> ## Documentation Index
> Fetch the complete documentation index at: https://docs.firebolt.io/llms.txt
> Use this file to discover all available pages before exploring further.

> Capabilities that warrant graduating from the firebolt-instance Helm chart to the Firebolt Kubernetes Operator, and the migration shape.

# Operator upgrade path

This page explains when to graduate to the [Firebolt Kubernetes Operator](https://github.com/firebolt-db/firebolt-kubernetes-operator) and what the migration looks like.

## When to use the operator

Reach for the operator when you need any of the following:

* **Zero-downtime engine rollouts** through blue-green generations.
* **Engine autoscaling** with scale-to-zero, schedule windows, and wake-on-traffic.
* **Continuous reconciliation**. Drift is corrected automatically, and controller restarts resume mid-flight transitions.
* **`EngineClass`**. A namespace-scoped pod template shared across engines for image, service account, sidecars, and scheduling.
* **Multi-replica engines under one StatefulSet**, with operator-managed blue-green transitions.
* **Status surfaces**. `kubectl get fire` and `kubectl get fireng` show phases, conditions, and events.

If none of those apply, the chart on its own is sufficient.

## Migration shape

The migration is staged so you can validate each step before the next.

1. **Install the operator and its CRDs** alongside the chart, in a separate namespace. See the operator's [installation guide](https://github.com/firebolt-db/firebolt-kubernetes-operator/blob/main/docs/installation.mdx).
2. **Create a `FireboltInstance` and one or more `FireboltEngine` resources** that mirror the engines in your `values.yaml`. The operator's [quickstart](https://github.com/firebolt-db/firebolt-kubernetes-operator/blob/main/docs/quickstart.mdx) shows a minimal example.
3. **Cut over traffic** by switching the gateway routing to the operator-managed engines, then uninstall the chart-managed engines once queries are flowing through the operator.
4. **Decommission the chart release** with `helm uninstall`.

Engine PVCs are retained by default (Kubernetes default `Retain` policy), so a `helm uninstall` does not lose engine state. Move or reattach the PVCs to the operator-managed StatefulSets as part of the cutover.