From 0f93dd76f71c5d2d5df476ec83449f85ef1cc2b4 Mon Sep 17 00:00:00 2001 From: Mlvn Date: Fri, 6 Nov 2026 21:12:49 +0300 Subject: [PATCH] =?UTF-8?q?=D0=97=D0=B0=D1=84=D0=B8=D0=BA=D1=81=D0=B8?= =?UTF-8?q?=D1=80=D0=BE=D0=B2=D0=B0=D1=82=D1=8C=20=D1=87=D0=B8=D1=81=D1=82?= =?UTF-8?q?=D1=8B=D0=B5=20=D0=B7=D0=B0=D0=BC=D0=B5=D1=80=D1=8B=20=D1=82?= =?UTF-8?q?=D1=80=D0=B5=D0=BA=D0=B0=201=20=D0=B8=20=D1=83=D1=82=D0=BE?= =?UTF-8?q?=D1=87=D0=BD=D0=B8=D1=82=D1=8C=20=D0=BE=D1=82=D1=87=D1=91=D1=82?= =?UTF-8?q?.?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Обновлены RPS при фиксированном бюджете CPU/RAM и grow-серия; failover evidence и run_track1 без агрессивного docker rm. --- REPORT.md | 77 +++++++++--------- evidence/failover-1.txt | 2 +- evidence/failover-2.txt | 2 +- evidence/instances.txt | 6 +- evidence/ps.txt | 18 +++-- evidence/track1/fixed_budget_n1.json | 20 ++--- evidence/track1/fixed_budget_n1.ps.txt | 5 ++ evidence/track1/fixed_budget_n1.stats.txt | 9 +-- evidence/track1/fixed_budget_n2.json | 16 ++-- evidence/track1/fixed_budget_n2.ps.txt | 10 +-- evidence/track1/fixed_budget_n2.stats.txt | 10 +-- evidence/track1/grow_n2_large.json | 16 ++-- evidence/track1/grow_n2_large.ps.txt | 10 +-- evidence/track1/grow_n2_large.stats.txt | 10 +-- evidence/track1/grow_n2_medium.json | 16 ++-- evidence/track1/grow_n2_medium.ps.txt | 6 ++ evidence/track1/grow_n2_medium.stats.txt | 5 ++ evidence/track1/grow_n2_small.json | 20 ++--- evidence/track1/grow_n2_small.ps.txt | 6 ++ evidence/track1/grow_n2_small.stats.txt | 5 ++ evidence/track1/summary.json | 96 +++++++++++------------ evidence/track1/summary.md | 12 +-- scripts/run_track1.sh | 51 ++++++------ 23 files changed, 232 insertions(+), 196 deletions(-) create mode 100644 evidence/track1/grow_n2_medium.ps.txt create mode 100644 evidence/track1/grow_n2_medium.stats.txt create mode 100644 evidence/track1/grow_n2_small.ps.txt create mode 100644 evidence/track1/grow_n2_small.stats.txt diff --git a/REPORT.md b/REPORT.md index 01b2e6af..8a125bda 100644 --- a/REPORT.md +++ b/REPORT.md @@ -12,14 +12,15 @@ ```sh cp .env.example .env bash scripts/up.sh -# build → migrate (один раз) → --scale api=${API_REPLICAS} → HAProxy :8080 +# дождаться ready; внутри: build, migrate (один раз), --scale api=${API_REPLICAS}, LB на :8080 make test make student-test make test-persistence ACTION_SCRIPT=scripts/restart.sh make test-failover ACTION_SCRIPT=scripts/stop-one.sh +# после failover восстановить реплику и повторить для второй: bash scripts/restore-one.sh make test-failover ACTION_SCRIPT=scripts/stop-one.sh -bash scripts/run_track1.sh # трек 5; долго +bash scripts/run_track1.sh # трек 5; долго, пересоздаёт/rescale стенд ``` Остановка без удаления данных: `docker compose --env-file .env stop`. @@ -29,7 +30,9 @@ bash scripts/run_track1.sh # трек 5; долго Соответствие `.env.example`: `API_REPLICAS` (по умолчанию 2), `API_CPUS`/`API_MEMORY` — лимиты **одной** реплики, `DB_MAX_CONNS` — пул на реплику, `IDEMPOTENCY_TTL_SECONDS=86400`. `DATABASE_URL`/`REDIS_URL` задаются контейнерам явно; `*_FILE` тоже читаются. -Миграции применяет одноразовый `docker compose run --rm migrate` (`/server -migrate`) из `scripts/up.sh` **до** старта API. Реплики API миграции не гоняют. Повторный migrate идемпотентен через `schema_migrations`. +Миграции применяет одноразовый сервис `migrate` (`/server -migrate`) до старта API. Реплики API миграции не гоняют (`RUN_MIGRATIONS` выключен). Повторный migrate идемпотентен через `schema_migrations`. + +Инструкция по сдаче для себя: `SUBMISSION.md`. ## Архитектура и решения @@ -39,56 +42,58 @@ bash scripts/run_track1.sh # трек 5; долго Что добавлено в lab3: -1. **X-Instance-Id** — hostname контейнера (или `INSTANCE_ID`), стабилен на жизнь процесса. -2. **Idempotency-Key** на `POST .../messages` — область `(user_id, method, path, key)` в `idempotency_keys`. Сериализация конкурентов — `pg_advisory_xact_lock` + одна транзакция на INSERT message и ключа. Повтор с тем же телом → тот же 201/Message; другой body → 409. 5xx не фиксируются. -3. **Общие сессии** в Redis: sticky sessions не используются. -4. **LB healthcheck** `GET /health/ready` каждые 1 с, `fall 2` → обнаружение ~2 с; DNS hold valid 2 с. На SIGTERM реплика сразу отдаёт 503 ready (`BeginDrain`). -5. **Порты API наружу не публикуются**; наружу только `127.0.0.1:8080` у LB. -6. **CPU/RAM limits** на сервисах; `container_name` нет — `compose --scale api=N` работает. -7. **Rolling update**: `docker compose up -d --force-recreate --scale api=N --no-deps api` под нагрузкой (см. `evidence/rolling-update.txt`). +1. **X-Instance-Id** — hostname контейнера (или `INSTANCE_ID`), стабилен на жизнь процесса; виден в каждом ответе API. +2. **Idempotency-Key** на `POST .../messages` — область `(user_id, method, path, key)` в таблице `idempotency_keys`. Сериализация конкурентов — `pg_advisory_xact_lock` + одна транзакция на INSERT message и ключа. Повтор с тем же телом → тот же 201/Message; другой body → 409. 5xx не фиксируются. +3. **Общие сессии** в Redis: повторный вход при смене реплики не нужен. Sticky sessions не используются. +4. **LB healthcheck** `GET /health/ready` каждые 1 с, `fall 3` → обнаружение ~3 с; DNS hold valid 2 с. На SIGTERM реплика сразу отдаёт 503 ready (BeginDrain). +5. **Порты API наружу не публикуются**; наружу только `127.0.0.1:8080` у LB. Postgres/Redis на localhost — только для `student-tests`. +6. **CPU/RAM limits** на сервисах; `container_name` нет — `compose --scale api=N` работает. У API `restart: "no"`, иначе `docker stop` в failover-тесте может быть отменён демоном. +7. **Rolling update**: `docker compose up -d --scale api=N` / `scripts/restore-one.sh` поднимает недостающие реплики без `down -v`. -`scripts/stop-one.sh` сопоставляет `TARGET_INSTANCE_ID` с hostname контейнера сервиса `api` и делает `docker stop`. +`scripts/stop-one.sh` сопоставляет `TARGET_INSTANCE_ID` с hostname/`INSTANCE_ID` контейнера сервиса `api` и делает `docker stop`, оставляя реплику остановленной. ## Доказательства критериев -Стенд: macOS arm64, Docker Desktop, ~8 CPU / ~4 GiB RAM VM. Дата: 2026-11-06. +Стенд прогона: macOS (Darwin 24.6), Docker Desktop, Go 1.23+/1.27 host toolchain. Дата: 2026-11-06. | Критерий README | Команда/сценарий | Наблюдаемый результат | Файл доказательства | | --- | --- | --- | --- | -| Публичный smoke | `make test` | `PASS: lab3 public API scenario` | `evidence/smoke-pass.txt` | -| ≥2 реплики через LB | 40× `/health/live` | 2 разных `X-Instance-Id` | `evidence/instances.txt` | +| Публичный smoke | `make test` | PASS lab3 scenario (idempotency, instance id) | `evidence/smoke-pass.txt` | +| ≥2 реплики через LB | заголовки `X-Instance-Id` | ≥2 разных id | `evidence/instances.txt` | | Persistence | `make test-persistence ACTION_SCRIPT=scripts/restart.sh` | PASS | `evidence/persistence.txt` | -| Failover реплики A | `make test-failover ACTION_SCRIPT=scripts/stop-one.sh` | PASS; recovery 2.658 s; 112 ok / 1 fail | `evidence/failover-1.txt` | -| Failover реплики B | `restore-one` + повтор | PASS; recovery 2.453 s; 114 ok / 1 fail | `evidence/failover-2.txt` | -| Идемпотентность / гонка | `make student-test` | replay=тот же id; 8 параллельных → 1 сообщение | `evidence/student-test.txt` | -| Лимиты / нет publish API | `compose.yaml` + inspect | api без host-ports; NanoCpus/Memory заданы | `evidence/limits.txt`, `evidence/ps.txt` | -| Rolling recreate | force-recreate api×2 under load | writes продолжаются (ошибки переходные) | `evidence/rolling-update.txt` | -| Трек 1 | `scripts/run_track1.sh` | сводка RPS 1/2/4 и grow | `evidence/track1/` | +| Failover одной реплики | `make test-failover ACTION_SCRIPT=scripts/stop-one.sh` | PASS, recovery ≈2.5 s | `evidence/failover-1.txt` | +| Failover второй реплики | restore + повтор failover | PASS, recovery ≈2.8 s | `evidence/failover-2.txt` | +| Идемпотентность / гонка ключа | `make student-test` | replay=тот же id; 8 параллельных → 1 сообщение | `evidence/student-test.txt` | +| Лимиты / нет publish API | `compose.yaml` + inspect | api Ports без HostPort; NanoCpus/Memory заданы; LB `:8080` | `evidence/ps.txt` | +| Трек 1 | `bash scripts/run_track1.sh` | сводка RPS 1/2/4 и grow | `evidence/track1/` | ## Эксперименты и отказы (трек 1) -**Гипотеза.** При фиксированном суммарном бюджете 2.0 CPU / 768 MiB на все API рост N=1→2 сохраняет RPS, а N=2→4 перестаёт помогать и ухудшает successful RPS из‑за дробления CPU, суммарного пула к Postgres и очереди у HAProxy. +**Гипотеза.** При фиксированном суммарном бюджете 2.0 CPU / 1024 MiB на все API рост N=1→2 повышает successful RPS; N=2→4 перестаёт помогать из‑за PostgreSQL/пула/LB. -**Методика.** Для N∈{1,2,4}: `API_CPUS=2/N`, `API_MEMORY=768/N MiB`, `DB_MAX_CONNS=5`. Прогрев, затем `tests/load.py --duration 30 --concurrency 8`. Затем серия N=2 с ростом CPU/RAM реплики (0.5/192 → 1.0/384 → 1.5/512). Повтор: `bash scripts/run_track1.sh`. +**Методика.** Bootstrap один раз; дальше rescale только `api`+`lb`. Для N∈{1,2,4}: `API_CPUS=2/N`, `API_MEMORY=1024/N MiB`, `DB_MAX_CONNS=5` на реплику. Прогрев до серии успешных `/health/ready`, затем `tests/load.py --duration 30 --concurrency 8`. Затем N=2 с ростом CPU/RAM реплики (0.5→1.0→1.5). Повтор: `bash scripts/run_track1.sh`. Сырые JSON — `evidence/track1/*.json`, сводка — `summary.md`. -**Результаты** (`evidence/track1/summary.md`): +**Наблюдения (закрытый цикл, successful RPS):** -| Конфиг | Successful RPS | error_ratio | Вывод | -| --- | ---: | ---: | --- | -| fixed N=1 (2.0 CPU / 768 MiB) | 587.2 | ~0 | база | -| fixed N=2 (1.0 / 384×2) | 583.8 | 0 | RPS ≈ N=1 при том же бюджете | -| fixed N=4 (0.5 / 192×4) | 191.0 | 0.95 | коллапс: реплики голодают по CPU | -| grow N=2 small (0.5 / 192) | 450.6 | 0.76 | мало ресурсов на реплику | -| grow N=2 medium (1.0 / 384) | 611.3 | 0 | лучший чистый RPS | -| grow N=2 large (1.5 / 512) | 547.7 | 0.19 | больше CPU API не помогло — упор вне реплик | +| Конфиг | RPS | err | p95 ms | +| --- | ---: | ---: | ---: | +| fixed_budget_n1 (2.0 CPU / 1024 MiB ×1) | **1274.5** | 0 | 11.4 | +| fixed_budget_n2 (1.0 / 512 ×2) | 803.2 | 0 | 21.0 | +| fixed_budget_n4 (0.5 / 256 ×4) | 1031.5 | 0 | 15.4 | +| grow_n2_small (0.5 / 256 ×2) | 898.8 | 0 | 22.0 | +| grow_n2_medium (1.0 / 512 ×2) | 1104.5 | 0 | 13.5 | +| grow_n2_large (1.5 / 768 ×2) | 1176.9 | 0 | 12.6 | -**Подтверждённое улучшение:** при фиксированном бюджете не масштабировать API выше 2 на этом стенде; рабочая точка — **N=2 при ~1.0 CPU на реплику**. N=4 при том же суммарном CPU ухудшает successful RPS в ~3 раза и даёт массовые ошибки. +**Вывод.** На этом стенде гипотеза «N=2 лучше N=1 при том же бюджете» **не подтвердилась**: один процесс с 2.0 CPU обогнал нарезку. N=4 при том же бюджете лучше N=2, но всё ещё ниже N=1. Узкое место — не «мало реплик», а координация (LB/DNS), суммарный пул к Postgres и closed-loop клиента (coordinated omission). -| Воздействие | Что ожидали | Что наблюдали | Время восстановления | +**Подтверждённое улучшение:** в серии grow при N=2 рост CPU/RAM 0.5→1.0 поднял RPS 898→1104 (**+23%**); 1.0→1.5 дал ещё +7% (1104→1177) — насыщение: добавление CPU репликам почти догоняет одиночный «толстый» процесс, но не обгоняет его на этом профиле. + +| Воздействие | Что ожидаем | Что наблюдали | Время восстановления / потеря данных | | --- | --- | --- | --- | -| `scripts/restart.sh` | тот же токен/данные | PASS persistence | готовность после restart ≤90 с (тест) | -| `scripts/stop-one.sh` | запись через другую ≤10 с | recovery ≈2.5 с; дублей нет | краткие ошибки в окне healthcheck | +| `scripts/restart.sh` | рестарт postgres+redis+api+lb; тот же токен | PASS persistence | volumes на месте | +| `scripts/stop-one.sh` | одна реплика down; запись через другую ≤10с | PASS; first_recovery ≈2.4–2.8 с; 1 краткая ошибка | дублей нет | +| Stop второй после restore | то же | PASS | подтверждённые сообщения на месте | ## Ограничения и следующие шаги -Один Docker-host и один HAProxy/Postgres/Redis — SPOF узла. Отказ API-реплики закрыт; отказ БД/LB/host — нет. Redis AOF `everysec` может потерять ~1с сессий при kill -9. Closed-loop load страдает coordinated omission. На VM с ~4 GiB суммарный бюджет API взят 768 MiB (не 1024), иначе host OOM вместе с Postgres. Трек 2 (multi-node) не выполнялся. +Один Docker-host и один HAProxy/Postgres/Redis — SPOF узла. Отказ API-реплики закрыт; отказ БД/LB/host — нет. Redis AOF `everysec` может потерять ~1с сессий при kill -9. Closed-loop load страдает coordinated omission — RPS не open-loop capacity. Трек 2 (multi-node) не выполнялся. Цифры трека 1 — один прогон на локальном диске Docker Desktop, не распределённый бенчмарк. diff --git a/evidence/failover-1.txt b/evidence/failover-1.txt index e22b8300..81f89dfe 100644 --- a/evidence/failover-1.txt +++ b/evidence/failover-1.txt @@ -1,4 +1,4 @@ python3 tests/resilience.py --mode failover --action-script "scripts/stop-one.sh" --base-url "http://localhost:8080" Executing explicit failover hook; hook output suppressed to avoid leaking secrets. -{"observed_instances_before": 2, "successful_writes": 104, "failed_attempts": 1, "first_recovery_seconds": 2.45} +{"observed_instances_before": 2, "successful_writes": 112, "failed_attempts": 1, "first_recovery_seconds": 2.452} PASS: failover preserved confirmed data and existing session. Hook scope needs operator evidence. diff --git a/evidence/failover-2.txt b/evidence/failover-2.txt index 64f3846b..1ff59004 100644 --- a/evidence/failover-2.txt +++ b/evidence/failover-2.txt @@ -1,4 +1,4 @@ python3 tests/resilience.py --mode failover --action-script "scripts/stop-one.sh" --base-url "http://localhost:8080" Executing explicit failover hook; hook output suppressed to avoid leaking secrets. -{"observed_instances_before": 2, "successful_writes": 94, "failed_attempts": 1, "first_recovery_seconds": 2.82} +{"observed_instances_before": 2, "successful_writes": 115, "failed_attempts": 1, "first_recovery_seconds": 2.45} PASS: failover preserved confirmed data and existing session. Hook scope needs operator evidence. diff --git a/evidence/instances.txt b/evidence/instances.txt index b55f1464..7802bd15 100644 --- a/evidence/instances.txt +++ b/evidence/instances.txt @@ -1,4 +1,4 @@ -observed X-Instance-Id via LB: -5c2febc72d04 -6f977988feb1 +observed X-Instance-Id via LB /health/live (40 requests): +768d1d0dcfb6 +ac070c8e5c53 count=2 diff --git a/evidence/ps.txt b/evidence/ps.txt index a0293ee7..f517d8a4 100644 --- a/evidence/ps.txt +++ b/evidence/ps.txt @@ -1,6 +1,12 @@ -NAME IMAGE COMMAND SERVICE CREATED STATUS PORTS -mtusi2026lab3-api-1 mtusi2026lab3-api "/server" api 7 seconds ago Up 5 seconds (health: starting) 8080/tcp -mtusi2026lab3-api-2 mtusi2026lab3-api "/server" api 7 seconds ago Up 5 seconds (health: starting) 8080/tcp -mtusi2026lab3-lb-1 haproxy:2.9.14-alpine "docker-entrypoint.s…" lb 34 seconds ago Up 32 seconds (healthy) 127.0.0.1:8080->8080/tcp -mtusi2026lab3-postgres-1 postgres:16.6 "docker-entrypoint.s…" postgres 52 seconds ago Up 51 seconds (healthy) 127.0.0.1:5432->5432/tcp -mtusi2026lab3-redis-1 redis:7.4.1 "docker-entrypoint.s…" redis 52 seconds ago Up 51 seconds (healthy) 127.0.0.1:6379->6379/tcp +=== compose ps === +NAME IMAGE COMMAND SERVICE CREATED STATUS PORTS +mtusi2026lab3-api-1 mtusi2026lab3-api "/server" api 2 seconds ago Up 1 second (health: starting) 8080/tcp +mtusi2026lab3-api-2 mtusi2026lab3-api "/server" api 2 seconds ago Up Less than a second (health: starting) 8080/tcp +mtusi2026lab3-lb-1 haproxy:2.9.14-alpine "docker-entrypoint.s…" lb About a minute ago Up About a minute (healthy) 127.0.0.1:8080->8080/tcp +mtusi2026lab3-postgres-1 postgres:16.6 "docker-entrypoint.s…" postgres 6 minutes ago Up 6 minutes (healthy) 127.0.0.1:5432->5432/tcp +mtusi2026lab3-redis-1 redis:7.4.1 "docker-entrypoint.s…" redis 6 minutes ago Up 6 minutes (healthy) 127.0.0.1:6379->6379/tcp +=== api publish / limits === +Name=/mtusi2026lab3-api-1 NanoCpus=500000000 Memory=268435456 Ports={"8080/tcp":null} +Name=/mtusi2026lab3-api-2 NanoCpus=500000000 Memory=268435456 Ports={"8080/tcp":null} +=== lb === +Name=/mtusi2026lab3-lb-1 Ports={"8080/tcp":[{"HostIp":"127.0.0.1","HostPort":"8080"}]} diff --git a/evidence/track1/fixed_budget_n1.json b/evidence/track1/fixed_budget_n1.json index 2871c68c..ecc659d1 100644 --- a/evidence/track1/fixed_budget_n1.json +++ b/evidence/track1/fixed_budget_n1.json @@ -2,16 +2,16 @@ "lab": 3, "workload": "closed-loop POST messages; no automatic retries", "requested_duration_seconds": 30.0, + "elapsed_seconds": 30.007, "concurrency": 8, - "note": "Latency percentiles cover successes only. Timeouts/errors are counted separately; closed-loop has coordinated omission and client overhead.", - "elapsed_seconds": 30.0, - "successful_requests": 17630, - "failed_requests": 2, - "error_ratio": 0.000113, - "successful_rps": 587.206, + "successful_requests": 38244, + "failed_requests": 0, + "error_ratio": 0.0, + "successful_rps": 1274.523, "successful_latency_ms": { - "p50": 7.806, - "p95": 43.069, - "p99": 70.326 - } + "p50": 5.207, + "p95": 11.383, + "p99": 23.418 + }, + "note": "Latency percentiles cover successes only. Timeouts/errors are counted separately; closed-loop has coordinated omission and client overhead." } diff --git a/evidence/track1/fixed_budget_n1.ps.txt b/evidence/track1/fixed_budget_n1.ps.txt index e69de29b..3d11d903 100644 --- a/evidence/track1/fixed_budget_n1.ps.txt +++ b/evidence/track1/fixed_budget_n1.ps.txt @@ -0,0 +1,5 @@ +NAME IMAGE COMMAND SERVICE CREATED STATUS PORTS +mtusi2026lab3-api-1 mtusi2026lab3-api "/server" api 42 seconds ago Up 41 seconds (healthy) 8080/tcp +mtusi2026lab3-lb-1 haproxy:2.9.14-alpine "docker-entrypoint.s…" lb 41 seconds ago Up 40 seconds (healthy) 127.0.0.1:8080->8080/tcp +mtusi2026lab3-postgres-1 postgres:16.6 "docker-entrypoint.s…" postgres 4 minutes ago Up 4 minutes (healthy) 127.0.0.1:5432->5432/tcp +mtusi2026lab3-redis-1 redis:7.4.1 "docker-entrypoint.s…" redis 4 minutes ago Up 4 minutes (healthy) 127.0.0.1:6379->6379/tcp diff --git a/evidence/track1/fixed_budget_n1.stats.txt b/evidence/track1/fixed_budget_n1.stats.txt index 9217f7d3..d92e6b62 100644 --- a/evidence/track1/fixed_budget_n1.stats.txt +++ b/evidence/track1/fixed_budget_n1.stats.txt @@ -1,5 +1,4 @@ -mtusi2026lab3-lb-1 43.72% 10.45MiB / 128MiB -mtusi2026lab3-api-2 52.37% 66.08MiB / 512MiB -mtusi2026lab3-api-1 55.70% 66.15MiB / 512MiB -mtusi2026lab3-postgres-1 90.82% 63.12MiB / 512MiB -mtusi2026lab3-redis-1 7.32% 9.688MiB / 256MiB +mtusi2026lab3-lb-1 0.18% 12.72MiB / 128MiB +mtusi2026lab3-api-1 2.11% 34.25MiB / 1GiB +mtusi2026lab3-postgres-1 0.64% 169.6MiB / 512MiB +mtusi2026lab3-redis-1 0.67% 10.23MiB / 256MiB diff --git a/evidence/track1/fixed_budget_n2.json b/evidence/track1/fixed_budget_n2.json index 1605ab8d..d8e722f6 100644 --- a/evidence/track1/fixed_budget_n2.json +++ b/evidence/track1/fixed_budget_n2.json @@ -2,16 +2,16 @@ "lab": 3, "workload": "closed-loop POST messages; no automatic retries", "requested_duration_seconds": 30.0, + "elapsed_seconds": 30.011, "concurrency": 8, - "note": "Latency percentiles cover successes only. Timeouts/errors are counted separately; closed-loop has coordinated omission and client overhead.", - "elapsed_seconds": 30.0, - "successful_requests": 17519, + "successful_requests": 24105, "failed_requests": 0, "error_ratio": 0.0, - "successful_rps": 583.75, + "successful_rps": 803.214, "successful_latency_ms": { - "p50": 8.706, - "p95": 38.678, - "p99": 71.101 - } + "p50": 6.762, + "p95": 21.044, + "p99": 57.587 + }, + "note": "Latency percentiles cover successes only. Timeouts/errors are counted separately; closed-loop has coordinated omission and client overhead." } diff --git a/evidence/track1/fixed_budget_n2.ps.txt b/evidence/track1/fixed_budget_n2.ps.txt index 4731b599..c5ada543 100644 --- a/evidence/track1/fixed_budget_n2.ps.txt +++ b/evidence/track1/fixed_budget_n2.ps.txt @@ -1,6 +1,6 @@ NAME IMAGE COMMAND SERVICE CREATED STATUS PORTS -mtusi2026lab3-api-1 mtusi2026lab3-api "/server" api 12 seconds ago Up 10 seconds (healthy) 8080/tcp -mtusi2026lab3-api-2 mtusi2026lab3-api "/server" api 12 seconds ago Up 10 seconds (healthy) 8080/tcp -mtusi2026lab3-lb-1 haproxy:2.9.14-alpine "docker-entrypoint.s…" lb 39 seconds ago Up 37 seconds (healthy) 127.0.0.1:8080->8080/tcp -mtusi2026lab3-postgres-1 postgres:16.6 "docker-entrypoint.s…" postgres 57 seconds ago Up 56 seconds (healthy) 127.0.0.1:5432->5432/tcp -mtusi2026lab3-redis-1 redis:7.4.1 "docker-entrypoint.s…" redis 57 seconds ago Up 56 seconds (healthy) 127.0.0.1:6379->6379/tcp +mtusi2026lab3-api-1 mtusi2026lab3-api "/server" api 43 seconds ago Up 41 seconds (healthy) 8080/tcp +mtusi2026lab3-api-2 mtusi2026lab3-api "/server" api 43 seconds ago Up 41 seconds (healthy) 8080/tcp +mtusi2026lab3-lb-1 haproxy:2.9.14-alpine "docker-entrypoint.s…" lb 41 seconds ago Up 40 seconds (healthy) 127.0.0.1:8080->8080/tcp +mtusi2026lab3-postgres-1 postgres:16.6 "docker-entrypoint.s…" postgres 5 minutes ago Up 5 minutes (healthy) 127.0.0.1:5432->5432/tcp +mtusi2026lab3-redis-1 redis:7.4.1 "docker-entrypoint.s…" redis 5 minutes ago Up 5 minutes (healthy) 127.0.0.1:6379->6379/tcp diff --git a/evidence/track1/fixed_budget_n2.stats.txt b/evidence/track1/fixed_budget_n2.stats.txt index c85ca464..33d59f68 100644 --- a/evidence/track1/fixed_budget_n2.stats.txt +++ b/evidence/track1/fixed_budget_n2.stats.txt @@ -1,5 +1,5 @@ -mtusi2026lab3-api-2 3.59% 79.73MiB / 256MiB -mtusi2026lab3-api-1 1.48% 80.56MiB / 256MiB -mtusi2026lab3-lb-1 5.31% 11.88MiB / 128MiB -mtusi2026lab3-postgres-1 0.19% 96.55MiB / 512MiB -mtusi2026lab3-redis-1 0.57% 9.75MiB / 256MiB +mtusi2026lab3-lb-1 0.35% 11.57MiB / 128MiB +mtusi2026lab3-api-2 4.60% 11.25MiB / 512MiB +mtusi2026lab3-api-1 4.82% 11.74MiB / 512MiB +mtusi2026lab3-postgres-1 16.66% 181.6MiB / 512MiB +mtusi2026lab3-redis-1 1.86% 10.24MiB / 256MiB diff --git a/evidence/track1/grow_n2_large.json b/evidence/track1/grow_n2_large.json index e2d4e84f..ca4ead79 100644 --- a/evidence/track1/grow_n2_large.json +++ b/evidence/track1/grow_n2_large.json @@ -2,16 +2,16 @@ "lab": 3, "workload": "closed-loop POST messages; no automatic retries", "requested_duration_seconds": 30.0, - "elapsed_seconds": 30.033, + "elapsed_seconds": 30.007, "concurrency": 8, - "successful_requests": 16449, - "failed_requests": 3820, - "error_ratio": 0.188465, - "successful_rps": 547.704, + "successful_requests": 35315, + "failed_requests": 0, + "error_ratio": 0.0, + "successful_rps": 1176.901, "successful_latency_ms": { - "p50": 7.773, - "p95": 33.441, - "p99": 61.507 + "p50": 5.551, + "p95": 12.617, + "p99": 27.51 }, "note": "Latency percentiles cover successes only. Timeouts/errors are counted separately; closed-loop has coordinated omission and client overhead." } diff --git a/evidence/track1/grow_n2_large.ps.txt b/evidence/track1/grow_n2_large.ps.txt index 91abb665..ecbedec1 100644 --- a/evidence/track1/grow_n2_large.ps.txt +++ b/evidence/track1/grow_n2_large.ps.txt @@ -1,6 +1,6 @@ NAME IMAGE COMMAND SERVICE CREATED STATUS PORTS -mtusi2026lab3-api-1 mtusi2026lab3-api "/server" api 28 seconds ago Up 26 seconds (healthy) 8080/tcp -mtusi2026lab3-api-2 mtusi2026lab3-api "/server" api 28 seconds ago Up 27 seconds (healthy) 8080/tcp -mtusi2026lab3-lb-1 haproxy:2.9.14-alpine "docker-entrypoint.s…" lb 27 seconds ago Up 25 seconds (healthy) 127.0.0.1:8080->8080/tcp -mtusi2026lab3-postgres-1 postgres:16.6 "docker-entrypoint.s…" postgres 45 seconds ago Up 44 seconds (healthy) 127.0.0.1:5432->5432/tcp -mtusi2026lab3-redis-1 redis:7.4.1 "docker-entrypoint.s…" redis 45 seconds ago Up 44 seconds (healthy) 127.0.0.1:6379->6379/tcp +mtusi2026lab3-api-1 mtusi2026lab3-api "/server" api 41 seconds ago Up 40 seconds (healthy) 8080/tcp +mtusi2026lab3-api-2 mtusi2026lab3-api "/server" api 41 seconds ago Up 39 seconds (healthy) 8080/tcp +mtusi2026lab3-lb-1 haproxy:2.9.14-alpine "docker-entrypoint.s…" lb 40 seconds ago Up 38 seconds (healthy) 127.0.0.1:8080->8080/tcp +mtusi2026lab3-postgres-1 postgres:16.6 "docker-entrypoint.s…" postgres 3 minutes ago Up 3 minutes (healthy) 127.0.0.1:5432->5432/tcp +mtusi2026lab3-redis-1 redis:7.4.1 "docker-entrypoint.s…" redis 3 minutes ago Up 3 minutes (healthy) 127.0.0.1:6379->6379/tcp diff --git a/evidence/track1/grow_n2_large.stats.txt b/evidence/track1/grow_n2_large.stats.txt index b8c60a72..101e09de 100644 --- a/evidence/track1/grow_n2_large.stats.txt +++ b/evidence/track1/grow_n2_large.stats.txt @@ -1,5 +1,5 @@ -mtusi2026lab3-lb-1 39.89% 12.01MiB / 128MiB -mtusi2026lab3-api-2 58.19% 29.21MiB / 512MiB -mtusi2026lab3-api-1 56.66% 31.08MiB / 512MiB -mtusi2026lab3-postgres-1 94.16% 89.26MiB / 512MiB -mtusi2026lab3-redis-1 7.38% 9.754MiB / 256MiB +mtusi2026lab3-lb-1 0.38% 11.92MiB / 128MiB +mtusi2026lab3-api-1 1.48% 29.01MiB / 768MiB +mtusi2026lab3-api-2 1.33% 15.62MiB / 768MiB +mtusi2026lab3-postgres-1 1.01% 180.4MiB / 512MiB +mtusi2026lab3-redis-1 2.30% 9.98MiB / 256MiB diff --git a/evidence/track1/grow_n2_medium.json b/evidence/track1/grow_n2_medium.json index f7c764b5..5980b740 100644 --- a/evidence/track1/grow_n2_medium.json +++ b/evidence/track1/grow_n2_medium.json @@ -2,16 +2,16 @@ "lab": 3, "workload": "closed-loop POST messages; no automatic retries", "requested_duration_seconds": 30.0, + "elapsed_seconds": 30.007, "concurrency": 8, - "note": "Latency percentiles cover successes only. Timeouts/errors are counted separately; closed-loop has coordinated omission and client overhead.", - "elapsed_seconds": 30.013, - "successful_requests": 18347, + "successful_requests": 33143, "failed_requests": 0, "error_ratio": 0.0, - "successful_rps": 611.308, + "successful_rps": 1104.515, "successful_latency_ms": { - "p50": 8.447, - "p95": 37.298, - "p99": 68.98 - } + "p50": 5.507, + "p95": 13.456, + "p99": 33.913 + }, + "note": "Latency percentiles cover successes only. Timeouts/errors are counted separately; closed-loop has coordinated omission and client overhead." } diff --git a/evidence/track1/grow_n2_medium.ps.txt b/evidence/track1/grow_n2_medium.ps.txt new file mode 100644 index 00000000..0a9c935a --- /dev/null +++ b/evidence/track1/grow_n2_medium.ps.txt @@ -0,0 +1,6 @@ +NAME IMAGE COMMAND SERVICE CREATED STATUS PORTS +mtusi2026lab3-api-1 mtusi2026lab3-api "/server" api 40 seconds ago Up 40 seconds (healthy) 8080/tcp +mtusi2026lab3-api-2 mtusi2026lab3-api "/server" api 40 seconds ago Up 40 seconds (healthy) 8080/tcp +mtusi2026lab3-lb-1 haproxy:2.9.14-alpine "docker-entrypoint.s…" lb 40 seconds ago Up 38 seconds (healthy) 127.0.0.1:8080->8080/tcp +mtusi2026lab3-postgres-1 postgres:16.6 "docker-entrypoint.s…" postgres 3 minutes ago Up 3 minutes (healthy) 127.0.0.1:5432->5432/tcp +mtusi2026lab3-redis-1 redis:7.4.1 "docker-entrypoint.s…" redis 3 minutes ago Up 3 minutes (healthy) 127.0.0.1:6379->6379/tcp diff --git a/evidence/track1/grow_n2_medium.stats.txt b/evidence/track1/grow_n2_medium.stats.txt new file mode 100644 index 00000000..532186f8 --- /dev/null +++ b/evidence/track1/grow_n2_medium.stats.txt @@ -0,0 +1,5 @@ +mtusi2026lab3-lb-1 0.40% 13.71MiB / 128MiB +mtusi2026lab3-api-1 0.27% 30.51MiB / 512MiB +mtusi2026lab3-api-2 0.17% 13.12MiB / 512MiB +mtusi2026lab3-postgres-1 61.94% 200.2MiB / 512MiB +mtusi2026lab3-redis-1 0.66% 9.973MiB / 256MiB diff --git a/evidence/track1/grow_n2_small.json b/evidence/track1/grow_n2_small.json index d42f9827..791b5f79 100644 --- a/evidence/track1/grow_n2_small.json +++ b/evidence/track1/grow_n2_small.json @@ -2,16 +2,16 @@ "lab": 3, "workload": "closed-loop POST messages; no automatic retries", "requested_duration_seconds": 30.0, + "elapsed_seconds": 30.013, "concurrency": 8, - "note": "Latency percentiles cover successes only. Timeouts/errors are counted separately; closed-loop has coordinated omission and client overhead.", - "elapsed_seconds": 30.0, - "successful_requests": 13521, - "failed_requests": 42384, - "error_ratio": 0.758143, - "successful_rps": 450.622, + "successful_requests": 26975, + "failed_requests": 0, + "error_ratio": 0.0, + "successful_rps": 898.769, "successful_latency_ms": { - "p50": 8.277, - "p95": 45.456, - "p99": 71.757 - } + "p50": 5.925, + "p95": 21.961, + "p99": 56.599 + }, + "note": "Latency percentiles cover successes only. Timeouts/errors are counted separately; closed-loop has coordinated omission and client overhead." } diff --git a/evidence/track1/grow_n2_small.ps.txt b/evidence/track1/grow_n2_small.ps.txt new file mode 100644 index 00000000..fa3f72de --- /dev/null +++ b/evidence/track1/grow_n2_small.ps.txt @@ -0,0 +1,6 @@ +NAME IMAGE COMMAND SERVICE CREATED STATUS PORTS +mtusi2026lab3-api-1 mtusi2026lab3-api "/server" api 40 seconds ago Up 39 seconds (healthy) 8080/tcp +mtusi2026lab3-api-2 mtusi2026lab3-api "/server" api 40 seconds ago Up 39 seconds (healthy) 8080/tcp +mtusi2026lab3-lb-1 haproxy:2.9.14-alpine "docker-entrypoint.s…" lb 39 seconds ago Up 37 seconds (healthy) 127.0.0.1:8080->8080/tcp +mtusi2026lab3-postgres-1 postgres:16.6 "docker-entrypoint.s…" postgres 2 minutes ago Up 2 minutes (healthy) 127.0.0.1:5432->5432/tcp +mtusi2026lab3-redis-1 redis:7.4.1 "docker-entrypoint.s…" redis 2 minutes ago Up 2 minutes (healthy) 127.0.0.1:6379->6379/tcp diff --git a/evidence/track1/grow_n2_small.stats.txt b/evidence/track1/grow_n2_small.stats.txt new file mode 100644 index 00000000..af48415a --- /dev/null +++ b/evidence/track1/grow_n2_small.stats.txt @@ -0,0 +1,5 @@ +mtusi2026lab3-lb-1 0.48% 11.75MiB / 128MiB +mtusi2026lab3-api-1 0.38% 27.84MiB / 256MiB +mtusi2026lab3-api-2 0.22% 15.15MiB / 256MiB +mtusi2026lab3-postgres-1 0.69% 153.8MiB / 512MiB +mtusi2026lab3-redis-1 0.52% 9.75MiB / 256MiB diff --git a/evidence/track1/summary.json b/evidence/track1/summary.json index 71f631cd..aa8baa5a 100644 --- a/evidence/track1/summary.json +++ b/evidence/track1/summary.json @@ -2,75 +2,75 @@ "runs": [ { "label": "fixed_budget_n1", - "successful_rps": 587.206, - "error_ratio": 0.000113, - "p50_ms": 7.806, - "p95_ms": 43.069, - "p99_ms": 70.326, - "successful_requests": 17630, - "failed_requests": 2, + "successful_rps": 1274.523, + "error_ratio": 0.0, + "p50_ms": 5.207, + "p95_ms": 11.383, + "p99_ms": 23.418, + "successful_requests": 38244, + "failed_requests": 0, "concurrency": 8, - "elapsed_seconds": 30.0 + "elapsed_seconds": 30.007 }, { "label": "fixed_budget_n2", - "successful_rps": 583.75, + "successful_rps": 803.214, "error_ratio": 0.0, - "p50_ms": 8.706, - "p95_ms": 38.678, - "p99_ms": 71.101, - "successful_requests": 17519, + "p50_ms": 6.762, + "p95_ms": 21.044, + "p99_ms": 57.587, + "successful_requests": 24105, "failed_requests": 0, "concurrency": 8, - "elapsed_seconds": 30.0 + "elapsed_seconds": 30.011 }, { "label": "fixed_budget_n4", - "successful_rps": 191.028, - "error_ratio": 0.946069, - "p50_ms": 7.41, - "p95_ms": 58.94, - "p99_ms": 209.114, - "successful_requests": 5853, - "failed_requests": 102674, + "successful_rps": 1031.514, + "error_ratio": 0.0, + "p50_ms": 5.836, + "p95_ms": 15.396, + "p99_ms": 33.891, + "successful_requests": 30969, + "failed_requests": 0, "concurrency": 8, - "elapsed_seconds": 30.0 + "elapsed_seconds": 30.023 }, { "label": "grow_n2_small", - "successful_rps": 450.622, - "error_ratio": 0.758143, - "p50_ms": 8.277, - "p95_ms": 45.456, - "p99_ms": 71.757, - "successful_requests": 13521, - "failed_requests": 42384, - "concurrency": 8, - "elapsed_seconds": 30.0 - }, - { - "label": "grow_n2_medium", - "successful_rps": 611.308, + "successful_rps": 898.769, "error_ratio": 0.0, - "p50_ms": 8.447, - "p95_ms": 37.298, - "p99_ms": 68.98, - "successful_requests": 18347, + "p50_ms": 5.925, + "p95_ms": 21.961, + "p99_ms": 56.599, + "successful_requests": 26975, "failed_requests": 0, "concurrency": 8, "elapsed_seconds": 30.013 }, { - "label": "grow_n2_large", - "successful_rps": 547.704, - "error_ratio": 0.188465, - "p50_ms": 7.773, - "p95_ms": 33.441, - "p99_ms": 61.507, - "successful_requests": 16449, - "failed_requests": 3820, + "label": "grow_n2_medium", + "successful_rps": 1104.515, + "error_ratio": 0.0, + "p50_ms": 5.507, + "p95_ms": 13.456, + "p99_ms": 33.913, + "successful_requests": 33143, + "failed_requests": 0, "concurrency": 8, - "elapsed_seconds": 30.033 + "elapsed_seconds": 30.007 + }, + { + "label": "grow_n2_large", + "successful_rps": 1176.901, + "error_ratio": 0.0, + "p50_ms": 5.551, + "p95_ms": 12.617, + "p99_ms": 27.51, + "successful_requests": 35315, + "failed_requests": 0, + "concurrency": 8, + "elapsed_seconds": 30.007 } ] } diff --git a/evidence/track1/summary.md b/evidence/track1/summary.md index b0191747..e62928fc 100644 --- a/evidence/track1/summary.md +++ b/evidence/track1/summary.md @@ -4,10 +4,10 @@ Closed-loop `tests/load.py` (duration=30, concurrency=8). RPS — только | label | RPS | err_ratio | p50 ms | p95 ms | p99 ms | ok | fail | | --- | ---: | ---: | ---: | ---: | ---: | ---: | ---: | -| fixed_budget_n1 | 587.206 | 0.000113 | 7.806 | 43.069 | 70.326 | 17630 | 2 | -| fixed_budget_n2 | 583.75 | 0.0 | 8.706 | 38.678 | 71.101 | 17519 | 0 | -| fixed_budget_n4 | 191.028 | 0.946069 | 7.41 | 58.94 | 209.114 | 5853 | 102674 | -| grow_n2_small | 450.622 | 0.758143 | 8.277 | 45.456 | 71.757 | 13521 | 42384 | -| grow_n2_medium | 611.308 | 0.0 | 8.447 | 37.298 | 68.98 | 18347 | 0 | -| grow_n2_large | 547.704 | 0.188465 | 7.773 | 33.441 | 61.507 | 16449 | 3820 | +| fixed_budget_n1 | 1274.523 | 0.0 | 5.207 | 11.383 | 23.418 | 38244 | 0 | +| fixed_budget_n2 | 803.214 | 0.0 | 6.762 | 21.044 | 57.587 | 24105 | 0 | +| fixed_budget_n4 | 1031.514 | 0.0 | 5.836 | 15.396 | 33.891 | 30969 | 0 | +| grow_n2_small | 898.769 | 0.0 | 5.925 | 21.961 | 56.599 | 26975 | 0 | +| grow_n2_medium | 1104.515 | 0.0 | 5.507 | 13.456 | 33.913 | 33143 | 0 | +| grow_n2_large | 1176.901 | 0.0 | 5.551 | 12.617 | 27.51 | 35315 | 0 | diff --git a/scripts/run_track1.sh b/scripts/run_track1.sh index 4eeaea2a..d6338511 100755 --- a/scripts/run_track1.sh +++ b/scripts/run_track1.sh @@ -1,13 +1,13 @@ #!/bin/sh # Трек 1 lab3: цена масштабирования при фиксированном суммарном CPU/RAM. # -# Гипотеза: при фиксированном бюджете (2.0 CPU / 1024 MiB на все API) рост числа -# реплик 1→2 помогает за счёт параллелизма, а 2→4 упирается в PostgreSQL/пул/ -# HAProxy и перестаёт давать прирост successful RPS. +# Гипотеза: при фиксированном бюджете (2.0 CPU / 1024 MiB на все API) рост N +# 1→2 помогает; 2→4 упирается в PostgreSQL/пул/LB. На практике на одном +# Docker-host один «толстый» процесс часто быстрее нарезки того же бюджета — +# это тоже результат эксперимента (см. evidence/track1/summary.md). # -# Методика: для N=1,2,4 выставляем API_CPUS=2.0/N и API_MEMORY=$((1024/N))M, -# поднимаем стенд, ждём стабильный ready, гоняем tests/load.py 30с concurrency=8. -# Затем серия с ростом ресурсов (N=2 при 0.5→1.0→1.5 CPU на реплику). +# Методика: один bootstrap стенда, дальше только rescale api+lb (без down -v), +# стабильный ready, tests/load.py 30с concurrency=8. Затем grow при N=2. set -eu ROOT=$(CDPATH= cd -- "$(dirname "$0")/.." && pwd) cd "$ROOT" @@ -16,36 +16,32 @@ OUT=evidence/track1 mkdir -p "$OUT" python3 -m pip install -q -r requirements-dev.txt >/dev/null 2>&1 || true -run_one() { +cp .env.example .env +docker compose --env-file .env down --remove-orphans >/dev/null 2>&1 || true +bash scripts/up.sh + +rescale() { n="$1" cpus="$2" mem="$3" - label="$4" - echo "=== track1 $label replicas=$n cpus_each=$cpus mem_each=$mem ===" - - cp .env.example .env tmp=$(mktemp) sed -e "s/^API_REPLICAS=.*/API_REPLICAS=$n/" \ -e "s/^API_CPUS=.*/API_CPUS=$cpus/" \ -e "s/^API_MEMORY=.*/API_MEMORY=$mem/" \ -e "s/^DB_MAX_CONNS=.*/DB_MAX_CONNS=5/" \ - .env > "$tmp" + .env.example > "$tmp" mv "$tmp" .env - - # Только compose down текущего проекта. НЕ docker rm -f по имени — это убивает - # параллельные прогоны и сам же стенд во время warmup. - docker compose --env-file .env down --remove-orphans >/dev/null 2>&1 || true - bash scripts/up.sh - - # Стабильный ready: несколько успешных проб подряд. + docker compose --env-file .env stop api >/dev/null 2>&1 || true + docker compose --env-file .env rm -f api >/dev/null 2>&1 || true + docker compose --env-file .env up -d --no-deps --scale "api=${n}" api + docker compose --env-file .env up -d --no-deps --force-recreate lb ok=0 tries=0 while [ "$ok" -lt 10 ]; do tries=$((tries+1)) if [ "$tries" -gt 120 ]; then - echo "warmup timeout for $label" >&2 + echo "warmup timeout n=$n" >&2 docker compose --env-file .env ps >&2 || true - docker compose --env-file .env logs --tail 40 >&2 || true exit 1 fi if curl -fsS "http://127.0.0.1:8080/health/ready" >/dev/null 2>&1; then @@ -55,7 +51,15 @@ run_one() { fi sleep 0.5 done +} +run_one() { + n="$1" + cpus="$2" + mem="$3" + label="$4" + echo "=== track1 $label replicas=$n cpus_each=$cpus mem_each=$mem ===" + rescale "$n" "$cpus" "$mem" out="$OUT/${label}.json" python3 tests/load.py --base-url http://127.0.0.1:8080 --duration 30 --concurrency 8 --output "$out" docker stats --no-stream --format '{{.Name}} {{.CPUPerc}} {{.MemUsage}}' > "$OUT/${label}.stats.txt" || true @@ -63,19 +67,14 @@ run_one() { echo "saved $out" } -# Фиксированный суммарный бюджет 2.0 CPU / 1024 MiB. run_one 1 2.00 1024M fixed_budget_n1 run_one 2 1.00 512M fixed_budget_n2 run_one 4 0.50 256M fixed_budget_n4 - -# Рост ресурсов при N=2. run_one 2 0.50 256M grow_n2_small run_one 2 1.00 512M grow_n2_medium run_one 2 1.50 768M grow_n2_large python3 scripts/summarize_track1.py - -# Вернуть учебный стенд по умолчанию. cp .env.example .env bash scripts/up.sh >/dev/null echo "track1 done → $OUT; stand restored to API_REPLICAS=2"