Here is my overall goal:
Have a MongoDB running
Persist the data through pod failures / updates etc
The approach I’ve taken:
K8S Provider: Digital Ocean
Nodes: 3
Create a PVC
Create a headless Service
Create a StatefulSet
Here’s a dumbed down version of the config:
apiVersion: v1
kind: PersistentVolumeClaim
metadata:
name: some-pvc
spec:
accessModes:
- ReadWriteOnce
resources:
requests:
storage: 5Gi
storageClassName: do-block-storage
---
apiVersion: v1
kind: Service
metadata:
name: some-headless-service
labels:
app: my-app
spec:
ports:
- port: 27017
name: my-app-database
clusterIP: None
selector:
app: my-app
tier: database
---
apiVersion: apps/v1
kind: StatefulSet
metadata:
name: my-app-database
labels:
app: my-app
tier: database
spec:
serviceName: some-headless-service
replicas: 1
selector:
matchLabels:
app: my-app
tier: database
template:
metadata:
labels:
app: my-app
tier: database
spec:
containers:
- name: my-app-database
image: mongo:latest
volumeMounts:
- name: some-volume
mountPath: /data
ports:
- containerPort: 27017
name: my-app-database
volumes:
- name: some-volume
persistentVolumeClaim:
claimName: some-pvcThis is working as expected. I can spin down the replicas to 0:
kubectl scale —replicas=0 statefulset/my-app-database
Spin it back up:
kubectl scale —replicas=1 statefulset/my-app-database
And the data will persist..
But one time, as I was messing around by scaling the statefulset up and down, I was met with this error:
Volume is already exclusively attached to one node and can't be attached to anotherBeing new to k8s, I deleted the PVC and “recreated” the same one:
kubectl delete pvc some-pvc
kubectl apply -f persistent-volume-claims/The statefulset spun back up with a new PV and the old PV was deleted as the persistentVolumeReclaimPolicy was set to Delete by default.
I set this new PV persistentVolumeReclaimPolicy to Retain to ensure that the data would not be automatically removed.. and I realized: I’m not sure how I’d reclaim that PV. Earlier to get through the “volume attachment” error, I deleted the PVC, which will just create another new PV with the setup I have, and now I’m left with my data in that Released PV.
My main questions are:
Does this overall sound like the right approach for my goal?
Should I look into adding a claimRef to the dynamically created PV and then recreating a new PVC with that claimRef, as mentioned here: Can a PVC be bound to a specific PV?
Should I be trying to get that fresh statefulset PVC to actually use that old PV?
Would it make sense to try to reattach the old PV to the correct node, and how would I do that?
If your want to use StatefulSet with scalability, your storage should also support this, there are two way to handle this:
If do-block-storage storage class supprt ReadWriteMany, then put all pod's data in single volume.
Each pod use a different volume, add volumeClaimTemplate to your StatefulSet.spec, then k8s will create PVC like some-pvc-{statefulset_name}-{idx} automatically:
spec:
volumeClaimTemplates:
- metadata:
name: some-pvc
spec:
accessModes:
- ReadWriteOnce
resources:
requests:
storage: 5Gi
storageClassName: do-block-storageUpdate:
StatefulSet replicas Must deploy with mongodb replication, then each pod in StatefulSet will has same data storage.
So when container run mongod command, you must add option --replSet={name}. when all pods up, execute command rs.initiate() to tell mongodb how to handle data replication. When you scale up or down StatefulSet, execute command rs.add() or rs.remove() to tell mongodb members has changed.