fluid-core/internal/runtime/telemetry_test.go

package runtime

import (
	"context"
	"errors"
	"testing"
	"time"

	"github.com/tegwick/fluid-core/internal/contract"
)

// stallingSink never returns, standing in for a wedged evidence store.
type stallingSink struct{ entered chan struct{} }

func (s *stallingSink) Write(ctx context.Context, _ contract.FluidTelemetry) error {
	select {
	case s.entered <- struct{}{}:
	default:
	}
	<-ctx.Done()
	return ctx.Err()
}

// TestEmitNeverBlocks is the test for Blueprint invariant 34.2. If it ever
// starts failing, the data plane has acquired a dependency on the observation
// plane and the isolation the whole architecture rests on is gone.
func TestEmitNeverBlocks(t *testing.T) {
	sink := &stallingSink{entered: make(chan struct{}, 1)}
	e := NewEmitter(sink, EmitterOptions{Buffer: 8, Workers: 1, WriteTimeout: time.Hour})
	defer func() {
		// The stalled worker cannot drain, so do not wait on Close.
		_ = e
	}()

	<-func() chan struct{} {
		e.Emit(contract.FluidTelemetry{ID: "warm"})
		return sink.entered
	}()

	done := make(chan struct{})
	go func() {
		// Far more than the buffer holds, against a sink that never completes.
		for i := 0; i < 10_000; i++ {
			e.Emit(contract.FluidTelemetry{ID: "ev"})
		}
		close(done)
	}()

	select {
	case <-done:
	case <-time.After(2 * time.Second):
		t.Fatal("Emit blocked while the sink was stalled; the request path is now coupled to telemetry")
	}

	if e.Stats().Dropped == 0 {
		t.Error("expected drops against a stalled sink, got none")
	}
}

// countingSink records deliveries.
type countingSink struct {
	ch  chan contract.FluidTelemetry
	err error
}

func (s *countingSink) Write(_ context.Context, ev contract.FluidTelemetry) error {
	if s.err != nil {
		return s.err
	}
	s.ch <- ev
	return nil
}

func TestEmitterDeliversAndDrains(t *testing.T) {
	sink := &countingSink{ch: make(chan contract.FluidTelemetry, 32)}
	e := NewEmitter(sink, EmitterOptions{Buffer: 32, Workers: 2})

	for i := 0; i < 10; i++ {
		e.Emit(contract.FluidTelemetry{ID: "ev", Kind: contract.FluidTelemetryKindRequest})
	}
	e.Close()

	if got := len(sink.ch); got != 10 {
		t.Errorf("delivered %d events, want 10", got)
	}
	if s := e.Stats(); s.Written != 10 || s.Dropped != 0 {
		t.Errorf("stats = %+v, want 10 written and 0 dropped", s)
	}
}

func TestEmitterCountsSinkFailures(t *testing.T) {
	e := NewEmitter(&countingSink{err: errors.New("store down")}, EmitterOptions{Buffer: 4, Workers: 1})
	for i := 0; i < 4; i++ {
		e.Emit(contract.FluidTelemetry{ID: "ev"})
	}
	e.Close()

	if s := e.Stats(); s.Failed == 0 {
		t.Errorf("sink failures not counted: %+v", s)
	}
}