[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Condor-users] Wrong hostname in Condor-G file staging



On Jan 30, 2008, at 2:22 AM, Jan Ploski wrote:

condor-users-bounces@xxxxxxxxxxx schrieb am 01/29/2008 10:50:38 PM:

On Jan 25, 2008, at 6:47 AM, Jan Ploski wrote:

condor-users-bounces@xxxxxxxxxxx schrieb am 01/24/2008 11:12:19 PM:

Hello,

Condor-G (from a 6.7.18 installation) is misbehaving on a machine
which
has two IP addresses and two different hostnames. The internal
hostname
(not mapped in the global DNS) keeps appearing in the
sourceUrl/destinationUrl fields of the SOAP message used to request
file

staging from WS-GRAM, causing the job to be held with the "Globus
error:

Staging error for RSL element fileStageIn" message.

How can we make Condor-G use the external hostname instead of the
internal one? Am I right in assuming that the wrong hostname
resolution
is performed somewhere in the Condor implementation and not in the
GAHP/Globus Java components (i.e., do they get the RSL with the
resolved

hostname from condor_schedd or how does it work)?

The problem is solved, so I will answer myself for the archive:

The hostname resolution seems to happen in the condor_gridmanager
process,
spawned by condor_schedd. The GAHP server which interfaces with the
Condor-shipped Globus components is in turn spawned by the
condor_gridmanager. GAHP most likely receives the (correctly or
incorrectly) resolved hostname from condor_gridmanager.

The submission host's IP address set on its "external world" network
interface was being resolved via DNS to an internal hostname. To
prevent
this, the /etc/hosts file was edited so that this IP address is
resolved
to the external hostname. I'm still puzzled a bit about the workings
of
the hostname resolution mechanism, but the SOAP messages now contain
the
correct hostname.


A correction for the archives:
The globus-gridftp-server launched by the gridmanager to handle the
file transfers picks the hostname to use.

Are you sure? I was having a wrong hostname in the RFT SOAP message
exchange, which I believe happens before the GridFTP server becomes
involved? Do you mean these SOAP messages are constructed by the
globus-gridftp-server?


Condor starts a globus-gridftp-server as the user to handle the client side of the file transfers. When the gridftp server starts up, it prints out its hostname and port. That is the hostname that the gridmanager uses in the RFT requests inside the GRAM job description.

+--------------------------------+-----------------------------------+
|           Jaime Frey           | I used to be a heavy gambler.     |
|       jfrey@xxxxxxxxxxx        | But now I just make mental bets.  |
| http://www.cs.wisc.edu/~jfrey/ | That's how I lost my mind.        |
+--------------------------------+-----------------------------------+