[BUG] Flyte spark tasks are not being executed with spark service account

[YmirKhang]

Describe the bug

Spark tasks are being executed with the "default" service account, which doesn't have the role binding to create spark executor pods. So the execution of the spark tasks fail, because only the driver pod is created and the executors can not be created.

Flyte spark tasks were being executed with the spark service account by default. On the documentation for spark jobs, one should create an additional k8s service account per namespace to enable spark jobs. The name of these service accounts should be "spark", and they have extra k8s priveleges.

Starting with flyteorg/flyteadmin#378, flyteadmin adds the "default" service account in the application config here, which changes the old behavior of spark tasks.

The spark backend plugin, checks the execution metadata of the submitted tasks, and if it is able to find a service account name, it doesn't assign "spark" service account. Before the aferomentioned change to flyteadmin, it was assigning "spark".

Also the documentation should be updated to state that the spark service account created by flyte deployments are not being used by default atm.

Expected behavior

The spark tasks are executed by the "spark" service account, if the user doesn't explicitly set another service account name. Or there should be a way to specify the service account name on the task level.

Additional context to reproduce

• Deploy flyte-core 1.0.1
• Do the environment setup according to this doc page
• Replicate and execute the pyspark pi example

Screenshots

No response

Are you sure this issue hasn't been raised already?

• Yes

Have you read the Code of Conduct?

• Yes