html_url,issue_url,id,node_id,user,created_at,updated_at,author_association,body,reactions,performed_via_github_app,issue
https://github.com/pydata/xarray/issues/4739#issuecomment-752261269,https://api.github.com/repos/pydata/xarray/issues/4739,752261269,MDEyOklzc3VlQ29tbWVudDc1MjI2MTI2OQ==,2448579,2020-12-29T22:19:29Z,2020-12-29T22:19:29Z,MEMBER,"> But some time could be saved if we could convert them to dask arrays in xr.Dataset.interp before the variable loop starts.

Now implemented. Runtime has dropped from 5.3s to 2.3s (!)","{""total_count"": 2, ""+1"": 0, ""-1"": 0, ""laugh"": 0, ""hooray"": 2, ""confused"": 0, ""heart"": 0, ""rocket"": 0, ""eyes"": 0}",,775875024
https://github.com/pydata/xarray/issues/4739#issuecomment-752210306,https://api.github.com/repos/pydata/xarray/issues/4739,752210306,MDEyOklzc3VlQ29tbWVudDc1MjIxMDMwNg==,2448579,2020-12-29T19:13:27Z,2020-12-29T19:33:07Z,MEMBER,"We don't support lazy index variables yet (#1603) so you can't interpolate to a dask variable.

> But some time could be saved if we could convert them to dask arrays in xr.Dataset.interp before the variable loop starts.

This may be true. I *think* we could convert `x` and `destination` to dask (only once) if any of the variables to be interpolated are dask-arrays and pass that to `interp_func` here rather than passing IndexVariables through.
https://github.com/pydata/xarray/blob/bf0fe2caca1d2ebc4f1298f019758baa12f68b94/xarray/core/missing.py#L641-L643

OTOH I found some easier optimizations. See #4740

1. Passing `meta` to `blockwise` saves 0.5s in your example. 
2. Another thing we can do is call `_localize` at the `Dataset` level rather than within the variable loop. This is taking 1.65s most of which is in 4000 calls to `get_loc`. At the `Dataset` level, this becomes just 2 calls to `get_loc`
","{""total_count"": 0, ""+1"": 0, ""-1"": 0, ""laugh"": 0, ""hooray"": 0, ""confused"": 0, ""heart"": 0, ""rocket"": 0, ""eyes"": 0}",,775875024