Uninstall

This guide describes how to remove TDP Kubernetes components in an organized and safe way.

Before you start

Distinguish what is being removed:

• components installed directly via Helm
• Applications managed by ArgoCD
• ArgoCD itself, when it is also part of the removal

This document clearly separates each scenario.

Attention

Removing Kubernetes resources does not automatically remove persisted data.
PVCs, PVs, snapshots and namespace must be evaluated separately.

Before performing any removal, identify how the component or environment was deployed.

1. Check if the component was installed via Helm

Use this command to discover existing Helm releases in the namespace:

Terminal input

helm list -n <NAMESPACE>

Figure 1 - Existing Helm Releases

If the component appears in this listing, the main removal is done with helm uninstall.

2. Check if the component is managed by ArgoCD

Use this command to discover Applications managed by ArgoCD:

Terminal input

argocd app list

Figure 2 - Applications managed by ArgoCD

If the component appears as an ArgoCD Application, the main removal is done with argocd app delete.

Rule of thumb

Do not mix flows unnecessarily.
If the component is managed by ArgoCD, the main path is to remove the Application.
If it was installed directly via Helm, the main path is to remove the Helm release.

Preparations before removal

Back up data

Before removing any stateful components, back up anything that needs to be preserved.

Common examples:

• PostgreSQL: database dump
• ClickHouse: export of tables and schemas
• Airflow: DAGs, variables and connections
• OpenMetadata: metadata and lineage
• Ranger: relevant policies and settings

Export Helm release configuration

Once the component has been installed via Helm, export the currently applied values:

Terminal input

helm get values <RELEASE_NAME> -n <NAMESPACE> -o yaml > <RELEASE_NAME>-values-backup.yaml

Figure 3 - Exporting the Helm release configuration

Note

This file typically contains values overwritten during installation or upgrade.
To consult all parameters accepted by the chart, use helm show values on the corresponding chart.

To export the values ​​for all releases in the namespace at once:

Terminal input

for release in $(helm list -n <NAMESPACE> -q); do
  helm get values "$release" -n <NAMESPACE> -o yaml > "${release}-values-backup.yaml"
  echo "$release backup completed"
done

Identify associated PVCs

Before removing the component, verify that there are persistent volumes associated with it:

Terminal input

kubectl get pvc -n <NAMESPACE>

Figure 4 - PVCs associated with the release

To restrict the query to a specific Helm release:

Terminal input

kubectl get pvc -n <NAMESPACE> -l app.kubernetes.io/instance=<RELEASE_NAME>

Create snapshot of persistent volumes, when supported

If the cluster supports CSI snapshots and there is VolumeSnapshotClass available, create snapshots of the critical PVCs before removal.

Check if the cluster supports snapshots:

Terminal input

kubectl get volumesnapshotclass

Create the snapshot:

Terminal input

kubectl apply -f - <<EOT
apiVersion: snapshot.storage.k8s.io/v1
kind: VolumeSnapshot
metadata:
  name: <pvc-name>-final-snapshot
  namespace: <NAMESPACE>
spec:
  volumeSnapshotClassName: <snapshot-class>
  source:
    persistentVolumeClaimName: <pvc-name>
EOT

Verify the result:

Terminal input

kubectl get volumesnapshot -n <NAMESPACE>

Evaluate CRDs before removing the platform

The tdp-crds chart installs cluster-scoped resources. These resources do not belong to a specific namespace and may remain after removing Helm releases or Argo CD Applications.

Before removing CRDs, confirm that all components that depend on them have already been removed and that no custom resources remain:

Terminal input

kubectl get crd

To evaluate remaining objects, check the types installed by the operators used in the platform, such as Argo CD, Strimzi, NiFiKop, Altinity ClickHouse Operator, OpenMetadata OMJob Operator and cert-manager.

caution

Remove CRDs only during full platform uninstallation or controlled administrative cleanup. Removing a CRD also removes the corresponding custom API and may affect resources that still exist in the cluster.

Remove components

Via Helm

Commands

Uninstall an individual component

Use this flow when the component was installed directly as a Helm release.

Uninstall the release

Terminal input

helm uninstall <RELEASE_NAME> -n <NAMESPACE>

Figure 5 - Uninstalling a component via Helm

Check if release features have been removed

Terminal input

kubectl get all -n <NAMESPACE> -l app.kubernetes.io/instance=<RELEASE_NAME>

Figure 6 - Release-managed resources after uninstallation

note

This command shows what is still associated with the release after helm uninstall.
If the result is empty, the resources were removed successfully.

Check remaining PVCs

Terminal input

kubectl get pvc -n <NAMESPACE> -l app.kubernetes.io/instance=<RELEASE_NAME>

Figure 7 - Remaining PVCs

Note

helm uninstall removes the release and its associated Kubernetes resources — Deployments, StatefulSets, Services, ConfigMaps, and Secrets.
PVCs are normally not removed automatically.

Remove complete environment via Helm

Use this flow when the entire environment has been deployed directly via Helm, without Applications managed by ArgoCD.

Remove releases in reverse dependency order

Terminal input

helm uninstall tdp-superset -n <NAMESPACE>
helm uninstall tdp-openmetadata -n <NAMESPACE>
helm uninstall tdp-cloudbeaver -n <NAMESPACE>
helm uninstall tdp-jupyter -n <NAMESPACE>
helm uninstall tdp-airflow -n <NAMESPACE>
helm uninstall tdp-clickhouse -n <NAMESPACE>
helm uninstall tdp-trino -n <NAMESPACE>
helm uninstall tdp-spark -n <NAMESPACE>
helm uninstall tdp-nifi -n <NAMESPACE>
helm uninstall tdp-ranger -n <NAMESPACE>
helm uninstall tdp-iceberg -n <NAMESPACE>
helm uninstall tdp-deltalake -n <NAMESPACE>
helm uninstall tdp-ozone -n <NAMESPACE>
helm uninstall tdp-hive-metastore -n <NAMESPACE>

Kafka and Strimzi CRDs

Before uninstalling tdp-kafka, verify that there are no pending Strimzi resources in the namespace:

Terminal input

kubectl get kafkas,kafkatopics,kafkausers,kafkaconnects,kafkaconnectors -n <NAMESPACE>

Remove custom resources before uninstalling the chart to avoid orphaned CRDs in the cluster.

Terminal input

helm uninstall tdp-kafka -n <NAMESPACE>
helm uninstall tdp-postgresql -n <NAMESPACE>
helm uninstall tdp-argo -n <NAMESPACE>
helm uninstall tdp-crds -n <NAMESPACE>

Check remaining PVCs

Terminal input

kubectl get pvc -n <NAMESPACE>

Remove PVCs, if removal is permanent

Terminal input

kubectl delete pvc --all -n <NAMESPACE>

Figure 8 - Permanent removal of PVCs

Data loss

Deleting PVCs permanently removes persisted data.
Do this only when disassembly is final and backups are complete.

Videos

helm list

helm get values

helm uninstall

kubectl get pvc

kubectl get pvc (por release)

kubectl get all

kubectl delete pvc

kubectl delete pvc --all

kubectl get pv

Full demo

Via ArgoCD

Commands

Remove an individual component via ArgoCD

Use this flow when the component is managed as an ArgoCD Application.

Remove the Application

Terminal input

argocd app delete <componente> --cascade

Figure 9 - Removing an Application on ArgoCD

Verify removal

Terminal input

argocd app list

Figure 10 - Verification of removal in ArgoCD

Check remaining resources in the namespace

Terminal input

kubectl get all -n <NAMESPACE>

Figure 11 - Checking for remaining resources in the Namespace

Check remaining PVCs

Terminal input

kubectl get pvc -n <NAMESPACE>

Figure 12 - Remaining PVCs

Attention

The --cascade option removes the Application and the Kubernetes resources managed by it.
Even so, PVCs, PVs and other infrastructure waste may require additional checking and cleaning.

Remove the complete environment via ArgoCD

Use this path when the TDP environment is fully deployed and managed by ArgoCD.

Option A — Remove a root app of type app-of-apps

If the environment uses a root Application that controls the others, the removal can start with it:

Terminal input

argocd app delete <root-application> --cascade

Use this option only when the app-of-apps model is validated in your environment and you are confident that deleting the root app will correctly remove child applications and their resources.

Option B — Remove Applications individually in reverse order of dependency

When there is no root app, or when you want full control of the disassembly, remove the Applications one by one:

Terminal input

argocd app delete tdp-superset --cascade
argocd app delete tdp-openmetadata --cascade
argocd app delete tdp-cloudbeaver --cascade
argocd app delete tdp-jupyter --cascade
argocd app delete tdp-airflow --cascade
argocd app delete tdp-clickhouse --cascade
argocd app delete tdp-trino --cascade
argocd app delete tdp-spark --cascade
argocd app delete tdp-nifi --cascade
argocd app delete tdp-ranger --cascade
argocd app delete tdp-iceberg --cascade
argocd app delete tdp-deltalake --cascade
argocd app delete tdp-ozone --cascade
argocd app delete tdp-hive-metastore --cascade
argocd app delete tdp-kafka --cascade
argocd app delete tdp-postgresql --cascade

Check for remaining Applications

Terminal input

argocd app list

Check for remaining resources in the namespace

Terminal input

kubectl get all -n <NAMESPACE>

Check remaining PVCs

Terminal input

kubectl get pvc -n <NAMESPACE>

Remove PVCs, if removal is permanent

Terminal input

kubectl delete pvc --all -n <NAMESPACE>

Data loss

Deleting PVCs permanently removes persisted data.
Do this only when disassembly is final and backups are complete.

Remove ArgoCD infrastructure

This step is separate from removing the TDP environment.

First remove all Applications and workloads managed by ArgoCD.
Only then remove ArgoCD itself, if that is part of the objective.

ArgoCD is removed via Helm regardless of how the TDP environment was managed.

Uninstall ArgoCD releases

Terminal input

helm uninstall tdp-argo -n <NAMESPACE>
helm uninstall tdp-crds -n <NAMESPACE>

Attention

helm uninstall tdp-crds does not remove CRDs automatically. CRDs are cluster-scope resources and can be shared by other installations or projects. tdp-crds manages multi-component CRDs (Argo CD, Kafka, NiFiKop, ClickHouse, OpenMetadata and cert-manager) — evaluate carefully before removing them.

Remove CRDs manually (when necessary)

Apply the cleanup job included in the chart:

Terminal input

kubectl apply -f stack/charts/tdp/tdp-crds/cleanup/delete-crds-job.yaml

The Job checks whether there are dependent resources before deleting each CRD — CRDs with resources in use are ignored.

Videos

argocd app delete

argocd app list

kubectl get all

Limpeza complementar

Check Persistent Orphan Volumes

After removing the PVCs, check if there are PVs in the Released state:

Terminal input

kubectl get pv | grep <NAMESPACE>

Figure 13 - Verification of Persistent Orphan Volumes

If there are orphaned PVs and you no longer need them:

Terminal input

kubectl delete pv <pv-name>

Figure 14 - Delete PVs

Remove namespace, when applicable

Only remove the namespace when it is dedicated exclusively to the environment being dismounted.

Terminal input

kubectl delete namespace <NAMESPACE>

Attention

Removing the namespace deletes all namespaced resources still in it, including PVCs, ConfigMaps, Secrets, and ServiceAccounts.

If the namespace gets stuck in Terminating, check finalizers:

Terminal input

kubectl get namespace <NAMESPACE> -o json | jq '.spec.finalizers'

To force termination:

Terminal input

kubectl get namespace <NAMESPACE> -o json | jq '.spec.finalizers = []' | \
kubectl replace --raw "/api/v1/namespaces/<NAMESPACE>/finalize" -f -

Final check

After removal, check whether there is anything left in the environment.

Check remaining resources in the namespace

Terminal input

kubectl get all -n <NAMESPACE>

Check remaining Helm releases

Terminal input

helm list -n <NAMESPACE>

Check remaining Applications in ArgoCD

Terminal input

argocd app list

Check remaining PVCs

Terminal input

kubectl get pvc -n <NAMESPACE>

Check remaining PVs

Terminal input

kubectl get pv | grep <NAMESPACE>

Resumo

Situation	Main path	Note
Component installed directly via Helm	helm uninstall <RELEASE_NAME> -n <NAMESPACE>	Use when there is actual Helm release
Component managed by ArgoCD	argocd app delete <component> --cascade	Removes the Application and managed resources
Complete removal from ArgoCD managed environment	argocd app delete ... --cascade in reverse order, or root app removal	PVCs and PVs may require manual cleaning
Removal of ArgoCD itself	helm uninstall tdp-argo -n <NAMESPACE>	It must occur after removing the Applications
Complete removal of the installed environment via Helm	sequence of helm uninstall in reverse order	Separate flow from ArgoCD model
Persistent data cleaning	kubectl delete pvc... and, if necessary, kubectl delete pv...	Manual and destructive step
Namespace cleanup	kubectl delete namespace <NAMESPACE>	Only when the namespace is dedicated to the removed environment