Optimization Constraints | Monitoring & Reporting

As part of the Cluster Score, Cast AI checks your workloads for configuration issues that affect reliability, availability, and cost optimization. These checks are called optimization constraints.

Each constraint flags a specific issue in the workload manifest and points to the location in the YAML where it was detected.

📘
Note
Optimization constraints are advisory — Cast AI does not modify your workloads to resolve them.

Constraint types

Missing health probes

Checks whether containers have liveness and readiness probes configured.

Probe type	Purpose
Liveness probe	Detects hung or deadlocked containers so Kubernetes can restart them.
Readiness probe	Prevents traffic from reaching containers that haven't finished starting.

Readiness probe checks are skipped for CronJob and Job workloads.

Pod Disruption Budget (PDB) issues

Checks whether PDBs matching your workloads could block voluntary disruptions such as node drains during rebalancing.

Issue	Description
Missing availability settings	A PDB exists but has neither `minAvailable` nor `maxUnavailable` set.
Too strict PDB	`maxUnavailable: 0` or `minAvailable` ≥ replica count blocks all voluntary disruptions.

Applies to workload types that have replicas: Deployments, ReplicaSets, StatefulSets, and ReplicationControllers.

Topology spread constraints

Checks whether workloads with 3 or more replicas have topology spread constraints and whether existing constraints have high skew values.

Issue	Description
Missing topology spread constraints	No `topologySpreadConstraints` configured, risking uneven pod distribution across zones or nodes.
High max skew	`maxSkew` exceeds the recommended threshold (2 for zone/region, 3 for hostname).

Skipped for Pod, Job, and CronJob workloads.

Suppressing constraints

To suppress all optimization constraint checks for a specific workload, add this annotation to its pod template (spec.template.metadata.annotations):

Annotation	Value	Where to apply	Description
`reporting.cast.ai/ignore-optimization-constraints`	`"true"` (case-sensitive — `"True"`, `"1"`, `"yes"` are ignored)	Pod template (`spec.template.metadata.annotations`) or bare Pod (`metadata.annotations`)	Suppresses all optimization constraint checks; you cannot suppress individual types. Removing the annotation re-enables checks on the next reporting cycle.

Example: the annotation in context, on a controller-managed workload and a bare Pod:

Add the annotation to the pod template at spec.template.metadata.annotations:

apiVersion: apps/v1
kind: Deployment
metadata:
  name: my-app
  namespace: default
spec:
  replicas: 3
  selector:
    matchLabels:
      app: my-app
  template:
    metadata:
      labels:
        app: my-app
      annotations:
        reporting.cast.ai/ignore-optimization-constraints: "true"
    spec:
      containers:
        - name: app
          image: my-app:latest

Without a pod template, place the annotation directly on metadata.annotations:

apiVersion: v1
kind: Pod
metadata:
  name: my-app
  namespace: default
  labels:
    app: my-app
  annotations:
    reporting.cast.ai/ignore-optimization-constraints: "true"
spec:
  containers:
    - name: app
      image: my-app:latest

To apply this annotation across many workloads at once, you can use Pod mutations to inject it automatically at admission time — useful for enforcing suppression consistently or in GitOps workflows where manifests aren't edited by hand.