[
https://issues.apache.org/jira/browse/JAMES-3803?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17578529#comment-17578529
]
Benoit Tellier commented on JAMES-3803:
---------------------------------------
>
No, not on master ;-)
{code:java}
private reactor.core.Disposable run(MailQueue queue) {
scheduler =
Schedulers.newBoundedElastic(configuration.getConcurrencyLevel() + 1,
DEFAULT_BOUNDED_ELASTIC_QUEUESIZE,
"spooler");
// ...
return Flux.from(queue.deQueue())
.flatMap(item ->
handleOnQueueItem(item).subscribeOn(scheduler),
configuration.getConcurrencyLevel())
.onErrorContinue((throwable, item) -> LOGGER.error("Exception
processing mail while spooling {}", item, throwable))
.subscribeOn(scheduler)
.subscribe();
}
{code}
We got rid of `Schedulers.elastic()` calls (which was quite a journey!) so now
this class follows a pattern very similar to DeliveryRunnable except that the
sizing of the scheduler is done taking the expected concurrency level +1. It
looks like this very difference is why we don't run into this issue for
JamesMailSpooler but do for DeliveryRunnable.
This means there is an alternative fix in DeliveryRunnable:
{code:java}
public void start() {
remoteDeliveryScheduler =
Schedulers.newBoundedElastic(configuration.getDeliveryThreads() + 1,
Schedulers.DEFAULT_BOUNDED_ELASTIC_QUEUESIZE, "RemoteDelivery");
disposable = Flux.from(queue.deQueue())
.flatMap(queueItem ->
runStep(queueItem).subscribeOn(remoteDeliveryScheduler),
configuration.getDeliveryThreads())
.onErrorContinue(((throwable, nothing) -> LOGGER.error("Exception
caught in RemoteDelivery", throwable)))
.subscribeOn(remoteDeliveryScheduler)
.subscribe();
}
{code}
(I'm fine with your fix proposal too, btw)
> DeliveryRunnable blocks email in BoundedElasticScheduler-queue if running out
> of threads
> ----------------------------------------------------------------------------------------
>
> Key: JAMES-3803
> URL: https://issues.apache.org/jira/browse/JAMES-3803
> Project: James Server
> Issue Type: Bug
> Components: Remote Delivery
> Affects Versions: 3.7.0
> Reporter: Adrian Bucher
> Priority: Major
> Attachments: jstack-of-the-bug.txt
>
> Time Spent: 40m
> Remaining Estimate: 0h
>
> The DeliveryRunnable does use the same BoundedElasticScheduler (ThreadPool)
> for dequeuing the messages (long running thread), and processing (delivering)
> them, if lots of messages are dequeued and the BoundedElasticScheduler starts
> to queue them, the one queueing behind the dequeuing - long running thread
> are never processed nor delivered.
>
> Messages are still visible in the embedded activemq delivery queues, a
> restart of the james will process them, as they will be reacknoledged by the
> dequeuer.
>
>
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: server-dev-unsubscr...@james.apache.org
For additional commands, e-mail: server-dev-h...@james.apache.org
