Confluence 8.0 request timeout transition and async indexing cleanup boundary

Direct answer

Confluence 8.5.x does not document guaranteed interrupt propagation to plugin-provided Callable tasks submitted via SpringTaskExecutor when confluence.request.timeout.seconds expires. Treat the interrupt as not propagated until you verify it in your environment.

Confirmed behavior vs. likely explanation

Well-established:

• When the timeout fires, Tomcat aborts the HTTP worker thread and returns a 504 Gateway Timeout.
• confluence.request.timeout.seconds and confluence.indexing.thread.pool.size are independent settings; changing one does not automatically adjust the other.
• There is no documented automatic cancellation of background tasks when a request times out.

Likely explanation (unconfirmed for 8.5.x):

• In earlier 8.x lines, the interrupt flag set on the aborted request thread was typically swallowed, so plugin Callable tasks ran to completion unless they manually checked Thread.interrupted(). Whether 8.5.x changed this is not explicitly documented.
• The searchIndexing executor is expected to retain any threads that were active when the timeout occurred; those threads return to the pool only when the current task finishes. Under sustained 504 conditions, this can lead to pool exhaustion.

Version-specific compatibility notes

There are no published compatibility notes that couple confluence.request.timeout.seconds with confluence.indexing.thread.pool.size. Because the settings are independent, a low timeout combined with a small indexing pool increases the risk of stranded threads. Raising the timeout without monitoring the pool can mask exhaustion.

Verification and safe steps

1. Add explicit cancellation. Do not rely on the request timeout to cancel indexing. Poll Thread.interrupted() or use a cooperative cancellation token inside your plugin Callable.
2. Capture a thread dump immediately after reproducing a 504 and inspect whether searchIndexing threads show an interrupted status.
3. Enable DEBUG logging for com.atlassian.confluence.spring.SpringTaskExecutor and org.springframework.scheduling.concurrent.ThreadPoolTaskExecutor.
4. Monitor the pool via JMX (Confluence → Monitoring → Thread Pools) before and after a 504 to confirm active counts return to baseline.
5. Test in staging with simulated 504 scenarios before changing either property in production.

One diagnostic detail that changes the recommendation

Can you confirm whether your plugin-provided Callable already polls Thread.interrupted(), or whether you are observing searchIndexing threads persisting after a 504? If the task does not check for interruption, add cooperative cancellation first. If it already does and threads still persist, inspect the SpringTaskExecutor configuration and rejection policy for your 8.5.x instance.