On Thu, May 12, 2016 at 5:33 PM, Steven Burgess <[email protected]>
wrote:

> Snapshots take up some space, but 2M per sound much higher than expected.
> I ran an experiment on an otherwise idle machine:
>
> zpool get -Hp -o value free testpool
>
> 7362543104
>
> zfs get -rHp -o value -t snapshot name testpool/testfs | wc -l
>
>    5000
>
> Zfs destroy testpool/testfs@first%last
>
> zpool get -Hp -o value free testpool
>
> 7351909376
>
> So by destroying those 5000 snapshots we freed up 10 MiB, that's 2126
> bytes (0.002 MiB) per snapshot.
>

Which is closer to what I expected.


> How did you confirm that the snapshots took no space? If files are shared
> between snapshots, it is possible to free space by destroying a large set
> of snapshots where each snapshots used is 0. You can use zfs destroy -nv to
> see this kind of space usage.
>

This is all after the fact. But beforehand zfs list said:

NAME   USED  AVAIL  REFER  MOUNTPOINT
tank/a   300M   169G   300M  /a

so USED and REFER were identical - there's no evidence of any significant
space used by snapshots
there. Which matches what I would expect based on usage.

I have 2 similar machines, in both cases I removed 35k snapshots; in both
cases I came back to find
that I had another 70G free in the pool. And yes, I can match up the steps
on the free space graph
with when the snapshots were destroyed.

This isn't really a problem, I was just curious because the result was
unexpected.


> On Thu, May 12, 2016 at 11:14 AM, Dan McDonald <[email protected]> wrote:
>
>>
>> > On May 12, 2016, at 7:01 AM, Peter Tribble <[email protected]>
>> wrote:
>> >
>> > Does anyone know what the overhead of a zfs snapshot is?
>> >
>> > I'm not talking about the data, I'm talking about whatever space
>> > zfs needs to allocate internally for housekeeping.
>> >
>> > A quick estimate indicates something around 2M of overhead.
>> >
>> > (I've just deleted ~35k snapshots on a system, each taking no space.
>> > I was a little surprised - although pleased - to get ~70G of space back
>> > in the pool.)
>> 
>> I'd highly recommend asking this question and sharing your observations
>> with the ZFS developer's list.  In fact, why don't I just do that for you...
>> 
>> Dan
>> 
>
>


-- 
-Peter Tribble
http://www.petertribble.co.uk/ - http://ptribble.blogspot.com/



-------------------------------------------
openzfs-developer
Archives: https://www.listbox.com/member/archive/274414/=now
RSS Feed: https://www.listbox.com/member/archive/rss/274414/28015062-cce53afa
Modify Your Subscription: 
https://www.listbox.com/member/?member_id=28015062&id_secret=28015062-f966d51c
Powered by Listbox: http://www.listbox.com

Reply via email to