[Swift-devel] 0.93 site testing

[Michael Wilde]

A question about 0.93 testing in general: We agreed yesterday to stay with the current 0.93 branch.  Should we be testing 0.93 with the tests in 0.93 itself, or the tests in trunk, or both?

I think if we do this rigorously, the answer is that we test 0.93 with the 0.93 test suite, but that we integrate relevant test corrections and enhancements in both directions. The same of course applies to code, but in the case of the tests we may find that more cross-branch integration is needed, especially if we are going to do a lot of improvement on site tests in the next 2 weeks.

Im wondering if we can afford the 2-way integration process, or should take the shortcut of testing 0.93 with trunk tests?  Im in favor of the two-way approach, but to keep an eye on the process and reconsider if it becomes too costly.

- Mike


----- Original Message -----
> From: "Michael Wilde" <wilde at mcs.anl.gov>
> To: "David Kelly" <davidk at ci.uchicago.edu>, "Alberto Chavez" <alberto_chavez at live.com>
> Cc: swift-devel at ci.uchicago.edu
> Sent: Thursday, August 4, 2011 4:44:29 AM
> Subject: Re: [Swift-devel] 0.93 site testing
> David, Alberto,
> 
> The test list looks good. We can maybe shuffle the names assigned:
> Alberto on PADS and Fusion; Justin can maybe help on the BG/P's and
> Frankin; can add the Cray crow test system.
> 
> It might be good to add tests of the plain ssh provider, and then the
> ssh:local coaster configuration. (Jon is using this eg to run on PADS
> and Beagle from Globus Online).(Aside: I see very excessive logging
> from the ssh provider - lets investigate, I'll file a ticket)
> 
> I noticed that in several past incidents our site tests were fooled
> into thinking they passed, when in fact the actual application
> invocations took place in an environment different than intended. Some
> cases that come to mind:
> 
> - thinking that we were running on a cluster via coasters when in fact
> the apps ran on localhost. This was the incorrect PADS sites entry you
> mention below.
> 
> - thinking that we were running on Cray compute nodes when in fact the
> apps ran on the Cray PBS service node (on Beagle, again this was a
> login node)
> 
> - asking to run N apps per compute node (1 per core) when in fact we
> ran 1 app per node
> 
> - asking to run N apps per compute node when in fact we ran N^2 apps
> per node
> 
> In this next round of testing, can we enhance the tests (or add new
> ones) so that:
> 
> 1) part of the app execution records the node(s) it executes on and
> ensures that we are running on a compute node (We can do this in a
> site-independent fashion by adding a "compute node hostname pattern"
> to the siteTester script:
> https://trac.ci.uchicago.edu/swift/browser/trunk/tests/sitetester?desc=1
> and passing the name pattern to the test.
> 
> 2) the expected number of apps are running on the compute node (sleep;
> do ps; count the number of app shells running, and ensure that there
> are >1 and <= N)
> 
> - Mike
> 
> ----- Original Message -----
> > From: "David Kelly" <davidk at ci.uchicago.edu>
> > To: swift-devel at ci.uchicago.edu
> > Sent: Wednesday, August 3, 2011 8:57:07 PM
> > Subject: [Swift-devel] 0.93 site testing
> > Hello,
> >
> > I updated the swift devel website tonight with plans for 0.93 site
> > testing. I am starting with the same site tests that we performed
> > with
> > the last release. The page is at:
> >
> > https://sites.google.com/site/swiftdevel/site-specific-testing
> >
> > Feel free to edit that page as the tests get run. The tests are
> > located in swift/tests/providers, but they likely need tweaked. I'm
> > pretty sure the PADS template is incorrect, others may be as well so
> > it is probably worthwhile to double check everything.
> >
> > David
> > _______________________________________________
> > Swift-devel mailing list
> > Swift-devel at ci.uchicago.edu
> > https://lists.ci.uchicago.edu/cgi-bin/mailman/listinfo/swift-devel
> 
> --
> Michael Wilde
> Computation Institute, University of Chicago
> Mathematics and Computer Science Division
> Argonne National Laboratory
> 
> _______________________________________________
> Swift-devel mailing list
> Swift-devel at ci.uchicago.edu
> https://lists.ci.uchicago.edu/cgi-bin/mailman/listinfo/swift-devel

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