On Sun, Jul 12, 2020 at 11:07:05AM +0200, Solene Rapenne wrote:

> On Sun, 12 Jul 2020 09:13:47 +0200
> Otto Moerbeek <o...@drijf.net>:
> 
> > On Mon, Jun 29, 2020 at 02:30:41PM +0200, Otto Moerbeek wrote:
> > 
> > > On Sun, Jun 21, 2020 at 03:35:21PM +0200, Otto Moerbeek wrote:
> > >   
> > > > Hi,
> > > > 
> > > > both phase 1 and phase 5 need cylinder group metadata.  This diff
> > > > keeps the cg data read in phase 1 in memory to be used by phase 5 if
> > > > possible. From FreeBSD. 
> > > > 
> > > >         -Otto
> > > > 
> > > > On an empty 30T fileystem:
> > > > 
> > > > $ time obj/fsck_ffs -f /dev/sd3a
> > > >     2m44.10s real     0m13.21s user     0m07.38s system
> > > > 
> > > > $ time doas obj/fsck_ffs -f /dev/sd3a
> > > >     1m32.81s real     0m12.86s user     0m05.25s system
> > > > 
> > > > The difference will be less if a fileystem is filled up, but still 
> > > > nice.  
> > > 
> > > Any takers?  
> > 
> > No feedback. I'm getting discouraged in doing more filesystem work...
> > 
> > What to do?
> > 
> > 1) Abondon the diff
> > 2) Commit without ok
> > 
> > I did quite extensive testing, but both options are unsatisfactory.
> > 
> >     -Otto
> 
> I'm not sure how to test your diff.
> Would running fsck on a sane filesystem enough?
> 
> Are you using Vms that you halt to force a
> fsck on them? Would this be a good test too?

I have used both large and small fieysystems, clean and with
inconsistencies, both ffs1 and ffs2. Sometimes I create
inconsistencies by power cycling a machine, buut I have created faulty
filesystems by carefully overwriting meta data with dd in the past as
well.

In this case running with a restricted ulimit -d to force the fallback
code to kick in is also an good idea.

        -Otto

Reply via email to