Compare commits
1
Commits
| Author | SHA1 | Date | |
|---|---|---|---|
 Nixevol
|
b6f3bafd1f |
+8
-8
@@ -1719,12 +1719,12 @@ issue #3 未关闭，`feat/fix-3-downlink-deadlock` 未合入 `main`。下面是
|
||||
- 备选方案：为每个未测子项补验收用例（本波不做，避免为变绿放松断言）。
|
||||
- 影响：汇总改为通过 19、部分通过 4（F03/F08/F21/F22）、失败 0。F19 仍引用仓库内 SDK 清单、本波不重跑。
|
||||
|
||||
### ¸´ÉóÐÞ¸´ R3-03
|
||||
### 复审修复 R3-07
|
||||
|
||||
1. **д¶ÓÁÐÂúʱ Close ²»ÔÙÓë enqueue ËÀËø**
|
||||
- ÈÕÆÚ£º2026-09-30
|
||||
- ÔÌõ¿î£ºissue #67£»Í£»úÐè Drain/Close ÔÚ³¬Ê±ÄÚ·µ»Ø£»ÒÑÈë¶ÓÈÎÎñÈÔÓÉд goroutine ´¦ÀíÍê¡£
|
||||
- ʵ¼Ê×ö·¨£ºQueue Ôö¼Ó closing Ðźš£enqueue ÔÚ³Ö sendMu.RLock ʱ select ͬʱµÈ´ý q.ch¡¢closing Óë ctx.Done()£¬Í¨µÀÂúʱ²»ÔÙÎÞÏÞ×èÈû¡£Close ÏÈ Swap(closed) ²¢ close(closing) »½ÐÑÔÚ;·¢ËÍ·½ÊͷŶÁËø£¬ÔÙÔÚдËøÄÚ close(q.ch)£¬×îºóµÈ loop Í˳ö¡£²»ÏòÒÑ¹Ø±Õ channel ·¢ËÍ¡£
|
||||
- ÔÒò£ºÍ¨µÀÂúʱ²¢·¢ Do Õ¼×ŶÁËø¶ÂÔÚ·¢ËÍÉÏ£¬Close µÄдËøÄò»µ½£¬Í£»ú¿¨ËÀ£»Î´Ìá½»µÄÔÚ;дҲ»á¶ª¡£
|
||||
- ±¸Ñ¡·½°¸£ºÈë¶Ó¸ÄΪ·Ç×èÈû£¬ÂúÔòÁ¢¼´ ErrBusy£¨·ñ¾ö£º¸Ä±ä±³Ñ¹ÓïÒ壬Õý³£¸ß·å»áÎóÉËÌá½»£©¡£
|
||||
- Ó°Ï죺½ö internal/store/queue.go Óë²âÊÔ£»²»¸ÄǨÒƱàºÅ¡£
|
||||
1. **客户端建群事务内复核创建者**
|
||||
- 日期：2026-09-30
|
||||
- 原条款：Gitea #71。`group.Create` 在 `Queue.Do` 里直接 `insertMemberTx` 群主，不跑 `endpointCheckTx`；停用/删除与建群抢在事务外对话密码窗口时，可插入已停用甚至刚删除的 `owner_id`。
|
||||
- 实际做法：`Create` 同一写事务里，插入群与群主成员之前对 `actorID` 做与 `createAdmin` 相同的 `endpointCheckTx`；已停用 `endpoint_disabled`，不存在 `invalid_target`，且不插入群。不改 `emit`、不改迁移编号。
|
||||
- 原因：成员侧已有事务内复核，群主侧缺对称校验，竞态会留下非法群主。
|
||||
- 备选方案：事务外再查一次创建者（否决，无法覆盖密码窗口与写事务之间的竞态）。
|
||||
- 影响：创建者在进入写事务前被停用/删除时建群失败且无新群行；与后台建群错误码对齐。
|
||||
|
||||
@@ -147,6 +147,15 @@ func (a *App) Create(ctx context.Context, actorID string, req *protocol.GroupCre
|
||||
|
||||
var inserted []string
|
||||
err := a.db.Queue.Do(ctx, func(tx *sql.Tx) error {
|
||||
if e := endpointCheckTx(tx, actorID); e != nil {
|
||||
if protoCode(e) == protocol.CodeInvalidTarget {
|
||||
return errCode(protocol.CodeInvalidTarget, "owner not found")
|
||||
}
|
||||
if protoCode(e) == protocol.CodeEndpointDisabled {
|
||||
return errCode(protocol.CodeEndpointDisabled, "owner disabled")
|
||||
}
|
||||
return e
|
||||
}
|
||||
var exists int
|
||||
qErr := tx.QueryRow(`SELECT 1 FROM groups WHERE id = ?`, gid).Scan(&exists)
|
||||
if qErr == nil {
|
||||
|
||||
@@ -604,6 +604,106 @@ func TestU02CreateDedupesMembers(t *testing.T) {
|
||||
}
|
||||
}
|
||||
|
||||
// raceTalk 在对话密码校验成功后执行一次 mutate，模拟事务外密码窗口里创建者被停用/删除。
|
||||
type raceTalk struct {
|
||||
inner group.TalkGate
|
||||
mutate func()
|
||||
once sync.Once
|
||||
}
|
||||
|
||||
func (r *raceTalk) CheckTalkPasswordForJoin(ctx context.Context, actorID, targetID, talkPassword, remoteIP string) error {
|
||||
err := r.inner.CheckTalkPasswordForJoin(ctx, actorID, targetID, talkPassword, remoteIP)
|
||||
if err == nil && r.mutate != nil {
|
||||
r.once.Do(r.mutate)
|
||||
}
|
||||
return err
|
||||
}
|
||||
|
||||
func TestR307CreateRejectsActorDisabledOrDeletedBeforeWrite(t *testing.T) {
|
||||
t.Parallel()
|
||||
cases := []struct {
|
||||
name string
|
||||
code string
|
||||
mutate func(ctx context.Context, t *testing.T, idApp *identity.App, db *store.DB)
|
||||
}{
|
||||
{
|
||||
name: "disabled",
|
||||
code: protocol.CodeEndpointDisabled,
|
||||
mutate: func(ctx context.Context, t *testing.T, idApp *identity.App, _ *store.DB) {
|
||||
t.Helper()
|
||||
if err := idApp.Disable(ctx, "alice"); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
},
|
||||
},
|
||||
{
|
||||
name: "deleted",
|
||||
code: protocol.CodeInvalidTarget,
|
||||
mutate: func(ctx context.Context, t *testing.T, idApp *identity.App, _ *store.DB) {
|
||||
t.Helper()
|
||||
if err := idApp.Delete(ctx, "alice"); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
},
|
||||
},
|
||||
}
|
||||
for _, tc := range cases {
|
||||
tc := tc
|
||||
t.Run(tc.name, func(t *testing.T) {
|
||||
t.Parallel()
|
||||
db, err := store.Open(filepath.Join(t.TempDir(), "data"), "FULL")
|
||||
if err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
t.Cleanup(func() { _ = db.Close() })
|
||||
fixed := time.UnixMilli(1_700_000_000_000)
|
||||
locks := auth.NewLoginLocks()
|
||||
idApp := identity.New(identity.Config{
|
||||
DB: db, Hash: auth.NewStubHashPool(), Locks: locks,
|
||||
Sessions: auth.NewSessionTokens(),
|
||||
Now: func() time.Time { return fixed },
|
||||
})
|
||||
gate := &raceTalk{inner: idApp}
|
||||
gApp := group.New(group.Config{
|
||||
DB: db, Talk: gate, MaxGroupMembers: 1000,
|
||||
Now: func() time.Time { return fixed }, DefaultRemoteIP: "1.1.1.1",
|
||||
})
|
||||
ctx := context.Background()
|
||||
insertEP(t, db, "alice", 1)
|
||||
insertEP(t, db, "bob", 1)
|
||||
if err := idApp.SelfSetTalkPassword(ctx, "bob", "secret"); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
gate.mutate = func() { tc.mutate(ctx, t, idApp, db) }
|
||||
|
||||
var before int
|
||||
if qErr := db.Read.QueryRow(`SELECT COUNT(*) FROM groups`).Scan(&before); qErr != nil {
|
||||
t.Fatal(qErr)
|
||||
}
|
||||
_, err = gApp.Create(ctx, "alice", &protocol.GroupCreate{
|
||||
V: protocol.Version, Type: protocol.TypeGroupCreate, RID: "1",
|
||||
ID: "g_r307_" + tc.name, Name: "竞态群",
|
||||
Members: []protocol.GroupMemberIn{{ID: "bob", TalkPassword: "secret"}},
|
||||
})
|
||||
if protoCode(err) != tc.code {
|
||||
t.Fatalf("want %s got %v", tc.code, err)
|
||||
}
|
||||
var after int
|
||||
if qErr := db.Read.QueryRow(`SELECT COUNT(*) FROM groups`).Scan(&after); qErr != nil {
|
||||
t.Fatal(qErr)
|
||||
}
|
||||
if after != before {
|
||||
t.Fatalf("groups leaked: before=%d after=%d", before, after)
|
||||
}
|
||||
var exists int
|
||||
qErr := db.Read.QueryRow(`SELECT 1 FROM groups WHERE id = ?`, "g_r307_"+tc.name).Scan(&exists)
|
||||
if !errors.Is(qErr, sql.ErrNoRows) {
|
||||
t.Fatalf("expected no group row, got exists=%d err=%v", exists, qErr)
|
||||
}
|
||||
})
|
||||
}
|
||||
}
|
||||
|
||||
func TestU02AdminCreateOwnerMustExistAndEnabled(t *testing.T) {
|
||||
t.Parallel()
|
||||
gApp, _, _, db, down := setup(t)
|
||||
|
||||
+9
-22
@@ -36,11 +36,10 @@ type writeJob struct {
|
||||
type Queue struct {
|
||||
db *sql.DB
|
||||
|
||||
ch chan writeJob
|
||||
done chan struct{}
|
||||
closing chan struct{} // Close 时关闭，唤醒持读锁阻塞在发送上的 enqueue
|
||||
closed atomic.Bool
|
||||
sendMu sync.RWMutex
|
||||
ch chan writeJob
|
||||
done chan struct{}
|
||||
closed atomic.Bool
|
||||
sendMu sync.RWMutex
|
||||
|
||||
mu sync.Mutex
|
||||
ready bool
|
||||
@@ -53,16 +52,11 @@ type Queue struct {
|
||||
|
||||
// NewQueue 创建合并写入队列并启动写 goroutine。
|
||||
func NewQueue(db *sql.DB) *Queue {
|
||||
return newQueue(db, queueBuffSize)
|
||||
}
|
||||
|
||||
func newQueue(db *sql.DB, buffSize int) *Queue {
|
||||
q := &Queue{
|
||||
db: db,
|
||||
ch: make(chan writeJob, buffSize),
|
||||
done: make(chan struct{}),
|
||||
closing: make(chan struct{}),
|
||||
ready: true,
|
||||
db: db,
|
||||
ch: make(chan writeJob, queueBuffSize),
|
||||
done: make(chan struct{}),
|
||||
ready: true,
|
||||
}
|
||||
go q.loop()
|
||||
return q
|
||||
@@ -121,15 +115,10 @@ func (q *Queue) enqueue(job writeJob) error {
|
||||
return ErrQueueClosed
|
||||
}
|
||||
q.addPending(1)
|
||||
// 通道满时不得只堵在发送上持有读锁：Close 需要写锁关闭 q.ch。
|
||||
select {
|
||||
case q.ch <- job:
|
||||
q.sendMu.RUnlock()
|
||||
return nil
|
||||
case <-q.closing:
|
||||
q.addPending(-1)
|
||||
q.sendMu.RUnlock()
|
||||
return ErrQueueClosed
|
||||
case <-job.ctx.Done():
|
||||
q.addPending(-1)
|
||||
q.sendMu.RUnlock()
|
||||
@@ -387,13 +376,11 @@ func (q *Queue) Drain(ctx context.Context) error {
|
||||
}
|
||||
|
||||
// Close 关闭队列：不再接受新任务，并等待写 goroutine 处理完已入队任务后退出。
|
||||
// 先关闭 closing 唤醒因通道满而阻塞的发送方并释放读锁，再在无发送者时关闭数据通道，
|
||||
// 避免向已关闭 channel 发送而 panic，也避免与持读锁的 enqueue 死锁。
|
||||
// 在写锁内关闭数据通道，避免并发 Do 向已关闭 channel 发送而 panic。
|
||||
func (q *Queue) Close() error {
|
||||
if q.closed.Swap(true) {
|
||||
return nil
|
||||
}
|
||||
close(q.closing)
|
||||
q.sendMu.Lock()
|
||||
close(q.ch)
|
||||
q.sendMu.Unlock()
|
||||
|
||||
@@ -289,137 +289,3 @@ ON CONFLICT(key) DO UPDATE SET value = excluded.value, updated_at = excluded.upd
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// openSmallQueue 用小缓冲队列替换默认队列，便于测满通道时的 Close/Drain。
|
||||
func openSmallQueue(t *testing.T, buf int) *DB {
|
||||
t.Helper()
|
||||
dir := t.TempDir()
|
||||
db, err := Open(dir, "FULL")
|
||||
if err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
if err := db.Queue.Close(); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
db.Queue = newQueue(db.Write, buf)
|
||||
return db
|
||||
}
|
||||
|
||||
func TestQueueCloseUnblocksFullChannel(t *testing.T) {
|
||||
t.Parallel()
|
||||
const buf = 4
|
||||
db := openSmallQueue(t, buf)
|
||||
defer func() { _ = db.Close() }()
|
||||
q := db.Queue
|
||||
|
||||
hold := make(chan struct{})
|
||||
blockerStarted := make(chan struct{})
|
||||
blockerErr := make(chan error, 1)
|
||||
go func() {
|
||||
blockerErr <- q.Do(context.Background(), func(tx *sql.Tx) error {
|
||||
close(blockerStarted)
|
||||
<-hold
|
||||
return nil
|
||||
})
|
||||
}()
|
||||
<-blockerStarted
|
||||
|
||||
ctx := context.Background()
|
||||
var fillWG sync.WaitGroup
|
||||
for i := 0; i < buf; i++ {
|
||||
fillWG.Add(1)
|
||||
go func() {
|
||||
defer fillWG.Done()
|
||||
_ = q.Do(ctx, func(tx *sql.Tx) error { return nil })
|
||||
}()
|
||||
}
|
||||
deadline := time.Now().Add(2 * time.Second)
|
||||
for q.Len() < buf+1 && time.Now().Before(deadline) {
|
||||
time.Sleep(2 * time.Millisecond)
|
||||
}
|
||||
if q.Len() < buf+1 {
|
||||
t.Fatalf("channel not full: len=%d", q.Len())
|
||||
}
|
||||
|
||||
blockedErr := make(chan error, 1)
|
||||
go func() {
|
||||
blockedErr <- q.Do(ctx, func(tx *sql.Tx) error { return nil })
|
||||
}()
|
||||
// 等额外 Do 堵在 enqueue（pending 超过通道容量+正在执行的一条）。
|
||||
deadline = time.Now().Add(2 * time.Second)
|
||||
for q.Len() < buf+2 && time.Now().Before(deadline) {
|
||||
time.Sleep(2 * time.Millisecond)
|
||||
}
|
||||
|
||||
closeDone := make(chan error, 1)
|
||||
go func() { closeDone <- q.Close() }()
|
||||
|
||||
select {
|
||||
case err := <-blockedErr:
|
||||
if !errors.Is(err, ErrQueueClosed) {
|
||||
t.Fatalf("blocked Do: %v", err)
|
||||
}
|
||||
case <-time.After(2 * time.Second):
|
||||
t.Fatal("Close did not unblock full-channel enqueue")
|
||||
}
|
||||
|
||||
close(hold)
|
||||
select {
|
||||
case err := <-closeDone:
|
||||
if err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
case <-time.After(2 * time.Second):
|
||||
t.Fatal("Close hung after writer released")
|
||||
}
|
||||
<-blockerErr
|
||||
fillWG.Wait()
|
||||
}
|
||||
|
||||
func TestQueueDrainTimeoutWhileWriterBlocked(t *testing.T) {
|
||||
t.Parallel()
|
||||
const buf = 4
|
||||
db := openSmallQueue(t, buf)
|
||||
defer func() { _ = db.Close() }()
|
||||
q := db.Queue
|
||||
|
||||
hold := make(chan struct{})
|
||||
started := make(chan struct{})
|
||||
go func() {
|
||||
_ = q.Do(context.Background(), func(tx *sql.Tx) error {
|
||||
close(started)
|
||||
<-hold
|
||||
return nil
|
||||
})
|
||||
}()
|
||||
<-started
|
||||
|
||||
ctx := context.Background()
|
||||
var wg sync.WaitGroup
|
||||
for i := 0; i < buf; i++ {
|
||||
wg.Add(1)
|
||||
go func() {
|
||||
defer wg.Done()
|
||||
_ = q.Do(ctx, func(tx *sql.Tx) error { return nil })
|
||||
}()
|
||||
}
|
||||
deadline := time.Now().Add(2 * time.Second)
|
||||
for q.Len() < buf+1 && time.Now().Before(deadline) {
|
||||
time.Sleep(2 * time.Millisecond)
|
||||
}
|
||||
|
||||
drainCtx, cancel := context.WithTimeout(context.Background(), 80*time.Millisecond)
|
||||
defer cancel()
|
||||
start := time.Now()
|
||||
err := q.Drain(drainCtx)
|
||||
elapsed := time.Since(start)
|
||||
if !errors.Is(err, context.DeadlineExceeded) {
|
||||
t.Fatalf("Drain err=%v want deadline", err)
|
||||
}
|
||||
if elapsed > 500*time.Millisecond {
|
||||
t.Fatalf("Drain took %s, should return on timeout", elapsed)
|
||||
}
|
||||
|
||||
close(hold)
|
||||
wg.Wait()
|
||||
}
|
||||
|
||||
Reference in New Issue
Block a user