Presto

Presto (also called PrestoDB) is an open-source, distributed SQL query engine for interactive analytics on large datasets. Like Trino, it separates compute from storage: a coordinator plans each query, workers scan data in parallel through connectors (Hive, Iceberg, Delta Lake, Hudi, MySQL, Kafka and more), and intermediate results are pipelined through memory, giving answers in seconds over petabyte-scale lakes. Facebook created Presto in 2012; since 2019 it is hosted by the Presto Foundation under the Linux Foundation, with Meta, IBM and Uber among its main contributors.

Presto logo

How it works

The architecture is the classic MPP query engine animated on the Query Engine page: the coordinator parses and optimizes SQL, splits the plan into stages and tasks, assigns splits of the data to workers, and workers exchange data to complete joins and aggregations. Tables are addressed as catalog.schema.table, where a catalog is a configured connector.

Presto and Trino: one origin, two projects

In January 2019 Presto’s original creators left Facebook and continued as PrestoSQL. Facebook kept the original repository and donated it, with the Presto trademark, to the Linux Foundation in September 2019. In December 2020 PrestoSQL was renamed Trino. Full timeline on the Trino page.

Both still share most concepts and much of their SQL, but they have diverged for over six years:

Presto (PrestoDB) Trino
Governance Presto Foundation (Linux Foundation) Trino Software Foundation
Main backers Meta, IBM (acquired Ahana in 2023), Uber, Alibaba Starburst, plus a broad user community
Execution direction Presto C++ native workers (“Prestissimo”) built on Meta’s Velox vectorized engine, replacing Java workers Java engine, frequent releases; fault-tolerant execution with spooled exchanges
Batch / fault tolerance Presto on Spark runs Presto plans as Spark jobs retry-policy=TASK inside Trino
Commercial / managed IBM watsonx.data, Meta and Uber internal platforms Starburst, Amazon Athena (engine v3)

Which one should you pick? For most new teams Trino has the larger community, more connectors and more managed options. Presto is a strong choice where its backers’ platforms are used, or where the native C++ execution path matters for cost at very large scale. Queries written for one usually port to the other with small changes.

Presto C++ and Velox

Velox, open-sourced by Meta in 2022, is a C++ library of vectorized operators (scans, filters, joins, aggregations) shared across engines. Presto C++ replaces the Java worker with a native process that runs Velox, keeping the Java coordinator. The aim is lower CPU and memory per query and fewer JVM garbage-collection pauses. Velox is also used outside Presto, for example by Apache Gluten to accelerate Spark SQL.

Common mistakes

  • Mixing up Presto and Trino documentation, clients and configuration. Property names, JDBC drivers and some functions differ. Check which project and version you run.
  • Using an interactive engine for very long batch jobs without Presto on Spark or retries: one lost worker fails the whole query.
  • Small files and missing partitions on the lake, which make even a fast engine slow.

Learning Resources

Trino Data Architecture