fix: 补三处推送漏唤醒让回执与后续 pending 继续推 (#65)

internal/app/message/ack.go

@@ -90,6 +90,7 @@ 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
@@ -156,6 +157,7 @@ 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,
@@ -190,6 +192,9 @@ 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

@@ -127,6 +127,7 @@ 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)
@@ -203,8 +204,10 @@ LIMIT ?`, endpointID, room)
 	}
 
 	var toClaim []pushItem
+	skipped := false
 	for _, it := range items {
 		if it.body == nil {
+			skipped = true
 			continue
 		}
 		msg := protocol.Msg{
@@ -227,6 +230,7 @@ LIMIT ?`, endpointID, room)
 				return rejErr
 			}
 			a.WakePush(it.senderID)
+			skipped = true
 			continue
 		}
 		it.payload = payload
@@ -252,6 +256,10 @@ 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

@@ -0,0 +1,180 @@
+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))
+}