Problem reading Enviroment variables when deploying with Docker

Phoenix Forum Questions / Help
, , ,
1

Hello Fellas,

I am working on a project that uses elixir 1.9 and phoenix 1.5.7 in an ubrella project.

Now in certain live views I need to access enviroment variables which I normally do through
System.get_env("AWS_BUCKET_NAME_IE")

The problem is that when I deploye through docker the variables that suppose to containing enviroment variable content are empty.

But wierdly enough (at least for me) if I log into the container. through

  1. docker exec -it {container_name} bash
  2. ./bin/appname remote

If I try to access the variable from the elixir shell I got, for example executing System.get_env("AWS_BUCKET_NAME_IE")
I get the expected results

The way I ussyally get the name of the variables in the live_view is
@bucket System.get_env("AWS_BUCKET_NAME_IE") and that always has an empty string as result when deploying with docker.

my Docker file looks like (app specific names are changed)

FROM bitwalker/alpine-elixir-phoenix:latest as builder

WORKDIR /app

ENV MIX_ENV=prod \
    SECRET_KEY_BASE=$SECRET_KEY_BASE \
    AWS_ACCESS_KEY_ID=$AWS_ACCESS_KEY_ID \
    AWS_SECRET_ACCESS_KEY=$AWS_SECRET_ACCESS_KEY \
    AWS_BUCKET_NAME=$AWS_BUCKET_NAME \
    AWS_ASSETS_BUCKET_NAME=$AWS_ASSETS_BUCKET_NAME \
    DATABASE_URL=$DATABASE_URL \
    SHELL=/bin/bash

# install hex + rebar
RUN mix do local.hex --force, local.rebar --force
          
COPY config config
COPY mix.* ./
COPY apps/app/mix.exs ./apps/app/
COPY apps/app_web/mix.exs ./apps/app_web/
          
RUN mix do deps.get --only $MIX_ENV, deps.compile
          
COPY . /app/
        
# build assets
WORKDIR /app/apps/app_web
    
RUN npm install --prefix ./assets/ ci --progress=false --no-audit --loglevel=error
RUN npm run deploy --prefix ./assets
RUN mix phx.digest
      
WORKDIR /app   

RUN PORT=4000 MIX_ENV=prod mix do compile, phx.swagger.generate, release

# prepare release image
FROM bitwalker/alpine-elixir:1.10.3
ARG VERSION=0.7.0
  
WORKDIR /app
COPY --from=builder /app/_build/prod/rel/app /app/
    
RUN ln -s ./app/lib/app_nameapi_web-${VERSION}/priv/static /app/static
ENV HOME=/app
  
EXPOSE 4000    
  
ENTRYPOINT ["./bin/app", "start"]

and then the docker-compose

services:
    app:
    image: eu-west-3.amazonaws.com/app/web:staging
    expose:       
      - "4000/tcp"             
    environment:               
      - AWS_ACCESS_KEY_ID=${AWS_ACCESS_KEY_ID}
      - AWS_SECRET_ACCESS_KEY=${AWS_SECRET_ACCESS_KEY}
      - AWS_BUCKET_NAME=${AWS_BUCKET_NAME}
      - DATABASE_URL=${DATABASE_URL}  
      - SECRET_KEY_BASE=${SECRET_KEY_BASE}
    env_file:
      - envfile
    volumes:
      - appweb_static_vol:/app/static

Any direction on where to look for the problem would be greatly appriciated !

2

Don’t know if it’s a typo but there’s an additional _IE in your elixir code.

3

Yes, it’s just an example the variable names are different.

4

I don’t recall how RUN works in docker but since that @module_attribute is being set at compilation time you need to make sure it’s available when you do mix compile/release

1 Like
5

So for example I can export them before issuing the docker build command?

6

You need to make sure that when running the task for building the release or compiling the app those values are accessible in the task.

2 Likes
7

That was exactly correct thank you for that really usefull comment.

8

You are setting the @bucket module attribute with whatever the AWS_BUCKET_NAME_IE contains during compiletime.

Please use config/release.exs or config/runtime.exs depending on your elixir version to configure the application environment and read it during runtime.

3 Likes
9

I’m not sure if “compile time configuration” is the proper way to go.

It requires a full rebuild including redeploy if the setting has to be changed.

1 Like
10

I usually prefer env vars readable at runtime too but being a docker container might also mean that you’re not changing them while it’s running.

11

True but for me usually compile time configuration for the most part is in files that I would want to keep in the revision control system so not the best part to keep secrets. Also what is the mechanism of elixir/phoenix to read enviroment variables during runtime ?

12

I have only used docker to generate releases and I basically used those releases in systemd, using the service definition to pass in env/env files to the release and from elixir just doing regular System.get_env. The only thing you can’t do in that case is to make module attributes out of the get_env calls as those will be baked during generation.

I think you can now do better with what @NobbZ mentioned, maybe this helps Configuration and releases - The Elixir programming language

1 Like