Start with the query
- Filter early and prune partitions. Predicates on partition columns let the engine skip data it never reads. Queries that scan whole fact tables without filters are the most common cause of slowness.
- Select only the columns you need. Wide
SELECT *over columnar data reads more than necessary. - Set a row limit while exploring. Use a
LIMITto preview results without running unbounded scans - see Run queries and view results.
Read the execution details
After a run, open the query’s execution details to see the Total Elapsed Time broken into planning, queueing, and execution, plus the Query Profile showing the most expensive steps. This tells you whether time is going to queueing (cluster under-provisioned), planning, or execution (the query itself). See Run queries and view results.Cluster-side levers
A simple workflow
- Run the query and open its execution details.
- If queueing dominates, scale the cluster up or enable autoscaling.
- If execution dominates, look at the Query Profile - add filters, prune partitions, or try native execution for scan-heavy patterns.
- For repeated reports, enable result caching.
- Measure again.