Skip site navigation (1)Skip section navigation (2)
Date:      Tue, 01 Oct 2013 11:14:02 +0100
From:      Johannes Totz <johannes@jo-t.de>
To:        freebsd-fs@freebsd.org
Subject:   Re: zfs: the exponential file system from hell
Message-ID:  <l2e78v$ipa$2@ger.gmane.org>
In-Reply-To: <20130930234401.GA68360@neutralgood.org>
References:  <52457A32.2090105@fsn.hu> <77F6465C-4E76-4EE9-88B5-238FFB4E0161@sarenet.es> <20130930234401.GA68360@neutralgood.org>

next in thread | previous in thread | raw e-mail | index | archive | help
On 01/10/2013 00:44, kpneal@pobox.com wrote:
> On Mon, Sep 30, 2013 at 11:07:33AM +0200, Borja Marcos wrote:
>>
>> On Sep 27, 2013, at 2:29 PM, Attila Nagy wrote:
>>
>>> Hi,
>>>
>>> Did anyone try to fill a zpool with multiple zfs in it and graph the space accounted by df and zpool list?
>>> If not, here it is:
>>> https://picasaweb.google.com/104147045962330059540/FreeBSDZfsVsDf#5928271443977601554
>>
>> There is a fundamental problem with "df" and ZFS. df is based on the assumption that each file system has
>> a fixed maximum size (generally the size of the disk partition on which it resides).
>
>> Anyway, in a system with variable datasets "df" is actually meaningless and you should rely on "zpool list", which gives you
>> the real size, allocated space, free space, etc.
>>
>>
>> % zpool list
>> NAME   SIZE  ALLOC   FREE    CAP  DEDUP  HEALTH  ALTROOT
>> pool  1.59T   500G  1.11T    30%  1.00x  ONLINE  -
>> %
>
> Well, not quite. The 'zpool' command works at a lower level of abstraction
> than the 'zfs' command. And zpool has a quirk where the amount of space
> used and available is only accurate for mirrors or single disk vdevs, but
> for raidz* it does not factor in space used for redundancy. (This does not
> make it _wrong_, you just have to understand what it is telling you.)

I'd say this is a desgin flow in zfs though. One motivation for having 
it was to do away with all the layering in the storage stack and have 
something integrated. Does somebody have a usecase where the numbers 
reported by zpool for (free/used) space are actually useful?


> For example, I have two pools here, one of which (aursys) is a two way
> mirror, and the other (aurd0) is a 6-drive raidz2.
>
> [kpn@aurora ~]$ zpool list
> NAME     SIZE  ALLOC   FREE    CAP  DEDUP  HEALTH  ALTROOT
> aurd0   4.91T  3.21T  1.70T    65%  1.00x  ONLINE  -
> aursys   278G  84.7G   193G    30%  1.00x  ONLINE  -
>
> [kpn@aurora ~]$ zfs list -o space aurd0 aursys
> NAME    AVAIL   USED  USEDSNAP  USEDDS  USEDREFRESERV  USEDCHILD
> aurd0   1.08T  2.14T     4.00K   59.9K             1G      2.14T
> aursys   189G  85.7G         0   44.5K             1G      84.7G
>
> See that the zfs command says aurd0 has used 2.14T of space while the zpool
> command says it has used 3.21T? But aursys (the mirror) has numbers that
> roughly match.
>
> Since 'zfs' works above the pool level it gives accurate sizes no matter
> what kind of redundancy (if any) you are using.
>
> Bottom line:
> The replacement for the 'df' command when using ZFS is 'zfs list'.
>





Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?l2e78v$ipa$2>