Is slim_handler necessary now?

[monkut]

slim_handler is a work-around for cases where the resulting zip package exceeds the size limit (50MB) for a lambda package.

It seems that this is not a hard enforced limit, as we have seen packages over 60MB be successfully deployed. However, this is enforced in the lambda AWS Console.

Since it's introduction the following have been introduced:

• Dockerfile support (250MB limit [custom runtime])
• Lambda Layers (layers + package <= 250MB)

While Dockerfile support is an attractive solution to get around the package size limit, it does not allow support for the newly added SnapStart feature.

Moving dependencies to a Lambda Layer (as suggested in the issue below) appears to be a good solution, taking advantage of SnapStart while also resolving the size issue. However, I'm not sure this should be applied to ALL projects, assuming that the addition of a layer will add additional startup time overhead.
#1366

I'd like to see a benchmark of our current slim_handler solution with/without SlimStart and comparing it to a Lambda Layers implementation.

At the moment, I'm leaning toward removing slim_handler or refactoring it's implementation to take the Lambda Layers approach.

Any thoughts?

Limits:
https://docs.aws.amazon.com/lambda/latest/dg/gettingstarted-limits.html