Integrated: 8306774: Make runtime/Monitor/GuaranteedAsyncDeflationIntervalTest.java more reliable
Aleksey Shipilev
shade at openjdk.org
Wed Apr 26 11:25:57 UTC 2023
On Tue, 25 Apr 2023 09:01:13 GMT, Aleksey Shipilev <shade at openjdk.org> wrote:
> As seen in some CI runs, doing `Object.wait(1)` to inflate 10K monitors can take quite a while. One of the test cases implicitly relies on threshold heuristics kicking in before the guaranteed interval of 5 seconds expires. Even if we wait for 1ms for each `wait` call, we only inflate 5K monitors in those 5 seconds. Which is okay for 5s and 10% threshold test, which would fire after about 1K monitors inflated. But we can also stall for longer -- because `wait(1)` is not guaranteed to return in 1ms, subject to locking granularity -- which means we can inflate less that 1K monitors, which would mean the threshold heuristics would not fire.
>
> This fix strengthens the test in two ways:
> 1. Uses multiple threads to inflate the monitors, so that we are able to eat the cost of `wait(1)`.
> 2. Drops the threshold heuristics threshold to minimum value to increase the chances of it firing.
>
> This also makes the test faster.
>
> @dcubed-ojdk, you might want to run this through your CI to see if it fixes the test troubles.
>
> Additional testing:
> - [x] Ad-hoc runs of the test, 100 iterations without trouble.
This pull request has now been integrated.
Changeset: 9ad6dc88
Author: Aleksey Shipilev <shade at openjdk.org>
URL: https://git.openjdk.org/jdk/commit/9ad6dc881d285cc26c136f0ef19af5bac0a75022
Stats: 34 lines in 2 files changed: 19 ins; 4 del; 11 mod
8306774: Make runtime/Monitor/GuaranteedAsyncDeflationIntervalTest.java more reliable
Reviewed-by: stuefe, dcubed
-------------
PR: https://git.openjdk.org/jdk/pull/13634
More information about the hotspot-runtime-dev
mailing list