SQL warehouses
Creating the compute that answers SQL, sizing it so it scales with load and stops when nobody is using it, and reading what it costs.
A SQL warehouse is a pool of workers that answers SQL queries, powered by Trino™. SQL notebooks, dashboards and the %%trino command in kernel notebooks all run on one. It grows and shrinks with load, and stops when nothing has queried it for a while.
Creating a warehouse
Open Warehouses under Compute and choose New warehouse. Owners, admins and members can create and run warehouses.
| Field | What it does |
|---|---|
| Name, Description | How people recognise it. |
| Worker size | The compute profile each worker runs as, which sets how big a worker is. Only profiles your role or group may use are listed — a warehouse reads the catalog as its profile, so choosing one is choosing what it can read. |
| Min workers | The floor while running; 1 by default. 0 scales to nothing while idle: cheaper, but the first query afterwards waits for a worker to start. |
| Max workers | The ceiling, and so the cost ceiling; 4 by default. Your plan may set a maximum, shown under the field. |
| Scale-down grace (seconds) | How long to hold the current size after load drops; 300 by default. |
| Idle timeout (seconds) | How long with no queries at all before the warehouse stops; 1800 by default. 0 means it never stops for being idle. |
Your organisation limits how many warehouses can run at once. The Warehouses page shows how many are running, the limit, and this month's cost.
How it scales and stops
Scale-down and the idle timeout are separate. Scale-down releases workers when load drops but keeps the warehouse running. The idle timeout stops it entirely when nothing is querying it. A warehouse can shrink without stopping, and stop without ever having shrunk.
States
| State | Means |
|---|---|
| Running | Answering queries. |
| Starting, Scaling, Stopping | Changing size or state. |
| Stopped | No workers and no cost. The page says why: stopped by someone, stopped after being idle, stopped because the organisation is over its limit, or stopped after an error. |
| Error | Something failed. The page shows what, what to do about it, and a reference to quote if you need help. |
The warehouse page
- Start, Stop, Settings and Delete.
- Workers now against the ceiling, this month's cost, and — while running — how long until it stops for being idle.
- Resize to changes the worker size.
- Recent usage lists each metered interval: workers, worker-seconds and cost.
The Cost page lists usage across every warehouse, for anyone sizing one to see what it costs before they do.
Using a warehouse
- SQL notebooks send their queries to the most recently started warehouse that is running. The notebook's Engine button names it. See The notebook editor.
- Dashboards read through a warehouse. See SQL and dashboards.
- Kernel notebooks reach it with
%%trino. See Kernels.
What a warehouse lets you read is decided per person: your own grants apply, and column masks and row filters are enforced on every query.