Tracking your AWS costs

The Problem

Amazon Web Services (AWS) offers an enormous amount of products, with reasonable pricing for a hobbiest (pay for what you use) or business (save on sysadmin wages). However, their monthly bill doesn't have any details: just the service, the charges, the taxes, and the total.

I wanted more granular look in my morning update email, and here's how I got it done.

Capturing the data

Amazon's Creating an AWS Cost and Usage Report guide offers a step-by-step breakdown to follow. I used "include resource IDs" so I can see costs of specific EC2 servers. I also selected "Enable report data integration for Amazon Athena" since it's a database that only charges when you query it.

You have to wait a day or so for Amazon to writing some billing data to your S3 bucket, then proceed to the next step.

Reading the schema

Now we turn to the aptly named AWS Glue to crawl our S3 bucket's new billing data and write its database schema to AWS Athena.

I configured the crawler to "Run Monthly" on the 1st, and allowed Glue to create a new IAM role with permissions to do the work it needs to do. All-in-all the crawler cost me $0.02 to run once. Refer to the previously noted Amazon guide‘s "Configuring Athena data integration" section for more details here.

Querying the costs

After the Glue crawler finishes, we're ready to start querying. Visit AWS Athena and you'll see a new database and its tables.

Play around with different queries inside Athena's UI. Here's what I use to query daily costs by resource:

SELECT line_item_product_code,
         line_item_resource_id,
         line_item_currency_code,
         SUM(line_item_blended_cost) as line_item_blended_cost
FROM billing
WHERE line_item_blended_cost > 0
        AND line_item_usage_start_date >= TIMESTAMP '2020-01-15 00:00:00.000'
        AND line_item_usage_start_date < TIMESTAMP '2020-01-16 00:00:00.000'
GROUP BY line_item_product_code, line_item_resource_id, line_item_currency_code

Adding costs to the email

Every morning my system prepares a report with important information from the previous day: active users, support statistics, and application health. I'm excited to add "application cost" to that list! Here's the relevant code:

class GatherAwsCosts < ApplicationService
  attr_reader :start_response
  attr_reader :wait_response
  attr_reader :finish_response

  attr_reader :csv
  attr_reader :costs

  def initialize(date, bucket_name, wait_delay: 1.second)
    @date = date
    @bucket_name = bucket_name
    @wait_delay = wait_delay
  end

  def call
    start_query_execution
    wait_for_query_to_finish
    download_results
    process_results
  end

  private

  def start_query_execution
    @start_response = athena.start_query_execution({
      query_string: sql_statement,
      query_execution_context: {
        database: 'billing',
      },
      result_configuration: {
        output_location: "s3://#{@bucket_name}/billing/results/",
      },
      work_group: 'primary',
    })
  end

  def wait_for_query_to_finish
    loop do
      sleep @wait_delay
      Rails.logger.info "waiting for Athena query to finish"

      @wait_response = athena.get_query_execution(
        query_execution_id: @start_response.query_execution_id,
      )
      status = @wait_response.query_execution.status

      case status.state
      when 'SUCCEEDED'
        @finish_response = @wait_response
        break
      when 'QUEUED', 'RUNNING'
        # continue waiting
      else
        raise "Unhandled query execution state: #{status.state.inspect} (#{status.state_change_reason})"
      end
    end
  end

  def download_results
    bucket = s3.bucket(@bucket_name)

    obj_url = @finish_response.query_execution.result_configuration.output_location
    obj_key = obj_url.split('/')[3..-1].join('/') # remove s3://bucket_name
    s3_object = bucket.object(obj_key)

    @csv = s3_object.get.body.read
  end

  def process_results
    @costs = []
    CSV.parse(@csv, headers: true) do |row|
      cost = row['line_item_blended_cost'].to_f
      next if cost < 0.01 # ignore costs that are less than a penny

      @costs << OpenStruct.new(
        product: row['line_item_product_code'],
        resource: row['line_item_resource_id'],
        cost: cost.round(2),
        currency: row['line_item_currency_code'],
      )
    end
    @costs
  end

  private

  def aws_credentials
    {
      access_key_id: Rails.application.credentials.dig(:aws, :access_key_id),
      secret_access_key: Rails.application.credentials.dig(:aws, :secret_access_key),
      region: Rails.application.credentials.dig(:aws, :region),
    }
  end

  def s3
    @s3 ||= Aws::S3::Resource.new(aws_credentials)
  end

  def athena
    @athena ||= Aws::Athena::Client.new(aws_credentials)
  end

  def sql_statement
    from = @date.to_s
    to = (@date + 1.day).to_s

    %(
      SELECT line_item_product_code,
               line_item_resource_id,
               line_item_currency_code,
               SUM(line_item_blended_cost) as line_item_blended_cost
      FROM billing
      WHERE line_item_blended_cost > 0
              AND line_item_usage_start_date >= TIMESTAMP '#{from} 00:00:00.000'
              AND line_item_usage_start_date < TIMESTAMP '#{to} 00:00:00.000'
      GROUP BY line_item_product_code, line_item_resource_id, line_item_currency_code;
    )
  end
end

# app/mailers/reports_mailer.rb
class ReportsMailer < ApplicationMailer
  def morning_report
    @date = Time.zone.yesterday
    @aws_costs = GatherAwsCosts.call(@date, 'your_bucket')
  end
end

<!-- app/views/reports_mailer/morning_report.html.erb -->
<table>
  <thead>
    <tr>
      <th>Product</th>
      <th>Resource</th>
      <th>Cost</th>
    </tr>
  </thead>
  <tbody>
    <% @aws_costs.each do |line_item| %>
      <tr>
        <td><%= line_item.product %></td>
        <td><%= line_item.resource %></td>
        <td><%= line_item.cost %> <%= line_item.currency %></td>
      </tr>
    <% end %>
  </tbody>
  <thead>
    <tr>
      <th>Total</th>
      <th></th>
      <th><%= @aws_costs.sum(:cost) %> <%= @aws_costs.map(&:currency).uniq.join(' ') %></th>
    </tr>
  </thead>
</table>

Wrap-Up

Now every morning I get a complete list of the previous day's AWS expenses. Eventually I may drop it back to every week or two, but I'm not tired of studying the fluctuations yet. Here's some ideas for future work:

Replace my EC2 instace resource IDs with its name via #describe_instances
Graph the per day costs in a monthly report
Turn this idea into a reuseable library