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/demand/discovery-platform/feed-aggregator-CONTROLLER_4XX_Alert_EKS.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: CONTROLLER_4XX_Alert_EKS
expr: sum by (service,uri,httpCode) (CONTROLLER_execTime_count{service="feed-aggregator",httpCode=~"[4].*"} ) >= 100
for: 1m
labels:
severity: critical
bu: demand
team: discovery-platform
service: feed-aggregator
env: prd
priority: p0
annotations:
summary: "EKS CONTROLLER_4XX_Alert on {{ $labels.service }}"
description: "EKS CONTROLLER_4XX_Alert on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/demand/discovery-platform/feed-aggregator-CONTROLLER_5XX_Alert_EKS.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: CONTROLLER_5XX_Alert_EKS
expr: sum by (service,uri,httpCode) (CONTROLLER_execTime_count{service="feed-aggregator",httpCode=~"[5].*"} ) >= 500
for: 1m
labels:
severity: critical
bu: demand
team: discovery-platform
service: feed-aggregator
env: prd
priority: p0
annotations:
summary: "EKS CONTROLLER_5XX_Alert on {{ $labels.service }}"
description: "EKS CONTROLLER_5XX_Alert on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/demand/discovery-platform/feed-aggregator-search-HTTP_4XX_Alert_Per_API_Contour.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: HTTP_4XX_Alert_Per_API_Contour_EKS
expr: sum(rate(envoy_cluster_upstream_rq{envoy_cluster_name=~"prd-feed-aggregator-search_prd-feed-aggregator-search-primary_80",envoy_response_code=~"[4].*"}[1m])) by (cluster, envoy_cluster_name, job) * 60 >= 50
for: 2m
labels:
severity: critical
bu: demand
team: discovery-platform
service: feed-aggregator-search
env: prd
priority: p1
annotations:
summary: "EKS HTTP_4XX_Alert_Per_API_Contour on {{ $labels.service }}"
description: "EKS HTTP_4XX_Alert_Per_API_Contour on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/demand/discovery-platform/feed-aggregator-search-HTTP_5XX_Alert_Per_API_Contour.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: HTTP_5XX_Alert_Per_API_Contour_EKS
expr: sum(rate(envoy_cluster_upstream_rq{envoy_cluster_name=~"prd-feed-aggregator-search_prd-feed-aggregator-search-primary_80",envoy_response_code=~"[5].*"}[1m])) by (cluster, envoy_cluster_name, job) * 60 >= 250
for: 2m
labels:
severity: critical
bu: demand
team: discovery-platform
service: feed-aggregator-search
env: prd
priority: p1
annotations:
summary: "EKS HTTP_5XX_Alert_Per_API_Contour on {{ $labels.service }}"
description: "EKS HTTP_5XX_Alert_Per_API_Contour on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/demand/discovery-platform/feed-aggregator-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-feed-aggregator-search_prd-feed-aggregator-search-primary_80"}[1m])) by (cluster, envoy_cluster_name, le, job)) >= 1500
for: 2m
labels:
severity: critical
bu: demand
team: discovery-platform
service: feed-aggregator-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/demand/discovery-platform/feed-aggregator-secondary-CONTROLLER_4XX_Alert_EKS.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: CONTROLLER_4XX_Alert_EKS
expr: sum by (service,uri,httpCode) (CONTROLLER_execTime_count{service="feed-aggregator-secondary",httpCode=~"[4].*"} ) >= 100
for: 1m
labels:
severity: critical
bu: demand
team: discovery-platform
service: feed-aggregator-secondary
env: prd
priority: p0
annotations:
summary: "EKS CONTROLLER_4XX_Alert on {{ $labels.service }}"
description: "EKS CONTROLLER_4XX_Alert on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/demand/discovery-platform/feed-aggregator-secondary-CONTROLLER_5XX_Alert_EKS.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: CONTROLLER_5XX_Alert_EKS
expr: sum by (service,uri,httpCode) (CONTROLLER_execTime_count{service="feed-aggregator-secondary",httpCode=~"[5].*"} ) >= 500
for: 1m
labels:
severity: critical
bu: demand
team: discovery-platform
service: feed-aggregator-secondary
env: prd
priority: p0
annotations:
summary: "EKS CONTROLLER_5XX_Alert on {{ $labels.service }}"
description: "EKS CONTROLLER_5XX_Alert on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/demand/discovery-platform/feed-aggregator-secondary-HTTP_4XX_Alert_Per_API_Contour.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: HTTP_4XX_Alert_Per_API_Contour_EKS
expr: sum(rate(envoy_cluster_upstream_rq{envoy_cluster_name=~"prd-feed-aggregator-secondary_prd-feed-aggregator-secondary-primary_80",envoy_response_code=~"[4].*"}[1m])) by (cluster, envoy_cluster_name, job) * 60 >= 100
for: 1m
labels:
severity: critical
bu: demand
team: discovery-platform
service: feed-aggregator-secondary
env: prd
priority: p1
annotations:
summary: "EKS HTTP_4XX_Alert_Per_API_Contour on {{ $labels.service }}"
description: "EKS HTTP_4XX_Alert_Per_API_Contour on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/demand/discovery-platform/feed-aggregator-secondary-HTTP_5XX_Alert_Per_API_Contour.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: HTTP_5XX_Alert_Per_API_Contour_EKS
expr: sum(rate(envoy_cluster_upstream_rq{envoy_cluster_name=~"prd-feed-aggregator-secondary_prd-feed-aggregator-secondary-primary_80",envoy_response_code=~"[5].*"}[1m])) by (cluster, envoy_cluster_name, job) * 60 >= 500
for: 5m
labels:
severity: critical
bu: demand
team: discovery-platform
service: feed-aggregator-secondary
env: prd
priority: p1
annotations:
summary: "EKS HTTP_5XX_Alert_Per_API_Contour on {{ $labels.service }}"
description: "EKS HTTP_5XX_Alert_Per_API_Contour on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/demand/discovery-platform/feed-aggregator-secondary-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(nginx_ingress_controller_request_duration_seconds_bucket{cluster!="prod-ops-cluster",service=~"prd-feed-aggregator-secondary", job=~"kubernetes-service-endpoints"}) by (cluster, service, le, ingress, namespace, job)) >= 0.75
for: 1m
labels:
severity: critical
bu: demand
team: discovery-platform
service: feed-aggregator-secondary
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/demand/discovery-platform/feed-aggregator-secondary-P99_Latency_Alert_Contour.yaml
+16
View File
@@ -0,0 +1,16 @@
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-feed-aggregator-secondary_prd-feed-aggregator-secondary-primary_80"}[1m])) by (cluster, envoy_cluster_name, le, job)) >= 2000
for: 1m
labels:
severity: critical
bu: demand
team: discovery-platform
service: feed-aggregator-secondary
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 }}\n LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/demand/discovery-platform/hero-product-web-HTTP_4XX_Alert_Per_API_Contour.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: HTTP_4XX_Alert_Per_API_Contour_EKS
expr: sum(rate(envoy_cluster_upstream_rq{envoy_cluster_name=~"prd-hero-product-web_prd-hero-product-web-primary_80",envoy_response_code=~"[4].*"}[1m])) by (cluster, envoy_cluster_name, job) * 60 >= 100
for: 1m
labels:
severity: critical
bu: demand
team: discovery-platform
service: hero-product-web
env: prd
priority: p0
annotations:
summary: "EKS HTTP_4XX_Alert_Per_API_Contour on {{ $labels.service }}"
description: "EKS HTTP_4XX_Alert_Per_API_Contour on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/demand/discovery-platform/hero-product-web-HTTP_5XX_Alert_Per_API_Contour.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: HTTP_5XX_Alert_Per_API_Contour_EKS
expr: sum(rate(envoy_cluster_upstream_rq{envoy_cluster_name=~"prd-hero-product-web_prd-hero-product-web-primary_80",envoy_response_code=~"[5].*"}[1m])) by (cluster, envoy_cluster_name, job) * 60 >= 100
for: 2m
labels:
severity: critical
bu: demand
team: discovery-platform
service: hero-product-web
env: prd
priority: p0
annotations:
summary: "EKS HTTP_5XX_Alert_Per_API_Contour on {{ $labels.service }}"
description: "EKS HTTP_5XX_Alert_Per_API_Contour on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
helm-templates/victoria-metrics-alerts-config/configmap/demand/discovery-platform/hero-product-web-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-hero-product-web_prd-hero-product-web-primary_80"}[1m])) by (cluster, envoy_cluster_name, le, job)) >= 200
for: 1m
labels:
severity: critical
bu: demand
team: discovery-platform
service: hero-product-web
env: prd
priority: p0
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/demand/discovery-platform/sort-and-filter-HTTP_4XX_Alert_Contour.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: HTTP_4XX_Alert_EKS
expr: sum(rate(envoy_cluster_upstream_rq{envoy_cluster_name=~"prd-sort-and-filter_prd-sort-and-filter-primary_80",envoy_response_code=~"[4].*"}[1m])) by (cluster, envoy_cluster_name, job) * 60 >= 150
for: 1m
labels:
severity: critical
bu: demand
team: discovery-platform
service: sort-and-filter
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/demand/discovery-platform/sort-and-filter-HTTP_5XX_Alert_Contour.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: HTTP_5XX_Alert_EKS
expr: sum(rate(envoy_cluster_upstream_rq{envoy_cluster_name=~"prd-sort-and-filter_prd-sort-and-filter-primary_80",envoy_response_code=~"[5].*"}[1m])) by (cluster, envoy_cluster_name, job) * 60 >= 90
for: 1m
labels:
severity: critical
bu: demand
team: discovery-platform
service: sort-and-filter
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/demand/discovery-platform/sort-and-filter-Latency_P99_Alert_Contour.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(envoy_cluster_upstream_rq_time_bucket{envoy_cluster_name=~"prd-sort-and-filter_prd-sort-and-filter-primary_80"}[1m])) by (cluster, envoy_cluster_name, le, job)) >= 750
for: 1m
labels:
severity: critical
bu: demand
team: discovery-platform
service: sort-and-filter
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/demand/discovery-platform/sort-and-filter-secondary-HTTP_4XX_Alert_contour.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: HTTP_4XX_Alert_EKS
expr: sum(rate(envoy_cluster_upstream_rq{envoy_cluster_name=~"prd-sort-and-filter-secondary_prd-sort-and-filter-secondary-primary_80",envoy_response_code=~"[4].*"}[1m])) by (cluster, envoy_cluster_name, job) * 60 >= 100
for: 2m
labels:
severity: critical
bu: demand
team: discovery-platform
service: sort-and-filter-secondary
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/demand/discovery-platform/sort-and-filter-secondary-HTTP_5XX_Alert_Contour.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: HTTP_5XX_Alert_EKS
expr: sum(rate(envoy_cluster_upstream_rq{envoy_cluster_name=~"prd-sort-and-filter-secondary_prd-sort-and-filter-secondary-primary_80",envoy_response_code=~"[5].*"}[1m])) by (cluster, envoy_cluster_name, job) * 60 >= 250
for: 1m
labels:
severity: critical
bu: demand
team: discovery-platform
service: sort-and-filter-secondary
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/demand/discovery-platform/sort-and-filter-secondary-Latency_P99_Alert_Contour.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(envoy_cluster_upstream_rq_time_bucket{envoy_cluster_name=~"prd-sort-and-filter-secondary_prd-sort-and-filter-secondary-primary_80"}[1m])) by (cluster, envoy_cluster_name, le, job)) >= 750
for: 1m
labels:
severity: critical
bu: demand
team: discovery-platform
service: sort-and-filter-secondary
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/demand/discovery-platform/sort-and-filter-vs-HTTP_4XX_Alert_contour.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: HTTP_4XX_Alert_EKS
expr: sum(rate(envoy_cluster_upstream_rq{envoy_cluster_name=~"prd-sort-and-filter-vs_prd-sort-and-filter-vs-primary_80",envoy_response_code=~"[4].*"}[1m])) by (cluster, envoy_cluster_name, job) * 60 >= 100
for: 1m
labels:
severity: critical
bu: demand
team: discovery-platform
service: sort-and-filter-vs
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/demand/discovery-platform/sort-and-filter-vs-HTTP_5XX_Alert_Contour.yaml
+16
View File
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: HTTP_5XX_Alert_EKS
expr: sum(rate(envoy_cluster_upstream_rq{envoy_cluster_name=~"prd-sort-and-filter-vs_prd-sort-and-filter-vs-primary_80",envoy_response_code=~"[5].*"}[1m])) by (cluster, envoy_cluster_name, job) * 60 >= 90
for: 1m
labels:
severity: critical
bu: demand
team: discovery-platform
service: sort-and-filter-vs
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/demand/discovery-platform/sort-and-filter-vs-Latency_P99_Alert_Contour.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(envoy_cluster_upstream_rq_time_bucket{envoy_cluster_name=~"prd-sort-and-filter-vs_prd-sort-and-filter-vs-primary_80"}[1m])) by (cluster, envoy_cluster_name, le, job)) >= 1200
for: 1m
labels:
severity: critical
bu: demand
team: discovery-platform
service: sort-and-filter-vs
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 }}"