Expose more cluster data in nodes/v2

9 files changed, 79 insertions, 38 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 5d5c6fdac5a..c7549a5ddee 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
@@ -80,8 +80,13 @@ public class Autoscaler {
                                "Have measurements from " + nodesMeasured + " nodes, but require from " + clusterNodes.size());
 
 
+        var scalingDuration = cluster.scalingDuration(clusterNodes.clusterSpec());
         var clusterTimeseries = metricsDb.getClusterTimeseries(application.id(), cluster.id());
-        var target = ResourceTarget.idealLoad(clusterTimeseries, clusterNodesTimeseries, currentAllocation, application);
+        var target = ResourceTarget.idealLoad(scalingDuration,
+                                              clusterTimeseries,
+                                              clusterNodesTimeseries,
+                                              currentAllocation,
+                                              application);
 
         Optional<AllocatableClusterResources> bestAllocation =
                 allocationOptimizer.findBestAllocation(target, currentAllocation, limits);
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 173d76e4c26..2d0e77742ec 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
@@ -16,14 +16,12 @@ import java.util.stream.Collectors;
  */
 public class ClusterNodesTimeseries {
 
-    private final Cluster cluster;
     private final NodeList clusterNodes;
 
     /** The measurements for all nodes in this snapshot */
     private final List<NodeTimeseries> timeseries;
 
     public ClusterNodesTimeseries(Duration period, Cluster cluster, NodeList clusterNodes, MetricsDb db) {
-        this.cluster = cluster;
         this.clusterNodes = clusterNodes;
         var timeseries = db.getNodeTimeseries(period, clusterNodes);
 
@@ -35,12 +33,6 @@ public class ClusterNodesTimeseries {
         this.timeseries = timeseries;
     }
 
-    /** The cluster this is a timeseries for */
-    public Cluster cluster() { return cluster; }
-
-    /** The nodes of the cluster this is a timeseries for */
-    public NodeList clusterNodes() { return clusterNodes; }
-
     /** Returns the average number of measurements per node */
     public int measurementsPerNode() {
         int measurementCount = timeseries.stream().mapToInt(m -> m.size()).sum();
diff --git a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/autoscale/ClusterTimeseries.java b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/autoscale/ClusterTimeseries.java
index 5b6ed43b713..a435814c21e 100644
--- a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/autoscale/ClusterTimeseries.java
+++ b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/autoscale/ClusterTimeseries.java
@@ -23,6 +23,8 @@ public class ClusterTimeseries {
     private final ClusterSpec.Id cluster;
     private final List<ClusterMetricSnapshot> snapshots;
 
+    private Double cachedMaxQueryGrowthRate = null;
+
     ClusterTimeseries(ClusterSpec.Id cluster, List<ClusterMetricSnapshot> snapshots) {
         this.cluster = cluster;
         List<ClusterMetricSnapshot> sortedSnapshots = new ArrayList<>(snapshots);
@@ -48,6 +50,12 @@ public class ClusterTimeseries {
 
     /** The max query growth rate we can predict from this time-series as a fraction of the current traffic per minute */
     public double maxQueryGrowthRate() {
+        if (cachedMaxQueryGrowthRate != null)
+            return cachedMaxQueryGrowthRate;
+        return cachedMaxQueryGrowthRate = computeMaxQueryGrowthRate();
+    }
+
+    private double computeMaxQueryGrowthRate() {
         if (snapshots.isEmpty()) return 0.1;
 
         // Find the period having the highest growth rate, where total growth exceeds 30% increase
diff --git a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/autoscale/ResourceTarget.java b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/autoscale/ResourceTarget.java
index f29181b8343..ab6a6d548e9 100644
--- a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/autoscale/ResourceTarget.java
+++ b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/autoscale/ResourceTarget.java
@@ -49,12 +49,13 @@ public class ResourceTarget {
     }
 
     /** Create a target of achieving ideal load given a current load */
-    public static ResourceTarget idealLoad(ClusterTimeseries clusterTimeseries,
+    public static ResourceTarget idealLoad(Duration scalingDuration,
+                                           ClusterTimeseries clusterTimeseries,
                                            ClusterNodesTimeseries clusterNodesTimeseries,
                                            AllocatableClusterResources current,
                                            Application application) {
         return new ResourceTarget(nodeUsage(Resource.cpu, clusterNodesTimeseries.averageLoad(Resource.cpu), current)
-                                  / idealCpuLoad(clusterTimeseries, clusterNodesTimeseries, application),
+                                  / idealCpuLoad(scalingDuration, clusterTimeseries, application),
                                   nodeUsage(Resource.memory, clusterNodesTimeseries.averageLoad(Resource.memory), current)
                                   / Resource.memory.idealAverageLoad(),
                                   nodeUsage(Resource.disk, clusterNodesTimeseries.averageLoad(Resource.disk), current)
@@ -71,12 +72,11 @@ public class ResourceTarget {
     }
 
     /** Ideal cpu load must take the application traffic fraction into account */
-    private static double idealCpuLoad(ClusterTimeseries clusterTimeseries,
-                                       ClusterNodesTimeseries clusterNodesTimeseries,
-                                       Application application) {
+    public static double idealCpuLoad(Duration scalingDuration,
+                                      ClusterTimeseries clusterTimeseries,
+                                      Application application) {
         // What's needed to have headroom for growth during scale-up as a fraction of current resources?
         double maxGrowthRate = clusterTimeseries.maxQueryGrowthRate(); // in fraction per minute of the current traffic
-        Duration scalingDuration = clusterNodesTimeseries.cluster().scalingDuration(clusterNodesTimeseries.clusterNodes().clusterSpec());
         double growthRateHeadroom = 1 + maxGrowthRate * scalingDuration.toMinutes();
         // Cap headroom at 10% above the historical observed peak
         double fractionOfMax = clusterTimeseries.currentQueryFractionOfMax();
@@ -96,4 +96,8 @@ public class ResourceTarget {
         return 1 / growthRateHeadroom * 1 / trafficShiftHeadroom * Resource.cpu.idealAverageLoad();
     }
 
+    public static double idealMemoryLoad() { return Resource.memory.idealAverageLoad(); }
+
+    public static double idealDiskLoad() { return Resource.disk.idealAverageLoad(); }
+
 }
diff --git a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/maintenance/SpareCapacityMaintainer.java b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/maintenance/SpareCapacityMaintainer.java
index 79f3dad75d3..0307ae13b24 100644
--- a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/maintenance/SpareCapacityMaintainer.java
+++ b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/maintenance/SpareCapacityMaintainer.java
@@ -163,7 +163,7 @@ public class SpareCapacityMaintainer extends NodeRepositoryMaintainer {
 
         ApplicationId application = nodeToRetire.get().allocation().get().owner();
         try (MaintenanceDeployment deployment = new MaintenanceDeployment(application, deployer, metric, nodeRepository())) {
-            if ( ! deployment.isValid()) return; // this will be done at another config server
+            if ( ! deployment.isValid()) return;
 
             Optional<Node> nodeWithWantToRetire = nodeRepository().nodes().node(nodeToRetire.get().hostname())
                     .map(node -> node.withWantToRetire(true, Agent.SpareCapacityMaintainer, nodeRepository().clock().instant()));
@@ -171,7 +171,7 @@ public class SpareCapacityMaintainer extends NodeRepositoryMaintainer {
 
             nodeRepository().nodes().write(nodeWithWantToRetire.get(), deployment.applicationLock().get());
             log.log(Level.INFO, String.format("Redeploying %s to move %s from overcommitted host",
-                    application, nodeToRetire.get().hostname()));
+                                              application, nodeToRetire.get().hostname()));
             deployment.activate();
         }
     }
diff --git a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/restapi/ApplicationSerializer.java b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/restapi/ApplicationSerializer.java
index 4235bae6850..4d1c963d8ea 100644
--- a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/restapi/ApplicationSerializer.java
+++ b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/restapi/ApplicationSerializer.java
@@ -1,6 +1,7 @@
 // Copyright Verizon Media. Licensed under the terms of the Apache 2.0 license. See LICENSE in the project root.
 package com.yahoo.vespa.hosted.provision.restapi;
 
+import com.yahoo.config.provision.ApplicationId;
 import com.yahoo.config.provision.ClusterResources;
 import com.yahoo.slime.Cursor;
 import com.yahoo.slime.Slime;
@@ -10,8 +11,10 @@ import com.yahoo.vespa.hosted.provision.applications.Application;
 import com.yahoo.vespa.hosted.provision.applications.Cluster;
 import com.yahoo.vespa.hosted.provision.applications.ScalingEvent;
 import com.yahoo.vespa.hosted.provision.autoscale.ClusterNodesTimeseries;
+import com.yahoo.vespa.hosted.provision.autoscale.ClusterTimeseries;
 import com.yahoo.vespa.hosted.provision.autoscale.MetricsDb;
 import com.yahoo.vespa.hosted.provision.autoscale.Resource;
+import com.yahoo.vespa.hosted.provision.autoscale.ResourceTarget;
 
 import java.net.URI;
 import java.time.Duration;
@@ -20,50 +23,60 @@ import java.util.List;
 
 /**
  * Serializes application information for nodes/v2/application responses
+ *
+ * @author bratseth
  */
 public class ApplicationSerializer {
 
-    public static Slime toSlime(Application application, List<Node> applicationNodes, MetricsDb metricsDb, URI applicationUri) {
+    public static Slime toSlime(Application application, NodeList applicationNodes, MetricsDb metricsDb, URI applicationUri) {
         Slime slime = new Slime();
         toSlime(application, applicationNodes, metricsDb, slime.setObject(), applicationUri);
         return slime;
     }
 
     private static void toSlime(Application application,
-                                List<Node> applicationNodes,
+                                NodeList applicationNodes,
                                 MetricsDb metricsDb,
                                 Cursor object,
                                 URI applicationUri) {
         object.setString("url", applicationUri.toString());
         object.setString("id", application.id().toFullString());
-        clustersToSlime(application.clusters().values(), applicationNodes, metricsDb, object.setObject("clusters"));
+        clustersToSlime(application, applicationNodes, metricsDb, object.setObject("clusters"));
     }
 
-    private static void clustersToSlime(Collection<Cluster> clusters,
-                                        List<Node> applicationNodes,
+    private static void clustersToSlime(Application application,
+                                        NodeList applicationNodes,
                                         MetricsDb metricsDb,
                                         Cursor clustersObject) {
-        clusters.forEach(cluster -> toSlime(cluster, applicationNodes, metricsDb, clustersObject));
+        application.clusters().values().forEach(cluster -> toSlime(application, cluster, applicationNodes, metricsDb, clustersObject));
     }
 
-    private static void toSlime(Cluster cluster,
-                                List<Node> applicationNodes,
+    private static void toSlime(Application application,
+                                Cluster cluster,
+                                NodeList applicationNodes,
                                 MetricsDb metricsDb,
                                 Cursor clustersObject) {
-        NodeList nodes = NodeList.copyOf(applicationNodes).not().retired().cluster(cluster.id());
+        NodeList nodes = applicationNodes.not().retired().cluster(cluster.id());
         if (nodes.isEmpty()) return;
         ClusterResources currentResources = nodes.toResources();
+        Duration scalingDuration = cluster.scalingDuration(nodes.clusterSpec());
+        var clusterNodesTimeseries = new ClusterNodesTimeseries(Duration.ofHours(1), cluster, nodes, metricsDb);
+        var clusterTimeseries = metricsDb.getClusterTimeseries(application.id(), cluster.id());
 
         Cursor clusterObject = clustersObject.setObject(cluster.id().value());
+        clusterObject.setString("type", nodes.clusterSpec().type().name());
         toSlime(cluster.minResources(), clusterObject.setObject("min"));
         toSlime(cluster.maxResources(), clusterObject.setObject("max"));
         toSlime(currentResources, clusterObject.setObject("current"));
         if (cluster.shouldSuggestResources(currentResources))
             cluster.suggestedResources().ifPresent(suggested -> toSlime(suggested.resources(), clusterObject.setObject("suggested")));
         cluster.targetResources().ifPresent(target -> toSlime(target, clusterObject.setObject("target")));
-        clusterUtilizationToSlime(cluster, NodeList.copyOf(applicationNodes), metricsDb, clusterObject.setObject("utilization"));
+        clusterUtilizationToSlime(application, scalingDuration, clusterTimeseries, clusterNodesTimeseries, clusterObject.setObject("utilization"));
         scalingEventsToSlime(cluster.scalingEvents(), clusterObject.setArray("scalingEvents"));
         clusterObject.setString("autoscalingStatus", cluster.autoscalingStatus());
+        clusterObject.setLong("scalingDuration", scalingDuration.toMillis());
+        clusterObject.setDouble("maxQueryGrowthRate", clusterTimeseries.maxQueryGrowthRate());
+        clusterObject.setDouble("currentQueryFractionOfMax", clusterTimeseries.currentQueryFractionOfMax());
     }
 
     private static void toSlime(ClusterResources resources, Cursor clusterResourcesObject) {
@@ -72,12 +85,17 @@ public class ApplicationSerializer {
         NodeResourcesSerializer.toSlime(resources.nodeResources(), clusterResourcesObject.setObject("resources"));
     }
 
-    private static void clusterUtilizationToSlime(Cluster cluster, NodeList nodes, MetricsDb metricsDb, Cursor utilizationObject) {
-        var timeseries = new ClusterNodesTimeseries(Duration.ofHours(1), cluster, nodes, metricsDb);
-
-        utilizationObject.setDouble("cpu", timeseries.averageLoad(Resource.cpu));
-        utilizationObject.setDouble("memory", timeseries.averageLoad(Resource.memory));
-        utilizationObject.setDouble("disk", timeseries.averageLoad(Resource.disk));
+    private static void clusterUtilizationToSlime(Application application,
+                                                  Duration scalingDuration,
+                                                  ClusterTimeseries clusterTimeseries,
+                                                  ClusterNodesTimeseries clusterNodesTimeseries,
+                                                  Cursor utilizationObject) {
+        utilizationObject.setDouble("cpu", clusterNodesTimeseries.averageLoad(Resource.cpu));
+        utilizationObject.setDouble("idealCpu", ResourceTarget.idealCpuLoad(scalingDuration, clusterTimeseries, application));
+        utilizationObject.setDouble("memory", clusterNodesTimeseries.averageLoad(Resource.memory));
+        utilizationObject.setDouble("idealMemory", ResourceTarget.idealMemoryLoad());
+        utilizationObject.setDouble("disk", clusterNodesTimeseries.averageLoad(Resource.disk));
+        utilizationObject.setDouble("idealDisk", ResourceTarget.idealDiskLoad());
     }
 
     private static void scalingEventsToSlime(List<ScalingEvent> scalingEvents, Cursor scalingEventsArray) {
diff --git a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/restapi/NodesV2ApiHandler.java b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/restapi/NodesV2ApiHandler.java
index 62c7f40f7da..2442ff9d565 100644
--- a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/restapi/NodesV2ApiHandler.java
+++ b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/restapi/NodesV2ApiHandler.java
@@ -445,7 +445,7 @@ public class NodesV2ApiHandler extends LoggingRequestHandler {
         if (application.isEmpty())
             return ErrorResponse.notFoundError("No application '" + id + "'");
         Slime slime = ApplicationSerializer.toSlime(application.get(),
-                                                    nodeRepository.nodes().list(Node.State.active).owner(id).asList(),
+                                                    nodeRepository.nodes().list(Node.State.active).owner(id),
                                                     metricsDb,
                                                     withPath("/nodes/v2/applications/" + id, uri));
         return new SlimeJsonResponse(slime);
diff --git a/node-repository/src/test/java/com/yahoo/vespa/hosted/provision/restapi/responses/application1.json b/node-repository/src/test/java/com/yahoo/vespa/hosted/provision/restapi/responses/application1.json
index a35c742bc2a..1e9a2d60837 100644
--- a/node-repository/src/test/java/com/yahoo/vespa/hosted/provision/restapi/responses/application1.json
+++ b/node-repository/src/test/java/com/yahoo/vespa/hosted/provision/restapi/responses/application1.json
@@ -3,6 +3,7 @@
   "id" : "tenant1.application1.instance1",
   "clusters" : {
     "id1" : {
+      "type": "container",
       "min" : {
         "nodes" : 2,
         "groups" : 1,
@@ -65,8 +66,11 @@
       },
       "utilization" : {
           "cpu" : 0.0,
+          "idealCpu": 0.2,
           "memory" : 0.0,
-          "disk" : 0.0
+          "idealMemory": 0.7,
+          "disk" : 0.0,
+          "idealDisk": 0.6
       },
       "scalingEvents" : [
         {
@@ -97,7 +101,10 @@
           "at" : 123
         }
       ],
-      "autoscalingStatus" : ""
+      "autoscalingStatus": "",
+      "scalingDuration": 600000,
+      "maxQueryGrowthRate": 0.1,
+      "currentQueryFractionOfMax": 0.5
     }
   }
 }
diff --git a/node-repository/src/test/java/com/yahoo/vespa/hosted/provision/restapi/responses/application2.json b/node-repository/src/test/java/com/yahoo/vespa/hosted/provision/restapi/responses/application2.json
index baf2528c74a..376b748ff8e 100644
--- a/node-repository/src/test/java/com/yahoo/vespa/hosted/provision/restapi/responses/application2.json
+++ b/node-repository/src/test/java/com/yahoo/vespa/hosted/provision/restapi/responses/application2.json
@@ -3,6 +3,7 @@
   "id": "tenant2.application2.instance2",
   "clusters": {
     "id2": {
+      "type": "content",
       "min": {
         "nodes": 2,
         "groups": 1,
@@ -41,8 +42,11 @@
       },
       "utilization" : {
         "cpu" : 0.0,
+        "idealCpu": 0.19047619047619047,
         "memory" : 0.0,
-        "disk" : 0.0
+        "idealMemory": 0.7,
+        "disk" : 0.0,
+        "idealDisk": 0.6
       },
       "scalingEvents" : [
         {
@@ -73,7 +77,10 @@
           "at" : 123
         }
       ],
-      "autoscalingStatus" : ""
+      "autoscalingStatus" : "",
+      "scalingDuration": 43200000,
+      "maxQueryGrowthRate": 0.1,
+      "currentQueryFractionOfMax": 0.5
     }
   }
 }