we run some openstack clusters and currently see issues with a lot connections in rabbitmq. So I would like to ask if someone here has some knowhow about running openstack on >100 nodes he is able to share.

Our setup:

* 3 Nodes Control-Plane

* RabbitMQ Cluster with "sync_mode: all" (had issues during node-failure in the past without this)

* transport_url contains all 3 nodes

Most of the connections are neutron-based. We already played a bit with max_pool and max_pool_overload, but didnt change anything.

As next try I would add some haproxy-lb and change the transport_url to the VIP, this should afaik reduce the amount of connections to 1/3, but I dont know if there are any drawbacks jet.

Any suggestions?

Thanks a lot,

 Fabian