Is there a way to enforce consistent time sorting when using append_dim arg in to_zarr()?

[cwerner]

Hi.

I just started using Zarr stores with xarray. Everything is fine, but now I had the problem that I need to prepend a chunk of data at the time dimension...

If I just call:

new_data_but_prior_in_time.to_zarr(store=store, append_dim="time")

with the new data batch (with prior time info) it get's added no problem but now the time coordinate is not continuous any more.

Demo:

import xarray as xr
import numpy as np

a = xr.DataArray(
    np.ones((10,100,100)), 
    coords=[('time', range(20,30)), ('lat', range(100)), ('lon', range(100))]
)
a.name = "DATA"

b = xr.DataArray(
    np.ones((10,100,100)), 
    coords=[('time', range(10,20)), ('lat', range(100)), ('lon', range(100))]
)
b.name = "DATA"

a.to_dataset().to_zarr("testitest.zarr")
b.to_dataset().to_zarr("testitest.zarr", append_dim="time")

c = xr.open_dataset("testitest.zarr", engine="zarr")
print(c.time)

Output:

array([20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19])

Is there a way around creating an entire new zarr store to accomplish this?
If that's not possible: What would be the most efficient way to create new/ joined and properly time-sorted zarr file from two zarr files that are already located in an S3 bucket (each >50GB in size)?

Cheers,
C