|
14 | 14 |
|
15 | 15 | package google.registry.mosapi; |
16 | 16 |
|
| 17 | +import com.google.api.client.util.DateTime; |
| 18 | +import com.google.api.services.monitoring.v3.Monitoring; |
| 19 | +import com.google.api.services.monitoring.v3.model.CreateTimeSeriesRequest; |
| 20 | +import com.google.api.services.monitoring.v3.model.Metric; |
| 21 | +import com.google.api.services.monitoring.v3.model.MonitoredResource; |
| 22 | +import com.google.api.services.monitoring.v3.model.Point; |
| 23 | +import com.google.api.services.monitoring.v3.model.TimeInterval; |
| 24 | +import com.google.api.services.monitoring.v3.model.TimeSeries; |
| 25 | +import com.google.api.services.monitoring.v3.model.TypedValue; |
| 26 | +import com.google.common.base.Ascii; |
| 27 | +import com.google.common.collect.ImmutableList; |
| 28 | +import com.google.common.collect.ImmutableMap; |
| 29 | +import com.google.common.collect.Lists; |
17 | 30 | import com.google.common.flogger.FluentLogger; |
| 31 | +import google.registry.config.RegistryConfig.Config; |
| 32 | +import google.registry.mosapi.MosApiModels.ServiceStatus; |
18 | 33 | import google.registry.mosapi.MosApiModels.TldServiceState; |
19 | 34 | import jakarta.inject.Inject; |
| 35 | +import jakarta.inject.Named; |
| 36 | +import java.io.IOException; |
| 37 | +import java.util.ArrayList; |
| 38 | +import java.util.Collections; |
20 | 39 | import java.util.List; |
| 40 | +import java.util.Map; |
| 41 | +import java.util.concurrent.ExecutorService; |
21 | 42 |
|
22 | 43 | /** Metrics Exporter for MoSAPI. */ |
23 | 44 | public class MosApiMetrics { |
24 | 45 |
|
25 | 46 | private static final FluentLogger logger = FluentLogger.forEnclosingClass(); |
26 | 47 |
|
| 48 | + // Google Cloud Monitoring Limit: Max 200 TimeSeries per request |
| 49 | + private static final int MAX_TIMESERIES_PER_REQUEST = 195; |
| 50 | + |
| 51 | + // Magic String Constants |
| 52 | + private static final String METRIC_DOMAIN = "custom.googleapis.com/mosapi/"; |
| 53 | + private static final String PROJECT_RESOURCE_PREFIX = "projects/"; |
| 54 | + private static final String RESOURCE_TYPE_GLOBAL = "global"; |
| 55 | + private static final String LABEL_PROJECT_ID = "project_id"; |
| 56 | + private static final String LABEL_TLD = "tld"; |
| 57 | + private static final String LABEL_SERVICE_TYPE = "service_type"; |
| 58 | + |
| 59 | + // Metric Names |
| 60 | + private static final String METRIC_TLD_STATUS = "tld_status"; |
| 61 | + private static final String METRIC_SERVICE_STATUS = "service_status"; |
| 62 | + private static final String METRIC_EMERGENCY_USAGE = "emergency_usage"; |
| 63 | + |
| 64 | + // MoSAPI Status Constants |
| 65 | + private static final String STATUS_UP_INCONCLUSIVE = "UP-INCONCLUSIVE"; |
| 66 | + private static final String STATUS_DOWN = "DOWN"; |
| 67 | + private static final String STATUS_DISABLED = "DISABLED"; |
| 68 | + |
| 69 | + private final Monitoring monitoringClient; |
| 70 | + private final String projectId; |
| 71 | + private final ExecutorService executor; |
| 72 | + |
27 | 73 | @Inject |
28 | | - public MosApiMetrics() {} |
| 74 | + public MosApiMetrics( |
| 75 | + Monitoring monitoringClient, |
| 76 | + @Config("projectId") String projectId, |
| 77 | + @Named("mosapiMetricsExecutor") ExecutorService executor) { |
| 78 | + this.monitoringClient = monitoringClient; |
| 79 | + this.projectId = projectId; |
| 80 | + this.executor = executor; |
| 81 | + } |
29 | 82 |
|
| 83 | + /** Accepts a list of states and processes them in a single async batch task. */ |
30 | 84 | public void recordStates(List<TldServiceState> states) { |
31 | | - // b/467541269: Logic to push status to Cloud Monitoring goes here |
32 | | - logger.atInfo().log("MoSAPI record metrics logic will be implemented from here"); |
| 85 | + executor.execute( |
| 86 | + () -> { |
| 87 | + try { |
| 88 | + pushBatchMetrics(states); |
| 89 | + } catch (Throwable t) { |
| 90 | + logger.atWarning().withCause(t).log("Async batch metric push failed."); |
| 91 | + } |
| 92 | + }); |
| 93 | + } |
| 94 | + |
| 95 | + private void pushBatchMetrics(List<TldServiceState> states) throws IOException { |
| 96 | + List<TimeSeries> allTimeSeries = new ArrayList<>(); |
| 97 | + TimeInterval interval = |
| 98 | + new TimeInterval().setEndTime(new DateTime(System.currentTimeMillis()).toString()); |
| 99 | + |
| 100 | + for (TldServiceState state : states) { |
| 101 | + // 1. TLD Status Metric |
| 102 | + allTimeSeries.add(createTldStatusTimeSeries(state, interval)); |
| 103 | + |
| 104 | + // 2. Service-level Metrics |
| 105 | + Map<String, ServiceStatus> services = state.serviceStatuses(); |
| 106 | + if (services != null) { |
| 107 | + for (Map.Entry<String, ServiceStatus> entry : services.entrySet()) { |
| 108 | + addServiceMetrics(allTimeSeries, state.tld(), entry.getKey(), entry.getValue(), interval); |
| 109 | + } |
| 110 | + } |
| 111 | + } |
| 112 | + |
| 113 | + for (List<TimeSeries> chunk : Lists.partition(allTimeSeries, MAX_TIMESERIES_PER_REQUEST)) { |
| 114 | + CreateTimeSeriesRequest request = new CreateTimeSeriesRequest().setTimeSeries(chunk); |
| 115 | + monitoringClient |
| 116 | + .projects() |
| 117 | + .timeSeries() |
| 118 | + .create(PROJECT_RESOURCE_PREFIX + projectId, request) |
| 119 | + .execute(); |
| 120 | + logger.atInfo().log( |
| 121 | + "Successfully pushed batch of %d time series to Cloud Monitoring.", chunk.size()); |
| 122 | + } |
| 123 | + } |
| 124 | + |
| 125 | + private void addServiceMetrics( |
| 126 | + List<TimeSeries> list, |
| 127 | + String tld, |
| 128 | + String serviceType, |
| 129 | + ServiceStatus statusObj, |
| 130 | + TimeInterval interval) { |
| 131 | + ImmutableMap<String, String> labels = |
| 132 | + ImmutableMap.of(LABEL_TLD, tld, LABEL_SERVICE_TYPE, serviceType); |
| 133 | + |
| 134 | + list.add( |
| 135 | + createTimeSeries( |
| 136 | + METRIC_SERVICE_STATUS, labels, parseServiceStatus(statusObj.status()), interval)); |
| 137 | + |
| 138 | + list.add( |
| 139 | + createTimeSeries(METRIC_EMERGENCY_USAGE, labels, statusObj.emergencyThreshold(), interval)); |
| 140 | + } |
| 141 | + |
| 142 | + private TimeSeries createTldStatusTimeSeries(TldServiceState state, TimeInterval interval) { |
| 143 | + return createTimeSeries( |
| 144 | + METRIC_TLD_STATUS, |
| 145 | + ImmutableMap.of(LABEL_TLD, state.tld()), |
| 146 | + parseTldStatus(state.status()), |
| 147 | + interval); |
| 148 | + } |
| 149 | + |
| 150 | + private TimeSeries createTimeSeries( |
| 151 | + String suffix, Map<String, String> labels, Number val, TimeInterval interval) { |
| 152 | + Metric metric = new Metric().setType(METRIC_DOMAIN + suffix).setLabels(labels); |
| 153 | + MonitoredResource resource = |
| 154 | + new MonitoredResource() |
| 155 | + .setType(RESOURCE_TYPE_GLOBAL) |
| 156 | + .setLabels(Collections.singletonMap(LABEL_PROJECT_ID, projectId)); |
| 157 | + |
| 158 | + TypedValue tv = new TypedValue(); |
| 159 | + if (val instanceof Double) { |
| 160 | + tv.setDoubleValue((Double) val); |
| 161 | + } else { |
| 162 | + tv.setInt64Value(val.longValue()); |
| 163 | + } |
| 164 | + |
| 165 | + return new TimeSeries() |
| 166 | + .setMetric(metric) |
| 167 | + .setResource(resource) |
| 168 | + .setPoints(ImmutableList.of(new Point().setInterval(interval).setValue(tv))); |
| 169 | + } |
| 170 | + |
| 171 | + /** |
| 172 | + * Translates MoSAPI status to a numeric metric. |
| 173 | + * |
| 174 | + * <p>Mappings: 1 (UP) = Healthy; 0 (DOWN) = Critical failure; 2 (UP-INCONCLUSIVE) = Disabled/Not |
| 175 | + * Monitored/In Maintenance. |
| 176 | + * |
| 177 | + * <p>A status of 2 indicates the SLA monitoring system is under maintenance. The TLD is |
| 178 | + * considered "UP" by default, but individual service checks are disabled. This distinguishes |
| 179 | + * maintenance windows from actual availability or outages. |
| 180 | + * |
| 181 | + * @see <a href="https://www.icann.org/mosapi-specification.pdf">ICANN MoSAPI Spec Sec 5.1</a> |
| 182 | + */ |
| 183 | + private long parseTldStatus(String status) { |
| 184 | + if (status == null) { |
| 185 | + return 1; |
| 186 | + } |
| 187 | + return switch (Ascii.toUpperCase(status)) { |
| 188 | + case STATUS_DOWN -> 0; |
| 189 | + case STATUS_UP_INCONCLUSIVE -> 2; |
| 190 | + default -> 1; // status is up |
| 191 | + }; |
| 192 | + } |
| 193 | + |
| 194 | + /** |
| 195 | + * Translates MoSAPI service status to a numeric metric. |
| 196 | + * |
| 197 | + * <p>Mappings: 1 (UP) = Healthy; 0 (DOWN) = Critical failure; 2 (DISABLED/UP-INCONCLUSIVE*) = |
| 198 | + * Disabled/Not Monitored/In Maintenance. |
| 199 | + * |
| 200 | + * @see <a href="https://www.icann.org/mosapi-specification.pdf">ICANN MoSAPI Spec Sec 5.1</a> |
| 201 | + */ |
| 202 | + private long parseServiceStatus(String status) { |
| 203 | + if (status == null) { |
| 204 | + return 1; |
| 205 | + } |
| 206 | + String serviceStatus = Ascii.toUpperCase(status); |
| 207 | + if (serviceStatus.startsWith(STATUS_UP_INCONCLUSIVE)) { |
| 208 | + return 2; |
| 209 | + } |
| 210 | + return switch (serviceStatus) { |
| 211 | + case STATUS_DOWN -> 0; |
| 212 | + case STATUS_DISABLED -> 2; |
| 213 | + default -> 1; // status is Up |
| 214 | + }; |
33 | 215 | } |
34 | 216 | } |
0 commit comments