/cc @vishh
xref #62747

Aren't existing conditions expected to be transitioned to taints in which case k'let will place and remove taints?
I have use cases around bootstrapping and handling termination where the current plan is to re-use k'let's credentials.
Why is this restriction necessary especially given that taints are yet to be fully adopted even in core?

Aren't existing conditions expected to be transitioned to taints

They already are, but are translated by the node controller, not set directly by the kubelet (which works really well to allow a limited set of conditions to affect taints without giving the kubelet full control over them)

Why is this restriction necessary especially given that taints are yet to be fully adopted even in core?

Because kubelets removing/modifying their own taints removes a primary tool an administrator has for preventing nodes from accepting certain workloads (imagine a "compromised" or "unverified" taint set on a node intended to keep the scheduler from sending workloads to it)

See kubernetes/community#911 for a fuller discussion of the issues with letting kubelets steer workloads via self-labeling or self-untainting

cc @kubernetes/sig-auth-pr-reviews @kubernetes/sig-node-pr-reviews @kubernetes/sig-scheduling-pr-reviews

The diff is ok to me :) There's a discussion (#62311) on replacing node condition with taints, will we add white list for node conditions?

Would it impact these scenario (cloud specific taint being applied on the fly/runtime): https://github.com/kubernetes/kubernetes/blob/master/pkg/cloudprovider/providers/aws/aws.go#L1777 ?

Would it impact these scenario (cloud specific taint being applied on the fly/runtime): https://github.com/kubernetes/kubernetes/blob/master/pkg/cloudprovider/providers/aws/aws.go#L1777 ?

No, that is only called from AttachDisk, which has been done by the attach/detach controller, not the kubelet, since 1.3

xref #55517

There's a discussion (#62311) on replacing node condition with taints, will we add white list for node conditions?

If we want the kubelet to start setting/updating/removing taints on itself directly, we can discuss the pros/cons of whitelisting specific taints at that point. I could see whitelisting the set that is automatically translated from node-reported conditions->taints, but would want that to be an intentional decision.

There's a discussion (#62311) on replacing node condition with taints, will we add white list for node conditions?

If we want the kubelet to start setting/updating/removing taints on itself directly, we can discuss the pros/cons of whitelisting specific taints at that point. I could see whitelisting the set that is automatically translated from node-reported conditions->taints, but would want that to be an intentional decision.

That's ok to me, and PR is LGTM; waiting for mikedanese@'s comments if any :)

I would like to restrict this since nothing uses it, and loosen the restriction as needed going forward. @vishh has a reasonable usecase. How should a node agent out of core request evacuation? Does the node agent need an accompanying controller that understands a new node condition?

I would like to restrict this since nothing uses it, and loosen the restriction as needed going forward.

I agree.

@vishh has a reasonable usecase. How should a node agent out of core request evacuation? Does the node agent need an accompanying controller that understands a new node condition?

I could see any of the following (roughly in order of preference):

• running a separate controller for managing a custom condition<->taint (part of the custom node agent setup)
• deciding on a well-known key or prefix for taints that nodes would be allowed to manage themselves
• options to the existing node controller to tell it additional custom conditions to translate into taints
• options to the admission plugin to configure specific taints to allow the node to manage

cc @kubernetes/sig-cluster-lifecycle-pr-reviews

/lgtm

Let's get this in and relax constraints when we need to.

rebased

Download error for bazel build

W0515 17:57:42.467] ERROR: /bazel-scratch/.cache/bazel/_bazel_root/e9f728bbd90b3fba632eb31b20e1dacd/external/io_bazel_rules_go/BUILD.bazel:9:1: GoStdlib external/io_bazel_rules_go/linux_amd64_stripped/stdlib~/pkg failed: Failed to delete output files after incomplete download. Cannot continue with local execution.: /bazel-scratch/.cache/bazel/_bazel_root/e9f728bbd90b3fba632eb31b20e1dacd/execroot/main/bazel-out/host/bin/external/io_bazel_rules_go/linux_amd64_stripped/stdlib~/pkg (Directory not empty)

/retest

/test pull-kubernetes-e2e-kops-aws

running a separate controller for managing a custom condition<->taint (part of the custom node agent setup)

Separate controller manager has race condition issue between condition <-> taints :( , well-known key or prefix for taints makes sense to me.

xref #63897

/lgtm

help to re-lgtm :)

[APPROVALNOTIFIER] This PR is APPROVED

This pull-request has been approved by: k82cn, liggitt, mikedanese

The full list of commands accepted by this bot can be found here.

The pull request process is described here

/test all [submit-queue is verifying that this PR is safe to merge]

Automatic merge from submit-queue (batch tested with PRs 63167, 63357). If you want to cherry-pick this change to another branch, please follow the instructions here.