Merge pull request #9748 from vespa-engine/hmusum/declare-and-update-jvm-heap-metrics-for-node-admin

Declare and update jvm heap metrics for node admin

3 files changed, 24 insertions, 2 deletions

diff --git a/node-admin/src/main/java/com/yahoo/vespa/hosted/node/admin/nodeadmin/NodeAdmin.java b/node-admin/src/main/java/com/yahoo/vespa/hosted/node/admin/nodeadmin/NodeAdmin.java
index ca9083e9d27..00ec985ba0c 100644
--- a/node-admin/src/main/java/com/yahoo/vespa/hosted/node/admin/nodeadmin/NodeAdmin.java
+++ b/node-admin/src/main/java/com/yahoo/vespa/hosted/node/admin/nodeadmin/NodeAdmin.java
@@ -9,7 +9,7 @@ import java.util.Set;
 
 /**
  * NodeAdmin manages the life cycle of NodeAgents.
- * @author dybis
+ * @author Haakon Dybdahl
  */
 public interface NodeAdmin {
 
@@ -19,6 +19,9 @@ public interface NodeAdmin {
     /** Gather node agent and its docker container metrics and forward them to the {@code MetricReceiverWrapper} */
     void updateNodeAgentMetrics();
 
+    /** Gather node admin metrics and forward them to the {@code MetricReceiverWrapper} */
+    void updateNodeAdminMetrics();
+
     /**
      * Attempts to freeze/unfreeze all NodeAgents and itself. To freeze a NodeAgent means that
      * they will not pick up any changes from NodeRepository.
@@ -29,7 +32,7 @@ public interface NodeAdmin {
     boolean setFrozen(boolean frozen);
 
     /**
-     * Returns whether the NodeAdmin itself is currently frozen, meaning it will not pick up any changes
+     * Returns whether NodeAdmin itself is currently frozen, meaning it will not pick up any changes
      * from NodeRepository.
      */
     boolean isFrozen();
diff --git a/node-admin/src/main/java/com/yahoo/vespa/hosted/node/admin/nodeadmin/NodeAdminImpl.java b/node-admin/src/main/java/com/yahoo/vespa/hosted/node/admin/nodeadmin/NodeAdminImpl.java
index 5b5d13ca346..0d520241ac8 100644
--- a/node-admin/src/main/java/com/yahoo/vespa/hosted/node/admin/nodeadmin/NodeAdminImpl.java
+++ b/node-admin/src/main/java/com/yahoo/vespa/hosted/node/admin/nodeadmin/NodeAdminImpl.java
@@ -43,6 +43,9 @@ public class NodeAdminImpl implements NodeAdmin {
     private final Map<String, NodeAgentWithScheduler> nodeAgentWithSchedulerByHostname = new ConcurrentHashMap<>();
 
     private final GaugeWrapper numberOfContainersInLoadImageState;
+    private final GaugeWrapper jvmHeapUsed;
+    private final GaugeWrapper jvmHeapFree;
+    private final GaugeWrapper jvmHeapTotal;
     private final CounterWrapper numberOfUnhandledExceptionsInNodeAgent;
 
     public NodeAdminImpl(NodeAgentFactory nodeAgentFactory, MetricReceiverWrapper metricReceiver, Clock clock) {
@@ -70,6 +73,10 @@ public class NodeAdminImpl implements NodeAdmin {
         Dimensions dimensions = new Dimensions.Builder().add("role", "docker").build();
         this.numberOfContainersInLoadImageState = metricReceiver.declareGauge(MetricReceiverWrapper.APPLICATION_DOCKER, dimensions, "nodes.image.loading");
         this.numberOfUnhandledExceptionsInNodeAgent = metricReceiver.declareCounter(MetricReceiverWrapper.APPLICATION_DOCKER, dimensions, "nodes.unhandled_exceptions");
+
+        this.jvmHeapUsed = metricReceiver.declareGauge(MetricReceiverWrapper.APPLICATION_HOST, new Dimensions.Builder().build(), "mem.heap.used");
+        this.jvmHeapFree = metricReceiver.declareGauge(MetricReceiverWrapper.APPLICATION_HOST, new Dimensions.Builder().build(), "mem.heap.free");
+        this.jvmHeapTotal = metricReceiver.declareGauge(MetricReceiverWrapper.APPLICATION_HOST, new Dimensions.Builder().build(), "mem.heap.total");
     }
 
     @Override
@@ -113,6 +120,17 @@ public class NodeAdminImpl implements NodeAdmin {
     }
 
     @Override
+    public void updateNodeAdminMetrics() {
+        Runtime runtime = Runtime.getRuntime();
+        long freeMemory = runtime.freeMemory();
+        long totalMemory = runtime.totalMemory();
+        long usedMemory = totalMemory - freeMemory;
+        jvmHeapFree.sample(freeMemory);
+        jvmHeapUsed.sample(usedMemory);
+        jvmHeapTotal.sample(totalMemory);
+    }
+
+    @Override
     public boolean setFrozen(boolean wantFrozen) {
         if (wantFrozen != previousWantFrozen) {
             if (wantFrozen) {
diff --git a/node-admin/src/main/java/com/yahoo/vespa/hosted/node/admin/nodeadmin/NodeAdminStateUpdater.java b/node-admin/src/main/java/com/yahoo/vespa/hosted/node/admin/nodeadmin/NodeAdminStateUpdater.java
index 41c4544c533..2cd15a3ebe4 100644
--- a/node-admin/src/main/java/com/yahoo/vespa/hosted/node/admin/nodeadmin/NodeAdminStateUpdater.java
+++ b/node-admin/src/main/java/com/yahoo/vespa/hosted/node/admin/nodeadmin/NodeAdminStateUpdater.java
@@ -81,6 +81,7 @@ public class NodeAdminStateUpdater {
             try {
                 if (suspendedStates.contains(currentState)) return;
                 nodeAdmin.updateNodeAgentMetrics();
+                nodeAdmin.updateNodeAdminMetrics();
             } catch (Throwable e) {
                 log.log(Level.WARNING, "Metric fetcher scheduler failed", e);
             }