Re: [Automated-testing] [RFC PATCH 1/1] API: Allow to use xfs filesystems < 300 MB

[Petr Vorel <pvorel@suse.cz>]

Hi Li, all,

> Hi Petr, All,

> On Sat, Aug 20, 2022 at 3:28 AM Petr Vorel <pvorel@suse.cz> wrote:

> > Hi all,

> > > Hi Cyril,

> > > On Thu, Aug 18, 2022 at 1:28 PM Cyril Hrubis <chrubis@suse.cz> wrote:
> > > > > > I'm starting to wonder if we should start tracking minimal FS size
> > per
> > > > > > filesystem since btrfs and xfs will likely to continue to grow and
> > with
> > > > > > that we will end up disabling the whole fs related testing on
> > embedded
> > > > > > boards with a little disk space. If we tracked that per filesystem
> > we
> > > > > > would be able to skip a subset of filesystems when there is not
> > enough
> > > > > > space. The downside is obviously that we would have to add a bit
> > more
> > > > > > complexity to the test library.

> > > > > Maybe I could for start rewrite v2 (I've sent it without Cc kernel
> > devs now it's
> > > > > mainly LTP internal thing) as it just to have 300 MB for XFS and 256
> > MB for the
> > > > > rest. That would require to specify filesystem when acquiring device
> > (NULL would
> > > > > be for the default filesystem), that's would be worth if embedded
> > folks counter
> > > > > each MB. It'd be nice to hear from them.

> > > > The 256MB limit was set previously due to btrfs, I bet that we can
> > > > create smaller images for ext filesytems for example.

> > Thanks for input, Geert!

> > > Yeah, we used to have ext2 root file systems that fit on 1440 KiB
> > floppies.
> > These nice times when everything simple hadn't been solved yet ... :).
> > > IIRC, ext3 does have a minimum size of 32 MiB or so.
> > Interesting, I was able to create smaller.

> > I did some testing minimal size (verified on chdir01 test):
> > XFS: 300 MB, btrfs: 109 MB, ntfs: 2 MB, ext3: 2 MB, ext[24]: 1 MB, vfat: 1
> > MB, exfat: 1 MB.

> > I guess using XFS: 300 MB, btrfs: 109 MB and 16 MB for the rest could be
> > enough.


> I think so, tracking minimal FS size per FS is a practical idea.
> But one thing we have to be aware of is that there may be different
> minimal sizes for each FS version.
> (so we'd better choose the maximum of minimal sizes).

> 16MB for general FS should be fine, I will help to test that if someone
> works out the patch.

So should we combine both: minimal FS size and those XFS variables which would
allow to use lower size for XFS? I wonder which which size would be relevant,
it might be safer to use 64 MB:

https://git.kernel.org/pub/scm/fs/xfs/xfsprogs-dev.git/tree/mkfs/xfs_mkfs.c
/*
 * Realistically, the log should never be smaller than 64MB.  Studies by the
 * kernel maintainer in early 2022 have shown a dramatic reduction in long tail
 * latency of the xlog grant head waitqueue when running a heavy metadata
 * update workload when the log size is at least 64MB.
 */

Because there is really not a big difference between 256MB and 300MB.

Kind regards,
Petr

> > But that would require to run all tests to see how many tests actually use
> > bigger data.


> Absolutely YES!