Mounts the /var/lib/postgresql/data directory again

[scipsycho]

I am using Kubernetes to spin up a postgres pod/container. I also want to make it persistent, so I am using a smb server (TrueNas) to create a PV / PVC to be used by this pod.

Here is my simple setup:

Config map to store some credentials.

apiVersion: v1
kind: ConfigMap
metadata:
  name: postgres-secret
  labels:
    app: postgres
data:
  POSTGRES_DB: ps_db
  POSTGRES_USER: postgres
  PGUSER: postgres
  POSTGRES_PASSWORD: <password>

---

Persistent volume and persistent volume claim to be used by the pod

apiVersion: v1
kind: PersistentVolume
metadata:
  name: services-postgres
spec:
  storageClassName: freenas-smb-csi
  capacity:
    storage: 50Gi
  accessModes:
    - ReadWriteMany
  persistentVolumeReclaimPolicy: Retain
  mountOptions:
    - username=postgres
    - password=<passoword>
    - uid=999 #Mapping the user running the postgres process with the one that is the owner of the SMB share 
    - dir_mode=0700 # Doing this as Postgres demands the pgdata directory to have 0700 or 0750 permissions
    - noperm # Truenas uses acl for permissions. Need to use this so that Postgres ignored it.
  csi:
    driver: org.democractic-csi.smb
    readOnly: false
    fsType: cifs
    volumeHandle: postgres

    volumeAttributes:
      server:  <ip_addr>
      share: postgres
      node_attach_driver: smb
      provisioner_driver: smb

---

apiVersion: v1
kind: PersistentVolumeClaim
metadata:
  name: services-postgres-claim
spec:
  storageClassName: freenas-smb-csi
  volumeName: services-postgres
  accessModes:
    - ReadWriteMany
  volumeMode: Filesystem
  resources:
    requests:
      storage: 50Gi

---

Service to expose the server

apiVersion: v1
kind: Service
metadata:
  name: postgres-service
spec:
  sessionAffinity: ClientIP
  selector:
    app: postgres
  ports:
  - port: 5432
    targetPort: 5432
    name: postgres-tcp
    protocol: TCP
 

Main deployment file.

apiVersion: apps/v1
kind: Deployment
metadata:
  name: postgres
spec:
  replicas: 1
  selector:
    matchLabels:
      app: postgres
  template:
    metadata:
      labels:
        app: postgres
    spec:
      securityContext:
        runAsUser: 999
        fsGroup: 1008
      containers:
        - name: postgres
          image: 'postgres:14'
          imagePullPolicy: IfNotPresent
          ports:
            - containerPort: 5432
          envFrom:
            - configMapRef:
                name: postgres-secret
          env:
          - name: PGDATA
            value: /var/lib/postgresql/mounted_data/pgdata # PGDATA directory is different from the normal /var/lib/postgresql/data

         # Mounting a subdirectory to the mount point as it has been said that
         # postgres doesn't like it. Something to do with permissions.
          volumeMounts:
            - mountPath: /var/lib/postgresql
              name: postgres
              subPath: var_lib_postgresql
                                    
         # Not necessarily needed but from the logs I could see that postgres was failing once before restarting
         # I tried to prevent that by adding this check
         # but it seems that this is an expected behavior from postgres (?)
          readinessProbe: 
            exec:
              command:
                - pg_isready
            initialDelaySeconds: 5
            timeoutSeconds: 5
            failureThreshold: 5

      volumes:
        - name: postgres
          persistentVolumeClaim:
            claimName: services-postgres-claim

Now, the issue is that for some reason,

Postgres container has the /var/lib/postgresql/data directory mounted to some other virtual device (probably created by containerd runtime) when the /var/lib/postgres directory is already mounted to the PVC.

From inside the container, I see the following:

$  df -hT /var/lib/postgresql
Filesystem               Type  Size  Used Avail Use% Mounted on
//<ip-addr>/postgres cifs  8.7T   96K  8.7T   1% /var/lib/postgresql
$ df -hT /var/lib/postgresql/data
Filesystem     Type  Size  Used Avail Use% Mounted on
/dev/sda6      xfs    99G   12G   88G  12% /var/lib/postgresql/data

Intially, I had the PGDATA set to /var/lib/postgresql/data/data but this was leading me to loose the entire data because of the exact same problem. Because of this, I cannot use the data driectory to store anything as it would be mounted later on.

Is this expected behavior? Is there anyway, I can resolve this issue or is this something that is not related to postgres but kubernetes?

[tianon]

Given that you're setting PGDATA explicitly, I think the extra mount at /var/lib/postgresql/data is a red herring -- it won't get used, and is an "anonymous volume" provided by the runtime in response to the image metadata.

I'm guessing you're running into issues running the service? I would imagine that PostgreSQL itself doesn't like having PGDATA stored on a Samba/SMB/cifs/"Windows File Sharing" mount. 😬

[scipsycho]

I'm guessing you're running into issues running the service?

Yeah, I was running into data loss issues and I just couldn't understand why. I had previously setup PGDATA to /var/lib/postgresql/data/data. It was only after I looked at mounts that I saw this data directory is being mounted to another device. Thankfully, all this is experimental and I didn't really loose any critical data.

I would imagine that PostgreSQL itself doesn't like having PGDATA stored on a Samba/SMB/cifs/"Windows File Sharing" mount.

I didn't see any better choice to make the service persistent. So, I had to use some network storage. Do you think this is an SMB. specific issue? If I were to use NFS or ISCSI, might I face a similar issue?