diff --git a/internal/web/service/host.go b/internal/web/service/host.go
index 4da687ebf..4343e9387 100644
--- a/internal/web/service/host.go
+++ b/internal/web/service/host.go
@@ -1,6 +1,7 @@
 package service
 
 import (
+	"encoding/json"
 	"slices"
 	"sort"
 	"strconv"
@@ -182,19 +183,51 @@ func buildHostRows(groupId string, req *entity.HostGroup) []*model.Host {
 	return rows
 }
 
+// A node's group is stored once per address and per adopted inbound, so the
+// response cap does not bound the write: the group's size and a per-node row budget do.
+const (
+	maxAdoptedHostGroupBytes = 16 << 10
+	maxAdoptedHostRows       = 2000
+)
+
+// adoptableHostGroups drops the node-reported groups too large to adopt.
+func adoptableHostGroups(groups []*entity.HostGroup) []*entity.HostGroup {
+	return slices.DeleteFunc(slices.Clone(groups), func(g *entity.HostGroup) bool {
+		if g == nil {
+			return true
+		}
+		wire, err := json.Marshal(g)
+		return err != nil || len(wire) > maxAdoptedHostGroupBytes
+	})
+}
+
+// adoptedHostRowBudget is how many more host rows a node's inbounds may adopt.
+func adoptedHostRowBudget(db *gorm.DB, nodeID int) (int, error) {
+	var held int64
+	err := db.Model(&model.Host{}).
+		Where("inbound_id IN (?)", db.Model(&model.Inbound{}).Select("id").Where("node_id = ?", nodeID)).
+		Count(&held).Error
+	return max(maxAdoptedHostRows-int(held), 0), err
+}
+
 // adoptedHostRows projects a node's host groups onto a freshly adopted central
-// inbound so TLS/SNI/fingerprint overrides survive the node-to-master import.
-func adoptedHostRows(groups []*entity.HostGroup, nodeInboundId, centralInboundId int) []*model.Host {
+// inbound, building at most budget rows; it also returns how many groups that left out.
+func adoptedHostRows(groups []*entity.HostGroup, nodeInboundId, centralInboundId, budget int) ([]*model.Host, int) {
 	var rows []*model.Host
+	skipped := 0
 	for _, g := range groups {
 		if g == nil || !slices.Contains(g.InboundIds, nodeInboundId) {
 			continue
 		}
+		if len(rows)+max(len(g.Hosts), 1) > budget {
+			skipped++
+			continue
+		}
 		scoped := *g
 		scoped.InboundIds = []int{centralInboundId}
 		rows = append(rows, buildHostRows(g.GroupId, &scoped)...)
 	}
-	return rows
+	return rows, skipped
 }
 
 func validateInboundsExist(tx *gorm.DB, inboundIds []int) error {
diff --git a/internal/web/service/inbound_node.go b/internal/web/service/inbound_node.go
index 76c352389..3c543b68c 100644
--- a/internal/web/service/inbound_node.go
+++ b/internal/web/service/inbound_node.go
@@ -654,6 +654,16 @@ func (s *InboundService) setRemoteTrafficLocked(nodeID int, snap *runtime.Traffi
 		}
 	}
 
+	// The host rows the node's inbounds already hold count against this merge's adoption.
+	hostGroups := adoptableHostGroups(snap.HostGroups)
+	hostGroupsSkipped := len(snap.HostGroups) - len(hostGroups)
+	hostRowBudget := 0
+	if len(hostGroups) > 0 {
+		if hostRowBudget, err = adoptedHostRowBudget(db, nodeID); err != nil {
+			return false, err
+		}
+	}
+
 	tx := db.Begin()
 	committed := false
 	defer func() {
@@ -821,9 +831,13 @@ func (s *InboundService) setRemoteTrafficLocked(nodeID int, snap *runtime.Traffi
 			if newIb.Tag != snapIb.Tag {
 				tagToCentral[newIb.Tag] = &newIb
 			}
-			if rows := adoptedHostRows(snap.HostGroups, snapIb.Id, newIb.Id); len(rows) > 0 {
+			rows, skipped := adoptedHostRows(hostGroups, snapIb.Id, newIb.Id, hostRowBudget)
+			hostGroupsSkipped += skipped
+			if len(rows) > 0 {
 				if err := tx.Create(&rows).Error; err != nil {
 					logger.Warningf("setRemoteTraffic: adopt host rows for tag %q failed: %v", newIb.Tag, err)
+				} else {
+					hostRowBudget -= len(rows)
 				}
 			}
 			newInboundIDs[newIb.Id] = struct{}{}
@@ -901,6 +915,13 @@ func (s *InboundService) setRemoteTrafficLocked(nodeID int, snap *runtime.Traffi
 		}
 	}
 
+	if hostGroupsSkipped > 0 {
+		logger.Warningf(
+			"setRemoteTraffic: node %d: left out %d host group adoption(s) — a group is limited to %d bytes and the node's inbounds to %d host rows",
+			nodeID, hostGroupsSkipped, maxAdoptedHostGroupBytes, maxAdoptedHostRows,
+		)
+	}
+
 	// The inbound each reported email now lives under, so a swept inbound's
 	// accumulator rows follow the email instead of being dropped and re-seeded at 0.
 	snapEmailHome := make(map[string]int, len(snapEmailsAll))
diff --git a/internal/web/service/node_host_adopt_test.go b/internal/web/service/node_host_adopt_test.go
index 53906744b..bac184472 100644
--- a/internal/web/service/node_host_adopt_test.go
+++ b/internal/web/service/node_host_adopt_test.go
@@ -1,6 +1,8 @@
 package service
 
 import (
+	"fmt"
+	"strings"
 	"testing"
 
 	"github.com/mhsanaei/3x-ui/v3/internal/database"
@@ -124,3 +126,71 @@ func TestSetRemoteTraffic_SweptInboundTakesItsDependentsAlong(t *testing.T) {
 		t.Fatalf("balancer members after the sweep = %v, want [%d]", balancer.InboundIds, kept.Id)
 	}
 }
+
+// floodSnapshot reports inboundCount new inbounds (ids and ports from firstPort)
+// and makes every group a member of all of them.
+func floodSnapshot(tagPrefix string, firstPort, inboundCount int, groups ...*entity.HostGroup) *runtime.TrafficSnapshot {
+	snap := &runtime.TrafficSnapshot{HostGroups: groups}
+	ids := make([]int, inboundCount)
+	for i := range ids {
+		ids[i] = firstPort + i
+		snap.Inbounds = append(snap.Inbounds, &model.Inbound{
+			Id: firstPort + i, Tag: fmt.Sprintf("%s-%d", tagPrefix, i), Enable: true, Port: firstPort + i, Protocol: model.VLESS, Settings: `{"clients":[]}`,
+		})
+	}
+	for _, g := range groups {
+		g.InboundIds = ids
+	}
+	return snap
+}
+
+func countHostRows(t *testing.T, query string, args ...any) int64 {
+	t.Helper()
+	var n int64
+	if err := database.GetDB().Model(&model.Host{}).Where(query, args...).Count(&n).Error; err != nil {
+		t.Fatalf("count hosts: %v", err)
+	}
+	return n
+}
+
+// A group is stored once per address and per inbound, so a small hosts/list answer
+// must not be able to write an unbounded number of rows or bytes, in one merge or many.
+func TestSetRemoteTraffic_BoundsAdoptedHostRows(t *testing.T) {
+	setupConflictDB(t)
+
+	const nodeID = 7
+	if err := database.GetDB().Create(&model.Node{
+		Id: nodeID, Name: "flood-node", Address: "10.0.0.7", Port: 2053, ApiToken: "t", Guid: "flood-node-guid",
+	}).Error; err != nil {
+		t.Fatalf("create node: %v", err)
+	}
+
+	addresses := make([]string, 50)
+	for i := range addresses {
+		addresses[i] = fmt.Sprintf("h%d.example", i)
+	}
+	snap := floodSnapshot("flood", 30000, 100,
+		&entity.HostGroup{GroupId: "g-big", Hosts: []string{"cdn.example"}, Remark: "big", Sni: strings.Repeat("s", 100_000)},
+		&entity.HostGroup{GroupId: "g-many", Hosts: addresses, Remark: "many", Sni: "sni.example"},
+	)
+
+	svc := InboundService{}
+	if _, err := svc.setRemoteTrafficLocked(nodeID, snap, false, false); err != nil {
+		t.Fatalf("setRemoteTrafficLocked: %v", err)
+	}
+	if big := countHostRows(t, "group_id = ?", "g-big"); big != 0 {
+		t.Fatalf("a 100000-byte group was stored %d times", big)
+	}
+	if total := countHostRows(t, "1 = 1"); total != maxAdoptedHostRows {
+		t.Fatalf("one merge adopted %d host rows, want %d", total, maxAdoptedHostRows)
+	}
+
+	more := floodSnapshot("more", 31000, 10, &entity.HostGroup{GroupId: "g-more", Hosts: addresses, Remark: "more"})
+	more.Inbounds = append(more.Inbounds, snap.Inbounds...)
+	if _, err := svc.setRemoteTrafficLocked(nodeID, more, false, false); err != nil {
+		t.Fatalf("second setRemoteTrafficLocked: %v", err)
+	}
+	if total := countHostRows(t, "1 = 1"); total != maxAdoptedHostRows {
+		t.Fatalf("a second merge raised the node's host rows to %d, want %d", total, maxAdoptedHostRows)
+	}
+}