added eviction-metric

Author: christophrichtersap

charts/openstack-hypervisor-operator/alerts/eviction.yaml

@@ -0,0 +1,42 @@
+# SPDX-FileCopyrightText: 2025 SAP SE or an SAP affiliate company and cobaltcore-dev contributors
+# SPDX-License-Identifier: Apache-2.0
+
+groups:
+- name: evictionLifecycle
+  rules:
+  - alert: EvictionFailed
+    expr: |
+      kube_customresource_eviction_condition{condition="Evicting", reason="Failed"} == 1
+    for: 5m
+    labels:
+      severity: warning
+      type: hypervisor_operator
+    annotations:
+      summary: "Eviction {{ $labels.name }} has failed"
+      description: "The eviction {{ $labels.name }} for hypervisor {{ $labels.hypervisor }} has reached a terminal failure state. Manual intervention is required — check if the hypervisor exists in OpenStack."
+
+  - alert: EvictionMigrationFailing
+    expr: |
+      kube_customresource_eviction_condition{condition="MigratingInstance", reason="Failed"} == 1
+        and on (name)
+      kube_customresource_eviction_condition{condition="Evicting", reason="Running"} == 1
+    for: 1h
+    labels:
+      severity: warning
+      type: hypervisor_operator
+    annotations:
+      summary: "Eviction {{ $labels.name }} has failing instance migrations for over 1 hour"
+      description: "The eviction {{ $labels.name }} has had MigratingInstance=Failed for more than 1 hour while still running. Instances may be in ERROR state, blocking eviction progress."
+
+  - alert: EvictionOutstandingRamHigh
+    expr: |
+      kube_customresource_eviction_outstanding_ram_mb > 0
+        and on (name)
+      kube_customresource_eviction_condition{condition="Evicting", reason="Running"} == 1
+    for: 6h
+    labels:
+      severity: warning
+      type: hypervisor_operator
+    annotations:
+      summary: "Eviction {{ $labels.name }} has outstanding RAM for over 6 hours"
+      description: "The eviction {{ $labels.name }} has had {{ $value }}MB of outstanding RAM for more than 6 hours. Check for stuck live-migrations or instances that cannot be moved."

charts/openstack-hypervisor-operator/templates/metrics-eviction-cm.yaml

@@ -0,0 +1,53 @@
+# SPDX-FileCopyrightText: 2025 SAP SE or an SAP affiliate company and cobaltcore-dev contributors
+# SPDX-License-Identifier: Apache-2.0
+
+{{- if .Values.customResourceMetrics.create }}
+{{- $disabled := .Values.customResourceMetrics.disabled | default dict }}
+{{- if not (index $disabled "eviction") }}
+apiVersion: v1
+kind: ConfigMap
+metadata:
+  name: {{ printf "%s-eviction-resources" .Release.Name | trunc 63 | trimSuffix "-" }}
+  labels:
+    dev.custom.kube-state-metrics: "true"
+{{ include "openstack-hypervisor-operator.labels" . | indent 4 }}
+data:
+  eviction-metrics.yaml: |
+    spec:
+      resources:
+        - groupVersionKind:
+            group: kvm.cloud.sap
+            version: v1
+            kind: Eviction
+          labelsFromPath:
+            name: [metadata, name]
+          metrics:
+            - name: eviction_info
+              help: "Info metric for eviction with hypervisor and reason labels (always 1)"
+              each:
+                type: Info
+                info:
+                  labelsFromPath:
+                    hypervisor: [spec, hypervisor]
+                    reason: [spec, reason]
+            - name: eviction_outstanding_ram_mb
+              help: "Outstanding RAM in MB to be migrated"
+              each:
+                type: Gauge
+                gauge:
+                  path: [status, outstandingRamMb]
+                  nilIsZero: true
+            - name: eviction_condition
+              help: "Eviction condition status (1=True, 0=False)"
+              each:
+                type: Gauge
+                gauge:
+                  path: [status, conditions]
+                  labelsFromPath:
+                    condition: [type]
+                    reason: [reason]
+                  valueFrom: [status]
+                  booleanTrue: "True"
+                  booleanFalse: "False"
+{{- end }}
+{{- end }}

…r/templates/custom-resource-metrics.yaml …tor/templates/metrics-hypervisor-cm.yamlcharts/openstack-hypervisor-operator/templates/custom-resource-metrics.yaml renamed to charts/openstack-hypervisor-operator/templates/metrics-hypervisor-cm.yaml charts/openstack-hypervisor-operator/templates/custom-resource-metrics.yaml renamed to charts/openstack-hypervisor-operator/templates/metrics-hypervisor-cm.yaml

@@ -2,6 +2,8 @@
 # SPDX-License-Identifier: Apache-2.0
 
 {{- if .Values.customResourceMetrics.create }}
+{{- $disabled := .Values.customResourceMetrics.disabled | default dict }}
+{{- if not (index $disabled "hypervisor") }}
 apiVersion: v1
 kind: ConfigMap
 metadata:
@@ -88,3 +90,4 @@ data:
                   booleanFalse: "false"
                   nilIsZero: true
 {{- end }}
+{{- end }}

charts/openstack-hypervisor-operator/values.yaml

@@ -67,6 +67,9 @@ prometheusRules:
   # disabled:
   #   HypervisorOnboardingStuck: true
   #   HypervisorEvictedTooLong: true
+  #   EvictionFailed: true
+  #   EvictionMigrationFailing: true
+  #   EvictionOutstandingRamHigh: true
   disabled: {}
 
 dashboards:
@@ -80,3 +83,8 @@ dashboards:
 
 customResourceMetrics:
   create: true
+  # Disable individual custom resource metrics ConfigMaps by name:
+  # disabled:
+  #   hypervisor: true
+  #   eviction: true
+  disabled: {}