Elasticsearch 9.5.0 contains a query-correctness defect

Further investigation confirmed the impact of this issue was limited to boolean queries containing a must, filter, or should clause, along with a must_not clause on unindexed fields. Such queries can return false-positives returning documents that should have been excluded and report higher document counts than expected.

We have identified the root cause, a fix is in progress, and we are preparing a patch release. We will provide a further update by 10:00 UTC August 6.

Elasticsearch 9.5.0 contains a defect that can cause searches to return incorrect results for any index or data stream that disables indexing on a queried field . A query that excludes values using a must_not clause may fail to exclude them when the targeted field has doc values enabled but is not indexed for search. Affected searches can return documents that should have been excluded and report higher document counts than expected, and results may vary between runs of the same query.

No error is raised, so affected queries appear to succeed. Dashboards, alerting rules, and anything else built on these searches may report inaccurate values. Both the query DSL and ES|QL are affected.

Time series (TSDB) data streams are the most likely to be affected, because indexing is disabled by default for these fields. Columnar indices, currently in technical preview, are affected for the same reason. Any index or data stream that disables indexing on a queried field can be affected.

This affects Elasticsearch 9.5.0 on Elastic Cloud Hosted, Elastic Cloud Enterprise, and self-managed deployments, as well as Elasticsearch Serverless projects. Elasticsearch 9.4.x and earlier are not affected. There is no impact to cluster availability, connectivity, or data ingestion.

What you can do:

- If you have not yet upgraded to 9.5.0, we recommend deferring the upgrade until a fixed version is available.
- If you are already running 9.5.0, contact Elastic Support if you need help determining whether your searches are affected.

We have identified the root cause, a fix is in progress, and we are preparing a patch release. We will provide a further update by 10:00 UTC August 6.