Singleton initialisation in a multi AWS Lambda scenario

[roamingthings]

I'm currently thinking about how to optimize AWS serverless applications that include multiple AWS Lambda functions. My aim is to reduce deployment artefact size, number of artefacts, number of Gradle/Maven modules.

Let's imagine I have multiple AWS Lambda handler classes extending something like MicronautRequestHandler in one application/jar. These functions require a different set of Singleton beans:

• Function 1 -> Service 1 -> [S3Client, Service 2]
• Function 2 (so simple it doesn't require any other bean)
• Function 3 -> S3Client

When I call Function 2 I would like to have none of the Singleton beans being created to optimize for short cold start times.
When I call Function 3 I would like to have only S3Client being created.

Currently there is a way to set eager Singleton creation. However, this doesn't help in these cases since it either moves Singleton initialisation to execution time.