Docker – How to use environment variables in docker-compose


I would like to be able to use env variables inside docker-compose.yml, with values passed in at the time of docker-compose up. This is the example.

I am doing this today with basic docker run command, which is wrapped around my own script.
Is there a way to achieve it with compose, without any such bash wrappers?

  hostname: $hostname
    - /mnt/data/logs/$hostname:/logs
    - /mnt/data/$hostname:/data

Best Answer

The DOCKER solution:

Docker-compose 1.5+ has enabled variables substitution:

The latest Docker Compose allows you to access environment variables from your compose file. So you can source your environment variables, then run Compose like so:

set -a
source .my-env
docker-compose up -d

For example, assume we have the following .my-env file:


(or pass them via command-line args when calling docker-compose, like so: POSTGRES_VERSION=14 docker-compose up -d)

Then you can reference the variables in docker-compose.yml using a ${VARIABLE} syntax, like so:

  image: "postgres:${POSTGRES_VERSION}"

And here is more info from the docs, taken here:

When you run docker-compose up with this configuration, Compose looks for the POSTGRES_VERSION environment variable in the shell and substitutes its value in. For this example, Compose resolves the image to postgres:9.3 before running the configuration.

If an environment variable is not set, Compose substitutes with an empty string. In the example above, if POSTGRES_VERSION is not set, the value for the image option is postgres:.

Both $VARIABLE and ${VARIABLE} syntax are supported. Extended shell-style features, such as ${VARIABLE-default} and ${VARIABLE/foo/bar}, are not supported.

If you need to put a literal dollar sign in a configuration value, use a double dollar sign ($$).

The feature was added in this pull request.

Alternative Docker-based solution: Implicitly sourcing an env vars file through the docker-compose command

If you want to avoid any bash wrappers, or having to source a env vars file explicitly (as demonstrated above), then you can pass a --env-file flag to the docker-compose command with the location of your env var file:

Then you can reference it within your docker-compose command without having to source it explicitly:

docker-compose --env-file .my-env  up -d 

If you don't pass a --env-file flag, the default env var file will be .env.

Note the following caveat with this approach:

Values present in the environment at runtime always override those defined inside the .env file. Similarly, values passed via command-line arguments take precedence as well.

So be careful about any env vars that may override the ones defined in the --env-file!

The BASH solution:

I notice that Docker's automated handling of environment variables can cause confusion. Instead of dealing with environment variables in Docker, let's go back to basics, like bash! Here is a method using a bash script and a .env file, with some extra flexibility to demonstrate the utility of env vars:

# Note that the variable below is commented out and will not be used:

# You can even define the compose file in an env variable like so:
# You can define other compose files, and just comment them out
# when not needed:
# COMPOSE_CONFIG=another-compose-file.yml

then run this bash script in the same directory, which should deploy everything properly:


docker rm -f `docker ps -aq -f name=myproject_*`
set -a
source .env
cat ${COMPOSE_CONFIG} | envsubst | docker-compose -f - -p "myproject" up -d

Just reference your env variables in your compose file with the usual bash syntax (ie ${POSTGRES_VERSION} to insert the POSTGRES_VERSION from the .env file).

While this solution involves bash, some may prefer it because it has better separation of concerns.

Note the COMPOSE_CONFIG is defined in my .env file and used in my bash script, but you can easily just replace {$COMPOSE_CONFIG} with the my-compose-file.yml in the bash script.

Also note that I labeled this deployment by naming all of my containers with the "myproject" prefix. You can use any name you want, but it helps identify your containers so you can easily reference them later. Assuming that your containers are stateless, as they should be, this script will quickly remove and redeploy your containers according to your .env file params and your compose YAML file.

Since this answer seems pretty popular, I wrote a blog post that describes my Docker deployment workflow in more depth: This might be helpful when you add more complexity to a deployment configuration, like Nginx configs, LetsEncrypt certs, and linked containers.

Related Question