GNet Solutions » GNU/Linux Administration

Why LXC?¶

So I'm doing web development, and I'm using Debian Wheezy as my development environment, which doesn't have the same version of software than stable, which is what we usually use as target servers. I used to use chroots for this, but I found them painful to manage, especially when running daemons on the same ports than on the host machine.

People like to use virtualization for this, such as VirtualBox (esp. with Vagrant) but I didn't want that since it forces you to start a whole virtual machine every time you want to develop. Running a virtual machine is quite a heavy process and they constantly use resources even if they don't do anything. The big plus of using containers is that you nearly don't have any performance hit, and your container doesn't take any resource to run. Only processes that run will use some resources. Not to mention my initial available computing architecture was PowerPC and many virtualization solutions just will not work due to the lack of virtualization support on the PowerPC processors.

The main drawback of using LXC is that you can only run systems that support the same kernel as your host. Basically it means that you won't be able to create a BSD, Solaris or Windows container on your Debian host. But you'll still be able to create any Linux container, whatever distribution you want to install.

I won't go in detail about what LXC is. If you want more information, check out this Wikipedia page: http://en.wikipedia.org/wiki/LXC.

New VPS Node Checklist¶

When adding a new VPS server, make sure to have the following puppet configuration parameters set:

1. Puppet Node SSL signed by the Puppet Master
2. Puppet Node defined
3. Puppet Classes Debian base, APT client, Kerberos client, LXC vServer, and Puppet client included
4. Unique Node /etc/fstab defined
5. Unique Node /etc/rc.local defined
6. Unique Node /etc/network/interfaces defined
7. Virtual Server template files defined

Cgroups configuration¶

Cgroups are a kernel feature that's needed for lxc to run properly. Start by adding the following to your /etc/fstab file:

cgroup  /sys/fs/cgroup  cgroup  defaults  0   0

Mount it:

sudo mount /sys/fs/cgroup

Installing the packages¶

We'll be using lxc to create and manage the containers, and libvirt to manage the network:

sudo apt-get install lxc libvirt-bin dnsmasq-base

First container creation¶

Do this for each container you want to create. Let's create our first container:

sudo lxc-create -n myfirstcontainer -t debian

Answer to the different questions (just go with the defaults, except when it asks you for a root password if you want to put a custom password).

Once the container is created (this takes some time because it needs to download all the base packages and install a base system), chroot in it and create some TTY devices. I found that the lxc-console command didn't work if I didn't create these devices first:

sudo chroot /var/lib/lxc/myfirstcontainer/rootfs
mknod -m 666 /dev/tty1 c 4 1
mknod -m 666 /dev/tty2 c 4 2
mknod -m 666 /dev/tty3 c 4 3
mknod -m 666 /dev/tty4 c 4 4
mknod -m 666 /dev/tty5 c 4 5
mknod -m 666 /dev/tty6 c 4 6

Exit the chroot (by pressing Ctrl-D or by typing exit) and try to start it:

sudo lxc-start -n myfirstcontainer

You'll get some warnings and permission errors. You can safely ignore them. Now open a new shell and try to open a console to the container:

sudo lxc-console -n myfirstcontainer

You should be brought to a login prompt. Log in as root using the password you provided during the container creation phase. At this point you already have a fully functional container. The next step is to setup an ip address for your container so you can ssh into it, make http requests, etc. To shut down the container cleanly, use the halt command like on your host machine (if it's stuck and the halt command doesn't work, use the stop command):

sudo lxc-stop -n myfirstcontainer

Network configuration¶

We'll be using libvirt to manage the network bridge that we'll create for our containers. Start by enabling the default network configuration that comes with libvirt (you can skip the -c option if you don't have any other virtualization solution installed, such as VirtualBox):

sudo virsh -c lxc:/// net-define /etc/libvirt/qemu/networks/default.xml
sudo virsh -c lxc:/// net-start default
sudo virsh -c lxc:/// net-autostart default

The default network libvirt uses is 192.168.122.0/24. If you want to change it, run:

virsh -c lxc:/// net-edit default

and adapt the settings accordingly.

The last line will allow the interface to be automatically started on boot, so you don't have to do it manually. Now if you run ifconfig you'll see you have a new interface named virbr0.

Now that our bridge is defined, we need to set the network interface of our container. To do this, add the following lines to /var/lib/lxc/myfirstcontainer/config (feel free to replace the ip address to match your bridge configuration):

lxc.network.type = veth
lxc.network.flags = up
lxc.network.link = virbr0
lxc.network.ipv4 = 192.168.122.2/24

Start your machine with the lxc-start command, lxc-console into it, run ifconfig and check that you have an eth0 interface defined. I still don't know why but it looks like the routing won't work by default. To solve this, you can edit your /etc/rc.local file and add this before the exit line (adapt it if your network configuration is different):

ip route add default via 192.168.122.1

or adding a gateway to the container configuration:

lxc.network.ipv4.gateway = 192.168.122.1

Now you can either restart your container or execute this command in a shell, and you should be able to connect to your container from your host machine.

SSH¶

I don't know why but the base ssh install seems to have a problem with the keys generation, making it unusable. To fix it, run lxc-console to go in your container and reinstall ssh:

apt-get install --reinstall openssh-server

Now you should be able to ssh to your container from your host:

ssh root@192.168.122.2

Mount Points¶

You'll probably want to mount some of your host directories in your container. Here's an example to mount the directory /home/you/directory_to_mount to /srv/mountpoint. Add the following lines to /var/lib/lxc/myfirstcontainer/config:

sudo mkdir /var/lib/lxc/myfirstcontainer/rootfs/srv/mountpoint
lxc.mount.entry = /home/you/directory_to_mount /var/lib/lxc/myfirstcontainer/rootfs/srv/mountpoint none defaults,bind 0 0

Automatically start your containers¶

To automatically start your containers at boot, all you have to do is to put symlinks to your containers config files in the /etc/lxc/auto/ directory. For example for your previously created container (it's important that your symlink has the same name as your container):

sudo ln -s /var/lib/lxc/myfirstcontainer/config /etc/lxc/auto/myfirstcontainer