Connect data

Give GraphLake read access to the tables, warehouse views, or files that contain your entities and relationships.

On this page

Choose a connector

Databases
PostgreSQL, MySQL, MariaDB, SQLite, SQL Server, Oracle, MongoDB, DuckDB, TiDB, and OceanBase in MySQL mode.
Warehouses and lakehouse
Snowflake, BigQuery, ClickHouse, Trino, Spanner, SAP HANA, Databricks SQL, Apache Doris, StarRocks, Iceberg, and Delta Lake.
Files and object storage
CSV, JSON Lines, and Parquet. Configure S3, MinIO, OSS, TOS, COS, Azure Blob Storage, or GCS as a connection to browse and map files.

Oracle direct access requires Oracle Instant Client on the server. Trino can also expose SQL Server, Oracle, Hive, Hudi, MongoDB, Elasticsearch, and Redshift through configured catalogs. See the connector reference for configuration and supported types.

Add a source in the workbench

  1. Open Data Sources (数据源) in the model draft and select a connector.
  2. Enter connection settings or map file paths to table names. For credentials, specify the environment variable names configured on the GraphLake server.
  3. Choose Add to draft (添加到草稿) and save. Browse tables and columns to confirm the fields you plan to map.

Use credentials with read access to the intended tables and views. Paths resolve on the GraphLake server. CSV and JSON Lines need explicit column types. Parquet supplies its own schema.

Map the tables

Identify a stable key for each entity and the columns that connect one entity to another. Continue to Model a graph. If you want queries to use a scheduled copy of selected columns, create a local Parquet table.