> For clean Markdown of any page, append .md to the page URL.
> For a complete documentation index, see https://docs.fabricgateway.ai/llms.txt.
> For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.fabricgateway.ai/_mcp/server.

# Uninstall the data plane

> Remove the data plane Helm release, and clean up the Secrets, volumes, database, and console nodes that the uninstall leaves behind.

This page applies to the Enterprise plan, where you run the data plane in your own VPC. `helm uninstall` removes the data plane's workloads, but it deliberately leaves behind your data, the Secrets you created, and a few other resources. This page shows how to uninstall, what's left afterward, and how to clean it up if you're decommissioning the data plane for good.

## Prerequisites

Before you begin, make sure you have the following:

* Access to the cluster and the release's namespace.
* If you might reinstall later, a backup of the secret store and of `FABRIC_GW_SECRET_KEY`. See [Maintain the data plane and its data](/run/maintenance#back-up-the-secret-store).
* A plan for client traffic. Once the gateway pods stop, requests to the data plane fail.

## Uninstall the release

1. Remove the release:

   ```bash
   helm uninstall fabric-dp -n pm-fabric-dp
   ```

2. Check that the gateway and embedder pods are gone:

   ```bash
   kubectl -n pm-fabric-dp get pods
   ```

This removes everything the release manages: the gateway and embedder workloads and Services, the environment and collector ConfigMaps, both PodDisruptionBudgets, the service account, a CA ConfigMap rendered from `syncCA.pem`, the headless Service and OPA ConfigMap when those features are on, and any autoscaling, monitoring, or NetworkPolicy objects you turned on.

> **Warning**
>
> A model-cache PVC that the chart created with `embedder.persistence.enabled` is part of the release, so `helm uninstall` deletes it and the cached model with it. A reinstall downloads the model again. The PVC is named `fabric-gateway-embedder-model-cache`. To keep the cache, use a PVC you own with `embedder.persistence.existingClaim`.

## What's left behind

| Left behind                                                                                  | Why                                                                      | To remove it                                                                                       |
| -------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------ | -------------------------------------------------------------------------------------------------- |
| Your MySQL database                                                                          | It's outside the release                                                 | Drop the database in MySQL, if you're decommissioning                                              |
| Your telemetry data                                                                          | It's in your backend                                                     | Use your backend's retention or deletion tools                                                     |
| The app Secret and the `fabric-dp-sync-ca` Secret                                            | You created them outside the release                                     | `kubectl -n pm-fabric-dp delete secret <name>`                                                     |
| The migration ConfigMap, `fabric-gateway-chart-dp-sql`                                       | It's a Helm hook, so the release doesn't track it                        | `kubectl -n pm-fabric-dp delete configmap fabric-gateway-chart-dp-sql`                             |
| The development Secret, if you used `secrets.dev`                                            | It's a Helm hook                                                         | `kubectl -n pm-fabric-dp delete secret fabric-gateway-chart-ext-secret`                            |
| With the distributed store, each pod's volume, named `fabric-cache-fabric-gateway-chart-<n>` | Kubernetes keeps a StatefulSet's volumes when the StatefulSet is deleted | List them with `kubectl -n pm-fabric-dp get pvc`, then `kubectl -n pm-fabric-dp delete pvc <name>` |
| An embedder PVC you supplied with `existingClaim`                                            | You own it                                                               | Delete it when you no longer need the cache                                                        |
| The nodes in **Admin** > **Gateway health**                                                  | The control plane keeps their records                                    | Retire the nodes in the console, as described below                                                |

The migration Job is already gone if it succeeded, because Helm deletes it on success. A failed Job isn't deleted, so remove it with `kubectl -n pm-fabric-dp delete job fabric-gateway-chart-dp-migrate`.

With the distributed store, the pod volumes hold the encrypted store snapshots and each pod's certificate. Keep them if you plan to reinstall and want the store back. Delete them if you're decommissioning.

## Retire the nodes in the console

After an uninstall, the old nodes show as **Disconnected** in **Admin** > **Gateway health**, and their certificates stay valid until they expire. If you're decommissioning, open each node and select **Revoke certificate**, so a copy of its certificate can't reconnect. See [Connect a data plane to the control plane](/run/connect-a-data-plane).

If you enrolled with a bootstrap token, delete the token from your secret manager as well.

## Reinstall

The release holds no state, so you can reinstall against the same MySQL database. The migration Job adopts the existing schema and applies only what's missing. Use the same `FABRIC_GW_SECRET_KEY`, or the data plane can't read the credentials already stored. New pods enroll again as they start. For the steps, see [Install the data plane](/run/install).

## Next steps

* To reinstall, see [Install the data plane](/run/install).
* To understand what data the data plane keeps, see [Maintain the data plane and its data](/run/maintenance).