shashank created CAMEL-27321:
--------------------------------

             Summary: camel-jms, camel-sjms - an InOut request registered or 
timed out around the reply manager stop is never completed (follow-up of 
CAMEL-24124)
                 Key: CAMEL-27321
                 URL: https://issues.apache.org/jira/browse/CAMEL-27321
             Project: Camel
          Issue Type: Bug
          Components: camel-jms, camel-sjms
            Reporter: shashank
            Assignee: shashank


h3. What happens

CAMEL-24124 completes the exchanges that wait for a reply when the reply 
manager stops ({{CorrelationTimeoutMap.doStop}} drains the map). A few 
exchanges are still never completed, so a synchronous caller waits for ever and 
the exchange stays in flight:

* a request registered between the drain and the {{map.clear()}} of 
{{DefaultTimeoutMap.doStop}} (cleared without an event), or after the stop (it 
stays in the stopped map, whose purge task is cancelled). {{JmsProducer}} 
checks {{isRunAllowed()}} once and registers the reply later, from the 
{{MessageCreator}};
* with {{useMessageIDAsCorrelationID}}, {{updateCorrelationId}} removes the 
handler and puts it back under the JMSMessageID, and the drain can run in 
between;
* at stop the drained handlers are handed to the onTimeout executor; when their 
continuations take longer than {{shutdownAwaitTermination}}, the forced 
shutdown drops the queued tasks;
* with {{camel.threadpool.rejectedPolicy=Abort}} and a full queue, a timeout 
task is rejected after its entry was removed from the map, and 
{{DefaultTimeoutMap.emitEvent}} swallows the exception without a log.

h3. How to reproduce

A reply manager set up as {{JmsProducer}} does (no broker needed), with the 
thread orders forced through the map's listener events: register during or 
after the stop, move a handler after the drain, stop with a busy onTimeout 
thread and a short {{shutdownAwaitTermination}}, or time out three requests on 
a pool of one thread with a queue of one and the Abort policy. In each case an 
exchange is never completed.

Found with a TLA+ model of the JMS request-reply correlation.

h3. Fix

Close the correlation map under its lock before the drain: refuse a new request 
({{RejectedExecutionException}}, the producer completes the exchange and does 
not send it) and complete a moved handler at once. Run each timeout task once: 
in the calling thread when the executor rejects it, and from the reply manager 
after the executor shutdown when the executor never ran it. Log an exception of 
a timeout map listener at WARN. Same in camel-jms and camel-sjms.

_Filed with Claude Code on behalf of allthingssecurity._



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

Reply via email to