Issues using ROMS readers over network drives (SSHFS/NIRD)

[AndersOpdal]

I have some challenges when trying to initialize OpenDrift readers for ROMS datasets (Norfjords 160m and NorKyst 800m) stored on a remote network drive (NIRD) mounted via SSHFS on a Windows machine. I have to the best of my abilities, and help from AI not managed to get this to work.

When using the standard initialization with a file pattern, Reader(path_to_files), I think the high latency of the SSHFS mount causes xr.open_mfdataset to struggle. I get the error ValueError: Could not find any dimension coordinates to use to order the datasets,

I tried to work around this by pre-loading the dataset manually using a different open_mfdataset configuration:

ds = xr.open_mfdataset(files, 
                       concat_dim='ocean_time', 
                       combine='nested', 
                       data_vars='minimal', 
                       coords='minimal', 
                       compat='override')

This works for creating an xarray Dataset. However, passing this ds object into OpenDrift readers does not work:

`reader_ROMS_native: `Even when passed a Dataset, the __init__ method seems to trigger a new open_mfdataset call on the underlying file paths, which brings back the network latency/ordering issue.

`reader_netCDF_CF_generic:` This reader correctly identifies the dataset as ROMS but then raises ValueError: 

ZeroDivisionError: When I try to strip the ROMS attributes to "trick" the CF reader, I get a `ZeroDivisionError` in `basereader/__init__.py `

Question

What is the recommended way to use OpenDrift readers over a network drive (NIRD) mounted via SSHFS on a Windows machine, or to use OpenDrift readers when the dataset is already opened as an xarray Dataset??

Is there a way to bypass the ROMS check in the generic reader, or a way to prevent the ROMS-native reader from re-scanning files when a Dataset is already provided?

Any help would be greatly appreciated!

[knutfrode]

This might be more of an Xarray- or filesystem/OS-issue than an OpenDrift issue.

However, it appears that your version of OpenDrift might be a little dated? If so, you can try after updating to latest version.
In newer version, both the ROMS- and generic readers use this common helper method to open datasets:
https://opendrift.github.io/_modules/opendrift/readers.html#open_dataset_opendrift

Thus if providing an Xarray Dataset, this should be used. However, lazy opening with Xarray depends much on on the arguments passed to open_mfdataset, and in particular chunks. Thus note in particular the default chunking of 1 for ocean_time in the ROMS reader:
https://github.com/OpenDrift/opendrift/blob/master/opendrift/readers/reader_ROMS_native.py#L150

The helper method is generally optimised, so you could try to simply pass the string with wildcards directly to the ROMS reader constructor, or use the convenience method o.add_readers_from_list()