On Mon, 6 Jan 2003 kpneal@pobox.com wrote:

> On Sun, Jan 05, 2003 at 01:07:24PM -0600, Frederick Bruckman wrote:
> > On the question of whether or not softdeps should be still be
> > considered experimental, I was engaged in a discussion in November in
> > news:comp.unix.bsd.netbsd.misc under the subject "[NetBSD 1.6] current
> > status of softdep". You might like to browse Google groups for that.
> > To sum up, it's not experimental, most users do have it enabled, but
> > it can't quite live up to it's guarantee of "never any file system
> > corruption" unless you turn off the drive's write cache, which kills
> > performance on ATA drives (yet the situation with write cache and
> > softdeps isn't particularly worse than with write cache and no
> > softdeps).
>
> Does this mean the problems with the system crashing when too many
> files are removed has been fixed?
>
> Want fun? On a machine with 64 megs of ram running 1.5 remove part
> of a news archive (with, say, 10,000 files or so). The rm -rf runs
> for a while and then BLAMMO it reboots. On the way back up fsck
> fills up lost+found and then complains a lot. It took my 233 MHz
> Alpha around an hour to finish an fsck.
>
> Has this been fixed? My personal opinion (FWIW) is that a softdep
> that has this bug is an experimental softdep.

Sure, I was bitten by that one myself. I didn't run into any such
problems with 1.6A-K, so I was enboldened to turn it back on on the
1.6 box (now 1.6_STABLE), and since then I've moved file systems
around and populated a new disk with no such crashes. A search turns
up about half a dozen PRs with "softdep" in the name, but they're all
against netbsd-1-4/netbsd-1-5, except for one against 1.6, requesting
that softdeps be turned on by default. But don't take my word for it
-- try it for yourself!

Frederick