`open_zarr` hangs if 's3://' at front of root s3fs string

[birdsarah]

The following code has an error in it:

import s3fs
import xarray as xr

S3_DIR = 's3://my_bucket'

s3 = s3fs.S3FileSystem(**storage_options)
store = s3fs.S3Map(root=f'{S3_DIR}/my_zarr_store', s3=s3)
array = xr.open_zarr(store)['data']

The presence of "s3://" at the beginning of the string causes to take a really really really long time (I don't have the time off hand but over 10 minutes) to return with a key error, that there is nothing at 'data', which is often a clue of a permissions error.

Without the "s3://" this returns quickly with my data.

This error occurred for me as I was opening other files with dask with code such as

df = dd.read_parquet(f'{S3_DIR}/my_data.parquet', storage_options=storage_options)

I know that this is not technically an xarray issue. However, it is the xarray line that suffers the user experience as the s3fs just returns without any checking.

I was wondering whether the open_zarr function could be generous and inspect the root argument in the case of s3fs access and warn if 's3://' is detected.

I am also wondering what the interaction issue is that causes it to take so long for the permission type error to be returned.

ping @martindurant in case you have thoughts from the s3fs side.

[martindurant]

Do you have any idea what is taking the extra time? s3fs ought to, in theory, treat URLs with and with the s3:// the same, but this may not have been tested with tea mapping.

[birdsarah]

I am no longer having this issue - see #2927 - so closing.