fix: shutdown admission, recovery identity/order, rollback journal retention, pickup merge

1 files changed, 102 insertions, 0 deletions

diff --git a/internal/persistence/txn_rollback_test.go b/internal/persistence/txn_rollback_test.go
new file mode 100644
index 0000000..2bd7974
--- /dev/null
+++ b/internal/persistence/txn_rollback_test.go
@@ -0,0 +1,102 @@
+package persistence
+
+// Regression: when a transaction's apply fails AND its rollback fails, the
+// journal is the only durable record of the pre-transaction before-images.
+// Deleting it at commit time made the corruption permanent; it must be kept
+// so the next Recover call (or operator repair) can retry the rollback.
+
+import (
+	"os"
+	"path/filepath"
+	"strings"
+	"testing"
+)
+
+func TestCommitKeepsJournalAfterFailedRollback(t *testing.T) {
+	dir := t.TempDir()
+	path := filepath.Join(dir, "character.yaml")
+	if err := os.WriteFile(path, []byte("before"), 0600); err != nil {
+		t.Fatal(err)
+	}
+	journalDir := filepath.Join(dir, "journals")
+	txn := NewTransaction(journalDir)
+	if err := txn.Replace(path, []byte("after"), 0600); err != nil {
+		t.Fatal(err)
+	}
+	// Deterministically force both the apply and the rollback writes to
+	// fail after the snapshot: the target becomes a nonempty directory,
+	// so WriteFile cannot create the file and rollback cannot restore it.
+	if err := os.Remove(path); err != nil {
+		t.Fatal(err)
+	}
+	if err := os.Mkdir(path, 0700); err != nil {
+		t.Fatal(err)
+	}
+	if err := os.WriteFile(filepath.Join(path, "blocker"), []byte("x"), 0600); err != nil {
+		t.Fatal(err)
+	}
+
+	err := txn.Commit()
+	if err == nil || !strings.Contains(err.Error(), "rollback failed") {
+		t.Fatalf("need an apply+rollback failure, got: %v", err)
+	}
+	entries, err := ListJournalFiles(journalDir)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if len(entries) == 0 {
+		t.Fatal("failed rollback deleted the only durable before-image; restart cannot recover the pre-transaction state")
+	}
+}
+
+// TestRecoverRetriesKeptRollbackJournal: after the operator repairs the
+// fault that broke the rollback, the next Recover completes it.
+func TestRecoverRetriesKeptRollbackJournal(t *testing.T) {
+	dir := t.TempDir()
+	path := filepath.Join(dir, "character.yaml")
+	if err := os.WriteFile(path, []byte("before"), 0600); err != nil {
+		t.Fatal(err)
+	}
+	journalDir := filepath.Join(dir, "journals")
+	txn := NewTransaction(journalDir)
+	if err := txn.Replace(path, []byte("after"), 0600); err != nil {
+		t.Fatal(err)
+	}
+	if err := os.Remove(path); err != nil {
+		t.Fatal(err)
+	}
+	if err := os.Mkdir(path, 0700); err != nil {
+		t.Fatal(err)
+	}
+	if err := os.WriteFile(filepath.Join(path, "blocker"), []byte("x"), 0600); err != nil {
+		t.Fatal(err)
+	}
+	if err := txn.Commit(); err == nil {
+		t.Fatal("expected the injected apply/rollback failure")
+	}
+
+	// Operator repair: put the pre-transaction file back.
+	if err := os.RemoveAll(path); err != nil {
+		t.Fatal(err)
+	}
+	if err := os.WriteFile(path, []byte("after"), 0600); err != nil {
+		t.Fatal(err)
+	}
+	if err := Recover(journalDir); err != nil {
+		t.Fatalf("recovery retry failed: %v", err)
+	}
+	data, err := os.ReadFile(path)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if string(data) != "before" {
+		t.Fatalf("recovery did not roll back to the before-image: %q", data)
+	}
+	entries, err := ListJournalFiles(journalDir)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if len(entries) != 0 {
+		t.Fatalf("completed recovery left journals behind: %v", entries)
+	}
+}