Access Ceph Cluster with librados

If you use Ceph as the underlying storage for JuiceFS, you can either use the standard S3 RESTful API to access the Ceph Object Gateway (RGW), or the more efficient librados to access Ceph storage.

JuiceFS CSI Driver supports adding extra files to Mount Pod, with this mechanism, we can add Ceph client configuration files (obtained from the host path /etc/ceph) into the Mount Pod.

Create JuiceFS volume using Ceph storage

Assume we have an Ceph cluster, and in one node of this cluster, list the content of /etc/ceph:

/etc/ceph/
├── ceph.client.admin.keyring
├── ceph.conf
├── ...
└── ...

With ceph.conf and ceph.client.admin.keyring, we can access Ceph cluster with librados.

On this node, we create an new JuiceFS volume ceph-volume

juicefs format --storage=ceph \
  --bucket=ceph://ceph-test \
  --access-key=ceph \
  --secret-key=client.admin \
  redis://juicefs-redis.example.com/2 \
  ceph-volume

note

Here we assume the Redis URL is redis://juicefs-redis.example.com/2, replace it with your own. For more details about the --access-key and --secret-key of Ceph RADOS, refer How to Setup Object Storage.

View Ceph storage status.

$ ceph osd pool ls
ceph-test

Create secret for Ceph configuration files

The following command creates a YAML file named ceph-conf.yaml on the node where Ceph is located, replacing CEPH_CLUSTER_NAME with the actual name.

$ cat > ceph-conf.yaml <<EOF
apiVersion: v1
kind: Secret
metadata:
  name: ceph-conf
  namespace: kube-system
type: Opaque
data:
  <CEPH_CLUSTER_NAME>.conf: $(base64 -w 0 /etc/ceph/ceph.conf)
  <CEPH_CLUSTER_NAME>.client.admin.keyring: $(base64 -w 0 /etc/ceph/ceph.client.admin.keyring)
EOF

note

The $ at the beginning of line is the shell prompt. base64 command is required, if it isn't present, try to install coreutils package with your OS package manager such as apt-get or yum.

Apply the generated ceph-conf.yaml to the Kubernetes cluster:

$ kubectl apply -f ceph-conf.yaml

$ kubectl -n kube-system describe secret ceph-conf
Name:         ceph-conf
Namespace:    kube-system
Labels:       <none>
Annotations:  <none>

Type:  Opaque

Data
====
ceph.client.admin.keyring:  63 bytes
ceph.conf:                  257 bytes

Create secret for JuiceFS CSI Driver

Create a Secret profile by referring to the following command.

$ cat > juicefs-secret.yaml <<EOF
apiVersion: v1
metadata:
  name: juicefs-secret
  namespace: kube-system
kind: Secret
type: Opaque
data:
  bucket: $(echo -n ceph://ceph-test | base64 -w 0)
  metaurl: $(echo -n redis://juicefs-redis.example.com/2 | base64 -w 0)
  name: $(echo -n ceph-volume | base64 -w 0)
  storage: $(echo -n ceph | base64 -w 0)
  access-key: $(echo -n ceph | base64 -w 0)
  secret-key: $(echo -n client.admin | base64 -w 0)
  configs: $(echo -n '{"ceph-conf": "/etc/ceph"}' | base64 -w 0)
EOF

Apply the configuration.

$ kubectl apply -f juicefs-secret.yaml
secret/juicefs-secret created

To see if the configuration is in effect.

$ kubectl -n kube-system describe secret juicefs-secret
Name:         juicefs-secret
Namespace:    kube-system
Labels:       <none>
Annotations:  <none>

Type:  Opaque

Data
====
access-key:  4 bytes
bucket:      16 bytes
configs:     26 bytes
metaurl:     35 bytes
name:        11 bytes
secret-key:  12 bytes
storage:     4 bytes

As we want the ceph-conf secret we created before to be mounted under /etc/ceph, we construct a JSON string {"ceph-conf": "/etc/ceph"} for the key configs .

Access JuiceFS volume in Kubernetes Pod

Dynamic provisioning

Please refer "Dynamic Provisioning" for how to access JuiceFS using storage class. Replace $(SECRET_NAME) with juicefs-secret and $(SECRET_NAMESPACE) with kube-system.

Static provisioning

Please refer "Static Provisioning" for how to access JuiceFS using static provisioning. Replace name and namespace of nodePublishSecretRef with juicefs-secret and kube-system.

Ceph compatibility

JuiceFS currently supports the following Ceph versions:

JuiceFS Version	Ceph Version
v1.0.x	v12, v13, v14, v15
v1.1.x	v15, v16, v17
v1.2.x	v15, v16, v17

If the Ceph version you are using is not listed above, please refer to the following method to build the image.

How to build Docker image

We use the official ceph/ceph as the base image. If we want to build JuiceFS CSI from Ceph Nautilus:

docker build --build-arg BASE_IMAGE=ceph/ceph:v14 --build-arg JUICEFS_REPO_TAG=v0.16.2 -f docker/ceph.Dockerfile -t juicefs-csi-driver:ceph-nautilus .

The ceph/ceph:v14 image is the official Ceph image for Ceph Nautilus. For other Ceph release base images, see the Ceph image repository .