Rediger

Del via

Storage options for applications in Azure Kubernetes Service (AKS)

  • Artikkel

Applications running in Azure Kubernetes Service (AKS) may need to store and retrieve data. While some application workloads can use local, fast storage on unneeded, emptied nodes, others require storage that persists on more regular data volumes within the Azure platform.

Multiple pods may need to:

  • Share the same data volumes.
  • Reattach data volumes if the pod is rescheduled on a different node.

You also might need to collect and store sensitive data or application configuration information into pods.

This article introduces the core concepts that provide storage to your applications in AKS:

Diagram of storage options for applications in an Azure Kubernetes Services (AKS) cluster.

Default OS disk sizing

When you create a new cluster or add a new node pool to an existing cluster, the number for vCPUs by default determines the OS disk size. The number of vCPUs is based on the VM SKU. The following table lists the default OS disk size for each VM SKU:

VM SKU Cores (vCPUs) Default OS Disk Tier Provisioned IOPS Provisioned Throughput (Mbps)
1 - 7 P10/128G 500 100
8 - 15 P15/256G 1100 125
16 - 63 P20/512G 2300 150
64+ P30/1024G 5000 200

Important

Default OS disk sizing is only used on new clusters or node pools when Ephemeral OS disks aren't supported and a default OS disk size isn't specified. The default OS disk size might impact the performance or cost of your cluster. You can't change the OS disk size after cluster or node pool creation. This default disk sizing affects clusters or node pools created on July 2022 or later.

Ephemeral OS disk

By default, Azure automatically replicates the operating system disk for a virtual machine to Azure Storage to avoid data loss when the VM is relocated to another host. However, since containers aren't designed to have local state persisted, this behavior offers limited value while providing some drawbacks. These drawbacks include, but aren't limited to, slower node provisioning and higher read/write latency.

By contrast, ephemeral OS disks are stored only on the host machine, just like a temporary disk. With this configuration, you get lower read/write latency, together with faster node scaling and cluster upgrades.

Note

When you don't explicitly request Azure managed disks for the OS, AKS defaults to ephemeral OS if possible for a given node pool configuration.

Size requirements and recommendations for ephemeral OS disks are available in the Azure VM documentation. The following are some general sizing considerations:

  • If you chose to use the AKS default VM size Standard_DS2_v2 SKU with the default OS disk size of 100 GiB, the default VM size supports ephemeral OS, but only has 86 GiB of cache size. This configuration would default to managed disks if you don't explicitly specify it. If you do request an ephemeral OS, you receive a validation error.

  • If you request the same Standard_DS2_v2 SKU with a 60-GiB OS disk, this configuration would default to ephemeral OS. The requested size of 60 GiB is smaller than the maximum cache size of 86 GiB.

  • If you select the Standard_D8s_v3 SKU with 100-GB OS disk, this VM size supports ephemeral OS and has 200 GiB of cache space. If you don't specify the OS disk type, the node pool would receive ephemeral OS by default.

The latest generation of VM series doesn't have a dedicated cache, but only temporary storage. For example, if you selected the Standard_E2bds_v5 VM size with the default OS disk size of 100 GiB, it supports ephemeral OS disks, but only has 75 GB of temporary storage. This configuration would default to managed OS disks if you don't explicitly specify it. If you do request an ephemeral OS disk, you receive a validation error.

  • If you request the same Standard_E2bds_v5 VM size with a 60-GiB OS disk, this configuration defaults to ephemeral OS disks. The requested size of 60 GiB is smaller than the maximum temporary storage of 75 GiB.

  • If you select Standard_E4bds_v5 SKU with 100-GiB OS disk, this VM size supports ephemeral OS and has 150 GiB of temporary storage. If you don't specify the OS disk type, by default Azure provisions an ephemeral OS disk to the node pool.

Customer-managed keys

You can manage encryption for your ephemeral OS disk with your own keys on an AKS cluster. For more information, see Use Customer Managed key with Azure disk on AKS.

Volumes

Kubernetes typically treats individual pods as ephemeral, disposable resources. Applications have different approaches available to them for using and persisting data. A volume represents a way to store, retrieve, and persist data across pods and through the application lifecycle.

Traditional volumes are created as Kubernetes resources backed by Azure Storage. You can manually create data volumes to be assigned to pods directly or have Kubernetes automatically create them. Data volumes can use: Azure Disk, Azure Files, Azure NetApp Files, or Azure Blobs.

Note

Depending on the VM SKU you're using, the Azure Disk CSI driver might have a per-node volume limit. For some high performance VMs (for example, 16 cores), the limit is 64 volumes per node. To identify the limit per VM SKU, review the Max data disks column for each VM SKU offered. For a list of VM SKUs offered and their corresponding detailed capacity limits, see General purpose virtual machine sizes.

To help determine best fit for your workload between Azure Files and Azure NetApp Files, review the information provided in the article Azure Files and Azure NetApp Files comparison.

Azure Disk

Use Azure Disk to create a Kubernetes DataDisk resource. Disks types include:

  • Premium SSDs (recommended for most workloads)
  • Ultra disks
  • Standard SSDs
  • Standard HDDs

Tip

For most production and development workloads, use Premium SSDs.

Because an Azure Disk is mounted as ReadWriteOnce, it's only available to a single node. For storage volumes accessible by pods on multiple nodes simultaneously, use Azure Files.

Azure Files

Use Azure Files to mount a Server Message Block (SMB) version 3.1.1 share or Network File System (NFS) version 4.1 share. Azure Files let you share data across multiple nodes and pods and can use:

  • Azure Premium storage backed by high-performance SSDs
  • Azure Standard storage backed by regular HDDs

Azure NetApp Files

  • Ultra Storage
  • Premium Storage
  • Standard Storage

Azure Blob Storage

Use Azure Blob Storage to create a blob storage container and mount it using the NFS v3.0 protocol or BlobFuse.

  • Block blobs

Volume types

Kubernetes volumes represent more than just a traditional disk for storing and retrieving information. Kubernetes volumes can also be used as a way to inject data into a pod for use by its containers.

Common volume types in Kubernetes include:

emptyDir

Commonly used as temporary space for a pod. All containers within a pod can access the data on the volume. Data written to this volume type persists only for the lifespan of the pod. Once you delete the pod, the volume is deleted. This volume typically uses the underlying local node disk storage, though it can also exist only in the node's memory.

secret

You can use secret volumes to inject sensitive data into pods, such as passwords.

  1. Create a secret using the Kubernetes API.
  2. Define your pod or deployment and request a specific secret.
    • Secrets are only provided to nodes with a scheduled pod that requires them.
    • The secret is stored in tmpfs, not written to disk.
  3. When you delete the last pod on a node requiring a secret, the secret is deleted from the node's tmpfs.
    • Secrets are stored within a given namespace and are only accessed by pods within the same namespace.

configMap

You can use configMap to inject key-value pair properties into pods, such as application configuration information. Define application configuration information as a Kubernetes resource, easily updated and applied to new instances of pods as they're deployed.

Like using a secret:

  1. Create a ConfigMap using the Kubernetes API.
  2. Request the ConfigMap when you define a pod or deployment.
    • ConfigMaps are stored within a given namespace and are only accessed by pods within the same namespace.

Persistent volumes

Volumes defined and created as part of the pod lifecycle only exist until you delete the pod. Pods often expect their storage to remain if a pod is rescheduled on a different host during a maintenance event, especially in StatefulSets. A persistent volume (PV) is a storage resource created and managed by the Kubernetes API that can exist beyond the lifetime of an individual pod.

You can use the following Azure Storage data services to provide the persistent volume:

As noted in the Volumes section, the choice of Azure Disks or Azure Files is often determined by the need for concurrent access to the data or the performance tier.

Diagram of persistent volumes in an Azure Kubernetes Services (AKS) cluster.

A cluster administrator can statically create a persistent volume, or a volume can be created dynamically by the Kubernetes API server. If a pod is scheduled and requests storage that is currently unavailable, Kubernetes can create the underlying Azure Disk or File storage and attach it to the pod. Dynamic provisioning uses a storage class to identify what type of resource needs to be created.

Important

Persistent volumes can't be shared by Windows and Linux pods due to differences in file system support between the two operating systems.

Storage classes

To specify different tiers of storage, such as premium or standard, you can create a storage class.

A storage class also defines a reclaim policy. When you delete the persistent volume, the reclaim policy controls the behavior of the underlying Azure Storage resource. The underlying resource can either be deleted or kept for use with a future pod.

For clusters using the Container Storage Interface (CSI) drivers the following extra storage classes are created:

Storage class Description
managed-csi Uses Azure Standard SSD locally redundant storage (LRS) to create a managed disk. The reclaim policy ensures that the underlying Azure Disk is deleted when the persistent volume that used it is deleted. The storage class also configures the persistent volumes to be expandable. You can edit the persistent volume claim to specify the new size. Effective starting with Kubernetes version 1.29, in Azure Kubernetes Service (AKS) clusters deployed across multiple availability zones, this storage class utilizes Azure Standard SSD zone-redundant storage (ZRS) to create managed disks.
managed-csi-premium Uses Azure Premium locally redundant storage (LRS) to create a managed disk. The reclaim policy again ensures that the underlying Azure Disk is deleted when the persistent volume that used it is deleted. Similarly, this storage class allows for persistent volumes to be expanded. Effective starting with Kubernetes version 1.29, in Azure Kubernetes Service (AKS) clusters deployed across multiple availability zones, this storage class utilizes Azure Premium zone-redundant storage (ZRS) to create managed disks.
azurefile-csi Uses Azure Standard storage to create an Azure file share. The reclaim policy ensures that the underlying Azure file share is deleted when the persistent volume that used it is deleted.
azurefile-csi-premium Uses Azure Premium storage to create an Azure file share. The reclaim policy ensures that the underlying Azure file share is deleted when the persistent volume that used it is deleted.
azureblob-nfs-premium Uses Azure Premium storage to create an Azure Blob storage container and connect using the NFS v3 protocol. The reclaim policy ensures that the underlying Azure Blob storage container is deleted when the persistent volume that used it is deleted.
azureblob-fuse-premium Uses Azure Premium storage to create an Azure Blob storage container and connect using BlobFuse. The reclaim policy ensures that the underlying Azure Blob storage container is deleted when the persistent volume that used it is deleted.

Unless you specify a storage class for a persistent volume, the default storage class is used. Ensure volumes use the appropriate storage you need when requesting persistent volumes.

Important

Starting with Kubernetes version 1.21, AKS only uses CSI drivers by default and CSI migration is enabled. While existing in-tree persistent volumes continue to function, starting with version 1.26, AKS will no longer support volumes created using in-tree driver and storage provisioned for files and disk.

The default class will be the same as managed-csi.

Effective starting with Kubernetes version 1.29, when you deploy Azure Kubernetes Service (AKS) clusters across multiple availability zones, AKS now utilizes zone-redundant storage (ZRS) to create managed disks within built-in storage classes. ZRS ensures synchronous replication of your Azure managed disks across multiple Azure availability zones in your chosen region. This redundancy strategy enhances the resilience of your applications and safeguards your data against datacenter failures.

However, it's important to note that zone-redundant storage (ZRS) comes at a higher cost compared to locally redundant storage (LRS). If cost optimization is a priority, you can create a new storage class with the skuname parameter set to LRS. You can then use the new storage class in your Persistent Volume Claim (PVC).

You can create a storage class for other needs using kubectl. The following example uses premium managed disks and specifies that the underlying Azure Disk should be retained when you delete the pod:

apiVersion: storage.k8s.io/v1
kind: StorageClass
metadata:
  name: managed-premium-retain
provisioner: disk.csi.azure.com
parameters:
  skuName: Premium_ZRS
reclaimPolicy: Retain
volumeBindingMode: WaitForFirstConsumer
allowVolumeExpansion: true

Note

AKS reconciles the default storage classes and will overwrite any changes you make to those storage classes.

For more information about storage classes, see StorageClass in Kubernetes.

Persistent volume claims

A persistent volume claim (PVC) requests storage of a particular storage class, access mode, and size. The Kubernetes API server can dynamically provision the underlying Azure Storage resource if no existing resource can fulfill the claim based on the defined storage class.

The pod definition includes the volume mount once the volume has been connected to the pod.

Diagram of persistent volume claims in an Azure Kubernetes Services (AKS) cluster.

Once an available storage resource has been assigned to the pod requesting storage, the persistent volume is bound to a persistent volume claim. Persistent volumes are mapped to claims in a 1:1 mapping.

The following example YAML manifest shows a persistent volume claim that uses the managed-premium storage class and requests an Azure Disk that is 5Gi in size:

apiVersion: v1
kind: PersistentVolumeClaim
metadata:
  name: azure-managed-disk
spec:
  accessModes:
  - ReadWriteOnce
  storageClassName: managed-premium-retain
  resources:
    requests:
      storage: 5Gi

When you create a pod definition, you also specify:

  • The persistent volume claim to request the desired storage.
  • The volume mount for your applications to read and write data.

The following example YAML manifest shows how the previous persistent volume claim can be used to mount a volume at /mnt/azure:

kind: Pod
apiVersion: v1
metadata:
  name: nginx
spec:
  containers:
    - name: myfrontend
      image: mcr.microsoft.com/oss/nginx/nginx:1.15.5-alpine
      volumeMounts:
      - mountPath: "/mnt/azure"
        name: volume
  volumes:
    - name: volume
      persistentVolumeClaim:
        claimName: azure-managed-disk

For mounting a volume in a Windows container, specify the drive letter and path. For example:

...      
      volumeMounts:
      - mountPath: "d:"
        name: volume
      - mountPath: "c:\k"
        name: k-dir
...

Next steps

For associated best practices, see Best practices for storage and backups in AKS and AKS Storage Considerations.

To see how to use CSI drivers, see the following how-to articles:

For more information on core Kubernetes and AKS concepts, see the following articles: