Add host switch updater

6 files changed, 149 insertions, 3 deletions

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 82f37c9bc93..be4c889cd75 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
@@ -52,6 +52,7 @@ public class ControllerMaintenance extends AbstractComponent {
     private final ApplicationMetaDataGarbageCollector applicationMetaDataGarbageCollector;
     private final HostRepairMaintainer hostRepairMaintainer;
     private final ContainerImageExpirer containerImageExpirer;
+    private final HostSwitchUpdater hostSwitchUpdater;
 
     @Inject
     @SuppressWarnings("unused") // instantiated by Dependency Injection
@@ -81,6 +82,7 @@ public class ControllerMaintenance extends AbstractComponent {
         applicationMetaDataGarbageCollector = new ApplicationMetaDataGarbageCollector(controller, intervals.applicationMetaDataGarbageCollector);
         hostRepairMaintainer = new HostRepairMaintainer(controller, intervals.hostRepairMaintainer);
         containerImageExpirer = new ContainerImageExpirer(controller, intervals.containerImageExpirer);
+        hostSwitchUpdater = new HostSwitchUpdater(controller, intervals.hostSwitchUpdater);
     }
 
     public Upgrader upgrader() { return upgrader; }
@@ -111,6 +113,7 @@ public class ControllerMaintenance extends AbstractComponent {
         applicationMetaDataGarbageCollector.close();
         hostRepairMaintainer.close();
         containerImageExpirer.close();
+        hostSwitchUpdater.close();
     }
 
     /** Create one OS upgrader per cloud found in the zone registry of controller */
@@ -148,6 +151,7 @@ public class ControllerMaintenance extends AbstractComponent {
         private final Duration applicationMetaDataGarbageCollector;
         private final Duration hostRepairMaintainer;
         private final Duration containerImageExpirer;
+        private final Duration hostSwitchUpdater;
 
         public Intervals(SystemName system) {
             this.system = Objects.requireNonNull(system);
@@ -170,6 +174,7 @@ public class ControllerMaintenance extends AbstractComponent {
             this.applicationMetaDataGarbageCollector = duration(12, HOURS);
             this.hostRepairMaintainer = duration(12, HOURS);
             this.containerImageExpirer = duration(2, HOURS);
+            this.hostSwitchUpdater = duration(12, HOURS);
         }
 
         private Duration duration(long amount, TemporalUnit unit) {
diff --git a/controller-server/src/main/java/com/yahoo/vespa/hosted/controller/maintenance/HostSwitchUpdater.java b/controller-server/src/main/java/com/yahoo/vespa/hosted/controller/maintenance/HostSwitchUpdater.java
new file mode 100644
index 00000000000..35b1407441e
--- /dev/null
+++ b/controller-server/src/main/java/com/yahoo/vespa/hosted/controller/maintenance/HostSwitchUpdater.java
@@ -0,0 +1,52 @@
+// Copyright Verizon Media. Licensed under the terms of the Apache 2.0 license. See LICENSE in the project root.
+package com.yahoo.vespa.hosted.controller.maintenance;
+
+import com.yahoo.vespa.hosted.controller.Controller;
+import com.yahoo.vespa.hosted.controller.api.integration.configserver.Node;
+import com.yahoo.vespa.hosted.controller.api.integration.configserver.NodeRepository;
+import com.yahoo.vespa.hosted.controller.api.integration.entity.NodeEntity;
+import com.yahoo.vespa.hosted.controller.api.integration.noderepository.NodeRepositoryNode;
+
+import java.time.Duration;
+import java.util.Map;
+import java.util.function.Function;
+import java.util.stream.Collectors;
+
+/**
+ * Ensures that the switch information for all hosts is up to date.
+ *
+ * @author mpolden
+ */
+public class HostSwitchUpdater extends ControllerMaintainer {
+
+    private final NodeRepository nodeRepository;
+
+    public HostSwitchUpdater(Controller controller, Duration interval) {
+        super(controller, interval);
+        this.nodeRepository = controller.serviceRegistry().configServer().nodeRepository();
+    }
+
+    @Override
+    protected boolean maintain() {
+        Map<String, NodeEntity> nodeEntities = controller().serviceRegistry().entityService().listNodes().stream()
+                                                           .collect(Collectors.toMap(NodeEntity::hostname,
+                                                                                     Function.identity()));
+        for (var zone : controller().zoneRegistry().zones().controllerUpgraded().all().ids()) {
+            for (var node : nodeRepository.list(zone)) {
+                NodeEntity nodeEntity = nodeEntities.get(node.hostname().value());
+                if (!shouldUpdate(node, nodeEntity)) continue;
+
+                NodeRepositoryNode updatedNode = new NodeRepositoryNode();
+                updatedNode.setSwitchHostname(nodeEntity.switchHostname().orElse(null));
+                nodeRepository.patchNode(zone, node.hostname().value(), updatedNode);
+            }
+        }
+        return true;
+    }
+
+    private static boolean shouldUpdate(Node node, NodeEntity nodeEntity) {
+        if (nodeEntity == null) return false;
+        return !node.switchHostname().equals(nodeEntity.switchHostname());
+    }
+
+}
diff --git a/controller-server/src/test/java/com/yahoo/vespa/hosted/controller/integration/NodeRepositoryMock.java b/controller-server/src/test/java/com/yahoo/vespa/hosted/controller/integration/NodeRepositoryMock.java
index 61bd7858da2..ca478905893 100644
--- a/controller-server/src/test/java/com/yahoo/vespa/hosted/controller/integration/NodeRepositoryMock.java
+++ b/controller-server/src/test/java/com/yahoo/vespa/hosted/controller/integration/NodeRepositoryMock.java
@@ -41,6 +41,8 @@ public class NodeRepositoryMock implements NodeRepository {
     private final Map<ZoneId, TargetVersions> targetVersions = new HashMap<>();
     private final Map<Integer, Duration> osUpgradeBudgets = new HashMap<>();
 
+    private boolean allowPatching = false;
+
     /** Add or update given nodes in zone */
     public void putNodes(ZoneId zone, List<Node> nodes) {
         nodeRepository.putIfAbsent(zone, new HashMap<>());
@@ -230,7 +232,14 @@ public class NodeRepositoryMock implements NodeRepository {
 
     @Override
     public void patchNode(ZoneId zoneId, String hostName, NodeRepositoryNode node) {
-        throw new UnsupportedOperationException();
+        if (!allowPatching) throw new UnsupportedOperationException();
+        List<Node> existing = list(zoneId, List.of(HostName.from(hostName)));
+        if (existing.size() != 1) throw new IllegalArgumentException("Node " + hostName + " not found in " + zoneId);
+
+        // Note: Only supports switchHostname
+        Node newNode = new Node.Builder(existing.get(0)).switchHostname(node.getSwitchHostname())
+                                                        .build();
+        putNodes(zoneId, newNode);
     }
 
     @Override
@@ -280,4 +289,9 @@ public class NodeRepositoryMock implements NodeRepository {
         nodeRepository.get(zoneId).get(hostName).reports().put(reportId, report);
     }
 
+    public NodeRepositoryMock allowPatching(boolean allowPatching) {
+        this.allowPatching = allowPatching;
+        return this;
+    }
+
 }
diff --git a/controller-server/src/test/java/com/yahoo/vespa/hosted/controller/integration/ServiceRegistryMock.java b/controller-server/src/test/java/com/yahoo/vespa/hosted/controller/integration/ServiceRegistryMock.java
index ea31667d249..96eb4b39510 100644
--- a/controller-server/src/test/java/com/yahoo/vespa/hosted/controller/integration/ServiceRegistryMock.java
+++ b/controller-server/src/test/java/com/yahoo/vespa/hosted/controller/integration/ServiceRegistryMock.java
@@ -15,7 +15,6 @@ import com.yahoo.vespa.hosted.controller.api.integration.aws.ResourceTagger;
 import com.yahoo.vespa.hosted.controller.api.integration.billing.BillingController;
 import com.yahoo.vespa.hosted.controller.api.integration.billing.MockBillingController;
 import com.yahoo.vespa.hosted.controller.api.integration.certificates.EndpointCertificateMock;
-import com.yahoo.vespa.hosted.controller.api.integration.configserver.ConfigServer;
 import com.yahoo.vespa.hosted.controller.api.integration.dns.MemoryNameService;
 import com.yahoo.vespa.hosted.controller.api.integration.entity.MemoryEntityService;
 import com.yahoo.vespa.hosted.controller.api.integration.organization.MockContactRetriever;
@@ -81,7 +80,7 @@ public class ServiceRegistryMock extends AbstractComponent implements ServiceReg
     }
 
     @Override
-    public ConfigServer configServer() {
+    public ConfigServerMock configServer() {
         return configServerMock;
     }
 
diff --git a/controller-server/src/test/java/com/yahoo/vespa/hosted/controller/maintenance/HostSwitchUpdaterTest.java b/controller-server/src/test/java/com/yahoo/vespa/hosted/controller/maintenance/HostSwitchUpdaterTest.java
new file mode 100644
index 00000000000..1688141fe2d
--- /dev/null
+++ b/controller-server/src/test/java/com/yahoo/vespa/hosted/controller/maintenance/HostSwitchUpdaterTest.java
@@ -0,0 +1,73 @@
+// Copyright Verizon Media. Licensed under the terms of the Apache 2.0 license. See LICENSE in the project root.
+package com.yahoo.vespa.hosted.controller.maintenance;
+
+import com.yahoo.vespa.hosted.controller.ControllerTester;
+import com.yahoo.vespa.hosted.controller.api.integration.configserver.Node;
+import com.yahoo.vespa.hosted.controller.api.integration.entity.NodeEntity;
+import org.junit.Test;
+
+import java.time.Duration;
+import java.util.ArrayList;
+import java.util.List;
+
+import static org.junit.Assert.assertEquals;
+import static org.junit.Assert.assertFalse;
+
+/**
+ * @author mpolden
+ */
+public class HostSwitchUpdaterTest {
+
+    @Test
+    public void maintain() {
+        ControllerTester tester = new ControllerTester();
+        tester.serviceRegistry().configServer().nodeRepository().allowPatching(true);
+        addNodeEntities(tester);
+
+        // First iteration patches all hosts
+        HostSwitchUpdater maintainer = new HostSwitchUpdater(tester.controller(), Duration.ofDays(1));
+        maintainer.maintain();
+        List<Node> nodes = allNodes(tester);
+        assertFalse(nodes.isEmpty());
+        for (var node : nodes) {
+            assertEquals("Node " + node.hostname().value() + (node.type().isHost() ? " has" : " does not have")
+                         + " switch hostname", node.type().isHost(), node.switchHostname().isPresent());
+            if (node.type().isHost()) {
+                assertEquals("tor-" + node.hostname().value(), node.switchHostname().get());
+            }
+        }
+
+        // Second iteration does not patch anything as all switch information is current
+        tester.serviceRegistry().configServer().nodeRepository().allowPatching(false);
+        maintainer.maintain();
+
+        // One host is moved to a different switch
+        Node host = allNodes(tester).stream().filter(node -> node.type().isHost()).findFirst().get();
+        String newSwitch = "tor2-" + host.hostname().value();
+        NodeEntity nodeEntity = new NodeEntity(host.hostname().value(), "", "", newSwitch);
+        tester.serviceRegistry().entityService().addNodeEntity(nodeEntity);
+
+        // Host is updated
+        tester.serviceRegistry().configServer().nodeRepository().allowPatching(true);
+        maintainer.maintain();
+        Node updatedHost = allNodes(tester).stream().filter(node -> node.hostname().equals(host.hostname())).findFirst().get();
+        assertEquals(newSwitch, updatedHost.switchHostname().get());
+    }
+
+    private static List<Node> allNodes(ControllerTester tester) {
+        List<Node> nodes = new ArrayList<>();
+        for (var zone : tester.zoneRegistry().zones().controllerUpgraded().all().ids()) {
+            nodes.addAll(tester.serviceRegistry().configServer().nodeRepository().list(zone));
+        }
+        return nodes;
+    }
+
+    private static void addNodeEntities(ControllerTester tester) {
+        for (var node : allNodes(tester)) {
+            if (!node.type().isHost()) continue;
+            NodeEntity nodeEntity = new NodeEntity(node.hostname().value(), "", "", "tor-" + node.hostname().value());
+            tester.serviceRegistry().entityService().addNodeEntity(nodeEntity);
+        }
+    }
+
+}
diff --git a/controller-server/src/test/java/com/yahoo/vespa/hosted/controller/restapi/controller/responses/maintenance.json b/controller-server/src/test/java/com/yahoo/vespa/hosted/controller/restapi/controller/responses/maintenance.json
index c1ee1489cd4..5ab087aab17 100644
--- a/controller-server/src/test/java/com/yahoo/vespa/hosted/controller/restapi/controller/responses/maintenance.json
+++ b/controller-server/src/test/java/com/yahoo/vespa/hosted/controller/restapi/controller/responses/maintenance.json
@@ -34,6 +34,9 @@
       "name": "HostRepairMaintainer"
     },
     {
+      "name": "HostSwitchUpdater"
+    },
+    {
       "name": "JobRunner"
     },
     {