diff options
author | Valerij Fredriksen <freva@users.noreply.github.com> | 2022-10-13 21:10:41 +0200 |
---|---|---|
committer | GitHub <noreply@github.com> | 2022-10-13 21:10:41 +0200 |
commit | a5ed12b351806b187613457b58982ca67f537594 (patch) | |
tree | 84e4beabd59ca6a1440607b5ba28468e92a83e48 /node-repository | |
parent | 6e7459c9eff635d9c8227cd8d1add320f298c0e2 (diff) |
Revert "Remove HostLivenessTracker"
Diffstat (limited to 'node-repository')
10 files changed, 307 insertions, 69 deletions
diff --git a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/maintenance/NodeFailer.java b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/maintenance/NodeFailer.java index 32eac49a288..3e7abe8f053 100644 --- a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/maintenance/NodeFailer.java +++ b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/maintenance/NodeFailer.java @@ -77,6 +77,23 @@ public class NodeFailer extends NodeRepositoryMaintainer { int throttledHostFailures = 0; int throttledNodeFailures = 0; + // Ready nodes + try (Mutex lock = nodeRepository().nodes().lockUnallocated()) { + for (FailingNode failing : findReadyFailingNodes()) { + attempts++; + if (throttle(failing.node())) { + failures++; + if (failing.node().type().isHost()) + throttledHostFailures++; + else + throttledNodeFailures++; + continue; + } + nodeRepository().nodes().fail(failing.node().hostname(), Agent.NodeFailer, failing.reason()); + } + } + + // Active nodes for (FailingNode failing : findActiveFailingNodes()) { attempts++; if (!failAllowedFor(failing.node().type())) continue; @@ -99,6 +116,22 @@ public class NodeFailer extends NodeRepositoryMaintainer { return asSuccessFactor(attempts, failures); } + private Collection<FailingNode> findReadyFailingNodes() { + Set<FailingNode> failingNodes = new HashSet<>(); + for (Node node : nodeRepository().nodes().list(Node.State.ready)) { + Node hostNode = node.parentHostname().flatMap(parent -> nodeRepository().nodes().node(parent)).orElse(node); + List<String> failureReports = reasonsToFailHost(hostNode); + if (failureReports.size() > 0) { + if (hostNode.equals(node)) { + failingNodes.add(new FailingNode(node, "Host has failure reports: " + failureReports)); + } else { + failingNodes.add(new FailingNode(node, "Parent (" + hostNode + ") has failure reports: " + failureReports)); + } + } + } + return failingNodes; + } + private Collection<FailingNode> findActiveFailingNodes() { Set<FailingNode> failingNodes = new HashSet<>(); NodeList activeNodes = nodeRepository().nodes().list(Node.State.active); @@ -117,7 +150,7 @@ public class NodeFailer extends NodeRepositoryMaintainer { for (Node node : activeNodes) { if (allSuspended(node, activeNodes)) { - Node host = node.parentHostname().flatMap(activeNodes::node).orElse(node); + Node host = node.parentHostname().flatMap(parent -> activeNodes.node(parent)).orElse(node); if (host.type().isHost()) { List<String> failureReports = reasonsToFailHost(host); if ( ! failureReports.isEmpty()) { @@ -142,7 +175,7 @@ public class NodeFailer extends NodeRepositoryMaintainer { /** Returns whether node has any kind of hardware issue */ static boolean hasHardwareIssue(Node node, NodeList allNodes) { - Node host = node.parentHostname().flatMap(allNodes::node).orElse(node); + Node host = node.parentHostname().flatMap(parent -> allNodes.node(parent)).orElse(node); return reasonsToFailHost(host).size() > 0; } @@ -311,6 +344,30 @@ public class NodeFailer extends NodeRepositoryMaintainer { } - private record FailingNode(Node node, String reason) { } + private static class FailingNode { + + private final Node node; + private final String reason; + + public FailingNode(Node node, String reason) { + this.node = node; + this.reason = reason; + } + + public Node node() { return node; } + public String reason() { return reason; } + + @Override + public boolean equals(Object other) { + if ( ! (other instanceof FailingNode)) return false; + return ((FailingNode)other).node().equals(this.node()); + } + + @Override + public int hashCode() { + return node.hashCode(); + } + + } } diff --git a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/maintenance/NodeHealthTracker.java b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/maintenance/NodeHealthTracker.java index 624492a14f3..b43e2ae051f 100644 --- a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/maintenance/NodeHealthTracker.java +++ b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/maintenance/NodeHealthTracker.java @@ -3,6 +3,7 @@ package com.yahoo.vespa.hosted.provision.maintenance; import com.yahoo.config.provision.ApplicationId; import com.yahoo.config.provision.ApplicationLockException; +import com.yahoo.config.provision.HostLivenessTracker; import com.yahoo.jdisc.Metric; import com.yahoo.lang.MutableInteger; import com.yahoo.transaction.Mutex; @@ -12,10 +13,12 @@ import com.yahoo.vespa.hosted.provision.Node; import com.yahoo.vespa.hosted.provision.NodeList; import com.yahoo.vespa.hosted.provision.NodeRepository; import com.yahoo.vespa.hosted.provision.node.Agent; +import com.yahoo.vespa.hosted.provision.node.History; import com.yahoo.vespa.service.monitor.ServiceMonitor; import com.yahoo.yolean.Exceptions; import java.time.Duration; +import java.time.Instant; import java.util.List; import java.util.Map; import java.util.Optional; @@ -31,18 +34,41 @@ import static java.util.stream.Collectors.counting; */ public class NodeHealthTracker extends NodeRepositoryMaintainer { + /** Provides information about the status of ready hosts */ + private final HostLivenessTracker hostLivenessTracker; + /** Provides (more accurate) information about the status of active hosts */ private final ServiceMonitor serviceMonitor; - public NodeHealthTracker(ServiceMonitor serviceMonitor, NodeRepository nodeRepository, + public NodeHealthTracker(HostLivenessTracker hostLivenessTracker, + ServiceMonitor serviceMonitor, NodeRepository nodeRepository, Duration interval, Metric metric) { super(nodeRepository, interval, metric); + this.hostLivenessTracker = hostLivenessTracker; this.serviceMonitor = serviceMonitor; } @Override protected double maintain() { - return updateActiveNodeDownState(); + return ( updateReadyNodeLivenessEvents() + updateActiveNodeDownState() ) / 2; + } + + private double updateReadyNodeLivenessEvents() { + // Update node last request events through ZooKeeper to collect request to all config servers. + // We do this here ("lazily") to avoid writing to zk for each config request. + try (Mutex lock = nodeRepository().nodes().lockUnallocated()) { + for (Node node : nodeRepository().nodes().list(Node.State.ready)) { + Optional<Instant> lastLocalRequest = hostLivenessTracker.lastRequestFrom(node.hostname()); + if (lastLocalRequest.isEmpty()) continue; + + if (!node.history().hasEventAfter(History.Event.Type.requested, lastLocalRequest.get())) { + History updatedHistory = node.history() + .with(new History.Event(History.Event.Type.requested, Agent.NodeHealthTracker, lastLocalRequest.get())); + nodeRepository().nodes().write(node.with(updatedHistory), lock); + } + } + } + return 1.0; } /** diff --git a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/maintenance/NodeRepositoryMaintenance.java b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/maintenance/NodeRepositoryMaintenance.java index 708d8b59eb0..dac6ee61ef3 100644 --- a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/maintenance/NodeRepositoryMaintenance.java +++ b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/maintenance/NodeRepositoryMaintenance.java @@ -5,6 +5,7 @@ import com.yahoo.component.AbstractComponent; import com.yahoo.component.annotation.Inject; import com.yahoo.concurrent.maintenance.Maintainer; import com.yahoo.config.provision.Deployer; +import com.yahoo.config.provision.HostLivenessTracker; import com.yahoo.config.provision.InfraDeployer; import com.yahoo.config.provision.NodeType; import com.yahoo.config.provision.Zone; @@ -32,7 +33,7 @@ public class NodeRepositoryMaintenance extends AbstractComponent { @SuppressWarnings("unused") @Inject public NodeRepositoryMaintenance(NodeRepository nodeRepository, Deployer deployer, InfraDeployer infraDeployer, - ServiceMonitor serviceMonitor, + HostLivenessTracker hostLivenessTracker, ServiceMonitor serviceMonitor, Zone zone, Metric metric, ProvisionServiceProvider provisionServiceProvider, FlagSource flagSource, MetricsFetcher metricsFetcher) { @@ -45,7 +46,7 @@ public class NodeRepositoryMaintenance extends AbstractComponent { maintainers.add(infrastructureProvisioner); maintainers.add(new NodeFailer(deployer, nodeRepository, defaults.failGrace, defaults.nodeFailerInterval, defaults.throttlePolicy, metric)); - maintainers.add(new NodeHealthTracker(serviceMonitor, nodeRepository, defaults.nodeFailureStatusUpdateInterval, metric)); + maintainers.add(new NodeHealthTracker(hostLivenessTracker, serviceMonitor, nodeRepository, defaults.nodeFailureStatusUpdateInterval, metric)); maintainers.add(new ExpeditedChangeApplicationMaintainer(deployer, metric, nodeRepository, defaults.expeditedChangeRedeployInterval)); maintainers.add(new ReservationExpirer(nodeRepository, defaults.reservationExpiry, metric)); maintainers.add(new RetiredExpirer(nodeRepository, deployer, metric, defaults.retiredInterval, defaults.retiredExpiry)); diff --git a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/node/History.java b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/node/History.java index 4aca1cbd056..14fa2e1c8ff 100644 --- a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/node/History.java +++ b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/node/History.java @@ -46,10 +46,8 @@ public class History { private static ImmutableMap<Event.Type, Event> toImmutableMap(Collection<Event> events) { ImmutableMap.Builder<Event.Type, Event> builder = new ImmutableMap.Builder<>(); - for (Event event : events) { - if (event.type() == Event.Type.requested) continue; // TODO (freva): Remove requested event after 8.70 + for (Event event : events) builder.put(event.type(), event); - } return builder.build(); } diff --git a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/persistence/CuratorDatabaseClient.java b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/persistence/CuratorDatabaseClient.java index a5ebc6b3efc..b4e304155a6 100644 --- a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/persistence/CuratorDatabaseClient.java +++ b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/persistence/CuratorDatabaseClient.java @@ -1,7 +1,6 @@ // Copyright Yahoo. Licensed under the terms of the Apache 2.0 license. See LICENSE in the project root. package com.yahoo.vespa.hosted.provision.persistence; -import ai.vespa.http.DomainName; import com.yahoo.component.Version; import com.yahoo.concurrent.UncheckedTimeoutException; import com.yahoo.config.provision.ApplicationId; @@ -485,12 +484,15 @@ public class CuratorDatabaseClient { transaction.onCommitted(() -> { for (var lb : loadBalancers) { if (lb.state() == fromState) continue; - Optional<String> target = lb.instance().flatMap(instance -> instance.hostname().map(DomainName::value).or(instance::ipAddress)); if (fromState == null) { - log.log(Level.INFO, () -> "Creating " + lb.id() + target.map(t -> " (" + t + ")").orElse("") + + log.log(Level.INFO, () -> "Creating " + lb.id() + lb.instance() + .map(instance -> " (" + instance.hostname() + ")") + .orElse("") + " in " + lb.state()); } else { - log.log(Level.INFO, () -> "Moving " + lb.id() + target.map(t -> " (" + t + ")").orElse("") + + log.log(Level.INFO, () -> "Moving " + lb.id() + lb.instance() + .map(instance -> " (" + instance.hostname() + ")") + .orElse("") + " from " + fromState + " to " + lb.state()); } diff --git a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/testutils/ContainerConfig.java b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/testutils/ContainerConfig.java index 024f071abb1..83a80847ec8 100644 --- a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/testutils/ContainerConfig.java +++ b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/testutils/ContainerConfig.java @@ -21,6 +21,7 @@ public class ContainerConfig { " <component id='com.yahoo.vespa.hosted.provision.testutils.MockDeployer'/>\n" + " <component id='com.yahoo.vespa.hosted.provision.testutils.MockInfraDeployer'/>\n" + " <component id='com.yahoo.vespa.hosted.provision.testutils.MockProvisioner'/>\n" + + " <component id='com.yahoo.vespa.hosted.provision.testutils.TestHostLivenessTracker'/>\n" + " <component id='com.yahoo.vespa.hosted.provision.testutils.ServiceMonitorStub'/>\n" + " <component id='com.yahoo.vespa.hosted.provision.testutils.MockDuperModel'/>\n" + " <component id='com.yahoo.vespa.hosted.provision.testutils.MockNodeFlavors'/>\n" + diff --git a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/testutils/TestHostLivenessTracker.java b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/testutils/TestHostLivenessTracker.java new file mode 100644 index 00000000000..28d0d5f89d7 --- /dev/null +++ b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/testutils/TestHostLivenessTracker.java @@ -0,0 +1,32 @@ +// Copyright Yahoo. Licensed under the terms of the Apache 2.0 license. See LICENSE in the project root. +package com.yahoo.vespa.hosted.provision.testutils; + +import com.yahoo.config.provision.HostLivenessTracker; + +import java.time.Clock; +import java.time.Instant; +import java.util.HashMap; +import java.util.Map; +import java.util.Optional; + +/** This is a fully functional implementation */ +public class TestHostLivenessTracker implements HostLivenessTracker { + + private final Clock clock; + private final Map<String, Instant> lastRequestFromHost = new HashMap<>(); + + public TestHostLivenessTracker(Clock clock) { + this.clock = clock; + } + + @Override + public void receivedRequestFrom(String hostname) { + lastRequestFromHost.put(hostname, clock.instant()); + } + + @Override + public Optional<Instant> lastRequestFrom(String hostname) { + return Optional.ofNullable(lastRequestFromHost.get(hostname)); + } + +} diff --git a/node-repository/src/test/java/com/yahoo/vespa/hosted/provision/maintenance/NodeFailTester.java b/node-repository/src/test/java/com/yahoo/vespa/hosted/provision/maintenance/NodeFailTester.java index 977d72c11ea..f67e9cd8345 100644 --- a/node-repository/src/test/java/com/yahoo/vespa/hosted/provision/maintenance/NodeFailTester.java +++ b/node-repository/src/test/java/com/yahoo/vespa/hosted/provision/maintenance/NodeFailTester.java @@ -27,6 +27,7 @@ import com.yahoo.vespa.hosted.provision.provisioning.NodeRepositoryProvisioner; import com.yahoo.vespa.hosted.provision.provisioning.ProvisioningTester; import com.yahoo.vespa.hosted.provision.testutils.MockDeployer; import com.yahoo.vespa.hosted.provision.testutils.ServiceMonitorStub; +import com.yahoo.vespa.hosted.provision.testutils.TestHostLivenessTracker; import java.time.Clock; import java.time.Duration; @@ -62,6 +63,7 @@ public class NodeFailTester { public ServiceMonitorStub serviceMonitor; public MockDeployer deployer; public TestMetric metric; + private final TestHostLivenessTracker hostLivenessTracker; private final NodeRepositoryProvisioner provisioner; private final Curator curator; @@ -72,6 +74,7 @@ public class NodeFailTester { curator = tester.getCurator(); nodeRepository = tester.nodeRepository(); provisioner = tester.provisioner(); + hostLivenessTracker = new TestHostLivenessTracker(clock); } private void initializeMaintainers(Map<ApplicationId, MockDeployer.ApplicationContext> apps) { @@ -109,7 +112,7 @@ public class NodeFailTester { /** Create hostCount hosts, one app with containerCount containers, and one app with contentCount content nodes. */ public static NodeFailTester withTwoApplications(int hostCount, int containerCount, int contentCount) { NodeFailTester tester = new NodeFailTester(); - tester.tester.makeReadyHosts(hostCount, new NodeResources(2, 8, 20, 10)); + tester.createHostNodes(hostCount); // Create tenant host application ClusterSpec clusterNodeAdminApp = ClusterSpec.request(ClusterSpec.Type.container, ClusterSpec.Id.from("node-admin")).vespaVersion("6.42").build(); @@ -136,7 +139,13 @@ public class NodeFailTester { public static NodeFailTester withTwoApplications(int numberOfHosts) { NodeFailTester tester = new NodeFailTester(); - tester.tester.makeReadyNodes(numberOfHosts, new NodeResources(4, 16, 400, 10), NodeType.host, 8); + + int nodesPerHost = 3; + List<Node> hosts = tester.createHostNodes(numberOfHosts); + for (int i = 0; i < hosts.size(); i++) { + tester.createReadyNodes(nodesPerHost, i * nodesPerHost, Optional.of("parent" + (i + 1)), + new NodeResources(1, 4, 100, 0.3), NodeType.tenant); + } // Create applications ClusterSpec clusterNodeAdminApp = ClusterSpec.request(ClusterSpec.Type.container, ClusterSpec.Id.from("node-admin")).vespaVersion("6.42").build(); @@ -221,11 +230,26 @@ public class NodeFailTester { } public NodeHealthTracker createUpdater() { - return new NodeHealthTracker(serviceMonitor, nodeRepository, Duration.ofMinutes(5), metric); + return new NodeHealthTracker(hostLivenessTracker, serviceMonitor, nodeRepository, Duration.ofMinutes(5), metric); + } + + public void allNodesMakeAConfigRequestExcept(Node ... deadNodeArray) { + allNodesMakeAConfigRequestExcept(List.of(deadNodeArray)); + } + + public void allNodesMakeAConfigRequestExcept(List<Node> deadNodes) { + for (Node node : nodeRepository.nodes().list()) { + if ( ! deadNodes.contains(node)) + hostLivenessTracker.receivedRequestFrom(node.hostname()); + } } public Clock clock() { return clock; } + public List<Node> createReadyNodes(int count) { + return createReadyNodes(count, 0); + } + public List<Node> createReadyNodes(int count, NodeResources resources) { return createReadyNodes(count, 0, resources); } @@ -234,10 +258,22 @@ public class NodeFailTester { return createReadyNodes(count, 0, Optional.empty(), hostFlavors.getFlavorOrThrow("default"), nodeType); } + public List<Node> createReadyNodes(int count, int startIndex) { + return createReadyNodes(count, startIndex, "default"); + } + + public List<Node> createReadyNodes(int count, int startIndex, String flavor) { + return createReadyNodes(count, startIndex, Optional.empty(), hostFlavors.getFlavorOrThrow(flavor), NodeType.tenant); + } + public List<Node> createReadyNodes(int count, int startIndex, NodeResources resources) { return createReadyNodes(count, startIndex, Optional.empty(), new Flavor(resources), NodeType.tenant); } + private List<Node> createReadyNodes(int count, int startIndex, Optional<String> parentHostname, NodeResources resources, NodeType nodeType) { + return createReadyNodes(count, startIndex, parentHostname, new Flavor(resources), nodeType); + } + private List<Node> createReadyNodes(int count, int startIndex, Optional<String> parentHostname, Flavor flavor, NodeType nodeType) { List<Node> nodes = new ArrayList<>(count); int lastOctetOfPoolAddress = 0; diff --git a/node-repository/src/test/java/com/yahoo/vespa/hosted/provision/maintenance/NodeFailerTest.java b/node-repository/src/test/java/com/yahoo/vespa/hosted/provision/maintenance/NodeFailerTest.java index a21edf31cb8..3ba536ee4d7 100644 --- a/node-repository/src/test/java/com/yahoo/vespa/hosted/provision/maintenance/NodeFailerTest.java +++ b/node-repository/src/test/java/com/yahoo/vespa/hosted/provision/maintenance/NodeFailerTest.java @@ -13,6 +13,7 @@ import com.yahoo.vespa.hosted.provision.NodeList; import com.yahoo.vespa.hosted.provision.NodeRepository; import com.yahoo.vespa.hosted.provision.node.Agent; import com.yahoo.vespa.hosted.provision.node.Report; +import com.yahoo.vespa.hosted.provision.node.Reports; import org.junit.Test; import java.time.Duration; @@ -20,7 +21,6 @@ import java.time.Instant; import java.util.ArrayList; import java.util.Arrays; import java.util.Collections; -import java.util.Comparator; import java.util.List; import java.util.Map; import java.util.Set; @@ -62,21 +62,23 @@ public class NodeFailerTest { } private void testNodeFailingWith(NodeFailTester tester, String hostWithHwFailure) { - // The host should have 2 nodes in active + // The host should have 2 nodes in active and 1 ready Map<Node.State, List<String>> hostnamesByState = tester.nodeRepository.nodes().list().childrenOf(hostWithHwFailure).asList().stream() .collect(Collectors.groupingBy(Node::state, Collectors.mapping(Node::hostname, Collectors.toList()))); assertEquals(2, hostnamesByState.get(Node.State.active).size()); + assertEquals(1, hostnamesByState.get(Node.State.ready).size()); // Suspend the first of the active nodes tester.suspend(hostnamesByState.get(Node.State.active).get(0)); tester.runMaintainers(); tester.clock.advance(Duration.ofHours(25)); + tester.allNodesMakeAConfigRequestExcept(); tester.runMaintainers(); // The first (and the only) ready node and the 1st active node that was allowed to fail should be failed Map<Node.State, List<String>> expectedHostnamesByState1Iter = Map.of( - Node.State.failed, List.of(hostnamesByState.get(Node.State.active).get(0)), + Node.State.failed, List.of(hostnamesByState.get(Node.State.ready).get(0), hostnamesByState.get(Node.State.active).get(0)), Node.State.active, hostnamesByState.get(Node.State.active).subList(1, 2)); Map<Node.State, List<String>> hostnamesByState1Iter = tester.nodeRepository.nodes().list().childrenOf(hostWithHwFailure).asList().stream() .collect(Collectors.groupingBy(Node::state, Collectors.mapping(Node::hostname, Collectors.toList()))); @@ -86,6 +88,7 @@ public class NodeFailerTest { tester.suspend(hostnamesByState.get(Node.State.active).get(1)); tester.clock.advance(Duration.ofHours(25)); + tester.allNodesMakeAConfigRequestExcept(); tester.runMaintainers(); // All of the children should be failed now @@ -98,7 +101,7 @@ public class NodeFailerTest { tester.suspend(hostWithHwFailure); tester.runMaintainers(); assertEquals(Node.State.failed, tester.nodeRepository.nodes().node(hostWithHwFailure).get().state()); - assertEquals(3, tester.nodeRepository.nodes().list(Node.State.failed).size()); + assertEquals(4, tester.nodeRepository.nodes().list(Node.State.failed).size()); } @Test @@ -107,12 +110,14 @@ public class NodeFailerTest { String hostWithFailureReports = selectFirstParentHostWithNActiveNodesExcept(tester.nodeRepository, 2); assertEquals(Node.State.active, tester.nodeRepository.nodes().node(hostWithFailureReports).get().state()); - // The host has 2 nodes in active + // The host has 2 nodes in active and 1 ready Map<Node.State, List<String>> hostnamesByState = tester.nodeRepository.nodes().list().childrenOf(hostWithFailureReports).asList().stream() .collect(Collectors.groupingBy(Node::state, Collectors.mapping(Node::hostname, Collectors.toList()))); assertEquals(2, hostnamesByState.get(Node.State.active).size()); String activeChild1 = hostnamesByState.get(Node.State.active).get(0); String activeChild2 = hostnamesByState.get(Node.State.active).get(1); + assertEquals(1, hostnamesByState.get(Node.State.ready).size()); + String readyChild = hostnamesByState.get(Node.State.ready).get(0); // Set failure report to the parent and all its children. Report badTotalMemorySizeReport = Report.basicReport("badTotalMemorySize", HARD_FAIL, Instant.now(), "too low"); @@ -123,16 +128,20 @@ public class NodeFailerTest { tester.nodeRepository.nodes().write(updatedNode, () -> {}); }); - // Neither the host nor the 2 active nodes are failed out because they have not been suspended + // The ready node will be failed, but neither the host nor the 2 active nodes since they have not been suspended + tester.allNodesMakeAConfigRequestExcept(); tester.runMaintainers(); + assertEquals(Node.State.failed, tester.nodeRepository.nodes().node(readyChild).get().state()); assertEquals(Node.State.active, tester.nodeRepository.nodes().node(hostWithFailureReports).get().state()); assertEquals(Node.State.active, tester.nodeRepository.nodes().node(activeChild1).get().state()); assertEquals(Node.State.active, tester.nodeRepository.nodes().node(activeChild2).get().state()); - // Suspending the host will not fail any more since none of the children are suspended + // Suspending the host will not fail any more since none of the children are suspened tester.suspend(hostWithFailureReports); tester.clock.advance(Duration.ofHours(25)); + tester.allNodesMakeAConfigRequestExcept(); tester.runMaintainers(); + assertEquals(Node.State.failed, tester.nodeRepository.nodes().node(readyChild).get().state()); assertEquals(Node.State.active, tester.nodeRepository.nodes().node(hostWithFailureReports).get().state()); assertEquals(Node.State.active, tester.nodeRepository.nodes().node(activeChild1).get().state()); assertEquals(Node.State.active, tester.nodeRepository.nodes().node(activeChild2).get().state()); @@ -140,7 +149,9 @@ public class NodeFailerTest { // Suspending one child node will fail that out. tester.suspend(activeChild1); tester.clock.advance(Duration.ofHours(25)); + tester.allNodesMakeAConfigRequestExcept(); tester.runMaintainers(); + assertEquals(Node.State.failed, tester.nodeRepository.nodes().node(readyChild).get().state()); assertEquals(Node.State.active, tester.nodeRepository.nodes().node(hostWithFailureReports).get().state()); assertEquals(Node.State.failed, tester.nodeRepository.nodes().node(activeChild1).get().state()); assertEquals(Node.State.active, tester.nodeRepository.nodes().node(activeChild2).get().state()); @@ -148,7 +159,9 @@ public class NodeFailerTest { // Suspending the second child node will fail that out and the host. tester.suspend(activeChild2); tester.clock.advance(Duration.ofHours(25)); + tester.allNodesMakeAConfigRequestExcept(); tester.runMaintainers(); + assertEquals(Node.State.failed, tester.nodeRepository.nodes().node(readyChild).get().state()); assertEquals(Node.State.failed, tester.nodeRepository.nodes().node(hostWithFailureReports).get().state()); assertEquals(Node.State.failed, tester.nodeRepository.nodes().node(activeChild1).get().state()); assertEquals(Node.State.failed, tester.nodeRepository.nodes().node(activeChild2).get().state()); @@ -197,18 +210,31 @@ public class NodeFailerTest { @Test public void node_failing() { - NodeFailTester tester = NodeFailTester.withTwoApplications(6); + NodeFailTester tester = NodeFailTester.withTwoApplications(); // For a day all nodes work so nothing happens for (int minutes = 0; minutes < 24 * 60; minutes +=5 ) { tester.runMaintainers(); tester.clock.advance(Duration.ofMinutes(5)); + tester.allNodesMakeAConfigRequestExcept(); - assertEquals(0, tester.deployer.redeployments); - assertEquals(8, tester.nodeRepository.nodes().list(Node.State.active).nodeType(NodeType.tenant).size()); - assertEquals(0, tester.nodeRepository.nodes().list(Node.State.failed).nodeType(NodeType.tenant).size()); + assertEquals( 0, tester.deployer.redeployments); + assertEquals(12, tester.nodeRepository.nodes().list(Node.State.active).nodeType(NodeType.tenant).size()); + assertEquals( 0, tester.nodeRepository.nodes().list(Node.State.failed).nodeType(NodeType.tenant).size()); + assertEquals( 4, tester.nodeRepository.nodes().list(Node.State.ready).nodeType(NodeType.tenant).size()); } + // Hardware failures are detected on two ready nodes, which are then failed + Node readyFail1 = tester.nodeRepository.nodes().list(Node.State.ready).nodeType(NodeType.tenant).asList().get(2); + Node readyFail2 = tester.nodeRepository.nodes().list(Node.State.ready).nodeType(NodeType.tenant).asList().get(3); + tester.nodeRepository.nodes().write(readyFail1.with(new Reports().withReport(badTotalMemorySizeReport)), () -> {}); + tester.nodeRepository.nodes().write(readyFail2.with(new Reports().withReport(badTotalMemorySizeReport)), () -> {}); + assertEquals(4, tester.nodeRepository.nodes().list(Node.State.ready).nodeType(NodeType.tenant).size()); + tester.runMaintainers(); + assertEquals(2, tester.nodeRepository.nodes().list(Node.State.ready).nodeType(NodeType.tenant).size()); + assertEquals(Node.State.failed, tester.nodeRepository.nodes().node(readyFail1.hostname()).get().state()); + assertEquals(Node.State.failed, tester.nodeRepository.nodes().node(readyFail2.hostname()).get().state()); + String downHost1 = tester.nodeRepository.nodes().list(Node.State.active).owner(NodeFailTester.app1).asList().get(1).hostname(); String downHost2 = tester.nodeRepository.nodes().list(Node.State.active).owner(NodeFailTester.app2).asList().get(3).hostname(); tester.serviceMonitor.setHostDown(downHost1); @@ -217,34 +243,41 @@ public class NodeFailerTest { for (int minutes = 0; minutes < 45; minutes +=5 ) { tester.runMaintainers(); tester.clock.advance(Duration.ofMinutes(5)); - assertEquals(0, tester.deployer.redeployments); - assertEquals(8, tester.nodeRepository.nodes().list(Node.State.active).nodeType(NodeType.tenant).size()); - assertEquals(0, tester.nodeRepository.nodes().list(Node.State.failed).nodeType(NodeType.tenant).size()); + tester.allNodesMakeAConfigRequestExcept(); + assertEquals( 0, tester.deployer.redeployments); + assertEquals(12, tester.nodeRepository.nodes().list(Node.State.active).nodeType(NodeType.tenant).size()); + assertEquals( 2, tester.nodeRepository.nodes().list(Node.State.failed).nodeType(NodeType.tenant).size()); + assertEquals( 2, tester.nodeRepository.nodes().list(Node.State.ready).nodeType(NodeType.tenant).size()); } tester.serviceMonitor.setHostUp(downHost1); // downHost2 should now be failed and replaced, but not downHost1 tester.clock.advance(Duration.ofDays(1)); + tester.allNodesMakeAConfigRequestExcept(); tester.runMaintainers(); - assertEquals(1, tester.deployer.redeployments); - assertEquals(8, tester.nodeRepository.nodes().list(Node.State.active).nodeType(NodeType.tenant).size()); - assertEquals(1, tester.nodeRepository.nodes().list(Node.State.failed).nodeType(NodeType.tenant).size()); + assertEquals( 1, tester.deployer.redeployments); + assertEquals(12, tester.nodeRepository.nodes().list(Node.State.active).nodeType(NodeType.tenant).size()); + assertEquals( 3, tester.nodeRepository.nodes().list(Node.State.failed).nodeType(NodeType.tenant).size()); + assertEquals( 1, tester.nodeRepository.nodes().list(Node.State.ready).nodeType(NodeType.tenant).size()); assertEquals(downHost2, tester.nodeRepository.nodes().list(Node.State.failed).nodeType(NodeType.tenant).asList().get(0).hostname()); // downHost1 fails again tester.serviceMonitor.setHostDown(downHost1); tester.runMaintainers(); tester.clock.advance(Duration.ofMinutes(5)); + tester.allNodesMakeAConfigRequestExcept(); // the system goes down tester.clock.advance(Duration.ofMinutes(120)); tester.failer = tester.createFailer(); tester.runMaintainers(); // the host is still down and fails tester.clock.advance(Duration.ofMinutes(5)); + tester.allNodesMakeAConfigRequestExcept(); tester.runMaintainers(); - assertEquals(2, tester.deployer.redeployments); - assertEquals(8, tester.nodeRepository.nodes().list(Node.State.active).nodeType(NodeType.tenant).size()); - assertEquals(2, tester.nodeRepository.nodes().list(Node.State.failed).nodeType(NodeType.tenant).size()); + assertEquals( 2, tester.deployer.redeployments); + assertEquals(12, tester.nodeRepository.nodes().list(Node.State.active).nodeType(NodeType.tenant).size()); + assertEquals( 4, tester.nodeRepository.nodes().list(Node.State.failed).nodeType(NodeType.tenant).size()); + assertEquals( 0, tester.nodeRepository.nodes().list(Node.State.ready).nodeType(NodeType.tenant).size()); // the last host goes down Node lastNode = tester.highestIndex(tester.nodeRepository.nodes().list(Node.State.active).owner(NodeFailTester.app1)); @@ -253,19 +286,23 @@ public class NodeFailerTest { for (int minutes = 0; minutes < 75; minutes +=5 ) { tester.runMaintainers(); tester.clock.advance(Duration.ofMinutes(5)); - assertEquals(2, tester.deployer.redeployments); - assertEquals(8, tester.nodeRepository.nodes().list(Node.State.active).nodeType(NodeType.tenant).size()); - assertEquals(2, tester.nodeRepository.nodes().list(Node.State.failed).nodeType(NodeType.tenant).size()); + tester.allNodesMakeAConfigRequestExcept(); + assertEquals( 2, tester.deployer.redeployments); + assertEquals(12, tester.nodeRepository.nodes().list(Node.State.active).nodeType(NodeType.tenant).size()); + assertEquals( 4, tester.nodeRepository.nodes().list(Node.State.failed).nodeType(NodeType.tenant).size()); + assertEquals( 0, tester.nodeRepository.nodes().list(Node.State.ready).nodeType(NodeType.tenant).size()); } // A new node is available tester.createReadyNodes(1, 16, NodeFailTester.nodeResources); tester.clock.advance(Duration.ofDays(1)); + tester.allNodesMakeAConfigRequestExcept(); tester.runMaintainers(); // The node is now failed - assertEquals(3, tester.deployer.redeployments); - assertEquals(8, tester.nodeRepository.nodes().list(Node.State.active).nodeType(NodeType.tenant).size()); - assertEquals(3, tester.nodeRepository.nodes().list(Node.State.failed).nodeType(NodeType.tenant).size()); + assertEquals( 3, tester.deployer.redeployments); + assertEquals(12, tester.nodeRepository.nodes().list(Node.State.active).nodeType(NodeType.tenant).size()); + assertEquals( 5, tester.nodeRepository.nodes().list(Node.State.failed).nodeType(NodeType.tenant).size()); + assertEquals( 0, tester.nodeRepository.nodes().list(Node.State.ready).nodeType(NodeType.tenant).size()); assertTrue("The index of the last failed node is not reused", tester.highestIndex(tester.nodeRepository.nodes().list(Node.State.active).owner(NodeFailTester.app1)).allocation().get().membership().index() > @@ -282,10 +319,12 @@ public class NodeFailerTest { String downNode = tester.nodeRepository.nodes().list(Node.State.active).owner(NodeFailTester.app1).asList().get(1).hostname(); tester.serviceMonitor.setHostDown(downNode); + tester.allNodesMakeAConfigRequestExcept(); tester.runMaintainers(); assertEquals(0, tester.nodeRepository.nodes().list(Node.State.failed).nodeType(NodeType.tenant).size()); tester.clock.advance(Duration.ofMinutes(75)); + tester.allNodesMakeAConfigRequestExcept(); tester.runMaintainers(); assertEquals(1, tester.nodeRepository.nodes().list(Node.State.failed).nodeType(NodeType.tenant).size()); assertEquals(Node.State.failed, tester.nodeRepository.nodes().node(downNode).get().state()); @@ -293,10 +332,12 @@ public class NodeFailerTest { // Re-activate the node. It is still down, but should not be failed out until the grace period has passed again tester.nodeRepository.nodes().reactivate(downNode, Agent.system, getClass().getSimpleName()); tester.clock.advance(Duration.ofMinutes(30)); + tester.allNodesMakeAConfigRequestExcept(); tester.runMaintainers(); assertEquals(0, tester.nodeRepository.nodes().list(Node.State.failed).nodeType(NodeType.tenant).size()); tester.clock.advance(Duration.ofMinutes(45)); + tester.allNodesMakeAConfigRequestExcept(); tester.runMaintainers(); assertEquals(1, tester.nodeRepository.nodes().list(Node.State.failed).nodeType(NodeType.tenant).size()); assertEquals(Node.State.failed, tester.nodeRepository.nodes().node(downNode).get().state()); @@ -319,6 +360,7 @@ public class NodeFailerTest { for (int minutes = 0; minutes < 45; minutes +=5 ) { tester.runMaintainers(); tester.clock.advance(Duration.ofMinutes(5)); + tester.allNodesMakeAConfigRequestExcept(); assertEquals(0, tester.deployer.redeployments); assertEquals(3, tester.nodeRepository.nodes().list(Node.State.active).nodeType(NodeType.tenant).size()); assertEquals(0, tester.nodeRepository.nodes().list(Node.State.failed).nodeType(NodeType.tenant).size()); @@ -326,6 +368,7 @@ public class NodeFailerTest { // downHost should now be failed and replaced tester.clock.advance(Duration.ofDays(1)); + tester.allNodesMakeAConfigRequestExcept(); tester.runMaintainers(); assertEquals(1, tester.deployer.redeployments); assertEquals(1, tester.nodeRepository.nodes().list(Node.State.failed).nodeType(NodeType.tenant).size()); @@ -360,15 +403,37 @@ public class NodeFailerTest { } @Test + public void host_not_failed_without_config_requests() { + NodeFailTester tester = NodeFailTester.withTwoApplications(); + + // For a day all nodes work so nothing happens + for (int minutes = 0, interval = 30; minutes < 24 * 60; minutes += interval) { + tester.clock.advance(Duration.ofMinutes(interval)); + tester.allNodesMakeAConfigRequestExcept(); + tester.runMaintainers(); + assertEquals( 3, tester.nodeRepository.nodes().list(Node.State.ready).nodeType(NodeType.host).size()); + assertEquals( 0, tester.nodeRepository.nodes().list(Node.State.failed).nodeType(NodeType.host).size()); + } + + tester.clock.advance(Duration.ofMinutes(180)); + Node host = tester.nodeRepository.nodes().list(Node.State.ready).nodeType(NodeType.host).first().get(); + tester.allNodesMakeAConfigRequestExcept(host); + tester.runMaintainers(); + assertEquals( 3, tester.nodeRepository.nodes().list(Node.State.ready).nodeType(NodeType.host).size()); + assertEquals( 0, tester.nodeRepository.nodes().list(Node.State.failed).nodeType(NodeType.host).size()); + } + + @Test public void failing_hosts() { NodeFailTester tester = NodeFailTester.withTwoApplications(7); // For a day all nodes work so nothing happens for (int minutes = 0, interval = 30; minutes < 24 * 60; minutes += interval) { tester.clock.advance(Duration.ofMinutes(interval)); + tester.allNodesMakeAConfigRequestExcept(); tester.runMaintainers(); assertEquals(8, tester.nodeRepository.nodes().list(Node.State.active).nodeType(NodeType.tenant).size()); - assertEquals(0, tester.nodeRepository.nodes().list(Node.State.ready).nodeType(NodeType.tenant).size()); + assertEquals(13, tester.nodeRepository.nodes().list(Node.State.ready).nodeType(NodeType.tenant).size()); assertEquals(7, tester.nodeRepository.nodes().list(Node.State.active).nodeType(NodeType.host).size()); } @@ -381,17 +446,21 @@ public class NodeFailerTest { for (int minutes = 0; minutes < 45; minutes += 5 ) { tester.runMaintainers(); tester.clock.advance(Duration.ofMinutes(5)); + tester.allNodesMakeAConfigRequestExcept(); assertEquals(0, tester.deployer.redeployments); assertEquals(8, tester.nodeRepository.nodes().list(Node.State.active).nodeType(NodeType.tenant).size()); + assertEquals(13, tester.nodeRepository.nodes().list(Node.State.ready).nodeType(NodeType.tenant).size()); assertEquals(7, tester.nodeRepository.nodes().list(Node.State.active).nodeType(NodeType.host).size()); } tester.clock.advance(Duration.ofMinutes(30)); + tester.allNodesMakeAConfigRequestExcept(); tester.runMaintainers(); assertEquals(2, tester.deployer.redeployments); - assertEquals(2, tester.nodeRepository.nodes().list(Node.State.failed).nodeType(NodeType.tenant).size()); + assertEquals(3, tester.nodeRepository.nodes().list(Node.State.failed).nodeType(NodeType.tenant).size()); assertEquals(8, tester.nodeRepository.nodes().list(Node.State.active).nodeType(NodeType.tenant).size()); + assertEquals(10, tester.nodeRepository.nodes().list(Node.State.ready).nodeType(NodeType.tenant).size()); assertEquals(7, tester.nodeRepository.nodes().list(Node.State.active).nodeType(NodeType.host).size()); assertEquals(0, tester.nodeRepository.nodes().list(Node.State.failed).nodeType(NodeType.host).size()); @@ -399,8 +468,9 @@ public class NodeFailerTest { tester.runMaintainers(); assertEquals(2 + 1, tester.deployer.redeployments); - assertEquals(2, tester.nodeRepository.nodes().list(Node.State.failed).nodeType(NodeType.tenant).size()); + assertEquals(3, tester.nodeRepository.nodes().list(Node.State.failed).nodeType(NodeType.tenant).size()); assertEquals(8, tester.nodeRepository.nodes().list(Node.State.active).nodeType(NodeType.tenant).size()); + assertEquals(10, tester.nodeRepository.nodes().list(Node.State.ready).nodeType(NodeType.tenant).size()); assertEquals(6, tester.nodeRepository.nodes().list(Node.State.active).nodeType(NodeType.host).size()); assertEquals(1, tester.nodeRepository.nodes().list(Node.State.failed).nodeType(NodeType.host).size()); @@ -413,15 +483,18 @@ public class NodeFailerTest { for (int minutes = 0, interval = 30; minutes < 24 * 60; minutes += interval) { tester.runMaintainers(); tester.clock.advance(Duration.ofMinutes(interval)); - assertEquals(2 + 1, tester.nodeRepository.nodes().list(Node.State.failed).size()); + tester.allNodesMakeAConfigRequestExcept(); + assertEquals(3 + 1, tester.nodeRepository.nodes().list(Node.State.failed).size()); } tester.clock.advance(Duration.ofMinutes(30)); + tester.allNodesMakeAConfigRequestExcept(); tester.runMaintainers(); assertEquals(3 + 1, tester.deployer.redeployments); - assertEquals(3, tester.nodeRepository.nodes().list(Node.State.failed).nodeType(NodeType.tenant).size()); + assertEquals(4, tester.nodeRepository.nodes().list(Node.State.failed).nodeType(NodeType.tenant).size()); assertEquals(8, tester.nodeRepository.nodes().list(Node.State.active).nodeType(NodeType.tenant).size()); + assertEquals(9, tester.nodeRepository.nodes().list(Node.State.ready).nodeType(NodeType.tenant).size()); assertEquals(6, tester.nodeRepository.nodes().list(Node.State.active).nodeType(NodeType.host).size()); @@ -430,12 +503,14 @@ public class NodeFailerTest { tester.serviceMonitor.setHostDown(downHost2); tester.runMaintainers(); tester.clock.advance(Duration.ofMinutes(90)); + tester.allNodesMakeAConfigRequestExcept(); tester.runMaintainers(); tester.runMaintainers(); // The host is failed in the 2. maintain() assertEquals(5 + 2, tester.deployer.redeployments); - assertEquals(5, tester.nodeRepository.nodes().list(Node.State.failed).nodeType(NodeType.tenant).size()); + assertEquals(7, tester.nodeRepository.nodes().list(Node.State.failed).nodeType(NodeType.tenant).size()); assertEquals(8, tester.nodeRepository.nodes().list(Node.State.active).nodeType(NodeType.tenant).size()); + assertEquals(6, tester.nodeRepository.nodes().list(Node.State.ready).nodeType(NodeType.tenant).size()); assertEquals(5, tester.nodeRepository.nodes().list(Node.State.active).nodeType(NodeType.host).size()); @@ -445,11 +520,13 @@ public class NodeFailerTest { tester.serviceMonitor.setHostDown(downHost3); tester.runMaintainers(); tester.clock.advance(Duration.ofDays(1)); + tester.allNodesMakeAConfigRequestExcept(); tester.runMaintainers(); assertEquals(6 + 2, tester.deployer.redeployments); - assertEquals(6, tester.nodeRepository.nodes().list(Node.State.failed).nodeType(NodeType.tenant).size()); + assertEquals(9, tester.nodeRepository.nodes().list(Node.State.failed).nodeType(NodeType.tenant).size()); assertEquals(8, tester.nodeRepository.nodes().list(Node.State.active).nodeType(NodeType.tenant).size()); + assertEquals(4, tester.nodeRepository.nodes().list(Node.State.ready).nodeType(NodeType.tenant).size()); assertEquals(5, tester.nodeRepository.nodes().list(Node.State.active).nodeType(NodeType.host).size()); } @@ -470,6 +547,7 @@ public class NodeFailerTest { for (int minutes = 0; minutes < 24 * 60; minutes +=5 ) { tester.runMaintainers(); tester.clock.advance(Duration.ofMinutes(5)); + tester.allNodesMakeAConfigRequestExcept(); assertEquals(count, tester.nodeRepository.nodes().list(Node.State.active).nodeType(nodeType).size()); } @@ -482,6 +560,7 @@ public class NodeFailerTest { for (int minutes = 0; minutes < 45; minutes +=5 ) { tester.runMaintainers(); tester.clock.advance(Duration.ofMinutes(5)); + tester.allNodesMakeAConfigRequestExcept(); assertEquals( 0, tester.deployer.redeployments); assertEquals(count, tester.nodeRepository.nodes().list(Node.State.active).nodeType(nodeType).size()); } @@ -503,30 +582,38 @@ public class NodeFailerTest { } @Test + public void failing_divergent_ready_nodes() { + NodeFailTester tester = NodeFailTester.withNoApplications(); + + Node readyNode = tester.createReadyNodes(1).get(0); + + tester.runMaintainers(); + assertEquals(Node.State.ready, readyNode.state()); + + tester.nodeRepository.nodes().write(readyNode.with(new Reports().withReport(badTotalMemorySizeReport)), () -> {}); + + tester.runMaintainers(); + assertEquals(1, tester.nodeRepository.nodes().list(Node.State.failed).size()); + } + + @Test public void node_failing_throttle() { // Throttles based on an absolute number in small zone { - // 10 hosts with 7 container and 7 content nodes, total 24 nodes - NodeFailTester tester = NodeFailTester.withTwoApplications(10, 7, 7); - - List<String> failedHostHostnames = tester.nodeRepository.nodes().list().stream() - .flatMap(node -> node.parentHostname().stream()) - .collect(Collectors.groupingBy(h -> h, Collectors.counting())) - .entrySet().stream() - .sorted(Comparator.comparingLong((Map.Entry<String, Long> e) -> e.getValue()).reversed()) - .limit(3) - .map(Map.Entry::getKey) - .toList(); + // 10 hosts with 3 tenant nodes each, total 40 nodes + NodeFailTester tester = NodeFailTester.withTwoApplications(10); + NodeList hosts = tester.nodeRepository.nodes().list().nodeType(NodeType.host); // 3 hosts fail. 2 of them and all of their children are allowed to fail - failedHostHostnames.forEach(hostname -> tester.serviceMonitor.setHostDown(hostname)); + List<Node> failedHosts = hosts.asList().subList(0, 3); + failedHosts.forEach(host -> tester.serviceMonitor.setHostDown(host.hostname())); tester.runMaintainers(); tester.clock.advance(Duration.ofMinutes(61)); tester.runMaintainers(); tester.runMaintainers(); // hosts are typically failed in the 2. maintain() assertEquals(2 + /* hosts */ - (2 * 2) /* containers per host */, + (2 * 3) /* containers per host */, tester.nodeRepository.nodes().list(Node.State.failed).size()); assertEquals("Throttling is indicated by the metric", 1, tester.metric.values.get(NodeFailer.throttlingActiveMetric)); assertEquals("Throttled host failures", 1, tester.metric.values.get(NodeFailer.throttledHostFailuresMetric)); @@ -536,7 +623,7 @@ public class NodeFailerTest { tester.clock.advance(Duration.ofMinutes(interval)); } tester.runMaintainers(); - assertEquals(6, tester.nodeRepository.nodes().list(Node.State.failed).size()); + assertEquals(8, tester.nodeRepository.nodes().list(Node.State.failed).size()); assertEquals("Throttling is indicated by the metric", 1, tester.metric.values.get(NodeFailer.throttlingActiveMetric)); assertEquals("Throttled host failures", 1, tester.metric.values.get(NodeFailer.throttledHostFailuresMetric)); @@ -544,14 +631,14 @@ public class NodeFailerTest { tester.clock.advance(Duration.ofMinutes(30)); tester.runMaintainers(); tester.runMaintainers(); // hosts are failed in the 2. maintain() - assertEquals(9, tester.nodeRepository.nodes().list(Node.State.failed).size()); + assertEquals(12, tester.nodeRepository.nodes().list(Node.State.failed).size()); assertEquals("Throttling is not indicated by the metric, as no throttled attempt is made", 0, tester.metric.values.get(NodeFailer.throttlingActiveMetric)); assertEquals("No throttled node failures", 0, tester.metric.values.get(NodeFailer.throttledNodeFailuresMetric)); // Nothing else to fail tester.clock.advance(Duration.ofHours(25)); tester.runMaintainers(); - assertEquals(9, tester.nodeRepository.nodes().list(Node.State.failed).size()); + assertEquals(12, tester.nodeRepository.nodes().list(Node.State.failed).size()); assertEquals("Throttling is not indicated by the metric", 0, tester.metric.values.get(NodeFailer.throttlingActiveMetric)); assertEquals("No throttled node failures", 0, tester.metric.values.get(NodeFailer.throttledNodeFailuresMetric)); } diff --git a/node-repository/src/test/java/com/yahoo/vespa/hosted/provision/provisioning/ProvisioningTester.java b/node-repository/src/test/java/com/yahoo/vespa/hosted/provision/provisioning/ProvisioningTester.java index faf3f60e8af..03ab18c15d9 100644 --- a/node-repository/src/test/java/com/yahoo/vespa/hosted/provision/provisioning/ProvisioningTester.java +++ b/node-repository/src/test/java/com/yahoo/vespa/hosted/provision/provisioning/ProvisioningTester.java @@ -74,8 +74,6 @@ import static org.junit.Assert.assertTrue; */ public class ProvisioningTester { - public static final ApplicationId tenantHostApp = ApplicationId.from("hosted-vespa", "tenant-host", "default"); - private final Curator curator; private final NodeFlavors nodeFlavors; private final ManualClock clock; @@ -545,7 +543,7 @@ public class ProvisioningTester { } public void activateTenantHosts() { - prepareAndActivateInfraApplication(tenantHostApp, NodeType.host); + prepareAndActivateInfraApplication(applicationId(), NodeType.host); } public static ClusterSpec containerClusterSpec() { |