MINOR: [Java] Reduce enum array allocation

[sullis]

Rationale for this change

reduce Java object allocation

context: https://www.gamlor.info/wordpress/2017/08/javas-enum-values-hidden-allocations/

What changes are included in this PR?

cache ArrowFlightConnectionProperty.values() in a static variable.

Are these changes tested?

ArrowFlightConnectionConfigImplTest unit test passes.

Are there any user-facing changes?

no

[laurentgo]

(nit) Maybe we can use a truly immutable type like EnumSet + Collections#unmodifiableSet() instead of an array.

[kou]

Could you open a new issue instead of using MINOR:?
See also our MINOR definition: https://github.com/apache/arrow/blob/main/CONTRIBUTING.md#Minor-Fixes

[jduo]

I'm curious -- did we observe a large number of allocations here? It looks to me like this is only called once-per-JDBC-connection. Perhaps this is a connection pooling scenario?

We may want to optimize body of the noneMatch() call to use a case-insensitive TreeSet if this is called frequently (instead of searching linearly for each key as we're currently doing).

[laurentgo]

We may want to optimize body of the noneMatch() call to use a case-insensitive TreeSet if this is called frequently (instead of searching linearly for each key as we're currently doing).

I agree this is likely to be of greater benefit. I would argue that instead of using TreeSet, the original enum values could be stored in upper case (which is already the case but better safe than sorry) and instead convert each header to uppercase once as well and check the set. This is likely to avoid extra comparisons/case conversion to be done when navigating the tree.
I also remember something about http/2 (used by gRPC) about headers possibly normalized as lower-case which (if true) could also be a better path for optimization

[lidavidm]

Yes, http/2 normalizes to lower-case headers so that would be best

https://datatracker.ietf.org/doc/html/rfc9113#section-8.2-2