fuyou001 commented on PR #10828:
URL: https://github.com/apache/rocketmq/pull/10828#issuecomment-5216008222
Follow-up on the P1 lock-wait issue: Thread.yield() cannot provide the
progress guarantees this loop relies on.
The JDK contract defines Thread.yield() only as a scheduler hint, and the
scheduler is free to ignore it. It does not park the waiter, guarantee that the
current lock holder will run, provide fairness, or impose any upper bound on
acquisition time. Consequently, while (!tryLock(...)) { Thread.yield(); } can
keep a pullMessageExecutor worker RUNNABLE indefinitely; with multiple
same-attemptId retries, the scheduler may simply rotate among spinning waiters
while they continue consuming CPU. It also adds no deadline, cancellation,
interruption, disconnect, or shutdown handling.
Please do not use yield as the synchronization/progress mechanism here.
Prefer an explicit asynchronous unlock notification/handoff, or at minimum a
bounded timed acquisition tied to the request deadline and lifecycle
cancellation, followed by state revalidation after acquisition. A regression
test should hold the lock beyond the request deadline and assert that the
request completes without occupying the worker indefinitely and cannot mutate
POP/OrderInfo state after expiration; a multi-waiter test should also verify
that the shared executor is not exhausted.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
