feat(update): pre-flight gates + deeper healthcheck + auto-rollback

Phase 8 of v0.3. Tightens the update lifecycle on both ends.

Pre-flight (apply.go, before any download):
- Free-space check on the passive partition: image size + 10% headroom must
  be available. Uses statfs(2) via the new pkg/partition.FreeBytes /
  HasFreeSpaceFor helpers (tests cover happy path, tiny request, huge
  request, missing path). Catches corrupted-FS and shrunk-partition cases
  before we destroy the existing slot data.
- Node-block-label check: refuses if the local K8s node carries the
  updates.kubesolo.io/block=true label. New pkg/health.CheckNodeBlocked
  shells out to kubectl per the project's zero-deps stance. Silently bypassed
  when no kubeconfig is reachable (air-gap case). Skipped by --force.

Healthcheck (extended via new pkg/health/extended.go + preflight.go):
- CheckKubeSystemReady waits until every kube-system pod has held the Running
  phase for >= N seconds (default 30). Catches "started ok, will crash-loop"
  bugs that a single-shot phase check misses.
- CheckProbeURL fetches an operator-supplied URL; 200 = pass. Wired through
  update.conf as healthcheck_url= and cloud-init updates.healthcheck_url.
- CheckDiskWritable writes/fsyncs/reads a 1-KiB probe under /var/lib/kubesolo.
  Always runs in healthcheck so a wedged data partition fails fast.
- pkg/health.Status grows KubeSystemReady, ProbeURL, DiskWritable booleans.
  Optional checks default to true in RunAll() so they don't block when
  unconfigured. health_test.go updated to the new 6-field shape.

Auto-rollback (healthcheck.go):
- state.UpdateState gains HealthCheckFailures (consecutive post-Activated
  failures). Reset on a clean pass.
- --auto-rollback-after N (also auto_rollback_after= in update.conf) triggers
  env.ForceRollback() when the failure count reaches the threshold. State
  transitions to RolledBack with a descriptive LastError. The command still
  exits with the healthcheck error; the operator/init is expected to reboot.
- Only fires while Phase == Activated. Doesn't second-guess a long-stable
  system that happens to fail one healthcheck.

config / opts / cloud-init plumbing:
- update.conf gains healthcheck_url= and auto_rollback_after= keys.
- New CLI flags: --healthcheck-url, --auto-rollback-after, --kube-system-settle.
- cloud-init full-config.yaml documents the new updates: subfields.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

update/pkg/health/extended.go

@@ -0,0 +1,125 @@
+package health
+
+import (
+	"context"
+	"fmt"
+	"net/http"
+	"os"
+	"os/exec"
+	"path/filepath"
+	"strings"
+	"time"
+)
+
+// kubeSystemSettleSeconds is how long all kube-system pods must hold a
+// Running phase before we consider the cluster genuinely up. Catches the
+// "pod just started, will crash-loop in 5s" case.
+const kubeSystemSettleSeconds = 30
+
+// CheckKubeSystemReady verifies that every pod in the kube-system namespace
+// is in Running phase and has been Running for at least settle. Returns
+// (ready, error). settle defaults to 30s when zero.
+func (c *Checker) CheckKubeSystemReady(settle time.Duration) bool {
+	if settle == 0 {
+		settle = kubeSystemSettleSeconds * time.Second
+	}
+	if _, err := os.Stat(c.kubeconfigPath); err != nil {
+		return false
+	}
+	ctx, cancel := context.WithTimeout(context.Background(), 10*time.Second)
+	defer cancel()
+
+	// jsonpath emits one line per pod: <phase>|<startTime>
+	cmd := exec.CommandContext(ctx, "kubectl",
+		"--kubeconfig", c.kubeconfigPath,
+		"get", "pods", "-n", "kube-system",
+		"-o", `jsonpath={range .items[*]}{.status.phase}|{.status.startTime}{"\n"}{end}`,
+	)
+	out, err := cmd.Output()
+	if err != nil {
+		return false
+	}
+	lines := strings.Split(strings.TrimSpace(string(out)), "\n")
+	if len(lines) == 0 || lines[0] == "" {
+		// No pods reported. Conservatively treat as not-ready: kube-system
+		// is expected to host at least CoreDNS + pause.
+		return false
+	}
+	now := time.Now()
+	for _, line := range lines {
+		parts := strings.SplitN(line, "|", 2)
+		phase := strings.TrimSpace(parts[0])
+		if phase != "Running" {
+			return false
+		}
+		if len(parts) < 2 {
+			return false
+		}
+		start, perr := time.Parse(time.RFC3339, strings.TrimSpace(parts[1]))
+		if perr != nil {
+			return false
+		}
+		if now.Sub(start) < settle {
+			return false
+		}
+	}
+	return true
+}
+
+// CheckProbeURL fetches the given URL and reports whether it returned 200.
+// Empty url returns (true, nil) — the check is opt-in.
+func CheckProbeURL(url string) (bool, error) {
+	if url == "" {
+		return true, nil
+	}
+	client := &http.Client{Timeout: 5 * time.Second}
+	resp, err := client.Get(url)
+	if err != nil {
+		return false, fmt.Errorf("probe URL %s: %w", url, err)
+	}
+	defer resp.Body.Close()
+	return resp.StatusCode == http.StatusOK, nil
+}
+
+// CheckDiskWritable writes a small file under dataDir, fsyncs, reads it back,
+// and removes it. Confirms the data partition is mounted read-write and the
+// underlying disk is responsive. Empty dataDir defaults to /var/lib/kubesolo.
+func CheckDiskWritable(dataDir string) (bool, error) {
+	if dataDir == "" {
+		dataDir = "/var/lib/kubesolo"
+	}
+	if _, err := os.Stat(dataDir); err != nil {
+		// Data partition not mounted? That's catastrophic but we shouldn't
+		// claim the disk is fine.
+		return false, fmt.Errorf("dataDir %s: %w", dataDir, err)
+	}
+	probe := filepath.Join(dataDir, ".update-probe")
+	want := []byte("kubesolo-os healthcheck probe")
+
+	f, err := os.Create(probe)
+	if err != nil {
+		return false, fmt.Errorf("create probe: %w", err)
+	}
+	defer os.Remove(probe)
+
+	if _, err := f.Write(want); err != nil {
+		f.Close()
+		return false, fmt.Errorf("write probe: %w", err)
+	}
+	if err := f.Sync(); err != nil {
+		f.Close()
+		return false, fmt.Errorf("fsync probe: %w", err)
+	}
+	if err := f.Close(); err != nil {
+		return false, fmt.Errorf("close probe: %w", err)
+	}
+
+	got, err := os.ReadFile(probe)
+	if err != nil {
+		return false, fmt.Errorf("read probe: %w", err)
+	}
+	if string(got) != string(want) {
+		return false, fmt.Errorf("probe content mismatch: got %q", got)
+	}
+	return true, nil
+}

update/pkg/health/extended_test.go

@@ -0,0 +1,77 @@
+package health
+
+import (
+	"net/http"
+	"net/http/httptest"
+	"os"
+	"path/filepath"
+	"testing"
+)
+
+func TestCheckProbeURLEmptyAlwaysPasses(t *testing.T) {
+	ok, err := CheckProbeURL("")
+	if err != nil {
+		t.Fatalf("CheckProbeURL(\"\"): %v", err)
+	}
+	if !ok {
+		t.Error("empty probe URL should return ok=true (check disabled)")
+	}
+}
+
+func TestCheckProbeURL200(t *testing.T) {
+	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		w.WriteHeader(http.StatusOK)
+	}))
+	defer srv.Close()
+	ok, err := CheckProbeURL(srv.URL)
+	if err != nil {
+		t.Fatalf("CheckProbeURL: %v", err)
+	}
+	if !ok {
+		t.Error("expected ok=true on 200")
+	}
+}
+
+func TestCheckProbeURLNon200(t *testing.T) {
+	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		w.WriteHeader(http.StatusServiceUnavailable)
+	}))
+	defer srv.Close()
+	ok, err := CheckProbeURL(srv.URL)
+	if err != nil {
+		t.Fatalf("CheckProbeURL: %v", err)
+	}
+	if ok {
+		t.Error("expected ok=false on 503")
+	}
+}
+
+func TestCheckProbeURLNetworkError(t *testing.T) {
+	// Port 1 is reserved (tcpmux) and never bound by Linux defaults.
+	_, err := CheckProbeURL("http://127.0.0.1:1")
+	if err == nil {
+		t.Error("expected error for unreachable URL, got nil")
+	}
+}
+
+func TestCheckDiskWritableHappyPath(t *testing.T) {
+	dir := t.TempDir()
+	ok, err := CheckDiskWritable(dir)
+	if err != nil {
+		t.Fatalf("CheckDiskWritable: %v", err)
+	}
+	if !ok {
+		t.Error("expected ok=true on writable temp dir")
+	}
+	// Probe file should have been cleaned up.
+	if _, err := os.Stat(filepath.Join(dir, ".update-probe")); !os.IsNotExist(err) {
+		t.Errorf("probe file not cleaned up: stat err=%v", err)
+	}
+}
+
+func TestCheckDiskWritableMissingDir(t *testing.T) {
+	_, err := CheckDiskWritable("/this/path/does/not/exist")
+	if err == nil {
+		t.Error("expected error for missing dataDir, got nil")
+	}
+}

update/pkg/health/health.go

@@ -24,15 +24,20 @@ import (
 
 // Status represents the result of a health check.
 type Status struct {
-	Containerd bool
-	APIServer  bool
-	NodeReady  bool
-	Message    string
+	Containerd       bool
+	APIServer        bool
+	NodeReady        bool
+	KubeSystemReady  bool // optional — true unless KubeSystemSettle is non-zero
+	ProbeURL         bool // optional — true unless ProbeURL is set
+	DiskWritable     bool // optional — true unless DataDir is set
+	Message          string
 }
 
-// IsHealthy returns true if all checks passed.
+// IsHealthy returns true if all required checks passed. Optional checks
+// default to true when not configured, so they don't block the result.
 func (s *Status) IsHealthy() bool {
-	return s.Containerd && s.APIServer && s.NodeReady
+	return s.Containerd && s.APIServer && s.NodeReady &&
+		s.KubeSystemReady && s.ProbeURL && s.DiskWritable
 }
 
 // Checker performs health checks against the local KubeSolo instance.
@@ -40,6 +45,11 @@ type Checker struct {
 	kubeconfigPath string
 	apiServerAddr  string
 	timeout        time.Duration
+
+	// Optional gates. Zero values disable the check (it reports true).
+	KubeSystemSettle time.Duration
+	ProbeURL         string
+	DataDir          string
 }
 
 // NewChecker creates a health checker.
@@ -149,12 +159,37 @@ func (c *Checker) CheckNodeReady() bool {
 }
 
 // RunAll performs all health checks and returns the combined status.
+//
+// Optional checks (kube-system settle, user probe URL, disk writability) are
+// only run if the corresponding Checker fields are set; otherwise they
+// report true so as not to block the result.
 func (c *Checker) RunAll() *Status {
-	return &Status{
-		Containerd: c.CheckContainerd(),
-		APIServer:  c.CheckAPIServer(),
-		NodeReady:  c.CheckNodeReady(),
+	s := &Status{
+		Containerd:      c.CheckContainerd(),
+		APIServer:       c.CheckAPIServer(),
+		NodeReady:       c.CheckNodeReady(),
+		KubeSystemReady: true,
+		ProbeURL:        true,
+		DiskWritable:    true,
 	}
+	if c.KubeSystemSettle > 0 {
+		s.KubeSystemReady = c.CheckKubeSystemReady(c.KubeSystemSettle)
+	}
+	if c.ProbeURL != "" {
+		ok, err := CheckProbeURL(c.ProbeURL)
+		if err != nil {
+			slog.Warn("probe URL check failed", "url", c.ProbeURL, "error", err)
+		}
+		s.ProbeURL = ok
+	}
+	if c.DataDir != "" {
+		ok, err := CheckDiskWritable(c.DataDir)
+		if err != nil {
+			slog.Warn("disk writability check failed", "dir", c.DataDir, "error", err)
+		}
+		s.DiskWritable = ok
+	}
+	return s
 }
 
 // WaitForHealthy polls health checks until all pass or timeout expires.

update/pkg/health/health_test.go

@@ -6,36 +6,42 @@ import (
 )
 
 func TestStatusIsHealthy(t *testing.T) {
+	// Helper for the new 6-field Status: all-true except the named one.
+	allBut := func(field string) Status {
+		s := Status{
+			Containerd: true, APIServer: true, NodeReady: true,
+			KubeSystemReady: true, ProbeURL: true, DiskWritable: true,
+		}
+		switch field {
+		case "Containerd":
+			s.Containerd = false
+		case "APIServer":
+			s.APIServer = false
+		case "NodeReady":
+			s.NodeReady = false
+		case "KubeSystemReady":
+			s.KubeSystemReady = false
+		case "ProbeURL":
+			s.ProbeURL = false
+		case "DiskWritable":
+			s.DiskWritable = false
+		}
+		return s
+	}
+
 	tests := []struct {
 		name       string
 		status     Status
 		wantHealth bool
 	}{
-		{
-			name:       "all healthy",
-			status:     Status{Containerd: true, APIServer: true, NodeReady: true},
-			wantHealth: true,
-		},
-		{
-			name:       "containerd down",
-			status:     Status{Containerd: false, APIServer: true, NodeReady: true},
-			wantHealth: false,
-		},
-		{
-			name:       "apiserver down",
-			status:     Status{Containerd: true, APIServer: false, NodeReady: true},
-			wantHealth: false,
-		},
-		{
-			name:       "node not ready",
-			status:     Status{Containerd: true, APIServer: true, NodeReady: false},
-			wantHealth: false,
-		},
-		{
-			name:       "all down",
-			status:     Status{Containerd: false, APIServer: false, NodeReady: false},
-			wantHealth: false,
-		},
+		{"all healthy", allBut(""), true},
+		{"containerd down", allBut("Containerd"), false},
+		{"apiserver down", allBut("APIServer"), false},
+		{"node not ready", allBut("NodeReady"), false},
+		{"kube-system not ready", allBut("KubeSystemReady"), false},
+		{"probe URL failed", allBut("ProbeURL"), false},
+		{"disk not writable", allBut("DiskWritable"), false},
+		{"all down", Status{}, false},
 	}
 
 	for _, tt := range tests {

update/pkg/health/preflight.go

@@ -0,0 +1,51 @@
+package health
+
+import (
+	"context"
+	"fmt"
+	"os"
+	"os/exec"
+	"strings"
+	"time"
+)
+
+// NodeBlockLabel is the well-known label that workload authors set on the
+// local node to defer an OS update. When present and "true", apply refuses.
+const NodeBlockLabel = "updates.kubesolo.io/block"
+
+// CheckNodeBlocked returns (blocked, error). blocked==true means the local
+// node carries the updates.kubesolo.io/block=true label and the caller should
+// refuse the update.
+//
+// If the kubeconfig is not available (offline / pre-boot / air-gap), this
+// returns (false, nil) — silently allowing the update. That's the safe
+// behaviour for the air-gap case where the node may not be reachable from
+// the agent's perspective.
+func CheckNodeBlocked(kubeconfigPath string) (bool, error) {
+	if kubeconfigPath == "" {
+		kubeconfigPath = "/var/lib/kubesolo/pki/admin/admin.kubeconfig"
+	}
+	if _, err := os.Stat(kubeconfigPath); err != nil {
+		// No kubeconfig — assume air-gap / pre-K8s. Don't block updates.
+		return false, nil
+	}
+
+	// Query the node label via kubectl. We don't know the node name a
+	// priori, so we use --kubeconfig on the local admin config and ask for
+	// "the only node" (KubeSolo is single-node by design).
+	ctx, cancel := context.WithTimeout(context.Background(), 10*time.Second)
+	defer cancel()
+
+	cmd := exec.CommandContext(ctx, "kubectl",
+		"--kubeconfig", kubeconfigPath,
+		"get", "node",
+		"-o", `jsonpath={.items[0].metadata.labels.updates\.kubesolo\.io/block}`)
+	out, err := cmd.Output()
+	if err != nil {
+		// API unreachable or no nodes — treat as not blocked (analogous to
+		// the kubeconfig-missing case). We still surface the error so the
+		// caller can decide to log it.
+		return false, fmt.Errorf("query node label: %w", err)
+	}
+	return strings.TrimSpace(string(out)) == "true", nil
+}