8 changed files with 128 additions and 207 deletions
+7 -7
View File
@@ -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。
### 复审修复 R3-06
### 复审修复 R3-07
1. **Python/Java 首次握手失败应停止重连;Java rate_limited 在途计数只减一次**
1. **客户端建群事务内复核创建者**
- 日期:2026-09-30
- 原条款:DEVELOPMENT 第 9 节附录「首次连接超时或握手失败应停止重连,并向 connect() 返回未连接」;issue #70。
- 实际做法:`sdk/python` 的 `connect()` 在 `_handshake_error` 或非 ONLINE 终态时置 `_stop_reconnect`/`_want_connected=False`;内部连接/握手超时改用 `not_connected`(不再用 `busy`)。`sdk/java` 的 `connectSync` 在 `handshakeError` 时同样置 `stopReconnect`;`rate_limited` 路径去掉第二次 `inflightSends--`,只换新 rid 再排队。Python `transport.py` 将 paho 改为惰性导入,便于本机无 paho 时仍跑 FakeTransport 单测。不改 Go/JS 重连公式。
- 原因:原先抛错后工作线程仍按退避重连;Java 限速把在途计数减了两次。
- 备选方案:在后台 `_attempt_connect`/`attemptConnect` 内首次失败即停(否决:应用再次 `connect()` 才应恢复,标志应在对外 `connect` 失败路径统一置位)。
- 影响:仅 `sdk/python`、`sdk/java`;应用需再次调用 `connect()` 才会重连。
- 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。
+9
View File
@@ -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 {
+100
View File
@@ -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)
@@ -143,12 +143,6 @@ public final class Client {
return lastStopCode == null ? "" : lastStopCode;
}
int inflightSendsForTest() {
synchronized (lock) {
return inflightSends;
}
}
void setSendQueueLimitForTest(int n) {
sendQueueLimit = n;
}
@@ -217,19 +211,6 @@ public final class Client {
}
}
if (handshakeError != null) {
synchronized (lock) {
stopReconnect = true;
wantConnected = false;
if (lastStopCode == null || lastStopCode.isEmpty()) {
if (handshakeError instanceof NixMsgException) {
lastStopCode = ((NixMsgException) handshakeError).getCode();
lastStopErr = (NixMsgException) handshakeError;
} else {
lastStopCode = "not_connected";
lastStopErr = new NixMsgException("not_connected", handshakeError.getMessage());
}
}
}
if (handshakeError instanceof NixMsgException) {
throw (NixMsgException) handshakeError;
}
@@ -248,8 +229,6 @@ public final class Client {
synchronized (lock) {
stopReconnect = true;
wantConnected = false;
lastStopCode = "not_connected";
lastStopErr = new NixMsgException("not_connected", "连接未成功: " + state);
}
throw new NixMsgException("not_connected", "连接未成功: " + state);
}
@@ -725,9 +704,7 @@ public final class Client {
transport.publish(Protocol.upTopic(endpointId), Protocol.dumps(hello));
await(p.future, connectTimeoutMs);
if (p.error != null) {
throw p.error instanceof RuntimeException
? (RuntimeException) p.error
: new NixMsgException("not_connected", p.error.getMessage());
throw p.error instanceof RuntimeException ? (RuntimeException) p.error : new NixMsgException("busy", p.error.getMessage());
}
if (p.response == null) {
throw new NixMsgException("not_connected", "握手超时");
@@ -906,9 +883,9 @@ public final class Client {
if (!Boolean.TRUE.equals(frame.get("ok"))) {
Map<String, Object> err = asMap(frame.get("error"));
if ("rate_limited".equals(str(err.get("code"), ""))) {
// 在途计数已在上方减过一次;只换新 rid 再排队,勿再减。
synchronized (lock) {
pending.remove(rid);
inflightSends = Math.max(0, inflightSends - 1);
p.rid = "";
p.response = null;
p.error = null;
@@ -41,7 +41,7 @@ public class K00Test {
}
@Test
public void testK00FirstConnectTimeout() throws Exception {
public void testK00FirstConnectTimeout() {
FakeTransport tr = new FakeTransport();
tr.autoHello = null;
client = new Client(tr, true, Types.DEFAULT_MAX_FRAME, Types.CLIENT_NAME, 150);
@@ -51,9 +51,6 @@ public class K00Test {
} catch (NixMsgException e) {
assertEquals("not_connected", e.getCode());
}
int n = tr.connects.size();
Thread.sleep(350);
assertEquals("首次握手失败后不得自动重连", n, tr.connects.size());
tr.autoHello = new LinkedHashMap<String, Object>();
tr.autoHello.put("server_time_ms", 1750000000000L);
tr.autoHello.put("server_version", "0.1.0");
@@ -66,108 +63,6 @@ public class K00Test {
tr.autoHello.put("session_token", "nst_test_token");
client.connectSync("ws://example.test/mqtt", "ep1", "p", null, false);
assertEquals(ConnectionState.ONLINE, client.getState());
assertTrue(tr.connects.size() > n);
}
@Test
public void testK00RateLimitedInflightOnce() throws Exception {
Types.ReconnectBackoff.disableJitterForTest();
FakeTransport tr = new FakeTransport();
tr.autoSendOk = false;
final List<String> holdRids = new CopyOnWriteArrayList<String>();
final List<String> limitedRids = new CopyOnWriteArrayList<String>();
tr.onUp(new java.util.function.Function<Map<String, Object>, Map<String, Object>>() {
@Override
public Map<String, Object> apply(Map<String, Object> frame) {
if (!"send".equals(String.valueOf(frame.get("type")))) {
return null;
}
String rid = String.valueOf(frame.get("rid"));
String id = String.valueOf(frame.get("id"));
if ("hold".equals(id)) {
holdRids.add(rid);
return null; // 保持在途
}
limitedRids.add(rid);
if (limitedRids.size() == 1) {
Map<String, Object> err = new LinkedHashMap<String, Object>();
err.put("code", "rate_limited");
err.put("message", "slow");
Map<String, Object> resp = new LinkedHashMap<String, Object>();
resp.put("v", 1);
resp.put("type", "resp");
resp.put("rid", rid);
resp.put("ok", false);
resp.put("error", err);
return resp;
}
Map<String, Object> data = new LinkedHashMap<String, Object>();
data.put("id", frame.get("id"));
data.put("send_at_ms", frame.get("send_at_ms"));
data.put("state", "scheduled");
Map<String, Object> resp = new LinkedHashMap<String, Object>();
resp.put("v", 1);
resp.put("type", "resp");
resp.put("rid", rid);
resp.put("ok", true);
resp.put("data", data);
return resp;
}
});
connectOnline(tr);
Thread holder = new Thread(new Runnable() {
@Override
public void run() {
try {
SendOptions opt = new SendOptions();
opt.messageId = "hold";
opt.sendAtMs = 1700000000001L;
client.sendSync(new Target("endpoint", "ep2"), new Body("hold"), opt);
} catch (Exception ignored) {
}
}
});
holder.setDaemon(true);
holder.start();
long deadline = System.currentTimeMillis() + 2000;
while (holdRids.isEmpty() && System.currentTimeMillis() < deadline) {
Thread.sleep(10);
}
assertFalse(holdRids.isEmpty());
assertEquals(1, client.inflightSendsForTest());
final SendOptions opt = new SendOptions();
opt.messageId = "lim";
opt.sendAtMs = 1700000000002L;
Thread sender = new Thread(new Runnable() {
@Override
public void run() {
try {
client.sendSync(new Target("endpoint", "ep2"), new Body("lim"), opt);
} catch (Exception ignored) {
}
}
});
sender.setDaemon(true);
sender.start();
deadline = System.currentTimeMillis() + 3000;
while (limitedRids.size() < 1 && System.currentTimeMillis() < deadline) {
Thread.sleep(10);
}
assertTrue(limitedRids.size() >= 1);
// rate_limited 后只应减 1:仍剩 hold 那一条在途
deadline = System.currentTimeMillis() + 500;
int seen = -1;
while (System.currentTimeMillis() < deadline) {
seen = client.inflightSendsForTest();
if (seen == 1) {
break;
}
Thread.sleep(10);
}
assertEquals("rate_limited 后在途应只减 1", 1, seen);
client.close();
holder.join(1000);
sender.join(1000);
}
@Test
+2 -16
View File
@@ -227,15 +227,6 @@ class Client:
raise NixMsgError("not_connected", "连接超时")
err = self._handshake_error
if err:
with self._lock:
self._stop_reconnect = True
self._want_connected = False
if not self._last_stop_code:
code = getattr(err, "code", None) or "not_connected"
self._last_stop_code = str(code)
self._last_stop_err = err if isinstance(err, NixMsgError) else NixMsgError(
"not_connected", str(err)
)
raise err
if self._state not in (ConnectionState.ONLINE,):
if self._state == ConnectionState.AUTH_FAILED:
@@ -245,11 +236,6 @@ class Client:
)
if self._state == ConnectionState.KICKED:
raise NixMsgError("taken_over", "会话被接管")
with self._lock:
self._stop_reconnect = True
self._want_connected = False
self._last_stop_code = "not_connected"
self._last_stop_err = NixMsgError("not_connected", f"连接未成功: {self._state.value}")
raise NixMsgError("not_connected", f"连接未成功: {self._state.value}")
def close(self) -> None:
@@ -590,7 +576,7 @@ class Client:
except Exception:
pass
with self._lock:
self._handshake_error = NixMsgError("not_connected", "连接超时")
self._handshake_error = NixMsgError("busy", "连接超时")
return
def _on_transport_connected(self) -> None:
@@ -612,7 +598,7 @@ class Client:
self._pending[rid] = pending
self._transport.publish(up_topic(self._endpoint_id), dumps(hello))
if not pending.event.wait(self._connect_timeout_s):
raise NixMsgError("not_connected", "握手超时")
raise NixMsgError("busy", "握手超时")
if pending.error:
raise pending.error
assert pending.response is not None
+7 -22
View File
@@ -8,23 +8,16 @@ from dataclasses import dataclass, field
from typing import Any, Callable, Optional, Protocol
from urllib.parse import urlparse
from paho.mqtt.client import CallbackAPIVersion, Client as PahoClient, MQTT_ERR_SUCCESS, MQTTv5
from paho.mqtt.enums import MQTTErrorCode
from paho.mqtt.reasoncodes import ReasonCode
DownHandler = Callable[[bytes], None]
ConnHandler = Callable[[], None]
DiscHandler = Callable[[Optional[str], bool], None]
# reason_code_str, stop_reconnect
def _require_paho():
"""真实 MQTT 路径才加载 paho;假传输单测不依赖。"""
try:
from paho.mqtt.client import CallbackAPIVersion, Client as PahoClient, MQTT_ERR_SUCCESS, MQTTv5
from paho.mqtt.enums import MQTTErrorCode
from paho.mqtt.reasoncodes import ReasonCode
except ImportError as e:
raise ImportError("需要 paho-mqtt>=2.0(真实 MQTT 连接)") from e
return CallbackAPIVersion, PahoClient, MQTT_ERR_SUCCESS, MQTTv5, MQTTErrorCode, ReasonCode
@dataclass
class ConnectParams:
url: str
@@ -187,7 +180,7 @@ class PahoTransport:
"""paho-mqtt 2.x CallbackAPIVersion.VERSION2。"""
def __init__(self) -> None:
self._client: Any = None
self._client: Optional[PahoClient] = None
self._on_connected: Optional[ConnHandler] = None
self._on_disconnected: Optional[DiscHandler] = None
self._on_down: Optional[DownHandler] = None
@@ -207,7 +200,6 @@ class PahoTransport:
self._on_down = on_down
def connect(self, params: ConnectParams) -> None:
CallbackAPIVersion, PahoClient, _, MQTTv5, _, _ = _require_paho()
self.disconnect()
url = params.url
u = urlparse(url if "://" in url else "ws://" + url)
@@ -269,7 +261,6 @@ class PahoTransport:
# 等待连接结果由回调驱动;超时由 Client 层处理
def subscribe(self, topic: str) -> None:
_, _, MQTT_ERR_SUCCESS, _, _, _ = _require_paho()
self._down_topic = topic
if not self._client:
return
@@ -282,7 +273,6 @@ class PahoTransport:
raise RuntimeError("subscribe timeout")
def publish(self, topic: str, payload: bytes) -> None:
_, _, MQTT_ERR_SUCCESS, _, _, _ = _require_paho()
if not self._client:
raise RuntimeError("not connected")
info = self._client.publish(topic, payload, qos=1)
@@ -348,14 +338,9 @@ def _reason_to_int(reason_code) -> Optional[int]:
return None
if isinstance(reason_code, int):
return reason_code
try:
_, _, _, _, MQTTErrorCode, ReasonCode = _require_paho()
except ImportError:
MQTTErrorCode = () # type: ignore[assignment,misc]
ReasonCode = () # type: ignore[assignment,misc]
if ReasonCode and isinstance(reason_code, ReasonCode):
if isinstance(reason_code, ReasonCode):
return int(reason_code.value)
if MQTTErrorCode and isinstance(reason_code, MQTTErrorCode):
if isinstance(reason_code, MQTTErrorCode):
return int(reason_code)
# paho 偶发其它包装
val = getattr(reason_code, "value", None)
-31
View File
@@ -30,42 +30,11 @@ class K00Tests(unittest.TestCase):
with self.assertRaises(NixMsgError) as cm:
c.connect("ws://example.test/mqtt", "ep1", password="p", wait=True)
self.assertEqual(cm.exception.code, "not_connected")
n = len(tr.connects)
time.sleep(0.35)
self.assertEqual(len(tr.connects), n, "首次失败后不得自动重连")
tr.auto_accept = True
c.connect("ws://example.test/mqtt", "ep1", password="p", wait=True)
self.assertEqual(c.state, ConnectionState.ONLINE)
c.close()
def test_k00_first_hello_fail_stops_reconnect(self) -> None:
"""MQTT 已通但 hello 失败:connect 返回未连接,且后台不再连。"""
tr = FakeTransport()
tr.auto_hello = None
c = Client(transport=tr, connect_timeout_s=0.2)
with self.assertRaises(NixMsgError) as cm:
c.connect("ws://example.test/mqtt", "ep1", password="p", wait=True)
self.assertEqual(cm.exception.code, "not_connected")
n = len(tr.connects)
self.assertGreaterEqual(n, 1)
time.sleep(0.45)
self.assertEqual(len(tr.connects), n, "hello 失败后不得自动重连")
tr.auto_hello = {
"server_time_ms": 1_750_000_000_000,
"server_version": "0.1.0",
"max_body_bytes": 262144,
"max_meta_bytes": 4096,
"max_frame_bytes": 786432,
"max_ttl_seconds": 2592000,
"max_schedule_seconds": 31536000,
"ack_timeout_seconds": 300,
"session_token": "nst_retry",
}
c.connect("ws://example.test/mqtt", "ep1", password="p", wait=True)
self.assertEqual(c.state, ConnectionState.ONLINE)
self.assertGreater(len(tr.connects), n)
c.close()
def test_k00_taken_over_reason(self) -> None:
c, tr = self._connect()
got = []