[
https://issues.apache.org/jira/browse/CAMEL-14327?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17002757#comment-17002757
]
Rafał Gała commented on CAMEL-14327:
------------------------------------
I think I got it.
Evictions are done in a separate thread. When the following method is executed:
{code:java}
protected void onEvict(S s) {
Endpoint e = getEndpoint.apply(s);
Pool<S> p = pool.get(e);
if (p != null) {
if (p.evict(s)) {
pool.remove(e);
}
} else {
ServicePool.stop(s);
try {
e.getCamelContext().removeService(s);
} catch (Exception ex) {
LOG.error("Error removing service {}", s, ex);
}
}
}
{code}
the call to *p.evict(s)* results in *doStop()* method of *Jt400PgmProducer*
instance being called. After the p.evict(s) method has finished, but before
pool.remove(e) call, the producer can still be acquired from the pool, but
because the *doStop* has already been called and the *iSeries* field is already
nulled, exchange processing fails on NPE.
I set breakpoints in *onEvict(S)* method of *ServicePool* and in
*acquire(Endpoint endpoint)* method, they evict/acquire the same
*Jt400PgmProducer* instance at the same time.
{noformat}
Daemon Thread [ForkJoinPool.commonPool-worker-6] (Suspended (breakpoint at line
85 in ServicePool))
ServicePool<S>.onEvict(S) line: 85
146690937.accept(Object) line: not available
CaffeineLRUCache<K,V>.onRemoval(K, V, RemovalCause) line: 239
SSLMS<K,V>(BoundedLocalCache<K,V>).lambda$notifyRemoval$1(Object,
Object, RemovalCause) line: 333
1106690494.run() line: not available
ForkJoinTask$RunnableExecuteAction.exec() line: 1402
ForkJoinTask$RunnableExecuteAction(ForkJoinTask<V>).doExec() line: 289
ForkJoinPool$WorkQueue.runTask(ForkJoinTask<?>) line: 1056
ForkJoinPool.runWorker(ForkJoinPool$WorkQueue) line: 1692
ForkJoinWorkerThread.run() line: 157
{noformat}
{noformat}
Daemon Thread [Camel (camel-1) thread #17 - seda://seda] (Suspended (breakpoint
at line 113 in ServicePool))
ServicePool<S>.acquire(Endpoint) line: 113
DefaultProducerCache.acquireProducer(Endpoint) line: 112
DefaultProducerCache.doInAsyncProducer(Endpoint, Exchange,
AsyncCallback, ProducerCache$AsyncProducerCallback) line: 264
SendProcessor.process(Exchange, AsyncCallback) line: 162
RedeliveryErrorHandler$RedeliveryState.run() line: 476
DefaultReactiveExecutor$Worker.schedule(Runnable, boolean, boolean,
boolean) line: 185
DefaultReactiveExecutor.scheduleMain(Runnable, String) line: 59
Pipeline.process(Exchange, AsyncCallback) line: 87
CamelInternalProcessor.process(Exchange, AsyncCallback) line: 228
SedaConsumer.sendToConsumers(Exchange) line: 266
SedaConsumer.doRun() line: 180
SedaConsumer.run() line: 125
RejectableThreadPoolExecutor(ThreadPoolExecutor).runWorker(ThreadPoolExecutor$Worker)
line: 1149
ThreadPoolExecutor$Worker.run() line: 624
Thread.run() line: 748
{noformat}
Maybe the p.evict() should be called after pool.remove()?
> Jt400PgmProducer doStop() method called on actively used instance
> -----------------------------------------------------------------
>
> Key: CAMEL-14327
> URL: https://issues.apache.org/jira/browse/CAMEL-14327
> Project: Camel
> Issue Type: Bug
> Components: camel-jt400
> Affects Versions: 3.0.0
> Reporter: Rafał Gała
> Priority: Major
>
> Today I migrated to 3.0.0 version and there seems to be an issue with Service
> pooling for the Jt400PgmpProducer.
>
> Here's what I have:
>
> {code:java}
> from(
> "seda:someName?concurrentConsumers=2&size=10")
>
> .to("jt400://{{as400.user}}:{{as400.password}}@{{as400.host}}/QSYS.LIB/PROGRAM.LIB/KFKEVR.SRVPGM?fieldsLength=200,2000,4,8,8,1000&outputFieldsIdx=0,1,2,3,4,5&connectionPool=#as400ConnectionPool&format=binary&procedureName=RECEIVEEVENT");
> {code}
> When concurrentConsumers attribute of seda endpoint is set to 1 everything
> works fine, but when it is greater than 1 then it looks like the evict method
> from MultiplePool class calls stop method on a Jt400PgmProducer instance that
> is still being used (the process method on it is still getting called). This
> results in nulling the iSeries object inside Jt400PgmProducer instance:
> {code:java}
> @Override
> protected void doStop() throws Exception {
> if (iSeries != null) {
> LOG.info("Releasing connection to {}", getISeriesEndpoint());
> getISeriesEndpoint().releaseSystem(iSeries);
> iSeries = null;
> }
> }
> {code}
> and when the process method gets called later on this instance, it fails with
> NPE while constructing ServiceProgramCall:
> {code:java}
> @Override
> public void process(Exchange exchange) throws Exception {
> ...
> pgmCall = new ServiceProgramCall(iSeries);
> ...
> {code}
>
> I believe this may be related to producer caching in ServicePool class, some
> sort of key issue in the cache Map maybe?
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
