apache spark - Difference between YARN deployment modes - CLUSTER and CLIENT modes on the machine with a node manager -
for me difference in first case driver inside app master jvm whereas next above client mode configuration. different in 2 cases ?
spark deployment mode - client vs cluster
client mode
- it's interactive. if want job result (dynamic analysis) @ machine (client - driver program), client deployment mode quite useful.
- easier developing/debugging.
- end user have control driver program running
- always application: expose spark job launcher rest service or web ui.
cluster mode
- easier resource allocation (let master decide): fire , forget
- due fire-n-forget behavior, it's submit long, resource hungry job in cluster mode.
- monitoring driver program done master web ui other workers. (not client machine)
- stop @ end: once job finished, allocated resources freed automatically.
- in cluster mode, if driver program has issue, yarn restarts driver without killing executors.
- in client mode, yarn automatically kills executors if driver killed.
Comments
Post a Comment