RFR: 8351601: [JMH] test UnixSocketChannelReadWrite failed for 2 threads config

[Volkan Yazici]

On Fri, 21 Mar 2025 16:07:19 GMT, Michael McMahon <michaelm at openjdk.org> wrote:

> Seems like the only explanation for why that is happening is someone cleaning out /tmp while the test is running

@Michael-Mc-Mahon No, the reason for the failure is multiple threads creating a Unix socket in the very same `tempDir`, and then trying to delete the `tempDir`. See the following execution flow involving multiple threads:

1. `tempDir` is statically assigned during class initialization (say `/tmp/readWriteTest2414375588689416060`)
2. `Thread1` runs `beforeRun()`, which creates the Unix socket `/tmp/readWriteTest2414375588689416060/1`
3. `Thread2` runs `beforeRun()`, which creates the Unix socket `/tmp/readWriteTest2414375588689416060/2`
4. Both `Thread1` and `Thread2` runs `afterRun()` in parallel
5. `Thread1` runs `Files.delete(/tmp/readWriteTest2414375588689416060/1)` and succeeds
6. `Thread2` runs `Files.delete(/tmp/readWriteTest2414375588689416060/2)` and succeeds
7. `Thread1` runs `Files.delete(/tmp/readWriteTest2414375588689416060)` and succeeds (since both `/1` and `/2` socket files are removed above, and, hence, the directory is empty)
8. `Thread2` runs `Files.delete(/tmp/readWriteTest2414375588689416060)` and fails, since the directory has already been deleted above

Ideally, `tempDir` deletion should be performed by a single thread. But replacing `Files::delete` with `Files::deleteIfExists` also does the job, at a lower cost.

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

PR Comment: https://git.openjdk.org/jdk/pull/24126#issuecomment-2747320538