Merge pull request #26111 from vespa-engine/bratseth/autoscaling-measurements

Bratseth/autoscaling measurements

2 files changed, 5 insertions, 32 deletions

diff --git a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/autoscale/ClusterModel.java b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/autoscale/ClusterModel.java
index a30c9b588c2..264664f91b2 100644
--- a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/autoscale/ClusterModel.java
+++ b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/autoscale/ClusterModel.java
@@ -119,7 +119,7 @@ public class ClusterModel {
 
     /** Returns the relative load adjustment that should be made to this cluster given available measurements. */
     public Load loadAdjustment() {
-        if (nodeTimeseries().isEmpty()) return Load.one();
+        if (nodeTimeseries().measurementsPerNode() < 0.5) return Load.one(); // Don't change based on very little data
         Load adjustment = peakLoad().divide(idealLoad());
         if (! safeToScaleDown())
             adjustment = adjustment.map(v -> v < 1 ? 1 : v);
diff --git a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/autoscale/ClusterNodesTimeseries.java b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/autoscale/ClusterNodesTimeseries.java
index b86a24af5c9..0be4175c2c1 100644
--- a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/autoscale/ClusterNodesTimeseries.java
+++ b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/autoscale/ClusterNodesTimeseries.java
@@ -28,10 +28,10 @@ public class ClusterNodesTimeseries {
     public ClusterNodesTimeseries(Duration period, Cluster cluster, NodeList clusterNodes, MetricsDb db) {
         this.clusterNodes = clusterNodes;
 
-        // See warmupSeconds*4 into the past to see any generation change in it
+        // See warmupDuration*4 into the past to see any generation change in it.
         // If none can be detected we assume the node is new/was down.
         // If either this is the case, or there is a generation change, we ignore
-        // the first warmupWindow metrics
+        // the first warmupWindow metrics.
         var timeseries = db.getNodeTimeseries(period.plus(warmupDuration.multipliedBy(4)), clusterNodes);
         if (cluster.lastScalingEvent().isPresent()) {
             long currentGeneration = cluster.lastScalingEvent().get().generation();
@@ -52,42 +52,15 @@ public class ClusterNodesTimeseries {
     }
 
     /** Returns the average number of measurements per node */
-    public int measurementsPerNode() {
+    public double measurementsPerNode() {
         if (clusterNodes.size() == 0) return 0;
         int measurementCount = timeseries.stream().mapToInt(m -> m.size()).sum();
-        return measurementCount / clusterNodes.size();
+        return (double)measurementCount / clusterNodes.size();
     }
 
     /** Returns the number of nodes measured in this */
     public int nodesMeasured() { return timeseries.size(); }
 
-    /** Returns the average load after the given instant */
-    public Load averageLoad() {
-        Load total = Load.zero();
-        int count = 0;
-        for (var nodeTimeseries : timeseries) {
-            for (var snapshot : nodeTimeseries.asList()) {
-                total = total.add(snapshot.load());
-                count++;
-            }
-        }
-        return total.divide(count);
-    }
-
-    /** Returns average of the latest load reading from each node */
-    public Load currentLoad() {
-        Load total = Load.zero();
-        int count = 0;
-        for (var nodeTimeseries : timeseries) {
-            Optional<NodeMetricSnapshot> last = nodeTimeseries.last();
-            if (last.isEmpty()) continue;
-
-            total = total.add(last.get().load());
-            count++;
-        }
-        return total.divide(count);
-    }
-
     /**
      * Returns the "peak load" in this: Which is for each load dimension,
      * the average of the highest reading for that dimension on each node.