docs(helm): align teardown release and namespace with install (NVBug 6622509)

nemo_retriever/helm/README.md

@@ -81,6 +81,13 @@ nemo_retriever/helm/
 
 ## Quick start
 
+The examples in this README use the Helm release name `retriever`.
+`nemo-retriever` is the chart name. It is not the release name. When a
+command omits `--namespace`, Helm installs into the current kubectl
+namespace. The [Recommended minimal install](#recommended-minimal-install-26081)
+and [Full teardown](#full-teardown) set `REL` and `NS` explicitly so
+cleanup targets the same release.
+
 ### Persistent storage prerequisite { #persistent-storage-prerequisite }
 
 The default chart creates **seven** PersistentVolumeClaims: three
@@ -458,10 +465,16 @@ Complete the [persistent storage prerequisite](#persistent-storage-prerequisite)
 and the [GPU scheduling prerequisite](#gpu-scheduling-prerequisite)
 before you install.
 
-Deploy only the four core NIMs that the retriever service auto-wires (`page_elements`, `table_structure`, `ocr`, `vlm_embed`):
+Deploy only the four core NIMs that the retriever service auto-wires
+(`page_elements`, `table_structure`, `ocr`, `vlm_embed`). Set `REL` and
+`NS` to the values you will reuse for upgrade and teardown. Replace
+`NS` if you install into a namespace other than `default`.
 
 ```bash
-helm install retriever ./nemo_retriever/helm \
+REL=retriever
+NS=default
+
+helm install "${REL}" ./nemo_retriever/helm -n "${NS}" --create-namespace \
   --set ngcImagePullSecret.create=true \
   --set ngcImagePullSecret.password=$NGC_API_KEY \
   --set ngcApiSecret.create=true \
@@ -527,29 +540,118 @@ Helm on uninstall.
 | `NIMService` CR remains | **Not expected** on a normal uninstall. Usually an **orphan** from a failed install/upgrade (release never recorded the resource, or the chart renamed a NIM). |
 | Deployments / GPU pods still running | Often the operator workload for a **kept** `NIMCache`, or a stale `NIMService` that Helm did not own. Check `kubectl get nimservice,nimcache -n <ns>`. |
 | `nemotron-*-job-*` pods in `Error` | The NIM Operator's **model-download Job** for a `NIMCache` (not the retriever service). Failed cache pulls retry and leave Error pods until the Job or `NIMCache` is deleted. Common after a failed `helm install` when the release is rolled back but `keep` retains the cache CR. |
-| `helm uninstall` appears to do nothing | Release may be missing or failed (`helm list -n <ns> -a`). CRs created before a failed install can be left without a release to clean them up. |
+| `helm uninstall` appears to do nothing | Wrong release name or namespace. The chart name is `nemo-retriever`. The example release name is `retriever`. Confirm with `helm list --all-namespaces`. A missing or failed release (`helm list -n <ns> --all`) can also leave CRs without a release to clean them up. |
 
 To change a NIM image on a later install or upgrade, delete the kept
 `NIMCache` first. Refer to
 [Changing a NIM image repository or tag](#changing-nim-image-repository-or-tag).
 
-**Full teardown** (dev cluster — deletes caches and PVCs Helm kept):
+### Full teardown { #full-teardown }
 
-```bash
-NS=retriever
-REL=nemo-retriever
+On a development cluster, remove the Helm release, kept `NIMCache`
+objects, and model PVCs. Set `REL` and `NS` to the same values you used
+at install time. The
+[Recommended minimal install](#recommended-minimal-install-26081) uses
+`REL=retriever` and `NS=default`. If you omitted `--namespace` at
+install time, Helm used the current kubectl namespace. Replace `NS` if
+that namespace is not `default`.
 
-helm uninstall "${REL}" -n "${NS}" 2>/dev/null || true
+1. Set the release identity and confirm the target:
 
-# Orphans and kept NIMCaches (Helm keep does not block kubectl delete):
-kubectl delete nimservice,nimcache -n "${NS}" --all
-# Optional: drop model PVCs if you will re-pull from NGC
-kubectl delete pvc -n "${NS}" -l 'app.kubernetes.io/managed-by=nvidia-nim-operator' 2>/dev/null || true
-```
+   ```bash
+   REL=retriever
+   NS=default
+
+   helm list -n "${NS}"
+   kubectl get deployment,nimservice,nimcache -n "${NS}"
+   ```
+
+   If `helm list` shows no matching release, stop. Run
+   `helm list --all-namespaces` and set `REL` and `NS` to the release
+   you intend to remove.
+
+2. Uninstall the release. Do not suppress Helm errors. If Helm reports
+   that the release was not found, correct `REL` and `NS` before you
+   delete NIM resources:
+
+   ```bash
+   helm uninstall "${REL}" -n "${NS}"
+   helm list -n "${NS}" --all
+   ```
+
+3. Delete kept `NIMCache` objects and leftover `NIMService` CRs. Helm
+   leaves **NIMCache** objects when
+   `nimOperator.nimCache.keepOnUninstall` is `true` (the default). Each
+   NIM uses a resource name from the chart, not the Helm release name.
+   The following command deletes every default name this chart can
+   create, including optional NIMs. `--ignore-not-found` skips names
+   you did not install.
+
+   If you overrode `nimOperator.ocr.nimServiceName`,
+   `nimOperator.vlm_embed.nimServiceName`, or
+   `nimOperator.answer_llm.nimServiceName`, replace those three default
+   names with the values you set.
+
+   ```bash
+   kubectl delete nimservice,nimcache -n "${NS}" \
+     nemotron-page-elements-v3 \
+     nemotron-table-structure-v1 \
+     nemotron-ocr-v2 \
+     llama-nemotron-embed-vl-1b-v2 \
+     llama-nemotron-rerank-vl-1b-v2 \
+     nemotron-parse \
+     nemotron-3-nano-omni-30b-a3b-reasoning \
+     audio \
+     answer-llm \
+     --ignore-not-found
+   ```
+
+   List what remains. Delete leftover CRs by the names in that list
+   before you continue. Remaining objects can be renamed caches or
+   resources that another product owns.
+
+   ```bash
+   kubectl get nimservice,nimcache -n "${NS}"
+   ```
+
+   Do not run `kubectl delete nimservice,nimcache -n "${NS}" --all`
+   unless this namespace contains only this install. That command
+   deletes every `NIMService` and `NIMCache` in the namespace, including
+   resources that another release or product owns.
+
+4. Optional: drop model PVCs when you re-pull weights from NGC. Confirm
+   claim names first. Default claim names use a `-pvc` suffix. Include
+   optional NIM claims. If you overrode a `nimServiceName` value, use
+   `<that-name>-pvc` instead of the default claim.
+
+   ```bash
+   kubectl get pvc -n "${NS}" -l 'app.kubernetes.io/managed-by=nvidia-nim-operator'
+   kubectl delete pvc -n "${NS}" \
+     nemotron-page-elements-v3-pvc \
+     nemotron-table-structure-v1-pvc \
+     nemotron-ocr-v2-pvc \
+     llama-nemotron-embed-vl-1b-v2-pvc \
+     llama-nemotron-rerank-vl-1b-v2-pvc \
+     nemotron-parse-pvc \
+     nemotron-3-nano-omni-30b-a3b-reasoning-pvc \
+     audio-pvc \
+     answer-llm-pvc \
+     --ignore-not-found
+   ```
+
+   List operator PVCs again and delete any leftover claim that belongs
+   to this install:
+
+   ```bash
+   kubectl get pvc -n "${NS}" -l 'app.kubernetes.io/managed-by=nvidia-nim-operator'
+   ```
 
 **Dev installs** that should not retain caches on uninstall:
 
 ```bash
+REL=retriever
+NS=default
+
 helm upgrade --install "${REL}" ./nemo_retriever/helm -n "${NS}" \
   --set nimOperator.nimCache.keepOnUninstall=false
 ```
@@ -1169,10 +1271,10 @@ The affected NIM is unavailable while the operator re-caches weights.
 
 1. Drain ingest traffic that depends on that NIM.
 2. Confirm the live `modelPuller` value differs from the new
-   `repository:tag`:
+   `repository:tag`. Set `NS` to the namespace of your Helm release:
 
    ```bash
-   NS=retriever
+   NS=default
    CACHE=nemotron-page-elements-v3
 
    kubectl get nimcache "${CACHE}" -n "${NS}" \