galaxy-game/lobby/internal/telemetry/runtime_test.go

package telemetry

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

	"github.com/stretchr/testify/assert"
	"github.com/stretchr/testify/require"
	sdkmetric "go.opentelemetry.io/otel/sdk/metric"
	"go.opentelemetry.io/otel/sdk/metric/metricdata"
)

func TestProcessConfigValidate(t *testing.T) {
	t.Parallel()

	tests := []struct {
		name    string
		cfg     ProcessConfig
		wantErr string
	}{
		{name: "defaults", cfg: ProcessConfig{TracesExporter: "none", MetricsExporter: "none"}},
		{name: "otlp", cfg: ProcessConfig{TracesExporter: "otlp", MetricsExporter: "otlp", TracesProtocol: "grpc", MetricsProtocol: "http/protobuf"}},
		{name: "bad traces exporter", cfg: ProcessConfig{TracesExporter: "weird", MetricsExporter: "none"}, wantErr: "unsupported traces exporter"},
		{name: "bad metrics exporter", cfg: ProcessConfig{TracesExporter: "none", MetricsExporter: "weird"}, wantErr: "unsupported metrics exporter"},
		{name: "bad traces protocol", cfg: ProcessConfig{TracesExporter: "none", MetricsExporter: "none", TracesProtocol: "xyz"}, wantErr: "OTLP traces protocol"},
		{name: "bad metrics protocol", cfg: ProcessConfig{TracesExporter: "none", MetricsExporter: "none", MetricsProtocol: "xyz"}, wantErr: "OTLP metrics protocol"},
	}

	for _, tt := range tests {
		tt := tt
		t.Run(tt.name, func(t *testing.T) {
			t.Parallel()

			err := tt.cfg.Validate()
			if tt.wantErr == "" {
				require.NoError(t, err)
				return
			}
			require.Error(t, err)
			require.Contains(t, err.Error(), tt.wantErr)
		})
	}
}

func TestNewProcessNoExporters(t *testing.T) {
	runtime, err := NewProcess(context.Background(), ProcessConfig{
		ServiceName:     "galaxy-lobby-test",
		TracesExporter:  "none",
		MetricsExporter: "none",
	}, nil)
	require.NoError(t, err)
	require.NotNil(t, runtime)

	assert.NotNil(t, runtime.TracerProvider())
	assert.NotNil(t, runtime.MeterProvider())

	require.NoError(t, runtime.Shutdown(context.Background()))
	require.NoError(t, runtime.Shutdown(context.Background())) // idempotent
}

func TestNewProcessNilContext(t *testing.T) {
	t.Parallel()

	_, err := NewProcess(nil, ProcessConfig{TracesExporter: "none", MetricsExporter: "none"}, nil) //nolint:staticcheck // test exercises the nil-context guard.
	require.Error(t, err)
	require.Contains(t, err.Error(), "nil context")
}

func TestNewProcessInvalidConfig(t *testing.T) {
	t.Parallel()

	_, err := NewProcess(context.Background(), ProcessConfig{TracesExporter: "weird", MetricsExporter: "none"}, nil)
	require.Error(t, err)
	require.Contains(t, err.Error(), "unsupported traces exporter")
}

func TestRuntimeNilSafeRecorders(t *testing.T) {
	t.Parallel()

	var runtime *Runtime
	runtime.RecordPublicHTTPRequest(context.Background(), nil, 0)
	runtime.RecordInternalHTTPRequest(context.Background(), nil, 0)
	runtime.RecordGameTransition(context.Background(), "draft", "enrollment_open", "command")
	runtime.RecordApplicationOutcome(context.Background(), "submitted")
	runtime.RecordInviteOutcome(context.Background(), "created")
	runtime.RecordMembershipChange(context.Background(), "activated")
	runtime.RecordStartFlowOutcome(context.Background(), "running")
	runtime.RecordNotificationPublish(context.Background(), "lobby.application.submitted", "ok")
	runtime.RecordEnrollmentAutomationCheck(context.Background(), "no_op")
	runtime.RecordRaceNameOutcome(context.Background(), "reserved")
	runtime.RecordPendingRegistrationExpiration(context.Background(), "tick")
	runtime.RecordUserLifecycleCascadeRelease(context.Background(), "permanent_blocked")
	runtime.RecordCapabilityEvaluation(context.Background(), "capable")
	require.Error(t, runtime.RegisterGauges(GaugeDependencies{}))
	require.NoError(t, runtime.Shutdown(context.Background()))
}

func TestRuntimeLobbyCountersRegistered(t *testing.T) {
	t.Parallel()

	reader := sdkmetric.NewManualReader()
	provider := sdkmetric.NewMeterProvider(sdkmetric.WithReader(reader))
	t.Cleanup(func() { _ = provider.Shutdown(context.Background()) })

	runtime, err := NewWithProviders(provider, nil)
	require.NoError(t, err)

	ctx := context.Background()
	runtime.RecordGameTransition(ctx, "draft", "enrollment_open", "command")
	runtime.RecordApplicationOutcome(ctx, "submitted")
	runtime.RecordInviteOutcome(ctx, "created")
	runtime.RecordMembershipChange(ctx, "activated")
	runtime.RecordStartFlowOutcome(ctx, "running")
	runtime.RecordNotificationPublish(ctx, "lobby.application.submitted", "ok")
	runtime.RecordEnrollmentAutomationCheck(ctx, "no_op")
	runtime.RecordRaceNameOutcome(ctx, "reserved")
	runtime.RecordPendingRegistrationExpiration(ctx, "tick")
	runtime.RecordUserLifecycleCascadeRelease(ctx, "permanent_blocked")
	runtime.RecordCapabilityEvaluation(ctx, "capable")

	var rm metricdata.ResourceMetrics
	require.NoError(t, reader.Collect(ctx, &rm))

	got := collectedMetricNames(rm)
	want := []string{
		"lobby.application.outcomes",
		"lobby.capability_evaluations",
		"lobby.enrollment_automation.checks",
		"lobby.game.transitions",
		"lobby.invite.outcomes",
		"lobby.membership.changes",
		"lobby.notification.publish_attempts",
		"lobby.pending_registration.expirations",
		"lobby.race_name.outcomes",
		"lobby.start_flow.outcomes",
		"lobby.user_lifecycle.cascade_releases",
	}
	for _, name := range want {
		assert.Contains(t, got, name)
	}
}

type stubActiveGames map[string]int

func (s stubActiveGames) CountByStatus(_ context.Context) (map[string]int, error) {
	return s, nil
}

type stubLag struct {
	age time.Duration
	ok  bool
	err error
}

func (s stubLag) OldestUnprocessedAge(_ context.Context, _ string, _ string) (time.Duration, bool, error) {
	return s.age, s.ok, s.err
}

type stubOffsets struct{}

func (stubOffsets) Load(_ context.Context, _ string) (string, bool, error) {
	return "0-0", true, nil
}

func TestRuntimeGaugesObserveCounts(t *testing.T) {
	t.Parallel()

	reader := sdkmetric.NewManualReader()
	provider := sdkmetric.NewMeterProvider(sdkmetric.WithReader(reader))
	t.Cleanup(func() { _ = provider.Shutdown(context.Background()) })

	runtime, err := NewWithProviders(provider, nil)
	require.NoError(t, err)

	require.NoError(t, runtime.RegisterGauges(GaugeDependencies{
		ActiveGames:    stubActiveGames{"running": 3, "paused": 1},
		StreamLag:      stubLag{age: 750 * time.Millisecond, ok: true},
		Offsets:        stubOffsets{},
		GMEvents:       StreamGaugeBinding{OffsetLabel: "gm_lobby_events", StreamName: "gm:lobby_events"},
		RuntimeResults: StreamGaugeBinding{OffsetLabel: "runtime_results", StreamName: "runtime:job_results"},
		UserLifecycle:  StreamGaugeBinding{OffsetLabel: "user_lifecycle", StreamName: "user:lifecycle_events"},
	}))

	ctx := context.Background()

	var rm metricdata.ResourceMetrics
	require.NoError(t, reader.Collect(ctx, &rm))

	gauges := collectedGauges(rm)
	require.Contains(t, gauges, "lobby.active_games")
	require.Contains(t, gauges, "lobby.gm_events.oldest_unprocessed_age_ms")
	require.Contains(t, gauges, "lobby.runtime_results.oldest_unprocessed_age_ms")
	require.Contains(t, gauges, "lobby.user_lifecycle.oldest_unprocessed_age_ms")
}

func TestRuntimeRegisterGaugesValidates(t *testing.T) {
	t.Parallel()

	runtime, err := NewWithProviders(sdkmetric.NewMeterProvider(), nil)
	require.NoError(t, err)

	cases := []struct {
		name string
		deps GaugeDependencies
	}{
		{"missing active games", GaugeDependencies{}},
		{"missing stream lag", GaugeDependencies{ActiveGames: stubActiveGames{}}},
		{"missing offsets", GaugeDependencies{ActiveGames: stubActiveGames{}, StreamLag: stubLag{}}},
		{"missing gm events stream", GaugeDependencies{ActiveGames: stubActiveGames{}, StreamLag: stubLag{}, Offsets: stubOffsets{}}},
	}
	for _, tc := range cases {
		err := runtime.RegisterGauges(tc.deps)
		require.Error(t, err, tc.name)
	}
}

func TestRuntimeGaugeProbeErrorsAreLogged(t *testing.T) {
	t.Parallel()

	reader := sdkmetric.NewManualReader()
	provider := sdkmetric.NewMeterProvider(sdkmetric.WithReader(reader))
	t.Cleanup(func() { _ = provider.Shutdown(context.Background()) })

	runtime, err := NewWithProviders(provider, nil)
	require.NoError(t, err)

	require.NoError(t, runtime.RegisterGauges(GaugeDependencies{
		ActiveGames:    stubActiveGames{"running": 0},
		StreamLag:      stubLag{err: errors.New("boom")},
		Offsets:        stubOffsets{},
		GMEvents:       StreamGaugeBinding{OffsetLabel: "gm_lobby_events", StreamName: "gm:lobby_events"},
		RuntimeResults: StreamGaugeBinding{OffsetLabel: "runtime_results", StreamName: "runtime:job_results"},
		UserLifecycle:  StreamGaugeBinding{OffsetLabel: "user_lifecycle", StreamName: "user:lifecycle_events"},
	}))

	var rm metricdata.ResourceMetrics
	require.NoError(t, reader.Collect(context.Background(), &rm))
}

func collectedMetricNames(rm metricdata.ResourceMetrics) []string {
	var names []string
	for _, scope := range rm.ScopeMetrics {
		for _, m := range scope.Metrics {
			names = append(names, m.Name)
		}
	}
	sort.Strings(names)
	return names
}

func collectedGauges(rm metricdata.ResourceMetrics) map[string]struct{} {
	gauges := map[string]struct{}{}
	for _, scope := range rm.ScopeMetrics {
		for _, m := range scope.Metrics {
			if _, ok := m.Data.(metricdata.Gauge[int64]); ok {
				gauges[m.Name] = struct{}{}
			}
		}
	}
	return gauges
}