Smells

The results of a KubePattern analysis are Smells: custom resources that each describe one defect on one resource. A Smell is produced by a Pattern, and it is kept up to date by every run until the defect is fixed.

apiVersion: kubepattern.dev/v1
kind: Smell
metadata:
  name: cnpg-cluster-without-scheduledbackup-9c1e4d2a-5b7f-4c1e-8a3d-2f6b0e91c7a4
  namespace: orders
spec:
  name: cnpg-cluster-without-scheduledbackup
  category: Reliability
  severity: HIGH
  message: "Cluster orders/pg-orders has no ScheduledBackup."
  suppress: false
  pattern:
    name: cnpg-cluster-without-scheduledbackup
    version: kubepattern.dev/v1
  target:
    apiVersion: postgresql.cnpg.io/v1
    kind: Cluster
    name: pg-orders
    namespace: orders
    uid: 9c1e4d2a-5b7f-4c1e-8a3d-2f6b0e91c7a4

Retrieving results

Smells are namespaced resources, and smell is also the short name.

# every smell in the cluster
kubectl get smells -A

# one smell in detail
kubectl describe smell <smell-name> -n <namespace>

# as YAML
kubectl get smell <smell-name> -n <namespace> -o yaml

# the smells of one pattern, using its UID
kubectl get smells -A -l kubepattern.dev/pattern-uid=<pattern-uid>

Patterns, in contrast, are cluster-scoped: kubectl get patterns needs no namespace.

Filtering by field

kubectl --field-selector does not work on custom resource fields such as spec.severity. Use labels, or output to JSON and filter with jq:

kubectl get smells -A -o json \
  | jq -r '.items[] | select(.spec.severity=="HIGH") | .metadata.namespace + "/" + .metadata.name'

Columns

kubectl get smells shows the following columns. The last three appear with -o wide.

Column	Field
Severity	spec.severity
Phase next release	spec.phase
Since next release	spec.since	shown as an age, such as 3d
Target Kind	spec.target.kind
Target Name	spec.target.name
Target Namespace	spec.target.namespace	-o wide
Message	spec.message	-o wide
Suppressed	spec.suppress	-o wide

Where a Smell is stored

By default, a Smell is stored in the namespace of the resource it concerns. Smells about cluster-scoped resources, or all Smells if so configured, go to a dedicated namespace. Set analysis.saveInNamespace and analysis.targetNamespace in the chart to choose. See Getting Started.

Naming and identity

The name of a Smell is <pattern-name>-<target-uid>. The same finding is therefore updated in place across runs instead of being created again, and kubectl get smells -A always lists the current problems.

A Smell raised by an absence pattern, which has no target resource, is named <pattern-name>-empty.

The Smell resource

Property	Description
apiVersion	kubepattern.dev/v1.
kind	Always Smell.
metadata.name	<pattern-name>-<target-uid>.
metadata.namespace	The namespace of the target, unless configured otherwise.
metadata.labels	kubepattern.dev/pattern-uid next release links the Smell to its pattern. kubepattern.dev/phase next release mirrors spec.phase, so a selector can keep only Active smells. lastScan is the id of the run that last wrote it.
spec.name	The name of the pattern that produced the Smell.
spec.category	The category of the pattern.
spec.severity	LOW, MEDIUM, HIGH or CRITICAL.
spec.message	The message of the pattern, with its placeholders filled in.
spec.reference	A link to documentation on how to fix the Smell.
spec.pattern	The name of the pattern and its API version.
spec.target	The apiVersion, kind, name, namespace and uid of the resource the Smell is reported on.
spec.phase next release	Pending until the condition has held for the pattern's for, Active afterwards.
spec.since next release	When the condition was first observed.
spec.suppress	Whether the Smell has been marked as a false positive.

Suppressing a false positive

A pattern can only see the references it describes, so now and then it flags something you know is fine. Mark that Smell as suppressed:

kubectl patch smell <smell-name> -n <namespace> \
  --type merge -p '{"spec":{"suppress":true}}'

KubePattern keeps the flag on the following runs, so the suppression survives. next release Earlier versions reset it on every update. A suppressed Smell is still listed by kubectl get smells, so filter on spec.suppress in your own dashboards and alerts.

When a Smell disappears

KubePattern removes a Smell as soon as its target no longer satisfies the pattern. Fix the problem, and the Smell is gone on the next run, with no manual clean-up. A Smell is also removed when its pattern is deleted.

If a pattern is skipped, for example because the analyzer cannot read one of its kinds, its Smells are kept rather than removed. next release Lifecycle has the details.