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/6799#issuecomment-1317516314,https://api.github.com/repos/pydata/xarray/issues/6799,1317516314,IC_kwDOAMm_X85Oh7Qa,2448579,2022-11-16T18:55:00Z,2022-11-16T18:55:00Z,MEMBER,Linking the dask issue: https://github.com/dask/dask/issues/6474,"{""total_count"": 0, ""+1"": 0, ""-1"": 0, ""laugh"": 0, ""hooray"": 0, ""confused"": 0, ""heart"": 0, ""rocket"": 0, ""eyes"": 0}",,1307112340
https://github.com/pydata/xarray/issues/6799#issuecomment-1317358777,https://api.github.com/repos/pydata/xarray/issues/6799,1317358777,IC_kwDOAMm_X85OhUy5,2448579,2022-11-16T17:04:23Z,2022-11-16T17:04:23Z,MEMBER,"The challenge is you could be interping to an unordered set of locations.

So perhaps we can sort the input locations, do the interp with map_overlap, then argsort the result back to expected order.","{""total_count"": 0, ""+1"": 0, ""-1"": 0, ""laugh"": 0, ""hooray"": 0, ""confused"": 0, ""heart"": 0, ""rocket"": 0, ""eyes"": 0}",,1307112340
https://github.com/pydata/xarray/issues/6799#issuecomment-1317352980,https://api.github.com/repos/pydata/xarray/issues/6799,1317352980,IC_kwDOAMm_X85OhTYU,3309802,2022-11-16T17:00:04Z,2022-11-16T17:00:04Z,NONE,"> The current code also has the unfortunate side-effect of merging all chunks too

Don't really know what I'm talking about here, but it looks to me like the current dask-interpolation routine [uses `blockwise`](https://github.com/pydata/xarray/blob/df909b444991c3d76210d018e5268de541b8e17b/xarray/core/missing.py#L750-L762). That is, it's trying to simply map a function over each chunk in the array. To get the chunks into a structure where this is correct to do, you have to first merge all the chunks along the interpolation axis.

I would have expected interpolation to use [`map_overlap`](https://docs.dask.org/en/stable/array-overlap.html). You'd add some padding to each chunk, map the interpolation over each chunk (without combining them), then trim off the extra. By using overlap, you don't need to combine all the chunks into one big array first, so the operation can actually be parallel.

FYI, fixing this would probably be a big deal to geospatial people—then you could do array reprojection without GDAL! Unfortunately not something I have time to work on right now, but perhaps someone else would be interested?","{""total_count"": 0, ""+1"": 0, ""-1"": 0, ""laugh"": 0, ""hooray"": 0, ""confused"": 0, ""heart"": 0, ""rocket"": 0, ""eyes"": 0}",,1307112340
https://github.com/pydata/xarray/issues/6799#issuecomment-1194294204,https://api.github.com/repos/pydata/xarray/issues/6799,1194294204,IC_kwDOAMm_X85HL3u8,2448579,2022-07-25T16:07:09Z,2022-07-25T16:07:09Z,MEMBER,"The current code also has the unfortunate side-effect of merging all chunks too.

I think we should instead think of generating a dask array of weights and then using `xr.dot`","{""total_count"": 0, ""+1"": 0, ""-1"": 0, ""laugh"": 0, ""hooray"": 0, ""confused"": 0, ""heart"": 0, ""rocket"": 0, ""eyes"": 0}",,1307112340
https://github.com/pydata/xarray/issues/6799#issuecomment-1188348238,https://api.github.com/repos/pydata/xarray/issues/6799,1188348238,IC_kwDOAMm_X85G1MFO,39069044,2022-07-18T21:43:11Z,2022-07-18T21:43:11Z,CONTRIBUTOR,"The chunking structure on disk is pretty instrumental to my application, which requires fast retrievals of full slices in the time dimension. The loop option in my first post only takes about 10 seconds with `ni=1000` which is fine for my use case, so I'll probably go with that for now. It would be interesting to dig deeper though and see if there is a way to handle this better in the interp logic.","{""total_count"": 0, ""+1"": 0, ""-1"": 0, ""laugh"": 0, ""hooray"": 0, ""confused"": 0, ""heart"": 0, ""rocket"": 0, ""eyes"": 0}",,1307112340
https://github.com/pydata/xarray/issues/6799#issuecomment-1187755023,https://api.github.com/repos/pydata/xarray/issues/6799,1187755023,IC_kwDOAMm_X85Gy7QP,1005109,2022-07-18T16:55:42Z,2022-07-18T16:55:42Z,CONTRIBUTOR,"You are right about the behavior of the code. I don't see any way to enhance that in the general case.

Maybe, in your case, rechunking before interpolating might be a good idea","{""total_count"": 0, ""+1"": 0, ""-1"": 0, ""laugh"": 0, ""hooray"": 0, ""confused"": 0, ""heart"": 0, ""rocket"": 0, ""eyes"": 0}",,1307112340
https://github.com/pydata/xarray/issues/6799#issuecomment-1187729258,https://api.github.com/repos/pydata/xarray/issues/6799,1187729258,IC_kwDOAMm_X85Gy09q,39069044,2022-07-18T16:44:40Z,2022-07-18T16:44:40Z,CONTRIBUTOR,Interpolating on chunked dimensions doesn't work at all prior to #4155. The changes in #4069 are also relevant.,"{""total_count"": 0, ""+1"": 0, ""-1"": 0, ""laugh"": 0, ""hooray"": 0, ""confused"": 0, ""heart"": 0, ""rocket"": 0, ""eyes"": 0}",,1307112340
https://github.com/pydata/xarray/issues/6799#issuecomment-1187699270,https://api.github.com/repos/pydata/xarray/issues/6799,1187699270,IC_kwDOAMm_X85GytpG,2448579,2022-07-18T16:18:09Z,2022-07-18T16:18:21Z,MEMBER,"> Given the performance behavior I'm guessing we may be doing sequntial interpolation for the dimensions, basically an interp1d call for all the xx points and from there another to the yy points, which for even a small number of points would require nearly all chunks to be loaded in. 

Yeah I think this is right.

You could check if it was better before https://github.com/pydata/xarray/pull/4155 (if it worked that is)

cc @pums974 @Illviljan ","{""total_count"": 0, ""+1"": 0, ""-1"": 0, ""laugh"": 0, ""hooray"": 0, ""confused"": 0, ""heart"": 0, ""rocket"": 0, ""eyes"": 0}",,1307112340