Merge branch 'fix/mnode' of https://github.com/taosdata/TDengine into mnode

source/dnode/mgmt/mgmt_dnode/src/dmHandle.c

@@ -92,6 +92,13 @@ void dmSendStatusReq(SDnodeMgmt *pMgmt) {
   SEpSet epSet = {0};
   dmGetMnodeEpSet(pMgmt->pData, &epSet);
   rpcSendRecv(pMgmt->msgCb.clientRpc, &epSet, &rpcMsg, &rpcRsp);
+  if (rpcRsp.code != 0) {
+    dError("failed to send status msg since %s, numOfEps:%d inUse:%d", tstrerror(rpcRsp.code), epSet.numOfEps,
+           epSet.inUse);
+    for (int32_t i = 0; i < epSet.numOfEps; ++i) {
+      dDebug("index:%d, mnode ep:%s:%u", i, epSet.eps[i].fqdn, epSet.eps[i].port);
+    }
+  }
   dmProcessStatusRsp(pMgmt, &rpcRsp);
 }
 

source/dnode/mnode/impl/src/mndCluster.c

@@ -144,6 +144,7 @@ _OVER:
 
 static int32_t mndClusterActionInsert(SSdb *pSdb, SClusterObj *pCluster) {
   mTrace("cluster:%" PRId64 ", perform insert action, row:%p", pCluster->id, pCluster);
+  pSdb->pMnode->clusterId = pCluster->id;
   return 0;
 }
 

source/dnode/mnode/impl/src/mndDnode.c

@@ -441,7 +441,7 @@ static int32_t mndProcessStatusReq(SRpcMsg *pReq) {
     pDnode->numOfSupportVnodes = statusReq.numOfSupportVnodes;
 
     SStatusRsp statusRsp = {0};
-    statusRsp.dnodeVer = sdbGetTableVer(pMnode->pSdb, SDB_DNODE);
+    statusRsp.dnodeVer = sdbGetTableVer(pMnode->pSdb, SDB_DNODE) + sdbGetTableVer(pMnode->pSdb, SDB_MNODE);
     statusRsp.dnodeCfg.dnodeId = pDnode->id;
     statusRsp.dnodeCfg.clusterId = pMnode->clusterId;
     statusRsp.pDnodeEps = taosArrayInit(mndGetDnodeSize(pMnode), sizeof(SDnodeEp));

source/dnode/mnode/impl/src/mndMnode.c

@@ -233,7 +233,7 @@ void mndGetMnodeEpSet(SMnode *pMnode, SEpSet *pEpSet) {
     if (pObj->pDnode == NULL) {
       mError("mnode:%d, no corresponding dnode exists", pObj->id);
     } else {
-      if (pObj->state == TAOS_SYNC_STATE_LEADER) {
+      if (pObj->id == pMnode->selfDnodeId || pObj->state == TAOS_SYNC_STATE_LEADER) {
         pEpSet->inUse = pEpSet->numOfEps;
       }
       addEpIntoEpSet(pEpSet, pObj->pDnode->fqdn, pObj->pDnode->port);

source/libs/sync/src/syncAppendEntries.c

@@ -363,28 +363,32 @@ int32_t syncNodeOnAppendEntriesCb(SSyncNode* ths, SyncAppendEntries* pMsg) {
                     break;
                   }
                 }
-                ASSERT(hit == true);
 
+                SReConfigCbMeta cbMeta = {0};
                 bool isDrop;
-                syncNodeUpdateConfig(ths, &newSyncCfg, &isDrop);
 
-                // change isStandBy to normal
+                // I am in newConfig
-                if (!isDrop) {
+                if (hit) {
-                  if (ths->state == TAOS_SYNC_STATE_LEADER) {
+                  syncNodeUpdateConfig(ths, &newSyncCfg, &isDrop);
-                    syncNodeBecomeLeader(ths);
+
-                  } else {
+                  // change isStandBy to normal
-                    syncNodeBecomeFollower(ths);
+                  if (!isDrop) {
+                    if (ths->state == TAOS_SYNC_STATE_LEADER) {
+                      syncNodeBecomeLeader(ths);
+                    } else {
+                      syncNodeBecomeFollower(ths);
+                    }
                   }
+
+                  char* sOld = syncCfg2Str(&oldSyncCfg);
+                  char* sNew = syncCfg2Str(&newSyncCfg);
+                  sInfo("==config change== 0x11 old:%s new:%s isDrop:%d \n", sOld, sNew, isDrop);
+                  taosMemoryFree(sOld);
+                  taosMemoryFree(sNew);
                 }
 
-                char* sOld = syncCfg2Str(&oldSyncCfg);
+                // always call FpReConfigCb
-                char* sNew = syncCfg2Str(&newSyncCfg);
-                sInfo("==config change== 0x11 old:%s new:%s isDrop:%d \n", sOld, sNew, isDrop);
-                taosMemoryFree(sOld);
-                taosMemoryFree(sNew);
-
                 if (ths->pFsm->FpReConfigCb != NULL) {     
-                  SReConfigCbMeta cbMeta = {0};
                   cbMeta.code = 0;
                   cbMeta.currentTerm = ths->pRaftStore->currentTerm;
                   cbMeta.index = pEntry->index;

source/libs/sync/src/syncCommit.c

@@ -141,7 +141,10 @@ void syncMaybeAdvanceCommitIndex(SSyncNode* pSyncNode) {
                 break;
               }
             }
-            ASSERT(hit == true);
+
+            if (pSyncNode->state == TAOS_SYNC_STATE_LEADER) {
+              ASSERT(hit == true);
+            }
 
             bool isDrop;
             syncNodeUpdateConfig(pSyncNode, &newSyncCfg, &isDrop);

tests/script/tsim/mnode/basic1.sim

@@ -6,15 +6,6 @@ system sh/exec.sh -n dnode2 -s start
 sql connect
 
 print =============== show dnodes
-sql show dnodes;
-if $rows != 1 then 
-  return -1
-endi
-
-if $data00 != 1 then 
-  return -1
-endi
-
 sql show mnodes;
 if $rows != 1 then 
   return -1
@@ -30,63 +21,55 @@ endi
 
 print =============== create dnodes
 sql create dnode $hostname port 7200
-sleep 2000
+$x = 0
-
+step1: 
-sql show dnodes;
+	$x = $x + 1
-if $rows != 2 then 
+	sleep 500
-  return -1
+	if $x == 20 then
+		return -1
+	endi
+sql show dnodes -x step1
+if $data(1)[4] != ready then
+  goto step1
 endi
-
+if $data(2)[4] != ready then
-if $data00 != 1 then 
+  goto step1
-  return -1
-endi
-
-if $data10 != 2 then 
-  return -1
-endi
-
-print $data02
-if $data02 != 0 then 
-  return -1
-endi
-
-if $data12 != 0 then 
-  return -1
-endi
-
-if $data04 != ready then 
-  return -1
-endi
-
-if $data14 != ready then 
-  return -1
-endi
-
-sql show mnodes;
-if $rows != 1 then 
-  return -1
-endi
-
-if $data00 != 1 then 
-  return -1
-endi
-
-if $data02 != LEADER then 
-  return -1
 endi
 
 print =============== create drop mnode 1
 sql_error create mnode on dnode 1
 sql_error drop mnode on dnode 1
-
-print =============== create drop mnode 2
 sql create mnode on dnode 2
+
+$x = 0
+step1: 
+	$x = $x + 1
+	sleep 1000
+	if $x == 20 then
+		return -1
+	endi
 sql show mnodes
+print $data(1)[0] $data(1)[1] $data(1)[2] 
+print $data(2)[0] $data(2)[1] $data(2)[2] 
+
 if $rows != 2 then 
   return -1
 endi
-sql_error create mnode on dnode 2
+if $data(1)[0] != 1 then 
+  return -1
+endi
+if $data(1)[2] != LEADER then 
+  return -1
+endi
+if $data(2)[0] != 2 then 
+  return -1
+endi
+if $data(2)[2] != FOLLOWER then 
+  goto step1
+endi
 
+sleep 2000
+print ============ drop mnodes
 sql drop mnode on dnode 2
 sql show mnodes
 if $rows != 1 then 
@@ -94,6 +77,35 @@ if $rows != 1 then
 endi
 sql_error drop mnode on dnode 2
 
+$x = 0
+step2: 
+	$x = $x + 1
+	sleep 1000
+	if $x == 20 then
+		return -1
+	endi
+sql show mnodes
+print $data(1)[0] $data(1)[1] $data(1)[2] 
+print $data(2)[0] $data(2)[1] $data(2)[2] 
+
+if $rows != 2 then 
+  return -1
+endi
+if $data(1)[0] != 1 then 
+  return -1
+endi
+if $data(1)[2] != LEADER then 
+  return -1
+endi
+if $data(2)[0] != NULL then 
+  goto step2
+endi
+if $data(2)[2] != NULL then 
+  goto step2
+endi
+
+sleep 2000
+
 print =============== create drop mnodes
 sql create mnode on dnode 2
 sql show mnodes
@@ -101,17 +113,32 @@ if $rows != 2 then
   return -1
 endi
 
-print =============== restart
+$x = 0
-system sh/exec.sh -n dnode1 -s stop -x SIGINT
+step3: 
-system sh/exec.sh -n dnode2 -s stop -x SIGINT
+	$x = $x + 1
-system sh/exec.sh -n dnode1 -s start
+	sleep 1000
-system sh/exec.sh -n dnode2 -s start
+	if $x == 20 then
-
+		return -1
-sleep 2000
+	endi
 sql show mnodes
+print $data(1)[0] $data(1)[1] $data(1)[2] 
+print $data(2)[0] $data(2)[1] $data(2)[2] 
+
 if $rows != 2 then 
   return -1
 endi
+if $data(1)[0] != 1 then 
+  return -1
+endi
+if $data(1)[2] != LEADER then 
+  return -1
+endi
+if $data(2)[0] != 2 then 
+  return -1
+endi
+if $data(2)[2] != FOLLOWER then 
+  goto step3
+endi
 
 system sh/exec.sh -n dnode1 -s stop -x SIGINT
 system sh/exec.sh -n dnode2 -s stop -x SIGINT

tests/script/tsim/mnode/basic2.sim

@@ -21,29 +21,31 @@ endi
 
 print =============== create dnodes
 sql create dnode $hostname port 7200
-sql create dnode $hostname port 7300
+$x = 0
-sleep 2000
+step1: 
-
+	$x = $x + 1
-sql show dnodes;
+	sleep 500
-if $rows != 3 then 
+	if $x == 20 then
-  return -1
+		return -1
+	endi
+sql show dnodes -x step1
+if $data(1)[4] != ready then
+  goto step1
 endi
-
+if $data(2)[4] != ready then
-sql show mnodes;
+  goto step1
-if $rows != 1 then 
-  return -1
-endi
-
-if $data00 != 1 then 
-  return -1
-endi
-
-if $data02 != LEADER then 
-  return -1
 endi
 
 print =============== create mnode 2
 sql create mnode on dnode 2
+
+$x = 0
+step1: 
+	$x = $x + 1
+	sleep 1000
+	if $x == 20 then
+		return -1
+	endi
 sql show mnodes
 print $data(1)[0] $data(1)[1] $data(1)[2] 
 print $data(2)[0] $data(2)[1] $data(2)[2] 
@@ -60,8 +62,8 @@ endi
 if $data(2)[0] != 2 then 
   return -1
 endi
-if $data(2)[2] == LEADER then 
+if $data(2)[2] != FOLLOWER then 
-  return -1
+  goto step1
 endi
 
 print =============== create user
@@ -71,45 +73,47 @@ if $rows != 2 then
   return -1
 endi
 
-#sql create database db
+sql create database db
-#sql show databases
+sql show databases
-#if $rows != 3 then 
+if $rows != 3 then 
-#  return -1
+  return -1
-#endi
+endi
 
 sleep 5000
 
+print =============== restart
 system sh/exec.sh -n dnode1 -s stop
 system sh/exec.sh -n dnode2 -s stop
-sleep 100
-return
 system sh/exec.sh -n dnode1 -s start
 system sh/exec.sh -n dnode2 -s start
 
 sql connect
-
 sql show mnodes
 if $rows != 2 then 
   return -1
 endi
-if $data(1)[0] != 1 then 
-  return -1
-endi
-if $data(1)[2] != LEADER then 
-  return -1
-endi
 
 sql show users
 if $rows != 2 then 
   return -1
 endi
 
-#sql show databases
+sql show databases
-#if $rows != 3 then 
+if $rows != 3 then 
-#  return -1
+  return -1
-#endi
+endi
 
-return
+sql show dnodes
+if $data(1)[4] != ready then
+  return -1
+endi
+if $data(2)[4] != ready then
+  return -1
+endi
+
+print =============== insert data
+sql create table db.stb (ts timestamp, i int) tags (j int)
+sql create table db.ctb using db.stb tags(1);
 
 system sh/exec.sh -n dnode1 -s stop
 system sh/exec.sh -n dnode2 -s stop

tests/script/tsim/mnode/basic3.sim

@@ -15,11 +15,10 @@ $x = 0
 step1: 
 	$x = $x + 1
 	sleep 1000
-	if $x == 10 then
+	if $x == 20 then
 		return -1
 	endi
-
+sql show dnodes -x step1
-sql show dnodes
 if $data(1)[4] != ready then
   goto step1
 endi
@@ -38,15 +37,10 @@ $x = 0
 step2: 
 	$x = $x + 1
 	sleep 1000
-	if $x == 10 then
+	if $x == 20 then
 		return -1
 	endi
-
+sql show mnodes -x step2
-sql show mnodes
-print $data(1)[0] $data(1)[1] $data(1)[2] 
-print $data(2)[0] $data(2)[1] $data(2)[2] 
-print $data(3)[0] $data(3)[1] $data(3)[2] 
-
 if $data(1)[2] != LEADER then
   goto step2
 endi
@@ -64,44 +58,80 @@ if $rows != 2 then
   return -1
 endi
 
+# wait mnode2 mnode3 recv data finish
 sleep 10000
 
 print =============== step4: stop dnode1
 system sh/exec.sh -n dnode1 -s stop
 
-
+$x = 0
-return
+step4: 
-system sh/exec.sh -n dnode2 -s stop
+	$x = $x + 1
-sleep 100
+	sleep 1000
-return
+	if $x == 20 then
-system sh/exec.sh -n dnode1 -s start
+		return -1
-system sh/exec.sh -n dnode2 -s start
+	endi
-
+sql show mnodes -x step4
-sleep 10000
+print $data(1)[0] $data(1)[1] $data(1)[2] 
-sql connect
+print $data(2)[0] $data(2)[1] $data(2)[2] 
-
+print $data(3)[0] $data(3)[1] $data(3)[2] 
-sql show mnodes
-if $rows != 2 then 
-  return -1
-endi
-if $data(1)[0] != 1 then 
-  return -1
-endi
-if $data(1)[2] != LEADER then 
-  return -1
-endi
 
 sql show users
 if $rows != 2 then 
   return -1
 endi
 
-#sql show databases
+sleep 1000
-#if $rows != 3 then 
+sql show dnodes
-#  return -1
+if $data(2)[4] != ready then
-#endi
+  return -1
+endi
+if $data(3)[4] != ready then
+  return -1
+endi
 
-return
+print =============== step5: stop dnode1
+system sh/exec.sh -n dnode1 -s start
+system sh/exec.sh -n dnode2 -s stop
+
+$x = 0
+step5: 
+	$x = $x + 1
+	sleep 1000
+	if $x == 20 then
+		return -1
+	endi
+sql show mnodes -x step5
+print $data(1)[0] $data(1)[1] $data(1)[2] 
+print $data(2)[0] $data(2)[1] $data(2)[2] 
+print $data(3)[0] $data(3)[1] $data(3)[2] 
+
+sql show users
+if $rows != 2 then 
+  return -1
+endi
+
+print =============== step6: stop dnode1
+system sh/exec.sh -n dnode2 -s start
+system sh/exec.sh -n dnode3 -s stop
+
+$x = 0
+step6: 
+	$x = $x + 1
+	sleep 1000
+	if $x == 20 then
+		return -1
+	endi
+sql show mnodes -x step6
+print $data(1)[0] $data(1)[1] $data(1)[2] 
+print $data(2)[0] $data(2)[1] $data(2)[2] 
+print $data(3)[0] $data(3)[1] $data(3)[2] 
+
+sql show users
+if $rows != 2 then 
+  return -1
+endi
 
 system sh/exec.sh -n dnode1 -s stop
 system sh/exec.sh -n dnode2 -s stop
+system sh/exec.sh -n dnode3 -s stop