Examples with JobControl com.yahoo.vespa.hosted.provision.maintenance.JobControl used on opensource projects

Example 1 with JobControl

use of com.yahoo.vespa.hosted.provision.maintenance.JobControl in project vespa by vespa-engine.

the class MetricsReporterTest method test_registered_metric.

@Test
public void test_registered_metric() throws Exception {
    NodeFlavors nodeFlavors = FlavorConfigBuilder.createDummies("default");
    Curator curator = new MockCurator();
    NodeRepository nodeRepository = new NodeRepository(nodeFlavors, curator, Clock.systemUTC(), Zone.defaultZone(), new MockNameResolver().mockAnyLookup(), new DockerImage("docker-registry.domain.tld:8080/dist/vespa"), true);
    Node node = nodeRepository.createNode("openStackId", "hostname", Optional.empty(), nodeFlavors.getFlavorOrThrow("default"), NodeType.tenant);
    nodeRepository.addNodes(Collections.singletonList(node));
    Node hostNode = nodeRepository.createNode("openStackId2", "parent", Optional.empty(), nodeFlavors.getFlavorOrThrow("default"), NodeType.proxy);
    nodeRepository.addNodes(Collections.singletonList(hostNode));
    Map<String, Number> expectedMetrics = new HashMap<>();
    expectedMetrics.put("hostedVespa.provisionedHosts", 1L);
    expectedMetrics.put("hostedVespa.parkedHosts", 0L);
    expectedMetrics.put("hostedVespa.readyHosts", 0L);
    expectedMetrics.put("hostedVespa.reservedHosts", 0L);
    expectedMetrics.put("hostedVespa.activeHosts", 0L);
    expectedMetrics.put("hostedVespa.inactiveHosts", 0L);
    expectedMetrics.put("hostedVespa.dirtyHosts", 0L);
    expectedMetrics.put("hostedVespa.failedHosts", 0L);
    expectedMetrics.put("hostedVespa.docker.totalCapacityDisk", 0.0);
    expectedMetrics.put("hostedVespa.docker.totalCapacityMem", 0.0);
    expectedMetrics.put("hostedVespa.docker.totalCapacityCpu", 0.0);
    expectedMetrics.put("hostedVespa.docker.freeCapacityDisk", 0.0);
    expectedMetrics.put("hostedVespa.docker.freeCapacityMem", 0.0);
    expectedMetrics.put("hostedVespa.docker.freeCapacityCpu", 0.0);
    expectedMetrics.put("wantedRebootGeneration", 0L);
    expectedMetrics.put("currentRebootGeneration", 0L);
    expectedMetrics.put("wantToReboot", 0);
    expectedMetrics.put("wantToRetire", 0);
    expectedMetrics.put("wantToDeprovision", 0);
    expectedMetrics.put("hardwareFailure", 0);
    expectedMetrics.put("hardwareDivergence", 0);
    expectedMetrics.put("allowedToBeDown", 0);
    expectedMetrics.put("numberOfServices", 0L);
    Orchestrator orchestrator = mock(Orchestrator.class);
    ServiceMonitor serviceMonitor = mock(ServiceMonitor.class);
    when(orchestrator.getNodeStatus(any())).thenReturn(HostStatus.NO_REMARKS);
    ServiceModel serviceModel = mock(ServiceModel.class);
    when(serviceMonitor.getServiceModelSnapshot()).thenReturn(serviceModel);
    when(serviceModel.getServiceInstancesByHostName()).thenReturn(Collections.emptyMap());
    TestMetric metric = new TestMetric();
    MetricsReporter metricsReporter = new MetricsReporter(nodeRepository, metric, orchestrator, serviceMonitor, Duration.ofMinutes(1), new JobControl(nodeRepository.database()));
    metricsReporter.maintain();
    assertEquals(expectedMetrics, metric.values);
}

Example 2 with JobControl

use of com.yahoo.vespa.hosted.provision.maintenance.JobControl in project vespa by vespa-engine.

the class MultigroupProvisioningTest method test_provisioning_of_multiple_groups_after_flavor_migration_and_exiration.

@Test
public void test_provisioning_of_multiple_groups_after_flavor_migration_and_exiration() {
    ProvisioningTester tester = new ProvisioningTester(new Zone(Environment.prod, RegionName.from("us-east")));
    ApplicationId application1 = tester.makeApplicationId();
    tester.makeReadyNodes(10, "small");
    tester.makeReadyNodes(10, "large");
    deploy(application1, 8, 1, "small", tester);
    deploy(application1, 8, 1, "large", tester);
    // Expire small nodes
    tester.advanceTime(Duration.ofDays(7));
    MockDeployer deployer = new MockDeployer(tester.provisioner(), Collections.singletonMap(application1, new MockDeployer.ApplicationContext(application1, cluster(), Capacity.fromNodeCount(8, Optional.of("large"), false), 1)));
    new RetiredExpirer(tester.nodeRepository(), tester.orchestrator(), deployer, tester.clock(), Duration.ofDays(30), Duration.ofHours(12), new JobControl(tester.nodeRepository().database())).run();
    assertEquals(8, tester.getNodes(application1, Node.State.inactive).flavor("small").size());
    deploy(application1, 8, 8, "large", tester);
}

Example 3 with JobControl

use of com.yahoo.vespa.hosted.provision.maintenance.JobControl in project vespa by vespa-engine.

the class NodeTypeProvisioningTest method retire_proxy.

@Test
public void retire_proxy() {
    MockDeployer deployer = new MockDeployer(tester.provisioner(), Collections.singletonMap(application, new MockDeployer.ApplicationContext(application, clusterSpec, capacity, 1)));
    RetiredExpirer retiredExpirer = new RetiredExpirer(tester.nodeRepository(), tester.orchestrator(), deployer, tester.clock(), Duration.ofDays(30), Duration.ofMinutes(10), new JobControl(tester.nodeRepository().database()));
    {
        // Deploy
        List<HostSpec> hosts = deployProxies(application, tester);
        assertEquals("Reserved all proxies", 11, hosts.size());
        tester.activate(application, new HashSet<>(hosts));
        List<Node> nodes = tester.nodeRepository().getNodes(NodeType.proxy, Node.State.active);
        assertEquals("Activated all proxies", 11, nodes.size());
    }
    Node nodeToRetire = tester.nodeRepository().getNodes(NodeType.proxy, Node.State.active).get(5);
    {
        // Pick out a node and retire it
        tester.nodeRepository().write(nodeToRetire.with(nodeToRetire.status().withWantToRetire(true)));
        List<HostSpec> hosts = deployProxies(application, tester);
        assertEquals(11, hosts.size());
        tester.activate(application, new HashSet<>(hosts));
        List<Node> nodes = tester.nodeRepository().getNodes(NodeType.proxy, Node.State.active);
        assertEquals(11, nodes.size());
        // Verify that wantToRetire has been propagated
        assertTrue(tester.nodeRepository().getNode(nodeToRetire.hostname()).flatMap(Node::allocation).map(allocation -> allocation.membership().retired()).orElseThrow(RuntimeException::new));
    }
    {
        // Redeploying while the node is still retiring has no effect
        List<HostSpec> hosts = deployProxies(application, tester);
        assertEquals(11, hosts.size());
        tester.activate(application, new HashSet<>(hosts));
        List<Node> nodes = tester.nodeRepository().getNodes(NodeType.proxy, Node.State.active);
        assertEquals(11, nodes.size());
        // Verify that the node is still marked as retired
        assertTrue(tester.nodeRepository().getNode(nodeToRetire.hostname()).flatMap(Node::allocation).map(allocation -> allocation.membership().retired()).orElseThrow(RuntimeException::new));
    }
    {
        tester.advanceTime(Duration.ofMinutes(11));
        retiredExpirer.run();
        List<HostSpec> hosts = deployProxies(application, tester);
        assertEquals(10, hosts.size());
        tester.activate(application, new HashSet<>(hosts));
        List<Node> nodes = tester.nodeRepository().getNodes(NodeType.proxy, Node.State.active);
        assertEquals(10, nodes.size());
        // Verify that the node is now inactive
        assertEquals(Node.State.inactive, tester.nodeRepository().getNode(nodeToRetire.hostname()).orElseThrow(RuntimeException::new).state());
    }
}

Example 4 with JobControl

use of com.yahoo.vespa.hosted.provision.maintenance.JobControl in project vespa by vespa-engine.

the class ProvisioningTest method application_deployment_extends_existing_reservations_on_deploy.

@Test
public void application_deployment_extends_existing_reservations_on_deploy() {
    ProvisioningTester tester = new ProvisioningTester(new Zone(Environment.prod, RegionName.from("us-east")));
    ApplicationId application = tester.makeApplicationId();
    tester.makeReadyNodes(2, "default");
    // Deploy fails with out of capacity
    try {
        prepare(application, 2, 0, 2, 0, "default", tester);
        fail("Expected exception");
    } catch (OutOfCapacityException ignored) {
    }
    assertEquals("Reserved a subset of required nodes", 2, tester.getNodes(application, Node.State.reserved).size());
    // Enough nodes become available
    tester.makeReadyNodes(2, "default");
    // Deploy is retried after a few minutes
    tester.clock().advance(Duration.ofMinutes(2));
    SystemState state = prepare(application, 2, 0, 2, 0, "default", tester);
    List<Node> reserved = tester.getNodes(application, Node.State.reserved).asList();
    assertEquals("Reserved required nodes", 4, reserved.size());
    assertTrue("Time of event is updated for all nodes", reserved.stream().allMatch(n -> n.history().event(History.Event.Type.reserved).get().at().equals(tester.clock().instant())));
    // Over 10 minutes pass since first reservation. First set of reserved nodes are not expired
    tester.clock().advance(Duration.ofMinutes(8).plus(Duration.ofSeconds(1)));
    ReservationExpirer expirer = new ReservationExpirer(tester.nodeRepository(), tester.clock(), Duration.ofMinutes(10), new JobControl(tester.nodeRepository().database()));
    expirer.run();
    assertEquals("Nodes remain reserved", 4, tester.getNodes(application, Node.State.reserved).size());
    tester.activate(application, state.allHosts);
    assertEquals(4, tester.getNodes(application, Node.State.active).size());
}

Example 5 with JobControl

use of com.yahoo.vespa.hosted.provision.maintenance.JobControl in project vespa by vespa-engine.

the class NodeTypeProvisioningTest method retire_multiple_proxy_simultaneously.

@Test
public void retire_multiple_proxy_simultaneously() {
    MockDeployer deployer = new MockDeployer(tester.provisioner(), Collections.singletonMap(application, new MockDeployer.ApplicationContext(application, clusterSpec, capacity, 1)));
    RetiredExpirer retiredExpirer = new RetiredExpirer(tester.nodeRepository(), tester.orchestrator(), deployer, tester.clock(), Duration.ofDays(30), Duration.ofMinutes(10), new JobControl(tester.nodeRepository().database()));
    final int numNodesToRetire = 5;
    {
        // Deploy
        List<HostSpec> hosts = deployProxies(application, tester);
        assertEquals("Reserved all proxies", 11, hosts.size());
        tester.activate(application, new HashSet<>(hosts));
        List<Node> nodes = tester.nodeRepository().getNodes(NodeType.proxy, Node.State.active);
        assertEquals("Activated all proxies", 11, nodes.size());
    }
    List<Node> nodesToRetire = tester.nodeRepository().getNodes(NodeType.proxy, Node.State.active).subList(3, 3 + numNodesToRetire);
    String currentyRetiringHostname;
    {
        nodesToRetire.forEach(nodeToRetire -> tester.nodeRepository().write(nodeToRetire.with(nodeToRetire.status().withWantToRetire(true))));
        List<HostSpec> hosts = deployProxies(application, tester);
        assertEquals(11, hosts.size());
        tester.activate(application, new HashSet<>(hosts));
        List<Node> nodes = tester.nodeRepository().getNodes(NodeType.proxy, Node.State.active);
        assertEquals(11, nodes.size());
        // Verify that wantToRetire has been propagated
        List<Node> nodesCurrentlyRetiring = nodes.stream().filter(node -> node.allocation().get().membership().retired()).collect(Collectors.toList());
        assertEquals(1, nodesCurrentlyRetiring.size());
        // The retiring node should be one of the nodes we marked for retirement
        currentyRetiringHostname = nodesCurrentlyRetiring.get(0).hostname();
        assertTrue(nodesToRetire.stream().map(Node::hostname).filter(hostname -> hostname.equals(currentyRetiringHostname)).count() == 1);
    }
    {
        // Redeploying while the node is still retiring has no effect
        List<HostSpec> hosts = deployProxies(application, tester);
        assertEquals(11, hosts.size());
        tester.activate(application, new HashSet<>(hosts));
        List<Node> nodes = tester.nodeRepository().getNodes(NodeType.proxy, Node.State.active);
        assertEquals(11, nodes.size());
        // Verify that wantToRetire has been propagated
        List<Node> nodesCurrentlyRetiring = nodes.stream().filter(node -> node.allocation().get().membership().retired()).collect(Collectors.toList());
        assertEquals(1, nodesCurrentlyRetiring.size());
        // The node that started retiring is still the only one retiring
        assertEquals(currentyRetiringHostname, nodesCurrentlyRetiring.get(0).hostname());
    }
    {
        tester.advanceTime(Duration.ofMinutes(11));
        retiredExpirer.run();
        List<HostSpec> hosts = deployProxies(application, tester);
        assertEquals(10, hosts.size());
        tester.activate(application, new HashSet<>(hosts));
        List<Node> nodes = tester.nodeRepository().getNodes(NodeType.proxy, Node.State.active);
        assertEquals(10, nodes.size());
        // Verify the node we previously set to retire has finished retiring
        assertEquals(Node.State.inactive, tester.nodeRepository().getNode(currentyRetiringHostname).orElseThrow(RuntimeException::new).state());
        // Verify that a node is currently retiring
        List<Node> nodesCurrentlyRetiring = nodes.stream().filter(node -> node.allocation().get().membership().retired()).collect(Collectors.toList());
        assertEquals(1, nodesCurrentlyRetiring.size());
        // This node is different from the one that was retiring previously
        String newRetiringHostname = nodesCurrentlyRetiring.get(0).hostname();
        assertNotEquals(currentyRetiringHostname, newRetiringHostname);
        // ... but is one of the nodes that were put to wantToRetire earlier
        assertTrue(nodesToRetire.stream().map(Node::hostname).filter(hostname -> hostname.equals(newRetiringHostname)).count() == 1);
    }
    for (int i = 0; i < 10; i++) {
        tester.advanceTime(Duration.ofMinutes(11));
        retiredExpirer.run();
        List<HostSpec> hosts = deployProxies(application, tester);
        tester.activate(application, new HashSet<>(hosts));
    }
    // After a long time, all currently active proxy nodes are not marked with wantToRetire or as retired
    long numRetiredActiveProxyNodes = tester.nodeRepository().getNodes(NodeType.proxy, Node.State.active).stream().filter(node -> !node.status().wantToRetire()).filter(node -> !node.allocation().get().membership().retired()).count();
    assertEquals(11 - numNodesToRetire, numRetiredActiveProxyNodes);
    // All the nodes that were marked with wantToRetire earlier are now inactive
    assertEquals(nodesToRetire.stream().map(Node::hostname).collect(Collectors.toSet()), tester.nodeRepository().getNodes(Node.State.inactive).stream().map(Node::hostname).collect(Collectors.toSet()));
}