ClickHouse 26.7 adds measured execution plans with EXPLAIN ANALYZE

ClickHouse’s August 6 release announcement introduces EXPLAIN ANALYZE in version 26.7, connecting a logical execution plan with measurements from running the query. The release article explains that the command executes the pipeline and discards result rows while collecting diagnostic information.

See where time and parallelism go

The output separates planning from execution and reports observed parallelism at individual stages. That makes it possible to distinguish a stage with available parallel capacity from one that actually uses it.

Our analysis is that this is valuable when tuning has stalled at a single end-to-end latency number. A slow request may spend time planning, reading, aggregating or moving through a stage with limited concurrency. Those explanations lead to different experiments; adding CPU is not a substitute for identifying the limiting stage.

Treat the plan as evidence from a particular run. Cache state, concurrent queries and input distribution should be recorded with it if the result will guide a capacity decision.

A diagnostic run is still a workload

Because EXPLAIN ANALYZE executes the pipeline, use it with the same care as the query itself. Begin on representative test data or during an appropriate observation window, and avoid turning a troubleshooting session into uncontrolled extra load.

A useful comparison pairs a baseline query with one deliberate change. Compare planning time, execution time, rows processed and memory, then check that the application result remains equivalent outside the diagnostic command.

Start with an expensive query whose bottleneck is unclear. Preserve its exact text, settings and measured plan, and use the stage-level evidence to choose the next experiment rather than changing several tuning options at once.