On Thu, 9 Apr 2015 04:39:59 -0400 (EDT) Kamil Paral kparal@redhat.com wrote:

I see people asking on #fedora-devel or #fedora-admin about depcheck inferior architecture issue [1] quite often. Most of them are very confused. Last time I saw someone asking about this was tonight [2].

We're not able fix this easily, but I think we should finally at least put some temporary measures to "work around" this issue and stop confusing people that much, or least start explaining better what this is and what it means. I have the following ideas:

1. In depcheck, detect if the output has "inferior architecture"

substring and add an explanatory note, like this:

not ok - depcheck for Bodhi update xforms-1.2.4-2.fc22 # FAIL

arch: x86_64 details: output: |- Build xforms-1.2.4-2.fc22 failed depcheck package xforms-devel-1.2.4-2.fc22.i686 requires xforms(x86-32) = 1.2.4-2.fc22, but none of the providers can be installed xforms-1.2.4-2.fc22.i686 has inferior architecture xforms-1.2.4-2.fc22.i686 has inferior architecture package xforms-devel-1.2.4-2.fc22.i686 requires xforms(x86-32) = 1.2.4-2.fc22, but none of the providers can be installed xforms-1.2.4-2.fc22.i686 has inferior architecture xforms-1.2.4-2.fc22.i686 has inferior architecture **** PLEASE NOTE ****: This failure is most probably invalid, caused by a bug we weren't able to fix yet: https://phab.qadevel.cloud.fedoraproject.org/T366 . Please test you package dependencies manually and if everything looks correct, ignore this error. We're sorry for this inconvenience. item: xforms-1.2.4-2.fc22 outcome: FAILED summary: xforms-1.2.4-2.fc22 into F22 testing type: bodhi_update ...

2. Do not submit this result (I'm mostly concerned about Bodhi, but

there's no easy way to disconnect ResultsDB reporting from Bodhi reporting, so it would affect both systems). That can be done by filtering out "inferior architecture" CheckDetails at the end of the depcheck run, before the final TAP is generated. We would print these CheckDetails to stdout instead, so that the results would still be visible in the log.

3. Alternatively to 2), Josef proposed setting these results to

ABORTED instead of FAILED. They would still show up in ResultsDB, and they would be easy to search for (we'll need to fix T458, but we'll need to fix that anyway). I've went through bodhi_comment_directive, and I believe it will report the ABORTED outcome to Bodhi the same way as any other outcome. I'd prefer either not to report ABORTED at all, or least not send emails for them. But either way, saying ABORTED is a bit less confusing than FAILED. And if we add the explanatory note as suggested in 1), it could be a substantial improvement. I think I prefer this to 2).

What do you think? Any better suggestions?

I like all three, to be honest - put a note in the logs, don't report to bodhi and mark as ABORTED in resultsdb.

From what I've seen thus far, it's not a consistent failure but for some reason, subsequent runs aren't updating bodhi and the bounceback to PASS only shows up in resultsdb. Given that, I don't think that we need to report the result to bodhi if it's a freak occurrence.

The other thing I'd like to do is start utilizing artifacts to store repodata for at least affected runs. I've not been able to reproduce the issue and I might get at least some more insight into what's going wrong if I can get my hands on the generated repodata used during the problematic runs.

I'm going to start working on some patches to grab that data and assuming there are no objections here, handle "inferior architecture" errors as follows: - add note to result with similar wording to what Kamil listed above - mark result as ABORTED - make sure that it isn't reported to bodhi

Tim