Right now, the metadata agent writes into the SB database to indicate that a network has been provisioned with metadata on a particular chassis. This triggers updates to all clients monitoring the Chassis columns and makes neutron-server wait for an event during a VM boot up with a 15s timeouts. On very loaded systems, it is very frequent to see the 15 seconds timeout expire so the mechanism becomes useless. Also, on non loaded systems, the wait is not needed since cloud-init retries the metadata requests up to 20 times which exceeds by far the 15 seconds. By eliminating this readiness mechanism we can set the grounds to make the metadata agent connect to slave instances for read-only operations.
This is a scale related enhancement
Verified on RHOS-16.1-RHEL-8-20211007.n.1 with python3-networking-ovn-7.3.1-1.20210714143309.el8ost.noarch Monitored SB DB and verified that there are no more writings to Chassis table external_ids when creating VMs. Used this command: sudo ovsdb-client monitor tcp:172.17.1.137:6642 OVN_Southbound Chassis external_ids. Before the fix there were neutron-metadata-proxy-networks parameter updates.
Since the problem described in this bug report should be resolved in a recent advisory, it has been closed with a resolution of ERRATA. For information on the advisory (Red Hat OpenStack Platform 16.1.7 (Train) bug fix and enhancement advisory), and where to find the updated files, follow the link below. If the solution does not work for you, open a new bug report. https://access.redhat.com/errata/RHBA-2021:3762