fix: use notifyContext to manage the operator exit

[jiuker]

fix: use notifyContext to manage the operator exit

Description

when operator lost leader role. Looks like we can't not exit operator
fix #2458

Related Issue

Type of Change

• Bug fix 🐛
• New feature 🚀
• Breaking change 🚨
• Documentation update 📖
• Refactor 🔨
• Other (please describe) ⬇️

Screenshots (if applicable e.g before/after)

Checklist

• I have tested these changes
• I have updated relevant documentation (if applicable)
• I have added necessary unit tests (if applicable)

Test Steps

Additional Notes / Context

[ramondeklein]

Can you please add the test steps, so this can be tested?

[jiuker]

Unless we support infrastructure disconnect testing, I am adding this code compilation test
@ramondeklein

// StartOperator starts the MinIO Operator controller
func StartOperator(kubeconfig string) {
	_ = v2.AddToScheme(scheme.Scheme)
	_ = stsv1beta1.AddToScheme(scheme.Scheme)
	_ = stsv1alpha1.AddToScheme(scheme.Scheme)
	klog.Info("Starting MinIO Operator")

	// set up signals, so we handle the first shutdown signal gracefully
	ctx, cancel := setupSignalHandler(context.Background())
	defer cancel()
	done := ctx.Done()

+	go func() {
+		time.Sleep(time.Second * 5)
+		cancel()
+	}()

I can see pod restart count +1.

[cniackz]

I’m approving this PR because it provides a pragmatic improvement to how the operator handles leadership loss, which currently results in a stuck and resource-hungry state. While it doesn’t solve the root cause (lack of probes or retry logic), it introduces a clean and safe exit mechanism using notifyContext, allowing Kubernetes to restart the pod. This is a net gain in resilience, especially given that the current behavior requires manual intervention. The code is minimal, targeted, and does not introduce complexity or regression risk.

[allanrogerr]

Minor change for clarity

[ramondeklein]

I did test this PR and the pod now properly terminates when it looses the lease. I do think we need to refactor this code, because it uses two mechanisms:

• context.Context that is initialized when starting the operator.
• chan struct{} that will be triggered when the context is cancelled.

I think it's better to have just a single method, because Controller.Start now receives both the context and the channel which serve exactly the same purpose. I think we need to fix that.