Scoreboard/server/internal/realtime/hub_test.go

package realtime

import (
	"context"
	"database/sql"
	"encoding/json"
	"fmt"
	"io"
	"log/slog"
	"net/http/httptest"
	"path/filepath"
	"reflect"
	"strings"
	"testing"
	"time"

	"github.com/coder/websocket"
	_ "modernc.org/sqlite"

	"git.yeij.top/AskaEth/Scoreboard/server/internal/store"
)

// ---------- 测试脚手架 ----------

type testEnv struct {
	hub *Hub
	db  *sql.DB
	srv *httptest.Server
}

func newTestEnv(t *testing.T) *testEnv {
	t.Helper()

	db := newTestDB(t)
	h := NewHub(db, slog.New(slog.NewTextHandler(io.Discard, &slog.HandlerOptions{Level: slog.LevelError})))
	srv := httptest.NewServer(h.Handler())

	t.Cleanup(func() {
		srv.Close()
		_ = h.Close()
		_ = db.Close()
	})
	return &testEnv{hub: h, db: db, srv: srv}
}

func newTestDB(t *testing.T) *sql.DB {
	t.Helper()

	path := filepath.Join(t.TempDir(), "event.db")
	db, err := sql.Open("sqlite", "file:"+path+
		"?_pragma=journal_mode(WAL)&_pragma=foreign_keys(1)&_pragma=busy_timeout(5000)")
	if err != nil {
		t.Fatalf("打开测试库: %v", err)
	}
	t.Cleanup(func() { _ = db.Close() })

	if err := store.MigrateEvent(db); err != nil {
		t.Fatalf("迁移测试库: %v", err)
	}
	return db
}

func (e *testEnv) dial(t *testing.T, hello *helloPayload) *wsClient {
	t.Helper()

	ctx, cancel := context.WithTimeout(context.Background(), 30*time.Second)
	url := "ws" + strings.TrimPrefix(e.srv.URL, "http") + "/ws"
	conn, _, err := websocket.Dial(ctx, url, nil)
	if err != nil {
		cancel()
		t.Fatalf("连接 /ws: %v", err)
	}

	c := &wsClient{t: t, ctx: ctx, cancel: cancel, conn: conn, msgs: make(chan wsMessage, 256)}
	go c.readLoop()
	t.Cleanup(c.close)
	if hello != nil {
		c.send(EventHello, *hello)
	}
	return c
}

type wsMessage struct {
	event string
	data  json.RawMessage
	err   error
}

type wsClient struct {
	t      *testing.T
	ctx    context.Context
	cancel context.CancelFunc
	conn   *websocket.Conn
	msgs   chan wsMessage
}

// readLoop 单独读取：测试里等待「没有消息」时不能靠取消 Read 的 context，
// 那会把整条连接关掉，后续用例就没法继续读了。
func (c *wsClient) readLoop() {
	defer close(c.msgs)
	for {
		_, data, err := c.conn.Read(c.ctx)
		if err != nil {
			c.msgs <- wsMessage{err: err}
			return
		}
		var env envelope
		if err := json.Unmarshal(data, &env); err != nil {
			c.msgs <- wsMessage{err: err}
			return
		}
		c.msgs <- wsMessage{event: env.Event, data: env.Data}
	}
}

func (c *wsClient) close() {
	_ = c.conn.CloseNow()
	c.cancel()
}

func (c *wsClient) send(event string, data any) {
	c.t.Helper()
	msg, err := encode(event, data)
	if err != nil {
		c.t.Fatalf("编码 %s: %v", event, err)
	}
	ctx, cancel := context.WithTimeout(c.ctx, 5*time.Second)
	defer cancel()
	if err := c.conn.Write(ctx, websocket.MessageText, msg); err != nil {
		c.t.Fatalf("发送 %s: %v", event, err)
	}
}

// recvWithin 在 d 内取一条消息；超时返回 ok=false（连接保持可用）。
func (c *wsClient) recvWithin(d time.Duration) (string, json.RawMessage, bool) {
	c.t.Helper()
	timer := time.NewTimer(d)
	defer timer.Stop()

	select {
	case msg, ok := <-c.msgs:
		if !ok || msg.err != nil {
			c.t.Fatalf("连接已断开: %v", msg.err)
		}
		return msg.event, msg.data, true
	case <-timer.C:
		return "", nil, false
	}
}

// recvEvent 等待指定事件，跳过心跳等无关消息。
func (c *wsClient) recvEvent(want string) json.RawMessage {
	c.t.Helper()
	deadline := time.Now().Add(5 * time.Second)
	for time.Now().Before(deadline) {
		ev, data, ok := c.recvWithin(time.Until(deadline))
		if !ok {
			break
		}
		if ev == want {
			return data
		}
	}
	c.t.Fatalf("等待事件 %s 超时", want)
	return nil
}

func decode[T any](t *testing.T, raw json.RawMessage) T {
	t.Helper()
	var v T
	if err := json.Unmarshal(raw, &v); err != nil {
		t.Fatalf("解析载荷 %s: %v", raw, err)
	}
	return v
}

func insertChange(t *testing.T, db *sql.DB, entity string, entityID int64, action, payload string) int64 {
	t.Helper()
	res, err := db.Exec(
		`INSERT INTO change_log(entity_type, entity_id, action, payload_json) VALUES (?, ?, ?, ?)`,
		entity, entityID, action, payload)
	if err != nil {
		t.Fatalf("写入 change_log: %v", err)
	}
	id, err := res.LastInsertId()
	if err != nil {
		t.Fatalf("读取 change_log id: %v", err)
	}
	return id
}

func loadChanges(t *testing.T, h *Hub, since int64, limit int) []Change {
	t.Helper()
	changes, err := h.LoadChanges(context.Background(), since, limit)
	if err != nil {
		t.Fatalf("读取增量变更: %v", err)
	}
	return changes
}

func waitFor(t *testing.T, what string, cond func() bool) {
	t.Helper()
	deadline := time.Now().Add(3 * time.Second)
	for time.Now().Before(deadline) {
		if cond() {
			return
		}
		time.Sleep(5 * time.Millisecond)
	}
	t.Fatalf("等待条件超时: %s", what)
}

func ptrInt64(v int64) *int64 { return &v }

// ---------- 客户端状态归约（模拟前端按 version 应用变更） ----------

type clientState map[string]json.RawMessage

func stateKey(ch Change) string { return fmt.Sprintf("%s/%d", ch.Entity, ch.ID) }

func applyChanges(s clientState, changes []Change) clientState {
	for _, ch := range changes {
		switch ch.Action {
		case ActionVoid:
			delete(s, stateKey(ch))
		case ActionReplace:
			s[ch.Entity] = ch.Data
		default:
			s[stateKey(ch)] = ch.Data
		}
	}
	return s
}

// ---------- 用例 ----------

func TestHelloSnapshotThenChange(t *testing.T) {
	env := newTestEnv(t)
	c := env.dial(t, &helloPayload{Role: RoleDisplay, InstanceID: "d1"})

	snap := decode[Snapshot](t, c.recvEvent(EventSnapshot))
	if snap.Version != 0 {
		t.Fatalf("空库快照版本应为 0，得到 %d", snap.Version)
	}

	v := insertChange(t, env.db, "result", 42, ActionUpsert, `{"rank":1}`)
	env.hub.Broadcast(loadChanges(t, env.hub, 0, 0))

	got := decode[changePayload](t, c.recvEvent(EventChange))
	if got.Version != v {
		t.Fatalf("change 版本应为 %d，得到 %d", v, got.Version)
	}
	if len(got.Changes) != 1 {
		t.Fatalf("change 条数应为 1，得到 %d", len(got.Changes))
	}
	ch := got.Changes[0]
	if ch.Entity != "result" || ch.ID != 42 || ch.Action != ActionUpsert {
		t.Fatalf("change 内容不符: %+v", ch)
	}
	if string(ch.Data) != `{"rank":1}` {
		t.Fatalf("change data 不符: %s", ch.Data)
	}
	if env.hub.Version() != v {
		t.Fatalf("Hub 版本应推进到 %d，得到 %d", v, env.hub.Version())
	}
}

// 断线期间的变更通过 catchup 补齐，且终态与全量重放的终态一致。
func TestReconnectCatchupTerminalState(t *testing.T) {
	env := newTestEnv(t)
	v1 := insertChange(t, env.db, "result", 1, ActionUpsert, `{"rank":1}`)
	v2 := insertChange(t, env.db, "standings", 0, ActionReplace, `[{"teamId":1,"points":10}]`)

	c1 := env.dial(t, &helloPayload{Role: RoleDisplay, InstanceID: "d1"})
	snap := decode[Snapshot](t, c1.recvEvent(EventSnapshot))
	if snap.Version != v2 || snap.Version != v1+1 {
		t.Fatalf("首次快照版本应为 %d，得到 %d", v2, snap.Version)
	}
	c1.close()

	v3 := insertChange(t, env.db, "result", 2, ActionUpsert, `{"rank":2}`)
	v4 := insertChange(t, env.db, "result", 1, ActionVoid, `{}`)
	v5 := insertChange(t, env.db, "standings", 0, ActionReplace, `[{"teamId":1,"points":12}]`)
	if v4 != v3+1 || v5 != v4+1 {
		t.Fatalf("change_log 版本应连续递增: %d %d %d", v3, v4, v5)
	}

	c2 := env.dial(t, &helloPayload{Role: RoleDisplay, InstanceID: "d1", LastVersion: ptrInt64(snap.Version)})
	cp := decode[catchupPayload](t, c2.recvEvent(EventCatchup))

	if cp.FromVersion != snap.Version {
		t.Fatalf("fromVersion 应为 %d，得到 %d", snap.Version, cp.FromVersion)
	}
	if cp.Version != v5 {
		t.Fatalf("catchup 版本应为 %d，得到 %d", v5, cp.Version)
	}
	// 版本号不上线路（协议只在消息层带 version），因此按内容核对补发的三条变更。
	wantChanges := []struct {
		entity string
		id     int64
		action string
	}{
		{"result", 2, ActionUpsert},
		{"result", 1, ActionVoid},
		{"standings", 0, ActionReplace},
	}
	if len(cp.Changes) != len(wantChanges) {
		t.Fatalf("catchup 条数应为 %d，得到 %d", len(wantChanges), len(cp.Changes))
	}
	for i, want := range wantChanges {
		got := cp.Changes[i]
		if got.Entity != want.entity || got.ID != want.id || got.Action != want.action {
			t.Fatalf("第 %d 条变更应为 %+v，得到 %+v", i, want, got)
		}
	}

	// 断线前已应用的变更 + catchup 的变更 == 从头全量重放，终态必须一致。
	base := applyChanges(clientState{}, loadChanges(t, env.hub, 0, int(snap.Version)))
	got := applyChanges(base, cp.Changes)
	want := applyChanges(clientState{}, loadChanges(t, env.hub, 0, 0))
	if !reflect.DeepEqual(got, want) {
		t.Fatalf("catchup 终态与全量重放不一致:\n got=%v\nwant=%v", got, want)
	}

	// 新连接的整包快照版本必须与 catchup 的终点一致。
	fresh, err := env.hub.Snapshot(context.Background())
	if err != nil {
		t.Fatalf("构建快照: %v", err)
	}
	if fresh.Version != cp.Version {
		t.Fatalf("快照版本 %d 与 catchup 版本 %d 不一致", fresh.Version, cp.Version)
	}
}

// 增量超过阈值时回落全量快照。
func TestCatchupFallsBackToSnapshotWhenTooOld(t *testing.T) {
	env := newTestEnv(t)
	env.hub.SetMaxCatchup(2)
	first := insertChange(t, env.db, "result", 1, ActionUpsert, `{}`)
	for i := int64(2); i <= 5; i++ {
		insertChange(t, env.db, "result", i, ActionUpsert, `{}`)
	}

	c := env.dial(t, &helloPayload{Role: RoleDisplay, InstanceID: "d1", LastVersion: ptrInt64(first - 1)})
	snap := decode[Snapshot](t, c.recvEvent(EventSnapshot))
	if snap.Version != 5 {
		t.Fatalf("应回落全量快照且版本为 5，得到 %d", snap.Version)
	}
}

func TestDuplicateDeliveryIgnored(t *testing.T) {
	env := newTestEnv(t)
	c := env.dial(t, &helloPayload{Role: RoleDisplay, InstanceID: "d1"})
	c.recvEvent(EventSnapshot)

	v1 := insertChange(t, env.db, "result", 1, ActionUpsert, `{"rank":1}`)
	changes := loadChanges(t, env.hub, 0, 0)
	env.hub.Broadcast(changes)

	got := decode[changePayload](t, c.recvEvent(EventChange))
	if got.Version != v1 {
		t.Fatalf("首投版本应为 %d，得到 %d", v1, got.Version)
	}
	c.send(EventAck, ackPayload{Version: v1})

	// 同一批变更再投一次，客户端不应再收到。
	env.hub.Broadcast(changes)
	if ev, _, ok := c.recvWithin(300 * time.Millisecond); ok {
		t.Fatalf("重复投递未被忽略，收到事件 %s", ev)
	}

	// 新变更仍能正常下发。
	v2 := insertChange(t, env.db, "result", 2, ActionUpsert, `{"rank":2}`)
	env.hub.Broadcast(loadChanges(t, env.hub, v1, 0))
	got = decode[changePayload](t, c.recvEvent(EventChange))
	if got.Version != v2 {
		t.Fatalf("新变更版本应为 %d，得到 %d", v2, got.Version)
	}
}

// 心跳携带版本号与服务器时间；版本落后时客户端 resync 能拿到最新快照。
func TestHeartbeatAndResync(t *testing.T) {
	env := newTestEnv(t)
	env.hub.SetHeartbeatInterval(40 * time.Millisecond)

	c := env.dial(t, &helloPayload{Role: RoleDisplay, InstanceID: "d1"})
	c.recvEvent(EventSnapshot)

	// 只写库不广播：心跳版本会领先于客户端，客户端据此发起 resync。
	insertChange(t, env.db, "result", 1, ActionUpsert, `{}`)
	insertChange(t, env.db, "result", 2, ActionUpsert, `{}`)

	hb := decode[heartbeatPayload](t, c.recvEvent(EventHeartbeat))
	if hb.Version != 2 {
		t.Fatalf("心跳版本应为 2，得到 %d", hb.Version)
	}
	if _, err := time.Parse(time.RFC3339Nano, hb.ServerTime); err != nil {
		t.Fatalf("serverTime 不是 RFC3339: %q (%v)", hb.ServerTime, err)
	}

	c.send(EventResync, struct{}{})
	snap := decode[Snapshot](t, c.recvEvent(EventSnapshot))
	if snap.Version != 2 {
		t.Fatalf("resync 快照版本应为 2，得到 %d", snap.Version)
	}
}

// 大屏实例注册表：instanceId、在线状态、当前页、最后心跳可查。
func TestDisplayInstancesRegistry(t *testing.T) {
	env := newTestEnv(t)
	c := env.dial(t, &helloPayload{Role: RoleDisplay, InstanceID: "screen-1", Page: "standings"})
	c.recvEvent(EventSnapshot)

	insts := env.hub.Instances()
	if len(insts) != 1 {
		t.Fatalf("实例数应为 1，得到 %d", len(insts))
	}
	if insts[0].ID != "screen-1" || !insts[0].Online || insts[0].Page != "standings" {
		t.Fatalf("实例注册内容不符: %+v", insts[0])
	}
	if insts[0].LastHeartbeat.IsZero() {
		t.Fatalf("最后心跳不应为零值: %+v", insts[0])
	}

	// 重复 hello 不应把实例引用计数加两次，否则断开后仍会显示在线。
	c.send(EventHello, helloPayload{Role: RoleDisplay, InstanceID: "screen-1", Page: "standings"})
	c.recvEvent(EventSnapshot)
	if insts := env.hub.Instances(); len(insts) != 1 || !insts[0].Online {
		t.Fatalf("重复 hello 后实例应仍只有一个且在线: %+v", insts)
	}

	gotoCmd, err := NewDisplayCommand(CmdGoto, map[string]string{"page": "medals"})
	if err != nil {
		t.Fatalf("构造 goto: %v", err)
	}
	env.hub.SendCommand(gotoCmd)
	cmd := decode[DisplayCommand](t, c.recvEvent(EventDisplayCommand))
	if cmd.Action != CmdGoto {
		t.Fatalf("指令 action 应为 goto，得到 %s", cmd.Action)
	}
	if insts := env.hub.Instances(); insts[0].Page != "medals" {
		t.Fatalf("goto 后当前页应为 medals，得到 %s", insts[0].Page)
	}

	detail, err := NewDisplayCommand(CmdShowDetail, map[string]any{"eventId": 7})
	if err != nil {
		t.Fatalf("构造 show-detail: %v", err)
	}
	env.hub.SendCommand(detail)
	if got := decode[DisplayCommand](t, c.recvEvent(EventDisplayCommand)); got.Action != CmdShowDetail {
		t.Fatalf("指令 action 应为 show-detail，得到 %s", got.Action)
	}

	c.close()
	waitFor(t, "实例离线", func() bool {
		insts := env.hub.Instances()
		return len(insts) == 1 && !insts[0].Online
	})
}

// 房间隔离：admin:<userId> 只发给对应管理端。
func TestAdminRoomTargeting(t *testing.T) {
	env := newTestEnv(t)
	display := env.dial(t, &helloPayload{Role: RoleDisplay, InstanceID: "d1"})
	display.recvEvent(EventSnapshot)
	admin := env.dial(t, &helloPayload{Role: RoleAdmin, UserID: 7})
	admin.recvEvent(EventSnapshot)

	insertChange(t, env.db, "result", 1, ActionUpsert, `{"rank":1}`)
	env.hub.BroadcastTo(AdminRoom(7), loadChanges(t, env.hub, 0, 0))

	got := decode[changePayload](t, admin.recvEvent(EventChange))
	if got.Version != 1 {
		t.Fatalf("定向 change 版本应为 1，得到 %d", got.Version)
	}
	if ev, _, ok := display.recvWithin(300 * time.Millisecond); ok {
		t.Fatalf("大屏不应收到 admin:<uid> 房间的消息，收到 %s", ev)
	}
}

// 慢客户端：ack 落后超过阈值即被识别并断开。
func TestSlowClientDetected(t *testing.T) {
	env := newTestEnv(t)
	env.hub.SetSlowLag(1)

	c := env.dial(t, &helloPayload{Role: RoleDisplay, InstanceID: "slow-1"})
	c.recvEvent(EventSnapshot)

	for i := int64(1); i <= 3; i++ {
		insertChange(t, env.db, "result", i, ActionUpsert, `{}`)
	}
	env.hub.Broadcast(loadChanges(t, env.hub, 0, 0))

	waitFor(t, "慢客户端被标记", func() bool {
		insts := env.hub.Instances()
		return len(insts) == 1 && insts[0].Slow
	})
	waitFor(t, "慢客户端连接被断开", func() bool {
		insts := env.hub.Instances()
		return len(insts) == 1 && !insts[0].Online
	})
}