How to decrease traffic when writing into region of zarr archive

[meteoDaniel]

we are initialize empty zarrs with shape:
(31, 96, 1000, 1000) and chunks : (31, 96, 50, 50) because we need all data from dim 1 and 2 in each query.
Now we write into region again:

    dataset.to_zarr(
        store,
        region=region,
        mode="r+",
        compute=True,
        consolidated=True if zarr_format < 3 else False,
        zarr_format=zarr_format,
        safe_chunks=safe_chunks,
    )

During writing along the first dimension day by day, the whole chunk (aka dataset) needs to be loaded to write into region. This causes a lot of traffic.

We would like to go into s3 object store with the process but we need to decrease the traffic. Do you know how to achive this?

The previous solution used append mode, which causes chunk sizes (1, 96, 50, 50) and very long queries against the dataset.

Best regards
Daniel

[meteoDaniel]

No one ? Am I the only facing issues with region write?

[rabernat]

Define "traffic".

[meteoDaniel]

I mean the fact, to load the whole chunk to write into it.
E.g. when I use a remote s3 this would result in additional traffic and i need to download it complete just to write some additional data into a region.
Or am I wrong with this assumption?