Book Image

Monitoring Docker

By : Russ McKendrick
Book Image

Monitoring Docker

By: Russ McKendrick

Overview of this book

This book will show you how monitoring containers and keeping a keen eye on the working of applications helps improve the overall performance of the applications that run on Docker. With the increased adoption of Docker containers, the need to monitor which containers are running, what resources they are consuming, and how these factors affect the overall performance of the system has become the need of the moment. This book covers monitoring containers using Docker's native monitoring functions, various plugins, as well as third-party tools that help in monitoring. Well start with how to obtain detailed stats for active containers, resources consumed, and container behavior. We also show you how to use these stats to improve the overall performance of the system. Next, you will learn how to use SysDig to both view your containers performance metrics in real time and record sessions to query later. By the end of this book, you will have a complete knowledge of how to implement monitoring for your containerized applications and make the most of the metrics you are collecting
Table of Contents (15 chapters)
Monitoring Docker
About the Author
About the Reviewer

Pets, Cattle, Chickens, and Snowflakes

Before we start discussing the various ways in which you can monitor your containers, we should get an understanding of what a SysAdmins world looks like these days and also where containers fit into it.

A typical SysAdmin will probably be looking after an estate of servers that are hosted in either an on-site or third-party data center, some may even manage instances hosted in a public cloud such as Amazon Web Services or Microsoft Azure, and some SysAdmins may juggle all their server estates across multiple hosting environments.

Each of these different environments has its own way of doing things, as well as performing best practices. Back in February 2012, Randy Bias gave a talk at Cloudscaling that discussed architectures for open and scalable clouds. Towards the end of the slide deck, Randy introduced the concept of Pets versus Cattle (which he attributes to Bill Baker, who was then an engineer at Microsoft).

You can view the original slide deck at

Pets versus Cattle is now widely accepted as a good analogy to describe modern hosting practices.


Pets are akin to traditional physical servers or virtual machines, as follows:

  • Each pet has a name; for example,

  • When they're not well, you take them to the vet to help them get better. You employ SysAdmins to look after them.

  • You pay close attention to them, sometimes for years. You take backups, patch them, and ensure that they are fully documented.


Cattle, on the other hand, represent more modern cloud computing instances, as follows:

  • You've got too many to name, so you give them numbers; for example, the URL could look something like

  • When they get sick, you shoot them and if your herd requires it, you replace anything you've killed: A server crashes or shows signs that it is having problems, you terminate it and your configuration automatically replaces it with an exact replica.

  • You put them in a field and watch them from far and you don't expect them to live long. Rather than monitoring the individual instances, you monitor the cluster. When more resources are needed, you add more instances and once the resource is no longer required, you terminate the instances to get you back to your base configuration.


Next up is a term that is a good way of describing how containers fit into the Pets versus Cattle world; in a blog post title "Cloud Computing: Pets, Cattle and ... Chickens?" on ActiveState, Bernard Golden describes containers as Chickens:

  • They're more efficient than cattle when it comes to resource use. A container can boot in seconds where a instance or server can take minutes; it also uses less CPU power than a typical virtual machine or cloud instance.

  • There are many more chickens than cattle. You can quite densely pack containers onto your instances or servers.

  • Chickens tend to have a shorter lifespan than cattle and pets. Containers lend themselves to running micros-services; these containers may only be active for a few minutes.

The original blog post can be found at


The final term is not animal-related and it describes a type of server that you defiantly don't want to have in your server estate, a Snowflake. This term was penned by Martin Fowler in a blog post titled "SnowflakeServer". Snowflakes is a term applied to "legacy" or "inherited" servers:

  • Snowflakes are delicate and are treated with kid gloves. Typically, the server has been in the data center since you started. No one knows who originally configured it and there is no documentation of it; all you know is that it is important.

  • Each one is unique and is impossible to exactly reproduce. Even the most hardened SysAdmin fears to reboot the machine incase it doesn't boot afterwards, as it is running end-of-life software that can not easily be reinstalled.

Martin's post can be found at

So what does this all mean?

Depending on your requirements and the application you want to deploy, your containers can be launched onto either pet or cattle style servers. You can also create a clutch of chickens and have your containers run micro-services.

Also, in theory, you can replace your feared snowflake servers with a container-based application that meets all the end-of-life software requirements while remaining deployable on a modern supportable platform.

Each of the different styles of server has different monitoring requirements, in the final chapter we will look at Pets, Cattle, Chickens, and Snowflakes again and discuss the tools we have covered in the coming chapters. We will also cover best practices you should take into consideration when planning your monitoring.