[TD-638]

src/dnode/src/dnodeMgmt.c

@@ -617,6 +617,16 @@ static void dnodeSendStatusMsg(void *handle, void *tmrId) {
   pStatus->diskAvailable    = tsAvailDataDirGB;
   pStatus->alternativeRole  = (uint8_t) tsAlternativeRole;
 
+  // fill cluster cfg parameters
+  pStatus->ClusterCfgPara.numOfMnodes        = tsNumOfMnodes;
+  pStatus->ClusterCfgPara.mnodeEqualVnodeNum = tsMnodeEqualVnodeNum;
+  pStatus->ClusterCfgPara.offlineThreshold   = tsOfflineThreshold;
+  pStatus->ClusterCfgPara.statusInterval     = tsStatusInterval;
+  strcpy(pStatus->ClusterCfgPara.arbitrator, tsArbitrator);
+  strcpy(pStatus->ClusterCfgPara.timezone, tsTimezone);
+  strcpy(pStatus->ClusterCfgPara.locale, tsLocale);
+  strcpy(pStatus->ClusterCfgPara.charset, tsCharset);  
+  
   vnodeBuildStatusMsg(pStatus);
   contLen = sizeof(SDMStatusMsg) + pStatus->openVnodes * sizeof(SVnodeLoad);
   pStatus->openVnodes = htons(pStatus->openVnodes);

src/inc/taoserror.h

@@ -121,6 +121,7 @@ TAOS_DEFINE_ERROR(TSDB_CODE_MND_DNODE_NOT_EXIST,          0, 0x0331, "mnode dnod
 TAOS_DEFINE_ERROR(TSDB_CODE_MND_VGROUP_NOT_EXIST,         0, 0x0332, "mnode vgroup not exist")
 TAOS_DEFINE_ERROR(TSDB_CODE_MND_NO_REMOVE_MASTER,         0, 0x0333, "mnode cant not remove master")
 TAOS_DEFINE_ERROR(TSDB_CODE_MND_NO_ENOUGH_DNODES,         0, 0x0334, "mnode no enough dnodes")
+TAOS_DEFINE_ERROR(TSDB_CODE_MND_CLUSTER_CFG_INCONSISTENT, 0, 0x0335, "mnode cluster cfg inconsistent")
 
 TAOS_DEFINE_ERROR(TSDB_CODE_MND_ACCT_ALREADY_EXIST,       0, 0x0340, "mnode accounts already exist")
 TAOS_DEFINE_ERROR(TSDB_CODE_MND_INVALID_ACCT,             0, 0x0341, "mnode invalid account")

src/inc/taosmsg.h

@@ -556,6 +556,17 @@ typedef struct {
   SDMMnodeInfo nodeInfos[TSDB_MAX_REPLICA];
 } SDMMnodeInfos;
 
+typedef struct {
+  int32_t  numOfMnodes;               // tsNumOfMnodes
+  int32_t  mnodeEqualVnodeNum;        // tsMnodeEqualVnodeNum
+  int32_t  offlineThreshold;          // tsOfflineThreshold
+  int32_t  statusInterval;            // tsStatusInterval
+  char     arbitrator[TSDB_EP_LEN];   // tsArbitrator
+  char     timezone[64];              // tsTimezone
+  char     locale[TSDB_LOCALE_LEN];   // tsLocale
+  char     charset[TSDB_LOCALE_LEN];  // tsCharset
+} SClusterCfg;
+
 typedef struct {
   uint32_t    version;
   int32_t     dnodeId;
@@ -568,6 +579,7 @@ typedef struct {
   float       diskAvailable;  // GB
   uint8_t     alternativeRole;
   uint8_t     reserve[15];
+  SClusterCfg ClusterCfgPara;
   SVnodeLoad  load[];
 } SDMStatusMsg;
 

src/mnode/src/mnodeDnode.c

@@ -277,6 +277,20 @@ static void mnodeProcessCfgDnodeMsgRsp(SRpcMsg *rpcMsg) {
   mPrint("cfg dnode rsp is received");
 }
 
+static bool mnodeCheckClusterCfgPara(const SClusterCfg *clusterCfg) {
+  if (clusterCfg->numOfMnodes        != tsNumOfMnodes)        return false;
+  if (clusterCfg->mnodeEqualVnodeNum != tsMnodeEqualVnodeNum) return false;
+  if (clusterCfg->offlineThreshold   != tsOfflineThreshold)   return false;
+  if (clusterCfg->statusInterval     != tsStatusInterval)     return false;
+
+  if (0 != strncasecmp(clusterCfg->arbitrator, tsArbitrator, strlen(tsArbitrator))) return false;
+  if (0 != strncasecmp(clusterCfg->timezone, tsTimezone, strlen(tsTimezone)))       return false;
+  if (0 != strncasecmp(clusterCfg->locale, tsLocale, strlen(tsLocale)))              return false;
+  if (0 != strncasecmp(clusterCfg->charset, tsCharset, strlen(tsCharset)))           return false;
+    
+  return true;
+}
+
 static int32_t mnodeProcessDnodeStatusMsg(SMnodeMsg *pMsg) {
   SDMStatusMsg *pStatus = pMsg->rpcMsg.pCont;
   pStatus->dnodeId      = htonl(pStatus->dnodeId);
@@ -312,7 +326,6 @@ static int32_t mnodeProcessDnodeStatusMsg(SMnodeMsg *pMsg) {
   pDnode->alternativeRole  = pStatus->alternativeRole;
   pDnode->totalVnodes      = pStatus->numOfTotalVnodes; 
   pDnode->moduleStatus     = pStatus->moduleStatus;
-  pDnode->lastAccess       = tsAccessSquence;
   
   if (pStatus->dnodeId == 0) {
     mTrace("dnode:%d %s, first access", pDnode->dnodeId, pDnode->dnodeEp);
@@ -338,6 +351,14 @@ static int32_t mnodeProcessDnodeStatusMsg(SMnodeMsg *pMsg) {
   }
 
   if (pDnode->status == TAOS_DN_STATUS_OFFLINE) {
+    // Verify whether the cluster parameters are consistent when status change from offline to ready
+    bool ret = mnodeCheckClusterCfgPara(&(pStatus->ClusterCfgPara));
+    if (false == ret) {
+      mnodeDecDnodeRef(pDnode);
+      mError("dnode %s cluster cfg parameters inconsistent", pStatus->dnodeEp);
+      return TSDB_CODE_MND_CLUSTER_CFG_INCONSISTENT;
+    }
+    
     mTrace("dnode:%d, from offline to online", pDnode->dnodeId);
     pDnode->status = TAOS_DN_STATUS_READY;
     balanceUpdateMnode();
@@ -352,6 +373,8 @@ static int32_t mnodeProcessDnodeStatusMsg(SMnodeMsg *pMsg) {
     return TSDB_CODE_MND_OUT_OF_MEMORY;
   }
 
+  pDnode->lastAccess = tsAccessSquence;
+
   mnodeGetMnodeInfos(&pRsp->mnodes);
 
   pRsp->dnodeCfg.dnodeId = htonl(pDnode->dnodeId);

tests/script/unique/arbitrator/check_cluster_cfg_para.sim

+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/deploy.sh -n dnode5 -i 5
+system sh/deploy.sh -n dnode6 -i 6
+system sh/deploy.sh -n dnode7 -i 7
+
+
+system sh/cfg.sh -n dnode1 -c numOfMnodes -v 2
+system sh/cfg.sh -n dnode1 -c mnodeEqualVnodeNum -v 4
+system sh/cfg.sh -n dnode1 -c offlineThreshold -v 15
+system sh/cfg.sh -n dnode1 -c statusInterval -v 3
+system sh/cfg.sh -n dnode1 -c arbitrator -v $arbitrator
+#system sh/cfg.sh -n dnode1 -c timezone -v ""
+#system sh/cfg.sh -n dnode1 -c locale -v ""
+#system sh/cfg.sh -n dnode1 -c charset -v ""
+system sh/cfg.sh -n dnode1 -c balanceInterval -v 10
+
+######## dnode 2 the same with dnode1
+system sh/cfg.sh -n dnode2 -c numOfMnodes -v 2
+system sh/cfg.sh -n dnode2 -c mnodeEqualVnodeNum -v 4
+system sh/cfg.sh -n dnode2 -c offlineThreshold -v 15
+system sh/cfg.sh -n dnode2 -c statusInterval -v 3
+system sh/cfg.sh -n dnode2 -c arbitrator -v $arbitrator
+#system sh/cfg.sh -n dnode2 -c timezone -v ""
+#system sh/cfg.sh -n dnode2 -c locale -v ""
+#system sh/cfg.sh -n dnode2 -c charset -v ""
+system sh/cfg.sh -n dnode2 -c balanceInterval -v 10
+
+######## dnode 3 one para no same with dnode1
+system sh/cfg.sh -n dnode3 -c numOfMnodes -v 3
+system sh/cfg.sh -n dnode3 -c mnodeEqualVnodeNum -v 4
+system sh/cfg.sh -n dnode3 -c offlineThreshold -v 15
+system sh/cfg.sh -n dnode3 -c statusInterval -v 3
+system sh/cfg.sh -n dnode3 -c arbitrator -v $arbitrator
+#system sh/cfg.sh -n dnode3 -c timezone -v ""
+#system sh/cfg.sh -n dnode3 -c locale -v ""
+#system sh/cfg.sh -n dnode3 -c charset -v ""
+system sh/cfg.sh -n dnode3 -c balanceInterval -v 10
+
+######## dnode 4 one para no same with dnode1
+system sh/cfg.sh -n dnode4 -c numOfMnodes -v 2
+system sh/cfg.sh -n dnode4 -c mnodeEqualVnodeNum -v 5
+system sh/cfg.sh -n dnode4 -c offlineThreshold -v 15
+system sh/cfg.sh -n dnode4 -c statusInterval -v 3
+system sh/cfg.sh -n dnode4 -c arbitrator -v $arbitrator
+#system sh/cfg.sh -n dnode4 -c timezone -v ""
+#system sh/cfg.sh -n dnode4 -c locale -v ""
+#system sh/cfg.sh -n dnode4 -c charset -v ""
+system sh/cfg.sh -n dnode4 -c balanceInterval -v 10
+
+######## dnode 5 one para no same with dnode1
+system sh/cfg.sh -n dnode5 -c numOfMnodes -v 2
+system sh/cfg.sh -n dnode5 -c mnodeEqualVnodeNum -v 4
+system sh/cfg.sh -n dnode5 -c offlineThreshold -v 16
+system sh/cfg.sh -n dnode5 -c statusInterval -v 3
+system sh/cfg.sh -n dnode5 -c arbitrator -v $arbitrator
+#system sh/cfg.sh -n dnode5 -c timezone -v ""
+#system sh/cfg.sh -n dnode5 -c locale -v ""
+#system sh/cfg.sh -n dnode5 -c charset -v ""
+system sh/cfg.sh -n dnode5 -c balanceInterval -v 10
+
+
+######## dnode 6 one para no same with dnode1
+system sh/cfg.sh -n dnode6 -c numOfMnodes -v 2
+system sh/cfg.sh -n dnode6 -c mnodeEqualVnodeNum -v 4
+system sh/cfg.sh -n dnode6 -c offlineThreshold -v 15 
+system sh/cfg.sh -n dnode6 -c statusInterval -v 2
+system sh/cfg.sh -n dnode6 -c arbitrator -v $arbitrator
+#system sh/cfg.sh -n dnode6 -c timezone -v ""
+#system sh/cfg.sh -n dnode6 -c locale -v ""
+#system sh/cfg.sh -n dnode6 -c charset -v ""
+system sh/cfg.sh -n dnode6 -c balanceInterval -v 10
+
+
+######## dnode 7 one para no same with dnode1
+system sh/cfg.sh -n dnode7 -c numOfMnodes -v 2
+system sh/cfg.sh -n dnode7 -c mnodeEqualVnodeNum -v 4
+system sh/cfg.sh -n dnode7 -c offlineThreshold -v 15
+system sh/cfg.sh -n dnode7 -c statusInterval -v 3
+system sh/cfg.sh -n dnode7 -c arbitrator -v "plum-VirtualBox:8001"
+#system sh/cfg.sh -n dnode7 -c timezone -v ""
+#system sh/cfg.sh -n dnode7 -c locale -v ""
+#system sh/cfg.sh -n dnode7 -c charset -v ""
+system sh/cfg.sh -n dnode7 -c balanceInterval -v 10
+
+print ============== step0: start tarbitrator
+system sh/exec_tarbitrator.sh -s start
+
+print ============== step1: start dnode1
+system sh/exec.sh -n dnode1 -s start
+sleep 3000
+sql connect
+
+print ============== step2: start dnode2~7 and add into cluster 
+system sh/exec.sh -n dnode2 -s start
+system sh/exec.sh -n dnode3 -s start
+system sh/exec.sh -n dnode4 -s start
+system sh/exec.sh -n dnode5 -s start
+system sh/exec.sh -n dnode6 -s start
+system sh/exec.sh -n dnode7 -s start
+sql create dnode $hostname2
+sql create dnode $hostname3
+sql create dnode $hostname4
+sql create dnode $hostname5
+sql create dnode $hostname6
+sql create dnode $hostname7
+sleep 10000
+
+wait_dnode_created:
+sql show dnodes
+if $rows != 7 then
+  sleep 2000
+  goto wait_dnode_created
+endi
+print $data0_1  $data1_1  $data2_1  $data3_1  $data4_1
+print $data0_2  $data1_2  $data2_2  $data3_2  $data4_2
+print $data0_3  $data1_3  $data2_3  $data3_3  $data4_3
+print $data0_4  $data1_4  $data2_4  $data3_4  $data4_4
+print $data0_5  $data1_5  $data2_5  $data3_5  $data4_5
+print $data0_6  $data1_6  $data2_6  $data3_6  $data4_6
+print $data0_7  $data1_7  $data2_7  $data3_7  $data4_7
+$dnode1Status = $data4_1  
+$dnode2Status = $data4_2
+$dnode3Status = $data4_3  
+$dnode4Status = $data4_4 
+$dnode5Status = $data4_5
+$dnode6Status = $data4_6  
+$dnode7Status = $data4_7 
+
+if $dnode1Status != ready then
+  return -1
+endi
+if $dnode2Status != ready then
+  return -1
+endi
+if $dnode3Status != offline then
+  return -1
+endi
+if $dnode4Status != offline then
+  return -1
+endi
+if $dnode5Status != offline then
+  return -1
+endi
+if $dnode6Status != offline then
+  return -1
+endi
+if $dnode7Status != offline then
+  return -1
+endi
+
+sleep 10000
+
+wait_dnode_offline_overtime_dropped:
+sql show dnodes
+print $data0_1  $data1_1  $data2_1  $data3_1  $data4_1
+print $data0_2  $data1_2  $data2_2  $data3_2  $data4_2
+print $data0_3  $data1_3  $data2_3  $data3_3  $data4_3
+print $data0_4  $data1_4  $data2_4  $data3_4  $data4_4
+print $data0_5  $data1_5  $data2_5  $data3_5  $data4_5
+print $data0_6  $data1_6  $data2_6  $data3_6  $data4_6
+print $data0_7  $data1_7  $data2_7  $data3_7  $data4_7
+if $rows != 2 then
+  sleep 2000
+  goto wait_dnode_offline_overtime_dropped
+endi
+print $data0_1  $data1_1  $data2_1  $data3_1  $data4_1
+print $data0_2  $data1_2  $data2_2  $data3_2  $data4_2
+print $data0_3  $data1_3  $data2_3  $data3_3  $data4_3
+print $data0_4  $data1_4  $data2_4  $data3_4  $data4_4
+print $data0_5  $data1_5  $data2_5  $data3_5  $data4_5
+print $data0_6  $data1_6  $data2_6  $data3_6  $data4_6
+print $data0_7  $data1_7  $data2_7  $data3_7  $data4_7
+$dnode1Status = $data4_1  
+$dnode2Status = $data4_2
+$dnode3Status = $data4_3  
+$dnode4Status = $data4_4 
+$dnode5Status = $data4_5
+$dnode6Status = $data4_6  
+$dnode7Status = $data4_7 
+
+if $dnode1Status != ready then
+  return -1
+endi
+if $dnode2Status != ready then
+  return -1
+endi

tests/script/unique/arbitrator/dn3_mn1_nw_disable_timeout_autoDropDnode.sim

@@ -5,11 +5,11 @@ system sh/deploy.sh -n dnode3 -i 3
 system sh/deploy.sh -n dnode4 -i 4
 system sh/deploy.sh -n dnode5 -i 5
 
-system sh/cfg.sh -n dnode1 -c numOfMPeers -v 1
-system sh/cfg.sh -n dnode2 -c numOfMPeers -v 1
-system sh/cfg.sh -n dnode3 -c numOfMPeers -v 1
-system sh/cfg.sh -n dnode4 -c numOfMPeers -v 1
-system sh/cfg.sh -n dnode5 -c numOfMPeers -v 1
+system sh/cfg.sh -n dnode1 -c numOfMnodes -v 1
+system sh/cfg.sh -n dnode2 -c numOfMnodes -v 1
+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 dnode5 -c numOfMnodes -v 1
 
 system sh/cfg.sh -n dnode1 -c walLevel -v 1 
 system sh/cfg.sh -n dnode2 -c walLevel -v 1

tests/script/unique/arbitrator/dn3_mn1_vnode_change.sim

@@ -96,7 +96,12 @@ endi
 print ============== step3: stop dnode4, and remove its vnodeX subdirector 
 system sh/exec.sh -n dnode4 -s stop -x SIGINT
 sleep $sleepTimer
+$loopCnt = 0
 wait_dnode4_offline_0:
+$loopCnt = $loopCnt + 1
+if $loopCnt == 10 then
+  return -1
+endi
 sql show dnodes
 if $rows != 4 then
   sleep 2000
@@ -148,7 +153,14 @@ sleep 1000
 print ============== step4: restart dnode4, waiting sync end 
 system sh/exec.sh -n dnode4 -s start
 sleep $sleepTimer
+
+$loopCnt = 0
 wait_dnode4_reready:
+$loopCnt = $loopCnt + 1
+if $loopCnt == 10 then
+  return -1
+endi
+
 sql show dnodes
 if $rows != 4 then
   sleep 2000
@@ -171,7 +183,13 @@ if $dnode4Status != ready then
   goto wait_dnode4_reready
 endi
 
+$loopCnt = 0
 wait_dnode4_vgroup_slave:
+$loopCnt = $loopCnt + 1
+if $loopCnt == 10 then
+  return -1
+endi
+
 sql show vgroups
 if $rows != 1 then
   sleep 2000
@@ -200,7 +218,13 @@ system sh/exec.sh -n dnode2 -s stop
 system sh/exec.sh -n dnode3 -s stop
 sleep $sleepTimer
 
+
+$loopCnt = 0
 wait_dnode23_offline:
+$loopCnt = $loopCnt + 1
+if $loopCnt == 10 then
+  return -1
+endi
 sql show dnodes
 if $rows != 4 then
   sleep 2000
@@ -231,7 +255,13 @@ if $dnode4Status != ready then
   goto wait_dnode23_offline
 endi
 
+$loopCnt = 0
 wait_dnode4_vgroup_master:
+$loopCnt = $loopCnt + 1
+if $loopCnt == 10 then
+  return -1
+endi
+
 sql show vgroups
 if $rows != 1 then
   sleep 2000