-
Notifications
You must be signed in to change notification settings - Fork 24
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Upgrading macOS version in the test_macos.yml (addresses #278) #279
Conversation
* upgrade the macOS version to the latest available * add in the workflow conda installation
The Newtonian fluid test fails for three procs, but not for one and four. The tolerances all look reasonable, so not sure why this test fails now only with three procs. We could try recomputing the test with a direct solver to rule out any linear solver problems? |
@mrp089 I am not sure I know how to use a direct solver on svFSIplus. For macOS we do not have PETSc or Trilinos libraries, so unless we have a direct solver implemented in svFSILS, I am not sure how to use it. |
If our PETSc interface can read configuration files, it's pretty easy, e.g.:
If not, we first would need to add a line or two to read input files. |
@mrp089 this is one point we do not read options from file. The main point is for macOS testing we do not use petsc or trilinos. We do not build the solver with them, so we don't have them available. |
Oops, I missed that part in your reply, sorry. We could still recompute the reference solution on Ubuntu with a direct solver. Or we could tune up the tolerances of the linear solver more (if the error goes lower) in the reference solution and in the test and see if that fixes it. As a last resort, coarsen the test tolerance |
@dcodoni Some comments from a casual observer
|
It turns out that this pipeline has different convergence behavior for 1 (good), 3 (bad), 4 (good) procs for
|
@dcodoni, quick fix: set |
@mrp089 I did and it passes! It is converging in 7 iterations, it seems stiffer for the first 3 iterations repsects to the 1proc and 4procs case. |
I'm good with the quickfix in the interest of moving this and other PRs along. I documented the problem in #284, and we can keep digging into it over there. |
The primary goal of CI is to ensure that the code works as expected after code modifications. Three things should ideally be checked
We really do need to understand what the linear algebra settings (e.g. number of maximum iterations, tolerances, etc.) are testing and to make sure that all CI tests are consistent. |
Current situation
Currently the workflow file running integration tests on macOS uses an old and specific version of macOS (macOS 12). This now is creating problems, in particular the installation of necessary packages is taking around 5 hours (6 hours is the limit to complete the testing).
Release Notes
I upgraded the macOS version in the workflow file and I set it to macOS-latest.
Testing
Code of Conduct & Contributing Guidelines