[Swift-devel] Re: sites.xml settings for Beagle in Swift trunk?

[ketan]

Mihael,

For PBS provider, Justin made enhancement through which multiple 
provider attributes could be specified in sites.xml, separated by 
semicolon using the "providerAttribute" key as follows:

<profile namespace="globus" key="providerAttributes">
pbs.aprun
pbs.mpp=true
</profile>

While the older ppn key parsing was modified to parse just the integer 
instead of the colon separated string of previous version:

<profile namespace="globus" key="ppn">24:cray:pack</profile>

I missed to notice this above change in trunk and retained the above 
line in my sites.xml. This was causing the sites.xml parsing to fail and 
hence nothing was written on the ~/.globus/script/PBS*.submit file.

Still, this empty submit file was getting submitted which made Swift 
think that jobs are submitted and so it was emitting a "nnn jobs 
submitted" status.

Only to find that nothing was submitted.

The problem disappeared simply by removing the above colon separated line.


I would suggest, we continue to support both versions for PBS provider 
until some later version, say 0.95 before phasing into the new setup.


Ketan



On 5/6/11 2:16 PM, Mihael Hategan wrote:
> It may be useful to know what the exact problem was.
>
> Mihael
>
> On Fri, 2011-05-06 at 11:50 -0500, ketan wrote:
>> On 5/6/11 11:35 AM, Michael Wilde wrote:
>>> Ketan, Justin,
>>>
>>> Did you work out yesterday the right sites.xml settings for Coasters-PBS-Cray-aprun?
>>>
>>> I think this page was the start of your documentation for those, Justin:
>>>     https://sites.google.com/site/swiftdevel/internals/providers/coasters-provider
>>>
>>> Ketan, can you move these to the new cookbook, and work out any issues to make sure they are correct?
>>>
>>> Did you both determine that the trunk code is working OK, or that it has bugs that still prevent it from running correctly on Beagle?
>> I tested yesterday with the trunk code on beagle with new coaster-site
>> attributes and this bug is not in anymore.
>> However, testing the trunk code on a large run is still not concluded. I
>> will report as soon as a big run gets underway on beagle.
>>> Thanks,
>>>
>>> - Mike
>> _______________________________________________
>> Swift-devel mailing list
>> Swift-devel at ci.uchicago.edu
>> http://mail.ci.uchicago.edu/mailman/listinfo/swift-devel
>