-
Enhancement
-
Resolution: Done
-
Major
-
3.6.2
-
None
Say we have 2 clusters connected via RELAY2, AB and CD. A is the site master of AB, C is the site master of CD, and A is the coordinator of the bridge cluster.
When node A is stopped, it first leaves the AB cluster, and then the bridge cluster. When B receives the new B view, it tries to join the bridge cluster. But because RELAY2 hasn't yet finished stopping the relay channel on A, B still sees A as the coordinator of the bridge channel, and sends the JOIN_REQ message to it.
The bridge channel stops on A before receiving B's join response, so B won't receive any response and will eventually time out. If the user tries to stop node B, it will block waiting for the JChannel lock:
"testng@1" prio=5 tid=0x1 nid=NA waiting for monitor entry java.lang.Thread.State: BLOCKED waiting for Incoming-1,NodeB@2894 to release lock on <0xd26> (a org.jgroups.JChannel) at org.jgroups.JChannel.close(JChannel.java:385) at org.jgroups.util.Util.close(Util.java:408) at org.jgroups.protocols.relay.Relayer$Bridge.stop(Relayer.java:256) at org.jgroups.protocols.relay.Relayer.stop(Relayer.java:109) at org.jgroups.protocols.relay.RELAY2.stop(RELAY2.java:280) at org.jgroups.stack.ProtocolStack.stopStack(ProtocolStack.java:1015) at org.jgroups.JChannel.stopStack(JChannel.java:1003) at org.jgroups.JChannel.disconnect(JChannel.java:373) - locked <0xd54> (a org.jgroups.JChannel) at org.infinispan.remoting.transport.jgroups.JGroupsTransport.stop(JGroupsTransport.java:262) at sun.reflect.NativeMethodAccessorImpl.invoke0(NativeMethodAccessorImpl.java:-1) at sun.reflect.NativeMethodAccessorImpl.invoke(NativeMethodAccessorImpl.java:62) at sun.reflect.DelegatingMethodAccessorImpl.invoke(DelegatingMethodAccessorImpl.java:43) at java.lang.reflect.Method.invoke(Method.java:497) "Incoming-1,B@2894" prio=5 tid=0x2f nid=NA runnable java.lang.Thread.State: RUNNABLE blocks testng@1 at org.jgroups.protocols.pbcast.ClientGmsImpl.joinInternal(ClientGmsImpl.java:114) at org.jgroups.protocols.pbcast.ClientGmsImpl.join(ClientGmsImpl.java:44) at org.jgroups.protocols.pbcast.GMS.down(GMS.java:1084) at org.jgroups.protocols.FlowControl.down(FlowControl.java:353) at org.jgroups.protocols.FRAG2.down(FRAG2.java:136) at org.jgroups.protocols.RSVP.down(RSVP.java:153) at org.jgroups.stack.ProtocolStack.down(ProtocolStack.java:1038) at org.jgroups.JChannel.down(JChannel.java:791) at org.jgroups.JChannel._connect(JChannel.java:564) at org.jgroups.JChannel.connect(JChannel.java:294) - locked <0xd26> (a org.jgroups.JChannel) at org.jgroups.JChannel.connect(JChannel.java:279) at org.jgroups.protocols.relay.Relayer$Bridge.start(Relayer.java:250) at org.jgroups.protocols.relay.Relayer.start(Relayer.java:86) at org.jgroups.protocols.relay.RELAY2.startRelayer(RELAY2.java:681) at org.jgroups.protocols.relay.RELAY2.handleView(RELAY2.java:663) at org.jgroups.protocols.relay.RELAY2.up(RELAY2.java:416) at org.jgroups.protocols.RSVP.up(RSVP.java:201) at org.jgroups.protocols.FRAG2.up(FRAG2.java:165) at org.jgroups.protocols.FlowControl.up(FlowControl.java:394) at org.jgroups.protocols.tom.TOA.up(TOA.java:121) at org.jgroups.protocols.pbcast.GMS.installView(GMS.java:732) - locked <0xd2e> (a org.jgroups.protocols.pbcast.GMS) at org.jgroups.protocols.pbcast.ParticipantGmsImpl.handleViewChange(ParticipantGmsImpl.java:146)
Ideally, I'd like B to realize that the coordinator of the bridge cluster has the same site id, and delay sending the join request until C becomes the coordinator. Failing that, maybe JChannel.stop could interrupt the thread doing the join?
