examples

examples

Some show cased howto operate workload based on carbon emission.

Precondition: A installed Prometheus Stack with Entsoe Exporter to provide the current carbon footprint of power generation.

Update resourcequotas per Cron

In this cronjob a Prometheus API will ask for current carbon state of power generation. On this decision resourcequotas for the target namespace will adjusted.

Works technically but has no effects on running workload or the workload won't start if the quota is reached

Modify cgroups/cpu.max in Pod

Resourcequotas are realized by cgroup settings in Kubelet and the underlying Cgroup Driver, which manifests by the underlying Container Runtime Interface(CRI). Usually, and without HostPath this resources are only read-only in the Pod and can't be modified. Alibaba Cloud has a cgroup controller running, so the user can do this and act as in the example above

In-Place update Pod resources

In Kubernetes 1.27 this Kep was realized, which makes the resouces in containers.spec writable.

Requires, like K3S start flag:

...
        --kube-apiserver-arg feature-gates="InPlacePodVerticalScaling=true" \
        --kube-controller-arg feature-gates="InPlacePodVerticalScaling=true" \
        --kubelet-arg feature-gates="InPlacePodVerticalScaling=true" \

Then you can use the carbon-cronjob.yaml and make a patch based on the current carbon emission:

kubectl -n carbon patch pod pod-demo --patch '{"spec":{"containers":[{"name":"pod-demo", "resources":{"requests":{"cpu":"550m"}}}]}}'
pod/pod-demo patched

Deployment Resources

see ./resources/

Patch deployment and adjust cpu resources based on eco power generation

Horizontal Pod Autoscaler (HPA)

see ./hpa/

Patch hpa and adjust replicas based on eco power generation

Keda Prometheus

see ./keda-prometheus/

Use ScaledObject from https://keda.io to fetch Prometheus metrics of carbon emission and act on workload deployment