Alibaba Cloud provides a variety of Elasticsearch versions. This topic describes the features that are supported in each version. You can select a version based on your business requirements.
Comparison among Elasticsearch editions
Alibaba Cloud Elasticsearch offers the following editions: Standard Edition, Kernel-enhanced Edition, and Vector Enhanced Edition. The editions support different cluster versions and features.
Item | Kernel-enhanced Edition | Vector Enhanced Edition and Standard Edition |
Supported versions | 7.16, 7.10, and 6.7 | Vector Enhanced Edition: 8.15 Standard Edition: 8.13, 8.9, 8.5, 7.7, 6.8, 6.3, 5.6, and 5.5 |
Main features |
|
|
Use scenarios | All use scenarios of Elasticsearch. Especially suitable for the following scenarios:
| All use scenarios of Elasticsearch. Examples: information retrieval, search, log analysis, and vector retrieval. |
User profiles |
|
|
Billable items | You are charged based on the specifications, storage space, and number of nodes in your cluster. The basic and advanced enhancements offered by the AliES kernel are provided free of charge as plug-ins. You can install the plug-ins based on your business requirements. | You are charged based on the specifications, storage space, and number of nodes in your cluster. |
Open source version features
Each Alibaba Cloud Elasticsearch edition supports all features provided by open source Elasticsearch, and the advanced features of open source Elasticsearch of the Platinum edition are free of charge. This section describes the feature changes in different versions:
V7.16, V7.10, and V6.7 clusters are of Kernel-enhanced Edition. These clusters use the deeply optimized kernel AliES. This enables the clusters to provide enhancements based on open source features. For more information, see Features supported by Kernel-enhanced Edition.
8.15
New features:
Fields are optimized for vector indexes. For more information, see dense-vector.
By default, the INT8_HNSW algorithm is used instead of the Hierarchical Navigable Small World (HNSW) algorithm, and INT8 quantization is enabled.
INT4 quantization is supported, which helps save memory by eight times.
The vector type
bit
is supported.
Single instruction, multiple data (SIMD) instructions can be used to improve the performance of merging INT8-quantized indexes in the AArch64 architecture by about three times.
A rerank phase is supported, and rerank models are supported by the text_similarity_reranker API. For more information, see text-similarity-reranker-retriever.
The retriever query syntax is added to support multimodal searches. For more information, see retriever.
The semantic_text field type is supported to facilitate semantic searches. For more information, see semantic-text.
The sparse_vector syntax instead of text_expansion is used for sparse queries. For more information, see query-dsl-sparse-vector-query.
The query_rules API reaches general availability (GA). For more information, see query-rules-apis.
Nested fields are supported by index sorting. For more information, see index-modules-index-sorting.
The
logsdb
index mode is available for logging scenarios. For more information, see logs-data-stream.Lucene 9.11 is released to improve query performance and memory efficiency. For more information, see apache-lucenetm-9110-available.
For more information about the feature changes, see What's new in 8.15 and What's new in 8.14.
8.13
New features:
The maximum number of vector dimensions is increased to 4,096. For more information, see 4096 dimension dense vector.
The scalar quantization feature is supported by vector indexes. This feature can enable a vector index to use approximately 75% less memory. For more information, see Understanding scalar quantization in Lucene.
Sparse vectors are supported, and the sparse_vector type is added. For more information, see Sparse vector.
Shard-level query parallelization is supported. For more information, see Query parallelization.
Nested vectors are supported. You can split large documents into chunks and create a vector index for each chunk. For more information, see Multiple results from the same doc with nested vectors.
The Learning To Rank (LTR) feature is added to enable query result re-ranking in the restore phase. For more information, see Learning To Rank.
New inference APIs are supported to integrate external model services. For more information, see Inference APIs.
SIMD is supported to improve vector query performance. For more information, see Accelerating vector search with SIMD instructions.
For more information about the feature changes, see What's new in 8.13.
8.9
New features:
Mixed sorting of text and vector recall results is supported. For more information, see Reciprocal rank fusion (RRF).
The maximum number of vector dimensions is increased to 2,048. For more information, see Increase max number of vector dims to 2048.
The performance for brute-force searches is improved. For more information, see Improve brute force vector search speed.
Multiple fields can be used at the same time in k-nearest neighbors (k-NN) queries. For more information, see Allow more than one KNN search clause.
The built-in model ELSER is provided. ELSER is short for Elastic Learned Sparse EncodeR. For more information, see ELSER.
Scheduling management of distributed natural language processing (NLP) models is supported. For more information, see Make native inference generally available.
The performance for writing data with primary keys is improved. For more information, see Optimize primary keys.
The query performance of constant keyword fields is improved. For more information, see Skip shards when querying constant keyword fields.
Time series data streams (TSDSs) and the downsampling feature are supported. For more information, see TSDS and Downsample.
The memory for raw text is optimized, and ThreadLocal is removed. For more information, see Remove uses of deprecated LeafReader.
For more information about the feature changes, see What's new in 8.9.
8.5
New features:
HNSW-based vector similarity searches are supported. HNSW is short for Hierarchical Navigable Small World. For more information, see kNN search.
The TSDS feature is supported. For more information, see TSDS.
PyTorch models can be uploaded. For more information, see start-trained-model-deployment.
geo_grid queries are supported. For more information, see Geo-grid query.
Security-related configurations are simplified. For more information, see Start the Elastic Stack with security enabled automatically.
Lucene compression algorithms are improved to reduce index size.
The performance of range queries is enhanced.
Runtime fields of the lookup type are supported. For more information, see lookup-runtime-fields.
random_sampler aggregation is supported. For more information, see Random sampler aggregation.
The heap memory consumption of master nodes and data nodes is reduced.
Mapping types are no longer supported. You can use RESTful API compatibility to use mapping types. For more information, see rest-api-compatibility.
The index protection feature is supported. By default, users whose usernames are elastic can read data only from Elasticsearch built-in indexes.
For more information about the feature changes, see Breaking changes in 8.5.
7.16
New features:
SQL-based cross-cluster searches are supported.
Enrich policies of the range type are supported by ingest pipelines.
Cache is optimized to improve query performance.
Indexes can be added to and removed from data streams.
Cluster UUIDs and names are added to audit logs.
For more information about the feature changes, see Breaking changes in 7.16.
7.10
New features:
The compression of storage fields is improved, which reduces storage costs.
Event Query Language (EQL) is used to improve security.
The default value of search.max_buckets is changed from 10000 to 65535.
Queries that are not case-sensitive are supported. To implement such queries, you must set the optional parameter case_insensitive to true.
For more information about the feature changes, see Breaking changes in 7.10.
7.7
New features:
The default number of shards in the index template is changed from 5 to 1.
Mapping types are removed. You do not need to specify a mapping type when you define a mapping or an index template. For more information, see Removal of mapping types.
By default, a maximum of 10,000 documents can be returned for each request. If more than 10,000 matching documents exist, Elasticsearch returns only 10,000 matching documents. For more information, see track_total_hits 10000 default.
By default, a single data node can store a maximum of 1,000 shards. You can use the cluster.max_shards_per_node parameter to change this limit. For more information, see Cluster Shard Limit.
By default, a maximum of 500 scrolls can be performed. You can use the search.max_open_scroll_context parameter to change this limit. For more information, see Scroll Search Context.
The parent circuit breaker works based on the current memory usage. This is controlled by the indices.breaker.total.use_real_memory parameter. By default, the parent circuit breaker starts to work when the current memory usage reaches 95% of JVM heap memory usage. This indicates that Elasticsearch uses the maximum memory availability to avoid out of memory (OOM) issues. For more information, see Circuit Breaker.
The _all field is removed to improve search performance.
Intervals queries are supported. Elasticsearch searches for and returns documents based on the order and proximity of matching terms.
After the audit logging feature is enabled, audit events are persisted to <clustername>_audit.json in the file system of each node. The audit events cannot be stored in indexes. For more information, see Enabling audit logging.
For more information about the feature changes, see Breaking changes in 7.0.
6.x (6.3, 6.7, and 6.8)
New features:
An index can have only one type, and the _doc type is recommended.
The index lifecycle management (ILM) feature is introduced from V6.6.0 to reduce index O&M costs.
The historical data rollup feature is introduced to help summarize historical data.
Elasticsearch SQL, an X-Pack component, is supported in V6.3 and later. It enables SQL statements to be converted to domain-specific language (DSL) statements. This reduces costs for learning DSL.
The Composite, Parent, and Weighted Avg aggregation functions are supported.
For more information about the feature changes, see Breaking changes in 6.0.
5.x (5.5 and 5.6)
New features:
An index can have multiple types, and custom types are supported.
The STRING data type is replaced by the TEXT or KEYWORD data type.
The values of fields in indexes are changed from not_analyzed or no to true or false.
The DOUBLE data type is replaced by the FLOAT data type to reduce storage costs.
Java High Level REST Client is launched to replace Transport Client.
For more information about the feature changes, see Breaking changes in 5.0.
References
You can view the edition and version of your Alibaba Cloud Elasticsearch cluster on the Basic Information page of the cluster in the Elasticsearch console. For more information, see View the basic information of a cluster.
For information about how to purchase an Alibaba Cloud Elasticsearch cluster, see Create an Alibaba Cloud Elasticsearch cluster.
For information about how to evaluate the specifications and storage capacity of an Alibaba Cloud Elasticsearch cluster, see Evaluate specifications and storage capacity.