didi-lot2-backend/backend/observability/prometheus/prometheus.yml
2026-07-10 03:39:53 -07:00

104 lines
3.3 KiB
YAML

global:
scrape_interval: 30s
evaluation_interval: 30s
external_labels:
cluster: didi-prod
environment: production
rule_files:
- /etc/prometheus/rules/*.yml
alerting:
alertmanagers:
- static_configs:
- targets:
- didi-alertmanager:9093
scrape_configs:
# Prometheus self-monitoring
- job_name: prometheus
static_configs:
- targets: ['localhost:9090']
# Backend Node.js services (expun /metrics via prom-client)
- job_name: agent-v3
metrics_path: /metrics
static_configs:
- targets: ['didi-agent-v3:24803']
labels: { service: agent-v3, layer: orchestration }
- job_name: didi-framework
metrics_path: /metrics
static_configs:
- targets: ['didi-framework:3005']
labels: { service: framework, layer: orchestration }
# AI Platform Python services (expun /metrics via prometheus_client)
- job_name: ai-platform
metrics_path: /metrics
static_configs:
- targets:
- 'didiAI-llm-api:14011'
- 'didiAI-embeddings-api:14100'
- 'didiAI-rerank-api:14200'
- 'didiAI-audio-api:54300'
- 'didiAI-video-api:54600'
- 'didiAI-web-api:51100'
- 'didiAI-catalog-api:11000'
- 'didiAI-dashboard:51300'
- 'didibrain-api:8090'
labels: { layer: ai-platform }
# Workers (expun /metrics pe portul intern)
- job_name: agent-v3-workers
metrics_path: /metrics
static_configs:
- targets:
- 'agent-v3-worker-techniques-1:24803'
- 'agent-v3-worker-techniques-2:24803'
- 'agent-v3-worker-ai-tampered-1:24803'
- 'agent-v3-worker-ai-tampered-2:24803'
- 'agent-v3-worker-claims-1:24803'
- 'agent-v3-worker-claims-2:24803'
- 'agent-v3-worker-claims-3:24803'
- 'agent-v3-worker-domain-1:24803'
- 'agent-v3-worker-domain-2:24803'
- 'agent-v3-worker-media-preprocess-1:24803'
- 'agent-v3-worker-media-preprocess-2:24803'
- 'agent-v3-verdict-aggregator-1:24803'
- 'agent-v3-verdict-aggregator-2:24803'
labels: { service: agent-v3, layer: workers }
# Infrastructure metrics (cadvisor + node-exporter already running on host)
- job_name: cadvisor
static_configs:
- targets: ['heimdall_cadvisor:8080']
labels: { layer: infrastructure }
- job_name: node-exporter
static_configs:
- targets: ['heimdall_node_exporter:9100']
labels: { layer: infrastructure }
# OTel Collector self-metrics + traces (metric forwarder)
- job_name: otel-collector
static_configs:
- targets: ['didi-otel-collector:8889']
labels: { layer: observability }
# PostgreSQL via postgres_exporter (opt-in — add postgres-exporter container if needed)
# - job_name: postgres-exporter
# static_configs:
# - targets: ['postgres-exporter:9187']
# RabbitMQ has built-in prometheus support since 3.10 — enable rabbitmq_prometheus plugin
- job_name: rabbitmq
metrics_path: /metrics
static_configs:
- targets: ['staging-dataLayer-rabbitmq:15692']
labels: { service: rabbitmq, layer: data }
# Redis via redis_exporter (opt-in)
# - job_name: redis-exporter
# static_configs:
# - targets: ['redis-exporter:9121']