golem-retry-policies-ts

Configure retry policies for transient HTTP failures in Golem agents by defining a declarative retryPolicyDefaults block in golem.yaml. You can target specific error context properties like error-type, uri-scheme, and status-code using prioritized predicates to select the right policy for each failure context.

Apply exponential backoff with jitter to TypeScript retry strategies by composing countBox and timeBox limits with composition operators like andThen, union, and intersect. You can add optional clamping and jitter to ensure retry timing matches your production needs without overwhelming the system.

Predicate routing for status-code retries in Golem works by matching specific error context properties such as error-type, uri-scheme, status-code, and trap-type. Prioritized predicates evaluate these properties to select the appropriate retry policy for each distinct failure context across HTTP, RPC, DB, and trap scenarios.

You can manage retry policies live via the CLI without restarting your Golem agents. The CLI management interface allows you to apply, inspect, and update policies at runtime, while the TypeScript SDK provides withRetryPolicy, getRetryPolicies, and getRetryPolicyByName functions for code-level control.

Set per-environment retry rules in golem.yaml by defining a retryPolicyDefaults block tailored to each environment. For production, you can select transient errors and HTTPS with an exponential backoff policy that is clamped and jittered, ensuring predictable, fine-grained retry behavior that matches each environment's specific requirements.

Use the intersect operator when you need a retry strategy that satisfies multiple policy conditions simultaneously, and use the union operator when any single policy condition should trigger retries. Combine these with andThen to sequence strategies, allowing precise control over countBox, timeBox, and backoff behaviors across different failure contexts.