On 2004-02-20T17:40:42, Heinz Mauelshagen <Mauelshagen@xxxxxxxxxx> said: > > > in a fabric, which come back online shortly later when all paths have been > > > failed by the multipath target. > > You think EIO is the wrong response to this ? > In case all paths come back after a reasonably short time it is. Can you define 'reasonably short time'? If they come back faster than a single SCSI timeout: we are covered. Otherwise, I'd say we should place the limit at 'We can only queue so much until we run out of memory'; alas, the failure case which you bring up is swapping to a fully disconnected m-p device under OOM, and that's pathological. Sincerely, Lars Marowsky-Brée <lmb@xxxxxxx> -- High Availability & Clustering \ ever tried. ever failed. no matter. SUSE Labs | try again. fail again. fail better. Research & Development, SUSE LINUX AG \ -- Samuel Beckett