yahoo-eng-team team mailing list archive

[OpenStack Infra <1888878@xxxxxxxxxxxxxxxxxx>]

Reviewed:  https://review.opendev.org/745746
Committed: https://git.openstack.org/cgit/openstack/ovsdbapp/commit/?id=3cf8a427c0a4dfc248e0778c3b1ee65671d1d07b
Submitter: Zuul
Branch:    master

commit 3cf8a427c0a4dfc248e0778c3b1ee65671d1d07b
Author: Terry Wilson <twilson@xxxxxxxxxx>
Date:   Tue Aug 11 13:12:47 2020 -0500

    Add lock when calling idl.run()
    
    python-ovs transparently handles DB reconnections during calls to
    Idl.run(). Since ovsdbapp uses a separate thread for the
    Connection (so it can handle responding to OVSDB keep-alive echo
    requests) and for processing RowEvent notifcations, it is possible
    that a thread could be trying to access OVSDB data at the same
    time that the Connection thread is updating it. Especially during
    a reconnect where the in-memory copy of the DB will be completely
    re-written, this causes a problem.
    
    Theoretically, an application that uses ovsdbapp shouldn't access
    anything in the backend code directly, but since there is now only
    one supported Backend, that restriction has slipped a bit. If
    all accesses were in Command objects, mostly there wouldn't be a
    problem as they would be handled in the Connection thread. But
    with both ReadOnlyCommands and use of lookup() and anything in
    idlutils, the main thread can access the in-memory db directly and
    potentially cause problems.
    
    The most common problematic method to be called outside of Command
    objects is lookup() and ReadOnlyCommands. This patch adds a lock
    around calls to idl.run() and lookup() and execute() for the
    ReadOnlyCommand case. row_by_value() is another target, but it is
    left untouched because it doesn't have access to the API instance
    and it is *mostly* called from inside Command objects. User code
    that doesn't will be easy to find because user code basically
    shouldn't use idlutils.
    
    Change-Id: I98c37771883103e1fb0468de9cf85364071993fa
    Closes-Bug: #1888878


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

-- 
You received this bug notification because you are a member of Yahoo!
Engineering Team, which is subscribed to neutron.
https://bugs.launchpad.net/bugs/1888878

Title:
  Data not present in OVN IDL when requested

Status in neutron:
  In Progress
Status in ovsdbapp:
  Fix Released

Bug description:
  During a test execution, we were expecting the NB "NB_Global" table to
  have the cfg values. After the test execution (and the error presented
  here), "NB_Global" table was consulted via cli command (ovn-nbctl) and
  it was correctly populated.

  Error: http://paste.openstack.org/show/796293/

  Between the reconnection and the command (and the exception), there
  are is a time gap of only 36ms. Is it possible that the in-memory DB
  was still being populated?

To manage notifications about this bug go to:
https://bugs.launchpad.net/neutron/+bug/1888878/+subscriptions