boto3-client-cache 2.1.19

What is boto3-client-cache?

boto3-client-cache provides a concurrency-safe, bounded cache for boto3 client and resource objects with deterministic identity semantics. LRU and LFU eviction are supported.

boto3-client-cache was authored by Mike Letts and is maintained by 61418.

Why it exists

boto3 clients and resources consume a large amount of memory. Many developers never notice this. At scale, however, the memory footprint of boto3 clients and resources often becomes clear through manifold consequences. Caching is an obvious choice for managing multiple clients and-or resources at scale.

boto3 does not cache client or resource objects natively. There are also, to my knowledge, no other open-source tools available which do what boto3-client-cache does. To compensate, bespoke caching solutions circulate online. boto3-client-cache exists to standardize and democratize client and resource caching for the Python AWS community.

Design

The most important but challenging design choice for client and resource caching is selecting and enforcing a robust and standardized methodology for unique keys. boto3-client-cache hashes according to boto3 client and resource signatures.

Setting and retrieving clients and resources from the client cache therefore requires an explicit declaration of intention — that is, the developer must explicitly pass client and resource initialization parameters to a ClientCacheKey or ResourceCacheKey object in order to set or retrieve boto3 clients. This ensures setting and retrieving clients and resources are unambiguous and deterministic operations. By locking the cache, as boto3-client-cache does, race conditions are prevented, enabling developers to confidently employ the cache at scale with predictable cache eviction behavior. Lastly, by designing the cache like a dict in the standard Python library, the cache is ergonomically familiar and thus easy to use.

These decisions reflect the core design goals of boto3-client-cache: safety at scale, deterministic behavior, ergonomic interfacing, and explicit identity.

Installation

pip install boto3-client-cache

High-level API

The high-level API is ergonomically identical to boto3's API, but with caching capabilities built in.

To use it, you can initialize a client or resource which is automatically cached.

from boto3_client_cache import client

# you can specify eviction_policy and max_size by passing them as args
# here, they are excluded so the defaults (eviction_policy="LRU", max_size=10) are used
s3 = client("s3", region_name="us-west-2")
s3_again = client("s3", region_name="us-west-2")
assert s3 is s3_again  # True

Or initialize a client or resource from a session directly.

from boto3_client_cache import Session

session = Session(profile_name="default")
s3 = session.client("s3")
s3_again = session.client("s3")
assert s3 is s3_again  # True

Low-level API

The low-level API offers more control and flexibility, allowing you to manage multiple caches with different eviction policies and configurations. The low-level API must be used in tandem with boto3.

from boto3_client_cache import ClientCache, ClientCacheKey
import boto3

# create an LRU client cache with a maximum size of 30
cache = ClientCache(max_size=30)

# store boto3 client params in an object
kwargs = {"service_name": "s3", "region_name": "us-west-2"}

# create a cache key using those params
key = ClientCacheKey(**kwargs)

# assign a client
cache[key] = boto3.client(**kwargs)

# and retrieve that client using the key
s3_client = cache[key]

Error semantics

Errors for resources are identical to errors for clients, except that the word "Client" is replaced with "Resource" in the exception name.

# raises ClientCacheExistsError b/c client(**kwargs) already exists
# (from the pseudo-code above)
cache[key] = boto3.client(**kwargs)

# raises ClientCacheNotFoundError b/c the specific client was not
# cached above
cache[ClientCacheKey(service_name="ec2", region_name="us-west-2")]

# returns None instead of raising ClientCacheNotFoundError
cache.get(ClientCacheKey(service_name="ec2", region_name="us-west-2"))

# raises ClientCacheError b/c the key is not a ClientCacheKey
cache["this is not a ClientCacheKey"]

# raises ClientCacheError b/c the object is not a client
cache[ClientCacheKey("s3")] = "this is not a boto3 client"

License

Beginning v2.1.3, boto3-client-cache is licensed under the Mozilla Public License 2.0 (MPL-2.0). Earlier versions remain licensed under the Apache Software License 2.0.

Contributing

Refer to the contributing guidelines for additional information on how to contribute to boto3-client-cache.

Special thanks

• Patrick Sanders
• Ben Kehoe