Categories

ID #1411

How do I configure CephFS?

This article explains how to configure CephFS on a Bright Cluster. This procedure has the following requirements:

 

  • Bright Cluster Manager 8.0 or higher version.
  • Ceph already deployed as described in the Bright Openstack Deployment Manual.
  • There are enough placement groups to create the necessary Ceph pools.

 

The examples about how to mount the filesystem assume that the server that going to have the filesystem mounted on it is in the Ceph public network (this is true for the head node and all compute nodes in a typical installation).

1 - Deploy Ceph Metadata Server

 

Choose which server will be used to run the Ceph Metadata Server. Then run the following commands on the head node.

 

# cmsh

% device use <MDSSERVER>

% roles

% assign cephmds

% commit

 

In the above commands, replace <MDSSERVER> with the name of the node which will run the Ceph Metadata Server.

 

2 - Create Ceph pools

 

Create two Ceph pools. One will be used for the filesystem itself and the other one will be used for the metadata. The pools can be created by running the following commands on the head node:

 

# cmsh

% ceph

% pools

% add cephfs_metadata

% set pgnum 32; set replicas 3

% commit

% add cephfs_data

% set pgnum 32; set replicas 3

% commit

 

The above commands can be modified to specify the desired number of placement groups and replicas.

3 - Create Ceph filesystem

 

Run the following command on the head node:

 

# ceph fs new cephfs cephfs_metadata cephfs_data

 

In the preceding command, cephfs is a name which will identify the created filesystem, while cephfs_metadata and cephfs_data are the pools created in step 2.

 

4 - Mounting the filesystem (optional)

 

The documentation at http://docs.ceph.com explains different ways to mount the filesystem (at http://docs.ceph.com/docs/jewel/cephfs/ for Ceph Jewel, at http://docs.ceph.com/docs/luminous/cephfs/ for Ceph Luminous, and so on).

 

The following examples assume that the filesystem is going to be mounted on the head node and that the client.admin Ceph identity will be used for authentication. Please note that additional steps will be needed for other cases.

 

CephFS can be mounted either by using a FUSE (user space) client or a kernel client. It is recommended to go over the CephFS documentation (and the http://docs.ceph.com/docs/master/cephfs/best-practices section in particular) to decide which one is better suited for the particular use case.

 

 

4.1 - Mounting the filesystem using FUSE

 

In order to mount the Ceph filesystem created in the previous steps, first install the Ceph FUSE client:

 

# yum install ceph-fuse


The head node already has the /etc/ceph/ceph.conf file with the Ceph configuration and the /etc/ceph/ceph.client.admin.keyring with the keyring for the client.admin

identity, so the filesystem can be mounted by running this command (assuming /mnt/cephfs exists):

 

# ceph-fuse /mnt/cephfs/

 

 

This configuration can be made persistent via an entry in /etc/fstab. This requires the version of the util-linux package which was included on RHEL 7.3. So if the following procedure doesn’t work, the administrator should try updating the util-linux package.

 

For Ceph Jewel (the version supported by Bright 8.0) the following entry has to be added to /etc/fstab:

 

id=admin,conf=/etc/ceph/ceph.conf       /mnt/cephfs     fuse.ceph       _netdev,defaults        0 0

 

For newer versions of Ceph, a different syntax can be used:


none    /mnt/cephfs  fuse.ceph ceph.id=admin,ceph.conf=/etc/ceph/ceph.conf,_netdev,defaults  0 0

 

For the preceding examples, the mount points can also be added to /etc/fstab by CMDaemon. Thus, the example for Ceph Jewel can be configured by running the following commands on the head node:

 

# cmsh

% device use master

% fsmounts

% add /mnt/cephfs

% set device id=admin,conf=/etc/ceph/ceph.conf

% set filesystem fuse.ceph

% set mountoptions _netdev,defaults

% commit

 

The example for newer versions of Ceph can be configured by running the following commands on the head node:

 

# cmsh

% device use master

% fsmounts

% add /mnt/cephfs

% set device none

% set filesystem fuse.ceph

% set mountoptions ceph.id=admin,ceph.conf=/etc/ceph/ceph.conf,_netdev,defaults

% commit

 


4.2 - Mounting the filesystem using FUSE (another method, using systemd unit)

  

Starting with Ceph Luminous (the version supported in Bright 8.1) and when using RHEL 7 or CentOS 7, the ceph-fuse package includes a systemd unit that can be used to mount the filesystem. For example, to mount the filesystem on /mnt/cephfs, the following command can be run (assuming /mnt/cephfs exists):


# systemctl start ceph-fuse@/mnt/cephfs


It is possible to make this configuration persistent by running the following command:


# systemctl enable ceph-fuse@-mnt-cephfs


In the above command, please note that even though the mount point is “/mnt/cephfs” the "/" in the path has to be substituted by a "-".

 

4.3 - Mounting the filesystem using the kernel driver

 

Run the following command on the head node to generate a file with the client.admin secret key:

 

# ceph auth get-key client.admin > /etc/ceph/ceph.admin.secret

 

Then run the following command (assuming /mnt/cephfs exists):

 

# mount -t ceph <monitor>:6789:/ /mnt/cephfs -o name=admin,secretfile=/etc/ceph/ceph.admin.secret

 

In the above command, substitute <monitor> for the IP address of one of the Ceph monitors.

 

 

To make this configuration persistent, the following line has to be added to /etc/fstab:


<monitor>:6789                     /mnt/cephfs          ceph         name=admin,secretfile=/etc/ceph/ceph.admin.secret 0 0

If the client node is a node managed by CMDaemon, like the head node, then the mountpoint can be added by running the following commands on the head node:

 

# cmsh

% device use master

% fsmounts

% add /mnt/cephfs

% set device <monitor>:6789

% set filesystem ceph

% set mountoptions name=admin,secretfile=/etc/ceph/ceph.admin.secret

% commit

Tags: -

Related entries:

You cannot comment on this entry