Loading

Troubleshoot machine learning with CPS

Use this section to diagnose and resolve common problems with anomaly detection datafeeds that search across linked projects on Elastic Cloud Serverless. These topics apply when CPS-enabled ML datafeeds are available for your project.

Tip

If you can't find your issue here, explore the other troubleshooting topics or contact us.

Gather context from the following sources before narrowing down a CPS datafeed issue:

  • ML job Messages in Kibana — Open Machine Learning → Anomaly Detection, select the job, and review the Messages tab for audit entries and warnings about linked projects, credentials, or scope changes.
  • .ml-notifications-* and .ml-annotations-* indices — Search these indices for the job or datafeed ID to find historical audit messages and scope-change annotations.
  • GET _ml/datafeeds/{id}/_stats — Inspect remote_cluster_stats for per-project availability, skip counts, and cross-project search health over recent extraction cycles.
  • GET _ml/datafeeds/{id} — Confirm the effective project_routing expression and whether the datafeed uses qualified index references.
project_routing value Effective search scope
Empty or omitted (null) Flat world — searches the origin project and all linked projects
_origin or _alias:_origin Local only — searches the origin project
_alias:prod-* (or other tag expression) Subset — searches only linked projects whose tags match the expression
Qualified index (project:index) Specific project — searches the named project and index

For syntax details and examples, refer to Project routing in CPS.

Contact Elastic support when:

  • A linked project or region appears unavailable across multiple jobs and you have confirmed project linking in Elastic Cloud.
  • The same datafeed fails repeatedly after you apply the fixes in these topics.
  • The origin project reports memory pressure or out-of-memory errors while CPS datafeeds are running, and narrowing project_routing does not relieve the symptoms.