diff --git a/helm/prometheus-rules/templates/_helpers.tpl b/helm/prometheus-rules/templates/_helpers.tpl index 549c7b21d..e0140af3d 100644 --- a/helm/prometheus-rules/templates/_helpers.tpl +++ b/helm/prometheus-rules/templates/_helpers.tpl @@ -53,14 +53,6 @@ true {{- end -}} {{- end -}} -{{- define "isClusterServiceInstalled" -}} -{{ not (eq .Values.managementCluster.provider.flavor "capi") }} -{{- end -}} - -{{- define "isVaultBeingMonitored" -}} -{{ not (eq .Values.managementCluster.provider.flavor "capi") }} -{{- end -}} - {{- define "isBastionBeingMonitored" -}} {{ not (eq .Values.managementCluster.provider.flavor "capi") }} {{- end -}} diff --git a/helm/prometheus-rules/templates/shared/alerting-rules/calico.rules.yml b/helm/prometheus-rules/templates/kaas/phoenix/alerting-rules/calico.rules.yml similarity index 94% rename from helm/prometheus-rules/templates/shared/alerting-rules/calico.rules.yml rename to helm/prometheus-rules/templates/kaas/phoenix/alerting-rules/calico.rules.yml index 9adbc20e6..c23c3a195 100644 --- a/helm/prometheus-rules/templates/shared/alerting-rules/calico.rules.yml +++ b/helm/prometheus-rules/templates/kaas/phoenix/alerting-rules/calico.rules.yml @@ -1,3 +1,4 @@ +## TODO Remove with vintage {{- if eq .Values.managementCluster.provider.flavor "vintage" }} apiVersion: monitoring.coreos.com/v1 kind: PrometheusRule @@ -21,7 +22,7 @@ spec: area: kaas cancel_if_outside_working_hours: {{ include "workingHoursOnly" . }} severity: notify - team: {{ include "providerTeam" . }} + team: phoenix topic: kubernetes - alert: CalicoNodeMemoryHighUtilization annotations: @@ -36,6 +37,6 @@ spec: area: kaas cancel_if_outside_working_hours: {{ include "workingHoursOnly" . }} severity: notify - team: {{ include "providerTeam" . }} + team: phoenix topic: kubernetes {{- end }} diff --git a/helm/prometheus-rules/templates/shared/alerting-rules/cluster-service.rules.yml b/helm/prometheus-rules/templates/kaas/phoenix/alerting-rules/cluster-service.rules.yml similarity index 90% rename from helm/prometheus-rules/templates/shared/alerting-rules/cluster-service.rules.yml rename to helm/prometheus-rules/templates/kaas/phoenix/alerting-rules/cluster-service.rules.yml index 9cbbeb887..8b332f2aa 100644 --- a/helm/prometheus-rules/templates/shared/alerting-rules/cluster-service.rules.yml +++ b/helm/prometheus-rules/templates/kaas/phoenix/alerting-rules/cluster-service.rules.yml @@ -1,4 +1,5 @@ -{{- if eq (include "isClusterServiceInstalled" .) "true" }} +## TODO Remove with vintage +{{- if eq .Values.managementCluster.provider.flavor "vintage" }} apiVersion: monitoring.coreos.com/v1 kind: PrometheusRule metadata: @@ -23,7 +24,7 @@ spec: labels: area: storage severity: page - team: {{ include "providerTeam" . }} + team: phoenix topic: managementcluster {{- if eq .Values.managementCluster.pipeline "testing" }} - alert: TestClusterTooOld @@ -33,5 +34,6 @@ spec: for: 5m labels: severity: notify + team: phoenix {{- end }} {{- end }} diff --git a/helm/prometheus-rules/templates/shared/alerting-rules/docker.rules.yml b/helm/prometheus-rules/templates/kaas/phoenix/alerting-rules/docker.rules.yml similarity index 94% rename from helm/prometheus-rules/templates/shared/alerting-rules/docker.rules.yml rename to helm/prometheus-rules/templates/kaas/phoenix/alerting-rules/docker.rules.yml index 9a189696c..9280e3467 100644 --- a/helm/prometheus-rules/templates/shared/alerting-rules/docker.rules.yml +++ b/helm/prometheus-rules/templates/kaas/phoenix/alerting-rules/docker.rules.yml @@ -1,3 +1,4 @@ +## TODO Remove with vintage {{- if eq .Values.managementCluster.provider.flavor "vintage" }} # newer clusters don't use docker anymore apiVersion: monitoring.coreos.com/v1 @@ -22,6 +23,6 @@ spec: area: kaas cancel_if_outside_working_hours: "true" severity: page - team: {{ include "providerTeam" . }} + team: phoenix topic: observability {{- end }} diff --git a/helm/prometheus-rules/templates/shared/alerting-rules/vault.rules.yml b/helm/prometheus-rules/templates/kaas/phoenix/alerting-rules/vault.rules.yml similarity index 89% rename from helm/prometheus-rules/templates/shared/alerting-rules/vault.rules.yml rename to helm/prometheus-rules/templates/kaas/phoenix/alerting-rules/vault.rules.yml index 9969d0cb9..7d4049d76 100644 --- a/helm/prometheus-rules/templates/shared/alerting-rules/vault.rules.yml +++ b/helm/prometheus-rules/templates/kaas/phoenix/alerting-rules/vault.rules.yml @@ -1,4 +1,5 @@ -{{- if eq (include "isVaultBeingMonitored" .) "true" }} +## TODO Remove with vintage +{{- if eq .Values.managementCluster.provider.flavor "vintage" }} apiVersion: monitoring.coreos.com/v1 kind: PrometheusRule metadata: @@ -23,7 +24,7 @@ spec: area: kaas cancel_if_outside_working_hours: {{ include "workingHoursOnly" . }} severity: page - team: {{ include "providerTeam" . }} + team: phoenix topic: vault - alert: VaultIsSealed annotations: @@ -35,7 +36,7 @@ spec: area: kaas cancel_if_outside_working_hours: {{ include "workingHoursOnly" . }} severity: page - team: {{ include "providerTeam" . }} + team: phoenix topic: vault - alert: ClusterServiceVaultTokenAlmostExpired annotations: @@ -47,7 +48,7 @@ spec: area: kaas cancel_if_outside_working_hours: "true" severity: page - team: {{ include "providerTeam" . }} + team: phoenix topic: vault - alert: ClusterServiceVaultTokenAlmostExpiredMissing annotations: @@ -60,7 +61,7 @@ spec: cancel_if_outside_working_hours: {{ include "workingHoursOnly" . }} cancel_if_prometheus_agent_down: "true" severity: page - team: {{ include "providerTeam" . }} + team: phoenix topic: vault - alert: CertOperatorVaultTokenAlmostExpired annotations: @@ -72,7 +73,7 @@ spec: area: kaas cancel_if_outside_working_hours: "true" severity: page - team: {{ include "providerTeam" . }} + team: phoenix topic: vault - alert: CertOperatorVaultTokenAlmostExpiredMissing annotations: @@ -84,7 +85,7 @@ spec: area: kaas cancel_if_outside_working_hours: {{ include "workingHoursOnly" . }} severity: page - team: {{ include "providerTeam" . }} + team: phoenix topic: vault - alert: VaultLatestEtcdBackupTooOld annotations: @@ -96,7 +97,7 @@ spec: area: kaas cancel_if_outside_working_hours: "true" severity: page - team: {{ include "providerTeam" . }} + team: phoenix topic: vault - alert: VaultLatestEtcdBackupMetricsMissing annotations: @@ -108,7 +109,6 @@ spec: area: kaas cancel_if_outside_working_hours: "true" severity: page - team: {{ include "providerTeam" . }} + team: phoenix topic: vault - {{- end }} diff --git a/helm/prometheus-rules/templates/shared/alerting-rules/service-level.rules.yml b/helm/prometheus-rules/templates/platform/atlas/alerting-rules/service-level.rules.yml similarity index 100% rename from helm/prometheus-rules/templates/shared/alerting-rules/service-level.rules.yml rename to helm/prometheus-rules/templates/platform/atlas/alerting-rules/service-level.rules.yml