Skip to content

Latest commit

 

History

History
79 lines (59 loc) · 3.62 KB

20221026-pot-created-hook.md

File metadata and controls

79 lines (59 loc) · 3.62 KB
Raw
title authors reviewers creation-date last-updated status
PreNormal Hook
@veophi
@furykerry
@FillZpp
@zmberg
2022-10-26
2022-10-26
implementable

PreNormal Hook

A Pod lifecycle hook allow user do something before Pod becomes available.

Table of Contents

A table of contents is helpful for quickly jumping to sections of a proposal and for highlighting any additional information provided beyond the standard proposal template. Tools for generating a table of contents from markdown are available.

Motivation

In Kubernetes world, pod lifecycle state transformation controlled by Deployment/ReplicaSet is only synchronized with containers status, rather than others, such as traffic routing. For example, when rolling update, the Pods will be treated as Available after it is Ready(assume that minReadySeconds=0), which means all its containers running, then the rolling update will be continued even though the Pod traffic routing has been not ready. In extreme cases, application traffic may be completely lost.

PreNormal hook helps users to avoid such terrible disaster. PreNormal ho~~~~ok will hold pod lifecycle transformation from PreparingNormal to Normal(Available) util users "say yes".

Proposal

To improve Kruise pod lifecycle state machine, we add a new lifecycle state called PreparingNormal, which means that the pod is still unavailable and wait some components to do something before it becomes Normal state. Also, we add a new lifecycle hook, namely PreNormalHook, which will holds the transformation from PreparingNormal (Unavailable State) to Normal (Available State). For example:

kind: CloneSet
spec:
  replicas: 3
  lifecycle:
    PreNormal:
      labelHandlers:
        example.com/traffic-online: "true"
  template:
    metadata:
      labels:
        example.com/traffic-online: "false"
  updateStrategy:
    type: ReCreate
    MaxSurge: 1
    MaxUnavailable: 0

seq-1

When users update the CloneSet template, it will create one new-version Pod, and keep the 3 old-version Pod:

  1. At first, the new-version Pod will be at PreparingNormal lifecycle state and has a false label of PreNormalHook when it is created by CloneSet.
  2. Due to the false label of PreNormalHook, Kruise will not allow the pod enter Normal state.
  3. CloneSet will treat the new-version Pod as Unavailable because of the PreparingNormal state, so any of the 3 old-version Pods will not be deleted.
  4. When users' component observed the newly-created Pod, and make sure that the pod ip has been added to SLB (maybe others), and it is ready to accept connections and requests, then update the Pod label from false to true;
  5. Pod lifecycle state will be transformed from PreparingNormal to Normal, and will be treated as Available by CloneSet.
  6. Then the rolling update can be continued...

Assume that there is a Kubernetes Cloud Controller Manager (CCM) operator, whose Service controller is designed to sync with remote SLB service, and Pod controller is designed to manage Pod lifecycle and coordinate with Kruise, the whole Pod lifecycle will be as follows: seq-1

Furthermore, users can combine other lifecycle hook, e.g., InPlaceUpdateHook and PreDeleteHook, to manage full lifecycle of Pods. With the PreNormalHook, the new lifecycle state machine will become:

seq-1