shashank created CAMEL-27338:
--------------------------------

             Summary: camel-netty - a request that fails while it is set up on 
its pooled channel (before it is sent) never gives the channel back to the 
producer pool
                 Key: CAMEL-27338
                 URL: https://issues.apache.org/jira/browse/CAMEL-27338
             Project: Camel
          Issue Type: Bug
          Components: camel-netty
            Reporter: shashank
            Assignee: shashank


h3. What happens

{{NettyProducer.processWithConnectedChannel}} sets up the request on the 
channel it borrowed from the producer pool: the request timeout (from 
{{CamelNettyRequestTimeout}} or {{requestTimeout}}) and the state of the 
correlation manager. If that throws, {{ChannelConnectedListener}} completes the 
exchange with the exception through {{BodyReleaseCallback}}, which does not 
release the channel: only the {{NettyProducerCallback}} created later releases 
it. The channel stays borrowed for ever. Triggers: a 
{{CamelNettyRequestTimeout}} header that is not a number, a request without a 
correlation id for a {{TimeoutCorrelationManagerSupport}} manager, or a custom 
pipeline without a {{handler}}.

With {{producerPoolMaxTotal}} set, the pool runs out after that many failed 
requests: the next requests wait for a channel ({{producerPoolMaxWait=-1}}, the 
default: for ever) or fail with {{NoSuchElementException}}.

h3. How to reproduce

{{netty:tcp://localhost:port?textline=true&sync=true&producerPoolMaxTotal=1&producerPoolMaxWait=2000}}:
 send one request with {{CamelNettyRequestTimeout=not a number}}, then a valid 
request: it fails with {{NoSuchElementException: Timeout waiting for idle 
object}}.

Found with a TLA+ model of the netty producer.

h3. Fix

When setting up the request fails, the borrowed channel of the producer pool is 
invalidated (closed and removed from the pool): activating it armed the request 
timeout, which would otherwise fire on the idle channel, and the pipeline may 
have been changed. The shared channel of {{producerPoolEnabled=false}} is 
released as usual. With {{reuseChannel=true}} nothing changes; there the 
channel is returned by an on-completion that only overrides {{onComplete}}, so 
a failed exchange never returns it (a separate issue).

_Filed with Claude Code on behalf of allthingssecurity._



--
This message was sent by Atlassian Jira
(v10.3.26#10030026)

Reply via email to