Terms

Term

Description

MaxCompute data source

A full data source. Raw data is stored in MaxCompute by partition. Incremental data can be pushed by using API operations.

API data source

A real-time incremental data source. Data is updated by pushing it through API operations.

OSS data source

A full data source. Raw data is stored in Object Storage Service (OSS) by bucket. Incremental data can be pushed by using API operations.

Data Lake Formation (DLF)

A full data source. Raw data is migrated to Data Lake Formation (DLF). Incremental data can be pushed by using API operations.

Document

A document is a searchable unit of structured data. A document contains one or more fields and must contain a primary key field. An OpenSearch Retrieval Engine Edition instance uses the primary key value to uniquely identify a document. If a primary key value is duplicated, the existing document is overwritten.

Field

A field is a component of a document. A field consists of a field name and field content.

Multi-value field

A field that contains multiple different values. Each value is independent.

Primary key

The field that uniquely identifies a document.

Term

Description

Query node (QRS)

A role in online retrieval that parses query requests and merges the results returned by Searchers.

Data node (Searcher)

A role in the online retrieval system that loads index data and provides retrieval services.

Cluster

A retrieval service that consists of a group of query nodes and data nodes.

Processor

A role in offline index building that parses the raw data of users.

Builder

A role in offline index building that builds raw data into indexes.

Merger

A role in offline index building that merges and organizes indexes.

Full data

The process of building data in MaxCompute into indexes is called a full build. The resulting index is called a full index, and the version of the index is called a full version.

Incremental data

When data is updated in real time, the offline index building process generates indexes and switches them online.

Real-time data

Data pushed by using API operations takes effect in real time. This process is called real-time processing, and real-time indexes are generated in the memory of data nodes.

Inverted index

A linked list that consists of mappings from terms to documents. The query clause uses this method, which provides high query efficiency. Example: term1->doc1,doc2,doc3; term2->doc1,doc2.

Forward index

A linked list that consists of mappings from documents to fields. The filter clause uses this method, whose performance is slightly lower than that of inverted indexes. Example: doc1->id,type,create_time….

Summary index

Stores the information to be displayed together. The information is retrieved by primary key or docid and is finally returned to the search results page for display.

Tokenization

Splits pushed documents into terms. Fields of the TEXT type are split by retrieval unit. For example, the Chinese phrase \u6d59\u6c5f\u5927\u5b66 (Zhejiang University) is split into two terms for the TEXT type: \u6d59\u6c5f (Zhejiang) and \u5927\u5b66 (University). The characters are shown as Unicode escape sequences; decode them to obtain the actual Chinese characters before you enter them.

term

The terms produced by tokenization are called terms.

Change type

Repeated triggering allowed

Description

Service discovery

Yes

This can be understood as mounting the IP addresses of the engine to a domain name so that the engine can be called. For the same cluster, all earlier changes are terminated before the latest change is executed.

ha3_biz_apend

No

Adds a biz. This change is automatically triggered by the system, and only one such change exists. The change may run for several days and ends automatically only after index tables are correctly added to the instance and indexes are successfully built.

update_biz_depend_index_fsm

No

Updates the indexes on which a biz depends. This change is automatically triggered by the system, and only one such change exists. The change may run for several days and ends automatically only after indexes are successfully built for the index tables.

Add online deployment

Yes

For the same cluster, all earlier changes are terminated before the latest change is executed.

multi_biz_activate

No

This can be understood as instance initialization.

Each instance has only one such change. The change may run for several days and ends automatically only after index tables are correctly added to the instance and indexes are successfully built.

Add index

Yes

For the same index, all earlier changes are terminated before the latest change is executed.

Automatically triggered full build

Yes

Automatically executed after a new data partition is detected. The latest change coexists with historical changes.

Manually triggered full build

Yes

The latest change coexists with historical changes.

Push configuration

Yes

All earlier changes are terminated before the latest change is executed.

Online resources

Yes

For the same zone, all earlier changes are terminated before the latest change is executed.

Index rollback

Yes

The latest change coexists with historical changes.