Fix parameters usage for Online Bin tests: nb ts and nb threads
The number of threads was not fixed and our config script which force the OMP default to 1 is not loaded for the test for the online binary (typical case for htis binary which does not need the ROOT env setup).
=> This lead to the test using all available CPUs on computers with modern OS due to the OMP library implementation, stalling our ctest environment (and probably the host machine)
The bonus joke: the number of TS was defined on the cmake side but not used in the test wrapper script, meaning that the test was always running with a single TS for all test models
=>A good hint that the timeout observed in the Nightly tests on Virgo was also linked to the above bug with the number of threads!
Merge request reports
Activity
requested review from @f.uhlig
assigned to @f.uhlig
- Resolved by Pierre-Alain Loizeau
One can also add
RUN_SERIAl
to the set_test_properties of the test. According to the documentation in this case the test should not run in parallel to any other test.
I will try this after when run4 has finished.
- Last reply by Pierre-Alain Loizeau
- Resolved by Pierre-Alain Loizeau
If you are correct that we always run on one time slice I wonder how slow the execution of a single times lice is. We did not finish the execution in the Nightly tests with 56 processes in parallel within 27 minutes.
- Last reply by Pierre-Alain Loizeau
added 1 commit
- 6eab5051 - Force the online binary test to run alone + use all test CPUs
added 1 commit
- d9c2cef0 - Force the online binary test to run alone + use all test CPUs
Looks okay now. I saw in htop that the main job only 400-500% CPU which should be something like 4-5 cores but beside that I saw many processes in htop. Currently I don't understand the logic here, probably something to check more carefully if everything works. I am not sure if the test really runs with no other test running in parallel.
enabled an automatic merge when all merge checks for d9c2cef0 pass
added CodeOwners label
