YARN Application Lifespan
- Can be varied from few seconds to months.
- (Simplest) One application per job.
- (Second) One Application per workflow.
- Container can be reused.
- Cache intermedia data between jobs.
- Spark and Tez are examples.
- (Third) Long-running application shared by different users.
- Acts as coordination role.
- Apache Slide
- A long-running master for launching other applications on the cluster.
- Impala: always on ApplicationMaster.
- Proxy application.
- Reduce the overhead of starting a new ApplicationMaster.
TODO
- A practice in Slide
- How does the Proxy Application works in Impala?
- Is there any possible to implement a Proxy Application in Slide?