Performance · Version 1.2.0 · Reviewed 2026-08-02
ClickHouse Performance Tuning Specialist
Locate and remove the dominant bottleneck in ClickHouse latency attribution and ClickHouse throughput optimization with evidence, explicit trade-offs, and a verification plan.
4 method steps
4 documented failure modes
4 diagnostic checks
7 quality gates
Finds the dominant measured bottleneck and designs representative benchmarks for ClickHouse using table engines, sort keys, partitions, queries, and materialized views and query logs, parts count, merges, bytes read, and memory use, with explicit attention to a wrong sort key or tiny-part explosion forcing broad scans and merge debt.
₹199 one-time
Get this skill archive
What it checks first
ClickHouse Performance Tuning Specialist finds the dominant measured bottleneck and designs representative benchmarks for ClickHouse using table engines, sort keys, partitions, queries, and materialized views and query logs, parts count, merges, bytes read, and memory use, with explicit attention to a wrong sort key or tiny-part explosion forcing broad scans and merge debt. Use it when the work involves ClickHouse latency attribution, ClickHouse throughput optimization, ClickHouse performance regression guard.
- A measured baseline and the user-visible target, since optimization without both is guesswork.
- Whether the cost is CPU, memory, I/O wait, or lock contention — they have opposite fixes.
- The p99 path and how many round trips it contains.
- Whether the bottleneck moves after a change, which determines if the gain is real.
Example task
Input
Apply the performance tuning specialist to our ClickHouse system before the next production change. We can provide table engines, sort keys, partitions, queries, and materialized views; the main concern is a wrong sort key or tiny-part explosion forcing broad scans and merge debt.
Expected output
Define the failing percentile and workload, then attribute time with query logs, parts count, merges, bytes read, and memory use. The likely mechanism to disprove first is a wrong sort key or tiny-part explosion forcing broad scans and merge debt. Change one constraint at a time and compare resource use, tail latency, and correctness against a pinned baseline.