Re: help debugging failures on riscv64
Hi,
On Thu, Apr 17, 2025 at 3:13 AM Michael Stone <mstone@debian.org> wrote:
>
> I've been trying to track down build failures in uqm. As you can see at
> https://ci.debian.net/packages/u/uqm/unstable/riscv64/
> https://ci.debian.net/packages/u/uqm/testing/riscv64/
> the builds fail a lot. I was able to duplicate an intermittant build
> failure locally, which went away when I increased the timeout in the
> test (originally 1 minute). But now even with a 5 minute test, the job
> is still failing. It may still be a timeout issue, but I noticed that
> when it succeeds it is usually running on debci-31 and when it fails it
> is on some other host (e.g., debci-nj-09). Is there some way to run the
> tests for uqm manually without a timeout on one of the failing hosts to
> see if it ever completes when even 5 minutes doesn't seem to be enough
> time?
Ok, I can test this locally.
> I also noticed there are architectural differences between the
> boxes, so it is possible that there really is a problem other than
> speed.
Hmm, I am afraid the latter is true.
>From these debci logs[0], I can confirm that these logs are
successfully from debci-3* as you said, the one(debci-31) is p550[1]
but lacks kernel upstream support now. Failures on debci-bj-* were
from Unmatched[2], used as Debian official buildd machies and
debci-nj-* were licheepi4a[3]. Of these, the p550 is the highest
performing until now.
If this issue unblocks something, we can ask for the debci team to
help with this, from the current point of view, this should not be a
package problem or an architecture problem.
BR,
Bo
[0]: https://ci.debian.net/packages/u/uqm/unstable/riscv64/
[1]:https://www.sifive.com/boards/hifive-premier-p550
[2]: https://www.sifive.com/boards/hifive-unmatched
[3]: https://sipeed.com/licheepi4a
>
Reply to: