K
Q

Persistent disk problem on Kubernetes GCP

2/15/2019

I'm working in Kubernetes in GCP and I'm having problems with volumes and persistent disks.

I'm using Directus 7 (CMS Headless), which saves most of its information in the database except the files that are uploaded, these files are in the /var/www/html/public/uploads folder (tested locally with docker-compose and works fine), and that folder is the one I'm trying to save on the persistent disk.

No error occurs but when restart the Kubernetes Pod i lose the uploaded images (they are not being saved on the disk).

This is my configuration:

apiVersion: v1
kind: PersistentVolume
metadata:
  name: directus-pv
  namespace: default
spec:
  storageClassName: ""
  capacity:
    storage: 100G
  accessModes:
    - ReadWriteOnce
  gcePersistentDisk:
    pdName: directus-disk
    fsType: ext4

apiVersion: v1
kind: PersistentVolumeClaim
metadata:
  name: directus-pvc
  namespace: default
  labels:
    app: .....
spec:
  storageClassName: ""
  volumeName: directus-pv
  accessModes:
    - ReadWriteOnce
  resources:
    requests:
      storage: 100G

And in the deploy.yaml:

    volumeMounts:
      - name: api-disk
        mountPath: /var/www/html/public/uploads
        readOnly: false

  volumes:
  - name: api-disk
    persistentVolumeClaim:
      claimName: directus-pvc

Thanks for the help

-- pdsm
directus
docker
google-cloud-platform
kubernetes
persistent-storage

Similar Questions

Cloud SQL connection for Kubernetes using proxy
Simple Getting Started Istio Helm Install
How to Log spanId in Google Kubernetes Engine
How to run Tiller on Kubernetes cluster on AWS EKS
How horizontal pod autoscaler treats not-ready pods when one changes max/min replicas limit
With Helm 3 and K8s cluster w/RBAC (X509 certs), what permissions does Helm need? I think it creates ConfigMaps for release info?

2 Answers

2/16/2019

Remove namespace property from pv and pvc manifest. They are shared resources in the cluster. Remove storage class property as well.

-- P Ekambaram
Source: StackOverflow
2/21/2019

I presume that your manually provisioned persistence volume directus-pv, is being created somehow with PersistentVolumeReclaimPolicy=*Recycle. That's the only possible reason that could cause data erase on each POD restart.

I'm not able to reproduce your case with the provided manifest files, but I tried the following test:

  1. Create gcePersistentDisk
  2. Create PersistentVolume
  3. Create PersistentVolumeClaim
  4. Create ReplicaSet (replicas=1) like this one
apiVersion: extensions/v1beta1
kind: ReplicaSet
metadata:
  name: busybox-list-uploads
spec:
  replicas: 1
  template:
    metadata:
      labels:
        app: busybox-list-uploads
        version: "2"
    spec:
      containers:
        - image: busybox
          args: [/bin/sh, -c, 'sleep 9999' ]
          volumeMounts:
            - mountPath: /var/www/html/public/uploads
              name: api-disk
          name: busybox
      volumes:
      - name: api-disk
        persistentVolumeClaim:
          claimName: directus-pvc
  1. Write some file into mounted folder /var/www/html/public/uploads
  2. Restart POD (=kill the POD) by resizing replica to 0 then to 1
  3. List content of /var/www/html/public/uploads on newly created POD

for i in busybox-list-uploads-dgfbc; do kubectl exec -it $i -- ls /var/www/html/public/uploads; done; lost+found picture_from_busybox-list-uploads-ng4t6.png

As you can see output shows clearly, that data survives POD restart

* you can verify it with cmd: kubectl get pv/directus-pv -o yaml

-- Nepomucen
Source: StackOverflow