Request Flow Through Kubernetes¶

Overview¶

This section explains what happens when you run:

kubectl apply -f deployment.yaml

End-to-End Flow¶

sequenceDiagram participant User participant APIServer participant etcd participant Scheduler participant Kubelet participant ContainerRuntime User->>APIServer: kubectl apply (Deployment) APIServer->>etcd: Store desired state Scheduler->>APIServer: Watch for unscheduled Pods Scheduler->>APIServer: Bind Pod to Node Kubelet->>APIServer: Watch for assigned Pods Kubelet->>ContainerRuntime: Pull image & start container ContainerRuntime-->>Kubelet: Container running Kubelet-->>APIServer: Report status APIServer->>etcd: Update current state

Step-by-Step Breakdown¶

1. User Request¶

kubectl sends request to API Server

2. API Server¶

Validates request
Stores object in etcd

3. Scheduler¶

Detects unscheduled Pod
Assigns it to a node

4. Kubelet¶

Watches API Server
Pulls assigned Pod spec
Starts containers

5. Continuous Reconciliation¶

Controllers ensure:
- Correct number of replicas
- Failed Pods are replaced

Real-World Implications¶

Delays in scheduling → resource constraints
Pod stuck Pending → scheduler issue
CrashLoopBackOff → runtime or app issue

Key Takeaways¶

Everything starts with the API Server
Scheduling is asynchronous
Kubernetes is event-driven and declarative