Use Image Cleaner to clean up vulnerable stale images on your Azure Kubernetes Service (AKS) cluster

Article
2024-12-05

It's common to use pipelines to build and deploy images on Azure Kubernetes Service (AKS) clusters. While great for image creation, this process often doesn't account for the stale images left behind and can lead to image bloat on cluster nodes. These images might contain vulnerabilities, which might create security issues. To remove security risks in your clusters, you can clean these unreferenced images. Manually cleaning images can be time intensive. Image Cleaner performs automatic image identification and removal, which mitigates the risk of stale images and reduces the time required to clean them up.

Note

Image Cleaner is a feature based on Eraser. On an AKS cluster, the feature name and property name is Image Cleaner, while the relevant Image Cleaner pods' names contain Eraser.

Prerequisites

An Azure subscription. If you don't have an Azure subscription, you can create a free account.
Azure CLI version 2.49.0 or later. Run az --version to find your version. If you need to install or upgrade, see Install Azure CLI.

Limitations

Image Cleaner doesn't yet support Windows node pools or AKS virtual nodes.

How Image Cleaner works

After you enable Image Cleaner, there will be a controller manager pod named eraser-controller-manager deployed to your cluster.

Screenshot of a diagram showing ImageCleaner's workflow. The ImageCleaner pods running on the cluster can generate an ImageList, or manual input can be provided.

With Image Cleaner, you can choose between manual and automatic mode and the following configuration options:

Configuration options

Name	Description	Required
`--enable-image-cleaner`	Enable the Image Cleaner feature for an AKS cluster	Yes, unless disable is specified
`--disable-image-cleaner`	Disable the Image Cleaner feature for an AKS cluster	Yes, unless enable is specified
`--image-cleaner-interval-hours`	This parameter determines the interval time (in hours) Image Cleaner uses to run. The default value for Azure CLI is one week, the minimum value is 24 hours and the maximum is three months.	Not required for Azure CLI, required for ARM template or other clients

Automatic mode

Once eraser-controller-manager is deployed, the following steps will be taken automatically:

It immediately starts the cleanup process and creates eraser-aks-xxxxx worker pods for each node.
There are three containers in each worker pod:
- A collector, which collects unused images.
- A trivy-scanner, which leverages trivy to scan image vulnerabilities.
- A remover, which removes unused images with vulnerabilities.
After the cleanup process completes, the worker pod is deleted and the next scheduled cleanup happens according to the --image-cleaner-interval-hours you define.

Manual mode

You can manually trigger the cleanup by defining a CRD object,ImageList. This triggers the eraser-contoller-manager to create eraser-aks-xxxxx worker pods for each node and complete the manual removal process.

Note

After disabling Image Cleaner, the old configuration still exists. This means if you enable the feature again without explicitly passing configuration, the existing value is used instead of the default.

Enable Image Cleaner on your AKS cluster

Enable Image Cleaner on a new cluster

Enable Image Cleaner on a new AKS cluster using the az aks create command with the --enable-image-cleaner parameter.

az aks create \
    --resource-group myResourceGroup \
    --name myManagedCluster \
    --enable-image-cleaner \
    --generate-ssh-keys

Enable Image Cleaner on an existing cluster

Enable Image Cleaner on an existing AKS cluster using the az aks update command.

az aks update \
  --resource-group myResourceGroup \
  --name myManagedCluster \
  --enable-image-cleaner

Update the Image Cleaner interval on a new or existing cluster

Update the Image Cleaner interval on a new or existing AKS cluster using the --image-cleaner-interval-hours parameter.

# Create a new cluster with specifying the interval
az aks create \
    --resource-group myResourceGroup \
    --name myManagedCluster \
    --enable-image-cleaner \
    --image-cleaner-interval-hours 48 \
    --generate-ssh-keys

# Update the interval on an existing cluster
az aks update \
    --resource-group myResourceGroup \
    --name myManagedCluster \
    --enable-image-cleaner \
    --image-cleaner-interval-hours 48

Manually remove images using Image Cleaner

Important

The name must be set to imagelist.

Manually remove an image using the following kubectl apply command. This example removes the docker.io/library/alpine:3.7.3 image if it's unused.

cat <<EOF | kubectl apply -f -
apiVersion: eraser.sh/v1
kind: ImageList
metadata:
  name: imagelist
spec:
  images:
    - docker.io/library/alpine:3.7.3
EOF

The manual cleanup is a one-time operation and is only triggered when a new imagelist is created or changes are made to the existing imagelist. After the image is deleted, the imagelist won't be deleted automatically.

If you need to trigger another manual cleanup, you have to create a new imagelist or make changes to an existing one. If you want to remove the same image again, you need to create a new imagelist.

Delete an existing ImageList and create a new one

Delete the old imagelist using the kubectl delete command.
```
kubectl delete ImageList imagelist
```

Create a new imagelist with the same image name. The following example uses the same image as the previous example.

cat <<EOF | kubectl apply -f -
apiVersion: eraser.sh/v1
kind: ImageList
metadata:
  name: imagelist
spec:
  images:
    - docker.io/library/alpine:3.7.3
EOF

Modify an existing ImageList

Modify the existing imagelist using the kubectl edit command.

kubectl edit ImageList imagelist

# Add a new image to the list
apiVersion: eraser.sh/v1
kind: ImageList
metadata:
  name: imagelist
spec:
  images:
      docker.io/library/python:alpine3.18

When using manual mode, the eraser-aks-xxxxx pod deletes within 10 minutes after work completion.

Image exclusion list

Images specified in the exclusion list aren't removed from the cluster. Image Cleaner supports system and user-defined exclusion lists. It's not supported to edit the system exclusion list.

Check the system exclusion list

Check the system exclusion list using the following kubectl get command.

kubectl get -n kube-system configmap eraser-system-exclusion -o yaml

Create a user-defined exclusion list

Create a sample JSON file to contain excluded images.

cat > sample.json <<EOF
{"excluded": ["excluded-image-name"]}
EOF

Create a configmap using the sample JSON file using the following kubectl create and kubectl label command.

kubectl create configmap excluded --from-file=sample.json --namespace=kube-system
kubectl label configmap excluded eraser.sh/exclude.list=true -n kube-system

Disable Image Cleaner

Disable Image Cleaner on your cluster using the az aks update command with the --disable-image-cleaner parameter.
```
az aks update \
  --resource-group myResourceGroup \
  --name myManagedCluster \
  --disable-image-cleaner
```

FAQ

How can I check which version Image Cleaner is using?

kubectl describe configmap -n kube-system eraser-manager-config | grep tag -C 3

Does Image Cleaner support other vulnerability scanners besides trivy-scanner?

No.

Can I specify vulnerability levels for images to clean?

No. The default settings for vulnerability levels include:

LOW,
MEDIUM,
HIGH, and
CRITICAL

You can't customize the default settings.

How to review images were cleaned up by Image Cleaner?

Image logs are stored in the eraser-aks-xxxxx worker pod. When eraser-aks-xxxxx is alive, you can run the following commands to view deletion logs:

kubectl logs -n kube-system <worker-pod-name> -c collector
kubectl logs -n kube-system <worker-pod-name> -c trivy-scanner
kubectl logs -n kube-system <worker-pod-name> -c remover

The eraser-aks-xxxxx pod deletes within 10 minutes after work completion. You can follow these steps to enable the Azure Monitor add-on and use the Container Insights pod log table. After that, historical logs will be stored and you can review them even eraser-aks-xxxxx is deleted.

Ensure Azure Monitoring is enabled on your cluster. For detailed steps, see Enable Container Insights on AKS clusters.
Logs for the containers running in kube-system namespace are not collected by default. Remove the kube-system namespace from exclude_namespaces in the configmap and apply the config map to enable collection of these logs. See Configure Container insights data collection for details.

Get the Log Analytics resource ID using the az aks show command.

  az aks show --resource-group myResourceGroup --name myManagedCluster

After a few minutes, the command returns JSON-formatted information about the solution, including the workspace resource ID:

"addonProfiles": {
  "omsagent": {
    "config": {
      "logAnalyticsWorkspaceResourceID": "/subscriptions/<WorkspaceSubscription>/resourceGroups/<DefaultWorkspaceRG>/providers/Microsoft.OperationalInsights/workspaces/<defaultWorkspaceName>"
    },
    "enabled": true
  }
}

In the Azure portal, search for the workspace resource ID, then select Logs.

Copy one of the following queries and paste into the query window.

Use the following query if your cluster is using the ContainerLogV2 schema. If you're still using ContainerLog, you should upgrade to ContainerlogV2.

ContainerLogV2
| where PodName startswith "eraser-aks-" and PodNamespace == "kube-system"
| project TimeGenerated, PodName, LogMessage, LogSource

If you want continue to use ContainerLog, use the following query instead:

let startTimestamp = ago(1h);
KubePodInventory
| where TimeGenerated > startTimestamp
| project ContainerID, PodName=Name, Namespace
| where PodName startswith "eraser-aks-" and Namespace == "kube-system"
| distinct ContainerID, PodName
| join
(
   ContainerLog
   | where TimeGenerated > startTimestamp
)
on ContainerID
// at this point before the next pipe, columns from both tables are available to be "projected". Due to both
// tables having a "Name" column, we assign an alias as PodName to one column which we actually want
| project TimeGenerated, PodName, LogEntry, LogEntrySource
| summarize by TimeGenerated, LogEntry
| order by TimeGenerated desc

Select Run. Any deleted image logs appear in the Results area.

Share via