I am running with a forecast split into dask array chunks. The evaluation runs fine if I do not use this function and manually rename the zarr variables myself, even with the same dataset and same chunks.
Hey, thanks for reporting this error. That is indeed suboptimal. It sounds like you have a workaround for now. I added this to our internal TODO list but can't promise we will push a fix in the immediate future.
Issue Description
If the forecast dataset exceeds RAM amount, this function seems to always give out-of-memory errors. I've narrowed it down to the line
ds = xr.merge(das)
I am running with a forecast split into dask array chunks. The evaluation runs fine if I do not use this function and manually rename the zarr variables myself, even with the same dataset and same chunks.
Execution
Conda environment