added repo

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
@@ -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-address-search_prd-address-search-primary_80",envoy_response_code=~"[4].*"}[1m])) by (cluster, envoy_cluster_name, job) * 60 >= 700
for: 1m
labels:
severity: critical
bu: demand
team: transact
service: address-search
env: prd
priority: p0
annotations:
summary: "EKS HTTP_4XX_Alert_Contour on {{ $labels.service }}"
description: "EKS HTTP_4XX_Alert_Contour on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: HTTP_5XX_Alert_Contour_EKS_EKS
expr: sum(rate(envoy_cluster_upstream_rq{envoy_cluster_name=~"prd-address-search_prd-address-search-primary_80",envoy_response_code=~"[5].*"}[1m])) by (cluster, envoy_cluster_name, job) * 60 >= 50
for: 1m
labels:
severity: critical
bu: demand
team: transact
service: address-search
env: prd
priority: p0
annotations:
summary: "EKS HTTP_5XX_Alert_Contour_EKS on {{ $labels.service }}"
description: "EKS HTTP_5XX_Alert_Contour_EKS on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
@@ -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-address-search_prd-address-search-primary_80"}[1m])) by (cluster, envoy_cluster_name, le, job)) >= 750
for: 1m
labels:
severity: critical
bu: demand
team: transact
service: address-search
env: prd
priority: p0
annotations:
summary: "EKS P99_Latency_Alert_Contour on {{ $labels.service }}"
description: "EKS P99_Latency_Alert_Contour on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: address_mmi_api_null_response_EKS
expr: sum by(cluster, service) (increase(address_mmi_api_null_response_value_count{kubernetes_namespace="prd-address-search"}[1m])) >= 20
for: 1m
labels:
severity: critical
bu: demand
team: transact
service: address-search
env: prd
priority: p0
annotations:
summary: "EKS address_mmi_api_null_response on {{ $labels.service }}"
description: "EKS address_mmi_api_null_response on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: address_mmi_failure_count_EKS
expr: sum by(cluster, service) (increase(address_mmi_api_failed_value_count{kubernetes_namespace="prd-address-search"}[1m])) >= 20
for: 1m
labels:
severity: critical
bu: demand
team: transact
service: address-search
env: prd
priority: p0
annotations:
summary: "EKS address_mmi_failure_count on {{ $labels.service }}"
description: "EKS address_mmi_failure_count on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
@@ -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-cart_prd-cart-primary_80",envoy_response_code=~"[4].*"}[1m])) by (cluster, envoy_cluster_name, job) * 60 >= 100
for: 1m
labels:
severity: critical
bu: demand
team: transact
service: cart
env: prd
priority: p0
annotations:
summary: "EKS HTTP_4XX_Alert_Contour on {{ $labels.service }}"
description: "EKS HTTP_4XX_Alert_Contour on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
@@ -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-cart_prd-cart-primary_80",envoy_response_code=~"[5].*"}[1m])) by (cluster, envoy_cluster_name, job) * 60 >= 70
for: 1m
labels:
severity: critical
bu: demand
team: transact
service: cart
env: prd
priority: p0
annotations:
summary: "EKS HTTP_5XX_Alert_Contour on {{ $labels.service }}"
description: "EKS HTTP_5XX_Alert_Contour on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
@@ -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-cart_prd-cart-primary_80"}[1m])) by (cluster, envoy_cluster_name, le, job)) >= 750
for: 1m
labels:
severity: critical
bu: demand
team: transact
service: cart
env: prd
priority: p0
annotations:
summary: "EKS P99_Latency_Alert_Contour on {{ $labels.service }}"
description: "EKS P99_Latency_Alert_Contour on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
@@ -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-cart-internal_prd-cart-internal_80",envoy_response_code=~"[4].*"}[1m])) by (cluster, envoy_cluster_name, job) * 60 >= 5
for: 1m
labels:
severity: critical
bu: demand
team: transact
service: cart-internal
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 }}"
@@ -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-cart-internal_prd-cart-internal_80",envoy_response_code=~"[5].*"}[1m])) by (cluster, envoy_cluster_name, job) * 60 >= 5
for: 1m
labels:
severity: critical
bu: demand
team: transact
service: cart-internal
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 }}"
@@ -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-cart-internal_prd-cart-internal_80"}[1m])) by (cluster, envoy_cluster_name, le, job)) >= 750
for: 1m
labels:
severity: critical
bu: demand
team: transact
service: cart-internal
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 }}"
@@ -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-epg_prd-epg-primary_80",envoy_response_code=~"[4].*"}[1m])) by (cluster, envoy_cluster_name, job) * 60 >= 25
for: 1m
labels:
severity: critical
bu: demand
team: transact
service: epg
env: prd
priority: p0
annotations:
summary: "EKS HTTP_4XX_Alert_Contour on {{ $labels.service }}"
description: "EKS HTTP_4XX_Alert_Contour on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
@@ -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-epg_prd-epg-primary_80",envoy_response_code=~"[5].*"}[1m])) by (cluster, envoy_cluster_name, job) * 60 >= 15
for: 1m
labels:
severity: critical
bu: demand
team: transact
service: epg
env: prd
priority: p0
annotations:
summary: "EKS HTTP_5XX_Alert_Contour on {{ $labels.service }}"
description: "EKS HTTP_5XX_Alert_Contour on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: epg_get_customer_api_failure_EKS
expr: sum by(cluster, service) (increase(juspay_get_customer_api_failure_value{kubernetes_namespace="prd-epg"}[1m])) >= 10
for: 1m
labels:
severity: critical
bu: demand
team: transact
service: epg
env: prd
priority: p0
annotations:
summary: "EKS epg_get_customer_api_failure on {{ $labels.service }}"
description: "EKS epg_get_customer_api_failure on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: epg_juspay_status_api_failure_EKS
expr: sum by(cluster, service) (increase(juspay_status_api_failure_value{kubernetes_namespace="prd-epg"}[1m])) >= 20
for: 1m
labels:
severity: critical
bu: demand
team: transact
service: epg
env: prd
priority: p0
annotations:
summary: "EKS epg_juspay_status_api_failure on {{ $labels.service }}"
description: "EKS epg_juspay_status_api_failure on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
@@ -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-order-service_prd-order-service-primary_80",envoy_response_code=~"[4].*"}[1m])) by (cluster, envoy_cluster_name, job) * 60 >= 40
for: 1m
labels:
severity: critical
bu: demand
team: transact
service: order-service
env: prd
priority: p0
annotations:
summary: "EKS HTTP_4XX_Alert_Contour on {{ $labels.service }}"
description: "EKS HTTP_4XX_Alert_Contour on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
@@ -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-order-service_prd-order-service-primary_80",envoy_response_code=~"[5].*"}[1m])) by (cluster, envoy_cluster_name, job) * 60 >= 10
for: 1m
labels:
severity: critical
bu: demand
team: transact
service: order-service
env: prd
priority: p0
annotations:
summary: "EKS HTTP_5XX_Alert_Contour on {{ $labels.service }}"
description: "EKS HTTP_5XX_Alert_Contour on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
@@ -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-order-service_prd-order-service-primary_80"}[1m])) by (cluster, envoy_cluster_name, le, job)) >= 1000
for: 1m
labels:
severity: critical
bu: demand
team: transact
service: order-service
env: prd
priority: p0
annotations:
summary: "EKS P99_Latency_Alert_Contour on {{ $labels.service }}"
description: "EKS P99_Latency_Alert_Contour on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
@@ -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-order-service-internal_prd-order-service-internal-primary_80",envoy_response_code=~"[4].*"}[1m])) by (cluster, envoy_cluster_name, job) * 60 >= 5
for: 1m
labels:
severity: critical
bu: demand
team: transact
service: order-service-internal
env: prd
priority: p0
annotations:
summary: "EKS HTTP_4XX_Alert_Contour on {{ $labels.service }}"
description: "EKS HTTP_4XX_Alert_Contour on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
@@ -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-order-service-internal_prd-order-service-internal-primary_80",envoy_response_code=~"[5].*"}[1m])) by (cluster, envoy_cluster_name, job) * 60 >= 5
for: 1m
labels:
severity: critical
bu: demand
team: transact
service: order-service-internal
env: prd
priority: p0
annotations:
summary: "EKS HTTP_5XX_Alert_Contour on {{ $labels.service }}"
description: "EKS HTTP_5XX_Alert_Contour on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
@@ -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-order-service-internal_prd-order-service-internal-primary_80"}[1m])) by (cluster, envoy_cluster_name, le, job)) >= 750
for: 1m
labels:
severity: critical
bu: demand
team: transact
service: order-service-internal
env: prd
priority: p0
annotations:
summary: "EKS P99_Latency_Alert_Contour on {{ $labels.service }}"
description: "EKS P99_Latency_Alert_Contour on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
@@ -0,0 +1,16 @@
groups:
- name: EKS-App-Alerts
rules:
- alert: order_service_ORDER_ENTITY_NOT_FOUND_FOR_ROOT_ORDER_EKS
expr: sum by(cluster, service) (increase(ORDER_ENTITY_NOT_FOUND_FOR_ROOT_ORDER_value{kubernetes_namespace="prd-order-service-internal"}[1m])) >= 20
for: 1m
labels:
severity: critical
bu: demand
team: transact
service: order-service-internal
env: prd
priority: p0
annotations:
summary: "EKS order_service_ORDER_ENTITY_NOT_FOUND_FOR_ROOT_ORDER on {{ $labels.service }}"
description: "EKS order_service_ORDER_ENTITY_NOT_FOUND_FOR_ROOT_ORDER on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
@@ -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-ors", job=~"kubernetes-pods"}[1m])) by (cluster, service, ingress, job) * 60 >= 10
for: 2m
labels:
severity: critical
bu: demand
team: transact
service: ors
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 }}"
@@ -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-ors", job=~"kubernetes-pods"}[1m])) by (cluster, service, ingress, job) * 60 >= 10
for: 1m
labels:
severity: critical
bu: demand
team: transact
service: ors
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 }}"
@@ -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-ors", job=~"kubernetes-pods"}[1m])) by (cluster, service, le, ingress, namespace, job)) >= 0.75
for: 1m
labels:
severity: critical
bu: demand
team: transact
service: ors
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 }}"
@@ -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-trust-orch-layer_prd-trust-orch-layer-primary_80",envoy_response_code=~"[4].*"}[1m])) by (cluster, envoy_cluster_name, job) * 60 >= 10
for: 1m
labels:
severity: critical
bu: demand
team: transact
service: trust-orch-layer
env: prd
priority: p0
annotations:
summary: "EKS HTTP_4XX_Alert_Contour on {{ $labels.service }}"
description: "EKS HTTP_4XX_Alert_Contour on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
@@ -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-trust-orch-layer_prd-trust-orch-layer-primary_80",envoy_response_code=~"[5].*"}[1m])) by (cluster, envoy_cluster_name, job) * 60 >= 5
for: 1m
labels:
severity: critical
bu: demand
team: transact
service: trust-orch-layer
env: prd
priority: p0
annotations:
summary: "EKS HTTP_5XX_Alert_Contour on {{ $labels.service }}"
description: "EKS HTTP_5XX_Alert_Contour on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
@@ -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-trust-orch-layer_prd-trust-orch-layer-primary_80"}[1m])) by (cluster, envoy_cluster_name, le, job)) >= 160
for: 3m
labels:
severity: critical
bu: demand
team: transact
service: trust-orch-layer
env: prd
priority: p0
annotations:
summary: "EKS P99_Latency_Alert_Contour on {{ $labels.service }}"
description: "EKS P99_Latency_Alert_Contour on {{ $labels.service }} VALUE = {{ $value }}\n LABELS: {{ $labels }}"
@@ -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-wallet-web_prd-wallet-web-primary_80",envoy_response_code=~"[4].*"}[1m])) by (cluster, envoy_cluster_name, job) * 60 >= 10
for: 1m
labels:
severity: critical
bu: demand
team: transact
service: wallet-web
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 }}"
@@ -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-wallet-web_prd-wallet-web-primary_80",envoy_response_code=~"[5].*"}[1m])) by (cluster, envoy_cluster_name, job) * 60 >= 10
for: 1m
labels:
severity: critical
bu: demand
team: transact
service: wallet-web
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 }}"
@@ -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-wallet-web_prd-wallet-web-primary_80"}[1m])) by (cluster, envoy_cluster_name, le, job)) >= 750
for: 1m
labels:
severity: critical
bu: demand
team: transact
service: wallet-web
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 }}"