fix(onlines): 恢复被 P9 掏空的在线状态存储(第11条)

根因: 1035d00 删本地核心时, inbound_node.go 的在线存储层被一并掏空 (GetOnlineClients 写死空/SetNodeOnlineTree 等空函数/SetRemoteTraffic 末尾喂树整块删除), 而节点轮询与 agent 上报两条到达路径完好, 数据送达即被扔, 客户端/入站在线永远全零。修复: 把原 xray.Process 上的在线存储(算法/锁/grace 20s 语义逐字移植)落到 service 包单例 onlineStore(sharedOnlineStore, 对应原 currentXrayProcess 的单一持有语义); 恢复六个 wrapper + 新增 SetNodeActiveInboundTree; setRemoteTrafficLocked 补回 guidShared 与喂树块(原 f727d04 逻辑, process. 换 s.)。偏离: 本地集在精简主控无本地核心喂养, 仅 mtproto job 折入(该 job 已不被调度, 路径保留无害); GetOnlineClients 对 nil 归一为 []。测试: online_store_test 9 例(-race 全过: 并集/按GUID/替换/清空/ grace 边界/跨实例共享/并发); SetRemoteTraffic 集成 2 例(传统节点 GUID 树+active 树落库、旧节点 flat 回退); 全量 go test 串行复跑失败集合与基线 9 红逐字一致(零新增); service 包单跑唯一红为基线既有的 TestGetXrayConfig 红, 其余全过。
2026-10-22 09:19:20 +00:00
parent f569680182
commit ac39370879
4 changed files with 683 additions and 10 deletions
@@ -0,0 +1,193 @@
+package service
+
+import (
+	"fmt"
+	"sync"
+	"testing"
+	"time"
+)
+
+func sortedSet(in []string) map[string]bool {
+	out := make(map[string]bool, len(in))
+	for _, s := range in {
+		out[s] = true
+	}
+	return out
+}
+
+func requireSet(t *testing.T, what string, got []string, want ...string) {
+	t.Helper()
+	if len(got) != len(want) {
+		t.Fatalf("%s: got %v, want set %v", what, got, want)
+	}
+	set := sortedSet(got)
+	for _, w := range want {
+		if !set[w] {
+			t.Fatalf("%s: got %v, missing %q (want %v)", what, got, w, want)
+		}
+	}
+}
+
+// A node's reported tree must surface both in the flat union and in the
+// GUID-keyed view, deduped against the local set and other nodes.
+func TestOnlineStoreSetTreeUnionAndByGuid(t *testing.T) {
+	st := &onlineStore{}
+	st.SetNodeOnlineTree(1, map[string][]string{
+		"guid-a": {"alice@example.com", "bob@example.com"},
+		"guid-b": {"carol@example.com"},
+	})
+	st.SetNodeOnlineTree(2, map[string][]string{
+		"guid-c": {"bob@example.com", "dave@example.com"},
+	})
+	// Local fold-in overlapping one node email (dedupe check).
+	st.RefreshLocalOnline([]string{"alice@example.com", "local@example.com"}, nil, 1000, 20000)
+
+	requireSet(t, "union", st.GetOnlineClients(),
+		"alice@example.com", "bob@example.com", "carol@example.com", "dave@example.com", "local@example.com")
+
+	merged := st.GetMergedNodeTrees()
+	requireSet(t, "byGuid guid-a", merged["guid-a"], "alice@example.com", "bob@example.com")
+	requireSet(t, "byGuid guid-b", merged["guid-b"], "carol@example.com")
+	requireSet(t, "byGuid guid-c", merged["guid-c"], "bob@example.com", "dave@example.com")
+	if _, ok := merged["guid-absent"]; ok {
+		t.Fatalf("byGuid: unexpected key for absent guid: %v", merged)
+	}
+	requireSet(t, "local", st.GetLocalOnlineClients(), "alice@example.com", "local@example.com")
+}
+
+// The agent controller reports onlines under the literal "agent" GUID key;
+// they must reach the union untouched.
+func TestOnlineStoreAgentKeyTree(t *testing.T) {
+	st := &onlineStore{}
+	st.SetNodeOnlineTree(7, map[string][]string{"agent": {"agentuser@example.com"}})
+	requireSet(t, "union", st.GetOnlineClients(), "agentuser@example.com")
+	requireSet(t, "byGuid agent", st.GetMergedNodeTrees()["agent"], "agentuser@example.com")
+}
+
+// Setting a tree replaces the node's previous contribution; an empty tree
+// deletes it outright.
+func TestOnlineStoreSetTreeReplacesAndEmptyDeletes(t *testing.T) {
+	st := &onlineStore{}
+	st.SetNodeOnlineTree(3, map[string][]string{"g": {"one@example.com"}})
+	st.SetNodeOnlineTree(3, map[string][]string{"g": {"two@example.com"}})
+	requireSet(t, "union after replace", st.GetOnlineClients(), "two@example.com")
+
+	st.SetNodeOnlineTree(3, map[string][]string{})
+	if got := st.GetOnlineClients(); len(got) != 0 {
+		t.Fatalf("union after empty set: got %v, want empty", got)
+	}
+	if got := st.GetMergedNodeTrees(); len(got) != 0 {
+		t.Fatalf("byGuid after empty set: got %v, want empty", got)
+	}
+}
+
+// Active-inbound trees behave like online trees, with per-GUID dedupe.
+func TestOnlineStoreActiveInboundTrees(t *testing.T) {
+	st := &onlineStore{}
+	st.SetNodeActiveInboundTree(1, map[string][]string{"g1": {"in-a", "in-b"}})
+	st.SetNodeActiveInboundTree(2, map[string][]string{"g1": {"in-b", "in-c"}})
+	merged := st.GetMergedActiveInboundTrees()
+	requireSet(t, "active g1", merged["g1"], "in-a", "in-b", "in-c")
+
+	st.RefreshLocalOnline(nil, []string{"local-in"}, 5000, 20000)
+	requireSet(t, "local active", st.GetLocalActiveInbounds(), "local-in")
+}
+
+// Clear drops BOTH of the node's trees and leaves other nodes untouched.
+func TestOnlineStoreClear(t *testing.T) {
+	st := &onlineStore{}
+	st.SetNodeOnlineTree(1, map[string][]string{"g1": {"a@example.com"}})
+	st.SetNodeActiveInboundTree(1, map[string][]string{"g1": {"in-a"}})
+	st.SetNodeOnlineTree(2, map[string][]string{"g2": {"b@example.com"}})
+
+	st.ClearNodeOnlineClients(1)
+	requireSet(t, "union after clear", st.GetOnlineClients(), "b@example.com")
+	if got := st.GetMergedActiveInboundTrees(); len(got) != 0 {
+		t.Fatalf("active after clear: got %v, want empty", got)
+	}
+}
+
+// Local entries live for the grace window, then a prune-only refresh
+// (nil, nil — what NodeTrafficSyncJob calls every tick) ages them out.
+func TestOnlineStoreGraceWindowPrune(t *testing.T) {
+	st := &onlineStore{}
+	const grace = int64(20000)
+	st.RefreshLocalOnline([]string{"stay@example.com"}, []string{"tag-stay"}, 10000, grace)
+	st.RefreshLocalOnline([]string{"fresh@example.com"}, nil, 25000, grace)
+
+	// At t=25000 the first entry is 15s old: still inside the window.
+	requireSet(t, "local within grace", st.GetLocalOnlineClients(), "stay@example.com", "fresh@example.com")
+
+	// Prune-only at t=45001: stay@10000 is 35s old (gone), fresh@25000 is
+	// 20001ms old (also past the strict < grace comparison), tag pruned.
+	st.RefreshLocalOnline(nil, nil, 45001, grace)
+	if got := st.GetLocalOnlineClients(); len(got) != 0 {
+		t.Fatalf("local after prune: got %v, want empty", got)
+	}
+	if got := st.GetLocalActiveInbounds(); len(got) != 0 {
+		t.Fatalf("local active after prune: got %v, want empty", got)
+	}
+}
+
+// Boundary: an entry exactly graceMs old is expired (original uses <).
+func TestOnlineStoreGraceBoundary(t *testing.T) {
+	st := &onlineStore{}
+	st.RefreshLocalOnline([]string{"edge@example.com"}, nil, 0, 20000)
+	st.RefreshLocalOnline(nil, nil, 19999, 20000)
+	requireSet(t, "just inside", st.GetLocalOnlineClients(), "edge@example.com")
+	st.RefreshLocalOnline(nil, nil, 20000, 20000)
+	if got := st.GetLocalOnlineClients(); len(got) != 0 {
+		t.Fatalf("at boundary: got %v, want empty", got)
+	}
+}
+
+// The store is shared across InboundService values: a write through one
+// instance must be visible through another (services are bare values; an
+// instance field would fragment state between writers and readers).
+func TestOnlineStoreSharedAcrossServiceInstances(t *testing.T) {
+	writer := &InboundService{}
+	reader := &InboundService{}
+	const nodeID = 987654
+	t.Cleanup(func() { writer.ClearNodeOnlineClients(nodeID) })
+
+	writer.SetNodeOnlineTree(nodeID, map[string][]string{"guid-shared": {"shared@example.com"}})
+	requireSet(t, "union via other instance", reader.GetOnlineClients(), "shared@example.com")
+	byGuid := reader.GetOnlineClientsByGuid()
+	requireSet(t, "byGuid via other instance", byGuid["guid-shared"], "shared@example.com")
+}
+
+// Concurrent writers/readers must not race or corrupt the union. Run with
+// -race for the full guarantee; even without it, the final state is exact.
+func TestOnlineStoreConcurrentAccess(t *testing.T) {
+	st := &onlineStore{}
+	const nodes = 8
+	var wg sync.WaitGroup
+	for n := 0; n < nodes; n++ {
+		n := n
+		wg.Add(1)
+		go func() {
+			defer wg.Done()
+			guid := fmt.Sprintf("guid-%d", n)
+			for i := 0; i < 50; i++ {
+				st.SetNodeOnlineTree(n, map[string][]string{guid: {fmt.Sprintf("u%d-%d@example.com", n, i)}})
+				_ = st.GetOnlineClients()
+				_ = st.GetMergedNodeTrees()
+				st.RefreshLocalOnline([]string{fmt.Sprintf("local-%d@example.com", n)}, nil, time.Now().UnixMilli(), 60000)
+			}
+		}()
+	}
+	wg.Wait()
+	// Every node's last write holds exactly one email; locals may or may
+	// not have been pruned by a same-ms refresh, so assert on node trees.
+	merged := st.GetMergedNodeTrees()
+	for n := 0; n < nodes; n++ {
+		requireSet(t, fmt.Sprintf("tree %d", n), merged[fmt.Sprintf("guid-%d", n)],
+			fmt.Sprintf("u%d-49@example.com", n))
+	}
+	for n := 0; n < nodes; n++ {
+		st.ClearNodeOnlineClients(n)
+	}
+	if got := st.GetMergedNodeTrees(); len(got) != 0 {
+		t.Fatalf("trees after mass clear: got %v, want empty", got)
+	}
+}