galaxy-game/rtmanager/internal/api/internalhttp/server.go

// Package internalhttp provides the trusted internal HTTP listener used
// by the runnable Runtime Manager process. It exposes `/healthz` and
// `/readyz` plus the GM/Admin REST surface backed by the lifecycle
// services in `internal/service/`.
package internalhttp

import (
	"context"
	"encoding/json"
	"errors"
	"fmt"
	"log/slog"
	"net"
	"net/http"
	"strconv"
	"sync"
	"time"

	"galaxy/rtmanager/internal/api/internalhttp/handlers"
	"galaxy/rtmanager/internal/ports"
	"galaxy/rtmanager/internal/telemetry"

	"go.opentelemetry.io/contrib/instrumentation/net/http/otelhttp"
	"go.opentelemetry.io/otel/attribute"
)

const jsonContentType = "application/json; charset=utf-8"

// errorCodeServiceUnavailable mirrors the stable error code declared in
// `rtmanager/api/internal-openapi.yaml` `§Error Model`.
const errorCodeServiceUnavailable = "service_unavailable"

// HealthzPath and ReadyzPath are the internal probe routes documented in
// `rtmanager/api/internal-openapi.yaml`.
const (
	HealthzPath = "/healthz"
	ReadyzPath  = "/readyz"
)

// ReadinessProbe reports whether the dependencies the listener guards
// (PostgreSQL, Redis, Docker) are reachable. A non-nil error is reported
// to the caller as `503 service_unavailable` with the wrapped message.
type ReadinessProbe interface {
	Check(ctx context.Context) error
}

// Config describes the trusted internal HTTP listener owned by Runtime
// Manager.
type Config struct {
	// Addr is the TCP listen address used by the internal HTTP server.
	Addr string

	// ReadHeaderTimeout bounds how long the listener may spend reading
	// request headers before the server rejects the connection.
	ReadHeaderTimeout time.Duration

	// ReadTimeout bounds how long the listener may spend reading one
	// request.
	ReadTimeout time.Duration

	// WriteTimeout bounds how long the listener may spend writing one
	// response.
	WriteTimeout time.Duration

	// IdleTimeout bounds how long the listener keeps an idle keep-alive
	// connection open.
	IdleTimeout time.Duration
}

// Validate reports whether cfg contains a usable internal HTTP listener
// configuration.
func (cfg Config) Validate() error {
	switch {
	case cfg.Addr == "":
		return errors.New("internal HTTP addr must not be empty")
	case cfg.ReadHeaderTimeout <= 0:
		return errors.New("internal HTTP read header timeout must be positive")
	case cfg.ReadTimeout <= 0:
		return errors.New("internal HTTP read timeout must be positive")
	case cfg.WriteTimeout <= 0:
		return errors.New("internal HTTP write timeout must be positive")
	case cfg.IdleTimeout <= 0:
		return errors.New("internal HTTP idle timeout must be positive")
	default:
		return nil
	}
}

// Dependencies describes the collaborators used by the internal HTTP
// transport layer. The listener still works when the lifecycle service
// fields are zero — handlers register but each returns
// `500 internal_error` until the runtime wires the real services.
type Dependencies struct {
	// Logger writes structured listener lifecycle logs. When nil,
	// slog.Default is used.
	Logger *slog.Logger

	// Telemetry records low-cardinality probe metrics and lifecycle
	// events.
	Telemetry *telemetry.Runtime

	// Readiness reports whether PG / Redis / Docker are reachable. A
	// nil readiness probe makes `/readyz` always answer `200`; the
	// runtime always supplies a real probe in production wiring.
	Readiness ReadinessProbe

	// RuntimeRecords backs the read-only list/get handlers. When nil
	// those routes return `500 internal_error`.
	RuntimeRecords ports.RuntimeRecordStore

	// StartRuntime, StopRuntime, RestartRuntime, PatchRuntime, and
	// CleanupContainer back the lifecycle handlers. Each accepts a
	// narrow interface so tests can pass `mockgen`-generated mocks;
	// production wiring passes the concrete `*<lifecycle>.Service`
	// pointer.
	StartRuntime     handlers.StartService
	StopRuntime      handlers.StopService
	RestartRuntime   handlers.RestartService
	PatchRuntime     handlers.PatchService
	CleanupContainer handlers.CleanupService
}

// Server owns the trusted internal HTTP listener exposed by Runtime
// Manager.
type Server struct {
	cfg Config

	handler http.Handler
	logger  *slog.Logger
	metrics *telemetry.Runtime

	stateMu  sync.RWMutex
	server   *http.Server
	listener net.Listener
}

// NewServer constructs one trusted internal HTTP server for cfg and deps.
func NewServer(cfg Config, deps Dependencies) (*Server, error) {
	if err := cfg.Validate(); err != nil {
		return nil, fmt.Errorf("new internal HTTP server: %w", err)
	}

	logger := deps.Logger
	if logger == nil {
		logger = slog.Default()
	}

	return &Server{
		cfg:     cfg,
		handler: newHandler(deps, logger),
		logger:  logger.With("component", "internal_http"),
		metrics: deps.Telemetry,
	}, nil
}

// Addr returns the currently bound listener address after Run is called.
// It returns an empty string if the server has not yet bound a listener.
func (server *Server) Addr() string {
	server.stateMu.RLock()
	defer server.stateMu.RUnlock()
	if server.listener == nil {
		return ""
	}

	return server.listener.Addr().String()
}

// Run binds the configured listener and serves the internal HTTP surface
// until Shutdown closes the server.
func (server *Server) Run(ctx context.Context) error {
	if ctx == nil {
		return errors.New("run internal HTTP server: nil context")
	}
	if err := ctx.Err(); err != nil {
		return err
	}

	listener, err := net.Listen("tcp", server.cfg.Addr)
	if err != nil {
		return fmt.Errorf("run internal HTTP server: listen on %q: %w", server.cfg.Addr, err)
	}

	httpServer := &http.Server{
		Handler:           server.handler,
		ReadHeaderTimeout: server.cfg.ReadHeaderTimeout,
		ReadTimeout:       server.cfg.ReadTimeout,
		WriteTimeout:      server.cfg.WriteTimeout,
		IdleTimeout:       server.cfg.IdleTimeout,
	}

	server.stateMu.Lock()
	server.server = httpServer
	server.listener = listener
	server.stateMu.Unlock()

	server.logger.Info("rtmanager internal HTTP server started", "addr", listener.Addr().String())

	defer func() {
		server.stateMu.Lock()
		server.server = nil
		server.listener = nil
		server.stateMu.Unlock()
	}()

	err = httpServer.Serve(listener)
	switch {
	case err == nil:
		return nil
	case errors.Is(err, http.ErrServerClosed):
		server.logger.Info("rtmanager internal HTTP server stopped")
		return nil
	default:
		return fmt.Errorf("run internal HTTP server: serve on %q: %w", server.cfg.Addr, err)
	}
}

// Shutdown gracefully stops the internal HTTP server within ctx.
func (server *Server) Shutdown(ctx context.Context) error {
	if ctx == nil {
		return errors.New("shutdown internal HTTP server: nil context")
	}

	server.stateMu.RLock()
	httpServer := server.server
	server.stateMu.RUnlock()

	if httpServer == nil {
		return nil
	}

	if err := httpServer.Shutdown(ctx); err != nil && !errors.Is(err, http.ErrServerClosed) {
		return fmt.Errorf("shutdown internal HTTP server: %w", err)
	}

	return nil
}

func newHandler(deps Dependencies, logger *slog.Logger) http.Handler {
	mux := http.NewServeMux()
	mux.HandleFunc("GET "+HealthzPath, handleHealthz)
	mux.HandleFunc("GET "+ReadyzPath, handleReadyz(deps.Readiness, logger))

	handlers.Register(mux, handlers.Dependencies{
		Logger:           logger,
		RuntimeRecords:   deps.RuntimeRecords,
		StartRuntime:     deps.StartRuntime,
		StopRuntime:      deps.StopRuntime,
		RestartRuntime:   deps.RestartRuntime,
		PatchRuntime:     deps.PatchRuntime,
		CleanupContainer: deps.CleanupContainer,
	})

	metrics := deps.Telemetry
	options := []otelhttp.Option{}
	if metrics != nil {
		options = append(options,
			otelhttp.WithTracerProvider(metrics.TracerProvider()),
			otelhttp.WithMeterProvider(metrics.MeterProvider()),
		)
	}

	return otelhttp.NewHandler(withObservability(mux, metrics), "rtmanager.internal_http", options...)
}

func withObservability(next http.Handler, metrics *telemetry.Runtime) http.Handler {
	return http.HandlerFunc(func(writer http.ResponseWriter, request *http.Request) {
		startedAt := time.Now()
		recorder := &statusRecorder{
			ResponseWriter: writer,
			statusCode:     http.StatusOK,
		}

		next.ServeHTTP(recorder, request)

		route := request.Pattern
		switch recorder.statusCode {
		case http.StatusMethodNotAllowed:
			route = "method_not_allowed"
		case http.StatusNotFound:
			route = "not_found"
		case 0:
			route = "unmatched"
		}
		if route == "" {
			route = "unmatched"
		}

		if metrics != nil {
			metrics.RecordInternalHTTPRequest(
				request.Context(),
				[]attribute.KeyValue{
					attribute.String("route", route),
					attribute.String("method", request.Method),
					attribute.String("status_code", strconv.Itoa(recorder.statusCode)),
				},
				time.Since(startedAt),
			)
		}
	})
}

func handleHealthz(writer http.ResponseWriter, _ *http.Request) {
	writeStatusResponse(writer, http.StatusOK, "ok")
}

func handleReadyz(probe ReadinessProbe, logger *slog.Logger) http.HandlerFunc {
	return func(writer http.ResponseWriter, request *http.Request) {
		if probe == nil {
			writeStatusResponse(writer, http.StatusOK, "ready")
			return
		}

		if err := probe.Check(request.Context()); err != nil {
			logger.WarnContext(request.Context(), "rtmanager readiness probe failed",
				"err", err.Error(),
			)
			writeServiceUnavailable(writer, err.Error())
			return
		}

		writeStatusResponse(writer, http.StatusOK, "ready")
	}
}

func writeStatusResponse(writer http.ResponseWriter, statusCode int, status string) {
	writer.Header().Set("Content-Type", jsonContentType)
	writer.WriteHeader(statusCode)
	_ = json.NewEncoder(writer).Encode(statusResponse{Status: status})
}

func writeServiceUnavailable(writer http.ResponseWriter, message string) {
	writer.Header().Set("Content-Type", jsonContentType)
	writer.WriteHeader(http.StatusServiceUnavailable)
	_ = json.NewEncoder(writer).Encode(errorResponse{
		Error: errorBody{
			Code:    errorCodeServiceUnavailable,
			Message: message,
		},
	})
}

type statusResponse struct {
	Status string `json:"status"`
}

type errorBody struct {
	Code    string `json:"code"`
	Message string `json:"message"`
}

type errorResponse struct {
	Error errorBody `json:"error"`
}

type statusRecorder struct {
	http.ResponseWriter
	statusCode int
}

func (recorder *statusRecorder) WriteHeader(statusCode int) {
	recorder.statusCode = statusCode
	recorder.ResponseWriter.WriteHeader(statusCode)
}