Files
mtusi2026lab4/tools/ScaleLoad/summarize.py
T

144 lines
8.0 KiB
Python
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
#!/usr/bin/env python3
"""Сводка и графики трека 1 из evidence/track1/results.jsonl (+ diag.jsonl).
Запуск: python tools/ScaleLoad/summarize.py (из корня репозитория)
Создаёт: evidence/track1/summary.md и evidence/track1/fig-*.png
"""
import collections
import json
import os
import re
import statistics as st
import matplotlib
matplotlib.use('Agg')
import matplotlib.pyplot as plt # noqa: E402
ROOT = os.path.dirname(os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
DIR = os.path.join(ROOT, 'evidence', 'track1')
def load(name):
path = os.path.join(DIR, name)
if not os.path.exists(path):
return []
return [json.loads(line) for line in open(path, encoding='utf-8') if line.strip()]
def key_of(label):
return re.sub(r'^([A-Z])-r\d+-', r'\1-', label)
def aggregate(rows):
groups = collections.defaultdict(list)
for r in rows:
groups[key_of(r['label'])].append(r)
result = {}
for k, v in groups.items():
rps = [r['rps'] for r in v]
api = st.mean(r['cpu_pct'].get('api', 0) for r in v)
db = st.mean(r['cpu_pct'].get('db', 0) for r in v)
result[k] = {
'n': len(v), 'rps': st.mean(rps), 'rps_min': min(rps), 'rps_max': max(rps),
'p50': st.mean(r['p50_ms'] for r in v), 'p95': st.mean(r['p95_ms'] for r in v),
'p99': st.mean(r['p99_ms'] for r in v), 'errors': sum(r['errors'] for r in v),
'api_cpu': api, 'db_cpu': db, 'lb_cpu': st.mean(r['cpu_pct'].get('lb', 0) for r in v),
'api_ms_per_req': api * 10 / st.mean(rps), 'db_ms_per_req': db * 10 / st.mean(rps),
}
return result
def table(title, agg, keys, names):
lines = [f'### {title}', '',
'| Конфигурация | повторов | RPS (среднее) | RPS мин–макс | p50, мс | p95, мс | p99, мс | ошибки | CPU API, % | CPU БД, % | CPU API мс/запрос | CPU БД мс/запрос |',
'|---|---|---|---|---|---|---|---|---|---|---|---|']
for k in keys:
if k not in agg:
continue
a = agg[k]
lines.append(f"| {names.get(k, k)} | {a['n']} | {a['rps']:.0f} | {a['rps_min']:.0f}–{a['rps_max']:.0f} | {a['p50']:.0f} | {a['p95']:.0f} | "
f"{a['p99']:.0f} | {a['errors']} | {a['api_cpu']:.0f} | {a['db_cpu']:.0f} | {a['api_ms_per_req']:.2f} | {a['db_ms_per_req']:.2f} |")
return '\n'.join(lines) + '\n'
def main():
rows = load('results.jsonl')
diag = load('diag.jsonl')
agg = aggregate(rows)
agg_diag = aggregate([r for r in diag if r['label'].startswith('E-')])
out = ['# Трек 1: сводка измерений', '',
'Сгенерировано `tools/ScaleLoad/summarize.py` из `results.jsonl` (повторы в случайном порядке; среднее по повторам).', '']
a_keys = ['A-1x2.0', 'A-2x1.0', 'A-4x0.5']
a_names = {'A-1x2.0': '1 × 2.0 CPU', 'A-2x1.0': '2 × 1.0 CPU', 'A-4x0.5': '4 × 0.5 CPU'}
out.append(table('A. Фиксированный бюджет API: 2 CPU / 1 GiB суммарно (БД 4 CPU, LB 2 CPU, 64 клиента)', agg, a_keys, a_names))
b_keys = ['B-1x1.0-db2', 'B-2x1.0-db2', 'B-4x1.0-db2', 'B-8x1.0-db2']
b_names = {k: k.split('-')[1].replace('x', ' × ') + ' CPU' for k in b_keys}
out.append(table('B. Рост ресурсов: реплики по 1 CPU / 512 MiB, БД 2 CPU, LB 1 CPU, 128 клиентов', agg, b_keys, b_names))
c_keys = ['C-base', 'C-pool4', 'C-pool8', 'C-db4cpu']
c_names = {'C-base': '8 реплик, пул 20, БД 2 CPU (контроль)', 'C-pool4': '8 реплик, пул 4, БД 2 CPU',
'C-pool8': '8 реплик, пул 8, БД 2 CPU', 'C-db4cpu': '8 реплик, пул 20, БД 4 CPU'}
out.append(table('C. Проверка узкого места: 8 × 1 CPU, 128 клиентов', agg, c_keys, c_names))
if agg_diag:
e_keys = sorted(agg_diag)
out.append(table('E. Диагностика: API закреплён за ядрами 0–1 (cpuset, без квоты CFS), 64 клиента', agg_diag, e_keys, {}))
conn_path = os.path.join(DIR, 'connections.log')
if os.path.exists(conn_path):
out.append('### Соединения с PostgreSQL после замера (эксперимент C)\n')
out.append('```\n' + open(conn_path, encoding='utf-8').read().strip() + '\n```\n')
with open(os.path.join(DIR, 'summary.md'), 'w', encoding='utf-8') as f:
f.write('\n'.join(out))
# Рис. 1: A — RPS и CPU на запрос при фиксированном бюджете.
ka = [k for k in a_keys if k in agg]
if ka:
fig, ax = plt.subplots(1, 2, figsize=(9, 3.6))
labels = [a_names[k] for k in ka]
ax[0].bar(labels, [agg[k]['rps'] for k in ka], color='#3b6ea5')
ax[0].set_ylabel('успешных запросов/с'); ax[0].set_title('Пропускная способность (бюджет API = 2 CPU)')
ax[1].bar(labels, [agg[k]['api_ms_per_req'] for k in ka], color='#c0504d')
ax[1].set_ylabel('мс CPU API на запрос'); ax[1].set_title('Цена запроса')
for a in ax:
a.tick_params(axis='x', labelsize=8)
fig.tight_layout(); fig.savefig(os.path.join(DIR, 'fig-a-fixed-budget.png'), dpi=130); plt.close(fig)
# Рис. 2: B — RPS, p95 и загрузка БД при росте числа реплик.
kb = [k for k in b_keys if k in agg]
if kb:
xs = [int(k.split('-')[1].split('x')[0]) for k in kb]
fig, ax = plt.subplots(1, 3, figsize=(12, 3.6))
ax[0].errorbar(xs, [agg[k]['rps'] for k in kb],
yerr=[[agg[k]['rps'] - agg[k]['rps_min'] for k in kb], [agg[k]['rps_max'] - agg[k]['rps'] for k in kb]],
marker='o', color='#3b6ea5', capsize=3, label='измерено')
ax[0].plot(xs, [agg[kb[0]]['rps'] * x for x in xs], '--', color='gray', label='линейный рост')
ax[0].set_xlabel('реплик API (по 1 CPU)'); ax[0].set_ylabel('RPS'); ax[0].set_title('Масштабирование'); ax[0].legend(fontsize=8)
ax[1].plot(xs, [agg[k]['db_cpu'] for k in kb], marker='s', color='#c0504d', label='CPU БД')
ax[1].axhline(200, ls='--', color='k', lw=1); ax[1].text(xs[0], 204, 'лимит БД 200 %', fontsize=8)
ax[1].plot(xs, [agg[k]['api_cpu'] / x for k, x in zip(kb, xs)], marker='^', color='#3b6ea5', label='CPU API на реплику')
ax[1].set_xlabel('реплик API'); ax[1].set_ylabel('% одного ядра'); ax[1].set_title('Загрузка'); ax[1].legend(fontsize=8)
ax[2].plot(xs, [agg[k]['p95'] for k in kb], marker='o', color='#8064a2')
ax[2].set_xlabel('реплик API'); ax[2].set_ylabel('p95, мс'); ax[2].set_title('Задержка p95')
for a in ax:
a.set_xscale('log', base=2); a.set_xticks(xs); a.set_xticklabels(xs)
fig.tight_layout(); fig.savefig(os.path.join(DIR, 'fig-b-scale-out.png'), dpi=130); plt.close(fig)
# Рис. 3: C — улучшение в точке насыщения.
kc = [k for k in c_keys if k in agg]
if kc:
fig, ax = plt.subplots(1, 2, figsize=(9.5, 3.6))
names = ['пул 20\n(контроль)', 'пул 4', 'пул 8', 'БД 4 CPU'][:len(kc)]
names = [{'C-base': 'пул 20\n(контроль)', 'C-pool4': 'пул 4', 'C-pool8': 'пул 8', 'C-db4cpu': 'БД 4 CPU\n(пул 20)'}[k] for k in kc]
ax[0].bar(names, [agg[k]['rps'] for k in kc], color=['#7f7f7f', '#3b6ea5', '#3b6ea5', '#9bbb59'][:len(kc)])
ax[0].set_ylabel('RPS'); ax[0].set_title('8 реплик × 1 CPU, БД 2 CPU')
ax[1].bar(names, [agg[k]['p95'] for k in kc], color=['#7f7f7f', '#3b6ea5', '#3b6ea5', '#9bbb59'][:len(kc)])
ax[1].set_ylabel('p95, мс'); ax[1].set_title('Задержка p95')
fig.tight_layout(); fig.savefig(os.path.join(DIR, 'fig-c-bottleneck.png'), dpi=130); plt.close(fig)
print('\n'.join(out))
if __name__ == '__main__':
main()