Python Plus AWS SDK: The Boto3 Patterns I Actually Use

Boto3 is the AWS SDK for Python. It is also a big surface area with a lot of ways to shoot yourself in the foot. After six months of daily boto3 use in production, I have settled on a short list of patterns that work. Everything else I avoid.

Use Resources, Not Clients (Mostly)

Boto3 has two abstractions: clients (low-level, one method per API call) and resources (higher-level, object-oriented). I default to resources for the basics and reach for clients when I need control:

import boto3
 
table = boto3.resource('dynamodb').Table('posts')
table.put_item(Item=payload)  # clean
 
ddb = boto3.client('dynamodb')
ddb.transact_write_items(TransactItems=[...])  # need client for this

Resources handle pagination, serialization, and most of the tedious bits. Clients are needed for features the resource layer does not expose, like transactions or batch operations at scale.

Single Client per Process

Creating a boto3 client is expensive. Each one opens connection pools and loads configuration. I create clients at module load, never inside request handlers:

# module load
DDB = boto3.resource('dynamodb')
POSTS = DDB.Table('posts')
 
def handler(event):
    return POSTS.get_item(Key={'slug': event['slug']})

In Lambda that means the client is reused across warm invocations. Cold starts pay once, warm invocations are free.

Always Paginate

Almost every list or scan operation returns a page, not the full result set. The paginator abstraction handles that cleanly:

paginator = ddb_client.get_paginator('scan')
for page in paginator.paginate(TableName='posts'):
    for item in page['Items']:
        process(item)

Forgetting to paginate is how you get silent data truncation at 1MB. Always use the paginator.

Handle ClientError Specifically

Boto3 raises ClientError for most AWS failures. The error code lives in a nested dict:

from botocore.exceptions import ClientError
try:
    table.put_item(Item=payload, ConditionExpression='attribute_not_exists(slug)')
except ClientError as e:
    if e.response['Error']['Code'] == 'ConditionalCheckFailedException':
        return already_exists()
    raise

Catching ClientError broadly and branching on the code is how you get clean business logic without losing visibility into which specific failure happened.

Read the boto3 documentation for the full surface area.