Non side-tag builds wait for incorrect repo

[Elliott Sales de Andrade]

When I do fbrnch parallel -B, I see:

Successfully waited 2:26 for a new f33-build repo
Successfully waited 0:00 for R-cpp11-0.2.2-1.fc33~bootstrap to appear in the f33-build repo
R-cpp11-0.2.2-1.fc33~bootstrap job completed (3 jobs left)
Successfully waited 2:32 for a new f32-build repo
Successfully waited 0:00 for R-cpp11-0.2.2-1.fc32~bootstrap to appear in the f32-build repo
R-cpp11-0.2.2-1.fc32~bootstrap job completed (2 jobs left)
Successfully waited 5:16 for a new f34-build repo
Successfully waited 7:17 for a new f34-build repo
Successfully waited 0:00 for R-cpp11-0.2.2-1.fc34~bootstrap to appear in the f34-build repo
R-cpp11-0.2.2-1.fc34~bootstrap job completed (1 jobs left)

Why is it waiting for a new f3?-build repo? This sometimes takes ages, even though waiting for the package to appear would have been finished.

[Jens Petersen]

Hmm, this was a (possibly bad?) workaround for wait-repo warning about nvr is being latest yet in the target tag.

Basically what it does now is first wait for the next new-repo and then immediately it checks for the build there.
Hence the Successfully waited 0:00 for R-cpp11-0.2.2-1.fc3?~bootstrap to appear.

Why is it waiting for a new f3?-build repo? This sometimes takes ages, even though waiting for the package to appear would have been finished.

Can you say more about what you are seeing? So far this had worked for me but it is quite possible there is some fault in the logic. Not sure if we are on the same page here.

[Jens Petersen]

Maybe I should combine those outputs in fbrnch to "normalize" it.
(The longer term plan is also to replace using the python client for wait-repo, since it is also flaky.)

[Elliott Sales de Andrade]

Waiting for f33-build (for example) takes much much longer. I'm building something right now and can run koji wait-repo f33-build --build=... which returns immediately, while fbrnch is still waiting.

[Elliott Sales de Andrade]

I left something running today; it waited on f32-build/f31-build for 2 hours and timed out. But the builds were in the repo way before then.

[Jens Petersen]

Okay I dropped the double wait-repo hack anyway in git - though it ought to have worked IMHO. Thanks for raising this issue.
(Still wonder why it failed sometimes.)

Anyway encouraged by this issue, I have now implemented and committed native wait-repo in fbrnch - and I am planning to build it soon in copr. :-)

[Jens Petersen]

This should be fixed now in the 0.6 build with native wait-repo.

[Elliott Sales de Andrade]

Seems to be working much better now.