Trino

What it is

Trino is a distributed SQL query engine built for interactive analytics against large datasets, wherever they live. It doesn't store data itself — it connects to your existing sources (S3/data lakes via Hive or Iceberg connectors, PostgreSQL, MySQL, Kafka, Cassandra, Elasticsearch, and dozens more) and federates queries across them in a single SQL statement if needed. It was forked from Presto in 2020 (originally PrestoSQL) after a split with Facebook over trademark and governance control of the Presto name. Written in Java, ANSI SQL compliant, MPP (massively parallel processing) architecture with a coordinator and worker nodes.

Who builds it and why

Trino is maintained by the Trino Software Foundation, a vendor-neutral nonprofit set up specifically to avoid a repeat of the Presto trademark dispute. The core team includes engineers from Starburst (the primary commercial vendor built on Trino), plus contributors from companies like LinkedIn, Amazon, and others who run it at scale internally. 1,159 contributors on GitHub is a healthy, broad contributor base — not a single-vendor project wearing an open-source mask, though Starburst clearly drives a large share of commits and roadmap direction since their business is built on it.

Production readiness signal

482 releases and a commit as recent as your data window indicates active, continuous development — this isn't a project coasting on momentum. 12,982 GitHub stars is modest for a project this operationally significant, which is typical for infrastructure tools used by data platform teams rather than app developers (low visibility, high dependency). Apache-2.0 license, no vendor lock-in risk at the license level. CNCF maturity is not publicly available — Trino is not a CNCF project; it sits under its own foundation. This is a mature, battle-tested engine: it powers query layers at Netflix, LinkedIn, Amazon, and other large-scale data shops, and has for years under both the Presto and Trino names combined.

Who should use this

Teams that need one SQL interface across heterogeneous data stores — a data lake in S3/Iceberg, a few operational Postgres databases, and a Kafka topic — without ETL-ing everything into one warehouse first. Also a strong fit if you're running interactive, ad-hoc analytics at scale and want separation between compute and storage (query engine that scales independently of where data sits). If you already have a data lakehouse (Iceberg/Delta) and need a fast, standards-compliant SQL engine on top, Trino is a default-shortlist candidate.

Who should NOT use this

Skip it if you need transactional (OLTP) workloads — Trino is read-heavy analytics only, not a database. Skip it if your data already lives entirely in a single modern cloud warehouse (Snowflake, BigQuery, Databricks SQL) and you have no cross-source federation need — you're adding operational overhead (cluster management, tuning, connector config) for no real gain. Skip it if you lack dedicated platform engineering capacity: Trino clusters need real tuning (memory config, worker sizing, connector-specific quirks) and someone on-call who understands query planning — it is not a "deploy and forget" system. Also reconsider if your workloads are small-data, low-concurrency — the distributed overhead isn't worth it below a certain scale, and a simple Postgres or DuckDB will outperform it on latency for small jobs.

Alternatives

• Apache Spark SQL — better fit if you also need batch ETL/ML pipelines, not just interactive query, at the cost of higher latency for ad-hoc queries.
• Snowflake / BigQuery / Databricks SQL — fully managed, less operational burden, but you give up the multi-source federation model and take on vendor lock-in and usage-based cost.
• Apache Druid — better for sub-second dashboards on time-series/event data specifically, not general-purpose federated SQL.

Pricing

Trino itself is fully open source under Apache-2.0 — no license cost, no usage tiers, no feature gating. You pay in infrastructure (compute for coordinator/worker nodes) and engineering time to operate it. Commercial support and a managed/enterprise distribution are available from Starburst if you want vendor support, additional connectors, or a hosted offering, but the core engine has no paywall.