Ceph

Right now a ceph osd crashes/ exits if it gets an io error. In many cases - i assume - ceph could simply continue running as the io error occured on data sent to the client and actually not used by ceph itself. So a switch should be added to make ceph ignore non-fatal io errors.

This could be really useful (life safer) if you have a degraded cluster and the last remaining osd crashes because of an io error. Right now the whole cluster (or at least a big part of it) is lost, as there's no chance to get the last osd running again. If ceph would ignore the io error, only a little chunk of data is lost which the filesystem of the guest will most probably be able to correct itself or just handle gracefully with a fs check/ repair.