test(ananke): make startup cooldown coverage deterministic

internal/state/intent.go

@@ -22,12 +22,23 @@ type Intent struct {
 	UpdatedAt time.Time `json:"updated_at"`
 }
 
-var writeIntentImpl = writeIntentDefault
+var (
+	readIntentImpl  = readIntentDefault
+	writeIntentImpl = writeIntentDefault
+)
 
 // ReadIntent runs one orchestration or CLI step.
 // Signature: ReadIntent(path string) (Intent, error).
 // Why: keeps behavior explicit so startup/shutdown workflows remain maintainable as services evolve.
 func ReadIntent(path string) (Intent, error) {
+	return readIntentImpl(path)
+}
+
+// readIntentDefault runs one orchestration or CLI step.
+// Signature: readIntentDefault(path string) (Intent, error).
+// Why: keeps production read behavior available while tests can override intent
+// reads deterministically without racing background file mutations.
+func readIntentDefault(path string) (Intent, error) {
 	b, err := os.ReadFile(path)
 	if err != nil {
 		if os.IsNotExist(err) {

internal/state/testhooks.go

@@ -22,6 +22,34 @@ func TestHookWriteIntentDefault(path string, in Intent) error {
 	return writeIntentDefault(path, in)
 }
 
+// TestHookReadIntentDefault runs one orchestration or CLI step.
+// Signature: TestHookReadIntentDefault(path string) (Intent, error).
+// Why: lets top-level tests delegate to production ReadIntent behavior while
+// selectively forcing deterministic read sequences for lifecycle branches.
+func TestHookReadIntentDefault(path string) (Intent, error) {
+	return readIntentDefault(path)
+}
+
+// TestHookSetReadIntentOverride runs one orchestration or CLI step.
+// Signature: TestHookSetReadIntentOverride(fn func(path string) (Intent, error)) (restore func()).
+// Why: enables deterministic intent-read failure injection without sleeping
+// goroutines that race slower CI agents.
+func TestHookSetReadIntentOverride(fn func(path string) (Intent, error)) (restore func()) {
+	testHookOverrideMu.Lock()
+	prev := readIntentImpl
+	if fn == nil {
+		readIntentImpl = readIntentDefault
+	} else {
+		readIntentImpl = fn
+	}
+	testHookOverrideMu.Unlock()
+	return func() {
+		testHookOverrideMu.Lock()
+		readIntentImpl = prev
+		testHookOverrideMu.Unlock()
+	}
+}
+
 // TestHookSetWriteIntentOverride runs one orchestration or CLI step.
 // Signature: TestHookSetWriteIntentOverride(fn func(path string, in Intent) error) (restore func()).
 // Why: enables deterministic intent-write failure injection from the top-level

testing/orchestrator/hooks_lifecycle_deep_matrix_test.go

@@ -4,7 +4,6 @@ import (
 	"context"
 	"errors"
 	"net"
-	"os"
 	"strings"
 	"testing"
 	"time"
@@ -127,19 +126,23 @@ func TestLifecycleDeepFailureMatrix(t *testing.T) {
 		cfg := lifecycleFastConfig(t)
 		cfg.Startup.RequireNodeInventoryReach = false
 		cfg.Startup.ShutdownCooldownSeconds = 5
-		if err := state.WriteIntent(cfg.State.IntentPath, state.Intent{
-			State:     state.IntentShutdownComplete,
-			Reason:    "recent",
-			Source:    "test",
-			UpdatedAt: time.Now().UTC(),
-		}); err != nil {
-			t.Fatalf("seed cooldown intent: %v", err)
-		}
-		go func(intentPath string) {
-			time.Sleep(2 * time.Second)
-			_ = os.Remove(intentPath)
-			_ = os.Mkdir(intentPath, 0o755)
-		}(cfg.State.IntentPath)
+		reads := 0
+		restoreRead := state.TestHookSetReadIntentOverride(func(path string) (state.Intent, error) {
+			if path != cfg.State.IntentPath {
+				return state.TestHookReadIntentDefault(path)
+			}
+			reads++
+			if reads == 1 {
+				return state.Intent{
+					State:     state.IntentShutdownComplete,
+					Reason:    "recent",
+					Source:    "test",
+					UpdatedAt: time.Now().UTC().Add(-4 * time.Second),
+				}, nil
+			}
+			return state.Intent{}, errors.New("forced reread failure")
+		})
+		t.Cleanup(restoreRead)
 		orch, _ := newHookOrchestrator(t, cfg, nil, nil)
 		err := orch.Startup(context.Background(), cluster.StartupOptions{Reason: "cooldown-reread"})
 		if err == nil || !strings.Contains(err.Error(), "re-read startup intent after cooldown wait") {
@@ -151,23 +154,28 @@ func TestLifecycleDeepFailureMatrix(t *testing.T) {
 		cfg := lifecycleFastConfig(t)
 		cfg.Startup.RequireNodeInventoryReach = false
 		cfg.Startup.ShutdownCooldownSeconds = 5
-		if err := state.WriteIntent(cfg.State.IntentPath, state.Intent{
-			State:     state.IntentShutdownComplete,
-			Reason:    "recent",
-			Source:    "test",
-			UpdatedAt: time.Now().UTC(),
-		}); err != nil {
-			t.Fatalf("seed cooldown intent: %v", err)
-		}
-		go func(intentPath string) {
-			time.Sleep(2 * time.Second)
-			_ = state.WriteIntent(intentPath, state.Intent{
+		reads := 0
+		restoreRead := state.TestHookSetReadIntentOverride(func(path string) (state.Intent, error) {
+			if path != cfg.State.IntentPath {
+				return state.TestHookReadIntentDefault(path)
+			}
+			reads++
+			if reads == 1 {
+				return state.Intent{
+					State:     state.IntentShutdownComplete,
+					Reason:    "recent",
+					Source:    "test",
+					UpdatedAt: time.Now().UTC().Add(-4 * time.Second),
+				}, nil
+			}
+			return state.Intent{
 				State:     state.IntentShuttingDown,
 				Reason:    "peer-shutdown",
 				Source:    "test",
 				UpdatedAt: time.Now().UTC(),
-			})
-		}(cfg.State.IntentPath)
+			}, nil
+		})
+		t.Cleanup(restoreRead)
 		orch, _ := newHookOrchestrator(t, cfg, nil, nil)
 		err := orch.Startup(context.Background(), cluster.StartupOptions{Reason: "cooldown-peer-active"})
 		if err == nil || !strings.Contains(err.Error(), "shutdown intent became active during cooldown wait") {

testing/state/state_testhooks_quality_test.go

@@ -24,12 +24,36 @@ func TestStateTestHookOverrideSetters(t *testing.T) {
 	}
 	restoreWriteNil()
 
+	restoreReadNil := state.TestHookSetReadIntentOverride(nil)
+	readAfterNil, err := state.ReadIntent(intentPath)
+	if err != nil || readAfterNil.State != state.IntentNormal {
+		t.Fatalf("expected default read intent path after nil override, got %v / %v", readAfterNil, err)
+	}
+	restoreReadNil()
+
+	readOverrideCalled := false
+	restoreRead := state.TestHookSetReadIntentOverride(func(path string) (state.Intent, error) {
+		readOverrideCalled = true
+		return state.Intent{}, errors.New("forced read override")
+	})
+	_, err = state.ReadIntent(intentPath)
+	if err == nil || !strings.Contains(err.Error(), "forced read override") {
+		t.Fatalf("expected forced read override error, got %v", err)
+	}
+	if !readOverrideCalled {
+		t.Fatalf("expected read override to be invoked")
+	}
+	restoreRead()
+	if _, err := state.TestHookReadIntentDefault(intentPath); err != nil {
+		t.Fatalf("expected explicit default read helper to succeed, got %v", err)
+	}
+
 	writeOverrideCalled := false
 	restoreWrite := state.TestHookSetWriteIntentOverride(func(path string, in state.Intent) error {
 		writeOverrideCalled = true
 		return errors.New("forced write override")
 	})
-	err := state.WriteIntent(intentPath, state.Intent{State: state.IntentNormal})
+	err = state.WriteIntent(intentPath, state.Intent{State: state.IntentNormal})
 	if err == nil || !strings.Contains(err.Error(), "forced write override") {
 		t.Fatalf("expected forced write override error, got %v", err)
 	}