Xref: utzoo comp.unix.wizards:5897 comp.mail.misc:745 news.software.b:1044
Path: utzoo!mnetor!uunet!husc6!necntc!ames!elroy!mahendo!jplgodo!wlbr!scgvaxd!trwrb!desint!geoff
From: geoff@desint.UUCP (Geoff Kuenning)
Newsgroups: comp.unix.wizards,comp.mail.misc,news.software.b
Subject: Re: Another way that news eats inodes
Message-ID: <1659@desint.UUCP>
Date: 28 Dec 87 22:10:23 GMT
References: <440@minya.UUCP> <375@octopus.UUCP>
Reply-To: geoff@desint.UUCP (Geoff Kuenning)
Organization: Interrupt Technology Corp., Manhattan Beach, CA
Lines: 36
Summary: it's an old old bug

In article <375@octopus.UUCP> pete@octopus.UUCP (Pete Holzmann) writes:

> In article <440@minya.UUCP> jc@minya.UUCP (John Chambers) writes:
> <[Discussion of how rnews sometimes eats all inodes in a filesystem...]

> It started happening to me recently as well. I wasn't watching the system
> closely, so I didn't get to it until things were in real bad shape: 12000
> inodes get eaten up, and I'm getting complaints all over the console screen.
> I reboot, fsck tells me that the inode count is wrong in my /usr/spool
> partition (where news gets stored and unpacked), and we're back to normal
> for another few days to few weeks. I've never been around when things started
> going bad, so I've never found the culprit. Maybe this is it? Something is
> very strange, since I can't imagine rnews having a valid need for so many
> inodes no matter *what*! We don't get more than around 6-7000 articles in a
> whole week!

Absolutely FASCINATING.  I've suffered this problem for at least a year,
maybe two, and there's a possibility it goes back as much as five years.
But I blamed it on a hardware problem.  I run a special crontab script
that watches the news i-nodes and kills incoming news if they get too
low.  Then I hand-unmount and fsck the disk.  Since inews leaves
"No space on device" errors in the errlog, I can then use sendme's
to re-request the articles that were dropped before the incoming
news got killed.

An interesting point here is that the number of i-nodes remains too
small even after rnews exits.  Only fsck will correct the problem.
This leads me to wonder if maybe rnews is triggering a kernel bug or
deficiency of some sort.

Since I now have evidence it's a bug and not a hardware problem, I'll
modify the script to collect a little more data and see if I can
trap something.  It'll probably take a while, since the occurrence
rate varies from several times a week to once every few months.
-- 
	Geoff Kuenning   geoff@ITcorp.com   {uunet,trwrb}!desint!geoff