What is Code Resilience?

Resilient code assumes failure is normal. Every external call has a timeout, because a call without one will eventually hang and consume a thread. Retries use exponential backoff with jitter so that a recovering dependency is not immediately overwhelmed by synchronized retry storms.

Beyond that, resilient code isolates failure. Bulkheads prevent one slow dependency from exhausting a shared connection pool. Circuit breakers stop calls to a failing service rather than queuing them. Fallbacks return degraded but useful responses when the ideal response is unavailable.

Resilience determines whether a dependency failure is contained or propagates. In a distributed system, that difference is usually the difference between a minor degradation and a full outage.

• Failure containment: one unhealthy service does not take down its callers.
• Cascade prevention: timeouts and breakers stop failures from spreading upward.
• Partial availability: core functionality survives when auxiliary features do not.
• Recovery behavior: the system returns to health without manual intervention.

Resilience patterns add complexity and can hide problems. A fallback that silently returns stale data may keep the error rate at zero while users receive wrong answers, which is harder to detect than an outright failure.

Retries are frequently misapplied. Retrying a request that failed because the downstream service is overloaded adds load to an overloaded service. Retry is correct for transient faults and actively harmful for saturation, and distinguishing between the two at runtime is not always easy.

Infrastructure resilience comes from redundancy: multiple replicas, multiple zones, load balancers that route around unhealthy instances. It handles the failure of a machine well.

Code resilience handles the failure of a behavior. If a dependency returns errors from every replica, redundancy does not help, because the problem is not that a host died. Infrastructure covers failures of the substrate, code covers failures of logic and dependencies, and a system needs both.