聊聊Spring AI Alibaba的ObsidianDocumentReader
序
本文主要研究一下Spring AI Alibaba的ObsidianDocumentReader
ObsidianDocumentReader
community/document-readers/spring-ai-alibaba-starter-document-reader-obsidian/src/main/java/com/alibaba/cloud/ai/reader/obsidian/ObsidianDocumentReader.java
public class ObsidianDocumentReader implements DocumentReader {private final Path vaultPath;private final MarkdownDocumentParser parser;/*** Constructor for reading all files in vault* @param vaultPath Path to Obsidian vault*/public ObsidianDocumentReader(Path vaultPath) {this.vaultPath = vaultPath;this.parser = new MarkdownDocumentParser();}@Overridepublic List<Document> get() {List<Document> allDocuments = new ArrayList<>();// Find all markdown files in vaultList<ObsidianResource> resources = ObsidianResource.findAllMarkdownFiles(vaultPath);// Parse each filefor (ObsidianResource resource : resources) {try {List<Document> documents = parser.parse(resource.getInputStream());String source = resource.getSource();// Add metadata to each documentfor (Document doc : documents) {doc.getMetadata().put(ObsidianResource.SOURCE, source);}allDocuments.addAll(documents);}catch (IOException e) {throw new RuntimeException("Failed to read Obsidian file: " + resource.getFilePath(), e);}}return allDocuments;}public static Builder builder() {return new Builder();}public static class Builder {private Path vaultPath;public Builder vaultPath(Path vaultPath) {this.vaultPath = vaultPath;return this;}public ObsidianDocumentReader build() {return new ObsidianDocumentReader(vaultPath);}}}
ObsidianDocumentReader的get方法通过ObsidianResource.findAllMarkdownFiles(vaultPath)来读取ObsidianResource，之后遍历resources使用MarkdownDocumentParser进行解析
ObsidianResource
community/document-readers/spring-ai-alibaba-starter-document-reader-obsidian/src/main/java/com/alibaba/cloud/ai/reader/obsidian/ObsidianResource.java
public class ObsidianResource implements Resource {public static final String SOURCE = "source";public static final String MARKDOWN_EXTENSION = ".md";private final Path vaultPath;private final Path filePath;private final InputStream inputStream;/*** Constructor for single file* @param vaultPath Path to Obsidian vault* @param filePath Path to markdown file*/public ObsidianResource(Path vaultPath, Path filePath) {Assert.notNull(vaultPath, "VaultPath must not be null");Assert.notNull(filePath, "FilePath must not be null");Assert.isTrue(Files.exists(vaultPath), "Vault directory does not exist: " + vaultPath);Assert.isTrue(Files.exists(filePath), "File does not exist: " + filePath);Assert.isTrue(filePath.toString().endsWith(MARKDOWN_EXTENSION), "File must be a markdown file: " + filePath);this.vaultPath = vaultPath;this.filePath = filePath;try {this.inputStream = new FileInputStream(filePath.toFile());}catch (IOException e) {throw new RuntimeException("Failed to create input stream for file: " + filePath, e);}}/*** Find all markdown files in the vault Recursively searches through all* subdirectories Only includes .md files and ignores hidden files/directories* @param vaultPath Root path of the Obsidian vault* @return List of ObsidianResource for each markdown file*/public static List<ObsidianResource> findAllMarkdownFiles(Path vaultPath) {Assert.notNull(vaultPath, "VaultPath must not be null");Assert.isTrue(Files.exists(vaultPath), "Vault directory does not exist: " + vaultPath);Assert.isTrue(Files.isDirectory(vaultPath), "VaultPath must be a directory: " + vaultPath);List<ObsidianResource> resources = new ArrayList<>();try (Stream<Path> paths = Files.walk(vaultPath)) {paths// Only include .md files.filter(path -> path.toString().endsWith(MARKDOWN_EXTENSION))// Ignore hidden files and files in hidden directories.filter(path -> {Path relativePath = vaultPath.relativize(path);String[] pathParts = relativePath.toString().split("/");for (String part : pathParts) {if (part.startsWith(".")) {return false;}}return true;})// Only include regular files (not directories).filter(Files::isRegularFile).forEach(path -> resources.add(new ObsidianResource(vaultPath, path)));}catch (IOException e) {throw new RuntimeException("Failed to walk vault directory: " + vaultPath, e);}return resources;}//......
}
ObsidianResource构造器要求输入vaultPath和filePath，其findAllMarkdownFiles方法会遍历vaultPath目录，找出
.md结尾的文件
示例
community/document-readers/spring-ai-alibaba-starter-document-reader-obsidian/src/test/java/com/alibaba/cloud/ai/reader/obsidian/ObsidianDocumentReaderIT.java
@EnabledIfEnvironmentVariable(named = "OBSIDIAN_VAULT_PATH", matches = ".+")
class ObsidianDocumentReaderIT {private static final String VAULT_PATH = System.getenv("OBSIDIAN_VAULT_PATH");// Static initializer to log a message if environment variable is not setstatic {if (VAULT_PATH == null || VAULT_PATH.isEmpty()) {System.out.println("Skipping Obsidian tests because OBSIDIAN_VAULT_PATH environment variable is not set.");}}ObsidianDocumentReader reader;@BeforeEachvoid setUp() {// Only initialize if VAULT_PATH is setif (VAULT_PATH != null && !VAULT_PATH.isEmpty()) {reader = ObsidianDocumentReader.builder().vaultPath(Path.of(VAULT_PATH)).build();}}@Testvoid should_read_markdown_files() {// Skip test if reader is nullAssumptions.assumeTrue(reader != null, "Skipping test because ObsidianDocumentReader could not be initialized");// whenList<Document> documents = reader.get();// thenassertThat(documents).isNotEmpty();// Verify document content and metadatafor (Document doc : documents) {// Verify source metadataassertThat(doc.getMetadata()).containsKey(ObsidianResource.SOURCE);String source = doc.getMetadata().get(ObsidianResource.SOURCE).toString();assertThat(source).isNotEmpty().endsWith(ObsidianResource.MARKDOWN_EXTENSION);// Verify contentassertThat(doc.getText()).isNotEmpty();// Print for debuggingSystem.out.println("Document source: " + source);if (doc.getMetadata().containsKey("category")) {System.out.println("Document category: " + doc.getMetadata().get("category"));}System.out.println("Document content: " + doc.getText());System.out.println("---");}}}
小结
spring-ai-alibaba-starter-document-reader-obsidian提供了ObsidianDocumentReader用于读取指定仓库(vaultPath)下的所有markdown文件，之后使用MarkdownDocumentParser去解析为List<Document>。
doc
- java2ai