How it works
The extension hooks into PostgreSQL’s query lifecycle to collect timing, row counts, and error data. Events are written to a lock-free shared-memory ring buffer and flushed to ClickHouse by a background worker. All aggregation (p50/p95/p99, top queries, errors) happens in ClickHouse via materialized views — not in the extension.Features
- Zero-copy capture — hooks write directly into shared memory with no heap allocation
- Background export — a background worker drains the queue and batches inserts to ClickHouse
- ClickHouse aggregation — materialized views handle percentiles, top-N, and error rollups
- Multi-version support — works with PostgreSQL 16, 17, and 18