On Thu, 9 Oct 2025 10:00:17 GMT, Stefan Karlsson <[email protected]> wrote:

> [JDK-8260555](https://bugs.openjdk.org/browse/JDK-8260555) changed the 
> default TIMEOUT_FACTOR from 1 to 4 to make it easier to understand how our 
> timeouts worked. Together with that small change, we also bumped a number of 
> tests that relied on the previous extended timeout. The anticipation was that 
> there would be some fallout from that change and that we for a short time 
> after it had been integrated would have to tweak some tests that 
> intermittently needed a larger timeout.
> 
> It turns out that the way we run tests concurrently causes the tests to 
> sometimes run in a resource-constrained manner. This has the effect that even 
> simple tests that usually don't take a long time to execute run the risk of 
> hitting the default 120s timeout limit. This invalidates the earlier plan to 
> fix the additional, few tests that now times out, because it's probably not 
> the tests themselves that are the problem, but rather how we run the tests.
> 
> Hunting down the reasons for the new set of timeouts we are seeing is good 
> and figuring out how to fix our testing to not over-strain our testing 
> machines is also something that we want to do. The problem is that there's 
> enough of these timeouts that it affects more than just a limited set of JDK 
> devs.
> 
> The proposal is that we, for now, revert back to the default timeout factor 
> of 4 to relive the pressure to investigate and fix these intermittent 
> timeouts. And revert back to 1 once enough investigation and tweaks have been 
> made to the test infrastructure.
> 
> I will run this through Oracle's tier1-tier8 testing.

Marked as reviewed by prr (Reviewer).

I don't have much visibility into how this affected other areas but I was 
surprised that a test that for me runs in a couple of seconds was apparently 
timing out on some systems  (not the Oracle CI but some external one).
In the PR to resolve this I suggested 
(https://github.com/openjdk/jdk/pull/27397#discussion_r2383687871) 
that we may be using too high a concurrency as chosen by the build as tests 
could be more sensitive to this.
But I also see that if you have a system with N CPUs it is a waste to only use 
(eg) N/2 because "some" test might time out.
A compromise may be needed.

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

PR Review: https://git.openjdk.org/jdk/pull/27721#pullrequestreview-3321344090
PR Comment: https://git.openjdk.org/jdk/pull/27721#issuecomment-3388293960

Reply via email to