62 lines
1.7 KiB
YAML
62 lines
1.7 KiB
YAML
|
# Prometheus metrics configuration for Bridge
|
||
|
|
|
||
|
|
global:
|
||
|
|
scrape_interval: 15s
|
||
|
|
evaluation_interval: 15s
|
||
|
|
|
||
|
|
scrape_configs:
|
||
|
|
- job_name: 'bridge-api'
|
||
|
|
static_configs:
|
||
|
|
- targets: ['localhost:8080']
|
||
|
|
metrics_path: '/metrics'
|
||
|
|
|
||
|
|
- job_name: 'bridge-orchestrator'
|
||
|
|
static_configs:
|
||
|
|
- targets: ['localhost:9090']
|
||
|
|
metrics_path: '/metrics'
|
||
|
|
|
||
|
|
- job_name: 'bridge-observability'
|
||
|
|
static_configs:
|
||
|
|
- targets: ['localhost:9091']
|
||
|
|
metrics_path: '/metrics'
|
||
|
|
|
||
|
|
# Alert rules
|
||
|
|
groups:
|
||
|
|
- name: bridge_alerts
|
||
|
|
interval: 30s
|
||
|
|
rules:
|
||
|
|
- alert: HighFailureRate
|
||
|
|
expr: rate(bridge_failure_count[5m]) > 0.1
|
||
|
|
for: 5m
|
||
|
|
annotations:
|
||
|
|
summary: "Bridge failure rate is high"
|
||
|
|
description: "Failure rate is {{ $value }} failures per second"
|
||
|
|
|
||
|
|
- alert: LowSuccessRate
|
||
|
|
expr: bridge_success_rate < 95
|
||
|
|
for: 10m
|
||
|
|
annotations:
|
||
|
|
summary: "Bridge success rate is below threshold"
|
||
|
|
description: "Success rate is {{ $value }}%"
|
||
|
|
|
||
|
|
- alert: HighSettlementTime
|
||
|
|
expr: bridge_avg_settlement_time > 600
|
||
|
|
for: 15m
|
||
|
|
annotations:
|
||
|
|
summary: "Average settlement time is high"
|
||
|
|
description: "Average settlement time is {{ $value }} seconds"
|
||
|
|
|
||
|
|
- alert: LiquidityFailure
|
||
|
|
expr: increase(bridge_liquidity_failures[1h]) > 5
|
||
|
|
for: 5m
|
||
|
|
annotations:
|
||
|
|
summary: "Multiple liquidity failures detected"
|
||
|
|
description: "{{ $value }} liquidity failures in the last hour"
|
||
|
|
|
||
|
|
- alert: BridgePaused
|
||
|
|
expr: bridge_paused == 1
|
||
|
|
for: 1m
|
||
|
|
annotations:
|
||
|
|
summary: "Bridge is paused"
|
||
|
|
description: "Bridge operations have been paused"
|