Potential 'output_writers' saving bug? about oceananigans.jl HOT 6 CLOSED

Our system seems to have forgotten it's got GPUs post power cut so it's CPU only for now, I assume the underlying save functions are the same between devices?

Yes. CPUs are even better for MWEs since we can run them from any laptop.

Could you please post the MWE on #3056? That way we centralize the discussion there.

from oceananigans.jl.

I've seen this before and yes I think it's a bug. Is there another issue open about this? I think there may be. @tomchor has a PR that is looking at this, we could use some extra eyes though.

from oceananigans.jl.

I also reported this on #3056, so maybe we should close this to avoid multiple issues of the same problem.

I don't have a PR specifically trying to fix that, but it is possible that #3606 fixes, or least ameliorates, the issue.

@mncrowe can you test your simulation on that branch and check if you see the same behavior?

from oceananigans.jl.

@mncrowe it sounds like this happened when you weren't using adaptive time-stepping --- which is nice because it will help us construct an MWE more easily. Do you think you can help with that? We just need a very simple simulation that reproduces the issue, hopefully something 0D with no dynamics that runs fast.

from oceananigans.jl.

@glwagner Will do. I'll remove the dynamics from my run and see if I can get a minimum not-working example.

@tomchor I've downloaded that branch and will test when I can.

Our system seems to have forgotten it's got GPUs post power cut so it's CPU only for now, I assume the underlying save functions are the same between devices?

from oceananigans.jl.

I don't think the MWE even requires saving output. Can't we also achieve an MWE using a simple Callback?

This issue should be independent of whether you are using the CPU or GPU.

from oceananigans.jl.