forked from mtusi2026/mtusi2026lab7
Перенесена реализация lab6; бизнес-API без изменений, транспорт — HTTPS. - Внешний HTTPS (public-ca) на 127.0.0.1:8443; HSTS, TLS >= 1.2. - Обязательное mTLS LB -> каждая API-реплика: API проверяет цепочку internal-ca и SAN клиента (allowlist: lb); health только на 127.0.0.1. - TLS с проверкой сервера ко всем зависимостям: PostgreSQL verify-full (pg_hba только hostssl), HTTPS S3, AMQPS, OTLP/HTTPS с client cert, mTLS Collector -> Prometheus/Loki/Tempo, Grafana -> datasources, scrape. - OpenBao 2.7.2 не в dev-mode: Raft на volume, init/unseal оператором, auth/cert с ролью и policy на сервис, TTL 15 мин + renew/re-login. - Отдельные роли PostgreSQL с грантами по таблицам; разные ключи S3 и пользователи брокера выдаются из OpenBao. - Argon2id (m=19 MiB, t=2, p=1, замер на стенде) и rehash при входе. - Случайный X-Instance-Id; hardening контейнеров (non-root, cap_drop ALL, no-new-privileges, read-only, лимиты); TCP_NODELAY на сокете API. - Модель угроз, trust matrix, отрицательные тесты TLS/mTLS, API, OpenBao, hardening; evidence, в т.ч. прогон из чистого clone.
70 lines
3.3 KiB
Python
70 lines
3.3 KiB
Python
#!/usr/bin/env python3
|
|
"""Lab6: скорость разбора backlog worker-ами (до/после исправления цикла AMQP в worker.run_job).
|
|
|
|
python3 student-tests/drain_bench.py --count 240 --label after-fix --output evidence/drain-bench-after.json
|
|
|
|
worker останавливаются, через API принимаются --count маленьких PNG (64x48, 202 queued), worker
|
|
запускаются; измеряется время, пока все вложения не станут ready, и по Prometheus — пропускная
|
|
способность (jobs/s), utilization потребителей и CPU worker в окне разбора.
|
|
"""
|
|
import argparse
|
|
import concurrent.futures
|
|
import sys
|
|
import time
|
|
|
|
from lab5lib import Client, Failure, Log, check, compose, png, sql, upload
|
|
import obslib
|
|
|
|
|
|
def main():
|
|
p = argparse.ArgumentParser()
|
|
p.add_argument("--count", type=int, default=240)
|
|
p.add_argument("--label", default="run")
|
|
p.add_argument("--output")
|
|
args = p.parse_args()
|
|
log = Log("worker drain benchmark (%s)" % args.label)
|
|
c = Client(timeout=30)
|
|
alice, _bob, _eve, chat = c.bootstrap()
|
|
compose("stop", "worker")
|
|
imgs = [png(64, 48, color=(i % 256, (i * 7) % 256, 90)) for i in range(args.count)]
|
|
with concurrent.futures.ThreadPoolExecutor(8) as ex:
|
|
ids = list(ex.map(lambda img: upload(c, alice, chat, img, "bench.png")["id"], imgs))
|
|
log.step("backlog created with workers stopped", uploads=len(ids))
|
|
t0 = time.time()
|
|
compose("up", "-d", "--no-deps", "--no-recreate", "--wait", "worker", timeout=200)
|
|
t_up = time.time()
|
|
while True:
|
|
left = int(sql("SELECT count(*) FROM attachments WHERE chat_id = '%s' AND status <> 'ready'" % chat["id"])[0][0])
|
|
if left == 0:
|
|
break
|
|
check(time.time() - t0 < 600, "drain timeout")
|
|
time.sleep(0.5)
|
|
t1 = time.time()
|
|
drain = t1 - t_up
|
|
time.sleep(25) # метрики экспортируются раз в 10 с
|
|
w = max(30, int(t1 - t_up) + 15)
|
|
res = {
|
|
"uploads": len(ids), "drain_seconds_after_workers_up": round(drain, 1),
|
|
"jobs_per_second_wallclock": round(len(ids) / drain, 2),
|
|
"workers_x_concurrency": obslib.prom_value("sum(last_over_time(messenger_worker_capacity[30s]))"),
|
|
"utilization_peak_1m": round(obslib.prom_value(
|
|
"max_over_time((sum(rate(messenger_worker_busy_time_seconds_total[30s])) / sum(last_over_time(messenger_worker_capacity[30s])))[%ds:5s])" % w), 3),
|
|
"job_duration_p50_s": round(obslib.prom_value(
|
|
'histogram_quantile(0.5, sum by (le) (increase(messenger_job_duration_seconds_bucket{messenger_job_outcome="ready"}[%ds])))' % w), 3),
|
|
"worker_cpu_cores_peak": round(obslib.prom_value(
|
|
'max_over_time(sum(rate(container_cpu_time_seconds_total{job="messenger-worker"}[30s]))[%ds:5s])' % w), 3),
|
|
}
|
|
log.step("drained", **res)
|
|
log.save(args.output, label=args.label, **res)
|
|
print("PASS: %s drained %d jobs in %.1fs (%.1f jobs/s)" % (args.label, len(ids), drain, len(ids) / drain))
|
|
return 0
|
|
|
|
|
|
if __name__ == "__main__":
|
|
try:
|
|
sys.exit(main())
|
|
except Failure as exc:
|
|
print("FAIL:", exc)
|
|
compose("up", "-d", "--no-deps", "--no-recreate", "worker", ok=False)
|
|
sys.exit(1)
|