[
https://issues.apache.org/jira/browse/CAMEL-24932?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18118336#comment-18118336
]
Andrea Cosentino commented on CAMEL-24932:
------------------------------------------
PR opened: https://github.com/apache/camel/pull/26778
Verified by reverting the fix: the new test then fails because the NPE kills
the polling thread before the handler is ever reached.
----
_Claude Code on behalf of oscerd (Andrea Cosentino)._
> camel-pulsar - the polling loop dies on a NullPointerException when a receive
> fails
> -----------------------------------------------------------------------------------
>
> Key: CAMEL-24932
> URL: https://issues.apache.org/jira/browse/CAMEL-24932
> Project: Camel
> Issue Type: Bug
> Components: camel-pulsar
> Reporter: Andrea Cosentino
> Assignee: Andrea Cosentino
> Priority: Major
>
> h3. Summary
> With {{messageListener=false}}, the polling loop reports its own errors
> through
> {{endpoint.getExceptionHandler()}}, which is {{null}} unless the route
> configures one. The resulting
> {{NullPointerException}} escapes {{run()}}, the consumer thread dies, and
> that consumer silently stops
> receiving messages.
> h3. Details
> {{PulsarConsumer.PulsarConsumerLoop.run()}}:
> {code:java}
> while (running && isRunAllowed()) {
> try {
> Message<byte[]> msg = consumer.receive();
> listener.received(consumer, msg);
> } catch (PulsarClientException e) {
> if (e.getCause() instanceof InterruptedException) {
> ...
> running = false;
> } else {
> endpoint.getExceptionHandler().handleException(e);
> }
> } catch (Exception e) {
> endpoint.getExceptionHandler().handleException(e);
> }
> }
> {code}
> {{getExceptionHandler()}} here resolves to {{DefaultEndpoint}}'s
> {{@UriParam}} field, which stays
> {{null}} unless the route sets {{?exceptionHandler=#bean}}.
> {{bridgeErrorHandler=true}} does not help
> either: {{DefaultEndpoint.configureConsumer}} installs both the bridging
> handler and a custom handler on
> the *consumer*, never on the endpoint.
> So any {{PulsarClientException}} from {{consumer.receive()}} whose cause is
> not an interrupt - a broker
> reconnect, a topic unload, a timeout - and any other {{Exception}}, throws an
> NPE from inside the catch
> block. The NPE propagates out of {{run()}} into the {{Future}} returned by
> {{executor.submit(...)}},
> which is discarded, so nothing is logged at all. The thread is gone and the
> loop is not restarted, so
> that consumer stops consuming for the lifetime of the route.
> h3. Scope
> Only reachable with {{messageListener=false}}; the default is {{true}}, which
> uses
> {{PulsarMessageListener}} instead. That one already does the right thing and
> calls
> {{pulsarConsumer.getExceptionHandler()}}.
> h3. Proposed fix
> Use the consumer's exception handler, which {{DefaultConsumer}} always
> initialises, exactly as
> {{PulsarMessageListener}} does. Worth also considering whether the loop
> should keep running after an
> unknown error, but the immediate defect is the NPE.
> ----
> _Reported by Claude Code on behalf of oscerd (Andrea Cosentino)._
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
