πŸššπŸ” a deploy pipeline for Cloudflare Workers Sites

Setting up a simple workflow with GitHub and Cloudflare Workers for this blog.

October 8, 2019
Workers , Cloudflare , blog , walkthrough

I converted this blog to Workers Sites a couple weeks ago and wrote about it here. The initial setup was pretty bare bones. I saved the repository locally without any version control and deployed directly to production. No one gets paged if I break this blog, so that was fine.

However, I did receive some questions on Twitter asking about how to integrate a deploy pipeline into this workflow. I decided to take some time this weekend and build out a staging flow and answer some of those questions. This work gave me an excuse to bring some maturity to this process, including a GitHub repo, which helps with situations like “what if I lose my laptop?”

Cloudflare Workers makes it easy to spin up a staging environment that mirrors production. I just need to deploy to a different subdomain. That deployment will use the same Workers and Workers KV stack as production and does not add any fixed cost to running the site since I don’t need to spin up a separate VM. However, that does leave me with a site that is effectively public.

I need a way to lock it down. I work at Cloudflare as the Product Manager for Cloudflare Access, a product that can help entire enterprises replace their VPN or, for this personal blog, lock down a staging URL. With Access, I can control who can reach the staging site by prompting users, including myself, to login with an identity provider. I’m excited to bring what I work on into this pipeline.

I have a few goals for this project:

  • Use GitHub for version control of my static site files
  • Use Wrangler environments for deploying to staging
  • Use Cloudflare Access to secure staging

⏲️Time to complete: ~1 hour (assuming you already completed the work in the last post).


Bringing GitHub to a static project

I have an admittedly quiet GitHub account. That said, the platform makes it easy to create a new repository; users can choose between public, which is shared to the world, or private. I’m going to keep the blog content repository private for the time being while I tune up this flow. To begin, I’ll create a new repository in the GitHub UI.

Git ignore

Before I push my files to that repository, I have some material that does not need to live in GitHub. I can use a .gitignore file to tell Git to exclude those files or folders.

blog-samrhea samrhea$ touch .gitignore

Inside of that file, I’m going to include the following paths:

/themes/hugo-cactus-themes
/public

Why?

Folder or File Why exclude?
/themes/hugo-cactus-themes This folder consists of the public theme I cloned for this project. The only modification I made was to replace my avatar image.
/public Contains the output of my Hugo build, which changes whenever I run hugo. As long as I have the remaining files in this repository and Hugo, I can generate this at any time.

Configuration and first commit

I have an empty GitHub repository, but a pretty complete local repository. I now need to take the static site content, past and present, and push it to GitHub.

blog-samrhea samrhea$ git init

Initializes a Git project in my local repository

blog-samrhea samrhea$ git remote add origin https://github.com/AustinCorridor/blog-samrhea.git

Configures the Git project to use my GitHub repository

blog-samrhea samrhea$ git add .

Once I have created my project and added my remote origin, all of the local files I have that are not included in .gitignore look like changes. I need to stage all of these.

blog-samrhea samrhea$ git commit

Commits the entire set of changes, which in this case is the current local repository.

blog-samrhea samrhea$ git push -u origin master

Pushes the changes to the GitHub repository.

I can now visit the link and confirm that the Hugo files for my personal blog are saved to GitHub.

Creating a staging environment

Hugo includes support to run a local web server of your project. The server updates each time you save a file, which makes it very convenient for reviewing changes in real-time in a browser alongside my text editor.

However, I also want to make sure that the Hugo output can be packaged and published by Wrangler to Cloudflare Workers without issue. To observe that, I want a real staging environment that mirrors production. In this case, the stack I need to mirror is Cloudflare and Cloudflare Workers. To get a staging environment, I need to deploy to the Internet. That gives me the benefit of running staging in the most equivalent setup to production possible. That opportunity also requires I find a way to limit who can reach staging, which I’ll do in the second topic in this section.

Wrangler environments

Wrangler supports deployment to multiple destinations using the wrangler.toml file and the commands you run. I’m going to create a new DNS entry in Cloudflare, “blog-staging.samrhea.com” that will serve as my staging environment. Now I need to tell Wrangler where staging lives. To do so, I’ll change my wrangler.toml file by adding the block in [env.staging].

account_id = "CF_ACCOUNT_ID"
name = "blog-samrhea"
type = "webpack"
route = "blog.samrhea.com/*"
workers_dev = false
zone_id = "CF_ZONE_ID"

[env.staging]
name = "blog-staging"
route = "blog-staging.samrhea.com/*"

[site]
bucket = "public"
entry-point = "workers-site"

βš™οΈ Setting variables. In my current namespace, I set my zone ID and account ID variables by running export CF_ZONE_ID="123" and expore CF_ACCOUNT_ID="456". With that set, I can keep my wrangler.toml file as it is above. I’ll leave those lines here to reduce confusion, but with the variables set you can leave them out altogether.

Now, I can run the following commands to build my project and push those to Cloudflare:

blog-samrhea samrhea$ hugo
blog-samrhea samrhea$ wrangler publish --env staging

⚠️ Don’t forget the Hugo base URL. In my config.toml file, I set a base URL for my project. When Hugo builds the site, it uses that to construct the links. This causes a problem when I want to preview my staging site since the links will take me to production. Instead, I need to update the baseURL value to the staging URL when I run hugo before publishing. This a clunky step; I’ll try and remove it down the road.

When I visit “blog-staging.samrhea.com” I can see the output of my current work in an environment that matches production 1-1 because, beyond Hugo, the only stack I need to use includes Workers and Workers KV.

Locking down staging with Cloudflare Access

Nothing I publish on this blog is confidential, but I do want to limit audience visibility into work-in-progress. Fortunately, Cloudflare offers a product that can protect subdomains or paths and only allow approved team members to reach them: Access.

Like I mentioned earlier, I’m the Product Manager for Access. I have written about Access on this blog before, as well as on the Cloudflare blog. I’ll try to be succinct here. With Access, I can force any visitor to the staging URL of my blog to first authenticate. In this case, I’ll just allow myself to view it, but I could add editors down the road.

Now, when I visit “blog-staging.samrhea.com”, I am presented with this prompt to login:

I’ll login with Google and, once authenticated, Access will redirect me to “blog-staging.samrhea.com”. I can review the site as it will appear in production.

The full picture

I now have the pieces for a full deployment pipeline. I’ve broken them out into the diagram below.

That pipeline gives me a contained staging environment that mirrors production and version control so I can save my work and track changes. Again, if a bad deploy took down this blog it wouldn’t constitute an incident, but this workflow helps me prevent that and better manage the site.

What’s next?

Automation. The 10 steps in that list are mostly manual. I want to get this down to one-click to staging and one-click to production. Most of that work is on the Hugo side with the baseURL requirement. I’ll try to tackle that next.