лр7: Docker-прогон, evidence, исправления

This commit is contained in:
kalibr456
2026-11-07 20:17:11 +03:00
parent 20afc2c790
commit cb38fd5231
249 changed files with 114500 additions and 82 deletions
+265
View File
@@ -0,0 +1,265 @@
"""Генерирует infra/grafana/dashboards/messenger.json (импортируемый dashboard, provisioning из Git).
python scripts/build_dashboard.py
Все запросы панелей — здесь, в одном месте. student-tests/check_dashboard.py выполняет каждый запрос
против живых Prometheus/Loki/Tempo и падает, если имя метрики/метки не существует или запрос с ошибкой.
Единицы: секунды (s) для длительностей, reqps/ops для частот, percentunit для долей, bytes для памяти.
"""
import json
from pathlib import Path
ROOT = Path(__file__).resolve().parents[1]
OUT = ROOT / "infra/grafana/dashboards/messenger.json"
PROM = {"type": "prometheus", "uid": "prometheus"}
LOKI = {"type": "loki", "uid": "loki"}
TEMPO = {"type": "tempo", "uid": "tempo"}
API = 'job="messenger-api", http_route!~"/health/.*"' # бизнес-запросы API (знаменатель SLO)
QUEUE = 'messaging_destination_name="attachments.process.v1"'
_panels = []
_next_id = [1]
_y = [0]
def _id():
_next_id[0] += 1
return _next_id[0]
def row(title):
_panels.append({"type": "row", "title": title, "id": _id(), "collapsed": False,
"gridPos": {"h": 1, "w": 24, "x": 0, "y": _y[0]}, "panels": []})
_y[0] += 1
class Line:
"""Раскладка панелей в строку сетки Grafana (24 колонки)."""
def __init__(self, height):
self.x, self.h = 0, height
def place(self, width):
pos = {"h": self.h, "w": width, "x": self.x, "y": _y[0]}
self.x += width
return pos
def end(self):
_y[0] += self.h
def prom(expr, legend="", instant=False, fmt=None):
target = {"datasource": PROM, "expr": expr, "legendFormat": legend or "__auto", "refId": "",
"range": not instant, "instant": instant}
if fmt:
target["format"] = fmt
return target
def _with_refs(targets):
for n, target in enumerate(targets):
target["refId"] = chr(ord("A") + n)
return targets
def timeseries(line, width, title, description, targets, unit, thresholds=None, stack=False, minimum=None):
defaults = {"unit": unit, "custom": {"lineWidth": 1, "fillOpacity": 10, "showPoints": "never",
"stacking": {"mode": "normal" if stack else "none"}}}
if minimum is not None:
defaults["min"] = minimum
if thresholds:
defaults["thresholds"] = {"mode": "absolute", "steps": [{"color": "green", "value": None}] + [
{"color": color, "value": value} for value, color in thresholds]}
defaults["custom"]["thresholdsStyle"] = {"mode": "line+area"}
_panels.append({"type": "timeseries", "id": _id(), "title": title, "description": description,
"datasource": PROM, "gridPos": line.place(width), "targets": _with_refs(targets),
"fieldConfig": {"defaults": defaults, "overrides": []},
"options": {"legend": {"displayMode": "list", "placement": "bottom"},
"tooltip": {"mode": "multi", "sort": "desc"}}})
def stat(line, width, title, description, expr, unit, steps, decimals=None):
defaults = {"unit": unit, "thresholds": {"mode": "absolute", "steps": [
{"color": color, "value": value} for value, color in steps]}, "color": {"mode": "thresholds"}}
if decimals is not None:
defaults["decimals"] = decimals
_panels.append({"type": "stat", "id": _id(), "title": title, "description": description, "datasource": PROM,
"gridPos": line.place(width), "targets": _with_refs([prom(expr, instant=True)]),
"fieldConfig": {"defaults": defaults, "overrides": []},
"options": {"reduceOptions": {"calcs": ["lastNotNull"], "fields": "", "values": False},
"colorMode": "background", "graphMode": "none", "textMode": "value"}})
def quantiles(metric, selector, by="le"):
return [prom(f"histogram_quantile({q}, sum by ({by}) (rate({metric}_bucket{{{selector}}}[$__rate_interval])))",
f"p{int(q * 100)}") for q in (0.5, 0.95, 0.99)]
def build():
row("Обзор и SLO (окно SLO — 1 час)")
line = Line(4)
stat(line, 3, "RPS API", "Бизнес-запросы API в секунду (без /health), все реплики.",
"messenger:http_requests:rate1m", "reqps", [(None, "blue")])
stat(line, 3, "Доля 5xx (1m)", "5xx / все бизнес-запросы за 1 минуту. Порог алерта 5%.",
"messenger:http_error_ratio:rate1m", "percentunit", [(None, "green"), (0.01, "orange"), (0.05, "red")], 2)
stat(line, 3, "p95 API (5m)", "95-й перцентиль из buckets гистограммы всех реплик (не среднее p95).",
"messenger:http_request_duration_seconds:p95_5m", "s", [(None, "green"), (0.25, "orange"), (1, "red")])
stat(line, 3, "Старейшее задание", "Возраст старейшего готового к обработке задания (queue lag). Порог алерта 30 с.",
f"max(messenger_queue_oldest_age_seconds{{{QUEUE}}})", "s", [(None, "green"), (10, "orange"), (30, "red")])
stat(line, 3, "SLO-1 доступность", "Доля бизнес-запросов без 5xx за 1 час. Цель 99%.",
"messenger:slo_availability:ratio_1h", "percentunit", [(None, "red"), (0.99, "green")], 2)
stat(line, 3, "Бюджет ошибок SLO-1", "Остаток бюджета ошибок (1% запросов) за 1 час; < 0 — SLO нарушен.",
"messenger:slo_availability:error_budget_remaining_1h", "percentunit",
[(None, "red"), (0, "orange"), (0.5, "green")], 0)
stat(line, 3, "SLO-2 POST < 250 мс", "Доля успешных POST /chats/{chat_id}/messages быстрее 250 мс за 1 час. Цель 95%.",
"messenger:slo_latency_messages:ratio_1h", "percentunit", [(None, "red"), (0.95, "green")], 2)
stat(line, 3, "SLO-3 ready ≤ 30 с", "Доля вложений, ставших ready не позже 30 с после 202, за 1 час. Цель 95%.",
"messenger:slo_time_to_ready:ratio_1h", "percentunit", [(None, "red"), (0.95, "green")], 2)
line.end()
line = Line(5)
_panels.append({"type": "table", "id": _id(), "title": "Алерты Prometheus (pending/firing)",
"description": "ALERTS из Prometheus: имя, состояние, severity. Пусто — все правила inactive.",
"datasource": PROM, "gridPos": line.place(24),
"targets": _with_refs([prom('ALERTS{alertstate=~"pending|firing"}', instant=True, fmt="table")]),
"transformations": [{"id": "organize", "options": {"excludeByName": {
"Time": True, "Value": True, "__name__": True}}}],
"fieldConfig": {"defaults": {}, "overrides": []}, "options": {"showHeader": True}})
line.end()
row("API: запросы, ошибки, задержка (RED)")
line = Line(8)
timeseries(line, 8, "RPS по маршрутам", "Шаблон маршрута, а не URL: метка ограниченной кардинальности.",
[prom(f"sum by (http_request_method, http_route) (rate(http_server_request_duration_seconds_count{{{API}}}[$__rate_interval]))",
"{{http_request_method}} {{http_route}}")], "reqps")
timeseries(line, 8, "RPS по классу статуса", "2xx/4xx/5xx. 4xx — ошибки клиента, в SLO не входят.",
[prom(f"sum by (http_response_status_class) (rate(http_server_request_duration_seconds_count{{{API}}}[$__rate_interval]))",
"{{http_response_status_class}}")], "reqps", stack=True)
timeseries(line, 8, "Доля 5xx", "Числитель: 5xx бизнес-запросов; знаменатель: все бизнес-запросы. Линия — порог алерта.",
[prom(f'(sum(rate(http_server_request_duration_seconds_count{{{API}, http_response_status_class="5xx"}}[$__rate_interval])) or on() vector(0)) '
f"/ sum(rate(http_server_request_duration_seconds_count{{{API}}}[$__rate_interval]))", "5xx ratio")],
"percentunit", thresholds=[(0.05, "red")], minimum=0)
line.end()
line = Line(8)
timeseries(line, 8, "Задержка API p50 / p95 / p99", "histogram_quantile по сумме buckets всех реплик.",
quantiles("http_server_request_duration_seconds", API), "s")
timeseries(line, 8, "p95 по маршрутам", "Где именно медленно.",
[prom(f"histogram_quantile(0.95, sum by (le, http_route) (rate(http_server_request_duration_seconds_bucket{{{API}}}[$__rate_interval])))",
"{{http_route}}")], "s")
timeseries(line, 8, "Запросы в обработке (по репликам)", "http.server.active_requests: насыщение реплик API.",
[prom('sum by (instance) (http_server_active_requests{job="messenger-api"})', "{{instance}}")], "short")
line.end()
row("Фоновая обработка: очередь, worker, время до ready")
line = Line(8)
timeseries(line, 8, "Попытки заданий в секунду по исходу", "ready / retry / failed / stale / lease_lost (messenger.job.duration count).",
[prom("sum by (outcome) (rate(messenger_job_duration_seconds_count[$__rate_interval]))", "{{outcome}}")],
"ops", stack=True)
timeseries(line, 8, "Длительность попытки p50 / p95 / p99 (ready)", "От claim до COMMIT ready в worker.",
quantiles("messenger_job_duration_seconds", 'outcome="ready"'), "s")
timeseries(line, 8, "Время до ready p50 / p95 / p99", "От устойчивого приёма (COMMIT задания в API) до COMMIT ready. Линия — SLO-3 (30 с).",
quantiles("messenger_attachment_time_to_ready_seconds", 'job_type="attachment.process.v1"'), "s",
thresholds=[(30, "red")])
line.end()
line = Line(8)
timeseries(line, 8, "Глубина очереди и backlog", "Сообщения в RabbitMQ (очередь и DLQ) и незавершённые задания в PostgreSQL.",
[prom("max by (messaging_destination_name) (messenger_queue_depth)", "broker {{messaging_destination_name}}"),
prom("max by (status) (messenger_jobs_backlog)", "jobs {{status}}")], "short")
timeseries(line, 8, "Возраст старейшего задания и outbox", "Queue lag (worker не успевает) и publish lag (dispatcher/брокер). Линия — порог алерта.",
[prom(f"max(messenger_queue_oldest_age_seconds{{{QUEUE}}})", "oldest queued job"),
prom("max(messenger_outbox_oldest_age_seconds)", "oldest unpublished outbox row")], "s",
thresholds=[(30, "red")])
timeseries(line, 8, "Доставки и публикации", "Worker: processed / noop / dead_lettered / requeued; dispatcher: ok / error.",
[prom("sum by (outcome) (rate(messenger_messages_consumed_total[$__rate_interval]))", "consumed {{outcome}}"),
prom("sum by (outcome) (rate(messenger_messages_published_total[$__rate_interval]))", "published {{outcome}}")],
"ops")
line.end()
row("Насыщение: пул БД, worker, CPU/RAM")
line = Line(8)
timeseries(line, 8, "Пул PostgreSQL API: занято / максимум / ждут", "used растёт до max и появляются ждущие — запросы ждут соединение (медленная БД или блокировки).",
[prom('sum by (instance) (db_client_connection_count{db_client_connection_state="used"})', "used {{instance}}"),
prom("max(db_client_connection_max)", "max per replica"),
prom("sum by (instance) (db_client_connection_pending_requests)", "waiting {{instance}}")], "short")
timeseries(line, 8, "Ожидание соединения из пула p95", "Время получения соединения из пула API (не время запроса).",
[prom("histogram_quantile(0.95, sum by (le, instance) (rate(db_client_connection_wait_time_seconds_bucket[$__rate_interval])))",
"p95 {{instance}}")], "s")
timeseries(line, 8, "Worker: занятость (busy / capacity)", "Доля занятых потоков-потребителей. 100% — worker насыщен, очередь растёт.",
[prom("sum(messenger_worker_busy) / sum(messenger_worker_capacity)", "saturation")], "percentunit",
thresholds=[(0.8, "orange"), (1, "red")], minimum=0)
line.end()
line = Line(8)
timeseries(line, 8, "CPU процессов (ядра)", "rate(process.cpu.time): ядер на процесс.",
[prom("sum by (job, instance) (rate(process_cpu_time_seconds_total[$__rate_interval]))", "{{job}} {{instance}}")],
"short")
timeseries(line, 8, "CPU контейнеров / лимит", "cgroup v2 (только в Docker): доля лимита CPU контейнера, включая дочерние процессы worker.",
[prom("sum by (job, instance) (rate(container_cpu_time_seconds_total[$__rate_interval])) / on (job, instance) max by (job, instance) (container_cpu_limit)",
"{{job}} {{instance}}")], "percentunit")
timeseries(line, 8, "Память: RSS процесса и контейнер", "process.memory.usage (RSS) и container.memory.usage (cgroup).",
[prom("max by (job, instance) (process_memory_usage_bytes)", "rss {{job}} {{instance}}"),
prom("max by (job, instance) (container_memory_usage_bytes)", "cgroup {{job}} {{instance}}")], "bytes")
line.end()
row("Конвейер телеметрии: Collector и SDK")
line = Line(8)
timeseries(line, 8, "Collector: принято и отправлено", "Спаны/записи логов/точки метрик в секунду.",
[prom("sum(rate(otelcol_receiver_accepted_spans[$__rate_interval]))", "accepted spans"),
prom("sum(rate(otelcol_receiver_accepted_log_records[$__rate_interval]))", "accepted logs"),
prom("sum(rate(otelcol_receiver_accepted_metric_points[$__rate_interval]))", "accepted metric points"),
prom("sum by (exporter) (rate(otelcol_exporter_sent_spans[$__rate_interval]))", "sent spans {{exporter}}"),
prom("sum by (exporter) (rate(otelcol_exporter_sent_log_records[$__rate_interval]))", "sent logs {{exporter}}")],
"ops")
timeseries(line, 8, "Collector: потери и отказы", "send_failed — истёк повтор; enqueue_failed — очередь полна; refused — memory_limiter.",
[prom("sum by (exporter) (rate(otelcol_exporter_send_failed_spans[$__rate_interval]))", "send failed spans {{exporter}}"),
prom("sum by (exporter) (rate(otelcol_exporter_send_failed_log_records[$__rate_interval]))", "send failed logs {{exporter}}"),
prom("sum by (exporter) (rate(otelcol_exporter_enqueue_failed_spans[$__rate_interval]))", "queue full spans {{exporter}}"),
prom("sum(rate(otelcol_receiver_refused_spans[$__rate_interval]))", "refused spans (memory_limiter)")],
"ops")
timeseries(line, 8, "Очереди exporter и RSS Collector", "Заполнение sending_queue (доля ёмкости) и память процесса Collector.",
[prom("max by (exporter) (otelcol_exporter_queue_size / otelcol_exporter_queue_capacity)", "queue {{exporter}}")],
"percentunit", minimum=0)
line.end()
line = Line(8)
timeseries(line, 12, "Collector RSS", "Ограничено memory_limiter (200 MiB) при недоступном хранилище.",
[prom("max(otelcol_process_memory_rss)", "collector rss")], "bytes")
timeseries(line, 12, "SDK приложений: экспорт и потери", "Потери в процессе приложения: processor error.type=queue_full (очередь 2048 полна), exporter error.type (Collector не принял пакет).",
[prom('sum by (job) (rate(otel_sdk_exporter_span_exported_total{error_type=""}[$__rate_interval]))', "exported spans {{job}}"),
prom('sum by (job, error_type) (rate(otel_sdk_exporter_span_exported_total{error_type!=""}[$__rate_interval]))', "export failed spans {{job}} {{error_type}}"),
prom('sum by (job, error_type) (rate(otel_sdk_exporter_log_exported_total{error_type!=""}[$__rate_interval]))', "export failed logs {{job}} {{error_type}}"),
prom('sum by (job) (rate(otel_sdk_processor_span_processed_total{error_type!=""}[$__rate_interval]))', "queue full spans {{job}}")],
"ops")
line.end()
row("Логи и трассы")
line = Line(9)
_panels.append({"type": "logs", "id": _id(), "title": "Предупреждения и ошибки (Loki)",
"description": "Структурные логи всех сервисов; trace_id в записи ведёт в Tempo.",
"datasource": LOKI, "gridPos": line.place(12),
"targets": _with_refs([{"datasource": LOKI, "refId": "", "queryType": "range",
"expr": '{service_name=~"messenger-.*"} | detected_level=~"warn|error"'}]),
"options": {"showTime": True, "wrapLogMessage": True, "sortOrder": "Descending",
"enableLogDetails": True}})
_panels.append({"type": "table", "id": _id(), "title": "Трассы с ошибками (Tempo, TraceQL)",
"description": "Последние трассы со статусом ERROR: 5xx API, неудачные попытки заданий.",
"datasource": TEMPO, "gridPos": line.place(12),
"targets": _with_refs([{"datasource": TEMPO, "refId": "", "queryType": "traceql",
"query": "{ status = error }", "limit": 20, "tableType": "traces"}]),
"fieldConfig": {"defaults": {}, "overrides": []}, "options": {"showHeader": True}})
line.end()
return {
"uid": "messenger-lab6", "title": "Messenger — lab6 observability", "tags": ["messenger", "lab6"],
"timezone": "browser", "editable": False, "graphTooltip": 1, "schemaVersion": 39, "version": 1,
"refresh": "10s", "time": {"from": "now-30m", "to": "now"},
"templating": {"list": []}, "annotations": {"list": [{
"builtIn": 1, "datasource": {"type": "grafana", "uid": "-- Grafana --"}, "enable": True, "hide": True,
"iconColor": "rgba(0, 211, 255, 1)", "name": "Annotations & Alerts", "type": "dashboard"}]},
"panels": _panels,
}
if __name__ == "__main__":
OUT.parent.mkdir(parents=True, exist_ok=True)
OUT.write_text(json.dumps(build(), ensure_ascii=False, indent=2) + "\n", encoding="utf-8")
print(f"written {OUT.relative_to(ROOT)}: {len(_panels)} panels")