Skip to content

Add environment variables for client OS/arch and client user UID/GID #11820

Open
@arnested

Description

Description

When using Docker Compose for development setup, you often need to tweak the configuration to vary on Linux, macOS, and Windows. And some services you would like to run with the same UID/GID as your current client user.

Our developers use both Linux and Mac (and we have one on Windows as well).

Currently, we cannot vary our Docker Compose files to take the platform differences into account.

To work around that, we have tried to introduce shell scripts that need to be run before using Docker Compose so that it can add .env files or compose.override.yaml. And we tried using conventions like “only a minority uses Linux, and they are quite experienced, so if they export their UID we can use it if set and default to 501 otherwise”.

Neither of those solutions are elegant or error proof, and they remove some intended flexibility of env files and override configs.

To help with that, I suggest introducing four new environment variables while parsing the configuration files:

  • COMPOSE_CLIENT_OS: set to Go's runtime.GOOS
  • COMPOSE_CLIENT_ARCH: set to Go's runtime.GOARCH
  • COMPOSE_CLIENT_UID: set to the current users UID
  • COMPOSE_CLIENT_GUID: set to the current users GID

This way, we can now have a Docker Compose setup like this:

compose.yaml:

services:
  php:
    image: php
    volumes:
      - .:/code
    user: ${COMPOSE_CLIENT_UID}:${COMPOSE_CLIENT_GID}
  web:
    image: apache
    extends:
      file: compose.${COMPOSE_CLIENT_OS}.yaml
      service: web

compose.linux.yaml:

services:
  web:
    environment:
      VIRTUAL_HOST: mysite.local

compose.darwin.yaml:

services:
  web:
    environment:
      VIRTUAL_HOST: mysite.docker

The change in #11821 is rather simple. But I'll acknowledge my lack of familiarity with the docker Compose code base and therefore there could be unintended side effects, and maybe it fits better into other parts of the code base. Possibly, it also needs test coverage. And as usual, we can discuss one of the two hard problems: naming things (there might be better names for the environment variables).

At least, I hope this can be given some thought. I know it would make our setups more elegant.

Metadata

Assignees

No one assigned

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions