On Tue, Mar 13, 2018 at 10:49:23AM -0700, Darrick J. Wong wrote:
> On Tue, Mar 13, 2018 at 11:08:11AM -0600, Andreas Dilger wrote:
> > On Mar 13, 2018, at 10:36 AM, Darrick J. Wong <darrick.wong@xxxxxxxxxx> wrote:
> > >
> > > On Tue, Mar 13, 2018 at 07:23:44AM +0100, Lukas Czerner wrote:
> > >> On Mon, Mar 12, 2018 at 09:14:53AM -0700, Darrick J. Wong wrote:
> > >>> From: Darrick J. Wong <darrick.wong@xxxxxxxxxx>
> > >>>
> > >>> Create an e2scrub_all command to find all ext* filesystems
> > >>> and run an online scrub against them all.
> > >>
> > >> Sorry for not bringing that up before, by why don't we have
> > >>
> > >> e2scrub -a
> > >>
> > >> instead of this ? Wouldn't it be better to have just one tool ?
> > >
> > > I'd rather have two simple tools that each do one thing ("scrub this
> > > ext4 lvm volume") ("find all ext4 lvm volumes and run scrub") than
> > > combine them into one less cohesive tool. There's precedence here with
> > > fsck.$fstype and fsck, where the first one performs an offline check of a
> > > single filesystem and the second one (if you fsck -A) finds all the
> > > individual filesystems and feeds them through fsck.$fstype. In the
> > > longer term it probably makes sense to set up a fsscrub wrapper to
> > > invoke the fs-specific scrub tools.
> > >
> > > Though now that I think about that, e2scrub probably ought to take a
> > > mount point and translate that into a lvm volume, which makes
> > > e2scrub_all mostly a dumb iterator of /proc/mounts.
> >
> > Except that won't scrub offline volumes, nor will all mounted ext4
> > filesystems be LVs that can be scrubbed, so I don't think that is
> > an improvement.
>
> Ok, I'll modify e2scrub so that you can pass it either (a) a lvm block
> device or (b) a mountpoint for a filesystem on a lvm block device.
> e2scrub_all will retain its ability to schedule a check even if the fs
> isn't mounted.
It occurred to me while making these changes that it's not necessarily
safe to assume that we can check all the offline LVs -- suppose you have
clustered storage being managed by LVM. If each node only e2scrubs the
LVs that it has mounted, we avoid a thundering stampede at 3:30am when
all the nodes all start trying to e2scrub any LV they have access to.
However, I'll grant you that we could (more) easily have offline local
LVs, so I think a reasonable compromise is to establish "e2scrub_all -A"
to mean "check all the LVs regardless of mount state".
--D
> --D
>
> > Cheers, Andreas
> >
> >
> >
> >
> >
>
>
