Re: snapshot space use

[Qu Wenruo <quwenruo@cn.fujitsu.com>]

-------- Original Message  --------
Subject: Re: snapshot space use
From: Justin Maggard <jmaggard10@gmail.com>
To: Dongsheng Yang <yangds.fnst@cn.fujitsu.com>
Date: 2015年04月10日 08:40

> On Thu, Apr 9, 2015 at 5:32 PM, Dongsheng Yang
> <yangds.fnst@cn.fujitsu.com> wrote:
>> On 04/10/2015 07:39 AM, Hugo Mills wrote:
>>>
>>> On Thu, Apr 09, 2015 at 04:27:08PM -0700, Justin Maggard wrote:
>>>>
>>>> On Thu, Apr 9, 2015 at 3:24 PM, Hugo Mills <hugo@carfax.org.uk> wrote:
>>>>>
>>>>> On Thu, Apr 09, 2015 at 03:01:55PM -0700, Justin Maggard wrote:
>>>>>>
>>>>>> On Wed, Apr 8, 2015 at 5:45 PM, Qu Wenruo <quwenruo@cn.fujitsu.com>
>>>>>> wrote:
>>>>>>>
>>>>>>> Hi,
>>>>>>>
>>>>>>> You can use btrfs quota feature to do it.
>>>>>>> Like this:
>>>>>>>
>>>>>>> # btrfs quota enable <MNT_POINT>
>>>>>>> # btrfs quota rescan -w <MNT_POINT>
>>>>>>> # btrfs qgroup show -prce <MNT_POINT>
>>>>>>> qgroupid rfer    excl    max_rfer max_excl parent  child
>>>>>>> -------- ----    ----    -------- -------- ------  -----
>>>>>>> 0/5      2248704 12288   0        0        ---     ---
>>>>>>> 0/256    5509120 3272704 0        0        ---     ---
>>>>>>>
>>>>>>>
>>>>>>> rfer is all the space the subvolume takes.
>>>>>>> excl is the exclusive space the subvolume takes.
>>>>>>>
>>>>>> Yes, but this isn't as useful as it sounds if you have more than one
>>>>>> snapshot.  Because if a file is included in at least two snapshots,
>>>>>> it's no longer exclusive.  So AFAICT even with btrfs qgroups, you
>>>>>> still cannot answer the question "How much space are my snapshots
>>>>>> using?" for a given subvolume, unless you have only one snapshot.  But
>>>>>> I'd be happy to be informed otherwise. :-)
>>>>>
>>>>>      There are basically two useful answers to that question, depending
>>>>> on how the question is specified, exactly:
>>>>>
>>>>> 1) How much space would I use if I copied this subvolume to a
>>>>>      different filesystem?
>>>>>
>>>>> 2) How much space would I free up if I deleted this subvolume?
>>>>>
>>>>>      Part 1 is the rfer answer. Part 2 is the excl answer.
>>>>>
>>>>>      Which of these do you mean by "How much space are my snapshots
>>>>> using?". The question as posed is highly ambiguous, and needs
>>>>> considerably more precision before it can be answered.
>>>>>
>>>>>      If you can pose the question more precisely, you might wish to ask
>>>>> one of those questions about N subvolumes as a group -- this is
>>>>> where(*) you would define a qgroup covering the subvols you're
>>>>> interested in, and then use the above interface to ask the question
>>>>> for the group of subvols as a whole.
>>>>
>>>> To be more precise, let's consider an example, where we have a
>>>> subvolume named "stuff"; and two snapshots of "stuff", "snap1" and
>>>> "snap2".  So the question I'm posing is:
>>>>
>>>> How much space would I free up if I were to remove my snapshots,
>>>> "snap1" and "snap2", but keep "stuff" intact?
>>>>
>>>> I can't use the rfer value from snap1 and snap2, because in general
>>>> most of that space is shared by "stuff" and would not be freed.  I
>>>> could look at the excl column for snap1 and snap2 and add them up; but
>>>> if they shared extents with each other (and not with "stuff"), those
>>>> extents are no longer exclusive, and thus not accounted for.  So I
>>>> would have to delete either snap1 or snap2 in order to answer my
>>>> question using qgroups.
>>>>
>>>> I can't think of a way to define a qgroup that would be able to answer
>>>> my question, although I'm admittedly no expert in qgroups either.
>>>
>>>      I think you'd define a qgroup comtaining snap1 and snap2, and look
>>> at the excl value.
>>
>>
>> Agreed, This is a possible method to get what you want. We can create a
>> qgroup
>> in a higher level, as a pool.
>> # btrfs qgroup create 1/0 /MNT
>>
>> And then assign your snapshots into this qgroup.
>> # btrfs qgroup assign 0/xxx 1/0 /MNT
>> # btrfs qgroup assign 0/xxx 1/0 /MNT
>>
>> *Considering we can not update the quota number automatically in moving
>> qgroup
>> we need a rescan here.
>> # btrfs quota rescan -w /MNT
>>
>> Then you can get the information you want by
>> #btrfs qgroup show -prce /MNT
>>
>
> Nice, thanks for the input Hugo and Yang!  I'll play with that a bit.
> I was assuming the excl value wouldn't be effective since the extents
> are still shared by the original subvolume.
>
> -Justin

The higher level qgroup should archive your goal, if you adds all 
qgroups(including top-level subvol 5) into it.

But the higher level one can only shows you how much the space all the 
subvolumes really take. But not exactly one qgroup takes(as explained, 
no good definition on qgroup space usage due to the complexity in shared 
extents).

BTW, as you step deeper in btrfs qgroup, more problem may shows up.
For higher level subvolume qgroup, IIRC, deleting a snapshot/subvolume 
may cause wrong higher level qgroup numbers.
Anyway, if you decides to use qgroup, run rescan at a interval is suggested.

Thanks,
Qu
>
>> Thanx
>> Yang
>>
>>> Anyone out there with qgroups experience who can
>>> tell us how/whether that would do the job?
>>>
>>>      Hugo.
>>>
>>>> -Justin
>>>>
>>>>>      Hugo.
>>>>>
>>>>> (*) I'm not an expert in qgroups. I may have the idea completely
>>>>> wrong, but I think this is the right kind of approach.
>>>>>
>>>>>
>>>>>> -Justin
>>>>>>
>>>>>>> You can also refer to 'btrfs-quota'(8) and 'btrfs-qgroup'(8),
>>>>>>> Also the following wiki can help:
>>>>>>> https://btrfs.wiki.kernel.org/index.php/Quota_support
>>>>>>>
>>>>>>> NOTE: quota is not so stable and has some problem, but should give
>>>>>>> you enough info.
>>>>>>>
>>>>>>> Thanks,
>>>>>>> Qu
>>>>>>>
>>>>>>>> # zfs list -t snapshot
>>>>>>>> NAME                        USED  AVAIL  REFER  MOUNTPOINT
>>>>>>>> hetz0/be0-mail@2015-03-10  2.88G      -   387G  -
>>>>>>>> hetz0/be0-mail@2015-03-11  1.12G      -   388G  -
>>>>>>>> hetz0/be0-mail@2015-03-12  1.11G      -   388G  -
>>>>>>>> hetz0/be0-mail@2015-03-13  1.19G      -   388G  -
>>>>>>>> hetz0/be0-mail@2015-03-14  1.02G      -   388G  -
>>>>>>>> hetz0/be0-mail@2015-03-15   989M      -   386G  -
>>>>>>>>
>>>>>>>> Is there any way to do something similar to the above ZFS command?
>>>>>>>> It's
>>>>>>>> handy
>>>>>>>> to know which snapshots are taking up the most space, especially when
>>>>>>>> multiple
>>>>>>>> subvols are being snapshotted.
>>>>>>>>
>>