The PR to scale up the size of the profiling test data to make the results less noisy has gone too far. Actions are crashing out and - although the errors are a bit opaque - it seems likely this is due to the runners running out of memory (16GB on ubuntu-latest). @MarionBWeinzierl has seen the same problem running the profiling suite on a 16GB Ubuntu laptop.
I've run some simple peak memory usage estimates locally and changing the pmodel scale up factor lon gives the following peak memory usage using /usr/bin/time -l outputs:
pmodel_profile_scaleup
peak memory footprint in GB
40 (current)
35.84
20
19.74
10
9.95
We need to scale down the tests and it would be useful to make the scaling factors configurable.
Describe the bug
The PR to scale up the size of the profiling test data to make the results less noisy has gone too far. Actions are crashing out and - although the errors are a bit opaque - it seems likely this is due to the runners running out of memory (16GB on
ubuntu-latest
). @MarionBWeinzierl has seen the same problem running the profiling suite on a 16GB Ubuntu laptop.I've run some simple peak memory usage estimates locally and changing the
pmodel
scale up factorlon
gives the following peak memory usage using/usr/bin/time -l
outputs:We need to scale down the tests and it would be useful to make the scaling factors configurable.
To Reproduce
See here:
https://github.com/ImperialCollegeLondon/pyrealm/actions/runs/8704343221