Occasional nacl/update-cpu failures with amdclang-hip-571
Summary
Every few days, our gromacs:amdclang-hip-571:regressiontest job fails with:
Re-running nacl using CPU-based update
Ran grompp in 0 seconds
Ran mdrun in 1 seconds
FAILED. Check checkforce.out (1 errors) file(s) in nacl/update-cpu for nacl-update-cpuExample of mismatching forces:
f[ 102] (-2.32801e+02 -7.58409e+02 1.41438e+02) - (-2.35271e+02 -7.61785e+02 9.58010e+01)Exact steps to reproduce
Look at CI history, run it manually in the container replicating the CI setup (around 10% chance of failure).
For developers: Why is this important?
Flaky CI is annoying and can indicate a potential bug.