@oneuptime/common 14.0.5 → 14.0.7

package/Models/DatabaseModels/InventoryItem.ts

@@ -12,6 +12,7 @@ import TableColumn from "../../Types/Database/TableColumn";
 import TableColumnType from "../../Types/Database/TableColumnType";
 import TableMetadata from "../../Types/Database/TableMetadata";
 import TenantColumn from "../../Types/Database/TenantColumn";
+import UNSYNCHRONIZED_INDEX from "../../Types/Database/UnsynchronizedIndex";
 import IconProp from "../../Types/Icon/IconProp";
 import { JSONObject } from "../../Types/JSON";
 import ObjectID from "../../Types/ObjectID";
@@ -20,8 +21,7 @@ import EntitySource from "../../Types/Telemetry/EntitySource";
 import EntityType from "../../Types/Telemetry/EntityType";
 import {
   InventoryLiveness,
-  INVENTORY_LIVE_WINDOW_MINUTES,
-  INVENTORY_STALE_AFTER_MINUTES,
+  getInventoryLivenessSql,
 } from "../../Types/Telemetry/InventoryLiveness";
 import {
   Column,
@@ -125,6 +125,24 @@ const UPDATE_PERMS: Array<Permission> = [
  * the tenant column rather than being scanned per row.
  */
 @Index(["projectId", "isArchived"])
+/*
+ * Lookup by entity key alone. Relationships name their ends by key, never by
+ * type, so every "what is at the other end" join (the Topology API's callee,
+ * placement and drawer lookups) resolves keys without knowing the type. The
+ * unique index above can only serve that as one probe per type in the
+ * project; this serves it in one.
+ *
+ * Created by migration 1795200000000-AddInventoryItemProjectEntityKeyIndex,
+ * which carries the timeouts and the CONCURRENTLY runbook for large installs
+ * that a generated migration cannot. `synchronize: false` keeps the schema
+ * builder from dropping it or generating a second copy (see
+ * UnsynchronizedIndex).
+ */
+@Index(
+  "IDX_INVENTORY_ITEM_PROJECT_ENTITY_KEY",
+  ["projectId", "entityKey"],
+  UNSYNCHRONIZED_INDEX,
+)
 export default class InventoryItem extends DatabaseBaseModel {
   @ColumnAccessControl({ create: CREATE_PERMS, read: READ_PERMS, update: [] })
   @TableColumn({
@@ -385,9 +403,8 @@ export default class InventoryItem extends DatabaseBaseModel {
    * status without persisting a value that becomes stale as time passes.
    * A virtual column needs no schema migration and is never written.
    *
-   * The badge floors elapsed minutes before applying inclusive thresholds:
-   * 30m59s is still Live, and 24h00m59s is still Recent. Strict comparisons
-   * against the following minute preserve those exact boundaries.
+   * The SQL is shared with the Inventory Overview's "Gone Quiet" count — see
+   * getInventoryLivenessSql for the boundary rules.
    */
   @ColumnAccessControl({ create: [], read: READ_PERMS, update: [] })
   @TableColumn({
@@ -400,13 +417,7 @@ export default class InventoryItem extends DatabaseBaseModel {
   @VirtualColumn({
     type: ColumnType.ShortText,
     query: (alias: string): string => {
-      return `CASE
-        WHEN ${alias}."source" IS DISTINCT FROM '${EntitySource.Discovered}' THEN '${InventoryLiveness.NotTracked}'
-        WHEN ${alias}."lastSeenAt" IS NULL THEN '${InventoryLiveness.Never}'
-        WHEN ${alias}."lastSeenAt" > CURRENT_TIMESTAMP - INTERVAL '${INVENTORY_LIVE_WINDOW_MINUTES + 1} minutes' THEN '${InventoryLiveness.Live}'
-        WHEN ${alias}."lastSeenAt" > CURRENT_TIMESTAMP - INTERVAL '${INVENTORY_STALE_AFTER_MINUTES + 1} minutes' THEN '${InventoryLiveness.Recent}'
-        ELSE '${InventoryLiveness.Stale}'
-      END`;
+      return getInventoryLivenessSql(alias);
     },
   })
   public inventoryStatus?: InventoryLiveness = undefined;

package/Server/Infrastructure/Postgres/SchemaMigrations/1795200000000-AddInventoryItemProjectEntityKeyIndex.ts

@@ -0,0 +1,217 @@
+import logger from "../../../Utils/Logger";
+import { MigrationInterface, QueryRunner } from "typeorm";
+
+export const INVENTORY_ITEM_PROJECT_ENTITY_KEY_INDEX: string =
+  "IDX_INVENTORY_ITEM_PROJECT_ENTITY_KEY";
+
+/*
+ * Above this many InventoryItem rows the index is not built here (see
+ * "LOCKING" below). Roughly ten seconds of build on modest hardware.
+ */
+export const INVENTORY_ITEM_INLINE_INDEX_BUILD_MAX_ROWS: number = 1_000_000;
+
+/* What an operator runs to build the index online, under exactly this name. */
+export const INVENTORY_ITEM_PROJECT_ENTITY_KEY_INDEX_RUNBOOK: string = `CREATE INDEX CONCURRENTLY IF NOT EXISTS "${INVENTORY_ITEM_PROJECT_ENTITY_KEY_INDEX}" ON "InventoryItem" ("projectId", "entityKey")`;
+
+/*
+ * Adds ("projectId", "entityKey") on InventoryItem.
+ *
+ * WHY
+ *
+ * Relationships name both ends by entity key only. The Topology API resolves
+ * those keys to items constantly — the things services call, where they run,
+ * the other end of every row in the detail drawer — and the only index that
+ * leads with the key today is the unique ("projectId", "entityType",
+ * "entityKey"), which a type-less lookup can use only as one probe per type in
+ * the project. The queries are written to stay correct and acceptable without
+ * this index (they constrain "entityType" to the project's types so the
+ * unique index is usable), because on Helm installs migrations run
+ * asynchronously and the new API can serve traffic before this has been
+ * built. This makes the same lookups one probe.
+ *
+ * Relationships get no new index on purpose: InventoryItemRelationship is the
+ * largest inventory table and a plain CREATE INDEX would block its ingest
+ * writes for the whole build.
+ *
+ * WHY THIS IS HAND-WRITTEN
+ *
+ * A generated migration cannot bound its own lock wait or runtime, nor decide
+ * to leave a build to the operator, and this one runs against a table that
+ * ingest writes continuously. The entity declares the index by name with
+ * `synchronize: false` (see Types/Database/UnsynchronizedIndex.ts), so the
+ * schema builder neither generates a second copy nor drops this one.
+ *
+ * LOCKING, AND THE RUNBOOK FOR LARGE INSTALLS
+ *
+ * Migrations run one transaction each (migrationsTransactionMode "each"), so
+ * CREATE INDEX CONCURRENTLY is not available here. A plain CREATE INDEX holds
+ * a SHARE lock on InventoryItem for as long as the build takes: reads
+ * continue, inventory writes queue behind it and fail once their own
+ * lock_timeout (3 s in the app) or statement_timeout runs out.
+ *
+ * The build cannot be given more time than the connection allows. The
+ * effective ceiling is the client-side DATABASE_QUERY_TIMEOUT_MS of
+ * node-postgres (35 s by default), which no SET can raise: past it the client
+ * abandons the query — and the ROLLBACK queued behind it — while the server
+ * keeps building and keeps the lock. So statement_timeout stays at the
+ * connection's own value (30 s by default, below that ceiling; `DEFAULT`
+ * undoes anything an earlier statement on the session may have set), which
+ * makes Postgres cancel an overlong build itself and release the lock at
+ * once. `lock_timeout` stops the migration from queueing behind a long
+ * transaction (and parking every writer behind itself) for more than five
+ * seconds; a timeout fails the migration, which the next start retries.
+ *
+ * A build that would not finish comfortably inside that window is not
+ * attempted: when InventoryItem holds more than
+ * INVENTORY_ITEM_INLINE_INDEX_BUILD_MAX_ROWS rows (the planner's estimate
+ * from pg_class; a bounded count when the table has never been analyzed),
+ * this migration logs a warning and completes without the index. The
+ * Topology API is correct without it, only slower on key lookups. Such an
+ * install should build the index online, at any time, under exactly this
+ * name:
+ *
+ *   CREATE INDEX CONCURRENTLY IF NOT EXISTS "IDX_INVENTORY_ITEM_PROJECT_ENTITY_KEY"
+ *     ON "InventoryItem" ("projectId", "entityKey");
+ *
+ * Built before the upgrade, this migration finds it and does nothing. A
+ * CONCURRENTLY build that failed leaves an INVALID index behind under that
+ * name, which IF NOT EXISTS would otherwise accept forever; such a leftover
+ * is dropped here first (so the runbook works as written afterwards) and
+ * rebuilt when the table is small enough.
+ *
+ * An online build that is still RUNNING looks exactly like that leftover
+ * (pg_index.indisvalid is false until it finishes). Dropping it would queue
+ * for an ACCESS EXCLUSIVE lock behind the build and stall every reader and
+ * writer of InventoryItem meanwhile, so a build in progress
+ * (pg_stat_progress_create_index) is left to finish and this migration
+ * completes without touching it.
+ */
+export class AddInventoryItemProjectEntityKeyIndex1795200000000
+  implements MigrationInterface
+{
+  public name: string = "AddInventoryItemProjectEntityKeyIndex1795200000000";
+
+  /* Lowered only by tests; TypeORM constructs migrations without arguments. */
+  public constructor(
+    private maxRowsForInlineBuild: number = INVENTORY_ITEM_INLINE_INDEX_BUILD_MAX_ROWS,
+  ) {}
+
+  public async up(queryRunner: QueryRunner): Promise<void> {
+    await queryRunner.query(`SET LOCAL statement_timeout = DEFAULT`);
+    await queryRunner.query(`SET LOCAL lock_timeout = '5s'`);
+
+    const existing: Array<{ isValid: boolean }> = await queryRunner.query(
+      `SELECT x.indisvalid AS "isValid" FROM pg_index x
+       JOIN pg_class c ON c.oid = x.indexrelid
+       JOIN pg_namespace n ON n.oid = c.relnamespace
+       WHERE c.relname = $1 AND n.nspname = current_schema()`,
+      [INVENTORY_ITEM_PROJECT_ENTITY_KEY_INDEX],
+    );
+
+    if (
+      existing.some((index: { isValid: boolean }): boolean => {
+        return index.isValid === true;
+      })
+    ) {
+      /* Pre-built by an operator (or already migrated): nothing to do. */
+      return;
+    }
+
+    if (existing.length > 0) {
+      if (await this.isBuildInProgress(queryRunner)) {
+        logger.warn(
+          `AddInventoryItemProjectEntityKeyIndex: an online build of ${INVENTORY_ITEM_PROJECT_ENTITY_KEY_INDEX} is in progress; leaving it to finish. If that build fails, drop the INVALID index it leaves behind and run again: ${INVENTORY_ITEM_PROJECT_ENTITY_KEY_INDEX_RUNBOOK};`,
+        );
+        return;
+      }
+      await queryRunner.query(
+        `DROP INDEX IF EXISTS "${INVENTORY_ITEM_PROJECT_ENTITY_KEY_INDEX}"`,
+      );
+    }
+
+    const rows: number = await this.estimateRows(queryRunner);
+
+    if (rows > this.maxRowsForInlineBuild) {
+      logger.warn(
+        `AddInventoryItemProjectEntityKeyIndex: InventoryItem holds about ${Math.round(
+          rows,
+        )} rows, more than the ${this.maxRowsForInlineBuild} this migration builds an index over while it blocks inventory writes, so the index ${INVENTORY_ITEM_PROJECT_ENTITY_KEY_INDEX} was NOT created. The Topology maps work without it but resolve entity keys more slowly. Build it online (it does not block writes) by running on the OneUptime database: ${INVENTORY_ITEM_PROJECT_ENTITY_KEY_INDEX_RUNBOOK};`,
+      );
+      return;
+    }
+
+    await queryRunner.query(
+      `CREATE INDEX IF NOT EXISTS "${INVENTORY_ITEM_PROJECT_ENTITY_KEY_INDEX}" ON "InventoryItem" ("projectId", "entityKey")`,
+    );
+  }
+
+  public async down(queryRunner: QueryRunner): Promise<void> {
+    /*
+     * DROP INDEX takes ACCESS EXCLUSIVE on the table; waiting for it would
+     * park every reader and writer of InventoryItem behind this statement.
+     */
+    await queryRunner.query(`SET LOCAL statement_timeout = DEFAULT`);
+    await queryRunner.query(`SET LOCAL lock_timeout = '5s'`);
+    await queryRunner.query(
+      `DROP INDEX IF EXISTS "${INVENTORY_ITEM_PROJECT_ENTITY_KEY_INDEX}"`,
+    );
+  }
+
+  /* Whether a CREATE INDEX (CONCURRENTLY) of this index is running now. */
+  private async isBuildInProgress(queryRunner: QueryRunner): Promise<boolean> {
+    const builds: Array<unknown> = await queryRunner.query(
+      `SELECT 1 FROM pg_stat_progress_create_index p
+       JOIN pg_class c ON c.oid = p.index_relid
+       JOIN pg_namespace n ON n.oid = c.relnamespace
+       WHERE c.relname = $1 AND n.nspname = current_schema()`,
+      [INVENTORY_ITEM_PROJECT_ENTITY_KEY_INDEX],
+    );
+    return builds.length > 0;
+  }
+
+  /*
+   * How many rows InventoryItem holds, from the statistics the planner uses:
+   * pg_class.reltuples scaled from the pages it was measured over to the
+   * table's current size (so growth since the last ANALYZE counts). A table
+   * that has never been analyzed (reltuples -1), or whose statistics saw no
+   * pages although it has some now, is counted instead — only up to just past
+   * the threshold, so the count stays cheap on any table.
+   */
+  private async estimateRows(queryRunner: QueryRunner): Promise<number> {
+    const statistics: Array<{
+      reltuples: number | string | null;
+      relpages: number | string | null;
+      pages: number | string | null;
+    }> = await queryRunner.query(
+      `SELECT c.reltuples::float8 AS "reltuples", c.relpages::float8 AS "relpages",
+         (pg_relation_size(c.oid) / current_setting('block_size')::int)::float8 AS "pages"
+       FROM pg_class c
+       JOIN pg_namespace n ON n.oid = c.relnamespace
+       WHERE c.relname = 'InventoryItem' AND n.nspname = current_schema()`,
+    );
+
+    const reltuples: number = Number(statistics[0]?.reltuples ?? -1);
+    const relpages: number = Number(statistics[0]?.relpages ?? 0);
+    const pages: number = Number(statistics[0]?.pages ?? 0);
+
+    const isKnown: boolean =
+      Number.isFinite(reltuples) &&
+      reltuples >= 0 &&
+      Number.isFinite(relpages) &&
+      Number.isFinite(pages) &&
+      (relpages > 0 || pages === 0);
+
+    if (isKnown) {
+      return relpages > 0
+        ? Math.max(reltuples, (reltuples / relpages) * pages)
+        : reltuples;
+    }
+
+    const counted: Array<{ count: number | string }> = await queryRunner.query(
+      `SELECT COUNT(*)::int AS "count" FROM (SELECT 1 FROM "InventoryItem" LIMIT $1) bounded`,
+      [this.maxRowsForInlineBuild + 1],
+    );
+
+    return Number(counted[0]?.count ?? 0);
+  }
+}

package/Server/Infrastructure/Postgres/SchemaMigrations/Index.ts

@@ -12,6 +12,7 @@ import { AddSessionReplaySameOriginTracePropagation1794800000000 } from "./17948
 import { AddIncidentAlert1794900000000 } from "./1794900000000-AddIncidentAlert";
 import { AddDatabaseServerTables1795000000000 } from "./1795000000000-AddDatabaseServerTables";
 import { AddUserProjectSsoConsent1795100000000 } from "./1795100000000-AddUserProjectSsoConsent";
+import { AddInventoryItemProjectEntityKeyIndex1795200000000 } from "./1795200000000-AddInventoryItemProjectEntityKeyIndex";
 import { AddConfigurableRuleCriteria1792500000000 } from "./1792500000000-AddConfigurableRuleCriteria";
 import { AddGoogleSecOpsDiagnostics1792300000000 } from "./1792300000000-AddGoogleSecOpsDiagnostics";
 import { AddIncomingCallPolicyPhoneNumbers1792600000000 } from "./1792600000000-AddIncomingCallPolicyPhoneNumbers";
@@ -1224,4 +1225,5 @@ export default [
   AddIncidentAlert1794900000000,
   AddDatabaseServerTables1795000000000,
   AddUserProjectSsoConsent1795100000000,
+  AddInventoryItemProjectEntityKeyIndex1795200000000,
 ];

package/Server/Services/AlertService.ts

@@ -20,7 +20,9 @@ import OnCallDutyPolicyService from "./OnCallDutyPolicyService";
 import TeamMemberService from "./TeamMemberService";
 import UserService from "./UserService";
 import URL from "../../Types/API/URL";
-import { getSloAffectedResourceMarkdownLines } from "../../Utils/Slo/SloAffectedResourceMarkdown";
+import LinkedAffectedResources, {
+  LinkedAffectedResource,
+} from "../Utils/AffectedResources/LinkedAffectedResources";
 import DatabaseCommonInteractionProps from "../../Types/BaseDatabase/DatabaseCommonInteractionProps";
 import SortOrder from "../../Types/BaseDatabase/SortOrder";
 import LIMIT_MAX, { LIMIT_PER_PROJECT } from "../../Types/Database/LimitMax";
@@ -73,7 +75,6 @@ import AlertReminderRule from "../../Models/DatabaseModels/AlertReminderRule";
 import WorkspaceType from "../../Types/Workspace/WorkspaceType";
 import NotificationRuleWorkspaceChannel from "../../Types/Workspace/NotificationRules/NotificationRuleWorkspaceChannel";
 import AlertWorkspaceMessages from "../Utils/Workspace/WorkspaceMessages/Alert";
-import Monitor from "../../Models/DatabaseModels/Monitor";
 import ServiceLevelObjective from "../../Models/DatabaseModels/ServiceLevelObjective";
 import MonitorService from "./MonitorService";
 import { MessageBlocksByWorkspaceType } from "./WorkspaceNotificationRuleService";
@@ -1024,19 +1025,6 @@ export class Service extends DatabaseService<Model> {
           labels: {
             name: true,
           },
-          monitor: {
-            name: true,
-            _id: true,
-          },
-          /*
-           * This read runs as root, so projectId comes along and the feed
-           * names only this project's SLOs (getSloAffectedResourceMarkdownLines).
-           */
-          serviceLevelObjectives: {
-            name: true,
-            _id: true,
-            projectId: true,
-          },
         },
         props: {
           isRoot: true,
@@ -1067,31 +1055,28 @@ ${alert.description || "No description provided."}
       }
 
       /*
-       * The monitor, then the SLOs this alert is linked to. A burn-rate alert
-       * has no monitor, so its SLO is the only resource there is to name -
-       * and the feed's only way back to the objective that raised it. The SLO
-       * link is built inline: ServiceLevelObjectiveService cannot be imported
-       * here (it reaches this service through the burn-rate rule service).
+       * Everything the alert's Affected Resources card lists: its monitor,
+       * the hosts, clusters and services it is attached to, then its SLOs. A
+       * burn-rate alert has no monitor, so its SLO is the only resource there
+       * is to name - and the feed's only way back to the objective that
+       * raised it.
        */
-      const sloLines: Array<string> =
-        alert.serviceLevelObjectives && alert.serviceLevelObjectives.length > 0
-          ? getSloAffectedResourceMarkdownLines({
-              dashboardUrl: await DatabaseConfig.getDashboardUrl(),
-              projectId: alert.projectId!,
-              serviceLevelObjectives: alert.serviceLevelObjectives,
-            })
-          : [];
+      const resources: Array<LinkedAffectedResource> =
+        await LinkedAffectedResources.readForAlert({
+          service: this,
+          projectId: alert.projectId!,
+          alertId: alert.id!,
+        });
 
-      if (alert.monitor || sloLines.length > 0) {
+      if (resources.length > 0) {
         feedInfoInMarkdown += `🌎 **Resources Affected**:\n`;
 
-        if (alert.monitor) {
-          const monitor: Monitor = alert.monitor;
-          feedInfoInMarkdown += `- [${monitor.name}](${(await MonitorService.getMonitorLinkInDashboard(alert.projectId!, monitor.id!)).toString()})\n`;
-        }
-
-        for (const sloLine of sloLines) {
-          feedInfoInMarkdown += `${sloLine}\n`;
+        for (const resourceLine of LinkedAffectedResources.getMarkdownLines({
+          dashboardUrl: await DatabaseConfig.getDashboardUrl(),
+          projectId: alert.projectId!,
+          resources: resources,
+        })) {
+          feedInfoInMarkdown += `${resourceLine}\n`;
         }
 
         feedInfoInMarkdown += `\n\n`;
@@ -1768,6 +1753,12 @@ ${alertSeverity.name}
     notifyOwners: boolean;
     rootCause: string | undefined;
     stateChangeLog: JSONObject | undefined;
+    /*
+     * The user the change is credited to (the alert's feed names them), for
+     * a change a user asked for that is written on their behalf. Unset for
+     * changes the system makes on its own.
+     */
+    createdByUserId?: ObjectID | undefined;
     props: DatabaseCommonInteractionProps | undefined;
   }): Promise<void> {
     const {
@@ -1777,6 +1768,7 @@ ${alertSeverity.name}
       notifyOwners,
       rootCause,
       stateChangeLog,
+      createdByUserId,
       props,
     } = data;
 
@@ -1822,6 +1814,10 @@ ${alertSeverity.name}
       statusTimeline.rootCause = rootCause;
     }
 
+    if (createdByUserId) {
+      statusTimeline.createdByUserId = createdByUserId;
+    }
+
     await AlertStateTimelineService.create({
       data: statusTimeline,
       props: props || {},