Erich Duda created ARTEMIS-1484:
-----------------------------------
Summary: Live's topology update may be ignored
Key: ARTEMIS-1484
URL: https://issues.apache.org/jira/browse/ARTEMIS-1484
Project: ActiveMQ Artemis
Issue Type: Bug
Components: Broker
Affects Versions: 2.3.0
Reporter: Erich Duda
In tests based on {{MultiServerTestBase}} it sometimes happens that after all
servers are started, the check waitForTopology fails with the following error.
{code}
Timed out waiting for cluster topology of live=5,backup=5 (received live=4,
backup=5) topology = topology on
Topology@5884a914[owner=ClusterConnectionImpl@405215542[nodeUUID=bbbae377-ba40-11e7-aff3-fa163e312a80,
connector=TransportConfiguration(name=bbbabc66-ba40-11e7-aff3-fa163e312a80,
factory=org-apache-activemq-artemis-core-remoting-impl-invm-InVMConnectorFactory)
?serverId=0, address=cluster-queues,
server=ActiveMQServerImpl::BridgeFailoverTest/Live(0)]]:
bbd79349-ba40-11e7-aff3-fa163e312a80 => TopologyMember[id =
bbd79349-ba40-11e7-aff3-fa163e312a80,
connector=Pair[a=TransportConfiguration(name=bbd79348-ba40-11e7-aff3-fa163e312a80,
factory=org-apache-activemq-artemis-core-remoting-impl-invm-InVMConnectorFactory)
?serverId=1,
b=TransportConfiguration(name=bbd79353-ba40-11e7-aff3-fa163e312a80,
factory=org-apache-activemq-artemis-core-remoting-impl-invm-InVMConnectorFactory)
?serverId=6], backupGroupName=null, scaleDownGroupName=null]
bbd7935b-ba40-11e7-aff3-fa163e312a80 => TopologyMember[id =
bbd7935b-ba40-11e7-aff3-fa163e312a80,
connector=Pair[a=TransportConfiguration(name=bbd7935a-ba40-11e7-aff3-fa163e312a80,
factory=org-apache-activemq-artemis-core-remoting-impl-invm-InVMConnectorFactory)
?serverId=2,
b=TransportConfiguration(name=bbd7ba75-ba40-11e7-aff3-fa163e312a80,
factory=org-apache-activemq-artemis-core-remoting-impl-invm-InVMConnectorFactory)
?serverId=7], backupGroupName=null, scaleDownGroupName=null]
bbbae377-ba40-11e7-aff3-fa163e312a80 => TopologyMember[id =
bbbae377-ba40-11e7-aff3-fa163e312a80,
connector=Pair[a=TransportConfiguration(name=bbbabc66-ba40-11e7-aff3-fa163e312a80,
factory=org-apache-activemq-artemis-core-remoting-impl-invm-InVMConnectorFactory)
?serverId=0,
b=TransportConfiguration(name=bbd76c31-ba40-11e7-aff3-fa163e312a80,
factory=org-apache-activemq-artemis-core-remoting-impl-invm-InVMConnectorFactory)
?serverId=5], backupGroupName=null, scaleDownGroupName=null]
bbd7ba8f-ba40-11e7-aff3-fa163e312a80 => TopologyMember[id =
bbd7ba8f-ba40-11e7-aff3-fa163e312a80, connector=Pair[a=null,
b=TransportConfiguration(name=bbd7ba99-ba40-11e7-aff3-fa163e312a80,
factory=org-apache-activemq-artemis-core-remoting-impl-invm-InVMConnectorFactory)
?serverId=9], backupGroupName=null, scaleDownGroupName=null]
bbd7ba7d-ba40-11e7-aff3-fa163e312a80 => TopologyMember[id =
bbd7ba7d-ba40-11e7-aff3-fa163e312a80,
connector=Pair[a=TransportConfiguration(name=bbd7ba7c-ba40-11e7-aff3-fa163e312a80,
factory=org-apache-activemq-artemis-core-remoting-impl-invm-InVMConnectorFactory)
?serverId=3,
b=TransportConfiguration(name=bbd7ba87-ba40-11e7-aff3-fa163e312a80,
factory=org-apache-activemq-artemis-core-remoting-impl-invm-InVMConnectorFactory)
?serverId=8], backupGroupName=null, scaleDownGroupName=null]
nodes=9 members=5)
{code}
I dug into this and found out that in some certain cases Live's topology update
message has older event ID than Backup's update message and it is also received
later. In these cases the Live's message is ignored, because it doesn't meet
the condition as it is shown below in the code snippet.
I think if the current node has no connector to Live, it shouldn't ignore
topology update from Live even if it is older than the current record.
{code:java}
public boolean updateMember(final long uniqueEventID, final String nodeId,
final TopologyMemberImpl memberInput) {
if (uniqueEventID > currentMember.getUniqueEventID()) {
...
}
/*
* always add the backup, better to try to reconnect to something that's not
there then to
* not know about it at all
*/
if (currentMember.getBackup() == null && memberInput.getBackup() != null) {
currentMember.setBackup(memberInput.getBackup());
}
}
{code}
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
