fix: PublishUp 失败时换 rid 并继续泵发送队列

sdk/go/client_test.go

@@ -156,6 +156,66 @@ func TestFrameTooLargeLocal(t *testing.T) {
 	}
 }
 
+func TestPublishUpFailRetriesWithNewRID(t *testing.T) {
+	fake := NewFakeTransport()
+	fake.FailSendPublishN(1)
+	c := connectFake(t, fake)
+	defer c.Close()
+
+	at := time.UnixMilli(1_700_000_000_000)
+	fixedID := "msg-publish-retry"
+	done := make(chan struct{})
+	var firstRID, secondRID string
+
+	go func() {
+		defer close(done)
+		deadline := time.Now().Add(3 * time.Second)
+		for time.Now().Before(deadline) {
+			sends := fake.FindUp("send")
+			if len(sends) < 2 {
+				time.Sleep(5 * time.Millisecond)
+				continue
+			}
+			first, second := sends[0], sends[1]
+			firstRID, _ = first["rid"].(string)
+			secondRID, _ = second["rid"].(string)
+			if first["id"] != fixedID || second["id"] != fixedID {
+				t.Errorf("id changed: %v -> %v", first["id"], second["id"])
+			}
+			if first["send_at_ms"] != second["send_at_ms"] {
+				t.Errorf("send_at_ms changed: %v -> %v", first["send_at_ms"], second["send_at_ms"])
+			}
+			body1, _ := json.Marshal(first["body"])
+			body2, _ := json.Marshal(second["body"])
+			if string(body1) != string(body2) {
+				t.Errorf("body changed: %s -> %s", body1, body2)
+			}
+			if firstRID == "" || firstRID == secondRID {
+				t.Errorf("rid not regenerated: %q -> %q", firstRID, secondRID)
+			}
+			fake.ReplyOK(secondRID, map[string]any{
+				"id": fixedID, "send_at_ms": second["send_at_ms"], "state": "scheduled",
+			})
+			return
+		}
+		t.Error("timed out waiting for send retry")
+	}()
+
+	ctx, cancel := context.WithTimeout(context.Background(), 3*time.Second)
+	defer cancel()
+	res, err := c.Send(ctx, Target{Kind: "endpoint", ID: "b"}, Body{Enc: "utf8", Data: "hi"}, SendOptions{ID: fixedID, SendAt: &at})
+	<-done
+	if err != nil {
+		t.Fatalf("Send hung or failed: %v", err)
+	}
+	if res.ID != fixedID {
+		t.Fatalf("result id=%q want %q", res.ID, fixedID)
+	}
+	if firstRID == "" || secondRID == "" || firstRID == secondRID {
+		t.Fatalf("rids=%q/%q", firstRID, secondRID)
+	}
+}
+
 func TestResendKeepsIDAndSendAt(t *testing.T) {
 	fake := NewFakeTransport()
 	c := connectFake(t, fake)

sdk/go/fake_transport_test.go

@@ -3,6 +3,7 @@ package nixmsg
 import (
 	"context"
 	"encoding/json"
+	"errors"
 	"sync"
 	"sync/atomic"
 	"time"
@@ -29,6 +30,8 @@ type FakeTransport struct {
 	MaxFrameBytes     int
 	HelloDelay        time.Duration
 	ReceiveMaximumSet bool
+	// failSendLeft 接下来若干次 type=send 的 PublishUp 返回错误（仍记入 up）。
+	failSendLeft int
 }
 
 type fakeConnect struct {
@@ -65,6 +68,13 @@ func (f *FakeTransport) Start(ctx context.Context, cfg transportConfig) error {
 	return nil
 }
 
+// FailSendPublishN 让接下来 n 次 send 帧 PublishUp 失败（hello 等其它类型不受影响）。
+func (f *FakeTransport) FailSendPublishN(n int) {
+	f.mu.Lock()
+	f.failSendLeft = n
+	f.mu.Unlock()
+}
+
 func (f *FakeTransport) PublishUp(payload []byte) error {
 	f.mu.Lock()
 	cp := append([]byte(nil), payload...)
@@ -80,6 +90,17 @@ func (f *FakeTransport) PublishUp(payload []byte) error {
 	if auto && head.Type == "hello" && head.RID != "" {
 		f.replyHello(head.RID)
 	}
+	if head.Type == "send" {
+		f.mu.Lock()
+		fail := f.failSendLeft > 0
+		if fail {
+			f.failSendLeft--
+		}
+		f.mu.Unlock()
+		if fail {
+			return errors.New("publish failed")
+		}
+	}
 	return nil
 }
 

sdk/go/send.go

@@ -202,19 +202,24 @@ func (c *Client) dispatchSend(tr transport, item *sendItem, rid string, payload
 	if err := tr.PublishUp(payload); err != nil {
 		c.mu.Lock()
 		delete(c.pending, rid)
-		if item.epoch == epoch && item.inflight {
-			item.inflight = false
-			if c.inflight > 0 {
-				c.inflight--
-			}
-			if c.stopReconnect {
-				errStop := c.stopErrLocked()
-				c.mu.Unlock()
-				c.finishSend(item, SendResult{}, errStop)
-				return
-			}
+		if item.epoch != epoch || !item.inflight {
+			c.mu.Unlock()
+			return
 		}
+		item.inflight = false
+		if c.inflight > 0 {
+			c.inflight--
+		}
+		if c.stopReconnect {
+			errStop := c.stopErrLocked()
+			c.mu.Unlock()
+			c.finishSend(item, SendResult{}, errStop)
+			return
+		}
+		// 保留 id/body/send_at_ms，换新 rid 后继续泵，避免 Send 永久挂起。
+		c.regenerateSendLocked(item)
 		c.mu.Unlock()
+		c.drainSendQueue()
 		return
 	}