tower-resilience
A comprehensive resilience and fault-tolerance toolkit for Tower services, inspired by Resilience4j.
About
Tower-resilience provides composable middleware for building robust distributed systems in Rust. Tower is a library of modular and reusable components for building robust networking clients and servers. This crate extends Tower with resilience patterns commonly needed in production systems.
Inspired by Resilience4j, a fault tolerance library for Java, tower-resilience adapts these battle-tested patterns to Rust's async ecosystem and Tower's middleware model.
Resilience Patterns
- Circuit Breaker - Prevents cascading failures by stopping calls to failing services
- Bulkhead - Isolates resources to prevent system-wide failures
- Time Limiter - Advanced timeout handling with cancellation support
- Retry - Intelligent retry with exponential backoff and jitter
- Rate Limiter - Controls request rate to protect services
- Cache - Response memoization to reduce load
Features
- Composable - Stack multiple resilience patterns using Tower's ServiceBuilder
- Observable - Event system for monitoring pattern behavior (retries, state changes, etc.)
- Configurable - Builder APIs with sensible defaults
- Async-first - Built on tokio for async Rust applications
- Zero-cost abstractions - Minimal overhead when patterns aren't triggered
Quick Start
[]
= "0.1"
= "0.5"
use ServiceBuilder;
use *;
let service = new
.layer
.layer
.service;
Examples
Circuit Breaker
Prevent cascading failures by opening the circuit when error rate exceeds threshold:
use CircuitBreakerLayer;
let layer = builder
.failure_rate_threshold // Open at 50% failure rate
.sliding_window_size // Track last 100 calls
.build;
See examples/circuitbreaker.rs for a complete example.
Bulkhead
Limit concurrent requests to prevent resource exhaustion:
use BulkheadLayer;
let layer = builder
.max_concurrent_calls
.wait_timeout
.build;
See examples/bulkhead.rs for a complete example.
Time Limiter
Enforce timeouts on operations:
use TimeLimiterLayer;
let layer = builder
.timeout_duration
.cancel_running_future
.build;
Retry
Retry failed requests with exponential backoff:
use RetryLayer;
let layer = builder
.max_attempts
.exponential_backoff
.build;
Rate Limiter
Control request rate to protect downstream services:
use RateLimiterLayer;
let layer = builder
.max_permits
.refresh_period
.build;
Cache
Cache responses to reduce load on expensive operations:
use CacheLayer;
let layer = builder
.max_size
.ttl
.key_extractor
.build;
Error Handling
Zero-Boilerplate with ResilienceError
When composing multiple resilience layers, use ResilienceError<E> to eliminate manual error conversion code:
use ResilienceError;
// Your application error
// That's it! No From implementations needed
type ServiceError = ;
// All resilience layer errors automatically convert
let service = new
.layer
.layer
.layer
.service;
Benefits:
- Zero boilerplate - no
Fromtrait implementations - Rich error context (layer names, counts, durations)
- Convenient helpers:
is_timeout(),is_rate_limited(), etc.
See the Layer Composition Guide for details.
Manual Error Handling
For specific use cases, you can still implement custom error types with manual From conversions. See examples for both approaches.
Pattern Composition
Stack multiple patterns for comprehensive resilience:
use ServiceBuilder;
// Client-side: timeout -> circuit breaker -> retry
let client = new
.layer
.layer
.layer
.service;
// Server-side: rate limit -> bulkhead -> timeout
let server = new
.layer
.layer
.layer
.service;
Performance
Benchmarks measure the overhead of each pattern in the happy path (no failures, circuit closed, permits available):
| Pattern | Overhead (ns) | vs Baseline |
|---|---|---|
| Baseline (no middleware) | ~10 ns | 1.0x |
| Retry (no retries) | ~80-100 ns | ~8-10x |
| Time Limiter | ~107 ns | ~10x |
| Rate Limiter | ~124 ns | ~12x |
| Bulkhead | ~162 ns | ~16x |
| Cache (hit) | ~250 ns | ~25x |
| Circuit Breaker (closed) | ~298 ns | ~29x |
| Circuit Breaker + Bulkhead | ~413 ns | ~40x |
Key Takeaways:
- All patterns add < 300ns overhead individually
- Overhead is additive when composing patterns
- Even the heaviest pattern (circuit breaker) is negligible for most use cases
- Retry and time limiter are the lightest weight options
Run benchmarks yourself:
Documentation
- API Documentation
- Pattern Guides - In-depth guides on when and how to use each pattern
Examples
Two sets of examples are provided:
- Top-level examples - Simple, getting-started examples matching this README (one per pattern)
- Module examples - Detailed examples in each crate's
examples/directory showing advanced features
Run top-level examples with:
# etc.
Why tower-resilience?
Tower provides some built-in resilience (timeout, retry, rate limiting), but tower-resilience offers:
- Circuit Breaker - Not available in Tower
- Advanced retry - More backoff strategies and better control
- Bulkhead - True resource isolation with async-aware semaphores
- Unified events - Consistent observability across all patterns
- Builder APIs - Ergonomic configuration with sensible defaults
- Production-ready - Patterns inspired by battle-tested Resilience4j
Minimum Supported Rust Version (MSRV)
This crate's MSRV is 1.64.0, matching Tower's MSRV policy.
We follow Tower's approach:
- MSRV bumps are not considered breaking changes
- When increasing MSRV, the new version must have been released at least 6 months ago
- MSRV is tested in CI to prevent unintentional increases
License
Licensed under either of:
- Apache License, Version 2.0 (LICENSE-APACHE or http://www.apache.org/licenses/LICENSE-2.0)
- MIT license (LICENSE-MIT or http://opensource.org/licenses/MIT)
at your option.
Contributing
Contributions are welcome! Please see the contributing guidelines for more information.