It is possible for an application to be
OutOfSync even immediately after a successful Sync operation. Some reasons for this might be:
- There is a bug in the manifest, where it contains extra/unknown fields from the actual K8s spec. These extra fields would get dropped when querying Kubernetes for the live state,
resulting in an
OutOfSyncstatus indicating a missing field was detected.
- The sync was performed (with pruning disabled), and there are resources which need to be deleted.
- A controller or mutating webhook is altering the object after it was submitted to Kubernetes in a manner which contradicts Git.
- A Helm chart is using a template function such as
randAlphaNum, which generates different data every time
helm templateis invoked.
- For Horizontal Pod Autoscaling (HPA) objects, the HPA controller is known to reorder
spec.metricsin a specific order. See kubernetes issue #74099. To work around this, you can order
spec.metricsin Git in the same order that the controller prefers.
In case it is impossible to fix the upstream issue, Argo CD allows you to optionally ignore differences of problematic resources. The diffing customization can be configured for single or multiple application resources or at a system level.
Application Level Configuration¶
Argo CD allows ignoring differences at a specific JSON path, using RFC6902 JSON patches and JQ path expressions. The following sample application is configured to ignore differences in
spec.replicas for all deployments:
spec: ignoreDifferences: - group: apps kind: Deployment jsonPointers: - /spec/replicas
The above customization could be narrowed to a resource with the specified name and optional namespace:
spec: ignoreDifferences: - group: apps kind: Deployment name: guestbook namespace: default jsonPointers: - /spec/replicas
To ignore elements of a list, you can use JQ path expressions to identify list items based on item content:
spec: ignoreDifferences: - group: apps kind: Deployment jqPathExpressions: - .spec.template.spec.initContainers | select(.name == "injected-init-container")
The comparison of resources with well-known issues can be customized at a system level. Ignored differences can be configured for a specified group and kind
resource.customizations key of
argocd-cm ConfigMap. Following is an example of a customization which ignores the
data: resource.customizations.ignoreDifferences.admissionregistration.k8s.io_MutatingWebhookConfiguration: | jqPathExpressions: - '.webhooks?.clientConfig.caBundle'
status field of
CustomResourceDefinitions is often stored in Git/Helm manifest and should be ignored during diffing. The
ignoreResourceStatusField setting simplifies
handling that edge case:
data: resource.compareoptions: | # disables status field diffing in specified resource types # 'crd' - CustomResourceDefinition-s (default) # 'all' - all resources # 'none' - disabled ignoreResourceStatusField: crd
status field is ignored during diffing for
CustomResourceDefinition resource. The behavior can be extended to all resources using
all value or disabled using
Known Kubernetes types in CRDs (Resource limits, Volume mounts etc)¶
Some CRDs are re-using data structures defined in the Kubernetes source base and therefore inheriting custom JSON/YAML marshaling. Custom marshalers might serialize CRDs in a slightly different format that causes false positives during drift detection.
A typical example is the
argoproj.io/Rollout CRD that re-using
core/v1/PodSpec data structure. Pod resource requests
might be reformatted by the custom marshaller of
IntOrString data type:
resources: requests: cpu: 100m
resources: requests: cpu: 0.1
The solution is to specify which CRDs fields are using built-in Kubernetes types in the
apiVersion: v1 kind: ConfigMap metadata: name: argocd-cm namespace: argocd labels: app.kubernetes.io/name: argocd-cm app.kubernetes.io/part-of: argocd data: resource.customizations.knownTypeFields.argoproj.io_Rollout: | - field: spec.template.spec type: core/v1/PodSpec
The list of supported Kubernetes types is available in diffing_known_types.txt