With https://gerrit.wikimedia.org/r/c/operations/puppet/+/1264557 trove-guestagent got rabbit transient queue durable enabled. This task tracks deploying the change to DB instances

The per-host procedure I (Filippo) worked out as of April 2026 is as follows:

1. Identify the instance name and trove ID from wmcs-openstack database instance list --all-projects
2. Enable ssh to the host: wmcs-openstack --os-project-id trove server add security group NAME ssh-from-anywhere
3. Login with the trove ssh key as user ubuntu
4. Apply the following patch to /etc/trove/conf.d/trove-guestagent.conf

@@ -95,8 +95,13 @@ log_dir = /var/log/trove/
 log_file = logfile.txt
 # log_config_append = /etc/trove/trove-logging-guestagent.conf
 
+[oslo_concurrency]
+lock_path = /var/lib/trove/tmp
+
 [oslo_messaging_rabbit]
+use_queue_manager=true
 rabbit_quorum_queue=true
+rabbit_transient_quorum_queue=true
 rabbit_retry_interval=1
 rabbit_retry_backoff=2
 

1. Delete the exchange and fanout queue from one of the cloudrabbit hosts:

rabbitmqctl list_exchanges | grep TROVE-ID
rabbitmqadmin delete exchange name=EXCHANGE-NAME
rabbitmqctl list_queues | grep TROVE-ID_fanout
rabbitmqctl delete_queue QUEUE-NAME

1. Restart guestagent on the host: systemctl restart guest-agent
2. The queue will be back under a different name: rabbitmqctl list_queues | grep TROVE-ID_fanout
3. Remove the ssh security group: wmcs-openstack --os-project-id trove server remove security group NAME ssh-from-anywhere