Trino + Iceberg + S3
An open lakehouse query stack where Trino runs interactive SQL over Apache Iceberg tables on S3. It maximizes engine independence and federation at the cost of self-managed operations.
Trino + Iceberg + S3
This stack provides interactive SQL analytics over an open lakehouse. Trino, a distributed MPP query engine, runs fast queries directly against Apache Iceberg tables whose data sits as open Parquet files in Amazon S3. There is no proprietary warehouse: storage, table format, and engine are all open and independently replaceable.
Components
- Trino: A massively parallel SQL engine that queries data where it lives. It separates compute from storage, federates across many sources, and delivers low-latency interactive queries.
- Apache Iceberg: An open table format with hidden partitioning, schema and partition evolution, snapshot isolation, and time travel. It tracks metadata so engines can plan efficiently without listing files.
- Amazon S3: Durable, cheap object storage holding the Parquet data and Iceberg metadata files.
- Catalog: AWS Glue, a REST catalog, or Nessie tracks Iceberg table metadata and enables multi-engine access.
Strengths
- Engine independence. Iceberg's open spec lets Trino, Spark, Flink, and others read the same tables, avoiding lock-in.
- Interactive speed. Trino is optimized for ad hoc analytical queries and BI dashboards.
- Federation. Trino joins lakehouse data with operational databases in a single query.
- Cost efficiency. Compute scales separately from cheap S3 storage, and you run only the engine you need.
Trade-offs
- Self-managed complexity. Running Trino clusters, catalogs, and maintenance jobs requires operations effort.
- No built-in storage optimization. Compaction and snapshot expiration must be scheduled.
- Consistency model. Trino is for analytics, not transactional workloads.
- Tuning. Memory and concurrency settings need attention under heavy load.
When to Use It
Choose this stack for an open, vendor-neutral lakehouse where you want fast interactive SQL and multi-engine access to the same tables. It fits organizations standardizing on Iceberg, running BI and ad hoc analytics, and federating across sources. If you prefer fully managed simplicity over openness, a hosted warehouse may suit. For teams prioritizing open formats and engine choice, Trino plus Iceberg on S3 is a leading pattern.