Feature request: Implement a context manager to bind variables to a logger

[wesleyhamburger-okta]

Use case

I wan't to add variables to the logger for the duration of a request and get clean up for free without needing to explicitly call delete.

This is similar to https://www.structlog.org/en/stable/api.html#structlog.contextvars.bound_contextvars

Solution/User Experience

@contextmanager
def bind(self, **args):
    logger_instance.append_keys(**args)
    yield logger_instance
    logger_instance.remove_keys(args.keys())

with logger.bind(user_id=123):
    perform_operation():

Alternative solutions

No response

Acknowledgment

• This feature request meets Lambda Powertools Tenets
• Should this be considered in other Lambda Powertools languages? i.e. Java, TypeScript

[boring-cyborg]

Thanks for opening your first issue here! We'll come back to you as soon as we can.
In the meantime, check out the #python channel on our AWS Lambda Powertools Discord: Invite link

[heitorlessa]

hey @wesleyhamburger-okta thanks for raising this feature request. Please allow me to ask some questions to make sure I understand it correctly:

1. Are you also intending to use with threads? see RFC: Support thread-local keys for logger #991
2. Are you simply looking for a shortcut to cut boilerplate? OR, a completely isolated logger instance within that operation?
   • 2A. If the latter, would you want existing keys to be carried over when you use bind()? e.g., lambda request id, etc.,

Reason I ask that is structlog.bind() returns an entirely new logger instance.

stateDiagram
    [*] --> Logger
    Logger --> new_logger
    new_logger --> append(user_id)
    append(user_id) --> perform_operation()
    perform_operation() --> remove(user_id)
    new_logger --> [*]

Loading

We, on the other hand, would append user_id to the registered Formatter state, and that isn't thread-safe (yet!).

stateDiagram
    [*] --> Logger
    Logger --> append(user_id)
    append(user_id) --> perform_operation()
    perform_operation() --> remove(user_id)
    remove(user_id) --> [*]

    [*] --> LoggerElsewhere
    LoggerElsewhere --> bi_directional_state
    LoggerElsewhere --> receives_user_id_too
	receives_user_id_too --> user_id_removed
    user_id_removed --> [*]

Loading

The former would need a better name to not confuse existing structlog cystiners, since they would be led to believe our with logger.bind would yield a completely isolated state.

Looking forward to hearing from you.

PS: If it's primarily boilerplate, we'd welcome a contribution and can guide/co-develop you all the way through to expedite this for the next release :)

Thanks a lot!!

---

proof of concept for a .bind that yields back a new isolated Logger instance while retaining the service etc.

Example for question 2

with logger.bind(user_id="123") as new_logger:
    # new_logger.info() would have `user_id` key... but if I do logger.info()... `user_id` would not be present
    # similarly, perform_operation() OR any other function with a `logger` instance wouldn't have `user_id`
    perform_operation()

[trallnag]

Related #3719

[leandrodamascena]

Hello everyone! We are implementing this feature in this PR: #5883.

A summary of what we'll be implementing:

1 - We'll implement a context manager method to allow customers to do with logger.appended_context_keys(**kwargs): and persist keys in that block.

2 - This won't be supported on thread-safe keys - for now.

3 - Customers can use isolated Logger instances by creating a new Logger instance with a different service_name or by creating a child logger.