@oneuptime/common 10.4.10 → 10.4.12

package/Models/DatabaseModels/Incident.ts

@@ -1,13 +1,19 @@
+import DockerHost from "./DockerHost";
+import DockerResource from "./DockerResource";
 import Host from "./Host";
 import IncidentEpisode from "./IncidentEpisode";
 import IncidentSeverity from "./IncidentSeverity";
 import IncidentState from "./IncidentState";
+import KubernetesCluster from "./KubernetesCluster";
+import KubernetesContainer from "./KubernetesContainer";
+import KubernetesResource from "./KubernetesResource";
 import Label from "./Label";
 import Monitor from "./Monitor";
 import MonitorStatus from "./MonitorStatus";
 import OnCallDutyPolicy from "./OnCallDutyPolicy";
 import Probe from "./Probe";
 import Project from "./Project";
+import Service from "./Service";
 import User from "./User";
 import File from "./File";
 import BaseModel from "./DatabaseBaseModel/DatabaseBaseModel";
@@ -588,6 +594,332 @@ export default class Incident extends BaseModel {
   })
   public hosts?: Array<Host> = undefined; // hosts affected by this incident.
 
+  @ColumnAccessControl({
+    create: [
+      Permission.ProjectOwner,
+      Permission.ProjectAdmin,
+      Permission.ProjectMember,
+      Permission.IncidentAdmin,
+      Permission.IncidentMember,
+      Permission.CreateProjectIncident,
+    ],
+    read: [
+      Permission.ProjectOwner,
+      Permission.ProjectAdmin,
+      Permission.ProjectMember,
+      Permission.Viewer,
+      Permission.IncidentAdmin,
+      Permission.IncidentMember,
+      Permission.IncidentViewer,
+      Permission.ReadProjectIncident,
+    ],
+    update: [
+      Permission.ProjectOwner,
+      Permission.ProjectAdmin,
+      Permission.ProjectMember,
+      Permission.IncidentAdmin,
+      Permission.IncidentMember,
+      Permission.EditProjectIncident,
+    ],
+  })
+  @TableColumn({
+    required: false,
+    type: TableColumnType.EntityArray,
+    modelType: KubernetesCluster,
+    title: "Kubernetes Clusters",
+    description: "List of Kubernetes clusters affected by this incident.",
+  })
+  @ManyToMany(
+    () => {
+      return KubernetesCluster;
+    },
+    { eager: false },
+  )
+  @JoinTable({
+    name: "IncidentKubernetesCluster",
+    inverseJoinColumn: {
+      name: "kubernetesClusterId",
+      referencedColumnName: "_id",
+    },
+    joinColumn: {
+      name: "incidentId",
+      referencedColumnName: "_id",
+    },
+  })
+  public kubernetesClusters?: Array<KubernetesCluster> = undefined;
+
+  @ColumnAccessControl({
+    create: [
+      Permission.ProjectOwner,
+      Permission.ProjectAdmin,
+      Permission.ProjectMember,
+      Permission.IncidentAdmin,
+      Permission.IncidentMember,
+      Permission.CreateProjectIncident,
+    ],
+    read: [
+      Permission.ProjectOwner,
+      Permission.ProjectAdmin,
+      Permission.ProjectMember,
+      Permission.Viewer,
+      Permission.IncidentAdmin,
+      Permission.IncidentMember,
+      Permission.IncidentViewer,
+      Permission.ReadProjectIncident,
+    ],
+    update: [
+      Permission.ProjectOwner,
+      Permission.ProjectAdmin,
+      Permission.ProjectMember,
+      Permission.IncidentAdmin,
+      Permission.IncidentMember,
+      Permission.EditProjectIncident,
+    ],
+  })
+  @TableColumn({
+    required: false,
+    type: TableColumnType.EntityArray,
+    modelType: KubernetesResource,
+    title: "Kubernetes Resources",
+    description:
+      "List of Kubernetes resources (pods, deployments, nodes, etc.) affected by this incident.",
+  })
+  @ManyToMany(
+    () => {
+      return KubernetesResource;
+    },
+    { eager: false },
+  )
+  @JoinTable({
+    name: "IncidentKubernetesResource",
+    inverseJoinColumn: {
+      name: "kubernetesResourceId",
+      referencedColumnName: "_id",
+    },
+    joinColumn: {
+      name: "incidentId",
+      referencedColumnName: "_id",
+    },
+  })
+  public kubernetesResources?: Array<KubernetesResource> = undefined;
+
+  @ColumnAccessControl({
+    create: [
+      Permission.ProjectOwner,
+      Permission.ProjectAdmin,
+      Permission.ProjectMember,
+      Permission.IncidentAdmin,
+      Permission.IncidentMember,
+      Permission.CreateProjectIncident,
+    ],
+    read: [
+      Permission.ProjectOwner,
+      Permission.ProjectAdmin,
+      Permission.ProjectMember,
+      Permission.Viewer,
+      Permission.IncidentAdmin,
+      Permission.IncidentMember,
+      Permission.IncidentViewer,
+      Permission.ReadProjectIncident,
+    ],
+    update: [
+      Permission.ProjectOwner,
+      Permission.ProjectAdmin,
+      Permission.ProjectMember,
+      Permission.IncidentAdmin,
+      Permission.IncidentMember,
+      Permission.EditProjectIncident,
+    ],
+  })
+  @TableColumn({
+    required: false,
+    type: TableColumnType.EntityArray,
+    modelType: KubernetesContainer,
+    title: "Kubernetes Containers",
+    description: "List of Kubernetes containers affected by this incident.",
+  })
+  @ManyToMany(
+    () => {
+      return KubernetesContainer;
+    },
+    { eager: false },
+  )
+  @JoinTable({
+    name: "IncidentKubernetesContainer",
+    inverseJoinColumn: {
+      name: "kubernetesContainerId",
+      referencedColumnName: "_id",
+    },
+    joinColumn: {
+      name: "incidentId",
+      referencedColumnName: "_id",
+    },
+  })
+  public kubernetesContainers?: Array<KubernetesContainer> = undefined;
+
+  @ColumnAccessControl({
+    create: [
+      Permission.ProjectOwner,
+      Permission.ProjectAdmin,
+      Permission.ProjectMember,
+      Permission.IncidentAdmin,
+      Permission.IncidentMember,
+      Permission.CreateProjectIncident,
+    ],
+    read: [
+      Permission.ProjectOwner,
+      Permission.ProjectAdmin,
+      Permission.ProjectMember,
+      Permission.Viewer,
+      Permission.IncidentAdmin,
+      Permission.IncidentMember,
+      Permission.IncidentViewer,
+      Permission.ReadProjectIncident,
+    ],
+    update: [
+      Permission.ProjectOwner,
+      Permission.ProjectAdmin,
+      Permission.ProjectMember,
+      Permission.IncidentAdmin,
+      Permission.IncidentMember,
+      Permission.EditProjectIncident,
+    ],
+  })
+  @TableColumn({
+    required: false,
+    type: TableColumnType.EntityArray,
+    modelType: DockerHost,
+    title: "Docker Hosts",
+    description: "List of Docker hosts affected by this incident.",
+  })
+  @ManyToMany(
+    () => {
+      return DockerHost;
+    },
+    { eager: false },
+  )
+  @JoinTable({
+    name: "IncidentDockerHost",
+    inverseJoinColumn: {
+      name: "dockerHostId",
+      referencedColumnName: "_id",
+    },
+    joinColumn: {
+      name: "incidentId",
+      referencedColumnName: "_id",
+    },
+  })
+  public dockerHosts?: Array<DockerHost> = undefined;
+
+  @ColumnAccessControl({
+    create: [
+      Permission.ProjectOwner,
+      Permission.ProjectAdmin,
+      Permission.ProjectMember,
+      Permission.IncidentAdmin,
+      Permission.IncidentMember,
+      Permission.CreateProjectIncident,
+    ],
+    read: [
+      Permission.ProjectOwner,
+      Permission.ProjectAdmin,
+      Permission.ProjectMember,
+      Permission.Viewer,
+      Permission.IncidentAdmin,
+      Permission.IncidentMember,
+      Permission.IncidentViewer,
+      Permission.ReadProjectIncident,
+    ],
+    update: [
+      Permission.ProjectOwner,
+      Permission.ProjectAdmin,
+      Permission.ProjectMember,
+      Permission.IncidentAdmin,
+      Permission.IncidentMember,
+      Permission.EditProjectIncident,
+    ],
+  })
+  @TableColumn({
+    required: false,
+    type: TableColumnType.EntityArray,
+    modelType: DockerResource,
+    title: "Docker Resources",
+    description:
+      "List of Docker resources (containers, images, networks, volumes) affected by this incident.",
+  })
+  @ManyToMany(
+    () => {
+      return DockerResource;
+    },
+    { eager: false },
+  )
+  @JoinTable({
+    name: "IncidentDockerResource",
+    inverseJoinColumn: {
+      name: "dockerResourceId",
+      referencedColumnName: "_id",
+    },
+    joinColumn: {
+      name: "incidentId",
+      referencedColumnName: "_id",
+    },
+  })
+  public dockerResources?: Array<DockerResource> = undefined;
+
+  @ColumnAccessControl({
+    create: [
+      Permission.ProjectOwner,
+      Permission.ProjectAdmin,
+      Permission.ProjectMember,
+      Permission.IncidentAdmin,
+      Permission.IncidentMember,
+      Permission.CreateProjectIncident,
+    ],
+    read: [
+      Permission.ProjectOwner,
+      Permission.ProjectAdmin,
+      Permission.ProjectMember,
+      Permission.Viewer,
+      Permission.IncidentAdmin,
+      Permission.IncidentMember,
+      Permission.IncidentViewer,
+      Permission.ReadProjectIncident,
+    ],
+    update: [
+      Permission.ProjectOwner,
+      Permission.ProjectAdmin,
+      Permission.ProjectMember,
+      Permission.IncidentAdmin,
+      Permission.IncidentMember,
+      Permission.EditProjectIncident,
+    ],
+  })
+  @TableColumn({
+    required: false,
+    type: TableColumnType.EntityArray,
+    modelType: Service,
+    title: "Services",
+    description: "List of services affected by this incident.",
+  })
+  @ManyToMany(
+    () => {
+      return Service;
+    },
+    { eager: false },
+  )
+  @JoinTable({
+    name: "IncidentService",
+    inverseJoinColumn: {
+      name: "serviceId",
+      referencedColumnName: "_id",
+    },
+    joinColumn: {
+      name: "incidentId",
+      referencedColumnName: "_id",
+    },
+  })
+  public services?: Array<Service> = undefined;
+
   @ColumnAccessControl({
     create: [
       Permission.ProjectOwner,

package/Models/DatabaseModels/Index.ts

@@ -226,7 +226,6 @@ import RunbookExecution from "./RunbookExecution";
 import RunbookOwnerTeam from "./RunbookOwnerTeam";
 import RunbookOwnerUser from "./RunbookOwnerUser";
 import RunbookRule from "./RunbookRule";
-import ServiceDependency from "./ServiceDependency";
 import ServiceMonitor from "./ServiceMonitor";
 
 import UserTotpAuth from "./UserTotpAuth";
@@ -618,7 +617,6 @@ const AllModelTypes: Array<{
   Service,
   ServiceOwnerTeam,
   ServiceOwnerUser,
-  ServiceDependency,
   ServiceMonitor,
   ServiceCodeRepository,
 

package/Models/DatabaseModels/KubernetesCluster.ts

@@ -19,6 +19,7 @@ import UniqueColumnBy from "../../Types/Database/UniqueColumnBy";
 import IconProp from "../../Types/Icon/IconProp";
 import ObjectID from "../../Types/ObjectID";
 import Permission from "../../Types/Permission";
+import TelemetryRetentionConfig from "../../Types/Telemetry/TelemetryRetentionConfig";
 import {
   Column,
   Entity,
@@ -725,4 +726,86 @@ export default class KubernetesCluster extends BaseModel {
     },
   })
   public labels?: Array<Label> = undefined;
+
+  @ColumnAccessControl({
+    create: [
+      Permission.ProjectOwner,
+      Permission.ProjectAdmin,
+      Permission.ProjectMember,
+      Permission.SettingsAdmin,
+      Permission.SettingsMember,
+      Permission.CreateKubernetesCluster,
+    ],
+    read: [
+      Permission.ProjectOwner,
+      Permission.ProjectAdmin,
+      Permission.ProjectMember,
+      Permission.Viewer,
+      Permission.SettingsAdmin,
+      Permission.SettingsMember,
+      Permission.SettingsViewer,
+      Permission.ReadKubernetesCluster,
+    ],
+    update: [
+      Permission.ProjectOwner,
+      Permission.ProjectAdmin,
+      Permission.ProjectMember,
+      Permission.SettingsAdmin,
+      Permission.SettingsMember,
+      Permission.EditKubernetesCluster,
+    ],
+  })
+  @TableColumn({
+    type: TableColumnType.Number,
+    title: "Retain Telemetry Data For Days",
+    description:
+      "Number of days to retain telemetry data for this Kubernetes cluster. Leave blank to use the project-wide default.",
+  })
+  @Column({
+    type: ColumnType.Number,
+    nullable: true,
+    unique: false,
+  })
+  public retainTelemetryDataForDays?: number = undefined;
+
+  @ColumnAccessControl({
+    create: [
+      Permission.ProjectOwner,
+      Permission.ProjectAdmin,
+      Permission.ProjectMember,
+      Permission.SettingsAdmin,
+      Permission.SettingsMember,
+      Permission.CreateKubernetesCluster,
+    ],
+    read: [
+      Permission.ProjectOwner,
+      Permission.ProjectAdmin,
+      Permission.ProjectMember,
+      Permission.Viewer,
+      Permission.SettingsAdmin,
+      Permission.SettingsMember,
+      Permission.SettingsViewer,
+      Permission.ReadKubernetesCluster,
+    ],
+    update: [
+      Permission.ProjectOwner,
+      Permission.ProjectAdmin,
+      Permission.ProjectMember,
+      Permission.SettingsAdmin,
+      Permission.SettingsMember,
+      Permission.EditKubernetesCluster,
+    ],
+  })
+  @TableColumn({
+    type: TableColumnType.JSON,
+    required: false,
+    title: "Telemetry Data Retention Overrides",
+    description:
+      "Per-pillar retention overrides for this Kubernetes cluster (logs by severity, traces by status, metrics, profiles). Unset fields fall back to the cluster default, then the project's retention settings.",
+  })
+  @Column({
+    type: ColumnType.JSON,
+    nullable: true,
+  })
+  public telemetryRetentionConfig?: TelemetryRetentionConfig = undefined;
 }

package/Models/DatabaseModels/RunbookAgent.ts

@@ -313,7 +313,14 @@ export default class RunbookAgent extends BaseModel {
   public key?: string = undefined;
 
   @ColumnAccessControl({
-    create: [],
+    create: [
+      Permission.ProjectOwner,
+      Permission.ProjectAdmin,
+      Permission.ProjectMember,
+      Permission.RunbookAdmin,
+      Permission.RunbookMember,
+      Permission.CreateRunbookAgent,
+    ],
     read: [
       Permission.ProjectOwner,
       Permission.ProjectAdmin,
@@ -369,7 +376,14 @@ export default class RunbookAgent extends BaseModel {
   public lastAlive?: Date = undefined;
 
   @ColumnAccessControl({
-    create: [],
+    create: [
+      Permission.ProjectOwner,
+      Permission.ProjectAdmin,
+      Permission.ProjectMember,
+      Permission.RunbookAdmin,
+      Permission.RunbookMember,
+      Permission.CreateRunbookAgent,
+    ],
     read: [
       Permission.ProjectOwner,
       Permission.ProjectAdmin,

package/Server/EnvironmentConfig.ts

@@ -285,6 +285,22 @@ export const ShouldClickhouseSslEnable: boolean = Boolean(
   ClickhouseTlsCa || (ClickhouseTlsCert && ClickhouseTlsKey),
 );
 
+export const MaxClickhouseConnections: number = parseInt(
+  process.env["CLICKHOUSE_MAX_OPEN_CONNECTIONS"] || "100",
+  10,
+);
+
+/*
+ * Ingest pool size. Falls back to MaxClickhouseConnections so single-knob
+ * setups still work; override only when the ingest pool needs to be sized
+ * independently from the query pool.
+ */
+export const MaxClickhouseIngestConnections: number = parseInt(
+  process.env["CLICKHOUSE_INGEST_MAX_OPEN_CONNECTIONS"] ||
+    String(MaxClickhouseConnections),
+  10,
+);
+
 export const GitSha: string = process.env["GIT_SHA"] || "unknown";
 
 export const AppVersion: string = process.env["APP_VERSION"] || "unknown";

package/Server/Infrastructure/ClickhouseConfig.ts

@@ -9,6 +9,8 @@ import {
   ClickhouseTlsCert,
   ClickhouseTlsKey,
   ClickhouseUsername,
+  MaxClickhouseConnections,
+  MaxClickhouseIngestConnections,
   ShouldClickhouseSslEnable,
 } from "../EnvironmentConfig";
 import Hostname from "../../Types/API/Hostname";
@@ -35,6 +37,13 @@ const options: ClickHouseClientConfigOptions = {
    * aggregation statements provides the hard server-side cap.
    */
   request_timeout: 58_000,
+  /*
+   * @clickhouse/client defaults max_open_connections to 10. Sized for the
+   * query pool (dashboard reads, DDL); ingest writes use a separate pool
+   * (see ingestDataSourceOptions) so a burst of inserts cannot starve
+   * user-facing queries of HTTP sockets.
+   */
+  max_open_connections: MaxClickhouseConnections,
 };
 
 if (ShouldClickhouseSslEnable && ClickhouseTlsCa) {
@@ -58,5 +67,10 @@ if (
 
 export const dataSourceOptions: ClickHouseClientConfigOptions = options;
 
+export const ingestDataSourceOptions: ClickHouseClientConfigOptions = {
+  ...options,
+  max_open_connections: MaxClickhouseIngestConnections,
+};
+
 export const testDataSourceOptions: ClickHouseClientConfigOptions =
   dataSourceOptions;

package/Server/Infrastructure/ClickhouseDatabase.ts

@@ -2,6 +2,7 @@ import logger from "../Utils/Logger";
 import {
   ClickHouseClientConfigOptions,
   dataSourceOptions,
+  ingestDataSourceOptions,
   testDataSourceOptions,
 } from "./ClickhouseConfig";
 import { PingResult, createClient, ClickHouseClient } from "@clickhouse/client";
@@ -18,9 +19,16 @@ export type ClickhouseClient = ClickHouseClient;
 
 export default class ClickhouseDatabase {
   private dataSource!: ClickhouseClient | null;
+  private options: ClickHouseClientConfigOptions;
+
+  public constructor(
+    options: ClickHouseClientConfigOptions = dataSourceOptions,
+  ) {
+    this.options = options;
+  }
 
   public getDatasourceOptions(): ClickHouseClientConfigOptions {
-    return dataSourceOptions;
+    return this.options;
   }
 
   public getTestDatasourceOptions(): ClickHouseClientConfigOptions {
@@ -161,5 +169,14 @@ export default class ClickhouseDatabase {
   }
 }
 
-export const ClickhouseAppInstance: ClickhouseDatabase =
-  new ClickhouseDatabase();
+export const ClickhouseAppInstance: ClickhouseDatabase = new ClickhouseDatabase(
+  dataSourceOptions,
+);
+
+/*
+ * Separate pool for high-volume telemetry inserts. Reads/DDL keep using
+ * ClickhouseAppInstance so dashboard queries are not starved of HTTP
+ * sockets when ingest is bursting.
+ */
+export const ClickhouseIngestInstance: ClickhouseDatabase =
+  new ClickhouseDatabase(ingestDataSourceOptions);

package/Server/Infrastructure/InMemoryTTLCache.ts

@@ -0,0 +1,61 @@
+interface CacheEntry<T> {
+  value: T;
+  expiresAt: number;
+}
+
+/**
+ * Bounded, in-process TTL cache. Entries expire on read after their TTL and the
+ * oldest entry is evicted when capacity is reached (Map preserves insertion
+ * order, which we use as a coarse LRU on writes).
+ *
+ * Each process holds its own copy — there is no cross-process invalidation.
+ * Callers that need stronger consistency should pair the TTL with an explicit
+ * `delete()` on writes to the underlying data.
+ */
+export default class InMemoryTTLCache<T> {
+  private store: Map<string, CacheEntry<T>> = new Map();
+  private maxEntries: number;
+
+  public constructor(maxEntries: number = 10_000) {
+    this.maxEntries = maxEntries;
+  }
+
+  public set(key: string, value: T, ttlMs: number): void {
+    if (this.store.size >= this.maxEntries && !this.store.has(key)) {
+      const oldest: string | undefined = this.store.keys().next().value;
+      if (oldest !== undefined) {
+        this.store.delete(oldest);
+      }
+    }
+    this.store.delete(key);
+    this.store.set(key, { value, expiresAt: Date.now() + ttlMs });
+  }
+
+  public get(key: string): T | undefined {
+    const entry: CacheEntry<T> | undefined = this.store.get(key);
+    if (!entry) {
+      return undefined;
+    }
+    if (Date.now() > entry.expiresAt) {
+      this.store.delete(key);
+      return undefined;
+    }
+    return entry.value;
+  }
+
+  public has(key: string): boolean {
+    return this.get(key) !== undefined;
+  }
+
+  public delete(key: string): void {
+    this.store.delete(key);
+  }
+
+  public clear(): void {
+    this.store.clear();
+  }
+
+  public size(): number {
+    return this.store.size;
+  }
+}

package/Server/Infrastructure/Postgres/SchemaMigrations/1779199346010-AddTelemetryRetentionConfig.ts

@@ -3,7 +3,7 @@ import { MigrationInterface, QueryRunner } from "typeorm";
 export class AddTelemetryRetentionConfig1779199346010
   implements MigrationInterface
 {
-  name = "AddTelemetryRetentionConfig1779199346010";
+  public name: string = "AddTelemetryRetentionConfig1779199346010";
 
   public async up(queryRunner: QueryRunner): Promise<void> {
     await queryRunner.query(