Category: Technology

In part 2 we coded our PDF generator as a background job. But the PDF is still being stored on the local file system. Let’s store it in S3 instead and give our users a URL so they can download it.

First let’s add the AWS SDK gem to our Gemfile:

gem "aws-sdk"

Let’s define environment variables for our AWS credentials:

AWS_ACCESS_KEY_ID=abc
AWS_SECRET_ACCESS_KEY=secret

Next we’ll modify our background job to connect to S3 and upload our PDF file instead of saving it to the local file system:

class PdfJob < ActiveJob::Base
  def perform(html)
    pdf = WickedPdf.new.pdf_from_string(html)
    s3 = AWS::S3.new
    bucket = s3.buckets['my-bucket'] # replace with your bucket name
    bucket.objects['output.pdf'].write(pdf)
  end
end

Nice! But how do we enable our users to download the file? S3 has several options for this. One option would be to make the bucket publicly accessible. The downside to this approach is that it would allow anyone to download any PDFs stored in the bucket, regardless of who originally uploaded them. Depending on what kind of data is being included in the PDFs, this could be a bad idea.

A better option is to generate a temporary URL. This URL can be given to a user so they can download the file, but the URL is only usable for the period of time we specify. This reduces the likelihood that the PDF will be exposed publicly. Here’s how it’s done:

class PdfJob < ActiveJob::Base
  def perform(html)
    # ...
    obj = bucket.objects['output.pdf'].write(pdf)
    url = obj.url_for(:get, expires: 3.minutes.from_now).to_s
  end
end

Looks good. But how do we get this URL back to the user? The background job is asynchronous so it’s not like we can generate the PDF and return the string to the user all in the same HTTP request.

A simple approach is to write the URL back into the database. Let’s introduce a new user param and update the user with the URL (this assumes the column exists on the users table):

class PdfJob < ActiveJob::Base
  def perform(html, user)
    # ...
    url = obj.url_for(:get, s3_url_options).to_s
    user.update_attribute(:pdf_url, url)
  end
end

Now that the URL is available in the database, we can display it on the user’s profile page.

If we want to get even fancier we can write some JavaScript that’s executed immediately after the user requests a PDF. This script would periodically poll an Ajax endpoint in our app to determine if the URL has been written to the users table yet. When it detects the URL, it would redirect the user to the URL. This would make the PDF generation process seamless from the user’s perspective.

An example in jQuery might look something like this:

function poll(btn) {
  $.get("http://www.our-app.com/users/123/pdf_url", function(data) {
    if (data.length > 0) {
      window.location = data;
    } else {
      setTimeout(function() { poll(btn) }, 2000);
    }
  });
}

Our controller action might look like this:

class UsersController < ApplicationController
  def pdf_url
    user = User.find(params[:id])
    render text: user.pdf_url
  end
end

And there you have it. I hope this gave you a good idea of just how easy it can be to generate PDFs in a background job. If your site isn’t getting much traffic, it’s probably not worth going this route. But if it’s a popular site (or you expect it to be one day) it would be well worth investing the time to background this process. It’ll go a long way towards keeping your HTTP response times short, and your app will feel much snappier as a result.

In part 1 we learned why backgrounding is important. Now let’s dive into some code.

First things first. Add wicked_pdf and delayed_job to your Gemfile:

gem "wicked_pdf"
gem "delayed_job"

Now we can generate a PDF from inside our Rails app with this simple command:

html = "<strong>Hello world!</strong>"
pdf = WickedPdf.new.pdf_from_string(html)
IO.write("output.pdf", pdf)</pre>

You’ll notice that the more complex the HTML, the longer it takes wicked_pdf to run. That’s exactly why it’s important to run this process as a background job instead of in a web server process. A complex PDF with embedded images can take several seconds to render. That translates into several seconds of unavailability for the web process handling that particular request.

Let’s move this code into a background job:

class PdfJob < ActiveJob::Base
  def perform
    html = "<strong>Hello world!</strong>"
    pdf = WickedPdf.new.pdf_from_string(html)
    IO.write("output.pdf", pdf)
  end
end

Now we can queue the background job from a Rails controller like this:

class PdfController < ApplicationController
  def generate_pdf
    PdfJob.perform_later
  end
end

The only problem is, our job isn’t doing anything particularly interesting yet. The HTML is statically defined and we’re writing out to the same file each time the job runs. Let’s make this more dynamic.

First, let’s consider the HTML we want to generate. In a Rails app, the controller is generally responsible for rendering HTML from a given ERB template using a specific layout. There are ways to render ERB templates outside controllers, but they tend to be messy and unwieldy. In this situation, it’s perfectly reasonable to render the HTML in the controller and pass it along when we queue a job:

class PdfController < ApplicationController
  def generate_pdf
    html = render_to_string template: "my_pdf"
    PdfJob.perform_later(html)
  end
end

This assumes an ERB template named “my_pdf.erb” exists and contains the HTML we want to convert into a PDF. Our method definition within our background job then becomes:

class PdfJob < ActiveJob::Base
  def perform(html)
    pdf = WickedPdf.new.pdf_from_string(html)
    IO.write("output.pdf", pdf)
  end
end

delayed_job actually persists the HTML passed to the job in a database table so the job can retrieve the HTML when it gets executed. Since the job is executed asynchronously, the HTML has to be stored somewhere temporarily.

So far, so good. The job will generate a PDF based on the HTML rendered in the controller. But how do we return this PDF back to the user when it’s ready? It turns out there are a variety of ways to do this. Saving the PDF to the file system in a publicly accessible folder is always an option. But why consume precious storage space on our own server when we can just upload to Amazon S3 instead for a few fractions of a cent?

What’s nice about S3 is that it can be configured to automatically delete PDFs within a bucket after 24 hours. Furthermore, we can generate a temporary URL to allow a user to download a PDF directly from the S3 bucket. This temporary URL expires after a given period of time, greatly reducing the chance that a third party might access sensitive information.

Next week I’ll demonstrate how to integrate S3 into our background job using the AWS SDK.

What do you get when you combine the slick PDF generation capabilities of wicked_pdf with the elegance and efficiency of delayed_job? A high performance way to convert HTML pages into beautiful PDF documents.

I’ve been leveraging wicked_pdf to generate high school transcripts from my SaaS app, Teascript, since 2009. Prior to that I had been using Prawn which ultimately proved to lack the flexibility I needed to produce beautiful PDFs.

wicked_pdf converts HTML pages into PDF documents using WebKit, the engine behind Apple’s Safari browser (among others). For the past few years, Teascript produced PDFs without any kind of backgrounding in place. This meant that if someone’s PDF took an unusually long time to generate, they were tying up a web server process for that entire duration.

If multiple users generated PDFs simultaneously, it might prevent other visitors from accessing the site. Not good. Furthermore, if the PDF generation process exceeded the web server’s default timeout, the user might not ever get the PDF, just an error page.

Any time your web app integrates with a third party API or a system process, it’s a viable candidate for backgrounding. delayed_job to the rescue. By offloading the long-running processes onto background workers, we free our web server to do what it’s best at: serving static HTML and images.

Backgrounding isn’t a silver bullet, though. It introduces added complexity into the app, making it more vulnerable to failures. This requires writing additional code to handle these failure scenarios gracefully. But at the cost of this added complexity, we can ensure our web server stays fast and lean while our users still get the pretty PDF they want.

Next week we’ll dive into some actual code. I’ll demonstrate how to integrate wicked_pdf with delayed_job and hook the entire thing up to your Rails app. Don’t touch that remote.