yahoo-eng-team team mailing list archive

[OpenStack Infra <1948963@xxxxxxxxxxxxxxxxxx>]

Reviewed:  https://review.opendev.org/c/openstack/nova/+/814735
Committed: https://opendev.org/openstack/nova/commit/a7bccff06b871765fb87ee3e9edd9458c7f73701
Submitter: "Zuul (22348)"
Branch:    master

commit a7bccff06b871765fb87ee3e9edd9458c7f73701
Author: Balazs Gibizer <balazs.gibizer@xxxxxxxx>
Date:   Wed Oct 27 16:29:29 2021 +0200

    Fix interference in db unit test
    
    We discovered that two unit test cases added in
    I0647bb8545c1464b521a1d866cf5ee674aea2eae cause errors like
      oslo_db.sqlalchemy.enginefacade.AlreadyStartedError:
      this TransactionFactory is already started
    
    when the db tests run selectively with tox -e py38 nova.tests.unit.db
    but does not cause errors if the whole unit test suit is run.
    
    This error happened because our db code uses two global transaction
    factory, one for the api DB and one for the main DB. There was a global
    flag SESSION_CONFIGURED in our Database fixture that guarded against
    double initialization of the factory. But the faulty test cases in
    question do not use our Database fixture but use the
    OpportunisticDBTestMixin from oslo_db. Obviously that fixture does not
    know about our SESSION_CONFIGURED global. So if one of the offending
    test case ran first in an executor then that initialized the
    transaction factory globally and a later test that uses our Database
    fixture tried to configure it again leading to the error. For some
    unknown reason if these tests were run in the opposite order the faulty
    re-initialization did not happen. Probably the OpportunisticDBTestMixin
    was able to prevent that.
    
    A previous patch already removed the global SESSION_CONFIGURED flag
    from our fixture and replaced it with a per DB specific patch_factory
    calls that allow resetting the state of the factory at the end of each
    test case. This would already solve the current test case issue as only
    our offending test cases would initialize the global factory without
    cleanup and we have one test case per DB. So there would be no
    interference. However if in the future we add similar test cases then
    those can still interfere through the global factory.
    
    So this patch fixes the two offending test case. Also it extends the
    DatabasePoisonFixture used for the NoDbTestCase tesst. The poison now
    detects if the test case starts any transaction factory.
    
    This poison caught another offender test case,
    test_db_sync_with_special_symbols_in_connection_string, that was marked
    NoDb but actually using the database. It is now changed to declare
    itself as a test that sets up the DB manually and also it is changed to
    use the Database fixture instead of touching the global factory
    directly.
    
    Closes-Bug: #1948963
    
    Change-Id: Id96f1795034490c13125ebbab49b029fb96af1c7


** Changed in: nova
       Status: In Progress => Fix Released

-- 
You received this bug notification because you are a member of Yahoo!
Engineering Team, which is subscribed to OpenStack Compute (nova).
https://bugs.launchpad.net/bugs/1948963

Title:
  runnig test_db_version_alembic unit test only with other unit.db tests
  causes: oslo_db.sqlalchemy.enginefacade.AlreadyStartedError:   this
  TransactionFactory is already started

Status in OpenStack Compute (nova):
  Fix Released

Bug description:
  If you run

    tox -e py38 nova.tests.unit.db

  then some unit.db tests will fail with

    oslo_db.sqlalchemy.enginefacade.AlreadyStartedError: this
  TransactionFactory is already started

  The exact failed tests and the amount of them depending on the number
  of executors and some random seed.

To manage notifications about this bug go to:
https://bugs.launchpad.net/nova/+bug/1948963/+subscriptions