A python-based multi-node devstack tool

Made by people that care.

How to use it

So a basic understanding of devstack is useful but not needed, what is required is a cloud that you can get virtual machine from (and currently we assume centos7 so it may not work elsewhere without changes).

Once you have that ensured the the following should be done:

1. Make a virtualenv, install the requirements.txt file into that.

2. Ensure that you have exported the following OS_AUTH_URL and OS_USERNAME and OS_TENANT_NAME and most importantly OS_PASSWORD.

   For example:

   $ env | grep OS_ | grep -v OS_PASS
   OS_AUTH_URL=https://openstack.int.godaddy.com:35357/v2.0
   OS_USERNAME=jxharlow
   OS_TENANT_NAME=user-jxharlow
   

3. Run ./builder.sh create and watch the steps trigger that once successful will have build yourself a cluster.

   • If it fails after the stack.sh commands have started to run ensure to run ./builder.sh destroy before re-running because those commands are not idempotent (the other actions are).

What it does (during create)

• Creates a cloud instance (configuring itself via the mechanisms supported by the shade library).
• Creates/opens a local sqlite3 database (used for state tracking and data retention; primarily used for resuming, allowing for this program to be mostly crash-tolerant).
• Validates provided create arguments against the matched cloud (for example to ensure the image provided is actually found).
• Creates a desired instance layout (and saves it).
• Scans current cloud servers and sees if layout is satisfied (if not servers are spawned to match the desired layout).
• Performs (and records what was done and what was not) remote server commands on all matched (or spawned) servers to turn them into a multi-node devstack cloud.

What it does (during destroy)

• Creates a cloud instance (configuring itself via the mechanisms supported by the shade library).
• Creates/opens a local sqlite3 database (used for state tracking and data retention; primarily used for resuming, allowing for this program to be mostly crash-tolerant).
• Extracts prior servers from local sqlite3 database and destroys them (by whatever mechanism the underlying cloud performs such actions).

What is not done (yet)

• Creating a overlay network (so that the VMs spawned can connect and communicate, likely in a private only fashion).
  • Perhaps using something inspired by and/or like https://github.com/openstack-infra/devstack-gate/blob/master/functions.sh#L1105 ?
• Ensuring the hypervisors spun up are all connected together correctly and actually work.
• Making sure the cells (parent and child nova cell) actually works.
• Using neutron (vs nova-net).
• Making it easier to plug-in new projects (in some modular manner?).
• Removing various godaddy artifacts (metadata for example).

Examples

• https://asciinema.org/a/86254
• https://asciinema.org/a/86253

Get involved

Contributions are always welcome, find me on IRC or openstack-dev (ML) or in person or just open a pull request.