prober web page improvements

Signed-off-by: Percy Wegmann <percy@tailscale.com>

cmd/derpprobe/derpprobe.go

@@ -106,7 +106,7 @@ func getOverallStatus(p *prober.Prober) (o overallStatus) {
 			// Do not show probes that have not finished yet.
 			continue
 		}
-		if i.Result {
+		if i.Status == prober.ProbeStatusSucceeded {
 			o.addGoodf("%s: %s", p, i.Latency)
 		} else {
 			o.addBadf("%s: %s", p, i.Error)

prober/prober.go

@@ -345,6 +345,11 @@ func (p *Probe) run() (pi ProbeInfo, err error) {
 		var cancel func()
 		ctx, cancel = context.WithTimeout(ctx, timeout)
 		defer cancel()
+	} else {
+		// For continuous probes, reset last error
+		p.mu.Lock()
+		p.lastErr = nil
+		p.mu.Unlock()
 	}
 
 	err = p.probeClass.Probe(ctx)
@@ -385,6 +390,16 @@ func (p *Probe) recordEnd(err error) {
 	p.successHist = p.successHist.Next()
 }
 
+// ProbeStatus indicates the status of a probe.
+type ProbeStatus string
+
+const (
+	ProbeStatusUnknown   = "unknown"
+	ProbeStatusRunning   = "running"
+	ProbeStatusFailed    = "failed"
+	ProbeStatusSucceeded = "succeeded"
+)
+
 // ProbeInfo is a snapshot of the configuration and state of a Probe.
 type ProbeInfo struct {
 	Name            string
@@ -394,7 +409,7 @@ type ProbeInfo struct {
 	Start           time.Time
 	End             time.Time
 	Latency         time.Duration
-	Result          bool
+	Status          ProbeStatus
 	Error           string
 	RecentResults   []bool
 	RecentLatencies []time.Duration
@@ -422,6 +437,10 @@ func (pb ProbeInfo) RecentMedianLatency() time.Duration {
 	return pb.RecentLatencies[len(pb.RecentLatencies)/2]
 }
 
+func (pb ProbeInfo) Continuous() bool {
+	return pb.Interval < 0
+}
+
 // ProbeInfo returns the state of all probes.
 func (p *Prober) ProbeInfo() map[string]ProbeInfo {
 	out := map[string]ProbeInfo{}
@@ -449,9 +468,14 @@ func (probe *Probe) probeInfoLocked() ProbeInfo {
 		Labels:   probe.metricLabels,
 		Start:    probe.start,
 		End:      probe.end,
-		Result:   probe.succeeded,
 	}
-	if probe.lastErr != nil {
+	inf.Status = ProbeStatusUnknown
+	if probe.end.Before(probe.start) {
+		inf.Status = ProbeStatusRunning
+	} else if probe.succeeded {
+		inf.Status = ProbeStatusSucceeded
+	} else if probe.lastErr != nil {
+		inf.Status = ProbeStatusFailed
 		inf.Error = probe.lastErr.Error()
 	}
 	if probe.latency > 0 {

prober/prober_test.go

@@ -316,7 +316,7 @@ func TestProberProbeInfo(t *testing.T) {
 			Interval:        probeInterval,
 			Labels:          map[string]string{"class": "", "name": "probe1"},
 			Latency:         500 * time.Millisecond,
-			Result:          true,
+			Status:          ProbeStatusSucceeded,
 			RecentResults:   []bool{true},
 			RecentLatencies: []time.Duration{500 * time.Millisecond},
 		},
@@ -324,6 +324,7 @@ func TestProberProbeInfo(t *testing.T) {
 			Name:            "probe2",
 			Interval:        probeInterval,
 			Labels:          map[string]string{"class": "", "name": "probe2"},
+			Status:          ProbeStatusFailed,
 			Error:           "error2",
 			RecentResults:   []bool{false},
 			RecentLatencies: nil, // no latency for failed probes
@@ -349,7 +350,7 @@ type probeResult struct {
 	}{
 		{
 			name:                    "no_runs",
-			wantProbeInfo:           ProbeInfo{},
+			wantProbeInfo:           ProbeInfo{Status: ProbeStatusUnknown},
 			wantRecentSuccessRatio:  0,
 			wantRecentMedianLatency: 0,
 		},
@@ -358,7 +359,7 @@ type probeResult struct {
 			results: []probeResult{{latency: 100 * time.Millisecond, err: nil}},
 			wantProbeInfo: ProbeInfo{
 				Latency:         100 * time.Millisecond,
-				Result:          true,
+				Status:          ProbeStatusSucceeded,
 				RecentResults:   []bool{true},
 				RecentLatencies: []time.Duration{100 * time.Millisecond},
 			},
@@ -369,7 +370,7 @@ type probeResult struct {
 			name:    "single_failure",
 			results: []probeResult{{latency: 100 * time.Millisecond, err: errors.New("error123")}},
 			wantProbeInfo: ProbeInfo{
-				Result:          false,
+				Status:          ProbeStatusFailed,
 				RecentResults:   []bool{false},
 				RecentLatencies: nil,
 				Error:           "error123",
@@ -390,7 +391,7 @@ type probeResult struct {
 				{latency: 80 * time.Millisecond, err: nil},
 			},
 			wantProbeInfo: ProbeInfo{
-				Result:        true,
+				Status:        ProbeStatusSucceeded,
 				Latency:       80 * time.Millisecond,
 				RecentResults: []bool{false, true, true, false, true, true, false, true},
 				RecentLatencies: []time.Duration{
@@ -420,7 +421,7 @@ type probeResult struct {
 				{latency: 110 * time.Millisecond, err: nil},
 			},
 			wantProbeInfo: ProbeInfo{
-				Result:        true,
+				Status:        ProbeStatusSucceeded,
 				Latency:       110 * time.Millisecond,
 				RecentResults: []bool{true, true, true, true, true, true, true, true, true, true},
 				RecentLatencies: []time.Duration{
@@ -483,7 +484,7 @@ func TestProberRunHandler(t *testing.T) {
 				ProbeInfo: ProbeInfo{
 					Name:          "success",
 					Interval:      probeInterval,
-					Result:        true,
+					Status:        ProbeStatusSucceeded,
 					RecentResults: []bool{true, true},
 				},
 				PreviousSuccessRatio: 1,
@@ -498,7 +499,7 @@ func TestProberRunHandler(t *testing.T) {
 				ProbeInfo: ProbeInfo{
 					Name:          "failure",
 					Interval:      probeInterval,
-					Result:        false,
+					Status:        ProbeStatusFailed,
 					Error:         "error123",
 					RecentResults: []bool{false, false},
 				},

prober/status.go

@@ -62,8 +62,9 @@ func (p *Prober) StatusHandler(opts ...statusHandlerOpt) tsweb.ReturnHandlerFunc
 	return func(w http.ResponseWriter, r *http.Request) error {
 		type probeStatus struct {
 			ProbeInfo
-			TimeSinceLast time.Duration
-			Links         map[string]template.URL
+			TimeSinceLastStart time.Duration
+			TimeSinceLastEnd   time.Duration
+			Links              map[string]template.URL
 		}
 		vars := struct {
 			Title           string
@@ -81,12 +82,15 @@ type probeStatus struct {
 
 		for name, info := range p.ProbeInfo() {
 			vars.TotalProbes++
-			if !info.Result {
+			if info.Error != "" {
 				vars.UnhealthyProbes++
 			}
 			s := probeStatus{ProbeInfo: info}
+			if !info.Start.IsZero() {
+				s.TimeSinceLastStart = time.Since(info.Start).Truncate(time.Second)
+			}
 			if !info.End.IsZero() {
-				s.TimeSinceLast = time.Since(info.End).Truncate(time.Second)
+				s.TimeSinceLastEnd = time.Since(info.End).Truncate(time.Second)
 			}
 			for textTpl, urlTpl := range params.probeLinks {
 				text, err := renderTemplate(textTpl, info)

prober/status.html

@@ -73,8 +73,9 @@
             <th>Name</th>
             <th>Probe Class & Labels</th>
             <th>Interval</th>
-            <th>Last Attempt</th>
-            <th>Success</th>
+            <th>Last Finished</th>
+            <th>Last Started</th>
+            <th>Status</th>
             <th>Latency</th>
             <th>Last Error</th>
         </tr></thead>
@@ -85,9 +86,11 @@
                 {{$name}}
                 {{range $text, $url := $probeInfo.Links}}
                 <br/>
-                <button onclick="location.href='{{$url}}';" type="button">
-                    {{$text}}
-                </button>
+                {{if not $probeInfo.Continuous}}
+                    <button onclick="location.href='{{$url}}';" type="button">
+                        {{$text}}
+                    </button>
+                {{end}}
                 {{end}}
             </td>
             <td>{{$probeInfo.Class}}<br/>
@@ -97,28 +100,48 @@
                 {{end}}
                 </div>
             </td>
-            <td>{{$probeInfo.Interval}}</td>
-            <td data-sort="{{$probeInfo.TimeSinceLast.Milliseconds}}">
-                {{if $probeInfo.TimeSinceLast}}
-                    {{$probeInfo.TimeSinceLast.String}} ago<br/>
+            <td>
+                {{if $probeInfo.Continuous}}
+                    Continuous
+                {{else}}
+                    {{$probeInfo.Interval}}
+                {{end}}
+            </td>
+            <td data-sort="{{$probeInfo.TimeSinceLastEnd.Milliseconds}}">
+                {{if $probeInfo.TimeSinceLastEnd}}
+                    {{$probeInfo.TimeSinceLastEnd.String}} ago<br/>
                     <span class="small">{{$probeInfo.End.Format "2006-01-02T15:04:05Z07:00"}}</span>
                 {{else}}
                     Never
                 {{end}}
             </td>
-            <td>
-                {{if $probeInfo.Result}}
-                    {{$probeInfo.Result}}
+            <td data-sort="{{$probeInfo.TimeSinceLastStart.Milliseconds}}">
+                {{if $probeInfo.TimeSinceLastStart}}
+                    {{$probeInfo.TimeSinceLastStart.String}} ago<br/>
+                    <span class="small">{{$probeInfo.Start.Format "2006-01-02T15:04:05Z07:00"}}</span>
                 {{else}}
-                    <span class="error">{{$probeInfo.Result}}</span>
+                    Never
+                {{end}}
+            </td>
+            <td>
+                {{if $probeInfo.Error}}
+                    <span class="error">{{$probeInfo.Status}}</span>
+                {{else}}
+                    {{$probeInfo.Status}}
                 {{end}}<br/>
-                <div class="small">Recent: {{$probeInfo.RecentResults}}</div>
-                <div class="small">Mean: {{$probeInfo.RecentSuccessRatio}}</div>
+                {{if not $probeInfo.Continuous}}
+                    <div class="small">Recent: {{$probeInfo.RecentResults}}</div>
+                    <div class="small">Mean: {{$probeInfo.RecentSuccessRatio}}</div>
+                {{end}}
             </td>
             <td data-sort="{{$probeInfo.Latency.Milliseconds}}">
-                {{$probeInfo.Latency.String}}
-                <div class="small">Recent: {{$probeInfo.RecentLatencies}}</div>
-                <div class="small">Median: {{$probeInfo.RecentMedianLatency}}</div>
+                {{if $probeInfo.Continuous}}
+                    n/a
+                {{else}}
+                    {{$probeInfo.Latency.String}}
+                    <div class="small">Recent: {{$probeInfo.RecentLatencies}}</div>
+                    <div class="small">Median: {{$probeInfo.RecentMedianLatency}}</div>
+                {{end}}
             </td>
             <td class="small">{{$probeInfo.Error}}</td>
         </tr>