Configuration¶

All configuration is done via values.yaml. Below are the key parameters.

Required¶

Key	Description
`clusterName`	Cluster identifier — used as the `cluster` label on all metrics and alerts
`datasources.victoriametrics.url`	Prometheus read endpoint (PromQL-compatible)
`datasources.loki.url`	Loki read endpoint (LogQL)
`datasources.alertmanager.url`	Alertmanager v2 API endpoint

Controller¶

Key	Default	Description
`controller.replicaCount`	`2`	Replicas (HA via Kubernetes Lease)
`controller.dryRun`	`true`	Log alerts without dispatching to Alertmanager
`controller.leaderElection.enabled`	`true`	K8s Lease leader election
`controller.tickInterval`	`30s`	Interval between detection cycles

Workers¶

Key	Default	Description
`worker.replicaCount`	`3`	Worker replicas (stateless, horizontally scalable)
`worker.concurrency`	`5`	Parallel detection jobs per worker

ML Service¶

Key	Default	Description
`ml.enabled`	`true`	Enable the Python ML service
`ml.image.repository`	`ghcr.io/staffops/anomaly-detection-ml`	ML service image

Redis¶

Key	Default	Description
`redis.enabled`	`true`	Deploy in-cluster Redis
`redis.persistence.enabled`	`false`	PVC for Redis data (survives pod restarts)
`redis.external.addr`	`""`	External Redis address when `redis.enabled=false`
`redis.external.existingSecret`	`""`	Secret with `redis-password` key

Detection rules¶

Static thresholds¶

detection:
  staticRules:
    - name: high-cpu
      expr: 'avg by (pod) (rate(container_cpu_usage_seconds_total[5m])) > 0.9'
      severity: warning
      labels:
        team: platform

Adaptive (Z-Score)¶

detection:
  adaptiveMetrics:
    - name: request-rate
      query: 'sum(rate(http_requests_total[5m]))'
      window: 1h
      threshold: 3.0   # standard deviations

Log patterns (Loki)¶

detection:
  logPatterns:
    - name: oom-kill
      query: '{namespace="production"} |= "OOMKilled"'
      threshold: 5     # occurrences per window
      window: 10m

Suppression¶

suppression:
  excludeNamespaces:
    - kube-system
    - monitoring
  excludeStaticOnly:
    - staging    # adaptive still fires, static rules suppressed

Observability integrations¶

Key	Default	Description
`serviceMonitor.enabled`	`false`	Prometheus Operator ServiceMonitor
`vmServiceScrape.enabled`	`false`	vm-operator ServiceMonitor
`vmRule.enabled`	`false`	vm-operator PrometheusRule (health alerts + recording rules)
`grafanaDashboard.enabled`	`false`	ConfigMap labelled for Grafana sidecar discovery