neutron/releasenotes/notes/bug-1921150-c02692e548a3750e.yaml
Bence Romsics 7f35e4e857 Physical NIC RP should be child of agent RP
In the fix for #1853840 I made a mistake and since then we created
the physical NIC resource providers as a child of the hypervisor
resource provider instead of the agent resource provider. Here:

https://review.opendev.org/c/openstack/neutron/+/696600/3/neutron/agent/common/placement_report.py#159

This *did not* break the minimum bandwidth aware scheduling.
But still there are multiple problems:

1) If you created your physical NIC RPs before the fix for #1853840
   but upgraded to after the fix for #1853840, then resource syncs
   will throw an error in neutron-server at each physical NIC RP
   update. That pollutes the logs and wastes some resources since
   the prohibited update will be forever retried.

2) If you created your physical NIC RPs after the fix for #1853840
   then your physical NIC RPs have the wrong parent. Which again
   does not break minimum bandwidth aware scheduling. But it may pose
   problems for later features wanting to build on the originally
   planned RP tree structure.

3) Cleanup of decommissioned RPs is a bit different than expected.
   This cleanup was always left to the admin, so it only affects a
   manual process.

The proper RP structure was and should be the following:

The hypervisor RP(s) must be the root(s).
As a child of each hypervisor RP, there should be an agent RP.
The physical NIC RPs should be the children of the agent RPs.

Unfortunately at the moment the Placement API generically prohibits
update of the parent resource provider id in a PUT request:

https://docs.openstack.org/api-ref/placement/?expanded=update-resource-provider-detail#update-resource-provider

Therefore without a later Placement change we cannot fix the RPs
already created with the wrong parent. However we can fix the RPs
to be created later. We do that here. We also fix a bug in the unit
tests that allowed the wrong parent to pass unnoticed. Plus we
add an extra log message to direct the user seeing the pollution
in the logs to the proper bug report.

There may be a follow up patch later, because not all RP re-parenting
operations are problematic, therefore we are thinking of relaxing
this blanket prohibition in Placement. When Placement allows updates
to the parent id we can fix RPs already created with the wrong parent
too.

Change-Id: I7caa8827d22103600ca685a58294640fc831dbd9
Closes-Bug: #1921150
Co-Authored-By: "Balazs Gibizer" <balazs.gibizer@est.tech>
Related-Bug: #1853840
2021-03-26 15:52:18 +01:00

18 lines
958 B
YAML

---
issues:
- |
When using the minimim-bandwidth QoS feature due to bug
https://launchpad.net/bugs/1921150 physical NIC resource providers
were for some time created with the wrong parent (i.e. the
hypervisor RP). This is now partially fixed and new resource
providers are created now with the expected parent (i.e. the agent
RP). However Placement does not allow re-parenting an already
existing resource provider, therefore the following Placement
DB update may be needed after the fix for bug 1921150 is applied:
neutron/tools/bug-1921150-re-parent-device-rps.sql
Until all resource providers have the proper parent, neutron-server
will retry the re-parenting update, which will be rejected every time,
therefore expect polluted logs and some wasted load on Placement.
However please note that the bandwidth-aware scheduling is supposed
to work even with the wrongly parented resource providers.