CephFS

CephFS provides a common filesystem that can be shared amongst multiple OpenStack VM hosts. Access to the service is granted via requests to cloud@tech.alliancecan.ca.

This is a fairly technical procedure that assumes basic Linux skills for creating/editing files, setting permissions, and creating mount points. For assistance in setting up this service, write to cloud@tech.alliancecan.ca.

Procedure

Request access to shares

If you do not already have a quota for the service, you will need to request this through cloud@tech.alliancecan.ca. In your request please provide the following:

• OpenStack project name
• amount of quota required in GB
• number of shares required

OpenStack configuration: Create a CephFS share

Create the share.

In Project --> Share --> Shares, click on +Create Share.

Share Name = enter a name that identifies your project (e.g. project-name-shareName)

Share Protocol = CephFS

Size = size you need for this share

Share Type = cephfs

Availability Zone = nova

Do not check Make visible for all, otherwise the share will be accessible by all users in all projects.

Click on the Create button.

Configuration of CephFS on Horizon GUI

Create an access rule to generate access key.

In Project --> Share --> Shares --> Actions column, select Manage Rules from the drop-down menu.

Click on the +Add Rule button (right of the page).

Access Type = cephx

Access Level = select read-write or read-only (you can create multiple rules for either access level if required)

Access To = select a key name that describes the key, this name is important, it will be used in the cephfs clien config on the VM, we will use MyCephFS-RW on this page.

Properly configured CephFS

Note the share details which you will need later.

In Project --> Share --> Shares, click on the name of the share.

In the Share Overview, note the three elements circled in red in the "Properly configured" image: Path, which will be used in the mount command on the VM, the Access Rules, which will be the client name and the Access Key that will let the VM's client connect.

Attach the CephFS network to your VM

On Arbutus

On Arbutus the cephFS network is already exposed to your VM, there is nothing to do here, go to VM configuration section.

On SD4H/Juno

On SD4H/Juno, you need to explicitly attach the cephFS network to the VM.

With the Web Gui

For each VM you need to attach, select Instance --> Action --> Attach interface select the CephFS-Network, leave the Fixed IP Address box empty.

With the Openstack client

List the servers and select the id of the server you need to attach to the CephFS

$ openstack  server list 
+--------------------------------------+--------------+--------+-------------------------------------------+--------------------------+----------+
| ID                                   | Name         | Status | Networks                                  | Image                    | Flavor   |
+--------------------------------------+--------------+--------+-------------------------------------------+--------------------------+----------+
| 1b2a3c21-c1b4-42b8-9016-d96fc8406e04 | prune-dtn1   | ACTIVE | test_network=172.16.1.86, 198.168.189.3   | N/A (booted from volume) | ha4-15gb |
| 0c6df8ea-9d6a-43a9-8f8b-85eb64ca882b | prune-mgmt1  | ACTIVE | test_network=172.16.1.64                  | N/A (booted from volume) | ha4-15gb |
| 2b7ebdfa-ee58-4919-bd12-647a382ec9f6 | prune-login1 | ACTIVE | test_network=172.16.1.111, 198.168.189.82 | N/A (booted from volume) | ha4-15gb |
+--------------------------------------+--------------+--------+----------------------------------------------+--------------------------+----------+

Select the ID of the VM you want to attach, will pick the first one here and run

$ openstack  server add network 1b2a3c21-c1b4-42b8-9016-d96fc8406e04 CephFS-Network
$ openstack  server list 
+--------------------------------------+--------------+--------+---------------------------------------------------------------------+--------------------------+----------+
| ID                                   | Name         | Status | Networks                                                            | Image                    | Flavor   |
+--------------------------------------+--------------+--------+---------------------------------------------------------------------+--------------------------+----------+
| 1b2a3c21-c1b4-42b8-9016-d96fc8406e04 | prune-dtn1   | ACTIVE | CephFS-Network=10.65.20.71; test_network=172.16.1.86, 198.168.189.3 | N/A (booted from volume) | ha4-15gb |
| 0c6df8ea-9d6a-43a9-8f8b-85eb64ca882b | prune-mgmt1  | ACTIVE | test_network=172.16.1.64                                            | N/A (booted from volume) | ha4-15gb |
| 2b7ebdfa-ee58-4919-bd12-647a382ec9f6 | prune-login1 | ACTIVE | test_network=172.16.1.111, 198.168.189.82                           | N/A (booted from volume) | ha4-15gb |
+--------------------------------------+--------------+--------+------------------------------------------------------------------------+--------------------------+----------+

We can see that the CephFS network is attached to the first VM.

VM configuration: install and configure CephFS client

Required packages for the Red Hat family (RHEL, CentOS, Fedora, Rocky, Alma )

Check the available releases at https://download.ceph.com/ and look for recent rpm-* directories. As of July 2024, quincy is the latest stable release. The compatible distributions (distros) are listed at https://download.ceph.com/rpm-quincy/. Here we will show the full installation for el8.

Install relevant repositories for access to ceph client packages

File : /etc/yum.repos.d/ceph.repo

[Ceph]
name=Ceph packages for $basearch
baseurl=http://download.ceph.com/rpm-quincy/el8/$basearch
enabled=1
gpgcheck=1
type=rpm-md
gpgkey=https://download.ceph.com/keys/release.asc

[Ceph-noarch]
name=Ceph noarch packages
baseurl=http://download.ceph.com/rpm-quincy/el8/noarch
enabled=1
gpgcheck=1
type=rpm-md
gpgkey=https://download.ceph.com/keys/release.asc

[ceph-source]
name=Ceph source packages
baseurl=http://download.ceph.com/rpm-quincy/el8/SRPMS
enabled=1
gpgcheck=1
type=rpm-md
gpgkey=https://download.ceph.com/keys/release.asc

The epel repo also needs to be in place

sudo dnf install epel-release

You can now install the ceph lib, cephfs client and other dependencies:

sudo dnf install -y libcephfs2 python3-cephfs ceph-common python3-ceph-argparse

Required packages for the Debian family (Debian, Ubuntu, Mint, etc.)

You can get the repository once you have figured out your distro {codename} with lsb_release -sc

sudo apt-add-repository 'deb https://download.ceph.com/debian-quincy/ {codename} main'

You can now install the ceph lib, cephfs client and other dependencies:

 sudo apt-get install -y libcephfs2 python3-cephfs ceph-common python3-ceph-argparse

Configure ceph client

Once the client is installed, you can create a ceph.conf file. Note the different mon host for the different cloud.

ArbutusSD4H/Juno

File : /etc/ceph/ceph.conf

[global]
admin socket = /var/run/ceph/$cluster-$name-$pid.asok
client reconnect stale = true
debug client = 0/2
fuse big writes = true
mon host = 10.30.201.3:6789,10.30.202.3:6789,10.30.203.3:6789
[client]
quota = true

File : /etc/ceph/ceph.conf

[global]
admin socket = /var/run/ceph/$cluster-$name-$pid.asok
client reconnect stale = true
debug client = 0/2
fuse big writes = true
mon host = 10.65.0.10:6789,10.65.0.12:6789,10.65.0.11:6789
[client]
quota = true

You can find the monitor information in the share details Path field that will be used to mount the volume. If the value of the web page is different than what is seen here, it means that the wiki page is out of date.

You also need to put your client name and secret in the ceph.keyring file

File : /etc/ceph/ceph.keyring

[client.MyCephFS-RW]
    key = <access Key>

Again, the access key and client name (here MyCephFS-RW) are found under access rules on your project web page. Look for Project --> Share --> Shares, then click on the name of the share.

Retrieve the connection information from the share page for your connection

Open up the share details by clicking the name of the share in the Shares page.

Copy the entire path of the share for mounting the filesystem.

Mount the filesystem

Create a mount point directory somewhere in your host (/cephfs, is used here)

 mkdir /cephfs

Via kernel mount using the ceph driver. You can do a permanent mount by adding the following in the VM fstab

ArbutusSD4H/Juno

File : /etc/fstab

:/volumes/_nogroup/f6cb8f06-f0a4-4b88-b261-f8bd6b03582c /cephfs/ ceph name=MyCephFS-RW 0  2

File : /etc/fstab

:/volumes/_nogroup/f6cb8f06-f0a4-4b88-b261-f8bd6b03582c /cephfs/ ceph name=MyCephFS-RW,mds_namespace=cephfs_4_2,x-systemd.device-timeout=30,x-systemd.mount-timeout=30,noatime,_netdev,rw 0  2

Notice the non-standard : before the device path. It is not a typo! The mount options are different on different systems. The namespace option is required for SD4H/Juno, while other options are performance tweaks.

It can also be done from the command line

ArbutusSD4H/Juno

sudo mount -t ceph :/volumes/_nogroup/f6cb8f06-f0a4-4b88-b261-f8bd6b03582c /cephfs/ -o name=MyCephFS-RW

sudo mount -t ceph :/volumes/_nogroup/f6cb8f06-f0a4-4b88-b261-f8bd6b03582c /cephfs/ -o name=MyCephFS-RW,mds_namespace=cephfs_4_2,x-systemd.device-timeout=30,x-systemd.mount-timeout=30,noatime,_netdev,rw

Or via ceph-fuse if the file system needs to be mounted in user space

No strange : here.

Install the ceph-fuse lib

sudo dnf install ceph-fuse

Let the fuse mount be accessible in userspace by uncommenting user_allow_other in the fuse.conf file.

File : /etc/fuse.conf

# mount_max = 1000
user_allow_other

You can now mount cephFS in a user’s home:

mkdir ~/my_cephfs
ceph-fuse my_cephfs/ --id=MyCephFS-RW --conf=~/ceph.conf --keyring=~/ceph.keyring   --client-mountpoint=/volumes/_nogroup/f6cb8f06-f0a4-4b88-b261-f8bd6b03582c

Note that the client name is here the --id. The ceph.conf and ceph.keyring content are exactly the same as for the ceph kernel mount.

Notes

A particular share can have more than one user key provisioned for it. This allows a more granular access to the filesystem, for example, if you needed some hosts to only access the filesystem in a read-only capacity. If you have multiple keys for a share, you can add the extra keys to your host and modify the above mounting procedure. This service is not available to hosts outside of the OpenStack cluster.