Skip to content

Latest commit

 

History

History

allocation

README.md

k8s-assessor

An agent which goes over the nodes in a Kubernetes cluster and calculates the total requested resources. We consider the following resources: "cpu", "memory", "pods", "gpu", and "ephemeral-storage". The total resource requested quantities are annotated as 'requested-xxx', where 'xxx' is the resource name. Further, the agent computes the available allocatable resources and annotates the nodes with 'free-xxx' quantities. The units for CPU are in milli-cores and for memory and storage in bytes.

The k8s-assessor is written in Python 3 and runs either externally to the cluster or internally in a pod.

External agent

Make sure to install the kubernetes package.

pip install kubernetes

The Kubernetes configuration file used is either the one set in the environment variable KUBECONFIG or (if not set) the default location ~/.kube/config.

Then, run the agent.

export CLUSTER_CONF=out-of-cluster

python k8s-assessor.py

Internal agent

Build the image.

docker build . -t k8s-assessor:0.0

Upload the image and run as a pod.

kubectl create -f k8s-assessor.yaml

If RBAC is enabled in the cluster, then you may use the following.

kubectl create -f k8s-assessor-rbac.yaml

A sample output is

Setting up environment ...
Using cluster_config = out-of-cluster
using freq = 60
Creating Kubernetes client ...
Using out-of-cluster config
Client created successfully! 
Format is node_name: resource[allocatable, requested, available]

2019-10-30 10:29:56.265725
minikube:  cpu=[2000, 855, 1145];   memory=[2033401856, 272629760, 1760772096];   pods=[110, 12, 98];   gpu=[0, 0, 0];   ephemeral-storage=[17361141760, 0, 17361141760];

A sample annotation is

Annotations:        free-cpu: 1145
                    free-ephemeral-storage: 17361141760
                    free-gpu: 0
                    free-memory: 1760772096
                    free-pods: 98
                    requested-cpu: 855
                    requested-ephemeral-storage: 0
                    requested-gpu: 0
                    requested-memory: 272629760
                    requested-pods: 12

The frequency of running the agent may be altered through the environment variable ANNOTATE_FREQ (units in seconds). The default is 60.

export ANNOTATE_FREQ=10
Notes:
  • The quantity.py script is added here since it is not included in older kubernetes client versions. It is needed to parse memory quantities and make the conversion to byte units.
  • May not run correctly with Python 2.
  • The k8s-assessor.py is built on the node-annotator of the kube-safe-scheduler extender project.
  • An alternative approach is to grep the block entitled Allocated resources: obtained from kubectl describe node and parsing it (or by parsing the obtained json file).