From fcf3c692bff7ea672b8017663a32929f36b01632 Mon Sep 17 00:00:00 2001 From: Mathod Date: Wed, 10 Dec 2025 14:58:12 +0100 Subject: [PATCH] rollback --- manifests/metrics-server/configmap.yaml | 347 ++++-------------------- 1 file changed, 48 insertions(+), 299 deletions(-) diff --git a/manifests/metrics-server/configmap.yaml b/manifests/metrics-server/configmap.yaml index 5bdd0d7..1eb4b20 100644 --- a/manifests/metrics-server/configmap.yaml +++ b/manifests/metrics-server/configmap.yaml @@ -3,302 +3,51 @@ kind: ConfigMap metadata: name: argocd-metrics-server-configmap data: - config.json: | - { - "prometheus": { - "applications": [ - { - "name": "default", - "default": true, - "dashboards": [ - { - "groupKind": "pod", - "tabs": ["Golden Signal"], - "rows": [ - { - "name": "pod", - "title": "Pods", - "tab": "Golden Signal", - "graphs": [ - { - "name": "pod_cpu_line", - "title": "CPU", - "description": "", - "graphType": "line", - "metricName": "pod", - "queryExpression": "sum(rate(container_cpu_usage_seconds_total{pod=~\"{{.name}}\", image!=\"\", container!=\"POD\", container!=\"\", container_name!=\"POD\"}[5m])) by (pod)" - }, - { - "name": "pod_cpu_pie", - "title": "CPU Avg", - "description": "", - "graphType": "pie", - "metricName": "pod", - "queryExpression": "sum(rate(container_cpu_usage_seconds_total{pod=~\"{{.name}}\", container!=\"POD\", image!=\"\", container!=\"\", container_name!=\"POD\"}[5m])) by (pod)" - }, - { - "name": "pod_memory_line", - "title": "Memory", - "description": "", - "graphType": "line", - "metricName": "pod", - "queryExpression": "sum(rate(container_memory_usage_bytes{pod=~\"{{.name}}\", container!=\"POD\", image!=\"\", container!=\"\", container_name!=\"POD\"}[5m])) by (pod)" - }, - { - "name": "pod_memory_pie", - "title": "Mem Avg", - "description": "", - "graphType": "pie", - "metricName": "pod", - "queryExpression": "sum(rate(container_memory_usage_bytes{pod=~\"{{.name}}\", container!=\"POD\", image!=\"\", container!=\"\", container_name!=\"POD\"}[5m])) by (pod)" - } - ] - }, - { - "name": "container", - "title": "Containers", - "tab": "Golden Signal", - "graphs": [ - { - "name": "container_cpu_line", - "title": "CPU", - "description": "", - "graphType": "line", - "metricName": "container", - "queryExpression": "sum(rate(container_cpu_usage_seconds_total{pod=~\"{{.name}}\", image!=\"\", container!=\"POD\", container!=\"\", container_name!=\"POD\"}[5m])) by (container)" - }, - { - "name": "container_cpu_pie", - "title": "CPU Avg", - "description": "", - "graphType": "pie", - "metricName": "container", - "queryExpression": "sum(rate(container_cpu_usage_seconds_total{pod=~\"{{.name}}\", image!=\"\",container!=\"POD\", container!=\"\", container_name!=\"POD\"}[5m])) by (container)" - }, - { - "name": "container_memory_line", - "title": "Memory", - "description": "", - "graphType": "line", - "metricName": "container", - "queryExpression": "sum(rate(container_memory_usage_bytes{pod=~\"{{.name}}\", image!=\"\", container!=\"POD\", container!=\"\", container_name!=\"POD\"}[5m])) by (container)" - }, - { - "name": "container_memory_pie", - "title": "Mem Avg", - "description": "", - "graphType": "pie", - "metricName": "container", - "queryExpression": "sum(rate(container_memory_usage_bytes{pod=~\"{{.name}}\", image!=\"\", container!=\"POD\", container!=\"\", container_name!=\"POD\"}[5m])) by (container)" - } - ] - } - ] - }, - { - "groupKind": "deployment", - "tabs": ["Golden Signal"], - "rows": [ - { - "name": "httplatency", - "title": "HTTP Latency", - "tab": "Golden Signal", - "graphs": [ - { - "name": "http_200_latency", - "title": "Latency", - "description": "", - "graphType": "line", - "metricName": "pod_template_hash", - "queryExpression": "sum(rate(http_server_requests_seconds_sum {namespace=\"{{.namespace}}\", status=\"200\"} [1m])) by (pod_template_hash)" - } - ] - }, - { - "name": "httperrortate", - "title": "HTTP Error Rate", - "tab": "Golden Signal", - "graphs": [ - { - "name": "http_error_rate_500", - "title": "HTTP Error 500", - "description": "", - "graphType": "line", - "metricName": "pod_template_hash", - "queryExpression": "sum(rate(http_server_requests_seconds_count {namespace=\"{{.namespace}}\", status=\"500\"} [1m])) by (pod_template_hash)" - }, - { - "name": "http_error_rate_400", - "title": "HTTP Error 400", - "description": "", - "graphType": "line", - "metricName": "pod_template_hash", - "queryExpression": "sum(rate(http_server_requests_seconds_count {namespace=\"{{.namespace}}\", status=\"404\"} [1m])) by (pod_template_hash)" - } - ] - }, - { - "name": "httptraffic", - "title": "HTTP Traffic", - "tab": "Golden Signal", - "graphs": [ - { - "name": "http_traffic", - "title": "Traffic", - "description": "", - "graphType": "line", - "metricName": "pod_template_hash", - "queryExpression": "sum(rate(http_server_requests_seconds_count {namespace=\"{{.namespace}}\"} [1m])) by (pod_template_hash)" - } - ] - }, - { - "name": "pod", - "title": "Pods", - "tab": "Golden Signal", - "graphs": [ - { - "name": "pod_cpu_line", - "title": "CPU", - "description": "", - "graphType": "line", - "metricName": "pod", - "queryExpression": "sum(rate(container_cpu_usage_seconds_total{pod=~\"{{.name}}\", image!=\"\", container!=\"POD\", container!=\"\", container_name!=\"POD\"}[5m])) by (pod)" - }, - { - "name": "pod_cpu_pie", - "title": "CPU Avg", - "description": "", - "graphType": "pie", - "metricName": "pod", - "queryExpression": "sum(rate(container_cpu_usage_seconds_total{pod=~\"{{.name}}\", container!=\"POD\", image!=\"\", container!=\"\", container_name!=\"POD\"}[5m])) by (pod)" - }, - { - "name": "pod_memory_line", - "title": "Memory", - "description": "", - "graphType": "line", - "metricName": "pod", - "queryExpression": "sum(rate(container_memory_usage_bytes{pod=~\"{{.name}}\", container!=\"POD\", image!=\"\", container!=\"\", container_name!=\"POD\"}[5m])) by (pod)" - }, - { - "name": "pod_memory_pie", - "title": "Mem Avg", - "description": "", - "graphType": "pie", - "metricName": "pod", - "queryExpression": "sum(rate(container_memory_usage_bytes{pod=~\"{{.name}}\", container!=\"POD\", image!=\"\", container!=\"\", container_name!=\"POD\"}[5m])) by (pod)" - } - ] - } - ] - }, - { - "groupKind": "rollout", - "tabs": ["Golden Signal"], - "rows": [ - { - "name": "httplatency", - "title": "HTTP Latency", - "tab": "Golden Signal", - "graphs": [ - { - "name": "http_200_latency", - "title": "Latency", - "description": "", - "graphType": "line", - "metricName": "rollout_template_hash", - "queryExpression": "sum(rate(http_server_requests_seconds_sum {namespace=\"{{.namespace}}\", status=\"200\"} [1m])) by (rollout_template_hash)" - } - ] - }, - { - "name": "httperrortate", - "title": "HTTP Error Rate", - "tab": "Golden Signal", - "graphs": [ - { - "name": "http_error_rate_500", - "title": "HTTP Error 500", - "description": "", - "graphType": "line", - "metricName": "rollout_template_hash", - "queryExpression": "sum(rate(http_server_requests_seconds_count {namespace=\"{{.namespace}}\", status=\"500\"} [1m])) by (rollout_template_hash)" - }, - { - "name": "http_error_rate_400", - "title": "HTTP Error 400", - "description": "", - "graphType": "line", - "metricName": "rollout_template_hash", - "queryExpression": "sum(rate(http_server_requests_seconds_count {namespace=\"{{.namespace}}\", status=\"404\"} [1m])) by (rollout_template_hash)" - } - ] - }, - { - "name": "httptraffic", - "title": "HTTP Traffic", - "tab": "Golden Signal", - "graphs": [ - { - "name": "http_traffic", - "title": "Traffic", - "description": "", - "graphType": "line", - "metricName": "rollout_template_hash", - "queryExpression": "sum(rate(http_server_requests_seconds_count {namespace=\"{{.namespace}}\"} [1m])) by (rollout_template_hash)" - } - ] - }, - { - "name": "pod", - "title": "Pods", - "tab": "Golden Signal", - "graphs": [ - { - "name": "pod_cpu_line", - "title": "CPU", - "description": "", - "graphType": "line", - "metricName": "pod", - "queryExpression": "sum(rate(container_cpu_usage_seconds_total{pod=~\"{{.name}}\", image!=\"\", container!=\"POD\", container!=\"\", container_name!=\"POD\"}[5m])) by (pod)" - }, - { - "name": "pod_cpu_pie", - "title": "CPU Avg", - "description": "", - "graphType": "pie", - "metricName": "pod", - "queryExpression": "sum(rate(container_cpu_usage_seconds_total{pod=~\"{{.name}}\", container!=\"POD\", image!=\"\", container!=\"\", container_name!=\"POD\"}[5m])) by (pod)" - }, - { - "name": "pod_memory_line", - "title": "Memory", - "description": "", - "graphType": "line", - "metricName": "pod", - "queryExpression": "sum(rate(container_memory_usage_bytes{pod=~\"{{.name}}\", container!=\"POD\", image!=\"\", container!=\"\", container_name!=\"POD\"}[5m])) by (pod)" - }, - { - "name": "pod_memory_pie", - "title": "Mem Avg", - "description": "", - "graphType": "pie", - "metricName": "pod", - "queryExpression": "sum(rate(container_memory_usage_bytes{pod=~\"{{.name}}\", container!=\"POD\", image!=\"\", container!=\"\", container_name!=\"POD\"}[5m])) by (pod)" - } - ] - } - ] - } - ] - } - ], - "provider": - { - "Name": "default", - "default": true, - "address": "http://prometheus-kube-prometheus-prometheus.monitoring.svc:9090" - } - } - } \ No newline at end of file + # Metrics for Deployments + extension.metrics.deployments: | + - name: "CPU Usage" + description: "CPU usage for the deployment" + type: "graph" + graphType: "area" + yAxisLabel: "CPU Cores" + query: 'sum(rate(container_cpu_usage_seconds_total{namespace="{{.metadata.namespace}}", pod=~"{{.metadata.name}}-.*"}[5m])) by (pod)' + + - name: "Memory Usage" + description: "Memory usage for the deployment" + type: "graph" + graphType: "area" + yAxisLabel: "Memory (MB)" + query: 'sum(container_memory_working_set_bytes{namespace="{{.metadata.namespace}}", pod=~"{{.metadata.name}}-.*"}) by (pod) / 1024 / 1024' + + - name: "Network Received" + description: "Network bytes received" + type: "graph" + graphType: "line" + yAxisLabel: "Bytes/sec" + query: 'sum(rate(container_network_receive_bytes_total{namespace="{{.metadata.namespace}}", pod=~"{{.metadata.name}}-.*"}[5m])) by (pod)' + + # Metrics for Pods + extension.metrics.pods: | + - name: "CPU Usage" + description: "CPU usage for the pod" + type: "graph" + graphType: "area" + yAxisLabel: "CPU Cores" + query: 'sum(rate(container_cpu_usage_seconds_total{namespace="{{.metadata.namespace}}", pod="{{.metadata.name}}"}[5m])) by (container)' + + - name: "Memory Usage" + description: "Memory usage for the pod" + type: "graph" + graphType: "area" + yAxisLabel: "Memory (MB)" + query: 'sum(container_memory_working_set_bytes{namespace="{{.metadata.namespace}}", pod="{{.metadata.name}}"}) by (container) / 1024 / 1024' + + - name: "Container Restarts" + description: "Number of container restarts" + type: "graph" + graphType: "line" + yAxisLabel: "Restarts" + query: 'kube_pod_container_status_restarts_total{namespace="{{.metadata.namespace}}", pod="{{.metadata.name}}"}' + + # Prometheus connection + prometheus.url: "http://prometheus-kube-prometheus-prometheus.monitoring.svc:9090" \ No newline at end of file