[FLINK-21153] yarn-per-job deployment target ignores yarn options

• Type: Bug
• Status: Open
• Priority: Not a Priority
• Resolution: Unresolved
• Affects Version/s: 1.12.1, 1.13.0
• Component/s: Command Line Client, Deployment / YARN
• Labels: auto-deprioritized-major auto-deprioritized-minor auto-unassigned usability
• Assignee: Unassigned
• Reporter: Till Rohrmann
• Created: 26/Jan/21 16:48
• Updated: 02/Jan/22 10:39

Description

While looking into the problem reported in FLINK-6949, I stumbled across an odd behaviour of Flink. I tried to deploy a Flink cluster on Yarn and ship some files to the cluster. Only the first command successfully shipped the additional files to the cluster:

1) bin/flink run -p 1 --yarnship ../flink-test-job/cluster -m yarn-cluster ../flink-test-job/target/flink-test-job-1.0-SNAPSHOT.jar
2) bin/flink run -p 1 --yarnship ../flink-test-job/cluster -t yarn-per-job ../flink-test-job/target/flink-test-job-1.0-SNAPSHOT.jar

The problem seems to be that the second command does not activate the FlinkYarnSessionCli but uses the GenericCLI.

kkl0u, aljoscha, tison what is the intended behaviour in this case. I always thought that -m yarn-cluster and -t yarn-per-job would be equivalent.

Issue Links

• is related to: FLINK-15179 Kubernetes should not have a CustomCommandLine.
• is related to: FLINK-15294 Migrate CustomCommandLine to Executor abstraction