Trino

← Back to Batch Processing

Open source distributed SQL query engine (formerly PrestoSQL, originated at Facebook) for interactive analytics across heterogeneous data sources. Queries data in place without requiring ETL into a separate system — a single SQL query can join across a data lake, a relational database, and a key-value store.

Key Properties

• Federated Queries
• Connector Architecture
• MPP (Massively Parallel Processing)
• Query Optimization

How It Works

A coordinator node parses SQL, plans the query, and distributes work across worker nodes. Connectors abstract data sources (Hive/S3, PostgreSQL, Kafka, Elasticsearch, etc.) behind a common interface. Executes entirely in memory with pipelined stages — no intermediate disk writes — optimized for low-latency interactive queries rather than long-running ETL.

Connectors

• Hive / S3 — query Parquet, ORC, Avro files in data lakes
• Iceberg / Delta Lake — table formats with ACID semantics
• RDBMS — PostgreSQL, MySQL, SQL Server
• NoSQL — Cassandra, Elasticsearch, MongoDB, Redis

Related

• Data Lakehouse (Trino commonly used as the query layer)
• Data Warehouse (alternative approach — Trino queries data in place instead)
• Apache Spark (complementary — Spark for heavy ETL, Trino for interactive queries)
• BigQuery (managed alternative on GCP)

---

data-pipelines batch trino