Release Configuration

How to configure your release

There are two forms of configuration I will deal with here. One is configuration for the release process itself, the latter is handling application configuration for your release. The following custom release configuration is supported:

  • rel/sys.config - This is the configuration file the release will use in production. I would use config/config.exs or config/myapp.conf (if using conform) instead of this, but it’s there if you want it.
  • rel/vm.args - This file contains line-separated arguments that the Erlang VM will use when booting up. Provide your own here and it will be used instead of the default one. Make sure you provide values for sname and cookie though, or you won’t be able to connect to your release!
  • rel/relx.config - This file is used to provide configuration to exrm’s underyling relx dependency. See the documentation at relx’s GitHub page for more information on what you can provide here. The default one should cover 99% of cases, but if you need to tweak values, you can provide your own relx configuration, and setting the config values you care about. You do not need to provide the entire configuration, as your customizations will be merged with the defaults exrm uses.

Elixir has support for providing configuration using Elixir terms in a config/config.exs file. While this is perfectly usable, it’s not very simple for your operations group to work with, and generally contains no useful documentation on what each setting is for or what they do. To help make configuration much more easy and maintainable, exrm bundles a dependency called conform. It is optional to use, but is there if you desire to use it.

Using Conform with Exrm

Conform relies primarily on two files: a <project>.schema.exs file, and a <project>.conf file. The .conf file is where you will configure your app, and the .schema.exs file is where you define what configuration is available in the .conf, and how it is translated to the final sys.config that your release loads up at runtime.

Conform itself has the best documentation on how to work with these files, and to see an example app which makes use of this, check out the exrm-test project.

Here’s a quick rundown on how it works. You probably already have a config.exs file, and if you don’t that’s fine, it’s not required. If you do have one already, you can compile your project and run mix conform.new to generate the conform schema from your current configuration. If you don’t have one, check out the conform README on how to create one. Once you have the schema file in your config directory, you can work off the definitions generated from your current config, and/or start adding definitions for config settings you wish to add.

Once your schema is all set, you can generate the default .conf file for your app using mix conform.configure. This will output a .conf file to config/yourapp.conf. This will be bundled with your release, and located in $DEPLOY_DIR/releases/$RELEASE_VER/myapp.conf per default (also it could be moved, using RELEASE_CONFIG_FILE or RELEASE_CONFIG_DIR environment variables). Your ops group can then do all their configuration in production via that file.

If you are wondering how that .conf file is usable by the VM, it’s very simple. When you run <deploy dir>/bin/<project> start, or any other command which boots your app, a conform escript is run which translates the .conf via the schema (also bundled with the release) to Elixir terms, that is then merged over the top of the sys.config which is also bundled with the release, and then saved over the top of the existing sys.config. Once the escript has finished executing, your app is booted using that sys.config file, and everything carries on like normal.

NOTE: Your config/config.exs file is still converted to the sys.config which is bundled with the release. If you wish to hide settings from your end users, put them in there, and remove the definitions for them from your schema file. The sys.config is merged with the configuration which is defined in the .conf, so your settings will still be applied, they just won’t be exposed for end users.

You can also change the directory of all your configuration files sys.config, vm.args and <project>.conf using RELEASE_CONFIG_DIR or only for conform config <project>.conf using RELEASE_CONFIG_FILE system environments like this:

RELEASE_CONFIG_DIR=/some_path_to_configs bin/<project> start

or

RELEASE_CONFIG_FILE=/some_path_to_configs/<project>.conf bin/<project> start

So you can have persistent configuration for your application.

The configs placed in $DEPLOY_DIR/releases/$RELEASE_VER will be used as persistent default configs. They will be used by first release start and placed in $DEPLOY_DIR/releases/$RELEASE_VER/running-config if no RELEASE_MUTABLE_DIR defined. It is also possible to move the running-config, logs and erl_pipes using RELEASE_MUTABLE_DIR system environment. The idea is to hold persistent and non-persistent data separately.

NOTE: If not using conform, and relying on config.exs, you cannot use dynamic code which relies on the runtime environment, i.e:

config :myapp,
  foo: System.get_env("FOOBAR")

The reason for this is that the Erlang VM uses sys.config for configuration, and sys.config can only contain static terms, not function calls or other dynamic code. When your config.exs is evaluated and converted to sys.config, the dynamic code in config.exs is executed, evaluated, and the result is persisted in sys.config. If you are relying on such things as environment variables in config.exs, the value stored in sys.config will be the value of those variables when the build was produced, not their values when the release is booted, which is almost certainly not what you intended. When running your app with iex -S mix or mix run --no-halt, the way configuration is evaluated is different, as Mix will load the config from config.exs, and overwrite whatever is in the default configuration. As neither Mix, nor your config.exs is present in a release, this is not possible. If you need to load configuration from the environment at runtime, you will need to do something like the following:

my_setting = Application.get_env(:myapp, :setting) || System.get_env("MY_SETTING") || default_val