retry

Retry refers to the repeated attempt to complete a task after a previous failed attempt, typically prompted by a transient error or delay. In computing, retries are common in distributed systems, network communication, and input/output operations to improve reliability in the face of temporary outages or timeouts. A retry policy governs when to retry, how many times, and how long to wait between attempts.

Retry is beneficial for transient faults such as short-lived network glitches or a temporarily unavailable service. However, retries can cause duplicate effects if the operation is not idempotent. Therefore, operations should be designed to be idempotent, or retries should be accompanied by compensating actions. Distinguishing retryable errors (transient) from non-retryable errors (permanent) is crucial.

Common strategies include fixed delays between attempts, exponential backoff (increasing the wait time after each failure), and capped backoff (limiting the maximum delay). Adding jitter, or randomness, to delays helps prevent synchronized retries across many clients. Retries are typically bounded by a maximum number of attempts or a total time limit.

To prevent retry storms and cascading failures, systems may combine retries with circuit breakers and rate limiting. Observability is important to distinguish successful and failed retries. In distributed systems, server-side retries and client-side retries may interact, so design decisions should consider idempotence, timeout policies, and eventual consistency.

Retry is related to backoff strategies, timeout policies, circuit breakers, and failover. It is a fundamental technique in resilience engineering for unreliable networks and services.