Hi taffit, On 11/12/25 23:27, David Prévot wrote:
Is there a way to provide a healthy timeout (or better capacity cap) for those tests?
$(timeout) exists, I'm not sure if it's the best option for timeouts. I'm worried though that adding a timeout is going to increase flakiness if I look at the variation of run times. I don't have experience with capability caps but a very recent similar bug report did solve it that way: bug 1120170.
DEP-8 doesn’t seem to document one but maybe you have some hints…
Indeed, this feels more like something for a tricks-and-tips wiki page under https://wiki.debian.org/ContinuousIntegration/ but currently there is no such information.
Maybe we can avoid the described situation by running the tests serially (instead of the current parallel way). It’ll take more time, but if it avoids human useless interaction, I’m happy to try.
Well, regularly the test runs fast and well behaved, it's the "sometimes" I worry about. But indeed, not having parallism might make the problem deterministic to always or never show up.
What do you think of trying to actually find where it goes wrong? I can offer you access to a testbed on our ppc64el host (we'd need to align the time/date, we can do that on #debci) or if you know of a command I could run I can collect some information for you.
Paul
OpenPGP_signature.asc
Description: OpenPGP digital signature
