single reply queue
Affects | Status | Importance | Assigned to | Milestone | |
---|---|---|---|---|---|
oslo.messaging |
New
|
Undecided
|
Unassigned |
Bug Description
we are experiencing a handful of rabbitmq related issues, especially under an unstable network. Sometimes the problem is tricky in that apparently mnesia database is inconsistent but it seems either it does not report such event to rabbitmq or rabbitmq somehow lost this information.
oslo.messaging's rpc call is effected most often especially using openstack-helm which creating a lot of rpc's for health check (honestly I don't know if it is worthy), when network is lagging seems mnesia has its own problem for syncing.
I am thinking if it is possible to have a single universal reply queue for all callbacks in vhost. In doing this, every callback message will be routed by a single queue and it is particularly beneficial for large deplyment where queue number is going beyond 10 thousands (most of which is reply queue which created by all kinds of agents dvr/dhcp etc)