Compiling Python into AWS Lambda / Step Function

Ben North (GitHub / blog), March 2018 (repository root)

Background

Among the components of Amazon Web Services are the following two parts of their 'serverless' approach:

• Lambda — a 'Lambda function' is a self-contained piece of code which AWS runs on your behalf in response to triggers you specify;
• Step Functions — a 'Step Function' is a mechanism for controlling the interlinked operation of multiple steps, including invocation of Lambda functions.

While Lambda functions can be written in many languages, to write a Step Function you describe the logic as a state machine in JSON. This seems cumbersome when compared to our normal way of describing how to control interlinked computations, which is to write some Python (or C#, or Java, or...).

Based on this observation, the tools presented here are a 'plausibility argument of concept' for the idea that you could write your top-level logic as a Python program and have it compiled into a Step Function state machine. (I haven't developed this far enough to call it a 'proof of concept'.)

One of the desired properties of the system is that the source program should be more or less 'normal Python'. It should be possible to use it in two ways:

• Run it as a Python program with the usual Python interpreter;
• Compile it into a Step Function and run in the AWS cloud.

The ability to run your logic as a normal Python program allows local development and testing.

General approach

Compile Python code to Step Function state machine

The 'Python to Step Function compiler' tool, pysfnc.py, reads in a file of Python code and emits JSON corresponding to the control flow of a specified 'entry point' function in that code. The resulting JSON is used for the creation of an AWS Step Function. Various supplied Python functions allow the programmer to express intent in terms of retry characteristics, parallel invocations, error handling, etc. Nonetheless the code is valid normal Python and executes with (mostly) equivalent semantics to those the resulting Step Function will have.

Wrap original Python code as Lambda function

The 'Python to Step Function wrapper compiler' tool, pysfnwc.py, constructs a zip-file containing the original Python code together with a small wrapper. The zip-file is suitable for uploading as an AWS Lambda function. This gives the top-level Step Function access to what were callees in the original Python code.

Example

In the below I have omitted details like creation of IAM users, creation of roles, etc. See Amazon's documentation on these points.

Prerequisites

Python 3.6, with

pip install click
pip install attrs
pip install pytest  # To be able to run unit tests

Run unit tests on original Python

The original Python source consists of a main driver function, with a collection of small functions used by the main function. It is very simple, performing a few computations on an input string, but serves the purpose of illustrating the compilation process. It has a suite of unit tests:

pytest test_analyse_text.py

Output:

# ... ======== 10 passed in 0.02 seconds ======== ...

Wrap original Python ready for Lambda

python pysfnwc.py analyse_text.py lambda-function.zip
unzip -l lambda-function.zip

Output:

Archive:  lambda-function.zip
  Length      Date    Time    Name
---------  ---------- -----   ----
      302  1980-01-01 00:00   handler.py
     1981  2018-03-25 20:01   inner/analyse_text.py
      452  2018-03-23 22:32   pysfn.py
---------                     -------
     2735                     3 files

Now upload lambda-function.zip as a new Lambda function with the Python 3.6 runtime, specify handler.dispatch as its entry point, and note its ARN for use in the next step.

Compile original Python into Step Function JSON

python pysfnc.py analyse_text.py LAMBDA-FUN-ARN > stepfun.json
cat stepfun.json

Output (the full output is 196 lines):

{
  "States": {
    "n0": {
      "Type": "Pass",
      "Result": {
        "function": "get_summary",
        "arg_names": [
          "text"
        ]
      },
      "ResultPath": "$.call_descr",
      "Next": "n1"
    },

    [...]

    "n19": {
      "Type": "Succeed",
      "InputPath": "$.locals.result"
    }
  },
  "StartAt": "n0"
}

Now copy-and-paste this as the JSON for a new Step Function.

Execute Step Function

You should now be able to perform an execution of this Step Function with, for example, the input

{
  "locals": {
    "text": "a short example"
  }
}

to get the output

{
  "output": "text starts with a, has 15 chars, 5 vowels, and 2 spaces"
}

More documentation

• Implementation notes
• Future work

---

This document: Copyright 2018 Ben North; licensed under CC BY-SA 4.0

See the file COPYING for full licensing details.