use vespalib metrics library

4 files changed, 39 insertions, 67 deletions

diff --git a/configd/src/apps/sentinel/config-handler.cpp b/configd/src/apps/sentinel/config-handler.cpp
index e5222b0350b..03ea1da9617 100644
--- a/configd/src/apps/sentinel/config-handler.cpp
+++ b/configd/src/apps/sentinel/config-handler.cpp
@@ -395,61 +395,12 @@ 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
-fillTotalRestarts(vespalib::SimpleMetricSnapshot &snapshot,
-		  unsigned long totalRestarts)
-{
-    snapshot.addGauge("sentinel.totalRestarts",
-                      "how many times sentinel restarted a service since "
-		      "sentinel start",
-                      totalRestarts);
-}
-
-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);
-    fillRestarts(snapshot, _startMetrics.totalRestartsLastSnapshot);
-    fillTotalRestarts(snapshot, _startMetrics.totalRestartsCounter);
-    fillRunning(snapshot, _startMetrics.currentlyRunningServices);
-    fillUptime(snapshot, now - _startMetrics.startedTime);
-    _stateApi.myMetrics.setMetrics(snapshot.asString());
-
-    vespalib::SimpleMetricSnapshot totals(_startMetrics.startedTime, now);
-    fillRestarts(totals, _startMetrics.totalRestartsCounter);
-    fillRunning(totals, _startMetrics.currentlyRunningServices);
-    fillUptime(totals, now - _startMetrics.startedTime);
-    _stateApi.myMetrics.setTotalMetrics(totals.asString());
-
+    _startMetrics.maybeLog();
+    _stateApi.myMetrics.setMetrics(_startMetrics.producer.getMetrics(""));
+    _stateApi.myMetrics.setTotalMetrics(_startMetrics.producer.getTotalMetrics(""));
 }
 
 void
diff --git a/configd/src/apps/sentinel/metrics.cpp b/configd/src/apps/sentinel/metrics.cpp
index 42c61c06b50..bc9d87d0e73 100644
--- a/configd/src/apps/sentinel/metrics.cpp
+++ b/configd/src/apps/sentinel/metrics.cpp
@@ -4,18 +4,28 @@
 #include <vespa/log/log.h>
 LOG_SETUP(".metrics");
 
+#include <vespa/vespalib/metrics/simple_metrics.h>
+
 namespace config {
 namespace sentinel {
 
+using vespalib::metrics::SimpleMetricsManager;
+using vespalib::metrics::SimpleManagerConfig;
+
 StartMetrics::StartMetrics()
-    : currentlyRunningServices(0), totalRestartsCounter(0), totalRestartsLastPeriod(1),
-      lastLoggedTime(0),
-      totalRestartsLastSnapshot(0),
-      snapshotStart(0),
-      snapshotEnd(0)
+    : metrics(SimpleMetricsManager::create(SimpleManagerConfig())),
+      producer(metrics),
+      currentlyRunningServices(0),
+      totalRestartsCounter(0),
+      totalRestartsLastPeriod(1),
+      startedTime(time(nullptr)),
+      lastLoggedTime(startedTime - 55),
+      sentinel_restarts(metrics->counter("sentinel.restarts")),
+      sentinel_totalRestarts(metrics->gauge("sentinel.totalRestarts")),
+      sentinel_running(metrics->gauge("sentinel.running")),
+      sentinel_uptime(metrics->gauge("sentinel.uptime"))
 {
-    snapshotEnd = time(nullptr);
-    lastLoggedTime = snapshotEnd - 55;
+    sentinel_restarts.add();
 }
 
 void
@@ -29,9 +39,6 @@ StartMetrics::output()
 void
 StartMetrics::reset(unsigned long curTime)
 {
-    totalRestartsLastSnapshot = totalRestartsLastPeriod;
-    snapshotStart = snapshotEnd;
-    snapshotEnd = curTime;
     totalRestartsLastPeriod = 0;
     lastLoggedTime = curTime;
 }
@@ -40,6 +47,9 @@ void
 StartMetrics::maybeLog()
 {
     uint32_t curTime = time(nullptr);
+    sentinel_totalRestarts.sample(totalRestartsCounter);
+    sentinel_running.sample(currentlyRunningServices);
+    sentinel_uptime.sample(curTime - startedTime);
     if (curTime > lastLoggedTime + 59) {
         output();
         reset(curTime);
diff --git a/configd/src/apps/sentinel/metrics.h b/configd/src/apps/sentinel/metrics.h
index dd24bb2280d..2378a055663 100644
--- a/configd/src/apps/sentinel/metrics.h
+++ b/configd/src/apps/sentinel/metrics.h
@@ -2,20 +2,29 @@
 #pragma once
 
 #include <sys/time.h>
+#include <vespa/vespalib/metrics/simple_metrics.h>
 
 namespace config::sentinel {
 
+using vespalib::metrics::Counter;
+using vespalib::metrics::Gauge;
+using vespalib::metrics::MetricsManager;
+
 struct StartMetrics {
+    std::shared_ptr<MetricsManager> metrics;
+    vespalib::metrics::Producer producer;
     unsigned long currentlyRunningServices;
     unsigned long totalRestartsCounter;
     unsigned long totalRestartsLastPeriod;
-    long lastLoggedTime;
-    unsigned long totalRestartsLastSnapshot;
-    long snapshotStart;
-    long snapshotEnd;
     long startedTime;
+    long lastLoggedTime;
+    Counter sentinel_restarts;
+    Gauge   sentinel_totalRestarts;
+    Gauge   sentinel_running;
+    Gauge   sentinel_uptime;
 
     StartMetrics();
+    ~StartMetrics() {}
 
     void output();
     void reset(unsigned long curTime);
@@ -23,4 +32,3 @@ struct StartMetrics {
 };
 
 }
-
diff --git a/configd/src/apps/sentinel/service.cpp b/configd/src/apps/sentinel/service.cpp
index 2f13a05eb4f..3c762a957ec 100644
--- a/configd/src/apps/sentinel/service.cpp
+++ b/configd/src/apps/sentinel/service.cpp
@@ -239,6 +239,7 @@ Service::start()
     // ensureChildRuns(pipes[0]); // This will wait until the execl goes through
     setState(RUNNING);
     _metrics.currentlyRunningServices++;
+    _metrics.sentinel_running.sample(_metrics.currentlyRunningServices);
     close(pipes[0]); // close reading end
 
     using ns_log::LLParser;
@@ -314,6 +315,7 @@ Service::youExited(int status)
         setState(FAILED);
     }
     _metrics.currentlyRunningServices--;
+    _metrics.sentinel_running.sample(_metrics.currentlyRunningServices);
 
     if (_state == TERMINATING) {
         setState(TERMINATED);
@@ -326,6 +328,7 @@ Service::youExited(int status)
         setState(READY);
         _metrics.totalRestartsCounter++;
         _metrics.totalRestartsLastPeriod++;
+        _metrics.sentinel_restarts.add();
         start();
     }
 }