Unit test speed issues

[rrsettgast]

There are a few unit test that take much too long to complete on the CI machines:

108/270 Test #108: testXmlWrapper ..........................................   Passed   11.74 sec
180/270 Test #180: testGraphColoringMPI_mpi ................................   Passed  129.23 sec
228/270 Test #228: testSurfaceGenerator_mpi ................................   Passed  334.74 sec
239/270 Test #239: testCompositionalMultiPhaseMFDPolyhedral ................   Passed   12.75 sec
242/270 Test #242: testMixedDimSinglePhaseFlow_mpi .........................   Passed   29.31 sec
244/270 Test #244: testMixedDimHydrostaticEquilibrium_mpi ..................   Passed  445.77 sec
268/270 Test #268: testWavePropagationAdjoint1 .............................   Passed    8.14 sec
270/270 Test #270: testFEMConsistency_mpi ..................................   Passed  621.07 sec

Obviously testing is important, but some of these cases must be able to gain coverage in a more efficient manner. If you "own" these tests please look into your code and figure out why it is taking so long.

a unit test should be a few seconds.

[OmarDuran]

@rrsettgast I’ll address this as soon as I can. I'm currently tied up with industrial cases for contact mechanics in both external and internal fracture workflows.

On a related note, what is the acceptable execution time range for integrated tests? I want to ensure that the test cases I'm developing remain within accepted CI limits.

[rrsettgast]

@OmarDuran I have not looked at the tests, but I think some of these may be "application/integration" tests. I think you mentioned having a 1000 dof system to solve, which is clearly not a unit test.

[bd713]

@rrsettgast @OmarDuran Just to confirm, I’m working on the coloring and the wave equation. I’ll update by the end of the week at the latest.

[OmarDuran]

@rrsettgast yes, some of those cases are slightly large. I agree that at that size ( #DoF > 1000), they should not be in the integrationTests folder. My approach will be to reduce those meshes as much as possible so they run in few seconds. I think it is crucial for the stability of the surface generator and the contact mechanic that they remain in the integrationTests rather than the application tests.

[rrsettgast]

@rrsettgast yes, some of those cases are slightly large. I agree that at that size ( #DoF > 1000), they should not be in the integrationTests folder. My approach will be to reduce those meshes as much as possible so they run in few seconds. I think it is crucial for the stability of the surface generator and the contact mechanic that they remain in the integrationTests rather than the application tests.

I have no objection ensuring robustness and stability. Getting the tests in place has been an important step. I am hoping that this can be achieved in less than 10 minutes.