no-server-november

Lets have some fun... https://serverless.com/blog/no-server-november-challenge/

Strategy

All challenges will be grouped into the same folder, under the same serverless.yml, but will be documented separately below.

Lorem Ipsum

For this challenge, we implement a simple lorem ipsum generator with custom text borrowed straight from stranger ipsum. A single function, using Server-Side Rendering, will provide a well-formatted HTML page, with custom text generated on every request. We use wizipsum to create the text, and he to encode it for HTML. You can see the generator in action here.

Dad Bot

Lets twitter up a bot, shall we? We create a new bot, whose name is @NDadbot, which is using the Twitter API to broadcast delightfully terrible jokes from the dadbot api. Fortunately for us, there already exists a dadbot client in nodejs, as does the twitter api, so this primarily involved creating a new twitter account, promoting that account to a developer by applying, and provisioning tokens. We're using credstash from my private account to manage tokens, because we are responsible adults.

GitHub Check

Alright, so we are tasked with creating a github check. Mercifully, we already have a github repo to work from — so let us create a check on that (this?). We've set up a github webhook to point to https://no-server-november.ulfhedinn.net/github_check, and you'll see checks appear when this piece is complete. We'll be using the official octokit client to talk to github, and we've created a new github application to authenticate against this. This was alleviated using github-app, which made the whole github app song-and-dance more palletable. We use a simple regex to match JIRA tickets, specifically /[A-Z]{2,}-[0-9]+:/g, and you can see the check in action here.

Cat Meme

A brutally simple website using the giphy API. We're using the giphy-api tool to ease the burden of querying giphy, and of course credstash for the API token.

One Direction Facts

okay so if we want to talk Boy Bands, why couldn't it be N'Sync? Oh well, so we're gonna scrape some facts off the internet, use vim to make this less painful:

:v/[0-9]\+)/d (deletes all lines not starting like 3))
:%s/"/\\"/g (turn every " into \", so this does not mess up JSON later)
:%s/^[0-9]\+) $.*$$/ "\1",/g (turn every line into a JSON-like string)
Add [ to the beginning of the file, remove the final trailing , and add ] to the end of the file. The final file should look similar to this cat facts JSON payload. Then we follow along with the example serverless app, register the app as a simple skill, and voila. Notably, the Alexa Skill setup took longer than expected (the setup instructions changed!).

Image Classifier Bot

Follow the same steps as above; this time the bot is named @ServerImage. This time however, we need to respond to direct messages — that means we're going to need to set up some webhooks using the Activity API. I set up an Account Activity API Sandbox, and added my new Image Classifier app to a dev environment there. Also, don't forget to change the permissions of the app to include read, write and direct messages. Do this before creating access tokens...

In order to register a webhook, we first need to implement a GET end-point to prove to Twitter that this endpoint is owned by this application; we borrow heavily from another codebase for their CRC implementation for that. Next time, most of the registration crap should probably be handled through twurl

In order to kick off the process, we manually set up the subscription with a PATCH call to the bot (so we can lazily avoid figuring out how to curl). Notably, this took a hell of a long time to get right.

For the love of... ok, we're only now getting to the actual meat of this problem. It's garbage issues like this crap that makes the Twitter API such a hassle to deal with. Also, the direct_messages event API just doesn't work with the [node-twitter] package (and a bunch of the other end-points), so we end up using the regular request package for like half of these requests.

Useful scripts here:

serverless invoke local -f image_classifier -d '{"httpMethod": "PATCH", "body": "{\"appId\": \"<APP_ID>\"}"}'
serverless invoke local -f image_classifier -d '{"httpMethod": "DELETE", "body": "{\"appId\": \"<APP_ID>\", \"id\": \"<ID>\"}"}'

And in the nodejs console...

request.get({ oauth: { token, token_secret, consumer_key, consumer_secret }, url: "https://api.twitter.com/1.1/account_activity/all/dev/webhooks.json" }, console.log);

The web-hook itself...

Payload { id: '1064008442954706944',
  url: 'https://no-server-november.ulfhedinn.net/imagebot',
  valid: true,
  created_timestamp: '2018-11-18 04:12:36 +0000' }

Twilio Integration - Wake Me Up!

Wake me up every morning at 8am PDT (4pm UTC). Super simple, deliciously straightforward, and now I have a new phone number I can text from!

Interestingly enough, with this addition we see a significant increase in memory pressure, so we are now deploying with more memory:

node --max-old-space-size=12000 ./node_modules/serverless/bin/serverless deploy

jontg / no-server-november Goto Github PK