Class MilvusDatabaseKind

java.lang.Object

dev.langchain4j.store.embedding.hibernate.milvus.MilvusDatabaseKind

All Implemented Interfaces:

DatabaseKind

public final class MilvusDatabaseKind
extends Object
implements DatabaseKind

The database kind for using Milvus.

See Also:

• HibernateEmbeddingStore.BaseBuilder.databaseKind(DatabaseKind)

Field Summary

Fields

Modifier and Type	Field	Description
static final DatabaseKind	INSTANCE

Fields inherited from interface DatabaseKind

COCKROACHDB, DB2, HANA, MARIADB, MSSQL, MYSQL, ORACLE, POSTGRESQL

Method Summary

Modifier and Type	Method	Description
String	createIndexDDL(DistanceFunction distanceFunction, String indexType, String table, String embeddingColumn, String indexOptions)
String	createJdbcUrl(String host, int port, String database)
Map<String,Object>	defaultQueryHints()	Returns the Milvus session consistency level, so that a search sees all embeddings that this application added or removed before it.
String	getSetupSql()
boolean	isJdbcUrl(String jdbcUrl)
String	tableIndexOptions(boolean createIndex, DistanceFunction distanceFunction, String indexType, String table, String embeddingColumn, String indexOptions)	Returns the options of the vector index of the collection.

Methods inherited from class Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Field Details

INSTANCE

public static final DatabaseKind INSTANCE

Method Details

createIndexDDL

public String createIndexDDL(DistanceFunction distanceFunction,
 String indexType,
 String table,
 String embeddingColumn,
 String indexOptions)

Specified by:

createIndexDDL in interface DatabaseKind

tableIndexOptions

public String tableIndexOptions(boolean createIndex,
 DistanceFunction distanceFunction,
 String indexType,
 String table,
 String embeddingColumn,
 String indexOptions)

Returns the options of the vector index of the collection.

A Milvus collection always has a vector index, and a search must use the metric of that index. So the metric of the distance function is always returned, also without createIndex(true). The index type and the index options are only added with createIndex(true), like for the relational databases.

Specified by:

tableIndexOptions in interface DatabaseKind

Parameters:

createIndex - whether the builder is configured to create a vector index, see HibernateEmbeddingStore.BaseBuilder.createIndex(Boolean)

distanceFunction - the distance function of the store

indexType - the index type set on the builder, or null

table - the name of the embedding table

embeddingColumn - the name of the embedding column

indexOptions - the index options set on the builder, or null

Returns:

the index options, or null to declare no vector index on the table

Throws:

IllegalArgumentException - if Milvus does not support the distance function

See Also:

• HibernateEmbeddingStore.BaseBuilder.indexOptions(String)

getSetupSql

public String getSetupSql()

Specified by:

getSetupSql in interface DatabaseKind

defaultQueryHints

public Map<String,Object> defaultQueryHints()

Returns the Milvus session consistency level, so that a search sees all embeddings that this application added or removed before it. Without a consistency level, Milvus uses the default level of the collection, and a search right after add(...) can miss the new embeddings.

Specified by:

defaultQueryHints in interface DatabaseKind

Returns:

the session consistency level as a HibernateHints.HINT_QUERY_DATABASE hint

See Also:

• HibernateEmbeddingStore.BaseBuilder.queryHints(Map)

createJdbcUrl

public String createJdbcUrl(String host,
 int port,
 String database)

Specified by:

createJdbcUrl in interface DatabaseKind

isJdbcUrl

public boolean isJdbcUrl(String jdbcUrl)

Specified by:

isJdbcUrl in interface DatabaseKind