GFS2

From Segfault
Jump to navigation Jump to search

Prerequisites

The kernel needs to support GFS:

$ grep GFS /boot/config*
CONFIG_GFS2_FS=m
CONFIG_GFS2_FS_LOCKING_DLM=y

Userspace needs to be prepared as well:

$ apt-get install gfs2-tools                 # Debian, Ubuntu
$ yum install gfs2-cluster                   # Fedora

We have to setup a real interface on both nodes too and make it resolvable:

$ cat /etc/network/interfaces 
[...]
auto eth0:1
iface eth0:1 inet static
      address 10.0.0.100
      netmask 255.255.255.0
 
$ grep ^10 /etc/hosts
10.0.0.100   node-00
10.0.0.101   node-01
 
$ ifup eth0:1

Installation

$ cat /etc/cluster/cluster.conf 
<source lang=xml>
<?xml version="1.0"?>
<cluster name="myGfs2" config_version="1">
      <clusternodes>
              <clusternode name="node-00" nodeid="1"/>
              <clusternode name="node-01" nodeid="2"/>
      </clusternodes>
</cluster>
</source>

With /etc/cluster/cluster.conf in place on both nodes, we can now start cman (Cluster Manager) on both nodes. Try to start them somewhat simultaneously, so that the quorum gets a timely response:

$ /etc/init.d/cman start
Starting cluster: 
  Checking Network Manager... [  OK  ]
  Global setup... [  OK  ]
  Loading kernel modules... [  OK  ]
  Mounting configfs... [  OK  ]
  Starting cman... [  OK  ]
  Waiting for quorum... [  OK  ]
  Starting fenced... [  OK  ]
  Starting dlm_controld... [  OK  ]
  Starting gfs_controld... [  OK  ]
  Unfencing self... [  OK  ]
  Joining fence domain... [  OK  ]

By now, 4 new daemons should be running:

$ ps -ef
[...]
root      1623     1  0 01:27 ?        00:00:00 corosync -f
root      1673     1  0 01:27 ?        00:00:00 fenced
root      1698     1  0 01:27 ?        00:00:00 dlm_controld
root      1743     1  0 01:27 ?        00:00:00 gfs_controld
$ cman_tool nodes
Node  Sts   Inc   Joined               Name
 1   M     44   2011-03-07 01:27:55  node-00
 2   M     40   2011-03-07 01:27:55  node-01

Usage

We're now able to use our shared disk. On one of the cluster nodes, we do:

$ mkfs.gfs2 -p lock_dlm -t myGfs2:mydisk -j 2 /dev/sdc
  • -p specifies the locking protocol to use, lock_dlm
  • -t specifies the (unique) name in a cluster configuration (ClusterName:FSName)
  • -j specifies the number of journals. One journal for each node is required.

Now, we can mount /dev/sdc on each cluster node:

$ mount -t gfs2 /dev/sdc /mnt/gfs

$ gfs_control -n ls
gfs mountgroups
name          sdc
id            0x94f0cbd2
flags         0x00000008 mounted
change        member 2 joined 1 remove 0 failed 0 seq 7,7
members       1 2 
all nodes
nodeid 1 jid 0 member 1 failed 0 start 1 seq_add 1 seq_rem 0 mount done
nodeid 2 jid 1 member 1 failed 0 start 1 seq_add 7 seq_rem 0 mount done

Of course, /dev/sdc should be some kind of SAN, iSCSI or DRBD blockdevice.

Links