[ML] Fix datafeed start time is incorrect when the job has trailing empty buckets (#71976)

Co-authored-by: Elastic Machine <elasticmachine@users.noreply.github.com>

x-pack/plugins/ml/common/types/anomaly_detection_jobs/summary_job.ts

@@ -30,6 +30,7 @@ export interface MlSummaryJob {
   isSingleMetricViewerJob: boolean;
   deleting?: boolean;
   latestTimestampSortValue?: number;
+  earliestStartTimestampMs?: number;
 }
 
 export interface AuditMessage {

x-pack/plugins/ml/common/util/job_utils.test.ts

@@ -18,8 +18,10 @@ import {
   prefixDatafeedId,
   getSafeAggregationName,
   getLatestDataOrBucketTimestamp,
+  getEarliestDatafeedStartTime,
 } from './job_utils';
 import { CombinedJob, Job } from '../types/anomaly_detection_jobs';
+import moment from 'moment';
 
 describe('ML - job utils', () => {
   describe('calculateDatafeedFrequencyDefaultSeconds', () => {
@@ -581,4 +583,22 @@ describe('ML - job utils', () => {
       expect(getLatestDataOrBucketTimestamp(undefined, undefined)).toBe(undefined);
     });
   });
+
+  describe('getEarliestDatafeedStartTime', () => {
+    test('returns expected value when no gap in data at end of bucket processing', () => {
+      expect(getEarliestDatafeedStartTime(1549929594000, 1549928700000)).toBe(1549929594000);
+    });
+    test('returns expected value when there is a gap in data at end of bucket processing', () => {
+      expect(getEarliestDatafeedStartTime(1549929594000, 1562256600000)).toBe(1562256600000);
+    });
+    test('returns expected value when bucket span is provided', () => {
+      expect(
+        getEarliestDatafeedStartTime(1549929594000, 1562256600000, moment.duration(1, 'h'))
+      ).toBe(1562260200000);
+    });
+
+    test('returns expected value when job has not run', () => {
+      expect(getLatestDataOrBucketTimestamp(undefined, undefined)).toBe(undefined);
+    });
+  });
 });

x-pack/plugins/ml/common/util/job_utils.ts

@@ -6,7 +6,7 @@
 
 import _ from 'lodash';
 import semver from 'semver';
-import { Duration } from 'moment';
+import moment, { Duration } from 'moment';
 // @ts-ignore
 import numeral from '@elastic/numeral';
 
@@ -621,6 +621,23 @@ function isValidTimeInterval(value: string | undefined): boolean {
   return parseTimeIntervalForJob(value) !== null;
 }
 
+// The earliest start time for the datafeed should be the max(latest_record_timestamp, latest_bucket.timestamp + bucket_span).
+export function getEarliestDatafeedStartTime(
+  latestRecordTimestamp: number | undefined,
+  latestBucketTimestamp: number | undefined,
+  bucketSpan?: Duration | null | undefined
+): number | undefined {
+  if (latestRecordTimestamp !== undefined && latestBucketTimestamp !== undefined) {
+    // if bucket span is available (e.g. 15m) add it to the latest bucket timestamp in ms
+    const adjustedBucketStartTime = bucketSpan
+      ? moment(latestBucketTimestamp).add(bucketSpan).valueOf()
+      : latestBucketTimestamp;
+    return Math.max(latestRecordTimestamp, adjustedBucketStartTime);
+  } else {
+    return latestRecordTimestamp !== undefined ? latestRecordTimestamp : latestBucketTimestamp;
+  }
+}
+
 // Returns the latest of the last source data and last processed bucket timestamp,
 // as used for example in setting the end time of results views for cases where
 // anomalies might have been raised after the point at which data ingest has stopped.

x-pack/plugins/ml/public/application/jobs/jobs_list/components/start_datafeed_modal/start_datafeed_modal.js

@@ -222,6 +222,6 @@ StartDatafeedModal.propTypes = {
 };
 
 function getLowestLatestTime(jobs) {
-  const times = jobs.map((j) => j.latestTimestampSortValue);
+  const times = jobs.map((j) => j.earliestStartTimestampMs || 0);
   return moment(Math.min(...times));
 }

x-pack/plugins/ml/server/models/job_service/jobs.ts

@@ -8,6 +8,7 @@ import { i18n } from '@kbn/i18n';
 import { uniq } from 'lodash';
 import Boom from 'boom';
 import { ILegacyScopedClusterClient } from 'kibana/server';
+import { parseTimeIntervalForJob } from '../../../common/util/job_utils';
 import { JOB_STATE, DATAFEED_STATE } from '../../../common/constants/states';
 import {
   MlSummaryJob,
@@ -24,11 +25,11 @@ import { resultsServiceProvider } from '../results_service';
 import { CalendarManager, Calendar } from '../calendar';
 import { fillResultsWithTimeouts, isRequestTimeout } from './error_utils';
 import {
+  getEarliestDatafeedStartTime,
   getLatestDataOrBucketTimestamp,
   isTimeSeriesViewJob,
 } from '../../../common/util/job_utils';
 import { groupsProvider } from './groups';
-
 export interface MlJobsResponse {
   jobs: Job[];
   count: number;
@@ -171,6 +172,11 @@ export function jobsProvider(mlClusterClient: ILegacyScopedClusterClient) {
         description: job.description || '',
         groups: Array.isArray(job.groups) ? job.groups.sort() : [],
         processed_record_count: job.data_counts?.processed_record_count,
+        earliestStartTimestampMs: getEarliestDatafeedStartTime(
+          dataCounts?.latest_record_timestamp,
+          dataCounts?.latest_bucket_timestamp,
+          parseTimeIntervalForJob(job.analysis_config?.bucket_span)
+        ),
         memory_status: job.model_size_stats ? job.model_size_stats.memory_status : '',
         jobState: job.deleting === true ? deletingStr : job.state,
         hasDatafeed,
@@ -182,8 +188,8 @@ export function jobsProvider(mlClusterClient: ILegacyScopedClusterClient) {
         latestTimestampMs: dataCounts?.latest_record_timestamp,
         earliestTimestampMs: dataCounts?.earliest_record_timestamp,
         latestResultsTimestampMs: getLatestDataOrBucketTimestamp(
-          dataCounts?.latest_record_timestamp as number,
-          dataCounts?.latest_bucket_timestamp as number
+          dataCounts?.latest_record_timestamp,
+          dataCounts?.latest_bucket_timestamp
         ),
         isSingleMetricViewerJob: isTimeSeriesViewJob(job),
         nodeName: job.node ? job.node.name : undefined,