How to use allow_rechunk for interpolate_na

[keves1]

I'm using xarray.DataArray.interpolate_na and when the interpolation dimension is not a single chunk, I get this error: ValueError: dimension time on 0th function argument to apply_ufunc with dask='parallelized' consists of multiple chunks, but is also a core dimension. To fix, either rechunk into a single array chunk along this dimension, i.e., ``.chunk(dict(time=-1))``, or pass ``allow_rechunk=True`` in ``dask_gufunc_kwargs`` but beware that this may significantly increase memory usage.

However, I don't see how to use allow_rechunk=True. I tried passing this to interpolate_na like this: da.interpolate_na("time", method="linear", dask_gufunc_kwargs={"allow_rechunk": True}) but it seems to have no effect because I get the same error.

If allow_rechunk is not available, how do I handle linear interpolation along a dimension that is too large to be loaded as a single chunk?

[dcherian]

ATM You'll have to rechunk so that there is only one chunk along the axes you are interpolating over.

Code-wise we could either migrate to using xr.dot (#6799 (comment)) or some clever tricks with map_overlap (#6799 (comment)) if you're up for it.

[keves1]

Got it, thanks.