diff options
author | Martin Polden <mpolden@mpolden.no> | 2023-07-18 10:41:05 +0200 |
---|---|---|
committer | GitHub <noreply@github.com> | 2023-07-18 10:41:05 +0200 |
commit | 1a7da42f40f7f7c664380fa646b048f8ee4e3523 (patch) | |
tree | 5ad9d6b0924d885e5589a878f42e250836c16a14 /node-repository | |
parent | e7e320d7e40f6a2dff5d10b5e0034c3c8a310d07 (diff) | |
parent | bfe317ab0a6ee6a05f309d5c59f974868e475a8b (diff) |
Merge pull request #27807 from vespa-engine/mpolden/empty-hosts-metric
Emit metric counting empty exclusive hosts
Diffstat (limited to 'node-repository')
2 files changed, 46 insertions, 3 deletions
diff --git a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/maintenance/MetricsReporter.java b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/maintenance/MetricsReporter.java index 3b846351b36..15913fec5ed 100644 --- a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/maintenance/MetricsReporter.java +++ b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/maintenance/MetricsReporter.java @@ -21,11 +21,13 @@ import com.yahoo.vespa.hosted.provision.NodeList; import com.yahoo.vespa.hosted.provision.NodeRepository; import com.yahoo.vespa.hosted.provision.node.Allocation; import com.yahoo.vespa.hosted.provision.node.ClusterId; +import com.yahoo.vespa.hosted.provision.node.History; import com.yahoo.vespa.hosted.provision.persistence.CacheStats; import com.yahoo.vespa.service.monitor.ServiceModel; import com.yahoo.vespa.service.monitor.ServiceMonitor; import java.time.Duration; +import java.time.Instant; import java.util.Comparator; import java.util.HashMap; import java.util.HashSet; @@ -64,7 +66,7 @@ public class MetricsReporter extends NodeRepositoryMaintainer { @Override public double maintain() { // Sort by hostname to get deterministic metric reporting order (and hopefully avoid changes - // to metric reporting time so we get double reporting or no reporting within a minute) + // to metric reporting time, so we get double reporting or no reporting within a minute) NodeList nodes = nodeRepository().nodes().list().sortedBy(Comparator.comparing(Node::hostname)); ServiceModel serviceModel = serviceMonitor.getServiceModelSnapshot(); @@ -79,6 +81,7 @@ public class MetricsReporter extends NodeRepositoryMaintainer { updateRepairTicketMetrics(nodes); updateAllocationMetrics(nodes); updateClusterMetrics(nodes); + updateEmptyExclusiveHosts(nodes); return 1.0; } @@ -386,6 +389,19 @@ public class MetricsReporter extends NodeRepositoryMaintainer { .forEach((status, number) -> metric.set(ConfigServerMetrics.HOSTED_VESPA_BREAKFIXED_HOSTS.baseName(), number, getContext(Map.of("status", status)))); } + private void updateEmptyExclusiveHosts(NodeList nodes) { + Instant now = nodeRepository().clock().instant(); + Duration minActivePeriod = Duration.ofMinutes(10); + int emptyHosts = nodes.parents().state(State.active) + .matching(node -> (node.type() != NodeType.host && node.type().isHost()) || + node.exclusiveToApplicationId().isPresent()) + .matching(host -> host.history().hasEventBefore(History.Event.Type.activated, + now.minus(minActivePeriod))) + .matching(host -> nodes.childrenOf(host).state(State.active).isEmpty()) + .size(); + metric.set(ConfigServerMetrics.NODES_EMPTY_EXCLUSIVE.baseName(), emptyHosts, null); + } + static Map<String, String> dimensions(ApplicationId application, ClusterSpec.Id cluster) { Map<String, String> dimensions = new HashMap<>(dimensions(application)); dimensions.put("clusterid", cluster.value()); diff --git a/node-repository/src/test/java/com/yahoo/vespa/hosted/provision/maintenance/MetricsReporterTest.java b/node-repository/src/test/java/com/yahoo/vespa/hosted/provision/maintenance/MetricsReporterTest.java index 2637666e643..3091f82143d 100644 --- a/node-repository/src/test/java/com/yahoo/vespa/hosted/provision/maintenance/MetricsReporterTest.java +++ b/node-repository/src/test/java/com/yahoo/vespa/hosted/provision/maintenance/MetricsReporterTest.java @@ -26,7 +26,6 @@ import com.yahoo.vespa.hosted.provision.autoscale.Autoscaling; import com.yahoo.vespa.hosted.provision.autoscale.Load; import com.yahoo.vespa.hosted.provision.node.Agent; import com.yahoo.vespa.hosted.provision.node.Allocation; -import com.yahoo.vespa.hosted.provision.node.ClusterId; import com.yahoo.vespa.hosted.provision.node.Generation; import com.yahoo.vespa.hosted.provision.node.IP; import com.yahoo.vespa.hosted.provision.provisioning.FlavorConfigBuilder; @@ -140,6 +139,7 @@ public class MetricsReporterTest { expectedMetrics.put("cache.curator.hitRate", 3D/5D); expectedMetrics.put("cache.curator.evictionCount", 0L); expectedMetrics.put("cache.curator.size", 2L); + expectedMetrics.put("nodes.emptyExclusive", 0); nodeRepository.nodes().list(); tester.clock().setInstant(Instant.ofEpochSecond(124)); @@ -278,7 +278,6 @@ public class MetricsReporterTest { assertEquals(4, getMetric("nodes.active", metric, dimensions)); assertEquals(0, getMetric("nodes.nonActive", metric, dimensions)); - Map<String, String> clusterDimensions = Map.of("applicationId", applicationId.toFullString(), "clusterid", clusterSpec.id().value()); assertEquals(1.392, getMetric("cluster.cost", metric, clusterDimensions)); @@ -341,6 +340,34 @@ public class MetricsReporterTest { assertEquals(1D, getMetric("nodes.exclusiveSwitchFraction", metric, MetricsReporter.dimensions(app, spec2.id())).doubleValue(), Double.MIN_VALUE); } + @Test + public void empty_exclusive_hosts() { + ProvisioningTester tester = new ProvisioningTester.Builder().build(); + ApplicationId app = ApplicationId.from("t1", "a1", "default"); + TestMetric metric = new TestMetric(); + MetricsReporter metricsReporter = metricsReporter(metric, tester); + NodeResources resources = new NodeResources(8, 32, 100, 10); + List<Node> hosts = tester.makeReadyNodes(4, resources, NodeType.host, 5); + tester.activateTenantHosts(); + tester.patchNodes(hosts, (host) -> host.withExclusiveToApplicationId(app)); + + // Hosts are not considered empty until enough time passes + metricsReporter.maintain(); + assertEquals(0, metric.values.get("nodes.emptyExclusive").intValue()); + tester.clock().advance(Duration.ofMinutes(10)); + metricsReporter.maintain(); + assertEquals(hosts.size(), metric.values.get("nodes.emptyExclusive").intValue()); + + // Deploy application + ClusterSpec spec = ClusterSpec.request(ClusterSpec.Type.container, ClusterSpec.Id.from("c1")).vespaVersion("1").build(); + Capacity capacity = Capacity.from(new ClusterResources(4, 1, resources)); + tester.deploy(app, spec, capacity); + + // Host are now in use + metricsReporter.maintain(); + assertEquals(0, metric.values.get("nodes.emptyExclusive").intValue()); + } + private Number getMetric(String name, TestMetric metric, Map<String, String> dimensions) { List<TestMetric.TestContext> metrics = metric.context.get(name).stream() .filter(ctx -> ctx.properties.entrySet().containsAll(dimensions.entrySet())) |