Hive on MR3 on Kubernetes inherits most of the features available in Hive on MR3 on Hadoop such as:
- Fault tolerance
- MR3 shuffle handler
- Speculative execution
- HiveServer2 in shared session mode and in individual session mode
- LLAP I/O
- Elastic allocation of cluster resources (namely Pods)
- ContainerGroup schemes: per-vertex, per-map-reduce, all-in-one
- Auto parallelism
High availability is achieved by exploiting StatefulSet and ReplicationController objects of Kubernetes. Additional features specific to Kubernetes, such as dynamic work rebalancing, are also under development.
This section describes additional new features in Hive on MR3 on Kubernetes.