[OpenAFS-devel] RE: du/find hang running in AFS space

[Touretsky, Gregory]

Hi Derrick,

first of all, it seems to be "once happens, once not" problem.

> Hi, > > we get a problem with Open AFS 1.2.2 running on RedHat 6.2 machine
(Linux > iwsl008 2.2.17-14smp #1 SMP Thu May 17 13:26:42 PDT 2001 i686
unknown). > > Both find and du stuck when run in AFS space. Any ideas why
this happens? I don't know if this happens in older > versions of OpenAFS. 

DB> What configuration parameters have you passed afsd? 
-stat 5000 -dcache 2400 -daemons 5 -volumes 128

DB> What does the directory structure you're walking look like (in broad
terms)? 
Nothing special. Set of files, directories and mount points, might be ~15
levels of sub-directories, with several thousands of files.
It hangs (when this happens) pretty fast, without going too deep.

DB> Does it work on non-SMP hardware (or with a non-SMP kernel)? 
I have no idea - we don't have non-SMP machines.

> time 803.510554, pid 1290: ProcessFS vp 0xd0a33238 old len 0x1000 new len
> 0x1000 
> time 803.510554, pid 1290: Access vp 0xd0a20f98 mode 0x40 len 0x800 
> time 803.510554, pid 1290: Access vp 0xd0a33238 mode 0x100 len 0x1000 
> time 803.510554, pid 1290: Open 0xd0a33238 flags 0x10800 
> time 803.510554, pid 1290: Open 0xd0a33238 flags 0xf423f 
DB> Does the output stop after this or was this just the last of the output
from this fstrace run? 
Well, it stops for this specific pid (find). Other commands that run in AFS
can proceed, generating more information. 
BTW, I left the machine with several stuck "find" processes, and found later
that AFS access completely hung - it took about an hour.

DB> It would also be interesting to know how many cache entries are
referenced. cmdebug (host) -long|grep "refcnt"|grep -v "refcnt 0"|wc -l is a
fine way to test this quickly. 
55


Gregory Touretsky
IDC Computing / Systems Engineering Group
Unix Server Platforms
gregory.touretsky@intel.com
> (+) 972-4-865-6377, Fax: 04-865-5999
iNET: 465-6377, M/S: IDC-1B