On Fri, 9 Feb 2024 02:17:01 GMT, Phil Race <[email protected]> wrote: > > I ran the test in our CI and I didn't notice an improvement. > > Did you run jdk:tier4 like Sergey was doing ?
No, I ran just this single test. I don't know how Sergey is running the test. > Our CI runs some number of (headless) tests in parallel (I am not sure but > think it is by default 4) and this might change the whole dynamics of "which > test is run last", compared to other test frameworks. Yep, I know that headless tests are run in parallel, which makes perfect sense. > I don't know if whatever Sergey is using does the same, but they could be > quite different in the number of CPUs per VM. > > If you ran just the _one_ test and it was no faster (wall-clock time) that > suggests the (sub-)tests were not being run in parallel which would be > interesting .. and we might want to ask jtreg folks for input. Yes, I submitted a job with one test only. At the same time, I didn't look thoroughly into the logs. This .java file contains four different tests which can be run in parallel, it could save the overall time; in the worst case, the same four tests are run consecutively. No negative impact, so I support the change. ------------- PR Comment: https://git.openjdk.org/jdk/pull/17719#issuecomment-1935920902
