Ceph » rgw

Currently persistent notifications uses cls 2pc queue to store the notification events which has the max size limit of 128Mb. In event of broker down for more than 30 minutes with continuous PUT requests on bucket with persistent notification enabled, the persistent queue gets full as the messages are not delivered due to broker outage resulting in 503 being sent to client (as per design for persistent notification)

To be more robust and allow more time for the broker to recover from outage, need to store the persistent events in a storage that does not have size limit of 128Mb, hence use cls FIFO as a backend storage to store the persistent events.

Following doc, lists down the broad design.