MLE-24374 Now using server-specific default namespace for embeddings

Author: rjrudin

marklogic-spark-connector/src/main/java/com/marklogic/spark/Options.java

@@ -453,7 +453,9 @@ public abstract class Options {
     public static final String WRITE_EMBEDDER_EMBEDDING_NAME = WRITE_EMBEDDER_PREFIX + "embedding.name";
 
     /**
-     * Allows for an optional namespace to be assigned to the embedding element in an XML chunk.
+     * Allows for an optional namespace to be assigned to the embedding element in an XML chunk. As of 3.0.0, if not
+     * set, then embedding elements in XML documents will be in the MarkLogic-specific vector namespace of
+     * "http://marklogic.com/vector".
      *
      * @since 2.5.0
      */

marklogic-spark-connector/src/main/java/com/marklogic/spark/Util.java

@@ -26,6 +26,13 @@ public interface Util {
 
     String DEFAULT_XML_NAMESPACE = "http://marklogic.com/appservices/model";
 
+    /**
+     * Aligns with the default vector namespace in the server.
+     *
+     * @since 3.0.0
+     */
+    String DEFAULT_VECTOR_NAMESPACE = "http://marklogic.com/vector";
+
     static boolean hasOption(Map<String, String> properties, String... options) {
         return Stream.of(options)
             .anyMatch(option -> properties.get(option) != null && !properties.get(option).trim().isEmpty());

marklogic-spark-connector/src/main/java/com/marklogic/spark/core/embedding/XmlChunkConfig.java

@@ -31,7 +31,7 @@ public XmlChunkConfig(String embeddingNamespace) {
     public XmlChunkConfig(String textExpression, String embeddingName, String embeddingNamespace, NamespaceContext namespaceContext, boolean base64EncodeVectors) {
         this.textExpression = textExpression != null ? textExpression : DEFAULT_TEXT_EXPRESSION;
         this.embeddingName = embeddingName != null ? embeddingName : "embedding";
-        this.embeddingNamespace = embeddingNamespace != null ? embeddingNamespace : Util.DEFAULT_XML_NAMESPACE;
+        this.embeddingNamespace = embeddingNamespace != null ? embeddingNamespace : Util.DEFAULT_VECTOR_NAMESPACE;
         this.namespaceContext = namespaceContext;
         this.base64EncodeVectors = base64EncodeVectors;
     }

marklogic-spark-connector/src/main/java/com/marklogic/spark/core/splitter/ChunkConfig.java

@@ -43,18 +43,13 @@ public static class Builder {
         private String documentType;
         private String rootName;
         private String xmlNamespace = Util.DEFAULT_XML_NAMESPACE;
-        private String embeddingXmlNamespace;
+        private String embeddingXmlNamespace = Util.DEFAULT_VECTOR_NAMESPACE;
         private String uriPrefix;
         private String uriSuffix;
         private boolean base64EncodeVectors = false;
 
         public ChunkConfig build() {
-            String tempNamespace = embeddingXmlNamespace;
-            if (tempNamespace == null) {
-                // If no embedding XML namespace is specified, default to the chunk namespace is defined.
-                tempNamespace = xmlNamespace != null ? xmlNamespace : Util.DEFAULT_XML_NAMESPACE;
-            }
-            return new ChunkConfig(metadata, maxChunks, documentType, rootName, xmlNamespace, tempNamespace, uriPrefix, uriSuffix, base64EncodeVectors);
+            return new ChunkConfig(metadata, maxChunks, documentType, rootName, xmlNamespace, embeddingXmlNamespace, uriPrefix, uriSuffix, base64EncodeVectors);
         }
 
         public Builder withMetadata(DocumentMetadataHandle metadata) {

marklogic-spark-connector/src/main/java/com/marklogic/spark/dom/NamespaceContextFactory.java

@@ -15,6 +15,7 @@ public interface NamespaceContextFactory {
     static NamespaceContext makeDefaultNamespaceContext() {
         Map<String, String> prefixesToNamespaces = new HashMap<>();
         prefixesToNamespaces.put("model", Util.DEFAULT_XML_NAMESPACE);
+        prefixesToNamespaces.put("vec", Util.DEFAULT_VECTOR_NAMESPACE);
         return new XPathNamespaceContext(prefixesToNamespaces);
     }
 

marklogic-spark-connector/src/test/java/com/marklogic/spark/AbstractIntegrationTest.java

@@ -185,6 +185,7 @@ protected XmlNode readXmlDocument(String uri) {
         // Registers frequently used namespaces in tests.
         return readXmlDocument(uri,
             Namespace.getNamespace("model", "http://marklogic.com/appservices/model"),
+            Namespace.getNamespace("vec", "http://marklogic.com/vector"),
             Namespace.getNamespace("ex", "org:example"),
             Namespace.getNamespace("acme", "org:acme"),
             Namespace.getNamespace("xml", "http://www.w3.org/XML/1998/namespace"),

marklogic-spark-connector/src/test/java/com/marklogic/spark/writer/embedding/AddEmbeddingsToXmlTest.java

@@ -88,8 +88,8 @@ void sidecarWithNamespace() {
         doc.assertElementCount("/ex:sidecar/ex:chunks/ex:chunk", 4);
         for (XmlNode chunk : doc.getXmlNodes("/ex:sidecar/ex:chunks/ex:chunk")) {
             chunk.assertElementExists("/ex:chunk/ex:text");
-            chunk.assertElementExists("When a namespace is specified for the sidecar XML document, that should " +
-                "override the default namespace for the embedding element.", "/ex:chunk/ex:embedding");
+            chunk.assertElementExists("As of 3.0.0, the embedding namespace should be the MarkLogic-specific one, " +
+                "unless the user explicitly overrides it.", "/ex:chunk/vec:embedding");
         }
     }
 
@@ -164,8 +164,8 @@ void sidecarWithNoNamespace() {
         XmlNode doc = readXmlDocument("/split-test.xml-chunks-1.xml");
         doc.assertElementValue("/root/source-uri", "/split-test.xml");
         doc.assertElementExists("/root/chunks/chunk[1]/text");
-        doc.assertElementExists("Since a namespace is specified for the document - no namespace - it should be " +
-            "applied to the embedding element too.", "/root/chunks/chunk[1]/embedding");
+        doc.assertElementExists("As of 3.0.0, the embedding element should default to the MarkLogic-specific " +
+            "namespace unless the user explicitly sets it.", "/root/chunks/chunk[1]/vec:embedding");
     }
 
     @Test
@@ -183,7 +183,7 @@ void customChunks() {
 
         XmlNode doc = readXmlDocument("/split-test.xml");
         doc.assertElementCount("Each of the 2 custom chunks should have an 'embedding' element.",
-            "/envelope/my-chunks/my-chunk[my-text and model:embedding]", 2);
+            "/envelope/my-chunks/my-chunk[my-text and vec:embedding]", 2);
     }
 
     @Test
@@ -201,8 +201,10 @@ void namespacedCustomChunks() {
             .save();
 
         XmlNode doc = readXmlDocument("/split-test.xml");
-        doc.assertElementCount("Each of the 2 custom chunks should have an 'embedding' element.",
-            "/ex:envelope/ex:my-chunks/ex:my-chunk[ex:my-text and model:embedding]", 2);
+        doc.assertElementCount("As of 3.0.0, the embedding element should default to the MarkLogic-specific " +
+                "namespace unless the user explicitly sets it. In this test, the chunks are in a custom namespace, " +
+                "but the embedding is not.",
+            "/ex:envelope/ex:my-chunks/ex:my-chunk[ex:my-text and vec:embedding]", 2);
     }
 
     @Test
@@ -299,8 +301,9 @@ private void verifyEachChunkOnDocumentHasAnEmbedding(String uri) {
         doc.getXmlNodes("/node()/model:chunks/model:chunk").forEach(chunk -> {
             chunk.assertElementExists("/model:chunk/model:text");
             chunk.assertElementExists(
-                "As of the 2.7.0 release, the embedding should have the zxx lang to disable stemming by MarkLogic.",
-                "/model:chunk/model:embedding[@xml:lang='zxx']"
+                "As of the 2.7.0 release, the embedding should have the zxx lang to disable stemming by MarkLogic. " +
+                    "And as of the 3.0.0 release, the embedding should default to the MarkLogic-specific vector namespace.",
+                "/model:chunk/vec:embedding[@xml:lang='zxx']"
             );
         });
     }
@@ -391,12 +394,13 @@ private void verifyDocumentHasTwoChunksWithEncodedVectors(XmlNode doc) {
         doc.assertElementCount("/root/model:chunks/model:chunk", 2);
 
         for (XmlNode chunk : doc.getXmlNodes("/root/model:chunks/model:chunk")) {
-            String embeddingValue = chunk.getElementValue("/model:chunk/model:embedding");
+            System.out.println(chunk.getPrettyXml());
+            String embeddingValue = chunk.getElementValue("/model:chunk/vec:embedding");
             assertEquals("AAAAAAMAAADD9UhAH4XLP5qZKUA=", embeddingValue,
                 "Base64 encoded vector should match expected encoding for test vector [3.14, 1.59, 2.65]");
 
             chunk.assertElementExists("xml:lang attribute should be 'zxx' to disable stemming",
-                "/model:chunk/model:embedding[@xml:lang='zxx']");
+                "/model:chunk/vec:embedding[@xml:lang='zxx']");
         }
     }
 }

test-app/src/main/ml-schemas-12/tde/xml-vector-chunks.json

@@ -23,7 +23,7 @@
           {
             "name": "embedding",
             "scalarType": "vector",
-            "val": "vec:vector(model:embedding)",
+            "val": "vec:vector(vec:embedding)",
             "dimension": "384",
             "invalidValues": "reject",
             "nullable": true