Changing the pipeline for qe2pert tests: downloading required files instead of QE calculations. #60
Conversation
…wnload the files from the cloud storage
|
This is great, Sergei! Two questions about Box:
|
|
There was a problem hiding this comment.
For legacy code, like run_scf, etc., can we keep it somewhere?
Maybe in a different file?..
It is not needed now, but it does the job of all the preliminary calculations that might be needed in future for other projects.
So, now it is all commented out, right?
Maybe we can leave it uncommented by in another file?..
Yes, that's a great idea! I have separated all legacy functions into a separate file in a separate folder. |
|
Cool! Feel free to merge when you like. |
Hi everyone!
In this PR we are completely changing the test approach for
qe2pert.x. In the original version, in order to compute epr files, we did a full computation starting from SCF. This allowed us not to waste memory storing heavy files, some of which are binary.However, this approach has 2 significant disadvantages:
qe2pert.xwas working correctly. This did not allow us to test this executable file normally.Therefore, it was decided to use the following approach:
qe2pert.xcalculations in remote storage (Box);qe2pert.xtests are run.Thanks to the answer from https://lists.quantum-espresso.org/pipermail/users/2024-August/051827.html, we learned that binaries are not a problem when switching between GNU and Intel, so for each epr file we just have a folder inside the archive. The archive itself weighs 412 MB, which is not so critical. The archive is stored in our group's Caltech Box, and the file is accessible to everyone, but I don't think that's a problem. At the same time, the time of tests on turbo was reduced to 3 minutes, in docker containers it became less than 5 minutes, and the problem with stability of calculations seems to have disappeared.
I'll be glad to receive any suggestions regarding the code and ideas on how to improve it!