Skip to content

Output

What setdrift export writes.

.gitignore keeps snapshots/ out of Git
snapshots/<time>-<cluster>.pb, .json what each export read (not in Git)
verify/<time>-<cluster>-report.md, .json each setdrift verify run
clusters/<cluster>/
README.md what the export found, and how to apply it
RESTORE.md the restore playbook
kustomization.yaml everything that can be applied, in order
crds/<name>.yaml
cluster-scoped/<kind>[.<group>]/<name>.yaml
namespaces/<ns>/<kind>[.<group>]/<name>.yaml
redacted/namespaces/<ns>/<kind>[.<group>]/<name>.yaml.redacted
external-secrets/namespaces/<ns>/externalsecret.external-secrets.io/<name>.yaml
restore/volumes/<ns>/<claim>.yaml
restore/storageclasses/<name>.yaml
platform/<cloud>/<distribution>/ e.g. platform/aws/eks, platform/local/kind
.setdrift-export where and when the export was taken

The cluster’s directory applies to any Kubernetes cluster. SetDrift also works out the platform it ran on (EKS, AKS, GKE, OpenShift, k3s, RKE2, kubeadm, kind, minikube, Docker Desktop, and the cloud underneath) from the kubeconfig, the API server and the cluster’s objects, and acts only when the evidence is clear. Then platform/<cloud>/<distribution>/ (or platform/<cloud>/vm/ for a self-managed cluster on a cloud’s VMs) is an overlay of the generic set plus the objects only that platform serves, with its own README and, for AWS and EKS, its own restore steps (platform/aws/RESTORE.md, platform/aws/eks/RESTORE.md).

CRDs are in crds/, first in the root kustomization.yaml. Argo CD and Flux apply them before the resources that use them; with kubectl, apply crds/ first (the cluster’s README.md has the commands), because kubectl rejects a custom resource whose definition does not exist yet. Custom resources whose definition the export lacks are listed in RESTORE.md, to install first.

Helm does not record which repository a chart came from. RESTORE.md takes it from Argo CD or Flux installing the same chart, or from a file passed with --helm-repos, and lists the charts left to fill in; see contrib/helm, which also has a post-renderer that records the repository at install time (setdrift.io/chart-repo).

--keep-source-defaults keeps the storage class a claim got from this cluster’s default class (normally left out), so claims ask for the same class by name in any cluster. RESTORE.md creates each class the claims use when the target lacks it, from this cluster’s definition, and says how to point it at another platform’s provisioner. Volumes another cluster can reach (EBS, and other CSI volumes) get a PersistentVolume in restore/volumes/, bound in advance to its claim; volumes on a node’s own disk (local, hostPath, kind’s local-path) are reported as not portable.

export writes into an empty directory. To export again into a repository, pass --force: it replaces only clusters/<cluster>/, so resources gone from the cluster don’t leave stale files, and keeps everything else, including the earlier snapshots and verify reports. It asks first; --yes skips the question in scripts.