@oneuptime/common 10.0.63 → 10.0.66

package/Server/Services/KubernetesResourceService.ts

@@ -0,0 +1,351 @@
+import DatabaseService from "./DatabaseService";
+import Model from "../../Models/DatabaseModels/KubernetesResource";
+import CaptureSpan from "../Utils/Telemetry/CaptureSpan";
+import ObjectID from "../../Types/ObjectID";
+import OneUptimeDate from "../../Types/Date";
+import { ParsedKubernetesResource } from "../../Types/Kubernetes/KubernetesInventoryExtractor";
+import logger from "../Utils/Logger";
+
+/*
+ * ------------------------------------------------------------------
+ * KubernetesResourceService
+ *
+ * Writes and reads the inventory table populated by the telemetry
+ * ingest path. Callers are either:
+ *   - OtelLogsIngestService (bulkUpsert, from processLogsAsync)
+ *   - CleanupStaleResources worker (deleteStaleForCluster)
+ *   - KubernetesResourceAPI / KubernetesObjectFetcher (reads via
+ *     the inherited DatabaseService CRUD)
+ *
+ * ------------------------------------------------------------------
+ */
+
+export type { ParsedKubernetesResource };
+
+export interface InventorySummary {
+  countsByKind: Record<string, number>;
+  /*
+   * Sum of `jsonb_array_length(spec->'containers')` across all pods in
+   * the cluster. Containers aren't a top-level kind in the inventory,
+   * so we derive the total server-side so the sidebar badge and the
+   * Containers page agree.
+   */
+  containerCount: number;
+  podPhaseCounts: {
+    running: number;
+    pending: number;
+    failed: number;
+    succeeded: number;
+    unknown: number;
+  };
+  nodeReadyCounts: {
+    ready: number;
+    notReady: number;
+  };
+  nodePressureCounts: {
+    memoryPressure: number;
+    diskPressure: number;
+    pidPressure: number;
+  };
+}
+
+const UPSERT_BATCH_SIZE: number = 500;
+const STALE_DELETE_WARN_THRESHOLD: number = 100;
+
+/*
+ * Column order used by both bulkUpsert() and its generated parameter tuples.
+ * Keep this and the INSERT column list in perfect sync.
+ */
+const UPSERT_COLUMNS: Array<keyof ParsedKubernetesResource | string> = [
+  "projectId",
+  "kubernetesClusterId",
+  "kind",
+  "namespaceKey",
+  "name",
+  "uid",
+  "phase",
+  "isReady",
+  "hasMemoryPressure",
+  "hasDiskPressure",
+  "hasPidPressure",
+  "labels",
+  "annotations",
+  "ownerReferences",
+  "spec",
+  "status",
+  "lastSeenAt",
+  "resourceCreationTimestamp",
+  "version",
+];
+
+export class Service extends DatabaseService<Model> {
+  public constructor() {
+    super(Model);
+  }
+
+  /**
+   * Upsert a batch of parsed resources for a single (project, cluster)
+   * pair. Uses ON CONFLICT on the UNIQUE (projectId, clusterId, kind,
+   * namespaceKey, name) index with a dominance guard on lastSeenAt
+   * so out-of-order ingest never regresses a newer snapshot.
+   */
+  @CaptureSpan()
+  public async bulkUpsert(data: {
+    projectId: ObjectID;
+    kubernetesClusterId: ObjectID;
+    resources: Array<ParsedKubernetesResource>;
+  }): Promise<void> {
+    if (data.resources.length === 0) {
+      return;
+    }
+
+    // Chunk to keep individual statement parameter counts reasonable.
+    for (let i: number = 0; i < data.resources.length; i += UPSERT_BATCH_SIZE) {
+      const chunk: Array<ParsedKubernetesResource> = data.resources.slice(
+        i,
+        i + UPSERT_BATCH_SIZE,
+      );
+
+      const valueFragments: Array<string> = [];
+      const params: Array<unknown> = [];
+      let paramIndex: number = 1;
+
+      for (const r of chunk) {
+        const placeholders: Array<string> = [];
+        for (let c: number = 0; c < UPSERT_COLUMNS.length; c++) {
+          placeholders.push(`$${paramIndex++}`);
+        }
+        valueFragments.push(`(${placeholders.join(", ")})`);
+
+        params.push(
+          data.projectId.toString(),
+          data.kubernetesClusterId.toString(),
+          r.kind,
+          r.namespaceKey,
+          r.name,
+          r.uid,
+          r.phase,
+          r.isReady,
+          r.hasMemoryPressure,
+          r.hasDiskPressure,
+          r.hasPidPressure,
+          r.labels ? JSON.stringify(r.labels) : null,
+          r.annotations ? JSON.stringify(r.annotations) : null,
+          r.ownerReferences ? JSON.stringify(r.ownerReferences) : null,
+          r.spec ? JSON.stringify(r.spec) : null,
+          r.status ? JSON.stringify(r.status) : null,
+          r.lastSeenAt,
+          r.resourceCreationTimestamp,
+          0, // version (BaseModel @VersionColumn)
+        );
+      }
+
+      const sql: string = `
+        INSERT INTO "KubernetesResource" (
+          "projectId", "kubernetesClusterId", "kind", "namespaceKey", "name",
+          "uid", "phase", "isReady",
+          "hasMemoryPressure", "hasDiskPressure", "hasPidPressure",
+          "labels", "annotations", "ownerReferences", "spec", "status",
+          "lastSeenAt", "resourceCreationTimestamp", "version"
+        )
+        VALUES ${valueFragments.join(", ")}
+        ON CONFLICT ("projectId", "kubernetesClusterId", "kind", "namespaceKey", "name")
+        DO UPDATE SET
+          "uid" = EXCLUDED."uid",
+          "phase" = EXCLUDED."phase",
+          "isReady" = EXCLUDED."isReady",
+          "hasMemoryPressure" = EXCLUDED."hasMemoryPressure",
+          "hasDiskPressure" = EXCLUDED."hasDiskPressure",
+          "hasPidPressure" = EXCLUDED."hasPidPressure",
+          "labels" = EXCLUDED."labels",
+          "annotations" = EXCLUDED."annotations",
+          "ownerReferences" = EXCLUDED."ownerReferences",
+          "spec" = EXCLUDED."spec",
+          "status" = EXCLUDED."status",
+          "lastSeenAt" = EXCLUDED."lastSeenAt",
+          "resourceCreationTimestamp" = EXCLUDED."resourceCreationTimestamp",
+          "updatedAt" = now()
+        WHERE EXCLUDED."lastSeenAt" >= "KubernetesResource"."lastSeenAt"
+      `;
+
+      await this.getRepository().manager.query(sql, params);
+    }
+  }
+
+  /**
+   * Hard-delete all resources in a cluster whose last snapshot is
+   * older than olderThan. Returns the number of deleted rows.
+   */
+  @CaptureSpan()
+  public async deleteStaleForCluster(data: {
+    kubernetesClusterId: ObjectID;
+    olderThan: Date;
+  }): Promise<number> {
+    const result: Array<{ affected?: number }> | { affected?: number } =
+      await this.getRepository().manager.query(
+        `DELETE FROM "KubernetesResource" WHERE "kubernetesClusterId" = $1 AND "lastSeenAt" < $2`,
+        [data.kubernetesClusterId.toString(), data.olderThan],
+      );
+
+    // Postgres driver returns [rows, affected] for DELETE — normalize.
+    let affected: number = 0;
+    if (Array.isArray(result) && result.length >= 2) {
+      const second: unknown = (result as Array<unknown>)[1];
+      if (typeof second === "number") {
+        affected = second;
+      }
+    }
+
+    if (affected > STALE_DELETE_WARN_THRESHOLD) {
+      logger.warn(
+        `KubernetesResource cleanup deleted ${affected} stale rows for cluster ${data.kubernetesClusterId.toString()} — larger than expected; investigate agent health.`,
+      );
+    }
+
+    return affected;
+  }
+
+  /**
+   * Compute the overview-page summary in Postgres. Returns counts per
+   * resource kind plus pod phase / node condition breakdowns in a
+   * single round-trip.
+   */
+  @CaptureSpan()
+  public async getInventorySummary(data: {
+    projectId: ObjectID;
+    kubernetesClusterId: ObjectID;
+  }): Promise<InventorySummary> {
+    const manager: ReturnType<Service["getRepository"]>["manager"] =
+      this.getRepository().manager;
+
+    const [kindRows, podRows, nodeRows, containerRows]: [
+      Array<{ kind: string; count: string }>,
+      Array<{ phase: string | null; count: string }>,
+      Array<{
+        ready: string;
+        notReady: string;
+        memoryPressure: string;
+        diskPressure: string;
+        pidPressure: string;
+      }>,
+      Array<{ total: string }>,
+    ] = await Promise.all([
+      manager.query(
+        `SELECT "kind", COUNT(*)::text AS count
+         FROM "KubernetesResource"
+         WHERE "projectId" = $1 AND "kubernetesClusterId" = $2 AND "deletedAt" IS NULL
+         GROUP BY "kind"`,
+        [data.projectId.toString(), data.kubernetesClusterId.toString()],
+      ),
+      manager.query(
+        `SELECT "phase", COUNT(*)::text AS count
+         FROM "KubernetesResource"
+         WHERE "projectId" = $1 AND "kubernetesClusterId" = $2 AND "kind" = 'Pod' AND "deletedAt" IS NULL
+         GROUP BY "phase"`,
+        [data.projectId.toString(), data.kubernetesClusterId.toString()],
+      ),
+      manager.query(
+        `SELECT
+           COUNT(*) FILTER (WHERE "isReady" IS TRUE)::text AS "ready",
+           COUNT(*) FILTER (WHERE "isReady" IS FALSE)::text AS "notReady",
+           COUNT(*) FILTER (WHERE "hasMemoryPressure" IS TRUE)::text AS "memoryPressure",
+           COUNT(*) FILTER (WHERE "hasDiskPressure" IS TRUE)::text AS "diskPressure",
+           COUNT(*) FILTER (WHERE "hasPidPressure" IS TRUE)::text AS "pidPressure"
+         FROM "KubernetesResource"
+         WHERE "projectId" = $1 AND "kubernetesClusterId" = $2 AND "kind" = 'Node' AND "deletedAt" IS NULL`,
+        [data.projectId.toString(), data.kubernetesClusterId.toString()],
+      ),
+      manager.query(
+        `SELECT COALESCE(SUM(
+           CASE WHEN jsonb_typeof("spec"->'containers') = 'array'
+                THEN jsonb_array_length("spec"->'containers')
+                ELSE 0 END
+         ), 0)::text AS total
+         FROM "KubernetesResource"
+         WHERE "projectId" = $1 AND "kubernetesClusterId" = $2 AND "kind" = 'Pod' AND "deletedAt" IS NULL`,
+        [data.projectId.toString(), data.kubernetesClusterId.toString()],
+      ),
+    ]);
+
+    const countsByKind: Record<string, number> = {};
+    for (const row of kindRows) {
+      countsByKind[row.kind] = parseInt(row.count, 10) || 0;
+    }
+
+    const podPhaseCounts: InventorySummary["podPhaseCounts"] = {
+      running: 0,
+      pending: 0,
+      failed: 0,
+      succeeded: 0,
+      unknown: 0,
+    };
+    for (const row of podRows) {
+      const phase: string = row.phase || "Unknown";
+      const count: number = parseInt(row.count, 10) || 0;
+      if (phase === "Running") {
+        podPhaseCounts.running = count;
+      } else if (phase === "Pending") {
+        podPhaseCounts.pending = count;
+      } else if (phase === "Failed") {
+        podPhaseCounts.failed = count;
+      } else if (phase === "Succeeded") {
+        podPhaseCounts.succeeded = count;
+      } else {
+        podPhaseCounts.unknown += count;
+      }
+    }
+
+    const nodeRow:
+      | {
+          ready: string;
+          notReady: string;
+          memoryPressure: string;
+          diskPressure: string;
+          pidPressure: string;
+        }
+      | undefined = nodeRows[0];
+
+    const containerCount: number =
+      parseInt(containerRows[0]?.total || "0", 10) || 0;
+
+    return {
+      countsByKind,
+      containerCount,
+      podPhaseCounts,
+      nodeReadyCounts: {
+        ready: parseInt(nodeRow?.ready || "0", 10) || 0,
+        notReady: parseInt(nodeRow?.notReady || "0", 10) || 0,
+      },
+      nodePressureCounts: {
+        memoryPressure: parseInt(nodeRow?.memoryPressure || "0", 10) || 0,
+        diskPressure: parseInt(nodeRow?.diskPressure || "0", 10) || 0,
+        pidPressure: parseInt(nodeRow?.pidPressure || "0", 10) || 0,
+      },
+    };
+  }
+
+  /**
+   * Helper for the cleanup worker: snapshot-interval aware cutoff.
+   * 3× the 5-minute snapshot interval. Tune via CLEANUP_THRESHOLD_MINUTES.
+   */
+  public getStaleThresholdDate(nowOverride?: Date): Date {
+    const minutes: number = this.getStaleThresholdMinutes();
+    return OneUptimeDate.addRemoveMinutes(
+      nowOverride || OneUptimeDate.getCurrentDate(),
+      -minutes,
+    );
+  }
+
+  public getStaleThresholdMinutes(): number {
+    const raw: string | undefined = process.env["K8S_INVENTORY_STALE_MINUTES"];
+    if (raw) {
+      const parsed: number = parseInt(raw, 10);
+      if (!isNaN(parsed) && parsed >= 5) {
+        return parsed;
+      }
+    }
+    return 15;
+  }
+}
+
+export default new Service();

package/Server/Services/MetricPipelineRuleService.ts

@@ -0,0 +1,10 @@
+import DatabaseService from "./DatabaseService";
+import Model from "../../Models/DatabaseModels/MetricPipelineRule";
+
+export class Service extends DatabaseService<Model> {
+  public constructor() {
+    super(Model);
+  }
+}
+
+export default new Service();

package/Server/Services/MetricRecordingRuleService.ts

@@ -0,0 +1,10 @@
+import DatabaseService from "./DatabaseService";
+import Model from "../../Models/DatabaseModels/MetricRecordingRule";
+
+export class Service extends DatabaseService<Model> {
+  public constructor() {
+    super(Model);
+  }
+}
+
+export default new Service();

package/Server/Services/TraceDropFilterService.ts

@@ -0,0 +1,10 @@
+import DatabaseService from "./DatabaseService";
+import Model from "../../Models/DatabaseModels/TraceDropFilter";
+
+export class Service extends DatabaseService<Model> {
+  public constructor() {
+    super(Model);
+  }
+}
+
+export default new Service();

package/Server/Services/TracePipelineProcessorService.ts

@@ -0,0 +1,10 @@
+import DatabaseService from "./DatabaseService";
+import Model from "../../Models/DatabaseModels/TracePipelineProcessor";
+
+export class Service extends DatabaseService<Model> {
+  public constructor() {
+    super(Model);
+  }
+}
+
+export default new Service();

package/Server/Services/TracePipelineService.ts

@@ -0,0 +1,10 @@
+import DatabaseService from "./DatabaseService";
+import Model from "../../Models/DatabaseModels/TracePipeline";
+
+export class Service extends DatabaseService<Model> {
+  public constructor() {
+    super(Model);
+  }
+}
+
+export default new Service();

package/Server/Services/TraceRecordingRuleService.ts

@@ -0,0 +1,10 @@
+import DatabaseService from "./DatabaseService";
+import Model from "../../Models/DatabaseModels/TraceRecordingRule";
+
+export class Service extends DatabaseService<Model> {
+  public constructor() {
+    super(Model);
+  }
+}
+
+export default new Service();

package/Server/Services/TraceScrubRuleService.ts

@@ -0,0 +1,10 @@
+import DatabaseService from "./DatabaseService";
+import Model from "../../Models/DatabaseModels/TraceScrubRule";
+
+export class Service extends DatabaseService<Model> {
+  public constructor() {
+    super(Model);
+  }
+}
+
+export default new Service();

package/Server/Utils/Monitor/Criteria/CompareCriteria.ts

@@ -389,6 +389,8 @@ export default class CompareCriteria {
     value: Array<number> | number;
     threshold: number;
     criteriaFilter: CriteriaFilter;
+    metricDisplayName?: string | undefined;
+    unit?: string | undefined;
   }): string | null {
     if (data.value === null || data.value === undefined) {
       return null;
@@ -412,6 +414,8 @@ export default class CompareCriteria {
           values: data.value,
           threshold: data.threshold as number,
           criteriaFilter: data.criteriaFilter,
+          metricDisplayName: data.metricDisplayName,
+          unit: data.unit,
         });
       }
 
@@ -432,6 +436,8 @@ export default class CompareCriteria {
           values: data.value,
           threshold: data.threshold as number,
           criteriaFilter: data.criteriaFilter,
+          metricDisplayName: data.metricDisplayName,
+          unit: data.unit,
         });
       }
 
@@ -452,6 +458,8 @@ export default class CompareCriteria {
           values: data.value,
           threshold: data.threshold as number,
           criteriaFilter: data.criteriaFilter,
+          metricDisplayName: data.metricDisplayName,
+          unit: data.unit,
         });
       }
 
@@ -472,6 +480,8 @@ export default class CompareCriteria {
           values: data.value,
           threshold: data.threshold as number,
           criteriaFilter: data.criteriaFilter,
+          metricDisplayName: data.metricDisplayName,
+          unit: data.unit,
         });
       }
 
@@ -492,6 +502,8 @@ export default class CompareCriteria {
           values: data.value,
           threshold: data.threshold as number,
           criteriaFilter: data.criteriaFilter,
+          metricDisplayName: data.metricDisplayName,
+          unit: data.unit,
         });
       }
 
@@ -512,6 +524,8 @@ export default class CompareCriteria {
           values: data.value,
           threshold: data.threshold as number,
           criteriaFilter: data.criteriaFilter,
+          metricDisplayName: data.metricDisplayName,
+          unit: data.unit,
         });
       }
 
@@ -526,6 +540,8 @@ export default class CompareCriteria {
     values: Array<number | boolean> | number | boolean | string;
     threshold: number | string | boolean;
     criteriaFilter: CriteriaFilter;
+    metricDisplayName?: string | undefined;
+    unit?: string | undefined;
   }): string {
     // CPU Percent over the last 5 minutes is 10 which is less than the threshold of 20
     let message: string = "";
@@ -546,7 +562,17 @@ export default class CompareCriteria {
       message += "All values of";
     }
 
-    message += ` ${data.criteriaFilter.checkOn}`;
+    /*
+     * Prefer a metric-specific display name over the generic "Metric Value"
+     * label when evaluating metric monitors.
+     */
+    const label: string =
+      data.metricDisplayName &&
+      data.criteriaFilter.checkOn === CheckOn.MetricValue
+        ? data.metricDisplayName
+        : data.criteriaFilter.checkOn;
+
+    message += ` ${label}`;
 
     if (data.criteriaFilter.checkOn === CheckOn.DiskUsagePercent) {
       const diskPath: string =
@@ -562,6 +588,8 @@ export default class CompareCriteria {
       message += ` over the last ${data.criteriaFilter.evaluateOverTimeOptions.timeValueInMinutes} minutes`;
     }
 
+    const unitSuffix: string = data.unit ? ` ${data.unit}` : "";
+
     if (
       data.criteriaFilter.filterType !== FilterType.True &&
       data.criteriaFilter.filterType !== FilterType.False
@@ -570,29 +598,29 @@ export default class CompareCriteria {
         data.values,
       );
 
-      message += ` is ${formattedValues}`;
+      message += ` is ${formattedValues}${unitSuffix}`;
 
       message += " which is";
     }
 
     switch (data.criteriaFilter.filterType) {
       case FilterType.GreaterThan:
-        message += ` greater than ${CompareCriteria.formatSingleValue(data.threshold)}. `;
+        message += ` greater than ${CompareCriteria.formatSingleValue(data.threshold)}${unitSuffix}. `;
         break;
       case FilterType.GreaterThanOrEqualTo:
-        message += ` greater than or equal to ${CompareCriteria.formatSingleValue(data.threshold)}. `;
+        message += ` greater than or equal to ${CompareCriteria.formatSingleValue(data.threshold)}${unitSuffix}. `;
         break;
       case FilterType.LessThan:
-        message += ` less than ${CompareCriteria.formatSingleValue(data.threshold)}. `;
+        message += ` less than ${CompareCriteria.formatSingleValue(data.threshold)}${unitSuffix}. `;
         break;
       case FilterType.LessThanOrEqualTo:
-        message += ` less than or equal to ${CompareCriteria.formatSingleValue(data.threshold)}. `;
+        message += ` less than or equal to ${CompareCriteria.formatSingleValue(data.threshold)}${unitSuffix}. `;
         break;
       case FilterType.NotEqualTo:
-        message += ` not equal to ${CompareCriteria.formatSingleValue(data.threshold)}. `;
+        message += ` not equal to ${CompareCriteria.formatSingleValue(data.threshold)}${unitSuffix}. `;
         break;
       case FilterType.EqualTo:
-        message += ` equal to ${CompareCriteria.formatSingleValue(data.threshold)}. `;
+        message += ` equal to ${CompareCriteria.formatSingleValue(data.threshold)}${unitSuffix}. `;
         break;
       case FilterType.Contains:
         message += ` contains ${CompareCriteria.formatSingleValue(data.threshold)}. `;
@@ -621,11 +649,45 @@ export default class CompareCriteria {
     values: Array<number | boolean> | number | boolean | string,
   ): string {
     if (Array.isArray(values)) {
-      return values
+      /*
+       * For a small number of values, list them verbatim so the message
+       * reads naturally ("is 42, 55, 60"). For larger arrays, summarize
+       * — dumping 30+ numbers on one line makes the root cause unreadable
+       * and the detailed breakdown is shown in the Breaching Samples
+       * table below.
+       */
+      const MAX_INLINE: number = 5;
+
+      if (values.length <= MAX_INLINE) {
+        return values
+          .map((value: number | boolean) => {
+            return CompareCriteria.formatSingleValue(value);
+          })
+          .join(", ");
+      }
+
+      const numericValues: Array<number> = values.filter(
+        (value: number | boolean): value is number => {
+          return typeof value === "number" && Number.isFinite(value);
+        },
+      );
+
+      if (numericValues.length === values.length && numericValues.length > 0) {
+        const min: number = Math.min(...numericValues);
+        const max: number = Math.max(...numericValues);
+        return `${numericValues.length} samples between ${CompareCriteria.formatSingleValue(
+          min,
+        )} and ${CompareCriteria.formatSingleValue(max)}`;
+      }
+
+      // Fall back to a truncated list when not all values are numeric
+      const head: string = values
+        .slice(0, MAX_INLINE)
         .map((value: number | boolean) => {
           return CompareCriteria.formatSingleValue(value);
         })
         .join(", ");
+      return `${head}, … (${values.length} values total)`;
     }
 
     return CompareCriteria.formatSingleValue(values);