descrybe/apps/api/cmd/api/main.go

package main

import (
	"context"
	"log"
	"log/slog"
	"net"
	"net/http"
	"os"
	"os/signal"
	"syscall"
	"time"

	"github.com/descrybe/descrybe-v2/apps/api/internal/auth"
	"github.com/descrybe/descrybe-v2/apps/api/internal/config"
	"github.com/descrybe/descrybe-v2/apps/api/internal/db"
	"github.com/descrybe/descrybe-v2/apps/api/internal/httpapi"
	"github.com/descrybe/descrybe-v2/apps/api/internal/logredact"
)

func main() {
	log.SetOutput(logredact.Writer(os.Stderr))
	slog.SetDefault(slog.New(logredact.NewJSONHandler(os.Stdout, &slog.HandlerOptions{Level: slog.LevelInfo})))

	cfg, err := config.Load()
	if err != nil {
		log.Fatalf("config: %v", err)
	}
	if cfg.InsecureLocalProductionActive() {
		slog.Warn("ALLOW_INSECURE_LOCAL_PRODUCTION=1 with loopback WEB_ORIGIN — not for public deploy",
			"web_origin", cfg.WebOrigin,
			"http_addr", cfg.HTTPAddr,
		)
	}
	if cfg.ShouldWarnRateLimits() {
		slog.Warn(cfg.RateLimitWarningMessage(),
			"rate_limit_replicas", cfg.RateLimitReplicas,
			"rate_limit_multi_replica", cfg.RateLimitMultiReplica,
			"rate_limit_backend", cfg.RateLimitBackend,
			"rate_limit_backend_requested", cfg.RateLimitBackendRequested,
		)
	}

	ctx := context.Background()
	pool, err := db.NewPool(ctx, cfg.DatabaseURL, db.PoolOptions{
		MaxConns:              int32(cfg.DBMaxConns),
		MinConns:              int32(cfg.DBMinConns),
		MaxConnLifetime:       cfg.DBMaxConnLifetime,
		MaxConnLifetimeJitter: cfg.DBMaxConnLifetimeJitter,
		MaxConnIdleTime:       cfg.DBMaxConnIdleTime,
		HealthCheckPeriod:     cfg.DBHealthCheckPeriod,
		StatementTimeout:      cfg.DBStatementTimeout,
	})
	if err != nil {
		log.Fatalf("db: %v", err)
	}
	defer pool.Close()

	sessions := auth.NewSessionManager(pool, cfg.SessionCookieName, cfg.SessionCookieDomain, cfg.CookieSecure(), cfg.SessionIdleHours)
	srv := httpapi.NewServer(cfg, pool, sessions)

	runCtx, runCancel := context.WithCancel(context.Background())
	defer runCancel()
	// Lightweight AI fallback poller so FAQ-miss tickets drain without a separate worker.
	if srv.Support != nil {
		go srv.Support.RunAutoJobsLoop(runCtx, 2*time.Second, 3)
	}

	httpServer := newHTTPServer(cfg.HTTPAddr, srv.Router())

	ln, err := net.Listen("tcp", cfg.HTTPAddr)
	if err != nil {
		log.Fatalf("listen: %v", err)
	}
	slog.Info("api_listening", "addr", cfg.HTTPAddr, "maintenance", cfg.MaintenanceMode, "read_only", cfg.ReadOnlyMode)

	go func() {
		if err := httpServer.Serve(ln); err != nil && err != http.ErrServerClosed {
			log.Fatalf("serve: %v", err)
		}
	}()

	stop := make(chan os.Signal, 1)
	signal.Notify(stop, syscall.SIGINT, syscall.SIGTERM)
	<-stop
	runCancel()

	shutdownCtx, cancel := context.WithTimeout(context.Background(), 10*time.Second)
	defer cancel()
	_ = httpServer.Shutdown(shutdownCtx)
}

// newHTTPServer configures net/http timeouts for the API listener.
//
// Tradeoff — WriteTimeout vs long sync/export:
// WriteTimeout bounds the whole ServeHTTP + response write. Feed export
// streams and sync-style handlers can run for many minutes; a short
// WriteTimeout aborts them mid-stream (clients see truncated/hanging
// responses). We use a long WriteTimeout ceiling instead of 0 (unlimited)
// so wedged handlers still release connections eventually. Finer per-route
// deadlines belong on request contexts / middleware for normal JSON APIs.
// Leaving WriteTimeout unset (0) would never reclaim a stuck writer.
func newHTTPServer(addr string, handler http.Handler) *http.Server {
	return &http.Server{
		Addr:    addr,
		Handler: handler,
		// Headers-only Slowloris guard (independent of ReadTimeout).
		ReadHeaderTimeout: 10 * time.Second,
		// Full request read (headers + body). Above typical API JSON uploads.
		ReadTimeout: 60 * time.Second,
		// Long ceiling so streaming exports/sync can finish; see comment above.
		WriteTimeout: 15 * time.Minute,
		// Close keep-alive connections idle between requests.
		IdleTimeout: 120 * time.Second,
	}
}