To Reproduce Steps to reproduce the behavior:
Customer is running RHOSP 17.1-based OpenStack environments with custom deployment framework and different implementation of workloads covered by SE (update from RHOSP 16.2 is in progress). Unlike RHOSP defaults, SR-IOV ports in this environment have propagate_uplink_status flag set to true.

When Neutron SR-IOV agent is restarted, connectivity is interrupted. We have reproduced this problem by running the following two commands on compute node (same commands are executed during Neutron SR-IOV agent initial sync:

ip link set p3p1 vf 55 spoofchk on
ip link set p3p1 vf 55 state auto

In VM the following message is logged:

iavf 0000:00:08.0: Reset indication received from the PF

I will try to figure out if this outcome is expected from SR-IOV perspective by collaborating with relevant support group and will let share their follow-up here. But it looks like second look at propagate_uplink_status is needed.

Expected behavior
Neutron SR-IOV agent restart doesn't break connectivity for workloads

Bug impact
Severe impact: all updates and some deployments cause downtime for workloads

Known workaround
WIP