Can't connect to server after using docker run

[cthulu]

🪲 Description

I'm following the instructions in doc/self-hosting.md to first build the Docker image (which works fine) and then run the container using docker run --rm -p 8080:80 --name shields shields.

It looks like the container hangs forever:

$ docker run --rm -p 8080:80 --name shields shields
raven@2.6.4 alert: no DSN provided, error reporting disabled
Configuration:
{ bind: { address: 'localhost' },
  metrics: { prometheus: { enabled: false } },
  ssl: { isSecure: false },
  cors: { allowedOrigin: [ 'http://localhost:3000' ] },
  persistence: { dir: './private' },
  services:
   { github:
      { baseUri: 'https://api.github.com/',
        debug: { enabled: false, intervalSeconds: 200 } },
     trace: false },
  profiling: { makeBadge: false },
  cacheHeaders: { defaultCacheLengthSeconds: 120 },
  rateLimit: false,
  handleInternalErrors: false,
  fetchLimit: '10MB' }
0305092655 FsTokenPersistence configured with /usr/src/app/private/github-user-tokens.json
0305092655 Server is starting up: http://localhost/

I think it expects a response from Gatsby or something (I'm not a Node expert).

If I run the app manually using npm start it works, but I'd like to containerize it.

[paulmelnikow]

What is the expected behavior?

[calebcartwright]

I haven't checked today, but my scheduled job successfully built the image and then ran it on Friday March 1st. That was with commit 56e71d7

[cthulu]

@paulmelnikow the expected behavior is that http://localhost:8080 will be exposing port :80 of the container and having shields running. However I get connection refused as the container starts but the application startup never finishes - I even have to manually kill the container docker kill shields.

I have no issues with running any other containers (docker engine 18.0.9.2, Mac).

When running the npm start (not in Docker, just locally), the startup continues:

[server] 0306104015 Server is starting up: http://localhost:8080/
success open and validate gatsby-configs — 0.008 s

In container I'm stuck on the line 0306104216 Server is starting up: http://localhost/

[cthulu]

I've tried also 56e71d7 with the same result.

Maybe the problem is the CMD node server. Shouldn't it be npm start?

[paulmelnikow]

npm start runs two processes, one of which is the Gatsby development server. In production you don't need the Gatsby server; instead the Dockerfile triggers a static build which does not need to be run in production. This gets served by the same node process that handles the badges.

I think the process inside the container needs to bind to all interfaces instead localhost. If that isn't the issue, then the port is wrong. Maybe it's supposed to be -p 80:8080 instead of -p 8080:80?

It would be helpful to document all this in self-hosting.md – would definitely welcome contributions there!

[paulmelnikow]

I'm guessing the only way the Dockerfile works today is by to pass an alternative command to run, like node server 80 0.0.0.0. I opened #3173 so it defaults to using 0.0.0.0 (all interfaces) in production, and set NODE_ENV=production in the Dockerfile, which should make the default command usable.

The reason the process isn't receiving a signal is that you need docker run --init, which tells Docker to runs the entrypoint inside a subprocess that properly handles signals. This has the added benefit of speeding up shutdown when using docker stop.

It would be great to document which CLI arguments are commended. I'm not sure what docker --rm does or whether --name shields should be in there, though I have gotten things working by using the changes from #3173 with docker run -p 8080:80 WHATEVER_IMAGE_ID.

[calebcartwright]

rm does some clean up stuff when the container's stopped if i recall correctly.

I'm personally running a docker run -d ... with the image, but I'm also passing in a lot of environment variable args to the container, including the bind address which I suspect is why mine was still working

[paulmelnikow]

Ah, right, I forgot you could set the bind through the env. Seems good to have it an entrypoint that works by default!

rm does some clean up stuff when the container's stopped if i recall correctly.

Sounds worth including then. 👍

[calebcartwright]

using the latest Dockerfile from master, I'm able to build the image, and run the container (on Ubuntu 18.04) successfully (the only environment variables I'm passing to the container are tokens/credentials for various badge services)

[paulmelnikow]

Would welcome a PR for the docs! Let's leave this open until that's taken care of.

[cthulu]

I can confirm, it's working now. Great work and thanks!

[calebcartwright]

@paulmelnikow - what's in scope for the doc updates? The self-hosting doc section for docker currently shows:

$ docker run --rm -p 8080:80 --name shields shields
# or if you have shields.env file, run the following instead
$ docker run --rm -p 8080:80 --env-file shields.env --name shields shields

So are you thinking the doc should include details/descriptions about the args shown (port mapping, clean up, etc.) and/or why the guide suggests those?