SMP: Background fsck deadlocks CURRENT and RELENG_5 solid

[Doug White]

On Fri, 20 Aug 2004, Martin Blapp wrote:

> > Perhaps you're short on disk space on / and the deltas from the snapshot
> > take you over the edge?  Or have some sort of large I/O operation run just
> > after boot?
>
> Hmm of course yes. But only about 10 seconds on one machine. On the other
> the workload just begins after startup. And, I start two kernel threaded
> programms at the same time.
>
> If you like to have a similiar setup you have to install:
>
> clamd
> mimedefang
> sendmail (from ports)
> dnsrbld
> qpopper
> apache
>
> I think that's enough and will lead to a crash.

How big is the mail queue?

There's some bugs in the snapshot code versus the background update, and
touching a file that gets adjusted later will explode like this.  Also
just plain massive changes will do it -- a while back it was cvsupping
ports while bgfsck was running.

You migth want to set your spool volume nobgfsck.

-- 
Doug White                    |  FreeBSD: The Power to Serve
dwhite at gumbysoft.com          |  www.FreeBSD.org