RFR: 8348402: PerfDataManager stalls shutdown for 1ms

[Aleksey Shipilev]

On Fri, 24 Jan 2025 09:27:07 GMT, Aleksey Shipilev <shade at openjdk.org> wrote:

> Found this when studying Leyden performance. [JDK-8049304](https://bugs.openjdk.org/browse/JDK-8049304) added 1ms sleep on destruction path to catch up with threads updating the counters. 
> 
> I was not able to confidently prove the deletion race is benign. Even though `sleep` is not really a fix for that race either, I think it is safer to use GlobalCounter to coordinate deletions.
> 
> This improves startup (roundtrip) tests for the expected 1ms:
> 
> 
> $ hyperfine -w 10 -r 100 ...
> 
> # Before
> Benchmark 1: build/linux-x86_64-server-release/images/jdk/bin/java -XX:+UnlockExperimentalVMOptions -XX:+UseEpsilonGC -Xmx128m Hello
>   Time (mean ± σ):      20.0 ms ±   0.3 ms    [User: 11.7 ms, System: 16.0 ms]
>   Range (min … max):    19.0 ms …  21.7 ms    1000 runs
> 
> # After
> Benchmark 1: build/linux-x86_64-server-release/images/jdk/bin/java -XX:+UnlockExperimentalVMOptions -XX:+UseEpsilonGC -Xmx128m Hello
>   Time (mean ± σ):      19.0 ms ±   0.3 ms    [User: 11.9 ms, System: 15.8 ms]
>   Range (min … max):    18.2 ms …  20.8 ms    1000 runs
> 
> 
> Additional testing:
>  - [x] Linux AArch64 server fastdebug, `all`
>  - [x] Linux x86_64 server fastdebug, `all`

I filed [JDK-8348829](https://bugs.openjdk.org/browse/JDK-8348829) for these counters removals. But I honestly do not see a squeaky clean way out of this mess. There might be other counters that need this kind of critical-section treatment, maybe? So I suppose we remove the counters with [JDK-8348829](https://bugs.openjdk.org/browse/JDK-8348829), and then we still do this PR, replacing the sleep with the GC::CS infrastructure, in case other counters need it.

-------------

PR Comment: https://git.openjdk.org/jdk/pull/23293#issuecomment-2618567685