Home

Creating a Price Tracker in Elixir

Elixir is a pretty neat language. It’s the first functional programming language that I’ve spent any amount of time with (unless you consider Java’s Functional Interfaces and C#’s LINQ to be functional programming). Elixir has cute syntax like pipes, cool features like pattern matching for everything, and sick capabilities like supervision trees. Pretending for a moment that most sites don’t already notify you when items go on sale, a price tracking application sounds like a great idea for an Elixir project.

Repository here

Extracting Price from Web Pages

The first thing we need to have is a way to grab html pages from the web that contain a product and its price. HTTPoison is the de facto standard HTTP client library for Elixir and what I’ll be using. Once we have the html, we need a way to extract the price from it. Floki is the HTML parsing library I’ll be using.

Let’s try to grab the contents of a web page. On success, HTTPoison.get/1 returns a tuple whose first element is the atom :ok and whose second element is a Response struct. Because all structs are maps in Elixir, we’ll use pattern patching to grab the body of the response.

{:ok, %{body: body}} = HTTPoison.get("https://www.escapefromtarkov.com/preorder-page")

Now that we have the HTML stored in body, we can use Floki.find/2 to extract what we want. The first argument is the raw HTML, and the second is a CSS attribute selector that Floki will use to match with elements in the document. Floki represents HTML in an html_tree struct where
<div class="hello">world</div> would be {"div", [{"class","hello"}], ["world"]}.

By inspecting the webpage in Chrome, I see the price for Escape From Tarkov's Edge of Darkness Edition is in a span with an attribute itemprop="price", so we'll see what Floki returns with that.

iex(2)> Floki.find(body, "[itemprop=price]")
[
  {"span", [{"itemprop", "price"}], ["44.99$"]},
  {"span", [{"itemprop", "price"}], ["74.99$"]},
  {"span", [{"itemprop", "price"}], ["99.99$"]},
  {"span", [{"itemprop", "price"}], ["139.99$"]}
]

That selector returned the prices for all the editions, but we can see the format Floki returns. I'm only interested in the crazy-expensive edition, so we'll adjust the selector to be more specific: "#preorder_edge_of_darkness [itemprop=price]" matches exactly what I want.

Now we'll grab the text from the node, remove any currency symbols, and parse the text as a float. Floki.text/2 will return the text from a node, and all its child nodes by default. The regex ~r|[$¢£€¥₿]| will help get just the number from the text. Before we put this all into a function, we need add a "User-Agent" header to our HTTP GET calls to trick sites like Amazon into thinking we're a browser.

def get_price(url, selector) do
  with  {:ok, %{body: body}}  <- HTTPoison.get(url, @headers, follow_redirect: true),
        html_node             <- Floki.find(body, selector),
        text                  <- Floki.text(html_node, sep: " "),
        num_str               <- String.replace(text, @currency_regex, ""),
        {price, _rest}        <- Float.parse(num_str),
        do: price
end

Our get_price/2 function leverages a 'with' statement with no else clause, so the calling process will exit if a match fails at any step. This is fine for our application where we'll just call that function in an unlinked process every hour.

Maintaining Product State

Next we need to define a struct to represent a product. We'll need a name to key the product, a URL, and a selector at a minimum. Besides that, we'll need a current price, a base price to compare it to, and a sale price or sale ratio to determine when we want to be notified of a price drop. Finally we'll need to keep track of the price of the item the last time we sent a sale notification.

defmodule Fyresale.Product do
  defstruct [
    :name,
    :url,
    :selector, 
    curr_price: 0.0,
    base_price: 0.0, 
    sale_price: 0.0,
    sale_ratio: 0.9,
    sent_price: 0.0
  ]
  ...
end

We'll say our product is on sale if either the current price is less than or equal to our sale price (if configured) or if the current price is less than or equal to the sale ratio (defaults to 0.9) multiplied by the base price. Because we don't want to send emails every hour something is on sale, we need to keep track of when we last sent a notification. We only want to care about a sale if the price drops even lower than when we last notified, or if the price has been at the base value since the last time we notified, in which case the value of 0.0 is assigned to sent_price.

def on_sale?(product) do
  product.curr_price <= product.sale_price or
  product.curr_price <= product.sale_ratio * product.base_price
end

def should_send_sale?(product) do
  product.sent_price == 0.0 or product.curr_price <= product.sent_price
end

Now we need to maintain the state of our products. A simple GenServer was the obvious choice for this purpose. The underlying state is a map of products keyed by their names. The client API and GenServer callbacks with logging removed are shown below.

def get_product(name) do
  GenServer.call(Products, {:get_product, name})
end

def set_product(name, product) do
  GenServer.cast(Products, {:set_product, name, product})
end

def handle_call({:get_product, name}, _from, state) do
  {:reply, Map.get(state, name), state}
end

def handle_cast({:set_product, name, product}, state) do
  {:noreply, Map.put(state, name, product)}
end

Sending Emails

Now that we have a concept of products and a way to maintain their state, we ought to be able to send emails. Bamboo is the most common email library for Elixir, and it has many standard and third party adapters for email services like Mailgun, Mandrill, Sendgrid, etc. These are probably very useful for a large scale Phoenix application, but I just want to be able to use my personal gmail, so I'm using Bamboo.SMTPAdapter. use Bamboo.Mailer gives your module access to the send_later/1 macro that will asynchronously send and an email created by Bamboo.Email.new_email/1. Though Bamboo recommends separating the module that creates emails from the module that sends them, I opted to do both in one module.

defmodule Fyresale.Mailer do
  use Bamboo.Mailer, otp_app: :fyresale
  
  import Bamboo.Email

  def sale_email(name) do
    product = Fyresale.ProductStore.get_product(name)
    new_email(
      to: get_recipients(),
      from: Application.get_env(:fyresale, Fyresale.Mailer)[:username],
      subject: "Fyresale - #{name} is on sale",
      html_body: "<h3>#{name} is on sale for $#{product.curr_price}</h3><br><a href=\"#{product.url}\">Link to product</a>",
      text_body: "#{name} is on sale for #{product.curr_price}. Link here: #{product.url}"
    )
  end

  defp get_recipients do
    unless Enum.empty?(Application.get_env(:fyresale, :recipients)) do
      Application.get_env(:fyresale, :recipients)
    else
      Application.get_env(:fyresale, Fyresale.Mailer)[:username]
    end
  end
end

Checking Price Periodically

With ability to check price, maintain product state, and send emails, we need a synchronous function to check a product's price, update it, and if needed, send an email notification and update again.

def check_price(name) do
  product = name |> ProductStore.get_product
  price = get_price(product.url, product.selector)
  product = product |> Product.update_price(price)
  name |> ProductStore.set_product(product)
  if Product.on_sale?(product) and Product.should_send_sale?(product) do
    Mailer.sale_email(name) |> Mailer.deliver_later
    ProductStore.set_product(name, Product.update_sent_to_curr(product))
  end
end

What's more, we can wrap that function in a fire-and-forget task to be run in its own unlinked, unsupervised process.

defp check_price_task(name), do: Task.start(fn -> check_price(name) end)

Looping in Elixir is accomplished through recursion. Tail call optimization allows you to do this without blowing up the stack. In Elixir, it's usually considered improper to use Process.sleep/1. The proper way to run a periodic task is to use Process.send_after/3 to send a message to a process after a timeout, very similar to setTimout() in JavaScript. In our loop, we want to spawn an independent task for each product every hour to check its price. Because the processes aren't linked to our main supervised process, if a product fails to check its price during the loop iteration, its process will die on its own without affecting any other process. Every hour a new process will start for each product regardless of whether or not the last price check was successful.

defp loop do
  receive do
    msg when is_list msg -> 
      msg |> Enum.each(&check_price_task(&1))
      Process.send_after(self(), msg, @loop_period)
      loop()
  end
end 

Conclusion

Great. With everything in place, we can throw our processes into a supervision tree and run our application. Thanks for reading and getting this far. Fyresale repository can be found here . The README has detailed instructions on how to use and configure it.