Our PathPartitionProvider picks names for new uploaded objects. These go into (descending...) partition prefixes that are periodically replaced.
Measure how such a scheme interacts with S3 access performance. Measure how many uploads/s we can get from 1, 3, and possibly more concurrent PathPartitionProviders. Do we get slowdowns from S3?