We need to enable customizing (add/delete) packages in fedora-atomic-docker-host.json. And make sure that the user should be able to use own customized OSTree.

We may want to remove reboot from Playbook as we do not have any task to perform after that.
Though having reboot in playbook makes our work fully automated but it ends up showing an ssh-connection error which doesn't look good to the user.

TASK [Perform a reboot] ********************************************************
fatal: [192.168.121.34]: UNREACHABLE! => {"changed": false, "msg": "Failed to connect to the host via ssh.", "unreachable": true}
 [WARNING]: Could not create retry file 'rebase.retry'.         [Errno 2] No such file or directory: ''


PLAY RECAP *********************************************************************
192.168.121.34             : ok=3    changed=2    unreachable=1    failed=0   

We can ask user to do the reboot manually instead.

Dockerfile

FROM fedora:25
MAINTAINER Gerard Braad <[email protected]>

# Run update and install packages
RUN dnf update -y && \
    dnf install -y ansible && \
    dnf clean all

# Create working dir
RUN mkdir -p /workspace
ADD ./compose.yml /workspace

# Compose (result will be in /srv/repo)
RUN cd /workspace && ansible-playbook compose.yml

# Expose web server port, set working dir
EXPOSE 8000
WORKDIR /workspace
VOLUME /workspace

# Start web server
CMD ostree trivial-httpd -p 8000 /srv/repo

compose.yml

#!/usr/bin/env ansible-playbook
---
- hosts: localhost

  tasks:
  - name: Install list of required packages
    package: name={{ item }} state=installed
    become: yes
    become_method: sudo
    with_items:
    - git
    - rpm-ostree

  - name: Clean package cache
    command: >-
      dnf clean all
      
  - name: Create directory structure
    file: path="{{ item }}" state=directory mode=750
    with_items:
    - "/srv/repo"
    - "/workspace"

  - name: Initialize repository
    command: >-
      ostree --repo=/srv/repo init --mode=archive-z2

  - name: Clone Fedora-Atomic buildscripts
    git: >-
      repo=https://pagure.io/fedora-atomic.git
      version=f25
      depth=1
      dest=/workspace/fedora-atomic

  - name: Compose Fedora-Atomic tree
    command: >-
      chdir=/workspace/fedora-atomic/
      rpm-ostree compose tree --repo=/srv/repo ./fedora-atomic-docker-host.json

#  - name: Start webserver
#    command: >-
#      ostree trivial-httpd -p 8000 /srv/repo

The Dockerfile might not run as it creates new namespaces during the compose step, however it describes what is needed. Maybe a Vagrantfile would have been better. Anyways, this will result in a Atomic host tree that can be served over http as static assets by nginx or by ostree trivial-httpd. Retested this process at: https://gitlab.com/gbraad/byo-atomic-fedora, Source: https://github.com/gbraad/byo-atomic-fedora

The tree can be used on the Atomic host as:

sudo ostree remote add byo-atomic-fedora http://[hostname]:8000 --no-gpg-verify
sudo rpm-ostree rebase byo-atomic-fedora:fedora-atomic/f25/x86_64/docker-host
systemctl reboot

Hi @trishnaguha ,
first of all I must say this is great work here and in the blogs, really helpful! Thank you :)

I have a use case where I'd like to roll my own ostree and use that to provision bare metal machines.

A playbook for that wouldn't really be overly complex, I hope:
One could create and serve a Kickstarter file to tell Anaconda to directly install the OS from OStree. No need to rebase later. How to do that in an (almost) unattended way is already documented here. You then only need to have grub on the machine and alter it manually.
This could be fully automated with addition of a PXEboot server.

Is this something you'd be willing to support here in this repo? I will likely do some work on this soon and would happily upstream to this repo.

IMO we should separate the HTTP server from the main playbook. Since the user needs to figure out the IP Address of the HTTP server and replace the httpserver var https://github.com/trishnaguha/build-atomic-host/blob/master/vars/atomic.yml#L4 with it. It will be difficult to that at one go.

So may be we should start the HTTP server first..then run the all-in-one playbook?

We should include requirements.yml in httpserver.yml. The requirements should be installed first.

F25 ships OSTree 2017.5 now. trivial-httpd has been removed.

@gbraad gave the idea to convert all shell scripts to playbooks and include all the playbooks in one single playbook, for instance main.yml in order to achieve automate all the things.