register custom resource events for efficient pod enqueuing #317
Conversation
k8s-ci-robot
added
the
cncf-cla: yes
|
[APPROVALNOTIFIER] This PR is APPROVED This pull-request has been approved by: Huang-Wei The full list of commands accepted by this bot can be found here. The pull request process is described here
Needs approval from an approver in each of these files:
Approvers can indicate their approval by writing |
k8s-ci-robot
added
approved
|
/check-cla |
| return []framework.ClusterEvent{ | ||
| {Resource: framework.Pod, ActionType: framework.Delete}, | ||
| // TODO: once bump the dependency to k8s 1.22, addd custom object events. | ||
| {Resource: framework.GVK(eqGVK), ActionType: framework.Add | framework.Update}, |
There was a problem hiding this comment.
We also need to include the Delete event for elasticquota. Pod may be scheduled successfully if the eq is deleted.
There was a problem hiding this comment.
Can you elaborate in which circumstance a deleted eq would potentially make a pod (failed by this plugin) schedulable? My understanding is that the unschedulable pods needs to be get its own label updated, and it's irrelevant with EQ's change.
There was a problem hiding this comment.
I wonder if there is a scenario.
Scheduling of PodA in namespace A fails because the used in eq of namespace A will exceed the max rather than a lack of cluster resources. Then we remove the eq in namespace A so that it is no longer restricted to pod in namespace A. At this point we may need to have the pod in namespace A rescheduled.
But in fact we are not allowed to delete eq in a production environment if there are pods in the namespace A, unless the pods are deleted first.
There was a problem hiding this comment.
sg, I will update the event to All.
Huang-Wei
force-pushed
the
reg-cr-event
branch
from
9a9b458 to
0d7aa7a
Compare
| return []framework.ClusterEvent{ | ||
| {Resource: framework.Pod, ActionType: framework.Delete}, | ||
| {Resource: framework.Node, ActionType: framework.Add | framework.UpdateNodeAllocatable}, | ||
| {Resource: framework.GVK(nrtGVK), ActionType: framework.Add | framework.Update}, |
There was a problem hiding this comment.
FYI: @swatisehgal @AlexeyPerevalov: this change is an optimization to give a pod failed by NRT filter plugin higher chance to be retried immediately. Here I think only NRT's add and update events are correlated.
k8s-ci-robot
added
the
needs-rebase
Huang-Wei
force-pushed
the
reg-cr-event
branch
from
0d7aa7a to
b61deb0
Compare
k8s-ci-robot
removed
the
needs-rebase
|
/lgtm Thanks @Huang-Wei |
k8s-ci-robot
added
the
lgtm
Along with kubernetes/kubernetes#101394, the downstream plugins can register custom resource event(s) so as to trigger efficient pod queueing.