Spring AI集成Milvus向量存储时抛出'distance'键不存在异常

问题场景 #

基于Spring AI 1.0.0.M1、Ollama、Docker运行的Milvus 2.3.0构建RAG应用，文档插入正常，但执行以下检索语句时抛出distance键未找到异常：

List documents = vectorStore.similaritySearch(SearchRequest.defaults().query(question.question()).withTopK(5));

异常栈信息 #

io.milvus.exception.ParamException: The key name is not found
at io.milvus.response.QueryResultsWrapper$RowRecord.get(QueryResultsWrapper.java:158) ~[milvus-sdk-java-2.4.2.jar!/:na]
at org.springframework.ai.vectorstore.MilvusVectorStore.getResultSimilarity(MilvusVectorStore.java:373) ~[spring-ai-milvus-store-1.0.0-M1.jar!/:na]
at org.springframework.ai.vectorstore.MilvusVectorStore.lambda$similaritySearch$1(MilvusVectorStore.java:360) ~[spring-ai-milvus-store-1.0.0-M1.jar!/:na]
at java.base/java.util.stream.ReferencePipeline$2$1.accept(ReferencePipeline.java:178) ~[na:na]
at java.base/java.util.ArrayList$ArrayListSpliterator.forEachRemaining(ArrayList.java:1708) ~[na:na]
at java.base/java.util.stream.AbstractPipeline.copyInto(AbstractPipeline.java:509) ~[na:na]
at java.base/java.util.stream.AbstractPipeline.wrapAndCopyInto(AbstractPipeline.java:499) ~[na:na]
at java.base/java.util.stream.AbstractPipeline.evaluate(AbstractPipeline.java:575) ~[na:na]
at java.base/java.util.stream.AbstractPipeline.evaluateToArrayNode(AbstractPipeline.java:260) ~[na:na]
at java.base/java.util.stream.ReferencePipeline.toArray(ReferencePipeline.java:616) ~[na:na]
at java.base/java.util.stream.ReferencePipeline.toArray(ReferencePipeline.java:622) ~[na:na]
at java.base/java.util.stream.ReferencePipeline.toList(ReferencePipeline.java:627) ~[na:na]
at org.springframework.ai.vectorstore.MilvusVectorStore.similaritySearch(MilvusVectorStore.java:369) ~[spring-ai-milvus-store-1.0.0-M1.jar!/:na]
at com.rbc.newton.ai.rag.services.OllamaAIServiceImpl.getAnswer(OllamaAIServiceImpl.java:59) ~[!/:0.0.1-SNAPSHOT]
at com.rbc.newton.ai.rag.controller.QueryController.askQuestion(QueryController.java:20) ~[!/:0.0.1-SNAPSHOT]
at java.base/jdk.internal.reflect.DirectMethodHandleAccessor.invoke(DirectMethodHandleAccessor.java:103) ~[na:na]
at java.base/java.lang.reflect.Method.invoke(Method.java:580) ~[na:na]
at org.springframework.web.method.support.InvocableHandlerMethod.doInvoke(InvocableHandlerMethod.java:255) ~[spring-web-6.1.11.jar!/:6.1.11]
at org.springframework.web.method.support.InvocableHandlerMethod.invokeForRequest(InvocableHandlerMethod.java:188) ~[spring-web-6.1.11.jar!/:6.1.11]
at org.springframework.web.servlet.mvc.method.annotation.ServletInvocableHandlerMethod.invokeAndHandle(ServletInvocableHandlerMethod.java:118) ~[spring-webmvc-6.1.11.jar!/:6.1.11]
at org.springframework.web.servlet.mvc.method.annotation.RequestMappingHandlerAdapter.invokeHandlerMethod(RequestMappingHandlerAdapter.java:926) ~[spring-webmvc-6.1.11.jar!/:6.1.11]
at org.springframework.web.servlet.mvc.method.annotation.RequestMappingHandlerAdapter.handleInternal(RequestMappingHandlerAdapter.java:831) ~[spring-webmvc-6.1.11.jar!/:6.1.11]
at org.springframework.web.servlet.mvc.method.AbstractHandlerMethodAdapter.handle(AbstractHandlerMethodAdapter.java:87) ~[spring-webmvc-6.1.11.jar!/:6.1.11]
at org.springframework.web.servlet.DispatcherServlet.doDispatch(DispatcherServlet.java:1089) ~[spring-webmvc-6.1.11.jar!/:6.1.11]
at org.springframework.web.servlet.DispatcherServlet.doService(DispatcherServlet.java:979) ~[spring-webmvc-6.1.11.jar!/:6.1.11]
at org.springframework.web.servlet.FrameworkServlet.processRequest(FrameworkServlet.java:1014) ~[spring-webmvc-6.1.11.jar!/:6.1.11]
at org.springframework.web.servlet.FrameworkServlet.doPost(FrameworkServlet.java:914) ~[spring-webmvc-6.1.11.jar!/:6.1.11]
at jakarta.servlet.http.HttpServlet.service(HttpServlet.java:590) ~[tomcat-embed-core-10.1.26.jar!/:na]
at org.springframework.web.servlet.FrameworkServlet.service(FrameworkServlet.java:885) ~[spring-webmvc-6.1.11.jar!/:6.1.11]
at jakarta.servlet.http.HttpServlet.service(HttpServlet.java:658) ~[tomcat-embed-core-10.1.26.jar!/:na]

用户创建Milvus Collection的代码 #

ConnectConfig connectConfig = ConnectConfig.builder()
    .uri(CLUSTER_ENDPOINT)
    .build();
MilvusClientV2 client = new MilvusClientV2(connectConfig);

CreateCollectionReq.CollectionSchema schema = client.createSchema();

schema.addField(AddFieldReq.builder()
        .fieldName("embedding")
        .dataType(DataType.FloatVector)
        .dimension(1024)
        .description("vector embedding")
        .build());

schema.addField(AddFieldReq.builder()
        .fieldName("id")
        .dataType(DataType.Int64)
        .isPrimaryKey(true)
        .autoID(true)
        .build());


CreateCollectionReq quickSetupReq = CreateCollectionReq.builder()
        .collectionName(collection_name)
        .collectionSchema(schema)
        .dimension(1024)
        .vectorFieldName("embedding")
        .metricType("COSINE")
        .enableDynamicField(true)
        .build();

client.createCollection(quickSetupReq);

IndexParam indexParamForVectorField = IndexParam.builder()
        .fieldName("embedding")
        .indexName("vector_index")
        .indexType(IndexParam.IndexType.IVF_FLAT)
        .metricType(IndexParam.MetricType.COSINE)
        .extraParams(Map.of("nlist", 1, "nprobe", 1))
        .build();

List<IndexParam> indexParams = new ArrayList<>();
indexParams.add(indexParamForVectorField);

CreateIndexReq createIndexReq = CreateIndexReq.builder()
        .collectionName(collection_name)
        .indexParams(indexParams)
        .build();

client.createIndex(createIndexReq);

问题原因与解决方法 #

原因 #

Spring AI 1.0.0.M1的MilvusVectorStore实现默认期望检索结果包含distance字段，但Milvus 2.3.0未显式指定返回该字段时，结果中不会生成该键；同时使用的milvus-sdk-java-2.4.2与Milvus 2.3.0版本不匹配，存在协议兼容性问题。

解决步骤 #

• 版本对齐：将milvus-sdk-java降级到2.3.x版本，保持客户端与服务端版本一致，避免协议差异导致的字段返回异常。
• 显式配置返回字段：在Spring AI配置中指定返回distance字段，例如在application.properties中添加：

  spring.ai.vectorstore.milvus.return-fields=id,embedding,distance
  

• 自定义搜索请求：若上述配置无效，可直接使用MilvusClientV2构建搜索请求，显式指定返回distance字段，再手动转换为Spring AI的Document对象：

  SearchReq searchReq = SearchReq.builder()
          .collectionName(collection_name)
          .data(List.of(questionEmbedding))
          .annsField("embedding")
          .topK(5)
          .metricType(MetricType.COSINE)
          .outputFields(List.of("id", "distance"))
          .build();
  SearchResults results = client.search(searchReq);
  // 手动转换结果为List<Document>
  

内容的提问来源于stack exchange，提问作者user26841525