Cloud storage selection and integration guide - Container Compute Service

Storage basics

Volumes

Container file systems are ephemeral. Two problems arise from this:

When a pod runs multiple containers, the containers cannot share files.
When a container crashes, files written during runtime are lost.

Kubernetes solves both problems through the volume abstraction. A volume is part of a pod — it cannot be created independently. All containers in a pod share access to the volume, but each must mount it to a specific directory. A volume is also a way to share data between containers in a pod, between pods, and between a pod and the external environment. Volume types include EmptyDir, ConfigMap, and PersistentVolumeClaim (PVC).

For more information, see Volumes.

PersistentVolumes and PersistentVolumeClaims

Not all volumes survive pod restarts. Persistent storage requires a remote storage service. Kubernetes represents this through two resource objects:

PersistentVolume (PV): An abstract storage volume backed by network or cloud storage. A PV exists independently of any pod and can be shared by multiple pods. A cluster administrator can create a PV manually, or a StorageClass (SC) can create one dynamically.
PersistentVolumeClaim (PVC): A pod's request for storage. A PVC specifies the capacity and access mode required and binds one-to-one with a PV. To use storage in a pod, declare a PVC as the pod's volume.

For more information, see Persistent Volumes and Storage Classes.

Choose a storage service

You can select a suitable cloud storage service based on the computing power type and storage requirements of your business workload. Consider factors such as data volume, data access frequency, IOPS, and throughput.

Use the table below to compare features and typical use cases across all supported storage types.

Storage support by pod type

Pod type	Supported storage
CPU-based (general-purpose, compute-optimized)	All storage types
GPU-based (GPU, GPU-HPN)	NAS, OSS
GPU-based — selected models only	Block storage, CPFS for Lingjun (submit a ticket to confirm availability)

Block storage and CPFS for Lingjun support on GPU-based pods depends on the specific GPU model. Submit a ticket to confirm whether your GPU model is supported before using these storage types.

Features and use cases

Storage type	Key characteristics	Use cases	Options
Block storage	Non-shared, low-latency, and highly reliable block-level storage. Functionally similar to a local disk: supports partitioning, formatting, and file system creation. Can only be mounted to one pod at a time.	High I/O, low-latency workloads: databases, middleware Single-pod storage: workloads that do not need to share data across pods	Disk categories: `cloud_essd`: Enterprise SSD (ESSD) — default `cloud_essd_entry`: ESSD Entry disk `cloud_auto`: ESSD AutoPL disk `cloud_ssd`: standard SSD `cloud_efficiency`: ultra disk For pricing and performance comparisons, see Prices of block storage devices and Block storage performance.
File Storage NAS	Distributed file system with shared access, scalability, high reliability, and high performance. Supports high throughput and IOPS, random read/write, and online file modification. Data persists after pods are deleted.	Data sharing: multiple pods reading and writing the same dataset Batch analytics: workloads that need high shared-storage throughput Web applications and CMS: shared content storage for stateless pods Persistent logs: centralized log storage that survives pod restarts	General-purpose NAS and Extreme NAS file systems are supported. Select the type and specification based on your cost and performance requirements. For more information, see General-purpose NAS, Extreme NAS, and Selection guide.
CPFS for Lingjun (invitational preview)	Designed for intelligent computing workloads. Provides ultra-high throughput and IOPS with end-to-end RDMA network support. Available in specific regions and zones only.	Intelligent computing: AIGC model training, autonomous driving data pipelines, and similar high-performance workloads	CPFS for Lingjun is in invitational preview and is available in specific regions and zones only. For more information, see CPFS for Lingjun.
Object Storage Service (OSS)	Cost-effective, massively scalable shared object storage. Best suited for data that is written once and read frequently. Data persists after pods are deleted.	Data sharing: multiple pods accessing the same dataset (configuration files, images, media) Read-intensive workloads: serving static assets such as images and video files. Use an ossfs 1.0 volume if your workload modifies file content and requires POSIX compatibility. High-concurrency batch processing: AI training, data analytics, autonomous driving pipelines. These workloads primarily perform sequential and random reads, and sequential (append-only) writes. Use an ossfs 2.0 volume for higher throughput. Security-sensitive workloads: OSS supports zone-redundant storage, server-side encryption (KMS), MD5 data integrity validation, and object-level authentication via RAM and OSS bucket policies.	Select a storage class based on access frequency and cost. For more information, see Selection guide and Storage classes.

Mount a storage service

Storage plug-in

ACS uses csi-provisioner as its CSI plug-in. It handles volume mounting, unmounting, creation, and deletion for all supported Alibaba Cloud storage types.

Important

The CSI plug-in requires permissions to access other Alibaba Cloud services. It uses a RAM role to request the creation of the AliyunCCCSIPluginRole role to access your resources in other cloud products. For details, see Authorization overview.

Mount methods

To mount cloud storage in a pod, describe a storage resource as a PV, bind the PV to a PVC, and declare the PVC in the pod's volume spec. ACS supports two provisioning methods:

Static provisioning: Create a PV manually from an existing storage resource, then bind it directly in a PVC. Use this method when the storage resource already exists and must be ready before the container starts.
Dynamic provisioning: Define the desired storage in a StorageClass and reference it in a PVC. ACS automatically creates and binds a PV based on the PVC and StorageClass configuration. This method requires no manual PV creation and is more flexible for scaling workloads.

For step-by-step instructions, see:

Container Compute Service:Use Alibaba Cloud storage in ACS clusters