[RFC][BUG #226769] strace test side-effect hangs selftest

[Vincent Ladeuil]

>>>>> "Andrew" == Andrew Bennetts <andrew at canonical.com> writes:

    Andrew> Vincent Ladeuil wrote:
    >> While submitting a merge request to pqm, selftest hanged.
    Andrew> [...]
    >> 
    >> 3) Implement a protection mechanism in selftest as proposed in
    >> https://bugs.launchpad.net/bzr/+bug/69978/comments/2

    Andrew> I think a better place to put a watchdog mechanism is
    Andrew> in the “make check” rule that PQM runs, rather than
    Andrew> selftest itself.

Agreed.

    Andrew> This avoids adding more signal/thread interactions
    Andrew> (which are part of the problem here!).

Indeed hence my doubts :-/

    Andrew> Also, for interactive use we don't really need this
    Andrew> (a human can always hit Ctrl-C), the main thing we
    Andrew> want to avoid is blocking PQM.

Good point.

    Andrew> One option for doing this in “make check” would be a
    Andrew> simple python wrapper script that uses the subprocess
    Andrew> module to watch for output.  If there's no output on
    Andrew> stdout for, say, 120s, we can assume that the tests
    Andrew> have hung (or are appallingly slow), and abort the
    Andrew> test run by killing the test process (and ideally the
    Andrew> rest of the process group).

I was thinking about a mechanism that can guard each test, but I
prefer your approach.

    Andrew> Ideally this feature would be builtin to PQM, I'm
    Andrew> sure we aren't the first or last project that would
    Andrew> find this feature useful...

Agreed.

I'll fix the bug by skipping the offending tests, have filled bug
#226856 for the ssh server (so that the existing FIXME get fixed
:) and noticed
https://blueprints.launchpad.net/pqm/+spec/robustness so I didn't
filled a bug against pqm.

You may want to update the specification with your comments,
since I'm not authorized to do so ;-)

      Vincent