Rework how install and run work internally

[vpavlin]

This is a post-Summit problem

Given an example of wordpress application:

wordpress-atomicapp
  ├── mariadb-atomicapp
  └── skydns-atomicapp

When install is called, atomicapp installs wordpress-atomicapp and, loads Nulecule and goes through graph. When it finds external dependency it recursively starts installation for this dependency with a destination path external/<dep-app-id>. Once the installation of dependency is finished, recursion ends and the graph cycle goes on.

Example:

atomicapp install wordpress-atomicapp
FOUND mariadb-atomicapp: 
    atomicapp install --destination external/mariadb-atomicapp mariadb-atomicapp
FOUND skydns-atomicapp:
    atomicapp install --destination external/skydns-atomicapp skydns-atomicapp

It works the same for run.

Problems

• Once install is finished for a dependency, atomicapp cannot access it's data easily
• We cannot easily inspect what a dependency provides (f.e. providers) and thus cannot give a user list of available providers or guess default provider
• It would be better if provider.deploy() happened from one "cycle" not from a recursively called instance cause that makes it hard to properly react if something fails.

Solution

Implement a Python class which will represent a Nulecule application (nulecule_base.py is sort of where to start) and

• Create object for the main app
• Create objects for all deps and reference them from main app object
• Rework install and run so that once all objects are created and filled with data, install/run gathers them to a single list and performs needed actions.

Comments?:)

[vpavlin]

This is needed for proper docker-compose provider

[rtnpro]

Fixed in https://github.com/projectatomic/atomicapp/tree/cdk2-beta3-rc1