We don't have a place that is consistent enough to test performance related changes, other than on production. We need a cluster, virtualized or bare-metal (TBD), that we can use to run performance tests against.
This cluster could (probably) also be used to do nightly testing that would get us closer to T281: True code pipeline.
- Set up a parallel OpenStack cluster that is independent of WMF Labs infrastructure
- This could probably "simply" reuse our Beta Cluster puppet configs
- Set up a cluster of real hardware