Go — Errors and Recovery#

Go surfaces failures as error values you check on every call — there's no exception path. The rule matches every binding: only backpressure is safe to retry blindly.

if err := bus.IngestRaw(payload); err != nil {
    // Inspect the error: backpressure is retryable; serialization/config are bugs;
    // a shutdown/not-connected error is a state change retrying won't fix.
    log.Printf("ingest failed: %v", err)
}

• Backpressure — the ring buffer/window was full. Retry with backoff, or slow the producer. The only case a blind retry fixes.
• Serialization / config — a bug. Retrying reruns it.
• Shutdown / not-connected — a state change; retrying won't undo it.

The binding maps the core error kinds to Go errors with stable prefixes (the RegistryClientError / FoldQueryClientError types carry a typed Kind), so you can branch on the kind where you need to. The full taxonomy is the Error Codes reference.

Recover an RPC call#

Retry, hedge, and circuit-breaker strategies apply the same way as the other bindings — and calling a tool or service by name lets the mesh pick a provider, so a substitutable capability fails over to a standby when the primary dies. The end-to-end patterns are in Recover a Failed Workflow.

The one rule#

Retry on backpressure. Treat serialization/config errors as bugs, auth errors as "get a new credential," and shutdown / not-connected as state changes retrying won't fix.