Bug #3774
Updated by Sage Weil over 11 years ago
These should get put at the top of the scrub queue in a way that still honors all the scheduling.
The problem is that they are immediately dumped into the work queue, and bypass the scrub reservation across OSDs, making them high-impact on a running cluster.
Currently we scheduled scrub off the last_scrub list, ordered by the last_scrub timestamp. One solution is to queue the PG based on either that, -or- time 0 if the pg is marked for immediate scrub.