[
https://issues.apache.org/jira/browse/CAMEL-26800?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Work on CAMEL-26800 started by shashank.
----------------------------------------
> camel-master - MasterConsumer gives up the leadership before its delegated
> consumer has stopped
> -----------------------------------------------------------------------------------------------
>
> Key: CAMEL-26800
> URL: https://issues.apache.org/jira/browse/CAMEL-26800
> Project: Camel
> Issue Type: Bug
> Components: camel-master
> Reporter: shashank
> Assignee: shashank
> Priority: Major
>
> *Leadership given up before the consumer stopped.* {{MasterConsumer.doStop}}
> removes its listener and releases the cluster view (which gives up the
> leadership) before it stops the delegated consumer. On every graceful stop of
> a leader (route stop, CamelContext shutdown, rolling restart), another node
> can take the leadership and start consuming while this node is still stopping
> its consumer (for example a JMS consumer that completes the messages it is
> processing). The {{master}} component is meant to have one active consumer in
> the cluster.
> *The route is stopped before the delegated consumer.* {{doStop}} first stops
> the processor of the route ({{super.doStop()}}), so a delegated consumer that
> still routes messages while it stops (such as JMS, which completes the
> messages it is processing) sends them into a stopped route, where they fail.
> *A failure to release the view leaves the consumer running.* There is no
> try/finally: when {{releaseView}} throws (for example the file lock cluster
> service when its data storage stalls), the delegated consumer is not stopped
> at all.
> Expected: the leadership is released, and the route processor stopped, only
> after the delegated consumer stopped, and the delegated consumer is stopped
> also when releasing the view fails.
> Suggested fix: in {{doStop}} stop the delegated consumer first, then the
> route processor, and release the view in a finally block.
> Found with Specula (TLA+ model checked against traces of the real code) and
> code review, and confirmed with tests on the real classes.
> _Filed with Claude Code on behalf of allthingssecurity._
--
This message was sent by Atlassian Jira
(v10.3.26#10030026)
