From c73b8fe8fd2d451f92083510109e8cdc35b4b7db Mon Sep 17 00:00:00 2001
From: codex <codex@bstein.dev>
Date: Sat, 3 Oct 2026 02:30:50 -0500
Subject: [PATCH] test: verify repair isolation without prolonged fixture waits

---
 .../daemon_poststart_autorepair_test.go       | 32 +++++++++++++++++++
 ...ooks_vault_lifecycle_branch_matrix_test.go |  9 ++++--
 2 files changed, 38 insertions(+), 3 deletions(-)

diff --git a/internal/service/daemon_poststart_autorepair_test.go b/internal/service/daemon_poststart_autorepair_test.go
index 9737ed1..9d81654 100644
--- a/internal/service/daemon_poststart_autorepair_test.go
+++ b/internal/service/daemon_poststart_autorepair_test.go
@@ -1,7 +1,14 @@
 package service
 
 import (
+	"bytes"
 	"context"
+	"errors"
+	"log"
+	"strings"
+
+	"scm.bstein.dev/bstein/ananke/internal/cluster"
+	"scm.bstein.dev/bstein/ananke/internal/execx"
 	"testing"
 	"time"
 
@@ -55,3 +62,28 @@ func TestDaemonMaybeRunPostStartAutoHeal(t *testing.T) {
 		t.Fatalf("expected second allowed auto-heal call, got %d", calls)
 	}
 }
+
+// TestDaemonRecoveryFailureDoesNotStopMonitoring exercises optional repair isolation.
+// Signature: TestDaemonRecoveryFailureDoesNotStopMonitoring(t *testing.T).
+// Why: a missing or failed repair helper must not terminate UPS monitoring.
+func TestDaemonRecoveryFailureDoesNotStopMonitoring(t *testing.T) {
+	var messages bytes.Buffer
+	d := &Daemon{cfg: config.Config{Startup: config.Startup{PostStartAutoHealSeconds: 60}}, log: log.New(&messages, "", 0)}
+	var last time.Time
+	d.maybeRunPostStartAutoHeal(context.Background(), &last, false)
+	if !last.IsZero() {
+		t.Fatal("missing helper must not consume an interval")
+	}
+	if err := d.runPostStartAutoHeal(context.Background()); err != nil {
+		t.Fatal(err)
+	}
+	d.orch = cluster.New(config.Config{}, &execx.Runner{DryRun: true}, nil, d.log)
+	if err := d.runPostStartAutoHeal(context.Background()); err != nil {
+		t.Fatal(err)
+	}
+	d.postStartAutoHealOverride = func(context.Context) error { return errors.New("synthetic repair failure") }
+	d.maybeRunPostStartAutoHeal(context.Background(), &last, false)
+	if last.IsZero() || !strings.Contains(messages.String(), "synthetic repair failure") {
+		t.Fatal("failed repair must be logged and rate limited")
+	}
+}
diff --git a/testing/orchestrator/hooks_vault_lifecycle_branch_matrix_test.go b/testing/orchestrator/hooks_vault_lifecycle_branch_matrix_test.go
index 88aacc1..622089d 100644
--- a/testing/orchestrator/hooks_vault_lifecycle_branch_matrix_test.go
+++ b/testing/orchestrator/hooks_vault_lifecycle_branch_matrix_test.go
@@ -96,7 +96,7 @@ func TestHookVaultLifecycleBranchMatrix(t *testing.T) {
 			}
 		})
 
-		t.Run("workload-ready-no-value-and-ensure-error", func(t *testing.T) {
+		t.Run("workload-ready-no-value-and-ensure-cancel", func(t *testing.T) {
 			cfg := lifecycleConfig(t)
 			runNoValue := func(ctx context.Context, timeout time.Duration, name string, args ...string) (string, error) {
 				command := name + " " + strings.Join(args, " ")
@@ -111,6 +111,8 @@ func TestHookVaultLifecycleBranchMatrix(t *testing.T) {
 				t.Fatalf("expected no-value readiness branch, ready=%v err=%v", ready, err)
 			}
 
+			waitCtx, cancelWait := context.WithCancel(context.Background())
+			defer cancelWait()
 			runEnsureErr := func(ctx context.Context, timeout time.Duration, name string, args ...string) (string, error) {
 				command := name + " " + strings.Join(args, " ")
 				switch {
@@ -119,14 +121,15 @@ func TestHookVaultLifecycleBranchMatrix(t *testing.T) {
 				case name == "kubectl" && strings.Contains(command, "get pods -o custom-columns"):
 					return "", nil
 				case name == "kubectl" && strings.Contains(command, "rollout status statefulset/victoria-metrics-single-server"):
+					cancelWait()
 					return "", errors.New("rollout failed")
 				default:
 					return lifecycleDispatcher(&commandRecorder{})(ctx, timeout, name, args...)
 				}
 			}
 			orchEnsureErr, _ := newHookOrchestratorWithRunnerMode(t, cfg, false, runEnsureErr, runEnsureErr)
-			if err := orchEnsureErr.TestHookEnsureCriticalStartupWorkloads(context.Background()); err == nil || !strings.Contains(err.Error(), "rollout failed") {
-				t.Fatalf("expected ensureCriticalStartupWorkloads wait error branch, got %v", err)
+			if err := orchEnsureErr.TestHookEnsureCriticalStartupWorkloads(waitCtx); !errors.Is(err, context.Canceled) {
+				t.Fatalf("expected ensureCriticalStartupWorkloads cancellation after failed rollout, got %v", err)
 			}
 		})