[
https://issues.apache.org/jira/browse/CAMEL-27335?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Work on CAMEL-27335 started by shashank.
----------------------------------------
> camel-netty - stopping a producer that uses a
> TimeoutCorrelationManagerSupport correlation manager leaves its pending
> exchanges uncompleted
> -------------------------------------------------------------------------------------------------------------------------------------------
>
> Key: CAMEL-27335
> URL: https://issues.apache.org/jira/browse/CAMEL-27335
> Project: Camel
> Issue Type: Bug
> Components: camel-netty
> Reporter: shashank
> Assignee: shashank
> Priority: Major
>
> h3. What happens
> When a netty producer stops, it closes its channels; {{channelInactive}} asks
> the correlation manager for the state with {{getState(ctx, channel,
> (Throwable) null)}}, which {{TimeoutCorrelationManagerSupport}} answers with
> {{null}}. The manager is stopped next: {{DefaultTimeoutMap.doStop}} clears
> the map without evicting the entries. The requests still waiting for a reply
> never get a callback (a synchronous caller blocks, the exchange stays in
> flight).
> The reply managers of camel-jms, camel-sjms and camel-rocketmq drain their
> timeout map ({{drainAndEvictAll}}) on stop for this case.
> Also, a request that the purge task evicts while the producer stops can be
> submitted to the worker pool after the pool was shut down: its CallerRuns
> policy drops the task, and the exchange is never completed.
> h3. How to reproduce
> A {{TimeoutCorrelationManagerSupport}} with a timeout of 60 s, a server that
> reads the request and never replies, stop the producer: the callback is not
> called 10 s after the stop. With the default correlation manager the exchange
> completes with "No response received from remote server".
> Found with a TLA+ model of the netty producer.
> h3. Fix
> The manager's timeout map evicts every pending entry when it stops (they
> complete as timed out on the worker pool, before it is shut down), and an
> eviction that reaches a shut-down (or missing) worker pool is completed in
> the calling thread. The completion stays at most once (markDone).
> _Filed with Claude Code on behalf of allthingssecurity._
--
This message was sent by Atlassian Jira
(v10.3.26#10030026)