Okera can be installed co-located or separately. Both models are supported. The model you pick depends on the following:
- If you're deploying in the data center where all services are collocated on a set of physical machines, Okera can be deployed on the cluster, collocated with the other services.
- If you're deploying in the cloud and have different kinds of access frameworks (Spark, Presto, Python) accessing data via Okera but not necessarily all collocated on the same machines, Okera can be deployed standalone as an independent service.
- With EMR, you can deploy Okera collocated with the EMR nodes.
More details of the various deployment models can be found in the Okera deployment reference architecture document