[Umbrella] Optimize Cluster Failover: import new effect taint and cluster failover strategy

[XiShanYongYe-Chang]

Background

We have completed some optimizations in v1.14: #6317, for the next work in version 1.15, let's continue to create an issue for tracking, which is more convenient and version management aware.

• Proposal @XiShanYongYe-Chang Optimize Cluster Failover #6274

release-1.15

• Introduce new taints, users actively configure migration
  • API Change
    • Add a new effect taint SelectiveNoExecute @XiShanYongYe-Chang API Change: add a new effect taint SelectiveNoExecute #6381
    • Application cluster failover strategy in PP/CPP @XiShanYongYe-Chang Add application-level cluster failover strategy API #6376
    • Update the comment of .spec.placement.clusterTolerations in the PP/CPP to explain its effective scope
    • ClusterTaintPolicy API supports time-dimension condition matching @XiShanYongYe-Chang API Change: support time-dimension condition matching in ClusterTaintPolicy API #6410
  • karmada-webhook
    • remove default-not-ready-toleration-seconds and default-unreachable-toleration-seconds flags @XiShanYongYe-Chang Remove default-not-ready-toleration-seconds and default-unreachable-toleration-seconds flags #6444
  • karmada-controller-manager
    • Add the controller to handle with SelectiveNoExecute effect taint @abhinav-1305 feat: add controller to handle taints with SelectiveNoExecute effect #6465
    • Import Directly, Gracefully PurgeMode, Deprecate the Immediately, Graciously PurgeMode @XiShanYongYe-Chang Introduced Directly, Gracefully PurgeMode, and deprecated Immediately and Graciously PurgeMode #6387
  • UT & E2E
  • Documentation
• GracefulEvictionTask clean issue
• karmada-controller-manager
  • remove failover-evition-timeout flag @XiShanYongYe-Chang Remove failover-eviction-timeout flag in the karmada-controller-manager #6450
  • The ClusterTaintPolicy controller performs time-dimension matching for conditions
  • When deleting Works from disconnected clusters, the system will automatically retry failed deletion operations at scheduled intervals.
  • In the graceful_eviction_controller, when cleaning up the GracefulEvctionTask, introduce a plugin, and each item of the plugin is used to check whether the elements can be cleaned
  • The start of the taint-manager controller is controlled by the controllers parameter
• Feature Gate
  • Deprecate the GracefulEviction FeatureGate
• Eviction Queue
  • Determine the resources affected by the cluster fault according to the taint type and enqueue
  • Configure multiple rate levels for the eviction queue
  • Enhanced migration success rate
  • Add metrics to the eviction queue
  • Handle the scenario of the failed cluster recovery

feature plan in release-1.xx

• API Change
  • When upgrading the PropagationPolicy version, convert the purgeMode value
    • Convert 'Immediately' to 'Directly'
    • Convert 'Graciously' to 'Gracefully'

[mszacillo]

Hi @XiShanYongYe-Chang, would it be possible to include the addition of state preservation for cluster failover within the scope of this feature? At the moment state preservation is tied explicitly to application failover, but the concept of state preservation should also be part of cluster failover.

[XiShanYongYe-Chang]

Hi @mszacillo, I submitted a new proposal for the state preservation for cluster failover: #6495, can you help take a review?