From owner-freebsd-hackers Thu Dec 7 18:18:19 1995 Return-Path: owner-hackers Received: (from root@localhost) by freefall.freebsd.org (8.7.3/8.7.3) id SAA15921 for hackers-outgoing; Thu, 7 Dec 1995 18:18:19 -0800 (PST) Received: from brasil.moneng.mei.com (brasil.moneng.mei.com [151.186.20.4]) by freefall.freebsd.org (8.7.3/8.7.3) with ESMTP id SAA15909 for ; Thu, 7 Dec 1995 18:18:10 -0800 (PST) Received: (from jgreco@localhost) by brasil.moneng.mei.com (8.7.Beta.1/8.7.Beta.1) id TAA11705; Thu, 7 Dec 1995 19:54:08 -0600 From: Joe Greco Message-Id: <199512080154.TAA11705@brasil.moneng.mei.com> Subject: Re: Any operational code using mmap() on FreeBSD ? To: davidg@root.com Date: Thu, 7 Dec 1995 19:54:07 -0600 (CST) Cc: hackers@freebsd.ORG In-Reply-To: <199512072217.OAA06221@corbin.Root.COM> from "David Greenman" at Dec 7, 95 02:17:20 pm X-Mailer: ELM [version 2.4 PL24] Content-Type: text Sender: owner-hackers@freebsd.ORG Precedence: bulk > >The question is what are the failure modes on 2.1. If there aren't any, > >you should upgrade to resolve the problem. If there are, and they are > >more severe, you should report them to be corrected, since 2.1 *is* > >being maintained (in the form of the 2.2-current). > > Uh, actually, 2.1 is being maintained seperately from 2.2-current. I have a "seat of my pants" feeling that there are still minor problems - i.e. I have a bug that is biting me in the butt. Symptom: INN news server compiled to access active file via mmap(). Every once in a while, an update to the active file apparently misses (doesn't get flushed to disk maybe?) and I end up with a symlinking error caused when INN goes to create a symbolic link to what SHOULD be the next free article number, and instead discovers that there is already a file with that number. I've checked the logs and files religiously each time this happens, all signs say that INN processed the article for which the active file increment does not seem to happen. Frequency: uptime is currently 8 days. It has happened twice in the last three days. I started testing mmap'ed accesses after discussing the issue with Mr. Dyson and I set up the system to catch any panics. Under 2.0.5R I saw the SAME behaviour but it was often followed very quickly by a crash and reboot. Two samples under 2.1.0R is not enough to rule out that this might still happen, of course, so I will continue to run the system until it either panics or something else interesting happens. Generally (under 2.0.5R) the system would run fine for several days after a reboot but would start going into shock around the 7-10 day mark, and I bet it never made it two weeks without a reboot. When I turned off mmap, operations became smooth as silk. (I just wanted to inject that there is someone else looking at this problem as well, and provide a status report..) ... Joe ------------------------------------------------------------------------------- Joe Greco - Systems Administrator jgreco@ns.sol.net Solaria Public Access UNIX - Milwaukee, WI 414/342-4847