Logger: log_event does not serialize classes

[kishaningithub]

What were you trying to accomplish?
I was trying to log the received S3 event. Please note that i am using the data classes present in this library for reading the event

@event_source(data_class=S3Event)
@log.inject_lambda_context(
    log_event=True
)
def lambda_handler(event: S3Event, context: LambdaContext):

Expected Behavior

The logged event should have all the information from the S3 event

Current Behavior

This is the output i get in the log (trimmed)

{
    "level": "INFO",
    "message": "<aws_lambda_powertools.utilities.data_classes.s3_event.S3Event object at 0x7f0be7efb2b0>",
    "timestamp": "2022-01-11 06:36:20,111+0000",
}

It looks to be that it was unable to properly represent the S3Event object as a string

Possible Solution

Implement repr and str methods in S3Event class or in the parent DictWrapper class

Steps to Reproduce (for bugs)

1. Implement a lambda which receives and S3 event like the following

@event_source(data_class=S3Event)
@log.inject_lambda_context(
    log_event=True
)
def lambda_handler(event: S3Event, context: LambdaContext):
   pass

2. Setup the lambda trigger as S3 object creation event
3. Upload a file in the S3 bucket where trigger is setup
4. See the logs in cloud watch

Environment

• Powertools version used: 1.24.0
• Packaging format (Layers, PyPi): PyPi
• AWS Lambda function runtime: 3.9
• Debugging logs

How to enable debug mode**

# paste logs here

[michaelbrewer]

Thanks @kishaningithub . I will take a look at that. We had a related fix for idempotency.

For now inverting the decorators:

@log.inject_lambda_context(log_event=True)
@event_source(data_class=S3Event)
def lambda_handler(event: S3Event, context: LambdaContext):
   pass

[heitorlessa]

hey @kishaningithub thanks for raising that. The main reason we didn't do this is that these are untrusted strings. We weren't certain of the potential attack vectors this could cause (similar but not the same to Log4j).

Besides the fix of inverting the decorator, I'd be cautious on doing this in production as you might dump sensitive data into the logs.

All that being said, you can bring your own JSON serializer for Logger via LambdaPowertoolsFormatter - this would be an explicit way to opt-in to serialize any data you want.

Keen to hear your thoughts

Thank you!

[heitorlessa]

Updating here as @michaelbrewer promptly created a PR to only log IF they're an instance of our built-in event source data classes.

I'm fine with this compromise since the only vector here would be a customer to override our data classes __repr__. Any other classes should be best dealt via the serializer option in the Formatter.

[github-actions]

This is now released under 1.25.0 version!