feat: notification service

2026-04-22 08:49:45 +02:00
parent 5b7593e6f6
commit 32dc29359a
135 changed files with 21828 additions and 130 deletions
@@ -0,0 +1,252 @@
+// Package internalhttp provides the private probe HTTP listener used by the
+// runnable Notification Service process.
+package internalhttp
+
+import (
+	"context"
+	"encoding/json"
+	"errors"
+	"fmt"
+	"log/slog"
+	"net"
+	"net/http"
+	"strconv"
+	"sync"
+	"time"
+
+	"galaxy/notification/internal/telemetry"
+
+	"go.opentelemetry.io/contrib/instrumentation/net/http/otelhttp"
+	"go.opentelemetry.io/otel/attribute"
+)
+
+const jsonContentType = "application/json; charset=utf-8"
+
+const (
+	// HealthzPath is the private liveness probe route.
+	HealthzPath = "/healthz"
+
+	// ReadyzPath is the private readiness probe route.
+	ReadyzPath = "/readyz"
+)
+
+// Config describes the private internal HTTP listener owned by Notification
+// Service.
+type Config struct {
+	// Addr is the TCP listen address used by the private probe 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
+
+	// IdleTimeout bounds how long the listener keeps an idle keep-alive
+	// connection open.
+	IdleTimeout time.Duration
+}
+
+// Validate reports whether cfg contains a usable private 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.IdleTimeout <= 0:
+		return errors.New("internal HTTP idle timeout must be positive")
+	default:
+		return nil
+	}
+}
+
+// Dependencies describes the collaborators used by the private probe
+// transport layer.
+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
+}
+
+// Server owns the private probe HTTP listener exposed by Notification
+// Service.
+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 private probe 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(logger, deps.Telemetry),
+		logger:  logger.With("component", "internal_http"),
+		metrics: deps.Telemetry,
+	}, nil
+}
+
+// Run binds the configured listener and serves the private probe 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,
+		IdleTimeout:       server.cfg.IdleTimeout,
+	}
+
+	server.stateMu.Lock()
+	server.server = httpServer
+	server.listener = listener
+	server.stateMu.Unlock()
+
+	server.logger.Info("notification internal HTTP server started", "addr", listener.Addr().String())
+	server.metrics.RecordInternalHTTPEvent(context.Background(), "started")
+
+	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("notification internal HTTP server stopped")
+		server.metrics.RecordInternalHTTPEvent(context.Background(), "stopped")
+		return nil
+	default:
+		return fmt.Errorf("run internal HTTP server: serve on %q: %w", server.cfg.Addr, err)
+	}
+}
+
+// Shutdown gracefully stops the private probe 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(logger *slog.Logger, metrics *telemetry.Runtime) http.Handler {
+	mux := http.NewServeMux()
+	mux.HandleFunc("GET "+HealthzPath, handleHealthz)
+	mux.HandleFunc("GET "+ReadyzPath, handleReadyz)
+
+	return otelhttp.NewHandler(withObservability(mux, metrics), "notification.internal_http")
+}
+
+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"
+		}
+
+		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(writer http.ResponseWriter, _ *http.Request) {
+	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})
+}
+
+type statusResponse struct {
+	Status string `json:"status"`
+}
+
+type statusRecorder struct {
+	http.ResponseWriter
+	statusCode int
+}
+
+func (recorder *statusRecorder) WriteHeader(statusCode int) {
+	recorder.statusCode = statusCode
+	recorder.ResponseWriter.WriteHeader(statusCode)
+}
@@ -0,0 +1,272 @@
+package internalhttp
+
+import (
+	"context"
+	"encoding/json"
+	"io"
+	"net"
+	"net/http"
+	"strings"
+	"testing"
+	"time"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+func TestNewServerRejectsInvalidConfiguration(t *testing.T) {
+	t.Parallel()
+
+	cfg := Config{
+		ReadHeaderTimeout: time.Second,
+		ReadTimeout:       time.Second,
+		IdleTimeout:       time.Second,
+	}
+
+	_, err := NewServer(cfg, Dependencies{})
+	require.Error(t, err)
+	assert.Contains(t, err.Error(), "addr")
+}
+
+func TestServerRunAndShutdown(t *testing.T) {
+	t.Parallel()
+
+	cfg := testConfig(t)
+	server, err := NewServer(cfg, Dependencies{})
+	require.NoError(t, err)
+
+	runErr := make(chan error, 1)
+	go func() {
+		runErr <- server.Run(context.Background())
+	}()
+
+	client := newTestHTTPClient(t)
+	waitForHealthzReady(t, client, cfg.Addr)
+
+	shutdownCtx, cancel := context.WithTimeout(context.Background(), time.Second)
+	defer cancel()
+	require.NoError(t, server.Shutdown(shutdownCtx))
+	waitForServerRunResult(t, runErr)
+}
+
+func TestProbeRoutesReturnStableJSON(t *testing.T) {
+	t.Parallel()
+
+	cfg := testConfig(t)
+	server, err := NewServer(cfg, Dependencies{})
+	require.NoError(t, err)
+
+	runErr := make(chan error, 1)
+	go func() {
+		runErr <- server.Run(context.Background())
+	}()
+
+	client := newTestHTTPClient(t)
+	waitForHealthzReady(t, client, cfg.Addr)
+
+	tests := []struct {
+		path   string
+		status string
+	}{
+		{path: HealthzPath, status: "ok"},
+		{path: ReadyzPath, status: "ready"},
+	}
+
+	for _, tt := range tests {
+		tt := tt
+
+		t.Run(tt.path, func(t *testing.T) {
+			request, err := http.NewRequest(http.MethodGet, "http://"+cfg.Addr+tt.path, nil)
+			require.NoError(t, err)
+
+			response, err := client.Do(request)
+			require.NoError(t, err)
+			defer response.Body.Close()
+
+			require.Equal(t, http.StatusOK, response.StatusCode)
+			require.Equal(t, "application/json; charset=utf-8", response.Header.Get("Content-Type"))
+
+			var payload statusResponse
+			require.NoError(t, json.NewDecoder(response.Body).Decode(&payload))
+			require.Equal(t, tt.status, payload.Status)
+		})
+	}
+
+	shutdownCtx, cancel := context.WithTimeout(context.Background(), time.Second)
+	defer cancel()
+	require.NoError(t, server.Shutdown(shutdownCtx))
+	waitForServerRunResult(t, runErr)
+}
+
+func TestServerDoesNotExposeMetricsOrUnknownRoutes(t *testing.T) {
+	t.Parallel()
+
+	cfg := testConfig(t)
+	server, err := NewServer(cfg, Dependencies{})
+	require.NoError(t, err)
+
+	runErr := make(chan error, 1)
+	go func() {
+		runErr <- server.Run(context.Background())
+	}()
+
+	client := newTestHTTPClient(t)
+	waitForHealthzReady(t, client, cfg.Addr)
+
+	for _, path := range []string{"/metrics", "/unknown"} {
+		request, err := http.NewRequest(http.MethodGet, "http://"+cfg.Addr+path, nil)
+		require.NoError(t, err)
+
+		response, err := client.Do(request)
+		require.NoError(t, err)
+		_, _ = io.ReadAll(response.Body)
+		response.Body.Close()
+
+		assert.Equalf(t, http.StatusNotFound, response.StatusCode, "path %s", path)
+	}
+
+	shutdownCtx, cancel := context.WithTimeout(context.Background(), time.Second)
+	defer cancel()
+	require.NoError(t, server.Shutdown(shutdownCtx))
+	waitForServerRunResult(t, runErr)
+}
+
+func TestServerPreservesStandardHEADBehavior(t *testing.T) {
+	t.Parallel()
+
+	cfg := testConfig(t)
+	server, err := NewServer(cfg, Dependencies{})
+	require.NoError(t, err)
+
+	runErr := make(chan error, 1)
+	go func() {
+		runErr <- server.Run(context.Background())
+	}()
+
+	client := newTestHTTPClient(t)
+	waitForHealthzReady(t, client, cfg.Addr)
+
+	request, err := http.NewRequest(http.MethodHead, "http://"+cfg.Addr+HealthzPath, nil)
+	require.NoError(t, err)
+
+	response, err := client.Do(request)
+	require.NoError(t, err)
+	defer response.Body.Close()
+
+	body, err := io.ReadAll(response.Body)
+	require.NoError(t, err)
+	require.Equal(t, http.StatusOK, response.StatusCode)
+	require.Empty(t, body)
+
+	shutdownCtx, cancel := context.WithTimeout(context.Background(), time.Second)
+	defer cancel()
+	require.NoError(t, server.Shutdown(shutdownCtx))
+	waitForServerRunResult(t, runErr)
+}
+
+func TestServerUsesStandardMethodNotAllowedBehavior(t *testing.T) {
+	t.Parallel()
+
+	cfg := testConfig(t)
+	server, err := NewServer(cfg, Dependencies{})
+	require.NoError(t, err)
+
+	runErr := make(chan error, 1)
+	go func() {
+		runErr <- server.Run(context.Background())
+	}()
+
+	client := newTestHTTPClient(t)
+	waitForHealthzReady(t, client, cfg.Addr)
+
+	request, err := http.NewRequest(http.MethodPost, "http://"+cfg.Addr+HealthzPath, nil)
+	require.NoError(t, err)
+
+	response, err := client.Do(request)
+	require.NoError(t, err)
+	defer response.Body.Close()
+	_, _ = io.ReadAll(response.Body)
+
+	require.Equal(t, http.StatusMethodNotAllowed, response.StatusCode)
+	require.Contains(t, response.Header.Get("Allow"), http.MethodGet)
+	require.Contains(t, response.Header.Get("Allow"), http.MethodHead)
+
+	shutdownCtx, cancel := context.WithTimeout(context.Background(), time.Second)
+	defer cancel()
+	require.NoError(t, server.Shutdown(shutdownCtx))
+	waitForServerRunResult(t, runErr)
+}
+
+func testConfig(t *testing.T) Config {
+	t.Helper()
+
+	return Config{
+		Addr:              mustFreeAddr(t),
+		ReadHeaderTimeout: time.Second,
+		ReadTimeout:       2 * time.Second,
+		IdleTimeout:       time.Minute,
+	}
+}
+
+func newTestHTTPClient(t *testing.T) *http.Client {
+	t.Helper()
+
+	transport := &http.Transport{DisableKeepAlives: true}
+	t.Cleanup(transport.CloseIdleConnections)
+
+	return &http.Client{
+		Timeout:   250 * time.Millisecond,
+		Transport: transport,
+	}
+}
+
+func waitForHealthzReady(t *testing.T, client *http.Client, addr string) {
+	t.Helper()
+
+	require.Eventually(t, func() bool {
+		request, err := http.NewRequest(http.MethodGet, "http://"+addr+HealthzPath, nil)
+		if err != nil {
+			return false
+		}
+
+		response, err := client.Do(request)
+		if err != nil {
+			return false
+		}
+		defer response.Body.Close()
+
+		payload, err := io.ReadAll(response.Body)
+		if err != nil {
+			return false
+		}
+
+		return response.StatusCode == http.StatusOK && strings.Contains(string(payload), `"status":"ok"`)
+	}, 5*time.Second, 25*time.Millisecond, "internal HTTP server did not become reachable")
+}
+
+func waitForServerRunResult(t *testing.T, runErr <-chan error) {
+	t.Helper()
+
+	var err error
+	require.Eventually(t, func() bool {
+		select {
+		case err = <-runErr:
+			return true
+		default:
+			return false
+		}
+	}, 5*time.Second, 10*time.Millisecond, "internal HTTP server did not stop")
+	require.NoError(t, err)
+}
+
+func mustFreeAddr(t *testing.T) string {
+	t.Helper()
+
+	listener, err := net.Listen("tcp", "127.0.0.1:0")
+	require.NoError(t, err)
+	defer func() {
+		assert.NoError(t, listener.Close())
+	}()
+
+	return listener.Addr().String()
+}