Christopher Baines <[email protected]> writes:

> Andreas Enge <[email protected]> writes:
>
>> for a day or two I have not seen any progress on the branch on QA, and
>> the build farm seems to be busy with other stuff. I have restarted the
>> build coordinator, bffe and qa-frontpage on bayfront, and the data
>> service for QA. So far to no avail, although the builds for this branch
>> are marked as scheduled:
>>    
>> https://data.qa.guix.gnu.org/revision/4af027127a87f508237bd40a9f683644c79e9f58/builds
>>
>> Chris, do you see what is going wrong?
>
> It's not immediately clear.
>
> Regarding the build coordinator and bffe, I don't know of any issues
> with them, so if they're running but you think something might be up,
> let me know instead of restarting them, as I'll try and investigate
> further.
>
> Unfortunately the same can't be said for the data service currently, it
> keeps running out of file descriptors, potentially due to the same
> problem that's been affecting Cuirass. There's an open Pull Request that
> I think might help: https://codeberg.org/guix/guix/pulls/11744
>

Perhaps the "Max open files" setting needs to be increased. I remember
the default settings were soft limit 1024 and hard limit 4096. In some
cases, these are simply not sufficient. For example, when I use Garage,
I often encounter the "Too many open files" error. To avoid this, I need
to set "make-forkexec-constructor #:resource-limits '((nofile 1048576
1048576))".

> The QA Frontpage is still pending at least partial replacement, so no
> harm in restarting it, and sometimes it can also be quite helpful.
>
> One action that can get builds going is running the following command on
> bayfront:
>
>   guix-build-coordinator setup-failures clear
>
> I think when the data service instances are low on file descriptors,
> that can lead to more errors and setup failures from builds. You can see
> the setup failures here https://bordeaux.guix.gnu.org/setup-failures but
> I'll look at putting a link to that on the activity page. I've already
> got it highlighting issues with agents and that's really helpful.
>
> The most likely issue for what you're seeing is that the builds are
> blocked, but for some reason the data service hasn't kept it's tracking
> of that up to date. If the builds for the branch have seemingly stopped,
> you can check this by going to the builds page on the revision,
> filtering by scheduled and clicking through to the builds. If it says
> blocked there, but it's in the "Unknown" column on the qa-frontpage,
> then it's probably a data service issue.

Attachment: signature.asc
Description: PGP signature

Reply via email to