Dockerfile PETSc version bump

[garth-wells]

Also add PETSc garbage collection to tests.

[jorgensd]

Are you sure about this? We downgraded due to the various issues listed in: #2483
Are all of them resolved ?

[garth-wells]

Are you sure about this? We downgraded due to the various issues listed in: #2483 Are all of them resolved ?

PETSc.garbage_cleanup() has been added. See https://gitlab.com/petsc/petsc/-/issues/1309.

[jorgensd]

Are you sure about this? We downgraded due to the various issues listed in: #2483 Are all of them resolved ?

PETSc.garbage_cleanup() has been added. See https://gitlab.com/petsc/petsc/-/issues/1309.

But have they merged in the fix for https://gitlab.com/petsc/petsc/-/issues/1288
into a release?

[garth-wells]

Tested inside Docker locally. Seems ok with latest MPI releases.

[francesco-ballarin]

Are we sure that all issues are solved?

I have heavily parametrized tests in multiphenicsx
https://github.com/multiphenics/multiphenicsx/blob/update/tests/unit/fem/test_assembler_restriction.py
and I am still getting failures like

internal_Dist_graph_create_adjacent(125).: MPI_Dist_graph_create_adjacent(comm=0xc40013e0, indegree=0, sources=(nil), sourceweights=0x7f5eeba42af4, outdegree=0, destinations=(nil), destweights=0x7f5eeba42af4, MPI_INFO_NULL, reorder=0, comm_dist_graph=0x7ffde86a6b08) failed
MPIR_Dist_graph_create_adjacent_impl(319): 
MPII_Comm_copy(913)......................: 
MPIR_Get_contextid_sparse_group(591).....: Too many communicators (0/2048 free on this process; ignore_id=0)
Abort(71919887) on node 0 (rank 0 in comm 368): application called MPI_Abort(comm=0xC40013E0, 71919887) - process 0

I am quite confident the error is not with my code, since:

• by installing pytest-repeat I can see the failed test moving to a different parameterization value depending on the value of --count.
• tests run ok on my local machine (with openmpi, rather than mpich)

[garth-wells]

The CI works with the latest OpenMPI and MPICH (4.1).

We can't sit on PETSc 3.17 forever, and there is little indication that any PETSc changes are forthcoming.

You can try destroying some PETSc objects when you're finished with them. This got me further with the test suite. Another possibility is running pytest in batches.