[Q-e-developers] [Pw_forum]

[Filippo Spiga]

On Aug 17, 2015, at 2:14 PM, Samuel Poncé <samuel.pon at gmail.com> wrote:
> So far I have 3 physical slaves (http://epw.org.uk/Main/Slaves) but would like to add some more to test the different architecture. 
> 
> The first slave contains the buildbot master that send the request to the other slaves. Buildbot allows you to define a series of steps you want to perform. 

Yes, I made my own experiments while ago (http://xiexie.syslab.disco.unimib.it:8010) and I scheduled some time early September to fix all build chains and write proper (not too long) documentation so people can provide their configuration and someone (probably me) can integrate a specific external build chains into the buildbot master config file and collect on the QE-FORGE statistics of all builds. 

The Buildbot master is very lightweight, I am planning to run a centralized instance from QE-FORGE without impact the performance of the website. Hopefully (cross-fingers) everything will be ready by the next scheduled release late September.

I am currently working on the deployment of a CI infrastructure for my own QE development and also other codes based both on Docker containers and OpenStack here in Cambridge (it is a good excuse for me to play DevOps in my spare time). We can share some best practices for buildbot configuration.


> Currently it's doing a svn update of the QE trunk, then configure with the correct env. variable and then make. If everything goes well, it then svn update our EPW code inside QE, then build the code.
> Then a series of test will be perform (not yet implemented) to be sure that the physical results are stable with different architecture & compilers (http://epw.org.uk/Main/TestFarm).

I noticed it is more flexible the GIT component than the SVN component, so my experimental instances were all fetching the source code from the git mirror of QE SVN repository hosted on Github. The mirror is sync automatically every hour.


> As an update to my previous mail, I've additionally tested pgfortran 15 and QE compiles without issue. Seems like there is a problem with pgfortran 12. 

I consider PGI 12.x too old to be included as a valid CI test. I deprecated on my HPC system all PGI version older than 14.1.


Continuous integration is not just perform builds against software stack or bare metal architectures. There is the testing side that is as important as "compile". I am working on it as well, testcode integrates very well with buildbot and it is successfully used by other quantum chemistry codes. What are you using for numerical regression? Do you have your own scripts?

We can follow-up privately, I do not want to bother people here with these technicalities... ;-)

F

--
Mr. Filippo SPIGA, M.Sc.
Quantum ESPRESSO Foundation
http://fspiga.github.io ~ skype: filippo.spiga

*****
Disclaimer: "Please note this message and any attachments are CONFIDENTIAL and may be privileged or otherwise protected from disclosure. The contents are not to be disclosed to anyone other than the addressee. Unauthorized recipients are requested to preserve this confidentiality and to advise the sender immediately of any error in transmission."