Rework serverless runtimes API

Description

Design a new serverless runtimes API based on the issues and conclusions coming from using the current ones.

Reasons

Bring more user-friendly runtimes that users could reconfigure, or easily replace with their own built without limitations.

Acceptance Criteria

• rework the handler's arguments. How it looks in other solutions:

  Tool	Nodejs Args	Python Args
  Google Cloud Functions nodejs/python	(req, res)	(flask.req) -> flask.resp
  AWS Lambdas nodejs/python	(event)	(event, context)
  Azure Functions nodejs/python	(request, context)	(req) -> resp
  Fission nodejs/python	(context) -> resp	()

def main(event):
  return "", 204

module.exports = {
  main: function(req, res) {
    // no support for tuples
    return [ "", 204 ]
  }
}

def main(req, res):
  res.setCode(204)

module.exports = {
  main: function(req, res) {
    data = req.body
    ...
    res.setCode(204)
  }
}

from flask import request
def main():
  data = request.getData()
  ...
  return '', 204

module.exports = {
  main: function(req, res) {
    res.setCode(204)
  }
}

• function's SDK should be defined in a separate, included package which every user can import in the handler file (now it's a part of the event object)

  import skd
  ...
  tracer = sdk.get_tracer()
  sdk.publishCloudEvent(...)
  

  and

  import {getTracer, publishCloudEvent} from sdk
  ...
  const tracer = getTracer()
  publishCloudEvent(...)
  

• redesign the ENVs architecture to include the whole configuration inside runtimes. The user should be able to reconfigure the runtime by passing custom environments to the Function CR, but the serverless-controller should not add any runtime-specific configuration through envs. Everything should be defaulted on the runtime level

  • function-controller should configure only the envs that runtime container could not calculate on its own (like source code, deps, path to sources, tracing endpoint, eventing endpoint

  • envs should have the right and unified naming convention, for example:

    • SERVER_* - server configuration
    • HANDLER_* - users code config
    • FUNC_* - function metadata (like name, namespace, annotations, labels)

  • all envs should have defaults defined insite the runtime container (expecting these ones that function-controller injects)

    Example env configured inside the server.py file:

    func_name=os.getenv('FUNC_NAME', '')
    func_namespace=os.getenv('FUNC_NAMESPACE', '')
    func_labels=os.getenv('FUNC_LABELS', '')
    func_annotations=os.getenv('FUNC_ANNOTATIONS', '')
    func_runtime=os.getenv('FUNC_RUNTIME', 'python314')
    server_host=os.getenv('SERVER_HOST', '0.0.0.0')
    server_port=int(os.getenv('SERVER_PORT', '8080'))
    server_numthreads=int(os.getenv('SERVER_NUMTHREADS', '50'))
    server_call_timeout=float(os.getenv('SERVER_CALL_TIMEOUT', '180'))
    handler_module_folder=os.getenv('HANDLER_FOLDER', '/')
    handler_module_name=os.getenv('HANDLER_MODULE_NAME', 'handler')
    handler_module_function=os.getenv('HANDLER_FUNCTION_NAME', 'main')
    tracecollector_endpoint = os.getenv('TRACE_COLLECTOR_ENDPOINT')
    publisher_proxy_address = os.getenv('PUBLISHER_PROXY_ADDRESS')
    

• runtime should print basic configuration on run:

  Importing function sources from <DIR>/<HANDLER_FILENAME>:<HANDLER_MAIN_FUNC>
  Tracing configured with endpoint <TRACECOLLECTOR_ENDPOINT>
  Publisher Proxy available on address <PUBLISHER_PROXY_ADDRESS>
  Starting <RUNTIME> server <HOST>:<PORT>
  

• runtime should return standard response when the handler's code failed (500), for example:

  $ curl localhost:8080
  Internal Server Error
  

• refactor should be applied for the python314 and the nodejs26 runtimes ONLY
  We should work on both runtimes at the same time to keep the same API for both runtime types

Related Issues

runtime releases and their lifecycles:

• https://nodejs.org/en/about/previous-releases
• https://devguide.python.org/versions/