fix: make WAL segment directory fsync failure fatal (C6)

Per design §3.2 line 248-272, segment directory fsync is a hard
requirement for durable-ready state, not best-effort. rename is atomic
in memory but not guaranteed to survive power loss without a directory
fsync. The previous code silently swallowed both os.Open(dir) and
dirFD.Sync() errors, leaving WAL writer to confirm batches as durable
when their segment might not exist after a crash.

Failure propagation:
- Initial segment creation: NewSegmentWriter fails -> NewSegmentManager
  fails -> DB.Open fails (user sees error, no data promise violated).
- Rotation during AppendBatch: NewSegmentWriter fails -> AppendBatch
  fails -> WalWriter.stopWithError(ErrCommitUnknown) -> write-stopped
  (per design line 272).

Changes:
- wal/segment_writer.go: extract dirFsync helper (Open -> f.Stat ->
  IsDir -> f.Sync, avoiding TOCTOU window), replace silent swallow with
  fatal error; on failure clean up resources (fd.Close + os.Remove) and
  surface cleanup errors via errors.Join so nothing is silently lost.
- wal/dir_fsync_test.go (new): unit test the helper with valid dir,
  non-existent dir (fails at os.Open), and not-a-dir (fails at IsDir).
- wal/segment_writer_test.go: add TestNewSegmentWriterDirFsyncFailure
  (injects failure via package-level dirFsyncFn override; documents the
  not-parallel-safe constraint), TestNewSegmentWriterNormalPathStillWorks
  (regression), and TestNewSegmentWriterRetryAfterDirFsyncFailure
  (verifies cleanup is effective for retry).
- wal/segment_manager_test.go: add TestSegmentManagerRotateFailsOnDirFsyncFailure
  (fills segment until rotation triggers, injects failure, verifies
  propagation through AppendBatch path) and TestNewSegmentManagerFailsOnDirFsyncFailure
  (covers the DB.Open failure path).

dirFsyncFn injection note: tests that override this package-level var
must not use t.Parallel(). All existing wal tests run serially within
the package; this is the lightest mechanism that doesn't require
interface indirection in production code.

Verified: each new test fails on pre-fix code (silent swallow returned
nil error) and passes after the fix. Full suite green including
go test -race ./... .

Audit context: docs/audit-3.2.md C6 (Oracle-verified bg_ef425776).
This commit is contained in:
dailz
2026-06-15 14:25:47 +08:00
parent 58d3bc92d7
commit 0739966e55
6 changed files with 775 additions and 4 deletions
+42
View File
@@ -0,0 +1,42 @@
package wal
import (
"fmt"
"os"
)
// dirFsyncFn is the package-level indirection over dirFsync so tests can
// inject failures without interface plumbing in production code.
//
// NOT PARALLEL-SAFE: tests that override this must not use t.Parallel().
// All existing wal tests run serially within the package.
var dirFsyncFn = dirFsync
// dirFsync opens the directory and fsyncs it. Required for durable-ready
// state per design §3.2 line 258. This is a hard requirement, not
// best-effort: rename is atomic in memory but not guaranteed to survive
// power loss without a directory fsync.
//
// Order: os.Open → f.Stat → IsDir → f.Sync. The open-then-stat sequence
// avoids the TOCTOU window between a separate os.Stat and os.Open, and
// ensures IsDir is checked against the actually-opened file.
func dirFsync(dir string) error {
f, err := os.Open(dir)
if err != nil {
return fmt.Errorf("open dir %q: %w", dir, err)
}
defer f.Close()
info, err := f.Stat()
if err != nil {
return fmt.Errorf("stat dir %q: %w", dir, err)
}
if !info.IsDir() {
return fmt.Errorf("path %q is not a directory", dir)
}
if err := f.Sync(); err != nil {
return fmt.Errorf("fsync dir %q: %w", dir, err)
}
return nil
}
+39
View File
@@ -0,0 +1,39 @@
package wal
import (
"os"
"strings"
"testing"
)
func TestDirFsyncSuccess(t *testing.T) {
dir := t.TempDir()
if err := dirFsync(dir); err != nil {
t.Errorf("dirFsync on valid dir: %v", err)
}
}
func TestDirFsyncNonExistentDir(t *testing.T) {
err := dirFsync("/nonexistent/path/that/should/not/exist")
if err == nil {
t.Fatal("expected error on non-existent dir")
}
if !strings.Contains(err.Error(), "open dir") {
t.Errorf("error should mention 'open dir', got: %v", err)
}
}
func TestDirFsyncNotADirectory(t *testing.T) {
dir := t.TempDir()
filePath := dir + "/notadir"
if err := os.WriteFile(filePath, []byte("x"), 0o644); err != nil {
t.Fatalf("WriteFile: %v", err)
}
err := dirFsync(filePath)
if err == nil {
t.Fatal("expected error when fsyncing a file as dir")
}
if !strings.Contains(err.Error(), "not a directory") {
t.Errorf("error should mention 'not a directory', got: %v", err)
}
}
+66
View File
@@ -1,8 +1,10 @@
package wal
import (
"errors"
"os"
"path/filepath"
"strings"
"testing"
"github.com/dailz/go-kv/config"
@@ -239,3 +241,67 @@ func itoa(n uint64) string {
}
return string(buf[i:])
}
// Regression guard for C6: NewSegmentManager must propagate dir fsync
// failure from initial segment creation. This is the DB.Open failure path.
func TestNewSegmentManagerFailsOnDirFsyncFailure(t *testing.T) {
dir := t.TempDir()
cfg := tinyWalConfig()
orig := dirFsyncFn
dirFsyncFn = func(string) error { return errors.New("simulated dir fsync failure") }
t.Cleanup(func() { dirFsyncFn = orig })
sm, err := NewSegmentManager(dir, 0, 0, cfg)
if err == nil {
if sm != nil {
sm.Close()
}
t.Fatal("NewSegmentManager: expected error on dir fsync failure, got nil")
}
if !strings.Contains(err.Error(), "create initial segment") {
t.Errorf("error should be wrapped as 'create initial segment', got: %v", err)
}
}
// Regression guard for C6: SegmentManager.AppendBatch must propagate
// rotation failure (which now includes dir fsync failure) as error.
//
// Note: C8 (segment_manager.go:64-66 passes byte offset as startSequence)
// makes multi-segment recovery broken, but this test only verifies error
// propagation through AppendBatch; it does not exercise recovery.
func TestSegmentManagerRotateFailsOnDirFsyncFailure(t *testing.T) {
dir := t.TempDir()
cfg := tinyWalConfig()
sm, err := NewSegmentManager(dir, 0, 0, cfg)
if err != nil {
t.Fatalf("NewSegmentManager: %v", err)
}
defer sm.Close()
encoded, err := EncodeWalBatch(0, []*WalEntry{
{OpType: OpPut, ValueKind: VKInline, Key: []byte("k"), Value: []byte("v")},
})
if err != nil {
t.Fatalf("EncodeWalBatch: %v", err)
}
// Fill the active segment until next AppendBatch would trigger rotation.
// segment_manager.go:62 triggers rotate when
// RemainingPayload() < len(encoded) + 2*PhysicalRecordHeaderSize
worstCaseSize := uint64(len(encoded)) + 2*uint64(PhysicalRecordHeaderSize)
for sm.RemainingPayload() >= worstCaseSize {
if err := sm.AppendBatch(encoded); err != nil {
t.Fatalf("fill AppendBatch: %v", err)
}
}
orig := dirFsyncFn
dirFsyncFn = func(string) error { return errors.New("simulated dir fsync failure") }
t.Cleanup(func() { dirFsyncFn = orig })
if err := sm.AppendBatch(encoded); err == nil {
t.Fatal("AppendBatch: expected rotation failure, got nil")
}
}
+13 -4
View File
@@ -1,6 +1,7 @@
package wal
import (
"errors"
"fmt"
"os"
"path/filepath"
@@ -82,10 +83,18 @@ func NewSegmentWriter(
return nil, fmt.Errorf("wal: open segment file for append: %w", err)
}
// Sync directory to make rename durable (best-effort on Linux).
if dirFD, derr := os.Open(dir); derr == nil {
dirFD.Sync()
dirFD.Close()
// Per design §3.2 line 258, directory fsync is a hard requirement for
// durable-ready. Without it, the rename above is not guaranteed to survive
// power loss, violating the Always-mode "no loss of acknowledged writes"
// promise.
if err := dirFsyncFn(dir); err != nil {
closeErr := fd.Close()
removeErr := os.Remove(finalPath)
if closeErr != nil || removeErr != nil {
cleanup := errors.Join(closeErr, removeErr)
return nil, fmt.Errorf("wal: fsync directory after segment rename (cleanup: %v): %w", cleanup, err)
}
return nil, fmt.Errorf("wal: fsync directory after segment rename: %w", err)
}
return &SegmentWriter{
+98
View File
@@ -1,9 +1,11 @@
package wal
import (
"errors"
"fmt"
"os"
"path/filepath"
"strings"
"testing"
"github.com/dailz/go-kv/config"
@@ -344,3 +346,99 @@ func TestSegmentWriterSync(t *testing.T) {
t.Fatalf("Close: %v", err)
}
}
// Regression guard for C6: NewSegmentWriter must return an error when
// directory fsync fails, instead of silently succeeding. Per design §3.2
// line 272, segment must NOT become active if durable-ready fails.
//
// dirFsyncFn is a package-level var; tests that override it must not use
// t.Parallel(). All wal tests run serially within the package.
func TestNewSegmentWriterDirFsyncFailure(t *testing.T) {
dir := t.TempDir()
cfg := config.Defaults()
orig := dirFsyncFn
dirFsyncFn = func(string) error { return errors.New("simulated dir fsync failure") }
t.Cleanup(func() { dirFsyncFn = orig })
sw, err := NewSegmentWriter(dir, 0, 0, &cfg)
if err == nil {
if sw != nil {
sw.Close()
}
t.Fatal("NewSegmentWriter: expected error on dir fsync failure, got nil")
}
if !strings.Contains(err.Error(), "fsync directory") {
t.Errorf("error should mention 'fsync directory', got: %v", err)
}
entries, err := os.ReadDir(dir)
if err != nil {
t.Fatalf("ReadDir: %v", err)
}
for _, e := range entries {
name := e.Name()
if strings.Contains(name, "segment-0") {
t.Errorf("segment file should be cleaned up, found: %s", name)
}
}
}
// Regression guard for C6: ensure normal path still works after the fix.
func TestNewSegmentWriterNormalPathStillWorks(t *testing.T) {
dir := t.TempDir()
cfg := config.Defaults()
sw, err := NewSegmentWriter(dir, 0, 0, &cfg)
if err != nil {
t.Fatalf("NewSegmentWriter normal path: %v", err)
}
defer sw.Close()
if _, err := os.Stat(sw.SegmentPath()); err != nil {
t.Errorf("segment file should exist: %v", err)
}
}
// Regression guard for C6: after a failed NewSegmentWriter due to dir
// fsync, retrying with fsync restored must succeed and not leak state.
func TestNewSegmentWriterRetryAfterDirFsyncFailure(t *testing.T) {
dir := t.TempDir()
cfg := config.Defaults()
orig := dirFsyncFn
dirFsyncFn = func(string) error { return errors.New("simulated") }
_, err := NewSegmentWriter(dir, 0, 0, &cfg)
if err == nil {
t.Fatal("expected first NewSegmentWriter to fail")
}
dirFsyncFn = orig
sw, err := NewSegmentWriter(dir, 0, 0, &cfg)
if err != nil {
t.Fatalf("retry NewSegmentWriter: %v", err)
}
defer sw.Close()
entries, err := os.ReadDir(dir)
if err != nil {
t.Fatalf("ReadDir: %v", err)
}
tmpCount := 0
walCount := 0
for _, e := range entries {
if strings.HasSuffix(e.Name(), ".tmp") {
tmpCount++
}
if strings.HasSuffix(e.Name(), ".wal") {
walCount++
}
}
if tmpCount != 0 {
t.Errorf("leftover .tmp files: %d", tmpCount)
}
if walCount != 1 {
t.Errorf("expected exactly 1 .wal file (from successful retry), got %d", walCount)
}
}