Open Stack

Hey team,

I am currently looking into using quorum queues (see patch [1]
introduced in oslo_messaging) for openstack services (nova, neutron
mostly).

After switching the value to true and bootstraping a new cluster, all
queues are quorum except the fanout and reply queues.

I understand that transient queues are managed differently in openstack,
but why is it so?

I am asking this because when using rabbit in cluster mode, the quorum
queues are replicated accross the cluster.
But because the fanout and reply queues are not replicated, if one of
the cluster is shutting down during an instance boot, the boot may endup
in error (the reply queue may be on the faulty node).

Is there any reason to not enable quorum on transient queues?


[1] https://github.com/openstack/oslo.messaging/commit/7e8acbf87070354cd2438f3ab0821d035c2e7d92