[kubernetes] kubernetes principle analysis and practical application (under update)

k8s The basic chapter
k8s The Internet

Kubernete How to control the resource level of the business ？

VPA

resource optimization ：Kubernete in GC（ Garbage collection ）

K8S Scheduling priority and preemption mechanism of

Official documents https://kubernetes.io/docs/concepts/scheduling-eviction/pod-priority-preemption/

Under normal circumstances
When one Pod After scheduling failure , It will be temporarily “ Put on hold ” get up , until Pod Be updated , Or the cluster state changes , That's what the scheduler does to this Pod Reschedule

Priority and preemption mechanism , The solution is Pod What to do when scheduling fails

In the scheduling process , There will be various pre selection and optimization algorithms , After so many thresholds , One POD To complete the dispatch , Provide services , In the scheduling process , Suppose a very important system service fails to schedule , It caused a breakdown , To avoid that , We can set a relatively high priority for the app , After a scheduling failure , Some relatively low priority applications that are not so important ” Squeeze away “, This is it. K8S The priority of scheduling and the role of preemption .

Scenarios with special requirements :
When a high priority Pod After scheduling failure , The Pod It's not going to be “ Put on hold ”, But will “ Squeeze away ” Some Node Some of the lower priority ones on Pod. This ensures this high priority Pod Dispatch of succeeded

PriorityClass Parameter limits

apiVersion: scheduling.k8s.io/v1
kind: PriorityClass
metadata:
  name: high-priority-nonpreempting
value: 1000000
preemptionPolicy: Never
globalDefault: false
description: "This priority class will not cause other pods to be preempted."

Kubernetes Setting priorities is a 32bit The integer of , The maximum value does not exceed 1000000000(10 Billion ,1 billion), And the higher the value, the higher the priority . And beyond 10 The value of 100 million , It's actually being Kubernetes Keep it and distribute it to the system Pod The use of . The purpose of this , It's the guarantee system Pod It won't be preempted by users .

tips:
K8S After a scheduling failure , Preemption will be based on priority , Preemption is not a simple way to Node Lower priority on pod” Squeeze away “, Preemptive design is relatively interesting .
https://blog.51cto.com/u_2010293/2781899

Kubernete Medium time task

https://kubernetes.io/zh-cn/docs/concepts/workloads/controllers/cron-jobs/

CronJob Spec

• .spec.schedule： Dispatch , Required fields , Specify the task run cycle , Same format Cron

• .spec.jobTemplate：Job Templates , Required fields , Specify the task to run , Same format Job

• .spec.startingDeadlineSeconds ： start-up Job Time limit of （ Second level ）, This field is optional . If the scheduled time is missed for any reason , So what's missing execution time Job Will be considered a failure . If not specified , There is no deadline

• .spec.concurrencyPolicy： Concurrency strategy , This field is also optional . It specifies how to deal with being Cron Job Created Job Concurrent execution of . Only one of the following policies is allowed to be specified ：

  • Allow（ Default ）： Allow concurrent runs Job
  • Forbid： Prohibit concurrent running , If the previous one hasn't been completed yet , Then skip the next one
  • Replace： Cancel the currently running Job, Replace... With a new one
    Be careful , The current policy can only be applied to the same Cron Job Created Job. If there are multiple Cron Job, They were created Job Always allow concurrent runs between .

• .spec.suspend ： Hang up , This field is also optional . If set to true, All subsequent execution will be suspended . It's about what's already being done Job It doesn't work . The default value is false.

• .spec.successfulJobsHistoryLimit and .spec.failedJobsHistoryLimit ： Historical limitations , Is an optional field . They specify how many completions and failures can be reserved Job.

  By default , They are set to 3 and 1. Set the limit to 0, Related types of Job It will not be retained after completion .