We can have additional schedulers on top of the default scheduler to schedule pods on nodes based on complex logic. We can also write a custom scheduler and use it as the default scheduler.
The default scheduler has the name default-scheduler by default, which is configured in the scheduler-config.yaml present at /etc/kubernetes/manifests/.
apiversion: kubescheduler.config.k8s.io/v1
kind: KubeSchedulerConfiguration
profiles:
- schedulerName: default-scheduler
When creating other schedulers, we can change the scheduler name in the scheduler config. Every scheduler in a cluster must have a unique name.
apiversion: kubescheduler.config.k8s.io/v1
kind: KubeSchedulerConfiguration
profiles:
- schedulerName: my-scheduler
<aside> 💡 Since most of K8s cluster is created using KubeAdmin, all the schedulers can be deployed as pods or deployments where the scheduler config file is passed as an argument (using ConfigMap).
</aside>
When we have multiple scheduler, we can specify which scheduler to use for a pod or a deployment. If a scheduler is not specified, the pod will remain in pending state.
apiVersion: v1
kind: Pod
metadata:
labels:
name: frontend
spec:
containers:
- name: httpd
image: httpd:2.4-alpine
schedulerName: my-scheduler
When running multiple schedulers in the cluster, there might be race conditions where multiple schedulers try to schedule a single pod. To avoid this, the latest K8s release has the concept of scheduler profiles.
It allows a single scheduler to have multiple profiles, each with a different set of plugins enabled and disabled. This way, a single scheduler acts like multiple schedulers and only one is used to schedule a pod.
We can also write custom plugins and use them.