Dataset.where performances regression.

What happened?

Hello,

I'm using the Dataset.where function to select data based on some fields values and it takes way to much time!
The dask dashboard seems to show some tasks repeating themselves many times.

The provided example uses a 1D array for which the selection could be done with Dataset.sel but with our real usecase we make selections on 2D variables.

This problem seems to have appeared with the 2022.6.0 xarray release, the 2022.3.0 is working as expected.

What did you expect to happen?

Using the 2022.3 release, this selection takes 1.37 seconds.
Using the 2022.6.0 up to the 2023.2.0 (the one from yesterday), this selection takes 8.47 seconds.

This example is a very simple and small one, with real data and use case we simply cannot use this function anymore.

Minimal Complete Verifiable Example

import dask.array as da
import distributed as dist
import xarray as xr


client = dist.Client()

# Using small chunks emphasis the problem
ds = xr.Dataset(
    {"field": xr.DataArray(data=da.empty(shape=10000, chunks=10), dims=("x"))}
)
sel = ds["field"] > 0

ds.where(sel, drop=True)

MVCE confirmation

• Minimal example — the example is as focused as reasonably possible to demonstrate the underlying issue in xarray.
• Complete example — the example is self-contained, including all data and the text of any traceback.
• Verifiable example — the example copy & pastes into an IPython prompt or Binder notebook, returning the result.
• New issue — a search of GitHub Issues suggests this is not a duplicate.

Relevant log output

No response

Anything else we need to know?

No response

Environment

Problematic version

Working version

Start by running the provided Dataset.where example with the listed xarray and Dask versions, then compare task graphs and timings between xarray 2022.3.0 and 2022.6.0 or later. Done means identifying and addressing the repeated tasks so the regression is resolved without losing support for 2D selections; the issue names no source files or tests.