yahoo-eng-team team mailing list archive

[Rodolfo Alonso <2020802@xxxxxxxxxxxxxxxxxx>]

Public bug reported:

This bug is related to [1] and the patch [2] (added in Yoga).

Some administrators, using older versions, are experiencing the issue
reported in [1]. For example, one user has an environment that has
several external networks with multiple projects (1K) and multiple
network RBACs (around 2.5K). The CLI command "openstack network list",
executed by a project member (not an admin), is timing out.

I've manually tested in a lab environments, based in stable/train (RH
OSP16). I've manually backported [2] and added the indexes to the
database. The network list API command improves from 2.5 seconds to 0.4
seconds [3]. In a loaded environment where the API times are slower,
that improvement (5x) could be critical.

This is why I'm proposing this bug, to make the DB migration in [2]
conditional. That means adding a check before adding any index. If the
index is already present in the DB, that means the admin has manually
added it. If not, the migration will add it.

The change I'm proposing **does NOT affect** any new deployment or any
upgrade from a system without the indexes manually added.

[1]https://bugs.launchpad.net/neutron/+bug/1918145
[2]https://review.opendev.org/c/openstack/neutron/+/810072
[3]https://paste.opendev.org/show/bTNKOkxldBrY1V4F9Ioj/

** Affects: neutron
     Importance: Undecided
         Status: New

-- 
You received this bug notification because you are a member of Yahoo!
Engineering Team, which is subscribed to neutron.
https://bugs.launchpad.net/bugs/2020802

Title:
  Make DB migration "Add indexes to RBACs" conditional

Status in neutron:
  New

Bug description:
  This bug is related to [1] and the patch [2] (added in Yoga).

  Some administrators, using older versions, are experiencing the issue
  reported in [1]. For example, one user has an environment that has
  several external networks with multiple projects (1K) and multiple
  network RBACs (around 2.5K). The CLI command "openstack network list",
  executed by a project member (not an admin), is timing out.

  I've manually tested in a lab environments, based in stable/train (RH
  OSP16). I've manually backported [2] and added the indexes to the
  database. The network list API command improves from 2.5 seconds to
  0.4 seconds [3]. In a loaded environment where the API times are
  slower, that improvement (5x) could be critical.

  This is why I'm proposing this bug, to make the DB migration in [2]
  conditional. That means adding a check before adding any index. If the
  index is already present in the DB, that means the admin has manually
  added it. If not, the migration will add it.

  The change I'm proposing **does NOT affect** any new deployment or any
  upgrade from a system without the indexes manually added.

  [1]https://bugs.launchpad.net/neutron/+bug/1918145
  [2]https://review.opendev.org/c/openstack/neutron/+/810072
  [3]https://paste.opendev.org/show/bTNKOkxldBrY1V4F9Ioj/

To manage notifications about this bug go to:
https://bugs.launchpad.net/neutron/+bug/2020802/+subscriptions