Scale down when we have sufficient confidence

2 files changed, 22 insertions, 10 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 9b9b7dcecb0..5dbc6465411 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
@@ -9,6 +9,7 @@ import com.yahoo.vespa.hosted.provision.applications.ScalingEvent;
 
 import java.time.Clock;
 import java.time.Duration;
+import java.time.Instant;
 import java.util.Optional;
 import java.util.OptionalDouble;
 import java.util.logging.Level;
@@ -94,15 +95,31 @@ public class ClusterModel {
     /** Returns the relative load adjustment that should be made to this cluster given available measurements. */
     public Load loadAdjustment() {
         if (nodeTimeseries().measurementsPerNode() == 0) return Load.one(); // No info, no change
+
+        Load peak = nodeTimeseries().peakLoad().divide(idealLoad()); // Peak relative to ideal
+
         // Should we scale up?
-        Load relativePeak = nodeTimeseries().peakLoad().divide(idealLoad());
-        if (relativePeak.any(v -> v > 1.01)) // "meaningful growth": 1% over status quo.
-            return relativePeak.max(Load.one()); // Don't downscale any dimension if we upscale
+        if (peak.any(v -> v > 1.01)) // "meaningful growth": 1% over status quo.
+            return peak.map(v -> v < 1 ? 1 : v); // Don't downscale any dimension if we upscale
 
         // Should we scale down?
-        // TODO
+        if (canScaleDown())
+            return averageLoad().divide(idealLoad());
+
+        return Load.one();
+    }
+
+    /** Are we in a position to make decisions to scale down at this point? */
+    private boolean canScaleDown() {
+        if (hasScaledIn(scalingDuration().multipliedBy(3))) return false;
+        if (nodeTimeseries().measurementsPerNode() < 4) return false;
+        if (nodeTimeseries().nodesMeasured() != nodeCount()) return false;
+        return true;
+    }
 
-        return averageLoad().divide(idealLoad());
+    private boolean hasScaledIn(Duration period) {
+        return cluster.lastScalingEvent().map(event -> event.at()).orElse(Instant.MIN)
+                      .isAfter(clock.instant().minus(period));
     }
 
     /** Returns the predicted duration of a rescaling of this cluster */
diff --git a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/autoscale/Load.java b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/autoscale/Load.java
index 88c7e70cd35..6ab5ff731d3 100644
--- a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/autoscale/Load.java
+++ b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/autoscale/Load.java
@@ -54,11 +54,6 @@ public class Load {
         return new Load(divide(cpu, resources.vcpu()), divide(memory, resources.memoryGb()), divide(disk, resources.diskGb()));
     }
 
-    /** Returns the load having the max value of this and the given load in each dimension. */
-    public Load max(Load other) {
-        return join(other, (a, b) -> Math.max(a, b));
-    }
-
     /** Returns the load where the given function is applied to each dimension of this. */
     public Load map(DoubleUnaryOperator f) {
         return new Load(f.applyAsDouble(cpu),