1318574 - Again: Artifact builds failing: Tried xx pushheads, no built artifacts found.

Status: RESOLVED FIXED

(Firefox Build System :: Android Studio and Gradle Integration, defect)

Description

[Sebastian Kaspari (:sebastian; :pocmo)]

+++ This bug was initially created as a clone of Bug #1311632 +++

Like in bug 1311632 artifact builds of mozilla-central are failing locally again:

> TheSilence:central sebastian$ ./mach artifact install
> 0:02.00 Attempting to find a pushhead containing 05e5b12f41df270b31955ff7e6d09245c1f83a7a on mozilla-central.
> 0:02.00 Attempting to find a pushhead containing 05e5b12f41df270b31955ff7e6d09245c1f83a7a on integration/mozilla-inbound.
> 0:02.83 Attempting to find a pushhead containing 05e5b12f41df270b31955ff7e6d09245c1f83a7a on releases/mozilla-aurora.
> 0:03.52 Retrieving the last 50 pushheads starting with id 84062 on integration/mozilla-inbound
> 0:04.13 Retrieving the last 50 pushheads starting with id 30965 on mozilla-central
> 0:36.73 Tried 14 pushheads, no built artifacts found.

Comment 1

[Chris Manchester (limited bugmail, email directly)]

It looks like we're finding buildbot opt jobs and taskcluster debug android builds at the route we're checking, so we have a build that produced "fennec-53.0a1.en-US.android-arm.apk" but we were expecting one that produced "target.apk".

Here's are the tasks for that revision: https://tools.taskcluster.net/index/#gecko.v2.mozilla-central.revision.05e5b12f41df270b31955ff7e6d09245c1f83a7a.mobile/gecko.v2.mozilla-central.revision.05e5b12f41df270b31955ff7e6d09245c1f83a7a.mobile

android-api-15-opt appears to come from buildbot, android-api-15-debug from TC. I know this worked as of bug 1311632, I'm not sure what happened.

Mike, can you point out where a buildbot build job would set its route? It seems like we shouldn't be doing this (or we should update our regex to accept either format).

Comment 2

[Michael Shal [:mshal]]

So it looks like debug jobs are tier-1 in Taskcluster, but opt jobs are tier-2 (so the buildbot opt job is technically still the source of truth). This means we'll only find a TC opt job in the index if we happen to skip building the buildbot job for a particular revision, which depends on scheduling intricacies. I think we should just update the regex to match either buildbot or TC naming schemes for now. If we switch to the buildbot naming scheme completely, then this would break again when opt builds become tier-1 in TC. Supporting both should work during the transition, and then we can remove the buildbot part of the regex once those builds are no longer being produced.

Patch upcoming...

Comment 3

[Michael Shal [:mshal]]

This is another example of why bug 1150976 would be very helpful. Artifact builds should just be grabbing a stable alias called "package" or some such, rather than doing a regex on artifact filenames.

Comment 4

[Michael Shal [:mshal]]

Attachment: Bug 1318574 - Support buildbot and TC jobs for opt android artifact builds;

Review commit: https://reviewboard.mozilla.org/r/94086/diff/#index_header
See other reviews: https://reviewboard.mozilla.org/r/94086/

Comment 5

[Chris Manchester (limited bugmail, email directly)]

Comment on attachment 8812290 [details]
Bug 1318574 - Support buildbot and TC jobs for opt android artifact builds;

https://reviewboard.mozilla.org/r/94086/#review94298

Thank you for the patch.

Comment 6

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/8b122a300411
Support buildbot and TC jobs for opt android artifact builds; r=chmanchester

Comment 7

[Sebastian Kaspari (:sebastian; :pocmo)]

Last time we had to fix the x86 builds too (bug 1312476) - this time again?

Comment 8

[Carsten Book [:Tomcat]]

https://hg.mozilla.org/mozilla-central/rev/8b122a300411

Comment 9

[Michael Shal [:mshal]]

(In reply to Sebastian Kaspari (:sebastian) from comment #7)
> Last time we had to fix the x86 builds too (bug 1312476) - this time again?

Bug 1293730 turned off the buildbot x86 jobs, so we only have the Taskcluster jobs for the regular builds (nightlies are still on buildbot, I believe). That means target.apk should be the only package name found with an android-x86 route, so I don't think we have to make a similar change here. Are you having issues with it currently?

Comment 10

[Sebastian Kaspari (:sebastian; :pocmo)]

(In reply to Michael Shal [:mshal] from comment #9)
> (In reply to Sebastian Kaspari (:sebastian) from comment #7)
> > Last time we had to fix the x86 builds too (bug 1312476) - this time again?
> 
> Bug 1293730 turned off the buildbot x86 jobs, so we only have the
> Taskcluster jobs for the regular builds (nightlies are still on buildbot, I
> believe). That means target.apk should be the only package name found with
> an android-x86 route, so I don't think we have to make a similar change
> here. Are you having issues with it currently?

No, I was only wondering if we missed that. Thanks! :)