[Swift-user] SWIFT Generates "incompatible" PBS scripts

Michael Wilde wilde at anl.gov
Wed Apr 2 18:25:17 CDT 2014


I've also run into the issue of PBS ignoring all directives after the 
first non-directive line in the file. I think that we either already 
have a ticket on this or need one; its indeed a Swift bug as far as I 
can tell.

Thanks for reporting it; as Mihael says, we'll create a fix and report 
back to the swift-user list.

- Mike

On 4/2/14, 6:21 PM, Mihael Hategan wrote:
> Hi,
>
> Yes, I suspect that the problem is that the script, as written
> originally by PBSExecutor is not sh-safe, but relies on some specific
> version of bash. Out of curiosity, on the machine this doesn't work on,
> can you let us know what the versions of /bin/sh and /bin/bash.
>
> Your workaround seems ok, in that I don't think it would cause problems
> unless you want to use the PBS provider directly and pass environment
> variables to jobs. In any event, we should fix this.
>
> Mihael
>
> On Thu, 2014-04-03 at 06:57 +0800, chen sui wrote:
>> Hello SWIFT,
>> I found a solution to the problem. It seems to be working for me in this specific situation so far.
>> The solution is to comment out lines 242 through 253 in file "cog/modules/provider-localscheduler/src/org/globus/cog/abstraction/impl/scheduler/pbs/PBSExecutor.java" and recompile the JAR file "cog-provider-localscheduler-0.4.jar".
>> After this change, PBSExecutor would no longer insert the starred lines and it now works on the cluster I'm using
>> Hope this may be useful for people who encounter the same issue.
>> Thanks!Tommy Chen
>>
>> From: baiypwup at hotmail.com
>> To: swift-user at ci.uchicago.edu
>> Subject: SWIFT Generates "incompatible" PBS scripts
>> Date: Thu, 3 Apr 2014 01:28:39 +0800
>>
>>
>>
>>
>> Hello SWIFT,I have been using SWIFT on two different clusters, it has always been running flawlessly on one of the clusters, but it fails on the second one.
>>
>> I think I have narrowed down the cause of the problem;
>>
>> When I have these lines (the 2 lines with stars in the following example) in a generated PBS script,
>> [begin example]
>> #PBS -S /bin/bash
>> #PBS -N B0402-0012360-0
>> #PBS -m n
>> (omitted)
>> export WORKER_LOGGING_LEVEL=NONE   (*)
>> #PBS -v WORKER_LOGGING_LEVEL  (*)
>> [end example]
>>
>> The last two lines is causing the PBS to be incompatible with one of the clusters I'm using. When I submit such a script, I get the following error.
>>
>> [begin Console output]
>> [cs900601 at eric2 scripts]$ qsub PBS4684660157119642383.submit
>> 506530.eric2
>> ** Job deleted, insufficient hours (0 < 96) left in allocation null **
>> [end Console output]
>>
>> Clearly, for some reason the qsub on this cluster has "forgotten" every "#PBS" directive prior to the "export" line.
>>
>> The reason seems to be
>>     1) the cluster I'm using does not support the "-v" switch.
>>     2) Does not accept PBS script whose "#PBS" lines have been "interrupted" by an non PBS directive (the "export" above).
>>
>> That means, if I am able to make Swift remove the 2 lines during script generation the problem would go away.
>>
>> So my question is, is it possible to do this through Swift configurations?
>> I have searched the configuration files but not seem to find a way to do it. The generated script says in the 1st line it's generated by a Java class "by class: class org.globus.cog.abstraction.impl.scheduler.pbs.PBSExecutor".
>>
>> Is there a way to check out the source of that "PBSExecutor" class to see what is adding these two lines with asterisks in the generated script and disable it?
>>
>> Thanks!Tommy Chen
>>   		 	   		   		 	   		
>> _______________________________________________
>> Swift-user mailing list
>> Swift-user at ci.uchicago.edu
>> https://lists.ci.uchicago.edu/cgi-bin/mailman/listinfo/swift-user
>
> _______________________________________________
> Swift-user mailing list
> Swift-user at ci.uchicago.edu
> https://lists.ci.uchicago.edu/cgi-bin/mailman/listinfo/swift-user

-- 
Michael Wilde
Mathematics and Computer Science          Computation Institute
Argonne National Laboratory               The University of Chicago




More information about the Swift-user mailing list