BigData / Apache StreamPark Interview questions
What is the difference between YARN Application mode and YARN Session mode in StreamPark?
Both submit Flink jobs to a YARN cluster, but they differ in how many jobs share a JobManager.
- Application mode — StreamPark starts a dedicated JobManager per job, and the job's
main()runs inside that JobManager on the cluster rather than on the client. Each job is fully isolated: one job's failure or resource spike can't starve another job's JobManager. - Session mode — A long-lived YARN session cluster is started once, and multiple jobs submit into that same shared JobManager. This uses fewer resources overall since jobs share the JobManager, but a JobManager crash or resource contention on that shared session can affect every job running in it.
The trade-off is isolation versus resource efficiency, which is why production guidance generally leans toward Application mode despite its higher per-job overhead.
In Application mode, how many JobManagers does each job get?
What is the main risk of running many jobs in one YARN Session?
More Related questions...
