Merge pull request #17609 from vespa-engine/bratseth/max-duration

Set a max scaling duration

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 e3622c8f076..f5eb67f0979 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
@@ -150,7 +150,8 @@ public class ClusterModel {
         for (ScalingEvent event : cluster.scalingEvents()) {
             if (event.duration().isEmpty()) continue;
             completedEventCount++;
-            totalDuration = totalDuration.plus(event.duration().get());
+            // Assume we have missed timely recording completion if it is longer than 4 days
+            totalDuration = totalDuration.plus(maximum(Duration.ofDays(4), event.duration().get()));
         }
 
         if (completedEventCount == 0) { // Use defaults
@@ -160,13 +161,25 @@ public class ClusterModel {
         else {
             Duration predictedDuration = totalDuration.dividedBy(completedEventCount);
 
-            // TODO: Remove when we have reliable completion for content clusters
-            if (clusterSpec.isStateful() && predictedDuration.minus(Duration.ofHours(12)).isNegative())
-                return Duration.ofHours(12);
+            if ( clusterSpec.isStateful() ) // TODO: Remove when we have reliable completion for content clusters
+                predictedDuration = minimum(Duration.ofHours(12), predictedDuration);
+
+            predictedDuration = minimum(Duration.ofMinutes(5), predictedDuration);
 
-            if (predictedDuration.minus(Duration.ofMinutes(5)).isNegative()) return Duration.ofMinutes(5); // minimum
             return predictedDuration;
         }
     }
 
+    private static Duration minimum(Duration smallestAllowed, Duration duration) {
+        if (duration.minus(smallestAllowed).isNegative())
+            return smallestAllowed;
+        return duration;
+    }
+
+    private static Duration maximum(Duration largestAllowed, Duration duration) {
+        if ( ! duration.minus(largestAllowed).isNegative())
+            return largestAllowed;
+        return duration;
+    }
+
 }
diff --git a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/maintenance/AutoscalingMaintainer.java b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/maintenance/AutoscalingMaintainer.java
index 17d33ef501c..7da6e0d3ebe 100644
--- a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/maintenance/AutoscalingMaintainer.java
+++ b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/maintenance/AutoscalingMaintainer.java
@@ -111,16 +111,15 @@ public class AutoscalingMaintainer extends NodeRepositoryMaintainer {
         if (clusterNodes.retired().stream()
                         .anyMatch(node -> node.history().hasEventAt(History.Event.Type.retired, event.at())))
             return cluster;
-        // - 2. all nodes have switched to the right config generation
+        // - 2. all nodes have switched to the right config generation (currently only measured on containers)
         for (var nodeTimeseries : nodeRepository().metricsDb().getNodeTimeseries(Duration.between(event.at(), clock().instant()),
                                                                                  clusterNodes)) {
-            Optional<NodeMetricSnapshot> firstOnNewGeneration =
+            Optional<NodeMetricSnapshot> onNewGeneration =
                     nodeTimeseries.asList().stream()
-                                           .filter(snapshot -> snapshot.generation() >= event.generation()).findFirst();
-            if (firstOnNewGeneration.isEmpty()) return cluster; // Not completed
+                                  .filter(snapshot -> snapshot.generation() >= event.generation()).findAny();
+            if (onNewGeneration.isEmpty()) return cluster; // Not completed
         }
 
-
         // Set the completion time to the instant we notice completion.
         Instant completionTime = nodeRepository().clock().instant();
         return cluster.with(event.withCompletion(completionTime));