Presto is a distributed SQL query engine for interactive analytics over data where it already lives - files in a data lake, tables in a warehouse, rows in a relational database - without copying it first. Built at Facebook to query petabyte-scale Hive warehouses in seconds, it splits into two projects in 2019: Trino (the original team's continuation, formerly PrestoSQL) and PrestoDB (maintained under the Linux Foundation). Both share the same architecture and most of the same SQL.

Query federation through connectors

The defining feature is federation. A pluggable connector exists for each data source - Hive/S3, PostgreSQL, MySQL, Kafka, MongoDB, and dozens more - and a single query can join across them. You can join a fact table in your lake against a dimension table in Postgres in one statement, with the engine pushing down filters to each source where possible.

SELECT u.name, SUM(e.amount)
FROM hive.analytics.events e
JOIN postgres.crm.users u ON e.user_id = u.id
WHERE e.dt = '2026-06-16'
GROUP BY u.name;

Built for interactive speed

Presto/Trino is a pure query engine with no storage of its own. A coordinator parses and plans a query, then distributes work to workers that process data in parallel, entirely in memory and streaming between stages - no writing intermediate results to disk between steps the way MapReduce does. That MPP, in-memory design is what delivers interactive latency on large datasets, and it commonly reads tables defined in the Hive Metastore.

Where it fits

Trino (the more actively developed fork) is the usual choice for an interactive SQL layer over a lakehouse, ad-hoc analytics across mixed sources, and BI tools that need fast responses. Because it holds working data in memory, very large joins can hit memory limits, so it complements rather than replaces batch engines for heavy transformations. Many platforms run Spark for ETL and Trino for interactive query over the same files.

Frequently asked questions

What is the difference between Presto and Trino? They split from a common codebase in 2019; Trino (formerly PrestoSQL) is the original developers' continuation and is the more actively developed, while PrestoDB continues under the Linux Foundation.

How does query federation work? Pluggable connectors let one query read and join across many sources - a data lake, relational databases, Kafka - with filters pushed down to each source.

Is it a database? No - it stores no data. It is a query engine that runs SQL over external sources, holding working data in memory for interactive speed.