WIP: Compose for multi-container installation refactorized

Related issue: None

This Merge Request includes:

• Tests
• A changelog fragment (cf https://docs.funkwhale.audio/contributing.html#changelog-management)

Why

The official docker-compose works, but is written in some very questionable way; creating serious issues in security, portability, modularity, readability and maintenability of the entire stack. The installation is also not automated, but requires extra steps (excluding configuration that doesn't count) to initialize for the first time the instance.

What changed

There has been a lot of changes, I will list the most notable ones:

Network

• Before: (see the appendix)
  • the containers were all under the same network which gave them internet access;
  • the api and nginx (the reverse-proxy) services were exposing to the public (using the port directive), respectively the 5000/tcp and 80/tcp ports;
  • the ports of postgres and redis were not documented (no expose directive used).
• Now: (see the appendix)
  • There are three distinct networks with precise purposes:
    1. An internal network (no Internet access) containing all the services;
    2. A network with Internet access with only the services that really need it, namely: api and celeryworker; if you want the whole app without internet access (but still reachable via web browser) set NO_INTERNET_ACC to true;
    3. An optional external network to hook the nginx service only (in case you have a main reverse-proxy on your server).
  • the api and nginx (the reverse-proxy) services no longer expose their ports to the public, they are only documented with the expose directive;
  • the ports of postgres and redis services are now documented with the expose directive.

Environment variables

Info: read this and this on the official Docker docs to understand the difference between .env file and env files imported with env_file.

• Before:
  • All the environment variables for all services were mixed together inside the same *.env file (.env);
  • The .env file (which is automatically loaded by Docker Compose to be used within the .yml file), is also directly injected in all serivices with the env_file directive, exposing information to services that doesn't (and shouldn't) need it.
• Now:
  • The environment variables are now separated in different files depending on the service or set of services that does need them; the .env file now has only variables concerning the configuration at the stack level, not service, except for the FUNKWHALE_HOSTNAME variable which is passed inside the containers via the environment directive.
  • The env_file is now used only when needed and import *.env files containing variables only needed for the target service.

Other

• Configurable paths for the volumes mounting point inside the containers has been removed, it's something that has no sense to exist and unnecessarily complicate things.
• Has been added the possibility to configure the restart and container_name directive on all the containers.
• Has been added the possibility to configure the tag on all services and image on some (nginx, postgres, redis).
• In the funkwhale.template has been removed all the unnecessary variables and replaced with static paths which reflects the static volume mountpoints I set for the service.
• The funkwhale_proxy.conf and funkwhale.template has been moved to a dedicated subfolder called nginx.
• In case the system administrator has intention to hook the nginx service to the network of its main reverse-proxy which happens to have a companion (for example docker-gen) I've set some variables with dummy values already in the nginx.env file.
• The directives in the docker-compose.yml file now are no longer sorted randomly; the order is the same for all and follows a criteria that (arguably) makes more sense.
• Except for DJANGO_SECRET_KEY, FUNKWHALE_HOSTNAME and REVERSE_PROXY_NETWORK which are mandatory variables to manually set, all the optional variables are left empty in the *.var files by default and has a correspondent default value defined in the docker-compose.yml file (except for funkwhale.env).
• postgres service image tag has been upgraded from 11 to 12.4 and redis from to 5 to 6.

What Didn't Change

• The overall stack structure remained the same, with the same container images.
• The name of the services (but they should be changed to a more generic ones, like: web, api, db, cache, etc...).
• The extra manual steps necessary to initialize for the first time the app
• The extra manual steps necessary to partially initialize the database in case to upgrade to a new version (tables or columns has been added/removed)
• The extra manual steps in case of bump to a new version of the service that is not backward compatible (for example posgres likes to break compatibility at every major release).

Appendix

Network Before

Network Now