The monitor starts the Varnish child process if it is stopped.

It should be configurable whether this happens -- an admin may want
to intentionally stop Varnish, and this feature would interfere with
that.

pkg/varnish/metrics.go

@@ -54,6 +54,8 @@ type instanceMetrics struct {
 	pings           prometheus.Counter
 	pingFails       prometheus.Counter
 	panics          prometheus.Counter
+	restarts        prometheus.Counter
+	restartFails    prometheus.Counter
 	childRunning    prometheus.Counter
 	childNotRunning prometheus.Counter
 	vclDiscards     prometheus.Counter
@@ -163,6 +165,18 @@ var (
 			Name:      "panics_total",
 			Help:      "Total number of panics detected",
 		}),
+		restarts: prometheus.NewCounter(prometheus.CounterOpts{
+			Namespace: namespace,
+			Subsystem: subsystem,
+			Name:      "restarts_total",
+			Help:      "Total number of child process restarts",
+		}),
+		restartFails: prometheus.NewCounter(prometheus.CounterOpts{
+			Namespace: namespace,
+			Subsystem: subsystem,
+			Name:      "restart_fails_total",
+			Help:      "Total number of child process restart failures",
+		}),
 		childRunning: prometheus.NewCounter(prometheus.CounterOpts{
 			Namespace: namespace,
 			Subsystem: subsystem,
@@ -219,6 +233,8 @@ func initMetrics() {
 	prometheus.Register(metrics.pings)
 	prometheus.Register(metrics.pingFails)
 	prometheus.Register(metrics.panics)
+	prometheus.Register(metrics.restarts)
+	prometheus.Register(metrics.restartFails)
 	prometheus.Register(metrics.childRunning)
 	prometheus.Register(metrics.childNotRunning)
 	prometheus.Register(metrics.vclDiscards)

pkg/varnish/monitor.go

@@ -41,6 +41,8 @@ const (
 	pingErr      = "PingFailure"
 	statusErr    = "StatusFailure"
 	statusNotRun = "StatusNotRunning"
+	restartErr   = "RestartFailure"
+	restart      = "Restart"
 	panicErr     = "PanicFailure"
 	panic        = "Panic"
 	vclListErr   = "VCLListFailure"
@@ -118,6 +120,19 @@ func (vc *Controller) checkInst() bool {
 	} else {
 		metrics.childNotRunning.Inc()
 		vc.warnEvt(statusNotRun, "Status: %s", state)
+		// XXX configure whether to auto-start
+		if state == admin.Stopped {
+			vc.log.Debug("Attempting restart")
+			if err := adm.Start(); err != nil {
+				vc.warnEvt(restartErr, "Cannot restart: %v",
+					err)
+				metrics.restartFails.Inc()
+				return false
+			} else {
+				vc.infoEvt(restart, "Child restarted")
+				metrics.restarts.Inc()
+			}
+		}
 	}
 
 	panic, err := adm.GetPanic()