yahoo-eng-team team mailing list archive

[OpenStack Infra <2089225@xxxxxxxxxxxxxxxxxx>]

Reviewed:  https://review.opendev.org/c/openstack/nova/+/935858
Committed: https://opendev.org/openstack/nova/commit/1d43a6116721456ee6cd582240267b14bd350ec3
Submitter: "Zuul (22348)"
Branch:    master

commit 1d43a6116721456ee6cd582240267b14bd350ec3
Author: Balazs Gibizer <gibi@xxxxxxxxxx>
Date:   Thu Nov 21 09:39:40 2024 +0100

    Improve heal_instance_info_cache periodic
    
    This change removes the slightly different behavior between value 0 and
    value < 0. Now in both case the periodic task will never be started.
    
    Also this change clarifies the doc of the config option about when to
    safe to turn this periodic off.
    
    Closes-Bug: #2089225
    
    Change-Id: Id9dc598c64e632c02ab686424fbaf39688a4c3a5


** 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/2089225

Title:
  The description of the heal_instance_info_cache_interval option is
  misleading

Status in OpenStack Compute (nova):
  Fix Released

Bug description:
  The current doc text is:

  Interval between instance network information cache updates.

  Number of seconds after which each compute node runs the task of
  querying Neutron for all of its instances networking information, then
  updates the Nova db with that information. Nova will never update it’s
  cache if this option is set to 0. If we don’t update the cache, the
  metadata service and nova-api endpoints will be proxying incorrect
  network data about the instance. So, it is not recommended to set this
  option to 0.

  Possible values:

  Any positive integer in seconds.

  Any value <=0 will disable the sync. This is not recommended.

  --

  There is a list of improvement we could make in the text / implementation of the periodic to make it clearer:
  * Currently setting it to 0 make the periodic run at the default interval and exit early[2]. While setting it to -1 disables the whole periodic. We should remove this complication and simply make sure that if it is set to <=0 then the periodic is not started.

  * Disabling the periodic is not dangerous any more with the most used
  neutron backends like the in tree ovs, ovn, sriov backends as they are
  properly updating nova's view with network-changed external events[1].
  So we should clarify this in the text.

  [1]https://github.com/openstack/nova/blob/e8a544f0e1134a4b260720cc634ad1997f936d6d/nova/compute/manager.py#L11135-L11140
  [2]https://github.com/openstack/nova/blame/e8a544f0e1134a4b260720cc634ad1997f936d6d/nova/compute/manager.py#L9909-L9924

To manage notifications about this bug go to:
https://bugs.launchpad.net/nova/+bug/2089225/+subscriptions