This article is now too long. I need to split it up.
Below image shows the key difference between traditional virtualization and containers,
Why Use It?
LXC is in between a VM and a pure application container. It will take up more resources than a pure container, but unlike a VM use as much memory as needed. It will provide a full kernel and OS (albeit it must be Linux). Unlike a VM, LXC container will have access to the resources and speed of the hardware without needing to emulate hardware. LXC is wrapped by Warden and the foundation for Cloud Foundry. Even if you are just interested in Docker or Cloud and not yet ready to give up your ssh console, start here. Learning LXC, you'll also really understand how Cloud Foundry, Docker and generally Cloud concepts actually works. There is also value in then using LXD to learn how to build your own bare metal, orchestrated Cloud infrastructure using technologies like Kubernetes.
When an container is created (last updated with Ubuntu 16.04.1) it is tiny,
|Disk Space||337 MB||That's the entire operating system because it can mount shared core kernel files from the Host. Better yet as Read-Only|
|Memory (externally viewing Container)||12.96 MiB ~ 1.7 MB||Using lxc-info which excludes the shared kernal memory|
|Memory (Inside Container)||5.6 MB||Smaller hence because the container leverages the host OS. Note, need to speak to LXC guys to understandy why external versus internal difference.|
|Processes (Inside Container)||15||Container leverages the host OS.|
There are some kernel functionality that cannot be used inside of a container,
- Mounting External File Systems - Instead Mount in the host OS and then share to the container
LXC is changing pretty fast, so have an idea of the version installed with your distribution,
Note, this article was last tested with LXC 1.x.
Best to use your own Linux machine, but if in a pinch, you can try it out on the Linux Containers website for free.
Create Regular Privileged Container
The simplest to implement container is a privileged container. In other words, the user and process running the container is root. This is not as bad as it seems from a security perspective and easiest to manage (so far). Tin to talk to Dickson to find out if Solaris (which is more mature with container technology) started using unprivileged and if so will look into that next for Linux.
Container Derived from Image File Downloaded to Base OS
This draws the files from the existing image matching downloaded from the Base OS to make your container. Note the image file must be compatible (for now) with the Base OS. I observed it may also be looking at details of your Host OS as the process creates accounts in the image that matched Host OS. The image folder was created/downloaded during the lxc setup and resides in /var/cache/lxc/trusty/rootfs-amd64. The template file to create the container resides in /usr/share/lxc/templates/,
The --name specified will also be the container's name.
The rootfs-amd64 folder is only 384MB and was created when lxc was installed. The OS version will match your Host OS.
You may also download additional LXC image templates which may be different distributions (for example, OpenSUSE, Cent OS, Fedora and Arch Linux) and various versions of your Host OS.
You can see your new container,
Sparse Container from Host OS
Not sure if this functionality exists yet. Solaris has a concept of a sparse root zone which I would call here sparse container. The idea to reduce scope overhead so that aside from basics (home, opt, temp) the rest of the file system is actually mounted as read-only from the Host OS. In this model we have tighter security with minimal packages.
You will need to use sudo or root to view the directories. I use root,
There are 3 key files,
See lxc.conf(5) for additional configuration options on the above files.
By default, LXC creates a private network namespace for each container using a DHCP server (dnsmasq), a NAT server (package name?) and configures IP Tables masquerade entries for outbound network access.
In other words, containers exist within a private network, may see each other, pull network data to whatever the host can access, but nothing outside aside from the host itself will see these servers. A good analogy is your home network behind 1 public IP. Your desktops can see each other, pull data from the Internet, but the Internet cannot see your systems.
Of course, you may expose containers to the hosts's network.
Enabling LXC DNS Resolution
By default, the containers can only see each other by IP addresses. DNS resolutions by default is not enabled.
I'm not going to show how to fix this because we will get this out of the box when graduating to using LXD.
Static LXC Assigned IP
When working with servers, without orchestration, it is best to use static IPs. This should be your first step if seriously using containers for production use. By default, LXC assigns an internal set of IPs within the range of 10.0.3.2 to 10.0.3.254 which is defined in /etc/default/lxc-net.
Change the range and enable reservations in /etc/default/lxc-net. Start by backing up per Bonsai Framework convention,
First change the range (LXC_DHCP_RANGE) and the maximum lease # (LXC_DHCP_MAX) to make available 10.0.3.2 to 10.0.3.99 for static assignment,
There are two keys to change, LXC_DHCP_RANGE and LXC_DHCP_MAX,
Somebody write a sed edit command here so fast and less room for errors.
Second, enable Reservations - Remove the # symble to enable "LXC_DHCP_CONFILE=/etc/lxc/dnsmasq.conf".
Create and edit the /etc/lxc/dnsmasq.conf,
Specify static IPs per container name (you can also use the container's mac address in situations of multiple network interfaces) (I have not tried or tested yet). To avoid conflicts, make sure the static IP addresses are outside of the range specified in /etc/default/lxc-net. In case of multiple ip addresses look at LXC Advanced Networking.
You need to shutdown your containers and then restart the Host OS or flush the DNS Mask,
So first shutdown all your containers (Aug 2016, there is an issue that lxc-net does not restart unless all containers are shut off). Then restart lxc-net,
The other option is to shutdown all containers and restart the host operating system.
Make sure everything restarted fine without errors. I had run into forgetting to change LXC_DHCP_MAX which resulted in lxc-dhcp not starting,
The nice thing about this approach is that the servers within the private network can see each other by assigned hostname. Try pinging by hostname and it will respond correctly. In the case two network cards though, there is some routing work that I have yet to figure out.
You may alternatively set the container OS itself to use static using the interfaces file.
Making Containers Available on the Network
Containers may also be made available on the larger network which is covered in 5.1 LXC Advanced Networking - Exposing Containers to the Network.
This starts the container as a process,
For troubleshooting, you may omit --daemon which will run the containers as an actual program where you can see the container booting up.
Replicate loading console,
Defaults for the template image are,
user = ubuntu
pass = ubuntu
To detach, CTRL+a, followed by ‘q’.
Make to change the password for the default user or delete the default user when you create your own accounts.
There is another way of getting in (good for initial setup without default accounts or emergencies),
lxc-attach is used to execute an arbitrary command inside a container that is already running from outside the container. Because nothing was specified, you end up running bash inside of the container.
Container Management Key Commands
Here are some of the key commands for container management,
There is more to research parked below,
Cloning a Container
One of the most exciting aspects of containers is being able to clone (duplicate).
Shutdown (not entirely sure if needed but I do as principle) your container and clone from the host,
The clone command copies the filesystem and also does some necessary house keeping,
- Update the host name
- Generate new unique mac address for your network card
If you want to just try things you, you will want to look at snapshots.
Clear Out Old Data
If you need to be enterprise class, clear out your new cloned containers log files which will reference the original container hostname. Here is an example search on a relatively new container,
To clear them,
There's more... user history...
Regenerate SSH Host Keys
Next boot up your container, log via the lxc console (is less steps) or ssh with a sudo enabled account and change your SSH host keys,
If you chose to log in using ssh, you will want to log out and update your fingerprint file otherwise you will receive a "REMOTE HOST IDENTIFICATION HAS CHANGED!" error and not be able to ssh in. Different ways of doing this. On a Unix, Linux or Mac OS X operating system,
Mounting a Network Folder
You can't at this moment (March 2016) and I understand this is because it is a shared kernel issue. Solaris does not allow this either if I recall.
Instead use your host to mount your network folder and then share your host folder as described in the next section.
Share Folders with Host
In this example, we want to mount a shared directory from the host system to be accessible to one or more containers,
You may actually use the same directory name and path for both host and container, but I have kept them different for clarity.
While in the lxc container, create your directory and setup any permission you may want to set and shutdown the container,
In the host either create the folder (you may use any user because root is the process accessing), but I use sudo. Note the folder you choose may also be a mount over a network share.
From the host modify the config files. In this example it is /var/lib/lxc/my-container/config and add the following line,
Not sure if the order matters, but I usually place under the lxc.mount entry. The bind.rw is a reference to read and write which you may adjust accordingly for example bind.ro would be read only.
I believe multiple containers may use the same host folder but you might start confusing yourself so suggest using your directory structure to keep track.
If your host is mounting another folder (ie a network share). You must mount in your host first then start your container.
Moving Container to New Host
... (this will probably be a separate article)
Setting Up Zero Footprint 32-bit Java on LXC 64-bit
By default, the apt-get packages inside of the the containers only use 64-bit architecture,
Then follow-up with installing the proper dependencies, but not sure if I need to install in the host... need to try this out in a VM.
Setup Name Resolution on Host
The containers know each other by name. However, the host has no awareness. This should be resolvable by adjusting the DNS resolution. However, first check if this is an issue that's already resolved in later implementations of Ubuntu (I think 17 has it solved) and also if LXD resolves this automatically.
Good overview of lxc - https://www.flockport.com/lxc-vs-docker/
Initial article reviewing - https://linuxcontainers.org/lxd/introduction/
Official Ubuntu Docs - https://help.ubuntu.com/lts/serverguide/lxc.html#lxc-installation
Changing dnsmasq IP range - http://skyhigh71.github.io/2013/09/29/change_ip_address_range.html
Still to read - appears to be distilled LXC theory - http://the.binbashtheory.com/before-you-start-with-lxc-and-docker/#more-26
How to change ssh host keys - http://www.cyberciti.biz/faq/howto-regenerate-openssh-host-keys/
Clean up log files - http://serverfault.com/questions/185253/delete-all-of-var-log
Clean up log files + more - https://lonesysadmin.net/2013/03/26/preparing-linux-template-vms/
Animated Gif of Traditional Virtualization vs Container - https://leandromoreira.com.br/2016/02/06/from-lxc-to-docker-machine-and-cloudery/
Running X Windows in LXC - http://unix.stackexchange.com/questions/18003/linux-lxc-deploying-images-with-tiniest-possible-x11
Good overview of namespaces and resource groups used to power container technology - https://content.pivotal.io/blog/cloud-foundrys-container-technology-a-garden-overview
LXC and LXD using host bridge - https://insights.ubuntu.com/2015/11/10/converting-eth0-to-br0-and-getting-all-your-lxc-or-lxd-onto-your-lan/