Currently for the timeseries insert path, when we have many writers and large batches, we spend a great deal of time waiting on the monolithic BucketCatalog mutex. We should make an effort to shrink our critical sections and possibly move to finer-grained locking within the BucketCatalog operations.