Date: Thu, 23 Jul 1998 23:26:49 -0500 From: Karl Denninger <karl@mcs.net> To: Gary Palmer <gpalmer@FreeBSD.ORG> Cc: "Daniel O'Connor" <doconnor@gsoft.com.au>, Adrian Penisoara <ady@warpnet.ro>, Gerald Ehritz <ehritz@dbai.tuwien.ac.at>, freebsd-current@FreeBSD.ORG Subject: Re: pine 3.96 locks 3.0-980621 completely Message-ID: <19980723232649.30224@mcs.net> In-Reply-To: <17356.901252917@gjp.erols.com>; from Gary Palmer on Fri, Jul 24, 1998 at 12:01:57AM -0400 References: <199807240143.LAA02750@cain.gsoft.com.au> <17356.901252917@gjp.erols.com>
next in thread | previous in thread | raw e-mail | index | archive | help
On Fri, Jul 24, 1998 at 12:01:57AM -0400, Gary Palmer wrote: > "Daniel O'Connor" wrote in message ID > <199807240143.LAA02750@cain.gsoft.com.au>: > > Err.. well isn't it kind of bad that the machine crashes? :) > > Yep > > > I mean it _pine_ crashed, then OK, upgrading would be fine, but when the > > machine crashes, maybe he should upgrade his version of -current.. > > It won't change the fact that it is a fundamentally bad idea to rely on NFS > locking for mail delivery/reading when we don't impliment said locking in the > NFS layer... > > I know I seem like I'm trying to cover up a bug in freebsd by saying "well, > don't do that", but I think anyone who knows anything about e-mail and NFS > would agree that it is a fundamentally *bad* idea. > > (Yes, I know of at least one *large* (like >400,000 users) ISP in the > US that uses NFS for mail & news stores. Doesn't mean I don't think they're > stupid :) ) There ARE safe ways to handle email via NFS. We do it here, and other than through abject user stupidity (ie: opening email via both POP3 and Pine at once) or programmatic bugs (ELM has known problems with the folder resync command) it doesn't lose messages or corrupt mailboxes. I get over 1200 emails a DAY. If it wasn't solid and stable you can bet I'd know about it within minutes around here. I will say this - FreeBSD has had problems in the past with NFS which made it impossible to get this to work reliably (primarily during the time period from roughly December '97 to April '98). For this reason we ran pretty old kernels around here for quite some time. Some of the semantics that we rely on to make sure that mail doesn't get smashed do have a couple of potential race conditions under extremely odd circumstances that could lead to trouble - but if we get bit by those, we have far more serious trouble afoot than a lost email message. HOWEVER, "mount and go" DOES NOT WORK SAFELY. In fact, "mount and go" won't work safely even WITH NFS locking on 90% of the machines out there, including Suns. I've yet to see a Sun implementation of rpc.lockd that can't be freaked out under less-than-pleasant circumstances; the usual symptom of such a freak out is a wedged lock that doesn't release, and thus wedged MTAs and MUAa. If you get nailed by that you're going to be rebooting things - not a pleasant thought on production machines. Now NFS for a news store is beyond insane. I don't know why you'd EVER do that on the news server. For *client* access its perfectly fine; client access is read-only, which is completely safe (there is only one *possible* writer). We do that here; the news spool is mounted *read only* on the cluster machines. Never had a single problem with that (nor would I expect to). However, you have to be insane not to have the spool disks LOCAL on the news *server* system. -- -- Karl Denninger (karl@MCS.Net)| MCSNet - Serving Chicagoland and Wisconsin http://www.mcs.net/ | T1's from $600 monthly / All Lines K56Flex/DOV | NEW! Corporate ISDN Prices dropped by up to 50%! Voice: [+1 312 803-MCS1 x219]| EXCLUSIVE NEW FEATURE ON ALL PERSONAL ACCOUNTS Fax: [+1 312 803-4929] | *SPAMBLOCK* Technology now included at no cost To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-current" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?19980723232649.30224>