Kubernetes 101 : Pod Affinity - Pods co-location -

The node affinity:

Node affinity means scheduling certain pods on specific nodes.
For example we could use node affinity to schedule resource-hungry pods on nodes that have more RAM or faster CPUs.

The pod affinity:

Pod affinity allows us to have pods that "work" together on the same node, same cloud region, ...
For example we could schedule a "web server pod" and a "database pod" on the same node.

Pod affinity example:

Below is a Yaml file of an "nginx" pod. 

We deploy the "nginx" pod first.

We also have a Yaml file of a "database pod
with pod affinity for the pods labeled "nginx_server":

We can create the above pods using the below command:


The "topologyKey: kubernetes.io/hostname" means that the pods with "pod affinity" attributes should be on the same host.

Pods affinity and anti-affinity:

The "requiredDuringSchedulingIgnoredDuringExecution" and
"preferredDuringSchedulingIgnoredDuringExecution" are the two types of pod affinity and anti-affinity.

We have a "required" and "preferred" version.

The "topologyKey" instruction:
In our example above, the "topologyKey" specifies that the pods with affinity need to be located on the same node.

Below some examples of topology keys:
  • kubernetes.io/hostname
  • beta.kubernetes.io/os 
  • beta.kubernetes.io/arch 
  • beta.kubernetes.io/instance-type 
  • -------

Whether Pods get scheduled on the same host, region, ..., depends on the values of both the topologyKey and the label selector.


Leave as a comment: