[
https://issues.apache.org/jira/browse/ARTEMIS-1353?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16130719#comment-16130719
]
ASF GitHub Bot commented on ARTEMIS-1353:
-----------------------------------------
Github user mtaylor commented on the issue:
https://github.com/apache/activemq-artemis/pull/1466
@clebertsuconic It appears the original issue was introduced when you
changed the ReplicationManager to skip the executor and send in the same
thread. I am trying to understand why this was required. From a quick read of
the code, I assume that when sending a large file there's potential for the
whole file to get scheduled on the ThreadPoolExecutor Queue, with potential to
cause an OOM? Is this assumption correct?
> Replication packets can get out of order
> ----------------------------------------
>
> Key: ARTEMIS-1353
> URL: https://issues.apache.org/jira/browse/ARTEMIS-1353
> Project: ActiveMQ Artemis
> Issue Type: Bug
> Affects Versions: 2.2.0
> Reporter: Justin Bertram
> Assignee: Justin Bertram
> Fix For: 2.3.0
>
>
> {{ReplicationEndpoint}} my throw a {{NullPointerException}} during initial
> synchronization.
> The issue is caused by incorrect ordering of replication packets. The NPE
> arises when {{ReplicationSyncFileMessage}} packets are sent before
> {{ReplicationStartSyncMessage}} packets.
> Incorrect ordering of replication packets may happen because of
> {{useExecutor}} parameter in the {{sendReplicatePacket}} method.
> {{ReplicationStartSyncMessage}} packets are sent as first, but they are sent
> with {{useExecutor=true}}. Although {{ReplicationSyncFileMessage}} packets
> are sent after {{ReplicationStartSyncMessage}} packets, they are sent with
> {{useExecutor=false}}. So sending of {{ReplicationStartSyncMessage}} packets
> is scheduled to executor and there is no guarantee when the task will be
> executed, whereas {{ReplicationStartSyncMessage}} packets are sent
> immediately.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
