Fedora CI enabled projects handled by both instances

[mfocko]

As has been discovered by @nforro, we have some projects listed in both prod and stage, which can lead to surprising behavior…

TODO

• Add a note to the config templates denoting that the projects should be added only to one instance at a time
• Clean up the lists to have projects handled at most by one instance

[lbarcziova]

which can lead to surprising behavior…

Can you describe this a bit more please?

[mfocko]

Can you describe this a bit more please?

AFAIK if you trigger both instances, it provides the commit statuses twice for everything (not to mention fast-forward commits), combined with the way Pagure displays it…

Example: https://src.fedoraproject.org/rpms/python-specfile/pull-request/606

I see Packit triggered 3 times:

• 2× stage
• 1× production

I'm not even sure why the production didn't get triggered 2nd time¹… Also unless I hover over the link to the dashboard, I can't even tell which one is prod and which one is stage

Footnotes

1. I have a feeling that's related to the overwriting of the statuses that @majamassarini has been fixing 👀 and it wasn't yet deployed ↩

[lbarcziova]

I see, to me it looks like we should rather fix the original issue (which as you wrote might be the one @majamassarini has been working on), rather than restricting having projects listed in both lists.

[nforro]

The issue should be fixed on stage, but not yet on prod, so that's probably fine. But another problem is that one scratch build can be triggered by stage and the other by prod, or if there is only one you don't have control over which instance will pick it up, which makes testing on stage harder.

[lbarcziova]

But another problem is that one scratch build can be triggered by stage and the other by prod, or if there is only one you don't have control over which instance will pick it up, which makes testing on stage harder.

If I remember correctly, this should be handled by checking the builds in the DB before processing the updated state, or am I missing something? If yes, can you please create an issue for that that we can fix in packit-service and close this issue?

[nforro]

If I remember correctly, this should be handled by checking the builds in the DB before processing the updated state, or am I missing something?

I'm talking about the original trigger (pagure.pr.Action). For example now both builds here were triggered by the staging instance, but on previous pushes one was triggered by staging and the other by prod, I haven't seem both triggered by prod yet, but it seems to be random.