[Entitlements] Fix Entitlement initialization to work across multiple versions

[ldematte]

This PR fixes an issue that become apparent with #121017: when we initialize Entitlements, and we get the list of methods to instrument, and instrument them, we need to get them from all the "versioned" interfaces compatible with the current runtime version. We also need to use the most specific (latest) "versioned" interface in the instrumentation code.

Alternative to #121132 which uses ASM and the actual inheritance chain to navigate the classes (interfaces) to extract check$ methods from.

[elasticsearchmachine]

Pinging @elastic/es-core-infra (Team:Core/Infra)

[prdoyle]

In general, reflection-based code is easier to read and maintain than ASM code. I'd lean toward reflection if there's no compelling reason to use ASM.

[ldematte]

In general, reflection-based code is easier to read and maintain than ASM code. I'd lean toward reflection if there's no compelling reason to use ASM.

I was thinking the same and thought that reflection could be equivalent but Iwas wrong in assuming that: as this test failure #120811 (comment) demonstrates, we have to be very careful when using reflection this early.
If the interface contains methods with types that are not available, things will break; that's the very reason we went with ASM here in the first place after all.

So I closed #121132 in favor of this one, "porting" all the comments here and addressing them.

[rjernst]

Looks fine but one question.

[rjernst]

Why do we need asm here? Since we have Class objects already, can't we use reflection to find all of the methods?

[ldematte]

I was thinking the same, but every time we use reflection (e.g. in #120811 (comment)), we end up regretting it and go back to ASM; for example in the linked issue I tried to use reflection first, only to end up with NoClassDefFoundError errors in tests due to the fact that our interface uses types that may not (are not) always available (at least at this time).
It might be fine in this case if we use it only for finding the base classes, but since we need to use ASM to find all the methods anyway, why risk?

[ldematte]

Serverless failure is due to a lack of policy - fixed in 3451

[prdoyle]

I'd rather see some protection against visiting interfaces multiple times, but we can merge this as-is I think.

Also, reflection would probably be simpler than asm, but that's also not a hill I want to die on.

[ldematte]

I'd rather see some protection against visiting interfaces multiple times, but we can merge this as-is I think.

That I've added, implementing your suggestion. See line 55 of InstrumentationServiceImpl.
As for reflection, I've tried and it gave trouble, see my previous comments.

[elasticsearchmachine]

💔 Backport failed

Status	Branch	Result
✅	9.0
❌	8.18	Commit could not be cherrypicked due to conflicts
❌	8.x	Commit could not be cherrypicked due to conflicts

You can use sqren/backport to manually backport by running backport --upstream elastic/elasticsearch --pr 121192