Re: Disk heads won't park
Hi
On Mon, Feb 17, 2014 at 01:09:35PM +0000, Nuno Magalhães wrote:
> Greetings,
>
> One of my 4 SATA HDDs is constantly making "write noise", as if it was
> under heavy load. However, the system is idle (and recently
> installed).
> I can't see any activity on iotop and this starts immediately when i
> turn on the box. The HDD activiy LED doesn't blink on this activity.
>
> All drives seem functional, smartctl -t short <device> only shows
> errors for sdd, the oldest drive, but by the noise it makes when
> running said test that's not the culprit and it doesn't even have
> filesystems yet (it's one big LVM VG with no LVs). That leaves me with
> 2 new drives (sda and sdc) that are part of a RAID1 array where / is
> an xfs in an LV; and sdb which has /boot (256MB ext2) and /stuff (50GB
> ext4 which is 1% full, also in an LV).
>
> I believe sdb may be the culprit. Also booting has become slow (from
> about 3sec to over 10, stopping on loading linux, loading ramdisk,
> mounting filesystems, etc). (Also GRUB delays on "error: unknown lvm
> metadata header" for /dev/sdb1, but that's because /boot is not on LVM
> - i've edited all 4 simlinks to it from /etc/lvm/lvm.conf but it still
> looks at it; GRUB proceeds normally after that).
>
> Short of unplugging the drives one by one, is there a way i can a)
> discover/confirm which one is thrashing about? and b) make it stop?
> I've seen hdparm -s is not recommended, and i'm not sure if whichever
> drive it is it supports the required power management feature (they
> all support "Power Management feature set", but other power-related
> features vary).
You may be able to get clues from sysstat - it can tell you how many
blocks have been transferred to each device. However, it is not very
granular in the default configuration: I believe it is every 10
minutes by default. But if you make the box idle for 30 minutes, and
you still see IO to a drive, then that would be a good clue.
Note that even read activity may cause writes - if the atime mount
option is set (not sure whether this applies for XFS though). Thus
reads from cache may still result in writes.
Another possibility is cron jobs - but e.g. stopping cron for a period
could be a useful diagnostic: if it still happens even with cron
stopped, then cron is unlikely to be the culprit. Or at least:
unlikely to be the *only* culprit.
Same goes for logging - e.g. (r)syslogd activity.
Hope this helps
--
Karl E. Jorgensen
Reply to: