Issue #, if available: #97

Description of changes:

This addresses a common issue when trying to use Logger across multiple modules while wanting to:

• Create a new child Logger to inherit parent Logger's configuration (handler, formatter, custom keys, sampling)
• Reuse the same Logger, where Logger is enriched iteratively across the code base

For this, we've introduced a new parameter, child, when initializing a new Logger, and some internal changes while keeping the same UX and backwards compatibility.

• [Internal] Direct access to methods, handlers, etc work as before, except that we proxy properties access to ._logger

UX

Key aspects:

• We support logging inheritance via the child parameter, and the Logger will be named after <service>.<caller_file_name>
• Calling Logger with the same service name (explicit or via environment variable) returns the same Logger
• Changes to either Child Logger or Parent Logger will be propagated to all instances

Creating a child Logger

Child logger can be created in any other as long as a parent is created too - This supports customers using Layers or different file names to create child loggers.

# lambda_handler.py
from aws_lambda_powertools import Logger
import collect

logger = Logger(service="payment") # parent Logger just like before, named 'payment'
...

# collect.py
from aws_lambda_powertools import Logger

logger = Logger(service="payment", child=True) # child Logger named 'payment.collect'

# a_new_key will now be available both in this child logger as well as the parent
logger.structure_logs(append=True, a_new_key="value") 
...

Reusing the same Logger

# lambda_handler.py
from aws_lambda_powertools import Logger
import collect

logger = Logger(service="payment") # parent Logger just like before, named 'payment'

# collect.py
from aws_lambda_powertools import Logger

logger = Logger(service="payment") # returns the existing 'payment' Logger
...

Checklist

• Meet tenets criteria
• Update tests
• Update docs
• PR title follows conventional commit semantics

Changes

• Register logger via getLogger to allow inheritance
• Encapsulate logger initialization/config in a new method
• Remove Logger sub-classing
  • Proxy calls to underneath Logger instead
• Create name parameter for hierarchy
  • Use service if name is empty
• Test on Lambda locally
  • Single Logger, no code changes
  • Two Loggers in different files
  • Additional Logger in a package that is imported on main entry-point
  • Lambda Layer
• Only add a handler if there isn't one
• Only add a handler if parent is root
• Update Formatter to append new keys
• Make handler private
• Remove log_keys
• Remove duplicate Parameters in docstring
• Double check Debug w/ getattr perf impact
• Docs: Document inheriting Loggers
• Docs: Changelog (doc random logger for tests)
• Test: Logger hierarchy via logger.parent
• Remove name parameter
• Allow logger child propagation
• Introduce child boolean param
• Don't add a handler nor evaluate sampling if it's a child Logger
• Name child Loggers based on <service>.<caller_file_name> to mimick getLogger(__name__)
• Update tests to remove Handler on every test
• Update tests to remove getRandomLogger utility function

Optional

• Ensure log level input is case insensitive

Benchmark

TL;DR: For the average use case we have a slight performance decrease, but all negligible in all cases (0.10 nano seconds).

Quick checks as we've removed logging.Logger sub-classing, respect Logging inheritance, and are proxying calls to inner Logger.

Basic logger

• Script: python -m timeit -s 'from aws_lambda_powertools import Logger; logger = Logger(); logger.info("hello world")'
• Before: 50000000 loops, best of 5: 6.12 nsec per loop
• After with Logger proxy object, no sub-classing: 50000000 loops, best of 5: 6.22 nsec per loop

Inheritance

• Script: python -m timeit -s 'from aws_lambda_powertools import Logger; logger = Logger(); l = Logger(); l.info("hello world")'
• Same logger: 50000000 loops, best of 5: 6.28 nsec per loop
• Random child logger name + fixup: 50000000 loops, best of 5: 6.14 nsec per loop
  • Script: python -m timeit -s 'from aws_lambda_powertools import Logger; logger = Logger(child=True); l = Logger(); l.info("hello world")'

Breaking change checklist

RFC issue #:

• Migration process documented
• Implement warnings (if it can live side by side)

By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.