TD-1257 mnode module startup check condition is wrong

src/mnode/src/mnodeMain.c

@@ -21,6 +21,7 @@
 #include "tgrant.h"
 #include "ttimer.h"
 #include "tglobal.h"
+#include "mnode.h"
 #include "dnode.h"
 #include "mnodeDef.h"
 #include "mnodeInt.h"
@@ -107,14 +108,19 @@ int32_t mnodeStartSystem() {
   tsMgmtIsRunning = true;
 
   mInfo("mnode is initialized successfully");
+
+  sdbUpdateSync();
+
   return 0;
 }
 
 int32_t mnodeInitSystem() {
   mnodeInitTimer();
-  if (!mnodeNeedStart()) return 0;
+  if (mnodeNeedStart()) {
     return mnodeStartSystem();
   }
+  return 0;
+}
 
 void mnodeCleanupSystem() {
   mInfo("starting to clean up mnode");
@@ -159,15 +165,20 @@ static void mnodeCleanupTimer() {
 
 static bool mnodeNeedStart() {
   struct stat dirstat;
-  bool fileExist = (stat(tsMnodeDir, &dirstat) == 0);
+  char mnodeFileName[TSDB_FILENAME_LEN * 2] = {0};
+  sprintf(mnodeFileName, "%s/wal0", tsMnodeDir);
+
+  bool fileExist = (stat(mnodeFileName, &dirstat) == 0);
   bool asMaster = (strcmp(tsFirst, tsLocalEp) == 0);
 
   if (asMaster || fileExist) {
+    mDebug("mnode module start, asMaster:%d fileExist:%d", asMaster, fileExist);
     return true;
-  }
+  } else {
-
+    mDebug("mnode module won't start, asMaster:%d fileExist:%d", asMaster, fileExist);
     return false;
   }
+}
 
 bool mnodeIsRunning() {
   return tsMgmtIsRunning;

src/mnode/src/mnodeSdb.c

@@ -291,6 +291,11 @@ static void sdbConfirmForward(void *ahandle, void *param, int32_t code) {
 }
 
 void sdbUpdateSync() {
+  if (!mnodeIsRunning()) {
+    mDebug("mnode not start yet, update sync info later");
+    return;
+  }
+
   SSyncCfg syncCfg = {0};
   int32_t  index = 0;
 

tests/script/jenkins/basic.txt

@@ -294,6 +294,7 @@ cd ../../../debug; make
 ./test.sh -f unique/stable/replica3_dnode6.sim
 ./test.sh -f unique/stable/replica3_vnode3.sim
 
+./test.sh -f unique/mnode/mgmt21.sim
 ./test.sh -f unique/mnode/mgmt22.sim
 ./test.sh -f unique/mnode/mgmt23.sim
 ./test.sh -f unique/mnode/mgmt24.sim

tests/script/jenkins/unique.txt

@@ -1,22 +1,11 @@
 cd ../../../debug; cmake ..
 cd ../../../debug; make
 
-./test.sh -f unique/account/account_create.sim
-./test.sh -f unique/account/account_delete.sim
-./test.sh -f unique/account/account_len.sim
-./test.sh -f unique/account/authority.sim
-./test.sh -f unique/account/basic.sim
-./test.sh -f unique/account/paras.sim
-./test.sh -f unique/account/pass_alter.sim
-./test.sh -f unique/account/pass_len.sim
-./test.sh -f unique/account/usage.sim
-./test.sh -f unique/account/user_create.sim
-./test.sh -f unique/account/user_len.sim
-
 ./test.sh -f unique/big/balance.sim
 ./test.sh -f unique/big/maxvnodes.sim
 ./test.sh -f unique/big/tcp.sim
 
+./test.sh -f unique/cluster/alter.sim
 ./test.sh -f unique/cluster/balance1.sim
 ./test.sh -f unique/cluster/balance2.sim
 ./test.sh -f unique/cluster/balance3.sim
@@ -61,6 +50,7 @@ cd ../../../debug; make
 ./test.sh -f unique/stable/replica3_dnode6.sim
 ./test.sh -f unique/stable/replica3_vnode3.sim
 
+./test.sh -f unique/mnode/mgmt21.sim
 ./test.sh -f unique/mnode/mgmt22.sim
 ./test.sh -f unique/mnode/mgmt23.sim
 ./test.sh -f unique/mnode/mgmt24.sim
@@ -77,31 +67,16 @@ cd ../../../debug; make
 ./test.sh -f unique/vnode/replica3_repeat.sim
 ./test.sh -f unique/vnode/replica3_vgroup.sim
 
-./test.sh -f general/stream/metrics_1.sim
-./test.sh -f general/stream/metrics_del.sim
-./test.sh -f general/stream/metrics_n.sim
-./test.sh -f general/stream/metrics_replica1_vnoden.sim
-#./test.sh -f general/stream/new_stream.sim
-./test.sh -f general/stream/restart_stream.sim
-./test.sh -f general/stream/stream_1.sim
-./test.sh -f general/stream/stream_2.sim
-./test.sh -f general/stream/stream_3.sim
-./test.sh -f general/stream/stream_restart.sim
-./test.sh -f general/stream/table_1.sim
-./test.sh -f general/stream/table_del.sim
-./test.sh -f general/stream/table_n.sim
-./test.sh -f general/stream/table_replica1_vnoden.sim
-
 ./test.sh -f unique/arbitrator/check_cluster_cfg_para.sim
-./test.sh -f unique/arbitrator/dn2_mn1_cache_file_sync.sim
+#./test.sh -f unique/arbitrator/dn2_mn1_cache_file_sync.sim
 ./test.sh -f unique/arbitrator/dn3_mn1_full_createTableFail.sim
-./test.sh -f unique/arbitrator/dn3_mn1_full_dropDnodeFail.sim
 ./test.sh -f unique/arbitrator/dn3_mn1_multiCreateDropTable.sim
-./test.sh -f unique/arbitrator/dn3_mn1_nw_disable_timeout_autoDropDnode.sim
+#./test.sh -f unique/arbitrator/dn3_mn1_nw_disable_timeout_autoDropDnode.sim
-./test.sh -f unique/arbitrator/dn3_mn1_replica2_wal1_AddDelDnode.sim
+#./test.sh -f unique/arbitrator/dn3_mn1_replica2_wal1_AddDelDnode.sim
 ./test.sh -f unique/arbitrator/dn3_mn1_replica_change_dropDnod.sim
 ./test.sh -f unique/arbitrator/dn3_mn1_replica_change.sim
-./test.sh -f unique/arbitrator/dn3_mn1_stopDnode_timeout.sim
+#./test.sh -f unique/arbitrator/dn3_mn1_stopDnode_timeout.sim
+# lower the priority while file corruption
 #./test.sh -f unique/arbitrator/dn3_mn1_vnode_change.sim
 #./test.sh -f unique/arbitrator/dn3_mn1_vnode_corruptFile_offline.sim
 #./test.sh -f unique/arbitrator/dn3_mn1_vnode_corruptFile_online.sim
@@ -126,6 +101,7 @@ cd ../../../debug; make
 ./test.sh -f unique/arbitrator/replica_changeWithArbitrator.sim
 ./test.sh -f unique/arbitrator/sync_replica2_alterTable_add.sim
 ./test.sh -f unique/arbitrator/sync_replica2_alterTable_drop.sim
+
 ./test.sh -f unique/arbitrator/sync_replica2_dropDb.sim
 ./test.sh -f unique/arbitrator/sync_replica2_dropTable.sim
 ./test.sh -f unique/arbitrator/sync_replica3_alterTable_add.sim

tests/script/unique/mnode/mgmt21.sim

@@ -0,0 +1,44 @@
+system sh/stop_dnodes.sh
+system sh/deploy.sh -n dnode1 -i 1
+system sh/deploy.sh -n dnode2 -i 2
+
+system sh/cfg.sh -n dnode1 -c numOfMnodes -v 2
+system sh/cfg.sh -n dnode2 -c numOfMnodes -v 2
+
+print ============== step1
+system sh/exec.sh -n dnode2 -s start
+sleep 10000
+
+system sh/exec.sh -n dnode1 -s start
+sql connect
+
+sql show mnodes
+print dnode1 ==> $data2_1
+print dnode2 ==> $data2_2
+if $data2_1 != master then
+  return -1
+endi
+
+print ============== step2
+sql create dnode $hostname2
+
+$x = 0
+show2: 
+	$x = $x + 1
+	sleep 2000
+	if $x == 10 then 
+		return -1
+	endi
+	
+sql show mnodes
+print dnode1 ==> $data2_1
+print dnode2 ==> $data2_2
+if $data2_1 != master then
+	goto show2
+endi
+if $data2_2 != slave then
+	goto show2
+endi
+
+system sh/exec.sh -n dnode1 -s stop  -x SIGINT
+system sh/exec.sh -n dnode2 -s stop  -x SIGINT