MINOR: clarify usage of stateful processor node

[vvcephei]

In recent PRs, we have been confused about the proper usage of
StatefulProcessorNode (#5731 , #5737 )

This change disambiguates it.

Committer Checklist (excluded from commit message)

• Verify design and implementation
• Verify test coverage and CI build status
• Verify documentation (including upgrade notes)

[vvcephei]

@guozhangwang @mjsax @bbejeck ,

In light of the recent confusion about this class, I was thinking this change might help avoid it in the future.

[vvcephei]

Having a separate method for getStatefulProcessorNode was unnecessary, so I've just inlined it to call the constructor directly instead. Perhaps that method did extra work in the past.

[vvcephei]

This is the core change. Instead of having one constructor and a contract that callers should specify either names or a store builder and set the other to null is confusing. Instead, we'll have two constructors, one for names and the other for the store builder.

[vvcephei]

This is an internal class, and the constructor is quite simple. Having a builder for it seems like extra code for no real gain. Plus, it makes it more difficult for callers to tell what arguments are expected.

[vvcephei]

The compiler can't tell which constructor to call, so we cast null to pick one. It doesn't matter which, since this argument is never evaluated.

[vvcephei]

Failure unrelated. (spotbugs couldn't write to some local socket?)

Retest this, please.

[bbejeck]

Thanks @vvcephei.

Having 2 constructors does clear things up. Just one comment about adding some Javadoc on the two constructors.

[bbejeck]

I agree having two separate constructors clears usage of StatefulProcessorNode up.

Maybe we could add some Javadoc comments that the constructor taking the StateStore parameter is for stateful operations that create a state store.
And the constructor taking storeNames is for stateful DSL methods process, transform and the other transform variants where the state store is already created (with a StreamsBuilder.addStateStore call) , and we pass in the names so the Processor can access the stores.

[guozhangwang]

Waiting for jenkins to pass.

[vvcephei]

Major hiccough in the core integration tests.

Retest this, please.

[mjsax]

LGTM.

Very nice cleanup @vvcephei!

[guozhangwang]

@vvcephei could you rebase please?

[vvcephei]

@guozhangwang, It is rebased.

[guozhangwang]

Cherry-picked to 2.1 as well.

[vvcephei]

Thank you!