Aryan Ganjoo created POOL-432:
---------------------------------
Summary: ResilientPooledObjectFactory monitor dies when recovery
has no adder
Key: POOL-432
URL: https://issues.apache.org/jira/browse/POOL-432
Project: Commons Pool
Issue Type: Bug
Reporter: Aryan Ganjoo
When ResilientPooledObjectFactory observes a factory failure while the pool has
no waiting borrowers, it marks the factory down but does not start an adder
thread. After the delegate recovers and successful additions move the failure
out of the recent-failure log, runChecks() calls adder.kill() even though adder
is null. The resulting NullPointerException terminates the monitor thread, so a
later outage is no longer detected.
Reproduction on current Commons Pool master
(0d91a3efb5f1530e5fcba1a15b9577b618950547):
1. Start a ResilientPooledObjectFactory monitor with a delegate that can fail
and recover, and no waiting borrowers.
2. Make the delegate fail; pool.addObject() throws and the factory is marked
down. No adder starts.
3. Recover the delegate and add enough objects to move the earlier failure out
of the log.
4. Observe the monitor-thread NullPointerException during recovery. Make the
delegate fail again and add another object. The second outage is not detected
because the monitor stopped.
Expected: recovery with no adder should not terminate the monitor; subsequent
outages should still be detected.
A focused regression was reproduced failing against the unmodified base and
passing with a local guard around adder.kill(). The patched test also passed 20
consecutive local runs. A prior full local Maven verify passed; hosted CI has
not run and no PR is open for this issue yet.
Prepared with AI assistance.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
