PROVIDER Β· Distributed SQL federation query engine
Trino
Trino is the distributed SQL engine maintained by Presto's core team, unifying lakes, warehouses, and relational stores behind one SQL dialect through over a hundred connectors for federated cross-source joins.
Pros
Cons
Product Information
Detailed Notes
Overview
Trino was renamed in 2019 from PrestoSQL by Presto's core team, the founders of Starburst, as a distributed MPP SQL engine for interactive analytics that stores no data itself but unifies Hive, Iceberg, MySQL, Kafka, MongoDB, and dozens more behind SQL through connectors.
For Backend & Database work the core value is data virtualisation, since cross-source joins happen without moving data, pairing Hive detail with MySQL dimensions or Kafka streams with Iceberg history, and decoupled compute scales query capacity independently of the sources.
The limits are performance and governance, because cross-source joins pay heavy data movement costs and poorly suit very large federated results, there is no native metadata or statistics to rely on beyond what connectors supply, and high concurrency needs careful resource group configuration with operations heavier than an all-in-one warehouse.
Core Strengths
- Over a hundred connectors put lakes, warehouses, relational stores, and queues behind one SQL dialect
- Fully decoupled compute scales query capacity independently of storage
- Cross-source joins virtualise data without movement
- An active community plus commercial support from Starburst with fast releases
Product Ecosystem
Connectors
Hive, Iceberg, Delta, MySQL, PostgreSQL, Kafka, MongoDB, Elasticsearch, and more loaded on demand.
Resource groups and security
Resource groups govern concurrency and queues with file and system access control and finer permissions in commercial editions.
Deployment
Self-managed clusters, Kubernetes, and the Starburst Galaxy managed service.
Drawbacks
- Cross-source joins move large amounts of data with poor performance on huge results
- No native metadata or statistics, leaving optimisation to connector capability
- High concurrency needs careful resource group tuning with a steep operations curve
- Stores no data and cannot replace a warehouse or database
Use Cases
- Unified queries across sources (β β β β β ) - Connectors with federated joins
- Interactive lake analytics (β β β β β ) - Query Iceberg and Hive in place
- Merging streams with history (β β β β ) - Kafka joined with lake tables
- Very large result exports (β β ) - Movement cost, use a warehouse
Pricing
| Plan | Price | Scope |
|---|---|---|
| Trino open source | $0 | Apache 2.0, self-managed |
| Starburst Galaxy | Per cluster and usage | Managed with optimisation and governance |
| Self-managed cloud cost | Cloud VMs and storage | Elastic scaling controls spend |
The open source edition has no licence fee with cluster compute as the cost, controlled by scaling nodes up under load and down when idle, while managed services bundle operations and optimisation billed per cluster for teams without a data platform.
FAQ
-
Trino or Presto? Same lineage split, with Trino maintained by the original core team at a faster cadence and a livelier community, while PrestoDB from Meta persists in some enterprises.
-
Can it serve as a warehouse? Not fully, since it stores no data or table lifecycles, and typically pairs with Iceberg as a query layer while warehousing stays on a lakehouse or cloud warehouse.
-
How is federated performance? Cross-source joins on small results work while large results slow down from data movement, so sync common dimensions into one source first.