Slack Bot

This notebook implements the Slack outgoing webhook API to parse certain messages as code to eval on a kernel. This notebook also interfaces with the Slack incoming webhook API to send text and image output from a kernel back to Slack.

To use this example notebook, it must be running in an environment with one additional port open for the HTTP server defined herein to receive webhook calls from Slack. The default is to listen for HTTP requests on port 9001 for demo purposes.

Open and run all the cells in the discussion notebook.
Create an incoming webhook integration in your Slack project.
1. Select a channel to which the bot will post responses.
2. Note its webhook URL.
Create an outcoming webhook integration in your Slack project.
1. Select which channel will be monitored for select trigger words.
2. State the URL of this notebook server, port 9001.
3. Note the token.
Put the URL for the incoming webhook in the SLACK_URL global variable below. This notebook will forward select output from the kernel associated with the discussion notebook to that URL for posting in your configured channel.
Put the token for the outgoing webhook in the SLACK_TOKEN global variable below. This notebook will only accept messages bearing that token to avoid spoofing.
Run this notebook top to bottom.
If all is well, the bot should report itself available in the configured output channel.
Interact with the bot by typing your configured trigger followed by code to eval.

To Slack

This section defines the logic that takes output from a kernel and forwards it to Slack in an appropriate manner.



In [ ]:

    
import json
import base64
import os
import hashlib
from pprint import pprint # for debug
from jupyter_client.ioloop import IOLoopKernelManager
from tornado.httpclient import AsyncHTTPClient, HTTPClient

We'll set some hard coded values up front. These should really get read from the environment but I'm lazy at the moment.



In [ ]:

    
SLACK_URL = 'https://hooks.slack.com/services/XXXXXX/XXXXXX/XXXXXXX'
SLACK_TOKEN = 'XXXXXXXXX'

Try connecting to an existing kernel connection file if one exists specifically for the defrag demo. If it's not there, Start kernel separate from this one. Keeps user code out of the namespace of this service. Avoids hard-to-reason-about async request handling all within a single kernel.



In [ ]:

    
if os.path.exists('/tmp/defrag_demo'):
    km = IOLoopKernelManager(connection_file='/tmp/defrag_demo')
    km.load_connection_file()
    print('connecting to existing kernel')
else:
    km = IOLoopKernelManager()
    km.start_kernel()

Slack doesn't accept base64 encoded blobs back as images. We can only send back a real URL. We'll make a local directory where we'll dump any plots generated by the kernel to disk. Later, we'll serve them up through the same web server that we'll use to handle the Slack outgoing webhook calls.



In [ ]:

    
PLOT_DIR = '/home/jovyan/plots'



In [ ]:

    
!mkdir -p $PLOT_DIR



In [ ]:

    
def b64_to_file(b64_str, ext):
    '''Dump a base64 encoded string to disk as a binary file with the given extension.'''
    # decode base64 image and write to disk under a unique ID
    img = base64.decodebytes(b64_str.encode('utf-8'))
    # hash to filename
    name = hashlib.sha1(img).hexdigest()
    with open(os.path.join(PLOT_DIR, name+'.'+ext), 'wb') as f:
        f.write(img)
    
    return name

Connect to the iopub socket to receive kernel output.



In [ ]:

    
if 'iopub' in locals():
    iopub.close()
iopub = km.connect_iopub()

Define functions to handle message types of interest. The generic on_reply below dispatches to these.



In [ ]:

    
def on_stream(content):
    '''Handles stdout, stderr.'''
    return dict(text=content['text'])



In [ ]:

    
def on_display_data(content):
    '''Handles rich output.'''
    data = content['data']
    response = {}
    
    # prefer images
    attachments = []
    for key in data.keys():
        if key.startswith('image'):
            _, ext = key.split('/')
            name = b64_to_file(data[key], ext)
            # point to plot on the web
            # TODO: don't hard code the server URL
            attachments.append({
                "fallback": "Oh noes! The plot didn't render!",
                'image_url': 'http://parente.cloudet.xyz:9001/static/{}.png'.format(name)
            })
        if len(attachments):
            response['attachments'] = attachments
    # fallback on text
    if 'text/plain' in data:
        response['text'] = data['text/plain']
    
    return response if len(response) else None

The doc says these two types of messages are equivalent. So just alias the function.



In [ ]:

    
on_execute_result = on_display_data

Build a HTTP client to use to push messages back to Slack.



In [ ]:

    
http_client = AsyncHTTPClient()

Process all kernel replies as they come in. Use the kernel manager methods for converting ZeroMQ stream byte strings to nice Python dictionaries.



In [ ]:

    
def on_reply(stream, msg_list):
    # process raw messages
    idents, msg_list = km.session.feed_identities(msg_list)
    msg = km.session.deserialize(msg_list)

    # get delegate based on message type
    func = globals().get('on_'+msg['msg_type'])
    if func is not None:
        # get an optional response
        response = func(msg['content'])
        if response:
            # dump the response as JSON to Slack
            http_client.fetch(SLACK_URL, method='POST', 
                              body=json.dumps(response), 
                              headers={'Content-Type' : 'application/json'})

Hook the on_reply to the iopub stream.



In [ ]:

    
iopub.on_recv_stream(on_reply)

Create a client that can be used to execute code on the kernel.



In [ ]:

    
kc = km.client()

From Slack

This section defines the logic that takes messages from Slack and forwards them to a kernel in an appropriate manner.



In [ ]:

    
import tornado.web
import tornado.httpserver
import json

Define a simple HTTP handler for Slack POSTs. Support a GET for liveliness checks too.



In [ ]:

    
class IncomingHandler(tornado.web.RequestHandler):
    def get(self):
        self.finish('{"status": "ok"}')
        
    def post(self):
        token = self.get_body_argument('token')
        if token != SLACK_TOKEN:
            return self.send_error(401)
        
        # get code to run
        code = self.get_body_argument('text')
        # remove command prefix, up to first space
        code = code[code.find(' ')+1:].strip()

        # execute the code in the other kernel
        kc.execute(code)
        
        # return nothing for now
        self.finish()

Map the handler and start listening.



In [ ]:

    
application = tornado.web.Application([
    (r"/", IncomingHandler)
], static_path=PLOT_DIR)



In [ ]:

    
if 'server' in locals():
    server.stop()
server = tornado.httpserver.HTTPServer(application)
server.listen(9001, '0.0.0.0')

Say hi.



In [ ]:

    
kc.execute('print("Hi, all")')