fix(v2): settleRenewal 重投补入队(通知不再永久丢)+ 无业务方首过返回 Processed(+同型路径核修)

settleRenewal 先 CreateRenewalPaid(建单已 paid)后 Enqueue,首次入队失败后重投走
created=false 的 duplicate 分支旧实现直接 return、不再入队,subscription.renewed
永久丢失(SyncPendingAttempts 救不了)。改为 duplicate 分支也无条件尝试入队,outbox
唯一键 ON CONFLICT DO NOTHING 幂等自愈;顺带对齐无业务方首过应返回 Processed。核查
发现 finalizeCanceled/markSubscriptionPastDue 有同型反纪律,同修。

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 18:30:11 +08:00
parent 62abdc1d65
commit ff65a06bf2
4 changed files with 186 additions and 26 deletions
+13 -1
View File
@@ -35,7 +35,11 @@ func (stubResolver) Resolve(sku, currency string) (int64, string, string, error)
type spyEnqueuer struct {
calls []map[string]any
failNext bool // 置 true 模拟 outbox 入队失败(settle 崩溃窗口测试用)
failNext bool // 置 true 模拟 outbox 入队失败(settle 崩溃窗口测试用)
seen map[string]bool // 镜像真实 WebhookStore.EnqueueDelivery 的 (out_trade_no,event_type,
// refund_id) 唯一键 ON CONFLICT DO NOTHING:重复 key 静默 no-op(不追加 calls,也不算失败)。
// 修复"重投补入队自愈"后,业务代码会在 duplicate/重投分支也调用 Enqueue,若 spy 仍是无脑
// 计数器就会把 outbox 天然幂等的重复行误判成"多发了一次 webhook",这里镜像真实幂等语义。
}
func (s *spyEnqueuer) Enqueue(outTradeNo, bizSystem, eventType, refundID string, data map[string]any) error {
@@ -43,6 +47,14 @@ func (s *spyEnqueuer) Enqueue(outTradeNo, bizSystem, eventType, refundID string,
s.failNext = false
return errors.New("outbox down")
}
key := outTradeNo + "|" + eventType + "|" + refundID
if s.seen == nil {
s.seen = make(map[string]bool)
}
if s.seen[key] {
return nil // 幂等 no-op,行已存在
}
s.seen[key] = true
s.calls = append(s.calls, data)
return nil
}
+27 -8
View File
@@ -138,7 +138,17 @@ func (g *Gateway) HandleCallback(ctx context.Context, method string, in provider
// settleRenewal 处理续费 invoice.paid(设计 §5/§4 决策记录):每期铸独立 renewal OrderV2
// (out_trade_no = 首购单号 + "-r-" + invoice id),建即 paid(续费不经收银台,无 pending 中间态)。
// 幂等靠 renewal attempt 的 (channel,provider_ref=invoice.ID) 唯一索引 + renewal order 的
// out_trade_no 唯一索引双保险;出问题让 Stripe 重投,重投时 created=false 不重复入队,自愈
// out_trade_no 唯一索引双保险。
//
// 入队纪律(与文件头 Settle 的顺序不变量同根同源,但形态不同——续费建单是"单步已 paid",
// 没有 Settle 那种"翻转严格发生在成功入队之后"的两段式可依赖):**duplicate 分支(created=
// false)也必须尝试入队**,不能像旧实现那样直接 return。理由:首次入队若瞬时失败,订单/attempt
// 已在同一事务内落为 paid(不回滚,幂等键护着),而 outbox 从未有行;Stripe 拿不到 200 会重投
// 同一 invoice.paid,此时 created 必为 false——若 duplicate 分支不入队,该续费通知永久丢失
// (SyncPendingAttempts 也救不了:续费 attempt 生来就是 AttemptPaid,不在 pending 轮询范围)。
// outbox 唯一键 (out_trade_no,event_type,refund_id) 上 ON CONFLICT DO NOTHING 天然幂等:行已
// 存在则本次 no-op,行曾丢失则本次补建——重投即自愈,任何一分支入队失败都仍返回 SettleFailed
// 交渠道再重投。
func (g *Gateway) settleRenewal(ctx context.Context, method string, ev *provider.PaidEvent) (SettleResult, error) {
// channel = 触发本次回调的 method(与 markSubscriptionPastDue/recordChargeback 同式);
// 订阅诞生时 Subscription.Channel 落的正是 att.Channel=method,查询须对齐,不能硬编码字面量
@@ -174,18 +184,27 @@ func (g *Gateway) settleRenewal(ctx context.Context, method string, ev *provider
if _, err := g.subs.Activate(sub.SubID, &nextEnd); err != nil {
return SettleFailed, err
}
if !created || sub.BizSystem == "" {
return SettleDuplicate, nil // 重投 / 独立收款
// 无论 created 与否都尝试入队(见函数注释的入队纪律);独立收款(BizSystem=="")无下游可发,跳过。
if sub.BizSystem != "" {
if err := g.enqueueRenewed(sub, renewalNo, ev, paidAt); err != nil {
return SettleFailed, err // 已建单的事实不回滚(幂等键护着);渠道重投时补入队
}
}
if err := g.webhook.Enqueue(renewalNo, sub.BizSystem, EvtSubscriptionRenewed, "", map[string]any{
if !created {
return SettleDuplicate, nil // 重投:订单/attempt 已在(幂等 no-op),outbox 已补齐
}
return SettleProcessed, nil // 首过(含无业务方的独立收款,对齐 enqueuePaymentSucceeded 语义)
}
// enqueueRenewed 组 subscription.renewed 领域 payload 并幂等入队(抽出供 settleRenewal 的
// created/duplicate 两分支共用,避免复制两份 payload 构造)。
func (g *Gateway) enqueueRenewed(sub *model.Subscription, renewalNo string, ev *provider.PaidEvent, paidAt time.Time) error {
return g.webhook.Enqueue(renewalNo, sub.BizSystem, EvtSubscriptionRenewed, "", map[string]any{
"event_type": EvtSubscriptionRenewed, "out_trade_no": renewalNo, "sub_id": sub.SubID,
"biz_system": sub.BizSystem, "biz_ref": sub.BizRef, "product_biz_code": sub.BizCode,
"amount_minor": ev.PaidAmountMinor, "currency": ev.PaidCurrency, "channel": sub.Channel,
"paid_at": paidAt.Format(time.RFC3339),
}); err != nil {
return SettleFailed, err
}
return SettleProcessed, nil
})
}
func (g *Gateway) recordChargeback(ctx context.Context, method string, ev *provider.PaidEvent) (SettleResult, error) {
+29 -17
View File
@@ -195,20 +195,24 @@ func (g *Gateway) settleSubscriptionCanceled(ctx context.Context, method string,
}
// finalizeCanceled 是取消的唯一落地点(主动取消 API + 入站 webhook 两路收敛于此):
// MarkCanceled 幂等翻转 canceled;仅首次真正翻转(flipped)且有业务方(BizSystem 非空)
// 才入队 subscription.canceled——重投/独立收款不重复发。
// MarkCanceled 幂等翻转 canceled,随后(有业务方时)入队 subscription.canceled。
//
// 入队不按 flipped 分叉(P8 Task4 审计发现的同型反纪律,同 commit 一并修):若第一次翻转
// 成功但入队瞬时失败,调用方拿到 err 后重投——此时 MarkCanceled 已是"已取消"故 flipped=
// false,若像旧实现那样 `!flipped→return nil` 直接跳过,outbox 永远补不上这一行,通知永久
// 丢失。改为无论 flipped 与否都尝试入队,outbox 唯一键 ON CONFLICT DO NOTHING 天然幂等:
// 行已存在则 no-op,行曾丢失则本次补建。canceled 每订阅只发生一次,不存在"同订阅多次 canceled
// 真被吞"的歧义(不同于 past_due 可能有多次不同 invoice 失败),无条件入队是安全的。
//
// outbox 键取真实首购单号 sub.OutTradeNo(不用合成后缀键):该单已 paid,Notifier 投递门禁
// (orderPaid)天然放行;event_type=subscription.canceled 与该单已有的 payment.succeeded /
// subscription.created 行不同 event_type,唯一键 (out_trade_no,event_type,refund_id) 不撞。
// canceled 每订阅只发生一次,不存在"同订阅多次 canceled 被吞"的问题(不同于 past_due)。
func (g *Gateway) finalizeCanceled(sub *model.Subscription) error {
flipped, err := g.subs.MarkCanceled(sub.SubID)
if err != nil {
if _, err := g.subs.MarkCanceled(sub.SubID); err != nil {
return err
}
if !flipped || sub.BizSystem == "" {
return nil // 已取消(重投)/ 独立收款无业务方回调
if sub.BizSystem == "" {
return nil // 独立收款无业务方回调
}
return g.webhook.Enqueue(sub.OutTradeNo, sub.BizSystem, EvtSubscriptionCanceled, "", map[string]any{
"event_type": EvtSubscriptionCanceled, "out_trade_no": sub.OutTradeNo, "sub_id": sub.SubID,
@@ -221,9 +225,15 @@ func (g *Gateway) finalizeCanceled(sub *model.Subscription) error {
// 当前 active 时翻转,已 past_due/已 canceled 不动),入队 subscription.past_due 供业务方提醒
// 用户换卡。下期 invoice.paid 成功由 settleRenewal 的 Activate 自动恢复 active。
//
// 入队不按 flipped 分叉(P8 Task4 审计发现的同型反纪律,同 commit 一并修):首次翻转成功但
// 入队瞬时失败时,重投这个同一事件会命中 flipped=false(已是 past_due),旧实现直接 return
// 会让这条 past_due 通知永久丢失。改为无论 flipped 与否都尝试入队,outbox 唯一键 ON CONFLICT
// DO NOTHING 天然幂等——重投即自愈。
//
// outbox 键取真实首购单号 sub.OutTradeNo(理由同 finalizeCanceled):Task5 最小可行接受
// "同订阅多次 past_due 只报首次"(第二次失败的行撞 (out_trade_no,event_type) 唯一键、
// EnqueueDelivery 幂等 no-op),Task 7 视需要再引入合成键 + Notifier 门禁旁路。
// "同订阅多次 past_due(不同 invoice 各自失败)只报首次"这一语义限制不变——第二次真实失败的
// 行同样撞 (out_trade_no,event_type) 唯一键、幂等 no-op,Task 7 视需要再引入合成键 + Notifier
// 门禁旁路;这里修的只是"同一事件重投也补不回入队"这一段自愈缺口,不扩大既有限制的范围。
func (g *Gateway) markSubscriptionPastDue(ctx context.Context, method string, ev *provider.PaidEvent) (SettleResult, error) {
sub, err := g.subs.GetByProviderRef(method, ev.SubscriptionRef)
if err != nil {
@@ -236,15 +246,17 @@ func (g *Gateway) markSubscriptionPastDue(ctx context.Context, method string, ev
if err != nil {
return SettleFailed, err
}
if !flipped || sub.BizSystem == "" {
return SettleDuplicate, nil
if sub.BizSystem != "" {
if err := g.webhook.Enqueue(sub.OutTradeNo, sub.BizSystem, EvtSubscriptionPastDue, "", map[string]any{
"event_type": EvtSubscriptionPastDue, "out_trade_no": sub.OutTradeNo, "sub_id": sub.SubID,
"biz_system": sub.BizSystem, "biz_ref": sub.BizRef, "product_biz_code": sub.BizCode,
"failed_at": time.Now().Format(time.RFC3339),
}); err != nil {
return SettleFailed, err
}
}
if err := g.webhook.Enqueue(sub.OutTradeNo, sub.BizSystem, EvtSubscriptionPastDue, "", map[string]any{
"event_type": EvtSubscriptionPastDue, "out_trade_no": sub.OutTradeNo, "sub_id": sub.SubID,
"biz_system": sub.BizSystem, "biz_ref": sub.BizRef, "product_biz_code": sub.BizCode,
"failed_at": time.Now().Format(time.RFC3339),
}); err != nil {
return SettleFailed, err
if !flipped {
return SettleDuplicate, nil
}
return SettleProcessed, nil
}
+117
View File
@@ -685,3 +685,120 @@ func TestMarkSubscriptionPastDueDuplicateInvoiceNotDoubleEnqueued(t *testing.T)
t.Fatalf("webhook calls after replay = %d, want still 1: %+v", len(spy.calls), spy.calls)
}
}
// TestSettleRenewalEnqueueFailureThenRetryRecovers 复现 Important 发现:settleRenewal 先
// CreateRenewalPaid(建单已 paid,单步无 pending 中间态)后 Enqueue——首次入队失败(瞬时)后,
// invoice.paid 重投走 created=false 的 duplicate 分支必须仍尝试入队(outbox 唯一键幂等,
// 行不存在则补建),否则 subscription.renewed 永久丢失(SyncPendingAttempts 救不了:续费
// attempt 生来就是 AttemptPaid,不在 pending 轮询范围)。
// 断言:①首次入队失败 → SettleFailed+err,订单已建为 paid(幂等键护着,不回滚)②同一
// invoice.paid 重投 → subscription.renewed 最终恰入队一次 ③renewal order 不因重投双铸。
func TestSettleRenewalEnqueueFailureThenRetryRecovers(t *testing.T) {
g, fp, spy, orders, _ := newSubGateway(t)
ctx := context.Background()
activateSub(t, g, fp)
spy.calls = nil
renewalRaw, err := json.Marshal(provider.PaidEvent{
Kind: provider.EventSubscriptionRenewal, SubscriptionRef: "sub_new", InvoiceRef: "in_flaky",
PaidAmountMinor: 2999, PaidCurrency: "USD", Status: provider.PaidSucceeded,
})
if err != nil {
t.Fatalf("marshal renewal event: %v", err)
}
spy.failNext = true
result, err := g.HandleCallback(ctx, "substripe", provider.CallbackInput{Raw: renewalRaw})
if err == nil || result != gateway.SettleFailed {
t.Fatalf("首次入队失败应 SettleFailed+err, got %v, %v", result, err)
}
allOrders, err := orders.ListOrders("pangolin", "u-1", 10)
if err != nil {
t.Fatalf("ListOrders: %v", err)
}
renewalCount := 0
for i := range allOrders {
if strings.Contains(allOrders[i].OutTradeNo, "-r-") {
renewalCount++
if allOrders[i].Status != model.OrderPaidV2 {
t.Fatalf("renewal order 应已建为 paid(幂等键护着,不因入队失败回滚), got %+v", allOrders[i])
}
}
}
if renewalCount != 1 {
t.Fatalf("renewal order count after first(失败) attempt = %d, want 1", renewalCount)
}
if len(spy.calls) != 0 {
t.Fatalf("入队失败不应留下 webhook 记录, got %+v", spy.calls)
}
// 渠道重投同一 invoice.paid(Stripe 拿不到 200 会重投):outbox 补建自愈。
result2, err := g.HandleCallback(ctx, "substripe", provider.CallbackInput{Raw: renewalRaw})
if err != nil {
t.Fatalf("HandleCallback retry: %v", err)
}
_ = result2 // duplicate(created=false,订单已在)——本用例只关心自愈,不断言具体 result 值
if len(spy.calls) != 1 {
t.Fatalf("重投后 subscription.renewed 应恰入队一次(自愈), got %d: %+v", len(spy.calls), spy.calls)
}
if spy.calls[0]["event_type"] != gateway.EvtSubscriptionRenewed {
t.Fatalf("event_type = %v, want %s", spy.calls[0]["event_type"], gateway.EvtSubscriptionRenewed)
}
allOrdersAfter, err := orders.ListOrders("pangolin", "u-1", 10)
if err != nil {
t.Fatalf("ListOrders after retry: %v", err)
}
renewalCountAfter := 0
for i := range allOrdersAfter {
if strings.Contains(allOrdersAfter[i].OutTradeNo, "-r-") {
renewalCountAfter++
}
}
if renewalCountAfter != 1 {
t.Fatalf("renewal order count after retry = %d, want still 1(不双铸)", renewalCountAfter)
}
}
// TestSettleRenewalNoBizSystemStillProcessed 独立收款(无业务方回调,BizSystem=="")的续费首过
// 应与 enqueuePaymentSucceeded 的"无业务方=跳过入队但仍 processed"语义对齐,不能误判 duplicate
// (created=true 是真正的首次成交,只是没有下游 webhook 可发)。
func TestSettleRenewalNoBizSystemStillProcessed(t *testing.T) {
g, fp, spy, _, subs := newSubGateway(t)
ctx := context.Background()
_, err := g.CreateSubscription(ctx, gateway.CreateSubscriptionInput{
SKU: "pro_monthly", Method: "substripe", // BizSystem/BizRef 留空 = 独立收款
})
if err != nil {
t.Fatalf("CreateSubscription: %v", err)
}
activateRaw, err := json.Marshal(provider.PaidEvent{
Kind: provider.EventPayment, ProviderRef: fp.sessionRef, Status: provider.PaidSucceeded,
PaidAmountMinor: 2999, PaidCurrency: "USD", SubscriptionRef: "sub_new",
})
if err != nil {
t.Fatalf("marshal activate event: %v", err)
}
if _, err := g.HandleCallback(ctx, "substripe", provider.CallbackInput{Raw: activateRaw}); err != nil {
t.Fatalf("activate: %v", err)
}
sub, err := subs.GetByProviderRef("substripe", "sub_new")
if err != nil || sub.BizSystem != "" {
t.Fatalf("subscription = %+v, %v, want BizSystem empty(独立收款)", sub, err)
}
renewalRaw, err := json.Marshal(provider.PaidEvent{
Kind: provider.EventSubscriptionRenewal, SubscriptionRef: "sub_new", InvoiceRef: "in_indep_1",
PaidAmountMinor: 2999, PaidCurrency: "USD", Status: provider.PaidSucceeded,
})
if err != nil {
t.Fatalf("marshal renewal event: %v", err)
}
result, err := g.HandleCallback(ctx, "substripe", provider.CallbackInput{Raw: renewalRaw})
if err != nil || result != gateway.SettleProcessed {
t.Fatalf("独立收款续费首过应 processed(非 duplicate), got %v, %v", result, err)
}
if len(spy.calls) != 0 {
t.Fatalf("独立收款不应入队 webhook, got %+v", spy.calls)
}
}