Deploying Bokeh+Flask Model to Heroku, Bokeh Doesn't Appear

Community Support
1

When I run my bokeh app locally, following the flask_embed.py example on the Bokeh Git, it works fine. But running it on Heroku, it shows the html template page but not the Bokeh app.
I can attach the heroku error logs if need be. Thanks.

 def bkapp(doc):
    textin = TextInput(title = "Submit Blog Post:")
    button = Button(label="Submit", button_type="success")
    p = Paragraph(text="Blog entry here")
    def update_data(event):
        data = str(textin.value)
        vector = transformer.transform([' '.join(clean(data))])
        result = model.predict(vector)
        if int(result) == 1:
            pred_text = 'Male'
        else:
            pred_text = 'Female'
        output = {'prediction': pred_text}
        p.text = "{}".format(output)
    button.on_click(update_data)
    box = WidgetBox(children = [textin, button, p])
    doc.add_root(box)
    
        
@app.route('/', methods=['GET'])
def bkapp_page():
    script = server_document('https://blog-nlp.herokuapp.com/bkapp')
    return render_template("embed.html", script=script, template="Flask")

def bk_worker():
    server = Server({'/bkapp': bkapp}, io_loop=IOLoop())
    server.start()
    server.io_loop.start()

Thread(target=bk_worker).start()

if __name__ == '__main__':
    app.run()
2

This is always advised. Browser JS console logs as well.

3

Thanks Bryan

2020-05-21T19:35:41.822243+00:00 app[web.1]: Exception in thread Thread-1:
2020-05-21T19:35:41.822268+00:00 app[web.1]: Traceback (most recent call last):
2020-05-21T19:35:41.822269+00:00 app[web.1]: File "/app/.heroku/python/lib/python3.7/threading.py", line 926, in _bootstrap_inner
2020-05-21T19:35:41.822270+00:00 app[web.1]: self.run()
2020-05-21T19:35:41.822271+00:00 app[web.1]: File "/app/.heroku/python/lib/python3.7/threading.py", line 870, in run
2020-05-21T19:35:41.822271+00:00 app[web.1]: self._target(*self._args, **self._kwargs)
2020-05-21T19:35:41.822271+00:00 app[web.1]: File "/app/bkflask.py", line 52, in bk_worker
2020-05-21T19:35:41.822272+00:00 app[web.1]: server = Server({'/bkapp': bkapp}, io_loop=IOLoop())
2020-05-21T19:35:41.822272+00:00 app[web.1]: File "/app/.heroku/python/lib/python3.7/site-packages/bokeh/server/server.py", line 390, in __init__
2020-05-21T19:35:41.822273+00:00 app[web.1]: sockets, self._port = bind_sockets(self.address, self.port)
2020-05-21T19:35:41.822273+00:00 app[web.1]: File "/app/.heroku/python/lib/python3.7/site-packages/bokeh/server/util.py", line 60, in bind_sockets
2020-05-21T19:35:41.822274+00:00 app[web.1]: ss = netutil.bind_sockets(port=port or 0, address=address)
2020-05-21T19:35:41.822274+00:00 app[web.1]: File "/app/.heroku/python/lib/python3.7/site-packages/tornado/netutil.py", line 174, in bind_sockets
2020-05-21T19:35:41.822274+00:00 app[web.1]: sock.bind(sockaddr)
2020-05-21T19:35:41.822275+00:00 app[web.1]: OSError: [Errno 98] Address already in use
2020-05-21T19:35:41.822275+00:00 app[web.1]: 
2020-05-21T19:36:33.641749+00:00 heroku[router]: at=info method=GET path="/bkapp/autoload.js?bokeh-autoload-element=1001&bokeh-app-path=/bkapp&bokeh-absolute-url=https://blog-nlp.herokuapp.com/bkapp" host=blog-nlp.herokuapp.com request_id=2b39ef47-a487-4957-9a7b-3a89e973607c fwd="68.129.186.174" dyno=web.1 connect=1ms service=7ms status=404 bytes=400 protocol=https
2020-05-21T19:36:33.641130+00:00 app[web.1]: 10.61.164.89 - - [21/May/2020:19:36:33 +0000] "GET /bkapp/autoload.js?bokeh-autoload-element=1001&bokeh-app-path=/bkapp&bokeh-absolute-url=https://blog-nlp.herokuapp.com/bkapp HTTP/1.1" 404 232 "http://127.0.0.1:5000/" "Mozilla/5.0 (X11; Ubuntu; Linux x86_64; rv:76.0) Gecko/20100101 Firefox/76.0"
^C
4

Well, this is certainly a problem:

Usually that occurs if you start a server process once, then try to start a second server process on the same port. I am afraid I don’t have any specific experience with Heroku to be able to say much more than that, though.

5

I don’t have much experience with embedding Bokeh like that, but I don’t think you can do it this particular way on Heroku. You’re using a public URL with server_document, and the Bokeh server cannot listen on the port associated with that URL on Heroku when your own Flask server listens on that port. Try replacing https://blog-nlp.herokuapp.com with http://localhost:5006. You may also need to explicitly pass address='localhost' to Server (port 5006 is the default one).

6

Sure, these commands work when deploying the app locally. But when I want to deploy it on Heroku it gives an AppError for that /bkapp page.