tune GUC defaults for remote ClickHouse targets

[iskakaushik]

Summary

• queue_capacity: 65,536 → 131,072 (~600MB shmem, buffers ~0.5s at 260K events/sec)
• batch_max: 10,000 → 200,000 (6x drain rate to remote ClickHouse)
• flush_interval_ms: 1,000 → 200 (faster wakeup for bursty workloads)

Previous defaults were tuned for local ClickHouse (sub-ms INSERT latency). With ClickHouse Cloud (~80ms RTT), the old defaults caused 53% event loss at 37K TPS. The new defaults reduce loss to ~9% for the same workload — still not zero (that needs architectural changes like disk buffering), but a significant improvement with no code changes needed from users.

Benchmark results (32 clients, 30s pgbench, CH Cloud us-west-2):

batch_max	Event Loss	Drain Rate
10,000 (old default)	53.2%	~20K events/s
100,000	23.6%	~80K events/s
200,000 (new default)	8.7%	~125K events/s
500,000	2.5%	~134K events/s

Users targeting remote ClickHouse with high-throughput workloads should further increase batch_max and queue_capacity based on their load profile.

Test plan

• Builds clean (mise run build)
• Regression tests pass
• Verified with local ClickHouse Docker (no behavior change expected)

🤖 Generated with Claude Code

[Copilot]

Pull request overview

This PR tunes the GUC (Grand Unified Configuration) defaults for pg_stat_ch to optimize performance for remote ClickHouse targets (e.g., ClickHouse Cloud) which have significantly higher network latency (~80ms RTT) compared to local ClickHouse instances (sub-ms latency). The changes increase queue capacity and batch size while reducing flush interval to improve throughput and reduce event loss under high-load scenarios.

Changes:

• Increased queue_capacity from 65,536 to 131,072 (2x increase, ~600MB shmem)
• Increased batch_max from 10,000 to 100,000 (10x increase for higher drain rate)
• Reduced flush_interval_ms from 1,000ms to 200ms (5x faster wake-up for bursty workloads)

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.