RFC: Powertools for AWS Lambda (Python) v3

[leandrodamascena]

Is this related to an existing feature request or issue?

#4067

Which Powertools for AWS Lambda (Python) utility does this relate to?

Other

Summary

We are excited to announce that we are starting a new major version (v3) for Powertools for AWS Lambda (Python). We are expediting a new major version to align with the scheduled Pydantic v1 end-of-life on June 30th, signalling the switch to much anticipated Pydantic v2. Pydantic plays a vital role and we care deeply about supply chain security. This release highlights our dedication to providing a secure and stable library that our customers trust.

Today, we already support both Pydantic v1 and v2. In this new major version, we will make Pydantic v2 the default, removing Pydantic v1. We will take this opportunity to introduce minor breaking changes to address common pain points. As always, we will provide an extensive upgrade guide to make this transition as smooth as possible.

Community

We invite the entire Powertools community, including users and contributors, to actively participate in this RFC. We greatly value the community's contribution, ideas, and willingness to solve problems or validate proposed changes.

Use case

The primary use case is to offer Pydantic v2. That includes a new Lambda Layer for each Python version since Pydantic v2 contains non-Python code (Rust). It future-proof Layers to contain any additional compiled dependency (like cryptography) without a new major version.

Additional key items on v3 backlog

• Distribute Lambda Layers in the AWS GovCloud regions
• Return sensible defaults over Optional in Event Source Data Classes to further reduce customers code
• Remove deprecated code, deprecate legacy code to be removed in v4, and reduce API surface

Proposal

The action plan for this release with scope and tasks is as follows:

• Update our documentation to comply with our Versioning Policy and inform customers that we have started work on a new major version
• Create a branch named v3
• Make adjustments to our GitHub actions to run checks on this new branch
• Update Powertools v3 milestone to reflect current situation
• Address and resolve the issues mapped out for the execution of this launch plan.
• If possible, release early alpha versions for testing

Quick summary (Work in progress - Items can be added or removed):

• ✅ = done
• 🚧 = Work in progress
• 🚫 = Not considered in v3

Item	Issue/PR	Status	Code change required	Utility	Notes	Cross-Language
refactor the SSMProvider class to implement the get_multiple method	#3252	✅	Yes	Parameters
Return empty Dict or List in Event Source Data Classes instead of None	#2605	✅	Yes	Event Source Data Class
event_parser envelopes should handle unions of BaseModels	#2734	✅	No	Parser
Pydantic v2 managed layer	#2939	✅	No	Layer
Deprecate tracing.base.BaseProvider	#4105	🚧	No	Tracer
resolved_headers_field is not Optional	#4147	✅	No	Event Source Data Class
Create a Powertools Lambda layer for each Python version	#3859	✅	No	Layer
Publish layer to Gov Cloud regions	#1166	🚧	No	Layer		✅
Remove any specific code for Pydantic v1	#4191	✅	Yes	Codebase
Update Powertools documentation to add banner about v3	#4198	✅	No	Documentation
Add the aws-encryption-sdk dependency in the Powertools layer	#4192	✅	No	Layer	Depends on #3859
Review the method extract_data_from_envelope in JMESPath Functions	#4218	✅	No	JMESPath Functions
Split the code from the api_gateway.py file into several other files to ease maintenance	#4194	🚫	Yes	Event Handler
Add support for v3 branch in the github actions	#4220	✅	No	CI
Release the new version of Tracer utility with support for external providers	#2342	🚧	Yes	Tracer
Unmarshal DynamoDB fields when using DynamoDBStreamModel	#4219	✅	Yes	Parser	Reference: #1619	✅
Remove legacy code in the Metrics utility?	-	🚫	Yes	Metrics	Needs discussion
Keep the compatibility with Pydantic v2 and Bedrock agent resolver	#4196	🚧	No	Event Handler
Change the default TTL (cache) to 5 minutes	#4193	✅	Yes	Parameters		✅
Update boto3 version to 1.34.32	#4195	✅	No	Dependencies
Change behavior to fail batch on first error when using DynamoDB and Kinesis in batch processing	NEED ISSUE	🚧	Yes	Batch Processing	Reference: https://tinyurl.com/yc335m3s	✅
Remove the flag log_uncaught_exceptions in the Logger utility	#4199	🚫	Yes	Logger	See: #1798 (comment)
Support for retrieving batch of secrets	#4200	🚧	No	Parameters

Deadlines

We plan to release the first version of v3 by the end of June/beginning of July.

Out of scope

We will not include in v3:

• Modularization to further reduce sizes
• Strict typing
• New utilities

Potential challenges

Adding Pydantic v2 in the Powertools layer requires layers per Python version, so the effort to implement this is not yet clear.

Dependencies and Integrations

No response

Alternative solutions

No response

Acknowledgment

• This feature request meets Powertools for AWS Lambda (Python) Tenets
• Should this be considered in other Powertools for AWS Lambda languages? i.e. Java, TypeScript, and .NET

[jplock]

There's a lot of overlap between the Event Handler, Event Source Data Classes, Parser, and Validation

Is there a nice way to unify it all somehow?

[ran-isenberg]

There's a lot of overlap between the Event Handler, Event Source Data Classes, Parser, and Validation

Is there a nice way to unify it all somehow?

I agree. Since v3 will use pydantic by default, is there a need to maintain event source data classes? why not move to the parser's pydantic schemas exclusively? I suppose there are some gaps in coverage and missing schemas, but that can be fixed :)
Event handler/validation already support pydantic/parser schemas.