-
Improvement
-
Resolution: Unresolved
-
Medium
-
None
-
None
-
3
-
9223372036854775807
zpool scrubs perform heavy I/O to the pool. Although there are mechanisms to stop or reduce scrub activity when lustre-initiated I/O is detected, they are reactive and so scrubs still affect latency and bandwidth. In addition, user I/O competes with the scrubs and so they take longer, and scrubs sometimes trigger device failures when a device is on the edge of failure.
For all these reasons, it can be useful to mark OSTs in a scrubbing pool as degraded.
This would essentially extend the existing ZFS zedlet which marks OSTs as degraded when device failures result in a DEGRADED zpool state.