Test services and mock the controller¶

Service lifecycle functions are ordinary functions and the controller is a concrete type you can drive directly, so most testing needs no mocks at all. When your own code depends on a controller, the Controllable interface (and the narrower interfaces beside it) let you substitute a fake.

Signals are already off. A controller installs no SIGINT/SIGTERM handler unless you ask for one, so a test needs no option to stay out of process-wide state. Just never pass WithSignals. Drive shutdown explicitly with Stop(), or by cancelling the context you constructed it with.

Test your Start/Stop/Status functions directly¶

StartFunc, StopFunc, and StatusFunc are plain functions, so the simplest test calls them without a controller at all:

func TestStatusProbe(t *testing.T) {
    srv := newServer()

    status := statusFunc(srv) // returns controls.StatusFunc == func() error
    require.NoError(t, status())

    srv.forceUnhealthy()
    require.Error(t, status())
}

Drive a real controller¶

For a lifecycle test, register against a real controller and drive it. This exercises concurrent startup, health aggregation, and shutdown for real:

func TestServiceLifecycle(t *testing.T) {
    ctx := context.Background()

    c := controls.NewController(ctx)

    var started, stopped atomic.Bool
    c.Register("worker",
        controls.WithStart(func(ctx context.Context) error {
            started.Store(true)
            <-ctx.Done() // serve until shutdown
            return ctx.Err()
        }),
        controls.WithStop(func(context.Context) { stopped.Store(true) }),
        controls.WithReadiness(func() error { return nil }),
    )

    c.Start()
    require.Eventually(t, started.Load, time.Second, time.Millisecond)

    // Health is read from the aggregated report, not a channel.
    require.True(t, c.Readiness().OverallHealthy)

    c.Stop()
    c.Wait() // blocks until the full shutdown sequence has finished
    require.True(t, stopped.Load())
}

Note that health is inspected with Status(), Liveness() and Readiness(), which return a HealthReport. A StatusFunc returns nil when healthy and an error otherwise; there is no health channel to read in application code.

The suite in this repository times shutdown and restart against the wall clock rather than a fake one, and a test of yours that does the same will be sensitive to a loaded CI runner. Give require.Eventually a generous ceiling and a short poll rather than a fixed time.Sleep.

Mock the controller with the Controllable interface¶

Production code should hold the concrete *controls.Controller. But code that merely registers services against a controller can depend on an interface and be tested with a fake. The module ships no mocks, so generate one for the interface you depend on (for example with mockery), or hand-write a small stub:

// Your code depends on the interface, not the concrete controller.
func wireServices(c controls.Controllable, deps Deps) {
    c.Register("api", controls.WithStart(deps.serve), controls.WithStop(deps.drain))
}

func TestWireServices(t *testing.T) {
    fake := NewMockControllable(t) // your generated mock
    fake.EXPECT().Register("api", mock.Anything, mock.Anything).Return()

    wireServices(fake, testDeps())
}

Depend on the narrowest interface¶

Controllable is the full surface. Prefer a narrower interface when your code only needs part of it, which makes the dependency (and the mock) smaller:

type waiter interface{ Wait() }

Reserve the concrete *controls.Controller for production wiring; reach for the interfaces at test seams.

Related¶

• Interfaces reference: every interface's methods, and the methods (Wait, WaitContext, RegisterHealthCheck, WaitGroup) that are on none of them.
• Register and run services
• Add health checks
• Handle graceful shutdown and signals