Carsten Ziegeler created SLING-13130:
----------------------------------------
Summary: JobQueueImpl.startJobsGuard and semaphore permit leak
when threadPool.execute() throws
Key: SLING-13130
URL: https://issues.apache.org/jira/browse/SLING-13130
Project: Sling
Issue Type: Bug
Components: Event
Reporter: Carsten Ziegeler
In JobQueueImpl.startJobs(), if threadPool.execute() throws (e.g.
RejectedExecutionException because the pool is full or shut down), two
resources are permanently leaked:
1. The startJobsGuard AtomicBoolean remains true because the exception
propagates past startJobsGuard.set(false) at the end of the method.
2. An available semaphore permit is lost because started=true is set before the
execute() call, so the finally block skips available.release().
The result is a permanently dead queue -- no new jobs can start because
startJobsGuard blocks all entry, and maintain() (every 60s) is also blocked.
This persists until JVM restart.
Proposed fix:
1. Wrap startJobsGuard.set(false) in a try/finally around the entire while-loop.
2. Move started=true to after threadPool.execute() succeeds, so the finally
block correctly releases the available permit on exception.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
