telemd-kubernetes-adapter

Daemon that watches Pods in Kubernetes that stores metadata, such that users are later able to associate the telemd cgroup metrics with the corresponding container based on the ID.

Build

To build the module's binaries using your local go installation run:

make

To build Docker images for local usage (without a go installation) run:

make docker

Details

telemd-kubernetes-adapter reports all Pods running and newly being created. It uses etcd's watch feature (via the kubernetes client lib) to monitor Pods that are created by Kubernetes.

Because the telemd reports for running containers (including Pods and their containers) only the container ID, we need to save some metadata in order to later associate Pods/Container Images to their respective usage. This functionality offers the telemd-kubernetes-adapter daemon.

See this SO thread for an explanation on how cgroups are structured for Pods.

Usage

Topic schema

The daemon will call for each pod the following command:

publish galileo/events "{timestamp} pod/[create|delete] {metadata}"

Whereas metadata:

{
    "podUid": "12b379f7-ef99-46e5-8125-0129a27bd799",
    "containerID": "containerd://11e9220bd2b450ed40132bcd5cdc3c115b93c69d02c37d844fec7574026edff3",
    "image": "docker.io/library/nginx:1.14.2",
    "podName": "nginx-deployment-66b6c48dd5-vd2jt",
    "hostIP": "10.0.1.2",
    "name": "nginx",
    "nodeName": "pod-host",
    "podIP": "10.42.1.95",
    "qosClass": "BestEffort",
    "startTime": "2021-08-17T18:29:58Z"
}

All fields are extracted from:

kubectl get pod --all-namespaces -o json | jq '.items[] | select(.metadata.uid == "12b379f7-ef99-46e5-8125-0129a27bd799")'

Environment variables

The daemon will read redis credentials from the following environment variables:

• telemd_kubernetes_adapter_redis_host
• telemd_kubernetes_adapter_redis_port
• telemd_kubernetes_adapter_location (either: local or cluster)
  • In case of local, will read kubeconfig from KUBECONFIG env variable