@oneuptime/common 11.1.2 → 11.2.0

package/Server/Services/IoTFleetOwnerRuleEngineService.ts

@@ -0,0 +1,216 @@
+import Label from "../../Models/DatabaseModels/Label";
+import IoTFleet from "../../Models/DatabaseModels/IoTFleet";
+import IoTFleetOwnerRule from "../../Models/DatabaseModels/IoTFleetOwnerRule";
+import IoTFleetOwnerUser from "../../Models/DatabaseModels/IoTFleetOwnerUser";
+import IoTFleetOwnerTeam from "../../Models/DatabaseModels/IoTFleetOwnerTeam";
+import IoTFleetOwnerRuleService from "./IoTFleetOwnerRuleService";
+import IoTFleetOwnerUserService from "./IoTFleetOwnerUserService";
+import IoTFleetOwnerTeamService from "./IoTFleetOwnerTeamService";
+import IoTFleetService from "./IoTFleetService";
+import ObjectID from "../../Types/ObjectID";
+import CaptureSpan from "../Utils/Telemetry/CaptureSpan";
+import logger, { LogAttributes } from "../Utils/Logger";
+
+class IoTFleetOwnerRuleEngineServiceClass {
+  /**
+   * Evaluates IoTFleetOwnerRule rows for the given IoT fleet and adds matched
+   * owner users / teams via IoTFleetOwnerUserService / IoTFleetOwnerTeamService. Rules
+   * with notifyOwners set notify the added owners; rules with notifyOwners off
+   * add silently.
+   */
+  @CaptureSpan()
+  public async applyRulesToIoTFleet(iotFleet: IoTFleet): Promise<void> {
+    if (!iotFleet.id || !iotFleet.projectId) {
+      return;
+    }
+
+    try {
+      const rules: Array<IoTFleetOwnerRule> =
+        await IoTFleetOwnerRuleService.findBy({
+          query: {
+            projectId: iotFleet.projectId,
+            isEnabled: true,
+          },
+          props: { isRoot: true },
+          select: {
+            _id: true,
+            name: true,
+            notifyOwners: true,
+            iotFleetLabels: { _id: true },
+            iotFleetNamePattern: true,
+            iotFleetDescriptionPattern: true,
+            ownerUsers: { _id: true },
+            ownerTeams: { _id: true },
+          },
+          limit: 100,
+          skip: 0,
+        });
+
+      if (rules.length === 0) {
+        return;
+      }
+
+      const iotFleetWithDetails: IoTFleet | null =
+        await IoTFleetService.findOneById({
+          id: iotFleet.id,
+          select: {
+            name: true,
+            description: true,
+            labels: { _id: true },
+          },
+          props: { isRoot: true },
+        });
+
+      if (!iotFleetWithDetails) {
+        return;
+      }
+
+      const usersByNotify: Map<boolean, Set<string>> = new Map([
+        [true, new Set()],
+        [false, new Set()],
+      ]);
+      const teamsByNotify: Map<boolean, Set<string>> = new Map([
+        [true, new Set()],
+        [false, new Set()],
+      ]);
+
+      const matchedRules: Array<IoTFleetOwnerRule> = [];
+
+      for (const rule of rules) {
+        const matches: boolean = this.doesIoTFleetMatchRule(
+          iotFleetWithDetails,
+          rule,
+        );
+        if (!matches) {
+          continue;
+        }
+        let ruleAddedAny: boolean = false;
+        const notify: boolean = rule.notifyOwners !== false;
+        for (const user of rule.ownerUsers || []) {
+          if (user.id) {
+            usersByNotify.get(notify)!.add(user.id.toString());
+            ruleAddedAny = true;
+          }
+        }
+        for (const team of rule.ownerTeams || []) {
+          if (team.id) {
+            teamsByNotify.get(notify)!.add(team.id.toString());
+            ruleAddedAny = true;
+          }
+        }
+        if (ruleAddedAny) {
+          matchedRules.push(rule);
+        }
+      }
+
+      if (matchedRules.length === 0) {
+        return;
+      }
+
+      for (const notify of [true, false]) {
+        const userIds: Set<string> = usersByNotify.get(notify)!;
+        const teamIds: Set<string> = teamsByNotify.get(notify)!;
+
+        for (const userId of userIds) {
+          const owner: IoTFleetOwnerUser = new IoTFleetOwnerUser();
+          owner.iotFleetId = iotFleet.id;
+          owner.projectId = iotFleet.projectId;
+          owner.userId = new ObjectID(userId);
+          owner.isOwnerNotified = !notify;
+          await IoTFleetOwnerUserService.create({
+            data: owner,
+            props: { isRoot: true },
+          });
+        }
+
+        for (const teamId of teamIds) {
+          const owner: IoTFleetOwnerTeam = new IoTFleetOwnerTeam();
+          owner.iotFleetId = iotFleet.id;
+          owner.projectId = iotFleet.projectId;
+          owner.teamId = new ObjectID(teamId);
+          owner.isOwnerNotified = !notify;
+          await IoTFleetOwnerTeamService.create({
+            data: owner,
+            props: { isRoot: true },
+          });
+        }
+      }
+
+      logger.debug(
+        `IoTFleetOwnerRuleEngine added owners to IoT fleet ${iotFleet.id}`,
+        { projectId: iotFleet.projectId.toString() } as LogAttributes,
+      );
+    } catch (error) {
+      logger.error(`Error applying IoT fleet owner rules: ${error}`, {
+        projectId: iotFleet.projectId?.toString(),
+        iotFleetId: iotFleet.id?.toString(),
+      } as LogAttributes);
+    }
+  }
+
+  private doesIoTFleetMatchRule(
+    iotFleet: IoTFleet,
+    rule: IoTFleetOwnerRule,
+  ): boolean {
+    if (rule.iotFleetLabels && rule.iotFleetLabels.length > 0) {
+      if (!iotFleet.labels || iotFleet.labels.length === 0) {
+        return false;
+      }
+      const ruleLabelIds: Array<string> = rule.iotFleetLabels.map(
+        (l: Label) => {
+          return l.id?.toString() || "";
+        },
+      );
+      const labelIds: Array<string> = iotFleet.labels.map((l: Label) => {
+        return l.id?.toString() || "";
+      });
+      if (
+        !ruleLabelIds.some((id: string) => {
+          return labelIds.includes(id);
+        })
+      ) {
+        return false;
+      }
+    }
+
+    if (
+      rule.iotFleetNamePattern &&
+      (!iotFleet.name ||
+        !this.testRegex(rule.iotFleetNamePattern, iotFleet.name, rule))
+    ) {
+      return false;
+    }
+
+    if (
+      rule.iotFleetDescriptionPattern &&
+      (!iotFleet.description ||
+        !this.testRegex(
+          rule.iotFleetDescriptionPattern,
+          iotFleet.description,
+          rule,
+        ))
+    ) {
+      return false;
+    }
+
+    return true;
+  }
+
+  private testRegex(
+    pattern: string,
+    value: string,
+    rule: IoTFleetOwnerRule,
+  ): boolean {
+    try {
+      const regex: RegExp = new RegExp(pattern, "i");
+      return regex.test(value);
+    } catch {
+      logger.warn(
+        `Invalid regex in IoT fleet owner rule ${rule.id}: ${pattern}`,
+      );
+      return false;
+    }
+  }
+}
+
+export default new IoTFleetOwnerRuleEngineServiceClass();

package/Server/Services/IoTFleetOwnerRuleService.ts

@@ -0,0 +1,14 @@
+import DatabaseService from "./DatabaseService";
+import Model from "../../Models/DatabaseModels/IoTFleetOwnerRule";
+import { IsBillingEnabled } from "../EnvironmentConfig";
+
+export class Service extends DatabaseService<Model> {
+  public constructor() {
+    super(Model);
+    if (IsBillingEnabled) {
+      this.hardDeleteItemsOlderThanInDays("createdAt", 3 * 365);
+    }
+  }
+}
+
+export default new Service();

package/Server/Services/IoTFleetOwnerTeamService.ts

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

package/Server/Services/IoTFleetOwnerUserService.ts

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

package/Server/Services/IoTFleetService.ts

@@ -0,0 +1,376 @@
+import DatabaseService from "./DatabaseService";
+import IoTFleetLabelRuleEngineService from "./IoTFleetLabelRuleEngineService";
+import IoTFleetOwnerRuleEngineService from "./IoTFleetOwnerRuleEngineService";
+import Model from "../../Models/DatabaseModels/IoTFleet";
+import Label from "../../Models/DatabaseModels/Label";
+import { OnCreate } from "../Types/Database/Hooks";
+import CaptureSpan from "../Utils/Telemetry/CaptureSpan";
+import ObjectID from "../../Types/ObjectID";
+import QueryHelper from "../Types/Database/QueryHelper";
+import OneUptimeDate from "../../Types/Date";
+import LIMIT_MAX from "../../Types/Database/LimitMax";
+import GlobalCache from "../Infrastructure/GlobalCache";
+import logger, { LogAttributes } from "../Utils/Logger";
+import crypto from "crypto";
+
+const LAST_SEEN_CACHE_NAMESPACE: string = "iot-fleet-last-seen";
+const LAST_SEEN_THROTTLE_SECONDS: number = 60;
+
+const LABELS_APPLIED_CACHE_NAMESPACE: string = "iot-fleet-labels-applied";
+const LABELS_APPLIED_CACHE_TTL_SECONDS: number = 60;
+
+export class Service extends DatabaseService<Model> {
+  public constructor() {
+    super(Model);
+  }
+
+  @CaptureSpan()
+  protected override async onCreateSuccess(
+    _onCreate: OnCreate<Model>,
+    createdItem: Model,
+  ): Promise<Model> {
+    /*
+     * Rules run once, on creation only — exact parity with
+     * KubernetesClusterService. Label engine first: it syncs the
+     * in-memory labels so the owner engine can match rule-added labels.
+     */
+    if (createdItem.projectId && createdItem.id) {
+      Promise.resolve()
+        .then(async () => {
+          await IoTFleetLabelRuleEngineService.applyRulesToIoTFleet(
+            createdItem,
+          );
+        })
+        .then(async () => {
+          await IoTFleetOwnerRuleEngineService.applyRulesToIoTFleet(
+            createdItem,
+          );
+        })
+        .catch((error: Error) => {
+          logger.error(
+            `Error applying iot fleet rules in IoTFleetService.onCreateSuccess: ${error}`,
+            {
+              projectId: createdItem.projectId?.toString(),
+              iotFleetId: createdItem.id?.toString(),
+            } as LogAttributes,
+          );
+        });
+    }
+    return createdItem;
+  }
+
+  @CaptureSpan()
+  public async findOrCreateByName(data: {
+    projectId: ObjectID;
+    name: string;
+  }): Promise<Model> {
+    /*
+     * An IoT fleet is keyed by the `iot.fleet.name` OTel resource
+     * attribute, which the user configures on the agent. Look it up
+     * case-insensitively: the unique guard (checkUniqueColumnBy ->
+     * findWithSameText) compares case-insensitively, so a case-sensitive
+     * lookup would miss an existing row that differs only by case, then
+     * fail to create it — wedging ingest for that fleet. Unlike
+     * DockerHost.hostIdentifier (host.name casing is unstable on Windows)
+     * the configured fleet name's casing is stable, so we preserve the
+     * user's casing on create instead of canonicalizing to lowercase.
+     */
+    const name: string = data.name.trim();
+
+    const existingFleet: Model | null = await this.findOneBy({
+      query: {
+        projectId: data.projectId,
+        name: QueryHelper.findWithSameText(name),
+      },
+      select: {
+        _id: true,
+        projectId: true,
+        name: true,
+      },
+      props: {
+        isRoot: true,
+      },
+    });
+
+    if (existingFleet) {
+      return existingFleet;
+    }
+
+    try {
+      // Create new fleet
+      const newFleet: Model = new Model();
+      newFleet.projectId = data.projectId;
+      newFleet.name = name;
+      newFleet.otelCollectorStatus = "connected";
+      newFleet.lastSeenAt = OneUptimeDate.getCurrentDate();
+
+      const createdFleet: Model = await this.create({
+        data: newFleet,
+        props: {
+          isRoot: true,
+        },
+      });
+
+      return createdFleet;
+    } catch {
+      /*
+       * Either two ingest workers raced to create the same fleet, or a
+       * fleet with this name in a different case already existed and the
+       * unique guard rejected the insert. Re-resolve case-insensitively so
+       * the caller still gets the existing row instead of throwing.
+       */
+      const reFetchedFleet: Model | null = await this.findOneBy({
+        query: {
+          projectId: data.projectId,
+          name: QueryHelper.findWithSameText(name),
+        },
+        select: {
+          _id: true,
+          projectId: true,
+          name: true,
+        },
+        props: {
+          isRoot: true,
+        },
+      });
+
+      if (reFetchedFleet) {
+        return reFetchedFleet;
+      }
+
+      throw new Error("Failed to create or find IoT fleet: " + name);
+    }
+  }
+
+  /*
+   * Refresh lastSeenAt / connection status and (optionally) the
+   * snapshot columns the list page renders. Count columns ride this
+   * extras path with COALESCE-per-column semantics: a key that is
+   * undefined is simply not written, so a partial batch (one that
+   * lacked the matching *_info series) never zeroes a count. The
+   * 60-second extras fingerprint cache is the write throttle — the
+   * steady state (identical snapshot every scrape) costs one Redis
+   * read per batch and at most one Postgres UPDATE per minute.
+   *
+   * Two callers share this throttle with DISJOINT extras shapes: the
+   * metrics snapshot flush (counts, every batch) and the fenced
+   * autoDiscoverIoTFleet maintenance path (agentVersion only — and
+   * usually an all-null fingerprint, since the shipped agent config
+   * does not stamp oneuptime.agent.version). The single fingerprint
+   * covers the whole extras object, so each alternation between the
+   * two shapes busts the throttle: at most one extra Postgres UPDATE
+   * per maintenance-fence window (~5 min), which is accepted. Do NOT
+   * key the cache per-caller — that would let two callers each refresh
+   * lastSeenAt under their own throttle and is not worth the
+   * complexity for one UPDATE per 5 minutes.
+   */
+  @CaptureSpan()
+  public async updateLastSeen(
+    fleetId: ObjectID,
+    extra?: {
+      agentVersion?: string | undefined;
+      deviceCount?: number | undefined;
+      onlineDeviceCount?: number | undefined;
+    },
+  ): Promise<void> {
+    const cacheKey: string = fleetId.toString();
+    const extrasFingerprint: string = crypto
+      .createHash("sha1")
+      .update(
+        JSON.stringify({
+          agentVersion: extra?.agentVersion ?? null,
+          deviceCount: extra?.deviceCount ?? null,
+          onlineDeviceCount: extra?.onlineDeviceCount ?? null,
+        }),
+      )
+      .digest("hex");
+
+    let cached: string | null = null;
+    try {
+      cached = await GlobalCache.getString(LAST_SEEN_CACHE_NAMESPACE, cacheKey);
+    } catch {
+      /*
+       * Cache unavailable — fail open and refresh lastSeenAt anyway. A
+       * cache error must never skip the DB write below, otherwise the
+       * resource is wrongly marked "disconnected" while telemetry is
+       * still flowing. Mirrors shouldRunMaintenance's fail-open stance.
+       */
+      cached = null;
+    }
+
+    if (cached === extrasFingerprint) {
+      return; // same data was written recently
+    }
+
+    try {
+      await GlobalCache.setString(
+        LAST_SEEN_CACHE_NAMESPACE,
+        cacheKey,
+        extrasFingerprint,
+        { expiresInSeconds: LAST_SEEN_THROTTLE_SECONDS },
+      );
+    } catch {
+      // Best-effort throttle write; proceed with the DB update regardless.
+    }
+
+    // eslint-disable-next-line @typescript-eslint/no-explicit-any
+    const data: any = {
+      lastSeenAt: OneUptimeDate.getCurrentDate(),
+      otelCollectorStatus: "connected",
+    };
+
+    if (extra?.agentVersion) {
+      data.agentVersion = extra.agentVersion;
+    }
+    // Counts: 0 is a legitimate value — gate on undefined, not falsiness.
+    if (extra?.deviceCount !== undefined) {
+      data.deviceCount = extra.deviceCount;
+    }
+    if (extra?.onlineDeviceCount !== undefined) {
+      data.onlineDeviceCount = extra.onlineDeviceCount;
+    }
+
+    /*
+     * Heartbeat write: a single-statement UPDATE with no hooks and no
+     * `version` bump, avoiding the hot-row Postgres lock convoy that the
+     * full updateOneById pipeline causes. See ServiceService.updateLastSeen.
+     */
+    await this.updateColumnsByIdWithoutHooks({
+      id: fleetId,
+      data: data,
+    });
+  }
+
+  /**
+   * Additively attach labels to an IoT fleet. Existing labels are
+   * never removed — manual labels set via the UI survive ingest. The
+   * set of labelIds passed in is fingerprinted and cached for 60s so
+   * the common case (steady-state collector pushing the same label
+   * set every batch) costs one in-memory lookup, not a join-table
+   * scan.
+   */
+  @CaptureSpan()
+  public async attachLabels(data: {
+    iotFleetId: ObjectID;
+    labelIds: Array<ObjectID>;
+  }): Promise<void> {
+    if (!data.labelIds || data.labelIds.length === 0) {
+      return;
+    }
+
+    const cacheKey: string = data.iotFleetId.toString();
+    const fingerprint: string = fingerprintLabelIds(data.labelIds);
+    const cached: string | null = await GlobalCache.getString(
+      LABELS_APPLIED_CACHE_NAMESPACE,
+      cacheKey,
+    );
+    if (cached === fingerprint) {
+      return;
+    }
+
+    try {
+      const iotFleetIdStr: string = data.iotFleetId.toString();
+      const existingLabels: Array<Label> = await this.getRepository()
+        .createQueryBuilder()
+        .relation(Model, "labels")
+        .of(iotFleetIdStr)
+        .loadMany();
+
+      const existingIds: Set<string> = new Set();
+      for (const lbl of existingLabels) {
+        const idStr: string | undefined = lbl._id?.toString();
+        if (idStr) {
+          existingIds.add(idStr);
+        }
+      }
+
+      const toAddIds: Array<string> = [];
+      const seen: Set<string> = new Set();
+      for (const id of data.labelIds) {
+        const idStr: string = id.toString();
+        if (existingIds.has(idStr) || seen.has(idStr)) {
+          continue;
+        }
+        seen.add(idStr);
+        toAddIds.push(idStr);
+      }
+
+      if (toAddIds.length > 0) {
+        await this.getRepository()
+          .createQueryBuilder()
+          .relation(Model, "labels")
+          .of(iotFleetIdStr)
+          .add(toAddIds);
+      }
+
+      await GlobalCache.setString(
+        LABELS_APPLIED_CACHE_NAMESPACE,
+        cacheKey,
+        fingerprint,
+        { expiresInSeconds: LABELS_APPLIED_CACHE_TTL_SECONDS },
+      );
+    } catch (err) {
+      logger.warn(
+        `IoTFleetService.attachLabels failed for iot fleet ${data.iotFleetId.toString()}: ${
+          err instanceof Error ? err.message : String(err)
+        }`,
+      );
+    }
+  }
+
+  @CaptureSpan()
+  public async markDisconnectedFleets(): Promise<void> {
+    /*
+     * Threshold must stay well above the 5-minute OTel ingest
+     * maintenance fence (MAINTENANCE_FENCE_TTL_SECONDS in
+     * OtelIngestBaseService) — lastSeenAt is legitimately up to
+     * ~5 minutes stale during continuous telemetry, so a threshold
+     * equal to the fence TTL flaps healthy resources. 15 minutes
+     * gives 3x headroom.
+     */
+    const fifteenMinutesAgo: Date = OneUptimeDate.addRemoveMinutes(
+      OneUptimeDate.getCurrentDate(),
+      -15,
+    );
+
+    const connectedFleets: Array<Model> = await this.findBy({
+      query: {
+        otelCollectorStatus: "connected",
+        lastSeenAt: QueryHelper.lessThan(fifteenMinutesAgo),
+      },
+      select: {
+        _id: true,
+      },
+      limit: LIMIT_MAX,
+      skip: 0,
+      props: {
+        isRoot: true,
+      },
+    });
+
+    for (const fleet of connectedFleets) {
+      if (fleet._id) {
+        await this.updateOneById({
+          id: new ObjectID(fleet._id.toString()),
+          data: {
+            otelCollectorStatus: "disconnected",
+          },
+          props: {
+            isRoot: true,
+          },
+        });
+      }
+    }
+  }
+}
+
+function fingerprintLabelIds(labelIds: Array<ObjectID>): string {
+  const sorted: Array<string> = labelIds
+    .map((id: ObjectID) => {
+      return id.toString();
+    })
+    .sort();
+  return crypto.createHash("sha1").update(sorted.join(",")).digest("hex");
+}
+
+export default new Service();

package/Server/Services/OpenTelemetryIngestService.ts

@@ -28,6 +28,7 @@ import DockerSwarmCluster from "../../Models/DatabaseModels/DockerSwarmCluster";
 import ServerlessFunction from "../../Models/DatabaseModels/ServerlessFunction";
 import CloudResource from "../../Models/DatabaseModels/CloudResource";
 import RumApplication from "../../Models/DatabaseModels/RumApplication";
+import IoTFleet from "../../Models/DatabaseModels/IoTFleet";
 import HostService from "./HostService";
 import DockerHostService from "./DockerHostService";
 import PodmanHostService from "./PodmanHostService";
@@ -38,6 +39,7 @@ import DockerSwarmClusterService from "./DockerSwarmClusterService";
 import ServerlessFunctionService from "./ServerlessFunctionService";
 import CloudResourceService from "./CloudResourceService";
 import RumApplicationService from "./RumApplicationService";
+import IoTFleetService from "./IoTFleetService";
 import GlobalCache from "../Infrastructure/GlobalCache";
 import ColumnLength from "../../Types/Database/ColumnLength";
 import EntityType from "../../Types/Telemetry/EntityType";
@@ -728,6 +730,20 @@ export default class OTelIngestService {
           telemetryRetentionConfig: cluster?.telemetryRetentionConfig ?? null,
         };
       }
+      if (primaryEntityType === ServiceType.IoTDevice) {
+        const fleet: IoTFleet | null = await IoTFleetService.findOneById({
+          id: resourceId,
+          select: {
+            retainTelemetryDataForDays: true,
+            telemetryRetentionConfig: true,
+          },
+          props: { isRoot: true },
+        });
+        return {
+          retainTelemetryDataForDays: fleet?.retainTelemetryDataForDays ?? null,
+          telemetryRetentionConfig: fleet?.telemetryRetentionConfig ?? null,
+        };
+      }
       if (primaryEntityType === ServiceType.CephCluster) {
         const cluster: CephCluster | null =
           await CephClusterService.findOneById({