Migrating to 8.16

Details
When a synonym or synonym_graph token filter is configured with updateable: true, the default lenient value will now be true.

Impact
synonym or synonym_graph token filters configured with updateable: true will ignore invalid synonyms by default. This prevents shard initialization errors on invalid synonyms.

Details
Elasticsearch 8.16 changes the version of the JDK that is included from version 22 to version 23. This changes the locale database that is used by Elasticsearch from the COMPAT database to the CLDR database. This change can cause significant differences to the textual date formats accepted by Elasticsearch, and to calculated week-dates.

If you run Elasticsearch 8.16 on JDK version 22 or below, it will use the COMPAT locale database to match the behavior of 8.15. However, starting with Elasticsearch 9.0, Elasticsearch will use the CLDR database regardless of JDK version it is run on.

Impact
This affects you if you use custom date formats using textual or week-date field specifiers. If you use date fields or calculated week-dates that change between the COMPAT and CLDR databases, then this change will cause Elasticsearch to reject previously valid date fields as invalid data. You might need to modify your ingest or output integration code to account for the differences between these two JDK versions.

Starting in version 8.15.2, Elasticsearch will log deprecation warnings if you are using date format specifiers that might change on upgrading to JDK 23. These warnings are visible in Kibana.

For detailed guidance, refer to Differences in locale information between JDK versions and the Elastic blog.

Details
Removes an undocumented syntax from ESQL: META FUNCTION. This was never reliable or really useful. Consult the documentation instead.

Impact
Removes an undocumented syntax from ESQL: META FUNCTION

Details
In this release (8.16), we have introduced major changes to the retrievers framework and how they can be evaluated, focusing mainly on compound retrievers like rrf and text_similarity_reranker, which allowed us to support full composability (i.e. any retriever can be nested under any compound retriever), as well as supporting additional search features like collapsing, explaining, aggregations, and highlighting.

To ensure consistency, and given that this rework is not available until 8.16, rrf and text_similarity_reranker retriever queries would now throw an exception in a mixed cluster scenario, where there are nodes both in current or later (i.e. >= 8.16) and previous ( ⇐ 8.15) versions.

As part of the rework, we have also removed the _rank property from the responses of an rrf retriever.

Impact
- Users will not be able to use the rrf and text_similarity_reranker retrievers in a mixed cluster scenario with previous releases (i.e. prior to 8.16), and the request will throw an IllegalArgumentException. - _rank has now been removed from the output of the rrf retrievers so trying to directly parse the field will throw an exception

Details
In this release we introduced global retention settings that fulfil the following criteria:

As a result, we defined different types of retention:

Considering the above changes, having a field named retention in the usage API was confusing. For this reason, we renamed it to data_retention and added telemetry about the other configurations too.

Impact
Users that use the field data_lifecycle.retention should use the data_lifecycle.data_retention

Details
kp, dutch_kp, dutchKp and lovins stemmers are deprecated and will be removed.

Impact
These stemmers will be removed and will be no longer supported.

Details
edge_ngram will no longer accept the side parameter.

Impact
Users will need to update any usage of edge_ngram token filter that utilizes side. If the back value was used, they can achieve the same behavior by using the reverse token filter.

Details
Indices beginning with a dot . are reserved for system and internal indices, and should not be used by and end-user. Additionally, composable index templates that contain patterns for dot-prefixed indices should also be avoided, as these patterns are meant for internal use only. In a future Elasticsearch version, creation of these dot-prefixed indices will no longer be allowed.

Impact
Requests performing an action that would create an index beginning with a dot (indexing a document, manual creation, reindex), or creating an index template with index patterns beginning with a dot, will contain a deprecation header warning about dot-prefixed indices in the response.

Details
Search API parameter sub_searches will no longer be a supported and will be removed in future releases. Similarly, rrf can only be used through the specified retriever and no longer though the rank parameter

Impact
Requests specifying rrf through rank and/or sub_searches elements will be disallowed in a future version. Users should instead utilize the new retriever parameter.

Details
Range query will not longer accept to, from, include_lower, and include_upper parameters.

Impact
Instead use gt, gte, lt and lte parameters.