[PYTHON] Findings when AWS Lambda gives "ELF load command address / offset not properly aligned" error

background

When developing AWS Lambda with Python, you may use the Serverless Framework and the serverless-python-requirements plugin. At that time, I was a little clogged with an error in the title, so I will summarize what was useful for solving it.

Operating environment

This article has been confirmed in the following environment.

• OS : Mac OS Mojave
• Serverless Framework : 1.54.0
• serverless-python-requirements : 5.0.1

Until you get an error

In this article, we will consider the following configuration.

requirements.txt

opencv-python

handler.py

import cv2 

def hello(event, context):
    response = {
        "statusCode": 200,
        "body": cv2.__version__
    }

    return response

serverless.yml

service: test

provider:
  name: aws
  runtime: python3.7

functions:
  hello:
    handler: handler.hello

custom:
  pythonRequirements:
    dockerizePip: true
    slim: true

plugins:
  - serverless-python-requirements

This is just a function that returns the version of OpenCV. If you deploy this and run it with sls invoke -f hello, you will get the following error.

{
    "errorMessage": "Unable to import module 'handler': /var/task/cv2/cv2.cpython-37m-x86_64-linux-gnu.so: ELF load command address/offset not properly aligned",
    "errorType": "Runtime.ImportModuleError"
}

In this article, I will summarize what to try in such situations.

Review the plugin settings

For this ʻELF load command address / offset not properly aligned` error, see the plugin README.md -not-to-strip-binaries)

custom:
  pythonRequirements:
    slim: true
    strip: false

It seems that you should write. So, rewrite serverless.yml and try again. If there is no error here, I think there is no problem. In this article, I will introduce the knowledge when an error occurs here.

Check the cache

serverless-python-requirements has a cache function so that you don't have to pip every time. https://github.com/UnitedIncome/serverless-python-requirements/blob/master/README.md#caching

The previous change to strip: false seems to use the cache, and the changed package will not be reflected. For example, sls remove does not remove the cache, so ʻELF load command address / offset not properly aligned` will continue to appear.

Therefore,

$ rm -rf /Users/xxx/Library/Caches/serverless-python-requirements

It may be solved by deleting the cache and then redeploying as in. (Of course, the path will change depending on the environment.)

Summary

If ELF load command address / offset not properly aligned occurs

• Check the plugin settings
• Check the cache

I think it's good. If you have any other knowledge, please let us know.