Pgbench kernel ab benchmark

[nmanthey]

Adding the pgbench postgresql benchmark as a new highlevel regression benchmark. Coefficient of variance is low enough to consider this benchmark as a stable benchmark to detect regressions (CV 0.7–1.7%). Since the same postgresql version used. Running multiple test entries in one pipeline works.

Testing Done

Ran the pgbench-kernel-regression test end-to-end on AWS (local-orchestrator path), base kernel 6.1.141-165.249.amzn2023 vs tip 6.1.150-174.273.amzn2023, on c8i.4xlarge in us-west-2 with the al2023-ami-kernel-6.1-x86_64 AMI.

Single-VM run

• 1/1 VM succeeded; full lifecycle exercised: install kernel A → reboot → pgbench (read-only + read-write) → install kernel B → reboot → pgbench → benchmark analysis.
• Produced benchmark-base-.csv / benchmark-tip-.csv with the expected schema and 4 metrics: postgresql.readonly.tps, postgresql.readonly.latency_avg, postgresql.readwrite.tps,
  postgresql.readwrite.latency_avg.
• Analyzer ran cleanly: "No regressions detected"

Multi-VM statistical comparison (5 VMs)

• 5/5 VMs succeeded; analyzer pooled 20 base + 20 tip rows (5 VMs × 4 metrics) and ran Welch's t-test, Mann-Whitney U, and Cohen's d per metric.
• Cross-VM variance was low (CV 0.7–1.7%), confirming stable measurements and that the per-VM aggregation works.

Combined config (all touched tests, 2 VMs each)

One config running pgbench-kernel-regression, unixbench-kernel-regression, and simple-unixbench at min_count: 2. 6/6 VMs succeeded. pgbench produced 8 base + 8 tip rows and compared correctly.

[nmanthey]

I rebased the pgbench, and re-executed a c8i.4xlarge × 5 VM run:

  pgbench A/B: 6.18 LTS vs 7.2 — done, 5/5 VMs green

  Both kernels booted cleanly, pgbench ran on each, all CSVs produced.
  
  Results (mean over 5 VMs, 6.18 LTS → 7.2.2) 
  ┌──────────────────────────┬──────────┬──────────┬────────┐
  │ Metric                   │ 6.18 LTS │    7.2.2 │ Change │
  ├──────────────────────────┼──────────┼──────────┼────────┤
  │ Read-only TPS            │  454,543 │  539,750 │ +18.7% │
  ├──────────────────────────┼──────────┼──────────┼────────┤
  │ Read-write TPS           │   56,622 │   68,445 │ +20.9% │
  ├──────────────────────────┼──────────┼──────────┼────────┤
  │ Read-only latency (avg)  │ 0.040 ms │ 0.034 ms │ −15.4% │
  ├──────────────────────────┼──────────┼──────────┼────────┤
  │ Read-write latency (avg) │ 0.276 ms │ 0.228 ms │ −17.3% │
  └──────────────────────────┴──────────┴──────────┴────────┘

In case we regress in the future, I'd like to use this test to notify authors/maintainer/...