chunks not updated from one zarr to another

[raybellwaves]

I'm creating a zarr store from another zarr store. Before I write to the final zarr store I call ds.compute(). I'm unsure on why the chunks metadata from the source zarr is copied over. I would have though chunks would have matched shape or not be present.

import xarray as xr

ds = xr.open_mfdataset(
    "s3://era5-pds/zarr/2020/01/data/air_temperature_at_2_metres.zarr", engine="zarr"
)
ds = ds.sel(lat=slice(2, 0), lon=slice(0, 2))
ds = ds.compute()
ds.to_zarr("ds.zarr", consolidated=True)

Looking at ds.zarr/air_temperature_at_2_metres/.zarray I see:

{
    "chunks": [
        372,
        150,
        150
    ],
    "compressor": {
        "blocksize": 0,
        "clevel": 3,
        "cname": "zstd",
        "id": "blosc",
        "shuffle": 2
    },
    "dtype": "<f4",
    "fill_value": "NaN",
    "filters": null,
    "order": "C",
    "shape": [
        744,
        9,
        9
    ],
    "zarr_format": 2
}

[raybellwaves]

I believe i'm getting caught out by the default encoding dict ds["air_temperature_at_2_metres"].encoding. I guess that's something to check when writing to a zarr store.