77 lines
2.4 KiB
YAML
77 lines
2.4 KiB
YAML
groups:
|
|
- name: Elastic-HA-Proxy-Alerts
|
|
rules:
|
|
- alert: haproxy_backend_status
|
|
expr: avg by(proxy)(haproxy_backend_status{job=~"elastic-fallback-haproxy.*", state="UP"}) < 1
|
|
for: 1m
|
|
labels:
|
|
severity: critical
|
|
bu: infra
|
|
team: devops
|
|
service: elastic-ha-proxy
|
|
env: prd
|
|
priority: p0
|
|
annotations:
|
|
summary: "haproxy backend status is down"
|
|
description: "haproxy backend status is down VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
- alert: Frontend_in_state_Elastic_HAProxy
|
|
expr: sum(rate(haproxy_frontend_bytes_in_total{job=~"elastic-fallback-haproxy.*"}[1m])) by (proxy) >= 650
|
|
for: 1m
|
|
labels:
|
|
severity: critical
|
|
bu: infra
|
|
team: devops
|
|
service: elastic-ha-proxy
|
|
env: prd
|
|
priority: p0
|
|
annotations:
|
|
summary: "Frontend in state Elastic HAProxy"
|
|
description: "Frontend_in_state_Elastic_HAProxy is greater than VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
|
|
- alert: haproxy_backend_5xx_total
|
|
expr: sum(rate(haproxy_backend_http_responses_total{job="elastic-fallback-haproxy.*",code=~"5xx|other"}[1m])) by (proxy,code) * 60 >= 20
|
|
for: 1m
|
|
labels:
|
|
severity: critical
|
|
bu: infra
|
|
team: devops
|
|
service: elastic-ha-proxy
|
|
env: prd
|
|
priority: p0
|
|
annotations:
|
|
summary: "haproxy backend 5xx is high"
|
|
description: "haproxy backend 5xx is greater VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: haproxy_backend_4xx_total
|
|
expr: sum(rate(haproxy_backend_http_responses_total{job="elastic-fallback-haproxy.*",code=~"4xx|other"}[1m])) by (proxy,code) * 60 >= 20
|
|
for: 1m
|
|
labels:
|
|
severity: critical
|
|
bu: infra
|
|
team: devops
|
|
service: elastic-ha-proxy
|
|
env: prd
|
|
priority: p0
|
|
annotations:
|
|
summary: "haproxy backend 4xx status is down"
|
|
description: "haproxy backend 4xx is down VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|
|
|
|
|
|
- alert: haproxy_frontend_bytes_out_total
|
|
expr: sum(rate(haproxy_frontend_bytes_out_total{job="elastic-fallback-haproxy.*"}[1m])) by (proxy) >= 650
|
|
for: 1m
|
|
labels:
|
|
severity: critical
|
|
bu: infra
|
|
team: devops
|
|
service: elastic-ha-proxy
|
|
env: prd
|
|
priority: p0
|
|
annotations:
|
|
summary: "haproxy frontend bytes out total is high"
|
|
description: "haproxy_frontend_bytes_out_total is greater than VALUE = {{ $value }}\n LABELS: {{ $labels }}"
|