QA Failure in structural mechanics application

[jrubiogonzalez]

When running QA, it is failing randomly specially in windows, but sometimes also in Liunux (Fedora). The failing test is

Ran 849 of 850 test cases in 2.45259s. 1 failed: TestConstitutiveLawIntegrateStressPlasticityFiniteStrainKinematic Failed with message: Error: Check failed because vector VMres with values [-1.80942e+07, -1.80942e+07, -1.80036e+07, 0, 0, 0] Is not near vector TestVM with values 6 Mismatch found in component 0: -1.80942e+07 not near -1.76859e+07 within relative tolerance 0.0001.

In fact both included in the same file fail alternatively

I went thorug valgrind adding unly this test and this is the output. valgrind_salida_kinematic_plasticity_constitutive_laws.txt

Looks like an uninitalised variable. @pooyan-dadvand @roigcarlo @RiccardoRossi @ipouplana @ddiezrod

[ipouplana]

I'll add @AlejandroCornejo just in case he knows something

[jrubiogonzalez]

It is stranger in Linux but if you run (at least in fedora) the test enough times it ends up failing

[roigcarlo]

Does it use openmp? it may be a race condition...

[loumalouomega]

Finite strain are my fault, many operations are involved, may be my fault

[jrubiogonzalez]

I thought the same, but OpenMP inside a constitutive law... Please check you can reproduce the issue

[loumalouomega]

My be a memory issue, finite strain makes mandatory backup intermediate values which are restores later in the values of the CL, as using pure references and raw pointers may be memory problematic

[loumalouomega]

We should use valgrind

[jrubiogonzalez]

I attached the file with the valgrind output, not very clear

[loumalouomega]

Valgrind is never clear :P

[loumalouomega]

Valgrind is never clear :P

Looks like a bug in the constitutive law utilities, I will check tonight

[ddiezrod]

Its complaining about this line it seems https://github.com/KratosMultiphysics/Kratos/blob/6bc6a85fc07acbd745a0ff721fe18ee7299e008f/applications/StructuralMechanicsApplication/custom_utilities/constitutive_law_utilities.cpp#L945

[ddiezrod]

I think I found the error, back stress vector is used unitialized in both finite_strains and small_strains kinematic plasticity CL.

[AlejandroCornejo]

Hi, I'm going to check what @ddiezrod says, one sec

[ddiezrod]

Here for example (its copied in many places) https://github.com/KratosMultiphysics/Kratos/blob/b6212cf7c7ce0c5f9c4b4a7514fb12615444e8e8/applications/StructuralMechanicsApplication/custom_advanced_constitutive/generic_finite_strain_kinematic_plasticity.cpp#L129

It is used later inside the integrator when you compute the plastic denominator

[AlejandroCornejo]

yes, I think It is smthg wrong with this CL, working on it

[ddiezrod]

yes, I think It is smthg wrong with this CL, working on it

Ok, if you do Vector back_stress_vector = ZeroVector(VoigtZise) tests pass

[AlejandroCornejo]

Sure but the algorithm should track the value of the BackStress historically... I forgot to create as member variable

[AlejandroCornejo]

I've done some changes in https://github.com/KratosMultiphysics/Kratos/pull/5871 I've run the tests and they run (maybe yo could run them just in case). As you were saying, the Back stress vector should be historical and I forgot to create the member variable.

[ddiezrod]

I think this can be closed with #5871 and #5878 . Just to mention it #5835