The workflow I use to create, publish, and update this blog is discussed in three of my earlier posts, namely docker-bootstrap Workflow, Building This Blog, and Developing This Blog. This workflow works nicely in the case of this blog, but my daughter and I created another site, Visualizing Abolition and Freedom, frequently referred to as simply VAF, where the same workflow doesn’t quite work. The problem, I believe is with the manner in which I attempted to obtain TLS certs from Let’s Encrypt.

In a nutshell, information security policy here at Grinnell College dictates that a site/server be scanned for vulnerabilities and deemed “safe” before we open the firewall to allow traffic from outside the campus network. Prudent practice indeed. However, Let’s Encrypt is one such agent and their HTTP-01 Challenge that I use needs access to a site in order to validate any certificate requests I make. So we sort of have a chicken/egg, Catch-22 scenario… I can’t get a trusted cert until the site/server is secure and open, and I can’t secure or open the site until I have a trusted cert.

The simple solution is to initially use Let’s Encrypt’sACME staging environment” to obtain a temporary, un-trusted cert. Once that cert is in place we run vulnerability scans and make necessary changes until the only remaining vulnerability is the un-trusted cert itself. When we arrive at that point the firewall is configured to accept traffic from off-campus, and we apply for a new, valid, and trusted cert from Let’s Encrypt’s production environment.

Simple. But in the case of VAF it doesn’t work, yet, with my workflow. If you examine the diagram I created for the docker-bootstrap Workflow post on May 15, 2019, you’ll see that deploying a site into production is really 2-step process:

  • First the server is prepared by “initializing” it via docker-bootstrap’s ./init command. This command creates services for Traefik, Watchtower, and Portainer. This step is performed only once, no matter the number of sites added to the server in each Step 2. I believe all of the ACME cert specifications are specified in this step as part of Traefik’s configuration; with the URL of each individual site added later in Step 2, as the sites themselves are added. Step 1 knows nothing of the individual sites, like VAF and this blog, that may follow.

    Typical ACME specifications included in Step 1 are:

    --acme \
    --acme.caserver="" \
    --acme.acmelogging \
    --acme.dnschallenge=false \
    --acme.entrypoint="https" \
    --acme.httpchallenge \
    --acme.httpChallenge.entryPoint="http" \
    --acme.onhostrule=true \"/root/acme.json" \"" \
  • The second step is repeated once for each site to be launched on the server, and it specifies a docker container run... command like this VAF example:

    docker container run -d --name vaf \
      --label traefik.backend=vaf \
      --label \
      --label "" \
      --label traefik.port=80 \
      --label com.centurylinklabs.watchtower.enable=true \
      --network traefik_webgateway \
      --restart always \
      --label acme.caserver="" \

Part of the problem here, I believe, is that I’m using a Traefik container built from “scratch”, so there’s no shell in the image, which means I can’t open a terminal inside the container to see what’s happening.

Exactly when is each cert generated?

That is the question I’m struggling with. It would seem impossible for certs to be generated in Step 1 when we haven’t even identified what an individual site’s URL will be yet. Still, all of the ACME parameters that control cert creation, including the specification of acme.caserver are specified in Step 1 as part of the Traefik configuration.

Are the ACME parameters in Step 1 buffered for later use in Step 2?

If that is what’s happening, how do I switch a site from using the “staging” CA environment to using the “production” CA?

Can I provide an acme.caserver label as part of Step 2 instead of Step 1?

Maybe the answer is specifying the “production” environment in an acme.caserver label during Step 2? Something like this:

docker container run -d --name ${NAME} \  
  --label acme.caserver="" ...

I’m going to put this last notion to the test now. Wish me luck, and I’ll report back here in a few minutes.

Nope. Still don’t have a trusted cert for VAF.

Next step I believe is to wipe the server clean, pull an :alpine image of Traefik (so I can shell in and see more detail), then try to rebuild it all piece-by-piece.

And that’s a wrap. Until next time…