Choose where to start
The first three guides form an end-to-end workflow, but you can also enter at the stage that matches what you already know.Diagnose slow queries
Start here when you need to identify a recurring slow-query pattern. Use local or cluster-wide query history to choose a representative run, inspect its resource usage, and form an initial hypothesis.
Isolate query bottlenecks
Start here when you have a slow query but do not yet know which operations dominate its performance. Compare progressively simpler forms of the query under controlled conditions.
Choose an optimization approach
Start here when you have evidence about the bottleneck and need to choose a targeted change. Compare approaches for reducing the data that is read, aligning the data layout with your queries, or precomputing repeatable work.
Example
Worked optimization example
Start with workload queries against the NYC Taxi dataset, then apply schema and ordering-key changes, measuring each stage against the same baseline.