Files
Edu/services/push-gateway/main.go

230 lines
7.9 KiB
Go

// Command push-gateway is the Edu platform's real-time WebSocket push gateway.
//
// It maintains per-user WebSocket connections, delivers directed and broadcast
// messages from msg (via HTTP /internal/* APIs and Kafka), and synchronizes
// online presence across instances via Redis Pub/Sub.
//
// Startup sequence:
// 1. config.Load (env vars, DevMode detection)
// 2. observability.InitLogger (slog JSON/text)
// 3. observability.InitTracer (shared-go/tracer → OTLP)
// 4. observability.NewMetrics (Prometheus)
// 5. hub.NewHub (in-memory connection registry)
// 6. redisclient.New + SetHub + RebuildPresenceOnStartup (ISSUE-058)
// 7. redisclient.SubscribeAll (cross-instance Pub/Sub fanout)
// 8. kafkaconsumer.New + goroutine Run (edu.notify.notification.sent, ARB-013)
// 9. ws.NewHandler (JWT RS256, X-Internal-Key, WebSocket upgrade)
// 10. health.NewReadyzer (/readyz soft-failure probe)
// 11. gin router + http.Server
//
// Graceful shutdown (SIGINT/SIGTERM):
// 1. Hub.CloseAll (send close frame 1001 to every live connection)
// 2. http.Server.Shutdown (stop accepting new HTTP/WebSocket requests)
// 3. Kafka consumer goroutine cancel + reader.Close
// 4. Redis Pub/Sub cancel + client.Close
// 5. Tracer shutdown (flush pending spans)
package main
import (
"context"
"errors"
"net/http"
"os"
"os/signal"
"syscall"
"time"
"github.com/edu-cloud/push-gateway/internal/config"
"github.com/edu-cloud/push-gateway/internal/health"
"github.com/edu-cloud/push-gateway/internal/hub"
"github.com/edu-cloud/push-gateway/internal/kafkaconsumer"
"github.com/edu-cloud/push-gateway/internal/observability"
"github.com/edu-cloud/push-gateway/internal/redisclient"
"github.com/edu-cloud/push-gateway/internal/ws"
"github.com/gin-gonic/gin"
"github.com/prometheus/client_golang/prometheus/promhttp"
"go.opentelemetry.io/contrib/instrumentation/github.com/gin-gonic/gin/otelgin"
)
const serviceName = "push-gateway"
// shutdownTimeout bounds the graceful shutdown of the HTTP server. After this
// window in-flight requests are forcibly closed. WebSocket long-poll
// connections are drained by Hub.CloseAll before this timer starts.
const shutdownTimeout = 10 * time.Second
func main() {
cfg := config.Load()
// 1. Structured logger (slog).
logger := observability.InitLogger(cfg.DevMode)
logger.Info("push-gateway starting",
"dev_mode", cfg.DevMode,
"instance_id", cfg.InstanceID,
"port", cfg.Port,
"max_conns_per_user", cfg.MaxConnsPerUser,
"jwks_url", cfg.JWKSURL,
"redis_url", cfg.RedisURL,
"kafka_brokers", cfg.KafkaBrokers,
"kafka_topic", cfg.KafkaNotificationTopic,
)
// 2. OpenTelemetry tracer (best-effort).
tracerShutdown := observability.InitTracer(serviceName, cfg.OTLPEndpoint)
defer tracerShutdown()
// 3. Prometheus metrics.
metrics := observability.NewMetrics()
// 4. Hub: in-memory connection registry.
h := hub.NewHub(cfg.MaxConnsPerUser)
// 5. Redis client (optional in DevMode). When present, wire presence
// hooks to the Hub and rebuild the online SET (ISSUE-058).
var redisClient *redisclient.Client
if cfg.RedisURL != "" {
rc, err := redisclient.New(cfg.RedisURL, cfg.InstanceID, metrics)
if err != nil {
logger.Error("redis connect failed; running without cross-instance fanout",
"err", err, "url", cfg.RedisURL)
} else {
redisClient = rc
redisClient.SetHub(h)
rebuildCtx, rebuildCancel := context.WithTimeout(context.Background(), 5*time.Second)
if err := redisClient.RebuildPresenceOnStartup(rebuildCtx); err != nil {
logger.Warn("redis presence rebuild failed (continuing)", "err", err)
}
rebuildCancel()
logger.Info("redis presence synced", "url", cfg.RedisURL)
}
} else {
logger.Warn("REDIS_URL empty; cross-instance fanout disabled")
}
// 6. Redis Pub/Sub subscriber (cross-instance message fanout).
var pubsubCancel func() error
if redisClient != nil {
subCtx, subCancel := context.WithCancel(context.Background())
defer subCancel()
var err error
pubsubCancel, err = redisClient.SubscribeAll(subCtx)
if err != nil {
logger.Error("redis SubscribeAll failed; cross-instance fanout disabled",
"err", err)
} else {
logger.Info("redis pubsub subscriber active")
}
}
// 7. Kafka consumer (edu.notify.notification.sent, ARB-013). Started only
// when at least one broker is configured; otherwise the HTTP /internal/push
// API is the only delivery channel.
var kafkaConsumer *kafkaconsumer.Consumer
var kafkaCancel context.CancelFunc
if len(cfg.KafkaBrokers) > 0 && cfg.KafkaNotificationTopic != "" {
kafkaConsumer = kafkaconsumer.New(kafkaconsumer.Config{
Brokers: cfg.KafkaBrokers,
Topic: cfg.KafkaNotificationTopic,
GroupID: cfg.KafkaConsumerGroup,
}, h, redisClient, metrics)
kafkaCtx, cancel := context.WithCancel(context.Background())
kafkaCancel = cancel
go func() {
logger.Info("kafka consumer starting",
"topic", cfg.KafkaNotificationTopic, "group", cfg.KafkaConsumerGroup)
if err := kafkaConsumer.Run(kafkaCtx); err != nil &&
!errors.Is(err, context.Canceled) {
logger.Error("kafka consumer exited with error", "err", err)
}
}()
} else {
logger.Warn("KAFKA_BROKERS empty; notification consumption disabled")
}
// 8. WebSocket + internal HTTP handlers.
wsHandler := ws.NewHandler(h, cfg, redisClient, metrics)
// 9. /readyz probe (soft failure per ARB-015 §17.4).
readyzer := health.NewReadyzer(h, redisClient, kafkaConsumer, serviceName, cfg.InstanceID)
// 10. Gin router.
gin.SetMode(gin.ReleaseMode)
r := gin.New()
r.Use(gin.Recovery())
r.Use(otelgin.Middleware(serviceName))
// Liveness (no auth, no dependency checks).
r.GET("/healthz", health.Healthz(serviceName))
// Readiness (soft failure on Redis/Kafka).
r.GET("/readyz", readyzer.Handler)
// Prometheus metrics.
r.GET("/metrics", gin.WrapH(promhttp.Handler()))
// WebSocket upgrade (JWT auth via query ?token= or Authorization header).
r.GET("/ws", wsHandler.HandleWebSocket)
// Internal HTTP APIs consumed by msg (X-Internal-Key auth, ARB-013).
internal := r.Group("/internal")
internal.POST("/push", wsHandler.PushHandler)
internal.POST("/broadcast", wsHandler.BroadcastHandler)
internal.GET("/online/:userID", wsHandler.OnlineHandler)
srv := &http.Server{
Addr: ":" + cfg.Port,
Handler: r,
ReadTimeout: 10 * time.Second,
WriteTimeout: 0, // WebSocket connections are long-lived; no write timeout.
}
// 11. Start HTTP server.
go func() {
logger.Info("http server listening", "addr", srv.Addr)
if err := srv.ListenAndServe(); err != nil && !errors.Is(err, http.ErrServerClosed) {
logger.Error("http server fatal error", "err", err)
os.Exit(1)
}
}()
// 12. Wait for termination signal.
quit := make(chan os.Signal, 1)
signal.Notify(quit, syscall.SIGINT, syscall.SIGTERM)
sig := <-quit
logger.Info("shutdown signal received", "signal", sig.String())
// 13. Graceful shutdown sequence (see package doc).
shutdownCtx, shutdownCancel := context.WithTimeout(context.Background(), shutdownTimeout)
defer shutdownCancel()
// 13a. Hub.CloseAll sends close frame 1001 to every live connection.
h.CloseAll()
logger.Info("hub closeAll complete; draining connections",
"active_connections", h.ActiveConnections())
// 13b. Stop accepting new HTTP requests; drain in-flight.
if err := srv.Shutdown(shutdownCtx); err != nil {
logger.Error("http server shutdown error", "err", err)
}
// 13c. Stop Kafka consumer (cancel context + close reader).
if kafkaCancel != nil {
kafkaCancel()
}
if kafkaConsumer != nil {
if err := kafkaConsumer.Close(); err != nil {
logger.Warn("kafka consumer close error", "err", err)
}
}
// 13d. Stop Redis Pub/Sub subscriber + close client.
if pubsubCancel != nil {
if err := pubsubCancel(); err != nil {
logger.Warn("redis pubsub cancel error", "err", err)
}
}
if redisClient != nil {
if err := redisClient.Close(); err != nil {
logger.Warn("redis close error", "err", err)
}
}
logger.Info("push-gateway shutdown complete")
}