#!/usr/bin/env python3 """Build CSV and figures from preserved measurements (no invented values).""" import csv, json, pathlib, statistics, os os.environ.setdefault('MPLCONFIGDIR','/private/tmp/track1-mpl') import matplotlib matplotlib.use('Agg') import matplotlib.pyplot as plt ROOT=pathlib.Path(__file__).resolve().parents[1] OUT=ROOT/'evidence/track1' rows=[] for path in sorted(OUT.glob('*-r[12].json')): r=json.loads(path.read_text()); label=r['label'].rsplit('-r',1)[0] samples={} seen=set() for line in path.with_name(path.stem+'-stats.jsonl').read_text().splitlines(): # Docker's streaming output may prepend terminal control sequences. line=line[line.find('{'):] try: s=json.loads(line) except ValueError: continue name=s['Name'] if name not in seen: seen.add(name) continue # Initial Docker sample predates the measurement interval. service=name.rsplit('-',1)[0].replace('mtusi-lab3-','') samples.setdefault(service,[]).append(float(s['CPUPerc'].rstrip('%'))) row=dict(case=label,repeat=path.stem[-1],rps=r['successful_rps'],read_p95_ms=r['read']['p95_ms'],write_p95_ms=r['write']['p95_ms'],error_share=r['error_share'],lost=r['verify']['lost'],duplicates=r['verify']['duplicate_ids']+r['verify']['duplicate_texts']) for s,values in samples.items(): row[s+'_cpu_mean']=statistics.mean(values) rows.append(row) if not rows: raise SystemExit('No results') fields=list(dict.fromkeys(k for r in rows for k in r)) with (OUT/'summary.csv').open('w') as f: w=csv.DictWriter(f,fields); w.writeheader(); w.writerows(rows) def avg(label,key): return statistics.mean(r[key] for r in rows if r['case']==label) def bounds(label): v=[r['rps'] for r in rows if r['case']==label]; m=statistics.mean(v) return [m-min(v),max(v)-m] plt.rcParams.update({'font.size':10}) fig,axes=plt.subplots(1,2,figsize=(10,3.5)) for labels,title,ax in [(['fixed-1','fixed-2','fixed-4'],'Общий бюджет API: 2 CPU / 1 ГиБ',axes[0]),(['grow-1','fixed-2','grow-4'],'На реплику: 1 CPU / 512 МиБ',axes[1])]: means=[avg(l,'rps') for l in labels] ax.errorbar([1,2,4],means,yerr=list(zip(*(bounds(l) for l in labels))),fmt='o-',capsize=5) ax.set(xticks=[1,2,4],xlabel='Реплики API',ylabel='Успешные запросы/с',title=title,ylim=(0,max(means)*1.2)); ax.grid(alpha=.25) fig.tight_layout(); fig.savefig(OUT/'scaling.png',dpi=160); plt.close(fig) labels=['grow-4','lb2-4','db4-4'] if all(any(r['case']==l for r in rows) for l in labels): fig,axes=plt.subplots(1,2,figsize=(10,3.5)) axes[0].bar(['Исходный','LB: 2 CPU','БД: 4 CPU'],[avg(l,'rps') for l in labels],yerr=list(zip(*(bounds(l) for l in labels))),capsize=5) axes[0].set(ylabel='Успешные запросы/с',title='Проверка узкого места: 4 API × 1 CPU') for s,limit in [('api',100),('lb',100),('postgres',200),('valkey',100)]: axes[1].bar(s,avg('grow-4',s+'_cpu_mean')/limit*100) axes[1].axhline(100,color='red',linestyle='--'); axes[1].set(ylabel='% от CPU-лимита',title='Средняя загрузка до улучшения') fig.tight_layout(); fig.savefig(OUT/'bottleneck.png',dpi=160); plt.close(fig) for l in dict.fromkeys(r['case'] for r in rows): print(l, 'RPS', round(avg(l,'rps')), 'read p95',round(avg(l,'read_p95_ms'),2),'write p95',round(avg(l,'write_p95_ms'),2), 'CPU', {s:round(avg(l,s+'_cpu_mean')) for s in ['api','lb','postgres','valkey']})