[
https://issues.apache.org/jira/browse/CAMEL-26305?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Work on CAMEL-26305 started by shashank.
----------------------------------------
> camel-jgroups-raft - removing the route that consumes from an endpoint closes
> the channel and the Raft log that its producer still uses, so the node leaves
> the cluster
> -----------------------------------------------------------------------------------------------------------------------------------------------------------------------
>
> Key: CAMEL-26305
> URL: https://issues.apache.org/jira/browse/CAMEL-26305
> Project: Camel
> Issue Type: Bug
> Components: camel-jgroups
> Reporter: shashank
> Assignee: shashank
> Priority: Major
>
> The producers and consumers of a {{jgroups-raft:}} endpoint share the
> {{RaftHandle}} the endpoint created (from {{channelProperties}} or the
> default configuration, no {{raftHandle}} on the component). When a route is
> removed, the core stops the endpoint of its consumer
> ({{RouteService.doShutdown}}), also when other routes still use the endpoint,
> and {{JGroupsRaftEndpoint.doStop}} closes the handle's channel and its Raft
> log regardless of the producers and consumers still connected (main
> c536c99caaaa, lines 122 and 127). The node leaves the Raft cluster and the
> producer route's {{set}} calls fail until the removed route is added again.
> As each endpoint creates its own handle with the endpoint's {{raftId}}, a
> node that both changes the state machine and listens to role changes uses the
> same uri for its producer and its consumer, so they share the endpoint:
> removing the role-change route (a route reload, a removed route template or
> Kamelet, JMX or the control bus) takes the node out of the cluster.
> CAMEL-25883 fixed this for the {{raftHandle}} of the component and left the
> handle an endpoint creates itself unchanged.
> *Fix*: the endpoint closes its own handle (channel and log) at stop only when
> no producer or consumer is connected; otherwise the last one that disconnects
> closes it once the endpoint is stopped. A restarted endpoint keeps its handle
> while the channel is still open. A connect is now counted once it succeeded
> (as in camel-jgroups), so a producer whose connect failed does not keep the
> handle open. The handles of the component or a {{raftHandle}} bean are
> unchanged.
> *Tests*: new {{JGroupsRaftRemoveRouteTest}} (three nodes A, B, C in the JVM
> with {{raftABC.xml}}; node A has a producer and a role-change consumer on the
> same uri): {{removeConsumerRouteKeepsTheHandleOfTheProducer}} and
> {{removeBothRoutesClosesTheHandle}} fail on main (3 runs: {{the producer
> still uses the channel ==> expected: <true> but was: <false>}}), and
> {{removeConsumerRouteThenAddItAgainKeepsTheHandle}} (3 runs: {{the re-added
> consumer uses the handle of the producer}}); all pass with the fix. With the
> fix camel-jgroups-raft has 22 tests, 0 failures.
> *How it was found*: a code sweep for endpoints whose {{doStop}} releases
> something that other routes on the same uri still use, after a random
> lifecycle-sequence harness found the same kind of defect in camel-disruptor
> (CAMEL-26286).
> *Affected*: main, camel-4.22.x, camel-4.18.x and camel-4.14.x (same
> {{doStop}}; read, not run).
> _Filed with Claude Code on behalf of allthingssecurity._
--
This message was sent by Atlassian Jira
(v10.3.26#10030026)
