address review

Signed-off-by: William Yao <william2000yao@gmail.com>

Author: willie-yao

README.md

@@ -108,86 +108,56 @@ And show the initial state of available GPU devices on the worker node:
 kubectl get resourceslice -o yaml
 ```
 
-You should see 2 GPUs (gpu-0, gpu-1) on the worker node, each with model
+You should see 8 GPUs (gpu-0 through gpu-7) on the worker node, each with model
 `LATEST-GPU-MODEL` and 80Gi of memory.
 
-Next, deploy some example apps to see DRA in action. The default configuration
-provides 2 GPUs per node, which is enough to run each example individually.
-Each example file has detailed comments at the top explaining what it
-demonstrates and how to verify the results.
-
-**Example 1: Exclusive GPU access**
-
-Two pods each requesting their own distinct GPU:
+Next, deploy some example apps to see DRA in action. Each example file has
+detailed comments at the top explaining what it demonstrates and how to verify
+the results:
 ```bash
-kubectl apply -f demo/two-pods-one-gpu-each.yaml
-kubectl wait --for=condition=Ready pod/pod0 pod/pod1 -n two-pods-one-gpu-each --timeout=60s
-```
-
-Check that each pod got a different GPU:
-```bash
-for pod in pod0 pod1; do
-  echo "${pod}:"
-  kubectl logs -n two-pods-one-gpu-each ${pod} -c ctr0 | grep -E "GPU_DEVICE_[0-9]+=" | grep -v "RESOURCE_CLAIM"
-done
+kubectl apply -f demo/basic-two-pods-one-gpu-each.yaml
+kubectl apply -f demo/basic-shared-gpu-across-containers.yaml
+kubectl apply -f demo/basic-gpu-sharing-strategies.yaml
 ```
 
-Clean up before the next example:
+Wait for all pods to be ready:
 ```bash
-kubectl delete -f demo/two-pods-one-gpu-each.yaml
+kubectl wait --for=condition=Ready pod/pod0 pod/pod1 -n basic-two-pods-one-gpu-each --timeout=60s
+kubectl wait --for=condition=Ready pod/pod0 -n basic-shared-gpu-across-containers --timeout=60s
+kubectl wait --for=condition=Ready pod/pod0 -n basic-gpu-sharing-strategies --timeout=60s
 ```
 
-**Example 2: Shared GPU across containers**
-
-Two containers in one pod sharing a single GPU:
+Then check the pod logs to see which GPUs were allocated:
 ```bash
-kubectl apply -f demo/shared-gpu-across-containers.yaml
-kubectl wait --for=condition=Ready pod/pod0 -n shared-gpu-across-containers --timeout=60s
-```
+# basic-two-pods-one-gpu-each: each pod should have 1 GPU with a distinct ID
+echo "basic-two-pods-one-gpu-each:"
+for pod in pod0 pod1; do
+  echo "  ${pod}:"
+  kubectl logs -n basic-two-pods-one-gpu-each ${pod} -c ctr0 | grep -E "GPU_DEVICE_[0-9]+=" | grep -v "RESOURCE_CLAIM"
+done
 
-Check that both containers see the same GPU with TimeSlicing:
-```bash
+# basic-shared-gpu-across-containers: both containers should show the same GPU ID
+echo "basic-shared-gpu-across-containers:"
 for ctr in ctr0 ctr1; do
-  echo "pod0 ${ctr}:"
-  kubectl logs -n shared-gpu-across-containers pod0 -c ${ctr} | grep -E "GPU_DEVICE_[0-9]+" | grep -v "RESOURCE_CLAIM"
+  echo "  pod0 ${ctr}:"
+  kubectl logs -n basic-shared-gpu-across-containers pod0 -c ${ctr} | grep -E "GPU_DEVICE_[0-9]+" | grep -v "RESOURCE_CLAIM"
 done
-```
 
-Clean up before the next example:
-```bash
-kubectl delete -f demo/shared-gpu-across-containers.yaml
-```
-
-**Example 3: GPU sharing strategies**
-
-Two GPUs configured with different sharing modes (TimeSlicing and SpacePartitioning):
-```bash
-kubectl apply -f demo/gpu-sharing-strategies.yaml
-kubectl wait --for=condition=Ready pod/pod0 -n gpu-sharing-strategies --timeout=60s
-```
-
-Check that ts-ctr0/ts-ctr1 share one GPU with TimeSlicing and sp-ctr0/sp-ctr1
-share another with SpacePartitioning:
-```bash
+# basic-gpu-sharing-strategies: ts-ctr0/ts-ctr1 share one GPU (TimeSlicing),
+# sp-ctr0/sp-ctr1 share another (SpacePartitioning)
+echo "basic-gpu-sharing-strategies:"
 for ctr in ts-ctr0 ts-ctr1 sp-ctr0 sp-ctr1; do
-  echo "pod0 ${ctr}:"
-  kubectl logs -n gpu-sharing-strategies pod0 -c ${ctr} | grep -E "GPU_DEVICE_[0-9]+" | grep -v "RESOURCE_CLAIM"
+  echo "  pod0 ${ctr}:"
+  kubectl logs -n basic-gpu-sharing-strategies pod0 -c ${ctr} | grep -E "GPU_DEVICE_[0-9]+" | grep -v "RESOURCE_CLAIM"
 done
 ```
 
-Clean up:
-```bash
-kubectl delete -f demo/gpu-sharing-strategies.yaml
-```
-
 In this example resource driver, no "actual" GPUs are made available to any
 containers. Instead, a set of environment variables are set in each container
 to indicate which GPUs *would* have been injected into them by a real resource
 driver and how they *would* have been configured.
 
 For the full list of all 8 available examples, see [`demo/README.md`](demo/README.md).
-To run multiple examples at the same time, increase `kubeletPlugin.numDevices`
-when installing the Helm chart.
 
 ### Demo DRA Admin Access Feature
 This example driver includes support for the [DRA AdminAccess feature](https://kubernetes.io/docs/concepts/scheduling-eviction/dynamic-resource-allocation/#admin-access), which allows administrators to gain privileged access to devices already in use by other users. This example demonstrates the end-to-end flow by setting the `DRA_ADMIN_ACCESS` environment variable. A driver managing real devices could use this to expose host hardware information.
@@ -205,7 +175,14 @@ To run this demo:
 
 ### Clean Up
 
-Once you are done, delete the `kind` cluster:
+Once you have verified everything is running correctly, delete the example apps:
+```bash
+kubectl delete -f demo/basic-two-pods-one-gpu-each.yaml
+kubectl delete -f demo/basic-shared-gpu-across-containers.yaml
+kubectl delete -f demo/basic-gpu-sharing-strategies.yaml
+```
+
+Finally, delete the `kind` cluster:
 ```bash
 ./demo/delete-cluster.sh
 ```

demo/README.md

@@ -1,31 +1,14 @@
 # Demo Examples
 
 This directory contains example workloads that demonstrate different ways to
-request and configure GPU devices using Dynamic Resource Allocation (DRA).
+request and configure devices using Dynamic Resource Allocation (DRA).
 
-## Quick Start
+Examples prefixed with `basic-` are featured in the
+[main README walkthrough](../README.md) and are a good starting point for
+learning about DRA.
 
-The following three examples are featured in the [main README walkthrough](../README.md)
-and are designed to run together with the default cluster configuration (2 GPUs):
-
-| Example | Description | GPUs |
-|---|---|---|
-| [two-pods-one-gpu-each.yaml](two-pods-one-gpu-each.yaml) | Two pods each get their own exclusive GPU | 2 |
-| [shared-gpu-across-containers.yaml](shared-gpu-across-containers.yaml) | Two containers in one pod share a single GPU | 1 |
-| [gpu-sharing-strategies.yaml](gpu-sharing-strategies.yaml) | TimeSlicing and SpacePartitioning on two GPUs | 2 |
-
-## All Examples
-
-| Example | Description | GPUs | Key Concept |
-|---|---|---|---|
-| [two-pods-one-gpu-each.yaml](two-pods-one-gpu-each.yaml) | Two pods, each requesting one exclusive GPU | 2 | ResourceClaimTemplate basics |
-| [one-pod-two-gpus.yaml](one-pod-two-gpus.yaml) | One container requesting two distinct GPUs | 2 | Multiple requests in a claim |
-| [shared-gpu-across-containers.yaml](shared-gpu-across-containers.yaml) | Two containers sharing one GPU within a pod | 1 | Intra-pod GPU sharing |
-| [shared-global-claim.yaml](shared-global-claim.yaml) | Two pods sharing a GPU via a pre-created ResourceClaim | 1 | ResourceClaim vs ResourceClaimTemplate |
-| [gpu-sharing-strategies.yaml](gpu-sharing-strategies.yaml) | TimeSlicing and SpacePartitioning configuration | 2 | Opaque driver config (GpuConfig) |
-| [initcontainer-shared-gpu.yaml](initcontainer-shared-gpu.yaml) | initContainer and container sharing a GPU | 1 | initContainer support |
-| [admin-access.yaml](admin-access.yaml) | Admin access to all GPUs with elevated privileges | All | DRA AdminAccess feature |
-| [cel-selector.yaml](cel-selector.yaml) | Selecting a GPU by model and memory using CEL | 1 | CEL expression selectors |
+Each example file has detailed comments at the top explaining what it
+demonstrates, what output to expect, and the driver and cluster requirements.
 
 ## Running Examples
 
@@ -43,9 +26,7 @@ kubectl delete -f demo/<example-name>.yaml
 
 ## Notes
 
-- The default Helm chart configures **2 GPUs** per node, which is enough to run
-  any single example (except `admin-access.yaml` which uses all available GPUs).
-- To run multiple examples simultaneously, increase `kubeletPlugin.numDevices`
-  in the Helm values.
+- The default Helm chart configures **8 GPUs** per node, which is enough to run
+  several examples simultaneously.
 - Each example creates its own namespace, so examples don't interfere with
   each other's resource names.

demo/admin-access.yaml

@@ -9,19 +9,20 @@
 #   - The namespace must have the label:
 #       resource.kubernetes.io/admin-access: "true"
 #   - The request must set adminAccess: true
-#   - allocationMode: All is used here to access all available GPUs
+#   - allocationMode: All is used here to access all available GPUs on a Node.
+#     Admins typically require access to all devices on a node to perform
+#     maintenance or monitoring.
 #
 # Expected: The container has DRA_ADMIN_ACCESS=true and GPU_DEVICE env vars
 # for all available GPUs. Check with:
 #   kubectl logs -n admin-access pod0 -c ctr0 | grep DRA_ADMIN_ACCESS
 #   kubectl logs -n admin-access pod0 -c ctr0 | grep GPU_DEVICE
 #
-# Resources created:
-#   - 1 Namespace (with admin-access label)
-#   - 1 ResourceClaimTemplate (multiple-gpus-admin)
-#   - 1 Pod (pod0) with 1 container
+# Cluster requirements:
+#   Kubernetes 1.34+
+#   Feature gate: DRAAdminAccess
 #
-# GPUs required: all available (uses allocationMode: All)
+# GPUs required: all available on a Node (uses allocationMode: All)
 
 ---
 apiVersion: v1

demo/gpu-sharing-strategies.yaml demo/basic-gpu-sharing-strategies.yamldemo/gpu-sharing-strategies.yaml renamed to demo/basic-gpu-sharing-strategies.yaml

@@ -11,26 +11,25 @@
 # Expected: ts-ctr0 and ts-ctr1 share one GPU with SHARING_STRATEGY=TimeSlicing
 # and TIMESLICE_INTERVAL=Long. sp-ctr0 and sp-ctr1 share a different GPU with
 # SHARING_STRATEGY=SpacePartitioning and PARTITION_COUNT=10. Check with:
-#   kubectl logs -n gpu-sharing-strategies pod0 -c ts-ctr0 | grep GPU_DEVICE
-#   kubectl logs -n gpu-sharing-strategies pod0 -c sp-ctr0 | grep GPU_DEVICE
+#   kubectl logs -n basic-gpu-sharing-strategies pod0 -c ts-ctr0 | grep GPU_DEVICE
+#   kubectl logs -n basic-gpu-sharing-strategies pod0 -c sp-ctr0 | grep GPU_DEVICE
 #
-# Resources created:
-#   - 1 ResourceClaimTemplate (multiple-gpus) with 2 requests + config
-#   - 1 Pod (pod0) with 4 containers
+# Cluster requirements:
+#   Kubernetes 1.34+
 #
 # GPUs required: 2
 
 ---
 apiVersion: v1
 kind: Namespace
 metadata:
-  name: gpu-sharing-strategies
+  name: basic-gpu-sharing-strategies
 
 ---
 apiVersion: resource.k8s.io/v1
 kind: ResourceClaimTemplate
 metadata:
-  namespace: gpu-sharing-strategies
+  namespace: basic-gpu-sharing-strategies
   name: multiple-gpus
 spec:
   spec:
@@ -68,7 +67,7 @@ spec:
 apiVersion: v1
 kind: Pod
 metadata:
-  namespace: gpu-sharing-strategies
+  namespace: basic-gpu-sharing-strategies
   name: pod0
 spec:
   containers:

demo/shared-gpu-across-containers.yaml …/basic-shared-gpu-across-containers.yamldemo/shared-gpu-across-containers.yaml renamed to demo/basic-shared-gpu-across-containers.yaml demo/shared-gpu-across-containers.yaml renamed to demo/basic-shared-gpu-across-containers.yaml

@@ -3,28 +3,27 @@
 # One pod, two containers.
 # Each asking for shared access to a single GPU.
 #
-# Expected: Both containers see the same GPU with TimeSlicing. Check with:
-#   kubectl logs -n shared-gpu-across-containers pod0 -c ctr0 | grep GPU_DEVICE
-#   kubectl logs -n shared-gpu-across-containers pod0 -c ctr1 | grep GPU_DEVICE
-# Both containers should show the same GPU ID with SHARING_STRATEGY=TimeSlicing.
+# Expected: Both containers see the same GPU. Check with:
+#   kubectl logs -n basic-shared-gpu-across-containers pod0 -c ctr0 | grep GPU_DEVICE
+#   kubectl logs -n basic-shared-gpu-across-containers pod0 -c ctr1 | grep GPU_DEVICE
+# Both containers should show the same GPU ID.
 #
-# Resources created:
-#   - 1 ResourceClaimTemplate (single-gpu)
-#   - 1 Pod (pod0) with 2 containers (ctr0, ctr1)
+# Cluster requirements:
+#   Kubernetes 1.34+
 #
 # GPUs required: 1
 
 ---
 apiVersion: v1
 kind: Namespace
 metadata:
-  name: shared-gpu-across-containers
+  name: basic-shared-gpu-across-containers
 
 ---
 apiVersion: resource.k8s.io/v1
 kind: ResourceClaimTemplate
 metadata:
-  namespace: shared-gpu-across-containers
+  namespace: basic-shared-gpu-across-containers
   name: single-gpu
 spec:
   spec:
@@ -38,7 +37,7 @@ spec:
 apiVersion: v1
 kind: Pod
 metadata:
-  namespace: shared-gpu-across-containers
+  namespace: basic-shared-gpu-across-containers
   name: pod0
 spec:
   containers:

demo/two-pods-one-gpu-each.yaml demo/basic-two-pods-one-gpu-each.yamldemo/two-pods-one-gpu-each.yaml renamed to demo/basic-two-pods-one-gpu-each.yaml

@@ -4,27 +4,26 @@
 # Each container asking for 1 distinct GPU.
 #
 # Expected: Each pod gets a different GPU. Check with:
-#   kubectl logs -n two-pods-one-gpu-each pod0 -c ctr0 | grep GPU_DEVICE
-#   kubectl logs -n two-pods-one-gpu-each pod1 -c ctr0 | grep GPU_DEVICE
+#   kubectl logs -n basic-two-pods-one-gpu-each pod0 -c ctr0 | grep GPU_DEVICE
+#   kubectl logs -n basic-two-pods-one-gpu-each pod1 -c ctr0 | grep GPU_DEVICE
 # Each container should have 1 GPU_DEVICE env var with a distinct GPU ID.
 #
-# Resources created:
-#   - 1 ResourceClaimTemplate (single-gpu)
-#   - 2 Pods (pod0, pod1), each with 1 container
+# Cluster requirements:
+#   Kubernetes 1.34+
 #
 # GPUs required: 2
 
 ---
 apiVersion: v1
 kind: Namespace
 metadata:
-  name: two-pods-one-gpu-each
+  name: basic-two-pods-one-gpu-each
 
 ---
 apiVersion: resource.k8s.io/v1
 kind: ResourceClaimTemplate
 metadata:
-  namespace: two-pods-one-gpu-each
+  namespace: basic-two-pods-one-gpu-each
   name: single-gpu
 spec:
   spec:
@@ -38,7 +37,7 @@ spec:
 apiVersion: v1
 kind: Pod
 metadata:
-  namespace: two-pods-one-gpu-each
+  namespace: basic-two-pods-one-gpu-each
   name: pod0
   labels:
     app: pod
@@ -59,7 +58,7 @@ spec:
 apiVersion: v1
 kind: Pod
 metadata:
-  namespace: two-pods-one-gpu-each
+  namespace: basic-two-pods-one-gpu-each
   name: pod1
   labels:
     app: pod

demo/cel-selector.yaml

@@ -8,9 +8,8 @@
 #   kubectl logs -n cel-selector pod0 -c ctr0 | grep GPU_DEVICE
 # The container should have 1 GPU_DEVICE env var.
 #
-# Resources created:
-#   - 1 ResourceClaimTemplate (single-gpu-cel) with CEL selectors
-#   - 1 Pod (pod0) with 1 container
+# Cluster requirements:
+#   Kubernetes 1.34+
 #
 # GPUs required: 1
 

demo/initcontainer-shared-gpu.yaml

@@ -8,9 +8,8 @@
 #   kubectl logs -n initcontainer-shared-gpu pod0 -c ctr0 | grep GPU_DEVICE
 # Both should show the same GPU ID.
 #
-# Resources created:
-#   - 1 ResourceClaimTemplate (single-gpu)
-#   - 1 Pod (pod0) with 1 initContainer + 1 container
+# Cluster requirements:
+#   Kubernetes 1.34+
 #
 # GPUs required: 1
 

demo/one-pod-two-gpus.yaml

@@ -7,9 +7,8 @@
 #   kubectl logs -n one-pod-two-gpus pod0 -c ctr0 | grep GPU_DEVICE
 # The container should have 2 GPU_DEVICE env vars with distinct GPU IDs.
 #
-# Resources created:
-#   - 1 ResourceClaimTemplate (multiple-gpus) with 2 requests
-#   - 1 Pod (pod0) with 1 container
+# Cluster requirements:
+#   Kubernetes 1.34+
 #
 # GPUs required: 2