Add support for apache httpclient5

[laurit]

Adds support for apache httpclient5 classic. Related to #2157 though the issue does not specify whether they are interested in class or async httpclient.

[anuraaga]

Hi @laurit - can we start with library instrumentation for this? We should generally use the library's native support for instrumentation instead of intercepting execute where it's possible - so if we start with library instrumentation registering interceptors, I think we'll know how to then wire that into the javaagent code.

https://hc.apache.org/httpcomponents-client-5.0.x/httpclient5/xref-test/org/apache/hc/client5/http/examples/ClientInterceptors.html

[laurit]

@anuraaga This is an adaption of httpclient-4.0 instrumentation and works exactly the same way. The approach you suggested would work when http client is built by HttpClientBuilder. HttpClients.createMinimal https://github.com/apache/httpcomponents-client/blob/master/httpclient5/src/main/java/org/apache/hc/client5/http/impl/classic/HttpClients.java does not use HttpClientBuilder and might need a different approach. I won't have time to work on this in the near future. If you believe that this approach is unacceptable then let me know and I'll close this pr.

[iNikem]

If this is exact adaption of the instrumentation for 4.0, then I am Ok with this. But I somewhat agree with @anuraaga sentiment, so will let him decide if we accept this without library instrumentation.

[anuraaga]

I do want to discourage agent-only PRs in general to not box out non-agent users, but at the same time I feel very bad to delete working code. If @laurit doesn't have time for it though that's totally fine and I would recommend merging if this is just copy paste from our other instrumentation with package rename, let's not waste it.

FYI the biggest thing on my mind is it's 2021 - if a new HTTP library is released without a proper instrumentation hook (which may be the case given the comment about createMinimal) we probably need to get in touch with the authors of the code 🤷‍♂️

[anuraaga]

Did a skim and I'm ok with this. @laurit just to be clear thanks for doing this, regardless of where this ends up long term it's for our users. I really appreciate it!

[trask]

can you also port #2276 to this new instrumentation?

[trask]

👍

[trask]

comment added to #2276 just before merging (if this is still the same in httpclient5?):

Suggested change

	try (Scope ignored = parentContext.makeCurrent()) {
	// ending the span before executing the callback handler (and scoping the callback handler to
	// the parent context), even though we are inside of a synchronous http client callback
	// underneath HttpClient.execute(..), in order to not attribute other CLIENT span timings that
	// may be performed in the callback handler to the http client span (and so we don't end up with
	// nested CLIENT spans, which we currently suppress)
	try (Scope ignored = parentContext.makeCurrent()) {