Bulk migration partial failure

When you migrate legacy machine learning anomaly detection jobs to cross-project search in Kibana, the Migrate all legacy jobs action updates each job's datafeeds in turn. Some jobs may succeed while others fail — for example, because validate-before-mint rejected a search probe or a job owner account is invalid.

Symptoms

• The bulk migration summary reports a mix of succeeded and failed jobs.
• Some jobs show an API key error indicator in the anomaly detection jobs list; others show as migrated.
• Failed jobs still use legacy credentials or lack project_routing while succeeded neighbors were updated.

Identify failed jobs

Note every job_id listed as failed in the migration summary. For each failed job:

1. Open the job in Machine Learning → Anomaly Detection and review Messages for the API error text.
2. Check GET _ml/datafeeds/{datafeed_id} for missing authorization.cloud_api_key.id on jobs that should have migrated.

For mint-time errors on individual jobs, see Cloud token mint failure. For routing or project-reference failures, see Routing matches no project and Stale project reference.

Record failed job ids

Keep the list of failed job_id values from the summary. Bulk migration does not automatically retry failed jobs.

Retry each job individually

For each failed job:

1. Fix the underlying error (project link, project_routing, ownership, or permissions).
2. Migrate that job alone from the jobs list flyout or run a single-job update:

				POST _ml/datafeeds/{datafeed_id}/_update
					{
  "project_routing": "_origin"
}
		

Adjust project_routing to the scope you selected during bulk migration (bulk migration defaults to _origin unless you chose another scope).

Use the single-job migration path

The per-job migration flyout lets you confirm search scope before retrying. Prefer this over re-running bulk migration for the entire list.

Verify recovery

After individual retries, confirm the API key error indicator clears and that GET _ml/datafeeds/{datafeed_id} shows the expected project_routing and cloud API key id for every previously failed job.