Date: Tue, 27 Jun 2000 17:38:11 -0700 (PDT) From: Matthew Jacob <mjacob@feral.com> To: Josef Karthauser <joe@pavilion.net> Cc: Akinori -Aki- MUSHA <knu@idaemons.org>, andrews@technologist.com, mckusick@FreeBSD.org, cvs-committers@FreeBSD.org, cvs-all@FreeBSD.org Subject: Re: cvs commit: src/sys/contrib/softupdates softdep.h ffs_softdep.c Message-ID: <Pine.BSF.4.05.10006271733510.25569-100000@semuta.feral.com> In-Reply-To: <20000628012956.B19558@pavilion.net>
next in thread | previous in thread | raw e-mail | index | archive | help
There were in fact, more issues at one point with readahead in UFS than in Softupdates code under SMP (I had it sysctl'd off for 3.3). As far as I know, this problem went at last 6 months ago- I think this was a Matt Dillon effort. Your comments are a bit unfair. *You* haven't been running the currently released product with Softupdates. You have excellent reasons to be concerned because of your previous experience, but your statement of "A number of times this year I've lost whole filesystems on an SMP..." is misleading at best. Your reference to a friend should be made clearer. I'm not trying to defend SOFTUPDATES/SMP or anything (which work fine for me, btw, on most of my filesystems) but I believe that your statements could be a bit more clearer to try and assist all of us to see whether there is a problem and how to fix it. On Wed, 28 Jun 2000, Josef Karthauser wrote: > On Tue, Jun 27, 2000 at 09:29:24AM -0700, Matthew Jacob wrote: > > > > > > I'm not sure.... I have a feeling that there are softupdate problems > > > running under SMP. A number of times this year I've lost whole filesystems > > > on an SMP machines. :( > > > > Really? Have you told Kirk? > > > > No because I've not had a chance to test it - it's on a production > webserver. It's running 3.4 currently, and I've had a couple of > crashes over the last year. The first one produced a lot of > 'SOFTUPDATE INCONSISTANCY' errors and lost a lot of files - we had > to recover from tape and diff to see which files had been lost. > (This was on a 16gb ccd containing customer web data). The second > time was more recently. The machine crashed and in the process > the / partition disappeared, so badly that the kernel wasn't > available to boot from! We brought it up on another drive and did > a fsck and all the files turned up in /lost+found. Not a single > file was left in the file system. Fortunately we had a fresh level > 0 dump from that evening and so I did a 'newfs' and 'restore' and > the machine's been running since. > > More recently a friend of mine was having crashes doing a make > world under 4.0 with SMP and softupdate. Switching softupdates > off caused the crashes to go away. > > The thing that worries me is that softupdates is supposed to render > the filesystem safe so that if a crash occurs we only loose recent > data. In both my cases we lost data that hadn't been touched for > years - i.e. ports tree data that hasn't ever been touched. It > definitely wasn't in the buffer cache before the crash. ! > > Is anyone else here using softupdates in an SMP environment? > > I'm not saying that softupdates doesnt work in SMP, but I'm worried > that if softupdates doesn't leave things in as consistant a state > as you'd expect in an SMP environment. > > Joe > To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe cvs-all" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?Pine.BSF.4.05.10006271733510.25569-100000>