fix: 首次握手失败停止重连并修正 Java 限速在途计数

docs/DEVIATIONS.md

@@ -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 R3-01
+### 复审修复 R3-06
 
-1. **推送 worker 漏唤醒：定时分发回执、超限后续、撤回腾窗**  
+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**  
    - 日期：2026-09-30  
-   - 原条款：Gitea #65；推送 worker 只在 `WakePush` 时跑一轮 `PushPending`。  
+   - 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：`dispatchDueBatch` 本条已分发时除 pending 接收方外始终 `WakePush` 发送方；`PushPending` 遇 `too_large` 等跳过且本轮未占满窗口时再 `WakePush` 当前接收方（不在同一次递归扫表）；`Recall` 成功改成 `recalled` 的接收方各 `WakePush` 一次，已推送撤回仍走 `flushRevokes`。不改 `PublishDown` 签名。  
+   - 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：终态回执、超限后的后续 pending、撤回腾出窗口后都依赖再唤醒，否则空闲在线端收不到。  
+   - 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：在同一次 `PushPending` 里循环扫完整 pending 表（否决，指令要求合并唤醒下一轮）。  
+   - 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 `internal/app/message/`；相关单测见 `review_r3_65_test.go`。
+   - 影响：仅 `sdk/python`、`sdk/java`；应用需再次调用 `connect()` 才会重连。

internal/app/message/ack.go

@@ -90,7 +90,6 @@ func (a *App) Recall(ctx context.Context, senderID string, req *protocol.Recall)
 	nowMs := a.now().UnixMilli()
 	var data protocol.RecallData
 	var revokes []revokeJob
-	var wakeReceivers []string
 	err := a.db.Queue.Do(ctx, func(tx *sql.Tx) error {
 		var seq int64
 		var state string
@@ -157,7 +156,6 @@ WHERE seq = ? AND endpoint_id = ? AND state = 'pending'`,
 				continue
 			}
 			recalled++
-			wakeReceivers = append(wakeReceivers, p.ep)
 			if p.pushed.Valid && p.pushed.String != "" {
 				revokes = append(revokes, revokeJob{
 					endpointID: p.ep,
@@ -192,9 +190,6 @@ SELECT COUNT(*) FROM deliveries WHERE seq = ? AND state IN ('expired','dropped',
 	a.pendingRevoke = append(a.pendingRevoke, revokes...)
 	a.mu.Unlock()
 	a.flushRevokes(ctx)
-	for _, ep := range wakeReceivers {
-		a.WakePush(ep)
-	}
 	return data, nil
 }
 

internal/app/message/push.go

@@ -127,7 +127,6 @@ LIMIT ?`, nowMs, limit)
 			}
 			mu.Lock()
 			n++
-			wake[d.senderID] = struct{}{}
 			mu.Unlock()
 			rows2, qErr := a.db.Read.QueryContext(ctx, `
 SELECT DISTINCT endpoint_id FROM deliveries WHERE seq = ? AND state = 'pending'`, d.seq)
@@ -204,10 +203,8 @@ LIMIT ?`, endpointID, room)
 	}
 
 	var toClaim []pushItem
-	skipped := false
 	for _, it := range items {
 		if it.body == nil {
-			skipped = true
 			continue
 		}
 		msg := protocol.Msg{
@@ -230,7 +227,6 @@ LIMIT ?`, endpointID, room)
 				return rejErr
 			}
 			a.WakePush(it.senderID)
-			skipped = true
 			continue
 		}
 		it.payload = payload
@@ -256,10 +252,6 @@ LIMIT ?`, endpointID, room)
 			a.observeDispatchToPush(it.sendAt, nowMs)
 		}
 	}
-	// 超限等跳过未占满窗口时再唤醒本端，让 worker 下一轮取后续 pending（不在本轮递归扫表）。
-	if skipped && len(claimed) < room {
-		a.WakePush(endpointID)
-	}
 	return a.pushReceipts(ctx, endpointID, connID, nowMs)
 }
 

internal/app/message/review_r3_65_test.go

@@ -1,180 +0,0 @@
-package message
-
-import (
-	"context"
-	"database/sql"
-	"encoding/json"
-	"testing"
-	"time"
-
-	"git.asio.asia/nixevol/NixMsg/internal/app/port"
-	"git.asio.asia/nixevol/NixMsg/internal/protocol"
-)
-
-// 定时到点且接收方离线成终态时，在线发送方经 WakePush 拿到回执。
-func TestR365DispatchDueWakesSenderForReceipt(t *testing.T) {
-	t.Parallel()
-	e := openDeliveryEnv(t, nil)
-	insertEndpoint(t, e.db, "alice", "", 1, 0)
-	insertEndpoint(t, e.db, "bob", "", 1, 0)
-	_ = e.db.Queue.Do(context.Background(), func(tx *sql.Tx) error {
-		_, err := tx.Exec(`UPDATE endpoints SET offline_since = ? WHERE id=?`, e.nowMs-120_000, "bob")
-		return err
-	})
-	ctx := context.Background()
-	aliceConn := LiveConn{ConnID: "c-alice", Ready: true}
-	e.conns.Set("alice", aliceConn)
-	if err := e.app.OnHandshakeComplete(ctx, "alice", aliceConn); err != nil {
-		t.Fatal(err)
-	}
-	t.Cleanup(func() { e.app.stopPushWorker("alice", "") })
-
-	delay := int64(5000)
-	req := baseSend("due-rcpt", "bob")
-	req.DelayMs = &delay
-	if _, err := e.app.Submit(ctx, "alice", port.ConnInfo{}, req); err != nil {
-		t.Fatal(err)
-	}
-	st, _ := e.msgState("alice", "due-rcpt")
-	if st != StateScheduled {
-		t.Fatalf("want scheduled got %s", st)
-	}
-	if e.down.FilterType(protocol.TypeReceipt) != 0 {
-		t.Fatal("receipt before due")
-	}
-
-	e.setNow(e.nowMs + 5000)
-	if _, err := e.app.DispatchDue(ctx, e.nowMs, 10); err != nil {
-		t.Fatal(err)
-	}
-	deadline := time.Now().Add(2 * time.Second)
-	for time.Now().Before(deadline) {
-		if e.down.FilterType(protocol.TypeReceipt) > 0 {
-			return
-		}
-		time.Sleep(20 * time.Millisecond)
-	}
-	t.Fatalf("sender got no receipt after scheduled dispatch, receipts=%d", e.down.FilterType(protocol.TypeReceipt))
-}
-
-// 窗口内首条超限拒收后，同连接后续小消息仍被 worker 推送。
-func TestR365TooLargeThenSmallContinuesPush(t *testing.T) {
-	t.Parallel()
-	e := openDeliveryEnv(t, func(l *Limits) { l.DeliveryWindow = 1 })
-	insertEndpoint(t, e.db, "alice", "", 1, 0)
-	insertEndpoint(t, e.db, "bob", "", 1, 0)
-	ctx := context.Background()
-	// 整帧上限：小消息能过，大正文整帧超限被拒。
-	bobConn := LiveConn{ConnID: "c-bob", Ready: true, MaxReceiveBytes: 200}
-	e.conns.Set("bob", bobConn)
-	if err := e.app.OnHandshakeComplete(ctx, "bob", bobConn); err != nil {
-		t.Fatal(err)
-	}
-	t.Cleanup(func() { e.app.stopPushWorker("bob", "") })
-
-	big := baseSend("big-skip", "bob")
-	b := make([]byte, 400)
-	for i := range b {
-		b[i] = 'A'
-	}
-	big.Body.Data = string(b)
-	if _, err := e.app.Submit(ctx, "alice", port.ConnInfo{}, big); err != nil {
-		t.Fatal(err)
-	}
-	small := baseSend("small-ok", "bob")
-	small.Body.Data = "hi"
-	if _, err := e.app.Submit(ctx, "alice", port.ConnInfo{}, small); err != nil {
-		t.Fatal(err)
-	}
-
-	deadline := time.Now().Add(2 * time.Second)
-	for time.Now().Before(deadline) {
-		foundSmall := false
-		for _, p := range e.down.Snapshots() {
-			if payloadType(p.Payload) != protocol.TypeMsg {
-				continue
-			}
-			var m struct {
-				ID string `json:"id"`
-			}
-			_ = json.Unmarshal(p.Payload, &m)
-			if m.ID == "small-ok" {
-				foundSmall = true
-			}
-		}
-		if foundSmall {
-			seqBig := e.seqOf("alice", "big-skip")
-			st, reason := e.deliveryState(seqBig, "bob")
-			if st != DeliveryRejected || reason != ReasonTooLarge {
-				t.Fatalf("big: %s/%s", st, reason)
-			}
-			return
-		}
-		time.Sleep(20 * time.Millisecond)
-	}
-	t.Fatalf("small msg not pushed after too_large; msgs=%d snapshots=%d",
-		e.down.FilterType(protocol.TypeMsg), len(e.down.Snapshots()))
-}
-
-// 撤回已推在途消息后，同连接更晚的 pending 继续推。
-func TestR365RecallFreesWindowForLaterPending(t *testing.T) {
-	t.Parallel()
-	e := openDeliveryEnv(t, func(l *Limits) { l.DeliveryWindow = 1 })
-	insertEndpoint(t, e.db, "alice", "", 1, 0)
-	insertEndpoint(t, e.db, "bob", "", 1, 0)
-	ctx := context.Background()
-	bobConn := LiveConn{ConnID: "c-bob", Ready: true}
-	e.conns.Set("bob", bobConn)
-	if err := e.app.OnHandshakeComplete(ctx, "bob", bobConn); err != nil {
-		t.Fatal(err)
-	}
-	t.Cleanup(func() { e.app.stopPushWorker("bob", "") })
-
-	if _, err := e.app.Submit(ctx, "alice", port.ConnInfo{}, baseSend("first-in-flight", "bob")); err != nil {
-		t.Fatal(err)
-	}
-	if _, err := e.app.Submit(ctx, "alice", port.ConnInfo{}, baseSend("second-wait", "bob")); err != nil {
-		t.Fatal(err)
-	}
-
-	deadline := time.Now().Add(2 * time.Second)
-	for time.Now().Before(deadline) {
-		if e.down.FilterType(protocol.TypeMsg) >= 1 {
-			break
-		}
-		time.Sleep(10 * time.Millisecond)
-	}
-	if e.down.FilterType(protocol.TypeMsg) < 1 {
-		t.Fatal("first msg not pushed")
-	}
-	seq1 := e.seqOf("alice", "first-in-flight")
-	var pushed sql.NullString
-	_ = e.db.Read.QueryRow(`SELECT pushed_conn FROM deliveries WHERE seq=? AND endpoint_id='bob'`, seq1).Scan(&pushed)
-	if !pushed.Valid {
-		t.Fatal("first not in-flight")
-	}
-
-	if _, err := e.app.Recall(ctx, "alice", &protocol.Recall{
-		V: protocol.Version, Type: protocol.TypeRecall, RID: "r1", ID: "first-in-flight",
-	}); err != nil {
-		t.Fatal(err)
-	}
-
-	deadline = time.Now().Add(2 * time.Second)
-	for time.Now().Before(deadline) {
-		for _, p := range e.down.Snapshots() {
-			if payloadType(p.Payload) != protocol.TypeMsg {
-				continue
-			}
-			var m struct {
-				ID string `json:"id"`
-			}
-			_ = json.Unmarshal(p.Payload, &m)
-			if m.ID == "second-wait" {
-				return
-			}
-		}
-		time.Sleep(20 * time.Millisecond)
-	}
-	t.Fatalf("second pending not pushed after recall; msgs=%d", e.down.FilterType(protocol.TypeMsg))
-}

sdk/java/src/main/java/asia/asio/nixmsg/Client.java

@@ -143,6 +143,12 @@ public final class Client {
         return lastStopCode == null ? "" : lastStopCode;
     }
 
+    int inflightSendsForTest() {
+        synchronized (lock) {
+            return inflightSends;
+        }
+    }
+
     void setSendQueueLimitForTest(int n) {
         sendQueueLimit = n;
     }
@@ -211,6 +217,19 @@ 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;
             }
@@ -229,6 +248,8 @@ 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);
         }
@@ -704,7 +725,9 @@ 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("busy", p.error.getMessage());
+                throw p.error instanceof RuntimeException
+                        ? (RuntimeException) p.error
+                        : new NixMsgException("not_connected", p.error.getMessage());
             }
             if (p.response == null) {
                 throw new NixMsgException("not_connected", "握手超时");
@@ -883,9 +906,9 @@ 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;

sdk/java/src/test/java/asia/asio/nixmsg/K00Test.java

@@ -41,7 +41,7 @@ public class K00Test {
     }
 
     @Test
-    public void testK00FirstConnectTimeout() {
+    public void testK00FirstConnectTimeout() throws Exception {
         FakeTransport tr = new FakeTransport();
         tr.autoHello = null;
         client = new Client(tr, true, Types.DEFAULT_MAX_FRAME, Types.CLIENT_NAME, 150);
@@ -51,6 +51,9 @@ 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");
@@ -63,6 +66,108 @@ 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

sdk/python/src/nixmsg/client.py

@@ -227,6 +227,15 @@ 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:
@@ -236,6 +245,11 @@ 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:
@@ -576,7 +590,7 @@ class Client:
             except Exception:
                 pass
             with self._lock:
-                self._handshake_error = NixMsgError("busy", "连接超时")
+                self._handshake_error = NixMsgError("not_connected", "连接超时")
             return
 
     def _on_transport_connected(self) -> None:
@@ -598,7 +612,7 @@ 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("busy", "握手超时")
+                raise NixMsgError("not_connected", "握手超时")
             if pending.error:
                 raise pending.error
             assert pending.response is not None

sdk/python/src/nixmsg/transport.py

@@ -8,16 +8,23 @@ 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
@@ -180,7 +187,7 @@ class PahoTransport:
     """paho-mqtt 2.x CallbackAPIVersion.VERSION2。"""
 
     def __init__(self) -> None:
-        self._client: Optional[PahoClient] = None
+        self._client: Any = None
         self._on_connected: Optional[ConnHandler] = None
         self._on_disconnected: Optional[DiscHandler] = None
         self._on_down: Optional[DownHandler] = None
@@ -200,6 +207,7 @@ 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)
@@ -261,6 +269,7 @@ 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
@@ -273,6 +282,7 @@ 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)
@@ -338,9 +348,14 @@ def _reason_to_int(reason_code) -> Optional[int]:
         return None
     if isinstance(reason_code, int):
         return reason_code
-    if isinstance(reason_code, ReasonCode):
+    try:
+        _, _, _, _, MQTTErrorCode, ReasonCode = _require_paho()
+    except ImportError:
+        MQTTErrorCode = ()  # type: ignore[assignment,misc]
+        ReasonCode = ()  # type: ignore[assignment,misc]
+    if ReasonCode and isinstance(reason_code, ReasonCode):
         return int(reason_code.value)
-    if isinstance(reason_code, MQTTErrorCode):
+    if MQTTErrorCode and isinstance(reason_code, MQTTErrorCode):
         return int(reason_code)
     # paho 偶发其它包装
     val = getattr(reason_code, "value", None)

sdk/python/tests/test_k00.py

@@ -30,11 +30,42 @@ 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 = []