Package copier doesn't restrict builds to the target distroseries' architectures

[Jamie Strandboge]

I cannot delete linux 2.6.24-28.77 from https://launchpad.net/~ubuntu-security-proposed/+archive/ppa/+packages. I have tried several times and it seems to work when using +delete-packages, but then it shows up some hours later as a newly uploaded version. See looking at https://launchpad.net/~ubuntu-security-proposed/+archive/ppa/+delete-packages, it shows several uploads that I did not actually do. This may be because I accidentally pocket copied it to 'maverick' even though it has the 'hardy-security' distribution name (and then tried to delete it).

OOPS-1709PPA849

[William Grant (wgrant)]

There is a delayed copy for that package (see https://edge.launchpad.net/api/devel/ubuntu/maverick?ws.op=getPackageUploads&archive=https://edge.launchpad.net/api/devel/~ubuntu-security-proposed/%2Barchive/ppa). It's still sitting in Accepted, which suggests that processing is failing repeatedly. This also suggests that process-accepted.py is committing too early.

[Julian Edwards (julian-edwards)]

QueueInconsistentStateError: linux_2.6.24-28.77.diff.gz is already published in archive for maverick

    Traceback (most recent call last):
  Module lp.soyuz.scripts.processaccepted, line 250, in main
    queue_item.realiseUpload(self.logger)
  Module lp.soyuz.model.queue, line 594, in realiseUpload
    queue_source.verifyBeforePublish()
  Module lp.soyuz.model.queue, line 1558, in verifyBeforePublish
    filename, self.packageupload.distroseries.name))
QueueInconsistentStateError: linux_2.6.24-28.77.diff.gz is already published in archive for maverick

[William Grant (wgrant)]

The QueueInconsistentStateError was a red herring. The real error was this:

 NotFoundError: u'Unknown architecture hppa for ubuntu maverick'

The bug is that builds are copied for architectures not in the target -- in this case for hppa and lpia, which no longer exist in Maverick. PackageUploadBuild.publish then dies with a NotFoundError (amusingly enough, a comment in the method already warns of this).

For future reference, here's what happened:

 - 2010-08-25 22:37:25: Delayed copy from Hardy to Maverick requested. hppa and lpia builds were incorrectly copied.
 - 2010-08-25 22:50:29: Delayed copy processed for the first time. Logs do not go far back enough, but this presumably fails with the NotFoundError above. Source and amd64 binaries published, but the rest (including the PackageUpload.status Accepted → Done transition) are missed. Somehow the source and amd64 binaries get committed, which suggests transaction issues somewhere.
 - Each process-accepted run now attempts to publish the copy, but fails with the QueueInconsistentStateError.
 - 2010-08-26 02:50:00 (approx): Deletion from target archive requested.
 - 2010-08-26 03:17:18: process-deathrow removes files.
 - 2010-08-26 03:35:27: process-accepted no longer experiences the QISE, since the files are gone. Tries to publish, but fails part-way with the hppa NotFoundError again.
 - Subsequent process-accepted runs fail with QISE again, until the source is deleted and removed again.

And again. And again. And again.

[Launchpad QA Bot (lpqabot)]

Fixed in stable r12220 <http://bazaar.launchpad.net/~launchpad-pqm/launchpad/stable/revision/12220>.