[
https://issues.apache.org/jira/browse/CAMEL-25260?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18126953#comment-18126953
]
Guillaume Nodet commented on CAMEL-25260:
-----------------------------------------
The fix has been merged via PR #27579.
*Root cause and fix:*
JUnit Platform 6.x added a warning when no {{executor-service}} is explicitly
configured for parallel test execution. The warning appeared thousands of times
in CI logs. The simplest fix is to set {{executor-service=FORK_JOIN_POOL}}
explicitly in the three {{junit-platform.properties}} files (camel-core,
camel-core-engine, camel-integration-test) — this silences the warning and
keeps existing behavior unchanged, since {{ForkJoinPool}} was already the
default.
We initially evaluated {{WORKER_THREAD_POOL}} (the other option introduced in
JUnit 6.1, {{@API(status = EXPERIMENTAL)}}), but rejected it because it uses
non-daemon threads by construction. This caused
{{DefaultExecutorServiceManagerTest}} to fail deterministically: our
{{ThreadFactoryListener}} test installs a bare factory that never calls
{{setDaemon(true)}}, and {{BaseExecutorServiceManager.newThread}} calls
{{createThreadFactory(..., daemon=true)}} — which the listener fully replaces,
so the created threads are non-daemon. The {{ForkJoinPool}} path already
creates daemon threads, so those tests pass unchanged.
*JUnit team feedback:*
The daemon-thread behavior of {{WORKER_THREAD_POOL}} may be worth reporting
upstream. If the worker-thread-pool executor is intended as a general
replacement for {{ForkJoinPool}}, it should either default to daemon threads or
expose a daemon flag. We did not file a JUnit issue; if someone wants to pursue
this, the relevant class is {{WorkerThreadPoolHierarchicalTestExecutorService}}.
*Follow-up:*
If a future JUnit version makes {{WORKER_THREAD_POOL}} stable and
daemon-thread-safe, switching to it would be a straightforward one-line change
in each of the three properties files. For now {{FORK_JOIN_POOL}} is the
correct choice — it keeps {{CamelParallelExecutionStrategy}} fully effective
(including the saturation predicate that logs pool-exhaustion warnings).
> Try
> junit.jupiter.execution.parallel.config.executor-service=WORKER_THREAD_POOL
> -------------------------------------------------------------------------------
>
> Key: CAMEL-25260
> URL: https://issues.apache.org/jira/browse/CAMEL-25260
> Project: Camel
> Issue Type: Test
> Components: build system
> Affects Versions: 4.23.0
> Reporter: Aurélien Pupier
> Assignee: Guillaume Nodet
> Priority: Major
>
> in logs, there are many times:
> {noformat}
> [INFO] Parallel test execution is enabled but the default ForkJoinPool-based
> executor service will be used. Please give the new implementation based on a
> regular thread pool a try by setting the
> 'junit.jupiter.execution.parallel.config.executor-service' configuration
> parameter to 'WORKER_THREAD_POOL' and report any issues to the JUnit team.
> Alternatively, set the configuration parameter to 'FORK_JOIN_POOL' to hide
> this message and keep using the original implementation.
> {noformat}
> could be worthy to give a try to see if there are improvements (maybe also
> with the other work in progress with -T1C)
> either it is better and use it, or not better and set
> junit.jupiter.execution.parallel.config.executor-service=FORK_JOIN_POOL to
> avoid the thousands of messages in the log
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
