Skip to content

Retrieving metadata from the Kubernetes API

Digger utilizes the Kubernetes API to gather metadata on these entities: pods, endpoints, and services. This process employs the method depicted as Efficient detection of changes in the official K8s documentation.

A dedicated ServiceAccount is created. This ServiceAccount is given cluster-wide permissions (ClusterRoleBinding) to get, list and watch on several apiGroups. Digger gets these access rights by using this ServiceAccount.

Here's how the communication between Digger and the Kubernetes API is structured:

How Digger interacts with the Kubernetes API

In more detail:

  1. During the initiation phase of the DIM Sensor, Digger conducts three list calls to collect initial data on collections of pods, endpoints, and services.

  2. Subsequently, Digger initiates three watch calls, essentially subscribing to notifications about configuration modifications in entities of these three types.

    • In scenarios where the Kubernetes API is unavailable due to any reason, Digger attempts to re-subscribe, with the waiting duration between attempts incrementing exponentially.

    • For the subscription process, Digger employs the resourceVersion so that Kubernetes will only transmit new events to it.

  3. Upon receiving a new entity in the notifications from Kubernetes, Digger processes the metadata, such as the name, namespace, labels, and annotations, storing these in the cache.

  4. Digger also identifies the higher-level Kubernetes construct that created this entity, in other words, the workload resource (such as DaemonSet, ReplicaSet, or Deployment, among others) that serves as the actual owner of this entity. This gives a broader context, for instance, to the pod name (which could be quite cryptic in isolation) of the discovered service.