publisher job is pushing empty commits
Affects | Status | Importance | Assigned to | Milestone | |
---|---|---|---|---|---|
CI Train [cu2d] |
Fix Released
|
Medium
|
Robert Bruce Park |
Bug Description
Saviq reported the publisher job pushed empty commits to qtmir/trunk and qtmir/gles, by CI Train Bot with "Resync trunk" as the commit message. There was a discussion that it could be bot permissions that should be fixed, but then Saviq pointed:
09:32:12 <Saviq> sil2100, it tried multiple times because it couldn't push to lp:cmake-extras
09:33:12 <Saviq> sil2100, Ursinha, this is what caused it https:/
09:33:42 <Saviq> sil2100, Ursinha, ah I know what happened
09:33:56 <Saviq> it pushed qtmir and qtmir-gles to trunks on the first try
09:34:08 <Saviq> and then noticed that trunks are ahead of what the silo branches were built on
09:34:14 <Saviq> so decided it needs to reconcile
09:34:18 <Saviq> and so on, and so on
Lukasz pointed that this should be fixed in such a way that it should register which merges it succeeded with and not try to push those again. Also, we have no feedback right now, so in this case it remembers which components succeeded and inform the frontend that some M&C's failed.
Changed in cupstream2distro: | |
assignee: | nobody → ℝobert 𝔹ruce ℙark (robru) |
importance: | Undecided → Medium |
status: | New → Triaged |
Changed in cupstream2distro: | |
assignee: | ℝobert 𝔹ruce ℙark (robru) → Łukasz Zemczak (sil2100) |
Ahhh, I see. This is caused by succeeding to push one branch but failing to push a different branch in the same silo. When this was first explained to me I didn't understand why it would try to push a successful merge more than once.
Now that the bot has all the right perms this shouldn't happen again, but yes a little safety check might be nice.