Github user tgravescs commented on the issue:
https://github.com/apache/spark/pull/18978
this part is specific to yarn because its all different on how the clients
for various cluster managers talk.
Security is different, this is very yarn specific dealing with tokens.
Somehow a client needs to know where the AM is and have a secret to be able to
connect to the AM. Without having our own mechanism for secrets that is pretty
hard.
I agree the interface for pushing tokens should be generic but its going to
be cluster manager specific implementation on how that happens from a client to
the spark application. In this case this client to am rpc can be used for the
pushing. once its at the driver it can be generic to reload and push to the
executors.
I also think things like status and kill are going to be different
depending on your cluster manager. Yes there is the generic spark driver type
kill but unregistering for yarn for instance will be cluster specific. The yarn
status is different from the status of the driver.
We could use the rest server here like mesos and standalone, the problem is
that doesn't satisfy all the use cases and setups. For instance if someone
doesn't have a good UI authentication like spnego its hard to use headless
users. so the rpc for yarn is a better fit.
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [email protected] or file a JIRA ticket
with INFRA.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
