Dump time issues

Kevin Oberman rkoberman at gmail.com
Mon Oct 27 18:09:41 UTC 2014


On Mon, Oct 27, 2014 at 9:25 AM, Chris H <bsd-lists at bsdforge.com> wrote:

> On Sat, 25 Oct 2014 11:26:44 -0700 Kevin Oberman <rkoberman at gmail.com>
> wrote
>
> > On Sat, Oct 25, 2014 at 4:38 AM, David Wolfskill <david at catwhisker.org>
> > wrote:
> >
> > > On Sat, Oct 25, 2014 at 01:26:43PM +0200, Alban Hertroys wrote:
> > > > I?m seeing something odd in my dump output that I didn?t notice
> before:
> > > >
> > > > Dumping /root...
> > > >   DUMP: WARNING: should use -L when dumping live read-write
> filesystems!
> > >
> > > (That, by the way, is pretty good advice in my experience.)
> > >
> >
> > As long as the volume is not SU+J. Then, in my experience, it's very bad
> > advice.
> This reminds me of a question that I've been meaning to ask for some
> time now;
> Why is the FreeBSD install media using SU+J by default?
> While I understand Journaling reduces retrieve times, which makes it
> all seem more responsive. IMHO the consequences, and limitations it
> imposes, seem to outweigh the pro's.
> Just wondering...
>
> --Chris
>
> I'm aware of two issues with SU+J, one of which is annoying and the other
is worse.
1. If the journal is not fully written on power fail or some other reason,
you may need to do a full fsck of the volume and the behavior of the system
until this is done can be very unpredictable.
2. You can't safely snapshot the system. This is what 'dump -L' does. This
means that some files dumped from a live FS may not be consistent (not
good!) or, if '-L' is used, the system may well hang.

While I love the fast fsck times (2 or 3 seconds) after a crash, I also
question the default. Still, it may be a preferred choice be used for very
large file systems where a full fsck would take a very long time as long as
the risks are understood. For these systems, ZFS might be a better choice.
These arguments do NOT favor it being the default, IMHO.
--
R. Kevin Oberman, Network Engineer, Retired
E-mail: rkoberman at gmail.com


More information about the freebsd-stable mailing list