RFC: Thread-local metrics object

[jaredcnance]

Purpose
It would be nice to create the metric context and logger at the beginning of the function execution and use that instance in multiple classes/methods without having to pass it as a parameter throughout.

Proposal Summary
Add an interface for accessing a thread-local metric logger.

Option A
Introduce a decorator that will provide access to the thread-local instance. The first time it gets called, it will initialize itself and close upon completion of the top-level method.

from aws_embedded_metrics import metric_scope

@metric_scope(local_scope=True)
def my_function(..):
  my_method()
  # metrics get closed when my_function exits

@metric_scope(local_scope=True)
def my_method(metrics):
  metrics.put_metric("Name", value)

Option B
Introduce a static accessor with decorator that initializes and closes the scope.

from aws_embedded_metrics import local_metric_scope, local_metrics

@local_metric_scope
def my_function(..):
  my_method()
  # metrics get closed when my_function exits

def my_method():
  local_metrics.put_metric("Name", value)

Option C
Explicit initialization of local metrics

from aws_embedded_metrics import local_metrics

def my_function(..):
  with local_metrics.init() as metrics:
    my_method()
    # metrics get closed when with statement completes

def my_method():
  local_metrics.put_metric("Name", value)

Considerations

• Metrics still need to be explicitly initialized and closed.
• Need to provide advice on unit testing code that consumes the thread-local metrics object.
• Currently, the flush for TCP is not async. If we make this async we can't rely on thread local and may need to use something like contextvars

[drduhe]

This would be an awesome feature! =)