[Swift-devel] OSG hang

[Jonathan Monette]

Ok.  I ran globus-job-run tuscany.med.harvard.edu /bin/sh -c '/bin/ps -g engage'  from communicado before the swift script and got this:
PID TTY          TIME CMD
  341 ?        00:00:00 globus-job-mana
  447 ?        00:00:00 ps
13407 ?        00:00:00 globus-job-mana
14200 ?        00:00:00 perl
14202 ?        00:00:03 perl
17440 ?        00:00:00 condor_shadow

I then ran the swift script from the VM and ran the globus-job-run command from above and got:
  PID TTY          TIME CMD
 9145 ?        00:00:00 globus-gridftp-
 9493 ?        00:00:00 globus-job-mana
 9588 ?        00:00:00 bash
 9602 ?        00:00:00 sleep
11734 ?        00:00:00 globus-job-mana
11828 ?        00:00:00 ps
13407 ?        00:00:00 globus-job-mana
14200 ?        00:00:00 perl
14202 ?        00:00:03 perl
17440 ?        00:00:00 condor_shadow


So it does seem like the job get run on the head node.  When the job is finished on the head node(I checked several times with globus-job-run) Swift continues to execute(well it just prints the Submitted: 1 line over and over again).

So the next step is to set those environment variables Mike mentioned and try again?  

On Jul 14, 2011, at 1:32 PM, Mihael Hategan wrote:

> On Thu, 2011-07-14 at 13:28 -0500, Jonathan Monette wrote:
>> What is the distinction between testing swift manually from the VM and then test the Swift script?
> 
> I think Mike means that you should do a simple sleep script that you
> would run by logging into the vm and typing "swift sleep.swift" on the
> command line vs. whatever automated system you may have to launch your
> real script.
> 
> More important, however, are the other steps.
> 
>> 
>> On Jul 14, 2011, at 1:25 PM, Mihael Hategan wrote:
>> 
>>> On Thu, 2011-07-14 at 13:22 -0500, Michael Wilde wrote:
>>>> Try this:
>>>> 
>>>> 
>>>> export GLOBUS_HOSTNAME=externalIPAddr # set it to the 50.x.y.z addr
>>>> export GLOBUS_TCP_PORT_RANGE=50000,51000
>>>> export GLOBUS_TCP_SOURCE_RANGE=50000,51000
>>>> 
>>>> 
>>>> Then test both globus-url-copy and globusrun from swift/bin (you will
>>>> need simple RSL for globusrun)
>>>> 
>>>> 
>>>> Then test Swift manually from the VM
>>>> 
>>>> 
>>>> Then test with a Karajan script
>>>> 
>>>> 
>>>> Then test the Swift script
>>>> 
>>>> 
>>>> Use a single sleep job, eg sleep 1234, so you can spot it running on
>>>> the head node.
>>>> 
>>>> 
>>>> Mihael, can you help walk Jon through the above steps or suggest
>>>> another approach?
>>> 
>>> I will. The steps you mention are what I would try.
>>> 
>>> 
>>> 
>> 
> 
>