From the email sent to the wrong account:
Don't forget that running the nightly tests successfully is necessary but not sufficient - there are some important tests that are not run by the nightlies, some because they haven't been setup to have automatic summaries, others because the nightlies only run on a single node.
We are planning on the special tests too. Are there more?
Yes. See the release wiki page http://wiki.mcs.anl.gov/mpich2/index.php/Creating_a_Release . *EVERYTHING* on this page is there because one time we didn't do it, and suffered as a result. All of the 7 items need to be checked. (As noted on the wiki page, we had a more comprehensive and valuable list, with more of those don't-forget-to-do-this-again items; what's here is the absolute minimum for anything that goes to the outside). In addition, some tests need to be run on more than one node (consider that part of #6) - the only such test is a partial test in the special tests, and it has been having some failures for quite some time now. Bill William Gropp Director, Parallel Computing Institute Deputy Director for Research Institute for Advanced Computing Applications and Technologies Paul and Cynthia Saylor Professor of Computer Science University of Illinois Urbana-Champaign
I've modified the LLNL MPI I/O test to integrate into the MPICH2 test suite, so it can be tested every night. Unfortunately, it looks like a part of the test does not pass correctly. I've created a ticket for it. Everything else on the wiki page seems to pass correctly (the failing tests are known issues for which tickets exist). I haven't tested MPE -- Anthony, can you check this? If it looks OK, Dave can cut the 1.5rc2 release on Monday. -- Pavan On 09/27/2012 12:43 PM, William Gropp wrote:
From the email sent to the wrong account:
Don't forget that running the nightly tests successfully is necessary but not sufficient - there are some important tests that are not run by the nightlies, some because they haven't been setup to have automatic summaries, others because the nightlies only run on a single node.
We are planning on the special tests too. Are there more?
Yes. See the release wiki page http://wiki.mcs.anl.gov/mpich2/index.php/Creating_a_Release . *EVERYTHING* on this page is there because one time we didn't do it, and suffered as a result. All of the 7 items need to be checked. (As noted on the wiki page, we had a more comprehensive and valuable list, with more of those don't-forget-to-do-this-again items; what's here is the absolute minimum for anything that goes to the outside).
In addition, some tests need to be run on more than one node (consider that part of #6) - the only such test is a partial test in the special tests, and it has been having some failures for quite some time now.
Bill
William Gropp Director, Parallel Computing Institute Deputy Director for Research Institute for Advanced Computing Applications and Technologies Paul and Cynthia Saylor Professor of Computer Science University of Illinois Urbana-Champaign
-- Pavan Balaji http://www.mcs.anl.gov/~balaji
I believe the LLNL MPI I/O test itself has some problems. Rob Latham can confirm. Rajeev On Sep 29, 2012, at 12:03 AM, Pavan Balaji wrote:
I've modified the LLNL MPI I/O test to integrate into the MPICH2 test suite, so it can be tested every night. Unfortunately, it looks like a part of the test does not pass correctly. I've created a ticket for it.
Everything else on the wiki page seems to pass correctly (the failing tests are known issues for which tickets exist).
I haven't tested MPE -- Anthony, can you check this?
If it looks OK, Dave can cut the 1.5rc2 release on Monday.
-- Pavan
On 09/27/2012 12:43 PM, William Gropp wrote:
From the email sent to the wrong account:
Don't forget that running the nightly tests successfully is necessary but not sufficient - there are some important tests that are not run by the nightlies, some because they haven't been setup to have automatic summaries, others because the nightlies only run on a single node.
We are planning on the special tests too. Are there more?
Yes. See the release wiki page http://wiki.mcs.anl.gov/mpich2/index.php/Creating_a_Release . *EVERYTHING* on this page is there because one time we didn't do it, and suffered as a result. All of the 7 items need to be checked. (As noted on the wiki page, we had a more comprehensive and valuable list, with more of those don't-forget-to-do-this-again items; what's here is the absolute minimum for anything that goes to the outside).
In addition, some tests need to be run on more than one node (consider that part of #6) - the only such test is a partial test in the special tests, and it has been having some failures for quite some time now.
Bill
William Gropp Director, Parallel Computing Institute Deputy Director for Research Institute for Advanced Computing Applications and Technologies Paul and Cynthia Saylor Professor of Computer Science University of Illinois Urbana-Champaign
-- Pavan Balaji http://www.mcs.anl.gov/~balaji
participants (3)
-
Pavan Balaji -
Rajeev Thakur -
William Gropp