How bad is the linux bug, was (Fwd) arla 0.34.2

[Dr A V Le Blanc]

On Sun, Aug 20, 2000 at 06:31:05PM +0200, Magnus Ahltorp wrote:
> The linux bug fixed between 0.34.1 and 0.34.2 is a directory
> consistency bug. Under certain circumstances, the directory is not
> invalidated correctly. It is not very serious, but it can be annoying,
> since files may appear when they are not there, and the other way
> around.

So this will not fix the bug which I see on very busy clients, in
which a read of one file (which does exist) returns the contents
of a completely different file (which, I presume, also exists)?
To fix this I can sometimes just do

     /etc/init.d/arla stop
     /etc/init.d/arla start

But sometimes even that does not work, and I need to do

     /etc/init.d/arla stop
     rm -rf /var/cache/arla/*
     /etc/init.d/arla start

The problem with this bug is being able to reproduce it reliably.
I do get lots of the messages which say, for example,

     kernel: XFS Panic: xfs_message_installdata didn't find node!

I also get quite a few of these:

     Aug 18 09:37:08 kernel: XFS SoftAssert: existing inode (cb421540,
	fid 166.1835387420.10066.5549) != installing earthsci.html(cb5a9870,
	fid 166.1835387420.10066.5946)

(You'll note that this is a web server.)

     -- Owen
     LeBlanc at mcc.ac.uk