use of sleep in nova_rootwrap test is a race condition

[Monty Taylor]

In nova/tests/test_nova_rootwrap.py, test_KillFilter spawns a sleep and then attempts to kill it. The problem is, a latency spike could cause the subsequent kill to wait long enough for the sleep to go away, and then boom, test suck.

Instead of sleep, perhaps cat with stdin=PIPE stdout=PIPE stderr=STDIN would sit there doing nothing indefinitely.

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: master
Review: https://review.openstack.org/14838

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.openstack.org/14838
Committed: http://github.com/openstack/nova/commit/b0529de679056935a48b6a1e9879765ffdc1799b
Submitter: Jenkins
Branch: master

commit b0529de679056935a48b6a1e9879765ffdc1799b
Author: Monty Taylor <email address hidden>
Date: Thu Oct 25 11:39:31 2012 -0700

    Use cat instead of sleep for rootwrap test.

    Sleep is a race condition. cat will block indefinitely.

    Fixes bug 1071443

    Change-Id: Ia731580ed335e53646aa67b2e7b9424118f0c9d5

[Thierry Carrez (ttx)]

Hrm. The idea behind using sleep was that the created process would always die after 5 seconds (the test does not kill the process at all). Here your "cat" will survive a lot longer (until all tests complete and parent process dies ?) which might or might not be an issue ?

Maybe consider terminating that subprocess at the end of test_KillFilter ?

[Monty Taylor (mordred)]

Yeah, ok. Fair enough.

[Thierry Carrez (ttx)]

Filed that as separate bug, see bug 1079629