placement functional tests can collide when synchronising the traits table

[Chris Dent]

The placement functional tests make use of the traits table. At the start of most requests to the objects in resource_provider.py some code is run to make sure that traits in the os-traits library are synchronised to the table.

A global flag is present which says "I've already synchronised". Functional tests are responsible for making sure this is in the right state.

It turns out that this management was not complete, and after a recent move of db/test_resource_provider.py and db/test_allocation_candidates.py within the placement hierarchy the delicate balance of how tests are split among processes by stestr was upset. This leads to test failures where no traits are in the traits table.

The fix is to ensure that functional tests manage the related db flags both during setup and teardown and not rely solely on one or the other (as people can easily get it wrong).

[Matt Riedemann (mriedem)]

Here is at least one fingerprint:

http://logstash.openstack.org/#dashboard/file/logstash.json?query=((message%3A%5C%22%5C%5C%5C%22DELETE%20%2Ftraits%2F%5C%22%20AND%20message%3A%5C%22status%3A%20404%5C%22)%20OR%20(message%3A%5C%22AssertionError%3A%20'MISC_SHARES_VIA_AGGREGATE'%20not%20found%20in%5C%22))%20AND%20tags%3A%5C%22console%5C%22&from=7d

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: master
Review: https://review.openstack.org/557722

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.openstack.org/557722
Committed: https://git.openstack.org/cgit/openstack/nova/commit/?id=bdf06d18aa4ad81e43bf0d1a22c851b43c4cd1c5
Submitter: Zuul
Branch: master

commit bdf06d18aa4ad81e43bf0d1a22c851b43c4cd1c5
Author: Chris Dent <email address hidden>
Date: Thu Mar 29 15:10:19 2018 +0100

    [placement] Fix bad management of _TRAITS_SYNCED flag

    Placement functional tests could, depending on run order, be unable
    to synchronise the os-traits because a previous test or this test
    has failed to reset the _TRAITS_SYNCED flag.

    This change fixes it so the tests will always reset the flag at
    both setUp and tearDown.

    In the process a nearby misleading comment which says a database is
    not being used is corrected.

    Change-Id: I595be7bca2c1bde86651b126ce501286b301d272
    Closes-Bug: #1759863

[OpenStack Infra (hudson-openstack)]

This issue was fixed in the openstack/nova 18.0.0.0b1 development milestone.