Kube-DNS: The Essential Service for Kubernetes Clusters
Discover why DNS is particularly important when dealing with large-scale, complex environments, like those based on Kubernetes. Get a breakdown of how DNS works in Kubernetes and gain some key tips and tricks on how to get the most out of Kubernetes DNS.
Can you recite your best friend's phone number from memory? Chances are that you can't – and it's not because you're a bad friend (at least not for that reason). It's that most people don't have to remember phone numbers anymore because smartphones do that hard work for us. When we want to call someone, we type their name in, and the phone automatically looks up and dials their number for us.
Computers do essentially the same thing, thanks to DNS, which makes it possible to map the names of applications, servers, services and basically any other entity that has a network identity onto IP addresses. It also assists in processes like service discovery, which enables entities on the network to find each other. DNS is important on virtually any type of network. But it's especially critical when you're dealing with large-scale, complex environments, like those based on Kubernetes. Indeed, without DNS, the life of Kubernetes admins would be considerably more tedious.
Keep reading for a breakdown of how DNS works in Kubernetes, along with tips on how to get the most out of Kubernetes DNS.
What is DNS?
Before diving into the particulars of DNS in Kubernetes, let's talk about what DNS means in general.
DNS – short for Domain Name Service – is a type of service that translates (or resolves, to use the technical term) human-readable names to network IP addresses. For example, DNS is how your computer figures out which IP address to connect to when you want to open google.com in your browser, or when you want to send a job to a local network printer.
DNS is, in other words, akin to a phonebook for the Internet. DNS servers store records of which human-readable network address names correspond to which IP addresses, and make that information available upon request to devices, applications or services that want it.
In general, applications and services that need to resolve network names using DNS have logic built into them for connecting to DNS servers. But you can also make DNS requests manually using tools like host, a CLI utility for Unix-like operating systems. For example, if you want to see the IP addresses of the server that hosts google.com, you could type:
It would return:
You can also use dig if you want to pull DNS records from a specific DNS server (rather than whichever one your computer is configured to use by default). For example, to run a DNS request on the server with IP address 8.8.8.8 (which happens to be a DNS server managed by Google), you'd do:
DNS and service discovery
In addition to mapping network names to IP addresses, DNS plays a role in service discovery, which is the process by which different resources on the network find out about each other.
By default, applications and servers don't usually know which other applications or servers exist on the network, unless they've been preconfigured with that information. Service discovery allows entities to poll the network, figure out what else is out there and determine its IP address.
It's kind of like walking down the street and yelling "HI THERE! WHAT'S YOUR NAME AND PHONE NUMBER?" at strangers as you pass them by, except in a less awkward and more systematic way.
DNS and large-scale environments
DNS is useful in any context, just as being able to save your friends' phone numbers is valuable no matter how many friends you have. But the importance of automatically mapping identities to numbers is especially critical when you’re dealing with lots of resources – or friends, as the case may be.
If you’re a loner with just three friends, you might be able to remember their phone numbers on your own, just as you could probably keep track of the IP addresses of just a handful of servers manually. But if you're super-duper cool with hundreds of friends, there's no way you can commit all of their numbers to memory. You need an automated system to track those numbers for you.
Kubernetes's need for DNS
A Kubernetes environment is like a person with lots of friends: There are dozens, hundreds or potentially even thousands of entities with unique network identities inside a Kubernetes cluster, and keeping track of the IP addresses of all of them by hand just doesn't work.
Think about it: A Kubernetes cluster is home to a large number of self-contained entities, which are constantly interacting with each other over a variety of internal networks. In addition, any cluster that hosts public-facing applications needs to manage requests from the Internet, which in most cases means it needs to know the IP addresses of even more endpoints.
If you wanted to hate your life, you could try to keep track of the IP addresses for the various resources inside Kubernetes manually. But that would be a nightmare because you'd not only have to spend hours and hours configuring IP addresses initially, but you'd also have to update your IP address lists constantly, since many IP addresses in Kubernetes are dynamically assigned and subject to constant change. The IP address of a Pod or node could change when the Pod or node restarts, for example.
By translating the service names of the various resources inside Kubernetes into IP addresses and keeping track of changes to IP addresses automatically, DNS services in Kubernetes make life much, much simpler for admins. They also help to ensure that the many moving parts inside Kubernetes can talk to each other smoothly – and that ingoing and outgoing requests associated with external networks are properly routed – even as the network configuration constantly changes.
To be clear, DNS is only one ingredient in a full-fledged Kubernetes networking stack. DNS doesn't handle other important network-related tasks, such as load-balancing or the enforcement of security rules via network policies. You'd need other tools (like service meshes and ingress controllers) to do that work. But DNS does handle one of the most important tasks required to operate a Kubernetes cluster at scale.
How Kubernetes DNS is implemented
Today, most Kubernetes clusters rely on CoreDNS, an open source DNS server written in Go, to provide DNS services across the environment. Earlier versions of Kubernetes used SkyDNS, but the Kubernetes developers switched to CoreDNS because the latter is designed specifically for cloud-native environments.
CoreDNS is deployed in a simple way: It runs as a cluster-level Pod and handles DNS queries within the cluster from there. By default, the service is named kube-dns.
Kubernetes external DNS
The CoreDNS DNS server is available in Kubernetes by default, so you don't need to do anything special to get it running. However, if you also need to resolve the names of external services, you might want to install an additional DNS server designed for external requests, such as the aptly named ExternalDNS.
Configuring and troubleshooting kube-DNS
Of course, just because Kubernetes provides an internal DNS server by default doesn't mean it always works the way it's supposed to. If you're experiencing issues like the failure of nodes, Pods or Services to communicate with each other, there's a chance it could be due to problems with your Kubernetes DNS server, so you should troubleshoot it and assess your DNS health.
The first step in doing so is making sure the kube-dns service is actually running. You can do this with kubectl:
Explore related posts
