From 59d23065fd469447717b2bc7da46ce2ec012c01e Mon Sep 17 00:00:00 2001
From: Shengliang Guan <slguan@taosdata.com>
Date: Tue, 11 Jan 2022 02:54:11 -0800
Subject: [PATCH 1/4] fix invalid write in mnode

---
 source/dnode/mnode/impl/src/mndDb.c     |  2 +-
 source/dnode/mnode/impl/src/mndDnode.c  |  2 +-
 source/dnode/mnode/impl/src/mndVgroup.c |  8 ++++
 tests/script/sh/exec.sh                 |  4 +-
 tests/script/sim/db/basic6.sim          | 56 ++++++++++---------------
 5 files changed, 33 insertions(+), 39 deletions(-)

diff --git a/source/dnode/mnode/impl/src/mndDb.c b/source/dnode/mnode/impl/src/mndDb.c
index 85b7fbbb42..bc4d890257 100644
--- a/source/dnode/mnode/impl/src/mndDb.c
+++ b/source/dnode/mnode/impl/src/mndDb.c
@@ -740,7 +740,7 @@ static int32_t mndBuildDropVgroupAction(SMnode *pMnode, STrans *pTrans, SDbObj *
     if (pReq == NULL) return -1;
 
     action.pCont = pReq;
-    action.contLen = sizeof(SCreateVnodeReq);
+    action.contLen = sizeof(SDropVnodeReq);
     action.msgType = TDMT_DND_DROP_VNODE;
     action.acceptableCode = TSDB_CODE_DND_VNODE_NOT_DEPLOYED;
     if (mndTransAppendRedoAction(pTrans, &action) != 0) {
diff --git a/source/dnode/mnode/impl/src/mndDnode.c b/source/dnode/mnode/impl/src/mndDnode.c
index 4bc570c11d..2a67e6a0f9 100644
--- a/source/dnode/mnode/impl/src/mndDnode.c
+++ b/source/dnode/mnode/impl/src/mndDnode.c
@@ -235,7 +235,7 @@ int32_t mndGetDnodeSize(SMnode *pMnode) {
 
 bool mndIsDnodeOnline(SMnode *pMnode, SDnodeObj *pDnode, int64_t curMs) {
   int64_t interval = ABS(pDnode->lastAccessTime - curMs);
-  if (interval > 3500 * pMnode->cfg.statusInterval) {
+  if (interval > 10000 * pMnode->cfg.statusInterval) {
     if (pDnode->rebootTime > 0) {
       pDnode->offlineReason = DND_REASON_STATUS_MSG_TIMEOUT;
     }
diff --git a/source/dnode/mnode/impl/src/mndVgroup.c b/source/dnode/mnode/impl/src/mndVgroup.c
index e0d6d3dd42..93d6d104ff 100644
--- a/source/dnode/mnode/impl/src/mndVgroup.c
+++ b/source/dnode/mnode/impl/src/mndVgroup.c
@@ -319,6 +319,14 @@ static int32_t mndGetAvailableDnode(SMnode *pMnode, SVgObj *pVgroup, SArray *pAr
 
   taosArraySort(pArray, (__compar_fn_t)mndCompareDnodeVnodes);
 
+  int32_t size = taosArrayGetSize(pArray);
+  if (size < pVgroup->replica) {
+    mError("db:%s, vgId:%d, no enough online dnodes:%d to alloc %d replica", pVgroup->dbName, pVgroup->vgId, size,
+           pVgroup->replica);
+    terrno = TSDB_CODE_MND_NO_ENOUGH_DNODES;
+    return -1;
+  }
+
   for (int32_t v = 0; v < pVgroup->replica; ++v) {
     SVnodeGid *pVgid = &pVgroup->vnodeGid[v];
     SDnodeObj *pDnode = taosArrayGet(pArray, v);
diff --git a/tests/script/sh/exec.sh b/tests/script/sh/exec.sh
index 2e95a740d0..05f756ebb6 100755
--- a/tests/script/sh/exec.sh
+++ b/tests/script/sh/exec.sh
@@ -30,7 +30,7 @@ do
       CLEAR_OPTION="clear"
       ;;
     v)
-      SHELL_OPTION="true"
+      VALGRIND_OPTION="true"
       ;;
     u)
       USERS=$OPTARG
@@ -99,7 +99,7 @@ fi
 if [ "$EXEC_OPTON" = "start" ]; then 
   echo "ExcuteCmd:" $EXE_DIR/taosd -c $CFG_DIR
   
-  if [ "$SHELL_OPTION" = "true" ]; then 
+  if [ "$VALGRIND_OPTION" = "true" ]; then 
     TT=`date +%s`
     mkdir ${LOG_DIR}/${TT}
     nohup valgrind --log-file=${LOG_DIR}/${TT}/valgrind.log --tool=memcheck --leak-check=full --show-reachable=no  --track-origins=yes --show-leak-kinds=all  -v  --workaround-gcc296-bugs=yes   $EXE_DIR/taosd -c $CFG_DIR > /dev/null 2>&1 &   
diff --git a/tests/script/sim/db/basic6.sim b/tests/script/sim/db/basic6.sim
index 684ce825fe..8e7d4f5ec1 100644
--- a/tests/script/sim/db/basic6.sim
+++ b/tests/script/sim/db/basic6.sim
@@ -1,29 +1,24 @@
 system sh/stop_dnodes.sh
 system sh/deploy.sh -n dnode1 -i 1
-system sh/cfg.sh -n dnode1 -c wallevel -v 0
-system sh/cfg.sh -n dnode1 -c maxVgroupsPerDb -v 4
-system sh/cfg.sh -n dnode1 -c maxTablesPerVnode -v 1000
-
 system sh/exec.sh -n dnode1 -s start
-
-sleep 2000
 sql connect
+
 print ============================ dnode1 start
 
 $i = 0
-$dbPrefix = ob_db_db
-$tbPrefix = ob_db_tb
+$dbPrefix = db
+$tbPrefix = tb
 $db = $dbPrefix . $i
 $tb = $tbPrefix . $i
 
 print =============== step1
-sql create database $db replica 1 days 20 keep 2000 cache 16
+sql create database $db replica 1 days 20 keep 2000 cache 16 vgroups 4
 sql show databases
 print $data00 $data01 $data02 $data03 $data04 $data05 $data06 $data07
 if $data00 != $db then 
   return -1
 endi
-if $data02 != 0 then 
+if $data02 != 4 then 
   return -1
 endi
 if $data03 != 0 then 
@@ -63,9 +58,6 @@ print $data00 $data01 $data02 $data03 $data04 $data05 $data06 $data07
 if $data00 != $db then 
   return -1
 endi
-if $data02 != 0 then 
-  return -1
-endi
 if $data03 != 0 then 
   return -1
 endi
@@ -77,30 +69,14 @@ if $data06 != 15 then
 endi
 
 print =============== step6
-sql use $db
-sql create table $tb (ts timestamp, speed int)
-$i = 1
-while $i < 4
+$i = $i + 1
+while $i < 5
   $db = $dbPrefix . $i
-  $tb = $tbPrefix . $i
   sql create database $db
   sql use $db
-  sql create table $tb (ts timestamp, speed int)
   $i = $i + 1
 endw
 
-sql show databases
-if $rows != 4 then 
-  return -1
-endi
-
-$i = 4
-$db = $dbPrefix . $i
-$tb = $tbPrefix . $i
-sql create database $db
-sql use $db
-sql create table $tb (ts timestamp, speed int)
-
 print =============== step7
 $i = 0
 while $i < 5
@@ -115,7 +91,8 @@ $db = $dbPrefix . $i
 $tb = $tbPrefix . $i
 sql create database $db
 sql use $db
-sql create table $tb (ts timestamp, speed int)
+sql create table st (ts timestamp, i int) tags (j int)
+sql create table $tb using st tags(1)
 sql show tables
 if $rows != 1 then 
   return -1
@@ -133,7 +110,8 @@ if $rows != 0 then
 endi
 
 print =============== step11
-sql create table $tb (ts timestamp, speed int)
+sql create table st (ts timestamp, i int) tags (j int)
+sql create table $tb using st tags(1)
 sql show tables
 if $rows != 1 then 
   return -1
@@ -149,16 +127,23 @@ sql show tables
 if $rows != 0 then 
   return -1
 endi
-sql create table $tb (ts timestamp, speed int)
+
+sql create table st (ts timestamp, i int) tags (j int)
+sql create table $tb using st tags(1)
 sql show tables
 if $rows != 1 then 
   return -1
 endi
+
 sql insert into $tb values (now+1a, 0)
 sql insert into $tb values (now+2a, 1)
 sql insert into $tb values (now+3a, 2)
 sql insert into $tb values (now+4a, 3)
 sql insert into $tb values (now+5a, 4)
+
+
+return 
+
 sql select * from $tb
 if $rows != 5 then 
   return -1
@@ -176,7 +161,8 @@ if $rows != 0 then
 endi
 
 print =============== step16
-sql create table $tb (ts timestamp, speed int)
+sql create table st (ts timestamp, i int) tags (j int)
+sql create table $tb using st tags(1)
 sql show tables
 if $rows != 1 then 
   return -1

From f94f8aadc0b00e15a0869eb7291ae53b8bec8d49 Mon Sep 17 00:00:00 2001
From: Shengliang Guan <slguan@taosdata.com>
Date: Tue, 11 Jan 2022 03:13:26 -0800
Subject: [PATCH 2/4] dnode may offline while create vnode

---
 source/dnode/mgmt/impl/inc/dndInt.h    |  1 +
 source/dnode/mgmt/impl/src/dndMgmt.c   | 13 ++++++++++++-
 source/dnode/mnode/impl/src/mndDnode.c |  2 +-
 3 files changed, 14 insertions(+), 2 deletions(-)

diff --git a/source/dnode/mgmt/impl/inc/dndInt.h b/source/dnode/mgmt/impl/inc/dndInt.h
index afdd678213..e637b38815 100644
--- a/source/dnode/mgmt/impl/inc/dndInt.h
+++ b/source/dnode/mgmt/impl/inc/dndInt.h
@@ -94,6 +94,7 @@ typedef struct {
   pthread_t   *threadId;
   SRWLatch     latch;
   SDnodeWorker mgmtWorker;
+  SDnodeWorker statusWorker;
 } SDnodeMgmt;
 
 typedef struct {
diff --git a/source/dnode/mgmt/impl/src/dndMgmt.c b/source/dnode/mgmt/impl/src/dndMgmt.c
index f252bffbbf..d9edf39b73 100644
--- a/source/dnode/mgmt/impl/src/dndMgmt.c
+++ b/source/dnode/mgmt/impl/src/dndMgmt.c
@@ -536,6 +536,11 @@ int32_t dndInitMgmt(SDnode *pDnode) {
     return -1;
   }
 
+  if (dndInitWorker(pDnode, &pMgmt->statusWorker, DND_WORKER_SINGLE, "dnode-status", 1, 1, dndProcessMgmtQueue) != 0) {
+    dError("failed to start dnode mgmt worker since %s", terrstr());
+    return -1;
+  }
+
   pMgmt->threadId = taosCreateThread(dnodeThreadRoutine, pDnode);
   if (pMgmt->threadId == NULL) {
     dError("failed to init dnode thread");
@@ -550,6 +555,7 @@ int32_t dndInitMgmt(SDnode *pDnode) {
 void dndStopMgmt(SDnode *pDnode) {
   SDnodeMgmt *pMgmt = &pDnode->dmgmt;
   dndCleanupWorker(&pMgmt->mgmtWorker);
+  dndCleanupWorker(&pMgmt->statusWorker);
 
   if (pMgmt->threadId != NULL) {
     taosDestoryThread(pMgmt->threadId);
@@ -587,7 +593,12 @@ void dndProcessMgmtMsg(SDnode *pDnode, SRpcMsg *pMsg, SEpSet *pEpSet) {
     dndUpdateMnodeEpSet(pDnode, pEpSet);
   }
 
-  if (dndWriteMsgToWorker(&pMgmt->mgmtWorker, pMsg, sizeof(SRpcMsg)) != 0) {
+  SDnodeWorker *pWorker = &pMgmt->mgmtWorker;
+  if (pMsg->msgType == TDMT_MND_STATUS_RSP) {
+    pWorker = &pMgmt->statusWorker;
+  }
+
+  if (dndWriteMsgToWorker(pWorker, pMsg, sizeof(SRpcMsg)) != 0) {
     if (pMsg->msgType & 1u) {
       SRpcMsg rsp = {.handle = pMsg->handle, .code = TSDB_CODE_OUT_OF_MEMORY};
       rpcSendResponse(&rsp);
diff --git a/source/dnode/mnode/impl/src/mndDnode.c b/source/dnode/mnode/impl/src/mndDnode.c
index 2a67e6a0f9..4bc570c11d 100644
--- a/source/dnode/mnode/impl/src/mndDnode.c
+++ b/source/dnode/mnode/impl/src/mndDnode.c
@@ -235,7 +235,7 @@ int32_t mndGetDnodeSize(SMnode *pMnode) {
 
 bool mndIsDnodeOnline(SMnode *pMnode, SDnodeObj *pDnode, int64_t curMs) {
   int64_t interval = ABS(pDnode->lastAccessTime - curMs);
-  if (interval > 10000 * pMnode->cfg.statusInterval) {
+  if (interval > 3500 * pMnode->cfg.statusInterval) {
     if (pDnode->rebootTime > 0) {
       pDnode->offlineReason = DND_REASON_STATUS_MSG_TIMEOUT;
     }

From 07a1b4acf21d25321819781d1ec9a2214d98e175 Mon Sep 17 00:00:00 2001
From: Shengliang Guan <slguan@taosdata.com>
Date: Tue, 11 Jan 2022 03:41:38 -0800
Subject: [PATCH 3/4] minor changes

---
 source/dnode/mnode/impl/test/profile/profile.cpp | 10 +++++-----
 tests/script/jenkins/basic.txt                   |  5 +++--
 tests/script/sim/db/basic6.sim                   |  4 ++++
 3 files changed, 12 insertions(+), 7 deletions(-)

diff --git a/source/dnode/mnode/impl/test/profile/profile.cpp b/source/dnode/mnode/impl/test/profile/profile.cpp
index bdffb6c72a..bf047517d3 100644
--- a/source/dnode/mnode/impl/test/profile/profile.cpp
+++ b/source/dnode/mnode/impl/test/profile/profile.cpp
@@ -13,7 +13,7 @@
 
 class MndTestProfile : public ::testing::Test {
  protected:
-  static void SetUpTestSuite() { test.Init("/tmp/mnode_test_profile", 9022); }
+  static void SetUpTestSuite() { test.Init("/tmp/mnode_test_profile", 9031); }
   static void TearDownTestSuite() { test.Cleanup(); }
 
   static Testbase test;
@@ -53,7 +53,7 @@ TEST_F(MndTestProfile, 01_ConnectMsg) {
 
   EXPECT_EQ(pRsp->epSet.inUse, 0);
   EXPECT_EQ(pRsp->epSet.numOfEps, 1);
-  EXPECT_EQ(pRsp->epSet.port[0], 9022);
+  EXPECT_EQ(pRsp->epSet.port[0], 9031);
   EXPECT_STREQ(pRsp->epSet.fqdn[0], "localhost");
 
   connId = pRsp->connId;
@@ -127,7 +127,7 @@ TEST_F(MndTestProfile, 04_HeartBeatMsg) {
 
   EXPECT_EQ(pRsp->epSet.inUse, 0);
   EXPECT_EQ(pRsp->epSet.numOfEps, 1);
-  EXPECT_EQ(pRsp->epSet.port[0], 9022);
+  EXPECT_EQ(pRsp->epSet.port[0], 9031);
   EXPECT_STREQ(pRsp->epSet.fqdn[0], "localhost");
 }
 
@@ -185,7 +185,7 @@ TEST_F(MndTestProfile, 05_KillConnMsg) {
 
     EXPECT_EQ(pRsp->epSet.inUse, 0);
     EXPECT_EQ(pRsp->epSet.numOfEps, 1);
-    EXPECT_EQ(pRsp->epSet.port[0], 9022);
+    EXPECT_EQ(pRsp->epSet.port[0], 9031);
     EXPECT_STREQ(pRsp->epSet.fqdn[0], "localhost");
 
     connId = pRsp->connId;
@@ -249,7 +249,7 @@ TEST_F(MndTestProfile, 07_KillQueryMsg) {
 
     EXPECT_EQ(pRsp->epSet.inUse, 0);
     EXPECT_EQ(pRsp->epSet.numOfEps, 1);
-    EXPECT_EQ(pRsp->epSet.port[0], 9022);
+    EXPECT_EQ(pRsp->epSet.port[0], 9031);
     EXPECT_STREQ(pRsp->epSet.fqdn[0], "localhost");
   }
 }
diff --git a/tests/script/jenkins/basic.txt b/tests/script/jenkins/basic.txt
index bc0c3a4f64..a84e473d9f 100644
--- a/tests/script/jenkins/basic.txt
+++ b/tests/script/jenkins/basic.txt
@@ -5,8 +5,9 @@
 ./test.sh -f sim/user/basic1.sim 
 
 # ---- db
-./test.sh -f sim/db/basic1.sim 
-./test.sh -f sim/db/error1.sim 
+./test.sh -f sim/db/basic1.sim
+./test.sh -f sim/db/error6.sim
+./test.sh -f sim/db/error1.sim
 
 # ---- table
 ./test.sh -f sim/table/basic1.sim 
diff --git a/tests/script/sim/db/basic6.sim b/tests/script/sim/db/basic6.sim
index 8e7d4f5ec1..a688b4c2f3 100644
--- a/tests/script/sim/db/basic6.sim
+++ b/tests/script/sim/db/basic6.sim
@@ -93,6 +93,10 @@ sql create database $db
 sql use $db
 sql create table st (ts timestamp, i int) tags (j int)
 sql create table $tb using st tags(1)
+
+return 
+system sh/exec.sh -n dnode1 -s stop -x SIGINT
+
 sql show tables
 if $rows != 1 then 
   return -1

From 530ffb6cbc6fc4b14a07c96388bc6e7fcf22c1fc Mon Sep 17 00:00:00 2001
From: Shengliang Guan <slguan@taosdata.com>
Date: Tue, 11 Jan 2022 03:46:05 -0800
Subject: [PATCH 4/4] minor changes

---
 tests/script/jenkins/basic.txt | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/tests/script/jenkins/basic.txt b/tests/script/jenkins/basic.txt
index a84e473d9f..bb6569deb0 100644
--- a/tests/script/jenkins/basic.txt
+++ b/tests/script/jenkins/basic.txt
@@ -6,7 +6,7 @@
 
 # ---- db
 ./test.sh -f sim/db/basic1.sim
-./test.sh -f sim/db/error6.sim
+./test.sh -f sim/db/basic6.sim
 ./test.sh -f sim/db/error1.sim
 
 # ---- table