asyncresil 0.1.0

Async resilience utilities — circuit breaker, rate limiter, cache, http client

asyncresil

Async resilience utilities for Python services. A small, focused toolkit extracted from production code so the same primitives can be shared across projects instead of re-implemented per repo.

The GitHub repo is named aura-utils for historical reasons (this library was extracted from the Aura project's mcp_server/utils/). The PyPI package and Python module are both asyncresil — installed via pip install asyncresil and imported as from asyncresil import ….

Modules

Module	What it does	Why it exists
http_client	httpx-based async client with retries, exponential backoff, and timeouts	Transient network failures shouldn't surface to callers — retry the obvious cases.
circuit_breaker	State machine (closed → open → half_open) that fails fast when a downstream is broken	Stops cascading failures and retry storms when a dependency is down.
rate_limiter	Token-bucket limiter for async code	Stay under third-party API limits, smooth bursts.
cache	Cache protocol with MemoryCache (default) and RedisCache (optional dep)	Skip repeat work — cut latency, cost, and downstream load.

The four compose into a typical outbound call chain:

caller → cache → rate_limiter → circuit_breaker → http_client → external service

Each layer protects the layers below from unnecessary load.

Install

pip install asyncresil                # core (in-memory cache only)
pip install 'asyncresil[redis]'       # adds Redis-backed cache

Requires Python 3.11+.

Intended consumers

Internal services that need a shared resilience layer:

• Aura — MCP server (the original source of these utilities)
• Eval Framework — uses the cache + rate limiter for batched LLM evaluation
• Agentic Engine — uses the circuit breaker around tool calls
• LLM Gateway — uses the http client + circuit breaker for upstream provider calls

Status

0.1.0 — early alpha. API is not yet stable.

License

MIT — see LICENSE.