[8.15] [Response Ops][Alerting] Deleting ad hoc run task if the last …

…schedule entry ends in a timeout (#187496) (#187810) # Backport This will backport the following commits from `main` to `8.15`: - [[Response Ops][Alerting] Deleting ad hoc run task if the last schedule entry ends in a timeout (#187496)](#187496)  ### Questions ? Please refer to the [Backport tool documentation](https://github.com/sqren/backport)  Co-authored-by: Ying Mao <[email protected]>
elastic · Jul 9, 2024 · bcdf5e9 · bcdf5e9
1 parent 056d64a
commit bcdf5e9
Show file tree

Hide file tree

Showing 2 changed files with 84 additions and 2 deletions.
diff --git a/x-pack/plugins/alerting/server/task_runner/ad_hoc_task_runner.test.ts b/x-pack/plugins/alerting/server/task_runner/ad_hoc_task_runner.test.ts
@@ -1244,6 +1244,84 @@ describe('Ad Hoc Task Runner', () => {
       expect(logger.error).not.toHaveBeenCalled();
     });
 
+    test('should handle task cancellation signal due to timeout when for last schedule entry', async () => {
+      encryptedSavedObjectsClient.getDecryptedAsInternalUser.mockResolvedValue({
+        ...mockedAdHocRunSO,
+        attributes: {
+          ...mockedAdHocRunSO.attributes,
+          schedule: [{ ...schedule1, status: adHocRunStatus.COMPLETE }, schedule2],
+        },
+      });
+      const taskRunner = new AdHocTaskRunner({
+        context: taskRunnerFactoryInitializerParams,
+        internalSavedObjectsRepository,
+        taskInstance: mockedTaskInstance,
+      });
+
+      const promise = taskRunner.run();
+      await Promise.resolve();
+      await taskRunner.cancel();
+      await promise;
+      await taskRunner.cleanup();
+
+      expect(encryptedSavedObjectsClient.getDecryptedAsInternalUser).toHaveBeenCalledWith(
+        AD_HOC_RUN_SAVED_OBJECT_TYPE,
+        'abc',
+        {}
+      );
+      expect(ruleTypeRegistry.get).toHaveBeenCalledWith('siem.queryRule');
+      expect(ruleTypeRegistry.ensureRuleTypeEnabled).toHaveBeenCalledWith('siem.queryRule');
+      expect(mockValidateRuleTypeParams).toHaveBeenCalledWith(
+        mockedAdHocRunSO.attributes.rule.params,
+        ruleTypeWithAlerts.validate.params
+      );
+
+      // @ts-ignore - accessing private variable
+      // should run the first entry in the schedule
+      expect(taskRunner.scheduleToRunIndex).toEqual(1);
+      expect(RuleRunMetricsStore).toHaveBeenCalledTimes(1);
+      expect(ruleTypeWithAlerts.executor).toHaveBeenCalledTimes(1);
+
+      expect(internalSavedObjectsRepository.update).toHaveBeenCalledWith(
+        AD_HOC_RUN_SAVED_OBJECT_TYPE,
+        mockedAdHocRunSO.id,
+        {
+          schedule: [
+            { ...schedule1, status: adHocRunStatus.COMPLETE },
+            { ...schedule2, status: adHocRunStatus.TIMEOUT },
+          ],
+        },
+        { namespace: undefined, refresh: false }
+      );
+
+      expect(internalSavedObjectsRepository.delete).toHaveBeenCalledWith(
+        AD_HOC_RUN_SAVED_OBJECT_TYPE,
+        mockedAdHocRunSO.id,
+        { namespace: undefined, refresh: false }
+      );
+
+      testAlertingEventLogCalls({
+        status: 'ok',
+        timeout: true,
+        backfillRunAt: schedule2.runAt,
+        backfillInterval: schedule2.interval,
+      });
+      expect(logger.debug).toHaveBeenCalledTimes(3);
+      expect(logger.debug).nthCalledWith(
+        1,
+        `Executing ad hoc run for rule test:rule-id for runAt ${schedule2.runAt}`
+      );
+      expect(logger.debug).nthCalledWith(
+        2,
+        `Cancelling execution for ad hoc run with id abc for rule type test with id rule-id - execution exceeded rule type timeout of 3m`
+      );
+      expect(logger.debug).nthCalledWith(
+        3,
+        `Aborting any in-progress ES searches for rule type test with id rule-id`
+      );
+      expect(logger.error).not.toHaveBeenCalled();
+    });
+
     test('should handle task cancellation that leads to executor throwing error', async () => {
       ruleTypeWithAlerts.executor.mockImplementationOnce(() => {
         throw new Error('Search has been aborted due to cancelled execution');

diff --git a/x-pack/plugins/alerting/server/task_runner/ad_hoc_task_runner.ts b/x-pack/plugins/alerting/server/task_runner/ad_hoc_task_runner.ts
@@ -20,7 +20,7 @@ import {
   TaskErrorSource,
 } from '@kbn/task-manager-plugin/server';
 import { nanosToMillis } from '@kbn/event-log-plugin/common';
-import { RunResult } from '@kbn/task-manager-plugin/server/task';
+import { CancellableTask, RunResult } from '@kbn/task-manager-plugin/server/task';
 import { AdHocRunStatus, adHocRunStatus } from '../../common/constants';
 import { RuleRunnerErrorStackTraceLog, RuleTaskStateAndMetrics, TaskRunnerContext } from './types';
 import { getExecutorServices } from './get_executor_services';
@@ -66,7 +66,7 @@ interface RunParams {
   validatedParams: RuleTypeParams;
 }
 
-export class AdHocTaskRunner {
+export class AdHocTaskRunner implements CancellableTask {
   private readonly context: TaskRunnerContext;
   private readonly executionId: string;
   private readonly internalSavedObjectsRepository: ISavedObjectsRepository;
@@ -573,6 +573,10 @@ export class AdHocTaskRunner {
             : undefined,
       },
     });
+    this.shouldDeleteTask = !this.hasAnyPendingRuns();
+
+    // cleanup function is not called for timed out tasks
+    await this.cleanup();
   }
 
   async cleanup() {