如何让基于Zbar的Python二维码识别函数在AWS Lambda中运行？

Hey Miguel, let's work through getting your QR code scanner running on AWS Lambda—this is a super common pitfall for folks new to Lambda, so let's break it down step by step, plus cover some easier alternatives if you don't want to mess with dependency packaging.

Your local code works because you're reading from a local file path, but Lambda has two key differences you're missing:

1. You can't directly read S3 objects with cv2.imread()—you need to download the file to Lambda's temporary /tmp/ directory first.
2. The dependencies you installed locally (pyzbar, opencv-python, pandas) are built for your machine's OS, but Lambda runs on Amazon Linux 2. Mismatched binaries are almost certainly causing your errors.

1. Update Your Code for S3 + Lambda #

Here's a revised version of your function that handles S3 events correctly:

import boto3
from pyzbar import pyzbar
import cv2
import pandas as pd

# Initialize AWS clients once (outside the handler for better performance)
s3 = boto3.client('s3')

def lambda_handler(event, context):
    # Pull S3 bucket/key from the trigger event
    try:
        bucket = event['Records'][0]['s3']['bucket']['name']
        key = event['Records'][0]['s3']['object']['key']
    except KeyError:
        return {"error": "Invalid S3 trigger event"}

    # Download the image to Lambda's temporary storage
    local_image_path = f"/tmp/{key.split('/')[-1]}"
    s3.download_file(bucket, key, local_image_path)

    # Scan for QR codes
    image = cv2.imread(local_image_path)
    barcodes = pyzbar.decode(image)

    # Filter and format results
    qr_results = []
    for barcode in barcodes:
        barcode_data = barcode.data.decode("utf-8")
        barcode_type = barcode.type
        if barcode_type == 'QRCODE':
            qr_results.append({"type": barcode_type, "url": barcode_data})

    return qr_results if qr_results else {"message": "No QR codes detected"}

2. Package Dependencies Correctly for Lambda #

You need to install dependencies built for Amazon Linux 2, not your local machine. Two easy ways to do this:

Option A: Use Docker to Build Dependencies #

If you have Docker installed locally, run this command in your project folder to install dependencies in a Lambda-like environment:

docker run -v "$PWD":/var/task public.ecr.aws/sam/build-python3.9:latest /bin/sh -c "pip install pyzbar opencv-python pandas -t .; exit"

Then add your revised lambda_function.py to the folder, zip everything up, and upload to Lambda.

Option B: Use an Amazon Linux 2 EC2 Instance #

1. Spin up a t2.micro Amazon Linux 2 instance (free tier eligible)
2. Install Python 3.9+ and pip:

   sudo yum install python39 python39-pip -y
   

3. Create a folder, install dependencies into it, and zip:

   mkdir lambda-package
   cd lambda-package
   pip3.9 install pyzbar opencv-python pandas --target .
   cp /path/to/your/lambda_function.py .
   zip -r lambda-deploy.zip .
   

   Download the zip to your local machine and upload to Lambda.

If you don't want to deal with dependency headaches, AWS Rekognition has a built-in barcode detection feature that supports QR codes. No third-party dependencies needed—just use the native boto3 client.

Here's a simplified function using Rekognition:

import boto3

# Initialize clients once
rekognition = boto3.client('rekognition')
s3 = boto3.client('s3')

def lambda_handler(event, context):
    try:
        bucket = event['Records'][0]['s3']['bucket']['name']
        key = event['Records'][0]['s3']['object']['key']
    except KeyError:
        return {"error": "Invalid S3 trigger event"}

    # Call Rekognition to detect barcodes
    response = rekognition.detect_barcodes(
        Image={
            'S3Object': {
                'Bucket': bucket,
                'Name': key
            }
        }
    )

    # Filter for QR codes
    qr_results = []
    for barcode in response['Barcodes']:
        if barcode['Type'] == 'QR_CODE':
            qr_results.append({
                "type": barcode['Type'],
                "url": barcode['Value']
            })

    return qr_results if qr_results else {"message": "No QR codes detected"}

Why this is better:

• No dependency packaging required—Lambda already includes boto3
• AWS maintains the barcode recognition model, so it's accurate and updated
• Faster to implement and debug

• Permissions: Make sure your Lambda execution role has:
  • s3:GetObject permission for your target S3 bucket
  • If using Rekognition, add rekognition:DetectBarcodes permission
• Memory/Timeout: For larger images, bump Lambda's memory to 512MB+ (more memory = faster CPU) and set a timeout of 10-15 seconds
• Temporary Storage: Lambda's /tmp/ folder has a 512MB limit, so avoid processing huge images

内容的提问来源于stack exchange，提问作者Miguel M.