On Tue, Dec 13 2016, Jure Erznožnik wrote: > First of all, I apologise if this mail list is not intended for layman > help, but this is what I am and I couldn't get an explanation > elsewhere. > > My problem is that (as it seems) mdadm is touching HDD superblocks > once per second, once at address 8 (sectors), next at address 16. > Total traffic is kilobytes per second, writes only, no other > detectable traffic. > > I have detailed the problem here: > http://unix.stackexchange.com/questions/329477/ > > Shortened: > kubuntu 16.10 4.8.0-30-generic #32, mdadm v3.4 2016-01-28 > My configuration: 4 spinning platters (/dev/sd[cdef]) assembled into a > raid5 array, then bcache set to cache (hopefully) everything > (cache_mode = writeback, sequential_cutoff = 0). On top of bcache > volume I have set up lvm. > > * iostat shows traffic on sd[cdef] and md0 > * iotop shows no traffic > * iosnoop shows COMM=[idle, md0_raid5, kworker] as processes working > on the disk. Blocks reported are 8, 16 (data size a few KB) and > 18446744073709500000 (data size 0). That last one must be some virtual > thingie as the disks are nowhere near that large. > * enabling block_dump shows md0_raid5 process writing to block 8 (1 > sectors) and 16 (8 sectors) > > This touching is caused by any write into the array and goes on for > quite a while after the write has been done (a couple of hours for > 60GB of writes). When services actually work with the array, this > becomes pretty much constant. > > What am I observing and is there any way of stopping it? Start with the uppermost layer which has I/O that you cannot explain. Presumably that is md0. Run 'blktrace' on that device for a little while, then 'blkparse' to look at the results. blktrace -w 10 md0 blkparse *blktrace* It will give the name of the process that initiated the request in [] at the end of some lines. NeilBrown
Attachment:
signature.asc
Description: PGP signature
