Rackspace's OnMetal launched in July as the first and only deployment of Ironic on the public cloud. Rackspace has since scaled Ironic to manage thousands of servers in a multitenant environment.
This talk will discuss the challenges we faced while deploying and scaling OnMetal. Ironic's architecture allows for a straightforward HA model and horizontal scaling, but there were (as always) performance issues that set our team back while growing our production environment. These issues came to life on both the Nova and Ironic side of the control plane.
This talk will also cover how we brought a new deploy model to Ironic that fixed some core architectural issues with the default deploy driver, while also allowing Ironic to do more interesting things with hardware that it manages.