Lakehouse Query

The Query service executes SQL against Iceberg tables registered in the Catalog, reading Parquet files directly from your blob storage. It's the read surface for the Lakehouse — used by the UI and by your own applications.

The engine is DuckDB, so the SQL dialect is standard DuckDB (Postgres-like, with DuckDB extensions).

REST endpoints

Query exposes a small HTTP API over your Lakehouse's public URL:

Endpoint	Purpose
`GET /tables`	List tables registered in the catalog
`GET /schema?table=<name>`	Schema (columns, types) for a table
`GET /partitions?table=<name>`	Partition layout for a table
`POST /query`	Run a SQL statement (body is SQL); returns rows
`GET /swagger`	OpenAPI spec for the above

All calls require a bearer token; the UI handles auth for you. For programmatic access, retrieve your Lakehouse URL and token from the portal.

SQL dialect

Supported out of the box:

Standard aggregates and window functions (AVG, SUM, COUNT, MIN, MAX, ROW_NUMBER, LAG, LEAD, …)
Time helpers — date_trunc, strftime, INTERVAL arithmetic, NOW()
Multi-table joins
Aggressive Parquet pruning — DuckDB's columnar reader uses Iceberg statistics from the catalog to skip files that can't match your predicates

Not supported:

TimescaleDB-style time_bucket, gap_fill, last — use date_trunc + window functions instead.

Examples

-- Last hour of telemetry for one device
SELECT timestamp, temperature, humidity
FROM telemetry
WHERE deviceId = 'sensor-42'
  AND timestamp >= NOW() - INTERVAL '1 hour'
ORDER BY timestamp;

-- 5-minute rollup over a day
SELECT
  date_trunc('minute', timestamp)
    - (EXTRACT(minute FROM timestamp)::INT % 5) * INTERVAL '1 minute' AS bucket,
  AVG(temperature) AS avg_temp,
  MAX(temperature) AS max_temp
FROM telemetry
WHERE timestamp >= DATE '2026-05-20'
  AND timestamp <  DATE '2026-05-21'
GROUP BY bucket
ORDER BY bucket;

How it works

Receives a SQL statement on the public URL.
Resolves table references against the Catalog.
Reads the current Iceberg snapshot, applies partition and column-statistics pruning to pick a minimal Parquet file set.
Reads Parquet from your blob storage; applies projection and filters at the columnar level.
Streams results back to the caller.

Consistency

Queries see a consistent Iceberg snapshot for their duration. Concurrent sink commits don't change a query mid-flight.