Trino is an optional, disabled-by-default SQL query engine for the Data Engineering track. Atlas wires Trino to the Iceberg REST catalog and the MinIO lakehouse buckets. Notebooks, Zeppelin, Airflow tasks and local tools can then query the same tables in SQL.

Image: trinodb/trino:482.

Atlas runs one coordinator, one lakehouse catalog and no workers. The catalog uses Iceberg REST for table metadata and MinIO for objects, the same path Spark uses.

TRINO_SOURCE=disabled                # container | disabled
TRINO_IMAGE=trinodb/trino:482
TRINO_PORT=                          # topology-assigned
TRINO_SCALE=                         # set by ./start.sh from TRINO_SOURCE

TRINO_SOURCE=container also needs MINIO_SOURCE=container and ICEBERG_REST_SOURCE=container. If either is disabled, ./start.sh reports a dependency violation and scales Trino to 0.

The mounted catalog file at services/trino/catalog/lakehouse.properties defines:

• connector.name=iceberg
• iceberg.catalog.type=rest
• iceberg.rest-catalog.uri=http://iceberg-rest:8181
• iceberg.rest-catalog.warehouse=s3://${ENV:MINIO_BUCKET_ICEBERG_LAKEHOUSE}/ (the bucket iceberg-rest, Spark and minio-init use; default lakehouse)
• fs.native-s3.enabled=true for Trino 482 native S3 access to MinIO at http://minio:9000
• scoped Iceberg MinIO credentials through ${ENV:MINIO_ICEBERG_ACCESS_KEY} and ${ENV:MINIO_ICEBERG_SECRET_KEY}

Atlas runs Trino with no Trino authenticator configured. The example user atlas is a convention shared by Atlas notebooks and clients, not an authentication boundary: the local coordinator accepts any user string. Do not use this setup as access control.

Minimal SQL smoke once Spark or another writer has created tables:

SHOW CATALOGS;
SHOW SCHEMAS FROM lakehouse;
SHOW TABLES FROM lakehouse.bronze;
SELECT * FROM lakehouse.bronze.<table_name> LIMIT 10;

Atlas seeds a Zeppelin JDBC interpreter when both ZEPPELIN_SOURCE=container and TRINO_SOURCE=container. Use the named Zeppelin 0.12.1 prefix %trino:

%trino
SHOW TABLES FROM lakehouse.bronze;

The seeded profile uses jdbc:trino://trino:8080/lakehouse, driver io.trino.jdbc.TrinoDriver and Maven artifact io.trino:trino-jdbc:482. Atlas registers it as the named interpreter %trino, not as a %jdbc(prefix) profile.

Python clients inside the Docker network can use the official trino DB-API package:

import trino

conn = trino.dbapi.connect(
    host="trino",
    port=8080,
    user="atlas",
    catalog="lakehouse",
    schema="gold",
)
cur = conn.cursor()
cur.execute("SHOW SCHEMAS FROM lakehouse")
print(cur.fetchall())

Host-side Python clients use the assigned host port:

import os
import trino

conn = trino.dbapi.connect(
    host="localhost",
    port=int(os.environ["TRINO_PORT"]),
    user="atlas",
    catalog="lakehouse",
    schema="gold",
)

Live CTAS smoke after the lakehouse path is up:

CREATE SCHEMA IF NOT EXISTS lakehouse.gold;
CREATE TABLE lakehouse.gold.atlas_trino_ctas_smoke AS
SELECT 1 AS id, 'atlas' AS note;
SELECT * FROM lakehouse.gold.atlas_trino_ctas_smoke;
DROP TABLE lakehouse.gold.atlas_trino_ctas_smoke;

Atlas does not create bronze/silver/gold namespaces at stack startup; a downstream project or the operator owns those namespaces. The CREATE SCHEMA line is part of the live smoke only.

Support tier: experimental — Capability contract declared; no cited cold-start, workflow, or upgrade qualification run yet (evidence at v0.1.0).

• Catalog missing: confirm ICEBERG_REST_SOURCE=container and that iceberg-rest is healthy before Trino starts.
• S3 access denied: confirm minio-init completed (it creates the iceberg service account) and .env has non-empty MINIO_ICEBERG_ACCESS_KEY / MINIO_ICEBERG_SECRET_KEY.
• Kong alias does not load: run ./start.sh --setup-hosts so trino.localhost resolves locally.

Open the full-size diagram for a full-screen view.

Add richer notebook examples that create a Spark-written Iceberg table and query it through Trino.

Superset or another BI UI can sit downstream of Trino once the lakehouse query path is stable.

Worker scaling, query resource groups, access-control files and additional catalogs are out of scope.