[release-4.20] OCPBUGS-77881: handle event socket broken pipe with automatic reconnection

cmd/main.go

@@ -159,7 +159,7 @@ func main() {
 		daemon.StartMetricsServer("0.0.0.0:9091")
 	}
 
-	daemon.StartReadyServer("0.0.0.0:8081", tracker)
+	daemon.StartReadyServer("0.0.0.0:8081", tracker, stdoutToSocket)
 
 	// Wait for one ticker interval before loading the profile
 	// This allows linuxptp-daemon connection to the cloud-event-proxy container to

pkg/daemon/daemon.go

@@ -85,6 +85,17 @@ var ptpTmpFiles = []string{
 	pmcSocketName,
 }
 
+const socketDialTimeout = 5 * time.Second
+
+func dialSocket() (net.Conn, error) {
+	c, err := net.DialTimeout("unix", eventSocket, socketDialTimeout)
+	if err != nil {
+		glog.Errorf("error trying to connect to event socket")
+		time.Sleep(connectionRetryInterval)
+	}
+	return c, err
+}
+
 // ProcessManager manages a set of ptpProcess
 // which could be ptp4l, phc2sys or timemaster.
 // Processes in ProcessManager will be started
@@ -171,6 +182,27 @@ func (p *ProcessManager) UpdateSynceConfig(config *synce.Relations) {
 
 }
 
+// EmitProcessStatusLogs emits process status logs using the EventHandler's
+// managed connection with reconnection support.
+func (p *ProcessManager) EmitProcessStatusLogs() {
+	for _, proc := range p.process {
+		status := PtpProcessUp
+		if proc.Stopped() {
+			status = PtpProcessDown
+		}
+		p.ptpEventHandler.EmitProcessStatusLog(proc.name, proc.configName, status)
+	}
+}
+
+// EmitClockClassLogs re-emits clock class via the EventHandler's managed connection.
+func (p *ProcessManager) EmitClockClassLogs() {
+	for _, proc := range p.process {
+		if proc.name == ptp4lProcessName {
+			p.ptpEventHandler.EmitClockClass(proc.configName)
+		}
+	}
+}
+
 type tBCProcessAttributes struct {
 	controlledPortsConfigFile string
 	// Time receiver interface name for T-BC clock monitoring

pkg/daemon/ready.go

@@ -76,10 +76,36 @@ func (h readyHandler) ServeHTTP(w http.ResponseWriter, r *http.Request) {
 	}
 }
 
-func StartReadyServer(bindAddress string, tracker *ReadyTracker) {
+type metricHandler struct {
+	tracker *ReadyTracker
+}
+
+func (h metricHandler) ServeHTTP(w http.ResponseWriter, _ *http.Request) {
+	if isReady, _ := h.tracker.Ready(); !isReady {
+		w.WriteHeader(http.StatusNoContent)
+		return
+	}
+	w.WriteHeader(http.StatusOK)
+
+	go func() {
+		eventHandler := h.tracker.processManager.ptpEventHandler
+		eventHandler.EmitClockSyncLogs()
+		eventHandler.EmitPortRoleLogs()
+
+		processManager := h.tracker.processManager
+		go processManager.EmitProcessStatusLogs()
+		go processManager.EmitClockClassLogs()
+	}()
+}
+
+// StartReadyServer ...
+func StartReadyServer(bindAddress string, tracker *ReadyTracker, serveInitMetrics bool) {
 	glog.Info("Starting Ready Server")
 	mux := http.NewServeMux()
 	mux.Handle("/ready", readyHandler{tracker: tracker})
+	if serveInitMetrics {
+		mux.Handle("/emit-logs", metricHandler{tracker: tracker})
+	}
 	go utilwait.Until(func() {
 		err := http.ListenAndServe(bindAddress, mux)
 		if err != nil {