OpenStack Folsom Design Summit

Currently Atlas-LB has adapters for certain kinds of s/w and h/w Load balancers like Zeus and Netscalar. The adapters however, directly configure the LBs. If Atlas LB has to work with more complex data center topologies and complex business rules of different organizations, which have their own workflows for provisioning Load balancers, then it needs to be able to talk to the external workflow provisioning systems. This way, it is easy to design interoperability between Atlas, and other workflow systems. This proposal adds the capability for Atlas-LB to support a workflow load balancer provisioning system at the backend to which it could delegate LB management requests. It accomplishes this by having a new message listener which invokes an asynchronous workflow adapter in Atlas-LB which acts as a proxy to a workflow system. The current listener invokes the adapter synchronously and marks the loadbalancer/Vip as Active or in Error depending on whether the adapter returns successfully or not. For a workflow based system, since the adapter can not complete its task in a short/known amount of time (due to manual steps in the workflow), the adapter has to be asynchronous and the listener should not mark the loadbalancer/Vip as active even if the workflow adapter returns successfully. The successful return of the adapter is taken to mean only the successful creation of a request in the workflow system. (Session lead is Kiran Lonikar)