NFS

CephFS namespaces can be exported over NFS protocol using theNFS-Ganesha NFS server.

Requirements

  • Ceph file system (preferably latest stable luminous or higher versions)

  • In the NFS server host machine, ‘libcephfs2’ (preferably latest stableluminous or higher), ‘nfs-ganesha’ and ‘nfs-ganesha-ceph’ packages (latestganesha v2.5 stable or higher versions)

  • NFS-Ganesha server host connected to the Ceph public network

Configuring NFS-Ganesha to export CephFS

NFS-Ganesha provides a File System Abstraction Layer (FSAL) to plug in differentstorage backends. FSAL_CEPHis the plugin FSAL for CephFS. For each NFS-Ganesha export, FSAL_CEPH uses alibcephfs client, user-space CephFS client, to mount the CephFS path thatNFS-Ganesha exports.

Setting up NFS-Ganesha with CephFS, involves setting up NFS-Ganesha’sconfiguration file, and also setting up a Ceph configuration file and cephxaccess credentials for the Ceph clients created by NFS-Ganesha to accessCephFS.

NFS-Ganesha configuration

A sample ganesha.conf configured with FSAL_CEPH can be found here,https://github.com/nfs-ganesha/nfs-ganesha/blob/next/src/config_samples/ceph.conf.It is suitable for a standalone NFS-Ganesha server, or an active/passiveconfiguration of NFS-Ganesha servers managed by some sort of clusteringsoftware (e.g., Pacemaker). Important details about the options areadded as comments in the sample conf. There are options to do the following:

  • minimize Ganesha caching wherever possible since the libcephfs clients(of FSAL_CEPH) also cache aggressively

  • read from Ganesha config files stored in RADOS objects

  • store client recovery data in RADOS OMAP key-value interface

  • mandate NFSv4.1+ access

  • enable read delegations (need at least v13.0.1 ‘libcephfs2’ packageand v2.6.0 stable ‘nfs-ganesha’ and ‘nfs-ganesha-ceph’ packages)

Configuration for libcephfs clients

Required ceph.conf for libcephfs clients includes:

  • a [client] section with mon_host option set to let the clients connectto the Ceph cluster’s monitors, usually generated via ceph config generate-minimal-conf, e.g.,
  1. [global]
  2.         mon host = [v2:192.168.1.7:3300,v1:192.168.1.7:6789], [v2:192.168.1.8:3300,v1:192.168.1.8:6789], [v2:192.168.1.9:3300,v1:192.168.1.9:6789]

Mount using NFSv4 clients

It is preferred to mount the NFS-Ganesha exports using NFSv4.1+ protocolsto get the benefit of sessions.

Conventions for mounting NFS resources are platform-specific. Thefollowing conventions work on Linux and some Unix platforms:

From the command line:

  1. mount -t nfs -o nfsvers=4.1,proto=tcp <ganesha-host-name>:<ganesha-pseudo-path> <mount-point>

Current limitations

  • Per running ganesha daemon, FSAL_CEPH can only export one Ceph file systemalthough multiple directories in a Ceph file system may be exported.

Exporting over NFS clusters deployed using rook

This tutorial assumes you have a kubernetes cluster deployed. If not minikube can be usedto setup a single node cluster. In this tutorial minikube is used.

Note

Configuration of this tutorial should not be used in a a realproduction cluster. For the purpose of simplification, the securityaspects of Ceph are overlooked in this setup.

Rook Setup And Cluster Deployment

Clone the rook repository:

  1. git clone https://github.com/rook/rook.git

Deploy the rook operator:

  1. cd cluster/examples/kubernetes/ceph
  2. kubectl create -f common.yaml
  3. kubectl create -f operator.yaml

Note

Nautilus release or latest Ceph image should be used.

Before proceding check if the pods are running:

  1. kubectl -n rook-ceph get pod

Note

For troubleshooting on any pod use:

  1. kubectl describe -n rook-ceph pod <pod-name>

If using minikube cluster change the dataDirHostPath to /data/rook incluster-test.yaml file. This is to make sure data persists across reboots.

Deploy the ceph cluster:

  1. kubectl create -f cluster-test.yaml

To interact with Ceph Daemons, let’s deploy toolbox:

  1. kubectl create -f ./toolbox.yaml

Exec into the rook-ceph-tools pod:

  1. kubectl -n rook-ceph exec -it $(kubectl -n rook-ceph get pod -l "app=rook-ceph-tools" -o jsonpath='{.items[0].metadata.name}') bash

Check if you have one Ceph monitor, manager, OSD running and cluster is healthy:

  1. [root@minikube /]# ceph -s
  2.    cluster:
  3.         id:     3a30f44c-a9ce-4c26-9f25-cc6fd23128d0
  4.         health: HEALTH_OK
  5.  
  6.    services:
  7.         mon: 1 daemons, quorum a (age 14m)
  8.         mgr: a(active, since 13m)
  9.         osd: 1 osds: 1 up (since 13m), 1 in (since 13m)
  10.  
  11.    data:
  12.         pools:   0 pools, 0 pgs
  13.         objects: 0 objects, 0 B
  14.         usage:   5.0 GiB used, 11 GiB / 16 GiB avail
  15.         pgs:

Note

Single monitor should never be used in real production deployment. Asit can cause single point of failure.

Create a Ceph File System

Using ceph-mgr volumes module, we will create a ceph file system:

  1. [root@minikube /]# ceph fs volume create myfs

By default replicated size for OSD is 3. Since we are using only one OSD. It can cause error. Let’s fix this up by setting replicated size to 1.:

  1. [root@minikube /]# ceph osd pool set cephfs.myfs.meta size 1
  2. [root@minikube /]# ceph osd pool set cephfs.myfs.data size 1

Note

The replicated size should never be less than 3 in real production deployment.

Check Cluster status again:

  1. [root@minikube /]# ceph -s
  2.   cluster:
  3.     id:     3a30f44c-a9ce-4c26-9f25-cc6fd23128d0
  4.     health: HEALTH_OK
  5.  
  6.   services:
  7.     mon: 1 daemons, quorum a (age 27m)
  8.     mgr: a(active, since 27m)
  9.     mds: myfs:1 {0=myfs-a=up:active} 1 up:standby-replay
  10.     osd: 1 osds: 1 up (since 56m), 1 in (since 56m)
  11.  
  12.   data:
  13.     pools:   2 pools, 24 pgs
  14.     objects: 22 objects, 2.2 KiB
  15.     usage:   5.1 GiB used, 11 GiB / 16 GiB avail
  16.     pgs:     24 active+clean
  17.  
  18.   io:
  19.     client:   639 B/s rd, 1 op/s rd, 0 op/s wr

Create a NFS-Ganesha Server Cluster

Add Storage for NFS-Ganesha Servers to prevent recovery conflicts:

  1. [root@minikube /]# ceph osd pool create nfs-ganesha 64
  2. pool 'nfs-ganesha' created
  3. [root@minikube /]# ceph osd pool set nfs-ganesha size 1
  4. [root@minikube /]# ceph orchestrator nfs add mynfs nfs-ganesha ganesha

Here we have created a NFS-Ganesha cluster called “mynfs” in “ganesha”namespace with “nfs-ganesha” OSD pool.

Scale out NFS-Ganesha cluster:

  1. [root@minikube /]# ceph orchestrator nfs update mynfs 2

Configure NFS-Ganesha Exports

Initially rook creates ClusterIP service for the dashboard. With this servicetype, only the pods in same kubernetes cluster can access it.

Expose Ceph Dashboard port:

  1. kubectl patch service -n rook-ceph -p '{"spec":{"type": "NodePort"}}' rook-ceph-mgr-dashboard
  2. kubectl get service -n rook-ceph rook-ceph-mgr-dashboard
  3. NAME                      TYPE       CLUSTER-IP       EXTERNAL-IP   PORT(S)          AGE
  4. rook-ceph-mgr-dashboard   NodePort   10.108.183.148   <none>        8443:31727/TCP   117m

This makes the dashboard reachable outside kubernetes cluster and the servicetype is changed to NodePort service.

Create JSON file for dashboard:

  1. $ cat ~/export.json
  2. {
  3.       "cluster_id": "mynfs",
  4.       "path": "/",
  5.       "fsal": {"name": "CEPH", "user_id":"admin", "fs_name": "myfs", "sec_label_xattr": null},
  6.       "pseudo": "/cephfs",
  7.       "tag": null,
  8.       "access_type": "RW",
  9.       "squash": "no_root_squash",
  10.       "protocols": [4],
  11.       "transports": ["TCP"],
  12.       "security_label": true,
  13.       "daemons": ["mynfs.a", "mynfs.b"],
  14.       "clients": []
  15. }

Note

Don’t use this JSON file for real production deployment. As here theganesha servers are given client-admin access rights.

We need to download and run this scriptto pass the JSON file contents. Dashboard creates NFS-Ganesha export filebased on this JSON file.:

  1. ./run-backend-rook-api-request.sh POST /api/nfs-ganesha/export "$(cat <json-file-path>)"

Expose the NFS Servers:

  1. kubectl patch service -n rook-ceph -p '{"spec":{"type": "NodePort"}}' rook-ceph-nfs-mynfs-a
  2. kubectl patch service -n rook-ceph -p '{"spec":{"type": "NodePort"}}' rook-ceph-nfs-mynfs-b
  3. kubectl get services -n rook-ceph rook-ceph-nfs-mynfs-a rook-ceph-nfs-mynfs-b
  4. NAME                    TYPE       CLUSTER-IP       EXTERNAL-IP   PORT(S)          AGE
  5. rook-ceph-nfs-mynfs-a   NodePort   10.101.186.111   <none>        2049:31013/TCP   72m
  6. rook-ceph-nfs-mynfs-b   NodePort   10.99.216.92     <none>        2049:31587/TCP   63m

Note

Ports are chosen at random by Kubernetes from a certain range.Specific port number can be added to nodePort field in spec.

Testing access to NFS Servers

Open a root shell on the host and mount one of the NFS servers:

  1. mkdir -p /mnt/rook
  2. mount -t nfs -o port=31013 $(minikube ip):/cephfs /mnt/rook

Normal file operations can be performed on /mnt/rook if the mount is successful.

Note

If minikube is used then VM host is the only client for the servers.In a real kubernetes cluster, multiple hosts can be used as clients,only when kubernetes cluster node IP addresses are accessible tothem.