[ 
https://issues.apache.org/jira/browse/FLINK-22698?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17349300#comment-17349300
 ] 

Austin Cawley-Edwards commented on FLINK-22698:
-----------------------------------------------

Hi [~cmick] – that cause makes perfect sense to me. And the proposed solution 
also sounds good, just a couple of questions:
 * What happens when a delivery times out?
 * How do other sources handle this, i.e. Kafka? Do they wait for messages on 
another thread than where the source's `run` is called?

 

Anyway, I'll assign it to you + thanks for looking into this! Also, I'm going 
to cc' [~fabian.paul], as he's been working with this connector (+ others) and 
its FLIP-27 upgrades. He might have some other ideas.

> RabbitMQ source does not stop unless message arrives in queue
> -------------------------------------------------------------
>
>                 Key: FLINK-22698
>                 URL: https://issues.apache.org/jira/browse/FLINK-22698
>             Project: Flink
>          Issue Type: Bug
>          Components: Connectors/ RabbitMQ
>    Affects Versions: 1.12.0
>            Reporter: Austin Cawley-Edwards
>            Priority: Major
>         Attachments: taskmanager_thread_dump.json
>
>
> In a streaming job with multiple RMQSources, a stop-with-savepoint request 
> has unexpected behavior. Regular checkpoints and savepoints complete 
> successfully, it is only the stop-with-savepoint request where this behavior 
> is seen.
>  
> *Expected Behavior:*
> The stop-with-savepoint request stops the job with a FINISHED state.
>  
> *Actual Behavior:*
> The stop-with-savepoint request either times out or hangs indefinitely unless 
> a message arrives in all the queues that the job consumes from after the 
> stop-with-savepoint request is made.
>  
> *Current workaround:*
> Send a sentinel value to each of the queues consumed by the job that the 
> deserialization schema checks in its isEndOfStream method. This is cumbersome 
> and makes it difficult to do stateful upgrades, as coordination with another 
> system is now necessary. 
>  
>  
> The TaskManager thread dump is attached.
>  



--
This message was sent by Atlassian Jira
(v8.3.4#803005)

Reply via email to