@oneuptime/common 10.7.2 → 10.8.1

package/Models/DatabaseModels/Service.ts

@@ -863,4 +863,272 @@ export default class Service extends BaseModel {
     type: ColumnType.Date,
   })
   public lastSeenAt?: Date = undefined;
+
+  /*
+   * The columns below are auto-populated from OpenTelemetry resource
+   * attributes at ingest time (see OTelIngestService.telemetryServiceFromName).
+   * They are system-managed — not user-editable — so create/update access
+   * control is empty and writes only happen via the root ingest path.
+   */
+
+  @ColumnAccessControl({
+    create: [],
+    read: [
+      Permission.ProjectOwner,
+      Permission.ProjectAdmin,
+      Permission.ProjectMember,
+      Permission.Viewer,
+      Permission.SettingsAdmin,
+      Permission.SettingsMember,
+      Permission.SettingsViewer,
+      Permission.ReadService,
+    ],
+    update: [],
+  })
+  @TableColumn({
+    required: false,
+    type: TableColumnType.ShortText,
+    canReadOnRelationQuery: true,
+    title: "Service Version",
+    description:
+      "Last-seen value of the service.version OpenTelemetry resource attribute.",
+  })
+  @Column({
+    nullable: true,
+    type: ColumnType.ShortText,
+    length: ColumnLength.ShortText,
+  })
+  public serviceVersion?: string = undefined;
+
+  @ColumnAccessControl({
+    create: [],
+    read: [
+      Permission.ProjectOwner,
+      Permission.ProjectAdmin,
+      Permission.ProjectMember,
+      Permission.Viewer,
+      Permission.SettingsAdmin,
+      Permission.SettingsMember,
+      Permission.SettingsViewer,
+      Permission.ReadService,
+    ],
+    update: [],
+  })
+  @TableColumn({
+    required: false,
+    type: TableColumnType.ShortText,
+    canReadOnRelationQuery: true,
+    title: "Deployment Environment",
+    description:
+      "Last-seen value of the deployment.environment.name (or deployment.environment) OpenTelemetry resource attribute, e.g. production, staging.",
+  })
+  @Column({
+    nullable: true,
+    type: ColumnType.ShortText,
+    length: ColumnLength.ShortText,
+  })
+  public deploymentEnvironment?: string = undefined;
+
+  @ColumnAccessControl({
+    create: [],
+    read: [
+      Permission.ProjectOwner,
+      Permission.ProjectAdmin,
+      Permission.ProjectMember,
+      Permission.Viewer,
+      Permission.SettingsAdmin,
+      Permission.SettingsMember,
+      Permission.SettingsViewer,
+      Permission.ReadService,
+    ],
+    update: [],
+  })
+  @TableColumn({
+    required: false,
+    type: TableColumnType.ShortText,
+    canReadOnRelationQuery: true,
+    title: "Service Namespace",
+    description:
+      "Last-seen value of the service.namespace OpenTelemetry resource attribute.",
+  })
+  @Column({
+    nullable: true,
+    type: ColumnType.ShortText,
+    length: ColumnLength.ShortText,
+  })
+  public serviceNamespace?: string = undefined;
+
+  @ColumnAccessControl({
+    create: [],
+    read: [
+      Permission.ProjectOwner,
+      Permission.ProjectAdmin,
+      Permission.ProjectMember,
+      Permission.Viewer,
+      Permission.SettingsAdmin,
+      Permission.SettingsMember,
+      Permission.SettingsViewer,
+      Permission.ReadService,
+    ],
+    update: [],
+  })
+  @TableColumn({
+    required: false,
+    type: TableColumnType.ShortText,
+    canReadOnRelationQuery: true,
+    title: "Runtime Name",
+    description:
+      "Last-seen value of the process.runtime.name OpenTelemetry resource attribute, e.g. nodejs, go, OpenJDK Runtime Environment.",
+  })
+  @Column({
+    nullable: true,
+    type: ColumnType.ShortText,
+    length: ColumnLength.ShortText,
+  })
+  public runtimeName?: string = undefined;
+
+  @ColumnAccessControl({
+    create: [],
+    read: [
+      Permission.ProjectOwner,
+      Permission.ProjectAdmin,
+      Permission.ProjectMember,
+      Permission.Viewer,
+      Permission.SettingsAdmin,
+      Permission.SettingsMember,
+      Permission.SettingsViewer,
+      Permission.ReadService,
+    ],
+    update: [],
+  })
+  @TableColumn({
+    required: false,
+    type: TableColumnType.ShortText,
+    canReadOnRelationQuery: true,
+    title: "Runtime Version",
+    description:
+      "Last-seen value of the process.runtime.version OpenTelemetry resource attribute.",
+  })
+  @Column({
+    nullable: true,
+    type: ColumnType.ShortText,
+    length: ColumnLength.ShortText,
+  })
+  public runtimeVersion?: string = undefined;
+
+  @ColumnAccessControl({
+    create: [],
+    read: [
+      Permission.ProjectOwner,
+      Permission.ProjectAdmin,
+      Permission.ProjectMember,
+      Permission.Viewer,
+      Permission.SettingsAdmin,
+      Permission.SettingsMember,
+      Permission.SettingsViewer,
+      Permission.ReadService,
+    ],
+    update: [],
+  })
+  @TableColumn({
+    required: false,
+    type: TableColumnType.ShortText,
+    canReadOnRelationQuery: true,
+    title: "Cloud Provider",
+    description:
+      "Last-seen value of the cloud.provider OpenTelemetry resource attribute, e.g. aws, gcp, azure.",
+  })
+  @Column({
+    nullable: true,
+    type: ColumnType.ShortText,
+    length: ColumnLength.ShortText,
+  })
+  public cloudProvider?: string = undefined;
+
+  @ColumnAccessControl({
+    create: [],
+    read: [
+      Permission.ProjectOwner,
+      Permission.ProjectAdmin,
+      Permission.ProjectMember,
+      Permission.Viewer,
+      Permission.SettingsAdmin,
+      Permission.SettingsMember,
+      Permission.SettingsViewer,
+      Permission.ReadService,
+    ],
+    update: [],
+  })
+  @TableColumn({
+    required: false,
+    type: TableColumnType.ShortText,
+    canReadOnRelationQuery: true,
+    title: "Cloud Platform",
+    description:
+      "Last-seen value of the cloud.platform OpenTelemetry resource attribute, e.g. aws_ecs, gcp_cloud_run, aws_lambda.",
+  })
+  @Column({
+    nullable: true,
+    type: ColumnType.ShortText,
+    length: ColumnLength.ShortText,
+  })
+  public cloudPlatform?: string = undefined;
+
+  @ColumnAccessControl({
+    create: [],
+    read: [
+      Permission.ProjectOwner,
+      Permission.ProjectAdmin,
+      Permission.ProjectMember,
+      Permission.Viewer,
+      Permission.SettingsAdmin,
+      Permission.SettingsMember,
+      Permission.SettingsViewer,
+      Permission.ReadService,
+    ],
+    update: [],
+  })
+  @TableColumn({
+    required: false,
+    type: TableColumnType.ShortText,
+    canReadOnRelationQuery: true,
+    title: "Cloud Region",
+    description:
+      "Last-seen value of the cloud.region OpenTelemetry resource attribute, e.g. us-east-1.",
+  })
+  @Column({
+    nullable: true,
+    type: ColumnType.ShortText,
+    length: ColumnLength.ShortText,
+  })
+  public cloudRegion?: string = undefined;
+
+  @ColumnAccessControl({
+    create: [],
+    read: [
+      Permission.ProjectOwner,
+      Permission.ProjectAdmin,
+      Permission.ProjectMember,
+      Permission.Viewer,
+      Permission.SettingsAdmin,
+      Permission.SettingsMember,
+      Permission.SettingsViewer,
+      Permission.ReadService,
+    ],
+    update: [],
+  })
+  @TableColumn({
+    required: false,
+    type: TableColumnType.ShortText,
+    canReadOnRelationQuery: true,
+    title: "Cloud Account ID",
+    description:
+      "Last-seen value of the cloud.account.id OpenTelemetry resource attribute.",
+  })
+  @Column({
+    nullable: true,
+    type: ColumnType.ShortText,
+    length: ColumnLength.ShortText,
+  })
+  public cloudAccountId?: string = undefined;
 }

package/Models/DatabaseModels/TelemetryException.ts

@@ -24,6 +24,9 @@ import Service from "./Service";
 import Host from "./Host";
 import DockerHost from "./DockerHost";
 import KubernetesCluster from "./KubernetesCluster";
+import ServerlessFunction from "./ServerlessFunction";
+import CloudResource from "./CloudResource";
+import RumApplication from "./RumApplication";
 
 @EnableDocumentation()
 @CanAccessIfCanReadOn("service")
@@ -71,7 +74,18 @@ import KubernetesCluster from "./KubernetesCluster";
  * of those resource types, and includeProjectScope lets in-project users
  * see the unattributed bucket (which has no owner resource).
  */
-@OwnedThrough("serviceId", [Service, Host, DockerHost, KubernetesCluster], {
+@OwnedThrough(
+  "serviceId",
+  [
+    Service,
+    Host,
+    DockerHost,
+    KubernetesCluster,
+    ServerlessFunction,
+    CloudResource,
+    RumApplication,
+  ],
+  {
   includeProjectScope: true,
 })
 @Entity({

package/Models/DatabaseModels/WorkflowLog.ts

@@ -15,6 +15,7 @@ import TableColumnType from "../../Types/Database/TableColumnType";
 import TableMetadata from "../../Types/Database/TableMetadata";
 import TenantColumn from "../../Types/Database/TenantColumn";
 import IconProp from "../../Types/Icon/IconProp";
+import { JSONObject } from "../../Types/JSON";
 import ObjectID from "../../Types/ObjectID";
 import Permission from "../../Types/Permission";
 import WorkflowStatus from "../../Types/Workflow/WorkflowStatus";
@@ -296,6 +297,57 @@ export default class WorkflowLog extends BaseModel {
   })
   public completedAt?: Date = undefined;
 
+  @ColumnAccessControl({
+    create: [],
+    read: [
+      Permission.ProjectOwner,
+      Permission.ProjectAdmin,
+      Permission.ProjectMember,
+      Permission.Viewer,
+      Permission.WorkflowAdmin,
+      Permission.WorkflowMember,
+      Permission.WorkflowViewer,
+      Permission.ReadWorkflowLog,
+    ],
+    update: [],
+  })
+  @TableColumn({
+    type: TableColumnType.Date,
+    title: "Resume At",
+    description:
+      "When this workflow run is scheduled to resume after a Sleep step (only set while the run is Waiting).",
+  })
+  @Column({
+    type: ColumnType.Date,
+    nullable: true,
+    unique: false,
+  })
+  public resumeAt?: Date = undefined;
+
+  /*
+   * Internal. Serialized execution state (pending steps, executed steps and
+   * accumulated component return values) used to resume a workflow run after a
+   * Sleep step. Never exposed via the API.
+   */
+  @ColumnAccessControl({
+    create: [],
+    read: [],
+    update: [],
+  })
+  @TableColumn({
+    required: false,
+    isDefaultValueColumn: false,
+    type: TableColumnType.JSON,
+    title: "Resume Data",
+    description:
+      "Internal serialized execution state used to resume a workflow run after a Sleep step.",
+  })
+  @Column({
+    type: ColumnType.JSON,
+    nullable: true,
+  })
+  public resumeData?: JSONObject = undefined;
+
   @ColumnAccessControl({
     create: [],
     read: [],

package/Server/API/TelemetryAPI.ts

@@ -688,11 +688,10 @@ router.post(
         : undefined;
 
       /*
-       * Compute all facets from a single sort-key-aligned sample query
-       * (ORDER BY startTime DESC LIMIT N) and count top-K in Node. This
-       * avoids ClickHouse GROUP BY aggregations that can't return partial
-       * results under max_execution_time 'break' mode, and returns in
-       * <1s even over 14-day windows.
+       * Shared window + active filters for both facet-counting paths below:
+       * the exact projection-backed GROUP BY (resource facets + statusCode)
+       * and the recent-N sample (kind + attribute facets, which have no cheap
+       * exact path).
        */
       const multiRequest: TraceMultiFacetRequest = {
         projectId: databaseProps.tenantId,
@@ -710,38 +709,92 @@ router.post(
         attributes,
       };
 
-      let facets: Record<string, Array<TraceFacetValue>>;
-      try {
-        facets =
-          await TraceAggregationService.getFacetValuesFromSample(multiRequest);
-      } catch {
-        facets = Object.fromEntries(
-          facetKeys.map((key: string): [string, Array<TraceFacetValue>] => {
-            return [key, []];
-          }),
-        );
+      /*
+       * Resource facets (serviceId / hostId / dockerHostId / k8s cluster ...)
+       * and statusCode are counted with an exact, projection-backed GROUP BY
+       * in getResourceFacetCounts(). The recent-N sample below saturates with
+       * whichever service is chattiest right now and reports 0 for every other
+       * service regardless of its true volume over the window — the "top 1000"
+       * symptom. Facets with no projection (kind, attribute keys) have no cheap
+       * exact path and stay on the sample.
+       */
+      const sampledKeys: Array<string> = facetKeys.filter(
+        (key: string): boolean => {
+          return (
+            !ResourceFacetResolver.isResourceFacet(key) && key !== "statusCode"
+          );
+        },
+      );
+
+      let facets: Record<string, Array<TraceFacetValue>> = {};
+      if (sampledKeys.length > 0) {
+        try {
+          facets = await TraceAggregationService.getFacetValuesFromSample({
+            ...multiRequest,
+            facetKeys: sampledKeys,
+          });
+        } catch {
+          facets = Object.fromEntries(
+            sampledKeys.map((key: string): [string, Array<TraceFacetValue>] => {
+              return [key, []];
+            }),
+          );
+        }
+      }
+
+      const needsAccurateCounts: boolean =
+        facetKeys.includes("statusCode") ||
+        facetKeys.some((key: string): boolean => {
+          return ResourceFacetResolver.isResourceFacet(key);
+        });
+
+      let serviceCounts: Map<string, number> = new Map<string, number>();
+      let statusCounts: Map<string, number> = new Map<string, number>();
+      if (needsAccurateCounts) {
+        try {
+          const accurate: {
+            serviceCounts: Map<string, number>;
+            statusCounts: Map<string, number>;
+          } =
+            await TraceAggregationService.getResourceFacetCounts(multiRequest);
+          serviceCounts = accurate.serviceCounts;
+          statusCounts = accurate.statusCounts;
+        } catch {
+          /*
+           * Degrade gracefully: resource facets still enumerate via Postgres
+           * (count 0), statusCode falls back to empty.
+           */
+        }
+      }
+
+      if (facetKeys.includes("statusCode")) {
+        facets["statusCode"] = Array.from(statusCounts.entries())
+          .map(([value, count]: [string, number]): TraceFacetValue => {
+            return { value, count };
+          })
+          .sort((a: TraceFacetValue, b: TraceFacetValue): number => {
+            return b.count - a.count;
+          })
+          .slice(0, limit);
       }
 
       /*
        * Replace resource-facet results with the Postgres source-of-truth list
-       * (filtered by facetSearchText and enriched with displayName). Counts
-       * come from the ClickHouse sample above — entities with no recent
-       * telemetry surface with count 0 instead of being hidden entirely. This
-       * means low-volume services / hosts still appear in the sidebar and the
-       * search box can find resources beyond the loaded subset.
+       * (filtered by facetSearchText and enriched with displayName). Every
+       * resource facet shares the same exact serviceId -> count map; resource
+       * ids are globally unique, so each facet only ever resolves its own
+       * entities. Entities with no telemetry in the window surface with count
+       * 0 instead of being hidden, and the search box can find resources
+       * beyond the loaded subset.
        */
       const resourceSpecs: Array<ResourceFacetSpec> = facetKeys
         .filter((key: string): boolean => {
           return ResourceFacetResolver.isResourceFacet(key);
         })
         .map((key: string): ResourceFacetSpec => {
-          const counts: Map<string, number> = new Map();
-          for (const fv of facets[key] || []) {
-            counts.set(fv.value, fv.count);
-          }
           return {
             facetKey: key,
-            counts,
+            counts: serviceCounts,
             searchText: facetSearchText?.[key],
             limit,
           };

package/Server/Infrastructure/Postgres/SchemaMigrations/1780931746908-AddResumeStateToWorkflowLog.ts

@@ -0,0 +1,21 @@
+import { MigrationInterface, QueryRunner } from "typeorm";
+
+export class AddResumeStateToWorkflowLog1780931746908
+  implements MigrationInterface
+{
+  public name = "AddResumeStateToWorkflowLog1780931746908";
+
+  public async up(queryRunner: QueryRunner): Promise<void> {
+    await queryRunner.query(
+      `ALTER TABLE "WorkflowLog" ADD "resumeAt" TIMESTAMP WITH TIME ZONE`,
+    );
+    await queryRunner.query(`ALTER TABLE "WorkflowLog" ADD "resumeData" jsonb`);
+  }
+
+  public async down(queryRunner: QueryRunner): Promise<void> {
+    await queryRunner.query(
+      `ALTER TABLE "WorkflowLog" DROP COLUMN "resumeData"`,
+    );
+    await queryRunner.query(`ALTER TABLE "WorkflowLog" DROP COLUMN "resumeAt"`);
+  }
+}

package/Server/Infrastructure/Postgres/SchemaMigrations/1780931863719-AddTelemetryResourceMetadataColumns.ts

@@ -0,0 +1,108 @@
+import { MigrationInterface, QueryRunner } from "typeorm";
+
+/*
+ * Phase 0 (capture dropped OTel attributes): add first-class columns to
+ * Service and Host for resource attributes that ingest previously dropped
+ * (they only lived as resource.* attributes on telemetry rows). These are
+ * auto-populated from OpenTelemetry resource attributes at ingest time and
+ * are system-managed (not user-editable).
+ */
+export class AddTelemetryResourceMetadataColumns1780931863719
+  implements MigrationInterface
+{
+  public name = "AddTelemetryResourceMetadataColumns1780931863719";
+
+  public async up(queryRunner: QueryRunner): Promise<void> {
+    // Service
+    await queryRunner.query(
+      `ALTER TABLE "Service" ADD "serviceVersion" character varying(100)`,
+    );
+    await queryRunner.query(
+      `ALTER TABLE "Service" ADD "deploymentEnvironment" character varying(100)`,
+    );
+    await queryRunner.query(
+      `ALTER TABLE "Service" ADD "serviceNamespace" character varying(100)`,
+    );
+    await queryRunner.query(
+      `ALTER TABLE "Service" ADD "runtimeName" character varying(100)`,
+    );
+    await queryRunner.query(
+      `ALTER TABLE "Service" ADD "runtimeVersion" character varying(100)`,
+    );
+    await queryRunner.query(
+      `ALTER TABLE "Service" ADD "cloudProvider" character varying(100)`,
+    );
+    await queryRunner.query(
+      `ALTER TABLE "Service" ADD "cloudPlatform" character varying(100)`,
+    );
+    await queryRunner.query(
+      `ALTER TABLE "Service" ADD "cloudRegion" character varying(100)`,
+    );
+    await queryRunner.query(
+      `ALTER TABLE "Service" ADD "cloudAccountId" character varying(100)`,
+    );
+
+    // Host
+    await queryRunner.query(
+      `ALTER TABLE "Host" ADD "deploymentEnvironment" character varying(100)`,
+    );
+    await queryRunner.query(
+      `ALTER TABLE "Host" ADD "runtimeName" character varying(100)`,
+    );
+    await queryRunner.query(
+      `ALTER TABLE "Host" ADD "runtimeVersion" character varying(100)`,
+    );
+    await queryRunner.query(
+      `ALTER TABLE "Host" ADD "cloudProvider" character varying(100)`,
+    );
+    await queryRunner.query(
+      `ALTER TABLE "Host" ADD "cloudPlatform" character varying(100)`,
+    );
+    await queryRunner.query(
+      `ALTER TABLE "Host" ADD "cloudRegion" character varying(100)`,
+    );
+    await queryRunner.query(
+      `ALTER TABLE "Host" ADD "cloudAccountId" character varying(100)`,
+    );
+  }
+
+  public async down(queryRunner: QueryRunner): Promise<void> {
+    // Host
+    await queryRunner.query(
+      `ALTER TABLE "Host" DROP COLUMN "cloudAccountId"`,
+    );
+    await queryRunner.query(`ALTER TABLE "Host" DROP COLUMN "cloudRegion"`);
+    await queryRunner.query(`ALTER TABLE "Host" DROP COLUMN "cloudPlatform"`);
+    await queryRunner.query(`ALTER TABLE "Host" DROP COLUMN "cloudProvider"`);
+    await queryRunner.query(`ALTER TABLE "Host" DROP COLUMN "runtimeVersion"`);
+    await queryRunner.query(`ALTER TABLE "Host" DROP COLUMN "runtimeName"`);
+    await queryRunner.query(
+      `ALTER TABLE "Host" DROP COLUMN "deploymentEnvironment"`,
+    );
+
+    // Service
+    await queryRunner.query(
+      `ALTER TABLE "Service" DROP COLUMN "cloudAccountId"`,
+    );
+    await queryRunner.query(`ALTER TABLE "Service" DROP COLUMN "cloudRegion"`);
+    await queryRunner.query(
+      `ALTER TABLE "Service" DROP COLUMN "cloudPlatform"`,
+    );
+    await queryRunner.query(
+      `ALTER TABLE "Service" DROP COLUMN "cloudProvider"`,
+    );
+    await queryRunner.query(
+      `ALTER TABLE "Service" DROP COLUMN "runtimeVersion"`,
+    );
+    await queryRunner.query(`ALTER TABLE "Service" DROP COLUMN "runtimeName"`);
+    await queryRunner.query(
+      `ALTER TABLE "Service" DROP COLUMN "serviceNamespace"`,
+    );
+    await queryRunner.query(
+      `ALTER TABLE "Service" DROP COLUMN "deploymentEnvironment"`,
+    );
+    await queryRunner.query(
+      `ALTER TABLE "Service" DROP COLUMN "serviceVersion"`,
+    );
+  }
+}