[OpenAFS] 1.3.70 comments?

Rainer Többicke rtb@mail.cern.nospam.ch
18 Aug 2004 10:15:16 +0200


On Tue, 2004-08-17 at 16:23, Kris Van Hees wrote:

> > . fstrace hangs the machine almost immediately after 'fstrace setset cm
> > -act'. This is on an SMP i386 Linux 2.4.21-15.0.3.EL. It did the same
> > BTW on everything after Openafs 1.2.6 that I tested.
> 
> Does the hang happen after a particular operation happening on the client, or
> just regardless of what is being done on it?  We solved one occurence where
> an 'fs flush' on a file could cause a hang when fstrace was enabled, and I am
> fairly sure that the patch for that made it into 1.3.70.  This might be another

The machine was running a test script creating 2GB files, no 'fs flush'
intended. After 'fstrace ... -act' it hung before I managed to type the
'p' in 'fstrace dump' - and I'm not a particularly slow typist.  


> (similar) case.  Since it has been hanging like this for you on everything you
> tested since 1.2.6, is there an open ticket in the OpenAFS RT for it that I
> can check out for more details?
> 

No sorry, no ticket: I ran into this a few weeks ago on 1.2.10 and then
quickly checked a few other machines running anything from 1.2.9 -
1.2.11. I'm pretty sure things were fine on 1.2.6, but I remember having
crashed a node under 1.2.7.

No problems on single processor machines, though. This is an SMP
problem.


BTW: I'm saying "hang" here because this is normally what it looks like
with the machine on our serial console server. But the first case I had
(1.2.10) was on a VGA screen and actually caused a kernel oops in
afs_icl_Event4 or so (lost my scribbles of the traceback). Tried to
check the code but the only conclusion was that the pointers must have
been mangled.

I just set up a kdb on a test system - as soon as the system finishes
the current test I'll try again and hopefully will get some info on the
serial console. However, my impression is that anybody can reproduce
this easily (with an SMP system).

-- 
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
Rainer Toebbicke 
European Laboratory for Particle Physics(CERN) - Geneva, Switzerland  
Phone: +41 22 767 8985       Fax: +41 22 767 7155