[OpenAFS-devel] Problem with CB_WhoAreYou

[Volker Lendecke]

--u3/rZRmxL6MmkK24
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
Content-Transfer-Encoding: quoted-printable

Hi!

In a highly loaded environment I get the error

	    ViceLog(0,("CB: WhoAreYou failed for %s:%d, error %d\n",=20
		       hoststr, ntohs(host->port), code));

in FileLog. The particular stress test to reproduce this reliably is to come
=66rom the following system as a client:

shadm:~ # rpm -qa|grep openafs
openafs-1.2.11-0
openafs-client-1.2.11-0
openafs-devel-1.2.11-0
openafs-server-1.2.11-0
shadm:~ # uname -a
Linux shadm 2.4.21-138-smp #1 SMP Fri Oct 31 00:51:31 UTC 2003 i686 unknown
shadm:~ # cat /etc/SuSE-release=20
SuSE SLES-8 (i386)
VERSION =3D 8.1
shadm:~ #=20

I precreate 200 processes under 200 different uid's with different tokens.=
=20

Then with a single signal I send them into a chdir to /afs/cell. Looking at=
 the
sniff shows that the client does not respond properly to the WhoAreYou call=
back
calls. I know this is an artificial workload, but the only way to reliably
reproduce the problem.

Looking at viced/host.c shows that it is not a fatal error if this call is =
not
answered properly. Replacing=20

	code =3D RXAFSCB_WhoAreYou(host->callback_rxcon, &interf);

with=20

	code =3D RXGEN_OPCODE;

in viced/host.c gives some more messages in FileLog, but my problem goes aw=
ay.
To me this seems like a client problem. Has anybody seen this yet? Will I
provoke severe problems if I make the above change to the file server?

Thanks,

Volker

--u3/rZRmxL6MmkK24
Content-Type: application/pgp-signature
Content-Disposition: inline

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.5 (GNU/Linux)

iD8DBQFCCQXRP/hMgvhmbo4RAir5AJ9rrfgmRNELtzNoSX4avAlszMDRIwCeINt/
Qn0VZNYqfDi5EAmzefV2KSk=
=MAmo
-----END PGP SIGNATURE-----

--u3/rZRmxL6MmkK24--