Maintain deploymentInfo

7 files changed, 75 insertions, 7 deletions

diff --git a/controller-server/src/main/java/com/yahoo/vespa/hosted/controller/ApplicationController.java b/controller-server/src/main/java/com/yahoo/vespa/hosted/controller/ApplicationController.java
index d053106fcee..a4bded314d9 100644
--- a/controller-server/src/main/java/com/yahoo/vespa/hosted/controller/ApplicationController.java
+++ b/controller-server/src/main/java/com/yahoo/vespa/hosted/controller/ApplicationController.java
@@ -93,6 +93,7 @@ import java.util.Optional;
 import java.util.OptionalInt;
 import java.util.Set;
 import java.util.TreeMap;
+import java.util.concurrent.ConcurrentHashMap;
 import java.util.concurrent.atomic.AtomicReference;
 import java.util.function.Consumer;
 import java.util.function.Predicate;
@@ -144,6 +145,7 @@ public class ApplicationController {
     private final ListFlag<String> incompatibleVersions;
     private final BillingController billingController;
     private final ListFlag<String> cloudAccountsFlag;
+    private final Map<DeploymentId, com.yahoo.vespa.hosted.controller.api.integration.configserver.Application> deploymentInfo = new ConcurrentHashMap<>();
 
     ApplicationController(Controller controller, CuratorDb curator, AccessControl accessControl, Clock clock,
                           FlagSource flagSource, BillingController billingController) {
@@ -206,6 +208,15 @@ public class ApplicationController {
     }
 
     /**
+     * Returns in-memory info for the given deployment pulled from the node repo.
+     * Info on any existing deployment can be missing if it has not yet been fetched since this instance was started.
+     * This is kept up to date by DeploymentInfoMaintainer.
+     * Accessing this is thread safe.
+     */
+    // TODO: Replace the wire level Application by a DeploymentInfo class in the model
+    public Map<DeploymentId, com.yahoo.vespa.hosted.controller.api.integration.configserver.Application> deploymentInfo() { return deploymentInfo; }
+
+    /**
      * Triggers reindexing for the given document types in the given clusters, for the given application.
      * <p>
      * If no clusters are given, reindexing is triggered for the entire application; otherwise
diff --git a/controller-server/src/main/java/com/yahoo/vespa/hosted/controller/maintenance/BcpGroupUpdater.java b/controller-server/src/main/java/com/yahoo/vespa/hosted/controller/maintenance/BcpGroupUpdater.java
index 1ef94ce527c..dc04a81fe4a 100644
--- a/controller-server/src/main/java/com/yahoo/vespa/hosted/controller/maintenance/BcpGroupUpdater.java
+++ b/controller-server/src/main/java/com/yahoo/vespa/hosted/controller/maintenance/BcpGroupUpdater.java
@@ -81,7 +81,7 @@ public class BcpGroupUpdater extends ControllerMaintainer {
         if ( successFactor == 0 )
             log.log(Level.WARNING, "Could not update traffic share on any applications", lastException);
         else if ( successFactor < 0.9 )
-            log.log(Level.FINE, "Could not update traffic share on any applications", lastException);
+            log.log(Level.FINE, "Could not update traffic share on all applications", lastException);
         return successFactor;
     }
 
diff --git a/controller-server/src/main/java/com/yahoo/vespa/hosted/controller/maintenance/ChangeManagementAssessor.java b/controller-server/src/main/java/com/yahoo/vespa/hosted/controller/maintenance/ChangeManagementAssessor.java
index fa44b1480e3..5be20f9a994 100644
--- a/controller-server/src/main/java/com/yahoo/vespa/hosted/controller/maintenance/ChangeManagementAssessor.java
+++ b/controller-server/src/main/java/com/yahoo/vespa/hosted/controller/maintenance/ChangeManagementAssessor.java
@@ -30,7 +30,6 @@ public class ChangeManagementAssessor {
     }
 
     Assessment assessmentInner(List<String> impactedHostnames, List<Node> allNodes, ZoneId zone) {
-
         List<String> impactedParentHosts = toParentHosts(impactedHostnames, allNodes);
         // Group impacted application nodes by parent host
         Map<Node, List<Node>> prParentHost = allNodes.stream()
@@ -53,10 +52,7 @@ public class ChangeManagementAssessor {
                 .map(node -> node.hostname())
                 .toList();
 
-        boolean allHostsReplacable = tenantHosts.isEmpty() || nodeRepository.isReplaceable(
-                zone,
-                tenantHosts
-        );
+        boolean allHostsReplacable = tenantHosts.isEmpty() || nodeRepository.isReplaceable(zone, tenantHosts);
 
         // Report assessment pr cluster
         var clusterAssessments = prCluster.entrySet().stream().map((entry) -> {
diff --git a/controller-server/src/main/java/com/yahoo/vespa/hosted/controller/maintenance/ControllerMaintenance.java b/controller-server/src/main/java/com/yahoo/vespa/hosted/controller/maintenance/ControllerMaintenance.java
index b64de5d5af4..bcc8296da2a 100644
--- a/controller-server/src/main/java/com/yahoo/vespa/hosted/controller/maintenance/ControllerMaintenance.java
+++ b/controller-server/src/main/java/com/yahoo/vespa/hosted/controller/maintenance/ControllerMaintenance.java
@@ -46,6 +46,7 @@ public class ControllerMaintenance extends AbstractComponent {
         maintainers.add(osUpgradeScheduler);
         maintainers.addAll(osUpgraders(controller, intervals.osUpgrader));
         maintainers.add(new DeploymentExpirer(controller, intervals.defaultInterval));
+        maintainers.add(new DeploymentInfoMaintainer(controller, intervals.deploymentInfoMaintainer));
         maintainers.add(new DeploymentUpgrader(controller, intervals.defaultInterval));
         maintainers.add(new DeploymentIssueReporter(controller, controller.serviceRegistry().deploymentIssues(), intervals.defaultInterval));
         maintainers.add(new MetricsReporter(controller, metric, athenzClientFactory.createZmsClient()));
@@ -109,6 +110,7 @@ public class ControllerMaintenance extends AbstractComponent {
         private final SystemName system;
 
         private final Duration defaultInterval;
+        private final Duration deploymentInfoMaintainer;
         private final Duration outstandingChangeDeployer;
         private final Duration versionStatusUpdater;
         private final Duration readyJobsTrigger;
@@ -143,6 +145,7 @@ public class ControllerMaintenance extends AbstractComponent {
         public Intervals(SystemName system) {
             this.system = Objects.requireNonNull(system);
             this.defaultInterval = duration(system.isCd() ? 1 : 5, MINUTES);
+            this.deploymentInfoMaintainer = duration(system.isCd() ? 1 : 10, MINUTES);
             this.outstandingChangeDeployer = duration(3, MINUTES);
             this.versionStatusUpdater = duration(3, MINUTES);
             this.readyJobsTrigger = duration(1, MINUTES);
diff --git a/controller-server/src/main/java/com/yahoo/vespa/hosted/controller/maintenance/DeploymentInfoMaintainer.java b/controller-server/src/main/java/com/yahoo/vespa/hosted/controller/maintenance/DeploymentInfoMaintainer.java
new file mode 100644
index 00000000000..cf9db1517a0
--- /dev/null
+++ b/controller-server/src/main/java/com/yahoo/vespa/hosted/controller/maintenance/DeploymentInfoMaintainer.java
@@ -0,0 +1,57 @@
+package com.yahoo.vespa.hosted.controller.maintenance;
+
+import com.yahoo.vespa.hosted.controller.Application;
+import com.yahoo.vespa.hosted.controller.Controller;
+import com.yahoo.vespa.hosted.controller.Instance;
+import com.yahoo.vespa.hosted.controller.api.identifiers.DeploymentId;
+import com.yahoo.vespa.hosted.controller.api.integration.configserver.ConfigServerException;
+import com.yahoo.vespa.hosted.controller.api.integration.configserver.NodeRepository;
+import com.yahoo.yolean.Exceptions;
+
+import java.time.Duration;
+import java.util.stream.Stream;
+
+/**
+ * This pulls application deployment information from the node repo on all config servers,
+ * and stores it in memory in controller.applications().deploymentInfo().
+ *
+ * @author bratseth
+ */
+public class DeploymentInfoMaintainer extends ControllerMaintainer {
+
+    private final NodeRepository nodeRepository;
+
+    public DeploymentInfoMaintainer(Controller controller, Duration duration) {
+        super(controller, duration);
+        this.nodeRepository = controller.serviceRegistry().configServer().nodeRepository();
+    }
+
+    @Override
+    protected double maintain() {
+        controller().applications().asList().stream()
+                                   .flatMap(this::mapApplicationToInstances)
+                                   .flatMap(this::mapInstanceToDeployments)
+                                   .forEach(this::updateDeploymentInfo);
+        return 1.0;
+    }
+
+    private Stream<Instance> mapApplicationToInstances(Application application) {
+        return application.instances().values().stream();
+    }
+
+    private Stream<DeploymentId> mapInstanceToDeployments(Instance instance) {
+        return instance.deployments().keySet().stream()
+                       .filter(zoneId -> !zoneId.environment().isTest())
+                       .map(zoneId -> new DeploymentId(instance.id(), zoneId));
+    }
+
+    private void updateDeploymentInfo(DeploymentId id) {
+        try {
+            controller().applications().deploymentInfo().put(id, nodeRepository.getApplication(id.zoneId(), id.applicationId()));
+        }
+        catch (ConfigServerException e) {
+            log.info("Could not retrieve deployment info for " + id + ": " + Exceptions.toMessageString(e));
+        }
+    }
+
+}
diff --git a/controller-server/src/main/java/com/yahoo/vespa/hosted/controller/maintenance/ResourceMeterMaintainer.java b/controller-server/src/main/java/com/yahoo/vespa/hosted/controller/maintenance/ResourceMeterMaintainer.java
index 44081081f73..d47b82a231f 100644
--- a/controller-server/src/main/java/com/yahoo/vespa/hosted/controller/maintenance/ResourceMeterMaintainer.java
+++ b/controller-server/src/main/java/com/yahoo/vespa/hosted/controller/maintenance/ResourceMeterMaintainer.java
@@ -168,6 +168,7 @@ public class ResourceMeterMaintainer extends ControllerMaintainer {
 
     private Stream<Map.Entry<ClusterId, List<Cluster.ScalingEvent>>> mapDeploymentToClusterScalingEvent(DeploymentId deploymentId) {
         try {
+            // TODO: get Application from controller.applications().deploymentInfo()
             return nodeRepository.getApplication(deploymentId.zoneId(), deploymentId.applicationId())
                     .clusters().entrySet().stream()
                     .map(cluster -> Map.entry(new ClusterId(deploymentId, cluster.getKey()), cluster.getValue().scalingEvents()));
diff --git a/controller-server/src/main/java/com/yahoo/vespa/hosted/controller/maintenance/VcmrMaintainer.java b/controller-server/src/main/java/com/yahoo/vespa/hosted/controller/maintenance/VcmrMaintainer.java
index 9211580abf1..45b3e4ef5dd 100644
--- a/controller-server/src/main/java/com/yahoo/vespa/hosted/controller/maintenance/VcmrMaintainer.java
+++ b/controller-server/src/main/java/com/yahoo/vespa/hosted/controller/maintenance/VcmrMaintainer.java
@@ -38,7 +38,7 @@ import java.util.stream.Stream;
  *
  * Maintains status and execution of Vespa CMRs.
  *
- * Currently this retires all affected tenant hosts if zone capacity allows it.
+ * Currently, this retires all affected tenant hosts if zone capacity allows it.
  *
  * @author olaa
  */