Jupyter-server-proxy, imported as python library, doesn't see config in jupyter_server_config.py

Niole_Nelson · July 16, 2021, 8:17pm

I’m using ProxyHandler from jupyter-server-proxy.handlers in order to proxy requests in a web server that I have built. I want to configure it to proxy to remote locations by configuring c.ServerApp.allow_remote_access = True in ~/.jupyter/jupyter_server_config.py, but it always ends up being false and jupyter-server-proxy throws a 403. I verified that jupyter-server outputs the configuration that I created when I call jupyter-server --Application.show_config=true, same for jupyter-server --ServerApp.show_config=true and jupyter-server --JupyterApp.show_config=true

Is this the right way to configure jupyter-server-proxy in this particular instance?

manics · July 16, 2021, 9:50pm

jupyter-server-proxy is an extension, so it has it’s own set of configuration properties, c.ServerApp.allow_remote_access only affects the jupyter-server component.

If you want jupyter-server-proxy running as a server/notebook extension to proxy remote hosts you’ll need to add the host(s) to host_allowlist, see: Accessing Arbitrary Ports or Hosts — Jupyter ServerProxy 1.0 documentation

In your title you’ve said it’s “imported as a Python library”. Can you explain a bit more, or share your code? Are you developing your own extension, or are you just importing it in a standalone script?

Niole_Nelson · July 16, 2021, 10:26pm

see this code. It’s being used to optionally integrate proxying into a web server. The code I’m talking about is actually in the dask.distributed python library.

Is it only possible to configure this by updating the host_allowlist, as in c.ServerProxy.host_allowlist = ['x']? in .jupyter/jupyter_notebook_config.py?

manics · July 21, 2021, 8:44pm

jupyter-server-proxy uses traitlets for configuration. Classes don’t automatically load their configuration from a file though, this has to be done explicitly.

Typically this is done in the main application, which then passes the configuration to other classes when they’re instantiated. Is this what you’re doing in dask.distributed?

If you’re not, you might be able to get away with just passing in host_allowlist=... when you construct the class:

github.com

jupyterhub/jupyter-server-proxy/blob/f89ed9de385846276a55a67aaa04cdcb50c6a23d/jupyter_server_proxy/handlers.py#L46

    
      
          used directly as a means of overriding CORS. This presents significant
          security risks, and could allow arbitrary remote code access. Instead, it is
          meant to be subclassed and used for proxying URLs from trusted sources.
          
          
Subclasses should implement open, http_get, post, put, delete, head, patch,
          and options.
          """
          def __init__(self, *args, **kwargs):
              self.proxy_base = ''
              self.absolute_url = kwargs.pop('absolute_url', False)
              self.host_allowlist = kwargs.pop('host_allowlist', ['localhost', '127.0.0.1'])
              self.subprotocols = None
              super().__init__(*args, **kwargs)
          
          
# Support all the methods that tornado does by default except for GET which
          # is passed to WebSocketHandlerMixin and then to WebSocketHandler.
          
          
async def open(self, port, proxied_path):
              raise NotImplementedError('Subclasses of ProxyHandler should implement open')
          
          
async def http_get(self, host, port, proxy_path=''):

Though from past experience things can get a bit hairy when you’re messing with traitlets derived classes. Probably no harm trying though

jacobtomlinson · August 4, 2022, 4:02pm

I’m running into this same issue. The docs for jupyter-server-proxy say to update the host_allowlist option in your configuration. But I can’t really figure out where that is.

My initial reaction was to create ~/.jupyter/jupyter_notebook_config.py with the contents c.ServerProxy.host_allowlist = ["x"] but that doesn’t seem to have taken effect.

jacobtomlinson · August 4, 2022, 4:49pm

Figured it out. It needed to be in jupyter_lab_config.py for my setup.

Zsailer · August 4, 2022, 5:16pm

Hey @jacobtomlinson, you can also use jupyter_server_config.py—which replaced jupyter_notebook_config.py.

I believe, if you have nbclassic enabled (a new package that we’re using to help with the transition from the classic Notebook to JupyterLab), it should automatically shim the jupyter_notebook_config.py settings to proper place. I’d be curious to know if that’s not working…

jacobtomlinson · August 4, 2022, 6:00pm

Thanks for getting back to me @Zsailer! I can import nbclassic so I guess that’s installed, but yeah it didn’t seem to pick up jupyter_notebook_config.py but did pick up jupyter_lab_config.py.

I can easily change is to jupyter_server_config.py which seems to make more sense, is that what you would recommend?

Topic		Replies	Views
Setting config options for jupyter-proxy-server via jupyterhub JupyterHub	1	847	June 24, 2019
Jupyter-server-proxy and port JupyterLab	0	891	February 9, 2022
Unable to make jupyter-server-proxy working - help really welcome! JupyterHub	4	1721	March 7, 2023
JupyterHub / JupyterServerProxy 403 to OPTIONS request (CORS preflight) JupyterHub how-to , help-wanted	2	479	March 18, 2024
Issues with url's server proxy JupyterHub	3	426	January 30, 2024

Jupyter-server-proxy, imported as python library, doesn't see config in jupyter_server_config.py

Related topics