Wordpress uses persistence in 2 ways:
Database(MySQL/MariaDB) for the data in the blog/site
Filesystem(wp-content directory off the root of the installation) for storing plugins and themes, which can include executable php code
The filesystem is more complicated.
How does AKS handle persistence?
AKS can use:
Each AKS cluster includes two pre-created storage classes both configured to work with Azure disks.
apiVersion: v1 kind: PersistentVolumeClaim metadata: name: charliehoverflylagoons-uploads-claim annotations: #volume.beta.kubernetes.io/storage-class: managed-premium volume.beta.kubernetes.io/storage-class: default labels: app: hoverflylagoons spec: accessModes: - ReadWriteOnce resources: requests: storage: 20Gi
To change the type of storage (HDD/SSD) change the storage class in the PVC. The performance of the default storage class is similar (after a slower intall) for normal operations.
PersistentVolume (PV) - representation of storage in the cluster that has been manually provisioned, or dynamically provisioned by Kubernetes using a StorageClass. I never manually create these.
PersistentVolmeClaim (PVC) - a request for storage by a user that can be fulfilled by a PersistentVolume
From here Dynamic volume provisioning allows storage volumes to be created on-demand. I can define a PVC and it automatically provisions storage
Azure Disk for all /var/www/html
The simplest strategy is to have an attached disk on the node VM for each wordpress deployment. This disk will hold all the source eg /var/www/html, and all user updated content eg wp-content
However there is a max data disks limit depending on which VM you have.
- good performance (attached disks can be SSD’s)
- easy to setup
- can only attach the disk to 1 node VM (so not horizontally scalable). ie if the node goes down so does the website.
- security - it is appealing having all source files baked into the image, so if anything is changed a restart will fix it.
Here is a real example:
# pvc.yaml apiVersion: v1 kind: PersistentVolumeClaim metadata: name: hoverflylagoons-uploads-claim annotations: volume.beta.kubernetes.io/storage-class: managed-premium labels: app: wordpress spec: accessModes: - ReadWriteOnce resources: requests: storage: 20Gi # deployment.yaml apiVersion: apps/v1 kind: Deployment metadata: name: hoverflylagoons spec: replicas: 1 selector: matchLabels: app: hoverflylagoons-dep strategy: type: Recreate template: metadata: labels: app: hoverflylagoons-dep spec: containers: - image: wordpress:4.9.6-apache name: wordpress env: - name: WORDPRESS_DB_HOST value: secretname.mysql.database.azure.com - name: WORDPRESS_DB_PASSWORD valueFrom: secretKeyRef: name: mysql-pass key: password - name: WORDPRESS_DB_USER value: dave@secretmysql # default name is wordpress - name: WORDPRESS_DB_NAME value: hoverflylagoons ports: - containerPort: 80 name: wordpress volumeMounts: - name: hoverflylagoons-wordpress-persistent-storage mountPath: /var/www/html volumes: - name: hoverflylagoons-wordpress-persistent-storage persistentVolumeClaim: claimName: hoverflylagoons-uploads-claim
This works well and is performant, and the only limiting factor is the cost and the number of disks you can attach to a VM.
Azure File Storage for wp-content
This seemed like a smart way to go, and I got so far with this:
az storage account create --resource-group MC_aksrg_aks_westeurope --name davenfstest2 --location westeurope --sku Standard_LRS
# azurefilestorage.yaml kind: StorageClass apiVersion: storage.k8s.io/v1 metadata: name: azurefile provisioner: kubernetes.io/azure-file mountOptions: - dir_mode=0777 - file_mode=0777 - uid=1000 - gid=1000 parameters: storageAccount: davenfstest2 # azurefilepvc.yaml apiVersion: v1 kind: PersistentVolumeClaim metadata: name: azurefile spec: accessModes: - ReadWriteMany storageClassName: azurefile resources: requests: storage: 5Gi
Whilst testing this it is important to delete the secret in Kubernetes for the fileshare, otherwise you’ll get permissions errors!
Also I noticed that pods were not being deleted in a timely manner when I deleted the fileshare manually on the Azure portal. Here is a good command to force the delete, however this may mean the pod is still on the cluster. It is much better to delete from kubernetes, which will dynamically delete the fileshare.
kubectl delete pod PODNAME --grace-period=0 --force
All went well with the wordpress installation and the expected files were in the fileshare.
However I couldn’t get the wordpress user to write to that file share. This is the cryptic error you see when the permissions are not correct.
kubectl exec -it shell-demo -- /bin/bash
After changing the wp-config.php to include
define('FS_METHOD', 'direct'); the permissions error went away. To get this working well, I’d need to bake this into my Dockerfile image.
However a bigger issue is performance:
Even in front end testing the load times are:
- Azure file 2s
- Azure disk HDD and SSD 850ms
A workaround could be to use WP Super Cache to generate html. This does work well and front end performance is fine (wwhen the cache is pointing to the container filesystem eg /var/www/wpsupercache
Care needs to be taken with FS_METHOD.
Performance on the edit screen was really bad too (this is with no plugins installed apart from defaults). I have measured a difference from 900ms on disk to 17s on file (actually this was for everything loaded from the file share). A more accurate difference is when only wp-content is on the fileshare and the difference is 1s on disk, 2s on fileshare.
Using Azure Files with Linux is a more detailed explanation. Perhaps there is a more performant way to access files, and to get the permissions correct than I have tried above.
Azure Disk for wp-content
This could be a good tactic, however the production website I’m dealing with has a caching plugin enabled (WP Super Cache) which needs access to wp-config.php in the root. So this tactic would give some better security (in that a restart would reset everything back to a base state), but I’d have to deal with plugin issues too.
Azure doesn’t yet have a performant distributed filesystem, so I am trialling GlusterFS using Azure Disks.
Kubepress implements Gluster.
Codeable tutorial uses Gluster
OSS Canada uses modified ARM templates from Azure quickstart.
Dysk looks promising but very early stages.