Currently we have this logic to persist the current queue size every 10 writes to ensure it can be restored after a crash. The problem that it's not reliable and can lead to sizing problems.

Instead, we should write all the information every time under one key: start/stop/inprogres/size.