running whoopsie during tests causes test failures

[Brian Murray]

I've switched the apport-noui.path systemd file from:

PathChanged=/var/crash

to

PathExistsGlob=/var/crash/*.crash

I've done this because with PathChanged it seemed to me that the whoopsie-upload-all, called by apport-noui.service, would be called when any file (e.g. .upload, .uploaded) were created in /var/crash. Naturally, this seemed excessive to me and PathExistsGlob seems like a better fit. Additionally, it is worth noting that the following from the systemd.path documentation:

"If a path already exists (in case of PathExists= and PathExistsGlob=) or a directory already is not empty (in case of DirectoryNotEmpty=) at the time the path unit is activated, then the configured unit is immediately activated as well. Something similar does not apply to PathChanged= and PathModified=."

Given that it appears with PathChanged the apport-noui service would never be run during boot up, again not what we want.

However, making these changes seems to have caused some strange test failures with the apport's tests.

https://jenkins.qa.ubuntu.com/job/wily-adt-apport/

runs 47 through 55

[Brian Murray (brian-murray)]

I haven't been able to locally recreate the same amd64 test failures, but have run into the following i386 failure (and only this failure which is odd):

======================================================================
FAIL: test_core_dump_packaged (__main__.T)
packaged executables create core dumps on proper ulimits
----------------------------------------------------------------------
Traceback (most recent call last):
  File "./test_signal_crashes.py", line 359, in test_core_dump_packaged
    self.check_report_coredump(self.test_report)
  File "./test_signal_crashes.py", line 918, in check_report_coredump
    self.assertTrue('CoreDump' in r)
AssertionError: False is not true

If I switch back to PathChanged from PathExistsGlob then the test passes.

[Brian Murray (brian-murray)]

It's actually a combination of the dependency upon whoopsie and the Path change.

If the package depends on whoopsie then tests FAIL if we use PathExistsGlob, if we use PathChanged then test PASS.

If the package does not depend on whoopsie then the tests PASS with either Path setting.

[Brian Murray (brian-murray)]

I've uploaded apport version 2.18-0ubuntu5 which removes the dependency of apport-noui on whoopsie as a stop gap measure.

[Martin Pitt (pitti)]

It's not all that surprising that whoopsie interferes with the test suite. The test suite assumes that new reports stay "unseen" and unprocessed, but whoopsie does process the reports. Also, it seems that whoopsie-upload-all itself crashes sometimes. Also, apport-noui.service is missing Type=oneshot.

We don't have tests for whoopsie and apport-noui, thus I'll just drop that from the test dependencies and restore the "whoopsie" dependency.

[Launchpad Janitor (janitor)]

This bug was fixed in the package apport - 2.18-0ubuntu6

---------------
apport (2.18-0ubuntu6) wily; urgency=medium

  * Drop apport-noui from test dependencies, as whoopsie interferes with the
    test crashes while the test suite runs. (LP: #1478115)
  * Restore whoopsie dependency of apport-noui.
  * apport-noui.service: Add missing Type=oneshot, to fix restart limits with
    crashes happening in rapid succession.
  * Merge test fixes from trunk.
  * data/package-hooks/source_linux.py: Fix PEP-8 error.

 -- Martin Pitt <email address hidden> Mon, 10 Aug 2015 11:25:07 +0200

[Brian Murray (brian-murray)]

Martin - Thanks for fixing this. Do you have any more information about whoopsie-upload-all crashing?

[Martin Pitt (pitti)]

Brian Murray [2015-08-10 17:14 -0000]:
> Do you have any more information about whoopsie-upload-all crashing?

I'm afraid not, I just noticed the .crash report for it in the
previous test runs.