Millised tegurid tagavad sõlme õige lõpetamise sõltuvalt pordi prioriteedist?

Correct node shutdown is achieved through the following factors:

1. Timeouts:
   • terminationGracePeriodSeconds: Defines the time during which the node will attempt to gracefully terminate pod operations before forcefully destroying them. Default is 30 seconds.
   • pod.spec.terminationGracePeriodSeconds: Can override the timeout for a specific pod.

   # Example pod manifest with extended termination timeout
   apiVersion: v1
   kind: Pod
   metadata:
     name: my-pod-with-grace-period
   spec:
     containers:
     - name: my-container
       image: my-image
     terminationGracePeriodSeconds: 60 # Extended timeout
   

2. Signals:
   • Kubernetes sends a SIGTERM signal to the pod's containers.
   • The application inside the container should be able to catch SIGTERM and handle it properly (e.g., close current connections, save state).
   • If the application does not terminate upon SIGTERM within terminationGracePeriodSeconds, a SIGKILL is sent.
3. Pod priority (priorityClassName):
   • During preemption (e.g., resource shortage or node eviction), lower priority pods are terminated first.
   • This ensures that critical (high-priority) pods run longer and have more time to terminate properly if the node becomes unavailable.

   # Example pod manifest with priority class
   apiVersion: v1
   kind: Pod
   metadata:
     name: my-priority-pod
   spec:
     containers:
     - name: my-container
       image: my-image
     priorityClassName: high-priority # Using priority class
   

4. PreStop hooks:
   • Additional scripts or calls executed inside the container just before SIGTERM is sent.
   • Can be used for synchronization, deregistration from services, or other tasks needed for proper shutdown.

   # Example PreStop hook
   apiVersion: v1
   kind: Pod
   metadata:
     name: my-pod-with-prestop
   spec:
     containers:
     - name: my-container
       image: my-image
       lifecycle:
         preStop:
           exec:
             command: ["/bin/sh", "-c", "echo 'Stopping container...' && sleep 5 && exit 0"] # Example command
   

5. Endpoint management:
   • Kubernetes removes the pod's IP from the Endpoints object at the start of shutdown so that new traffic is not directed to the terminating pod.
   • This allows existing connections to finish during the grace period.

Together, these mechanisms enable Kubernetes to manage the pod termination process on the node, considering their priority, and give them a chance to terminate properly, avoiding forced destruction where possible.