diff options
author | Martin Polden <mpolden@mpolden.no> | 2019-10-07 13:02:27 +0200 |
---|---|---|
committer | Martin Polden <mpolden@mpolden.no> | 2019-10-07 13:59:40 +0200 |
commit | 374b324b2f5fdbcc99761e064bb4a182e0f6aa07 (patch) | |
tree | 074cc669ec6ded569d81aeba6c041c108e8b5bb0 /controller-server/src/test/java/com | |
parent | dd2739ec3abce6ca0ab3341b0ef6968f48f26492 (diff) |
Report metric for nodes failing OS upgrade
Diffstat (limited to 'controller-server/src/test/java/com')
2 files changed, 88 insertions, 3 deletions
diff --git a/controller-server/src/test/java/com/yahoo/vespa/hosted/controller/integration/ConfigServerMock.java b/controller-server/src/test/java/com/yahoo/vespa/hosted/controller/integration/ConfigServerMock.java index 6da77a967f1..6e7a50b5f81 100644 --- a/controller-server/src/test/java/com/yahoo/vespa/hosted/controller/integration/ConfigServerMock.java +++ b/controller-server/src/test/java/com/yahoo/vespa/hosted/controller/integration/ConfigServerMock.java @@ -151,15 +151,44 @@ public class ConfigServerMock extends AbstractComponent implements ConfigServer /** Set version for an application in a given zone */ public void setVersion(ApplicationId application, ZoneId zone, Version version) { - setVersion(application, zone, version, -1); + setVersion(application, zone, version, -1, false); } /** Set version for nodeCount number of nodes in application in a given zone */ public void setVersion(ApplicationId application, ZoneId zone, Version version, int nodeCount) { + setVersion(application, zone, version, nodeCount, false); + } + + /** Set OS version for an application in a given zone */ + public void setOsVersion(ApplicationId application, ZoneId zone, Version version) { + setOsVersion(application, zone, version, -1); + } + + /** Set OS version for an application in a given zone */ + public void setOsVersion(ApplicationId application, ZoneId zone, Version version, int nodeCount) { + setVersion(application, zone, version, nodeCount, true); + } + + private void setVersion(ApplicationId application, ZoneId zone, Version version, int nodeCount, boolean osVersion) { int n = 0; for (Node node : nodeRepository().list(zone, application)) { - nodeRepository().putByHostname(zone, new Node(node.hostname(), node.state(), node.type(), node.owner(), - version, version)); + Node newNode; + if (osVersion) { + newNode = new Node(node.hostname(), node.state(), node.type(), node.owner(), node.currentVersion(), + node.wantedVersion(), version, version, node.serviceState(), + node.restartGeneration(), node.wantedRestartGeneration(), node.rebootGeneration(), + node.wantedRebootGeneration(), node.vcpu(), node.memoryGb(), node.diskGb(), + node.bandwidthGbps(), node.fastDisk(), node.cost(), node.canonicalFlavor(), + node.clusterId(), node.clusterType()); + } else { + newNode = new Node(node.hostname(), node.state(), node.type(), node.owner(), version, + version, node.currentOsVersion(), node.wantedOsVersion(), node.serviceState(), + node.restartGeneration(), node.wantedRestartGeneration(), node.rebootGeneration(), + node.wantedRebootGeneration(), node.vcpu(), node.memoryGb(), node.diskGb(), + node.bandwidthGbps(), node.fastDisk(), node.cost(), node.canonicalFlavor(), + node.clusterId(), node.clusterType()); + } + nodeRepository().putByHostname(zone, newNode); if (++n == nodeCount) break; } } diff --git a/controller-server/src/test/java/com/yahoo/vespa/hosted/controller/maintenance/MetricsReporterTest.java b/controller-server/src/test/java/com/yahoo/vespa/hosted/controller/maintenance/MetricsReporterTest.java index 9cb40d60677..44785407874 100644 --- a/controller-server/src/test/java/com/yahoo/vespa/hosted/controller/maintenance/MetricsReporterTest.java +++ b/controller-server/src/test/java/com/yahoo/vespa/hosted/controller/maintenance/MetricsReporterTest.java @@ -3,6 +3,7 @@ package com.yahoo.vespa.hosted.controller.maintenance; import com.yahoo.component.Version; import com.yahoo.config.provision.ApplicationId; +import com.yahoo.config.provision.CloudName; import com.yahoo.config.provision.Environment; import com.yahoo.config.provision.zone.UpgradePolicy; import com.yahoo.config.provision.zone.ZoneId; @@ -262,6 +263,57 @@ public class MetricsReporterTest { } } + @Test + public void test_nodes_failing_os_upgrade() { + var tester = new DeploymentTester(); + var reporter = createReporter(tester.controller()); + var zone = ZoneApiMock.fromId("prod.eu-west-1"); + var cloud = CloudName.defaultName(); + tester.controllerTester().zoneRegistry().setOsUpgradePolicy(cloud, UpgradePolicy.create().upgrade(zone)); + var osUpgrader = new OsUpgrader(tester.controller(), Duration.ofDays(1), + new JobControl(tester.controllerTester().curator()), CloudName.defaultName());; + var statusUpdater = new OsVersionStatusUpdater(tester.controller(), Duration.ofDays(1), + new JobControl(tester.controller().curator())); + tester.configServer().bootstrap(List.of(zone.getId()), SystemApplication.tenantHost); + + // All nodes upgrade to initial OS version + var version0 = Version.fromString("8.0"); + tester.controller().upgradeOsIn(cloud, version0, false); + osUpgrader.maintain(); + tester.configServer().setOsVersion(SystemApplication.tenantHost.id(), zone.getId(), version0); + statusUpdater.maintain(); + reporter.maintain(); + assertEquals(0, getNodesFailingOsUpgrade()); + + for (var version : List.of(Version.fromString("8.1"), Version.fromString("8.2"))) { + // System starts upgrading to next OS version + tester.controller().upgradeOsIn(cloud, version, false); + osUpgrader.maintain(); + statusUpdater.maintain(); + reporter.maintain(); + assertEquals(0, getNodesFailingOsUpgrade()); + + // 30 minutes pass and nothing happens + tester.clock().advance(Duration.ofMinutes(30)); + statusUpdater.maintain(); + reporter.maintain(); + assertEquals(0, getNodesFailingOsUpgrade()); + + // 1/3 nodes upgrade within timeout + tester.configServer().setOsVersion(SystemApplication.tenantHost.id(), zone.getId(), version, 1); + tester.clock().advance(Duration.ofMinutes(30).plus(Duration.ofSeconds(1))); + statusUpdater.maintain(); + reporter.maintain(); + assertEquals(2, getNodesFailingOsUpgrade()); + + // 3/3 nodes upgrade + tester.configServer().setOsVersion(SystemApplication.tenantHost.id(), zone.getId(), version); + statusUpdater.maintain(); + reporter.maintain(); + assertEquals(0, getNodesFailingOsUpgrade()); + } + } + private Duration getAverageDeploymentDuration(ApplicationId id) { return Duration.ofSeconds(getMetric(MetricsReporter.DEPLOYMENT_AVERAGE_DURATION, id).longValue()); } @@ -278,6 +330,10 @@ public class MetricsReporterTest { return metrics.getMetric(MetricsReporter.NODES_FAILING_SYSTEM_UPGRADE).intValue(); } + private int getNodesFailingOsUpgrade() { + return metrics.getMetric(MetricsReporter.NODES_FAILING_OS_UPGRADE).intValue(); + } + private Number getMetric(String name, ApplicationId id) { return metrics.getMetric((dimensions) -> id.tenant().value().equals(dimensions.get("tenant")) && appDimension(id).equals(dimensions.get("app")), |