@oneuptime/common 10.4.11 → 10.4.13

package/Types/Dashboard/DashboardTemplates.ts

@@ -5,13 +5,13 @@ import DashboardComponentType from "./DashboardComponentType";
 import DashboardChartType from "./Chart/ChartType";
 import ObjectID from "../ObjectID";
 import DashboardBaseComponent from "./DashboardComponents/DashboardBaseComponent";
+import DashboardVariable, { DashboardVariableType } from "./DashboardVariable";
 import IconProp from "../Icon/IconProp";
 import MetricsAggregationType from "../Metrics/MetricsAggregationType";
 import IncidentMetricType from "../Incident/IncidentMetricType";
 import MonitorMetricType from "../Monitor/MonitorMetricType";
 import MetricDashboardMetricType from "../Metrics/MetricDashboardMetricType";
 import { DashboardValueTrendDirection } from "./DashboardComponents/DashboardValueComponent";
-import { HostMetricKind } from "./DashboardComponents/DashboardHostMetricChartComponent";
 
 /*
  * Trace / Exception / Profiles entries are intentionally not in this
@@ -88,6 +88,19 @@ interface MetricConfig {
   aggregationType: MetricsAggregationType;
   legend?: string;
   legendUnit?: string;
+  /*
+   * OpenTelemetry attribute keys to fan the query out across (e.g.
+   * ["host.name"] for one series per host). When set, the chart renders
+   * one series per unique value combination.
+   */
+  groupByAttributeKeys?: Array<string>;
+  /*
+   * Plot the per-second rate of change instead of the raw cumulative
+   * counter. Required for OTel cumulative counters such as
+   * `system.disk.io` and `system.network.io` so the chart shows I/O rate
+   * rather than bytes-since-boot.
+   */
+  transformAsRate?: boolean;
 }
 
 function buildMetricQueryConfig(config: MetricConfig): Record<string, unknown> {
@@ -104,8 +117,10 @@ function buildMetricQueryConfig(config: MetricConfig): Record<string, unknown> {
         metricName: config.metricName,
         aggegationType: config.aggregationType,
       },
-      groupBy: undefined,
+      groupBy: config.groupByAttributeKeys ? { attributes: true } : undefined,
+      groupByAttributeKeys: config.groupByAttributeKeys,
     },
+    transformAsRate: config.transformAsRate,
   };
 }
 
@@ -383,6 +398,31 @@ function createKubernetesNodeListComponent(data: {
   };
 }
 
+/*
+ * Template variables are pre-built TelemetryAttribute variables that
+ * ship with the template — the toolbar renders a "Cluster" /
+ * "Namespace" / "Host" picker without the user having to open the
+ * Variables modal. Resource attributes from the OTel collector are
+ * stored in ClickHouse under the `resource.` prefix (see the comment
+ * in MonitorAlert.ts), so the binding keys here mirror what the
+ * Kubernetes / Host detail pages already filter on.
+ */
+function createTelemetryAttributeVariable(data: {
+  name: string;
+  label: string;
+  attributeKey: string;
+  isMultiSelect?: boolean;
+}): DashboardVariable {
+  return {
+    id: ObjectID.generate().toString(),
+    name: data.name,
+    label: data.label,
+    type: DashboardVariableType.TelemetryAttribute,
+    attributeKey: data.attributeKey,
+    isMultiSelect: data.isMultiSelect ?? false,
+  };
+}
+
 function createHostListComponent(data: {
   title: string;
   top: number;
@@ -412,33 +452,6 @@ function createHostListComponent(data: {
   };
 }
 
-function createHostMetricChartComponent(data: {
-  title: string;
-  description?: string;
-  metricKind: HostMetricKind;
-  top: number;
-  left: number;
-  width: number;
-  height: number;
-}): DashboardBaseComponent {
-  return {
-    _type: ObjectType.DashboardComponent,
-    componentType: DashboardComponentType.HostMetricChart,
-    componentId: ObjectID.generate(),
-    topInDashboardUnits: data.top,
-    leftInDashboardUnits: data.left,
-    widthInDashboardUnits: data.width,
-    heightInDashboardUnits: data.height,
-    minHeightInDashboardUnits: 3,
-    minWidthInDashboardUnits: 6,
-    arguments: {
-      title: data.title,
-      description: data.description,
-      metricKind: data.metricKind,
-    },
-  };
-}
-
 // -- Dashboard configs --
 
 function createMonitorDashboardConfig(): DashboardViewConfig {
@@ -630,9 +643,25 @@ function createMonitorDashboardConfig(): DashboardViewConfig {
     }),
   ];
 
+  /*
+   * Monitor metrics are stored with bare attribute keys (`monitorName`,
+   * `probeName`) rather than the OTel `resource.*` prefix — see
+   * MonitorMetricUtil.buildAttributes — so the variable binds to the
+   * bare key. Multi-select lets users pin a small group of monitors.
+   */
+  const variables: Array<DashboardVariable> = [
+    createTelemetryAttributeVariable({
+      name: "monitor",
+      label: "Monitor",
+      attributeKey: "monitorName",
+      isMultiSelect: true,
+    }),
+  ];
+
   return {
     _type: ObjectType.DashboardViewConfig,
     components,
+    variables,
     heightInDashboardUnits: Math.max(DashboardSize.heightInDashboardUnits, 13),
   };
 }
@@ -1122,9 +1151,31 @@ function createKubernetesDashboardConfig(): DashboardViewConfig {
     }),
   ];
 
+  /*
+   * Pre-built variables let the user scope every widget on the dashboard
+   * to a single cluster / namespace from the toolbar. Multi-select is on
+   * for namespace so users can pick a couple of namespaces at once;
+   * cluster stays single-select since the typical "compare two clusters"
+   * workflow lives on the Compare page.
+   */
+  const variables: Array<DashboardVariable> = [
+    createTelemetryAttributeVariable({
+      name: "cluster",
+      label: "Cluster",
+      attributeKey: "resource.k8s.cluster.name",
+    }),
+    createTelemetryAttributeVariable({
+      name: "namespace",
+      label: "Namespace",
+      attributeKey: "resource.k8s.namespace.name",
+      isMultiSelect: true,
+    }),
+  ];
+
   return {
     _type: ObjectType.DashboardViewConfig,
     components,
+    variables,
     heightInDashboardUnits: Math.max(DashboardSize.heightInDashboardUnits, 21),
   };
 }
@@ -1442,9 +1493,25 @@ function createMetricsDashboardConfig(): DashboardViewConfig {
     }),
   ];
 
+  /*
+   * Most HTTP / runtime metrics on this dashboard carry the standard
+   * OTel `resource.service.name` attribute, so scoping by service is
+   * the most useful default. Multi-select keeps the cross-service view
+   * available — e.g. compare API and worker on one chart.
+   */
+  const variables: Array<DashboardVariable> = [
+    createTelemetryAttributeVariable({
+      name: "service",
+      label: "Service",
+      attributeKey: "resource.service.name",
+      isMultiSelect: true,
+    }),
+  ];
+
   return {
     _type: ObjectType.DashboardViewConfig,
     components,
+    variables,
     heightInDashboardUnits: Math.max(DashboardSize.heightInDashboardUnits, 24),
   };
 }
@@ -1453,11 +1520,10 @@ function createHostDashboardConfig(): DashboardViewConfig {
   /*
    * Layout notes:
    *
-   * - The HostMetricChart widget groups by `host.name` and plots one
-   *   series per host (or a single series when filtered to one host).
-   *   The OTel host receiver emits `system.cpu.utilization` and
-   *   `system.memory.utilization` as [0, 1] ratios, which Value/Gauge
-   *   widgets auto-scale to percent at render time (see
+   * - Per-host charts fan out via `groupByAttributeKeys: ["host.name"]`
+   *   so a single chart renders one series per host. The OTel host
+   *   receiver emits `system.cpu.utilization` as a [0, 1] ratio, which
+   *   Value/Gauge widgets auto-scale to percent at render time (see
    *   splitFormattedValue / isFractionScale).
    *
    * - The Value widget for "Avg Memory" uses `system.memory.usage`
@@ -1465,11 +1531,10 @@ function createHostDashboardConfig(): DashboardViewConfig {
    *   `system.memory.utilization` is not always emitted by default OTel
    *   host metrics.
    *
-   * - Disk and Network I/O Value widgets aggregate the raw byte counters
-   *   with Sum. HostMetricChart turns these into per-second rates for
-   *   the chart view; the Value tile shows the unrated total so users
-   *   reading "Disk I/O" against the time window get an absolute byte
-   *   figure rather than a noisy snapshot.
+   * - Disk and Network I/O charts set `transformAsRate: true` so the
+   *   cumulative byte counters render as per-second rates rather than
+   *   bytes-since-boot. The matching Value tiles in row 1 keep the
+   *   unrated Sum so users see an absolute byte figure over the window.
    */
   const components: Array<DashboardBaseComponent> = [
     // Row 0: Title
@@ -1529,21 +1594,31 @@ function createHostDashboardConfig(): DashboardViewConfig {
     }),
 
     // Row 2-4: Per-host CPU and Memory charts
-    createHostMetricChartComponent({
+    createChartComponent({
       title: "CPU Utilization by Host",
-      metricKind: HostMetricKind.CpuUtilization,
+      chartType: DashboardChartType.Line,
       top: 2,
       left: 0,
       width: 6,
       height: 3,
+      metricConfig: {
+        metricName: MetricDashboardMetricType.SystemCpuUtilization,
+        aggregationType: MetricsAggregationType.Avg,
+        groupByAttributeKeys: ["host.name"],
+      },
     }),
-    createHostMetricChartComponent({
+    createChartComponent({
       title: "Memory Usage by Host",
-      metricKind: HostMetricKind.MemoryUsage,
+      chartType: DashboardChartType.Line,
       top: 2,
       left: 6,
       width: 6,
       height: 3,
+      metricConfig: {
+        metricName: MetricDashboardMetricType.SystemMemoryUsage,
+        aggregationType: MetricsAggregationType.Avg,
+        groupByAttributeKeys: ["host.name"],
+      },
     }),
 
     // Row 5: Section header
@@ -1600,13 +1675,18 @@ function createHostDashboardConfig(): DashboardViewConfig {
         aggregationType: MetricsAggregationType.Avg,
       },
     }),
-    createHostMetricChartComponent({
+    createChartComponent({
       title: "Filesystem Usage by Host",
-      metricKind: HostMetricKind.Filesystem,
+      chartType: DashboardChartType.Line,
       top: 11,
       left: 4,
       width: 8,
       height: 3,
+      metricConfig: {
+        metricName: "system.filesystem.usage",
+        aggregationType: MetricsAggregationType.Avg,
+        groupByAttributeKeys: ["host.name"],
+      },
     }),
 
     // Row 14: Section header
@@ -1620,21 +1700,33 @@ function createHostDashboardConfig(): DashboardViewConfig {
     }),
 
     // Row 15-17: Disk and network I/O rates per host
-    createHostMetricChartComponent({
+    createChartComponent({
       title: "Disk I/O by Host",
-      metricKind: HostMetricKind.DiskIo,
+      chartType: DashboardChartType.Line,
       top: 15,
       left: 0,
       width: 6,
       height: 3,
+      metricConfig: {
+        metricName: MetricDashboardMetricType.SystemDiskIo,
+        aggregationType: MetricsAggregationType.Sum,
+        groupByAttributeKeys: ["host.name"],
+        transformAsRate: true,
+      },
     }),
-    createHostMetricChartComponent({
+    createChartComponent({
       title: "Network I/O by Host",
-      metricKind: HostMetricKind.NetworkIo,
+      chartType: DashboardChartType.Line,
       top: 15,
       left: 6,
       width: 6,
       height: 3,
+      metricConfig: {
+        metricName: MetricDashboardMetricType.SystemNetworkIo,
+        aggregationType: MetricsAggregationType.Sum,
+        groupByAttributeKeys: ["host.name"],
+        transformAsRate: true,
+      },
     }),
 
     // Row 18: Section header
@@ -1648,13 +1740,18 @@ function createHostDashboardConfig(): DashboardViewConfig {
     }),
 
     // Row 19-21: Process count chart and recent logs
-    createHostMetricChartComponent({
+    createChartComponent({
       title: "Process Count by Host",
-      metricKind: HostMetricKind.ProcessCount,
+      chartType: DashboardChartType.Line,
       top: 19,
       left: 0,
       width: 6,
       height: 3,
+      metricConfig: {
+        metricName: "system.processes.count",
+        aggregationType: MetricsAggregationType.Avg,
+        groupByAttributeKeys: ["host.name"],
+      },
     }),
     createLogStreamComponent({
       title: "Recent Logs",
@@ -1665,9 +1762,25 @@ function createHostDashboardConfig(): DashboardViewConfig {
     }),
   ];
 
+  /*
+   * Per-host scoping: the variable defaults to multi-select because the
+   * dashboard's by-host charts already split rendering per host; the
+   * selector lets users narrow to a subset (e.g. just two prod hosts)
+   * without losing the per-host breakdown.
+   */
+  const variables: Array<DashboardVariable> = [
+    createTelemetryAttributeVariable({
+      name: "host",
+      label: "Host",
+      attributeKey: "resource.host.name",
+      isMultiSelect: true,
+    }),
+  ];
+
   return {
     _type: ObjectType.DashboardViewConfig,
     components,
+    variables,
     heightInDashboardUnits: Math.max(DashboardSize.heightInDashboardUnits, 22),
   };
 }

package/Types/Monitor/CriteriaFilter.ts

@@ -84,6 +84,14 @@ export enum CheckOn {
   DomainStatusCode = "Domain Status Code",
   DomainIsExpired = "Domain Is Expired",
 
+  // DNSSEC monitors.
+  DnssecChainValid = "DNSSEC Chain Is Valid",
+  DnssecDnskeyExists = "DNSSEC DNSKEY Record Exists",
+  DnssecDsExists = "DNSSEC DS Record Exists At Parent",
+  DnssecSignatureExpiresInDays = "DNSSEC Signature Expires In Days",
+  DnssecResolverConsensus = "DNSSEC Resolver Consensus (AD Flag)",
+  DnssecNameserverConsistent = "DNSSEC Nameservers Are Consistent",
+
   // External Status Page monitors.
   ExternalStatusPageIsOnline = "External Status Page Is Online",
   ExternalStatusPageOverallStatus = "External Status Page Overall Status",
@@ -279,6 +287,11 @@ export class CriteriaFilterUtil {
       checkOn === CheckOn.SnmpIsOnline ||
       checkOn === CheckOn.DnsIsOnline ||
       checkOn === CheckOn.DomainIsExpired ||
+      checkOn === CheckOn.DnssecChainValid ||
+      checkOn === CheckOn.DnssecDnskeyExists ||
+      checkOn === CheckOn.DnssecDsExists ||
+      checkOn === CheckOn.DnssecResolverConsensus ||
+      checkOn === CheckOn.DnssecNameserverConsistent ||
       checkOn === CheckOn.ExternalStatusPageIsOnline
     ) {
       return false;

package/Types/Monitor/DnssecMonitor/DnssecMonitorResponse.ts

@@ -0,0 +1,69 @@
+export interface DnssecKeyRecord {
+  flags: number;
+  algorithm: number;
+  keyTag?: number | undefined;
+}
+
+export interface DnssecDsRecord {
+  keyTag: number;
+  algorithm: number;
+  digestType: number;
+  digest: string;
+}
+
+export interface DnssecRrsigRecord {
+  typeCovered: string;
+  algorithm: number;
+  signerName: string;
+  keyTag: number;
+  inception?: string | undefined;
+  expiration?: string | undefined;
+}
+
+export interface DnssecResolverCheck {
+  resolver: string;
+  adFlag: boolean;
+  servfailWhenValidating: boolean;
+  error?: string | undefined;
+}
+
+export interface DnssecNameserverCheck {
+  nameServer: string;
+  soaSerial?: string | undefined;
+  rrsigExpiration?: string | undefined;
+  error?: string | undefined;
+}
+
+export default interface DnssecMonitorResponse {
+  isOnline: boolean;
+  responseTimeInMs: number;
+  failureCause: string;
+  domainName: string;
+  isTimeout?: boolean | undefined;
+
+  // Zone signed?
+  isZoneSigned: boolean;
+
+  // DNSKEY presence
+  dnskeys: Array<DnssecKeyRecord>;
+
+  // DS at parent
+  parentDsRecords: Array<DnssecDsRecord>;
+  isParentDsPresent: boolean;
+
+  // RRSIG over the A record (zone apex by default)
+  rrsigs: Array<DnssecRrsigRecord>;
+  earliestSignatureExpiration?: string | undefined;
+  daysUntilSignatureExpiry?: number | undefined;
+
+  // Resolver consensus (AD flag + CD-bit SERVFAIL test)
+  resolverChecks: Array<DnssecResolverCheck>;
+  resolverConsensusAd: boolean;
+
+  // Primary/secondary nameserver consistency
+  nameserverChecks: Array<DnssecNameserverCheck>;
+  isNameserverConsistent: boolean;
+
+  // Overall chain validity (DNSKEY exists, DS exists, RRSIG valid, AD across resolvers)
+  isChainValid: boolean;
+}

package/Types/Monitor/MonitorCriteriaInstance.ts

@@ -487,6 +487,33 @@ export default class MonitorCriteriaInstance extends DatabaseProperty {
       return monitorCriteriaInstance;
     }
 
+    if (arg.monitorType === MonitorType.DNSSEC) {
+      const monitorCriteriaInstance: MonitorCriteriaInstance =
+        new MonitorCriteriaInstance();
+
+      monitorCriteriaInstance.data = {
+        id: ObjectID.generate().toString(),
+        monitorStatusId: arg.monitorStatusId,
+        filterCondition: FilterCondition.All,
+        filters: [
+          {
+            checkOn: CheckOn.DnssecChainValid,
+            filterType: FilterType.True,
+            value: undefined,
+          },
+        ],
+        incidents: [],
+        alerts: [],
+        createAlerts: false,
+        changeMonitorStatus: true,
+        createIncidents: false,
+        name: `Check if ${arg.monitorName} DNSSEC chain is valid`,
+        description: `This criteria checks if the ${arg.monitorName} DNSSEC chain is valid end-to-end`,
+      };
+
+      return monitorCriteriaInstance;
+    }
+
     if (arg.monitorType === MonitorType.ExternalStatusPage) {
       const monitorCriteriaInstance: MonitorCriteriaInstance =
         new MonitorCriteriaInstance();
@@ -695,6 +722,46 @@ export default class MonitorCriteriaInstance extends DatabaseProperty {
       };
     }
 
+    if (arg.monitorType === MonitorType.DNSSEC) {
+      monitorCriteriaInstance.data = {
+        id: ObjectID.generate().toString(),
+        monitorStatusId: arg.monitorStatusId,
+        filterCondition: FilterCondition.Any,
+        filters: [
+          {
+            checkOn: CheckOn.DnssecChainValid,
+            filterType: FilterType.False,
+            value: undefined,
+          },
+        ],
+        incidents: [
+          {
+            title: `${arg.monitorName} DNSSEC chain is broken`,
+            description: `${arg.monitorName} DNSSEC validation is currently failing.`,
+            incidentSeverityId: arg.incidentSeverityId,
+            autoResolveIncident: true,
+            id: ObjectID.generate().toString(),
+            onCallPolicyIds: [],
+          },
+        ],
+        changeMonitorStatus: true,
+        createIncidents: true,
+        createAlerts: false,
+        alerts: [
+          {
+            title: `${arg.monitorName} DNSSEC chain is broken`,
+            description: `${arg.monitorName} DNSSEC validation is currently failing.`,
+            alertSeverityId: arg.alertSeverityId,
+            autoResolveAlert: true,
+            id: ObjectID.generate().toString(),
+            onCallPolicyIds: [],
+          },
+        ],
+        name: `Check if ${arg.monitorName} DNSSEC chain is broken`,
+        description: `This criteria checks if the ${arg.monitorName} DNSSEC chain is broken`,
+      };
+    }
+
     if (arg.monitorType === MonitorType.ExternalStatusPage) {
       monitorCriteriaInstance.data = {
         id: ObjectID.generate().toString(),

package/Types/Monitor/MonitorStep.ts

@@ -38,6 +38,9 @@ import MonitorStepDnsMonitor, {
 import MonitorStepDomainMonitor, {
   MonitorStepDomainMonitorUtil,
 } from "./MonitorStepDomainMonitor";
+import MonitorStepDnssecMonitor, {
+  MonitorStepDnssecMonitorUtil,
+} from "./MonitorStepDnssecMonitor";
 import MonitorStepExternalStatusPageMonitor, {
   MonitorStepExternalStatusPageMonitorUtil,
 } from "./MonitorStepExternalStatusPageMonitor";
@@ -109,6 +112,9 @@ export interface MonitorStepType {
   // Domain monitor
   domainMonitor?: MonitorStepDomainMonitor | undefined;
 
+  // DNSSEC monitor
+  dnssecMonitor?: MonitorStepDnssecMonitor | undefined;
+
   // External Status Page monitor
   externalStatusPageMonitor?: MonitorStepExternalStatusPageMonitor | undefined;
 
@@ -150,6 +156,7 @@ export default class MonitorStep extends DatabaseProperty {
       snmpMonitor: undefined,
       dnsMonitor: undefined,
       domainMonitor: undefined,
+      dnssecMonitor: undefined,
       externalStatusPageMonitor: undefined,
       kubernetesMonitor: undefined,
       dockerMonitor: undefined,
@@ -191,6 +198,7 @@ export default class MonitorStep extends DatabaseProperty {
       snmpMonitor: undefined,
       dnsMonitor: undefined,
       domainMonitor: undefined,
+      dnssecMonitor: undefined,
       externalStatusPageMonitor: undefined,
       kubernetesMonitor: undefined,
       dockerMonitor: undefined,
@@ -334,6 +342,13 @@ export default class MonitorStep extends DatabaseProperty {
     return this;
   }
 
+  public setDnssecMonitor(
+    dnssecMonitor: MonitorStepDnssecMonitor,
+  ): MonitorStep {
+    this.data!.dnssecMonitor = dnssecMonitor;
+    return this;
+  }
+
   public setExternalStatusPageMonitor(
     externalStatusPageMonitor: MonitorStepExternalStatusPageMonitor,
   ): MonitorStep {
@@ -508,6 +523,23 @@ export default class MonitorStep extends DatabaseProperty {
       }
     }
 
+    if (monitorType === MonitorType.DNSSEC) {
+      if (!value.data.dnssecMonitor) {
+        return "DNSSEC configuration is required";
+      }
+
+      if (!value.data.dnssecMonitor.domainName) {
+        return "Domain name is required";
+      }
+
+      if (
+        !value.data.dnssecMonitor.resolvers ||
+        value.data.dnssecMonitor.resolvers.length === 0
+      ) {
+        return "At least one resolver is required";
+      }
+    }
+
     if (monitorType === MonitorType.ExternalStatusPage) {
       if (!value.data.externalStatusPageMonitor) {
         return "External status page configuration is required";
@@ -600,6 +632,9 @@ export default class MonitorStep extends DatabaseProperty {
           domainMonitor: this.data.domainMonitor
             ? MonitorStepDomainMonitorUtil.toJSON(this.data.domainMonitor)
             : undefined,
+          dnssecMonitor: this.data.dnssecMonitor
+            ? MonitorStepDnssecMonitorUtil.toJSON(this.data.dnssecMonitor)
+            : undefined,
           externalStatusPageMonitor: this.data.externalStatusPageMonitor
             ? MonitorStepExternalStatusPageMonitorUtil.toJSON(
                 this.data.externalStatusPageMonitor,
@@ -734,6 +769,9 @@ export default class MonitorStep extends DatabaseProperty {
       domainMonitor: json["domainMonitor"]
         ? (json["domainMonitor"] as JSONObject)
         : undefined,
+      dnssecMonitor: json["dnssecMonitor"]
+        ? (json["dnssecMonitor"] as JSONObject)
+        : undefined,
       externalStatusPageMonitor: json["externalStatusPageMonitor"]
         ? (json["externalStatusPageMonitor"] as JSONObject)
         : undefined,
@@ -775,6 +813,7 @@ export default class MonitorStep extends DatabaseProperty {
         snmpMonitor: Zod.any().optional(),
         dnsMonitor: Zod.any().optional(),
         domainMonitor: Zod.any().optional(),
+        dnssecMonitor: Zod.any().optional(),
         externalStatusPageMonitor: Zod.any().optional(),
         kubernetesMonitor: Zod.any().optional(),
         dockerMonitor: Zod.any().optional(),