Today we simply purge any history older than 10 seconds as the storage engine does not support snapshots older than that. As part of this ticket, we should instead use the minimum timestamp across the cluster, which is known to the config server.