Other Tests getting stuck in pipdwt state.

I'm running gitlab-runner instance on FreeBSD14.2. It builds without a problem but when tests are run, one or two (always the same) of them get stuck in "pipdwt" state. They are run concurrently (by python based driver), in my case 24 (threads) at once. Both produce rather big output (stdout), that should then be consumed/processed. I've monitored kern.ipc.pipekva but it's far from kern.ipc.maxpipekva: 1207250944 - so it's not about reaching the limit. I've reduced thread count to 10 - it's the same. It doesn't happen always but pretty frequently.
I've found similar issue on julia lang github https://github.com/JuliaLang/julia/issues/23143. Any help appreciated, maybe someone can enlighten me how to interpret pipdwt.
 
Ok. I think I know what is going on. It's the scheduler starving the processes.
Single (problematic) test is timed out after 300 seconds. Meanwhile test driver spawns as many test threads as it can see computing caps (threads), in my case 24.
If these threads execute fast they are replaced by the next ones. Whole suite execute in ~20m, enough to timeout single sleeping thread (test). Still not sure if it's related to pipes writes (child test -> parent driver).
Now is the time to get PhD in ULE tuning and make it prefer "old" tasks.
 
Back
Top