum
+    $ts = $tsStart + $x
+    sql insert into $tb values ( $ts + 0a , $x ) ( $ts + 1a , $x ) ( $ts + 2a , $x ) ( $ts + 3a , $x ) ( $ts + 4a , $x ) ( $ts + 5a , $x ) ( $ts + 6a , $x ) ( $ts + 7a , $x ) ( $ts + 8a , $x ) ( $ts + 9a , $x ) ( $ts + 10a , $x ) ( $ts + 11a , $x ) ( $ts + 12a , $x ) ( $ts + 13a , $x ) ( $ts + 14a , $x ) ( $ts + 15a , $x ) ( $ts + 16a , $x ) ( $ts + 17a , $x ) ( $ts + 18a , $x ) ( $ts + 19a , $x ) ( $ts + 20a , $x ) ( $ts + 21a , $x ) ( $ts + 22a , $x ) ( $ts + 23a , $x ) ( $ts + 24a , $x ) ( $ts + 25a , $x ) ( $ts + 26a , $x ) ( $ts + 27a , $x ) ( $ts + 28a , $x ) ( $ts + 29a , $x ) ( $ts + 30a , $x ) ( $ts + 31a , $x ) ( $ts + 32a , $x ) ( $ts + 33a , $x ) ( $ts + 34a , $x ) ( $ts + 25a , $x ) ( $ts + 26a , $x ) ( $ts + 27a , $x ) ( $ts + 28a , $x ) ( $ts + 29a , $x ) ( $ts + 30a , $x ) ( $ts + 31a , $x ) ( $ts + 32a , $x ) ( $ts + 33a , $x ) ( $ts + 34a , $x ) ( $ts + 35a , $x ) ( $ts + 36a , $x ) ( $ts + 37a , $x ) ( $ts + 38a , $x ) ( $ts + 39a , $x ) ( $ts + 40a , $x ) ( $ts + 41a , $x ) ( $ts + 42a , $x ) ( $ts + 43a , $x ) ( $ts + 44a , $x ) ( $ts + 45a , $x ) ( $ts + 46a , $x ) ( $ts + 47a , $x ) ( $ts + 48a , $x ) ( $ts + 49a , $x ) ( $ts + 50a , $x ) ( $ts + 51a , $x ) ( $ts + 52a , $x ) ( $ts + 53a , $x ) ( $ts + 54a , $x ) ( $ts + 55a , $x ) ( $ts + 56a , $x ) ( $ts + 57a , $x ) ( $ts + 58a , $x ) ( $ts + 59a , $x ) 
+    $x = $x + 60
+  endw 
+  $totalRows = $totalRows + $x
+  print       info: inserted $x rows into $tb and totalRows: $totalRows
+  $i = $i + 1
+endw 
+
+sql select count(*) from $stb
+print rows:$rows    data00:$data00    totalRows:$totalRows
+if $rows != 1 then
+  return -1
+endi
+
+if $data00 != $totalRows then
+  return -1
+endi
+
+
+print ============== step3: insert old data(now-15d) and new data(now+15d), control data rows  in order to save in cache, not falling disc
+sql insert into $tb values ( now - 20d , -20 )
+sql insert into $tb values ( now - 40d , -40 )
+$totalRows = $totalRows + 2
+
+print ============== step4: stop dnode1
+system sh/exec.sh -n dnode1 -s stop -x SIGINT
+
+$loopCnt = 0
+wait_dnode1_offline:
+$loopCnt = $loopCnt + 1
+if $loopCnt == 10 then 
+  return -1
+endi
+
+sql show dnodes
+if $rows != 2 then
+  sleep 2000
+  goto wait_dnode1_offline
+endi
+print $data0_1  $data1_1  $data2_1  $data3_1  $data4_1
+print $data0_2  $data1_2  $data2_2  $data3_2  $data4_2
+$dnode1Status = $data4_1  
+$dnode2Status = $data4_2
+
+if $dnode1Status != offline then
+  sleep 2000
+  goto wait_dnode1_offline
+endi
+if $dnode2Status != ready then
+  sleep 2000
+  goto wait_dnode1_offline
+endi
+
+# check using select
+sql select count(*) from $stb
+print data00 $data00
+if $data00 != $totalRows then
+	return -1
+endi
+
+#sql show vgroups
+#print show vgroups:
+#print $data0_1  $data1_1  $data2_1  $data3_1  $data4_1  $data5_1  $data6_1  $data7_1  $data8_1  $data9_1
+#print $data0_2  $data1_2  $data2_2  $data3_2  $data4_2  $data5_2  $data6_2  $data7_2  $data8_2  $data9_2
+#print $data0_3  $data1_3  $data2_3  $data3_3  $data4_3  $data5_3  $data6_3  $data7_3  $data8_3  $data9_3
+
+print ============== step5: remove the mnode dir of dnode1, then copy the monde dir of dnode2
+system_content rm -rf ../../../sim/dnode1/data/vnode
+system_content rm -rf ../../../sim/dnode1/data/mnode
+system_content cp -rf ../../../sim/dnode2/data/vnode ../../../sim/dnode1/data/
+system_content cp -rf ../../../sim/dnode2/data/mnode ../../../sim/dnode1/data/
+
+print ============== step6: restart dnode1, waiting sync end
+system sh/exec.sh -n dnode1 -s start
+sleep 1000
+
+$loopCnt = 0
+wait_dnode1_ready:
+$loopCnt = $loopCnt + 1
+if $loopCnt == 20 then 
+  return -1
+endi
+
+sql show dnodes -x wait_dnode1_ready
+if $rows != 2 then
+  sleep 2000
+  goto wait_dnode1_ready
+endi
+print $data0_1  $data1_1  $data2_1  $data3_1  $data4_1
+print $data0_2  $data1_2  $data2_2  $data3_2  $data4_2
+$dnode1Status = $data4_1  
+$dnode2Status = $data4_2
+
+if $dnode1Status != ready then
+  sleep 2000
+  goto wait_dnode1_ready
+endi
+if $dnode2Status != ready then
+  sleep 2000
+  goto wait_dnode1_ready
+endi
+
+$loopCnt = 0
+wait_dnode1_vgroup_slave:
+$loopCnt = $loopCnt + 1
+if $loopCnt == 10 then 
+  return -1
+endi
+
+sql show vgroups
+if $rows != 3 then
+  sleep 2000
+  goto wait_dnode1_vgroup_slave
+endi
+print show vgroups:
+print $data0_1  $data1_1  $data2_1  $data3_1  $data4_1  $data5_1  $data6_1  $data7_1  $data8_1  $data9_1
+print $data0_2  $data1_2  $data2_2  $data3_2  $data4_2  $data5_2  $data6_2  $data7_2  $data8_2  $data9_2
+print $data0_3  $data1_3  $data2_3  $data3_3  $data4_3  $data5_3  $data6_3  $data7_3  $data8_3  $data9_3
+print $data0_4  $data1_4  $data2_4  $data3_4  $data4_4  $data5_4  $data6_4  $data7_4  $data8_4  $data9_4
+$d2v2status = $data4_4
+$d2v3status = $data4_2
+$d2v4status = $data4_3
+
+$d1v2status = $data7_4
+$d1v3status = $data7_2
+$d1v4status = $data7_3
+
+if $d2v2status != master then
+  sleep 2000
+  goto wait_dnode1_vgroup_slave
+endi
+if $d2v3status != master then
+  sleep 2000
+  goto wait_dnode1_vgroup_slave
+endi
+if $d2v4status != master then
+  sleep 2000
+  goto wait_dnode1_vgroup_slave
+endi
+
+if $d1v2status != slave then
+  sleep 2000
+  goto wait_dnode1_vgroup_slave
+endi
+if $d1v3status != slave then
+  sleep 2000
+  goto wait_dnode1_vgroup_slave
+endi
+if $d1v4status != slave then
+  sleep 2000
+  goto wait_dnode1_vgroup_slave
+endi
+
+print ============== step7: stop dnode2
+system sh/exec.sh -n dnode2 -s stop -x SIGINT
+
+$loopCnt = 0
+wait_dnode2_offline:
+$loopCnt = $loopCnt + 1
+if $loopCnt == 10 then 
+  return -1
+endi
+
+sql show dnodes
+if $rows != 2 then
+  sleep 2000
+  goto wait_dnode2_offline
+endi
+print $data0_1  $data1_1  $data2_1  $data3_1  $data4_1
+print $data0_2  $data1_2  $data2_2  $data3_2  $data4_2
+$dnode1Status = $data4_1  
+$dnode2Status = $data4_2
+
+if $dnode1Status != ready then
+  sleep 2000
+  goto wait_dnode2_offline
+endi
+if $dnode2Status != offline then
+  sleep 2000
+  goto wait_dnode2_offline
+endi
+
+sql reset query cache 
+
+# check using select
+sql select count(*) from $stb
+print data00 $data00
+if $data00 != $totalRows then
+	return -1
+endi
\ No newline at end of file
diff --git a/tests/script/unique/migrate/mn2_vn2_repl2_rmMnodeVnodeDir_stopAll_starAll.sim b/tests/script/unique/migrate/mn2_vn2_repl2_rmMnodeVnodeDir_stopAll_starAll.sim
new file mode 100644
index 0000000000..dc9bc62696
--- /dev/null
+++ b/tests/script/unique/migrate/mn2_vn2_repl2_rmMnodeVnodeDir_stopAll_starAll.sim
@@ -0,0 +1,210 @@
+# Test case describe: dnode1/dnode2 include mnode and vnode roles
+# step 1: start dnode1/dnode2, and added into cluster
+# step 2: create db(repl = 2), table, insert data, 
+# step 4: stop dnode1, remove its mnode and vnode dir, and copy mnode and vnode dir of dnode2 to dnode1
+# step 5: restart dnode1, waiting sync end
+# step 6: stop dnode2, reset query cache, and query
+
+system sh/stop_dnodes.sh
+system sh/deploy.sh -n dnode1 -i 1
+system sh/deploy.sh -n dnode2 -i 2
+#system sh/deploy.sh -n dnode3 -i 3
+#system sh/deploy.sh -n dnode4 -i 4
+
+system sh/cfg.sh -n dnode1 -c numOfMnodes -v 2
+system sh/cfg.sh -n dnode2 -c numOfMnodes -v 2
+#system sh/cfg.sh -n dnode3 -c numOfMnodes -v 1
+#system sh/cfg.sh -n dnode4 -c numOfMnodes -v 1
+
+system sh/cfg.sh -n dnode1 -c walLevel -v 2
+system sh/cfg.sh -n dnode2 -c walLevel -v 2
+#system sh/cfg.sh -n dnode3 -c walLevel -v 2
+#system sh/cfg.sh -n dnode4 -c walLevel -v 2
+
+system sh/cfg.sh -n dnode1 -c balanceInterval -v 10
+system sh/cfg.sh -n dnode2 -c balanceInterval -v 10
+#system sh/cfg.sh -n dnode3 -c balanceInterval -v 10
+#system sh/cfg.sh -n dnode4 -c balanceInterval -v 10
+
+system sh/cfg.sh -n dnode1 -c numOfTotalVnodes -v 4
+system sh/cfg.sh -n dnode2 -c numOfTotalVnodes -v 4
+#system sh/cfg.sh -n dnode3 -c numOfTotalVnodes -v 4
+#system sh/cfg.sh -n dnode4 -c numOfTotalVnodes -v 4
+
+system sh/cfg.sh -n dnode1 -c alternativeRole -v 0
+system sh/cfg.sh -n dnode2 -c alternativeRole -v 0
+#system sh/cfg.sh -n dnode3 -c alternativeRole -v 2
+#system sh/cfg.sh -n dnode4 -c alternativeRole -v 2
+
+system sh/cfg.sh -n dnode1 -c maxtablesPerVnode -v 4
+system sh/cfg.sh -n dnode2 -c maxtablesPerVnode -v 4
+#system sh/cfg.sh -n dnode3 -c maxtablesPerVnode -v 4
+#system sh/cfg.sh -n dnode4 -c maxtablesPerVnode -v 4
+#system sh/cfg.sh -n dnode5 -c maxtablesPerVnode -v 4
+
+system sh/cfg.sh -n dnode1 -c arbitrator -v $arbitrator
+system sh/cfg.sh -n dnode2 -c arbitrator -v $arbitrator
+#system sh/cfg.sh -n dnode3 -c arbitrator -v $arbitrator
+
+print ============== step0: start tarbitrator
+system sh/exec_tarbitrator.sh -s start
+
+print ============== step1: start dnode1/dnode2 and add into cluster
+system sh/exec.sh -n dnode1 -s start
+system sh/exec.sh -n dnode2 -s start
+sleep 1000
+sql connect
+sleep 1000
+sql create dnode $hostname2
+sleep 1000
+
+print ============== step2: create database with replica 2, and create table, insert data
+$totalTableNum = 10
+$sleepTimer = 3000
+
+$db = db
+sql create database $db replica 2 cache 1
+sql use $db
+
+# create table , insert data
+$stb = stb
+sql create table $stb (ts timestamp, c1 double) tags(t1 int)
+$rowNum = 1200
+$tblNum = $totalTableNum
+$totalRows = 0
+$tsStart = 1577808000000  # 2020-01-01 00:00:00.000
+
+$i = 0
+while $i < $tblNum
+  $tb = tb . $i
+  sql create table $tb using $stb tags( $i )
+ 
+  $x = 0
+  while $x < $rowNum
+    $ts = $tsStart + $x
+    sql insert into $tb values ( $ts + 0a , $x ) ( $ts + 1a , $x ) ( $ts + 2a , $x ) ( $ts + 3a , $x ) ( $ts + 4a , $x ) ( $ts + 5a , $x ) ( $ts + 6a , $x ) ( $ts + 7a , $x ) ( $ts + 8a , $x ) ( $ts + 9a , $x ) ( $ts + 10a , $x ) ( $ts + 11a , $x ) ( $ts + 12a , $x ) ( $ts + 13a , $x ) ( $ts + 14a , $x ) ( $ts + 15a , $x ) ( $ts + 16a , $x ) ( $ts + 17a , $x ) ( $ts + 18a , $x ) ( $ts + 19a , $x ) ( $ts + 20a , $x ) ( $ts + 21a , $x ) ( $ts + 22a , $x ) ( $ts + 23a , $x ) ( $ts + 24a , $x ) ( $ts + 25a , $x ) ( $ts + 26a , $x ) ( $ts + 27a , $x ) ( $ts + 28a , $x ) ( $ts + 29a , $x ) ( $ts + 30a , $x ) ( $ts + 31a , $x ) ( $ts + 32a , $x ) ( $ts + 33a , $x ) ( $ts + 34a , $x ) ( $ts + 25a , $x ) ( $ts + 26a , $x ) ( $ts + 27a , $x ) ( $ts + 28a , $x ) ( $ts + 29a , $x ) ( $ts + 30a , $x ) ( $ts + 31a , $x ) ( $ts + 32a , $x ) ( $ts + 33a , $x ) ( $ts + 34a , $x ) ( $ts + 35a , $x ) ( $ts + 36a , $x ) ( $ts + 37a , $x ) ( $ts + 38a , $x ) ( $ts + 39a , $x ) ( $ts + 40a , $x ) ( $ts + 41a , $x ) ( $ts + 42a , $x ) ( $ts + 43a , $x ) ( $ts + 44a , $x ) ( $ts + 45a , $x ) ( $ts + 46a , $x ) ( $ts + 47a , $x ) ( $ts + 48a , $x ) ( $ts + 49a , $x ) ( $ts + 50a , $x ) ( $ts + 51a , $x ) ( $ts + 52a , $x ) ( $ts + 53a , $x ) ( $ts + 54a , $x ) ( $ts + 55a , $x ) ( $ts + 56a , $x ) ( $ts + 57a , $x ) ( $ts + 58a , $x ) ( $ts + 59a , $x ) 
+    $x = $x + 60
+  endw 
+  $totalRows = $totalRows + $x
+  print       info: inserted $x rows into $tb and totalRows: $totalRows
+  $i = $i + 1
+endw 
+
+sql select count(*) from $stb
+print rows:$rows    data00:$data00    totalRows:$totalRows
+if $rows != 1 then
+  return -1
+endi
+
+if $data00 != $totalRows then
+  return -1
+endi
+
+
+print ============== step3: insert old data(now-15d) and new data(now+15d), control data rows  in order to save in cache, not falling disc
+sql insert into $tb values ( now - 20d , -20 )
+sql insert into $tb values ( now - 40d , -40 )
+$totalRows = $totalRows + 2
+
+print ============== step4: stop dnode1
+system sh/exec.sh -n dnode1 -s stop -x SIGINT
+system sh/exec.sh -n dnode2 -s stop -x SIGINT
+
+print ============== step5: remove the mnode dir of dnode1, then copy the monde dir of dnode2
+system_content rm -rf ../../../sim/dnode1/data/vnode
+system_content rm -rf ../../../sim/dnode1/data/mnode
+system_content cp -rf ../../../sim/dnode2/data/vnode ../../../sim/dnode1/data/
+system_content cp -rf ../../../sim/dnode2/data/mnode ../../../sim/dnode1/data/
+
+print ============== step6: restart dnode1/dnode2
+system sh/exec.sh -n dnode1 -s start
+system sh/exec.sh -n dnode2 -s start
+sleep 1000
+sql connect
+sql use $db
+
+$loopCnt = 0
+wait_dnode1_ready:
+$loopCnt = $loopCnt + 1
+if $loopCnt == 20 then 
+  return -1
+endi
+
+sql show dnodes -x wait_dnode1_ready
+if $rows != 2 then
+  sleep 2000
+  goto wait_dnode1_ready
+endi
+print $data0_1  $data1_1  $data2_1  $data3_1  $data4_1
+print $data0_2  $data1_2  $data2_2  $data3_2  $data4_2
+$dnode1Status = $data4_1  
+$dnode2Status = $data4_2
+
+if $dnode1Status != ready then
+  sleep 2000
+  goto wait_dnode1_ready
+endi
+if $dnode2Status != ready then
+  sleep 2000
+  goto wait_dnode1_ready
+endi
+
+$loopCnt = 0
+wait_dnode1_vgroup_slave:
+$loopCnt = $loopCnt + 1
+if $loopCnt == 10 then 
+  return -1
+endi
+
+sql show vgroups
+if $rows != 3 then
+  sleep 2000
+  goto wait_dnode1_vgroup_slave
+endi
+print show vgroups:
+print $data0_1  $data1_1  $data2_1  $data3_1  $data4_1  $data5_1  $data6_1  $data7_1  $data8_1  $data9_1
+print $data0_2  $data1_2  $data2_2  $data3_2  $data4_2  $data5_2  $data6_2  $data7_2  $data8_2  $data9_2
+print $data0_3  $data1_3  $data2_3  $data3_3  $data4_3  $data5_3  $data6_3  $data7_3  $data8_3  $data9_3
+print $data0_4  $data1_4  $data2_4  $data3_4  $data4_4  $data5_4  $data6_4  $data7_4  $data8_4  $data9_4
+$d2v2status = $data4_4
+$d2v3status = $data4_2
+$d2v4status = $data4_3
+
+$d1v2status = $data7_4
+$d1v3status = $data7_2
+$d1v4status = $data7_3
+
+if $d2v2status != master then
+  sleep 2000
+  goto wait_dnode1_vgroup_slave
+endi
+if $d2v3status != master then
+  sleep 2000
+  goto wait_dnode1_vgroup_slave
+endi
+if $d2v4status != master then
+  sleep 2000
+  goto wait_dnode1_vgroup_slave
+endi
+
+if $d1v2status != slave then
+  sleep 2000
+  goto wait_dnode1_vgroup_slave
+endi
+if $d1v3status != slave then
+  sleep 2000
+  goto wait_dnode1_vgroup_slave
+endi
+if $d1v4status != slave then
+  sleep 2000
+  goto wait_dnode1_vgroup_slave
+endi
+
+sql reset query cache 
+
+# check using select
+sql select count(*) from $stb
+print data00 $data00
+if $data00 != $totalRows then
+	return -1
+endi
\ No newline at end of file
diff --git a/tests/script/unique/migrate/mn2_vn2_repl2_rmVnodeDir.sim b/tests/script/unique/migrate/mn2_vn2_repl2_rmVnodeDir.sim
new file mode 100644
index 0000000000..b754dc7a49
--- /dev/null
+++ b/tests/script/unique/migrate/mn2_vn2_repl2_rmVnodeDir.sim
@@ -0,0 +1,272 @@
+# Test case describe: dnode1/dnode2 include mnode and vnode roles
+# step 1: start dnode1/dnode2, and added into cluster
+# step 2: create db(repl = 2), table, insert data, 
+# step 4: stop dnode1, remove its vnode dir, and copy vnode dir of dnode2 to dnode1
+# step 5: restart dnode1, waiting sync end
+# step 6: stop dnode2, reset query cache, and query
+
+system sh/stop_dnodes.sh
+system sh/deploy.sh -n dnode1 -i 1
+system sh/deploy.sh -n dnode2 -i 2
+#system sh/deploy.sh -n dnode3 -i 3
+#system sh/deploy.sh -n dnode4 -i 4
+
+system sh/cfg.sh -n dnode1 -c numOfMnodes -v 2
+system sh/cfg.sh -n dnode2 -c numOfMnodes -v 2
+#system sh/cfg.sh -n dnode3 -c numOfMnodes -v 1
+#system sh/cfg.sh -n dnode4 -c numOfMnodes -v 1
+
+system sh/cfg.sh -n dnode1 -c walLevel -v 2
+system sh/cfg.sh -n dnode2 -c walLevel -v 2
+#system sh/cfg.sh -n dnode3 -c walLevel -v 2
+#system sh/cfg.sh -n dnode4 -c walLevel -v 2
+
+system sh/cfg.sh -n dnode1 -c balanceInterval -v 10
+system sh/cfg.sh -n dnode2 -c balanceInterval -v 10
+#system sh/cfg.sh -n dnode3 -c balanceInterval -v 10
+#system sh/cfg.sh -n dnode4 -c balanceInterval -v 10
+
+system sh/cfg.sh -n dnode1 -c numOfTotalVnodes -v 4
+system sh/cfg.sh -n dnode2 -c numOfTotalVnodes -v 4
+#system sh/cfg.sh -n dnode3 -c numOfTotalVnodes -v 4
+#system sh/cfg.sh -n dnode4 -c numOfTotalVnodes -v 4
+
+system sh/cfg.sh -n dnode1 -c alternativeRole -v 0
+system sh/cfg.sh -n dnode2 -c alternativeRole -v 0
+#system sh/cfg.sh -n dnode3 -c alternativeRole -v 2
+#system sh/cfg.sh -n dnode4 -c alternativeRole -v 2
+
+system sh/cfg.sh -n dnode1 -c maxtablesPerVnode -v 4
+system sh/cfg.sh -n dnode2 -c maxtablesPerVnode -v 4
+#system sh/cfg.sh -n dnode3 -c maxtablesPerVnode -v 4
+#system sh/cfg.sh -n dnode4 -c maxtablesPerVnode -v 4
+#system sh/cfg.sh -n dnode5 -c maxtablesPerVnode -v 4
+
+system sh/cfg.sh -n dnode1 -c arbitrator -v $arbitrator
+system sh/cfg.sh -n dnode2 -c arbitrator -v $arbitrator
+#system sh/cfg.sh -n dnode3 -c arbitrator -v $arbitrator
+
+print ============== step0: start tarbitrator
+system sh/exec_tarbitrator.sh -s start
+
+print ============== step1: start dnode1/dnode2 and add into cluster
+system sh/exec.sh -n dnode1 -s start
+system sh/exec.sh -n dnode2 -s start
+sleep 1000
+sql connect
+sleep 1000
+sql create dnode $hostname2
+sleep 1000
+
+print ============== step2: create database with replica 2, and create table, insert data
+$totalTableNum = 10
+$sleepTimer = 3000
+
+$db = db
+sql create database $db replica 2 cache 1
+sql use $db
+
+# create table , insert data
+$stb = stb
+sql create table $stb (ts timestamp, c1 double) tags(t1 int)
+$rowNum = 1200
+$tblNum = $totalTableNum
+$totalRows = 0
+$tsStart = 1577808000000  # 2020-01-01 00:00:00.000
+
+$i = 0
+while $i < $tblNum
+  $tb = tb . $i
+  sql create table $tb using $stb tags( $i )
+ 
+  $x = 0
+  while $x < $rowNum
+    $ts = $tsStart + $x
+    sql insert into $tb values ( $ts + 0a , $x ) ( $ts + 1a , $x ) ( $ts + 2a , $x ) ( $ts + 3a , $x ) ( $ts + 4a , $x ) ( $ts + 5a , $x ) ( $ts + 6a , $x ) ( $ts + 7a , $x ) ( $ts + 8a , $x ) ( $ts + 9a , $x ) ( $ts + 10a , $x ) ( $ts + 11a , $x ) ( $ts + 12a , $x ) ( $ts + 13a , $x ) ( $ts + 14a , $x ) ( $ts + 15a , $x ) ( $ts + 16a , $x ) ( $ts + 17a , $x ) ( $ts + 18a , $x ) ( $ts + 19a , $x ) ( $ts + 20a , $x ) ( $ts + 21a , $x ) ( $ts + 22a , $x ) ( $ts + 23a , $x ) ( $ts + 24a , $x ) ( $ts + 25a , $x ) ( $ts + 26a , $x ) ( $ts + 27a , $x ) ( $ts + 28a , $x ) ( $ts + 29a , $x ) ( $ts + 30a , $x ) ( $ts + 31a , $x ) ( $ts + 32a , $x ) ( $ts + 33a , $x ) ( $ts + 34a , $x ) ( $ts + 25a , $x ) ( $ts + 26a , $x ) ( $ts + 27a , $x ) ( $ts + 28a , $x ) ( $ts + 29a , $x ) ( $ts + 30a , $x ) ( $ts + 31a , $x ) ( $ts + 32a , $x ) ( $ts + 33a , $x ) ( $ts + 34a , $x ) ( $ts + 35a , $x ) ( $ts + 36a , $x ) ( $ts + 37a , $x ) ( $ts + 38a , $x ) ( $ts + 39a , $x ) ( $ts + 40a , $x ) ( $ts + 41a , $x ) ( $ts + 42a , $x ) ( $ts + 43a , $x ) ( $ts + 44a , $x ) ( $ts + 45a , $x ) ( $ts + 46a , $x ) ( $ts + 47a , $x ) ( $ts + 48a , $x ) ( $ts + 49a , $x ) ( $ts + 50a , $x ) ( $ts + 51a , $x ) ( $ts + 52a , $x ) ( $ts + 53a , $x ) ( $ts + 54a , $x ) ( $ts + 55a , $x ) ( $ts + 56a , $x ) ( $ts + 57a , $x ) ( $ts + 58a , $x ) ( $ts + 59a , $x ) 
+    $x = $x + 60
+  endw 
+  $totalRows = $totalRows + $x
+  print       info: inserted $x rows into $tb and totalRows: $totalRows
+  $i = $i + 1
+endw 
+
+sql select count(*) from $stb
+print rows:$rows    data00:$data00    totalRows:$totalRows
+if $rows != 1 then
+  return -1
+endi
+
+if $data00 != $totalRows then
+  return -1
+endi
+
+
+print ============== step3: insert old data(now-15d) and new data(now+15d), control data rows  in order to save in cache, not falling disc
+sql insert into $tb values ( now - 20d , -20 )
+sql insert into $tb values ( now - 40d , -40 )
+$totalRows = $totalRows + 2
+
+print ============== step4: stop dnode1
+system sh/exec.sh -n dnode1 -s stop -x SIGINT
+
+$loopCnt = 0
+wait_dnode1_offline:
+$loopCnt = $loopCnt + 1
+if $loopCnt == 10 then 
+  return -1
+endi
+
+sql show dnodes
+if $rows != 2 then
+  sleep 2000
+  goto wait_dnode1_offline
+endi
+print $data0_1  $data1_1  $data2_1  $data3_1  $data4_1
+print $data0_2  $data1_2  $data2_2  $data3_2  $data4_2
+$dnode1Status = $data4_1  
+$dnode2Status = $data4_2
+
+if $dnode1Status != offline then
+  sleep 2000
+  goto wait_dnode1_offline
+endi
+if $dnode2Status != ready then
+  sleep 2000
+  goto wait_dnode1_offline
+endi
+
+# check using select
+sql select count(*) from $stb
+print data00 $data00
+if $data00 != $totalRows then
+	return -1
+endi
+
+#sql show vgroups
+#print show vgroups:
+#print $data0_1  $data1_1  $data2_1  $data3_1  $data4_1  $data5_1  $data6_1  $data7_1  $data8_1  $data9_1
+#print $data0_2  $data1_2  $data2_2  $data3_2  $data4_2  $data5_2  $data6_2  $data7_2  $data8_2  $data9_2
+#print $data0_3  $data1_3  $data2_3  $data3_3  $data4_3  $data5_3  $data6_3  $data7_3  $data8_3  $data9_3
+
+print ============== step5: remove the mnode dir of dnode1, then copy the monde dir of dnode2
+system_content rm -rf ../../../sim/dnode1/data/vnode
+system_content cp -rf ../../../sim/dnode2/data/vnode ../../../sim/dnode1/data/
+
+print ============== step6: restart dnode1, waiting sync end
+system sh/exec.sh -n dnode1 -s start
+sleep 1000
+
+$loopCnt = 0
+wait_dnode1_ready:
+$loopCnt = $loopCnt + 1
+if $loopCnt == 20 then 
+  return -1
+endi
+
+sql show dnodes -x wait_dnode1_ready
+if $rows != 2 then
+  sleep 2000
+  goto wait_dnode1_ready
+endi
+print $data0_1  $data1_1  $data2_1  $data3_1  $data4_1
+print $data0_2  $data1_2  $data2_2  $data3_2  $data4_2
+$dnode1Status = $data4_1  
+$dnode2Status = $data4_2
+
+if $dnode1Status != ready then
+  sleep 2000
+  goto wait_dnode1_ready
+endi
+if $dnode2Status != ready then
+  sleep 2000
+  goto wait_dnode1_ready
+endi
+
+$loopCnt = 0
+wait_dnode1_vgroup_slave:
+$loopCnt = $loopCnt + 1
+if $loopCnt == 10 then 
+  return -1
+endi
+
+sql show vgroups
+if $rows != 3 then
+  sleep 2000
+  goto wait_dnode1_vgroup_slave
+endi
+print show vgroups:
+print $data0_1  $data1_1  $data2_1  $data3_1  $data4_1  $data5_1  $data6_1  $data7_1  $data8_1  $data9_1
+print $data0_2  $data1_2  $data2_2  $data3_2  $data4_2  $data5_2  $data6_2  $data7_2  $data8_2  $data9_2
+print $data0_3  $data1_3  $data2_3  $data3_3  $data4_3  $data5_3  $data6_3  $data7_3  $data8_3  $data9_3
+print $data0_4  $data1_4  $data2_4  $data3_4  $data4_4  $data5_4  $data6_4  $data7_4  $data8_4  $data9_4
+$d2v2status = $data4_4
+$d2v3status = $data4_2
+$d2v4status = $data4_3
+
+$d1v2status = $data7_4
+$d1v3status = $data7_2
+$d1v4status = $data7_3
+
+if $d2v2status != master then
+  sleep 2000
+  goto wait_dnode1_vgroup_slave
+endi
+if $d2v3status != master then
+  sleep 2000
+  goto wait_dnode1_vgroup_slave
+endi
+if $d2v4status != master then
+  sleep 2000
+  goto wait_dnode1_vgroup_slave
+endi
+
+if $d1v2status != slave then
+  sleep 2000
+  goto wait_dnode1_vgroup_slave
+endi
+if $d1v3status != slave then
+  sleep 2000
+  goto wait_dnode1_vgroup_slave
+endi
+if $d1v4status != slave then
+  sleep 2000
+  goto wait_dnode1_vgroup_slave
+endi
+
+print ============== step7: stop dnode2
+system sh/exec.sh -n dnode2 -s stop -x SIGINT
+
+$loopCnt = 0
+wait_dnode2_offline:
+$loopCnt = $loopCnt + 1
+if $loopCnt == 10 then 
+  return -1
+endi
+
+sql show dnodes
+if $rows != 2 then
+  sleep 2000
+  goto wait_dnode2_offline
+endi
+print $data0_1  $data1_1  $data2_1  $data3_1  $data4_1
+print $data0_2  $data1_2  $data2_2  $data3_2  $data4_2
+$dnode1Status = $data4_1  
+$dnode2Status = $data4_2
+
+if $dnode1Status != ready then
+  sleep 2000
+  goto wait_dnode2_offline
+endi
+if $dnode2Status != offline then
+  sleep 2000
+  goto wait_dnode2_offline
+endi
+
+sql reset query cache 
+
+# check using select
+sql select count(*) from $stb
+print data00 $data00
+if $data00 != $totalRows then
+	return -1
+endi
\ No newline at end of file

From 1405d825be8091c759af7fdd29889213498eb5e6 Mon Sep 17 00:00:00 2001
From: Shuduo Sang <sdsang@taosdata.com>
Date: Sat, 4 Jul 2020 17:45:38 +0800
Subject: [PATCH 04/19] emulate file operation random fail.

[TD-792]
---
 cmake/define.inc     |  4 ++++
 cmake/input.inc      |  7 +++++-
 src/util/inc/tfile.h | 29 ++++++++++++++++++++++++
 src/util/src/tfile.c | 53 ++++++++++++++++++++++++++++++++++++++++++++
 4 files changed, 92 insertions(+), 1 deletion(-)
 create mode 100644 src/util/inc/tfile.h
 create mode 100644 src/util/src/tfile.c

diff --git a/cmake/define.inc b/cmake/define.inc
index da100f4260..8d6a398709 100755
--- a/cmake/define.inc
+++ b/cmake/define.inc
@@ -24,3 +24,7 @@ ENDIF ()
 IF (TD_MEM_CHECK)
   ADD_DEFINITIONS(-DTAOS_MEM_CHECK)
 ENDIF ()
+
+IF (TD_RANDOM_FILE_FAIL)
+  ADD_DEFINITIONS(-DTAOS_RANDOM_FILE_FAIL)
+ENDIF ()
diff --git a/cmake/input.inc b/cmake/input.inc
index 5a17e0319c..574eac5b45 100755
--- a/cmake/input.inc
+++ b/cmake/input.inc
@@ -30,4 +30,9 @@ ENDIF ()
 IF (${MEM_CHECK} MATCHES "true")
   SET(TD_MEM_CHECK TRUE)
   MESSAGE(STATUS "build with memory check")
-ENDIF ()
\ No newline at end of file
+ENDIF ()
+
+IF (${RANDOM_FILE_FAIL} MATCHES "true")
+  SET(TD_RANDOM_FILE_FAIL TRUE)
+  MESSAGE(STATUS "build with random-file-fail enabled")
+ENDIF ()
diff --git a/src/util/inc/tfile.h b/src/util/inc/tfile.h
new file mode 100644
index 0000000000..566a429d32
--- /dev/null
+++ b/src/util/inc/tfile.h
@@ -0,0 +1,29 @@
+/*
+ * 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_TFILE_H
+#define TDENGINE_TFILE_H
+
+#ifdef TAOS_RANDOM_FILE_FAIL
+
+ssize_t taos_tread(int fd, void *buf, size_t count);
+ssize_t taos_twrite(int fd, void *buf, size_t count);
+
+#define tread(fd, buf, count)  taos_tread(fd, buf, count)
+#define twrite(fd, buf, count)  taos_twrite(fd, buf, count)
+
+#endif  // TAOS_RANDOM_FILE_FAIL
+
+#endif  // TDENGINE_TFILE_H
diff --git a/src/util/src/tfile.c b/src/util/src/tfile.c
new file mode 100644
index 0000000000..200d0f8af5
--- /dev/null
+++ b/src/util/src/tfile.c
@@ -0,0 +1,53 @@
+/*
+ * 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 <stdio.h>
+#include <stdlib.h>
+#include <error.h>
+#include <errno.h>
+#include <stdarg.h>
+#include <unistd.h>
+
+#include <sys/types.h>
+#include <sys/stat.h>
+#include <fcntl.h>
+
+#include "os.h"
+
+#define RANDOM_FACTOR 5
+
+ssize_t taos_tread(int fd, void *buf, size_t count)
+{
+#ifdef TAOS_RANDOM_FILE_FAIL
+  if (rand() % RANDOM_FACTOR == 0) {
+    errno = EIO;
+    return -1;
+  }
+#endif
+
+  return tread(fd, buf, count);
+}
+
+ssize_t taos_twrite(int fd, void *buf, size_t count)
+{
+#ifdef TAOS_RANDOM_FILE_FAIL
+  if (rand() % RANDOM_FACTOR == 0) {
+    errno = EIO;
+    return -1;
+  }
+#endif
+
+  return twrite(fd, buf, count);
+}

From 88dd38aac11f9b83312c582583cb5731e31fbd39 Mon Sep 17 00:00:00 2001
From: Shuduo Sang <sdsang@taosdata.com>
Date: Sat, 4 Jul 2020 18:10:42 +0800
Subject: [PATCH 05/19] add lseek random failure.

---
 src/query/src/qExecutor.c   |  1 +
 src/tsdb/src/tsdbFile.c     |  1 +
 src/tsdb/src/tsdbRWHelper.c |  1 +
 src/util/inc/tfile.h        |  2 ++
 src/util/src/tfile.c        | 12 ++++++++++++
 src/util/src/tkvstore.c     |  3 ++-
 6 files changed, 19 insertions(+), 1 deletion(-)

diff --git a/src/query/src/qExecutor.c b/src/query/src/qExecutor.c
index 8156967d5d..9f0b0b67b1 100644
--- a/src/query/src/qExecutor.c
+++ b/src/query/src/qExecutor.c
@@ -27,6 +27,7 @@
 #include "exception.h"
 #include "tscompression.h"
 #include "ttime.h"
+#include "tfile.h"
 
 /**
  * check if the primary column is load by default, otherwise, the program will
diff --git a/src/tsdb/src/tsdbFile.c b/src/tsdb/src/tsdbFile.c
index 33eae639b8..95cc47292b 100644
--- a/src/tsdb/src/tsdbFile.c
+++ b/src/tsdb/src/tsdbFile.c
@@ -28,6 +28,7 @@
 #include "tsdbMain.h"
 #include "tutil.h"
 #include "ttime.h"
+#include "tfile.h"
 
 const char *tsdbFileSuffix[] = {".head", ".data", ".last", "", ".h", ".l"};
 
diff --git a/src/tsdb/src/tsdbRWHelper.c b/src/tsdb/src/tsdbRWHelper.c
index 934fa8e733..eab9a5e056 100644
--- a/src/tsdb/src/tsdbRWHelper.c
+++ b/src/tsdb/src/tsdbRWHelper.c
@@ -19,6 +19,7 @@
 #include "tcoding.h"
 #include "tscompression.h"
 #include "tsdbMain.h"
+#include "tfile.h"
 
 #define TSDB_GET_COMPCOL_LEN(nCols) (sizeof(SCompData) + sizeof(SCompCol) * (nCols) + sizeof(TSCKSUM))
 
diff --git a/src/util/inc/tfile.h b/src/util/inc/tfile.h
index 566a429d32..5bddc76266 100644
--- a/src/util/inc/tfile.h
+++ b/src/util/inc/tfile.h
@@ -20,9 +20,11 @@
 
 ssize_t taos_tread(int fd, void *buf, size_t count);
 ssize_t taos_twrite(int fd, void *buf, size_t count);
+off_t taos_lseek(int fd, off_t offset, int whence);
 
 #define tread(fd, buf, count)  taos_tread(fd, buf, count)
 #define twrite(fd, buf, count)  taos_twrite(fd, buf, count)
+#define lseek(fd, offset, whence)  taos_lseek(fd, offset, whence)
 
 #endif  // TAOS_RANDOM_FILE_FAIL
 
diff --git a/src/util/src/tfile.c b/src/util/src/tfile.c
index 200d0f8af5..97eeda010e 100644
--- a/src/util/src/tfile.c
+++ b/src/util/src/tfile.c
@@ -51,3 +51,15 @@ ssize_t taos_twrite(int fd, void *buf, size_t count)
 
   return twrite(fd, buf, count);
 }
+
+off_t taos_lseek(int fd, off_t offset, int whence)
+{
+#ifdef TAOS_RANDOM_FILE_FAIL
+  if (rand() % RANDOM_FACTOR == 0) {
+    errno = EIO;
+    return -1;
+  }
+#endif
+
+  return lseek(fd, offset, whence);
+}
diff --git a/src/util/src/tkvstore.c b/src/util/src/tkvstore.c
index f33941376f..2a24a59742 100644
--- a/src/util/src/tkvstore.c
+++ b/src/util/src/tkvstore.c
@@ -27,6 +27,7 @@
 #include "tcoding.h"
 #include "tkvstore.h"
 #include "tulog.h"
+#include "tfile.h"
 
 #define TD_KVSTORE_HEADER_SIZE 512
 #define TD_KVSTORE_MAJOR_VERSION 1
@@ -581,4 +582,4 @@ _err:
   taosHashDestroyIter(pIter);
   tfree(buf);
   return -1;
-}
\ No newline at end of file
+}

From a4ede13384a6381823190935d5d3a1bba8e5a337 Mon Sep 17 00:00:00 2001
From: Hui Li <huili@taosdata.com>
Date: Sat, 4 Jul 2020 18:11:43 +0800
Subject: [PATCH 06/19] [TD-842]

---
 src/kit/taosdemo/taosdemo.c | 253 +++++++++++++++++++++++++++++++-----
 1 file changed, 218 insertions(+), 35 deletions(-)

diff --git a/src/kit/taosdemo/taosdemo.c b/src/kit/taosdemo/taosdemo.c
index 3265285cca..9a5aedcdb7 100644
--- a/src/kit/taosdemo/taosdemo.c
+++ b/src/kit/taosdemo/taosdemo.c
@@ -32,6 +32,7 @@
 #include <time.h>
 #include <unistd.h>
 #include <wordexp.h>
+#include <regex.h>
 
 #include "taos.h"
 #include "tutil.h"
@@ -54,6 +55,7 @@ static struct argp_option options[] = {
   {0, 'P', "password",                 0, "The password to use when connecting to the server. Default is 'taosdata'.",                                        3},
   {0, 'd', "database",                 0, "Destination database. Default is 'test'.",                                                                         3},
   {0, 'm', "table_prefix",             0, "Table prefix name. Default is 't'.",                                                                               3},
+  {0, 's', "sql file",                 0, "The select sql file.",                                                                                             3},
   {0, 'M', 0,                          0, "Use metric flag.",                                                                                                 13},
   {0, 'o', "outputfile",               0, "Direct output to the named file. Default is './output.txt'.",                                                      14},
   {0, 'q', "query_mode",               0, "Query mode--0: SYNC, 1: ASYNC. Default is SYNC.",                                                                  6},
@@ -79,6 +81,7 @@ typedef struct DemoArguments {
   char  *password;
   char  *database;
   char  *tb_prefix;
+  char  *sqlFile;
   bool   use_metric;
   bool   insert_only;
   char  *output_file;
@@ -120,6 +123,9 @@ static error_t parse_opt(int key, char *arg, struct argp_state *state) {
     case 'o':
       arguments->output_file = arg;
       break;
+    case 's':
+      arguments->sqlFile = arg;
+      break;
     case 'q':
       arguments->mode = atoi(arg);
       break;
@@ -179,10 +185,10 @@ static error_t parse_opt(int key, char *arg, struct argp_state *state) {
       arguments->tb_prefix = arg;
       break;
     case 'M':
-      arguments->use_metric = true;
+      arguments->use_metric = false;
       break;
     case 'x':
-      arguments->insert_only = true;
+      arguments->insert_only = false;
       break;
     case 'c':
       if (wordexp(arg, &full_path, 0) != 0) {
@@ -253,6 +259,9 @@ typedef struct {
   int data_of_rate;
   int64_t start_time;
   bool do_aggreFunc;
+  
+  char* cols;  
+  bool  use_metric;
 
   sem_t mutex_sem;
   int notFinished;
@@ -305,6 +314,8 @@ void rand_string(char *str, int size);
 double getCurrentTime();
 
 void callBack(void *param, TAOS_RES *res, int code);
+void multiThreadCreateTable(char* cols, bool use_metric, int threads, int ntables, char* db_name, char* tb_prefix, char *ip_addr, uint16_t port, char *user, char *pass);
+void querySqlFile(TAOS* taos, char* sqlFile);
 
 int main(int argc, char *argv[]) {
   SDemoArguments arguments = {  NULL,            // host
@@ -313,6 +324,7 @@ int main(int argc, char *argv[]) {
                                 "taosdata",      // password
                                 "test",          // database
                                 "t",             // tb_prefix
+                                NULL,
                                 false,           // use_metric
                                 false,           // insert_only
                                 "./output.txt",  // output_file
@@ -361,7 +373,7 @@ int main(int argc, char *argv[]) {
       abort();
     #endif
   }
-  
+
   enum MODE query_mode = arguments.mode;
   char *ip_addr = arguments.host;
   uint16_t port = arguments.port;
@@ -385,6 +397,13 @@ int main(int argc, char *argv[]) {
   char dataString[STRING_LEN];
   bool do_aggreFunc = true;
 
+  if (NULL != arguments.sqlFile) {
+    TAOS* qtaos = taos_connect(ip_addr, user, pass, db_name, port);
+    querySqlFile(qtaos, arguments.sqlFile);  
+    taos_close(qtaos);
+    return 0;
+  }
+
   memset(dataString, 0, STRING_LEN);
   int len = 0;
 
@@ -495,47 +514,19 @@ int main(int argc, char *argv[]) {
     len += snprintf(cols + len, STRING_LEN - len, ",f%d %s(%d))", colIndex + 1, data_type[colIndex % count_data_type], len_of_binary);
   }
 
-  if (!use_metric) {
-    /* Create all the tables; */
-    printf("Creating %d table(s)......\n", ntables);
-    for (int i = 0; i < ntables; i++) {
-      snprintf(command, BUFFER_SIZE, "create table if not exists %s.%s%d (ts timestamp%s;", db_name, tb_prefix, i, cols);
-      queryDB(taos, command);
-    }
-
-    printf("Table(s) created!\n");
-    taos_close(taos);
-
-  } else {
+  if (use_metric) {
     /* Create metric table */
     printf("Creating meters super table...\n");
     snprintf(command, BUFFER_SIZE, "create table if not exists %s.meters (ts timestamp%s tags (areaid int, loc binary(10))", db_name, cols);
     queryDB(taos, command);
     printf("meters created!\n");
 
-    /* Create all the tables; */
-    printf("Creating %d table(s)......\n", ntables);
-    for (int i = 0; i < ntables; i++) {
-      int j;
-      if (i % 10 == 0) {
-        j = 10;
-      } else {
-        j = i % 10;
-      }
-    if (j % 2 == 0) {
-      snprintf(command, BUFFER_SIZE, "create table if not exists %s.%s%d using %s.meters tags (%d,\"%s\");", db_name, tb_prefix, i, db_name, j, "shanghai");
-    } else {
-      snprintf(command, BUFFER_SIZE, "create table if not exists %s.%s%d using %s.meters tags (%d,\"%s\");", db_name, tb_prefix, i, db_name, j, "beijing");
-    }
-      queryDB(taos, command);
-    }
-
-    printf("Table(s) created!\n");
     taos_close(taos);
   }
-  /* Wait for table to create  */
   
-
+  /* Wait for table to create  */
+  multiThreadCreateTable(cols, use_metric, threads, ntables, db_name, tb_prefix, ip_addr, port, user, pass);
+  
   /* Insert data */
   double ts = getCurrentTime();
   printf("Inserting data......\n");
@@ -685,6 +676,198 @@ int main(int argc, char *argv[]) {
   return 0;
 }
 
+#define MAX_SQL_SIZE       65536
+void selectSql(TAOS* taos, char* sqlcmd)
+{    
+  TAOS_RES *pSql = taos_query(taos, sqlcmd);
+  int32_t code = taos_errno(pSql);
+  
+  if (code != 0) {
+    printf("Failed to sqlcmd:%s, reason:%s\n", sqlcmd, taos_errstr(pSql));
+    taos_free_result(pSql);
+    exit(1);
+  }
+  
+  int count = 0;
+  while (taos_fetch_row(pSql) != NULL) {
+    count++;
+  }
+  
+  taos_free_result(pSql);
+  return;
+}
+
+
+/* Function to do regular expression check */
+static int regexMatch(const char *s, const char *reg, int cflags) {
+  regex_t regex;
+  char    msgbuf[100] = {0};
+
+  /* Compile regular expression */
+  if (regcomp(&regex, reg, cflags) != 0) {
+    printf("Fail to compile regex\n");
+    exit(-1);
+  }
+
+  /* Execute regular expression */
+  int reti = regexec(&regex, s, 0, NULL, 0);
+  if (!reti) {
+    regfree(&regex);
+    return 1;
+  } else if (reti == REG_NOMATCH) {
+    regfree(&regex);
+    return 0;
+  } else {
+    regerror(reti, &regex, msgbuf, sizeof(msgbuf));
+    printf("Regex match failed: %s\n", msgbuf);
+    regfree(&regex);
+    exit(-1);
+  }
+
+  return 0;
+}
+
+static int isCommentLine(char *line) {
+  if (line == NULL) return 1;
+
+  return regexMatch(line, "^\\s*#.*", REG_EXTENDED);
+}
+
+void querySqlFile(TAOS* taos, char* sqlFile)
+{
+  FILE *fp = fopen(sqlFile, "r");
+  if (fp == NULL) {
+    printf("failed to open file %s, reason:%s\n", sqlFile, strerror(errno));
+    exit(-1);
+  }
+  
+  int       read_len = 0;
+  char *    cmd = calloc(1, MAX_SQL_SIZE);
+  size_t    cmd_len = 0;
+  char *    line = NULL;
+  size_t    line_len = 0;
+
+  double t = getCurrentTime();
+  
+  while ((read_len = getline(&line, &line_len, fp)) != -1) {
+    if (read_len >= MAX_SQL_SIZE) continue;
+    line[--read_len] = '\0';
+
+    if (read_len == 0 || isCommentLine(line)) {  // line starts with #
+      continue;
+    }
+
+    if (line[read_len - 1] == '\\') {
+      line[read_len - 1] = ' ';
+      memcpy(cmd + cmd_len, line, read_len);
+      cmd_len += read_len;
+      continue;
+    }
+
+    memcpy(cmd + cmd_len, line, read_len);
+    selectSql(taos, cmd);    
+    memset(cmd, 0, MAX_SQL_SIZE);
+    cmd_len = 0;
+  }
+
+  t = getCurrentTime() - t;
+  printf("run %s took %.6f second(s)\n\n", sqlFile, t);
+
+  free(cmd);
+  if (line) free(line);
+  fclose(fp);
+  return;
+}
+
+void * createTable(void *sarg) 
+{
+  char command[BUFFER_SIZE] = "\0";
+  
+  info *winfo = (info *)sarg;  
+
+  if (!winfo->use_metric) {
+    /* Create all the tables; */
+    printf("Creating table from %d to %d\n", winfo->start_table_id, winfo->end_table_id);
+    for (int i = winfo->start_table_id; i <= winfo->end_table_id; i++) {
+      snprintf(command, BUFFER_SIZE, "create table if not exists %s.%s%d (ts timestamp%s;", winfo->db_name, winfo->tb_prefix, i, winfo->cols);
+      queryDB(winfo->taos, command);
+    }
+
+    taos_close(winfo->taos);
+
+  } else {
+    /* Create all the tables; */
+    printf("Creating table from %d to %d\n", winfo->start_table_id, winfo->end_table_id);
+    for (int i = winfo->start_table_id; i <= winfo->end_table_id; i++) {
+      int j;
+      if (i % 10 == 0) {
+        j = 10;
+      } else {
+        j = i % 10;
+      }
+    if (j % 2 == 0) {
+      snprintf(command, BUFFER_SIZE, "create table if not exists %s.%s%d using %s.meters tags (%d,\"%s\");", winfo->db_name, winfo->tb_prefix, i, winfo->db_name, j, "shanghai");
+    } else {
+      snprintf(command, BUFFER_SIZE, "create table if not exists %s.%s%d using %s.meters tags (%d,\"%s\");", winfo->db_name, winfo->tb_prefix, i, winfo->db_name, j, "beijing");
+    }
+      queryDB(winfo->taos, command);
+    }
+    taos_close(winfo->taos);
+  }
+
+  return NULL;
+}
+
+void multiThreadCreateTable(char* cols, bool use_metric, int threads, int ntables, char* db_name, char* tb_prefix, char *ip_addr, uint16_t port, char *user, char *pass) {
+  double ts = getCurrentTime();
+  printf("create table......\n");
+  pthread_t *pids = malloc(threads * sizeof(pthread_t));
+  info *infos = malloc(threads * sizeof(info));
+
+  int a = ntables / threads;
+  if (a < 1) {
+    threads = ntables;
+    a = 1;
+  }
+
+  int b = 0;
+  if (threads != 0)
+    b = ntables % threads;
+  int last = 0;
+  for (int i = 0; i < threads; i++) {
+    info *t_info = infos + i;
+    t_info->threadID = i;
+    tstrncpy(t_info->db_name, db_name, MAX_DB_NAME_SIZE);
+    tstrncpy(t_info->tb_prefix, tb_prefix, MAX_TB_NAME_SIZE);
+    t_info->taos = taos_connect(ip_addr, user, pass, db_name, port);
+    t_info->start_table_id = last;
+    t_info->end_table_id = i < b ? last + a : last + a - 1;
+    last = t_info->end_table_id + 1;
+    t_info->use_metric = use_metric;
+    t_info->cols = cols;
+    pthread_create(pids + i, NULL, createTable, t_info);
+  }
+  
+  for (int i = 0; i < threads; i++) {
+    pthread_join(pids[i], NULL);
+  }
+
+  double t = getCurrentTime() - ts;
+  printf("Spent %.4f seconds to create %d tables with %d connections\n", t, ntables, threads);
+
+  for (int i = 0; i < threads; i++) {
+    info *t_info = infos + i;
+    taos_close(t_info->taos);
+    sem_destroy(&(t_info->mutex_sem));
+    sem_destroy(&(t_info->lock_sem));
+  }
+
+  free(pids);
+  free(infos);  
+
+  return ;
+}
+
 void *readTable(void *sarg) {
   info *rinfo = (info *)sarg;
   TAOS *taos = rinfo->taos;

From 764002212b0930a3cd14632661d013e5611888de Mon Sep 17 00:00:00 2001
From: Jeff Tao <jhtao@taosdata.com>
Date: Sat, 4 Jul 2020 11:28:50 +0000
Subject: [PATCH 07/19] signature

---
 src/rpc/src/rpcMain.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/src/rpc/src/rpcMain.c b/src/rpc/src/rpcMain.c
index 86b41747c4..c05c8c76e1 100644
--- a/src/rpc/src/rpcMain.c
+++ b/src/rpc/src/rpcMain.c
@@ -73,7 +73,7 @@ typedef struct {
   SRpcInfo *pRpc;       // associated SRpcInfo
   SRpcIpSet ipSet;      // ip list provided by app
   void     *ahandle;    // handle provided by app
-  void     *signature   // for validation
+  void     *signature;  // for validation
   struct SRpcConn *pConn; // pConn allocated
   char      msgType;    // message type
   uint8_t  *pCont;      // content provided by app

From 3e632cf7e148ab650aa140b392b0200bcc169adc Mon Sep 17 00:00:00 2001
From: Shengliang Guan <slguan@taosdata.com>
Date: Sat, 4 Jul 2020 21:12:44 +0800
Subject: [PATCH 08/19] fix crash while close vnodes

---
 src/dnode/src/dnodeMgmt.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/src/dnode/src/dnodeMgmt.c b/src/dnode/src/dnodeMgmt.c
index ece5aeb341..1ae1287888 100644
--- a/src/dnode/src/dnodeMgmt.c
+++ b/src/dnode/src/dnodeMgmt.c
@@ -283,7 +283,7 @@ static void *dnodeOpenVnode(void *param) {
 
 static int32_t dnodeOpenVnodes() {
   int32_t *vnodeList = calloc(TSDB_MAX_VNODES, sizeof(int32_t));
-  int32_t numOfVnodes;
+  int32_t numOfVnodes = 0;
   int32_t status = dnodeGetVnodeList(vnodeList, &numOfVnodes);
 
   if (status != TSDB_CODE_SUCCESS) {
@@ -360,7 +360,7 @@ void dnodeStartStream() {
 
 static void dnodeCloseVnodes() {
   int32_t vnodeList[TSDB_MAX_VNODES];
-  int32_t numOfVnodes;
+  int32_t numOfVnodes = 0;
   int32_t status;
 
   status = vnodeGetVnodeList(vnodeList, &numOfVnodes);

From 2976b629b248e2ee5fa7cd06ee29c664e1c75840 Mon Sep 17 00:00:00 2001
From: Shuduo Sang <sdsang@taosdata.com>
Date: Sun, 5 Jul 2020 18:06:13 +0800
Subject: [PATCH 09/19] emulate network read/write random failure.

[TD-791]
---
 cmake/define.inc                 |  4 ++++
 cmake/input.inc                  |  5 +++++
 src/os/linux/inc/os.h            | 12 ++++++++++++
 src/os/linux/src/linuxPlatform.c | 25 +++++++++++++++++++++++++
 4 files changed, 46 insertions(+)

diff --git a/cmake/define.inc b/cmake/define.inc
index 8d6a398709..93bf602610 100755
--- a/cmake/define.inc
+++ b/cmake/define.inc
@@ -28,3 +28,7 @@ ENDIF ()
 IF (TD_RANDOM_FILE_FAIL)
   ADD_DEFINITIONS(-DTAOS_RANDOM_FILE_FAIL)
 ENDIF ()
+
+IF (TD_RANDOM_NETWORK_FAIL)
+  ADD_DEFINITIONS(-DTAOS_RANDOM_NETWORK_FAIL)
+ENDIF ()
diff --git a/cmake/input.inc b/cmake/input.inc
index 574eac5b45..e963e20240 100755
--- a/cmake/input.inc
+++ b/cmake/input.inc
@@ -36,3 +36,8 @@ IF (${RANDOM_FILE_FAIL} MATCHES "true")
   SET(TD_RANDOM_FILE_FAIL TRUE)
   MESSAGE(STATUS "build with random-file-fail enabled")
 ENDIF ()
+
+IF (${RANDOM_NETWORK_FAIL} MATCHES "true")
+  SET(TD_RANDOM_NETWORK_FAIL TRUE)
+  MESSAGE(STATUS "build with random-network-fail enabled")
+ENDIF ()
diff --git a/src/os/linux/inc/os.h b/src/os/linux/inc/os.h
index 58e255f7bc..16fe56ac48 100644
--- a/src/os/linux/inc/os.h
+++ b/src/os/linux/inc/os.h
@@ -86,9 +86,21 @@ extern "C" {
     }                      \
   }
   
+#ifdef TAOS_RANDOM_NETWORK_FAIL
+
+ssize_t taos_read_random_fail(int fd, void *buf, size_t count);
+ssize_t taos_write_random_fail(int fd, const void *buf, size_t count);
+
+#define taosWriteSocket(fd, buf, len) taos_write_random_fail(fd, buf, len)
+#define taosReadSocket(fd, buf, len) taos_read_random_fail(fd, buf, len)
+
+#else
+
 #define taosWriteSocket(fd, buf, len) write(fd, buf, len)
 #define taosReadSocket(fd, buf, len) read(fd, buf, len)
 
+#endif  /* TAOS_RANDOM_NETWORK_FAIL */
+
 #define atomic_load_8(ptr) __atomic_load_n((ptr), __ATOMIC_SEQ_CST)
 #define atomic_load_16(ptr) __atomic_load_n((ptr), __ATOMIC_SEQ_CST)
 #define atomic_load_32(ptr) __atomic_load_n((ptr), __ATOMIC_SEQ_CST)
diff --git a/src/os/linux/src/linuxPlatform.c b/src/os/linux/src/linuxPlatform.c
index 9a38c98f81..073aae97bc 100644
--- a/src/os/linux/src/linuxPlatform.c
+++ b/src/os/linux/src/linuxPlatform.c
@@ -270,3 +270,28 @@ int tSystem(const char * cmd)
   }
 }
 
+#ifdef TAOS_RANDOM_NETWORK_FAIL
+
+#define RANDOM_NETWORK_FAIL_FACTOR  5
+
+ssize_t taos_read_random_fail(int fd, void *buf, size_t count)
+{
+  if (rand() % RANDOM_NETWORK_FAIL_FACTOR == 0) {
+    errno = EINTR;
+    return -1;
+  }
+
+  return read(fd, buf, count);
+}
+
+ssize_t taos_write_random_fail(int fd, const void *buf, size_t count)
+{
+  if (rand() % RANDOM_NETWORK_FAIL_FACTOR == 0) {
+    errno = EINTR;
+    return -1;
+  }
+
+  return write(fd, buf, count);
+}
+
+#endif /* TAOS_RANDOM_NETWORK_FAIL */

From 15e7b0d688d9b9011f94630a1c2b7f123e8ffc8c Mon Sep 17 00:00:00 2001
From: Shuduo Sang <sdsang@taosdata.com>
Date: Sun, 5 Jul 2020 18:07:59 +0800
Subject: [PATCH 10/19] change random factor name.

---
 src/util/src/tfile.c | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

diff --git a/src/util/src/tfile.c b/src/util/src/tfile.c
index 97eeda010e..eb7a2d5a66 100644
--- a/src/util/src/tfile.c
+++ b/src/util/src/tfile.c
@@ -26,12 +26,12 @@
 
 #include "os.h"
 
-#define RANDOM_FACTOR 5
+#define RANDOM_FILE_FAIL_FACTOR 5
 
 ssize_t taos_tread(int fd, void *buf, size_t count)
 {
 #ifdef TAOS_RANDOM_FILE_FAIL
-  if (rand() % RANDOM_FACTOR == 0) {
+  if (rand() % RANDOM_FILE_FAIL_FACTOR == 0) {
     errno = EIO;
     return -1;
   }
@@ -43,7 +43,7 @@ ssize_t taos_tread(int fd, void *buf, size_t count)
 ssize_t taos_twrite(int fd, void *buf, size_t count)
 {
 #ifdef TAOS_RANDOM_FILE_FAIL
-  if (rand() % RANDOM_FACTOR == 0) {
+  if (rand() % RANDOM_FILE_FAIL_FACTOR == 0) {
     errno = EIO;
     return -1;
   }
@@ -55,7 +55,7 @@ ssize_t taos_twrite(int fd, void *buf, size_t count)
 off_t taos_lseek(int fd, off_t offset, int whence)
 {
 #ifdef TAOS_RANDOM_FILE_FAIL
-  if (rand() % RANDOM_FACTOR == 0) {
+  if (rand() % RANDOM_FILE_FAIL_FACTOR == 0) {
     errno = EIO;
     return -1;
   }

From 10145efdd34bb6a19e3ca1bd8163a7a5177e1c9b Mon Sep 17 00:00:00 2001
From: Shuduo Sang <sdsang@taosdata.com>
Date: Sun, 5 Jul 2020 18:30:17 +0800
Subject: [PATCH 11/19] change error to ECONNRESET.

---
 src/os/linux/src/linuxPlatform.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/src/os/linux/src/linuxPlatform.c b/src/os/linux/src/linuxPlatform.c
index 073aae97bc..5edae8f9b2 100644
--- a/src/os/linux/src/linuxPlatform.c
+++ b/src/os/linux/src/linuxPlatform.c
@@ -277,7 +277,7 @@ int tSystem(const char * cmd)
 ssize_t taos_read_random_fail(int fd, void *buf, size_t count)
 {
   if (rand() % RANDOM_NETWORK_FAIL_FACTOR == 0) {
-    errno = EINTR;
+    errno = ECONNRESET;
     return -1;
   }
 

From c3f82496f42aff1f03c57d3a697d97c783c974bd Mon Sep 17 00:00:00 2001
From: colin <Colin.CHINA@outlook.com>
Date: Sun, 5 Jul 2020 18:38:58 +0800
Subject: [PATCH 12/19] update google-guava version

---
 tests/comparisonTest/opentsdb/opentsdbtest/pom.xml | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/tests/comparisonTest/opentsdb/opentsdbtest/pom.xml b/tests/comparisonTest/opentsdb/opentsdbtest/pom.xml
index f6728359e5..4e307db079 100644
--- a/tests/comparisonTest/opentsdb/opentsdbtest/pom.xml
+++ b/tests/comparisonTest/opentsdb/opentsdbtest/pom.xml
@@ -94,7 +94,7 @@
     <dependency>
       <groupId>com.google.guava</groupId>
       <artifactId>guava</artifactId>
-      <version>18.0</version>
+      <version>24.1.1</version>
     </dependency>
 
     <dependency>

From d2169f9b2d216afeec91137bebcf0594571f6599 Mon Sep 17 00:00:00 2001
From: Shuduo Sang <sdsang@taosdata.com>
Date: Sun, 5 Jul 2020 19:04:07 +0800
Subject: [PATCH 13/19] emulate send() and sendto() random fail too.

---
 src/os/linux/inc/os.h            |  7 +++++++
 src/os/linux/src/linuxPlatform.c | 23 ++++++++++++++++++++++-
 2 files changed, 29 insertions(+), 1 deletion(-)

diff --git a/src/os/linux/inc/os.h b/src/os/linux/inc/os.h
index 16fe56ac48..00b9f33f1b 100644
--- a/src/os/linux/inc/os.h
+++ b/src/os/linux/inc/os.h
@@ -88,9 +88,16 @@ extern "C" {
   
 #ifdef TAOS_RANDOM_NETWORK_FAIL
 
+ssize_t taos_send_random_fail(int sockfd, const void *buf, size_t len, int flags);
+
+ssize_t taos_sendto_random_fail(int sockfd, const void *buf, size_t len, int flags,
+                      const struct sockaddr *dest_addr, socklen_t addrlen);
 ssize_t taos_read_random_fail(int fd, void *buf, size_t count);
 ssize_t taos_write_random_fail(int fd, const void *buf, size_t count);
 
+#define send(sockfd, buf, len, flags) taos_send_random_fail(sockfd, buf, len, flags)
+#define sendto(sockfd, buf, len, flags, dest_addr, addrlen) \
+          taos_sendto_random_fail(sockfd, buf, len, flags, dest_addr, addrlen)
 #define taosWriteSocket(fd, buf, len) taos_write_random_fail(fd, buf, len)
 #define taosReadSocket(fd, buf, len) taos_read_random_fail(fd, buf, len)
 
diff --git a/src/os/linux/src/linuxPlatform.c b/src/os/linux/src/linuxPlatform.c
index 5edae8f9b2..216d8942bc 100644
--- a/src/os/linux/src/linuxPlatform.c
+++ b/src/os/linux/src/linuxPlatform.c
@@ -272,7 +272,28 @@ int tSystem(const char * cmd)
 
 #ifdef TAOS_RANDOM_NETWORK_FAIL
 
-#define RANDOM_NETWORK_FAIL_FACTOR  5
+#define RANDOM_NETWORK_FAIL_FACTOR  20
+
+ssize_t taos_send_random_fail(int sockfd, const void *buf, size_t len, int flags)
+{
+  if (rand() % RANDOM_NETWORK_FAIL_FACTOR == 0) {
+    errno = ECONNRESET;
+    return -1;
+  }
+
+  return send(sockfd, buf, len, flags);
+}
+
+ssize_t taos_sendto_random_fail(int sockfd, const void *buf, size_t len, int flags,
+                      const struct sockaddr *dest_addr, socklen_t addrlen)
+{
+  if (rand() % RANDOM_NETWORK_FAIL_FACTOR == 0) {
+    errno = ECONNRESET;
+    return -1;
+  }
+
+  return sendto(sockfd, buf, len, flags, dest_addr, addrlen);
+}
 
 ssize_t taos_read_random_fail(int fd, void *buf, size_t count)
 {

From d122bf5729965e993af4c67165d0465ad592f138 Mon Sep 17 00:00:00 2001
From: Shuduo Sang <sdsang@taosdata.com>
Date: Sun, 5 Jul 2020 23:20:34 +0800
Subject: [PATCH 14/19] make develop branch same as coverity-scan.

---
 src/os/linux/src/linuxSysPara.c | 19 +++++++++----------
 1 file changed, 9 insertions(+), 10 deletions(-)

diff --git a/src/os/linux/src/linuxSysPara.c b/src/os/linux/src/linuxSysPara.c
index c2134765df..1331503619 100644
--- a/src/os/linux/src/linuxSysPara.c
+++ b/src/os/linux/src/linuxSysPara.c
@@ -160,7 +160,7 @@ static void taosGetSystemTimezone() {
 
   /* load time zone string from /etc/timezone */
   FILE *f = fopen("/etc/timezone", "r");
-  char  buf[65] = {0};
+  char  buf[68] = {0};
   if (f != NULL) {
     int len = fread(buf, 64, 1, f);
     if(len < 64 && ferror(f)) {
@@ -170,18 +170,17 @@ static void taosGetSystemTimezone() {
     }
     
     fclose(f);
-  }
 
-  char *lineEnd = strstr(buf, "\n");
-  if (lineEnd != NULL) {
-    *lineEnd = 0;
-  }
+    char *lineEnd = strstr(buf, "\n");
+    if (lineEnd != NULL) {
+      *lineEnd = 0;
+    }
 
-  // for CentOS system, /etc/timezone does not exist. Ignore the TZ environment variables
-  if (strlen(buf) > 0) {
-    setenv("TZ", buf, 1);
+    // for CentOS system, /etc/timezone does not exist. Ignore the TZ environment variables
+    if (strlen(buf) > 0) {
+      setenv("TZ", buf, 1);
+    }
   }
-
   // get and set default timezone
   tzset();
 

From 648c85f0178591c3e0e5b2e01d596943593b114e Mon Sep 17 00:00:00 2001
From: Shengliang Guan <slguan@taosdata.com>
Date: Sun, 5 Jul 2020 23:25:32 +0800
Subject: [PATCH 15/19] vnode init/cleanup resource

---
 src/dnode/src/dnodeMgmt.c |  6 ++++++
 src/inc/vnode.h           |  2 ++
 src/vnode/src/vnodeMain.c | 31 ++++++++++---------------------
 3 files changed, 18 insertions(+), 21 deletions(-)

diff --git a/src/dnode/src/dnodeMgmt.c b/src/dnode/src/dnodeMgmt.c
index 10eb77058b..eaf34c9629 100644
--- a/src/dnode/src/dnodeMgmt.c
+++ b/src/dnode/src/dnodeMgmt.c
@@ -135,6 +135,12 @@ int32_t dnodeInitMgmt() {
     return -1; 
   }
 
+  code = vonodeInitResources();
+  if (code != TSDB_CODE_SUCCESS) {
+    dnodeCleanupMgmt();
+    return -1;
+  }
+
   code = dnodeOpenVnodes();
   if (code != TSDB_CODE_SUCCESS) {
     dnodeCleanupMgmt();
diff --git a/src/inc/vnode.h b/src/inc/vnode.h
index 971f341258..0c41485bb1 100644
--- a/src/inc/vnode.h
+++ b/src/inc/vnode.h
@@ -60,6 +60,8 @@ int32_t vnodeProcessWrite(void *pVnode, int qtype, void *pHead, void *item);
 int32_t vnodeGetVnodeList(int32_t vnodeList[], int32_t *numOfVnodes);
 void    vnodeBuildStatusMsg(void *param);
 void    vnodeSetAccess(SDMVgroupAccess *pAccess, int32_t numOfVnodes);
+
+int32_t vonodeInitResources();
 void    vnodeCleanupResources();
 
 int32_t vnodeProcessRead(void *pVnode, SReadMsg *pReadMsg);
diff --git a/src/vnode/src/vnodeMain.c b/src/vnode/src/vnodeMain.c
index ec97cbc48c..d431eafcb5 100644
--- a/src/vnode/src/vnodeMain.c
+++ b/src/vnode/src/vnodeMain.c
@@ -34,8 +34,7 @@
 
 #define TSDB_VNODE_VERSION_CONTENT_LEN 31
 
-static int32_t  tsOpennedVnodes;
-static void    *tsDnodeVnodesHash;
+static SHashObj*tsDnodeVnodesHash;
 static void     vnodeCleanUp(SVnodeObj *pVnode);
 static int32_t  vnodeSaveCfg(SMDCreateVnodeMsg *pVnodeCfg);
 static int32_t  vnodeReadCfg(SVnodeObj *pVnode);
@@ -48,8 +47,6 @@ static void     vnodeNotifyRole(void *ahandle, int8_t role);
 static void     vnodeNotifyFileSynced(void *ahandle, uint64_t fversion);
 static void     vnodeFreeqHandle(void* phandle);
 
-static pthread_once_t  vnodeModuleInit = PTHREAD_ONCE_INIT;
-
 #ifndef _SYNC
 tsync_h syncStart(const SSyncInfo *info) { return NULL; }
 int32_t syncForwardToPeer(tsync_h shandle, void *pHead, void *mhandle, int qtype) { return 0; }
@@ -59,25 +56,28 @@ int     syncGetNodesRole(tsync_h shandle, SNodesRole * cfg) { return 0; }
 void    syncConfirmForward(tsync_h shandle, uint64_t version, int32_t code) {}
 #endif
 
-static void vnodeInit() {
+int32_t vonodeInitResources() {
   vnodeInitWriteFp();
   vnodeInitReadFp();
 
   tsDnodeVnodesHash = taosHashInit(TSDB_MAX_VNODES, taosGetDefaultHashFunction(TSDB_DATA_TYPE_INT), true);
   if (tsDnodeVnodesHash == NULL) {
     vError("failed to init vnode list");
+    return TSDB_CODE_VND_OUT_OF_MEMORY;
   }
+
+  return TSDB_CODE_SUCCESS;
 }
 
 void vnodeCleanupResources() {
-  taosHashCleanup(tsDnodeVnodesHash);
-  vnodeModuleInit = PTHREAD_ONCE_INIT;
-  tsDnodeVnodesHash = NULL;
+  if (tsDnodeVnodesHash != NULL) {
+    taosHashCleanup(tsDnodeVnodesHash);
+    tsDnodeVnodesHash = NULL;
+  }
 }
 
 int32_t vnodeCreate(SMDCreateVnodeMsg *pVnodeCfg) {
   int32_t code;
-  pthread_once(&vnodeModuleInit, vnodeInit);
 
   SVnodeObj *pTemp = (SVnodeObj *)taosHashGet(tsDnodeVnodesHash, (const char *)&pVnodeCfg->cfg.vgId, sizeof(int32_t));
   if (pTemp != NULL) {
@@ -145,11 +145,6 @@ int32_t vnodeCreate(SMDCreateVnodeMsg *pVnodeCfg) {
 }
 
 int32_t vnodeDrop(int32_t vgId) {
-  if (tsDnodeVnodesHash == NULL) {
-    vDebug("vgId:%d, failed to drop, vgId not exist", vgId);
-    return TSDB_CODE_VND_INVALID_VGROUP_ID;
-  }
-
   SVnodeObj **ppVnode = (SVnodeObj **)taosHashGet(tsDnodeVnodesHash, (const char *)&vgId, sizeof(int32_t));
   if (ppVnode == NULL || *ppVnode == NULL) {
     vDebug("vgId:%d, failed to drop, vgId not find", vgId);
@@ -188,7 +183,6 @@ int32_t vnodeAlter(void *param, SMDCreateVnodeMsg *pVnodeCfg) {
 
 int32_t vnodeOpen(int32_t vnode, char *rootDir) {
   char temp[TSDB_FILENAME_LEN];
-  pthread_once(&vnodeModuleInit, vnodeInit);
 
   SVnodeObj *pVnode = calloc(sizeof(SVnodeObj), 1);
   if (pVnode == NULL) {
@@ -196,7 +190,6 @@ int32_t vnodeOpen(int32_t vnode, char *rootDir) {
     return TAOS_SYSTEM_ERROR(errno);
   }
 
-  atomic_add_fetch_32(&tsOpennedVnodes, 1);
   atomic_add_fetch_32(&pVnode->refCount, 1);
 
   pVnode->vgId     = vnode;
@@ -366,13 +359,11 @@ void vnodeRelease(void *pVnodeRaw) {
 
   free(pVnode);
 
-  int32_t count = atomic_sub_fetch_32(&tsOpennedVnodes, 1);
+  int32_t count = taosHashGetSize(tsDnodeVnodesHash);
   vDebug("vgId:%d, vnode is released, vnodes:%d", vgId, count);
 }
 
 void *vnodeGetVnode(int32_t vgId) {
-  if (tsDnodeVnodesHash == NULL) return NULL;
-
   SVnodeObj **ppVnode = (SVnodeObj **)taosHashGet(tsDnodeVnodesHash, (const char *)&vgId, sizeof(int32_t));
   if (ppVnode == NULL || *ppVnode == NULL) {
     terrno = TSDB_CODE_VND_INVALID_VGROUP_ID;
@@ -425,8 +416,6 @@ static void vnodeBuildVloadMsg(SVnodeObj *pVnode, SDMStatusMsg *pStatus) {
 }
 
 int32_t vnodeGetVnodeList(int32_t vnodeList[], int32_t *numOfVnodes) {
-  if (tsDnodeVnodesHash == NULL) return TSDB_CODE_SUCCESS;
-
   SHashMutableIterator *pIter = taosHashCreateIter(tsDnodeVnodesHash);
   while (taosHashIterNext(pIter)) {
     SVnodeObj **pVnode = taosHashIterGet(pIter);

From a567e04ab6c34f7f1a3bee73c073a7ccb9a73fda Mon Sep 17 00:00:00 2001
From: Shengliang Guan <slguan@taosdata.com>
Date: Sun, 5 Jul 2020 23:26:58 +0800
Subject: [PATCH 16/19] rename func

---
 src/dnode/src/dnodeMgmt.c | 2 +-
 src/inc/vnode.h           | 2 +-
 src/vnode/src/vnodeMain.c | 2 +-
 3 files changed, 3 insertions(+), 3 deletions(-)

diff --git a/src/dnode/src/dnodeMgmt.c b/src/dnode/src/dnodeMgmt.c
index eaf34c9629..2b809e60be 100644
--- a/src/dnode/src/dnodeMgmt.c
+++ b/src/dnode/src/dnodeMgmt.c
@@ -135,7 +135,7 @@ int32_t dnodeInitMgmt() {
     return -1; 
   }
 
-  code = vonodeInitResources();
+  code = vnodeInitResources();
   if (code != TSDB_CODE_SUCCESS) {
     dnodeCleanupMgmt();
     return -1;
diff --git a/src/inc/vnode.h b/src/inc/vnode.h
index 0c41485bb1..f4ed7dbfaf 100644
--- a/src/inc/vnode.h
+++ b/src/inc/vnode.h
@@ -61,7 +61,7 @@ int32_t vnodeGetVnodeList(int32_t vnodeList[], int32_t *numOfVnodes);
 void    vnodeBuildStatusMsg(void *param);
 void    vnodeSetAccess(SDMVgroupAccess *pAccess, int32_t numOfVnodes);
 
-int32_t vonodeInitResources();
+int32_t vnodeInitResources();
 void    vnodeCleanupResources();
 
 int32_t vnodeProcessRead(void *pVnode, SReadMsg *pReadMsg);
diff --git a/src/vnode/src/vnodeMain.c b/src/vnode/src/vnodeMain.c
index d431eafcb5..0815a741c4 100644
--- a/src/vnode/src/vnodeMain.c
+++ b/src/vnode/src/vnodeMain.c
@@ -56,7 +56,7 @@ int     syncGetNodesRole(tsync_h shandle, SNodesRole * cfg) { return 0; }
 void    syncConfirmForward(tsync_h shandle, uint64_t version, int32_t code) {}
 #endif
 
-int32_t vonodeInitResources() {
+int32_t vnodeInitResources() {
   vnodeInitWriteFp();
   vnodeInitReadFp();
 

From 3a2d7bc0abd63d0b960cce0588e26be998f39f82 Mon Sep 17 00:00:00 2001
From: Shengliang Guan <slguan@taosdata.com>
Date: Sun, 5 Jul 2020 23:37:12 +0800
Subject: [PATCH 17/19] init vnode list memory

---
 src/dnode/src/dnodeMgmt.c | 10 ++++------
 1 file changed, 4 insertions(+), 6 deletions(-)

diff --git a/src/dnode/src/dnodeMgmt.c b/src/dnode/src/dnodeMgmt.c
index 2b809e60be..7110e34411 100644
--- a/src/dnode/src/dnodeMgmt.c
+++ b/src/dnode/src/dnodeMgmt.c
@@ -288,13 +288,12 @@ static void *dnodeOpenVnode(void *param) {
 }
 
 static int32_t dnodeOpenVnodes() {
-  int32_t *vnodeList = calloc(TSDB_MAX_VNODES, sizeof(int32_t));
-  int32_t numOfVnodes;
+  int32_t vnodeList[TSDB_MAX_VNODES] = {0};
+  int32_t numOfVnodes = 0;
   int32_t status = dnodeGetVnodeList(vnodeList, &numOfVnodes);
 
   if (status != TSDB_CODE_SUCCESS) {
     dInfo("get dnode list failed");
-    free(vnodeList);
     return status;
   }
 
@@ -340,7 +339,6 @@ static int32_t dnodeOpenVnodes() {
     free(pThread->vnodeList);
   }
 
-  free(vnodeList);
   free(threads);
   dInfo("there are total vnodes:%d, openned:%d failed:%d", numOfVnodes, openVnodes, failedVnodes);
 
@@ -348,7 +346,7 @@ static int32_t dnodeOpenVnodes() {
 }
 
 void dnodeStartStream() {
-  int32_t vnodeList[TSDB_MAX_VNODES];
+  int32_t vnodeList[TSDB_MAX_VNODES] = {0};
   int32_t numOfVnodes = 0;
   int32_t status = vnodeGetVnodeList(vnodeList, &numOfVnodes);
 
@@ -365,7 +363,7 @@ void dnodeStartStream() {
 }
 
 static void dnodeCloseVnodes() {
-  int32_t vnodeList[TSDB_MAX_VNODES];
+  int32_t vnodeList[TSDB_MAX_VNODES]= {0};
   int32_t numOfVnodes;
   int32_t status;
 

From 9c96634b9481189b199f6b57cb3e1533dff20e07 Mon Sep 17 00:00:00 2001
From: Shengliang Guan <slguan@taosdata.com>
Date: Sun, 5 Jul 2020 23:47:13 +0800
Subject: [PATCH 18/19] un-init variables

---
 src/dnode/src/dnodeMgmt.c | 16 ++++++++--------
 1 file changed, 8 insertions(+), 8 deletions(-)

diff --git a/src/dnode/src/dnodeMgmt.c b/src/dnode/src/dnodeMgmt.c
index 7110e34411..81608ac8e2 100644
--- a/src/dnode/src/dnodeMgmt.c
+++ b/src/dnode/src/dnodeMgmt.c
@@ -106,6 +106,12 @@ int32_t dnodeInitMgmt() {
     }
   }
 
+  int32_t code = vnodeInitResources();
+  if (code != TSDB_CODE_SUCCESS) {
+    dnodeCleanupMgmt();
+    return -1;
+  }
+
   // create the queue and thread to handle the message 
   tsMgmtQset = taosOpenQset();
   if (tsMgmtQset == NULL) {
@@ -127,7 +133,7 @@ int32_t dnodeInitMgmt() {
   pthread_attr_init(&thAttr);
   pthread_attr_setdetachstate(&thAttr, PTHREAD_CREATE_JOINABLE);
 
-  int32_t code = pthread_create(&tsQthread, &thAttr, dnodeProcessMgmtQueue, NULL);
+  code = pthread_create(&tsQthread, &thAttr, dnodeProcessMgmtQueue, NULL);
   pthread_attr_destroy(&thAttr);
   if (code != 0) {
     dError("failed to create thread to process mgmt queue, reason:%s", strerror(errno));
@@ -135,12 +141,6 @@ int32_t dnodeInitMgmt() {
     return -1; 
   }
 
-  code = vnodeInitResources();
-  if (code != TSDB_CODE_SUCCESS) {
-    dnodeCleanupMgmt();
-    return -1;
-  }
-
   code = dnodeOpenVnodes();
   if (code != TSDB_CODE_SUCCESS) {
     dnodeCleanupMgmt();
@@ -364,7 +364,7 @@ void dnodeStartStream() {
 
 static void dnodeCloseVnodes() {
   int32_t vnodeList[TSDB_MAX_VNODES]= {0};
-  int32_t numOfVnodes;
+  int32_t numOfVnodes = 0;
   int32_t status;
 
   status = vnodeGetVnodeList(vnodeList, &numOfVnodes);

From 6e06701aad9b5648d32af47411e36673fd38d9e6 Mon Sep 17 00:00:00 2001
From: Hongze Cheng <hzcheng@taosdata.com>
Date: Mon, 6 Jul 2020 16:35:13 +0800
Subject: [PATCH 19/19] fix unitiailized error

---
 src/query/src/qExecutor.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/src/query/src/qExecutor.c b/src/query/src/qExecutor.c
index bd8c9951e1..d41c9fbba8 100644
--- a/src/query/src/qExecutor.c
+++ b/src/query/src/qExecutor.c
@@ -2173,6 +2173,7 @@ static void ensureOutputBuffer(SQueryRuntimeEnv* pRuntimeEnv, SDataBlockInfo* pB
         if (tmp == NULL) {  // todo handle the oom
           assert(0);
         } else {
+          memset(tmp + sizeof(tFilePage) + bytes * pRec->rows, 0, (newSize - pRec->rows) * bytes);
           pQuery->sdata[i] = (tFilePage *)tmp;
         }