Issue link: https://insights.oneneck.com/i/1293607
CHAPTER 3 The Impact of Containers 21 These materials are © 2017 John Wiley & Sons, Inc. Any dissemination, distribution, or unauthorized use is strictly prohibited. A full Docker container deployment fully negates the risk of down- time that accompanies a hypervisor change and may allow orga- nizations to adopt less expensive hypervisor solutions. The reason is the Docker Engine. By ensuring that there is a Docker Engine running on both the old hypervisor system and the new one, you can simply shift those containers across to their new home. The management and monitoring plane Overall application management and monitoring remains a chal- lenge even today. All kinds of companies are working really hard to provide complete solutions, but the sheer vastness of the appli- cation landscape makes a one-size-fits-all approach incredibly challenging. Containers brings with them some new options for ongoing man- agement, but you may be required to make a couple of upfront (but pretty easy!) choices: » Choose an engine. Again, Docker is recommended thanks to the fact that it's widely available and supported. » Choose a tool for ongoing management. I discuss these tools in a bit more depth later in this book, but the big ones are Docker Swarm and Universal Control Plane (UCP), Kubernetes, or Mesosphere DC/OS. By making these choices early and sticking to them, you can dras- tically reduce tool sprawl. Doing so allows developers and opera- tions to work from the same toolkit and promotes collaborative working styles. On the monitoring front, there is another key element to bear in mind: Container-based applications are typically distributed applications. In some cases, containers are big and almost per- manent members of the data center, but in many cases, they're also short-lived and run for just a few seconds, carry out their tasks, and are then recycled. In cases like this, using traditional approaches for monitoring is not appropriate because many tra- ditional approaches would see the "death" of a container as a system-down issue, when, in fact, it's perfectly normal. A mind shift needs to take place. You're not necessarily monitoring availability. Instead, you're using centralized logs so that key performance indicators (KPIs) focusing on end-user applica- tion performance or speed of transactions within the system are meeting or exceeding established requirements.