A simple Python package for refreshing the temporary security credentials in a boto3.session.Session object automatically.

Features

• Drop-in replacement for boto3.session.Session
• Supports automatic credential refresh methods for various AWS services:
  • STS
  • ECS
• Supports assume_role configuration, custom STS clients, and profile / region configuration, as well as all other parameters supported by boto3.session.Session
• Tested, documented, and published to PyPI

Recognition, Adoption, and Testimonials

Featured in TL;DR Sec.

Featured in CloudSecList.

Recognized during AWS Community Day Midwest on June 5th, 2025.

A testimonial from a Cyber Security Engineer at a FAANG company:

Most of my work is on tooling related to AWS security, so I'm pretty choosy about boto3 credentials-adjacent code. I often opt to just write this sort of thing myself so I at least know that I can reason about it. But I found boto3-refresh-session to be very clean and intuitive [...] We're using the RefreshableSession class as part of a client cache construct [...] We're using AWS Lambda to perform lots of operations across several regions in hundreds of accounts, over and over again, all day every day. And it turns out that there's a surprising amount of overhead to creating boto3 clients (mostly deserializing service definition json), so we can run MUCH more efficiently if we keep a cache of clients, all equipped with automatically refreshing sessions.

The following line plot illustrates the adoption of BRS over the last three months in terms of average daily downloads over a rolling seven day window.

Installation

pip install boto3-refresh-session

Usage

import boto3_refresh_session as brs

# you can pass all of the params associated with boto3.session.Session
profile_name = '<your-profile-name>'
region_name = 'us-east-1'
...

# as well as all of the params associated with STS.Client.assume_role
assume_role_kwargs = {
  'RoleArn': '<your-role-arn>',
  'RoleSessionName': '<your-role-session-name>',
  'DurationSeconds': '<your-selection>',
  ...
}

# as well as all of the params associated with STS.Client, except for 'service_name'
sts_client_kwargs = {
  'region_name': region_name,
  ...
}

# basic initialization of boto3.session.Session
session = brs.RefreshableSession(
  assume_role_kwargs=assume_role_kwargs, # required
  sts_client_kwargs=sts_client_kwargs,
  region_name=region_name,
  profile_name=profile_name,
  ...
)

# now you can create clients, resources, etc. without worrying about expired temporary 
# security credentials
s3 = session.client(service_name='s3')
buckets = s3.list_buckets()

Raison d'être

Long-running data pipelines, security tooling, ETL jobs, and cloud automation scripts frequently interact with the AWS API using boto3 — and often run into the same problem:

Temporary credentials expire.

When that happens, engineers typically fall back on one of two strategies:

• Wrapping AWS calls in try/except blocks that catch ClientError exceptions
• Writing ad hoc logic to refresh credentials using botocore credentials internals

Both approaches are fragile, tedious to maintain, and error-prone at scale.

Over the years, I noticed that every company I worked for — whether a scrappy startup or FAANG — ended up with some variation of the same pattern:
a small in-house module to manage credential refresh, written in haste, duplicated across services, and riddled with edge cases. Things only got more strange and difficult when I needed to run things in parallel.

Eventually, I decided to build boto3-refresh-session as a proper open-source Python package:

• Fully tested
• Extensible
• Integrated with boto3 idioms
• Equipped with automatic documentation and CI tooling

The goal: to solve a real, recurring problem once — cleanly, consistently, and for everyone - with multiple refresh strategies.

If you've ever written the same AWS credential-refresh boilerplate more than once, this library is for you.