Storage in object storage, compute spun up per query, and nothing running while you sleep.
Data sits in object storage in your region as compressed columnar files. Compute attaches to it per query.
Workers start in under two seconds and stop when the query finishes. There is nothing to size or leave on.
Identical queries against unchanged data return from cache and are not billed as a scan.
Partition and sort keys are picked automatically from query patterns. You can override them; most teams do not.
Read Parquet and Iceberg from your S3 in place. No ingestion step and no second copy to keep in sync.
Per-user and per-project caps in dollars. A runaway query is stopped, not discovered on the invoice.
This is a columnar analytics engine. Point lookups on a primary key are slower than Postgres, by design. Keep an operational database and use Quarry for the analytics on top.
Per-scan billing punishes queries that read everything. If your dashboards read the entire dataset every fifteen minutes, a reserved cluster will be cheaper and we will say so.
Inserts and merges are supported and are batched. Row-by-row OLTP writes are not what this is for.
Not available. Data residency in the UAE, EU, UK, Singapore or us-east is, but the control plane is ours.
