Kourier installation fails on Kind with "no matching resources found" — kubectl wait doesn't wait for pods to appear

[kekhuai]

Bug description

kn quickstart kind --registry fails intermittently during Kourier installation:

🕸️ Installing Kourier networking layer v1.22.0 ...
error: no matching resources found

Error: failed to install kourier to kind cluster knative: kourier: exit status 1

Root cause

The waitForPodsReady function in pkg/install/install.go uses kubectl wait which fails immediately when no pods match the selector — even with --timeout set:

func waitForPodsReady(ns string) error {
    return runCommand(exec.Command("kubectl", "wait", "pod",
        "--timeout=10m", "--for=condition=Ready",
        "-l", "!job-name", "-n", ns))
}

kubectl wait only waits for existing resources to meet a condition. When no pods exist yet (deployment controller still reconciling), it exits immediately with exit code 1.

Sequence

1. retryingApply applies kourier.yaml → deployments created
2. waitForPodsReady("kourier-system") called immediately
3. 3scale-kourier-gateway pod hasn't been created yet
4. kubectl wait exits instantly: error: no matching resources found
5. Installation fails

Reproducibility

This is a race condition — reproduces when pod creation is slower than code execution. Confirmed with:

kubectl wait pod --timeout=5s --for=condition=Ready -l '!job-name' -n kourier-system
# error: no matching resources found (exit 1)

Environment

• kn-quickstart: v1.22.0
• Kind: 0.32.0
• Kubernetes: v1.36.1 (kindest/node:v1.36.1)
• kubectl: v1.36.2
• OS: macOS

Suggested fix

Add a retry loop to waitForPodsReady that handles the "no matching resources found" case:

func waitForPodsReady(ns string) error {
    cmd := exec.Command("kubectl", "wait", "pod",
        "--timeout=10m", "--for=condition=Ready",
        "-l", "!job-name", "-n", ns)
    for i := 0; i < 6; i++ {
        if err := runCommand(cmd); err == nil {
            return nil
        }
        time.Sleep(10 * time.Second)
    }
    return fmt.Errorf("timeout waiting for pods in %s", ns)
}

Or alternatively, use kubectl wait --for=condition=Available deployment/<name> first, which properly handles the deployment lifecycle.

Note

The same latent bug exists in the waitForPodsReady("knative-serving") call in Kourier() and all three calls in Eventing(), but those are less likely to trigger because pods in those namespaces are typically already created by earlier steps.

[psschwei]

Thanks for reporting! #669 should fix it.