Open Stack

I ran a few tests yesterday to see how much the latest flattening [1]
efforts effect performance on the TripleO Heat templates. I only ran a
few of each type but the results looked towards a minor but positive
improvement in the speed with which the flattened templates get
processed via Heat.

Using an all-in-one standalone TripleO deployment I timed how long it
took for 'openstack tripleo deploy --standalone --output-only' to run
using the latest Heat binaries. I then compared the results switching
between t-h-t from master and rocky. On my laptop the results pointed
to about a 30-45 second improvement in wall time of 5 minutes or so. A
reasonably minor improvement but it is definately in the right
direction. The more complex number of services and roles you have the
greater this number is likely to be.

I don't think performance was the primary reason we undertook this
restructuring but I had several people ask about potential performance
improvements so I wanted to provide some feedback here.

Disclosure: In order to test rocky I had to make a few minor changes to
t-h-t in order to get the latest all-in-one to run that similarly.
Basically just backporting tools/process_templates.py and removing a
few invalid network variable references. Here is the branch [2] if you
are interested.

Dan

[1] 
https://blueprints.launchpad.net/tripleo/+spec/services-yaml-flattening
[2] https://github.com/dprince/tripleo-heat-templates/tree/rocky_test