On October 15th 2020, we upgraded the Nectar Allocation System to implement quotas for the Container Orchestration Engine (COE).
The Allocation form now shows a new Container Orchestration Service in the Services section containing a quota on the number of COE clusters. When you request quota for clusters, the Dashboard will check that you have also requested quotas for the other resources that are required; i.e. Advanced Networking quotas for 3 load balancers, 2 floating ips, a network and a router, and at least 2 Compute Service instances. The results of these checks will be presented to you as warnings. The form gives you the option of ignoring the warnings and submitting anyway.
To make the transition as smooth as possible, we have pre-approved COE cluster quotas for all projects that are currently running COE clusters. If need extra quotas, please submit an allocation amendment request in the normal way.
In the few months since the introduction of the Nectar COE service (also known as Magnum), we have seen a number of cases where users have attempted to launch a COE cluster only to run into problems. Often the problem turns out to be that the user's project does not have quotas for various networking resources required for their cluster; i.e. "network", "router", "load balancer" and "floating IP" quotas. These failures are difficult to diagnose, and sometimes lead to users giving up on the COE service.
According to our analysis, the real problem is that (particularly) new users of this service are not aware of the interrelatedness of COE and other Nectar services. In particular, it is not at all obvious exactly which resources are needed to build a basic cluster. As a result, users often don't know that they need to request these quotas. This typically results in an extra "go around" with allocation amendment. We have concluded that the best way to avoid this is to make COE clusters a quota'd resource so that we can cross-check with the other quotas.