worker: kernel installation should prefer using Provides
We are starting to add Provides: linux-{image,headers}-<srcpkgtail> to the
appropriate installation meta-package for a source package. This means
we will have a consistent installation target for testing the contents
of a kernel package. Add support for trying these new names first
and then fall back to the existing behaviour.
Signed-off-by: Andy Whitcroft <email address hidden>
For about two weeks this is now failing and retried by different people.
I found that all version combinations work well actually:
New or Old cffi against these three sets all work
- old python3-defaults
- new python3-defaults
- all-proposed
But what is important is that this runs with a lot of memory.
as soon as I set smaller memory on the VM guest the local retries seem
to crash the same way as the test logs we see on
autopgkgtest.ubuntu.com.
We now have a robots.txt being served properly and these bots should
respect that, so we don't need to block them in the web server's own
configuration.
We're seeing tests time out due to starvation of network and IO
resources when many tests are copying in parallel. Then they can restart
and enter the same loop over again. We should instead let them take a
bit longer in the hope that we will take overall less time because more
tests get through the timeout.