Nerdalize is building a different cloud. Rather than putting thousands of servers into a datacenter, we're distributing them over homes. There, homeowners make good use of the residual heat: to heat their home in winter and their shower water in summer.
The HeatScheduler is a scheduler for Kubernetes that takes boiler temperatures into account. Most important is that it shows how to run an extra scheduler with your own scheduling logic on any Kubernetes cluster.
- Change you Docker image name in the
Makefile. - Build and push the Docker image to a container registry:
make push. - Deploy the policy config:
kubectl create -f deployments/scheduler-policy-config.yaml. - Deploy the scheduler:
kubectl create -f deployments/heat-scheduler.yaml. - Deploy a pod that uses
schedulerName: heat-scheduler. Seedeployments/pod-example.yamlfor an example.
Feel free to use this code for your own custom Kubernetes scheduler. Including your own business logic can be done by changing the selectNode function in util.go.
- In the
deploymentsfolder you find a Kubernetes Deployment that launches a new kube-scheduler. More information about this can be found in the Kubernetes docs. - In
deployments/scheduler-policy-config.yamlyou can change the predicates that the scheduler should use by default, before passing the pod to the extender. A list of these predicates can be found here. - We currently only use the
filterendpoint, which returns a filtered subset of the original list of Nodes. It is also possible to return a weighted list of Nodes that can be used to indicate Node priorities by using theprioritizeendpoint. See the official docs for more information.