Arvados

Requirements:

• Primary configuration used by the playbook should come from config.yml, i.e. user creates an Arvados config and the installer figures out what needs to go where based on that. The installer might still process the config.yml to produce the final one that is sent to the nodes but the basic structure of the input should match config.yml.

We want to avoid creating additional configuration layers that just get turned into config.yml values.

"Don't force the user to reconfigure anything that is straightforwardly configured in config.yml"

However stuff that isn't adequately described in the current config.yml, such as cluster topography, should be expressed separately.

Should be able to reason about roles, e.g. node X is the load balancer node, so it needs to write a nginx config that refers to nodes Y and Z which are backend controller nodes.

Also, secret values should be kept in a separate file, so we have the option of keeping them in a secret store. This could be as simple as a separate "secrets" file that also has the same shape as "config.yml" and is merged with the main config.yml at build time.