IBM Cloud Kubernetes Service storage overview

Virtual Private Cloud Classic infrastructure Satellite

Review the following sections for an overview of the available storage options for your cluster.

Before you can decide what type of storage is the correct solution for your IBM Cloud® Kubernetes Service clusters, you must understand the IBM Cloud infrastructure provider, your app requirements, the type of data that you want to store, and how often you want to access this data.

Decide whether your data must be permanently stored.

Persistent storage: Data stored on persistent storage persists even when the container, the worker node, or the cluster is removed. Use persistent storage in the for stateful apps, core business data or data that must be available due to legal requirements, such as a defined retention period. Persistent storage is also a good option for auditing.

Non-persistent storage: Your data can be removed when the container, the worker node, or the cluster is removed. Non-persistent storage is typically used for logging information, such as system logs or container logs, development testing, or when you want to access data from the host's file system.

If you must persist your data, analyze if your app requires a specific type of storage. When you use an existing app, the app might be designed to store data in one of the following ways.

In a file system: The data can be stored as a file in a directory. For example, you could store this file on your local hard disk. Some apps require data to be stored in a specific file system, such as nfs or ext4 to optimize the data store and achieve performance goals.

In a database: The data must be stored in a database that follows a specific schema. Some apps come with a database interface that you can use to store your data. For example, WordPress is optimized to store data in a MySQL database. In these cases, the type of storage is selected for you.

Determine the type of data that you want to store.

Structured data: Data that you can store in a relational database where you have a table with columns and rows. Data in tables can be connected by using keys and is usually easy to access due to the pre-defined data model. Examples are phone numbers, account numbers, Social Security numbers, or postal codes.

Semi-structured data: Data that does not fit into a relational database, but that comes with some organizational properties that you can use to read and analyze this data more easily. Examples are markup language files such as CSV, XML, or JSON.

Unstructured data: Data that does not follow an organizational pattern and that is so complex that you can't store it in a relational database with pre-defined data models. To access this data, you need advanced tools and software. Examples are e-mail messages, videos, photos, audio files, presentations, social media data, or web pages.

If you have structured and unstructured data, try to store each data type separately in a storage solution that is designed for this data type. Using an appropriate storage solution for your data type eases up access to your data and gives you the benefits of performance, scalability, durability, and consistency.

Analyze how you want to access your data. Storage solutions are usually designed and optimized to support read or write operations.

Read-only: You don't want to write or change your data. Your data is read-only.

Read and write: You want to read, write, and change your data. For data that is read and written, it is important to understand if the operations are read-heavy, write-heavy, or balanced.

Determine the frequency that your data is accessed.

Hot data: Data that is accessed frequently. Common use cases are web or mobile apps.

Cool or warm data: Data that is accessed infrequently, such as once a month or less. Common use cases are archives, short-term data retention, or disaster recovery.

Cold data: Data that is rarely accessed, if at all. Common use cases are archives, long-term backups, historical data.

Frozen data: Data that is not accessed and that you need to keep due to legal reasons.

If you can't predict the frequency or the frequency does not follow a strict pattern, determine whether your workloads are read-heavy, write-heavy, or balanced. Then, look at the storage option that fits your workload and investigate what storage tier gives you the flexibility that you need. For example, IBM Cloud Object Storage provides a flex storage class that considers how frequent data is accessed in a month and takes into account this measurement to optimize your monthly billing.

Investigate if your data must be shared across multiple app instances, zones, or regions.

Access across pods: When you use Kubernetes persistent volumes to access your storage, you can determine the number of pods that can mount the volume at the same time. Some storage solutions can be accessed by one pod at a time only. With other storage solutions, you can share volume across multiple pods.

Access across zones and regions: You might require your data to be accessible across zones or regions. Some storage solutions, such as file and block storage, are data center-specific and can't be shared across zones in a multizone cluster setup.

If you want to make your data accessible across zones or regions, make sure to consult your legal department to verify that your data can be stored in multiple zones or a different country.

Understand other storage characteristics that impact your choice.

Consistency: The guarantee that a read operation returns the latest version of a file. Storage solutions can provide strong consistency when you are guaranteed to always receive the latest version of a file, or eventual consistency when the read operation might not return the latest version. You often find eventual consistency in geographically distributed systems where a write operation first must be replicated across all instances.

Performance: The time that it takes to complete a read or write operation.

Durability: The guarantee that a write operation that is committed to your storage survives permanently and does not get corrupted or lost, even if gigabytes or terabytes of data are written to your storage at the same time.

Resiliency: The ability to recover from an outage and continue operations, even if a hardware or software component failed. For example, your physical storage experiences a power outage, a network outage, or is destroyed during a natural disaster.

Availability: The ability to provide access to your data, even if a data center or a region is unavailable. Availability for your data is usually achieved by adding redundancy and setting up failover mechanisms.

Scalability: The ability to extend capacity and customize performance based on your needs.

Encryption: The masking of data to prevent visibility when data is accessed by an unauthorized user.

Non-persistent storage options

You can use non-persistent storage options if your data is not required to be persistently stored or if you want to unit-test your app components. The following image shows available non-persistent data storage options in IBM Cloud Kubernetes Service.

Figure 1. Non-persistent storage options

Characteristics	Inside the container	On the worker node's primary or secondary disk
Multizone capable	No	No
Data types	All	All
Capacity	Limited to the worker node's available secondary disk. To limit the amount of secondary storage that is consumed by your pod, use resource requests and limits for ephemeral storage.	Limited to the worker node's available space on the primary (hostPath) or secondary disk (emptyDir). To limit the amount of secondary storage that is consumed by your pod, use resource requests and limits for ephemeral storage.
Data access pattern	Read and write operations of any frequency	Read and write operations of any frequency
Access	Via the container's local file system	Via Kubernetes hostPath for access to worker node primary storage. Via Kubernetes emptyDir volume for access to worker node secondary storage.
Performance	High	High with lower latency when you use SSD
Resiliency	Low	Low
Availability	Specific to the container	Specific to the worker node
Scalability	Difficult to extend as limited to the worker node's secondary disk capacity	Difficult to extend as limited to the worker node's primary and secondary disk capacity
Durability	Data is lost when the container crashes or is removed.	Data in hostPath or emptyDir volumes is lost when the worker node is deleted, the worker node is reloaded or updated, the cluster is deleted, the IBM Cloud account reaches a suspended state. In addition, data in an emptyDir volume is removed when the assigned pod is permanently deleted from the worker node, the assigned pod is scheduled on another worker node.
Common use cases	Local image cache or container logs	Setting up a high-performance local cache, accessing files from the worker node file system, or running unit tests.
Non-ideal use cases	Persistent data storage or sharing data between containers	Persistent data storage

Single zone clusters

If you have a single zone cluster, you can choose between the following options in IBM Cloud Kubernetes Service that provide fast access to your data. For higher availability, use a storage option that is designed for geographically distributed data and, if possible for your requirements, create a multizone cluster.

The following image shows the options that you have in IBM Cloud Kubernetes Service to permanently store your data in a single cluster.

Table 1. Storage options for single zone clusters

Characteristics	Description
Deployment guide	Setting up File Storage for Classic.
Ideal data types	All
Supported provisioning type	Dynamic and static
Data usage pattern	Random read-write operations, sequential read-write operations, or write-intensive workloads
Access	Via file system on mounted volume
Supported Kubernetes access modes	ReadWriteMany (RWX) ReadOnlyMany (ROX) ReadWriteOnce (RWO)
Performance	Predictable due to assigned IOPS and size. IOPS are shared between the pods that access the volume.
Consistency	Strong
Durability	High
Resiliency	Medium as specific to a data center. File storage server is clustered by IBM with redundant networking.
Availability	Medium as specific to a data center.
Scalability	Difficult to extend beyond the data center. You can't change an existing storage tier.
Encryption	At rest
Backup and recovery	Set up periodic snapshots, replicate snapshots, duplicate storage, back up data to IBM Cloud Object Storage, or copy data to and from pod and containers.
Common use cases	Mass or single file storage or file sharing across a single zone cluster.
Non-ideal use cases	Multizone clusters or geographically distributed data.

Table 1. Storage options for single zone clusters

Characteristics	Description
Deployment guide	Setting up Block Storage for Classic.
Ideal data types	All
Supported provisioning type	Dynamic and static
Data usage pattern	Random read-write operations, sequential read-write operations, or write-intensive workloads
Access	Via file system on mounted volume.
Supported Kubernetes access modes	ReadWriteOnce (RWO)
Performance	Predictable due to assigned IOPS and size. IOPS are not shared between pods.
Consistency	Strong
Durability	High
Resiliency	Medium as specific to a data center. Block storage server is clustered by IBM with redundant networking.
Availability	Medium as specific to a data center.
Scalability	Difficult to extend beyond the data center. You can't change an existing storage tier.
Encryption	At rest.
Backup and recovery	Set up periodic snapshots, replicate snapshots, duplicate storage, back up data to IBM Cloud Object Storage, or copy data to and from pod and containers.
Common use cases	Stateful sets, backing storage when you run your own database, or high-performance access for single pods.
Non-ideal use cases	Multizone clusters, geographically distributed data, or sharing data across multiple app instances.

Table 1. Storage options for single zone clusters

Characteristic	Description
Deployment guide	Setting up File Storage for VPC.
Ideal data types	All
Supported provisioning type	Dynamic and static
Data usage pattern	Random read-write operations, sequential read-write operations, or write-intensive workloads
Access	Via file system on mounted volume
Supported Kubernetes access modes	ReadWriteMany (RWX) ReadOnlyMany (ROX) ReadWriteOnce (RWO) Version 1.2 and later.
Performance	Predictable due to assigned IOPS and size. IOPS are not shared between pods.
Consistency	Strong
Durability	High
Resiliency	Medium as specific to a data center. File storage server is clustered by IBM with redundant networking.
Availability	Medium as specific to a data center.
Scalability	Difficult to extend beyond the data center. You can't change an existing storage tier.
Encryption	None
Backup and recovery	Run kubectl cp or copy data to and from pod and containers.
Common use cases	Mass or single file storage or file sharing across a single zone cluster.
Non-ideal use cases	Multizone clusters, geographically distributed data, or sharing data across multiple app instances.

Table 1. Storage options for single zone clusters

Characteristics	Description
Deployment guide	Setting up Block Storage for VPC.
Multizone-capable	No, as specific to a data center. Data can't be shared across zones, unless you implement your own data replication.
Ideal data types	All
Data usage pattern	Random read-write operations, sequential read-write operations, or write-intensive workloads
Access	Via file system on mounted volume
Supported Kubernetes access writes	ReadWriteOnce (RWO)
Performance	Predictable due to assigned IOPS and size. IOPS are not shared between pods.
Consistency	Strong
Durability	High
Resiliency	Medium as specific to a data center. Block storage server is clustered by IBM with redundant networking.
Availability	Medium as specific to a data center.
Scalability	Difficult to extend beyond the data center. You can't change an existing storage tier.
Encryption	Encryption in transit with Key Protect
Backup and recovery	Set up periodic snapshots, replicate snapshots, duplicate storage, back up data to IBM Cloud Object Storage, or copy data to and from pod and containers.
Common use cases	Stateful sets, backing storage when you run your own database, or high-performance access for single pods.
Non-ideal use cases	Multizone clusters, geographically distributed data, or sharing data across multiple app instances.

Multizone clusters

The following sections show the options that you have in IBM Cloud Kubernetes Service to permanently store your data in a multizone cluster and make your data highly available. You can use these options in a single zone cluster, but you might not get the high availability benefits that your app requires.

Table 2. Storage options for multizone clusters

Characteristic	Description
Deployment guide	Setting up IBM Cloud Object Storage.
Supported infrastructure providers	Classic, VPC, Satellite
Ideal data types	Semi-structured and unstructured data
Data usage pattern	Read-intensive workloads. Few or no write operations.
Access	Via file system on mounted volume (plug-in) or via REST API from your app
Supported Kubernetes access modes	ReadWriteMany (RWX)
Performance	High for read operations. Predictable due to assigned IOPS and size when you use non-SDS machines.
Consistency	Eventual
Durability	Very high as data slices are dispersed across a cluster of storage nodes. Every node stores only a part of the data.
Resiliency	High as data slices are dispersed across three zones or regions. Medium, when set up in a single zone only.
Availability	High due to the distribution across zones or regions.
Scalability	Scales automatically
Encryption	In transit and at rest
Backup and recovery	Data is automatically replicated across multiple nodes for high durability. For more information, see the SLA in the IBM Cloud Object Storage service terms.
Common use cases	Geographically distributed data, static big data, static multimedia content, web apps, backups, archives, stateful sets.
Non-ideal use cases	Write-intensive workloads, random write operations, incremental data updates, or transaction databases.

Table 2. Storage options for multizone clusters

Characteristics	Description
Deployment guide	Setting up Portworx.
Supported infrastructure providers	Classic, VPC, Satellite
Ideal data types	Any
Data usage pattern	Read-intensive workloads. Few or no write operations.
Access	Via file system on mounted volume (plug-in) or via REST API from your app
Supported Kubernetes access modes	ReadWriteMany (RWX) ReadOnlyMany (ROX) ReadWriteOnce (RWO)
Performance	High for read operations. Predictable due to assigned IOPS and size when you use non-SDS machines.
Consistency	Eventual
Durability	Very high as data slices are dispersed across a cluster of storage nodes. Every node stores only a part of the data.
Resiliency	High as data slices are dispersed across three zones or regions. Medium, when set up in a single zone only.
Availability	High due to the distribution across zones or regions.
Scalability	Scales automatically
Encryption	Bring your own key to protect your data in transit and at rest with IBM Key Protect.
Backup and recovery	Data is automatically replicated across multiple nodes for high durability. For more information, see the SLA in the IBM Cloud Object Storage service terms. Use local or cloud snapshots to save the current state of a volume. For more information, see Create and use local snapshots.
Common use cases	Multizone clusters. Geographically distributed data. Static big data. Static multimedia content
Non-ideal use cases	Write-intensive workloads, random write operations, incremental data updates, or transaction databases.

Table 2. Storage options for multizone clusters

Characteristics	Description
Deployment guide	Connect a Cloud Databases deployment to an IBM Cloud Kubernetes Service application.
Supported infrastructure providers	Classic, VPC, Satellite
Ideal data types	Depends on the DBaaS
Data usage pattern	Read-write-intensive workloads
Access	Via REST API from your app.
Supported Kubernetes access writes	N/A as accessed from the app directly.
Performance	High if deployed to the same data center as your app.
Consistency	Depends on the DBaaS
Durability	High
Resiliency	Depends on the DBaaS and your setup.
Availability	High if you set up multiple instances.
Scalability	Scales automatically
Encryption	At rest
Backup and recovery	Depends on the DBaaS
Common use cases	Multizone clusters, relational and non-relational databases, or geographically distributed data.
Non-ideal use cases	App that is designed to write to a file system.