PVCs and multiple namespaces - kubernetes

Question

I have a small Saas App that customers can sign up for, and they get their own instance completely separated from the rest of the clients. It's a simple REST API, with a DB (postgres) and caddy that gets deployed using a docker-compose.
This works fine, but requires me to create the VPS, deploy the different services, and essentially is really hard to manage as I most of the work is manual.
I have decided to use kubernetes, and I have gotten to the point where I can create a separate instance of the system on it's own, isolated namespace for each client, fully automated. This creates the different deployments, services, and pods. Also, I create a PVC for each namespace/client.
The issue has to do with Persistent Volume Claims, and how they work in namespaces. As I want to keep the data completely separate from other instances, I wanted to create a PVC for each client, so that only the DB from that client can access it (and the server, as it requires some data to be written to disk).
This works fine in minikube, but the issue comes with the hosting provider. I use DigitalOcean's managed cluster and they do not allow multiple PVCs to be created, therefore making it impossible to achieve the level of isolation that I want. They allow you to mount a block storage (whatever size you need), and then use it. This would mean that the data is all stored on the "same disk", and all namespaces can access it.
My question is: Is there a way to achieve the same level of isolation, i.e. separate the mount points for each of the DB instances, in such a way that I can still achieve (or at least get close) to the level of separation that I require? The idea would be something along the lines of:
/pvc-root
/client1
/server
/db
/client2
/server
/db
...
This is what I have for now:
apiVersion: v1
kind: PersistentVolumeClaim
metadata:
creationTimestamp: null
labels:
io.kompose.service: database-claim
name: database-claim
spec:
accessModes:
- ReadWriteOnce
resources:
requests:
storage: 5Gi
---
apiVersion: v1
kind: Service
metadata:
labels:
io.kompose.service: database
name: database
spec:
ports:
- name: "5432"
port: 5432
targetPort: 5432
selector:
io.kompose.service: database
---
apiVersion: apps/v1
kind: Deployment
metadata:
labels:
io.kompose.service: database
name: database
spec:
replicas: 1
selector:
matchLabels:
io.kompose.service: database
strategy:
type: Recreate
template:
metadata:
labels:
io.kompose.service: database
spec:
containers:
- env:
- name: POSTGRES_DB
value: db_name
- name: POSTGRES_PASSWORD
value: db_password
- name: POSTGRES_USER
value: db_user
image: postgres:10
imagePullPolicy: ""
name: postgres
ports:
- containerPort: 5432
resources: {}
volumeMounts:
- mountPath: /var/lib/postgresql/data
name: database-claim
restartPolicy: Always
serviceAccountName: ""
volumes:
- name: database-claim
persistentVolumeClaim:
claimName: database-claim
---
apiVersion: v1
kind: Service
metadata:
labels:
io.kompose.service: server
name: server
spec:
ports:
- name: "8080"
port: 8080
targetPort: 8080
selector:
io.kompose.service: server
---
apiVersion: apps/v1
kind: Deployment
metadata:
creationTimestamp: null
labels:
io.kompose.service: server
name: server
spec:
replicas: 1
selector:
matchLabels:
io.kompose.service: server
template:
metadata:
labels:
io.kompose.service: server
spec:
containers:
- env:
- name: DB_HOST
value: database
- name: DB_NAME
value: db_name
- name: DB_PASSWORD
value: db_password
- name: DB_PORT
value: "5432"
- name: DB_USERNAME
value: db_user
image: <REDACTED>
name: server-image
ports:
- containerPort: 8080
restartPolicy: Always
volumes: null
EDIT Feb 02, 2021
I have been in contact with DO's customer support and they clarified a few things:
You need to manually attach a volume to the cluster, so the PVC deployment file is ignored. The volume is then mounted and available to the cluster but NOT in a ReadWriteMany config, which could have served this case quite well
They provide an API, so in theory I could create the volume (for each client) programmatically and then attach a volume for a specific client, keeping a ReadWriteOnce
This of course locks me in to them as a vendor, and makes things a bit harder to configure and migrate
I am still looking for suggestions whether this is the correct approach for my case. If you have a better way let me know!
in theory this should be achievabe

Answer 1

Is there a way to achieve the same level of isolation, i.e. separate the mount points for each of the DB instances, in such a way that I can still achieve (or at least get close) to the level of separation that I require?
Don't run a production database with a single volume. You want to run a database with some form of replication, in case a volume or node crash.
Either run the database in a distributed setup e.g. using Crunchy PostgreSQL for Kubernetes or use a managed database e.g. DigitalOcean Managed Database
Within that DBMS, create logical databases or schemas for each customer - if you really need that strong isolation. Hint: it is probably easier to maintain with less isolation e.g. using multi-tenancy within the tables.

Answer 2

Late to the party, but here's a solution that may bode well for your use case - if you still haven't found a solution.
Create a highly available NFS/CEPH server then export that in a way that your pods can attach to it. Then create PV's and PVC's as you'd like and bypass all that DO blockage.
You have an application very similar to what you describe I support and I went with a highly available NFS server using DRBD, Corosync, and Pacemaker and it all works as expected. No issues so far.