Closed kevink-sq closed 1 year ago
@kevink-sq Can you please give a little more detail, it's not actionable without additional information. Do you know which junixsocket objects were causing these GC pause times?
What Java VM on what platform are you using? What are the VM arguments? (which garbage collector, etc.). Can you post the output of junixsocket-selftest?
@kevink-sq Please try the latest 2.7.1-SNAPSHOT (2.7.1-20230828.103553 or newer). Please note that I can't release 2.7.1 on a hunch, so verifying that snapshot version is imperative if you want a proper fix.
@kevink-sq Please verify with 2.7.1.
Please verify with 2.7.2, just released. Thanks again for reporting, @kevink-sq !
Hi @kohlschuetter,
We did extensive prod testing with junixsocket 2.7.2 and found no regression. Its been baking in prod for the last 3 weeks now and things look fine.
2.7.1 however did have an issue with NativeUnixSocket.poll
timing out but that seems to have been fixed in #139.
Overall thanks for the work on junixsocket. We initially went on this path because we found jnr-unixsocket to have load issues during high concurrency and our infra could not support JEP 380.
@kevink-sq Thanks for the feedback, @kevink-sq, really appreciated!
Describe the bug
This is a follow-up to https://github.com/kohlschutter/junixsocket/issues/135 as it was found rolling out junixsocket to all services in prod caused gc spikes leading to p99/max latency. Cannot share heap dumps but the symptoms were as follows:
We had to rollback usage of junixsocket across our services.