OCPNODE-3029: WIP: handle required minimum kubelet version featuregate rollout

[haircommander]

- What I did
When a cluster admin rolls out a minimum kubelet verison, the cluster-config-operator (eventually) will render
a new set of featuregates based on the new minimum version (enabling ones that are safe given the new minimum).

However, on a rollback, we need to make sure the MCS won't return a rendered config using the old features, in the very
odd case a cluster admin creates a newly old node and somehow overrides the osImage (unlikely). Thus, we use the similar condition
as MCS to serve the config--once one node has updated, we treat the MCP as using the new config. Then, we write the status to the
node config object so the node authorization plugin can allow older nodes that are now deemed safe.

- How to verify it

- Description for the changelog

[openshift-ci]

[APPROVALNOTIFIER] This PR is NOT APPROVED

This pull-request has been approved by: haircommander
Once this PR has been reviewed and has the lgtm label, please assign djoshy for approval. For more information see the Code Review Process.

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

Needs approval from an approver in each of these files:

• OWNERS

Approvers can indicate their approval by writing /approve in a comment
Approvers can cancel approval by writing /approve cancel in a comment

[haircommander]

some background openshift/enhancements#1766

[openshift-ci-robot]

@haircommander: This pull request references OCPNODE-3029 which is a valid jira issue.

Warning: The referenced jira issue has an invalid target version for the target branch this PR targets: expected the spike to target the "4.19.0" version, but no target version was set.

In response to this:

- What I did
When a cluster admin rolls out a minimum kubelet verison, the cluster-config-operator (eventually) will render
a new set of featuregates based on the new minimum version (enabling ones that are safe given the new minimum).

However, on a rollback, we need to make sure the MCS won't return a rendered config using the old features, in the very
odd case a cluster admin creates a newly old node and somehow overrides the osImage (unlikely). Thus, we use the similar condition
as MCS to serve the config--once one node has updated, we treat the MCP as using the new config. Then, we write the status to the
node config object so the node authorization plugin can allow older nodes that are now deemed safe.

- How to verify it

- Description for the changelog

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 openshift-eng/jira-lifecycle-plugin repository.

[yuqi-zhang]

any particular reason this is a separate goroutine?

[haircommander]

I thought it could be async in case it takes a while to have the mcps rollback and I didn't know what could be blocked by having it sync

[cheesesashimi]

I think a better solution might be to push this onto one of the work queues by doing something like this:

func (ctrl *Controller) syncKubeletConfig(key string) error {
	// Key lookup stuff above here.
	// Here, we detect that we need to do this for the current kubeletconfig, so we just kick that off.
	if ctrl.writeMinimumKubeletVersion[kubeletCfg.Name] {
		defer func() {
			delete(ctrl.writeMinimumKubeletVersion, kubeletCfg.Name)
		}()
		return ctrl.writebackMinimumKubeletVersionIfAppropriate(...)
	}

	//
	// Bulk of function above here.
	//

	if ctrl.isMinimumKubeletVersionWritebackNeeded(...) {
		// Here, we update our internal state to indicate that we need to perform this action
		// and enqueue the action.
		ctrl.writeMinimumKubeletVersion[kubeletcfg.Name] = true
		return ctrl.enqueue(kubeletCfg)
	}

	// End of function
}

We'll probably want to use a sync.Map since the work queue has multiple workers that could mutate the controller struct state at any given time.

[yuqi-zhang]

question: why are these conditions not correct? isn't this first one just saying that the spec and status match what we expect?

[haircommander]

if the status matches the spec, no need to do the update (though I probably should actually check renderedKubeletVersion there too)
for the second, if the status already matches the rendered then we've already done the write back for this rendered version
for the third: we're rendering a different version than what is set in the spec..
I think I have to rework this condition though

[yuqi-zhang]

I'm not sure I understand the logic here. The if condition checks if at least 1 node is updated, and if not, whether a previous loop had updated it? Wouldn't it be easier to check if the mcp itself is updated directly, and then break out of the loop if any pool doesn't meet either condition?

Note that we should be careful of pools with no nodes in them, which I think would be covered if we copy the server check

[cheesesashimi]

I think a better solution might be to push this onto one of the work queues by doing something like this:

func (ctrl *Controller) syncKubeletConfig(key string) error {
	// Key lookup stuff above here.
	// Here, we detect that we need to do this for the current kubeletconfig, so we just kick that off.
	if ctrl.writeMinimumKubeletVersion[kubeletCfg.Name] {
		defer func() {
			delete(ctrl.writeMinimumKubeletVersion, kubeletCfg.Name)
		}()
		return ctrl.writebackMinimumKubeletVersionIfAppropriate(...)
	}

	//
	// Bulk of function above here.
	//

	if ctrl.isMinimumKubeletVersionWritebackNeeded(...) {
		// Here, we update our internal state to indicate that we need to perform this action
		// and enqueue the action.
		ctrl.writeMinimumKubeletVersion[kubeletcfg.Name] = true
		return ctrl.enqueue(kubeletCfg)
	}

	// End of function
}

We'll probably want to use a sync.Map since the work queue has multiple workers that could mutate the controller struct state at any given time.

[openshift-ci]

@haircommander: The following tests failed, say /retest to rerun all failed tests or /retest-required to rerun all mandatory failed tests:

Test name	Commit	Details	Required	Rerun command
ci/prow/images	1a7271d	link	true	/test images
ci/prow/okd-scos-e2e-aws-ovn	1a7271d	link	false	/test okd-scos-e2e-aws-ovn
ci/prow/unit	1a7271d	link	true	/test unit
ci/prow/bootstrap-unit	1a7271d	link	false	/test bootstrap-unit
ci/prow/e2e-azure-ovn-upgrade-out-of-change	1a7271d	link	false	/test e2e-azure-ovn-upgrade-out-of-change
ci/prow/e2e-gcp-op	1a7271d	link	true	/test e2e-gcp-op
ci/prow/e2e-hypershift	1a7271d	link	true	/test e2e-hypershift
ci/prow/e2e-aws-ovn	1a7271d	link	true	/test e2e-aws-ovn
ci/prow/verify	1a7271d	link	true	/test verify
ci/prow/e2e-aws-ovn-upgrade-out-of-change	1a7271d	link	false	/test e2e-aws-ovn-upgrade-out-of-change
ci/prow/e2e-aws-ovn-upgrade	1a7271d	link	true	/test e2e-aws-ovn-upgrade
ci/prow/e2e-agent-compact-ipv4	1a7271d	link	false	/test e2e-agent-compact-ipv4
ci/prow/e2e-gcp-op-single-node	1a7271d	link	true	/test e2e-gcp-op-single-node
ci/prow/e2e-gcp-op-techpreview	1a7271d	link	false	/test e2e-gcp-op-techpreview
ci/prow/e2e-gcp-op-ocl	1a7271d	link	false	/test e2e-gcp-op-ocl

Full PR test history. Your PR dashboard.

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-sigs/prow repository. I understand the commands that are listed here.