Over the history in some cases when RESTBase can't connect to Cassandra it can log at insane rates up to 200.000 messages per second. We need to be ably to rate-limit the logs per log class, but we don't want to cut off all the logging completely when the limit is exceeded, but instead probabilistically sample the logs so that the overall rate is close to the target rate.

Looking through the available npm packages wasn't successful so far, so some custom code might need to be implemented.

Ideally that should be solved before T112648 puppet is landed on March 07 2017.