#!/usr/bin/env python3 """Rebuild static research figures and aggregate metrics from raw summaries.""" from collections import defaultdict import argparse import json from pathlib import Path import statistics import matplotlib matplotlib.use('Agg') import matplotlib.pyplot as plt ROOT = Path(__file__).resolve().parents[1] def main(): parser = argparse.ArgumentParser() parser.add_argument('--directory', default='evidence/scaling') args = parser.parse_args() dest = ROOT / args.directory source_caption = 'Source: ' + args.directory + '/summary.json; 2026-11-06' records = json.loads((dest / 'summary.json').read_text(encoding='utf-8')) grouped = defaultdict(list) for record in records: grouped[(record['series'], record['replicas'], record['concurrency'])].append(record) aggregate = [] for (series, replicas, concurrency), runs in grouped.items(): cpu, ram = defaultdict(list), defaultdict(list) for run in runs: resource = json.loads((dest / (run['name'] + '.resources.json')).read_text(encoding='utf-8')) for sample in resource['resource_samples']: for c in sample['containers']: service = next((s for s in ('api', 'db', 'lb', 'tester') if '-' + s + '-' in c['Name']), None) if not service: continue cpu[service].append(float(c['CPUPerc'].rstrip('%'))) value = c['MemUsage'].split('/')[0].strip() for unit, factor in [('KiB', 1 / 1024), ('MiB', 1), ('GiB', 1024), ('B', 1 / 1048576)]: if value.endswith(unit): ram[service].append(float(value[:-len(unit)]) * factor) break aggregate.append({'series': series, 'replicas': replicas, 'concurrency': concurrency, 'runs': len(runs), 'api_cpu_each': runs[0]['api_cpu_each'], 'api_memory_each': runs[0]['api_memory_each'], 'db_cpu': runs[0]['db_cpu'], 'db_pool_size': runs[0].get('db_pool_size', 4), 'median_rps': statistics.median(r['successful_rps'] for r in runs), 'min_rps': min(r['successful_rps'] for r in runs), 'max_rps': max(r['successful_rps'] for r in runs), 'failed_requests_total': sum(r['failed_requests'] for r in runs), 'attempts_total': sum(r['successful_requests'] + r['failed_requests'] for r in runs), 'median_success_latency_ms': {p: statistics.median(r['successful_latency_ms'][p] for r in runs) for p in ('p50', 'p95', 'p99')}, 'sampled_cpu_percent_one_core': {s: {'mean': round(statistics.mean(v), 2), 'max': max(v)} for s, v in cpu.items()}, 'sampled_ram_mib_max_per_container': {s: round(max(v), 2) for s, v in ram.items() if v}, 'resource_note': 'Samples span bootstrap/warmup/measurement; CPU 100%=one core. API samples are per replica, not summed.'}) order = {s: i for i, s in enumerate(('fixed', 'growing', 'db-improved', 'saturation-baseline', 'saturation-improved', 'pool-four', 'pool-one', 'lb-unbounded', 'lb-bounded'))} aggregate.sort(key=lambda r: (order[r['series']], r['replicas'], r['concurrency'])) (dest / 'aggregate.json').write_text(json.dumps(aggregate, indent=2) + '\n', encoding='utf-8') plt.rcParams.update({'figure.dpi': 150, 'axes.spines.top': False, 'axes.spines.right': False}) fig, axes = plt.subplots(1, 2, figsize=(11, 4.5), layout='constrained') for series, label, color in [('fixed', 'Fixed total API: 1 CPU / 512 MiB', '#2563eb'), ('growing', 'Each API: 0.5 CPU / 256 MiB', '#b45309')]: rows = sorted((r for r in aggregate if r['series'] == series), key=lambda r: r['replicas']) x, y = [r['replicas'] for r in rows], [r['median_rps'] for r in rows] axes[0].errorbar(x, y, yerr=[[r['median_rps'] - r['min_rps'] for r in rows], [r['max_rps'] - r['median_rps'] for r in rows]], marker='o', capsize=4, label=label, color=color) axes[1].plot(x, [r['median_success_latency_ms']['p95'] for r in rows], marker='o', label=label, color=color) for ax in axes: ax.set_xlabel('API replicas (count)') ax.set_xticks([1, 2, 4]) ax.grid(alpha=.2) ax.legend(fontsize=8) axes[0].set_ylabel('Successful HTTP 201 acknowledgements / second') axes[0].set_title('Scaling: median RPS; whiskers=min/max across repeats') axes[1].set_ylabel('Successful request p95 latency (ms)') axes[1].set_title('Scaling: median of run-level p95') fig.suptitle('POST messages via HAProxy; concurrency 32; DB 0.25 CPU\n' + source_caption, fontsize=11) fig.savefig(dest / 'scaling.png') plt.close(fig) fig, axes = plt.subplots(1, 2, figsize=(11, 4.5), layout='constrained') for series, label, color in [('saturation-baseline', 'DB 0.25 CPU', '#b45309'), ('saturation-improved', 'DB 1.0 CPU', '#2563eb')]: rows = sorted((r for r in aggregate if r['series'] == series), key=lambda r: r['concurrency']) x = [r['concurrency'] for r in rows] axes[0].errorbar(x, [r['median_rps'] for r in rows], yerr=[[r['median_rps'] - r['min_rps'] for r in rows], [r['max_rps'] - r['median_rps'] for r in rows]], marker='o', capsize=4, label=label, color=color) axes[1].plot(x, [r['median_success_latency_ms']['p95'] for r in rows], marker='o', label=label, color=color) for ax in axes: ax.set_xlabel('Concurrent closed-loop clients (count)') ax.set_xticks([1, 4, 8, 32, 64]) ax.grid(alpha=.2) ax.legend() axes[0].set_ylabel('Successful HTTP 201 acknowledgements / second') axes[0].set_title('Shared dependency saturation: median and min/max') axes[1].set_ylabel('Successful request p95 latency (ms)') axes[1].set_title('Cost of concurrency: median of run-level p95') fig.suptitle('4 API x 0.5 CPU / 256 MiB; only DB CPU changes\n' + source_caption, fontsize=11) fig.savefig(dest / 'saturation.png') plt.close(fig) controls = [r for r in records if r['series'] == 'pool-four'] candidates = [r for r in records if r['series'] == 'pool-one'] if controls and candidates: controls.sort(key=lambda r: r['repeat']) candidates.sort(key=lambda r: r['repeat']) fig, axes = plt.subplots(1, 2, figsize=(10, 4.5), layout='constrained') for series, label, offset, color in [(controls, 'Pool 4 per API', -.18, '#b45309'), (candidates, 'Pool 1 per API', .18, '#2563eb')]: x = [r['repeat'] + offset for r in series] axes[0].bar(x, [r['successful_rps'] for r in series], width=.35, label=label, color=color) axes[1].bar(x, [r['successful_latency_ms']['p95'] for r in series], width=.35, label=label, color=color) for ax in axes: ax.set_xlabel('Alternating experiment pair (number)') ax.set_xticks([r['repeat'] for r in controls]) ax.legend() ax.grid(axis='y', alpha=.2) axes[0].set_ylabel('Successful HTTP 201 acknowledgements / second') axes[0].set_title('Connection pool control: observed RPS per run') axes[1].set_ylabel('Successful request p95 latency (ms)') axes[1].set_title('Connection pool control: observed p95 per run') fig.suptitle('Only pool limit changes: 4 API; DB 0.25 CPU; concurrency 32\n' + source_caption, fontsize=10) fig.savefig(dest / 'pool.png') plt.close(fig) print(json.dumps(aggregate, indent=2)) if __name__ == '__main__': main()