package metrics import ( "context" "fmt" "net" "net/http" _ "net/http/pprof" "runtime" "sync" "time" "github.com/facebookgo/grace/gracenet" "github.com/prometheus/client_golang/prometheus" "github.com/prometheus/client_golang/prometheus/promhttp" "github.com/rs/zerolog" "golang.org/x/net/trace" "github.com/cloudflare/cloudflared/diagnostic" ) const ( startupTime = time.Millisecond * 500 defaultShutdownTimeout = time.Second * 15 ) // This variable is set at compile time to allow the default local address to change. var Runtime = "host" func GetMetricsDefaultAddress(runtimeType string) string { // When issuing the diagnostic command we may have to reach a server that is // running in a virtual enviroment and in that case we must bind to 0.0.0.0 // otherwise the server won't be reachable. switch runtimeType { case "virtual": return "0.0.0.0:0" default: return "localhost:0" } } // GetMetricsKnownAddresses returns the addresses used by the metrics server to bind at // startup time to allow a semi-deterministic approach to know where the server is listening at. // The ports were selected because at the time we are in 2024 and they do not collide with any // know/registered port according https://en.wikipedia.org/wiki/List_of_TCP_and_UDP_port_numbers. func GetMetricsKnownAddresses(runtimeType string) []string { switch runtimeType { case "virtual": return []string{"0.0.0.0:20241", "0.0.0.0:20242", "0.0.0.0:20243", "0.0.0.0:20244", "0.0.0.0:20245"} default: return []string{"localhost:20241", "localhost:20242", "localhost:20243", "localhost:20244", "localhost:20245"} } } type Config struct { ReadyServer *ReadyServer DiagnosticHandler *diagnostic.Handler QuickTunnelHostname string Orchestrator orchestrator ShutdownTimeout time.Duration } type orchestrator interface { GetVersionedConfigJSON() ([]byte, error) } func newMetricsHandler( config Config, log *zerolog.Logger, ) *http.ServeMux { router := http.NewServeMux() router.Handle("/debug/", http.DefaultServeMux) router.Handle("/metrics", promhttp.Handler()) router.HandleFunc("/healthcheck", func(w http.ResponseWriter, r *http.Request) { _, _ = fmt.Fprintf(w, "OK\n") }) if config.ReadyServer != nil { router.Handle("/ready", config.ReadyServer) } router.HandleFunc("/quicktunnel", func(w http.ResponseWriter, r *http.Request) { _, _ = fmt.Fprintf(w, `{"hostname":"%s"}`, config.QuickTunnelHostname) }) if config.Orchestrator != nil { router.HandleFunc("/config", func(w http.ResponseWriter, r *http.Request) { json, err := config.Orchestrator.GetVersionedConfigJSON() if err != nil { w.WriteHeader(500) _, _ = fmt.Fprintf(w, "ERR: %v", err) log.Err(err).Msg("Failed to serve config") return } _, _ = w.Write(json) }) } config.DiagnosticHandler.InstallEndpoints(router) return router } // CreateMetricsListener will create a new [net.Listener] by using an // known set of ports when the default address is passed with the fallback // of choosing a random port when none is available. // // In case the provided address is not the default one then it will be used // as is. func CreateMetricsListener(listeners *gracenet.Net, laddr string) (net.Listener, error) { if laddr == GetMetricsDefaultAddress(Runtime) { // On the presence of the default address select // a port from the known set of addresses iteratively. addresses := GetMetricsKnownAddresses(Runtime) for _, address := range addresses { listener, err := listeners.Listen("tcp", address) if err == nil { return listener, nil } } // When no port is available then bind to a random one listener, err := listeners.Listen("tcp", laddr) if err != nil { return nil, fmt.Errorf("failed to listen to default metrics address: %w", err) } return listener, nil } // Explicitly got a local address then bind to it listener, err := listeners.Listen("tcp", laddr) if err != nil { return nil, fmt.Errorf("failed to bind to address (%s): %w", laddr, err) } return listener, nil } func ServeMetrics( l net.Listener, ctx context.Context, config Config, log *zerolog.Logger, ) (err error) { var wg sync.WaitGroup // Metrics port is privileged, so no need for further access control trace.AuthRequest = func(*http.Request) (bool, bool) { return true, true } // TODO: parameterize ReadTimeout and WriteTimeout. The maximum time we can // profile CPU usage depends on WriteTimeout h := newMetricsHandler(config, log) server := &http.Server{ ReadTimeout: 10 * time.Second, WriteTimeout: 10 * time.Second, Handler: h, } wg.Add(1) go func() { defer wg.Done() err = server.Serve(l) }() log.Info().Msgf("Starting metrics server on %s", fmt.Sprintf("%v/metrics", l.Addr())) // server.Serve will hang if server.Shutdown is called before the server is // fully started up. So add artificial delay. time.Sleep(startupTime) <-ctx.Done() shutdownTimeout := config.ShutdownTimeout if shutdownTimeout == 0 { shutdownTimeout = defaultShutdownTimeout } ctx, cancel := context.WithTimeout(context.Background(), shutdownTimeout) _ = server.Shutdown(ctx) cancel() wg.Wait() if err == http.ErrServerClosed { log.Info().Msg("Metrics server stopped") return nil } log.Err(err).Msg("Metrics server failed") return err } func RegisterBuildInfo(buildType, buildTime, version string) { buildInfo := prometheus.NewGaugeVec( prometheus.GaugeOpts{ // Don't namespace build_info, since we want it to be consistent across all Cloudflare services Name: "build_info", Help: "Build and version information", }, []string{"goversion", "type", "revision", "version"}, ) prometheus.MustRegister(buildInfo) buildInfo.WithLabelValues(runtime.Version(), buildType, buildTime, version).Set(1) }