Version: 0.9.68.dev.260504

后端： 1. 阶段 3 notification 服务边界落地，新增 `cmd/notification`、`services/notification`、`gateway/notification`、`shared/contracts/notification` 和 notification port，按 userauth 同款最小手搓 zrpc 样板收口 2. notification outbox consumer、relay 和 retry loop 迁入独立服务入口，处理 `notification.feishu.requested`，gateway 改为通过 zrpc client 调用 notification 3. 清退旧单体 notification DAO/model/service/provider/runner 和 `service/events/notification_feishu.go`，旧实现不再作为活跃编译路径 4. 修复 outbox 路由归属、dispatch 启动扫描、Kafka topic 探测/投递超时、sending 租约恢复、毒消息 MarkDead 错误回传和 RPC timeout 边界 5. 同步调整 active-scheduler 触发通知事件、核心 outbox handler、MySQL 迁移边界和 notification 配置文档： 1. 更新微服务迁移计划，将阶段 3 notification 标记为已完成，并明确下一阶段从 active-scheduler 开始
2026-05-04 18:40:39 +08:00
parent 9742dc8b1c
commit abe3b4960e
41 changed files with 2178 additions and 889 deletions
--- a/backend/services/notification/dao/channel.go
+++ b/backend/services/notification/dao/channel.go
@@ -0,0 +1,127 @@
+package dao
+
+import (
+	"context"
+	"errors"
+	"time"
+
+	notificationmodel "github.com/LoveLosita/smartflow/backend/services/notification/model"
+	"gorm.io/gorm"
+	"gorm.io/gorm/clause"
+)
+
+// ChannelDAO 管理用户外部通知通道配置。
+//
+// 职责边界：
+// 1. 只负责 user_notification_channels 的基础读写；
+// 2. 不负责 webhook 请求发送、notification_records 状态机或 outbox 消费；
+// 3. webhook_url / bearer_token 的脱敏由 service 层处理，DAO 保持真实持久化值。
+type ChannelDAO struct {
+	db *gorm.DB
+}
+
+func NewChannelDAO(db *gorm.DB) *ChannelDAO {
+	return &ChannelDAO{db: db}
+}
+
+func (d *ChannelDAO) WithTx(tx *gorm.DB) *ChannelDAO {
+	return &ChannelDAO{db: tx}
+}
+
+func (d *ChannelDAO) ensureDB() error {
+	if d == nil || d.db == nil {
+		return errors.New("notification channel dao 未初始化")
+	}
+	return nil
+}
+
+// UpsertUserNotificationChannel 按 user_id + channel 幂等保存用户通知配置。
+//
+// 说明：
+// 1. 只覆盖开关、webhook、鉴权配置和 updated_at；
+// 2. 不清空 last_test_*，避免用户保存配置后丢掉最近一次测试结果；
+// 3. channel.ID 由数据库自增，调用方不应依赖传入 ID。
+func (d *ChannelDAO) UpsertUserNotificationChannel(ctx context.Context, channel *notificationmodel.UserNotificationChannel) error {
+	if err := d.ensureDB(); err != nil {
+		return err
+	}
+	if channel == nil || channel.UserID <= 0 || channel.Channel == "" {
+		return errors.New("notification channel 必须包含 user_id 和 channel")
+	}
+	now := time.Now()
+	values := map[string]any{
+		"user_id":      channel.UserID,
+		"channel":      channel.Channel,
+		"enabled":      channel.Enabled,
+		"webhook_url":  channel.WebhookURL,
+		"auth_type":    channel.AuthType,
+		"bearer_token": channel.BearerToken,
+		"created_at":   now,
+		"updated_at":   now,
+	}
+	return d.db.WithContext(ctx).
+		Model(&notificationmodel.UserNotificationChannel{}).
+		Clauses(clause.OnConflict{
+			Columns: []clause.Column{{Name: "user_id"}, {Name: "channel"}},
+			DoUpdates: clause.Assignments(map[string]any{
+				"enabled":      channel.Enabled,
+				"webhook_url":  channel.WebhookURL,
+				"auth_type":    channel.AuthType,
+				"bearer_token": channel.BearerToken,
+				"updated_at":   now,
+			}),
+		}).
+		Create(values).Error
+}
+
+// GetUserNotificationChannel 查询用户指定通知通道配置。
+func (d *ChannelDAO) GetUserNotificationChannel(ctx context.Context, userID int, channel string) (*notificationmodel.UserNotificationChannel, error) {
+	if err := d.ensureDB(); err != nil {
+		return nil, err
+	}
+	if userID <= 0 || channel == "" {
+		return nil, gorm.ErrRecordNotFound
+	}
+	var row notificationmodel.UserNotificationChannel
+	err := d.db.WithContext(ctx).
+		Where("user_id = ? AND channel = ?", userID, channel).
+		First(&row).Error
+	if err != nil {
+		return nil, err
+	}
+	return &row, nil
+}
+
+// DeleteUserNotificationChannel 删除用户指定通知通道配置。
+//
+// 说明：当前表不保留软删除列；删除后再次保存会重新创建配置。
+func (d *ChannelDAO) DeleteUserNotificationChannel(ctx context.Context, userID int, channel string) error {
+	if err := d.ensureDB(); err != nil {
+		return err
+	}
+	if userID <= 0 || channel == "" {
+		return nil
+	}
+	return d.db.WithContext(ctx).
+		Where("user_id = ? AND channel = ?", userID, channel).
+		Delete(&notificationmodel.UserNotificationChannel{}).Error
+}
+
+// UpdateUserNotificationChannelTestResult 回写用户 webhook 测试结果。
+func (d *ChannelDAO) UpdateUserNotificationChannelTestResult(ctx context.Context, userID int, channel string, status string, testErr string, testedAt time.Time) error {
+	if err := d.ensureDB(); err != nil {
+		return err
+	}
+	if userID <= 0 || channel == "" {
+		return errors.New("user_id 和 channel 不能为空")
+	}
+	updates := map[string]any{
+		"last_test_status": status,
+		"last_test_error":  testErr,
+		"last_test_at":     &testedAt,
+	}
+	return d.db.WithContext(ctx).
+		Model(&notificationmodel.UserNotificationChannel{}).
+		Where("user_id = ? AND channel = ?", userID, channel).
+		Updates(updates).Error
+}
--- a/backend/services/notification/dao/connect.go
+++ b/backend/services/notification/dao/connect.go
@@ -0,0 +1,60 @@
+package dao
+
+import (
+	"fmt"
+
+	outboxinfra "github.com/LoveLosita/smartflow/backend/infra/outbox"
+	coremodel "github.com/LoveLosita/smartflow/backend/model"
+	notificationmodel "github.com/LoveLosita/smartflow/backend/services/notification/model"
+	"github.com/spf13/viper"
+	"gorm.io/driver/mysql"
+	"gorm.io/gorm"
+)
+
+// OpenDBFromConfig 创建 notification 服务自己的数据库句柄。
+//
+// 职责边界：
+// 1. 只迁移 notification_records 与 user_notification_channels；
+// 2. 不迁移主动调度、agent、userauth 或其它服务表；
+// 3. 返回的 *gorm.DB 供 notification 服务内 DAO 和 outbox consumer 复用。
+func OpenDBFromConfig() (*gorm.DB, error) {
+	host := viper.GetString("database.host")
+	port := viper.GetString("database.port")
+	user := viper.GetString("database.user")
+	password := viper.GetString("database.password")
+	dbname := viper.GetString("database.dbname")
+
+	dsn := fmt.Sprintf(
+		"%s:%s@tcp(%s:%s)/%s?charset=utf8mb4&parseTime=True&loc=Local",
+		user, password, host, port, dbname,
+	)
+
+	db, err := gorm.Open(mysql.Open(dsn), &gorm.Config{})
+	if err != nil {
+		return nil, err
+	}
+	if err = db.AutoMigrate(&notificationmodel.NotificationRecord{}, &notificationmodel.UserNotificationChannel{}); err != nil {
+		return nil, fmt.Errorf("auto migrate notification tables failed: %w", err)
+	}
+	if err = autoMigrateNotificationOutboxTable(db); err != nil {
+		return nil, err
+	}
+	return db, nil
+}
+
+// autoMigrateNotificationOutboxTable 只迁移 notification 服务自己的 outbox 物理表。
+//
+// 职责边界：
+// 1. 只负责 notification.outbox 对应表，不碰单体残留的其他业务表；
+// 2. 让独立 notification 服务可以单独启动和消费 outbox，不依赖 backend/inits 的全量迁移；
+// 3. 若后续调整 outbox 表名，只改 service catalog，不在这里硬编码。
+func autoMigrateNotificationOutboxTable(db *gorm.DB) error {
+	cfg, ok := outboxinfra.ResolveServiceConfig(outboxinfra.ServiceNotification)
+	if !ok {
+		return fmt.Errorf("resolve notification outbox config failed")
+	}
+	if err := db.Table(cfg.TableName).AutoMigrate(&coremodel.AgentOutboxMessage{}); err != nil {
+		return fmt.Errorf("auto migrate notification outbox table failed for %s (%s): %w", cfg.Name, cfg.TableName, err)
+	}
+	return nil
+}
--- a/backend/services/notification/dao/record.go
+++ b/backend/services/notification/dao/record.go
@@ -0,0 +1,170 @@
+package dao
+
+import (
+	"context"
+	"errors"
+	"time"
+
+	notificationmodel "github.com/LoveLosita/smartflow/backend/services/notification/model"
+	"gorm.io/gorm"
+)
+
+// RecordDAO 管理 notification_records 投递状态机持久化。
+//
+// 职责边界：
+// 1. 只负责通知记录的创建、去重查询、状态更新和重试扫描；
+// 2. 不负责 provider 发送、幂等锁或 outbox consumed 标记；
+// 3. 不读写 active_schedule_* 表，避免 notification 服务反向持有主动调度内部状态。
+type RecordDAO struct {
+	db *gorm.DB
+}
+
+func NewRecordDAO(db *gorm.DB) *RecordDAO {
+	return &RecordDAO{db: db}
+}
+
+func (d *RecordDAO) WithTx(tx *gorm.DB) *RecordDAO {
+	return &RecordDAO{db: tx}
+}
+
+func (d *RecordDAO) ensureDB() error {
+	if d == nil || d.db == nil {
+		return errors.New("notification record dao 未初始化")
+	}
+	return nil
+}
+
+func (d *RecordDAO) CreateNotificationRecord(ctx context.Context, record *notificationmodel.NotificationRecord) error {
+	if err := d.ensureDB(); err != nil {
+		return err
+	}
+	if record == nil {
+		return errors.New("notification record 不能为空")
+	}
+	return d.db.WithContext(ctx).Create(record).Error
+}
+
+func (d *RecordDAO) UpdateNotificationRecordFields(ctx context.Context, notificationID int64, updates map[string]any) error {
+	if err := d.ensureDB(); err != nil {
+		return err
+	}
+	if notificationID <= 0 {
+		return errors.New("notification record id 不能为空")
+	}
+	if len(updates) == 0 {
+		return nil
+	}
+	return d.db.WithContext(ctx).
+		Model(&notificationmodel.NotificationRecord{}).
+		Where("id = ?", notificationID).
+		Updates(updates).Error
+}
+
+func (d *RecordDAO) GetNotificationRecordByID(ctx context.Context, notificationID int64) (*notificationmodel.NotificationRecord, error) {
+	if err := d.ensureDB(); err != nil {
+		return nil, err
+	}
+	if notificationID <= 0 {
+		return nil, gorm.ErrRecordNotFound
+	}
+	var record notificationmodel.NotificationRecord
+	err := d.db.WithContext(ctx).Where("id = ?", notificationID).First(&record).Error
+	if err != nil {
+		return nil, err
+	}
+	return &record, nil
+}
+
+// FindNotificationRecordByDedupeKey 查询通知去重记录。
+//
+// 说明：
+// 1. notification 第一版按 channel + dedupe_key 聚合去重；
+// 2. 若返回 pending/sending/sent，上层应避免重复投递；
+// 3. 若返回 failed，上层可以复用同一条记录进入 provider retry。
+func (d *RecordDAO) FindNotificationRecordByDedupeKey(ctx context.Context, channel string, dedupeKey string) (*notificationmodel.NotificationRecord, error) {
+	if err := d.ensureDB(); err != nil {
+		return nil, err
+	}
+	if channel == "" || dedupeKey == "" {
+		return nil, gorm.ErrRecordNotFound
+	}
+	var record notificationmodel.NotificationRecord
+	err := d.db.WithContext(ctx).
+		Where("channel = ? AND dedupe_key = ?", channel, dedupeKey).
+		Order("created_at DESC, id DESC").
+		First(&record).Error
+	if err != nil {
+		return nil, err
+	}
+	return &record, nil
+}
+
+// ListRetryableNotificationRecords 查询到达重试时间的通知记录。
+//
+// 1. failed 记录按 next_retry_at 进入重试队列；
+// 2. sending 记录只有超过租约才会回收，避免仍在执行的 provider 调用被重复放大；
+// 3. 这让 retry scanner 同时覆盖显式失败重试和“发送中崩溃恢复”。
+func (d *RecordDAO) ListRetryableNotificationRecords(ctx context.Context, now time.Time, sendingStaleBefore time.Time, limit int) ([]notificationmodel.NotificationRecord, error) {
+	if err := d.ensureDB(); err != nil {
+		return nil, err
+	}
+	if limit <= 0 || now.IsZero() {
+		return []notificationmodel.NotificationRecord{}, nil
+	}
+	if sendingStaleBefore.IsZero() {
+		sendingStaleBefore = now.Add(-10 * time.Minute)
+	}
+	var records []notificationmodel.NotificationRecord
+	err := d.db.WithContext(ctx).
+		Where(
+			"(status = ? AND next_retry_at IS NOT NULL AND next_retry_at <= ?) OR (status = ? AND updated_at <= ?)",
+			notificationmodel.RecordStatusFailed,
+			now,
+			notificationmodel.RecordStatusSending,
+			sendingStaleBefore,
+		).
+		Order("next_retry_at ASC, id ASC").
+		Limit(limit).
+		Find(&records).Error
+	if err != nil {
+		return nil, err
+	}
+	return records, nil
+}
+
+// ClaimRetryableNotificationRecord 抢占一条到期失败通知，避免多实例重复调用 provider。
+//
+// 职责边界：
+// 1. 只做跨进程 claim，不发送通知、不推进最终投递状态；
+// 2. failed 到期记录和 stale sending 记录都可以被回收为 sending；
+// 3. 返回 claimed=false 表示记录已被其它实例抢走或状态已变化，调用方应跳过本次重试。
+func (d *RecordDAO) ClaimRetryableNotificationRecord(ctx context.Context, notificationID int64, now time.Time, sendingStaleBefore time.Time) (bool, error) {
+	if err := d.ensureDB(); err != nil {
+		return false, err
+	}
+	if notificationID <= 0 || now.IsZero() {
+		return false, nil
+	}
+	if sendingStaleBefore.IsZero() {
+		sendingStaleBefore = now.Add(-10 * time.Minute)
+	}
+	result := d.db.WithContext(ctx).
+		Model(&notificationmodel.NotificationRecord{}).
+		Where(
+			"id = ? AND ((status = ? AND next_retry_at IS NOT NULL AND next_retry_at <= ?) OR (status = ? AND updated_at <= ?))",
+			notificationID,
+			notificationmodel.RecordStatusFailed,
+			now,
+			notificationmodel.RecordStatusSending,
+			sendingStaleBefore,
+		).
+		Updates(map[string]any{
+			"status":        notificationmodel.RecordStatusSending,
+			"next_retry_at": nil,
+			"updated_at":    now,
+		})
+	if result.Error != nil {
+		return false, result.Error
+	}
+	return result.RowsAffected == 1, nil
+}