Source: dolfin Followup-For: Bug #920546 Took a while, but we eventually got the timeout error again in a debci test for 2019.1.0. One fail after 32 successful tests. That's evidently much less frequent than the timeout errors for 2018.1.0, indicating the new version has fixed some of the race conditions but not all.
debci provides only 2 CPUs, but the dolfin MPI test uses 3 processes, so the test is oversubscribed. However upstream says that oversubscribed MPI tests are in fact useful, and can help iron out bugs. They say that the MPI tests are not expected to time out the way we've been experiencing, so I'll draw their attention to the ongoing problem in 2019.1. This first C++ test fail in 2019.1 occured during demo_gmg-poisson_mpi.
