Is overriding custom pythonRequirements at function level allowed?

[0aprl1]

Hello All

SETUP

• python: 3.8.10
• serverless: 1.80.0
• serverless-python-requirements: 5.1.0

ISSUE

configuration.yml is a file I use to set custom properties in serverless.yml, where I define the following noDeploy as a consequence of installing some required library that have unused sub-dependencies:

pythonRequirements:
  slim: false
  dockerizePip: true
  usePoetry: false
  noDeploy:
    - boto3
    - botocore
    - docutils
    - jante
    - phonenumbers
    - unidecode
    - sqlmodel
    - sqlalchemy
    - sqlalchemy-stubs
    - psycopg2
    - psycopg2_binary.libs

I have several lambdas which use the above default pythonRequirements definition. For one particular lambda I have defined, I am attempting the following:

test-service:
    name: function-name
    handler: service.handler.handler
    events:
      - http:
          path: /test
          method: get
          authorizer:
            name: authorizer
            resultTtlInSeconds: 3600
          cors: ${self:custom.CORS_CONFIG}
    role: ServiceLambdaExecutionRole
    custom:
      pythonRequirements:
          slim: false
          dockerizePip: true
          usePoetry: false
          noDeploy:
            - boto3
            - botocore
            - docutils
            - jante
            - phonenumbers
            - unidecode
    package: ${self:custom.SERVICE_PACKAGING} 

Basically for this function, I try to override pythonRequirements at function definition level - the two noDeploy differ in the second not having listed

  - sqlmodel
  - sqlalchemy
  - sqlalchemy-stubs
  - psycopg2
  - psycopg2_binary.libs 

When I run sls package --region eu-west-1 --stage stg, I am able to package all functions correctly, except for the one defined above test-service, whose zip is excluding everything (11 modules in total) as per the default pythonRequirements, while I would have expected it to exclude only the ones defined at function level (6 in total).

So my question is whether what I am attempting is allowed or not, or if I might have misconfigured something.

Thanks in advance for the support.

[pgrzesik]

Hey @0aprl1, custom on the function level would be totally ignored, you're also on quite an old version of the Framework, in newer ones it would error out (or at least return a warning). The noDeploy option is global and cannot be applied on function level. There's another feature for scoping down dependencies to specific functions which is modules. You can read more about them in readme.

[0aprl1]

Thanks for the reply. I have tested, and the issue is that I also have a custom library which is common across all lambdas and I want to remove its subdependencies as well when packaging - so it's no just what is defined directly in the requirements.txt.

I have resorted to adding layers for those lambda that need the removed libraries, and still use noDeploy.