Eygene Ryabinkin wrote:
>>>Because my worker nodes are behind NAT and I am expiriencing hangs of
>>>globus-url-copy during the stage-in and stage-out of files between WN and RB:
>>>http://www.listserv.rl.ac.uk/cgi-bin/webadmin?A2=ind0605&L=lcg-rollout&D=1&I=3&O=D&P=39097
>>>It is likely due to the bug in the globus-io:
>>>http://goc.grid.sinica.edu.tw/gocwiki/gridftp_works_only_once_within_a_minute_or_so
>>>as was kindly pointed by Steve Traylen. So I will try to make it work -- I
>>>need it.
>>
>>Please make sure you do _not_ define GLOBUS_TCP_PORT_RANGE on your WN.
>>Then the abovementioned problem does not occur.
>>YAIM should take care of that for you...
>
> GLOBUS_TCP_PORT_RANGE is defined when my jobs are run on the WN:
> -----
> gatekeeper$ qstat -f | grep RANGE | head -2
> le-10481.1148709395,GLOBUS_TCP_PORT_RANGE=20000 25000,
> le-11137.1148709425,GLOBUS_TCP_PORT_RANGE=20000 25000,
> -----
> I do not do it manually per every task, so some LCG components are doing
> this for me. Not sure what are they, but I can see them. If
> GLOBUS_TCP_PORT_RANGE is unset then the problem does not show up. But
> I am really want this to be set, because I am only allowing this range
> of ports on the firewall that is taking care of worker nodes. Without this
> I should open all ports for the outgoing access: it is not that I want.
> Possible I will be able to live with it, but I prefer not to.
OK, I see the problem. Please apply this work-around on your WNs:
-------------------------------------------------------------------------
echo unsetenv GLOBUS_TCP_PORT_RANGE > /etc/profile.d/unset_port_range.csh
echo unset GLOBUS_TCP_PORT_RANGE > /etc/profile.d/unset_port_range.sh
-------------------------------------------------------------------------
That should fix it. We will let YAIM do that automatically.
|