[OmniOS-discuss] zfs diskusage

[Jim Klimov]

16 мая 2014 г. 11:07:52 CEST, Tobias Oetiker <tobi at oetiker.ch> пишет:
>Hi Dan,
>
>Yesterday Dan McDonald wrote:
>
>> On May 15, 2014, at 8:05 AM, Tobias Oetiker <tobi at oetiker.ch> wrote:
>>
>> > Today we were out of diskspace on one of our pools ... a few
>removed
>> > snapshots later all is fine, except that I find that I don't realy
>> > understand the numbers ... can anyone elighten me?
>> >
>> > # zpool list fast
>> > NAME   SIZE  ALLOC   FREE  EXPANDSZ    CAP  DEDUP  HEALTH  ALTROOT
>> > fast  4.34T  1.74T  2.61T         -    39%  1.22x  ONLINE  -
>> >
>> > # zfs list fast
>> > NAME   USED  AVAIL  REFER  MOUNTPOINT
>> > fast  2.59T   716G  78.5K  /fast
>> >
>> > Why does the 'zpool list' claim that 2.61T is free (61%)
>> > while 'zfs list' sees 716G free (27%)
>> >
>> > I know there is raidz2 and compression so the numbers  don't match
>> > up, but I don't understand why the ratio is so different between
>> > the two.
>>
>> Richard Elling addressed something similar on a different thread:
>>
>> http://lists.omniti.com/pipermail/omnios-discuss/2014-May/002609.html
>>
>> You're running raidz2, and that's likely why you're seeing the
>discrepency between zpool and zfs.
>>
>> Try Richard's advice of running "zfs list -o space" for the
>breakdown.
>
>well that looks nicer, but the numbers don't change ... the way it
>is, it seems very difficult to judge how much space is still
>available ... 61% free vs 27% free seems to be quite a big
>difference in my eyes.
>
>cheers
>tobi
>
>
>> Dan
>>
>>

Do you have volumes or other sparrse reservations? If there are no allocated bytes, that's in zpool free space (times the overhead factor of raidz parities), but if space is reserved for these datasets - that is not in free space (not available for writing) of the pool's root filesystem dataset.
With this consideration in mind, do the numbers fit?

Alternatively,  i have a similar questionably sized pool - a raidz1 over 4*4tb disks, with about 1tb free in zfs list and 2.7tb unallocated in zpool list, and almost no volumes (none big anyway). I did not yet look deeper (i.e. knto quotas ans reservations), but now that i remembered of it - something does not add up too ;)

//jim
--
Typos courtesy of K-9 Mail on my Samsung Android