Performance Optimizations targeting to reduce CPU churn

[alexeykudinkin]

Rationale for the change could be found in T6248475.

In a nutshell:

1. Avoiding string format in the hotpath
2. Replacing CHM traversal with single queue traversal (this will be leveraged in the follow-up)

[CLAassistant]

All committers have signed the CLA.

[alexeykudinkin]

Currently failing test M3ReporterTest is fixed in #71

[prateek]

few questions, mainly looking fine

[prateek]

why comment this out? i'd prefer if you either add in or remove entirely.

[alexeykudinkin]

This is cleaned up in #74

[prateek]

am I missing something or is this an unused import?

[alexeykudinkin]

Cleaning up in #74

[prateek]

I don't live in Java code so this more a curiosity: why override this method if you're going to only call the parent's method? Is it required by the Java type system?

[alexeykudinkin]

This is a refactoring aberration, cleaned up in #74

[prateek]

Can remove the TODO, it needs the concurrency because metrics can be both created concurrently, and created and reported upon concurrently

[alexeykudinkin]

Cleaned up in #74

[prateek]

why remove this? does the m3 module inherit the jmh settings from the parent?

[alexeykudinkin]

There's no benchmarks in this project

[ChuntaoLu]

Tally client already has heavy memory footprint, what's the memory impact of the additional queue?
Also how much cpu reduction does switching from traversing the hashmaps to traversing the queue? It is also not clear to me why that would reduce cpu usage.

[alexeykudinkin]

To reduce the traversal scope to only those metrics that had actually changed. This is leveraged in #74.