PerlStalker’s SysAdmin Notes

Notes from the life of a systems administrator

Gfs on Gentoo

Originally posted at [2006-03-24 Fri 10:40]

1 Installing

Many of the needed packages are masked. Add the following to /etc/portage/ packages.keywords

sys-fs/gfs ~x86
sys-cluster/gfs-kernel ~x86
sys-cluster/cman ~x86
sys-cluster/cman-headers ~x86
sys-cluster/cman-kernel ~x86
sys-cluster/ccs ~x86
sys-cluster/magma ~x86
sys-cluster/magma-plugins ~x86
sys-cluster/dlm ~x86
sys-cluster/dlm-headers ~x86
sys-cluster/dlm-kernel ~x86
sys-cluster/iddev ~x86
sys-cluster/gfs-headers ~x86
sys-cluster/fence ~x86
sys-fs/clvm ~x86
sys-cluster/gnbd ~x86
sys-cluster/gnbd-headers ~x86
sys-cluster/gnbd-kernel ~x86

Now you can emerge the packages. All of these packages are needed for the client machines except for sys-fs/clvm.

emerge gfs cman-kernel gfs-kernel dlm-kernel clvm ntp magma-plugins gnbd gnbd-kernel

Note: Remember to re-emerge the *-kernel packages when you upgrade your kernel.

2 Configuration

3 /etc/cluster/cluster.conf

Edit /etc/cluster/cluster.conf.

<?xml version="1.0"?>
<cluster name="mailfs" config_version="1">
<cman two_node="1" expected_votes="1"/>

<clusternodes>
  <clusternode name="server1">
   <fence>
    <method name="single">
     <device name="gnbd" nodename="server1"/>
    </method>
   </fence>
  </clusternode>

  <clusternode name="client1">
   <fence>
    <method name="single">
     <device name="gnbd" nodename="server1"/>
    </method>
   </fence>
  </clusternode>

</clusternodes>

<fencedevices>
<fencedevice name="gnbd" agent="fence_gnbd"/>
</fencedevices>

</cluster>

Note: The name for each clusternode must be a resolvable hostname. Also note that the config above is for a two node cluster. If you have more than two nodes, you can remove the cman tag.

3.1 /etc/ntp.conf

The defaults should work for most people. You may want to set the server option if you have your own ntp server.

4 Setup LVM

The setup for LVM will vary greatly depending on your setup, storage needs and if you are using md to do software RAID. The basic steps are:

  1. Use pvcreate to mark each physical volume, i.e. partition, disk, etc., as an device suitable for use by LVM. If you are using md, you must create those devices first and then use pvcreate to make them ready for LVM.
  2. Use vgcreate to create the volume group, i.e. virtual device, for use with each physical volume created above. You can add additional physical volumes to the group later with vgextend.
  3. Use lvcreate to create the a logical volume within a volume group. The logical volume can use some or all of the volume group for use as usable devices. These logical volumes can be mounted and used as a regular block device, i.e. it looks like a regular disk. The device will be named /var /vgname/lvname. On the system I use for the email cluser at amigo.net, I named the volume group vg01 and the logical volume mail so the device is named /var/vg01/mail.

When you’re done, add clvmd to your startup.

rc-update add clvmd start

4.1 Example Configuration Using md

Here is an example of how you might setup a logical volume using md devices. This is not meant to be an all-inclusive explanation of how md or LVM work. It’s simply presented as an example of how things can work.

4.1.1 /etc/raidtab

/etc/raidtab is where you configure your RAID setup. I setup my drives in a RAID10 (or 1+0, take your pick) configuration. This is not the only way to setup a RAID10. I’ve heard rummors of an option to set raid-level to 10 and be able to do the whole config one block but I have not seen docs on it. You can also use LVM to do the striping instead of md. That gives you a bit more flexibility by letting you add pairs to the RAID easily but, from what I’ve see in places around the Web, the performance isn’t as good.

# Mirrors
raiddev                 /dev/md0
raid-level              1
nr-raid-disks           2
chunk-size              32
persistent-superblock   1
device                  /dev/sdb1
raid-disk               0
device                  /dev/sdc1
raid-disk               1

raiddev                 /dev/md1
raid-level              1
nr-raid-disks           2
chunk-size              32
persistent-superblock   1
device                  /dev/sdd1
raid-disk               0
device                  /dev/sde1
raid-disk               1

raiddev                 /dev/md2
raid-level              1
nr-raid-disks           2
chunk-size              32
persistent-superblock   1
device                  /dev/sdf1
raid-disk               0
device                  /dev/sdg1
raid-disk               1

raiddev                 /dev/md3
raid-level              1
nr-raid-disks           2
chunk-size              32
persistent-superblock   1
device                  /dev/sdh1
raid-disk               0
device                  /dev/sdi1
raid-disk               1

## Stripe (Main RAID-10)
raiddev                 /dev/md4
raid-level              0
nr-raid-disks           4
chunk-size              32
persistent-superblock   1
device                  /dev/md0
raid-disk               0
device                  /dev/md1
raid-disk               1
device                  /dev/md2
raid-disk               2
device                  /dev/md3
raid-disk               3

If you have installed the raidtools package, you can use mkraid to fire up the RAID and let it get building. Note: The rebuild can take a while. You can check on the status by running cat /proc/mdstat.

mkraid /dev/md0 /dev/md1 /dev/md3
mkraid /dev/md4

Once the RAID is done building, you can add it as a physical volume for LVM

pvcreate /dev/md4

and create the volume and logical groups.

vgcreate vg01 /dev/md4
lvcreate -L 20G -n mail vg01

That will create a 20GB logical volume. You can create a logical volume that uses all of the space on the volume group but it takes an extra to get the number of physical entries on the volume.

$ vgdisplay vg01 | grep PE
  PE Size               4.00 MB
  Total PE              17393
  Alloc PE / Size       0 / 0 GB
  Free  PE / Size       17393 / 67.94 GB

Then use the first number on the Free PE / Size line when you run lvcreate.

lvcreate -l 17393 -n mail vg01

Note: That’s a lower case l instead of a capital L to let lvcreate know that you are spcifying the size in PEs and not bytes.

5 Create the gfs

gfs_mkfs -p lock_dlm -t mailfs:maildirs -j 4 /dev/vg01/mail

Note: mailfs must match the cluster name from cluster.conf.

6 Enable Services on Boot

for serv in ntp-client ntpd ccsd cman fenced clvmd gfs gnbd-srv gnbd-client;
do
  rc-update add $serv default
done

Note: Clients can omit the gnbd-srv service.

7 gnbd Configuration

Both clients and servers use /etc/gnbdtab. The file uses four columns. The first is the type of record this is; either import or export. The second column, host, is only used by the import type. It specifies the servers to import gnbd mounts from. The last two columns are used by the export type. The first gives a name to the export. It must be unique across your systems because, when imported, it will show up as /dev/gnbd/ name on the client machines. (You can do multipath with gnbd but I’m going to get into that at this time.) The forth, and final column, is the device (or file path) to export.

7.1 Server

7.1.1 /etc/gnbdtab

# type          host            name            device
export          NOTUSED         maildirs        /dev/vg01/mail

Note: It’s possible for the server to import devices that it is exporting. Add the client line (below) but set the host to localhost or 127.0.0.1. You will also need to make a small change to /etc/init.d/gnbd-client because Gentoo tries to start the client before the server. Edit /etc/init.d/gnbd-client and add

need gnbd-srv

to depend(). The entire block should look something like this:

depend() {
        use dns logger
        need net
        need fenced
        need gnbd-srv
}

7.2 Clients

7.2.1 /etc/gnbdtab

# type          host            name            device
import          gndb_server       NOTUSED         NOTUSED

7.2.2 /etc/fstab

/dev/gnbd/maildirs    /var/mail/virtual    gfs        noatime    0 0

Note: maildirs (in /dev/gnbd/maildirs) is the name of the gnbd export from above.

Comments