You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Why is this needed: #1762 addresses the eviction feature that focuses on trigger eviction based on taints toleration and introduces a taint manager to perform the eviction.
Once the taint manager performs the eviction (means remove a specific cluster from the RB(ResourceBinding) target list(.spec.clusters)), the resource referenced by the RB will be `removed immediately from the target cluster.
For example, there is an online workload(like an Nginx server) that was running on cluster-1, and then a network broken issue happens, Karmada starts to evict it from cluster-1 and probably karmada-scheuler scheduled it to cluster-2. The thing is the workload would be removed immediately and no service available until the workload runs on cluster-2.
The text was updated successfully, but these errors were encountered:
Instructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the kubernetes/test-infra repository.
What would you like to be added:
We hope to introduce a
graceful
way to delay the removal until the resource is available on the substitute cluster.Iterated task:
GracefulEviction
feature gate. #2340karmada-scheduler
to markschedulerObservedGeneration
(RB) #2330karmada-scheduler
to markschedulerObservedGeneration
(CRB) #2345grace-eviction-controller
to implementers the API. (#2319)binding controller
to adopt the changes. (#2339)taint manager
to trigger graceful eviction. (#2346)Why is this needed:
#1762 addresses the eviction feature that focuses on trigger eviction based on taints toleration and introduces a
taint manager
to perform the eviction.Once the
taint manager
performs the eviction (means remove a specific cluster from the RB(ResourceBinding
) target list(.spec.clusters)), the resource referenced by the RB will be `removed immediately from the target cluster.For example, there is an online workload(like an Nginx server) that was running on
cluster-1
, and then a network broken issue happens, Karmada starts to evict it fromcluster-1
and probablykarmada-scheuler
scheduled it tocluster-2
. The thing is the workload would be removed immediately and no service available until the workload runs oncluster-2
.The text was updated successfully, but these errors were encountered: