frequent cache corruption with arla 0.21 on linux 2.2.1

[Neulinger, Nathan R.]

> > Here are a few examples:
> > 	1) Created a file, and it immediately shows up, but it 
> reports "no
> > such file or directory" when doing ls. (Same as if you mkm 
> a non-existent
> > volume.) Problem is, the file can't be removed or recreated. 
> 
> This normally works. Could you give a more thorough explanation of
> what you do when this happens? 

Not sure... I'll try and pay a bit more attention next time. But basically,
I create a file in some fashion, and then right afterwards, the file is
there, but innaccessible. 

I think around that time, arlad often goes a bit haywire. I can't flushv or
anything. (Does arla support flushv/checkv?) I'm not positive about this
though.

> > 	2) sigpending lied - I am getting so many of these that it is
> > becoming a performance problem. The login time for an AFS account is
> > significantly different when these are occurring.
> 
> Are they always occuring? I have never seen these.

TONS of them. I _always_ get one of them when starting arlad, and then
usually get them in bursts whenever doing afs accesses - logins or file
copies, etc.

Feb  4 10:14:35 infinity telnetd[624]: connect from falcon.cc.umr.edu
Feb  4 10:14:37 infinity login: nneul login on ttyp1 from falcon.cc.umr.edu
Feb  4 10:14:37 infinity kernel: SIG: sigpending lied
Feb  4 10:14:38 infinity last message repeated 68 times
Feb  4 10:14:46 infinity telnetd[656]: connect from falcon.cc.umr.edu
Feb  4 10:14:46 infinity login: ROOT LOGIN ttyp1 FROM falcon.cc.umr.edu
Feb  4 10:14:46 infinity kernel: SIG: sigpending lied
Feb  4 10:14:47 infinity last message repeated 48 times
Feb  4 10:14:53 infinity telnetd[678]: connect from falcon.cc.umr.edu
Feb  4 10:14:54 infinity login: unixdb login on ttyp1 from falcon.cc.umr.edu
Feb  4 10:14:54 infinity kernel: SIG: sigpending lied
Feb  4 10:14:56 infinity last message repeated 53 times


> > 	4) Complete lockups and oops's: In some cases, the 
> lockups appear to
> > be an extreme case of sigpending lied. In others, they are oopses.
> 
> Here is a patch that might work:
> 

Will apply and let you know.

You don't perhaps have anon-cvs set up do you? Or is your arla cvs
repository in AFS (i.e. can it be mirrored)

-- Nathan