Kubernetes Classroom Series – 28/Jun/2021

Managed Life Cycle

• Overview: Containerized applications managed by cloud-native platforms have no control over their lifecycle & to be good cloud native citizens they have to listen to the events emitted by managing platforms and adapt their lifecylcles according.

• Problem:

  • In additon to monitoring the state of container, the platform sometimes might issue commands and expect the application to react to those
  • Driven by external factors and policies, platform may decide to start or stop the applications it is managing at any moment & it is up to the containerized application to determine which events are important to react & how to react.

• Solution:

  • For this usecase some events are emitted by the platform, that the container can listen to and react to if desired
  • SIGTERM signal: whenever k8s decides to shut down a container (failed liveness probe or Pod it belongs to is shutting), the container recieves a SIGTERM signal. Once the SIGTERM signal has been recived, the application should shutdown as quickly as possible.
  • SIGKILL signal: If a container process has not shutdown after SIGTERM signal, it is shutdown forcefully by following the SIGKILL signal. k8s doesn’t send the SIGKILL signal immemdietly after SIGTERM it will wait for a grace period (default is 30 second). The grace period can be defined per pod .spec.terminationGracePeriodSeconds
  • Container hooks: Refer Here for offical docs and Refer Here for the api reference of Lifecycle hooks
    • PostStart Hook
    • PreStop Hook

• Refer Here for the article terminating with grace

apiVersion: v1
kind: Pod
metadata:
  name: hooks-demo
spec:
  containers:
    - name: hooks-container
      image: nginx
      ports:
        - containerPort: 80
      livenessProbe:
        httpGet: 
          port: 80
          path: /
        initialDelaySeconds: 30
      lifecycle:
        preStop:
          httpGet:
            port: 80
            path: /shutdown
        

Automated Placement

• Overview: Automated Placement is the core function of the k8s scheduler for signing new Pods to nodes. This pattern is about exploring the ways to influence placement decisions from outside.
• Problem: A decent microservice will consist of tens to hunderds of isolated processes (Containers in a Pod). With large and ever growing number of microservices, assigning and placing them individually to nodes is not a managable activity
• Solution:
  • In k8s, assining Pods to nodes is done by kube-scheduler.
  • kube-scheduler whenever it finds a newly created Pod defintion for API Server assigns the pod to node.
  • Allocatable Node Resources: “` Allocatable = Node Capacity – kube-Reserved (kubelet, container) – System-Reserved (sshd, udev)