85 lines
2.8 KiB
YAML
85 lines
2.8 KiB
YAML
|
groups:
|
||
|
|
- name: ccip_alerts
|
||
|
|
interval: 30s
|
||
|
|
rules:
|
||
|
|
# CCIP message failure rate
|
||
|
|
- alert: CCIPHighFailureRate
|
||
|
|
expr: rate(ccip_messages_failed_total[5m]) / rate(ccip_messages_sent_total[5m]) > 0.05
|
||
|
|
for: 5m
|
||
|
|
labels:
|
||
|
|
severity: critical
|
||
|
|
annotations:
|
||
|
|
summary: "CCIP has high message failure rate"
|
||
|
|
description: "CCIP message failure rate is {{ $value | humanizePercentage }} (threshold: 5%)"
|
||
|
|
|
||
|
|
# CCIP message latency
|
||
|
|
- alert: CCIPHighLatency
|
||
|
|
expr: histogram_quantile(0.95, rate(ccip_message_latency_seconds_bucket[5m])) > 300
|
||
|
|
for: 5m
|
||
|
|
labels:
|
||
|
|
severity: warning
|
||
|
|
annotations:
|
||
|
|
summary: "CCIP message latency is high"
|
||
|
|
description: "P95 CCIP message latency is {{ $value }}s (threshold: 300s)"
|
||
|
|
|
||
|
|
# CCIP router unavailable
|
||
|
|
- alert: CCIPRouterUnavailable
|
||
|
|
expr: up{job="ccip-router"} == 0
|
||
|
|
for: 2m
|
||
|
|
labels:
|
||
|
|
severity: critical
|
||
|
|
annotations:
|
||
|
|
summary: "CCIP Router is unavailable"
|
||
|
|
description: "CCIP Router {{ $labels.instance }} is down"
|
||
|
|
|
||
|
|
# CCIP low success rate
|
||
|
|
- alert: CCIPLowSuccessRate
|
||
|
|
expr: rate(ccip_messages_success_total[5m]) / rate(ccip_messages_sent_total[5m]) < 0.95
|
||
|
|
for: 5m
|
||
|
|
labels:
|
||
|
|
severity: warning
|
||
|
|
annotations:
|
||
|
|
summary: "CCIP message success rate is low"
|
||
|
|
description: "CCIP success rate is {{ $value | humanizePercentage }} (threshold: 95%)"
|
||
|
|
|
||
|
|
# CCIP fee balance low
|
||
|
|
- alert: CCIPLowLinkBalance
|
||
|
|
expr: ccip_link_balance < 10
|
||
|
|
for: 5m
|
||
|
|
labels:
|
||
|
|
severity: warning
|
||
|
|
annotations:
|
||
|
|
summary: "CCIP LINK balance is low"
|
||
|
|
description: "CCIP contract has only {{ $value }} LINK remaining (threshold: 10 LINK)"
|
||
|
|
|
||
|
|
# CCIP fee balance critical
|
||
|
|
- alert: CCIPCriticalLinkBalance
|
||
|
|
expr: ccip_link_balance < 5
|
||
|
|
for: 2m
|
||
|
|
labels:
|
||
|
|
severity: critical
|
||
|
|
annotations:
|
||
|
|
summary: "CCIP LINK balance is critically low"
|
||
|
|
description: "CCIP contract has only {{ $value }} LINK remaining (threshold: 5 LINK)"
|
||
|
|
|
||
|
|
# CCIP message queue backlog
|
||
|
|
- alert: CCIPMessageBacklog
|
||
|
|
expr: ccip_messages_pending_total > 100
|
||
|
|
for: 10m
|
||
|
|
labels:
|
||
|
|
severity: warning
|
||
|
|
annotations:
|
||
|
|
summary: "CCIP message backlog detected"
|
||
|
|
description: "{{ $value }} messages are pending in CCIP queue"
|
||
|
|
|
||
|
|
# CCIP fee calculation errors
|
||
|
|
- alert: CCIPFeeCalculationErrors
|
||
|
|
expr: rate(ccip_fee_calculation_errors_total[5m]) > 0.1
|
||
|
|
for: 5m
|
||
|
|
labels:
|
||
|
|
severity: warning
|
||
|
|
annotations:
|
||
|
|
summary: "CCIP fee calculation errors detected"
|
||
|
|
description: "CCIP fee calculation error rate is {{ $value }} errors/second"
|
||
|
|
|