Merge pull request #3781 from vespa-engine/arnej/add-uptime-metric-for-sentinel

Arnej/add uptime metric for sentinel

3 files changed, 38 insertions, 10 deletions

diff --git a/config-model/src/main/java/com/yahoo/vespa/model/admin/monitoring/VespaMetricSet.java b/config-model/src/main/java/com/yahoo/vespa/model/admin/monitoring/VespaMetricSet.java
index ccf7974d381..4995e238879 100644
--- a/config-model/src/main/java/com/yahoo/vespa/model/admin/monitoring/VespaMetricSet.java
+++ b/config-model/src/main/java/com/yahoo/vespa/model/admin/monitoring/VespaMetricSet.java
@@ -40,6 +40,7 @@ public class VespaMetricSet {
         Set<Metric> metrics = new LinkedHashSet<>();
 
         metrics.add(new Metric("sentinel.restarts.count"));
+        metrics.add(new Metric("sentinel.uptime.last", "sentinel.uptime"));
 
         metrics.add(new Metric("sentinel.running.count"));
         metrics.add(new Metric("sentinel.running.last"));
diff --git a/configd/src/apps/sentinel/config-handler.cpp b/configd/src/apps/sentinel/config-handler.cpp
index 1075bc4d7ce..0527982c99f 100644
--- a/configd/src/apps/sentinel/config-handler.cpp
+++ b/configd/src/apps/sentinel/config-handler.cpp
@@ -395,21 +395,48 @@ ConfigHandler::handleCommand(CommandConnection *c)
     }
 }
 
+namespace {
+
+void
+fillRestarts(vespalib::SimpleMetricSnapshot &snapshot, unsigned long restarts)
+{
+    snapshot.addCount("sentinel.restarts",
+                      "how many times sentinel restarted a service",
+                      restarts);
+}
+
+void
+fillRunning(vespalib::SimpleMetricSnapshot &snapshot, unsigned long running)
+{
+    snapshot.addGauge("sentinel.running",
+                      "how many services the sentinel has running currently",
+                      running);
+}
+
+void
+fillUptime(vespalib::SimpleMetricSnapshot &snapshot, long uptime)
+{
+    snapshot.addGauge("sentinel.uptime",
+                      "how many seconds has the sentinel been running",
+                      uptime);
+}
+
+} // namespace <unnamed>
+
 void
 ConfigHandler::updateMetrics()
 {
+    time_t now = time(nullptr);
     vespalib::SimpleMetricSnapshot snapshot(_startMetrics.snapshotStart, _startMetrics.snapshotEnd);
-    snapshot.addCount("sentinel.restarts", "how many times sentinel restarted a service",
-                      _startMetrics.totalRestartsLastSnapshot);
-    snapshot.addGauge("sentinel.running", "how many services the sentinel has running currently",
-                      _startMetrics.currentlyRunningServices);
+    fillRestarts(snapshot, _startMetrics.totalRestartsLastSnapshot);
+    fillRunning(snapshot, _startMetrics.currentlyRunningServices);
+    fillUptime(snapshot, now - _startMetrics.startedTime);
     _stateApi.myMetrics.setMetrics(snapshot.asString());
 
-    vespalib::SimpleMetricSnapshot totals(_startMetrics.startedTime, time(nullptr));
-    totals.addCount("sentinel.restarts", "how many times sentinel restarted a service",
-                    _startMetrics.totalRestartsCounter);
-    totals.addGauge("sentinel.running", "how many services the sentinel has running currently",
-                    _startMetrics.currentlyRunningServices);
+    vespalib::SimpleMetricSnapshot totals(_startMetrics.startedTime, now);
+    fillRestarts(totals, _startMetrics.totalRestartsCounter);
+    fillRunning(totals, _startMetrics.currentlyRunningServices);
+    fillUptime(totals, now - _startMetrics.startedTime);
     _stateApi.myMetrics.setTotalMetrics(totals.asString());
 
 }
diff --git a/configd/src/apps/sentinel/metrics.cpp b/configd/src/apps/sentinel/metrics.cpp
index 078dfa275d9..42c61c06b50 100644
--- a/configd/src/apps/sentinel/metrics.cpp
+++ b/configd/src/apps/sentinel/metrics.cpp
@@ -8,7 +8,7 @@ namespace config {
 namespace sentinel {
 
 StartMetrics::StartMetrics()
-    : currentlyRunningServices(0), totalRestartsCounter(0), totalRestartsLastPeriod(0),
+    : currentlyRunningServices(0), totalRestartsCounter(0), totalRestartsLastPeriod(1),
       lastLoggedTime(0),
       totalRestartsLastSnapshot(0),
       snapshotStart(0),