Currently the autobuilders take about 7 hours to execute a single run of oe-selftest, one way to improve the time is to parallelize the tests.
Mariano, now that you've been working on the OEQA2 effort, how do you see this coming for 2.3? can you estimate a Milestone for this feature?
I don't think this is doable in 2.3, selftest would require a lot of effort to migrate to the new framework.
Moving to 2.4 because this bug depends on pending work for the selftest script.
Implementation branch, http://git.yoctoproject.org/cgit/cgit.cgi/poky-contrib/log/?h=alimon/oe_selftest_threaded
Created attachment 3862 [details] oe-selftest batch run
Created attachment 3863 [details] oe-selftest 4 pythreads with 16 bb process/parallel make
Comment on attachment 3862 [details] oe-selftest batch run Wrong log
Created attachment 3864 [details] oe-selftest batch run
Review at, http://lists.openembedded.org/pipermail/openembedded-core/2017-July/139527.html
Anibal will keep working on it, but while he gets an external account, I am the temporal owner of this bug.
there is a general agreement to use different approach with the idea to make it simpler and more maintainable: Use a wrapper script for oe-selftest, which will detect test cases and call each one on a separate build directory. One we have this setup, the next point is to figure out how much can be parallelized based on test cases dependencies and system resources.
The solution proposed on bug 11429 allows oe-selftest parallelization as demonstrated in [1]. However, the proposed oe-selftest does not **any** parallelization by default (just one job) so this bug is not fully solved. The final solution is to create **another** (I am not really sure if this is the path to go) script which basically does the following $ echo $ALLTESTS | time parallel --jobs4 oe-selftest -r where ALLTESTS is a variable containing all tests cases (modules) found by the the runner (oe-selftest-internal) (i.e. ALLTESTS="$(oe-selftest -m | awk '{ print $NF }' | grep -v ':')"). [1] http://lists.openembedded.org/pipermail/openembedded-core/2017-October/143697.html
I have submitted a patch with does the parallelization using GNU parallel [1] but in a internal discussion with RP, this is not the approach we want. Ideally, we should do the parallelization inside the meta/lib/oeqa/core as Anibal started but not concluded AFAIK. [1] http://lists.openembedded.org/pipermail/openembedded-core/2017-December/145742.html
Patches to parallelise oe-selftest merged for 2.6.