[
https://issues.apache.org/jira/browse/AMQ-6086?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15056268#comment-15056268
]
Gary Tully commented on AMQ-6086:
---------------------------------
having some sort of stack of started components is needed here. Pushing (start)
aborting if popping has started. A blocking component, like a persistence
adapter that will block on a lock acquire - would need to push early and
respond to a stop during eventual blocking.
This implies some symmetry between start/stop which makes sense and some
consistency in applying the service pattern. Something to keep in mind for the
future.
> Broker stop and start are not at all thread safe - we can do better
> -------------------------------------------------------------------
>
> Key: AMQ-6086
> URL: https://issues.apache.org/jira/browse/AMQ-6086
> Project: ActiveMQ
> Issue Type: Bug
> Components: Broker
> Affects Versions: 5.13.0
> Reporter: Gary Tully
> Assignee: Gary Tully
> Fix For: 5.14.0
>
>
> If one thread starts a broker and a second tries to stop we can get in a
> mess. Particularly if there is blocking or locking in the mix. In some cases
> we are ok but there is no determinism.
> In a simple case, the stop can complete before the start gets going at all
> and we miss the stop, leaving a dangling broker.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
