gitadmin/devops-infra-helm-charts-gcp
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

added repo

Browse Source
This commit is contained in:
Your Name
2026-08-26 03:39:42 +05:30
parent 45c25a95af
commit b8575bb8b9
6889 changed files with 1217125 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/ads-credit-HTTP_4XX_Alert_Contour.yaml
+17
View File
@@ -0,0 +1,17 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: HTTP_4XX_Alert_Contour_EKS
expr: sum(rate(envoy_cluster_upstream_rq{envoy_cluster_name=~"prd-ads-credit_prd-ads-credit_80",envoy_response_code=~"[4].*"}[1m])) by (cluster, envoy_cluster_name, job) * 60 >= 1
for: 1m
labels:
severity: critical
bu: supply
team: supplier-ads
service: ads-credit
env: prd
priority: p1
annotations:
summary: "EKS HTTP_4XX_Alert_Contour on {{ $labels.service }}"
description: "EKS HTTP_4XX_Alert_Contour on {{ $labels.service }} VALUE = {{ $value }}
LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/ads-credit-HTTP_5XX_Alert_Contour.yaml
+17
View File
@@ -0,0 +1,17 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: HTTP_5XX_Alert_Contour_EKS
expr: sum(rate(envoy_cluster_upstream_rq{envoy_cluster_name=~"prd-ads-credit_prd-ads-credit_80",envoy_response_code=~"[5].*"}[1m])) by (cluster, envoy_cluster_name, job) * 60 >= 1
for: 1m
labels:
severity: critical
bu: supply
team: supplier-ads
service: ads-credit
env: prd
priority: p1
annotations:
summary: "EKS HTTP_5XX_Alert_Contour on {{ $labels.service }}"
description: "EKS HTTP_5XX_Alert_Contour on {{ $labels.service }} VALUE = {{ $value }}
LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/ads-credit-P99_Latency_Alert_Contour.yaml
+17
View File
@@ -0,0 +1,17 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: P99_Latency_Alert_Contour_EKS
expr: histogram_quantile(0.99, sum(rate(envoy_cluster_upstream_rq_time_bucket{envoy_cluster_name=~"prd-ads-credit_prd-ads-credit_80"}[1m])) by (cluster, envoy_cluster_name, le, job)) >= 750
for: 1m
labels:
severity: critical
bu: supply
team: supplier-ads
service: ads-credit
env: prd
priority: p1
annotations:
summary: "EKS P99_Latency_Alert_Contour on {{ $labels.service }}"
description: "EKS P99_Latency_Alert_Contour on {{ $labels.service }} VALUE = {{ $value }}
LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/ads-payment-HTTP_4XX_Alert_Contour.yaml
+17
View File
@@ -0,0 +1,17 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: HTTP_4XX_Alert_Contour_EKS
expr: sum(rate(envoy_cluster_upstream_rq{envoy_cluster_name=~"prd-ads-payment_prd-ads-payment_8",envoy_response_code=~"[4].*"}[1m])) by (cluster, envoy_cluster_name, job) * 60 >= 10
for: 2m
labels:
severity: critical
bu: supply
team: supplier-ads
service: ads-payment
env: prd
priority: p1
annotations:
summary: "EKS HTTP_4XX_Alert_Contour on {{ $labels.service }}"
description: "EKS HTTP_4XX_Alert_Contour on {{ $labels.service }} VALUE = {{ $value }}
LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/ads-payment-HTTP_5XX_Alert.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: HTTP_5XX_Alert_EKS
expr: sum(rate(nginx_ingress_controller_requests{cluster!="prod-ops-cluster", status=~"[5].*", service=~"prd-ads-payment", job=~"kubernetes-pods"}[1m])) by (cluster, service, ingress, job) * 60 >= 10
for: 2m
labels:
severity: critical
bu: supply
team: supplier-ads
service: ads-payment
env: prd
priority: p1
annotations:
summary: "EKS HTTP 5XX Alert on {{ $labels.service }}"
description: "EKS HTTP 5XX Alert on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/ads-payment-HTTP_5XX_Alert_Contour.yaml
+17
View File
@@ -0,0 +1,17 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: HTTP_5XX_Alert_Contour_EKS
expr: sum(rate(envoy_cluster_upstream_rq{envoy_cluster_name=~"prd-ads-payment_prd-ads-payment_8",envoy_response_code=~"[5].*"}[1m])) by (cluster, envoy_cluster_name, job) * 60 >= 10
for: 2m
labels:
severity: critical
bu: supply
team: supplier-ads
service: ads-payment
env: prd
priority: p1
annotations:
summary: "EKS HTTP_5XX_Alert_Contour on {{ $labels.service }}"
description: "EKS HTTP_5XX_Alert_Contour on {{ $labels.service }} VALUE = {{ $value }}
LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/ads-payment-Latency_P99_Alert.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: Latency_P99_Alert_EKS
expr: histogram_quantile(0.99, sum(rate(nginx_ingress_controller_request_duration_seconds_bucket{cluster!="prod-ops-cluster", service=~"prd-ads-payment", job=~"kubernetes-pods"}[1m])) by (cluster, service, le, ingress, namespace, job)) >= 0.75
for: 1m
labels:
severity: critical
bu: supply
team: supplier-ads
service: ads-payment
env: prd
priority: p1
annotations:
summary: "EKS Latency/Response Time P99 Alert on {{ $labels.service }}"
description: "EKS Latency/Response Time P99 Alert on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/ads-payment-P99_Latency_Alert_Contour.yaml
+17
View File
@@ -0,0 +1,17 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: P99_Latency_Alert_Contour_EKS
expr: histogram_quantile(0.99, sum(rate(envoy_cluster_upstream_rq_time_bucket{envoy_cluster_name=~"prd-ads-payment_prd-ads-payment_8"}[1m])) by (cluster, envoy_cluster_name, le, job)) >= 750
for: 1m
labels:
severity: critical
bu: supply
team: supplier-ads
service: ads-payment
env: prd
priority: p1
annotations:
summary: "EKS P99_Latency_Alert_Contour on {{ $labels.service }}"
description: "EKS P99_Latency_Alert_Contour on {{ $labels.service }} VALUE = {{ $value }}
LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/ads-rti-enricher-listener-HTTP_4XX_Alert_Contour.yaml
+17
View File
@@ -0,0 +1,17 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: HTTP_4XX_Alert_Contour_EKS
expr: sum(rate(envoy_cluster_upstream_rq{envoy_cluster_name=~"prd-ads-rti-enricher-listener_prd-ads-rti-enricher-listener_80",envoy_response_code=~"[4].*"}[1m])) by (cluster, envoy_cluster_name, job) * 60 >= 200
for: 10m
labels:
severity: critical
bu: supply
team: supplier-ads
service: ads-rti-enricher-listener
env: prd
priority: p1
annotations:
summary: "EKS HTTP_4XX_Alert_Contour on {{ $labels.service }}"
description: "EKS HTTP_4XX_Alert_Contour on {{ $labels.service }} VALUE = {{ $value }}
LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/ads-rti-enricher-listener-HTTP_5XX_Alert_Contour.yaml
+17
View File
@@ -0,0 +1,17 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: HTTP_5XX_Alert_Contour_EKS
expr: sum(rate(envoy_cluster_upstream_rq{envoy_cluster_name=~"prd-ads-rti-enricher-listener_prd-ads-rti-enricher-listener_80",envoy_response_code=~"[5].*"}[1m])) by (cluster, envoy_cluster_name, job) * 60 >= 13
for: 10m
labels:
severity: critical
bu: supply
team: supplier-ads
service: ads-rti-enricher-listener
env: prd
priority: p1
annotations:
summary: "EKS HTTP_5XX_Alert_Contour on {{ $labels.service }}"
description: "EKS HTTP_5XX_Alert_Contour on {{ $labels.service }} VALUE = {{ $value }}
LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/ads-rti-enricher-listener-P99_Latency_Alert_Contour.yaml
+17
View File
@@ -0,0 +1,17 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: P99_Latency_Alert_Contour_EKS
expr: histogram_quantile(0.99, sum(rate(envoy_cluster_upstream_rq_time_bucket{envoy_cluster_name=~"prd-ads-rti-enricher-listener_prd-ads-rti-enricher-listener_80"}[1m])) by (cluster, envoy_cluster_name, le, job)) >= 750
for: 1m
labels:
severity: critical
bu: supply
team: supplier-ads
service: ads-rti-enricher-listener
env: prd
priority: p1
annotations:
summary: "EKS P99_Latency_Alert_Contour on {{ $labels.service }}"
description: "EKS P99_Latency_Alert_Contour on {{ $labels.service }} VALUE = {{ $value }}
LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/ads-rti-es-indexer-eks-HTTP_4XX_Alert_Contour.yaml
+17
View File
@@ -0,0 +1,17 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: HTTP_4XX_Alert_Contour_EKS
expr: sum(rate(envoy_cluster_upstream_rq{envoy_cluster_name=~"prd-ads-rti-es-indexer-eks_prd-ads-rti-es-indexer-eks_80",envoy_response_code=~"[4].*"}[1m])) by (cluster, envoy_cluster_name, job) * 60 >= 200
for: 10m
labels:
severity: critical
bu: supply
team: supplier-ads
service: ads-rti-es-indexer-eks
env: prd
priority: p1
annotations:
summary: "EKS HTTP_4XX_Alert_Contour on {{ $labels.service }}"
description: "EKS HTTP_4XX_Alert_Contour on {{ $labels.service }} VALUE = {{ $value }}
LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/ads-rti-es-indexer-eks-HTTP_5XX_Alert_Contour.yaml
+17
View File
@@ -0,0 +1,17 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: HTTP_5XX_Alert_Contour_EKS
expr: sum(rate(envoy_cluster_upstream_rq{envoy_cluster_name=~"prd-ads-rti-es-indexer-eks_prd-ads-rti-es-indexer-eks_80",envoy_response_code=~"[5].*"}[1m])) by (cluster, envoy_cluster_name, job) * 60 >= 13
for: 10m
labels:
severity: critical
bu: supply
team: supplier-ads
service: ads-rti-es-indexer-eks
env: prd
priority: p1
annotations:
summary: "EKS HTTP_5XX_Alert_Contour on {{ $labels.service }}"
description: "EKS HTTP_5XX_Alert_Contour on {{ $labels.service }} VALUE = {{ $value }}
LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/ads-rti-es-indexer-eks-P99_Latency_Alert_Contour.yaml
+17
View File
@@ -0,0 +1,17 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: P99_Latency_Alert_Contour_EKS
expr: histogram_quantile(0.99, sum(rate(envoy_cluster_upstream_rq_time_bucket{envoy_cluster_name=~"prd-ads-rti-es-indexer-eks_prd-ads-rti-es-indexer-eks_80"}[1m])) by (cluster, envoy_cluster_name, le, job)) >= 750
for: 1m
labels:
severity: critical
bu: supply
team: supplier-ads
service: ads-rti-es-indexer-eks
env: prd
priority: p1
annotations:
summary: "EKS P99_Latency_Alert_Contour on {{ $labels.service }}"
description: "EKS P99_Latency_Alert_Contour on {{ $labels.service }} VALUE = {{ $value }}
LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/advertisement-HTTP_4XX_Alert.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: HTTP_4XX_Alert_EKS
expr: sum(rate(nginx_ingress_controller_requests{cluster!="prod-ops-cluster", status=~"[4].*", service=~"prd-advertisement", job=~"kubernetes-pods"}[1m])) by (cluster, service, ingress, job) * 60 >= 40
for: 1m
labels:
severity: critical
bu: supply
team: supplier-ads
service: advertisement
env: prd
priority: p1
annotations:
summary: "EKS HTTP 4XX Alert on {{ $labels.service }}"
description: "EKS HTTP 4XX Alert on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/advertisement-HTTP_4XX_Alert_Contour.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: HTTP_4XX_Alert_Contour_EKS
expr: sum(rate(envoy_cluster_upstream_rq{envoy_cluster_name=~"prd-advertisement_prd-advertisement_80",envoy_response_code=~"[4].*"}[1m])) by (cluster, envoy_cluster_name, job) * 60 >= 40
for: 1m
labels:
severity: critical
bu: supply
team: supplier-ads
service: advertisement
env: prd
priority: p1
annotations:
summary: "EKS HTTP_4XX_Alert_Contour on {{ $labels.service }}"
description: "EKS HTTP_4XX_Alert_Contour on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/advertisement-HTTP_5XX_Alert.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: HTTP_5XX_Alert_EKS
expr: sum(rate(nginx_ingress_controller_requests{cluster!="prod-ops-cluster", status=~"[5].*", service=~"prd-advertisement", job=~"kubernetes-pods"}[1m])) by (cluster, service, ingress, job) * 60 >= 150
for: 1m
labels:
severity: critical
bu: supply
team: supplier-ads
service: advertisement
env: prd
priority: p1
annotations:
summary: "EKS HTTP 5XX Alert on {{ $labels.service }}"
description: "EKS HTTP 5XX Alert on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/advertisement-HTTP_5XX_Alert_Contour.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: HTTP_5XX_Alert_Contour_EKS
expr: sum(rate(envoy_cluster_upstream_rq{envoy_cluster_name=~"prd-advertisement_prd-advertisement_80",envoy_response_code=~"[5].*"}[1m])) by (cluster, envoy_cluster_name, job) * 60 >= 150
for: 1m
labels:
severity: critical
bu: supply
team: supplier-ads
service: advertisement
env: prd
priority: p1
annotations:
summary: "EKS HTTP_5XX_Alert_Contour on {{ $labels.service }}"
description: "EKS HTTP_5XX_Alert_Contour on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/advertisement-Latency_P99_Alert.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: Latency_P99_Alert_EKS
expr: histogram_quantile(0.99, sum(rate(nginx_ingress_controller_request_duration_seconds_bucket{cluster!="prod-ops-cluster", service=~"prd-advertisement", job=~"kubernetes-pods"}[1m])) by (cluster, service, le, ingress, namespace, job)) >= 0.75
for: 1m
labels:
severity: critical
bu: supply
team: supplier-ads
service: advertisement
env: prd
priority: p1
annotations:
summary: "EKS Latency/Response Time P99 Alert on {{ $labels.service }}"
description: "EKS Latency/Response Time P99 Alert on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/advertisement-Latency_P99_Alert_Contour.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: Latency_P99_Alert_Contour_EKS
expr: histogram_quantile(0.99, sum(rate(envoy_cluster_upstream_rq_time_bucket{envoy_cluster_name=~"prd-advertisement_prd-advertisement_80"}[1m])) by (cluster, envoy_cluster_name, le, job)) >= 200
for: 1m
labels:
severity: critical
bu: supply
team: supplier-ads
service: advertisement
env: prd
priority: p1
annotations:
summary: "EKS Latency_P99_Alert_Contour on {{ $labels.service }}"
description: "EKS Latency_P99_Alert_Contour on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/advertisement-admin-HTTP_4XX_Alert.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: HTTP_4XX_Alert_EKS
expr: sum(rate(nginx_ingress_controller_requests{cluster!="prod-ops-cluster", status=~"[4].*", service=~"prd-advertisement-admin", job=~"kubernetes-pods"}[1m])) by (cluster, service, ingress, job) * 60 >= 200
for: 10m
labels:
severity: critical
bu: supply
team: supplier-ads
service: advertisement-admin
env: prd
priority: p1
annotations:
summary: "EKS HTTP 4XX Alert on {{ $labels.service }}"
description: "EKS HTTP 4XX Alert on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/advertisement-admin-HTTP_4XX_Alert_Contour.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: HTTP_4XX_Alert_Contour_EKS
expr: sum(rate(envoy_cluster_upstream_rq{envoy_cluster_name=~"prd-advertisement-admin_prd-advertisement-admin_80",envoy_response_code=~"[4].*"}[1m])) by (cluster, envoy_cluster_name, job) * 60 >= 200
for: 10m
labels:
severity: critical
bu: supply
team: supplier-ads
service: advertisement-admin
env: prd
priority: p1
annotations:
summary: "EKS HTTP_4XX_Alert_Contour on {{ $labels.service }}"
description: "EKS HTTP_4XX_Alert_Contour on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/advertisement-admin-HTTP_5XX_Alert.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: HTTP_5XX_Alert_EKS
expr: sum(rate(nginx_ingress_controller_requests{cluster!="prod-ops-cluster", status=~"[5].*", service=~"prd-advertisement-admin", job=~"kubernetes-pods"}[1m])) by (cluster, service, ingress, job) * 60 >= 65
for: 10m
labels:
severity: critical
bu: supply
team: supplier-ads
service: advertisement-admin
env: prd
priority: p1
annotations:
summary: "EKS HTTP 5XX Alert on {{ $labels.service }}"
description: "EKS HTTP 5XX Alert on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/advertisement-admin-HTTP_5XX_Alert_Contour.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: HTTP_5XX_Alert_Contour_EKS
expr: sum(rate(envoy_cluster_upstream_rq{envoy_cluster_name=~"prd-advertisement-admin_prd-advertisement-admin_80",envoy_response_code=~"[5].*"}[1m])) by (cluster, envoy_cluster_name, job) * 60 >= 65
for: 10m
labels:
severity: critical
bu: supply
team: supplier-ads
service: advertisement-admin
env: prd
priority: p1
annotations:
summary: "EKS HTTP_5XX_Alert_Contour on {{ $labels.service }}"
description: "EKS HTTP_5XX_Alert_Contour on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/advertisement-admin-Latency_P99_Alert.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: Latency_P99_Alert_EKS
expr: histogram_quantile(0.99, sum(rate(nginx_ingress_controller_request_duration_seconds_bucket{cluster!="prod-ops-cluster", service=~"prd-advertisement-admin", job=~"kubernetes-pods"}[1m])) by (cluster, service, le, ingress, namespace, job)) >= 0.75
for: 1m
labels:
severity: critical
bu: supply
team: supplier-ads
service: advertisement-admin
env: prd
priority: p1
annotations:
summary: "EKS Latency/Response Time P99 Alert on {{ $labels.service }}"
description: "EKS Latency/Response Time P99 Alert on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/advertisement-admin-Latency_P99_Alert_Contour.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: Latency_P99_Alert_Contour_EKS
expr: histogram_quantile(0.99, sum(rate(envoy_cluster_upstream_rq_time_bucket{envoy_cluster_name=~"prd-advertisement-admin_prd-advertisement-admin_80"}[1m])) by (cluster, envoy_cluster_name, le, job)) >= 0.75
for: 1m
labels:
severity: critical
bu: supply
team: supplier-ads
service: advertisement-admin
env: prd
priority: p1
annotations:
summary: "EKS Latency_P99_Alert_Contour on {{ $labels.service }}"
description: "EKS Latency_P99_Alert_Contour on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/advertisement-admin/custom_alerts/Controller_latency_p99_alert.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: Controller_latency_p99_alert_EKS
expr: histogram_quantile(0.99, sum(rate(CONTROLLER_execTime_bucket{uri!~".*health.*",podNamespace="prd-advertisement-admin"}[1m])) by (le)) >= 200
for: 1m
labels:
severity: critical
bu: supply
team: supplier-ads
service: advertisement-admin
env: prd
priority: p1
annotations:
summary: "EKS Controller_latency_p99_alert on {{ $labels.service }}"
description: "EKS Controller_latency_p99_alert on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/advertisement-file-tasks-service-HTTP_4XX_Alert.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: HTTP_4XX_Alert_EKS
expr: sum(rate(nginx_ingress_controller_requests{cluster!="prod-ops-cluster", status=~"[4].*", service=~"prd-advertisement-file-tasks-service", job=~"kubernetes-pods"}[1m])) by (cluster, service, ingress, job) * 60 >= 10
for: 2m
labels:
severity: critical
bu: supply
team: supplier-ads
service: advertisement-file-tasks-service
env: prd
priority: p1
annotations:
summary: "EKS HTTP 4XX Alert on {{ $labels.service }}"
description: "EKS HTTP 4XX Alert on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/advertisement-file-tasks-service-HTTP_4XX_Alert_Contour.yaml
+17
View File
@@ -0,0 +1,17 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: HTTP_4XX_Alert_Contour_EKS
expr: sum(rate(envoy_cluster_upstream_rq{envoy_cluster_name=~"prd-advertisement-file-tasks-service_prd-advertisement-file-tasks-service_80",envoy_response_code=~"[4].*"}[1m])) by (cluster, envoy_cluster_name, job) * 60 >= 10
for: 2m
labels:
severity: critical
bu: supply
team: supplier-ads
service: advertisement-file-tasks-service
env: prd
priority: p1
annotations:
summary: "EKS HTTP_4XX_Alert_Contour on {{ $labels.service }}"
description: "EKS HTTP_4XX_Alert_Contour on {{ $labels.service }} VALUE = {{ $value }}
LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/advertisement-file-tasks-service-HTTP_5XX_Alert.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: HTTP_5XX_Alert_EKS
expr: sum(rate(nginx_ingress_controller_requests{cluster!="prod-ops-cluster", status=~"[5].*", service=~"prd-advertisement-file-tasks-service", job=~"kubernetes-pods"}[1m])) by (cluster, service, ingress, job) * 60 >= 10
for: 1m
labels:
severity: critical
bu: supply
team: supplier-ads
service: advertisement-file-tasks-service
env: prd
priority: p1
annotations:
summary: "EKS HTTP 5XX Alert on {{ $labels.service }}"
description: "EKS HTTP 5XX Alert on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/advertisement-file-tasks-service-HTTP_5XX_Alert_Contour.yaml
+17
View File
@@ -0,0 +1,17 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: HTTP_5XX_Alert_Contour_EKS
expr: sum(rate(envoy_cluster_upstream_rq{envoy_cluster_name=~"prd-advertisement-file-tasks-service_prd-advertisement-file-tasks-service_80",envoy_response_code=~"[5].*"}[1m])) by (cluster, envoy_cluster_name, job) * 60 >= 10
for: 1m
labels:
severity: critical
bu: supply
team: supplier-ads
service: advertisement-file-tasks-service
env: prd
priority: p1
annotations:
summary: "EKS HTTP_5XX_Alert_Contour on {{ $labels.service }}"
description: "EKS HTTP_5XX_Alert_Contour on {{ $labels.service }} VALUE = {{ $value }}
LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/advertisement-file-tasks-service-Latency_P99_Alert.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: Latency_P99_Alert_EKS
expr: histogram_quantile(0.99, sum(rate(nginx_ingress_controller_request_duration_seconds_bucket{cluster!="prod-ops-cluster", service=~"prd-advertisement-file-tasks-service", job=~"kubernetes-pods"}[1m])) by (cluster, service, le, ingress, namespace, job)) >= 0.75
for: 1m
labels:
severity: critical
bu: supply
team: supplier-ads
service: advertisement-file-tasks-service
env: prd
priority: p1
annotations:
summary: "EKS Latency/Response Time P99 Alert on {{ $labels.service }}"
description: "EKS Latency/Response Time P99 Alert on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/advertisement-file-tasks-service-P99_Latency_Alert_Contour.yaml
+17
View File
@@ -0,0 +1,17 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: P99_Latency_Alert_Contour_EKS
expr: histogram_quantile(0.99, sum(rate(envoy_cluster_upstream_rq_time_bucket{envoy_cluster_name=~"prd-advertisement-file-tasks-service_prd-advertisement-file-tasks-service_80"}[1m])) by (cluster, envoy_cluster_name, le, job)) >= 750
for: 1m
labels:
severity: critical
bu: supply
team: supplier-ads
service: advertisement-file-tasks-service
env: prd
priority: p1
annotations:
summary: "EKS P99_Latency_Alert_Contour on {{ $labels.service }}"
description: "EKS P99_Latency_Alert_Contour on {{ $labels.service }} VALUE = {{ $value }}
LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/advertisement-fy-HTTP_4XX_Alert.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: HTTP_4XX_Alert_EKS
expr: sum(rate(nginx_ingress_controller_requests{cluster!="prod-ops-cluster", status=~"[4].*", service=~"prd-advertisement-fy", job=~"kubernetes-pods"}[1m])) by (cluster, service, ingress, job) * 60 >= 150
for: 1m
labels:
severity: critical
bu: supply
team: supplier-ads
service: advertisement-fy
env: prd
priority: p1
annotations:
summary: "EKS HTTP 4XX Alert on {{ $labels.service }}"
description: "EKS HTTP 4XX Alert on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/advertisement-fy-HTTP_4XX_Alert_Contour.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: HTTP_4XX_Alert_Contour_EKS
expr: sum(rate(envoy_cluster_upstream_rq{envoy_cluster_name=~"prd-advertisement-fy_prd-advertisement-fy_80",envoy_response_code=~"[4].*"}[1m])) by (cluster, envoy_cluster_name, job) * 60 >= 150
for: 1m
labels:
severity: critical
bu: supply
team: supplier-ads
service: advertisement-fy
env: prd
priority: p1
annotations:
summary: "EKS HTTP_4XX_Alert_Contour on {{ $labels.service }}"
description: "EKS HTTP_4XX_Alert_Contour on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/advertisement-fy-HTTP_5XX_Alert.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: HTTP_5XX_Alert_EKS
expr: sum(rate(nginx_ingress_controller_requests{cluster!="prod-ops-cluster", status=~"[5].*", service=~"prd-advertisement-fy", job=~"kubernetes-pods"}[1m])) by (cluster, service, ingress, job) * 60 >= 150
for: 1m
labels:
severity: critical
bu: supply
team: supplier-ads
service: advertisement-fy
env: prd
priority: p1
annotations:
summary: "EKS HTTP 5XX Alert on {{ $labels.service }}"
description: "EKS HTTP 5XX Alert on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/advertisement-fy-HTTP_5XX_Alert_Contour.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: HTTP_5XX_Alert_Contour_EKS
expr: sum(rate(envoy_cluster_upstream_rq{envoy_cluster_name=~"prd-advertisement-fy_prd-advertisement-fy_80",envoy_response_code=~"[5].*"}[1m])) by (cluster, envoy_cluster_name, job) * 60 >= 150
for: 1m
labels:
severity: critical
bu: supply
team: supplier-ads
service: advertisement-fy
env: prd
priority: p1
annotations:
summary: "EKS HTTP_5XX_Alert_Contour on {{ $labels.service }}"
description: "EKS HTTP_5XX_Alert_Contour on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/advertisement-fy-Latency_P99_Alert.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: Latency_P99_Alert_EKS
expr: histogram_quantile(0.99, sum(rate(nginx_ingress_controller_request_duration_seconds_bucket{cluster!="prod-ops-cluster", service=~"prd-advertisement-fy", job=~"kubernetes-pods"}[1m])) by (cluster, service, le, ingress, namespace, job)) >= 0.75
for: 1m
labels:
severity: critical
bu: supply
team: supplier-ads
service: advertisement-fy
env: prd
priority: p1
annotations:
summary: "EKS Latency/Response Time P99 Alert on {{ $labels.service }}"
description: "EKS Latency/Response Time P99 Alert on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/advertisement-fy-Latency_P99_Alert_Contour.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: Latency_P99_Alert_Contour_EKS
expr: histogram_quantile(0.99, sum(rate(envoy_cluster_upstream_rq_time_bucket{envoy_cluster_name=~"prd-advertisement-fy_prd-advertisement-fy_80"}[1m])) by (cluster, envoy_cluster_name, le, job)) >= 0.75
for: 1m
labels:
severity: critical
bu: supply
team: supplier-ads
service: advertisement-fy
env: prd
priority: p1
annotations:
summary: "EKS Latency_P99_Alert_Contour on {{ $labels.service }}"
description: "EKS Latency_P99_Alert_Contour on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/advertisement-fy/custom_alerts/Controller_Http_Non_2xx_alert.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: Controller_Http_Non_2xx_alert_EKS
expr: avg(sum by (service,uri,httpCode) (CONTROLLER_execTime_count{podNamespace=~"prd-advertisement-fy",httpCode!~"2[0-9]{2}"} )) >= 50
for: 1m
labels:
severity: critical
bu: supply
team: supplier-ads
service: advertisement-fy
env: prd
priority: p1
annotations:
summary: "EKS Controller_Http_Non_2xx_alert on {{ $labels.service }}"
description: "EKS Controller_Http_Non_2xx_alert on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/advertisement-fy/custom_alerts/Controller_latency_p99_alert.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: Controller_latency_p99_alert_EKS
expr: avg(histogram_quantile(0.99, sum(rate(CONTROLLER_execTime_bucket{uri!~".*health.*",podNamespace=~"prd-advertisement-fy"}[1m])) by (le))) >= 200
for: 1m
labels:
severity: critical
bu: supply
team: supplier-ads
service: advertisement-fy
env: prd
priority: p1
annotations:
summary: "EKS Controller_latency_p99_alert on {{ $labels.service }}"
description: "EKS Controller_latency_p99_alert on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/advertisement-fy/custom_alerts/Feed_Aggregator_Ads_Failures_Alert.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: Feed_Aggregator_Ads_Failures_Alert_EKS
expr: max(sum(resilience4j_circuitbreaker_buffered_calls{kind="failed",name=~".*fy-ad-service-breaker*"})) >= 1000
for: 1m
labels:
severity: critical
bu: supply
team: supplier-ads
service: advertisement-fy
env: prd
priority: p1
annotations:
summary: "EKS Feed_Aggregator_Ads_Failures_Alert on {{ $labels.service }}"
description: "EKS Feed_Aggregator_Ads_Failures_Alert on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/advertisement-fy/custom_alerts/Feed_Aggregator_Ads_Short_Circuits_Alert.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: Feed_Aggregator_Ads_Short_Circuits_Alert_EKS
expr: max(sum(resilience4j_circuitbreaker_state{name=~".*ad-service-breaker*",state!="closed",name=~ ".*fy-ad-service.*"}) by (name,state)) >= 1
for: 5m
labels:
severity: critical
bu: supply
team: supplier-ads
service: advertisement-fy
env: prd
priority: p1
annotations:
summary: "EKS Feed_Aggregator_Ads_Short_Circuits_Alert on {{ $labels.service }}"
description: "EKS Feed_Aggregator_Ads_Short_Circuits_Alert on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/advertisement-fy/custom_alerts/Feed_Aggregator_Ads_Slow_Call_Percentage_Alert_EKS.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: Feed_Aggregator_Ads_Slow_Call_Percentage_Alert_EKS_EKS
expr: max(resilience4j_circuitbreaker_slow_call_rate{name=~".*fy-ad-service-breaker*"}) >= 10
for: 1m
labels:
severity: critical
bu: supply
team: supplier-ads
service: advertisement-fy
env: prd
priority: p1
annotations:
summary: "EKS Feed_Aggregator_Ads_Slow_Call_Percentage_Alert_EKS on {{ $labels.service }}"
description: "EKS Feed_Aggregator_Ads_Slow_Call_Percentage_Alert_EKS on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/advertisement-reco-HTTP_4XX_Alert.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: HTTP_4XX_Alert_EKS
expr: sum(rate(nginx_ingress_controller_requests{cluster!="prod-ops-cluster", status=~"[4].*", service=~"prd-advertisement-reco", job=~"kubernetes-pods"}[1m])) by (cluster, service, ingress, job) * 60 >= 150
for: 1m
labels:
severity: critical
bu: supply
team: supplier-ads
service: advertisement-reco
env: prd
priority: p1
annotations:
summary: "EKS HTTP 4XX Alert on {{ $labels.service }}"
description: "EKS HTTP 4XX Alert on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/advertisement-reco-HTTP_4XX_Alert_Contour.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: HTTP_4XX_Alert_Contour_EKS
expr: sum(rate(envoy_cluster_upstream_rq{envoy_cluster_name=~"prd-advertisement-reco_prd-advertisement-reco_80",envoy_response_code=~"[4].*"}[1m])) by (cluster, envoy_cluster_name, job) * 60 >= 150
for: 1m
labels:
severity: critical
bu: supply
team: supplier-ads
service: advertisement-reco
env: prd
priority: p1
annotations:
summary: "EKS HTTP_4XX_Alert_Contour on {{ $labels.service }}"
description: "EKS HTTP_4XX_Alert_Contour on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/advertisement-reco-HTTP_5XX_Alert.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: HTTP_5XX_Alert_EKS
expr: sum(rate(nginx_ingress_controller_requests{cluster!="prod-ops-cluster", status=~"[5].*", service=~"prd-advertisement-reco", job=~"kubernetes-pods"}[1m])) by (cluster, service, ingress, job) * 60 >= 150
for: 1m
labels:
severity: critical
bu: supply
team: supplier-ads
service: advertisement-reco
env: prd
priority: p1
annotations:
summary: "EKS HTTP 5XX Alert on {{ $labels.service }}"
description: "EKS HTTP 5XX Alert on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/advertisement-reco-HTTP_5XX_Alert_Contour.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: HTTP_5XX_Alert_Contour_EKS
expr: sum(rate(envoy_cluster_upstream_rq{envoy_cluster_name=~"prd-advertisement-reco_prd-advertisement-reco_80",envoy_response_code=~"[5].*"}[1m])) by (cluster, envoy_cluster_name, job) * 60 >= 150
for: 1m
labels:
severity: critical
bu: supply
team: supplier-ads
service: advertisement-reco
env: prd
priority: p1
annotations:
summary: "EKS HTTP_5XX_Alert_Contour on {{ $labels.service }}"
description: "EKS HTTP_5XX_Alert_Contour on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/advertisement-reco-Latency_P99_Alert.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: Latency_P99_Alert_EKS
expr: histogram_quantile(0.99, sum(rate(nginx_ingress_controller_request_duration_seconds_bucket{cluster!="prod-ops-cluster", service=~"prd-advertisement-reco", job=~"kubernetes-pods"}[1m])) by (cluster, service, le, ingress, namespace, job)) >= 0.75
for: 1m
labels:
severity: critical
bu: supply
team: supplier-ads
service: advertisement-reco
env: prd
priority: p1
annotations:
summary: "EKS Latency/Response Time P99 Alert on {{ $labels.service }}"
description: "EKS Latency/Response Time P99 Alert on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/advertisement-reco-Latency_P99_Alert_Contour.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: Latency_P99_Alert_Contour_EKS
expr: histogram_quantile(0.99, sum(rate(envoy_cluster_upstream_rq_time_bucket{envoy_cluster_name=~"prd-advertisement-reco_prd-advertisement-reco_80"}[1m])) by (cluster, envoy_cluster_name, le, job)) >= 0.75
for: 1m
labels:
severity: critical
bu: supply
team: supplier-ads
service: advertisement-reco
env: prd
priority: p1
annotations:
summary: "EKS Latency_P99_Alert_Contour on {{ $labels.service }}"
description: "EKS Latency_P99_Alert_Contour on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/advertisement-reco/custom_alerts/Controller_Http_Non_2xx_alert.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: Controller_Http_Non_2xx_alert_EKS
expr: avg(sum by (service,uri,httpCode) (CONTROLLER_execTime_count{podNamespace=~"prd-advertisement-reco",httpCode!~"2[0-9]{2}"} )) >= 50
for: 1m
labels:
severity: critical
bu: supply
team: supplier-ads
service: advertisement-reco
env: prd
priority: p1
annotations:
summary: "EKS Controller_Http_Non_2xx_alert on {{ $labels.service }}"
description: "EKS Controller_Http_Non_2xx_alert on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/advertisement-reco/custom_alerts/Controller_latency_p99_alert.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: Controller_latency_p99_alert_EKS
expr: avg(histogram_quantile(0.99, sum(rate(CONTROLLER_execTime_bucket{uri!~".*health.*",podNamespace=~"prd-advertisement-reco"}[1m])) by (le))) >= 150
for: 1m
labels:
severity: critical
bu: supply
team: supplier-ads
service: advertisement-reco
env: prd
priority: p1
annotations:
summary: "EKS Controller_latency_p99_alert on {{ $labels.service }}"
description: "EKS Controller_latency_p99_alert on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/advertisement-reco/custom_alerts/Feed_Aggregator_Ads_Failures_Alert.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: Feed_Aggregator_Ads_Failures_Alert_EKS
expr: max(sum(resilience4j_circuitbreaker_buffered_calls{kind="failed",name=~".*recommendation-ad-service-breaker*"})) >= 1000
for: 1m
labels:
severity: critical
bu: supply
team: supplier-ads
service: advertisement-reco
env: prd
priority: p1
annotations:
summary: "EKS Feed_Aggregator_Ads_Failures_Alert on {{ $labels.service }}"
description: "EKS Feed_Aggregator_Ads_Failures_Alert on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/advertisement-reco/custom_alerts/Feed_Aggregator_Ads_Short_Circuits_Alert.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: Feed_Aggregator_Ads_Short_Circuits_Alert_EKS
expr: max(sum(resilience4j_circuitbreaker_state{name=~".*ad-service-breaker*",state!="closed",name=~ ".*recommendation-ad-service.*"}) by (name,state)) >= 1
for: 5m
labels:
severity: critical
bu: supply
team: supplier-ads
service: advertisement-reco
env: prd
priority: p1
annotations:
summary: "EKS Feed_Aggregator_Ads_Short_Circuits_Alert on {{ $labels.service }}"
description: "EKS Feed_Aggregator_Ads_Short_Circuits_Alert on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/advertisement-reco/custom_alerts/Feed_Aggregator_Ads_Slow_Call_Percentage_Alert_EKS.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: Feed_Aggregator_Ads_Slow_Call_Percentage_Alert_EKS_EKS
expr: max(resilience4j_circuitbreaker_slow_call_rate{name=~".*recommendation-ad-service-breaker*"}) >= 10
for: 1m
labels:
severity: critical
bu: supply
team: supplier-ads
service: advertisement-reco
env: prd
priority: p1
annotations:
summary: "EKS Feed_Aggregator_Ads_Slow_Call_Percentage_Alert_EKS on {{ $labels.service }}"
description: "EKS Feed_Aggregator_Ads_Slow_Call_Percentage_Alert_EKS on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/advertisement-text-search-HTTP_4XX_Alert.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: HTTP_4XX_Alert_EKS
expr: sum(rate(nginx_ingress_controller_requests{cluster!="prod-ops-cluster", status=~"[4].*", service=~"prd-advertisement-text-search", job=~"kubernetes-pods"}[1m])) by (cluster, service, ingress, job) * 60 >= 50
for: 1m
labels:
severity: critical
bu: supply
team: supplier-ads
service: advertisement-text-search
env: prd
priority: p1
annotations:
summary: "EKS HTTP 4XX Alert on {{ $labels.service }}"
description: "EKS HTTP 4XX Alert on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/advertisement-text-search-HTTP_4XX_Alert_Contour.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: HTTP_4XX_Alert_Contour_EKS
expr: sum(rate(envoy_cluster_upstream_rq{envoy_cluster_name=~"prd-advertisement-text-search_prd-advertisement-text-search_80",envoy_response_code=~"[4].*"}[1m])) by (cluster, envoy_cluster_name, job) * 60 >= 50
for: 1m
labels:
severity: critical
bu: supply
team: supplier-ads
service: advertisement-text-search
env: prd
priority: p1
annotations:
summary: "EKS HTTP_4XX_Alert_Contour on {{ $labels.service }}"
description: "EKS HTTP_4XX_Alert_Contour on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/advertisement-text-search-HTTP_5XX_Alert.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: HTTP_5XX_Alert_EKS
expr: sum(rate(nginx_ingress_controller_requests{cluster!="prod-ops-cluster", status=~"[5].*", service=~"prd-advertisement-text-search", job=~"kubernetes-pods"}[1m])) by (cluster, service, ingress, job) * 60 >= 100
for: 1m
labels:
severity: critical
bu: supply
team: supplier-ads
service: advertisement-text-search
env: prd
priority: p1
annotations:
summary: "EKS HTTP 5XX Alert on {{ $labels.service }}"
description: "EKS HTTP 5XX Alert on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/advertisement-text-search-HTTP_5XX_Alert_Contour.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: HTTP_5XX_Alert_Contour_EKS
expr: sum(rate(envoy_cluster_upstream_rq{envoy_cluster_name=~"prd-advertisement-text-search_prd-advertisement-text-search_80",envoy_response_code=~"[5].*"}[1m])) by (cluster, envoy_cluster_name, job) * 60 >= 100
for: 1m
labels:
severity: critical
bu: supply
team: supplier-ads
service: advertisement-text-search
env: prd
priority: p1
annotations:
summary: "EKS HTTP_5XX_Alert_Contour on {{ $labels.service }}"
description: "EKS HTTP_5XX_Alert_Contour on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/advertisement-text-search-Latency_P99_Alert.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: Latency_P99_Alert_EKS
expr: histogram_quantile(0.99, sum(rate(nginx_ingress_controller_request_duration_seconds_bucket{cluster!="prod-ops-cluster", service=~"prd-advertisement-text-search", job=~"kubernetes-pods"}[1m])) by (cluster, service, le, ingress, namespace, job)) >= 0.75
for: 1m
labels:
severity: critical
bu: supply
team: supplier-ads
service: advertisement-text-search
env: prd
priority: p1
annotations:
summary: "EKS Latency/Response Time P99 Alert on {{ $labels.service }}"
description: "EKS Latency/Response Time P99 Alert on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/advertisement-text-search-Latency_P99_Alert_Contour.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: Latency_P99_Alert_Contour_EKS
expr: histogram_quantile(0.99, sum(rate(envoy_cluster_upstream_rq_time_bucket{envoy_cluster_name=~"prd-advertisement-text-search_prd-advertisement-text-search_80"}[1m])) by (cluster, envoy_cluster_name, le, job)) >= 0.75
for: 1m
labels:
severity: critical
bu: supply
team: supplier-ads
service: advertisement-text-search
env: prd
priority: p1
annotations:
summary: "EKS Latency_P99_Alert_Contour on {{ $labels.service }}"
description: "EKS Latency_P99_Alert_Contour on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/advertisement-text-search/custom_alerts/Ads_feed_aggregator_shortcircuits.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: Ads_feed_aggregator_shortcircuits_EKS
expr: sum(resilience4j_circuitbreaker_state{name=~".*ad-service-breaker*", name=~".*visual-search.*",state!="closed"}) by (name, state) >= 1
for: 1m
labels:
severity: critical
bu: supply
team: supplier-ads
service: advertisement-text-search
env: prd
priority: p1
annotations:
summary: "EKS Ads_feed_aggregator_shortcircuits on {{ $labels.service }}"
description: "EKS Ads_feed_aggregator_shortcircuits on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/advertisement-text-search/custom_alerts/Controller_Http_Non_2xx_alert_EKS.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: Controller_Http_Non_2xx_alert_EKS_EKS
expr: avg(sum by (service,uri,httpCode) (CONTROLLER_execTime_count{podNamespace=~"prd-advertisement-text-search",httpCode!~"2[0-9]{2}"} )) >= 50
for: 1m
labels:
severity: critical
bu: supply
team: supplier-ads
service: advertisement-text-search
env: prd
priority: p1
annotations:
summary: "EKS Controller_Http_Non_2xx_alert_EKS on {{ $labels.service }}"
description: "EKS Controller_Http_Non_2xx_alert_EKS on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/advertisement-text-search/custom_alerts/Controller_latency_p99_alert.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: Controller_latency_p99_alert_EKS
expr: avg(histogram_quantile(0.99, sum(rate(CONTROLLER_execTime_bucket{uri!~".*health.*",podNamespace=~"prd-advertisement-text-search"}[1m])) by (le))) >= 450
for: 1m
labels:
severity: critical
bu: supply
team: supplier-ads
service: advertisement-text-search
env: prd
priority: p1
annotations:
summary: "EKS Controller_latency_p99_alert on {{ $labels.service }}"
description: "EKS Controller_latency_p99_alert on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/advertisement-text-search/custom_alerts/Feed_Aggregator_Ads_Failures_Alert.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: Feed_Aggregator_Ads_Failures_Alert_EKS
expr: max(sum(resilience4j_circuitbreaker_buffered_calls{kind="failed",name=~".*text-search-ad-service-breaker*"})) >= 1000
for: 1m
labels:
severity: critical
bu: supply
team: supplier-ads
service: advertisement-text-search
env: prd
priority: p1
annotations:
summary: "EKS Feed_Aggregator_Ads_Failures_Alert on {{ $labels.service }}"
description: "EKS Feed_Aggregator_Ads_Failures_Alert on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/advertisement-text-search/custom_alerts/Feed_Aggregator_Ads_Failures_Alerts.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: Feed_Aggregator_Ads_Failures_Alerts_EKS
expr: max(sum(resilience4j_circuitbreaker_buffered_calls{kind="failed",name=~".*visual-search-ad-service-breaker*"})) >= 1000
for: 1m
labels:
severity: critical
bu: supply
team: supplier-ads
service: advertisement-text-search
env: prd
priority: p1
annotations:
summary: "EKS Feed_Aggregator_Ads_Failures_Alerts on {{ $labels.service }}"
description: "EKS Feed_Aggregator_Ads_Failures_Alerts on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/advertisement-text-search/custom_alerts/Feed_Aggregator_Ads_Short_Circuits_Alert.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: Feed_Aggregator_Ads_Short_Circuits_Alert_EKS
expr: max(sum(resilience4j_circuitbreaker_state{name=~".*ad-service-breaker*",state!="closed",name=~ ".*text-search-ad-service.*"}) by (name,state)) >= 1
for: 5m
labels:
severity: critical
bu: supply
team: supplier-ads
service: advertisement-text-search
env: prd
priority: p1
annotations:
summary: "EKS Feed_Aggregator_Ads_Short_Circuits_Alert on {{ $labels.service }}"
description: "EKS Feed_Aggregator_Ads_Short_Circuits_Alert on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/advertisement-text-search/custom_alerts/Feed_Aggregator_Ads_Slow_Call_Percentage_Alerts.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: Feed_Aggregator_Ads_Slow_Call_Percentage_Alerts_EKS
expr: max(resilience4j_circuitbreaker_slow_call_rate{name=~".*visual-search-ad-service-breaker*"}) >= 10
for: 1m
labels:
severity: critical
bu: supply
team: supplier-ads
service: advertisement-text-search
env: prd
priority: p1
annotations:
summary: "EKS Feed_Aggregator_Ads_Slow_Call_Percentage_Alerts on {{ $labels.service }}"
description: "EKS Feed_Aggregator_Ads_Slow_Call_Percentage_Alerts on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/advertisement-text-search/custom_alerts/Feed_Aggregator_Ads_Slow_Call_Percentage_alert.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: Feed_Aggregator_Ads_Slow_Call_Percentage_alert_EKS
expr: max(resilience4j_circuitbreaker_slow_call_rate{name=~".*text-search-ad-service-breaker*"}) >= 10
for: 1m
labels:
severity: critical
bu: supply
team: supplier-ads
service: advertisement-text-search
env: prd
priority: p1
annotations:
summary: "EKS Feed_Aggregator_Ads_Slow_Call_Percentage_alert on {{ $labels.service }}"
description: "EKS Feed_Aggregator_Ads_Slow_Call_Percentage_alert on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/advertisement/custom_alerts/Controller_Http_Non_2xx_alert.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: Controller_Http_Non_2xx_alert_EKS
expr: avg(sum by (service,uri,httpCode) (CONTROLLER_execTime_count{podNamespace=~"prd-advertisement",httpCode!~"2[0-9]{2}"} )) >= 200
for: 1m
labels:
severity: critical
bu: supply
team: supplier-ads
service: advertisement
env: prd
priority: p1
annotations:
summary: "EKS Controller_Http_Non_2xx_alert on {{ $labels.service }}"
description: "EKS Controller_Http_Non_2xx_alert on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/advertisement/custom_alerts/Controller_latency_p99_alert.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: Controller_latency_p99_alert_EKS
expr: avg(histogram_quantile(0.99, sum(rate(CONTROLLER_execTime_bucket{uri!~".*health.*",podNamespace=~"prd-advertisement"}[1m])) by (le))) >= 200
for: 1m
labels:
severity: critical
bu: supply
team: supplier-ads
service: advertisement
env: prd
priority: p1
annotations:
summary: "EKS Controller_latency_p99_alert on {{ $labels.service }}"
description: "EKS Controller_latency_p99_alert on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/advertisement/custom_alerts/Feed_Aggregator_Ads_Failures_Alert.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: Feed_Aggregator_Ads_Failures_Alert_EKS
expr: max(sum(resilience4j_circuitbreaker_buffered_calls{kind="failed",name=~".*clp-ad-service-breaker*"})) >= 1000
for: 1m
labels:
severity: critical
bu: supply
team: supplier-ads
service: advertisement
env: prd
priority: p1
annotations:
summary: "EKS Feed_Aggregator_Ads_Failures_Alert on {{ $labels.service }}"
description: "EKS Feed_Aggregator_Ads_Failures_Alert on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/advertisement/custom_alerts/Feed_Aggregator_Ads_Failures_Alerts.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: Feed_Aggregator_Ads_Failures_Alerts_EKS
expr: max(sum(resilience4j_circuitbreaker_buffered_calls{kind="failed",name=~".*clp-ad-service-breaker*"})) >= 1000
for: 1m
labels:
severity: critical
bu: supply
team: supplier-ads
service: advertisement
env: prd
priority: p1
annotations:
summary: "EKS Feed_Aggregator_Ads_Failures_Alerts on {{ $labels.service }}"
description: "EKS Feed_Aggregator_Ads_Failures_Alerts on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/advertisement/custom_alerts/Feed_Aggregator_Ads_Short_Circuits_Alert.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: Feed_Aggregator_Ads_Short_Circuits_Alert_EKS
expr: max(sum(resilience4j_circuitbreaker_state{name=~".*ad-service-breaker*",state!="closed",name=~ ".*collection-ad-service.*"}) by (name,state)) >= 1
for: 5m
labels:
severity: critical
bu: supply
team: supplier-ads
service: advertisement
env: prd
priority: p1
annotations:
summary: "EKS Feed_Aggregator_Ads_Short_Circuits_Alert on {{ $labels.service }}"
description: "EKS Feed_Aggregator_Ads_Short_Circuits_Alert on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/advertisement/custom_alerts/Feed_Aggregator_Ads_Short_Circuits_Alerts.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: Feed_Aggregator_Ads_Short_Circuits_Alerts_EKS
expr: max(sum(resilience4j_circuitbreaker_state{name=~".*ad-service-breaker*",state!="closed",name=~ ".*collection-ad-service.*"}) by (name,state)) >= 1
for: 5m
labels:
severity: critical
bu: supply
team: supplier-ads
service: advertisement
env: prd
priority: p1
annotations:
summary: "EKS Feed_Aggregator_Ads_Short_Circuits_Alerts on {{ $labels.service }}"
description: "EKS Feed_Aggregator_Ads_Short_Circuits_Alerts on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/advertisement/custom_alerts/Feed_Aggregator_Ads_Slow_Call_Percentage_Alert_EKS.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: Feed_Aggregator_Ads_Slow_Call_Percentage_Alert_EKS_EKS
expr: max(resilience4j_circuitbreaker_slow_call_rate{name=~".*clp-ad-service-breaker*"}) >= 10
for: 1m
labels:
severity: critical
bu: supply
team: supplier-ads
service: advertisement
env: prd
priority: p1
annotations:
summary: "EKS Feed_Aggregator_Ads_Slow_Call_Percentage_Alert_EKS on {{ $labels.service }}"
description: "EKS Feed_Aggregator_Ads_Slow_Call_Percentage_Alert_EKS on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/advertisement/custom_alerts/Feed_Aggregator_Ads_Slow_Call_Percentage_Alerts_EKS.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: Feed_Aggregator_Ads_Slow_Call_Percentage_Alerts_EKS_EKS
expr: max(resilience4j_circuitbreaker_slow_call_rate{name=~".*collection-ad-service-breaker*"}) >= 10
for: 1m
labels:
severity: critical
bu: supply
team: supplier-ads
service: advertisement
env: prd
priority: p1
annotations:
summary: "EKS Feed_Aggregator_Ads_Slow_Call_Percentage_Alerts_EKS on {{ $labels.service }}"
description: "EKS Feed_Aggregator_Ads_Slow_Call_Percentage_Alerts_EKS on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/campaign-performance-HTTP_4XX_Alert.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: HTTP_4XX_Alert_EKS
expr: sum(rate(nginx_ingress_controller_requests{cluster!="prod-ops-cluster", status=~"[4].*", service=~"prd-campaign-performance", job=~"kubernetes-pods"}[1m])) by (cluster, service, ingress, job) * 60 >= 10
for: 2m
labels:
severity: critical
bu: supply
team: supplier-ads
service: campaign-performance
env: prd
priority: p1
annotations:
summary: "EKS HTTP 4XX Alert on {{ $labels.service }}"
description: "EKS HTTP 4XX Alert on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/campaign-performance-HTTP_5XX_Alert.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: HTTP_5XX_Alert_EKS
expr: sum(rate(nginx_ingress_controller_requests{cluster!="prod-ops-cluster", status=~"[5].*", service=~"prd-campaign-performance", job=~"kubernetes-pods"}[1m])) by (cluster, service, ingress, job) * 60 >= 20
for: 1m
labels:
severity: critical
bu: supply
team: supplier-ads
service: campaign-performance
env: prd
priority: p1
annotations:
summary: "EKS HTTP 5XX Alert on {{ $labels.service }}"
description: "EKS HTTP 5XX Alert on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/campaign-performance-Latency_P99_Alert.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: Latency_P99_Alert_EKS
expr: histogram_quantile(0.99, sum(rate(nginx_ingress_controller_request_duration_seconds_bucket{cluster!="prod-ops-cluster", service=~"prd-campaign-performance", job=~"kubernetes-pods"}[1m])) by (cluster, service, le, ingress, namespace, job)) >= 0.75
for: 1m
labels:
severity: critical
bu: supply
team: supplier-ads
service: campaign-performance
env: prd
priority: p1
annotations:
summary: "EKS Latency/Response Time P99 Alert on {{ $labels.service }}"
description: "EKS Latency/Response Time P99 Alert on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/campaign-performance-P99_Alert_Per_API_Campaign_Performance.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: P99_Alert_Per_API_Campaign_Performance_EKS
expr: histogram_quantile(0.99, sum(rate(envoy_cluster_upstream_rq_time_bucket{envoy_cluster_name=~"prd-campaign-performance_prd-campaign-performance_80"}[1m])) by (cluster, envoy_cluster_name, le, job)) >= 100
for: 1m
labels:
severity: critical
bu: supply
team: supplier-ads
service: campaign-performance
env: prd
priority: p1
annotations:
summary: "EKS P99_Alert_Per_API_Campaign_Performance on {{ $labels.service }}"
description: "EKS P99_Alert_Per_API_Campaign_Performance on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/supplier-platform-ads-HTTP_4XX_Alert.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: HTTP_4XX_Alert_EKS
expr: sum(rate(nginx_ingress_controller_requests{cluster!="prod-ops-cluster", status=~"[4].*", service=~"prd-supplier-platform-ads", job=~"kubernetes-pods"}[1m])) by (cluster, service, ingress, job) * 60 >= 150
for: 2m
labels:
severity: critical
bu: supply
team: supplier-ads
service: supplier-platform-ads
env: prd
priority: p0
annotations:
summary: "EKS HTTP 4XX Alert on {{ $labels.service }}"
description: "EKS HTTP 4XX Alert on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/supplier-platform-ads-HTTP_5XX_Alert.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: HTTP_5XX_Alert_EKS
expr: sum(rate(nginx_ingress_controller_requests{cluster!="prod-ops-cluster", status=~"[5].*", service=~"prd-supplier-platform-ads", job=~"kubernetes-pods"}[1m])) by (cluster, service, ingress, job) * 60 >= 30
for: 1m
labels:
severity: critical
bu: supply
team: supplier-ads
service: supplier-platform-ads
env: prd
priority: p0
annotations:
summary: "EKS HTTP 5XX Alert on {{ $labels.service }}"
description: "EKS HTTP 5XX Alert on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/supply/supplier-ads/supplier-platform-ads-Latency_P99_Alert.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: Latency_P99_Alert_EKS
expr: histogram_quantile(0.99, sum(rate(nginx_ingress_controller_request_duration_seconds_bucket{cluster!="prod-ops-cluster", service=~"prd-supplier-platform-ads", job=~"kubernetes-pods"}[1m])) by (cluster, service, le, ingress, namespace, job)) >= 0.75
for: 1m
labels:
severity: critical
bu: supply
team: supplier-ads
service: supplier-platform-ads
env: prd
priority: p0
annotations:
summary: "EKS Latency/Response Time P99 Alert on {{ $labels.service }}"
description: "EKS Latency/Response Time P99 Alert on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"