On jeu, 2005-08-11 at 16:44 +0200, Hannes Reinecke wrote: > Hi Christophe, > > I've implemented yet another path checker, based on O_DIRECT. > Rationale: On S/390 it's possible to activate multipath support for DASD > devices (called Parallel Access Volume (PAV)). As these are no SCSI > devices the readsector0 path checker wouldn't work. > And as this checker is quite generic it'll work for just about anybody. > There seem to be cases where this checker can "hang forever" on failing paths, and thus block the whole daemon. If I understand correctly the open() manual, O_DIRECT reads by-design block until the data is sent back. If there is a trick to unblock the checker I'd like to learn it, because as-is this checker is really more dangerous than useful :/ I hit this problem easily by bringing down the network interface supporting the iSCSI trafic to an sfnet-IET target. Regards, cvaroqui PS: I have to check if other checkers are not affected too. If I remember we needed some SG infrastructure to let our checkers ask the kernel to FAILFAST its requests ... -- dm-devel@xxxxxxxxxx https://www.redhat.com/mailman/listinfo/dm-devel
