导入JDBC元数据到Apache Atlas

前言

前期实现了导入MySQL元数据到Apache Atlas, 由于是初步版本，且功能参照Atlas Hive Hook，实现的不够完美

本期对功能进行改进，实现了导入多种关系型数据库元数据到Apache Atlas

数据库schema与catalog

按照SQL标准的解释，在SQL环境下Catalog和Schema都属于抽象概念，可以把它们理解为一个容器或者数据库对象命名空间中的一个层次，主要用来解决命名冲突问题 。从概念上说，一个数据库系统包含多个Catalog，每个Catalog又包含多个Schema，而每个Schema又包含多个数据库对象（表、视图、字段等），反过来讲一个数据库对象必然属于一个Schema，而该Schema又必然属于一个Catalog，这样我们就可以得到该数据库对象的完全限定名称，从而解决命名冲突的问题了；例如数据库对象表的完全限定名称就可以表示为：Catalog名称.Schema名称.表名称。这里还有一点需要注意的是，SQL标准并不要求每个数据库对象的完全限定名称是唯一的。

从实现的角度来看，各种数据库系统对Catalog和Schema的支持和实现方式千差万别，针对具体问题需要参考具体的产品说明书，比较简单而常用的实现方式是使用数据库名作为Catalog名，使用用户名作为Schema名，具体可参见下表：

表1 常用数据库

供应商	Catalog支持	Schema支持
Oracle	不支持	Oracle User ID
MySQL	不支持	数据库名
MS SQL Server	数据库名	对象属主名，2005版开始有变
DB2	指定数据库对象时，Catalog部分省略	Catalog属主名
Sybase	数据库名	数据库属主名
Informix	不支持	不需要
PointBase	不支持	数据库名

原文：https://www.cnblogs.com/ECNB/p/4611309.html

元数据模型层级抽象

不同的关系型数据库，其数据库模式有所区别，对应与下面的层级关系

• Datasource -> Catalog -> Schema -> Table -> Column
• Datasource -> Catalog -> Table -> Column
• Datasource -> Schema -> Table -> Column

元数据转换设计

提供元数据

借鉴Apache DolphinScheduler中获取Connection的方式，不多赘述。

public Connection getConnection(DbType dbType, ConnectionParam connectionParam) throws ExecutionException {
        BaseConnectionParam baseConnectionParam = (BaseConnectionParam) connectionParam;
        String datasourceUniqueId = DataSourceUtils.getDatasourceUniqueId(baseConnectionParam, dbType);
        logger.info("Get connection from datasource {}", datasourceUniqueId);

        DataSourceClient dataSourceClient = uniqueId2dataSourceClientCache.get(datasourceUniqueId, () -> {
            Map<String, DataSourceChannel> dataSourceChannelMap = dataSourcePluginManager.getDataSourceChannelMap();
            DataSourceChannel dataSourceChannel = dataSourceChannelMap.get(dbType.getDescp());
            if (null == dataSourceChannel) {
                throw new RuntimeException(String.format("datasource plugin '%s' is not found", dbType.getDescp()));
            }
            return dataSourceChannel.createDataSourceClient(baseConnectionParam, dbType);
        });
        return dataSourceClient.getConnection();
    }

转换元数据

0. 元数据模型

创建数据库的元数据模型

private AtlasEntityDef createJdbcDatabaseDef() {
   AtlasEntityDef typeDef = createClassTypeDef(DatabaseProperties.JDBC_TYPE_DATABASE,
           Collections.singleton(DatabaseProperties.ENTITY_TYPE_DATASET),
           createOptionalAttrDef(DatabaseProperties.ATTR_URL, "string"),
           createOptionalAttrDef(DatabaseProperties.ATTR_DRIVER_NAME, "string"),
           createOptionalAttrDef(DatabaseProperties.ATTR_PRODUCT_NAME, "string"),
           createOptionalAttrDef(DatabaseProperties.ATTR_PRODUCT_VERSION, "string")
   );

   typeDef.setServiceType(DatabaseProperties.ENTITY_SERVICE_TYPE);

   return typeDef;
}

创建数据库模式的元数据模型

private AtlasEntityDef createJdbcSchemaDef() {
    AtlasEntityDef typeDef = AtlasTypeUtil.createClassTypeDef(
            SchemaProperties.JDBC_TYPE_SCHEMA,
            Collections.singleton(SchemaProperties.ENTITY_TYPE_DATASET)
    );

    typeDef.setServiceType(SchemaProperties.ENTITY_SERVICE_TYPE);
    typeDef.setOptions(new HashMap<>() {{
        put("schemaElementsAttribute", "tables");
    }});

    return typeDef;
}

创建数据库表的元数据模型

private AtlasEntityDef createJdbcTableDef() {
    AtlasEntityDef typeDef = createClassTypeDef(
            TableProperties.JDBC_TYPE_TABLE,
            Collections.singleton(TableProperties.ENTITY_TYPE_DATASET),
            createOptionalAttrDef(TableProperties.ATTR_TABLE_TYPE, "string")
    );

    typeDef.setServiceType(BaseProperties.ENTITY_SERVICE_TYPE);
    typeDef.setOptions(new HashMap<>() {{
        put("schemaElementsAttribute", "columns");
    }});

    return typeDef;
}

创建数据库列的元数据模型

private AtlasEntityDef createJdbcColumnDef() {
    AtlasEntityDef typeDef = createClassTypeDef(
            ColumnProperties.JDBC_TYPE_COLUMN,
            Collections.singleton(ColumnProperties.ENTITY_TYPE_DATASET),
            createOptionalAttrDef(ColumnProperties.ATTR_COLUMN_TYPE, "string"),
            createOptionalAttrDef(ColumnProperties.ATTR_IS_PRIMARY_KEY, "string"),
            createOptionalAttrDef(ColumnProperties.ATTR_COLUMN_IS_NULLABLE, "string"),
            createOptionalAttrDef(ColumnProperties.ATTR_COLUMN_DEFAULT_VALUE, "string"),
            createOptionalAttrDef(ColumnProperties.ATTR_COLUMN_AUTO_INCREMENT, "string")
    );

    typeDef.setServiceType(BaseProperties.ENTITY_SERVICE_TYPE);
    HashMap<String, String> options = new HashMap<>() {{
        put("schemaAttributes", "[\"name\", \"isPrimaryKey\", \"columnType\", \"isNullable\" , \"isAutoIncrement\", \"description\"]");
    }};
    typeDef.setOptions(options);

    return typeDef;
}

创建实体之间的关系模型

private List<AtlasRelationshipDef> createAtlasRelationshipDef() {
    String version = "1.0";
    // 数据库和模式的关系
    AtlasRelationshipDef databaseSchemasDef = createRelationshipTypeDef(
            BaseProperties.RELATIONSHIP_DATABASE_SCHEMAS,
            BaseProperties.RELATIONSHIP_DATABASE_SCHEMAS,
            version, COMPOSITION, AtlasRelationshipDef.PropagateTags.NONE,
            createRelationshipEndDef(BaseProperties.JDBC_TYPE_DATABASE, "schemas", SET, true),
            createRelationshipEndDef(BaseProperties.JDBC_TYPE_SCHEMA, "database", SINGLE, false)
    );
    databaseSchemasDef.setServiceType(BaseProperties.ENTITY_SERVICE_TYPE);

    AtlasRelationshipDef databaseTablesDef = createRelationshipTypeDef(
            BaseProperties.RELATIONSHIP_DATABASE_TABLES,
            BaseProperties.RELATIONSHIP_DATABASE_TABLES,
            version, AGGREGATION, AtlasRelationshipDef.PropagateTags.NONE,
            createRelationshipEndDef(BaseProperties.JDBC_TYPE_DATABASE, "tables", SET, true),
            createRelationshipEndDef(BaseProperties.JDBC_TYPE_TABLE, "database", SINGLE, false)
    );
    databaseTablesDef.setServiceType(BaseProperties.ENTITY_SERVICE_TYPE);

    // 模式和数据表的关系
    // 注意 schema 已经被使用, 需要更换否则会冲突, 例如改为 Jschema(jdbc_schema)
    AtlasRelationshipDef schemaTablesDef = createRelationshipTypeDef(
            BaseProperties.RELATIONSHIP_SCHEMA_TABLES,
            BaseProperties.RELATIONSHIP_SCHEMA_TABLES,
            version, AGGREGATION, AtlasRelationshipDef.PropagateTags.NONE,
            createRelationshipEndDef(BaseProperties.JDBC_TYPE_SCHEMA, "tables", SET, true),
            createRelationshipEndDef(BaseProperties.JDBC_TYPE_TABLE, "Jschema", SINGLE, false)
    );
    schemaTablesDef.setServiceType(BaseProperties.ENTITY_SERVICE_TYPE);

    // 表和数据列的关系
    AtlasRelationshipDef tableColumnsDef = createRelationshipTypeDef(
            BaseProperties.RELATIONSHIP_TABLE_COLUMNS,
            BaseProperties.RELATIONSHIP_TABLE_COLUMNS,
            version, COMPOSITION, AtlasRelationshipDef.PropagateTags.NONE,
            createRelationshipEndDef(BaseProperties.JDBC_TYPE_TABLE, "columns", SET, true),
            createRelationshipEndDef(BaseProperties.JDBC_TYPE_COLUMN, "table", SINGLE, false)
    );
    tableColumnsDef.setServiceType(BaseProperties.ENTITY_SERVICE_TYPE);

    return Arrays.asList(databaseSchemasDef, databaseTablesDef, schemaTablesDef, tableColumnsDef);
}

1. 提取元数据

   不再赘述

2. 转换元数据

使用工厂模式，提供不同类型的元数据转换方式

public interface JdbcTransferFactory {

    JdbcTransfer getTransfer(DatabaseMetaData metaData, AtlasClientV2 client);

    boolean supportType(String type);

    String getName();
}

List ignorePatterns 用来过滤不想导入的数据库元数据，例如mysql的information_schema

public interface JdbcTransfer {

    void transfer();

    JdbcTransfer setIgnorePatterns(List<Pattern> ignorePatterns);
}

举例：JdbcMysqlTransfer 和 MysqlTransferFactory

@AutoService(JdbcTransferFactory.class)
public class MysqlTransferFactory implements JdbcTransferFactory {

    public static final String MYSQL = "mysql";

    @Override
    public JdbcTransfer getTransfer(DatabaseMetaData metaData, AtlasClientV2 client) {
        return new JdbcMysqlTransfer(metaData, client);
    }

    @Override
    public boolean supportType(String type) {
        return MYSQL.equalsIgnoreCase(type);
    }

    @Override
    public String getName() {

        return MYSQL;
    }
}

public class JdbcMysqlTransfer implements JdbcTransfer {

    private final Jdbc jdbc;
    private final AtlasService atlasService;
    private List<Pattern> ignorePatterns;

    public JdbcMysqlTransfer(DatabaseMetaData metaData, AtlasClientV2 client) {
        this.jdbc = new Jdbc(new JdbcMetadata(metaData));
        this.atlasService = new AtlasService(client);
        this.ignorePatterns = Collections.emptyList();
    }


    @Override
    public JdbcTransfer setIgnorePatterns(List<Pattern> ignorePatterns) {
        this.ignorePatterns = ignorePatterns;
        return this;
    }

    private boolean tableIsNotIgnored(String tableName) {
        return ignorePatterns.stream().noneMatch(regex -> regex.matcher(tableName).matches());
    }

    @Override
    public void transfer() {
        // 1.数据库实体转换
        DatabaseTransfer databaseTransfer = new DatabaseTransfer(atlasService);
        AtlasEntity databaseEntity = databaseTransfer.apply(jdbc);

        // 2.表实体转换
        String catalog = (String) databaseEntity.getAttribute(BaseProperties.ATTR_NAME);
        List<AtlasEntity> tableEntities = jdbc.getTables(catalog, catalog).parallelStream()
                .filter(jdbcTable -> tableIsNotIgnored(jdbcTable.getTableName()))
                .map(new TableTransfer(atlasService, databaseEntity))
                .toList();

        // 3.列转换
        for (AtlasEntity tableEntity : tableEntities) {
            String tableName = (String) tableEntity.getAttribute(BaseProperties.ATTR_NAME);
            List<JdbcPrimaryKey> primaryKeys = jdbc.getPrimaryKeys(catalog, tableName);
            jdbc.getColumns(catalog, catalog, tableName).parallelStream()
                    .forEach(new ColumnTransfer(atlasService, tableEntity, primaryKeys));
        }
    }

}

3. 元数据存入Atlas

public class DatabaseTransfer implements Function<Jdbc, AtlasEntity> {

    private final AtlasService atlasService;

    public DatabaseTransfer(AtlasService atlasService) {
        this.atlasService = atlasService;
    }

    @Override
    public AtlasEntity apply(Jdbc jdbc) {
        String userName = jdbc.getUserName();
        String driverName = jdbc.getDriverName();
        String productName = jdbc.getDatabaseProductName();
        String productVersion = jdbc.getDatabaseProductVersion();

        String url = jdbc.getUrl();
        String urlWithNoParams = url.contains("?") ? url.substring(0, url.indexOf("?")) : url;
        String catalogName = urlWithNoParams.substring(urlWithNoParams.lastIndexOf("/") + 1);
        // 特殊处理 Oracle
        if (productName.equalsIgnoreCase("oracle")){
            catalogName = userName.toUpperCase();
            urlWithNoParams = urlWithNoParams + "/" + catalogName;
        }

        DatabaseProperties properties = new DatabaseProperties();
        properties.setQualifiedName(urlWithNoParams);
        properties.setDisplayName(catalogName);
        properties.setOwner(userName);
        properties.setUrl(url);
        properties.setDriverName(driverName);
        properties.setProductName(productName);
        properties.setProductVersion(productVersion);

        // 1.创建Atlas Entity
        AtlasEntity atlasEntity = new AtlasEntity(DatabaseProperties.JDBC_TYPE_DATABASE, properties.getAttributes());

        // 2.判断是否存在实体, 存在则填充GUID
        Map<String, String> searchParam = Collections.singletonMap(DatabaseProperties.ATTR_QUALIFIED_NAME, urlWithNoParams);
        Optional<AtlasEntityHeader> entityHeader = atlasService.checkAtlasEntityExists(DatabaseProperties.JDBC_TYPE_DATABASE, searchParam);
        entityHeader.ifPresent(header -> atlasEntity.setGuid(header.getGuid()));

        // 3，存储或者更新到Atlas中
        if (entityHeader.isPresent()){
            atlasService.createAtlasEntity(new AtlasEntity.AtlasEntityWithExtInfo(atlasEntity));
        }else {
            AtlasEntityHeader header = atlasService.createAtlasEntity(new AtlasEntity.AtlasEntityWithExtInfo(atlasEntity));
            atlasEntity.setGuid(header.getGuid());
        }
        return atlasEntity;
    }
}

效果展示

1. 元数据类型定义

2. 测试导入元数据

由于mysql没有采用schema，因此jdbc_schema为空

如图所示，可以清晰的了解mysql数据库中demo数据库的数据表内容

数据表元数据，qualifiedName使用数据库连接url.表名

如同所示，数据表内各个列的元数据；可以清晰的了解该数据表的各个字段信息