What’s New in Elasticsearch 8.0? Key Features and Migration Tips

Elasticsearch is a Lucene‑based distributed full‑text search engine with HTTP and schema‑free JSON APIs, written in Java and released under the Apache license. Official clients exist for Java, .NET, PHP, Python, Groovy, Ruby, and many other languages.

7.x REST API Compatibility

Version 8.0 adds optional compatibility headers that let you send 7.x‑compatible requests to an 8.0 cluster and receive 7.x‑compatible responses, easing migration.

Security Features Enabled by Default

Security (authentication, authorization, TLS) is now enabled and configured automatically on first start. A registration token is generated for Kibana or other nodes, eliminating the need to create certificates or edit YAML files.

On Linux ARM or macOS M1, the elastic password and Kibana token are not generated automatically. Use bin/elasticsearch-reset-password -u elastic to create the password, then bin/elasticsearch-create-enrollment-token -s kibana to generate a token.

Better Protection of System Indices

System indices now require the allow_restricted_indices permission set to true. The superuser role no longer has write access to system indices, and the elastic superuser cannot modify them by default.

New KNN Search API (Technical Preview)

The KNN API uses the dense_vector field to find the k nearest vectors, supporting recommendation and NLP relevance ranking. It offers faster approximate searches compared to the previous exact script_score approach.

Storage‑Saving Field Encodings

Updates to the inverted index reduce storage for keyword, match_only_text, and text fields. In benchmark tests on the message field, index size dropped 14.4% for match_only_text and overall disk usage fell 3.5%.

Faster Indexing for geo_point, geo_shape, and Range Fields

Multi‑dimensional point indexing speed for geo_point, geo_shape, and range fields improved by 10‑15% in Lucene‑level benchmarks.

PyTorch Model Support for NLP

External PyTorch models can now be uploaded and used for inference, bringing modern NLP and search use cases to the Elastic Stack.

Other Changes

Aggregations: removed adjacency matrix, deprecated MovingAverage pipeline, removed deprecated sorting fields _time and _term, and deleted deprecated date‑histogram intervals.

Allocation: removed include_relocations setting.

Analysis: cleaned up versioned deprecations, removed pre‑configured delimited_payload_filter.

Authentication: added file and native realms by default, stopped setting NameID format in policies, enforced order for realms.

Cluster Coordination: removed connection timeout and delayed state recovery support.

Distributed: removed sync flush and cluster.remote.connect setting.

Engine: rejected only_expunge_deletes and max_num_segments on force merge, removed per‑type index stats, removed translog retention setting.

Features/CAT APIs: removed deprecated local parameter from _cat/indices and _cat/shards.

Features/ILM+SLM: defaulted cluster.routing.allocation.enforce_default_tier_preference to true.

Features/Indices APIs: defaulted prefer_v2_templates to true, removed _upgrade API, removed include_type_name parameter, deleted template field from index templates.

Infra/Core: removed nodes/0 folder prefix, deleted bootstrap.system_call_filter, node.max_local_storage_nodes, Joda dependency, and camel‑case date/time format names.

Packaging: removed SysV init support, dropped JAVA_HOME support, and now requires Java 17 to run Elasticsearch.