Stateful Kubernetes Applications Made Easier: PSO and FlashBlade

For stateful applications on Kubernetes, Pure Service Orchestrator automates and productionizes those services with FlashBlade as shared storage. This repository presents three illustrative example applications: a simplistic shared scratch space as well as file-sharing services NextCloud and OwnCloud.

I assume the reader has only a basic understanding of Kubernetes and is interested in how to use FlashBlade with Kubernetes.

For a more general introduction and guide to PSO for both FlashArray and FlashBlade, see this explanation of Kubernetes Storage, Kubernetes documentation, and Containers-as-a-service architecture.

Introduction

The Pure Service Orchestrator (PSO) automates the process of creating filesystems on FlashBlade and attaching them to the running applications in Kubernetes.

But first, why run applications in Kubernetes? There is a generic set of problems that almost all applications need to solve: recovering from hardware failures, scaling up or down resources, orchestrating connectivity between inter-linked services, securing and isolating environments, and provisioning resources for running applications. Kubernetes helps solve these problems, simplifying the operation of production services.

Why Use PSO and FlashBlade

In Kubernetes, applications run inside a set of pods, which are ephemeral and not tied to physical compute or storage resources. If the pod needs persistent storage for any reason (and there are many!), then PersistentVolumes are needed. But manual storage administration is complicated: creating volumes, attaching them to running pods regardless of protocol, and finally deleting volumes. In a fully automated system, none of these steps should involve a human!

PSO automates the process of provisioning storage and hides the details of storage creation and attachment to each pod. The result is self-service storage that matches cloud-native workflows and applications.

FlashBlade provides shared filesystems for Read-Write-Many (RWX) volumes. These are critical for scale-out applications that spawn multiple pods; each additional pod is able to automatically share access to a common data store. FlashBlade can also support Read-Write-Once (RWO) volumes.

While I focus here on FlashBlade filesystems as volumes, PSO also serves FlashArray and block devices. The details of storage administration are different for block (FlashArray) and file (FlashBlade) volumes, and PSO hides all of these differences from the end-user. For example, a developer creating an application that uses a ReadWriteOnce volume can choose between latency sensitive or bandwidth sensitive performance simply by changing the text between “pure-file” and “pure-block.” All of the annoying differences between iSCSI, Fibre Channel, and NFS mounts are hidden automatically by PSO.

How to Use PSO with FlashBlade

Compared to traditional storage workflows, PSO automates most of the requests and interactions between users and administrators for file system creation and mounting. The result is a self-service storage infrastructure; no direct interaction is necessary between administrators and users.

Administrators install and occasionally upgrade the PSO software and users (developers) interact only via standard Kubernetes PersistentVolume mechanisms.

For the Kubernetes Administrator

PSO is a software layer for Kubernetes that utilizes the public REST API provided by FlashBlade to automate storage-related tasks and is installed via helm.

The one-time PSO configuration requires a FlashBlade management IP, data VIP, and the corresponding API token. These API tokens prove that PSO has permission to create and delete volumes or filesystems so safeguard them appropriately.

To get the API token, use the following command from the FlashBlade CLI:

Then, add the following to the PSO values.yaml (example) file to add a FlashBlade under management:

The PSO works by fulfilling any PVC claims with a StorageClass with corresponding name, “pure-file,” by creating a matching filesystem.

Update the default storageClass to use the “pure-file” class if you want FlashBlade to be the default for PVCs that do not specify StorageClass.

After this change, view the configured StorageClasses and their associated provisioner with the following command:

Finally, to view the PersistentVolumes in the system which were automatically created by PSO:

The PersistentVolumes reported here should match the filesystems created on the FlashBlade, as seen with the following CLI command in the FlashBlade:

Adding a StorageQuota (example) allows the administrator to limit the number of claims or the total amount of storage requested. These limits cause graceful failures in the case of buggy jobs that consume too much storage.

For the Kubernetes User

The following examples create a PersistentVolumeClaim that will be satisfied by the PSO by using a storageClassName of “pure-file”. The storageClass signals to Kubernetes how to satisfy the claim request. Switching to use a FlashArray block device instead can be done by using “pure-block” as StorageClass.

A PersistentVolumeClaim is a request for a PersistentVolume of certain type and size. If there is a plugin for the specified storageClass, then it will create a matching PersistentVolume. To attach this storage to a container at a specific path, refer back to the claim as follows:

What did you NOT need to do?

  • Create a filesystem via a UI of any kind (GUI or CLI)
  • Log in to the node and issue the commands to mount the filesystem
  • Think about what happens if the physical node fails and you need to move the app
  • Track who uses which filesystem so you know when to cleanup

And what you do get is that the volume automatically follows your container if Kubernetes restarts or moves the container to a different physical host.

Example 1: Shared Scratch Space

The first example to illustrate PSO is a simple Deployment and PersistentVolumeClaim to create a shared scratch workspace. In other words, the application is a standard Linux shell and command line tools.

A shared scratch workspace is useful when a small team needs to work together, for example for forensics investigation of log files. The goal is to automatically create a shared scratch space to download necessary data to work with and produce derivative datasets. PSO automates the creation, connection, and cleanup of the workspace on FlashBlade.

scratch-space.yaml.

The yaml file creates two pods that both mount a shared “/scratch” directory for collaborative work. This illustrates the core steps necessary to use a persistentVolumeClaim to attach a volume to pods in a Deployment.

To use this shared scratch space, each user connects to a pod and spawns a shell by running ‘exec’ on one of the pod instances:

Once this shell has been started, the user can collaborate using the ‘/scratch’ directory and all results will be visible to users in other pods.

What if a third user wants to also collaborate? Adding additional workspaces that share the same filesystem is as simple as:

Each additional replica pod is automatically connected to the filesystem, i.e., the mounting and unmounting of the filesystem is automatically handled by PSO.

There are many directions to make this example even more useful: 1) add a unique name to the resources based on a ticket number, 2) automate cleanup of resources when finished, and 3) use a prebuilt container image with already installed tools.

Example 2: NextCloud

NextCloud

NextCloud and OwnCloud are both open-source file sharing applications providing an alternative to cloud services like DropBox. These applications provide similar file sharing services across multiple client platforms while keeping the data in-house instead of an offsite 3rd party cloud; either performance, cost, or data governance may drive this desire for an in-house service instead of external. Both applications rely upon persistent storage for the user’s data and metadata.

The NextCloud configuration borrows heavily from existing documentation for NextCloud. This configuration is kept simple in order to highlight the usage of persistent storage.

The config file combines three major elements:

  • Service ‘nextcloud’ that exposes the application externally
  • Deployment ‘nextcloud’ that starts the application server and mounts a filesystem
  • PersistentVolumeClaim that PSO will use to create a matching filesystem on the FlashBlade

Create the nextcloud service as follows:

To make this application accessible externally, I used a NodePort service, which is not recommended in production. To access the NextCloud application, find the port assigned by Kubernetes:

Connect to the NextCloud server by using that port number and the ip address of any Kubernetes node. Production environments should configure load balancers instead.

Using InitContainers for Setup

The NextCloud example utilizes an initContainer to solve a common problem: configuring the volume to meet the conditions expected by the application. An initContainer is a container that is run to completion before the application container is started and must complete successfully for the application to begin. InitContainers encode the steps necessary to satisfy pre-conditions or dependencies that the user does not want to or cannot build into the main application container. As an example, an initContainer can download a dataset to the persistent volume for the primary application to use.

The NextCloud server expects the “data/” directory to be owned by the www-data user and to have permissions of 770, otherwise the application fails. To accomplish this, an initContainer performs two simple commands to achieve this precondition:

Many applications, especially legacy applications not originally built for containers, expect preconditions on the volumes and initContainers are the easiest way to achieve this. Beyond the simple example here with busybox, different container images can be used to place data or install necessary software on the volume.

Example 3: OwnCloud

OwnCloud

The OwnCloud deployment involves multiple applications working together: mariadb and redis alongside the owncloud server. These applications use a mix of block devices and filesystems. For example, MariaDB is an RDBMS service that requires a block device backed volume whereas OwnCloud stores user data on a filesystem-backed volume. Derived from docker documentation for OwnCloud.

With PSO, multiple applications and volume types can be mixed and modified easily. For example, switching the Redis application between “pure-file” and “pure-block” is as simple as changing the value in the yaml config before deploying.

Running multiple different applications on FlashBlade works well because the system is 1) built to support high concurrency and many simultaneous applications, 2) scales-out seamlessly and non-disruptively, and 3) is architected natively to leverage the mixed IO performance of NAND flash.

Deploy this set of applications and volumes together with the following command:

Note that there is a helm chart for OwnCloud here but it is currently not working due to a failed connection between OwnCloud and mariadb. With the default storageClass set to “pure-file,” this helm chart would automatically use FlashBlade as the backing store.

Quick Tips for Usability

Below are some quick-tips that I found useful in running the OwnCloud application.

Run OwnCloud reporting commands internally:

Log in to the server and look around at the contents of the filesystem.

Summary

A good software engineer always looks to automate tedious and error-prone tasks in order to increase the team’s productivity. Running Kubernetes and the Pure Service Orchestrator automates almost all storage related tasks: filesystem creation, mounting, and deletion. The result is simple and agile infrastructure that just works.

Legal stuff — Please note: I am an employee of Pure Storage. My statements and opinions on this site are my own and do not necessarily represent those of Pure Storage.

Written by

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store