diff --git a/REPORT.md b/REPORT.md
index 55d92527..8e17dadb 100644
--- a/REPORT.md
+++ b/REPORT.md
@@ -1,7 +1,7 @@
# Отчёт — лабораторная 3
- Автор: Самохвалов Андрей Ремович, БПИ-2302.
-- Цель: оценка 5, трек 1 (цена масштабирования). Уровень 4 проверен ниже; исследование трека 1 выполняется отдельно.
+- Цель: оценка 5, трек 1 (цена масштабирования). Уровень 4 и выбранный трек проверены ниже.
- Ветка: `solution/bpi2302-samohvalov`; commit/PR после публикации.
- Источники: `README.md`, `COURSE.md`, `contracts/`, публичные тесты репозитория, [HAProxy 3.0 configuration](https://docs.haproxy.org/3.0/configuration.html), [Docker Compose services](https://docs.docker.com/reference/compose-file/services/). Код и текст подготовлены с помощью Codex; результаты ниже получены реальным запуском.
@@ -81,6 +81,38 @@ python3 student-tests/test_operations.py --rolling --base-url http://127.0.0.1:8
Это **поочерёдная замена той же версии образа**: ни совместимость двух разных бинарных версий, ни миграция схемы во время нагрузки не проверялись. `migrate` выполняется до старта API, а скрипт замены запускает только API без повторной миграции. Для будущей смены схемы нужна отдельная совместимая миграция; данный опыт её не доказывает.
-## Ограничения и дальнейшие измерения
+## Уровень 5: цена масштабирования
-Стенд имеет один Docker-host, один LB и один PostgreSQL: отказ узла или БД не переживает. `depends_on` гарантирует порядок запуска, а маршрутизацию выполняет HAProxy. Сценарии U4 показывают работоспособность и ошибки во время замены, но не предельную RPS при 1/2/4 репликах; отдельный эксперимент трека 1 будет добавлен для уровня 5. Генератор `tests/load.py` замкнутый и не устраняет coordinated omission. Текущий локальный HTTP разрешён только для изолированного стенда с синтетическими данными; удалённая демонстрация требует TLS.
+**Гипотеза.** При фиксированном суммарном лимите API рост с двух до четырёх процессов не даст линейного выигрыша: каждому останется только 0.25 CPU, а конкуренция за пул и планировщик увеличит задержку. Если процессам разрешить суммарно больше CPU, пропускная способность вырастет, но на этой машине рост должен замедлиться около её двух виртуальных CPU.
+
+**Профиль и воспроизведение.** Измерения выполнены 2026-10-02 на том же macOS arm64 / Colima Docker-host с 2 CPU и 2.05 GB RAM, Docker Engine 29.5.2, Compose 5.3.1, образами и Python из раздела запуска. PostgreSQL оставался при 1 CPU/512 MiB, HAProxy при 0.5 CPU/128 MiB; pool каждой реплики ограничен восемью соединениями. Использован один синтетический чат с 1000 заранее записанными сообщениями; общий счётчик БД во время матрицы оставался 30969. Само измерение только читает последнюю страницу `/messages?limit=20` через LB с действующим Bearer. Файл `secrets/scale_fixture.json` хранит локальный тестовый токен и не включён в Git. Набор сообщений и запрос не менялись между точками и контрольным сравнением.
+
+`scripts/benchmark.sh` использует Python stdlib и задаёт *открытый* график поступления GET по монотонным часам: 64 потока, очередь максимум 256, HTTP keep-alive, timeout 4 с. У каждой точки 2 с прогрева и три независимых повтора по 8 с; заданная интенсивность 250/750/1500 запросов/с. Задержка успеха считается от запланированного момента поступления до полного ответа, включая ожидание в очереди; p50/p95/p99 относятся только к успешным ответам. Неуспешные поступления учитывают как HTTP/transport ошибки, так и переполнение очереди отдельно. Во всей матрице `errors_by_type` пуст, поэтому все потери на 1500 запросах/с — **отброшенные генератором поступления**, а не ответы API 5xx. Это доля необслуженных поступлений при заданном ограничении очереди, не точный максимум сервера. Показатель CPU самого генератора не превысил 0.115 с процессорного времени на 1 с наблюдения.
+
+Для чистого clone после настройки `.env` и secrets из раздела запуска (по умолчанию порт 8080):
+
+```sh
+./scripts/benchmark.sh --base-url http://127.0.0.1:8080 --rates 250,750,1500 --duration 8 --warmup 2 --output evidence/scale/baseline.jsonl
+python3 scripts/plot_scale.py evidence/scale/baseline.jsonl
+./scripts/benchmark.sh --base-url http://127.0.0.1:8080 --modes fixed --replicas 4,2 --rates 1500 --duration 8 --warmup 2 --output evidence/scale/confirmation-4-to-2.jsonl
+docker compose --env-file .env down
+```
+
+Скрипт отказывается дописывать существующий output: для нового опыта укажите новое имя файла. Он сам поднимает Compose с временным override лимитов и меняет число реплик. В локальном опыте использован порт 8083, так как 8080 уже занят другим учебным стендом. Матрица `fixed` делила общий лимит API 1 CPU/512 MiB между 1/2/4 репликами (соответственно 1/512, 0.5/256, 0.25/128 на процесс). Серия `growth` оставляла каждому процессу 0.5 CPU/256 MiB, поэтому сумма росла от 0.5/256 до 2 CPU/1024 MiB. Docker-host при этом **не** получил дополнительных физических ресурсов; 2 CPU у четырёх API конкурировали с БД и LB в VM на 2 CPU. Конфигурация override и фактические квоты зафиксированы в каждой JSONL-строке.
+
+Полные 54 результата с индивидуальными CPU/RAM API, PostgreSQL и LB, PostgreSQL counters/размером, pool waits, длительностью, RPS, ошибками и p50/p95/p99: [`evidence/scale/baseline.jsonl`](evidence/scale/baseline.jsonl). Медианы трёх повторов всех 18 точек: [`summary.csv`](evidence/scale/summary.csv). Графики медианных RPS и p95: [`fixed-rps.svg`](evidence/scale/fixed-rps.svg), [`fixed-p95.svg`](evidence/scale/fixed-p95.svg), [`growth-rps.svg`](evidence/scale/growth-rps.svg), [`growth-p95.svg`](evidence/scale/growth-p95.svg).
+
+| Серия, API | Успешные RPS при 750 | Успешные RPS при 1500 | Необслуженные при 1500 | p50 / p95 / p99 при 1500, мс | API / DB / LB CPU при 1500 |
+| --- | ---: | ---: | ---: | ---: | ---: |
+| fixed, 1 | 747 | 780 | 42.85% | 379 / 467 / 725 | 96 / 13 / 4% |
+| fixed, 2 | 750 | 957 | 33.88% | 315 / 468 / 516 | 99 / 14 / 5% |
+| fixed, 4 | 733 | 694 | 50.82% | 414 / 748 / 851 | 87 / 14 / 4% |
+| growth, 1 | 362 | 385 | 72.09% | 797 / 1018 / 1070 | 51 / 6 / 2% |
+| growth, 2 | 750 | 938 | 35.12% | 321 / 494 / 548 | 98 / 14 / 5% |
+| growth, 4 | 747 | 1128 | 20.58% | 231 / 457 / 547 | 143 / 20 / 7% |
+
+**Насыщение и улучшение.** При суммарном 1 CPU рост заданной интенсивности с 750 до 1500 почти не увеличивает успешный RPS у одной реплики, а у четырёх уменьшает его; p95 и отброшенные поступления резко растут. API-группа использует около 87–99% одного CPU, PostgreSQL 13–14%, LB 4–5%; в этом профиле узкое место — CPU квота API и сопутствующее ожидание пула, **не** достигнутый предел CPU БД или LB. Например, `pool_wait_ms` у fixed/1 вырос с 51465 при 750 до 182073 при 1500; это суммарное время ожидания всех запросов за окно, не длительность отдельного запроса. Добавление ресурсов до growth/4 увеличило успешный RPS, но не линейно: API заняли 143% CPU при двух доступных VM CPU, DB 20%, LB 7%, а очередь всё ещё отбрасывала 20.58% поступлений.
+
+Измеримое эксплуатационное изменение — **выбрать 2 API вместо 4 при том же общем 1 CPU/512 MiB**. Отдельный контроль после возврата к неизменённому коду и тому же чату, 3 повтора сначала с 4, затем с 2 репликами по 8 с при 1500 заданных GET/с: [`confirmation-4-to-2.jsonl`](evidence/scale/confirmation-4-to-2.jsonl). В целевом чате по-прежнему 1000 сообщений; общий счётчик БД стал 30970 из-за одного независимого синтетического сообщения в проверке ACL между сериями. Медиана успешного RPS 856.033 → 952.317 (**+11.2%**), необслуженная доля 39.18% → 33.94% (−5.24 п.п.), p95 605.467 → 489.999 мс (−19.1%). Все три RPS с 2 API (940.679–972.715) выше всех трёх с 4 API (805.135–895.216). В `compose.yaml` по умолчанию уже стоят 2 API; для применения настройки после опыта с четырьмя: `docker compose --env-file .env up -d --wait --scale api=2`. Проверка публичного API после контроля — `make test BASE_URL=http://127.0.0.1:8083`, PASS. Это настройка топологии, а не ускорение SQL или исходного кода.
+
+**Пределы вывода.** Открытый график устраняет coordinated omission замкнутого `tests/load.py`, но конечные 64 потока, очередь 256 и локальный клиент влияют на измеренную границу. Отброшенные задания не дошли до HAProxy, а HTTP 5xx в серии не наблюдались. Каждый результат описывает этот GET-профиль на одном Docker-host, не запись сообщений и не универсальную ёмкость API. Кратковременные эффекты CPU throttling и фоновая нагрузка объясняют разброс между первой матрицей и контрольной серией; поэтому эффект выбора 2 API основан на отдельном сравнении с неперекрывающимися тремя повторами. Стенд имеет один LB и один PostgreSQL, не переживает потерю хоста/БД. Текущий локальный HTTP разрешён только для изолированного стенда с синтетическими данными; удалённая демонстрация требует TLS.
diff --git a/evidence/scale/baseline.jsonl b/evidence/scale/baseline.jsonl
new file mode 100644
index 00000000..f44615e9
--- /dev/null
+++ b/evidence/scale/baseline.jsonl
@@ -0,0 +1,54 @@
+{"mode": "fixed", "replicas": 1, "rate": 250, "repeat": 1, "duration": 8, "warmup": 2, "dataset_messages": 1000, "api_cpu_each": 1.0, "api_memory_mib_each": 512, "pool_max_size": 8, "offered_rate": 250, "scheduled_requests": 2000, "successful_requests": 2000, "failed_requests": 0, "error_ratio": 0.0, "successful_rps": 249.95, "elapsed_seconds": 8.002, "successful_latency_ms": {"p50": 2.831, "p95": 37.915, "p99": 211.151}, "errors_by_type": {}, "queue_dropped": 0, "client_cpu_seconds": 0.384, "docker": {"api": {"avg_cpu_percent": 39.223, "max_memory_mib": 46.79}, "postgres": {"avg_cpu_percent": 6.993, "max_memory_mib": 53.62}, "lb": {"avg_cpu_percent": 2.253, "max_memory_mib": 41.58}}, "pool": {"observed_replicas": 1, "requests_queued": 75, "wait_ms": 600, "max_size": 8}, "postgres_delta": {"commits": 4018, "rollbacks": 0, "blocks_read": 0, "blocks_hit": 65527, "temp_files": 0, "deadlocks": 0, "messages": 0}, "database_bytes": 25767059, "database_messages": 30969}
+{"mode": "fixed", "replicas": 1, "rate": 250, "repeat": 2, "duration": 8, "warmup": 2, "dataset_messages": 1000, "api_cpu_each": 1.0, "api_memory_mib_each": 512, "pool_max_size": 8, "offered_rate": 250, "scheduled_requests": 2000, "successful_requests": 2000, "failed_requests": 0, "error_ratio": 0.0, "successful_rps": 249.951, "elapsed_seconds": 8.002, "successful_latency_ms": {"p50": 2.787, "p95": 3.429, "p99": 5.526}, "errors_by_type": {}, "queue_dropped": 0, "client_cpu_seconds": 0.355, "docker": {"api": {"avg_cpu_percent": 26.653, "max_memory_mib": 47.33}, "postgres": {"avg_cpu_percent": 5.803, "max_memory_mib": 49.23}, "lb": {"avg_cpu_percent": 1.86, "max_memory_mib": 41.47}}, "pool": {"observed_replicas": 1, "requests_queued": 0, "wait_ms": 0, "max_size": 8}, "postgres_delta": {"commits": 4016, "rollbacks": 0, "blocks_read": 0, "blocks_hit": 61557, "temp_files": 0, "deadlocks": 0, "messages": 0}, "database_bytes": 25767059, "database_messages": 30969}
+{"mode": "fixed", "replicas": 1, "rate": 250, "repeat": 3, "duration": 8, "warmup": 2, "dataset_messages": 1000, "api_cpu_each": 1.0, "api_memory_mib_each": 512, "pool_max_size": 8, "offered_rate": 250, "scheduled_requests": 2000, "successful_requests": 2000, "failed_requests": 0, "error_ratio": 0.0, "successful_rps": 249.971, "elapsed_seconds": 8.001, "successful_latency_ms": {"p50": 2.772, "p95": 3.703, "p99": 5.882}, "errors_by_type": {}, "queue_dropped": 0, "client_cpu_seconds": 0.353, "docker": {"api": {"avg_cpu_percent": 29.56, "max_memory_mib": 45.91}, "postgres": {"avg_cpu_percent": 5.72, "max_memory_mib": 49.26}, "lb": {"avg_cpu_percent": 1.75, "max_memory_mib": 41.62}}, "pool": {"observed_replicas": 1, "requests_queued": 0, "wait_ms": 0, "max_size": 8}, "postgres_delta": {"commits": 4022, "rollbacks": 0, "blocks_read": 0, "blocks_hit": 61093, "temp_files": 0, "deadlocks": 0, "messages": 0}, "database_bytes": 25767059, "database_messages": 30969}
+{"mode": "fixed", "replicas": 1, "rate": 750, "repeat": 1, "duration": 8, "warmup": 2, "dataset_messages": 1000, "api_cpu_each": 1.0, "api_memory_mib_each": 512, "pool_max_size": 8, "offered_rate": 750, "scheduled_requests": 6000, "successful_requests": 6000, "failed_requests": 0, "error_ratio": 0.0, "successful_rps": 746.986, "elapsed_seconds": 8.032, "successful_latency_ms": {"p50": 4.909, "p95": 121.369, "p99": 137.788}, "errors_by_type": {}, "queue_dropped": 0, "client_cpu_seconds": 0.707, "docker": {"api": {"avg_cpu_percent": 88.997, "max_memory_mib": 51.91}, "postgres": {"avg_cpu_percent": 12.417, "max_memory_mib": 49.31}, "lb": {"avg_cpu_percent": 4.683, "max_memory_mib": 41.64}}, "pool": {"observed_replicas": 1, "requests_queued": 4570, "wait_ms": 55096, "max_size": 8}, "postgres_delta": {"commits": 12605, "rollbacks": 0, "blocks_read": 0, "blocks_hit": 191767, "temp_files": 0, "deadlocks": 0, "messages": 0}, "database_bytes": 25767059, "database_messages": 30969}
+{"mode": "fixed", "replicas": 1, "rate": 750, "repeat": 2, "duration": 8, "warmup": 2, "dataset_messages": 1000, "api_cpu_each": 1.0, "api_memory_mib_each": 512, "pool_max_size": 8, "offered_rate": 750, "scheduled_requests": 6000, "successful_requests": 6000, "failed_requests": 0, "error_ratio": 0.0, "successful_rps": 749.783, "elapsed_seconds": 8.002, "successful_latency_ms": {"p50": 1.679, "p95": 85.479, "p99": 125.898}, "errors_by_type": {}, "queue_dropped": 0, "client_cpu_seconds": 0.636, "docker": {"api": {"avg_cpu_percent": 60.08, "max_memory_mib": 52.52}, "postgres": {"avg_cpu_percent": 10.823, "max_memory_mib": 49.39}, "lb": {"avg_cpu_percent": 3.9, "max_memory_mib": 41.6}}, "pool": {"observed_replicas": 1, "requests_queued": 1631, "wait_ms": 17243, "max_size": 8}, "postgres_delta": {"commits": 11926, "rollbacks": 0, "blocks_read": 0, "blocks_hit": 181241, "temp_files": 0, "deadlocks": 0, "messages": 0}, "database_bytes": 25767059, "database_messages": 30969}
+{"mode": "fixed", "replicas": 1, "rate": 750, "repeat": 3, "duration": 8, "warmup": 2, "dataset_messages": 1000, "api_cpu_each": 1.0, "api_memory_mib_each": 512, "pool_max_size": 8, "offered_rate": 750, "scheduled_requests": 6000, "successful_requests": 6000, "failed_requests": 0, "error_ratio": 0.0, "successful_rps": 746.631, "elapsed_seconds": 8.036, "successful_latency_ms": {"p50": 3.533, "p95": 103.334, "p99": 138.015}, "errors_by_type": {}, "queue_dropped": 0, "client_cpu_seconds": 0.695, "docker": {"api": {"avg_cpu_percent": 84.993, "max_memory_mib": 52.01}, "postgres": {"avg_cpu_percent": 12.933, "max_memory_mib": 49.41}, "lb": {"avg_cpu_percent": 4.407, "max_memory_mib": 41.71}}, "pool": {"observed_replicas": 1, "requests_queued": 4243, "wait_ms": 51465, "max_size": 8}, "postgres_delta": {"commits": 11896, "rollbacks": 0, "blocks_read": 0, "blocks_hit": 178481, "temp_files": 0, "deadlocks": 0, "messages": 0}, "database_bytes": 25767059, "database_messages": 30969}
+{"mode": "fixed", "replicas": 1, "rate": 1500, "repeat": 1, "duration": 8, "warmup": 2, "dataset_messages": 1000, "api_cpu_each": 1.0, "api_memory_mib_each": 512, "pool_max_size": 8, "offered_rate": 1500, "scheduled_requests": 12000, "successful_requests": 6890, "failed_requests": 5110, "error_ratio": 0.425833, "successful_rps": 823.026, "elapsed_seconds": 8.372, "successful_latency_ms": {"p50": 378.91, "p95": 447.591, "p99": 470.365}, "errors_by_type": {}, "queue_dropped": 5110, "client_cpu_seconds": 0.72, "docker": {"api": {"avg_cpu_percent": 97.463, "max_memory_mib": 64.79}, "postgres": {"avg_cpu_percent": 12.573, "max_memory_mib": 49.45}, "lb": {"avg_cpu_percent": 4.42, "max_memory_mib": 41.61}}, "pool": {"observed_replicas": 1, "requests_queued": 12335, "wait_ms": 188633, "max_size": 8}, "postgres_delta": {"commits": 14324, "rollbacks": 0, "blocks_read": 0, "blocks_hit": 215405, "temp_files": 0, "deadlocks": 0, "messages": 0}, "database_bytes": 25767059, "database_messages": 30969}
+{"mode": "fixed", "replicas": 1, "rate": 1500, "repeat": 2, "duration": 8, "warmup": 2, "dataset_messages": 1000, "api_cpu_each": 1.0, "api_memory_mib_each": 512, "pool_max_size": 8, "offered_rate": 1500, "scheduled_requests": 12000, "successful_requests": 6858, "failed_requests": 5142, "error_ratio": 0.4285, "successful_rps": 780.299, "elapsed_seconds": 8.789, "successful_latency_ms": {"p50": 372.063, "p95": 466.704, "p99": 1071.218}, "errors_by_type": {}, "queue_dropped": 5142, "client_cpu_seconds": 0.757, "docker": {"api": {"avg_cpu_percent": 96.24, "max_memory_mib": 53.53}, "postgres": {"avg_cpu_percent": 13.42, "max_memory_mib": 49.68}, "lb": {"avg_cpu_percent": 4.423, "max_memory_mib": 41.64}}, "pool": {"observed_replicas": 1, "requests_queued": 11597, "wait_ms": 182073, "max_size": 8}, "postgres_delta": {"commits": 14394, "rollbacks": 0, "blocks_read": 0, "blocks_hit": 216827, "temp_files": 0, "deadlocks": 0, "messages": 0}, "database_bytes": 25767059, "database_messages": 30969}
+{"mode": "fixed", "replicas": 1, "rate": 1500, "repeat": 3, "duration": 8, "warmup": 2, "dataset_messages": 1000, "api_cpu_each": 1.0, "api_memory_mib_each": 512, "pool_max_size": 8, "offered_rate": 1500, "scheduled_requests": 12000, "successful_requests": 6445, "failed_requests": 5555, "error_ratio": 0.462917, "successful_rps": 766.917, "elapsed_seconds": 8.404, "successful_latency_ms": {"p50": 382.659, "p95": 600.998, "p99": 725.297}, "errors_by_type": {}, "queue_dropped": 5555, "client_cpu_seconds": 0.75, "docker": {"api": {"avg_cpu_percent": 96.137, "max_memory_mib": 53.47}, "postgres": {"avg_cpu_percent": 15.257, "max_memory_mib": 49.48}, "lb": {"avg_cpu_percent": 4.627, "max_memory_mib": 41.67}}, "pool": {"observed_replicas": 1, "requests_queued": 11447, "wait_ms": 180076, "max_size": 8}, "postgres_delta": {"commits": 13337, "rollbacks": 0, "blocks_read": 0, "blocks_hit": 200363, "temp_files": 0, "deadlocks": 0, "messages": 0}, "database_bytes": 25767059, "database_messages": 30969}
+{"mode": "fixed", "replicas": 2, "rate": 250, "repeat": 1, "duration": 8, "warmup": 2, "dataset_messages": 1000, "api_cpu_each": 0.5, "api_memory_mib_each": 256, "pool_max_size": 8, "offered_rate": 250, "scheduled_requests": 2000, "successful_requests": 2000, "failed_requests": 0, "error_ratio": 0.0, "successful_rps": 249.96, "elapsed_seconds": 8.001, "successful_latency_ms": {"p50": 2.769, "p95": 13.026, "p99": 51.133}, "errors_by_type": {}, "queue_dropped": 0, "client_cpu_seconds": 0.342, "docker": {"api": {"avg_cpu_percent": 38.633, "max_memory_mib": 85.02}, "postgres": {"avg_cpu_percent": 5.927, "max_memory_mib": 62.68}, "lb": {"avg_cpu_percent": 1.71, "max_memory_mib": 41.82}}, "pool": {"observed_replicas": 2, "requests_queued": 5, "wait_ms": 4, "max_size": 8}, "postgres_delta": {"commits": 4198, "rollbacks": 0, "blocks_read": 0, "blocks_hit": 71471, "temp_files": 0, "deadlocks": 0, "messages": 0}, "database_bytes": 25767059, "database_messages": 30969}
+{"mode": "fixed", "replicas": 2, "rate": 250, "repeat": 2, "duration": 8, "warmup": 2, "dataset_messages": 1000, "api_cpu_each": 0.5, "api_memory_mib_each": 256, "pool_max_size": 8, "offered_rate": 250, "scheduled_requests": 2000, "successful_requests": 2000, "failed_requests": 0, "error_ratio": 0.0, "successful_rps": 249.959, "elapsed_seconds": 8.001, "successful_latency_ms": {"p50": 2.875, "p95": 33.197, "p99": 100.083}, "errors_by_type": {}, "queue_dropped": 0, "client_cpu_seconds": 0.356, "docker": {"api": {"avg_cpu_percent": 43.583, "max_memory_mib": 112.32}, "postgres": {"avg_cpu_percent": 5.957, "max_memory_mib": 64.63}, "lb": {"avg_cpu_percent": 1.9, "max_memory_mib": 41.56}}, "pool": {"observed_replicas": 2, "requests_queued": 3, "wait_ms": 14, "max_size": 8}, "postgres_delta": {"commits": 4072, "rollbacks": 0, "blocks_read": 0, "blocks_hit": 61681, "temp_files": 0, "deadlocks": 0, "messages": 0}, "database_bytes": 25767059, "database_messages": 30969}
+{"mode": "fixed", "replicas": 2, "rate": 250, "repeat": 3, "duration": 8, "warmup": 2, "dataset_messages": 1000, "api_cpu_each": 0.5, "api_memory_mib_each": 256, "pool_max_size": 8, "offered_rate": 250, "scheduled_requests": 2000, "successful_requests": 2000, "failed_requests": 0, "error_ratio": 0.0, "successful_rps": 249.937, "elapsed_seconds": 8.002, "successful_latency_ms": {"p50": 2.873, "p95": 25.949, "p99": 138.238}, "errors_by_type": {}, "queue_dropped": 0, "client_cpu_seconds": 0.372, "docker": {"api": {"avg_cpu_percent": 42.983, "max_memory_mib": 102.69}, "postgres": {"avg_cpu_percent": 6.063, "max_memory_mib": 64.66}, "lb": {"avg_cpu_percent": 1.753, "max_memory_mib": 41.64}}, "pool": {"observed_replicas": 2, "requests_queued": 31, "wait_ms": 78, "max_size": 8}, "postgres_delta": {"commits": 3777, "rollbacks": 0, "blocks_read": 0, "blocks_hit": 57121, "temp_files": 0, "deadlocks": 0, "messages": 0}, "database_bytes": 25767059, "database_messages": 30969}
+{"mode": "fixed", "replicas": 2, "rate": 750, "repeat": 1, "duration": 8, "warmup": 2, "dataset_messages": 1000, "api_cpu_each": 0.5, "api_memory_mib_each": 256, "pool_max_size": 8, "offered_rate": 750, "scheduled_requests": 6000, "successful_requests": 6000, "failed_requests": 0, "error_ratio": 0.0, "successful_rps": 749.804, "elapsed_seconds": 8.002, "successful_latency_ms": {"p50": 1.755, "p95": 108.359, "p99": 175.477}, "errors_by_type": {}, "queue_dropped": 0, "client_cpu_seconds": 0.623, "docker": {"api": {"avg_cpu_percent": 69.88, "max_memory_mib": 108.39}, "postgres": {"avg_cpu_percent": 12.08, "max_memory_mib": 64.74}, "lb": {"avg_cpu_percent": 3.88, "max_memory_mib": 41.91}}, "pool": {"observed_replicas": 2, "requests_queued": 1294, "wait_ms": 7867, "max_size": 8}, "postgres_delta": {"commits": 12232, "rollbacks": 0, "blocks_read": 0, "blocks_hit": 184771, "temp_files": 0, "deadlocks": 0, "messages": 0}, "database_bytes": 25767059, "database_messages": 30969}
+{"mode": "fixed", "replicas": 2, "rate": 750, "repeat": 2, "duration": 8, "warmup": 2, "dataset_messages": 1000, "api_cpu_each": 0.5, "api_memory_mib_each": 256, "pool_max_size": 8, "offered_rate": 750, "scheduled_requests": 6000, "successful_requests": 6000, "failed_requests": 0, "error_ratio": 0.0, "successful_rps": 749.772, "elapsed_seconds": 8.002, "successful_latency_ms": {"p50": 1.731, "p95": 100.322, "p99": 162.856}, "errors_by_type": {}, "queue_dropped": 0, "client_cpu_seconds": 0.61, "docker": {"api": {"avg_cpu_percent": 70.053, "max_memory_mib": 96.87}, "postgres": {"avg_cpu_percent": 12.173, "max_memory_mib": 65.0}, "lb": {"avg_cpu_percent": 3.93, "max_memory_mib": 41.97}}, "pool": {"observed_replicas": 2, "requests_queued": 1118, "wait_ms": 5573, "max_size": 8}, "postgres_delta": {"commits": 12156, "rollbacks": 0, "blocks_read": 0, "blocks_hit": 183345, "temp_files": 0, "deadlocks": 0, "messages": 0}, "database_bytes": 25767059, "database_messages": 30969}
+{"mode": "fixed", "replicas": 2, "rate": 750, "repeat": 3, "duration": 8, "warmup": 2, "dataset_messages": 1000, "api_cpu_each": 0.5, "api_memory_mib_each": 256, "pool_max_size": 8, "offered_rate": 750, "scheduled_requests": 6000, "successful_requests": 6000, "failed_requests": 0, "error_ratio": 0.0, "successful_rps": 749.734, "elapsed_seconds": 8.003, "successful_latency_ms": {"p50": 1.816, "p95": 178.504, "p99": 224.107}, "errors_by_type": {}, "queue_dropped": 0, "client_cpu_seconds": 0.627, "docker": {"api": {"avg_cpu_percent": 68.573, "max_memory_mib": 97.41}, "postgres": {"avg_cpu_percent": 12.37, "max_memory_mib": 64.76}, "lb": {"avg_cpu_percent": 3.767, "max_memory_mib": 41.91}}, "pool": {"observed_replicas": 2, "requests_queued": 1324, "wait_ms": 11493, "max_size": 8}, "postgres_delta": {"commits": 12142, "rollbacks": 0, "blocks_read": 0, "blocks_hit": 183157, "temp_files": 0, "deadlocks": 0, "messages": 0}, "database_bytes": 25767059, "database_messages": 30969}
+{"mode": "fixed", "replicas": 2, "rate": 1500, "repeat": 1, "duration": 8, "warmup": 2, "dataset_messages": 1000, "api_cpu_each": 0.5, "api_memory_mib_each": 256, "pool_max_size": 8, "offered_rate": 1500, "scheduled_requests": 12000, "successful_requests": 7863, "failed_requests": 4137, "error_ratio": 0.34475, "successful_rps": 941.177, "elapsed_seconds": 8.354, "successful_latency_ms": {"p50": 319.914, "p95": 473.58, "p99": 530.577}, "errors_by_type": {}, "queue_dropped": 4137, "client_cpu_seconds": 0.785, "docker": {"api": {"avg_cpu_percent": 98.37, "max_memory_mib": 109.51}, "postgres": {"avg_cpu_percent": 14.65, "max_memory_mib": 64.77}, "lb": {"avg_cpu_percent": 5.09, "max_memory_mib": 42.0}}, "pool": {"observed_replicas": 2, "requests_queued": 6822, "wait_ms": 77320, "max_size": 8}, "postgres_delta": {"commits": 14978, "rollbacks": 0, "blocks_read": 0, "blocks_hit": 225169, "temp_files": 0, "deadlocks": 0, "messages": 0}, "database_bytes": 25767059, "database_messages": 30969}
+{"mode": "fixed", "replicas": 2, "rate": 1500, "repeat": 2, "duration": 8, "warmup": 2, "dataset_messages": 1000, "api_cpu_each": 0.5, "api_memory_mib_each": 256, "pool_max_size": 8, "offered_rate": 1500, "scheduled_requests": 12000, "successful_requests": 8293, "failed_requests": 3707, "error_ratio": 0.308917, "successful_rps": 993.827, "elapsed_seconds": 8.345, "successful_latency_ms": {"p50": 297.841, "p95": 454.544, "p99": 505.25}, "errors_by_type": {}, "queue_dropped": 3707, "client_cpu_seconds": 0.762, "docker": {"api": {"avg_cpu_percent": 100.243, "max_memory_mib": 113.26}, "postgres": {"avg_cpu_percent": 14.147, "max_memory_mib": 65.02}, "lb": {"avg_cpu_percent": 4.87, "max_memory_mib": 42.0}}, "pool": {"observed_replicas": 2, "requests_queued": 5427, "wait_ms": 49740, "max_size": 8}, "postgres_delta": {"commits": 15964, "rollbacks": 0, "blocks_read": 0, "blocks_hit": 240127, "temp_files": 0, "deadlocks": 0, "messages": 0}, "database_bytes": 25767059, "database_messages": 30969}
+{"mode": "fixed", "replicas": 2, "rate": 1500, "repeat": 3, "duration": 8, "warmup": 2, "dataset_messages": 1000, "api_cpu_each": 0.5, "api_memory_mib_each": 256, "pool_max_size": 8, "offered_rate": 1500, "scheduled_requests": 12000, "successful_requests": 7934, "failed_requests": 4066, "error_ratio": 0.338833, "successful_rps": 956.631, "elapsed_seconds": 8.294, "successful_latency_ms": {"p50": 314.99, "p95": 467.748, "p99": 516.41}, "errors_by_type": {}, "queue_dropped": 4066, "client_cpu_seconds": 0.757, "docker": {"api": {"avg_cpu_percent": 98.82, "max_memory_mib": 122.91}, "postgres": {"avg_cpu_percent": 13.203, "max_memory_mib": 64.77}, "lb": {"avg_cpu_percent": 4.807, "max_memory_mib": 41.96}}, "pool": {"observed_replicas": 2, "requests_queued": 5707, "wait_ms": 72376, "max_size": 8}, "postgres_delta": {"commits": 16857, "rollbacks": 0, "blocks_read": 0, "blocks_hit": 253657, "temp_files": 0, "deadlocks": 0, "messages": 0}, "database_bytes": 25767059, "database_messages": 30969}
+{"mode": "fixed", "replicas": 4, "rate": 250, "repeat": 1, "duration": 8, "warmup": 2, "dataset_messages": 1000, "api_cpu_each": 0.25, "api_memory_mib_each": 128, "pool_max_size": 8, "offered_rate": 250, "scheduled_requests": 2000, "successful_requests": 2000, "failed_requests": 0, "error_ratio": 0.0, "successful_rps": 249.958, "elapsed_seconds": 8.001, "successful_latency_ms": {"p50": 3.016, "p95": 61.385, "p99": 81.718}, "errors_by_type": {}, "queue_dropped": 0, "client_cpu_seconds": 0.313, "docker": {"api": {"avg_cpu_percent": 64.537, "max_memory_mib": 205.98}, "postgres": {"avg_cpu_percent": 5.16, "max_memory_mib": 79.56}, "lb": {"avg_cpu_percent": 1.72, "max_memory_mib": 41.74}}, "pool": {"observed_replicas": 4, "requests_queued": 4, "wait_ms": 0, "max_size": 6}, "postgres_delta": {"commits": 4223, "rollbacks": 0, "blocks_read": 0, "blocks_hit": 76658, "temp_files": 0, "deadlocks": 0, "messages": 0}, "database_bytes": 25767059, "database_messages": 30969}
+{"mode": "fixed", "replicas": 4, "rate": 250, "repeat": 2, "duration": 8, "warmup": 2, "dataset_messages": 1000, "api_cpu_each": 0.25, "api_memory_mib_each": 128, "pool_max_size": 8, "offered_rate": 250, "scheduled_requests": 2000, "successful_requests": 2000, "failed_requests": 0, "error_ratio": 0.0, "successful_rps": 249.944, "elapsed_seconds": 8.002, "successful_latency_ms": {"p50": 2.944, "p95": 59.768, "p99": 80.401}, "errors_by_type": {}, "queue_dropped": 0, "client_cpu_seconds": 0.323, "docker": {"api": {"avg_cpu_percent": 60.467, "max_memory_mib": 205.43}, "postgres": {"avg_cpu_percent": 5.073, "max_memory_mib": 79.71}, "lb": {"avg_cpu_percent": 1.657, "max_memory_mib": 41.74}}, "pool": {"observed_replicas": 4, "requests_queued": 0, "wait_ms": 0, "max_size": 6}, "postgres_delta": {"commits": 4020, "rollbacks": 0, "blocks_read": 0, "blocks_hit": 60455, "temp_files": 0, "deadlocks": 0, "messages": 0}, "database_bytes": 25767059, "database_messages": 30969}
+{"mode": "fixed", "replicas": 4, "rate": 250, "repeat": 3, "duration": 8, "warmup": 2, "dataset_messages": 1000, "api_cpu_each": 0.25, "api_memory_mib_each": 128, "pool_max_size": 8, "offered_rate": 250, "scheduled_requests": 2000, "successful_requests": 2000, "failed_requests": 0, "error_ratio": 0.0, "successful_rps": 249.971, "elapsed_seconds": 8.001, "successful_latency_ms": {"p50": 3.021, "p95": 60.482, "p99": 79.776}, "errors_by_type": {}, "queue_dropped": 0, "client_cpu_seconds": 0.316, "docker": {"api": {"avg_cpu_percent": 51.723, "max_memory_mib": 203.98}, "postgres": {"avg_cpu_percent": 5.303, "max_memory_mib": 79.81}, "lb": {"avg_cpu_percent": 1.68, "max_memory_mib": 41.55}}, "pool": {"observed_replicas": 4, "requests_queued": 0, "wait_ms": 0, "max_size": 6}, "postgres_delta": {"commits": 4101, "rollbacks": 0, "blocks_read": 0, "blocks_hit": 61639, "temp_files": 0, "deadlocks": 0, "messages": 0}, "database_bytes": 25767059, "database_messages": 30969}
+{"mode": "fixed", "replicas": 4, "rate": 750, "repeat": 1, "duration": 8, "warmup": 2, "dataset_messages": 1000, "api_cpu_each": 0.25, "api_memory_mib_each": 128, "pool_max_size": 8, "offered_rate": 750, "scheduled_requests": 6000, "successful_requests": 6000, "failed_requests": 0, "error_ratio": 0.0, "successful_rps": 744.222, "elapsed_seconds": 8.062, "successful_latency_ms": {"p50": 3.319, "p95": 308.301, "p99": 415.861}, "errors_by_type": {}, "queue_dropped": 0, "client_cpu_seconds": 0.613, "docker": {"api": {"avg_cpu_percent": 83.097, "max_memory_mib": 205.59}, "postgres": {"avg_cpu_percent": 13.187, "max_memory_mib": 95.02}, "lb": {"avg_cpu_percent": 3.927, "max_memory_mib": 42.05}}, "pool": {"observed_replicas": 4, "requests_queued": 952, "wait_ms": 12771, "max_size": 8}, "postgres_delta": {"commits": 11710, "rollbacks": 0, "blocks_read": 0, "blocks_hit": 181824, "temp_files": 0, "deadlocks": 0, "messages": 0}, "database_bytes": 25767059, "database_messages": 30969}
+{"mode": "fixed", "replicas": 4, "rate": 750, "repeat": 2, "duration": 8, "warmup": 2, "dataset_messages": 1000, "api_cpu_each": 0.25, "api_memory_mib_each": 128, "pool_max_size": 8, "offered_rate": 750, "scheduled_requests": 6000, "successful_requests": 6000, "failed_requests": 0, "error_ratio": 0.0, "successful_rps": 713.504, "elapsed_seconds": 8.409, "successful_latency_ms": {"p50": 126.994, "p95": 414.136, "p99": 480.195}, "errors_by_type": {}, "queue_dropped": 0, "client_cpu_seconds": 0.582, "docker": {"api": {"avg_cpu_percent": 92.41, "max_memory_mib": 207.79}, "postgres": {"avg_cpu_percent": 12.73, "max_memory_mib": 95.05}, "lb": {"avg_cpu_percent": 3.723, "max_memory_mib": 42.23}}, "pool": {"observed_replicas": 4, "requests_queued": 2167, "wait_ms": 26550, "max_size": 8}, "postgres_delta": {"commits": 12191, "rollbacks": 0, "blocks_read": 0, "blocks_hit": 183838, "temp_files": 0, "deadlocks": 0, "messages": 0}, "database_bytes": 25767059, "database_messages": 30969}
+{"mode": "fixed", "replicas": 4, "rate": 750, "repeat": 3, "duration": 8, "warmup": 2, "dataset_messages": 1000, "api_cpu_each": 0.25, "api_memory_mib_each": 128, "pool_max_size": 8, "offered_rate": 750, "scheduled_requests": 6000, "successful_requests": 6000, "failed_requests": 0, "error_ratio": 0.0, "successful_rps": 732.997, "elapsed_seconds": 8.186, "successful_latency_ms": {"p50": 110.445, "p95": 397.681, "p99": 486.245}, "errors_by_type": {}, "queue_dropped": 0, "client_cpu_seconds": 0.585, "docker": {"api": {"avg_cpu_percent": 93.953, "max_memory_mib": 217.25}, "postgres": {"avg_cpu_percent": 12.887, "max_memory_mib": 95.12}, "lb": {"avg_cpu_percent": 3.673, "max_memory_mib": 42.28}}, "pool": {"observed_replicas": 4, "requests_queued": 1901, "wait_ms": 24258, "max_size": 8}, "postgres_delta": {"commits": 11844, "rollbacks": 0, "blocks_read": 0, "blocks_hit": 178211, "temp_files": 0, "deadlocks": 0, "messages": 0}, "database_bytes": 25767059, "database_messages": 30969}
+{"mode": "fixed", "replicas": 4, "rate": 1500, "repeat": 1, "duration": 8, "warmup": 2, "dataset_messages": 1000, "api_cpu_each": 0.25, "api_memory_mib_each": 128, "pool_max_size": 8, "offered_rate": 1500, "scheduled_requests": 12000, "successful_requests": 6950, "failed_requests": 5050, "error_ratio": 0.420833, "successful_rps": 811.023, "elapsed_seconds": 8.569, "successful_latency_ms": {"p50": 315.472, "p95": 633.293, "p99": 771.213}, "errors_by_type": {}, "queue_dropped": 5050, "client_cpu_seconds": 0.698, "docker": {"api": {"avg_cpu_percent": 98.617, "max_memory_mib": 216.33}, "postgres": {"avg_cpu_percent": 15.563, "max_memory_mib": 95.12}, "lb": {"avg_cpu_percent": 4.53, "max_memory_mib": 42.35}}, "pool": {"observed_replicas": 4, "requests_queued": 3048, "wait_ms": 32524, "max_size": 8}, "postgres_delta": {"commits": 14083, "rollbacks": 0, "blocks_read": 0, "blocks_hit": 211553, "temp_files": 0, "deadlocks": 0, "messages": 0}, "database_bytes": 25767059, "database_messages": 30969}
+{"mode": "fixed", "replicas": 4, "rate": 1500, "repeat": 2, "duration": 8, "warmup": 2, "dataset_messages": 1000, "api_cpu_each": 0.25, "api_memory_mib_each": 128, "pool_max_size": 8, "offered_rate": 1500, "scheduled_requests": 12000, "successful_requests": 5902, "failed_requests": 6098, "error_ratio": 0.508167, "successful_rps": 693.857, "elapsed_seconds": 8.506, "successful_latency_ms": {"p50": 413.811, "p95": 769.612, "p99": 855.228}, "errors_by_type": {}, "queue_dropped": 6098, "client_cpu_seconds": 0.674, "docker": {"api": {"avg_cpu_percent": 80.183, "max_memory_mib": 204.21}, "postgres": {"avg_cpu_percent": 11.61, "max_memory_mib": 95.16}, "lb": {"avg_cpu_percent": 3.993, "max_memory_mib": 42.35}}, "pool": {"observed_replicas": 4, "requests_queued": 2088, "wait_ms": 33271, "max_size": 8}, "postgres_delta": {"commits": 12393, "rollbacks": 0, "blocks_read": 0, "blocks_hit": 186123, "temp_files": 0, "deadlocks": 0, "messages": 0}, "database_bytes": 25767059, "database_messages": 30969}
+{"mode": "fixed", "replicas": 4, "rate": 1500, "repeat": 3, "duration": 8, "warmup": 2, "dataset_messages": 1000, "api_cpu_each": 0.25, "api_memory_mib_each": 128, "pool_max_size": 8, "offered_rate": 1500, "scheduled_requests": 12000, "successful_requests": 5756, "failed_requests": 6244, "error_ratio": 0.520333, "successful_rps": 681.037, "elapsed_seconds": 8.452, "successful_latency_ms": {"p50": 420.287, "p95": 747.7, "p99": 851.226}, "errors_by_type": {}, "queue_dropped": 6244, "client_cpu_seconds": 0.604, "docker": {"api": {"avg_cpu_percent": 87.06, "max_memory_mib": 194.64}, "postgres": {"avg_cpu_percent": 13.59, "max_memory_mib": 95.16}, "lb": {"avg_cpu_percent": 3.353, "max_memory_mib": 42.36}}, "pool": {"observed_replicas": 4, "requests_queued": 2203, "wait_ms": 38980, "max_size": 8}, "postgres_delta": {"commits": 11550, "rollbacks": 0, "blocks_read": 0, "blocks_hit": 173489, "temp_files": 0, "deadlocks": 0, "messages": 0}, "database_bytes": 25767059, "database_messages": 30969}
+{"mode": "growth", "replicas": 1, "rate": 250, "repeat": 1, "duration": 8, "warmup": 2, "dataset_messages": 1000, "api_cpu_each": 0.5, "api_memory_mib_each": 256, "pool_max_size": 8, "offered_rate": 250, "scheduled_requests": 2000, "successful_requests": 2000, "failed_requests": 0, "error_ratio": 0.0, "successful_rps": 249.946, "elapsed_seconds": 8.002, "successful_latency_ms": {"p50": 2.572, "p95": 206.839, "p99": 277.387}, "errors_by_type": {}, "queue_dropped": 0, "client_cpu_seconds": 0.316, "docker": {"api": {"avg_cpu_percent": 33.947, "max_memory_mib": 48.01}, "postgres": {"avg_cpu_percent": 6.093, "max_memory_mib": 49.68}, "lb": {"avg_cpu_percent": 1.697, "max_memory_mib": 41.68}}, "pool": {"observed_replicas": 1, "requests_queued": 463, "wait_ms": 4901, "max_size": 8}, "postgres_delta": {"commits": 4114, "rollbacks": 0, "blocks_read": 0, "blocks_hit": 66849, "temp_files": 0, "deadlocks": 0, "messages": 0}, "database_bytes": 25767059, "database_messages": 30969}
+{"mode": "growth", "replicas": 1, "rate": 250, "repeat": 2, "duration": 8, "warmup": 2, "dataset_messages": 1000, "api_cpu_each": 0.5, "api_memory_mib_each": 256, "pool_max_size": 8, "offered_rate": 250, "scheduled_requests": 2000, "successful_requests": 2000, "failed_requests": 0, "error_ratio": 0.0, "successful_rps": 249.97, "elapsed_seconds": 8.001, "successful_latency_ms": {"p50": 2.365, "p95": 39.928, "p99": 72.726}, "errors_by_type": {}, "queue_dropped": 0, "client_cpu_seconds": 0.262, "docker": {"api": {"avg_cpu_percent": 28.06, "max_memory_mib": 48.11}, "postgres": {"avg_cpu_percent": 5.183, "max_memory_mib": 49.45}, "lb": {"avg_cpu_percent": 1.587, "max_memory_mib": 41.71}}, "pool": {"observed_replicas": 1, "requests_queued": 135, "wait_ms": 196, "max_size": 8}, "postgres_delta": {"commits": 4078, "rollbacks": 0, "blocks_read": 0, "blocks_hit": 61300, "temp_files": 0, "deadlocks": 0, "messages": 0}, "database_bytes": 25767059, "database_messages": 30969}
+{"mode": "growth", "replicas": 1, "rate": 250, "repeat": 3, "duration": 8, "warmup": 2, "dataset_messages": 1000, "api_cpu_each": 0.5, "api_memory_mib_each": 256, "pool_max_size": 8, "offered_rate": 250, "scheduled_requests": 2000, "successful_requests": 2000, "failed_requests": 0, "error_ratio": 0.0, "successful_rps": 249.977, "elapsed_seconds": 8.001, "successful_latency_ms": {"p50": 2.684, "p95": 59.465, "p99": 80.414}, "errors_by_type": {}, "queue_dropped": 0, "client_cpu_seconds": 0.345, "docker": {"api": {"avg_cpu_percent": 34.943, "max_memory_mib": 47.95}, "postgres": {"avg_cpu_percent": 6.29, "max_memory_mib": 49.47}, "lb": {"avg_cpu_percent": 2.08, "max_memory_mib": 41.66}}, "pool": {"observed_replicas": 1, "requests_queued": 210, "wait_ms": 410, "max_size": 8}, "postgres_delta": {"commits": 3956, "rollbacks": 0, "blocks_read": 0, "blocks_hit": 60059, "temp_files": 0, "deadlocks": 0, "messages": 0}, "database_bytes": 25767059, "database_messages": 30969}
+{"mode": "growth", "replicas": 1, "rate": 750, "repeat": 1, "duration": 8, "warmup": 2, "dataset_messages": 1000, "api_cpu_each": 0.5, "api_memory_mib_each": 256, "pool_max_size": 8, "offered_rate": 750, "scheduled_requests": 6000, "successful_requests": 3164, "failed_requests": 2836, "error_ratio": 0.472667, "successful_rps": 362.33, "elapsed_seconds": 8.732, "successful_latency_ms": {"p50": 823.553, "p95": 1071.351, "p99": 1182.579}, "errors_by_type": {}, "queue_dropped": 2836, "client_cpu_seconds": 0.398, "docker": {"api": {"avg_cpu_percent": 50.06, "max_memory_mib": 64.7}, "postgres": {"avg_cpu_percent": 6.51, "max_memory_mib": 49.75}, "lb": {"avg_cpu_percent": 2.317, "max_memory_mib": 41.69}}, "pool": {"observed_replicas": 1, "requests_queued": 5298, "wait_ms": 184862, "max_size": 8}, "postgres_delta": {"commits": 5828, "rollbacks": 0, "blocks_read": 0, "blocks_hit": 88215, "temp_files": 0, "deadlocks": 0, "messages": 0}, "database_bytes": 25767059, "database_messages": 30969}
+{"mode": "growth", "replicas": 1, "rate": 750, "repeat": 2, "duration": 8, "warmup": 2, "dataset_messages": 1000, "api_cpu_each": 0.5, "api_memory_mib_each": 256, "pool_max_size": 8, "offered_rate": 750, "scheduled_requests": 6000, "successful_requests": 3140, "failed_requests": 2860, "error_ratio": 0.476667, "successful_rps": 347.987, "elapsed_seconds": 9.023, "successful_latency_ms": {"p50": 809.855, "p95": 1190.388, "p99": 1238.353}, "errors_by_type": {}, "queue_dropped": 2860, "client_cpu_seconds": 0.393, "docker": {"api": {"avg_cpu_percent": 50.673, "max_memory_mib": 64.82}, "postgres": {"avg_cpu_percent": 7.463, "max_memory_mib": 50.64}, "lb": {"avg_cpu_percent": 2.083, "max_memory_mib": 41.68}}, "pool": {"observed_replicas": 1, "requests_queued": 5266, "wait_ms": 191351, "max_size": 8}, "postgres_delta": {"commits": 6465, "rollbacks": 0, "blocks_read": 0, "blocks_hit": 97643, "temp_files": 0, "deadlocks": 0, "messages": 0}, "database_bytes": 25767059, "database_messages": 30969}
+{"mode": "growth", "replicas": 1, "rate": 750, "repeat": 3, "duration": 8, "warmup": 2, "dataset_messages": 1000, "api_cpu_each": 0.5, "api_memory_mib_each": 256, "pool_max_size": 8, "offered_rate": 750, "scheduled_requests": 6000, "successful_requests": 3410, "failed_requests": 2590, "error_ratio": 0.431667, "successful_rps": 391.052, "elapsed_seconds": 8.72, "successful_latency_ms": {"p50": 795.636, "p95": 976.194, "p99": 988.076}, "errors_by_type": {}, "queue_dropped": 2590, "client_cpu_seconds": 0.38, "docker": {"api": {"avg_cpu_percent": 50.493, "max_memory_mib": 61.84}, "postgres": {"avg_cpu_percent": 6.57, "max_memory_mib": 49.52}, "lb": {"avg_cpu_percent": 2.033, "max_memory_mib": 41.93}}, "pool": {"observed_replicas": 1, "requests_queued": 5799, "wait_ms": 185658, "max_size": 8}, "postgres_delta": {"commits": 7011, "rollbacks": 0, "blocks_read": 0, "blocks_hit": 105675, "temp_files": 0, "deadlocks": 0, "messages": 0}, "database_bytes": 25767059, "database_messages": 30969}
+{"mode": "growth", "replicas": 1, "rate": 1500, "repeat": 1, "duration": 8, "warmup": 2, "dataset_messages": 1000, "api_cpu_each": 0.5, "api_memory_mib_each": 256, "pool_max_size": 8, "offered_rate": 1500, "scheduled_requests": 12000, "successful_requests": 3475, "failed_requests": 8525, "error_ratio": 0.710417, "successful_rps": 399.011, "elapsed_seconds": 8.709, "successful_latency_ms": {"p50": 795.926, "p95": 918.481, "p99": 980.371}, "errors_by_type": {}, "queue_dropped": 8525, "client_cpu_seconds": 0.419, "docker": {"api": {"avg_cpu_percent": 50.17, "max_memory_mib": 65.2}, "postgres": {"avg_cpu_percent": 6.02, "max_memory_mib": 49.52}, "lb": {"avg_cpu_percent": 1.923, "max_memory_mib": 41.7}}, "pool": {"observed_replicas": 1, "requests_queued": 6018, "wait_ms": 195259, "max_size": 8}, "postgres_delta": {"commits": 6592, "rollbacks": 0, "blocks_read": 0, "blocks_hit": 99497, "temp_files": 0, "deadlocks": 0, "messages": 0}, "database_bytes": 25767059, "database_messages": 30969}
+{"mode": "growth", "replicas": 1, "rate": 1500, "repeat": 2, "duration": 8, "warmup": 2, "dataset_messages": 1000, "api_cpu_each": 0.5, "api_memory_mib_each": 256, "pool_max_size": 8, "offered_rate": 1500, "scheduled_requests": 12000, "successful_requests": 3152, "failed_requests": 8848, "error_ratio": 0.737333, "successful_rps": 354.029, "elapsed_seconds": 8.903, "successful_latency_ms": {"p50": 878.869, "p95": 1018.139, "p99": 1070.497}, "errors_by_type": {}, "queue_dropped": 8848, "client_cpu_seconds": 0.44, "docker": {"api": {"avg_cpu_percent": 51.427, "max_memory_mib": 52.96}, "postgres": {"avg_cpu_percent": 6.83, "max_memory_mib": 49.52}, "lb": {"avg_cpu_percent": 2.257, "max_memory_mib": 41.69}}, "pool": {"observed_replicas": 1, "requests_queued": 5362, "wait_ms": 190945, "max_size": 8}, "postgres_delta": {"commits": 6204, "rollbacks": 0, "blocks_read": 0, "blocks_hit": 93635, "temp_files": 0, "deadlocks": 0, "messages": 0}, "database_bytes": 25767059, "database_messages": 30969}
+{"mode": "growth", "replicas": 1, "rate": 1500, "repeat": 3, "duration": 8, "warmup": 2, "dataset_messages": 1000, "api_cpu_each": 0.5, "api_memory_mib_each": 256, "pool_max_size": 8, "offered_rate": 1500, "scheduled_requests": 12000, "successful_requests": 3349, "failed_requests": 8651, "error_ratio": 0.720917, "successful_rps": 384.74, "elapsed_seconds": 8.705, "successful_latency_ms": {"p50": 797.336, "p95": 1034.655, "p99": 1084.912}, "errors_by_type": {}, "queue_dropped": 8651, "client_cpu_seconds": 0.442, "docker": {"api": {"avg_cpu_percent": 50.96, "max_memory_mib": 64.4}, "postgres": {"avg_cpu_percent": 5.623, "max_memory_mib": 49.52}, "lb": {"avg_cpu_percent": 2.29, "max_memory_mib": 41.71}}, "pool": {"observed_replicas": 1, "requests_queued": 5539, "wait_ms": 185287, "max_size": 8}, "postgres_delta": {"commits": 6874, "rollbacks": 0, "blocks_read": 0, "blocks_hit": 103703, "temp_files": 0, "deadlocks": 0, "messages": 0}, "database_bytes": 25767059, "database_messages": 30969}
+{"mode": "growth", "replicas": 2, "rate": 250, "repeat": 1, "duration": 8, "warmup": 2, "dataset_messages": 1000, "api_cpu_each": 0.5, "api_memory_mib_each": 256, "pool_max_size": 8, "offered_rate": 250, "scheduled_requests": 2000, "successful_requests": 2000, "failed_requests": 0, "error_ratio": 0.0, "successful_rps": 249.98, "elapsed_seconds": 8.001, "successful_latency_ms": {"p50": 2.87, "p95": 10.076, "p99": 50.451}, "errors_by_type": {}, "queue_dropped": 0, "client_cpu_seconds": 0.319, "docker": {"api": {"avg_cpu_percent": 34.92, "max_memory_mib": 85.09}, "postgres": {"avg_cpu_percent": 6.023, "max_memory_mib": 58.95}, "lb": {"avg_cpu_percent": 1.613, "max_memory_mib": 41.94}}, "pool": {"observed_replicas": 2, "requests_queued": 6, "wait_ms": 1, "max_size": 7}, "postgres_delta": {"commits": 4102, "rollbacks": 0, "blocks_read": 0, "blocks_hit": 67511, "temp_files": 0, "deadlocks": 0, "messages": 0}, "database_bytes": 25767059, "database_messages": 30969}
+{"mode": "growth", "replicas": 2, "rate": 250, "repeat": 2, "duration": 8, "warmup": 2, "dataset_messages": 1000, "api_cpu_each": 0.5, "api_memory_mib_each": 256, "pool_max_size": 8, "offered_rate": 250, "scheduled_requests": 2000, "successful_requests": 2000, "failed_requests": 0, "error_ratio": 0.0, "successful_rps": 249.97, "elapsed_seconds": 8.001, "successful_latency_ms": {"p50": 2.936, "p95": 55.026, "p99": 129.173}, "errors_by_type": {}, "queue_dropped": 0, "client_cpu_seconds": 0.398, "docker": {"api": {"avg_cpu_percent": 41.043, "max_memory_mib": 96.67}, "postgres": {"avg_cpu_percent": 8.053, "max_memory_mib": 64.64}, "lb": {"avg_cpu_percent": 1.753, "max_memory_mib": 41.7}}, "pool": {"observed_replicas": 2, "requests_queued": 15, "wait_ms": 308, "max_size": 8}, "postgres_delta": {"commits": 4081, "rollbacks": 0, "blocks_read": 0, "blocks_hit": 61880, "temp_files": 0, "deadlocks": 0, "messages": 0}, "database_bytes": 25767059, "database_messages": 30969}
+{"mode": "growth", "replicas": 2, "rate": 250, "repeat": 3, "duration": 8, "warmup": 2, "dataset_messages": 1000, "api_cpu_each": 0.5, "api_memory_mib_each": 256, "pool_max_size": 8, "offered_rate": 250, "scheduled_requests": 2000, "successful_requests": 2000, "failed_requests": 0, "error_ratio": 0.0, "successful_rps": 249.982, "elapsed_seconds": 8.001, "successful_latency_ms": {"p50": 2.886, "p95": 15.37, "p99": 54.353}, "errors_by_type": {}, "queue_dropped": 0, "client_cpu_seconds": 0.341, "docker": {"api": {"avg_cpu_percent": 38.787, "max_memory_mib": 98.52}, "postgres": {"avg_cpu_percent": 5.4, "max_memory_mib": 64.68}, "lb": {"avg_cpu_percent": 1.707, "max_memory_mib": 41.59}}, "pool": {"observed_replicas": 2, "requests_queued": 1, "wait_ms": 2, "max_size": 8}, "postgres_delta": {"commits": 4079, "rollbacks": 0, "blocks_read": 0, "blocks_hit": 61581, "temp_files": 0, "deadlocks": 0, "messages": 0}, "database_bytes": 25767059, "database_messages": 30969}
+{"mode": "growth", "replicas": 2, "rate": 750, "repeat": 1, "duration": 8, "warmup": 2, "dataset_messages": 1000, "api_cpu_each": 0.5, "api_memory_mib_each": 256, "pool_max_size": 8, "offered_rate": 750, "scheduled_requests": 6000, "successful_requests": 5903, "failed_requests": 97, "error_ratio": 0.016167, "successful_rps": 728.191, "elapsed_seconds": 8.106, "successful_latency_ms": {"p50": 2.148, "p95": 468.108, "p99": 599.437}, "errors_by_type": {}, "queue_dropped": 97, "client_cpu_seconds": 0.647, "docker": {"api": {"avg_cpu_percent": 79.57, "max_memory_mib": 107.99}, "postgres": {"avg_cpu_percent": 13.603, "max_memory_mib": 64.71}, "lb": {"avg_cpu_percent": 4.013, "max_memory_mib": 41.94}}, "pool": {"observed_replicas": 2, "requests_queued": 1869, "wait_ms": 27621, "max_size": 8}, "postgres_delta": {"commits": 11975, "rollbacks": 0, "blocks_read": 0, "blocks_hit": 180645, "temp_files": 0, "deadlocks": 0, "messages": 0}, "database_bytes": 25767059, "database_messages": 30969}
+{"mode": "growth", "replicas": 2, "rate": 750, "repeat": 2, "duration": 8, "warmup": 2, "dataset_messages": 1000, "api_cpu_each": 0.5, "api_memory_mib_each": 256, "pool_max_size": 8, "offered_rate": 750, "scheduled_requests": 6000, "successful_requests": 6000, "failed_requests": 0, "error_ratio": 0.0, "successful_rps": 749.669, "elapsed_seconds": 8.004, "successful_latency_ms": {"p50": 2.292, "p95": 144.278, "p99": 193.517}, "errors_by_type": {}, "queue_dropped": 0, "client_cpu_seconds": 0.716, "docker": {"api": {"avg_cpu_percent": 92.29, "max_memory_mib": 108.7}, "postgres": {"avg_cpu_percent": 13.147, "max_memory_mib": 64.73}, "lb": {"avg_cpu_percent": 4.19, "max_memory_mib": 41.94}}, "pool": {"observed_replicas": 2, "requests_queued": 2115, "wait_ms": 15549, "max_size": 8}, "postgres_delta": {"commits": 12108, "rollbacks": 0, "blocks_read": 0, "blocks_hit": 183127, "temp_files": 0, "deadlocks": 0, "messages": 0}, "database_bytes": 25767059, "database_messages": 30969}
+{"mode": "growth", "replicas": 2, "rate": 750, "repeat": 3, "duration": 8, "warmup": 2, "dataset_messages": 1000, "api_cpu_each": 0.5, "api_memory_mib_each": 256, "pool_max_size": 8, "offered_rate": 750, "scheduled_requests": 6000, "successful_requests": 6000, "failed_requests": 0, "error_ratio": 0.0, "successful_rps": 749.762, "elapsed_seconds": 8.003, "successful_latency_ms": {"p50": 2.224, "p95": 167.623, "p99": 254.554}, "errors_by_type": {}, "queue_dropped": 0, "client_cpu_seconds": 0.731, "docker": {"api": {"avg_cpu_percent": 86.027, "max_memory_mib": 98.05}, "postgres": {"avg_cpu_percent": 13.06, "max_memory_mib": 64.73}, "lb": {"avg_cpu_percent": 4.49, "max_memory_mib": 41.98}}, "pool": {"observed_replicas": 2, "requests_queued": 1856, "wait_ms": 14945, "max_size": 8}, "postgres_delta": {"commits": 11997, "rollbacks": 0, "blocks_read": 0, "blocks_hit": 179629, "temp_files": 0, "deadlocks": 0, "messages": 0}, "database_bytes": 25767059, "database_messages": 30969}
+{"mode": "growth", "replicas": 2, "rate": 1500, "repeat": 1, "duration": 8, "warmup": 2, "dataset_messages": 1000, "api_cpu_each": 0.5, "api_memory_mib_each": 256, "pool_max_size": 8, "offered_rate": 1500, "scheduled_requests": 12000, "successful_requests": 7786, "failed_requests": 4214, "error_ratio": 0.351167, "successful_rps": 938.324, "elapsed_seconds": 8.298, "successful_latency_ms": {"p50": 321.188, "p95": 493.5, "p99": 548.275}, "errors_by_type": {}, "queue_dropped": 4214, "client_cpu_seconds": 0.739, "docker": {"api": {"avg_cpu_percent": 98.067, "max_memory_mib": 99.7}, "postgres": {"avg_cpu_percent": 14.437, "max_memory_mib": 64.74}, "lb": {"avg_cpu_percent": 4.373, "max_memory_mib": 41.98}}, "pool": {"observed_replicas": 2, "requests_queued": 7974, "wait_ms": 76567, "max_size": 8}, "postgres_delta": {"commits": 15558, "rollbacks": 0, "blocks_read": 0, "blocks_hit": 233671, "temp_files": 0, "deadlocks": 0, "messages": 0}, "database_bytes": 25767059, "database_messages": 30969}
+{"mode": "growth", "replicas": 2, "rate": 1500, "repeat": 2, "duration": 8, "warmup": 2, "dataset_messages": 1000, "api_cpu_each": 0.5, "api_memory_mib_each": 256, "pool_max_size": 8, "offered_rate": 1500, "scheduled_requests": 12000, "successful_requests": 7921, "failed_requests": 4079, "error_ratio": 0.339917, "successful_rps": 953.293, "elapsed_seconds": 8.309, "successful_latency_ms": {"p50": 321.964, "p95": 454.678, "p99": 509.653}, "errors_by_type": {}, "queue_dropped": 4079, "client_cpu_seconds": 0.758, "docker": {"api": {"avg_cpu_percent": 100.757, "max_memory_mib": 100.03}, "postgres": {"avg_cpu_percent": 15.21, "max_memory_mib": 64.75}, "lb": {"avg_cpu_percent": 4.527, "max_memory_mib": 42.23}}, "pool": {"observed_replicas": 2, "requests_queued": 9150, "wait_ms": 89177, "max_size": 8}, "postgres_delta": {"commits": 16078, "rollbacks": 0, "blocks_read": 0, "blocks_hit": 241647, "temp_files": 0, "deadlocks": 0, "messages": 0}, "database_bytes": 25767059, "database_messages": 30969}
+{"mode": "growth", "replicas": 2, "rate": 1500, "repeat": 3, "duration": 8, "warmup": 2, "dataset_messages": 1000, "api_cpu_each": 0.5, "api_memory_mib_each": 256, "pool_max_size": 8, "offered_rate": 1500, "scheduled_requests": 12000, "successful_requests": 7486, "failed_requests": 4514, "error_ratio": 0.376167, "successful_rps": 900.488, "elapsed_seconds": 8.313, "successful_latency_ms": {"p50": 321.365, "p95": 527.564, "p99": 574.568}, "errors_by_type": {}, "queue_dropped": 4514, "client_cpu_seconds": 0.76, "docker": {"api": {"avg_cpu_percent": 96.313, "max_memory_mib": 114.13}, "postgres": {"avg_cpu_percent": 13.923, "max_memory_mib": 64.75}, "lb": {"avg_cpu_percent": 4.857, "max_memory_mib": 41.98}}, "pool": {"observed_replicas": 2, "requests_queued": 7416, "wait_ms": 85412, "max_size": 8}, "postgres_delta": {"commits": 15026, "rollbacks": 0, "blocks_read": 0, "blocks_hit": 226057, "temp_files": 0, "deadlocks": 0, "messages": 0}, "database_bytes": 25767059, "database_messages": 30969}
+{"mode": "growth", "replicas": 4, "rate": 250, "repeat": 1, "duration": 8, "warmup": 2, "dataset_messages": 1000, "api_cpu_each": 0.5, "api_memory_mib_each": 256, "pool_max_size": 8, "offered_rate": 250, "scheduled_requests": 2000, "successful_requests": 2000, "failed_requests": 0, "error_ratio": 0.0, "successful_rps": 249.9, "elapsed_seconds": 8.003, "successful_latency_ms": {"p50": 2.847, "p95": 17.892, "p99": 116.852}, "errors_by_type": {}, "queue_dropped": 0, "client_cpu_seconds": 0.335, "docker": {"api": {"avg_cpu_percent": 73.133, "max_memory_mib": 165.93}, "postgres": {"avg_cpu_percent": 5.717, "max_memory_mib": 70.38}, "lb": {"avg_cpu_percent": 1.697, "max_memory_mib": 41.96}}, "pool": {"observed_replicas": 4, "requests_queued": 45, "wait_ms": 237, "max_size": 5}, "postgres_delta": {"commits": 4090, "rollbacks": 0, "blocks_read": 0, "blocks_hit": 71220, "temp_files": 0, "deadlocks": 0, "messages": 0}, "database_bytes": 25767059, "database_messages": 30969}
+{"mode": "growth", "replicas": 4, "rate": 250, "repeat": 2, "duration": 8, "warmup": 2, "dataset_messages": 1000, "api_cpu_each": 0.5, "api_memory_mib_each": 256, "pool_max_size": 8, "offered_rate": 250, "scheduled_requests": 2000, "successful_requests": 2000, "failed_requests": 0, "error_ratio": 0.0, "successful_rps": 249.719, "elapsed_seconds": 8.009, "successful_latency_ms": {"p50": 2.907, "p95": 9.687, "p99": 21.273}, "errors_by_type": {}, "queue_dropped": 0, "client_cpu_seconds": 0.328, "docker": {"api": {"avg_cpu_percent": 49.73, "max_memory_mib": 191.97}, "postgres": {"avg_cpu_percent": 5.503, "max_memory_mib": 70.47}, "lb": {"avg_cpu_percent": 1.753, "max_memory_mib": 41.91}}, "pool": {"observed_replicas": 4, "requests_queued": 0, "wait_ms": 0, "max_size": 5}, "postgres_delta": {"commits": 4110, "rollbacks": 0, "blocks_read": 0, "blocks_hit": 61812, "temp_files": 0, "deadlocks": 0, "messages": 0}, "database_bytes": 25767059, "database_messages": 30969}
+{"mode": "growth", "replicas": 4, "rate": 250, "repeat": 3, "duration": 8, "warmup": 2, "dataset_messages": 1000, "api_cpu_each": 0.5, "api_memory_mib_each": 256, "pool_max_size": 8, "offered_rate": 250, "scheduled_requests": 2000, "successful_requests": 2000, "failed_requests": 0, "error_ratio": 0.0, "successful_rps": 249.939, "elapsed_seconds": 8.002, "successful_latency_ms": {"p50": 3.098, "p95": 17.675, "p99": 76.487}, "errors_by_type": {}, "queue_dropped": 0, "client_cpu_seconds": 0.402, "docker": {"api": {"avg_cpu_percent": 74.683, "max_memory_mib": 177.49}, "postgres": {"avg_cpu_percent": 7.597, "max_memory_mib": 72.36}, "lb": {"avg_cpu_percent": 1.89, "max_memory_mib": 41.57}}, "pool": {"observed_replicas": 4, "requests_queued": 1, "wait_ms": 22, "max_size": 5}, "postgres_delta": {"commits": 4078, "rollbacks": 0, "blocks_read": 0, "blocks_hit": 62186, "temp_files": 0, "deadlocks": 0, "messages": 0}, "database_bytes": 25767059, "database_messages": 30969}
+{"mode": "growth", "replicas": 4, "rate": 750, "repeat": 1, "duration": 8, "warmup": 2, "dataset_messages": 1000, "api_cpu_each": 0.5, "api_memory_mib_each": 256, "pool_max_size": 8, "offered_rate": 750, "scheduled_requests": 6000, "successful_requests": 6000, "failed_requests": 0, "error_ratio": 0.0, "successful_rps": 747.495, "elapsed_seconds": 8.027, "successful_latency_ms": {"p50": 4.456, "p95": 312.888, "p99": 369.613}, "errors_by_type": {}, "queue_dropped": 0, "client_cpu_seconds": 0.793, "docker": {"api": {"avg_cpu_percent": 109.91, "max_memory_mib": 213.65}, "postgres": {"avg_cpu_percent": 16.54, "max_memory_mib": 95.11}, "lb": {"avg_cpu_percent": 4.963, "max_memory_mib": 42.07}}, "pool": {"observed_replicas": 4, "requests_queued": 350, "wait_ms": 3249, "max_size": 8}, "postgres_delta": {"commits": 12366, "rollbacks": 0, "blocks_read": 0, "blocks_hit": 193680, "temp_files": 0, "deadlocks": 0, "messages": 0}, "database_bytes": 25767059, "database_messages": 30969}
+{"mode": "growth", "replicas": 4, "rate": 750, "repeat": 2, "duration": 8, "warmup": 2, "dataset_messages": 1000, "api_cpu_each": 0.5, "api_memory_mib_each": 256, "pool_max_size": 8, "offered_rate": 750, "scheduled_requests": 6000, "successful_requests": 6000, "failed_requests": 0, "error_ratio": 0.0, "successful_rps": 748.081, "elapsed_seconds": 8.021, "successful_latency_ms": {"p50": 2.363, "p95": 25.407, "p99": 53.156}, "errors_by_type": {}, "queue_dropped": 0, "client_cpu_seconds": 0.616, "docker": {"api": {"avg_cpu_percent": 111.363, "max_memory_mib": 207.52}, "postgres": {"avg_cpu_percent": 13.387, "max_memory_mib": 95.12}, "lb": {"avg_cpu_percent": 3.747, "max_memory_mib": 41.89}}, "pool": {"observed_replicas": 4, "requests_queued": 23, "wait_ms": 57, "max_size": 8}, "postgres_delta": {"commits": 12044, "rollbacks": 0, "blocks_read": 0, "blocks_hit": 180901, "temp_files": 0, "deadlocks": 0, "messages": 0}, "database_bytes": 25767059, "database_messages": 30969}
+{"mode": "growth", "replicas": 4, "rate": 750, "repeat": 3, "duration": 8, "warmup": 2, "dataset_messages": 1000, "api_cpu_each": 0.5, "api_memory_mib_each": 256, "pool_max_size": 8, "offered_rate": 750, "scheduled_requests": 6000, "successful_requests": 6000, "failed_requests": 0, "error_ratio": 0.0, "successful_rps": 741.745, "elapsed_seconds": 8.089, "successful_latency_ms": {"p50": 2.595, "p95": 182.257, "p99": 375.515}, "errors_by_type": {}, "queue_dropped": 0, "client_cpu_seconds": 0.74, "docker": {"api": {"avg_cpu_percent": 117.407, "max_memory_mib": 220.91}, "postgres": {"avg_cpu_percent": 15.23, "max_memory_mib": 95.18}, "lb": {"avg_cpu_percent": 4.083, "max_memory_mib": 42.04}}, "pool": {"observed_replicas": 4, "requests_queued": 357, "wait_ms": 3544, "max_size": 8}, "postgres_delta": {"commits": 12159, "rollbacks": 0, "blocks_read": 0, "blocks_hit": 184013, "temp_files": 0, "deadlocks": 0, "messages": 0}, "database_bytes": 25767059, "database_messages": 30969}
+{"mode": "growth", "replicas": 4, "rate": 1500, "repeat": 1, "duration": 8, "warmup": 2, "dataset_messages": 1000, "api_cpu_each": 0.5, "api_memory_mib_each": 256, "pool_max_size": 8, "offered_rate": 1500, "scheduled_requests": 12000, "successful_requests": 9320, "failed_requests": 2680, "error_ratio": 0.223333, "successful_rps": 1119.586, "elapsed_seconds": 8.325, "successful_latency_ms": {"p50": 230.705, "p95": 466.032, "p99": 733.365}, "errors_by_type": {}, "queue_dropped": 2680, "client_cpu_seconds": 0.938, "docker": {"api": {"avg_cpu_percent": 142.593, "max_memory_mib": 218.29}, "postgres": {"avg_cpu_percent": 20.28, "max_memory_mib": 96.34}, "lb": {"avg_cpu_percent": 6.71, "max_memory_mib": 42.43}}, "pool": {"observed_replicas": 4, "requests_queued": 1799, "wait_ms": 17211, "max_size": 8}, "postgres_delta": {"commits": 17785, "rollbacks": 0, "blocks_read": 0, "blocks_hit": 267049, "temp_files": 0, "deadlocks": 0, "messages": 0}, "database_bytes": 25767059, "database_messages": 30969}
+{"mode": "growth", "replicas": 4, "rate": 1500, "repeat": 2, "duration": 8, "warmup": 2, "dataset_messages": 1000, "api_cpu_each": 0.5, "api_memory_mib_each": 256, "pool_max_size": 8, "offered_rate": 1500, "scheduled_requests": 12000, "successful_requests": 9886, "failed_requests": 2114, "error_ratio": 0.176167, "successful_rps": 1170.613, "elapsed_seconds": 8.445, "successful_latency_ms": {"p50": 226.066, "p95": 413.16, "p99": 489.537}, "errors_by_type": {}, "queue_dropped": 2114, "client_cpu_seconds": 0.941, "docker": {"api": {"avg_cpu_percent": 142.217, "max_memory_mib": 221.61}, "postgres": {"avg_cpu_percent": 20.977, "max_memory_mib": 95.24}, "lb": {"avg_cpu_percent": 6.13, "max_memory_mib": 42.45}}, "pool": {"observed_replicas": 4, "requests_queued": 1972, "wait_ms": 27198, "max_size": 8}, "postgres_delta": {"commits": 20148, "rollbacks": 0, "blocks_read": 0, "blocks_hit": 302329, "temp_files": 0, "deadlocks": 0, "messages": 0}, "database_bytes": 25767059, "database_messages": 30969}
+{"mode": "growth", "replicas": 4, "rate": 1500, "repeat": 3, "duration": 8, "warmup": 2, "dataset_messages": 1000, "api_cpu_each": 0.5, "api_memory_mib_each": 256, "pool_max_size": 8, "offered_rate": 1500, "scheduled_requests": 12000, "successful_requests": 9531, "failed_requests": 2469, "error_ratio": 0.20575, "successful_rps": 1127.648, "elapsed_seconds": 8.452, "successful_latency_ms": {"p50": 246.905, "p95": 456.887, "p99": 547.176}, "errors_by_type": {}, "queue_dropped": 2469, "client_cpu_seconds": 0.97, "docker": {"api": {"avg_cpu_percent": 144.297, "max_memory_mib": 235.65}, "postgres": {"avg_cpu_percent": 19.94, "max_memory_mib": 97.48}, "lb": {"avg_cpu_percent": 6.687, "max_memory_mib": 42.46}}, "pool": {"observed_replicas": 4, "requests_queued": 1540, "wait_ms": 18841, "max_size": 8}, "postgres_delta": {"commits": 19840, "rollbacks": 0, "blocks_read": 0, "blocks_hit": 297789, "temp_files": 0, "deadlocks": 0, "messages": 0}, "database_bytes": 25767059, "database_messages": 30969}
diff --git a/evidence/scale/confirmation-4-to-2.jsonl b/evidence/scale/confirmation-4-to-2.jsonl
new file mode 100644
index 00000000..9bf14525
--- /dev/null
+++ b/evidence/scale/confirmation-4-to-2.jsonl
@@ -0,0 +1,6 @@
+{"mode": "fixed", "replicas": 4, "rate": 1500, "repeat": 1, "duration": 8, "warmup": 2, "dataset_messages": 1000, "api_cpu_each": 0.25, "api_memory_mib_each": 128, "pool_max_size": 8, "offered_rate": 1500, "scheduled_requests": 12000, "successful_requests": 7298, "failed_requests": 4702, "error_ratio": 0.391833, "successful_rps": 856.033, "elapsed_seconds": 8.525, "successful_latency_ms": {"p50": 315.394, "p95": 605.467, "p99": 723.579}, "errors_by_type": {}, "queue_dropped": 4702, "client_cpu_seconds": 0.699, "docker": {"api": {"avg_cpu_percent": 101.373, "max_memory_mib": 230.97}, "postgres": {"avg_cpu_percent": 10.917, "max_memory_mib": 96.28}, "lb": {"avg_cpu_percent": 4.033, "max_memory_mib": 42.7}}, "pool": {"observed_replicas": 4, "requests_queued": 2961, "wait_ms": 49012, "max_size": 8}, "postgres_delta": {"commits": 14598, "rollbacks": 0, "blocks_read": 0, "blocks_hit": 211976, "temp_files": 0, "deadlocks": 0, "messages": 0}, "database_bytes": 25767059, "database_messages": 30970}
+{"mode": "fixed", "replicas": 4, "rate": 1500, "repeat": 2, "duration": 8, "warmup": 2, "dataset_messages": 1000, "api_cpu_each": 0.25, "api_memory_mib_each": 128, "pool_max_size": 8, "offered_rate": 1500, "scheduled_requests": 12000, "successful_requests": 7569, "failed_requests": 4431, "error_ratio": 0.36925, "successful_rps": 895.216, "elapsed_seconds": 8.455, "successful_latency_ms": {"p50": 302.528, "p95": 551.032, "p99": 821.366}, "errors_by_type": {}, "queue_dropped": 4431, "client_cpu_seconds": 0.705, "docker": {"api": {"avg_cpu_percent": 99.517, "max_memory_mib": 205.15}, "postgres": {"avg_cpu_percent": 15.823, "max_memory_mib": 96.31}, "lb": {"avg_cpu_percent": 4.787, "max_memory_mib": 42.63}}, "pool": {"observed_replicas": 4, "requests_queued": 4193, "wait_ms": 33869, "max_size": 8}, "postgres_delta": {"commits": 14895, "rollbacks": 0, "blocks_read": 0, "blocks_hit": 216728, "temp_files": 0, "deadlocks": 0, "messages": 0}, "database_bytes": 25767059, "database_messages": 30970}
+{"mode": "fixed", "replicas": 4, "rate": 1500, "repeat": 3, "duration": 8, "warmup": 2, "dataset_messages": 1000, "api_cpu_each": 0.25, "api_memory_mib_each": 128, "pool_max_size": 8, "offered_rate": 1500, "scheduled_requests": 12000, "successful_requests": 6745, "failed_requests": 5255, "error_ratio": 0.437917, "successful_rps": 805.135, "elapsed_seconds": 8.377, "successful_latency_ms": {"p50": 335.741, "p95": 691.364, "p99": 835.845}, "errors_by_type": {}, "queue_dropped": 5255, "client_cpu_seconds": 0.704, "docker": {"api": {"avg_cpu_percent": 99.893, "max_memory_mib": 202.03}, "postgres": {"avg_cpu_percent": 15.907, "max_memory_mib": 96.36}, "lb": {"avg_cpu_percent": 4.783, "max_memory_mib": 42.64}}, "pool": {"observed_replicas": 4, "requests_queued": 3062, "wait_ms": 48677, "max_size": 8}, "postgres_delta": {"commits": 13421, "rollbacks": 0, "blocks_read": 0, "blocks_hit": 194846, "temp_files": 0, "deadlocks": 0, "messages": 0}, "database_bytes": 25767059, "database_messages": 30970}
+{"mode": "fixed", "replicas": 2, "rate": 1500, "repeat": 1, "duration": 8, "warmup": 2, "dataset_messages": 1000, "api_cpu_each": 0.5, "api_memory_mib_each": 256, "pool_max_size": 8, "offered_rate": 1500, "scheduled_requests": 12000, "successful_requests": 7927, "failed_requests": 4073, "error_ratio": 0.339417, "successful_rps": 952.317, "elapsed_seconds": 8.324, "successful_latency_ms": {"p50": 316.048, "p95": 495.87, "p99": 540.128}, "errors_by_type": {}, "queue_dropped": 4073, "client_cpu_seconds": 0.749, "docker": {"api": {"avg_cpu_percent": 96.897, "max_memory_mib": 98.05}, "postgres": {"avg_cpu_percent": 13.917, "max_memory_mib": 66.3}, "lb": {"avg_cpu_percent": 3.933, "max_memory_mib": 42.29}}, "pool": {"observed_replicas": 2, "requests_queued": 6441, "wait_ms": 59041, "max_size": 8}, "postgres_delta": {"commits": 15327, "rollbacks": 0, "blocks_read": 0, "blocks_hit": 222787, "temp_files": 0, "deadlocks": 0, "messages": 0}, "database_bytes": 25767059, "database_messages": 30970}
+{"mode": "fixed", "replicas": 2, "rate": 1500, "repeat": 2, "duration": 8, "warmup": 2, "dataset_messages": 1000, "api_cpu_each": 0.5, "api_memory_mib_each": 256, "pool_max_size": 8, "offered_rate": 1500, "scheduled_requests": 12000, "successful_requests": 8055, "failed_requests": 3945, "error_ratio": 0.32875, "successful_rps": 972.715, "elapsed_seconds": 8.281, "successful_latency_ms": {"p50": 309.538, "p95": 465.748, "p99": 575.029}, "errors_by_type": {}, "queue_dropped": 3945, "client_cpu_seconds": 0.738, "docker": {"api": {"avg_cpu_percent": 94.727, "max_memory_mib": 100.38}, "postgres": {"avg_cpu_percent": 15.02, "max_memory_mib": 65.29}, "lb": {"avg_cpu_percent": 4.33, "max_memory_mib": 42.54}}, "pool": {"observed_replicas": 2, "requests_queued": 5546, "wait_ms": 59665, "max_size": 8}, "postgres_delta": {"commits": 17391, "rollbacks": 0, "blocks_read": 0, "blocks_hit": 253125, "temp_files": 0, "deadlocks": 0, "messages": 0}, "database_bytes": 25767059, "database_messages": 30970}
+{"mode": "fixed", "replicas": 2, "rate": 1500, "repeat": 3, "duration": 8, "warmup": 2, "dataset_messages": 1000, "api_cpu_each": 0.5, "api_memory_mib_each": 256, "pool_max_size": 8, "offered_rate": 1500, "scheduled_requests": 12000, "successful_requests": 7862, "failed_requests": 4138, "error_ratio": 0.344833, "successful_rps": 940.679, "elapsed_seconds": 8.358, "successful_latency_ms": {"p50": 319.584, "p95": 489.999, "p99": 595.768}, "errors_by_type": {}, "queue_dropped": 4138, "client_cpu_seconds": 0.725, "docker": {"api": {"avg_cpu_percent": 99.007, "max_memory_mib": 113.54}, "postgres": {"avg_cpu_percent": 15.377, "max_memory_mib": 68.17}, "lb": {"avg_cpu_percent": 4.427, "max_memory_mib": 42.29}}, "pool": {"observed_replicas": 2, "requests_queued": 7232, "wait_ms": 85079, "max_size": 8}, "postgres_delta": {"commits": 15296, "rollbacks": 0, "blocks_read": 0, "blocks_hit": 222495, "temp_files": 0, "deadlocks": 0, "messages": 0}, "database_bytes": 25767059, "database_messages": 30970}
diff --git a/evidence/scale/fixed-p95.svg b/evidence/scale/fixed-p95.svg
new file mode 100644
index 00000000..2b887a48
--- /dev/null
+++ b/evidence/scale/fixed-p95.svg
@@ -0,0 +1,33 @@
+
diff --git a/evidence/scale/fixed-rps.svg b/evidence/scale/fixed-rps.svg
new file mode 100644
index 00000000..98ab80b2
--- /dev/null
+++ b/evidence/scale/fixed-rps.svg
@@ -0,0 +1,33 @@
+
diff --git a/evidence/scale/growth-p95.svg b/evidence/scale/growth-p95.svg
new file mode 100644
index 00000000..0d9e327a
--- /dev/null
+++ b/evidence/scale/growth-p95.svg
@@ -0,0 +1,33 @@
+
diff --git a/evidence/scale/growth-rps.svg b/evidence/scale/growth-rps.svg
new file mode 100644
index 00000000..153d0b28
--- /dev/null
+++ b/evidence/scale/growth-rps.svg
@@ -0,0 +1,33 @@
+
diff --git a/evidence/scale/summary.csv b/evidence/scale/summary.csv
new file mode 100644
index 00000000..9209371f
--- /dev/null
+++ b/evidence/scale/summary.csv
@@ -0,0 +1,19 @@
+mode,replicas,rate,successful_rps,error_ratio,p50_ms,p95_ms,p99_ms,api_cpu_percent,postgres_cpu_percent,lb_cpu_percent,api_memory_mib,postgres_memory_mib,lb_memory_mib,pool_wait_ms,queue_dropped
+fixed,1,250,249.951,0.0,2.787,3.703,5.882,29.56,5.803,1.86,46.79,49.26,41.58,0,0
+fixed,1,750,746.986,0.0,3.533,103.334,137.788,84.993,12.417,4.407,52.01,49.39,41.64,51465,0
+fixed,1,1500,780.299,0.4285,378.91,466.704,725.297,96.24,13.42,4.423,53.53,49.48,41.64,182073,5142
+fixed,2,250,249.959,0.0,2.873,25.949,100.083,42.983,5.957,1.753,102.69,64.63,41.64,14,0
+fixed,2,750,749.772,0.0,1.755,108.359,175.477,69.88,12.173,3.88,97.41,64.76,41.91,7867,0
+fixed,2,1500,956.631,0.338833,314.99,467.748,516.41,98.82,14.147,4.87,113.26,64.77,42.0,72376,4066
+fixed,4,250,249.958,0.0,3.016,60.482,80.401,60.467,5.16,1.68,205.43,79.71,41.74,0,0
+fixed,4,750,732.997,0.0,110.445,397.681,480.195,92.41,12.887,3.723,207.79,95.05,42.23,24258,0
+fixed,4,1500,693.857,0.508167,413.811,747.7,851.226,87.06,13.59,3.993,204.21,95.16,42.35,33271,6098
+growth,1,250,249.97,0.0,2.572,59.465,80.414,33.947,6.093,1.697,48.01,49.47,41.68,410,0
+growth,1,750,362.33,0.472667,809.855,1071.351,1182.579,50.493,6.57,2.083,64.7,49.75,41.69,185658,2836
+growth,1,1500,384.74,0.720917,797.336,1018.139,1070.497,50.96,6.02,2.257,64.4,49.52,41.7,190945,8651
+growth,2,250,249.98,0.0,2.886,15.37,54.353,38.787,6.023,1.707,96.67,64.64,41.7,2,0
+growth,2,750,749.669,0.0,2.224,167.623,254.554,86.027,13.147,4.19,107.99,64.73,41.94,15549,0
+growth,2,1500,938.324,0.351167,321.365,493.5,548.275,98.067,14.437,4.527,100.03,64.75,41.98,85412,4214
+growth,4,250,249.9,0.0,2.907,17.675,76.487,73.133,5.717,1.753,177.49,70.47,41.91,22,0
+growth,4,750,747.495,0.0,2.595,182.257,369.613,111.363,15.23,4.083,213.65,95.12,42.04,3249,0
+growth,4,1500,1127.648,0.20575,230.705,456.887,547.176,142.593,20.28,6.687,221.61,96.34,42.45,18841,2469
diff --git a/scripts/benchmark.py b/scripts/benchmark.py
new file mode 100644
index 00000000..cdcdf2eb
--- /dev/null
+++ b/scripts/benchmark.py
@@ -0,0 +1,322 @@
+#!/usr/bin/env python3
+"""Reproducible, rate-controlled GET benchmark for the lab3 scaling track."""
+
+import argparse
+from collections import defaultdict
+import http.client
+import json
+import math
+import os
+from pathlib import Path
+from queue import Full, Queue
+import statistics
+import subprocess
+import sys
+import tempfile
+import threading
+import time
+from urllib.parse import urlsplit
+from urllib.request import urlopen
+
+ROOT = Path(__file__).resolve().parents[1]
+sys.path.insert(0, str(ROOT / "tests"))
+from client import Client # noqa: E402 - the repository's public bootstrap client
+
+
+def topology(mode, replicas):
+ if mode not in ("fixed", "growth") or replicas not in (1, 2, 4):
+ raise ValueError("mode must be fixed/growth and replicas must be 1/2/4")
+ return replicas, (1 / replicas if mode == "fixed" else 0.5), (512 // replicas if mode == "fixed" else 256)
+
+
+def validate_matrix(rows):
+ if not rows:
+ raise ValueError("no benchmark rows")
+ rates = {row["rate"] for row in rows}
+ if len(rates) < 3:
+ raise ValueError("need at least three offered rates")
+ profiles = {(row["duration"], row["warmup"], row["dataset_messages"]) for row in rows}
+ if len(profiles) != 1:
+ raise ValueError("workload profile changed across measurements")
+ for mode in ("fixed", "growth"):
+ for replicas in (1, 2, 4):
+ _, cpu, memory = topology(mode, replicas)
+ for rate in rates:
+ repeats = [row["repeat"] for row in rows if (row["mode"], row["replicas"], row["rate"]) == (mode, replicas, rate)]
+ if sorted(repeats) != [1, 2, 3]:
+ raise ValueError(f"need three repeats for {mode}/{replicas}/{rate}")
+ for row in rows:
+ if (row["mode"], row["replicas"]) == (mode, replicas) and (row["api_cpu_each"], row["api_memory_mib_each"]) != (cpu, memory):
+ raise ValueError("incorrect API resource quota")
+
+
+def percentile(values, fraction):
+ if not values:
+ return None
+ ordered = sorted(values)
+ return round(ordered[math.ceil(fraction * len(ordered)) - 1], 3)
+
+
+def command(*args):
+ return subprocess.run(args, cwd=ROOT, text=True, capture_output=True, check=True).stdout.strip()
+
+
+def compose(*args, override=None):
+ prefix = ["docker", "compose", "--env-file", ".env", "-f", "compose.yaml"]
+ if override:
+ prefix += ["-f", str(override)]
+ return command(*prefix, *args)
+
+
+def pool_snapshot(base_url, replicas):
+ found = {}
+ for _ in range(8 * replicas):
+ with urlopen(base_url + "/health/ready", timeout=4) as response:
+ headers = response.headers
+ found[headers["X-Instance-Id"]] = {
+ "requests": int(headers["X-Pool-Requests"]),
+ "queued": int(headers["X-Pool-Queued"]),
+ "wait_ms": int(headers["X-Pool-Wait-Ms"]),
+ "size": int(headers["X-Pool-Size"]),
+ "available": int(headers["X-Pool-Available"]),
+ }
+ if len(found) == replicas:
+ break
+ return found
+
+
+def db_snapshot(override):
+ sql = ("SELECT pg_database_size(current_database()),xact_commit,xact_rollback,"
+ "blks_read,blks_hit,temp_files,deadlocks,(SELECT count(*) FROM messages) "
+ "FROM pg_stat_database WHERE datname=current_database()")
+ values = command("docker", "compose", "--env-file", ".env", "-f", "compose.yaml", "-f", str(override),
+ "exec", "-T", "postgres", "psql", "-U", "messenger", "-d", "messenger", "-At", "-F", "|", "-c", sql)
+ return dict(zip(("database_bytes", "commits", "rollbacks", "blocks_read", "blocks_hit", "temp_files", "deadlocks", "messages"),
+ map(int, values.split("|"))))
+
+
+def docker_ids(override):
+ ids = {}
+ for service in ("api", "postgres", "lb"):
+ ids[service] = compose("ps", "-q", service, override=override).splitlines()
+ return ids
+
+
+def stats_sample(ids):
+ all_ids = [id for service_ids in ids.values() for id in service_ids]
+ if not all_ids:
+ return {}
+ output = command("docker", "stats", "--no-stream", "--format", "{{json .}}", *all_ids)
+ by_id = {}
+ for line in output.splitlines():
+ item = json.loads(line)
+ by_id[item["ID"]] = item
+ result = {}
+ for service, service_ids in ids.items():
+ cpu = []
+ memory = []
+ for id in service_ids:
+ item = by_id.get(id[:12]) or by_id.get(id)
+ if not item:
+ raise ValueError("Docker stats omitted a benchmark container")
+ cpu.append(float(item["CPUPerc"].rstrip("%")))
+ amount = item["MemUsage"].split("/")[0].strip()
+ value, unit = float("".join(c for c in amount if c.isdigit() or c == ".")), "".join(c for c in amount if c.isalpha())
+ memory.append(round(value * {"B": 1 / 2**20, "KiB": 1 / 1024, "MiB": 1, "GiB": 1024,
+ "kB": 1000 / 2**20, "MB": 10**6 / 2**20, "GB": 10**9 / 2**20}[unit], 3))
+ result[service] = {"cpu_percent": round(sum(cpu), 3), "memory_mib": round(sum(memory), 3)}
+ return result
+
+
+def open_loop(base_url, path, token, rate, duration, workers=64):
+ """Schedule arrivals by clock; latency includes queue wait and HTTP time."""
+ address = urlsplit(base_url)
+ if address.scheme != "http":
+ raise ValueError("this local lab3 benchmark expects HTTP")
+ jobs = Queue(maxsize=256)
+ latencies = []
+ errors = defaultdict(int)
+ lock = threading.Lock()
+ starts = time.monotonic()
+ cpu_start = time.process_time()
+ count = math.ceil(rate * duration)
+
+ def worker():
+ connection = None
+ while True:
+ due = jobs.get()
+ if due is None:
+ jobs.task_done()
+ break
+ try:
+ if connection is None:
+ connection = http.client.HTTPConnection(address.hostname, address.port or 80, timeout=4)
+ connection.request("GET", path, headers={"Authorization": "Bearer " + token, "Accept": "application/json"})
+ response = connection.getresponse()
+ payload = response.read()
+ if response.status != 200:
+ with lock:
+ errors[f"http_{response.status}"] += 1
+ elif len(json.loads(payload)["items"]) != 20:
+ with lock:
+ errors["wrong_page"] += 1
+ else:
+ with lock:
+ latencies.append((time.monotonic() - due) * 1000)
+ except (OSError, TimeoutError, ValueError, KeyError, http.client.HTTPException):
+ with lock:
+ errors["transport_or_response"] += 1
+ if connection:
+ connection.close()
+ connection = None
+ finally:
+ jobs.task_done()
+ if connection:
+ connection.close()
+
+ threads = [threading.Thread(target=worker, daemon=True) for _ in range(workers)]
+ for thread in threads:
+ thread.start()
+ dropped = 0
+ for index in range(count):
+ due = starts + index / rate
+ delay = due - time.monotonic()
+ if delay > 0:
+ time.sleep(delay)
+ try:
+ jobs.put_nowait(due)
+ except Full:
+ dropped += 1
+ jobs.join()
+ for _ in threads:
+ jobs.put(None)
+ for thread in threads:
+ thread.join()
+ elapsed = time.monotonic() - starts
+ error_count = sum(errors.values()) + dropped
+ return {
+ "offered_rate": rate, "scheduled_requests": count, "successful_requests": len(latencies),
+ "failed_requests": error_count, "error_ratio": round(error_count / count, 6),
+ "successful_rps": round(len(latencies) / elapsed, 3), "elapsed_seconds": round(elapsed, 3),
+ "successful_latency_ms": {f"p{p}": percentile(latencies, p / 100) for p in (50, 95, 99)},
+ "errors_by_type": dict(errors), "queue_dropped": dropped,
+ "client_cpu_seconds": round(time.process_time() - cpu_start, 3),
+ }
+
+
+def fixture(base_url, size, fixture_file):
+ if fixture_file.exists():
+ saved = json.loads(fixture_file.read_text())
+ if saved["base_url"] != base_url or saved["dataset_messages"] != size:
+ raise ValueError("saved fixture URL or size differs from requested profile")
+ Client(base_url).call("GET", saved["path"], actor={"token": saved["token"]})
+ return saved["token"], saved["path"]
+ client = Client(base_url)
+ alice, _, _, chat = client.bootstrap()
+ for index in range(size):
+ client.message(alice, chat, f"synthetic-scale-{index}")
+ path = "/api/v1/chats/" + chat["id"] + "/messages?limit=20"
+ fixture_file.parent.mkdir(parents=True, exist_ok=True)
+ with os.fdopen(os.open(fixture_file, os.O_WRONLY | os.O_CREAT | os.O_EXCL, 0o600), "w") as file:
+ file.write(json.dumps({"base_url": base_url, "dataset_messages": size,
+ "token": alice["token"], "path": path}) + "\n")
+ return alice["token"], path
+
+
+def run(args):
+ full_matrix = args.modes == ["fixed", "growth"] and args.replicas == [1, 2, 4]
+ if args.repeats != 3 or (full_matrix and len(args.rates) < 3) or args.dataset_messages < 20:
+ raise ValueError("require 3 repeats, at least 3 rates for a full matrix, and at least 20 fixture messages")
+ args.output.parent.mkdir(parents=True, exist_ok=True)
+ if args.output.exists():
+ raise ValueError(f"output already exists: {args.output}")
+ token = path = None
+ rows = []
+ with tempfile.TemporaryDirectory() as temporary:
+ override = Path(temporary) / "scale.yaml"
+ for mode in args.modes:
+ for replicas in args.replicas:
+ _, cpu, memory = topology(mode, replicas)
+ override.write_text("services:\n api:\n cpus: " + str(cpu) + "\n mem_limit: " + str(memory) + "m\n")
+ compose("up", "-d", "--build", "--wait", "--scale", f"api={replicas}", override=override)
+ if token is None:
+ token, path = fixture(args.base_url, args.dataset_messages, args.fixture)
+ ids = docker_ids(override)
+ for rate in args.rates:
+ for repeat in range(1, args.repeats + 1):
+ open_loop(args.base_url, path, token, rate, args.warmup)
+ pool_before = pool_snapshot(args.base_url, replicas)
+ db_before = db_snapshot(override)
+ samples = []
+ sampling_errors = []
+ stop = threading.Event()
+
+ def sample():
+ while not stop.is_set():
+ try:
+ samples.append(stats_sample(ids))
+ except (subprocess.CalledProcessError, ValueError, KeyError) as error:
+ sampling_errors.append(type(error).__name__)
+ return
+ stop.wait(1)
+
+ sampler = threading.Thread(target=sample, daemon=True)
+ sampler.start()
+ result = open_loop(args.base_url, path, token, rate, args.duration)
+ stop.set()
+ sampler.join()
+ if sampling_errors or not samples:
+ raise RuntimeError(f"Docker CPU/RAM sampler failed: {sampling_errors}")
+ db_after = db_snapshot(override)
+ if db_after["messages"] != db_before["messages"]:
+ raise RuntimeError("message dataset changed during read-only benchmark")
+ pool_after = pool_snapshot(args.base_url, replicas)
+ pool_delta = {
+ "observed_replicas": len(set(pool_before) & set(pool_after)),
+ "requests_queued": sum(pool_after[i]["queued"] - pool_before[i]["queued"] for i in pool_before.keys() & pool_after.keys()),
+ "wait_ms": sum(pool_after[i]["wait_ms"] - pool_before[i]["wait_ms"] for i in pool_before.keys() & pool_after.keys()),
+ "max_size": max((v["size"] for v in pool_after.values()), default=0),
+ }
+ services = {name: {
+ "avg_cpu_percent": round(statistics.mean(s[name]["cpu_percent"] for s in samples if name in s), 3),
+ "max_memory_mib": round(max(s[name]["memory_mib"] for s in samples if name in s), 3),
+ } for name in ("api", "postgres", "lb") if any(name in s for s in samples)}
+ row = {
+ "mode": mode, "replicas": replicas, "rate": rate, "repeat": repeat,
+ "duration": args.duration, "warmup": args.warmup, "dataset_messages": args.dataset_messages,
+ "api_cpu_each": cpu, "api_memory_mib_each": memory, "pool_max_size": 8,
+ **result, "docker": services, "pool": pool_delta,
+ "postgres_delta": {key: db_after[key] - db_before[key] for key in db_before if key != "database_bytes"},
+ "database_bytes": db_after["database_bytes"], "database_messages": db_after["messages"],
+ }
+ rows.append(row)
+ with args.output.open("a") as output:
+ output.write(json.dumps(row, ensure_ascii=False) + "\n")
+ print(f"{mode} replicas={replicas} rate={rate} repeat={repeat} rps={result['successful_rps']} errors={result['error_ratio']}", flush=True)
+ if full_matrix:
+ validate_matrix(rows)
+
+
+def main():
+ parser = argparse.ArgumentParser(description=__doc__)
+ parser.add_argument("--base-url", default="http://127.0.0.1:8080")
+ parser.add_argument("--rates", type=lambda value: [int(x) for x in value.split(",")], default=[100, 250, 500])
+ parser.add_argument("--modes", type=lambda value: value.split(","), default=["fixed", "growth"])
+ parser.add_argument("--replicas", type=lambda value: [int(x) for x in value.split(",")], default=[1, 2, 4])
+ parser.add_argument("--duration", type=int, default=8)
+ parser.add_argument("--warmup", type=int, default=2)
+ parser.add_argument("--repeats", type=int, default=3)
+ parser.add_argument("--dataset-messages", type=int, default=1000)
+ parser.add_argument("--fixture", type=Path, default=ROOT / "secrets/scale_fixture.json")
+ parser.add_argument("--output", type=Path, default=ROOT / "evidence/scale/baseline.jsonl")
+ args = parser.parse_args()
+ if (not args.rates or min(args.rates) <= 0 or not args.modes or not args.replicas or
+ any(mode not in ("fixed", "growth") for mode in args.modes) or
+ any(replica not in (1, 2, 4) for replica in args.replicas) or
+ args.duration <= 0 or args.warmup <= 0):
+ raise ValueError("rates and durations must be positive")
+ run(args)
+
+
+if __name__ == "__main__":
+ main()
diff --git a/scripts/benchmark.sh b/scripts/benchmark.sh
new file mode 100755
index 00000000..d9c755cb
--- /dev/null
+++ b/scripts/benchmark.sh
@@ -0,0 +1,4 @@
+#!/bin/sh
+set -eu
+cd "$(dirname "$0")/.."
+exec python3 scripts/benchmark.py "$@"
diff --git a/scripts/plot_scale.py b/scripts/plot_scale.py
new file mode 100644
index 00000000..87e58612
--- /dev/null
+++ b/scripts/plot_scale.py
@@ -0,0 +1,86 @@
+#!/usr/bin/env python3
+"""Render the three-repeat scale matrix as small, dependency-free SVG plots."""
+
+import argparse
+import csv
+import json
+from pathlib import Path
+import statistics
+
+from benchmark import validate_matrix
+
+
+COLORS = {1: "#2563eb", 2: "#16a34a", 4: "#dc2626"}
+
+
+def svg(mode, metric, rows, output):
+ width, height = 720, 420
+ left, right, top, bottom = 75, 25, 35, 65
+ plot_w, plot_h = width - left - right, height - top - bottom
+ rates = sorted({row["rate"] for row in rows})
+ max_y = max((row[metric] or 0 for row in rows), default=0) * 1.1 or 1
+ x = lambda rate: left + (rate - rates[0]) / (rates[-1] - rates[0]) * plot_w
+ y = lambda value: top + plot_h * (1 - value / max_y)
+ label = "Успешные запросы/с" if metric == "successful_rps" else "p95, мс"
+ parts = [f'']
+ output.write_text("\n".join(parts) + "\n", encoding="utf-8")
+
+
+def main():
+ parser = argparse.ArgumentParser(description=__doc__)
+ parser.add_argument("input", type=Path)
+ args = parser.parse_args()
+ rows = [json.loads(line) for line in args.input.read_text().splitlines() if line.strip()]
+ validate_matrix(rows)
+ summary = []
+ for mode in ("fixed", "growth"):
+ for replicas in (1, 2, 4):
+ for rate in sorted({row["rate"] for row in rows}):
+ group = [row for row in rows if (row["mode"], row["replicas"], row["rate"]) == (mode, replicas, rate)]
+ def latency(p):
+ values = [row["successful_latency_ms"][p] for row in group if row["successful_latency_ms"][p] is not None]
+ return round(statistics.median(values), 3) if values else None
+
+ summary.append({"mode": mode, "replicas": replicas, "rate": rate,
+ "successful_rps": round(statistics.median(row["successful_rps"] for row in group), 3),
+ "error_ratio": round(statistics.median(row["error_ratio"] for row in group), 6),
+ "p50_ms": latency("p50"), "p95_ms": latency("p95"), "p99_ms": latency("p99"),
+ "api_cpu_percent": round(statistics.median(row["docker"]["api"]["avg_cpu_percent"] for row in group), 3),
+ "postgres_cpu_percent": round(statistics.median(row["docker"]["postgres"]["avg_cpu_percent"] for row in group), 3),
+ "lb_cpu_percent": round(statistics.median(row["docker"]["lb"]["avg_cpu_percent"] for row in group), 3),
+ "api_memory_mib": round(statistics.median(row["docker"]["api"]["max_memory_mib"] for row in group), 3),
+ "postgres_memory_mib": round(statistics.median(row["docker"]["postgres"]["max_memory_mib"] for row in group), 3),
+ "lb_memory_mib": round(statistics.median(row["docker"]["lb"]["max_memory_mib"] for row in group), 3),
+ "pool_wait_ms": round(statistics.median(row["pool"]["wait_ms"] for row in group), 3),
+ "queue_dropped": statistics.median(row["queue_dropped"] for row in group)})
+ with (args.input.parent / "summary.csv").open("w", newline="") as file:
+ writer = csv.DictWriter(file, fieldnames=summary[0].keys(), lineterminator="\n")
+ writer.writeheader()
+ writer.writerows(summary)
+ for mode in ("fixed", "growth"):
+ group = [row for row in summary if row["mode"] == mode]
+ svg(mode, "successful_rps", group, args.input.parent / f"{mode}-rps.svg")
+ svg(mode, "p95_ms", group, args.input.parent / f"{mode}-p95.svg")
+
+
+if __name__ == "__main__":
+ main()
diff --git a/src/api.py b/src/api.py
index e9b7a8f8..b4418118 100644
--- a/src/api.py
+++ b/src/api.py
@@ -6,6 +6,7 @@ from uuid import UUID
from fastapi import APIRouter, Depends, HTTPException, Query, Request, Response
from .auth import basic_credentials, bearer_token
+from . import db
from .models import CreateChat, CreateMessage, CreateUser
from .store import store
@@ -18,8 +19,14 @@ def live():
@router.get("/health/ready")
-def ready():
- return store.ready()
+def ready(response: Response):
+ result = store.ready()
+ stats = db.pool().get_stats()
+ for header, key in (("X-Pool-Requests", "requests_num"), ("X-Pool-Queued", "requests_queued"),
+ ("X-Pool-Wait-Ms", "requests_wait_ms"), ("X-Pool-Size", "pool_size"),
+ ("X-Pool-Available", "pool_available")):
+ response.headers[header] = str(stats.get(key, 0))
+ return result
@router.post("/api/v1/users", status_code=201)
diff --git a/student-tests/test_benchmark_profile.py b/student-tests/test_benchmark_profile.py
new file mode 100644
index 00000000..660dea8e
--- /dev/null
+++ b/student-tests/test_benchmark_profile.py
@@ -0,0 +1,72 @@
+"""The scaling matrix must compare the same workload and resource budgets."""
+
+import sys
+import json
+import unittest
+from unittest.mock import patch
+from pathlib import Path
+
+sys.path.insert(0, str(Path(__file__).resolve().parents[1] / "scripts"))
+
+from benchmark import open_loop, topology, validate_matrix
+
+
+class BenchmarkProfileTests(unittest.TestCase):
+ def test_fixed_budget_is_identical_across_replica_counts(self):
+ budgets = [topology("fixed", replicas) for replicas in (1, 2, 4)]
+ self.assertEqual([replicas * cpu for replicas, cpu, _ in budgets], [1.0] * 3)
+ self.assertEqual([replicas * memory for replicas, _, memory in budgets], [512] * 3)
+
+ def test_growth_budget_increases_with_replicas(self):
+ budgets = [topology("growth", replicas) for replicas in (1, 2, 4)]
+ self.assertEqual([replicas * cpu for replicas, cpu, _ in budgets], [0.5, 1.0, 2.0])
+ self.assertEqual([replicas * memory for replicas, _, memory in budgets], [256, 512, 1024])
+
+ def test_matrix_requires_three_repeats_and_same_workload(self):
+ rows = [
+ {"mode": mode, "replicas": replicas, "rate": rate, "repeat": repeat,
+ "duration": 8, "warmup": 2, "dataset_messages": 1000,
+ "api_cpu_each": topology(mode, replicas)[1],
+ "api_memory_mib_each": topology(mode, replicas)[2]}
+ for mode in ("fixed", "growth")
+ for replicas in (1, 2, 4)
+ for rate in (100, 250, 500)
+ for repeat in (1, 2, 3)
+ ]
+ validate_matrix(rows)
+ with self.assertRaises(ValueError):
+ validate_matrix(rows[:-1])
+ rows[-1]["dataset_messages"] = 999
+ with self.assertRaises(ValueError):
+ validate_matrix(rows)
+
+ def test_open_loop_counts_responses_at_a_scheduled_rate(self):
+ class Response:
+ status = 200
+
+ def read(self):
+ return json.dumps({"items": [{}] * 20}).encode()
+
+ class Connection:
+ def __init__(self, *args, **kwargs):
+ pass
+
+ def request(self, *args, **kwargs):
+ pass
+
+ def getresponse(self):
+ return Response()
+
+ def close(self):
+ pass
+
+ with patch("benchmark.http.client.HTTPConnection", Connection):
+ result = open_loop("http://127.0.0.1:8080", "/history", "synthetic", 20, 0.2, workers=2)
+ self.assertEqual(result["scheduled_requests"], 4)
+ self.assertEqual(result["successful_requests"], 4)
+ self.assertEqual(result["failed_requests"], 0)
+ self.assertGreater(result["successful_latency_ms"]["p95"], 0)
+
+
+if __name__ == "__main__":
+ unittest.main()