序

本文主要研究一下Spring AI的PgVectorStore

示例

pom.xml

		<dependency>
			<groupId>org.springframework.ai</groupId>
			<artifactId>spring-ai-starter-vector-store-pgvector</artifactId>
		</dependency>

pgvector

docker run -it --rm --name postgres -p 5432:5432 -e POSTGRES_USER=postgres -e POSTGRES_PASSWORD=postgres pgvector/pgvector:pg16

配置

spring:
  datasource:
    name: pgvector
    driverClassName: org.postgresql.Driver
    url: jdbc:postgresql://localhost:5432/postgres?currentSchema=public&connectTimeout=60&socketTimeout=60
    username: postgres
    password: postgres
  ai:
    vectorstore:
      type: pgvector
      pgvector:
        initialize-schema: true
        index-type: HNSW
        distance-type: COSINE_DISTANCE
        dimensions: 1024
        max-document-batch-size: 10000
        schema-name: public
        table-name: vector_store

设置initialize-schema为true，默认会执行如下初始化脚本：

CREATE EXTENSION IF NOT EXISTS vector;
CREATE EXTENSION IF NOT EXISTS hstore;
CREATE EXTENSION IF NOT EXISTS "uuid-ossp";

CREATE TABLE IF NOT EXISTS vector_store (
	id uuid DEFAULT uuid_generate_v4() PRIMARY KEY,
	content text,
	metadata json,
	embedding vector(1536) // 1536 is the default embedding dimension
);

CREATE INDEX ON vector_store USING HNSW (embedding vector_cosine_ops);

脚本源码：

org/springframework/ai/vectorstore/pgvector/PgVectorStore.java

	public void afterPropertiesSet() {

		logger.info("Initializing PGVectorStore schema for table: {} in schema: {}", this.getVectorTableName(),
				this.getSchemaName());

		logger.info("vectorTableValidationsEnabled {}", this.schemaValidation);

		if (this.schemaValidation) {
			this.schemaValidator.validateTableSchema(this.getSchemaName(), this.getVectorTableName());
		}

		if (!this.initializeSchema) {
			logger.debug("Skipping the schema initialization for the table: {}", this.getFullyQualifiedTableName());
			return;
		}

		// Enable the PGVector, JSONB and UUID support.
		this.jdbcTemplate.execute("CREATE EXTENSION IF NOT EXISTS vector");
		this.jdbcTemplate.execute("CREATE EXTENSION IF NOT EXISTS hstore");

		if (this.idType == PgIdType.UUID) {
			this.jdbcTemplate.execute("CREATE EXTENSION IF NOT EXISTS \"uuid-ossp\"");
		}

		this.jdbcTemplate.execute(String.format("CREATE SCHEMA IF NOT EXISTS %s", this.getSchemaName()));

		// Remove existing VectorStoreTable
		if (this.removeExistingVectorStoreTable) {
			this.jdbcTemplate.execute(String.format("DROP TABLE IF EXISTS %s", this.getFullyQualifiedTableName()));
		}

		this.jdbcTemplate.execute(String.format("""
				CREATE TABLE IF NOT EXISTS %s (
					id %s PRIMARY KEY,
					content text,
					metadata json,
					embedding vector(%d)
				)
				""", this.getFullyQualifiedTableName(), this.getColumnTypeName(), this.embeddingDimensions()));

		if (this.createIndexMethod != PgIndexType.NONE) {
			this.jdbcTemplate.execute(String.format("""
					CREATE INDEX IF NOT EXISTS %s ON %s USING %s (embedding %s)
					""", this.getVectorIndexName(), this.getFullyQualifiedTableName(), this.createIndexMethod,
					this.getDistanceType().index));
		}
	}

代码

    @Test
    public void testAddAndSearch() {
        List<Document> documents = List.of(
                new Document("Spring AI rocks!! Spring AI rocks!! Spring AI rocks!! Spring AI rocks!! Spring AI rocks!!", Map.of("meta1", "meta1")),
                new Document("The World is Big and Salvation Lurks Around the Corner"),
                new Document("You walk forward facing the past and you turn back toward the future.", Map.of("meta2", "meta2")));

        // Add the documents to Milvus Vector Store
        pgVectorStore.add(documents);

        // Retrieve documents similar to a query
        List<Document> results = this.pgVectorStore.similaritySearch(SearchRequest.builder().query("Spring").topK(5).build());
        log.info("results:{}", JSON.toJSONString(results));
    }

输出如下：

results:[{"contentFormatter":{"excludedEmbedMetadataKeys":[],"excludedInferenceMetadataKeys":[],"metadataSeparator":"\n","metadataTemplate":"{key}: {value}","textTemplate":"{metadata_string}\n\n{content}"},"formattedContent":"distance: 0.43509135\nmeta1: meta1\n\nSpring AI rocks!! Spring AI rocks!! Spring AI rocks!! Spring AI rocks!! Spring AI rocks!!","id":"9dbce9af-0451-4bdb-8f03-1f8b8c4d696f","metadata":{"distance":0.43509135,"meta1":"meta1"},"score":0.5649086534976959,"text":"Spring AI rocks!! Spring AI rocks!! Spring AI rocks!! Spring AI rocks!! Spring AI rocks!!"},{"contentFormatter":{"$ref":"$[0].contentFormatter"},"formattedContent":"distance: 0.57093126\n\nThe World is Big and Salvation Lurks Around the Corner","id":"92a45683-11fc-48b7-8676-dcca3b518dd4","metadata":{"distance":0.57093126},"score":0.42906874418258667,"text":"The World is Big and Salvation Lurks Around the Corner"},{"contentFormatter":{"$ref":"$[0].contentFormatter"},"formattedContent":"distance: 0.5936024\nmeta2: meta2\n\nYou walk forward facing the past and you turn back toward the future.","id":"298f6565-bcc7-4cbc-8552-4c0e2d021dbf","metadata":{"distance":0.5936024,"meta2":"meta2"},"score":0.40639758110046387,"text":"You walk forward facing the past and you turn back toward the future."}]

源码

PgVectorStoreAutoConfiguration

org/springframework/ai/vectorstore/pgvector/autoconfigure/PgVectorStoreAutoConfiguration.java

@AutoConfiguration(after = JdbcTemplateAutoConfiguration.class)
@ConditionalOnClass({ PgVectorStore.class, DataSource.class, JdbcTemplate.class })
@EnableConfigurationProperties(PgVectorStoreProperties.class)
@ConditionalOnProperty(name = SpringAIVectorStoreTypes.TYPE, havingValue = SpringAIVectorStoreTypes.PGVECTOR,
		matchIfMissing = true)
public class PgVectorStoreAutoConfiguration {

	@Bean
	@ConditionalOnMissingBean(BatchingStrategy.class)
	BatchingStrategy pgVectorStoreBatchingStrategy() {
		return new TokenCountBatchingStrategy();
	}

	@Bean
	@ConditionalOnMissingBean
	public PgVectorStore vectorStore(JdbcTemplate jdbcTemplate, EmbeddingModel embeddingModel,
			PgVectorStoreProperties properties, ObjectProvider<ObservationRegistry> observationRegistry,
			ObjectProvider<VectorStoreObservationConvention> customObservationConvention,
			BatchingStrategy batchingStrategy) {

		var initializeSchema = properties.isInitializeSchema();

		return PgVectorStore.builder(jdbcTemplate, embeddingModel)
			.schemaName(properties.getSchemaName())
			.idType(properties.getIdType())
			.vectorTableName(properties.getTableName())
			.vectorTableValidationsEnabled(properties.isSchemaValidation())
			.dimensions(properties.getDimensions())
			.distanceType(properties.getDistanceType())
			.removeExistingVectorStoreTable(properties.isRemoveExistingVectorStoreTable())
			.indexType(properties.getIndexType())
			.initializeSchema(initializeSchema)
			.observationRegistry(observationRegistry.getIfUnique(() -> ObservationRegistry.NOOP))
			.customObservationConvention(customObservationConvention.getIfAvailable(() -> null))
			.batchingStrategy(batchingStrategy)
			.maxDocumentBatchSize(properties.getMaxDocumentBatchSize())
			.build();
	}

}

PgVectorStoreAutoConfiguration在
spring.ai.vectorstore.type为pgvector时会自动装配PgVectorStore，它依赖PgVectorStoreProperties及
JdbcTemplateAutoConfiguration

PgVectorStoreProperties

org/springframework/ai/vectorstore/pgvector/autoconfigure/PgVectorStoreProperties.java

@ConfigurationProperties(PgVectorStoreProperties.CONFIG_PREFIX)
public class PgVectorStoreProperties extends CommonVectorStoreProperties {

	public static final String CONFIG_PREFIX = "spring.ai.vectorstore.pgvector";

	private int dimensions = PgVectorStore.INVALID_EMBEDDING_DIMENSION;

	private PgIndexType indexType = PgIndexType.HNSW;

	private PgDistanceType distanceType = PgDistanceType.COSINE_DISTANCE;

	private boolean removeExistingVectorStoreTable = false;

	// Dynamically generate table name in PgVectorStore to allow backward compatibility
	private String tableName = PgVectorStore.DEFAULT_TABLE_NAME;

	private String schemaName = PgVectorStore.DEFAULT_SCHEMA_NAME;

	private PgVectorStore.PgIdType idType = PgVectorStore.PgIdType.UUID;

	private boolean schemaValidation = PgVectorStore.DEFAULT_SCHEMA_VALIDATION;

	private int maxDocumentBatchSize = PgVectorStore.MAX_DOCUMENT_BATCH_SIZE;

	//......
}	

PgVectorStoreProperties继承了
CommonVectorStoreProperties的initializeSchema配置，它提供了
spring.ai.vectorstore.pgvector的配置，主要有dimensions、indexType、distanceType、
removeExistingVectorStoreTable、tableName、schemaName、idType、schemaValidation、maxDocumentBatchSize这几个属性

JdbcTemplateAutoConfiguration

org/springframework/boot/autoconfigure/jdbc/JdbcTemplateAutoConfiguration.java

@AutoConfiguration(after = DataSourceAutoConfiguration.class)
@ConditionalOnClass({ DataSource.class, JdbcTemplate.class })
@ConditionalOnSingleCandidate(DataSource.class)
@EnableConfigurationProperties(JdbcProperties.class)
@Import({ DatabaseInitializationDependencyConfigurer.class, JdbcTemplateConfiguration.class,
		NamedParameterJdbcTemplateConfiguration.class })
public class JdbcTemplateAutoConfiguration {

}

JdbcTemplateAutoConfiguration引入了
DatabaseInitializationDependencyConfigurer、JdbcTemplateConfiguration、
NamedParameterJdbcTemplateConfiguration

小结

Spring AI提供了
spring-ai-starter-vector-store-pgvector用于自动装配PgVectorStore。除了
spring.ai.vectorstore.pgvector的配置，还需要配置spring.datasource。

doc

• vectordbs/pgvector