You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Many extended.openjdk tests timing out on aarch64 #2002
There are many tests timing out in the extended.openjdk target with messages like:
06:58:21 TEST RESULT: Error. Program `/home/jenkins/workspace/Test_openjdk16_hs_extended.openjdk_aarch64_linux/openjdkbinary/j2sdk-image/bin/java' timed out (timeout set to 960000ms, elapsed time including timeout handling was 1006476ms).
Current theory: The docker containers are restricted, however the tests are trying to spawn significantly more processes than there are CPUs available to the container. This is causing it to reach the memory limits of the container and slow things down. It would be good to know how the tests (jdk_time_1 being a good example) decide how much processes to spawn off. At the moment on my container restricted to 8 CPUs I have 29 javac process and 21 jstack ones
The above PR seems to resolve this problem based on the jdk_time-1 Grinders I've been running (Docker | Non-docker)
I've initiated a full extended.openjdk on a docker machine and a real machine to compare but once that PR is merged I'm confident this particular issue can be closed so I'll be able to re-enable all the machines again.
Activity
sxa commentedon Mar 4, 2021
Re-running at https://ci.adoptopenjdk.net/job/Grinder/7435 on the same system just now while the host system should otherwise be idle.
sxa commentedon Mar 8, 2021
@smlambert Are the extended openjdk suites similar to the system test suites in that they can chew up all the CPU cycles available to them?
sxa commentedon Mar 9, 2021
After running a few other Grinders I'm now running
jdk_time_1with 10 iterations on two affected machines (It's not 100% consistent):sxa commentedon Mar 9, 2021
Current theory: The docker containers are restricted, however the tests are trying to spawn significantly more processes than there are CPUs available to the container. This is causing it to reach the memory limits of the container and slow things down. It would be good to know how the tests (
jdk_time_1being a good example) decide how much processes to spawn off. At the moment on my container restricted to 8 CPUs I have 29javacprocess and 21jstackonessxa commentedon Mar 9, 2021
ping @smlambert Is this expected? How many cores would you expect these tests to use up generally?
smlambert commentedon Mar 9, 2021
A while back, I believe @jerboaa added this PR adoptium/aqa-tests#1427 which may be relevant to the question 'how many cores'?
sxa commentedon Mar 9, 2021
That would explain it :-) I'll see what I can do ...
Thanks for the pointer!! (Worthy of two
!I feel)sxa commentedon Mar 10, 2021
The above PR seems to resolve this problem based on the
jdk_time-1Grinders I've been running (Docker | Non-docker)I've initiated a full extended.openjdk on a docker machine and a real machine to compare but once that PR is merged I'm confident this particular issue can be closed so I'll be able to re-enable all the machines again.