Re: [PATCH] btrfs-progs: docs: update the qgroup performance hit

[Max Chernoff <git@maxchernoff.ca>]

Hi Qu,

On Sat, 2026-05-02 at 19:00 +0930, Qu Wenruo wrote:
> 在 2026/5/2 17:06, Max Chernoff 写道:
> > Would it maybe be worth adding something like the following?
> >
> >      If quota support is absolutely required for a filesystem with many
> >      snapshots, then "simple quotas" (``squota``) should be preferred
> >      over "quota groups" (``qgroups``).
> >
> > I've never used simple quotas myself, so I have no idea if that is
> > actually true, but the documentation suggests that that's the case at
> > least.
>
> Squota has one special quirk that makes it pretty anti-instinctual, thus
> I intentionally skipped that recommendation.
>
> E.g. We have subvolume 256, which is the initial subvolume, have around
> 128MiB data. Then a snapshot of subvol 256 is created, that snapshot has
> the subvolid 257.
>
> Now we have something like: (squota has no distinguish between reference
> and exclusive, it treat them all as semi-exclusive)
>
>    Subvolid         Reference/Exclusive
>         256         128MiB
>         257         0
>
> Here we ignore metadata to make it easier to explain.
>
> If you know subvol 257 is a snapshot of 256, it makes sense since there
> is no new data in subvol 257 yet.
>
> Then if subvol 256 is deleted, things go ugly immediately, after subvol
> 256 fully deleted, the result is:
>
>    Subvolid         Reference/Exclusive
>         256         128MiB
>         257         0
>
> Nothing changed.
>
> I do not believe most end users expect results like this at all.

Ok yeah, I agree that that would probably be too confusing to recommend
to unsuspecting users. Thanks again for the explanation.

-- Max