Tomáš Kouba wrote:
> Hi,
>
> I have a problem with job running too long. I have checked what part of
> job is taking so long on WN and found a wrapper script:
> /home/auger002/globus-tmp.skurut30.26711.0/globus-tmp.skurut30.26711.0/local/md5/9d/eaca21b7c538ea92de922d19368d1d/md5/c7/fadc3889ecb54d5082ab45c8dede98/data
>
>
> The script is created on our glite-WMS from template file
> /opt/glite/etc/templates/template.sh and it is stuck while
> executing sleep in function globus_url_retry_copy()
> called from function doExit():
>
> globus_url_retry_copy "file://${workdir}/${maradona}"
> "${__maradonaprotocol}"
>
> The function globus_url_retry_copy() tries a globus_url_copy in a loop
> until it succeeds or the count of retries is too high.
>
> In my case the globus_url_copy failes because the destination
> file is
> __maradonaprotocol="gsiftp:///var/glite/SandboxDir/yQ/https_3a_2f_2fskurut1.cesnet.cz_3a9000_2fyQv3D32IIfbQxTkKBlZ3hg/Maradona.output"
>
>
> Which is very strange to me, because it lacks the <hostname> part.
> The missing <hostname> is also the cause of globus_url_copy failure.
> But I cannot find where this __maradonaprotocol is set so I am not able
> to find out why this is set incorrectly.
Probably this bug:
https://savannah.cern.ch/bugs/index.php?20667
On WMS skurut1.cesnet.cz the HOSTNAME environment variable probably is
not set for daemons. See the bug for work-arounds.
|