diff --git a/k8s/monitoring/templates/node-alerts.yaml b/k8s/monitoring/templates/node-alerts.yaml index b78006b..6578bfe 100644 --- a/k8s/monitoring/templates/node-alerts.yaml +++ b/k8s/monitoring/templates/node-alerts.yaml @@ -10,7 +10,7 @@ spec: - name: node.alerts rules: - alert: NodeHighCPU - expr: (1 - avg by (instance, job) (rate(node_cpu_seconds_total{mode="idle"}[5m]))) * 100 > 20 + expr: (1 - avg without (cpu, mode) (rate(node_cpu_seconds_total{mode="idle"}[5m]))) * 100 > 20 for: 2m labels: severity: warning diff --git a/k8s/monitoring/templates/session-telemetry-dashboard.yaml b/k8s/monitoring/templates/session-telemetry-dashboard.yaml index 988ddcd..253e119 100644 --- a/k8s/monitoring/templates/session-telemetry-dashboard.yaml +++ b/k8s/monitoring/templates/session-telemetry-dashboard.yaml @@ -268,21 +268,33 @@ data: }, "targets": [ { - "expr": "{namespace=\"webrtc-be\", container=\"webrtc-be\"} |= \"type=peer_telemetry\" | logfmt", + "expr": "{namespace=\"webrtc-be\", container=\"webrtc-be\"} |= \"type=peer_telemetry\" | pattern \"[TELEMETRY] \" | line_format \"{{.data}}\" | logfmt", "refId": "A" } ], "title": "Raw Peer Telemetry", "transformations": [ + { + "id": "labelsToFields", + "options": { + "mode": "columns" + } + }, { "id": "organize", "options": { "excludeByName": { "app": true, + "app_kubernetes_io_component": true, + "app_kubernetes_io_name": true, + "cluster": true, "container": true, + "filename": true, "job": true, "level": true, "namespace": true, + "pod": true, + "pod_template_hash": true, "stream": true, "type": true }, @@ -332,21 +344,33 @@ data: }, "targets": [ { - "expr": "{namespace=\"webrtc-be\", container=\"webrtc-be\"} |= \"type=join_telemetry\" | logfmt", + "expr": "{namespace=\"webrtc-be\", container=\"webrtc-be\"} |= \"type=join_telemetry\" | pattern \"[TELEMETRY] \" | line_format \"{{.data}}\" | logfmt", "refId": "A" } ], "title": "Raw Join Telemetry", "transformations": [ + { + "id": "labelsToFields", + "options": { + "mode": "columns" + } + }, { "id": "organize", "options": { "excludeByName": { "app": true, + "app_kubernetes_io_component": true, + "app_kubernetes_io_name": true, + "cluster": true, "container": true, + "filename": true, "job": true, "level": true, "namespace": true, + "pod": true, + "pod_template_hash": true, "stream": true, "type": true }, @@ -393,21 +417,33 @@ data: }, "targets": [ { - "expr": "{namespace=\"webrtc-be\", container=\"webrtc-be\"} |= \"type=join_summary_telemetry\" | logfmt", + "expr": "{namespace=\"webrtc-be\", container=\"webrtc-be\"} |= \"type=join_summary_telemetry\" | pattern \"[TELEMETRY] \" | line_format \"{{.data}}\" | logfmt", "refId": "A" } ], "title": "Raw Join Summary Telemetry", "transformations": [ + { + "id": "labelsToFields", + "options": { + "mode": "columns" + } + }, { "id": "organize", "options": { "excludeByName": { "app": true, + "app_kubernetes_io_component": true, + "app_kubernetes_io_name": true, + "cluster": true, "container": true, + "filename": true, "job": true, "level": true, "namespace": true, + "pod": true, + "pod_template_hash": true, "stream": true, "type": true }, @@ -453,21 +489,33 @@ data: }, "targets": [ { - "expr": "{namespace=\"webrtc-be\", container=\"webrtc-be\"} |= \"type=leave_summary_telemetry\" | logfmt", + "expr": "{namespace=\"webrtc-be\", container=\"webrtc-be\"} |= \"type=leave_summary_telemetry\" | pattern \"[TELEMETRY] \" | line_format \"{{.data}}\" | logfmt", "refId": "A" } ], "title": "Raw Leave Summary Telemetry", "transformations": [ + { + "id": "labelsToFields", + "options": { + "mode": "columns" + } + }, { "id": "organize", "options": { "excludeByName": { "app": true, + "app_kubernetes_io_component": true, + "app_kubernetes_io_name": true, + "cluster": true, "container": true, + "filename": true, "job": true, "level": true, "namespace": true, + "pod": true, + "pod_template_hash": true, "stream": true, "type": true },