Allow configure orca to send requests to remote server

[cooperok]

According to code static image export will work only if orca is installed on same server as plotly.py

server_url = 'http://{hostname}:{port}'.format(hostname='localhost', port=orca_state['port'])

But, there is an option to run orca as standalone server. Would be greate if param hostname can be specified in options same as port.

My case is that I have Google App Engine application, and it's problematic to install orca. So I want to set up standalone image server and still use plotly to get images like plotly.io.to_image(...)

[jonmmease]

Hi @cooperok, yes this is a great idea.
Happy to help someone through a PR on this if anyone is interested in tackling this.

I think we could probably add a plotly.io.orca.config.server_url parameter. Setting this would then unset (or override) the executable, port, and timeout, mathjax, topojson, and mapbox_access_token properties.

[maresb]

For a quick-and-dirty workaround to specify a custom hostname, I found the following two possibilities.

1. If you are able to edit the source code, then edit _orca.py and manually set the desired hostname and port in the server_url = command. Since we won't be using the orca executable, at the end of _orca.py add the line ensure_server = lambda: None to disable the corresponding checks. (This sets ensure_server to the do-nothing function.)

2. If for some reason you are unable or unwilling to edit the plotly.py source code, then on the machine running plotly.py, run the command

    socat tcp-listen:9091,reuseaddr,fork,bind=127.0.0.1 tcp:orca-host:9091
   

   where orca-host is the hostname of the orca server. This creates a tunnel to access orca-host from localhost of the machine running plotly.py, using the default port 9091. Finally, import plotly.io with the following extra commands:

    import plotly.io as pio
    pio._orca.ensure_server = lambda: None
    pio._orca.orca_state["port"] = 9091
   

With either of these techniques, I can set up docker-compose to run Jupyter + plotly.py, and orca.

[miriad]

@jonmmease , Any chance you could review the PR on this?

[jonmmease]

done in #1850