Apache Iceberg tables
Transactions, schema evolution and rollback to an earlier state, on your S3, in a format the whole ecosystem can read.
← Overview
Lakehouse and SQL
An open lakehouse, on your storage
Your data stays in your object storage, as Apache Iceberg tables. Trino queries it in SQL, together with your existing databases, without copying. A shared catalog tells what each table holds, where it comes from and who may read it.
Key capabilities
Open tables, one SQL
Federated SQL with Trino
One query joins your Iceberg tables and your PostgreSQL, Oracle or SQL Server databases, without moving the data.
A SQL editor in the console
Read-only queries, paginated results, bounded time and volume: exploration never brings the cluster to its knees.
A layered catalog
Layers, databases, tables and fields, with their descriptions and full-text search. It updates at the end of every run.
Lineage down to the column
For each column, the source columns and the processing that produced it, captured automatically at run time.
Documentation drafted by AI
graal profiles a sample and proposes a description for a table and its fields. A human validates it before it is published.
How it works
From raw data to a documented table
Step 01
Write
Pipelines and jobs write Iceberg, layer by layer: raw, cleaned, ready to use.
Step 02
Catalog
Every table written joins the catalog with its schema, owner and lineage.
Step 03
Query
Analysts, notebooks, applications and agents query the same tables in SQL, under the same permissions.
Storage and compute stay separate
The data is Parquet files in your object storage; Iceberg adds the metadata that turns them into transactional tables. Compute comes on demand: Trino for interactive SQL, Spark for heavy processing. You size one without touching the other.
One catalog for every team
The graal catalog is the one read by the SQL editor, notebooks, pipelines and agents. A table described once is described for everyone, and its permissions apply whichever way it is accessed.
Standards and integrations
Formats everyone reads
- Apache Iceberg
- Parquet
- Trino
- SQL
- S3
- Apache Spark
- PyIceberg
- OpenLineage
Governance
Every table has its permissions
- Read and write permissions per project, per database and per table
- Every query is attributed to a person, an application or an agent
- Tables stay readable outside graal, by any Iceberg-compatible engine
Frequently asked questions
Is our data copied into graal?
No. Tables live in your S3-compatible storage, and metadata in the PostgreSQL database of your installation.
Can another engine read our tables?
Yes. Iceberg is an open format, and graal exposes an Iceberg REST catalog: Spark, PyIceberg or another compatible engine reads them directly.
Do we have to migrate our databases to query them?
No. Trino queries them where they are. You move to Iceberg what benefits from it, at your own pace.
Query your tables in SQL
A query joining regional consumption and temperature: it is one of the acts of the demonstration.