We are currently creating a fairly big openstack setup and our goal is to have everything in high-availability. However, when it comes to Ceilometer and it's database, I have a few questions.

Here's a quick overview of the setup : we have two controller nodes and I would like each of these controller nodes to run a ceilometer instance. Then, I would like to have each of these ceilometer to have a replicated database.

I understand that Mongodb is recommended for ceilometer (or at least that's what is used in the installation instructions). However, sharding it for production requires what seems to be a fairly complex setup with 3 configuration servers and a certain number of replicated datasets. Obviously, that would take more than 2 controllers to meet the "production level" recommendations.

Has anybody experimented with such a setup for ceilometer? What would be your recommendations? Would you suggest going with a very small mongodb setup with the config instances split between the two nodes?