When reading a block, the disagg block manager will retry up to 100 times if it gets errors from the underlying service.

Currently this is a tight loop. But to handle transient failures it would be better to back off – add some sleep time between retries.