JuiceFS CSI Driver Strange Behavior

[fedestritt3]

The JuiceFS CSI driver is ignoring the volumeHandle parameter in PersistentVolumes and mounting all pods to the same JuiceFS volume (influxdb), regardless of the different volume configurations specified in their respective secrets.

Environment

• Storage Backend: S3 Object Storage(http://s3-object-storage:8333/juicefs)
• Metadata Engine: etcd cluster ([etcd://<etcd-node-1>:2379,<etcd-node-2>:2379,<etcd-node-3>:2379]/jfs)
• Kubernetes: Multiple namespaces with separate PVs and secrets

Expected Behavior

Each pod should mount its own JuiceFS volume as specified by:

1. The volumeHandle in the PersistentVolume
2. The name field in the corresponding secret

Actual Behavior

All pods mount the same JuiceFS volume (influxdb), even when they have:

• Different PersistentVolumes with unique volumeHandle values
• Different secrets with unique name fields
• Different namespaces

Evidence

Pod Mount Points

# Pod 1 (namespace: influxdb)
$ kubectl exec -n influxdb influxdb-influxdb2-0 -- cat /proc/mounts | grep juicefs
JuiceFS:influxdb /var/lib/influxdb2 fuse.juicefs rw,relatime,user_id=0,group_id=0,default_permissions,allow_other 0 0

# Pod 2 (namespace: influxdb-recp-time)
$ kubectl exec -n influxdb-recp-time influxdb-recp-time-influxdb2-0 -- cat /proc/mounts | grep juicefs
JuiceFS:influxdb /var/lib/influxdb2 fuse.juicefs rw,relatime,user_id=0,group_id=0,default_permissions,allow_other 0 0

# Pod 3 (namespace: monitoring - Grafana)
$ kubectl exec -n monitoring kube-prometheus-stack-grafana-76bf7544df-rzr65 -- cat /proc/mounts | grep juicefs
JuiceFS:influxdb /var/lib/grafana fuse.juicefs rw,relatime,user_id=0,group_id=0,default_permissions,allow_other 0 0

All three pods mount JuiceFS:influxdb despite having different configurations.

PersistentVolume Configuration

$ kubectl get pv influxdb -o jsonpath='{.spec.csi.volumeHandle}'
influxdb

$ kubectl get pv influxdb-recp-time -o jsonpath='{.spec.csi.volumeHandle}'
influxdb-recp-time

$ kubectl get pv grafana -o jsonpath='{.spec.csi.volumeHandle}'
grafana

volumeHandle values are unique and correct.

Secret Configuration

Each pod references a different secret with unique name fields:

Pod 1 Secret (namespace: influxdb)

name: influxdb
metaurl: etcd://<etcd-cluster>/jfs
bucket: http://<s3-endpoint>/juicefs
access-key: <redacted>
secret-key: <redacted>
storage: s3

Pod 2 Secret (namespace: influxdb-recp-time)

name: influxdb-recp-time
metaurl: etcd://<etcd-cluster>/jfs
bucket: http://<s3-endpoint>/juicefs
access-key: <redacted>
secret-key: <redacted>
storage: s3

Pod 3 Secret (namespace: monitoring)

name: grafana
metaurl: etcd://<etcd-cluster>/jfs
bucket: http://<s3-endpoint>/juicefs
access-key: <redacted>
secret-key: <redacted>
storage: s3

Do you believe there’s a mistake in my configuration? Could you help me identify and fix it? Thanks for the support