On Sun, Sep 20, 2009 at 14:46, Robert Hancock <hancockrwd@xxxxxxxxx> wrote: > On 09/17/2009 09:44 AM, Tejun Heo wrote: >>> >>> Thanks Neil. This implies that when we see these fifteen second >>> hangs reading /proc/mdstat without write errors, there are genuinely >>> successful superblock writes which are taking fifteen seconds to >>> complete, presumably corresponding to flushes which complete but >>> take a full 15s to do so. >>> >>> Would such very slow (but ultimately successful) flushes be >>> consistent with the theory of power supply issues affecting the >>> drives? It feels like the 30s timeouts on flush could be just a more >>> severe version of the 15s very slow flushes. >> >> Probably not. Power problems usually don't resolve themselves with >> longer timeout. If the drive genuinely takes longer than 30s to >> flush, it would be very interesting tho. That's something people have >> been worrying about but hasn't materialized yet. The timeout is >> controlled by SD_TIMEOUT in drivers/scsi/sd.h. You might want to bump >> it up to, say, 60s and see whether anything changes. > > It's possible if the power dip only slightly disrupted the drive it might > just take longer to complete the write. I've also seen reports of vibration > issues causing problems in RAID arrays (there's a video on Youtube of a guy > yelling at a Sun disk array during heavy I/O and the resulting vibrations > causing an immediate spike in I/O service times). Could be something like > that causing issues with simultaneous media access to all drives in the > array, too.. There have been a rather large number of reported firmware problems lately with various models of Seagate SATA drives; typically they cause command timeouts and occasionally they completely brick the drive (restart does not fix it). I possessed 3 of these for a while and they pretty consistently fell over (even with just 3 in a low-power-CPU box with a good PSU rated for 8 drives). You might check with the various Seagate tech support lines to see if your drive firmwares are affected by the bugs (Some were related to NCQ command processing, others were just single-command failures). Cheers, Kyle Moffett -- To unsubscribe from this list: send the line "unsubscribe linux-ide" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html