.. _application_isolation: Lesson 18: Application Isolation ================================ ============= ============= ============= ========== `Homepage`_ `Content`_ `Slides`_ `Video`_ ============= ============= ============= ========== .. _Homepage: http://devopsbootcamp.osuosl.org .. _Content: http://devopsbootcamp.osuosl.org/application-isolation.html .. _Slides: http://slides.osuosl.org/devopsbootcamp/application-isolation.html .. _Video: .. include:: unfinished.txt .. ifnotslides:: .. contents:: Overview .. ifslides:: Overview -------- - Application Isolation - Virtual Machines - Containers - Pros - Cons - Tools Application Isolation --------------------- .. ifslides:: **The seperation of one program or application stack from others.** **The oldest way to do this is to run your application on a separate computer, but that gets very expensive very quickly.** .. ifnotslides:: Application isolation is the separation of one program or application stack from others. The oldest way to do this is to run your application on a separate computer, but that gets very expensive very quickly. There are two main ways to tackle Application Isolation on one computer: Virtual Machines and Containers. They both achieve similar end results but achieve that end in differnt ways and offer different advantages / disadvantages. Virtual Machines ---------------- .. image:: /static/vm-diagram.png :align: center :alt: Anatomical Diagram of a VM :target: https://commons.wikimedia.org/wiki/File:Hardware_Virtualization_%28copy%29.svg :width: 50% .. nextslide:: .. ifnotslides:: Virtual Machines are programs running on your operating system which emulate the entire computer and operating system. This is good because it completely isolates the programs running on the VM from the host operating system, but that advantage is also a problem. The core disadvantage of a VM is that it is resoruce intensive. There is a lot of overhead in emulating an Operating System. While it offers complete and thorough isolation, that comes at a cost. Here is a demonstration showing the processes inside of a VM versus a host OS :: [vm] # ps aux USER PID %CPU %MEM VSZ RSS TTY STAT START TIME COMMAND root 1 0.0 0.6 110564 3164 ? Ss 2015 11:17 /lib/systemd/systemd --system --deserialize 15 root 2 0.0 0.0 0 0 ? S 2015 0:00 [kthreadd] root 3 0.0 0.0 0 0 ? S 2015 3:55 [ksoftirqd/0] root 5 0.0 0.0 0 0 ? S< 2015 0:00 [kworker/0:0H] [... 120+ more lines ...] :: [host] # ps aux USER PID %CPU %MEM VSZ RSS TTY STAT START TIME COMMAND root 1 0.0 0.1 200328 5208 ? Ss Aug25 0:44 /sbin/init root 2 0.0 0.0 0 0 ? S Aug25 0:00 [kthreadd] root 3 0.0 0.0 0 0 ? S Aug25 0:05 [ksoftirqd/0] root 5 0.0 0.0 0 0 ? S< Aug25 0:00 [kworker/0:0H] [... 240+ more lines ...] OS Emulation ~~~~~~~~~~~~ .. ifnotslides:: Let's talk about that emulation problem. Imagine you are running a computer that has an X86_64 CPU on it. You want to emulate a computer with an ARM5 CPU. The emulated operating system makes system calls to the *'hardware'* it thinks it is running on. The host operating system translates those system calls into *real* system calls on the *actual* hardware. This translation is cost intensive and usually slow, but it's gotten a lot better over the decades. When emulating a separate CPU architecture, optimizations usually have to be made for the emulated OS to be usable. **Note:** When you are emulating an X86_64 VM on an X86_64 piece of hardware, an optimization that can be made in which the host OS passes the system calls directly to the hardware without having to *translate* anything. This is done by a *hypervisor* which enforces certain security protocols so the two operating systems (host and guest) are still isolated, but things go much faster. Containers ---------- .. ifnotslides:: Containers approach application isolation from a different angle. Instead of emulating an entire operating system they use the same host kernel (operating system) to run the guest operating system. This entirely by-passes the emulation problem. Containers isolate applicatiosn using two technologies on Linux: CGroups and Systemd. Together these isolate the guest OS' processes from the host, and limit the guest's resources respectively. Instead of emulating the guest OS containers use the host kernel and *lie* to the guest process and tell it that it's the only application running on that OS. Containers avoid the emulation problem by not using a hypervisor and instead using a combination of technologies to get the same job done. Containers have very become popular recently, but their underlying technologies aren't new. Many application developers and system administrators have begun migrating toward using Containers over VMs as they tend to be more performant, but the industry as a whole is waiting for them to get a bit more battle-tested. Here is the previous demonstration in a container: :: $ ps aux # Lists all processes running on an OS PID USER TIME COMMAND 1 root 0:00 sh 6 root 0:00 ps aux .. ifnotslides:: As you can see, instead of emulating an entire OS (running 100+ processes), the container is told that it's processes (``sh`` and ``ps`` in this case) are the only one in this environment. In theory this prevents a malicious attack from inside the container from invading the host OS. Not a Virtual Machine ~~~~~~~~~~~~~~~~~~~~~ .. ifnotslides:: One key thing to remember is that **a container is not a virtual machine**. It may act like a VM, and isolate like a VM, but they're different and for now they are not 100% interchangeable. .. image:: /static/hypervisor-vs-containers.png :align: center :alt: Diagram of Containers vs Virtual Machines CGroups + Systemd ~~~~~~~~~~~~~~~~~ .. ifnotslides:: CGroups and Systemd are the two technologies that allow Containers to exist. CGruops A linux kernel-level technology that name-spaces processes. It basically allows a host OS to convince a process running on it that it is running in it's own environment. This is what isolates a process from other processes, if they think they're the only thing running they can't tamper with the host OS. Systemd The service manager for most Linux distributions. Systemd starts services like Apache, but can also limit an application's resoruces. This allows you to limit a container from using all of your computer's resources, a common paradigm in VM management. Pros ---- .. TODO: Flesh this section out a bit ======================================= ======================================= **Virtual Machines** **Containers** --------------------------------------- --------------------------------------- Complete process isolation Fast startup 'Battle Tested' Little overhead ======================================= ======================================= Cons ---- ======================================= ======================================= **Virtual Machines** **Containers** --------------------------------------- --------------------------------------- Slightly more overhead. Security concerns. Slow startup. No cross-kernel emulation. Cross-OS emulation. ======================================= ======================================= .. ifnotslides:: Many of these concerns are temporary or negligible. For instance, many organizations that run virtual machines aren't concerned with startup times. And security concerns are a temporary issue in that the industry is already developing solutions for the existing security problems with containers. One thing that we haven't touched on is the **No cross-kernel emulation** for containers. One advantage to Virtual Machines is that a Windows OS can emulate Linux, and vice-verse. Unfortunately it's not the same with containers: A Linux OS can only (natively) run a Linux container, and Windows can only containerize Windows. The industry solution to this problem is to run a container in a small VM. Tools ----- .. ifnotslides:: There are a mutlitude of tools for using and managing VMs and Containers. ======================================= ======================================= **Virtual Machines** **Containers** --------------------------------------- --------------------------------------- VirtualBox Docker VMWare Rkt ======================================= ======================================= Virtual Machines ~~~~~~~~~~~~~~~~ .. ifnotslides:: There are a lot of tools for managing Virtual Machines in many different ways. VirtualBox An Open Source VM Manager. Widely used and supported on Linux, Mac, and Windows. VMWare A closed source VM Manager. VMWare is a widely used and tends to have better performance than Virtual Box. While it can emulate Linux it does not work natively on Linux. KVM The Kernel-based Virtual Machine. Linux's native infrastructure for handling Virtual Machines and emulation. Usually used in a larger emulation program, not alone. Containers ~~~~~~~~~~ .. ifnotslides:: Containers have only recently gained popularity, but there are also many tools avaliable for container management. Docker The defacto CLI tool for creating and using containers. Very popular and well integrated into other tools. RKT A competitor to Docker created by CoreOS. Approaches container management from a different angle which has it's advantages and disadvantages. ``chroot`` The *oldschool* way to use containers. Not a container in the modern sense, but achieves similar isolation. Jails The BSD Unix form of containerization. Offers a level of secure isolation not really possible in Linux. TODO ---- .. TODO: Add activity. Further Reading --------------- .. TODO: Add further reading. .. Suggested: - Docker docs - RKT Docs - Virtualbox docs - etc...