ComponentInstallers should run after LogManager if a custom one is detected

[mateuszrzeszutek]

Fixes #2591

[trask]

I wonder if we can keep the Java 8 condition. IIRC (and I likely do not), starting in Java 9, Java classes themselves no longer use j.u.l., instead using a new internal logging API.

[trask]

Suggested change

	// Once we see the LogManager class loading, it's safe to run ComponentInstaller#after() because
	// Once we see the LogManager class loading, it's safe to run ComponentInstaller#afterByteBuddyAgent() because

[iNikem]

@trask @mateuszrzeszutek what is the drawback of late installation of components?

[mateuszrzeszutek]

I wonder if we can keep the Java 8 condition. IIRC (and I likely do not), starting in Java 9, Java classes themselves no longer use j.u.l., instead using a new internal logging API.

That's true, Java 9 introduced the new System.Logger interface that's supposed to be used internally (https://openjdk.java.net/jeps/264) - let's keep the Java 8 condition then.

@trask @mateuszrzeszutek what is the drawback of late installation of components?

Hmm, runtime/JVM metrics starting a bit later? The only risk that I can think of is that LogManager never loads and ComponentInstaller#after...() never gets called, but with all the conditions and checks that we have this is a very minor risk.

[iNikem]

OpenTelemetryInstaller is also ComponentInstaller. Anything bad happens if its installation is delayed?

[mateuszrzeszutek]

No, because it doesn't override the afterByteBuddyAgent() method - this "delay after LogManager" thing only affects the after... method.