oslo.db's master breaks unittest in OS projects

[Viktor Serhieiev]

When I run the unittests in different OpennStack projects using the latest oslo.db there were a lot of failures on Nova, Neutron and Keystone. The lot of tests raises OperationalError: (OperationalError) cannot start a transaction within a transaction 'BEGIN' () The right approach is to fix these projects, but we are in the end of J release, so I'm not sure, that we can merge these fixes fast.

This issue was caused by commit [1], so the faster and simplest approach is - to revert this commit an continue our fork on this in K

[1] https://github.com/openstack/oslo.db/commit/78fd290a89545de31e5c13f3085df23368a8afaa

[Viktor Serhieiev (vsergeyev)]

See attempt to fiz this issue in Nova - https://review.openstack.org/#/c/115255/

[Viktor Serhieiev (vsergeyev)]

See also WIP fix to Keystone - https://review.openstack.org/#/c/120146/

[Mike Bayer (zzzeek)]

Here's the issue: the fact that we currently use the StaticPool implementation with SQLite, which shares a single SQLite connection for all requests; any API method which makes use of multiple sessions at once (not a good idea, but this is prevalent throughout nova, neutron) will therefore share the same SQLite connection with multiple SQLAlchemy Connection wrappers that are not aware of the existing state.

The fix proposed at https://review.openstack.org/#/c/120148/ adds a marker into the .info dictionary to track this condition.

Ultimately we absolutely should not be creating ad-hoc sessions throughout nova, neutron, etc., this is an obvious performance and concurrency issue. The StaticPool then is kind of a hack on top of a hack.

[Viktor Serhieiev (vsergeyev)]

Fixed in Keystone by https://review.openstack.org/#/c/120146/

[Roman Podoliaka (rpodolyaka)]

Hmm, it seems that we've got another broken test in Nova in a change on review - https://review.openstack.org/#/c/119100/ I'm wondering if it's related to this bug, as I'm seeing a really weird behaviour here - http://xsnippet.org/360192/raw/

[Launchpad Janitor (janitor)]

This bug was fixed in the package keystone - 1:2014.2~b3-0ubuntu2

---------------
keystone (1:2014.2~b3-0ubuntu2) utopic; urgency=medium

  * Start failing package builds on unit test failures again:
    - d/rules: Fail build on any test failures.
    - d/p/skip-pysaml2.patch: Skip federation tests for now as pysaml2 is
      not yet in Ubuntu main and federation is only a contrib module.
    - d/p/bug-1371620.patch: Cherry pick fix from upstream VCS for db locking
      issues with sqlite during package install (LP: #1371620).
    - d/p/bug-1367354.patch: Cherry pick fix from upstream VCS to ensure that
      downgrade tests on sqlite complete successfully (LP: #1367354).
    - d/control: Add missing python-ldappool to BD's.
  * d/control: Align version requirements for pycadf and six with upstream.
  * d/p/series: Re-enable disabled add-version-info.patch.
 -- James Page <email address hidden> Mon, 29 Sep 2014 15:34:45 +0100