ClassLoaderMatcher put to Guava Cache outside critical section

[lpriima]

1. move to Guava Cache with weak references
2. split cache supplier(loader) and matcher

petclinic 100 runs avg start time:
before: 51ca429 avgTime : 18.49126000000000000000
after: 01b32d814fdcf1de9df46a7185ed78daf981434d avgTime : 17.60960000000000000000

[dougqh]

We might also want to set the concurrency level. I picked 8 for the type cache.
Although, we should add a configuration option in a separate PR -- and maybe auto-select by availableProcessors.

[mar-kolya]

Be careful with availableProcessors - it may be a lie in k8s land...

[dougqh]

Yes, that is true. We probably want to a cap on it. Although, I suspect for concurrency level we're better off err-ing slightly on the high-side.

[lpriima]

capped on 8.

[lpriima]

Did we measure a difference by making the critical section smaller?

I suspect we could make this lighter by eliminating the Supplier which will cost an extra allocation at least while still running in the interpreter. We could either create a single instance and reuse the surrounding matcher -- or just eliminate the Supplier altogether and inline the code.

I'd previously tried the inline approach, but I'll admit the gains were small. Maybe, this just isn't worth the effort?
However, I think the switch away from WeakMap/WeakHashMap is still worthwhile.

Yes, I've tried w/o Supplier and it slightly better. But when we move to generic WeakCache abstraction, it will have the supplier (some anonymous class to load the cache) anyway: https://github.com/DataDog/dd-trace-java/blob/lpriima/WeakCache/dd-java-agent/agent-bootstrap/src/main/java/datadog/trace/bootstrap/WeakCache.java#L38

Updated this PR without Supplier anyway.

[dougqh]

Updated this PR without Supplier anyway.
Thanks for the nice summary of the intermediate progress.

Yes, I've tried w/o Supplier and it slightly better. But when we move to generic WeakCache abstraction, it will have the supplier (some anonymous class to load the cache) anyway: https://github.com/DataDog/dd-trace-java/blob/lpriima/WeakCache/dd-java-agent/agent-bootstrap/src/main/java/datadog/trace/bootstrap/WeakCache.java#L38

Looks like this gives us a nice win right now, so I'm satisfied with this as is.
However, we will want to revisit this once WeakCache is in place.

We'll also want to look at any places using WeakMap and see if they are candidates for using WeakCache instead.

[lpriima]

We'll also want to look at any places using WeakMap and see if they are candidates for using WeakCache instead.

it will be in 2 places here in this PR plus in all other places here: https://github.com/DataDog/dd-trace-java/pull/1256/files

[dougqh]

While performing a perf integration test, I accidentally merged this prematurely to master. Since this has passed reviews, we should get this merged through the normal means.

[lpriima]

ok. Merging