#39 Deal with timing issues re network install repos somehow
Closed: Fixed None Opened 7 years ago by adamwill.

So especially with the new worker host, we're having some annoying timing issues with network install tests.

When we schedule the tests, the compose is complete, but it is not yet synced to dl.fp.o. So network install tests that run soon after scheduling will likely get packages from the previous compose, not the new one.

There is a fedmsg when the sync completes, and we just got the compose ID added to it, but that's not enough either, as there's still the checksum thing that librepo does to make sure the repo is 'good', when we use mirror manager. The checksums aren't updated to include the one for the new compose until some time after the sync is complete, and I don't think there's a fedmsg for that either. So even if the new compose is synced, librepo will likely reject it because the checksum for its repo metadata will not be in the 'known good' list.

There are...a couple of things we can do here. We can adjust the tests to look for the 'Custom Operating System' env group that's now present on the Server DVD, then set the universal tests back to preferring the server DVD. That would avoid this problem any day the server DVD exists.

We could also adjust the universal tests to explicitly specify inst.repo (the location would have to be passed in by the scheduler) when running on network install images, and maybe just have the network install image install_default tests not do that, so they act as tests of the 'default' mirror manager based functionality.

I'll try and wrap my head around this and come up with a plan soon, as we kinda need to fix it.


This ticket had assigned some Differential requests:
D844
D845

Well, we've done the 'use the DVD when possible' thing now, at least - D844 and D845. I think we can say that's sufficient for now.

Login to comment on this ticket.

Metadata