summer-resilience Plugin

summer-resilience provides named resilience policies for asynchronous operations

summer-resilience provides named resilience policies for asynchronous application operations.

Retry configuration

[resilience.retry.instances.inventory]
max_attempts = 3
wait_duration = 100
enable_exponential_backoff = true
exponential_backoff_multiplier = 2.0
exponential_max_wait_duration = 2000
enable_randomized_wait = true
randomized_wait_factor = 0.5

Durations are expressed in milliseconds. max_attempts includes the initial call.

Retry macro

Add ResiliencePlugin to the application, then refer to a configured policy by name:

use summer::App;
use summer_resilience::{retry, ResiliencePlugin};

#[retry(name = "inventory")]
async fn load_inventory(id: String) -> Result<String, InventoryError> {
    inventory_client().load(id).await
}

App::new().add_plugin(ResiliencePlugin);

All errors are retried by default. A predicate can classify errors that are safe to retry:

#[retry(name = "inventory", retry_if = is_transient)]
async fn load_inventory(id: String) -> Result<String, InventoryError> {
    inventory_client().load(id).await
}

fn is_transient(error: &InventoryError) -> bool {
    error.is_timeout() || error.is_unavailable()
}

The operation's final error is returned unchanged. Retried arguments are cloned before every attempt, so owned function parameters must implement Clone. Mutable receivers and destructured parameter patterns are not supported.

Retrying can repeat side effects. Operations such as payments, writes, and message publication need their own idempotency guarantees.