Gluster, CIFS, ZFS - kind of part 2
By Jon Archer
A while ago I put together a post detailing the installation and configuration of 2 hosts running glusterfs, which was then presented as CIFS based storage.
http://jonarcher.info/2014/06/windows-cifs-fileshares-using-glusterfs-ctdb-highly-available-data/
This post gained a bit of interest through the comments and social networks, one of the comments I got was from John Mark Walker suggesting I look at the samba-gluster vfs method instead of mounting the filesystem using fuse (directly access the volume from samba, instead of mounting then presenting). On top of this I’ve also been looking quite a bit at ZFS, whereas previously I had a Linux RAID as the base filesystem. So here is a slightly different approach to my previous post.
Getting prepared
As before, we’re looking at 2 hosts, virtual in the case of this build but more than likely physical in a real world scenario, either way it’s irrelevant. Both of these hosts are running CentOS 6 minimal installs (I’ll update to 7 at a later date), static IP addresses assigned and DNS entries created. I’ll also be running everything under a root session, if you don’t do the same just prefix the commands with sudo. For purposes of this I have also disabled SELINUX and removed all firewall rules. I will one day leave SELINUX enabled in this configuration but for now lets leave it out of the equation.
In my case these names and addresses are as follows:
arcstor01 - 192.168.1.210
arcstor02 - 192.168.1.211
First off lets get the relevant repositories installed (EPEL, ZFS and Gluster)
yum localinstall –nogpgcheck http://download.fedoraproject.org/pub/epel/6/i386/epel-release-6-8.noarch.rpm yum localinstall –nogpgcheck http://archive.zfsonlinux.org/epel/zfs-release.el6.noarch.rpm curl -o /etc/yum.repos.d/gluster.repo http://download.gluster.org/pub/gluster/glusterfs/LATEST/EPEL.repo/glusterfs-epel.repo curl -o /etc/yum.repos.d/glusterfs-samba-epel.repo http://download.gluster.org/pub/gluster/glusterfs/samba/EPEL.repo/glusterfs-samba-epel.repo
Local filesystem
As previously mentioned, this configuration will be hosted from 2 virtual machines, each will have 3 disks. 1 for the OS, and the other 2 to be used in a ZFS pool.
First off we need to install ZFS itself, once you have the above zfs-release repo installed this can be done with the following command:
yum install kernel-devel zfs
Perform this on both hosts.
We can now create a zfs pool. In my case the disk device names are vdX but they could be sdX,
fdisk -l
can help you identify the device names, whatever they are just replace them in the following commands.
Create a ZFS pool
zpool create -f -m /gluster gluster mirror /dev/vdb /dev/vdc
this command will create a zfs pool mounted at /gluster, without -m /gluster it would mount at /{poolname} while in this case it’s the same I just added the option for clarity. The volume name is gluster, the redundancy level is mirrored which is similar to RAID1, there are a number of raid levels available in ZFS all are best explained here: http://www.zfsbuild.com/2010/05/26/zfs-raid-levels/. The final element to the command is where to host the pool, in our case on /dev/vdb and /dev/vdc. The -f option specified is to force creation of the pool, this is required remove the need to create partitions prior to the creation of the pool.
Running the command
zpool status
Will return the status of the created pool, which if successful should look something similar to:
[root@arcstor01 ~]# zpool status pool: gluster state: ONLINE scan: none requested config:
NAME STATE READ WRITE CKSUM gluster ONLINE 0 0 0 mirror-0 ONLINE 0 0 0 vdb1 ONLINE 0 0 0 vdc1 ONLINE 0 0 0
errors: No known data errors
A quick ls and df will also show us that the /gluster mountpoint is present and the pool is mounted, the df should show the size as being half the sum of both drives in the pool:
[root@arcstor01 ~]# ls / bin boot cgroup dev etc gluster home lib lib64 lost+found media mnt opt proc root sbin selinux srv sys tmp usr var [root@arcstor01 ~]# df -h Filesystem Size Used Avail Use% Mounted on /dev/vda1 15G 1.2G 13G 9% / tmpfs 498M 0 498M 0% /dev/shm gluster 20G 0 20G 0% /gluster
If this is the case, rinse and repeat on host 2. If this is also successful then we now have a resilient base filesystem on which to host our gluster volumes. There is a bucket load more to ZFS and it’s capabilities but it’s way outside the confines of this configuration, well worth looking into though.
Glusterising our pool
So now we have a filesystem, lets make it better. Next up, installing glusterfs, enabling it then preparing the directories, for this part it is pretty much identical to the previous post:
yum install glusterfs-server -y
chkconfig glusterd on
service glusterd start
mkdir -p /gluster/bricks/share/brick1
This needs to be done on both hosts.
Now only on host1 lets make the two nodes friends, create and then start the gluster volume:
# gluster peer probe arcstor02 peer probe: success.
gluster vol create share replica 2 arcstor01:/gluster/bricks/share/brick1 arcstor02:/gluster/bricks/share/brick1
volume create: share: success: please start the volume to access data
gluster vol start share
volume start: share: success
[root@arcstor01 ~]# gluster vol info share
Volume Name: data1 Type: Replicate Volume ID: 73df25d6-1689-430d-9da8-bff8b43d0e8b Status: Started Number of Bricks: 1 x 2 = 2 Transport-type: tcp Bricks: Brick1: arcstor01:/gluster/bricks/share1/brick1 Brick2: arcstor02:/gluster/bricks/share1/brick1
If all goes well above we should have a gluster volume ready to go, this volume will be presented via samba directly. For this configuration a locally available shared area is required, for this we will create another gluster volume to mount locally in which to store lockfiles and shared config files.
mkdir -p /gluster/bricks/config/brick1 gluster vol create config replica 2 arcstor01:/gluster/bricks/config/brick1 arcstor02:/gluster/bricks/config/brick1 gluster vol start config mkdir /opt/samba-config mount -t glusterfs localhost:config /opt/samba-config
The share volume could probably be used by using a different path in the samba config but for simplicity we’ll keep them seperate for now. The mountpoint for /opt/samba-config will need to be added to fstab to ensure it mounts at boot time.
echo “localhost:config /opt/samba-config glusterfs defaults,_netdev 0 0” »/etc/fstab
Should take care of that, remember that needs to be on both hosts.
Samba and CTDB
We now have a highly resilient datastore which could withstand both disk and host downtime, but we need to make that datastore available for consumption and also highly available in the process, for this we will use CTDB, as in the previous post. CTDB is a cluster version of the TDB database which sits under Samba. The majority of this section will be the same as the previous post except for the extra packages and a slightly different config for samba. Lets install the required packages:
yum -y install ctdb samba samba-common samba-winbind-clients samba-client samba-vfs-glusterfs
For the majority of config files we will create them in our shared config volume and symlink them to their expected location. First file we need to create is /etc/sysconfig/ctdb but we will do this as /opt/samba-config/ctdb then link it afterwards
Note: The files which are created in the shared area should be done only on one host, but the linking needs to be done on both.
vi /opt/samba-config/ctdb
CTDB_RECOVERY_LOCK=/opt/samba-config/lockfile #CIFS only CTDB_PUBLIC_ADDRESSES=/etc/ctdb/public_addresses CTDB_MANAGES_SAMBA=yes #CIFS only CTDB_NODES=/etc/ctdb/nodes
We’ll need to remove the existing file in /etc/sysconfig then we can create the symlink
rm /etc/sysconfig/ctdb
ln -s /opt/samba-config/ctdb /etc/sysconfig/ctdb
Although we are using Samba the service we will be using is CTDB which allows for the extra clustering components, we need to stop and disable the samba services and enable the ctdb ones:
service smb stop chkconfig smb off chkconfig ctdb on
With this configuration being a cluster of essentially a single datapoint we should really use a single entry point, for this a 3rd “floating” or virtual IP address is employed, more than one could be used but lets keep this simple - 192.168.1.212. We also need to create a ctdb config file which contains a list of all the nodes in the cluster. Both these files need to be created in the shared location:
vi /opt/samba-config/public_addresses
192.168.1.212/24 eth0
vi /opt/samba-config/nodes
192.168.1.210
192.168.1.211
They both then need to be linked to their expected locations - neither of these exist so don’t need to be removed.
ln -s /opt/samba-config/nodes /etc/ctdb/nodes
ln -s /opt/samba-config/public_addresses /etc/ctdb/public_addresses
The last step is to modify the samba configuration to present the volume via cifs, I seemed to have issues using a linked file for samba so will only use the shared area for storing a copy of the config which can then be copied to both nodes to keep them identical.
cp /etc/samba/smb.conf /opt/samba-config/
Lets edit that file:
vi /opt/samba-config/smb.conf
Near the top add the following options
clustering = yes
idmap backend = tdb2
private dir = /opt/samba-config/
These turn the clustering (CTDB) features on and specify the shared directory where samba will create lockfiles. You can test starting ctdb at this point to ensure all is working, on both hosts:
cp /opt/samba-config/smb.conf /etc/samba/
service ctdb start
It should start OK, then health status of the cluster can be checked with
ctdb status
At this point I was finding that CTDB was not starting correctly, after a little bit of logwatching I found an error in the samba logs suggesting:
Failed to create pipe directory /run/samba/ncalrpc - No such file or directory
Also, to be search engine friendly the CTDB logfile was outputting
50.samba OUTPUT:ERROR: Samba tcp port 445 is not responding
This was a red herring, the port wasn’t responding as the samba part of CTDB wasn’t starting, 50.samba is a script in /etc/ctdb/events/ which actually starts the smb process.
So I created the directory /run/samba and restarted ctdb and the issue seems to have disappeared.
Now we have a started service, we can go ahead and add the configuration for the share. A regular samba share would look something like:
[share] comment = just a share path = /share read only = no guest ok = yes valid users = jon
In the previous post this would have been ideal if our gluster volume was mounted at share, but for this we are removing a layer and want samba to talk directly to gluster rather than via the fuse layer. This is achieved using a VFS object, we installed the samba-vfs-glusterfs package earlier. The configuration is slightly different within the smb.conf file also. Adding the following to our file should enable access to the share volume we created:
[share] comment = gluster vfs share path = / read only = No guest ok = Yes kernel share modes = No vfs objects = glusterfs glusterfs:loglevel = 7 glusterfs:logfile = /var/log/samba/glusterfs-testvol.log glusterfs:volume = share
Notice the glusterfs: options near the bottom, these are specific to the glusterfs vfs object which is called further up (vfs objects = glusterfs). Another point to note is that the path is / this is relative to the volume rather than the filesystem, so a path to /test would be a test directory inside the gluster volume.
We can now reload the samba config, lets restart for completeness (on both nodes)
service ctdb restart
From a cifs client you should now be able to browse to \\192.168.1.212\share (or whatever IP you specified as the floating IP).
All done!
To conclude, here we have created a highly resilient, highly available, very scalable storage solution using some fantastic technologies. We have created a single access method (Cifs on a floating IP) to a datastore which is then stored on multiple hosts, which in turn store upon multiple disks. Talk about redundancy!
Useful links: