preface
- The declarative apis of K8S are mainly implemented in Reconcile cycles of various controllers
- Tune the loop to get the current and user-expected state of the cluster
- By adjusting the actual state, the desired state is constantly approached
- This article is based on the Kubernetes controller evolution journey: from resource control to open application model
- Because the text is easy to understand, after reading deeply inspired, the text content is displayed in the form of pictures
Phase 1: The controller accesses the API-server directly
- Too many requests cause the apI-server to be overloaded
Phase 2: The controller accesses the API-server through the Informer
- Informer provides List And Watch mechanism, incremental request API-server
- When wath, only watch specific resources
Phase 3: Multiple controllers share informer to access apI-server
- Resource objects managed by multiple controllers share Informer to further improve efficiency. For example, Deployment and DaemonSet both controllers manage POD resources
- The DeltaFIFO queue is used to handle event notifications and update the local cache
- WorkQueue Queues are used to notify each controller to process events
Stage 4: Customize controller + customize resource access API-server
- CRD and CRD Controller are used to customize business logic
- The Operator mode provides a simple programming framework for quickly implementing custom controllers
- The commonly used operator frameworks are:
- kubebuilder
- operator-sdk
Stage 5: Open Application Model(OAM)
- OAM is a set of specifications led by Ali and Microsoft
- The purpose is to decouple the responsibilities of development, operation and maintenance, and each performs its own duties
