How it works - 1.1 - What happens when you run kubectl apply command

We have started a Study Group at our company for PipeCD members.
Each week someone can take a short talk to share something with other members.
So I started a series of talks with the title "How it works" to explain to other members how the things in use work internally.
The "How it works" series consists of many seasons, episodes like the movie.

Transcript

1. nghialv

   Sig-Build Study
   Dec 17, 2021
   ”How It Works” Series
   Season 1: Kubernetes
   Episode 1: What happens when you run kubectl apply command
   

   View Slide

2. kubectl run myapp --image=myapp:1.0.0 --replicas=3
   What happens when you run
   

   View Slide

3. 1
   kubectl apply -f deployment.yaml
   Kubernetes Cluster
   apiVersion: apps/v1
   
   
   kind: Deployment
   
   
   metadata:
   
   
   name: myapp
   
   
   labels:
   
   
   app: myapp
   
   
   spec:
   
   
   replicas: 3
   
   
   selector:
   
   
   matchLabels:
   
   
   app: myapp
   
   
   template:
   
   
   metadata:
   
   
   labels:
   
   
   app: myapp
   
   
   spec:
   
   
   containers:
   
   
   - name: myapp
   
   
   image: myapp:1.0.0
   
   
   ports:
   
   
   - containerPort: 80
   
   
   kubectl run myapp --image=myapp:1.0.0 --replicas=3
   

   View Slide

4. 1
   kubectl apply -f deployment.yaml
   Control Plane Node
   Kubernetes Cluster
   apiVersion: apps/v1
   
   
   kind: Deployment
   
   
   metadata:
   
   
   name: myapp
   
   
   labels:
   
   
   app: myapp
   
   
   spec:
   
   
   replicas: 3
   
   
   selector:
   
   
   matchLabels:
   
   
   app: myapp
   
   
   template:
   
   
   metadata:
   
   
   labels:
   
   
   app: myapp
   
   
   spec:
   
   
   containers:
   
   
   - name: myapp
   
   
   image: myapp:1.0.0
   
   
   ports:
   
   
   - containerPort: 80
   
   
   kubectl run myapp --image=myapp:1.0.0 --replicas=3
   Node
   Means that there are many nodes are ommited in the diagram
   

   View Slide

5. kube-scheduler
   1
   kubectl apply -f deployment.yaml
   Control Plane Node
   Kubernetes Cluster
   etcd
   etcd etcd
   Persistence Store
   kube-apiserver
   kube-apiserver
   kube-apiserver
   Each controller is run in parallel

   by kube-controller-manager
   node-controller
   replicaset-controller
   ...
   deployment-controller
   kube-controller-manager
   apiVersion: apps/v1
   
   
   kind: Deployment
   
   
   metadata:
   
   
   name: myapp
   
   
   labels:
   
   
   app: myapp
   
   
   spec:
   
   
   replicas: 3
   
   
   selector:
   
   
   matchLabels:
   
   
   app: myapp
   
   
   template:
   
   
   metadata:
   
   
   labels:
   
   
   app: myapp
   
   
   spec:
   
   
   containers:
   
   
   - name: myapp
   
   
   image: myapp:1.0.0
   
   
   ports:
   
   
   - containerPort: 80
   
   
   kubectl run myapp --image=myapp:1.0.0 --replicas=3
   kubelet
   kube-proxy
   container-runtime
   kubelet
   kube-proxy
   container-runtime
   Node
   • kube-apiserver is a stateless component and can be easily scaled by adding more instances
   
   
   • kube-controller-manager runs as a single process that spawns many controllers to run in parallel
   

   View Slide

6. kube-scheduler
   1
   kubectl apply -f deployment.yaml
   Control Plane Node
   Kubernetes Cluster
   etcd
   etcd etcd
   Persistence Store
   kube-apiserver
   kube-apiserver
   kube-apiserver
   Each controller is run in parallel

   by kube-controller-manager
   node-controller
   replicaset-controller
   ...
   deployment-controller
   kube-controller-manager
   apiVersion: apps/v1
   
   
   kind: Deployment
   
   
   metadata:
   
   
   name: myapp
   
   
   labels:
   
   
   app: myapp
   
   
   spec:
   
   
   replicas: 3
   
   
   selector:
   
   
   matchLabels:
   
   
   app: myapp
   
   
   template:
   
   
   metadata:
   
   
   labels:
   
   
   app: myapp
   
   
   spec:
   
   
   containers:
   
   
   - name: myapp
   
   
   image: myapp:1.0.0
   
   
   ports:
   
   
   - containerPort: 80
   
   
   kubectl run myapp --image=myapp:1.0.0 --replicas=3
   kubelet
   kube-proxy
   container-runtime
   kubelet
   kube-proxy
   container-runtime
   Node
   All external requests go through this hub
   
   All internal requests for Object API go though this hub
   

   View Slide

7. kube-scheduler
   1
   kubectl apply -f deployment.yaml
   Control Plane Node
   Kubernetes Cluster
   etcd
   etcd etcd
   Persistence Store
   kube-apiserver
   kube-apiserver
   kube-apiserver
   Each controller is run in parallel

   by kube-controller-manager
   node-controller
   replicaset-controller
   ...
   deployment-controller
   kube-controller-manager
   apiVersion: apps/v1
   
   
   kind: Deployment
   
   
   metadata:
   
   
   name: myapp
   
   
   labels:
   
   
   app: myapp
   
   
   spec:
   
   
   replicas: 3
   
   
   selector:
   
   
   matchLabels:
   
   
   app: myapp
   
   
   template:
   
   
   metadata:
   
   
   labels:
   
   
   app: myapp
   
   
   spec:
   
   
   containers:
   
   
   - name: myapp
   
   
   image: myapp:1.0.0
   
   
   ports:
   
   
   - containerPort: 80
   
   
   kubectl run myapp --image=myapp:1.0.0 --replicas=3
   kubelet
   kube-proxy
   container-runtime
   kubelet
   kube-proxy
   container-runtime
   Node
   All external requests go through this hub
   
   All internal requests for Object API go though this hub
   2
   Save Objects
   

   View Slide

8. kube-scheduler
   1
   kubectl apply -f deployment.yaml
   Control Plane Node
   Kubernetes Cluster
   etcd
   etcd etcd
   Persistence Store
   kube-apiserver
   kube-apiserver
   kube-apiserver
   Each controller is run in parallel

   by kube-controller-manager
   node-controller
   replicaset-controller
   ...
   deployment-controller
   kube-controller-manager
   apiVersion: apps/v1
   
   
   kind: Deployment
   
   
   metadata:
   
   
   name: myapp
   
   
   labels:
   
   
   app: myapp
   
   
   spec:
   
   
   replicas: 3
   
   
   selector:
   
   
   matchLabels:
   
   
   app: myapp
   
   
   template:
   
   
   metadata:
   
   
   labels:
   
   
   app: myapp
   
   
   spec:
   
   
   containers:
   
   
   - name: myapp
   
   
   image: myapp:1.0.0
   
   
   ports:
   
   
   - containerPort: 80
   
   
   kubectl run myapp --image=myapp:1.0.0 --replicas=3
   kubelet
   kube-proxy
   container-runtime
   kubelet
   kube-proxy
   container-runtime
   Node
   All external requests go through this hub
   
   All internal requests for Object API go though this hub
   2
   Save Objects
   3
   Fetch Objects
   5 Handle Objects
   6
   Update Objects
   4
   Fetch Objects
   • deployment-controller fetches Deployment objects and handle (creating new ReplicaSet object) then write back their status to etcd
   
   
   • replicaset-controller fetches ReplicaSet objects and handle (creating new Pod object) then write back their status to etcd
   

   View Slide

9. 1
   kubectl apply -f deployment.yaml
   Control Plane Node
   Kubernetes Cluster
   etcd
   etcd etcd
   Persistence Store
   kube-apiserver
   kube-apiserver
   kube-apiserver
   Each controller is run in parallel

   by kube-controller-manager
   node-controller
   replicaset-controller
   ...
   deployment-controller
   kube-controller-manager
   apiVersion: apps/v1
   
   
   kind: Deployment
   
   
   metadata:
   
   
   name: myapp
   
   
   labels:
   
   
   app: myapp
   
   
   spec:
   
   
   replicas: 3
   
   
   selector:
   
   
   matchLabels:
   
   
   app: myapp
   
   
   template:
   
   
   metadata:
   
   
   labels:
   
   
   app: myapp
   
   
   spec:
   
   
   containers:
   
   
   - name: myapp
   
   
   image: myapp:1.0.0
   
   
   ports:
   
   
   - containerPort: 80
   
   
   kubectl run myapp --image=myapp:1.0.0 --replicas=3
   kubelet
   kube-proxy
   container-runtime
   kubelet
   kube-proxy
   container-runtime
   Node
   All external requests go through this hub
   
   All internal requests for Object API go though this hub
   2
   Save Objects
   3
   Fetch Objects
   5 Handle Objects
   6
   Update Objects
   4
   Fetch Objects
   kube-scheduler
   7
   Create a Binding object to assign 

   newly created Pod to a suitable Node
   Fetch Pods 8 Handle Pods
   9
   Update Pods
   • Node selection is based on node available resources (memory, cpu...), balancing, data locality, or user's speci
   fi
   ed conditions...
   

   View Slide

10. 1
    kubectl apply -f deployment.yaml
    Control Plane Node
    Kubernetes Cluster
    etcd
    etcd etcd
    Persistence Store
    kube-apiserver
    kube-apiserver
    kube-apiserver
    Each controller is run in parallel

    by kube-controller-manager
    node-controller
    replicaset-controller
    ...
    deployment-controller
    kube-controller-manager
    apiVersion: apps/v1
    
    
    kind: Deployment
    
    
    metadata:
    
    
    name: myapp
    
    
    labels:
    
    
    app: myapp
    
    
    spec:
    
    
    replicas: 3
    
    
    selector:
    
    
    matchLabels:
    
    
    app: myapp
    
    
    template:
    
    
    metadata:
    
    
    labels:
    
    
    app: myapp
    
    
    spec:
    
    
    containers:
    
    
    - name: myapp
    
    
    image: myapp:1.0.0
    
    
    ports:
    
    
    - containerPort: 80
    
    
    kubectl run myapp --image=myapp:1.0.0 --replicas=3
    kubelet
    kube-proxy
    container-runtime
    kubelet
    kube-proxy
    container-runtime
    Node
    All external requests go through this hub
    
    All internal requests for Object API go though this hub
    2
    Save Objects
    3
    Fetch Objects
    5 Handle Objects
    6
    Update Objects
    4
    Fetch Objects
    kube-scheduler
    7
    Create a Binding object to assign 

    newly created Pod to a suitable Node
    Fetch Pods 8 Handle Pods
    9
    Update Pods
    9
    container 1
    container 2
    pod-1
    9
    container 1
    container 2
    pod-2
    

    View Slide

11. kube-scheduler
    1
    kubectl apply -f deployment.yaml
    2
    7
    Control Plane Node
    Kubernetes Cluster
    etcd
    etcd etcd
    Persistence Store
    kube-apiserver
    kube-apiserver
    kube-apiserver
    Each controller is run in parallel

    by kube-controller-manager
    All external requests go through this hub
    
    All internal requests for Object API go though this hub
    Create a Binding object to assign 

    newly created Pod to a suitable Node
    node-controller
    replica-set-controller
    ...
    deployment-controller
    kube-controller-manager
    apiVersion: apps/v1
    
    
    kind: Deployment
    
    
    metadata:
    
    
    name: myapp
    
    
    labels:
    
    
    app: myapp
    
    
    spec:
    
    
    replicas: 3
    
    
    selector:
    
    
    matchLabels:
    
    
    app: myapp
    
    
    template:
    
    
    metadata:
    
    
    labels:
    
    
    app: myapp
    
    
    spec:
    
    
    containers:
    
    
    - name: myapp
    
    
    image: myapp:1.0.0
    
    
    ports:
    
    
    - containerPort: 80
    
    
    3
    Fetch Objects
    Save Objects
    5 Handle Objects
    6
    Update Objects
    Fetch Pods 8 Handle Pods
    9
    Update Pods
    4
    Fetch Objects
    kubectl run myapp --image=myapp:1.0.0 --replicas=3
    kubelet
    kube-proxy
    container-runtime
    container 1
    container 2
    pod-1
    container 1
    container 2
    pod-3
    kubelet
    kube-proxy
    container-runtime
    container 1
    container 2
    pod-2
    Node
    9
    9
    

    View Slide

12. What we have learned from K8s' design?
    

    View Slide

13. kube-scheduler
    1
    kubectl apply -f deployment.yaml
    2
    7
    Control Plane Node
    Kubernetes Cluster
    etcd
    etcd etcd
    Persistence Store
    kube-apiserver
    kube-apiserver
    kube-apiserver
    Each controller is run in parallel

    by kube-controller-manager
    All external requests go through this hub
    
    All internal requests for Object API go though this hub
    Create a Binding object to assign 

    newly created Pod to a suitable Node
    node-controller
    replica-set-controller
    ...
    deployment-controller
    kube-controller-manager
    apiVersion: apps/v1
    
    
    kind: Deployment
    
    
    metadata:
    
    
    name: myapp
    
    
    labels:
    
    
    app: myapp
    
    
    spec:
    
    
    replicas: 3
    
    
    selector:
    
    
    matchLabels:
    
    
    app: myapp
    
    
    template:
    
    
    metadata:
    
    
    labels:
    
    
    app: myapp
    
    
    spec:
    
    
    containers:
    
    
    - name: myapp
    
    
    image: myapp:1.0.0
    
    
    ports:
    
    
    - containerPort: 80
    
    
    3
    Fetch Objects
    Save Objects
    5 Handle Objects
    6
    Update Objects
    Fetch Pods 8 Handle Pods
    9
    Update Pods
    4
    Fetch Objects
    kubectl run myapp --image=myapp:1.0.0 --replicas=3
    kubelet
    kube-proxy
    container-runtime
    container 1
    container 2
    pod-1
    container 1
    container 2
    pod-3
    kubelet
    kube-proxy
    container-runtime
    container 1
    container 2
    pod-2
    Node
    9
    9
    

    View Slide

14. To be continued
    

    View Slide