What causes "no connection to cached dial" in derailed/k9s?

Run CheckConnectivity() and only call CachedDiscovery() when it returns true

no connection to cached dial

Error message

no connection to cached dial

What it means

Returned by APIClient.CachedDiscovery (internal/client/client.go:491) when connOK is false. This method builds a disk-cached discovery client (cache dir from KUBECACHEDIR or the default http cache dir) used for GVR resolution and resource mapping. Like Dial/DialLogs it is gated on the shared connection-health flag, so any earlier failed probe blocks it.

Source

Thrown at internal/client/client.go:491

	}
	c, err := kubernetes.NewForConfig(cfg)
	if err != nil {
		return nil, err
	}
	a.setClient(c)

	return a.getClient(), nil
}

// RestConfig returns a rest api client.
func (a *APIClient) RestConfig() (*restclient.Config, error) {
	return a.config.RESTConfig()
}

// CachedDiscovery returns a cached discovery client.
func (a *APIClient) CachedDiscovery() (*disk.CachedDiscoveryClient, error) {
	if !a.getConnOK() {
		return nil, errors.New("no connection to cached dial")
	}

	if c := a.getCachedClient(); c != nil {
		return c, nil
	}

	cfg, err := a.RestConfig()
	if err != nil {
		return nil, err
	}

	baseCacheDir := os.Getenv("KUBECACHEDIR")
	if baseCacheDir == "" {
		baseCacheDir = filepath.Join(mustHomeDir(), ".kube", "cache")
	}

	httpCacheDir := filepath.Join(baseCacheDir, "http")
	discCacheDir := filepath.Join(baseCacheDir, "discovery", toHostDir(cfg.Host))

View on GitHub (pinned to 2d3ccc6ba2)

Solutions

Run CheckConnectivity() and only call CachedDiscovery() when it returns true
Fix the kubeconfig/credentials that made the initial probe fail, then re-init the client
Optionally set KUBECACHEDIR to a writable directory so the disk cache itself cannot fail afterwards

Example fix

// before
dd, err := apiClient.CachedDiscovery()

// after
if apiClient.CheckConnectivity() {
    dd, err := apiClient.CachedDiscovery()
}

Defensive patterns

Strategy: validation

Validate before calling

if !apiClient.CheckConnectivity() {
    return nil, errors.New("discovery unavailable: no connection")
}
dd, err := apiClient.CachedDiscovery()

Try / catch

dd, err := apiClient.CachedDiscovery()
if err != nil && strings.Contains(err.Error(), "no connection to cached dial") {
    // reconnect, then retry once; discovery is safe to retry (idempotent)
}

Prevention

Resolve GVRs once at startup via CachedDiscovery after connectivity is confirmed
Set KUBECACHEDIR to a writable path so cache writes never compound the failure
Watch for apiservice churn and re-run discovery lazily, not per keystroke

When it happens

Trigger: Calling CachedDiscovery() after the connection was marked down, or before the first successful connection. Also triggered when RESTConfig() fails (bad kubeconfig) because CheckConnectivity sets connOK=false in that path.

Common situations: Resource listing or GVR lookups immediately after a network interruption; read-only kubeconfig with a broken certificate path; stale KUBECACHEDIR on a read-only filesystem; session resumed after cluster credentials rotated.

Related errors

AI-assisted analysis of derailed/k9s@2d3ccc6ba2 (2026-08-15). Data as JSON: /api/errors/1934cb80e5352058. Report an issue: GitHub.