Cloudflare Identifies Query Planning Bottleneck in ClickHouse
Summary
Cloudflare traced a billing pipeline slowdown to query-planning contention in ClickHouse. The team found that lock contention and repeated copying of the parts list were driving latency even though basic CPU, I/O, and memory metrics looked normal. It patched ClickHouse to use a shared lock, remove per-query list copying, and improve part filtering, then deployed the fix in March 2026. Query durations dropped by 50%, and Cloudflare contributed the changes upstream to ClickHouse 25.11. The article also highlights a broader operational warning: partition growth and metadata load can create long-term scaling risk even when surface-level monitoring looks healthy.
Classifications
industries
No industries detected
applications
Business Intelligence
AskAI Classifications
Labels
Cloud Infrastructure Software
Cybersecurity Software
Developer Tools
Linked Companies
Cloudflare
$1B+
ClickHouse
$25M to $50M