Dataset from url not found

[aaronspring]

What happened:

I tried to open a remote url and got OSError, but !wget url works

What you expected to happen:

open the remote netcdf file

Minimal Complete Verifiable Example:

from netCDF4 import Dataset

import netCDF4
netCDF4.__version__


url='https://crudata.uea.ac.uk/cru/data/temperature/HadCRUT.4.6.0.0.median.nc'
# working_url='https://thredds.ucar.edu/thredds/dodsC/grib/NCEP/GFS/Global_0p5deg/GFS_Global_0p5deg_20200923_0000.grib2'

Dataset(url)
---------------------------------------------------------------------------
OSError                                   Traceback (most recent call last)
<ipython-input-14-265839034cee> in <module>
----> 1 Dataset(url)

netCDF4/_netCDF4.pyx in netCDF4._netCDF4.Dataset.__init__()

netCDF4/_netCDF4.pyx in netCDF4._netCDF4._ensure_nc_success()

OSError: [Errno -90] NetCDF: file not found: b'https://crudata.uea.ac.uk/cru/data/temperature/HadCRUT.4.6.0.0.median.nc'

Environment:

Output of xr.show_versions()

INSTALLED VERSIONS

commit: None
python: 3.7.6 | packaged by conda-forge | (default, Jan 7 2020, 22:33:48)
[GCC 7.3.0]
python-bits: 64
OS: Linux
OS-release: 2.6.32-754.29.2.el6.x86_64
machine: x86_64
processor: x86_64
byteorder: little
LC_ALL: en_US.UTF-8
LANG: en_US.UTF-8
LOCALE: en_US.UTF-8
libhdf5: 1.10.5
libnetcdf: 4.6.2

xarray: 0.16.1
pandas: 1.1.2
numpy: 1.19.1
scipy: 1.5.2
netCDF4: 1.5.1.2
pydap: installed
h5netcdf: 0.8.0
h5py: 2.10.0
Nio: 1.5.5
zarr: 2.4.0
cftime: 1.2.1
nc_time_axis: 1.2.0
PseudoNetCDF: None
rasterio: 1.1.0
cfgrib: 0.9.7.6
iris: 2.2.0
bottleneck: 1.3.1
dask: 2.15.0
distributed: 2.20.0
matplotlib: 3.1.2
cartopy: 0.17.0
seaborn: 0.10.1
numbagg: None
pint: 0.11
setuptools: 47.1.1.post20200529
pip: 20.2.3
conda: None
pytest: 5.3.5
IPython: 7.15.0
sphinx: None

[rabernat]

Is url an opendap endpoint or just a netCDF file on an http server?

Until recently, the netCDF4 C library (which underlies netcdf4-python) has not been able to open the latter. However, I seem to remember recently @dopplershift posting about how this capability has recently been added; however, you need to add a special query string to the URL, which I can't remember or find in the docs.

[aaronspring]

Solution: xr.open_dataset('https://crudata.uea.ac.uk/cru/data/temperature/HadCRUT.4.6.0.0.median.nc#mode=bytes')

thanks @rabernat and @dopplershift

I am absolutely convinced (but cannot prove it) that a month ago and way before, xr.open_dataset('https://crudata.uea.ac.uk/cru/data/temperature/HadCRUT.4.6.0.0.median.nc') used to work without adding #mode=bytes.

[rabernat]

The only way I have ever made this work in the past was with h5netcdf (which supports opening file-like objects) + fsspec (which provides file-like objects for remote URLs).

[aaronspring]

you are right @rabernat, I used intake-xarray and fsspec (not specifying directly h5netcdf but probably internally somehow) to open remote files: https://gist.github.com/aaronspring/30c904009cfffecb15ab54b247e0472a

[rabernat]

On thing I don't like about intake is that it is sometimes capable of hiding too much. This is more a problem for developers than pure "users", although I think that distinction is not super useful.

[rsignell-usgs]

@aaronspring, it's worth noting that prepending simplecache to the URL you pass to fsspec causes the whole file to be downloaded, while using fsspec without simplecache or using the new netcdf4 #mode=bytes does not.

I tried these out with URL you provided, but it was very slow accessing the data on this server from AWS us-west-2. So I downloaded the 21MB file using wget (which took 15 minutes!), and put it on S3.

On S3 using these two methods I get these results:

[dopplershift]

Well that's just silly. I've opened the netCDF-c performance issue at Unidata/netcdf-c#1848.

[barronh]

Does netcdf4-python work with fsspec? This would be an amazing feature!

[dopplershift]

@barronh No it does not. Since netcdf4-python is a thin wrapper around the netCDF C library, it only supports I/O methods supported there. That list has grown to include things like S3, Zarr, opendap, and HTTP byte-range requests, though.

[barronh]

@dopplershift - Thanks. That makes sense. The S3 bucket was my primary interest, so it sounds like the limitation for me is the backend library. If I updated libnetcdf4 and linked netcdf4-python to it, then I would have s3 support already.

Thanks!

[rabernat]

What would be great would be a C (or Rust?) implementation of fsspec that could be shared among many C libraries (e.g. NetCDF + GDAL) that need flexible cloud-style I/O. Kind of like GDAL VSI but decoupled from GDAL.

[guigrpa]

Is #mode=bytes still supported? On netdf4-python v1.6.2, it returns an error:

>>> ds = netCDF4.Dataset('https://crudata.uea.ac.uk/cru/data/temperature/HadCRUT.4.6.0.0.median.nc#mode=bytes')
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "src/netCDF4/_netCDF4.pyx", line 2463, in netCDF4._netCDF4.Dataset.__init__
  File "src/netCDF4/_netCDF4.pyx", line 2026, in netCDF4._netCDF4._ensure_nc_success
FileNotFoundError: [Errno 2] No such file or directory: b'https://crudata.uea.ac.uk/cru/data/temperature/HadCRUT.4.6.0.0.median.nc#mode=bytes'

Tried with netCDF4 v1.5.8 and v.1.6.2.

[dopplershift]

@guigrpa I've opened an issue upstream in the C library to dig into why that URL isn't working (it should be).

[khouakhi]

I get
OSError: [Errno -67] NetCDF: libcurl failure: 'https://crudata.uea.ac.uk/cru/data/temperature/HadCRUT.4.6.0.0.median.nc#mode=bytes'
using recent netCDF4 1.6.3 and xarray 2023.3.0

[dopplershift]

@khouakhi This should be fixed with netCDF-c 4.9.2. The latest netcdf4 packages on conda-forge should pull it in, not sure what, if anything, needs to be done for the PyPI packages.