Serverless 3, Fastapi 0.104+ and Layers! Error: typing-extension

[Verde1705]

For fastapi version 0.104+ I have encountered the following error:

cannot import name 'Doc' from 'typing_extensions'

After much trial and error I realized that the problem was in the merge order of the lambda layers, since serverless adds a layer called "sls-sdk-python-v0-3-2" used for observability, which is merged after the layer created for the application, which affects the loading order of the libraries, I assume that the version of typing-extensions is older in this layer and therefore not compatible with the version of fastapi.

So, my question, how can I modify the loading order or tell serverless to load its observability layer before adding the layer for the code that we have implemented?

Edit:

Revised sls-sdk-python-v0-3-2 layer and typing-extensions version is 4.7.1 and fastapi 0.104 and later require typing-extension >=4.8.0

[rickythefox]

Just spent a few frustrating hours trying to fix this...

Here is a python script that you can run in your CI after deploy to move the sls layer to the first place and thus prevent it from overwriting typing-extensions with an older version. Make sure to pip install boto3 first.

#!/usr/bin/env python3

import sys
import boto3

def reorder_layers(lambda_name):
    # Initialize a Boto3 client for Lambda
    lambda_client = boto3.client('lambda')

    # Get the current configuration of the Lambda function, particularly the Layers
    response = lambda_client.get_function_configuration(FunctionName=lambda_name)
    layers = response.get('Layers', [])

    # Separate the desired layer from others
    reordered_layers = []
    sls_sdk_layer = None
    for layer in layers:
        if "sls-sdk-python-v0-3-2" in layer['Arn']:
            sls_sdk_layer = layer['Arn']
        else:
            reordered_layers.append(layer['Arn'])

    # Prepend 'sls-sdk-python-v0-3-2' if it exists
    if sls_sdk_layer:
        reordered_layers.insert(0, sls_sdk_layer)

    # Update the Lambda function configuration with the new layers order, if there's any change
    if reordered_layers:
        lambda_client.update_function_configuration(
            FunctionName=lambda_name,
            Layers=reordered_layers
        )
        print(f"Updated layers for Lambda function '{lambda_name}' successfully.")
    else:
        print(f"No layers to update for Lambda function '{lambda_name}'.")

if __name__ == "__main__":
    if len(sys.argv) != 2:
        print("Usage: {} <lambda_function_name>".format(sys.argv[0]))
        sys.exit(1)

    lambda_function_name = sys.argv[1]
    reorder_layers(lambda_function_name)

Other approaches:

• don't put your dependencies in a layer (fast but not always desireable)
• modify and reapply the CloudFormation template for the lambda (slower)
• write a serverless framework plugin (not sure you can modify the generated template before applying?)

[Verde1705]

I have been able to test this script and it works quite well, thanks for your help