fix(v2): 退款自愈扫描加回溯窗(防 LIMIT 尾部饥饿)+ 清死行

RefundApplyTask 的候选查询按 out_trade_no ASC/id ASC 取前 200,succeeded 退款历史
只增不减、partially_refunded 长期驻留,超过 limit 后最新的崩溃窗口永远排在候选外
扫不到。两路查询都加 since 回溯窗:RefundStore.ListDistinctOutTradeNosByStatusSince
(completed_at>=since)、OrderStore.ListOrdersByStatusSince(updated_at>=since),旧的
无窗方法原样保留。RefundApplyTask 新增 lookback/now 参数,config.ReconcileConfig 新增
RefundApplyLookbackMin(默认 48h)。顺带按 T6 复审补 OrderTTLMin 字段注释(须 ≫ crypto
orderTTL 15min),清 sync_test.go 死行 `_ = time.Second`。

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013nMthbVEmQquxBRKb9Fj8u
This commit is contained in:
wangjia
2026-07-10 17:59:57 +08:00
parent 727ed74899
commit 9117de7dcf
9 changed files with 212 additions and 14 deletions
+6 -1
View File
@@ -102,7 +102,7 @@ type QuerySyncConfig struct {
// ReconcileConfig v2 后台守护/对账周期任务开关与间隔。
type ReconcileConfig struct {
Enabled bool `mapstructure:"enabled"`
OrderTTLMin int `mapstructure:"order_ttl_min"` // pending 订单存活 TTL(分钟),超则关闭
OrderTTLMin int `mapstructure:"order_ttl_min"` // pending 订单存活 TTL(分钟),超则关闭;须 ≫ crypto 支付窗 orderTTL(15min,internal/provider/crypto/crypto.go)——否则在途 crypto 支付可能被订单级过期误杀
ExpireEverySec int `mapstructure:"expire_every_sec"` // 过期清理间隔
SyncEverySec int `mapstructure:"sync_every_sec"` // 查单对账间隔
UsageEverySec int `mapstructure:"usage_every_sec"` // 用量快照刷新间隔
@@ -114,6 +114,10 @@ type ReconcileConfig struct {
// 本期归到本任务(对账主体)一起装配,brief 原表未列。
RefundApplyEverySec int `mapstructure:"refund_apply_every_sec"` // 退款修复扫描 + 卡滞告警 共用间隔
RefundStuckWarnMin int `mapstructure:"refund_stuck_warn_min"` // 退款卡滞 processing/manual_pending 告警阈值(分钟)
// RefundApplyLookbackMin 退款修复扫描的候选回溯窗(分钟):只扫 completed_at/updated_at
// 落在 [now-lookback, now] 内的候选,防 LIMIT 尾部饥饿(见 store.ListDistinctOutTradeNosByStatusSince
// / ListOrdersByStatusSince)。默认 48 小时——自愈针对近期崩溃窗口,历史一致性由已收敛状态保证。
RefundApplyLookbackMin int `mapstructure:"refund_apply_lookback_min"`
}
var C Config
@@ -159,6 +163,7 @@ func Load() {
viper.SetDefault("reconcile.orphan_window_min", 180)
viper.SetDefault("reconcile.refund_apply_every_sec", 300)
viper.SetDefault("reconcile.refund_stuck_warn_min", 30)
viper.SetDefault("reconcile.refund_apply_lookback_min", 48*60)
if err := viper.ReadInConfig(); err != nil {
log.Println("[config] 未找到 config.yaml,使用默认值 + 环境变量")
+10 -4
View File
@@ -23,14 +23,20 @@ import (
// 候选订单 = distinct(有 succeeded 退款的订单) (当前处于 refunding/partially_refunded
// 态的订单):前者直接命中"退款成功但订单未跟上"的崩溃窗口;后者兜住"已在退款流程
// 中、但后续又有退款 succeeded 未被重算"的情形。
func RefundApplyTask(orders *store.OrderStore, refunds *store.RefundStore, limit int) func(ctx context.Context) error {
//
// 两路查询都带 lookback 回溯窗(since = now-lookback),防「LIMIT 尾部饥饿」:succeeded
// 退款历史只增不减、partially_refunded 订单长期驻留,若不设窗口按 ASC 取前 limit 条,
// 超过 limit 后最新的崩溃窗口(本扫描要自愈的目标)会被挤出候选、永远扫不到。lookback
// 只覆盖近期崩溃窗口——历史一致性由已收敛状态保证,不需要扫全表。
func RefundApplyTask(orders *store.OrderStore, refunds *store.RefundStore, lookback time.Duration, now func() time.Time, limit int) func(ctx context.Context) error {
return func(ctx context.Context) error {
succeededNos, err := refunds.ListDistinctOutTradeNosByStatus(model.RefundSucceeded, limit)
since := now().Add(-lookback)
succeededNos, err := refunds.ListDistinctOutTradeNosByStatusSince(model.RefundSucceeded, since, limit)
if err != nil {
return err
}
refundingOrders, err := orders.ListOrdersByStatus(
[]model.OrderStatusV2{model.OrderRefundingV2, model.OrderPartRefundedV2}, limit)
refundingOrders, err := orders.ListOrdersByStatusSince(
[]model.OrderStatusV2{model.OrderRefundingV2, model.OrderPartRefundedV2}, since, limit)
if err != nil {
return err
}
+104 -7
View File
@@ -25,12 +25,13 @@ func TestRefundApplyTaskSelfHealsStuckPaidOrder(t *testing.T) {
if err := orders.CreateOrder(&model.OrderV2{OutTradeNo: "STUCK-1", AmountMinor: 10000, Currency: "CNY", Status: model.OrderPaidV2}); err != nil {
t.Fatal(err)
}
completedAt := time.Now()
if err := refunds.CreateRefund(&model.Refund{RefundID: "rf-stuck-1", OutTradeNo: "STUCK-1", AttemptProviderRef: "STUCK-1",
AmountMinor: 10000, Currency: "CNY", Status: model.RefundSucceeded, InitiatedBy: "business"}); err != nil {
AmountMinor: 10000, Currency: "CNY", Status: model.RefundSucceeded, InitiatedBy: "business", CompletedAt: &completedAt}); err != nil {
t.Fatal(err)
}
task := reconcile.RefundApplyTask(orders, refunds, 200)
task := reconcile.RefundApplyTask(orders, refunds, 48*time.Hour, time.Now, 200)
if err := task(context.Background()); err != nil {
t.Fatalf("task: %v", err)
}
@@ -51,12 +52,13 @@ func TestRefundApplyTaskPartialStaysPartial(t *testing.T) {
if err := orders.CreateOrder(&model.OrderV2{OutTradeNo: "STUCK-2", AmountMinor: 10000, Currency: "CNY", Status: model.OrderPaidV2}); err != nil {
t.Fatal(err)
}
completedAt := time.Now()
if err := refunds.CreateRefund(&model.Refund{RefundID: "rf-stuck-2", OutTradeNo: "STUCK-2", AttemptProviderRef: "STUCK-2",
AmountMinor: 4000, Currency: "CNY", Status: model.RefundSucceeded, InitiatedBy: "business"}); err != nil {
AmountMinor: 4000, Currency: "CNY", Status: model.RefundSucceeded, InitiatedBy: "business", CompletedAt: &completedAt}); err != nil {
t.Fatal(err)
}
task := reconcile.RefundApplyTask(orders, refunds, 200)
task := reconcile.RefundApplyTask(orders, refunds, 48*time.Hour, time.Now, 200)
if err := task(context.Background()); err != nil {
t.Fatalf("task: %v", err)
}
@@ -76,12 +78,13 @@ func TestRefundApplyTaskIdempotentNoOpOnRerun(t *testing.T) {
if err := orders.CreateOrder(&model.OrderV2{OutTradeNo: "STUCK-3", AmountMinor: 5000, Currency: "CNY", Status: model.OrderPaidV2}); err != nil {
t.Fatal(err)
}
completedAt := time.Now()
if err := refunds.CreateRefund(&model.Refund{RefundID: "rf-stuck-3", OutTradeNo: "STUCK-3", AttemptProviderRef: "STUCK-3",
AmountMinor: 5000, Currency: "CNY", Status: model.RefundSucceeded, InitiatedBy: "business"}); err != nil {
AmountMinor: 5000, Currency: "CNY", Status: model.RefundSucceeded, InitiatedBy: "business", CompletedAt: &completedAt}); err != nil {
t.Fatal(err)
}
task := reconcile.RefundApplyTask(orders, refunds, 200)
task := reconcile.RefundApplyTask(orders, refunds, 48*time.Hour, time.Now, 200)
for i := 0; i < 3; i++ {
if err := task(context.Background()); err != nil {
t.Fatalf("run %d: %v", i, err)
@@ -109,7 +112,7 @@ func TestRefundApplyTaskConsistentOrderUntouched(t *testing.T) {
t.Fatal(err)
}
task := reconcile.RefundApplyTask(orders, refunds, 200)
task := reconcile.RefundApplyTask(orders, refunds, 48*time.Hour, time.Now, 200)
if err := task(context.Background()); err != nil {
t.Fatalf("task: %v", err)
}
@@ -119,6 +122,100 @@ func TestRefundApplyTaskConsistentOrderUntouched(t *testing.T) {
}
}
// TestRefundApplyTaskLookbackWindowExcludesStaleSucceeded 钉住 LIMIT 尾部饥饿修复:
// succeeded 退款侧的候选查询带回溯窗,completed_at 超窗的行不再入候选(窗外订单卡在
// paid 不被自愈),只有窗内的才被扫到并自愈。
func TestRefundApplyTaskLookbackWindowExcludesStaleSucceeded(t *testing.T) {
db := model.OpenTestDB(t)
orders := store.NewOrderStore(db)
refunds := store.NewRefundStore(db)
now := time.Date(2026, 7, 10, 12, 0, 0, 0, time.UTC)
lookback := 48 * time.Hour
if err := orders.CreateOrder(&model.OrderV2{OutTradeNo: "WIN-OLD", AmountMinor: 10000, Currency: "CNY", Status: model.OrderPaidV2}); err != nil {
t.Fatal(err)
}
staleCompleted := now.Add(-lookback - time.Hour) // 窗外
if err := refunds.CreateRefund(&model.Refund{RefundID: "rf-win-old", OutTradeNo: "WIN-OLD", AttemptProviderRef: "WIN-OLD",
AmountMinor: 10000, Currency: "CNY", Status: model.RefundSucceeded, InitiatedBy: "business", CompletedAt: &staleCompleted}); err != nil {
t.Fatal(err)
}
if err := orders.CreateOrder(&model.OrderV2{OutTradeNo: "WIN-NEW", AmountMinor: 10000, Currency: "CNY", Status: model.OrderPaidV2}); err != nil {
t.Fatal(err)
}
freshCompleted := now.Add(-time.Hour) // 窗内
if err := refunds.CreateRefund(&model.Refund{RefundID: "rf-win-new", OutTradeNo: "WIN-NEW", AttemptProviderRef: "WIN-NEW",
AmountMinor: 10000, Currency: "CNY", Status: model.RefundSucceeded, InitiatedBy: "business", CompletedAt: &freshCompleted}); err != nil {
t.Fatal(err)
}
task := reconcile.RefundApplyTask(orders, refunds, lookback, func() time.Time { return now }, 200)
if err := task(context.Background()); err != nil {
t.Fatalf("task: %v", err)
}
oldOrder, err := orders.GetOrder("WIN-OLD")
if err != nil || oldOrder.Status != model.OrderPaidV2 {
t.Fatalf("窗外行不应入候选、不应被自愈, got %+v, err=%v", oldOrder, err)
}
newOrder, err := orders.GetOrder("WIN-NEW")
if err != nil || newOrder.Status != model.OrderRefundedV2 {
t.Fatalf("窗内行应被自愈为 refunded, got %+v, err=%v", newOrder, err)
}
}
// TestRefundApplyTaskLookbackWindowExcludesStaleRefundingOrder 同上,针对
// refunding/partially_refunded 订单侧的候选查询:updated_at 超窗的订单不再入候选
// (即便它挂着可自愈的 succeeded 退款,只要两路查询都被窗口挡在外面就不会被扫到)。
func TestRefundApplyTaskLookbackWindowExcludesStaleRefundingOrder(t *testing.T) {
db := model.OpenTestDB(t)
orders := store.NewOrderStore(db)
refunds := store.NewRefundStore(db)
now := time.Date(2026, 7, 10, 12, 0, 0, 0, time.UTC)
lookback := 48 * time.Hour
staleAt := now.Add(-lookback - time.Hour) // 窗外
freshAt := now.Add(-time.Hour) // 窗内
// 两笔订单的 succeeded 退款 completed_at 都设窗外,隔离掉「succeeded 侧候选」这条路径
// ——只让 refunding/partially_refunded 订单侧的时间窗单独决定是否入候选。
if err := orders.CreateOrder(&model.OrderV2{OutTradeNo: "ORD-OLD", AmountMinor: 10000, Currency: "CNY", Status: model.OrderRefundingV2}); err != nil {
t.Fatal(err)
}
if err := refunds.CreateRefund(&model.Refund{RefundID: "rf-ord-old", OutTradeNo: "ORD-OLD", AttemptProviderRef: "ORD-OLD",
AmountMinor: 10000, Currency: "CNY", Status: model.RefundSucceeded, InitiatedBy: "business", CompletedAt: &staleAt}); err != nil {
t.Fatal(err)
}
if err := db.Model(&model.OrderV2{}).Where("out_trade_no = ?", "ORD-OLD").Update("updated_at", staleAt).Error; err != nil {
t.Fatal(err)
}
if err := orders.CreateOrder(&model.OrderV2{OutTradeNo: "ORD-NEW", AmountMinor: 10000, Currency: "CNY", Status: model.OrderRefundingV2}); err != nil {
t.Fatal(err)
}
if err := refunds.CreateRefund(&model.Refund{RefundID: "rf-ord-new", OutTradeNo: "ORD-NEW", AttemptProviderRef: "ORD-NEW",
AmountMinor: 10000, Currency: "CNY", Status: model.RefundSucceeded, InitiatedBy: "business", CompletedAt: &staleAt}); err != nil {
t.Fatal(err)
}
if err := db.Model(&model.OrderV2{}).Where("out_trade_no = ?", "ORD-NEW").Update("updated_at", freshAt).Error; err != nil {
t.Fatal(err)
}
task := reconcile.RefundApplyTask(orders, refunds, lookback, func() time.Time { return now }, 200)
if err := task(context.Background()); err != nil {
t.Fatalf("task: %v", err)
}
oldOrder, err := orders.GetOrder("ORD-OLD")
if err != nil || oldOrder.Status != model.OrderRefundingV2 {
t.Fatalf("窗外订单不应入候选、不应被自愈, got %+v, err=%v", oldOrder, err)
}
newOrder, err := orders.GetOrder("ORD-NEW")
if err != nil || newOrder.Status != model.OrderRefundedV2 {
t.Fatalf("窗内订单应被自愈为 refunded, got %+v, err=%v", newOrder, err)
}
}
// TestRefundStuckAlertTaskLogsWarnWithoutChangingState 卡滞 processing/manual_pending
// 超阈值只应打 WARN 日志,不改任何状态(观测型)。
func TestRefundStuckAlertTaskLogsWarnWithoutChangingState(t *testing.T) {
-1
View File
@@ -49,7 +49,6 @@ func TestSyncPendingTaskSettlesViaQuery(t *testing.T) {
if o.Status != model.OrderPaidV2 {
t.Fatalf("查单对账后应 paid, got %v", o.Status)
}
_ = time.Second
}
func TestPaidSpotCheckTaskRunsCleanOnConsistent(t *testing.T) {
+16
View File
@@ -175,6 +175,22 @@ func (s *OrderStore) ListOrdersByStatus(statuses []model.OrderStatusV2, limit in
return out, nil
}
// ListOrdersByStatusSince 同 ListOrdersByStatus,但加 updated_at>=since 回溯窗——
// refunding/partially_refunded 是长期驻留态(订单进入后可能停留很久),不设窗口时
// 会占满 limit 名额,把「近期才卡滞、需要自愈」的订单挤出候选(与
// RefundStore.ListDistinctOutTradeNosByStatusSince 同一治法,同一 lookback 语义)。
func (s *OrderStore) ListOrdersByStatusSince(statuses []model.OrderStatusV2, since time.Time, limit int) ([]model.OrderV2, error) {
if limit <= 0 || limit > 500 {
limit = 200
}
var out []model.OrderV2
if err := s.db.Where("status IN ? AND updated_at >= ?", statuses, since).
Order("id ASC").Limit(limit).Find(&out).Error; err != nil {
return nil, fmt.Errorf("store.ListOrdersByStatusSince: %w", err)
}
return out, nil
}
// ListAttemptsByChannelSince 列某渠道 created_at>=since 的 attempt(任意状态),
// 供 orphan 扫描构造"已知期望金额集"(凡 pay 合法签发过的金额都不算孤儿)。
func (s *OrderStore) ListAttemptsByChannelSince(channel string, since time.Time, limit int) ([]model.Attempt, error) {
+31
View File
@@ -177,6 +177,37 @@ func TestListOrdersByStatus(t *testing.T) {
}
}
// TestListOrdersByStatusSinceExcludesStale 钉住回溯窗:updated_at 早于 since 的订单
// 不应入候选(防 refunding/partially_refunded 长期驻留占满 LIMIT 名额——见 RefundApplyTask)。
func TestListOrdersByStatusSinceExcludesStale(t *testing.T) {
db := model.OpenTestDB(t)
s := store.NewOrderStore(db)
now := time.Now()
stale := now.Add(-72 * time.Hour)
fresh := now.Add(-time.Hour)
if err := s.CreateOrder(&model.OrderV2{OutTradeNo: "O-STALE", AmountMinor: 100, Currency: "CNY", Status: model.OrderRefundingV2}); err != nil {
t.Fatal(err)
}
if err := db.Model(&model.OrderV2{}).Where("out_trade_no = ?", "O-STALE").Update("updated_at", stale).Error; err != nil {
t.Fatal(err)
}
if err := s.CreateOrder(&model.OrderV2{OutTradeNo: "O-FRESH", AmountMinor: 100, Currency: "CNY", Status: model.OrderRefundingV2}); err != nil {
t.Fatal(err)
}
if err := db.Model(&model.OrderV2{}).Where("out_trade_no = ?", "O-FRESH").Update("updated_at", fresh).Error; err != nil {
t.Fatal(err)
}
got, err := s.ListOrdersByStatusSince([]model.OrderStatusV2{model.OrderRefundingV2}, now.Add(-48*time.Hour), 50)
if err != nil {
t.Fatalf("list: %v", err)
}
if len(got) != 1 || got[0].OutTradeNo != "O-FRESH" {
t.Fatalf("只应命中窗内的 O-FRESH, got %+v", got)
}
}
// TestListAttemptsByChannelSince 供 orphan 扫描构造"已知期望金额集":同渠道、
// created_at>=since、任意状态的 attempt(pending/paid/expired 都算——凡 pay
// 合法签发过的金额都不算孤儿)。
+21
View File
@@ -144,6 +144,27 @@ func (s *RefundStore) ListDistinctOutTradeNosByStatus(status model.RefundStatus,
return out, nil
}
// ListDistinctOutTradeNosByStatusSince 同 ListDistinctOutTradeNosByStatus,但加
// completed_at>=since 回溯窗——防「LIMIT 尾部饥饿」:succeeded 退款历史只增不减,
// 若不设窗口,按 out_trade_no ASC 取前 limit 条,超过 limit 后最新出现的"退款已成但
// 订单卡 paid"崩溃窗口(本扫描要自愈的目标)会被排在候选之外、永远扫不到。completed_at
// 是 MarkRefundStatus 翻成 succeeded/failed 时打的时间戳(即"进入该终态"的时刻),succeeded
// 是终态、之后不会再被改动,用它做窗口锚点比 updated_at 更精确。调用方(RefundApplyTask)
// 用 now-lookback 算 since,lookback 只覆盖近期崩溃窗口——历史一致性由已收敛状态保证,
// 不需要扫全表。
func (s *RefundStore) ListDistinctOutTradeNosByStatusSince(status model.RefundStatus, since time.Time, limit int) ([]string, error) {
if limit <= 0 || limit > 500 {
limit = 200
}
var out []string
if err := s.db.Model(&model.Refund{}).Where("status = ? AND completed_at >= ?", status, since).
Group("out_trade_no").Order("out_trade_no ASC").Limit(limit).
Pluck("out_trade_no", &out).Error; err != nil {
return nil, fmt.Errorf("store.ListDistinctOutTradeNosByStatusSince: %w", err)
}
return out, nil
}
// ListStuckRefunds 列 status 落在给定集合、且 updated_at 早于 before 的退款行,供
// 「卡滞 processing/manual_pending 退款告警」只读观测扫描用(不改状态)。updated_at
// 用作「进入当前状态」的近似时刻——本表除 MarkRefundStatus/CreateRefundGuarded 外
+23
View File
@@ -126,6 +126,29 @@ func TestListDistinctOutTradeNosByStatus(t *testing.T) {
}
}
// TestListDistinctOutTradeNosByStatusSinceExcludesStale 钉住回溯窗:completed_at
// 早于 since 的 succeeded 退款不应入候选(防 LIMIT 尾部饥饿——见 RefundApplyTask)。
func TestListDistinctOutTradeNosByStatusSinceExcludesStale(t *testing.T) {
db := model.OpenTestDB(t)
rs := NewRefundStore(db)
now := time.Now()
stale := now.Add(-72 * time.Hour)
fresh := now.Add(-time.Hour)
_ = rs.CreateRefund(&model.Refund{RefundID: "s-stale", OutTradeNo: "PAY-STALE", AmountMinor: 100, Currency: "CNY",
Status: model.RefundSucceeded, CompletedAt: &stale})
_ = rs.CreateRefund(&model.Refund{RefundID: "s-fresh", OutTradeNo: "PAY-FRESH", AmountMinor: 100, Currency: "CNY",
Status: model.RefundSucceeded, CompletedAt: &fresh})
got, err := rs.ListDistinctOutTradeNosByStatusSince(model.RefundSucceeded, now.Add(-48*time.Hour), 50)
if err != nil {
t.Fatalf("list: %v", err)
}
if len(got) != 1 || got[0] != "PAY-FRESH" {
t.Fatalf("只应命中窗内的 PAY-FRESH, got %+v", got)
}
}
// TestListStuckRefunds 供「卡滞 processing/manual_pending 退款告警」:只挑 updated_at
// 早于阈值的 processing/manual_pending 行,requested/succeeded/failed 不命中。
func TestListStuckRefunds(t *testing.T) {
+1 -1
View File
@@ -77,7 +77,7 @@ func main() {
runner.Add("paid-spotcheck", time.Duration(rc.SpotCheckEverySec)*time.Second,
reconcile.PaidSpotCheckTask(orderStore, pReg, time.Duration(rc.SpotCheckWindowMin)*time.Minute, time.Now))
runner.Add("refund-apply-sweep", time.Duration(rc.RefundApplyEverySec)*time.Second,
reconcile.RefundApplyTask(orderStore, refundStore, 200))
reconcile.RefundApplyTask(orderStore, refundStore, time.Duration(rc.RefundApplyLookbackMin)*time.Minute, time.Now, 200))
runner.Add("refund-stuck-alert", time.Duration(rc.RefundApplyEverySec)*time.Second,
reconcile.RefundStuckAlertTask(refundStore, time.Duration(rc.RefundStuckWarnMin)*time.Minute, time.Now))
orphanStore := store.NewOrphanStore(db)