It seems this may only present itself if neutron-server, etc are not upgraded and database migrations (ie. neutron-db-manage) are not run prior to network nodes being upgraded.
In a juju deployment this means that neutron-api units need to be upgraded prior to neutron-gateway units.
I'll test this theory again tomorrow from scratch to confirm as my current deployment has seen a lot of manual manipulation today.
I'm curious if anyone who's hit this remembers the order of their upgrades.
I came across this as I was bisecting upgrades of neutron-gateway to various upstream commits and HA routers started misbehaving when I upgraded to this commit https://opendev.org/openstack/neutron/commit/843b5ffd9a8ee3f4d9d8830f43aa3d517cc11e07 which has alembic migrations, and it dawned on me that the database hadn't been upgraded to include them.
It seems this may only present itself if neutron-server, etc are not upgraded and database migrations (ie. neutron-db-manage) are not run prior to network nodes being upgraded.
In a juju deployment this means that neutron-api units need to be upgraded prior to neutron-gateway units.
I'll test this theory again tomorrow from scratch to confirm as my current deployment has seen a lot of manual manipulation today.
I'm curious if anyone who's hit this remembers the order of their upgrades.
I came across this as I was bisecting upgrades of neutron-gateway to various upstream commits and HA routers started misbehaving when I upgraded to this commit https:/ /opendev. org/openstack/ neutron/ commit/ 843b5ffd9a8ee3f 4d9d8830f43aa3d 517cc11e07 which has alembic migrations, and it dawned on me that the database hadn't been upgraded to include them.