ML2 delete_network can loop infinitely if called in transaction

[Kevin Benton]

Both delete_network and delete_subnet in ML2 use a while True to continuously attempt to delete a network or subnet. This logic is fine assuming its DB lookups are fresh on each iteration to get the latest state. However, it fails when these functions are called inside of another transaction because its lookups will get stale data and continue to loop over that data forever.

Thanks to Armando who spotted it: http://paste.openstack.org/show/488834/

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: master
Review: https://review.openstack.org/286843

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: master
Review: https://review.openstack.org/286920

[Armando Migliaccio (armando-migliaccio)]

You're welcome.

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.openstack.org/286920
Committed: https://git.openstack.org/cgit/openstack/neutron/commit/?id=d1c501cd1d31ed3f859975b40c01cb7aaa06d26d
Submitter: Jenkins
Branch: master

commit d1c501cd1d31ed3f859975b40c01cb7aaa06d26d
Author: Kevin Benton <email address hidden>
Date: Tue Mar 1 13:49:48 2016 -0800

    L3HA: Do not wrap create/delete in transaction

    This is unsafe when calling ML2 because ML2 assumes that its
    functions will not be called inside of a transaction. This is
    not only an issue for drivers that try to do DB lookups using
    a different session in the post commit operation, but it's a
    big issue for the delete methods.

    The delete subnet and network methods in ML2 have 'while True'
    loops that catch concurrency errors and retry the operation after
    looking up info. If these are called inside a transaction, the
    lookups will contain stale information and it can lead to the
    while True loop never terminating!

    Closes-Bug: #1551958
    Change-Id: I33dc084ed15e5491fdda19da712a746ca87fbc8c

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.openstack.org/286843
Committed: https://git.openstack.org/cgit/openstack/neutron/commit/?id=afe1a834000d33900b8646d308fa26fa807a2ca0
Submitter: Jenkins
Branch: master

commit afe1a834000d33900b8646d308fa26fa807a2ca0
Author: Kevin Benton <email address hidden>
Date: Tue Mar 1 10:47:00 2016 -0800

    Block delete_(network|subnet) transactioned calls

    These functions assume that they won't be called inside of a
    transaction for two reasons. The first is an infinite loop that
    will only terminate if it can actually retrieve the latest
    information from the database. The second is that ML2's post-commit
    operations are expected to occur after data was actually committed
    so mechanism drivers that do DB lookups using a different session
    may break.

    This adds an explicit guard that prevents these two functions from
    being called within an active DB session. The rest of the ML2
    functions should eventually be converted to this style as well
    (see the partial bug).

    Related-Bug: #1551958
    Partial-Bug: #1540844
    Change-Id: I5c00b186585369ef6c8e2b9cb5a43b8bba0e5a7c

[Thierry Carrez (ttx)]

This issue was fixed in the openstack/neutron 8.0.0.0rc1 release candidate.

[OpenStack Infra (hudson-openstack)]

Related fix proposed to branch: stable/liberty
Review: https://review.openstack.org/314022

[OpenStack Infra (hudson-openstack)]

Change abandoned by John Schwarz (<email address hidden>) on branch: stable/liberty
Review: https://review.openstack.org/314022

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: stable/liberty
Review: https://review.openstack.org/321633

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.openstack.org/321633
Committed: https://git.openstack.org/cgit/openstack/neutron/commit/?id=7130d4c5013e6cc37f3b0861424c7766ad5a66a4
Submitter: Jenkins
Branch: stable/liberty

commit 7130d4c5013e6cc37f3b0861424c7766ad5a66a4
Author: Kevin Benton <email address hidden>
Date: Tue Mar 1 13:49:48 2016 -0800

    L3HA: Do not wrap create/delete in transaction

    This is unsafe when calling ML2 because ML2 assumes that its
    functions will not be called inside of a transaction. This is
    not only an issue for drivers that try to do DB lookups using
    a different session in the post commit operation, but it's a
    big issue for the delete methods.

    The delete subnet and network methods in ML2 have 'while True'
    loops that catch concurrency errors and retry the operation after
    looking up info. If these are called inside a transaction, the
    lookups will contain stale information and it can lead to the
    while True loop never terminating!

    Conflicts:
     doc/source/devref/effective_neutron.rst
     neutron/db/l3_hamode_db.py

    Closes-Bug: #1551958
    Change-Id: I33dc084ed15e5491fdda19da712a746ca87fbc8c
    (cherry picked from commit d1c501cd1d31ed3f859975b40c01cb7aaa06d26d)

[Doug Hellmann (doug-hellmann)]

This issue was fixed in the openstack/neutron 7.1.2 release.