Make pod name unique using the submission timestamp

[foxish]

Addressing part of #17
This should ensure we create unique driver/service names in every invocation.

[mccheah]

This would make it difficult for the user to look up the pod by name. Could it not be a requirement that spark.app.name needs to be a unique Kubernetes application? Or perhaps we don't anticipate users looking up the spawned pod that often, in which case this is fine.

[iyanuobidele]

I'm guessing @foxish is thinking ahead about exposing the status of the spark job using the TPR, removing the need for direct contact with the pods created for a specific spark job.

Or perhaps we don't anticipate users looking up the spawned pod that often, in which case this is fine.

Either way, this is correct.

[ash211]

If someone wants to look up a driver pod via an identifier they could use labels instead after #27 merges. We probably shouldn't be trying to bake extra meaning into names anyway.

As a related suggestion, does it make sense to apply spark.app.name and spark.app.id as labels to the created pods for that lookup purpose? Is this a pattern seen elsewhere in k8s?

[foxish]

This pattern fits in with the other controllers, such as Job.

As a related suggestion, does it make sense to apply spark.app.name and spark.app.id as labels to the created pods for that lookup purpose? Is this a pattern seen elsewhere in k8s?

Yes, the "name = foo" pattern is used quite a bit, and it would help find these pods like you said. kubectl takes the --selector commandline option to allow filtering by label in case one is using the CLI.

[ash211]

I'll file a separate ticket for that and look into adding it. Let's merge this one so we get unique names now.