galaxy-game/lobby/internal/app/runtime.go

package app

import (
	"context"
	"errors"
	"fmt"
	"log/slog"
	"time"

	"galaxy/lobby/internal/adapters/postgres/migrations"
	"galaxy/lobby/internal/adapters/redisstate"
	"galaxy/lobby/internal/api/internalhttp"
	"galaxy/lobby/internal/api/publichttp"
	"galaxy/lobby/internal/config"
	"galaxy/lobby/internal/domain/game"
	"galaxy/lobby/internal/ports"
	"galaxy/lobby/internal/telemetry"
	"galaxy/postgres"
)

// activeGamesProbe adapts ports.GameStore to telemetry.ActiveGamesProbe by
// converting domain status keys into the string-typed map the telemetry
// runtime consumes.
type activeGamesProbe struct {
	games ports.GameStore
}

func (probe activeGamesProbe) CountByStatus(ctx context.Context) (map[string]int, error) {
	counts, err := probe.games.CountByStatus(ctx)
	if err != nil {
		return nil, err
	}
	out := make(map[string]int, len(counts))
	for status, count := range counts {
		out[string(status)] = count
	}
	return out, nil
}

var _ telemetry.ActiveGamesProbe = activeGamesProbe{}

// Compile-time assertion that the active-games probe key set matches the
// frozen game.Status vocabulary; helps surface drift if a new status is
// introduced without updating telemetry attribute documentation.
var _ = game.AllStatuses

// Runtime owns the runnable Game Lobby Service process plus the cleanup
// functions that release runtime resources after shutdown.
type Runtime struct {
	cfg config.Config

	app *App

	wiring *wiring

	publicServer   *publichttp.Server
	internalServer *internalhttp.Server

	cleanupFns []func() error
}

// NewRuntime constructs the runnable Game Lobby Service process from cfg.
func NewRuntime(ctx context.Context, cfg config.Config, logger *slog.Logger) (*Runtime, error) {
	if ctx == nil {
		return nil, errors.New("new lobby runtime: nil context")
	}
	if err := cfg.Validate(); err != nil {
		return nil, fmt.Errorf("new lobby runtime: %w", err)
	}
	if logger == nil {
		logger = slog.Default()
	}

	runtime := &Runtime{
		cfg: cfg,
	}

	cleanupOnError := func(err error) (*Runtime, error) {
		if cleanupErr := runtime.Close(); cleanupErr != nil {
			return nil, fmt.Errorf("%w; cleanup: %w", err, cleanupErr)
		}

		return nil, err
	}

	telemetryRuntime, err := telemetry.NewProcess(ctx, telemetry.ProcessConfig{
		ServiceName:          cfg.Telemetry.ServiceName,
		TracesExporter:       cfg.Telemetry.TracesExporter,
		MetricsExporter:      cfg.Telemetry.MetricsExporter,
		TracesProtocol:       cfg.Telemetry.TracesProtocol,
		MetricsProtocol:      cfg.Telemetry.MetricsProtocol,
		StdoutTracesEnabled:  cfg.Telemetry.StdoutTracesEnabled,
		StdoutMetricsEnabled: cfg.Telemetry.StdoutMetricsEnabled,
	}, logger)
	if err != nil {
		return cleanupOnError(fmt.Errorf("new lobby runtime: telemetry: %w", err))
	}
	runtime.cleanupFns = append(runtime.cleanupFns, func() error {
		shutdownCtx, cancel := context.WithTimeout(context.Background(), cfg.ShutdownTimeout)
		defer cancel()
		return telemetryRuntime.Shutdown(shutdownCtx)
	})

	redisClient := newRedisClient(cfg.Redis)
	if err := instrumentRedisClient(redisClient, telemetryRuntime); err != nil {
		return cleanupOnError(fmt.Errorf("new lobby runtime: %w", err))
	}
	runtime.cleanupFns = append(runtime.cleanupFns, func() error {
		return redisClient.Close()
	})
	if err := pingRedis(ctx, cfg.Redis, redisClient); err != nil {
		return cleanupOnError(fmt.Errorf("new lobby runtime: %w", err))
	}

	pgPool, err := postgres.OpenPrimary(ctx, cfg.Postgres.Conn,
		postgres.WithTracerProvider(telemetryRuntime.TracerProvider()),
		postgres.WithMeterProvider(telemetryRuntime.MeterProvider()),
	)
	if err != nil {
		return cleanupOnError(fmt.Errorf("new lobby runtime: open postgres: %w", err))
	}
	runtime.cleanupFns = append(runtime.cleanupFns, pgPool.Close)
	unregisterPGStats, err := postgres.InstrumentDBStats(pgPool,
		postgres.WithMeterProvider(telemetryRuntime.MeterProvider()),
	)
	if err != nil {
		return cleanupOnError(fmt.Errorf("new lobby runtime: instrument postgres: %w", err))
	}
	runtime.cleanupFns = append(runtime.cleanupFns, func() error {
		return unregisterPGStats()
	})
	if err := postgres.Ping(ctx, pgPool, cfg.Postgres.Conn.OperationTimeout); err != nil {
		return cleanupOnError(fmt.Errorf("new lobby runtime: ping postgres: %w", err))
	}
	if err := postgres.RunMigrations(ctx, pgPool, migrations.FS(), "."); err != nil {
		return cleanupOnError(fmt.Errorf("new lobby runtime: run postgres migrations: %w", err))
	}

	wiring, err := newWiring(cfg, redisClient, pgPool, time.Now, logger, telemetryRuntime)
	if err != nil {
		return cleanupOnError(fmt.Errorf("new lobby runtime: wiring: %w", err))
	}
	runtime.wiring = wiring

	streamLagProbe, err := redisstate.NewStreamLagProbe(redisClient, time.Now)
	if err != nil {
		return cleanupOnError(fmt.Errorf("new lobby runtime: stream lag probe: %w", err))
	}
	if err := telemetryRuntime.RegisterGauges(telemetry.GaugeDependencies{
		ActiveGames: activeGamesProbe{games: wiring.gameStore},
		StreamLag:   streamLagProbe,
		Offsets:     wiring.streamOffsetStore,
		GMEvents: telemetry.StreamGaugeBinding{
			OffsetLabel: "gm_lobby_events",
			StreamName:  cfg.Redis.GMEventsStream,
		},
		RuntimeResults: telemetry.StreamGaugeBinding{
			OffsetLabel: "runtime_results",
			StreamName:  cfg.Redis.RuntimeJobResultsStream,
		},
		UserLifecycle: telemetry.StreamGaugeBinding{
			OffsetLabel: "user_lifecycle",
			StreamName:  cfg.Redis.UserLifecycleStream,
		},
		Logger: logger,
	}); err != nil {
		return cleanupOnError(fmt.Errorf("new lobby runtime: register gauges: %w", err))
	}

	publicServer, err := publichttp.NewServer(publichttp.Config{
		Addr:              cfg.PublicHTTP.Addr,
		ReadHeaderTimeout: cfg.PublicHTTP.ReadHeaderTimeout,
		ReadTimeout:       cfg.PublicHTTP.ReadTimeout,
		IdleTimeout:       cfg.PublicHTTP.IdleTimeout,
	}, publichttp.Dependencies{
		Logger:             logger,
		Telemetry:          telemetryRuntime,
		CreateGame:         wiring.createGame,
		UpdateGame:         wiring.updateGame,
		OpenEnrollment:     wiring.openEnrollment,
		CancelGame:         wiring.cancelGame,
		ManualReadyToStart: wiring.manualReadyToStart,
		StartGame:          wiring.startGame,
		RetryStartGame:     wiring.retryStartGame,
		PauseGame:          wiring.pauseGame,
		ResumeGame:         wiring.resumeGame,
		SubmitApplication:  wiring.submitApplication,
		ApproveApplication: wiring.approveApplication,
		RejectApplication:  wiring.rejectApplication,
		CreateInvite:       wiring.createInvite,
		RedeemInvite:       wiring.redeemInvite,
		DeclineInvite:      wiring.declineInvite,
		RevokeInvite:       wiring.revokeInvite,
		RemoveMember:       wiring.removeMember,
		BlockMember:        wiring.blockMember,
		RegisterRaceName:   wiring.registerRaceName,
		ListMyRaceNames:    wiring.listMyRaceNames,
		GetGame:            wiring.getGame,
		ListGames:          wiring.listGames,
		ListMemberships:    wiring.listMemberships,
		ListMyGames:        wiring.listMyGames,
		ListMyApplications: wiring.listMyApplications,
		ListMyInvites:      wiring.listMyInvites,
	})
	if err != nil {
		return cleanupOnError(fmt.Errorf("new lobby runtime: public HTTP server: %w", err))
	}
	runtime.publicServer = publicServer

	internalServer, err := internalhttp.NewServer(internalhttp.Config{
		Addr:              cfg.InternalHTTP.Addr,
		ReadHeaderTimeout: cfg.InternalHTTP.ReadHeaderTimeout,
		ReadTimeout:       cfg.InternalHTTP.ReadTimeout,
		IdleTimeout:       cfg.InternalHTTP.IdleTimeout,
	}, internalhttp.Dependencies{
		Logger:             logger,
		Telemetry:          telemetryRuntime,
		CreateGame:         wiring.createGame,
		UpdateGame:         wiring.updateGame,
		OpenEnrollment:     wiring.openEnrollment,
		CancelGame:         wiring.cancelGame,
		ManualReadyToStart: wiring.manualReadyToStart,
		StartGame:          wiring.startGame,
		RetryStartGame:     wiring.retryStartGame,
		PauseGame:          wiring.pauseGame,
		ResumeGame:         wiring.resumeGame,
		ApproveApplication: wiring.approveApplication,
		RejectApplication:  wiring.rejectApplication,
		RemoveMember:       wiring.removeMember,
		BlockMember:        wiring.blockMember,
		GetGame:            wiring.getGame,
		ListGames:          wiring.listGames,
		ListMemberships:    wiring.listMemberships,
	})
	if err != nil {
		return cleanupOnError(fmt.Errorf("new lobby runtime: internal HTTP server: %w", err))
	}
	runtime.internalServer = internalServer

	runtime.app = New(
		cfg,
		publicServer,
		internalServer,
		wiring.enrollmentAutomation,
		wiring.runtimeJobResultConsumer,
		wiring.gmEventsConsumer,
		wiring.pendingRegistration,
		wiring.userLifecycleConsumer,
	)

	return runtime, nil
}

// PublicServer returns the public HTTP server owned by runtime. It is
// primarily exposed for tests; production code should not depend on it.
func (runtime *Runtime) PublicServer() *publichttp.Server {
	if runtime == nil {
		return nil
	}

	return runtime.publicServer
}

// InternalServer returns the internal HTTP server owned by runtime. It is
// primarily exposed for tests; production code should not depend on it.
func (runtime *Runtime) InternalServer() *internalhttp.Server {
	if runtime == nil {
		return nil
	}

	return runtime.internalServer
}

// Run serves the public and internal HTTP listeners until ctx is canceled or
// one component fails.
func (runtime *Runtime) Run(ctx context.Context) error {
	if ctx == nil {
		return errors.New("run lobby runtime: nil context")
	}
	if runtime == nil {
		return errors.New("run lobby runtime: nil runtime")
	}
	if runtime.app == nil {
		return errors.New("run lobby runtime: nil app")
	}

	return runtime.app.Run(ctx)
}

// Close releases every runtime dependency in reverse construction order.
// Close is safe to call multiple times.
func (runtime *Runtime) Close() error {
	if runtime == nil {
		return nil
	}

	var joined error
	for index := len(runtime.cleanupFns) - 1; index >= 0; index-- {
		if err := runtime.cleanupFns[index](); err != nil {
			joined = errors.Join(joined, err)
		}
	}
	runtime.cleanupFns = nil

	return joined
}