A service should use a service specific control exchange for rpc

[Bogdan Dobrelya]

The RPC code from Oslo registers the control_exchange option
as a generic for openstack, but provides an API for changing it.

The generic 'openstack' value may conflict with other services
that configure the same default exchange with for its own use,
like enabling amqp_durable_queues. It would be also wise to reduce the failure domain by splitting off the service-specific exchanges.

Allow Cinder/Manila (at least) to configure the control exchange for its specific use without conflicts.

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: master
Review: https://review.opendev.org/c/openstack/cinder/+/820189

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: master
Review: https://review.opendev.org/c/openstack/manila/+/820197

[OpenStack Infra (hudson-openstack)]

Change abandoned by "Bogdan Dobrelya <email address hidden>" on branch: master
Review: https://review.opendev.org/c/openstack/manila/+/820197

[Bogdan Dobrelya (bogdando)]

There is upgrade impact associated with the change, if "newer" services have to pick up and process messages from "older" services - with this change, they will be on a different exchange, so they won't see any messages.

To avoid this, olso.m should provide a fallback mechanism to cover cases when that "old" default exchange disappears. For example, some dispatcher there could introduce by default twice shared control exchanges, one durable by default and the other not, and a switch to decide to publish on one or the other (a dispatcher).

[Bogdan Dobrelya (bogdando)]

Related https://bugs.launchpad.net/oslo.messaging/+bug/1953351

[OpenStack Infra (hudson-openstack)]

Change abandoned by "Bogdan Dobrelya <email address hidden>" on branch: master
Review: https://review.opendev.org/c/openstack/cinder/+/820189
Reason: This should be addressed in https://bugs.launchpad.net/oslo.messaging/+bug/1953351