[ 
https://issues.apache.org/jira/browse/CAMEL-27330?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
 ]

Work on CAMEL-27330 started by shashank.
----------------------------------------
> camel-mllp - maxConcurrentConsumers no longer limits the connections: extra 
> connections are served instead of reset
> -------------------------------------------------------------------------------------------------------------------
>
>                 Key: CAMEL-27330
>                 URL: https://issues.apache.org/jira/browse/CAMEL-27330
>             Project: Camel
>          Issue Type: Bug
>          Components: camel-mllp
>            Reporter: shashank
>            Assignee: shashank
>            Priority: Major
>
> h3. What happens
> {{maxConcurrentConsumers}} is documented as: if a new connection is received 
> and the maximum number are already established, the new connection will be 
> reset immediately.
> Since 4.18 (CAMEL-22957, virtual threads support, and its follow-up commit 
> a47b4c7 that added the profile with {{maxQueueSize=0}}) the consumer executor 
> is created from a {{ThreadPoolProfile}} that sets no rejected policy, so it 
> gets the default profile's {{CallerRuns}}. When the pool is full, {{submit}} 
> does not throw: the consumer runnable runs on the validation thread (an 
> unbounded cached pool). The reset in {{MllpTcpServerConsumer.startConsumer}} 
> is never reached, and every connection is served. Before 4.18 the pool was a 
> {{ThreadPoolExecutor}} with the JDK default {{AbortPolicy}}.
> {{MllpMaxConcurrentConsumersTest.testConcurrentConsumersMoreThanMaxConsumers}}
>  does not catch it: its second client connects to another port, where nothing 
> listens.
> When a runnable is rejected (with an Abort policy), {{startConsumer}} also 
> leaves it in {{consumerRunnables}} (with its socket and the buffer holding 
> the first message), as only the runnable itself removes its entry.
> h3. How to reproduce
> {{mllp://localhost:port?maxConcurrentConsumers=1&autoAck=true}}: three 
> clients each send a message and each gets an AA acknowledgement; three 
> consumer runnables are active.
> Found with a TLA+ model of the MLLP consumer.
> h3. Fix
> Set the Abort policy on the consumer profile, remove the rejected runnable 
> from {{consumerRunnables}}, and connect the test's second client to the 
> consumer's port. Upgrade guide entry for 4.24 (extra connections are reset 
> again; raise {{maxConcurrentConsumers}} to serve more).
> _Filed with Claude Code on behalf of allthingssecurity._



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

Reply via email to