Merge pull request #14321 from vespa-engine/bratseth/autoscale-reaction-time

Cluster size independent autoscaling reaction time

3 files changed, 6 insertions, 13 deletions

diff --git a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/autoscale/Autoscaler.java b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/autoscale/Autoscaler.java
index a02043aea78..3dd60fe440c 100644
--- a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/autoscale/Autoscaler.java
+++ b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/autoscale/Autoscaler.java
@@ -20,15 +20,7 @@ import java.util.stream.Collectors;
  */
 public class Autoscaler {
 
-    /*
-     TODO:
-     - Scale group size
-     - Consider taking spikes/variance into account
-     - Measure observed regulation lag (startup+redistribution) and take it into account when deciding regulation observation window
-     - Scale by performance not just load+cost
-     */
-
-    private static final int minimumMeasurements = 500; // TODO: Per node instead? Also say something about interval?
+    private static final int minimumMeasurementsPerNode = 60; // 1 hour
 
     /** What cost difference factor is worth a reallocation? */
     private static final double costDifferenceWorthReallocation = 0.1;
@@ -113,8 +105,10 @@ public class Autoscaler {
                                                           resource,
                                                           clusterNodes.stream().map(Node::hostname).collect(Collectors.toList()));
 
-        if (window.measurementCount() < minimumMeasurements) return Optional.empty();
-        if (window.hostnames() != clusterNodes.size()) return Optional.empty(); // Regulate only when all nodes are measured
+        // Require a total number of measurements scaling with the number of nodes,
+        // but don't require that we have at least that many from every node
+        if (window.measurementCount()/clusterNodes.size() < minimumMeasurementsPerNode) return Optional.empty();
+        if (window.hostnames() != clusterNodes.size()) return Optional.empty();
 
         return Optional.of(window.average());
     }
diff --git a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/autoscale/NodeMetricsDb.java b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/autoscale/NodeMetricsDb.java
index 75bc73df5b0..7fa81f0b78e 100644
--- a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/autoscale/NodeMetricsDb.java
+++ b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/autoscale/NodeMetricsDb.java
@@ -22,7 +22,6 @@ import java.util.stream.Collectors;
  */
 public class NodeMetricsDb {
 
-    private Logger log = Logger.getLogger(NodeMetricsDb.class.getName());
     private static final Duration dbWindow = Duration.ofHours(24);
 
     /** Measurements by key. Each list of measurements is sorted by increasing timestamp */
diff --git a/node-repository/src/test/java/com/yahoo/vespa/hosted/provision/autoscale/AutoscalingTest.java b/node-repository/src/test/java/com/yahoo/vespa/hosted/provision/autoscale/AutoscalingTest.java
index a0a44e4f342..2bcbac828b8 100644
--- a/node-repository/src/test/java/com/yahoo/vespa/hosted/provision/autoscale/AutoscalingTest.java
+++ b/node-repository/src/test/java/com/yahoo/vespa/hosted/provision/autoscale/AutoscalingTest.java
@@ -46,7 +46,7 @@ public class AutoscalingTest {
 
         assertTrue("No measurements -> No change", tester.autoscale(application1, cluster1.id(), min, max).isEmpty());
 
-        tester.addMeasurements(Resource.cpu, 0.25f, 1f, 60, application1);
+        tester.addMeasurements(Resource.cpu, 0.25f, 1f, 59, application1);
         assertTrue("Too few measurements -> No change", tester.autoscale(application1, cluster1.id(), min, max).isEmpty());
 
         tester.addMeasurements(Resource.cpu, 0.25f, 1f, 60, application1);