This project was initially thought of as a tool for people that were looking to rent apartments in Montreal. Since the most popular site for this purpose is Kijiji, I wanted to make a friendly front end that filtered out the spam posts and displayed relevant information about the listing. I also wanted to add a notification system that sent an email to the user anytime a new ad that matched their search terms was posted. Finally, I wanted to display additional information that wasn’t available directly from kijiji. I decided to add the Google Maps API that allows for a more intuitive navigation and the Walkscore API rates the listing’s access to public transportation and bike friendliness.
My goal was to learn to use third party APIs and to make a robust full stack application.
I used a firebase realtime database to store the scraped data and to eventually provide a public facing API.
For the backend I wanted to learn to develop and deploy serverless functions. This meant that I would have to split the functionality into simple modules. I used Netlify functions, a Heroku instance as a master clock for the lambda functions and lastly a Cloud Scheduler to keep everything synced.
After trying a few other providers, I decided that Netlify was the best choice for my application. It’s a powerful and fast platform for deployment, which allowed me to use (and learn) serverless functions.
Like most projects, the initial idea in my head was simpler and more elegant than what the actual solutions ended up being. I encountered a few bumps along the way that set back my progress significantly. Looking back, all these caveats forced me to learn much more than I initially challenged myself to.
When I started to develop the search function I encountered that kijiji doesn’t provide a public API for their ads. To solve this, I created a simple express app that scraped the site. The results would be displayed in real time on the application. Everything ran smoothly on my local machine, but when I deployed the application I soon realized that kijiji doesn’t like repeated requests from servers. I would need a more robust infrastructure to work around this limitation. Ultimately, I wanted to keep the app’s cost low (free) and this meant that I would have to make things more complicated for me.
I compiled the most popular search-terms in the Montreal area for real estate rentals and added them to my database. Whenever the function ran, it would read the keywords, do the search and store the results. That way, whenever a user interacted with the frontend, they would get up to date results.
Since Netlify functions run on the AWS lambda infrastructure, they have a 10 second limit on execution. This meant that I would have to trigger the function multiple times per minute if I wanted to have the scraper run non-stop. For this, I made a simple express app and ran it on Heroku. The app gets an HTTP request from the Netlify function, then waits 10 seconds until it makes a request back to the Netlify function.
My motivation to keep costs low led me to design an overly complex system to make a responsive backend API. In spite of a few headaches, I learned a lot about cloud functions and serverless infrastructure. Since I already use Netlify to deploy most of my projects, I will take advantage of the built-in lambda functions they provide. It’s a simple way to make requests that can bypass the browser’s CORS policies without the need of proxying requests.