multiple reads of nasa-nex-gddp-cmip6 dataset from MultiZarrToZarr concatenated metadata returns all nans

[solomon-negusse]

New to kerchunk and encountered this issue running the MSFT Planetary Computer example notebook. Simply running twice cells 13 & 14 that read and plot point single variable time-series for a point will reproduce this issue where the first run will have the valid values but second will be all nans. Is this usage issue or bug? Thanks.

fsspec == v2023.6.0 kerchunk == v0.1.2 xarray == v2023.5.0 zarr == v2.15.0

[martindurant]

There is a lot going on in that notebook, so hard for me to say. I recommend turning on the loggers for "fsspec.reference" and whatever abfs's one is called. You might want to get a planetary computer person on this too, to see if they have anything to say for example about how long your signed session is expected to last.