Serving is the read path of the platform. Analysts, dashboards and external API consumers never scan raw lake files directly — they submit SQL to the distributed query engine or request governed metrics from the semantic layer, which translates business definitions into warehouse SQL with row-level policies applied.
Responsibility in one sentence
Execute federated SQL over the lakehouse, cache hot results, throttle abusive queries, and resolve governed metrics into policy-aware SQL.
Could not build diagram for model "example-data-processing".
Trino/Presto cluster — plans, executes and caches distributed SQL over lake + warehouse.
Governed metrics and dimensions — one definition, many consumers.
Both containers expose REST APIs documented on their Overview tabs:
Live dashboard tiles query materialized stream outputs through the engine:
Analysts iterate in notebooks before publishing governed dashboard tiles:
Fair use
When a query exceeds its time budget, the Query Throttler kills it and suggests an equivalent semantic-layer metric instead.