Components, dependencies and lifecycle

The service state is an explicit tree under backplane.Root[Config]. Components inherit a named logger, tracer and meter. Providers acquire resources; dependency wrappers determine startup and readiness behavior.

type Greeter struct {
    deps.Component
    db deps.Dependency[*store.DB]
}

func New(parent deps.Scope, db deps.Dependency[*store.DB]) *Greeter {
    g := &Greeter{
        Component: deps.NewComponent(parent, "greeter"),
        db: db,
    }
    g.Go(g.report)
    return g
}

This fragment assumes your store.DB and report(context.Context) error. Background functions should stop when their context is canceled. Use the component's lifecycle instead of launching untracked goroutines.

Choose the correct wrapper

Wrapper	Startup and use
deps.NewDependency	Required resource; startup waits while acquisition retries, readiness follows the resource probe
deps.NewOptional	Startup can continue without the resource; acquisition keeps retrying
deps.NewSingleton	Lazily acquired on first use, released during shutdown; not automatically required for readiness

The hello state constructs a required store and optional cache:

Store: deps.NewDependency(root, store.New(&cfg.Store), deps.ProbeTimeout(time.Second)),
Cache: deps.NewOptional(root, store.New(&cfg.Cache), deps.Name("cache"),
    deps.ProbeOptional(), deps.Backoff(time.Second, 10*time.Second)),

Optional absence must be handled by business logic. Do not dereference a missing optional value or silently convert a required dependency into an optional one to make readiness green.

Probes

Readiness answers whether this instance can serve its declared work. Liveness answers whether the process is progressing. A temporary downstream outage usually affects readiness, not liveness. Attach additional probes through svc.ReadinessProbe and svc.LivenessProbe; hello wraps errors with probe.FromError.

Return a useful error when not ready. The probe integration preserves a reason, allowing operators to distinguish startup, a failed resource and an application-specific rejection. Avoid credentials or payloads in those reasons.

Telemetry and shutdown

Use component.Log(), Tracer() and Meter() so component identity remains attached. Span and deps.SpanValue wrap work with the component tracer; carry the provided context into downstream calls.

Resources should be released by their provider/lifecycle contract. The SDK uses the shared xshutdown/lifecycle machinery rather than a second service-specific lifecycle implementation. Verify that cancellation unblocks network calls, loops and dependency acquisition; a goroutine waiting forever can prevent clean shutdown.

The hello store and greeter demonstrate providers, lazy templates, optional cache recovery, probes and a background reporting loop.