BigData / Apache StreamPark Interview questions
What is the difference between Kubernetes Application mode and Kubernetes Session mode?
The distinction mirrors the YARN case, but on Kubernetes the isolation boundary is the pod rather than the JobManager process alone.
- Kubernetes Application mode — StreamPark deploys a dedicated Flink cluster (JobManager and TaskManager pods) for a single job. When the job ends, its pods are torn down. This gives clean CPU and memory isolation between jobs, since each job's pods are separate Kubernetes resources.
- Kubernetes Session mode — A Flink cluster is started once as a long-lived session, and multiple jobs are submitted into it, sharing the same JobManager and TaskManager pods.
Because Kubernetes already gives strong pod-level isolation and elastic scheduling, teams running on K8s often lean toward Application mode even more readily than on YARN, since the per-job overhead is smaller relative to the isolation gained.
In Kubernetes Application mode, what happens to pods when the job ends?
What do multiple jobs share in Kubernetes Session mode?
More Related questions...
