Files
oci-portal/internal/service/tenantdelete.go
T
2026-07-22 16:51:23 +08:00

376 lines
11 KiB
Go

package service
import (
"context"
"encoding/json"
"fmt"
"log"
"strings"
"gorm.io/gorm"
"gorm.io/gorm/clause"
"oci-portal/internal/model"
)
type tenantDeleteResult struct {
config model.OciConfig
deletedTaskIDs []uint
channelsGone bool
}
type tenantTaskAction struct {
task model.Task
deleteTask bool
payload string
}
type tenancyCacheInvalidator interface {
InvalidateTenancy(tenancyOCID string)
}
func (s *OciConfigService) deleteTenant(ctx context.Context, id uint) (*tenantDeleteResult, error) {
result := &tenantDeleteResult{}
err := s.db.WithContext(ctx).Transaction(func(tx *gorm.DB) error {
return s.deleteTenantInTx(tx, id, result)
})
if err != nil {
return nil, fmt.Errorf("delete oci config %d: %w", id, err)
}
return result, nil
}
func (s *OciConfigService) deleteTenantInTx(tx *gorm.DB, id uint, result *tenantDeleteResult) error {
if err := lockTenant(tx, id, &result.config); err != nil {
return err
}
if err := s.deleteTenantTasks(tx, id, result); err != nil {
return err
}
if err := deleteTenantEvents(tx, id); err != nil {
return err
}
if err := deleteTenantAI(tx, id, result); err != nil {
return err
}
if err := deleteTenantSnapshots(tx, id); err != nil {
return err
}
return deleteTenantConfig(tx, id)
}
func lockTenant(tx *gorm.DB, id uint, cfg *model.OciConfig) error {
err := tx.Clauses(clause.Locking{Strength: "UPDATE"}).First(cfg, id).Error
if err != nil {
return fmt.Errorf("load tenant: %w", err)
}
return nil
}
func (s *OciConfigService) deleteTenantTasks(tx *gorm.DB, id uint, result *tenantDeleteResult) error {
var tasks []model.Task
err := tx.Where("type IN ?", []string{
model.TaskTypeSnatch, model.TaskTypeHealthCheck, model.TaskTypeCost,
}).Find(&tasks).Error
if err != nil {
return fmt.Errorf("load tenant tasks: %w", err)
}
for i := range tasks {
action, ok, err := planTenantTask(tasks[i], id)
if err != nil {
return err
}
if !ok {
continue
}
if err := applyTenantTask(tx, action, result); err != nil {
return err
}
}
return nil
}
func planTenantTask(task model.Task, id uint) (tenantTaskAction, bool, error) {
switch task.Type {
case model.TaskTypeSnatch:
return planSnatchTask(task, id)
case model.TaskTypeHealthCheck, model.TaskTypeCost:
return planMultiTenantTask(task, id)
default:
return tenantTaskAction{}, false, nil
}
}
func planSnatchTask(task model.Task, id uint) (tenantTaskAction, bool, error) {
var payload snatchPayload
if err := json.Unmarshal([]byte(task.Payload), &payload); err != nil {
logSkippedTask(task, err)
return tenantTaskAction{}, false, nil
}
if payload.OciConfigID != id {
return tenantTaskAction{}, false, nil
}
return tenantTaskAction{task: task, deleteTask: true}, true, nil
}
// logSkippedTask 记录 payload 无法解析而被跳过的任务:坏数据无法归属租户,
// fail-closed 会永久阻断删除且只能手工修库;改为保留原任务(不删不改)并放行删除。
func logSkippedTask(task model.Task, err error) {
log.Printf("tenant delete: 任务 %d(%s)payload 无法解析,跳过处理: %v", task.ID, task.Type, err)
}
func planMultiTenantTask(task model.Task, id uint) (tenantTaskAction, bool, error) {
payload, err := decodeConfigIDs(task)
if err != nil {
logSkippedTask(task, err)
return tenantTaskAction{}, false, nil
}
if len(payload.OciConfigIDs) == 0 {
return tenantTaskAction{task: task, payload: task.Payload}, true, nil
}
remaining, found := removeConfigID(payload.OciConfigIDs, id)
if !found {
return tenantTaskAction{}, false, nil
}
if len(remaining) == 0 {
return tenantTaskAction{task: task, deleteTask: true}, true, nil
}
payload.OciConfigIDs = remaining
raw, err := json.Marshal(payload)
if err != nil {
return tenantTaskAction{}, false, fmt.Errorf("encode task %d payload: %w", task.ID, err)
}
return tenantTaskAction{task: task, payload: string(raw)}, true, nil
}
func decodeConfigIDs(task model.Task) (healthCheckPayload, error) {
var payload healthCheckPayload
if strings.TrimSpace(task.Payload) == "" {
return payload, nil
}
if err := json.Unmarshal([]byte(task.Payload), &payload); err != nil {
return payload, fmt.Errorf("parse task %d payload: %w", task.ID, err)
}
return payload, nil
}
func removeConfigID(ids []uint, id uint) ([]uint, bool) {
remaining := make([]uint, 0, len(ids))
found := false
for _, candidate := range ids {
if candidate == id {
found = true
continue
}
remaining = append(remaining, candidate)
}
return remaining, found
}
func applyTenantTask(tx *gorm.DB, action tenantTaskAction, result *tenantDeleteResult) error {
if err := deleteTaskHistory(tx, action.task.ID); err != nil {
return err
}
if action.deleteTask {
if err := tx.Delete(&model.Task{}, action.task.ID).Error; err != nil {
return fmt.Errorf("delete task %d: %w", action.task.ID, err)
}
result.deletedTaskIDs = append(result.deletedTaskIDs, action.task.ID)
return nil
}
return resetTenantTask(tx, action)
}
func deleteTaskHistory(tx *gorm.DB, taskID uint) error {
if err := tx.Where("task_id = ?", taskID).Delete(&model.TaskLog{}).Error; err != nil {
return fmt.Errorf("delete task %d logs: %w", taskID, err)
}
key := deadAliasKey(taskID)
if err := tx.Where("key = ?", key).Delete(&model.Setting{}).Error; err != nil {
return fmt.Errorf("delete task %d state: %w", taskID, err)
}
return nil
}
func resetTenantTask(tx *gorm.DB, action tenantTaskAction) error {
updates := map[string]any{
"payload": action.payload, "last_run_at": nil,
"last_error": "", "run_count": 0,
}
err := tx.Model(&model.Task{}).Where("id = ?", action.task.ID).Updates(updates).Error
if err != nil {
return fmt.Errorf("reset task %d: %w", action.task.ID, err)
}
return nil
}
func deleteTenantEvents(tx *gorm.DB, id uint) error {
if err := lockTenantEventRows(tx, id); err != nil {
return fmt.Errorf("load tenant log events: %w", err)
}
if err := deleteWhere(tx, &model.LogEvent{}, "oci_config_id = ?", id); err != nil {
return fmt.Errorf("delete tenant log events: %w", err)
}
return nil
}
// lockTenantEventRows 对租户全部日志事件行加 FOR UPDATE 锁(SQLite 忽略,
// MySQL/PG 阻塞并发写入);事件可达数万,ID 不回传拼接 SQL,删除用条件语句,
// 避免绑定变量上限。
func lockTenantEventRows(tx *gorm.DB, id uint) error {
var rows []model.LogEvent
return tx.Clauses(clause.Locking{Strength: "UPDATE"}).Select("id").
Where("oci_config_id = ?", id).Order("id").Find(&rows).Error
}
func deleteTenantAI(tx *gorm.DB, id uint, result *tenantDeleteResult) error {
channelIDs, err := lockedAiChannelIDs(tx, id)
if err != nil {
return fmt.Errorf("load tenant AI channels: %w", err)
}
if len(channelIDs) == 0 {
return nil
}
if err := lockAiCallRows(tx, channelIDs); err != nil {
return fmt.Errorf("load tenant AI calls: %w", err)
}
if err := deleteTenantAIRows(tx, channelIDs); err != nil {
return err
}
result.channelsGone = true
return reconcileAiProbeRows(tx, result)
}
func lockedAiChannelIDs(tx *gorm.DB, id uint) ([]uint, error) {
var rows []model.AiChannel
err := tx.Clauses(clause.Locking{Strength: "UPDATE"}).Select("id").
Where("oci_config_id = ?", id).Order("id").Find(&rows).Error
ids := make([]uint, 0, len(rows))
for _, row := range rows {
ids = append(ids, row.ID)
}
return ids, err
}
// lockAiCallRows 对租户渠道的全部调用日志行加 FOR UPDATE 锁(SQLite 忽略,
// MySQL/PG 阻塞并发写入);调用日志可达数万,ID 不回传拼接 SQL,
// 内容日志删除用子查询,避免绑定变量上限。
func lockAiCallRows(tx *gorm.DB, channelIDs []uint) error {
var rows []model.AiCallLog
return tx.Clauses(clause.Locking{Strength: "UPDATE"}).Select("id").
Where("channel_id IN ?", channelIDs).Order("id").Find(&rows).Error
}
func deleteTenantAIRows(tx *gorm.DB, channelIDs []uint) error {
callIDs := tx.Model(&model.AiCallLog{}).Select("id").Where("channel_id IN ?", channelIDs)
if err := deleteWhere(tx, &model.AiContentLog{}, "call_log_id IN (?)", callIDs); err != nil {
return fmt.Errorf("delete tenant AI content logs: %w", err)
}
steps := []struct {
name string
value any
}{
{"AI call logs", &model.AiCallLog{}},
{"AI model cache", &model.AiModelCache{}},
{"AI channels", &model.AiChannel{}},
}
for _, step := range steps {
if err := deleteAIChannelRows(tx, step.value, channelIDs); err != nil {
return fmt.Errorf("delete tenant %s: %w", step.name, err)
}
}
return nil
}
func deleteAIChannelRows(tx *gorm.DB, value any, channelIDs []uint) error {
column := "channel_id"
if _, ok := value.(*model.AiChannel); ok {
column = "id"
}
return deleteWhere(tx, value, column+" IN ?", channelIDs)
}
func reconcileAiProbeRows(tx *gorm.DB, result *tenantDeleteResult) error {
var channelCount int64
if err := tx.Model(&model.AiChannel{}).Count(&channelCount).Error; err != nil {
return fmt.Errorf("count remaining AI channels: %w", err)
}
var ids []uint
err := tx.Model(&model.Task{}).Where("type = ?", model.TaskTypeAiProbe).Pluck("id", &ids).Error
if err != nil {
return fmt.Errorf("load AI probe task: %w", err)
}
for _, id := range ids {
if err := deleteTaskHistory(tx, id); err != nil {
return err
}
if channelCount == 0 {
if err := tx.Delete(&model.Task{}, id).Error; err != nil {
return fmt.Errorf("delete AI probe task %d: %w", id, err)
}
result.deletedTaskIDs = append(result.deletedTaskIDs, id)
continue
}
if err := resetTaskHistoryFields(tx, id); err != nil {
return err
}
}
return nil
}
func resetTaskHistoryFields(tx *gorm.DB, id uint) error {
updates := map[string]any{"last_run_at": nil, "last_error": "", "run_count": 0}
if err := tx.Model(&model.Task{}).Where("id = ?", id).Updates(updates).Error; err != nil {
return fmt.Errorf("reset task %d history: %w", id, err)
}
return nil
}
func deleteTenantSnapshots(tx *gorm.DB, id uint) error {
steps := []struct {
name string
value any
}{
{"check snapshots", &model.CheckSnapshot{}},
{"cost snapshots", &model.CostSnapshot{}},
{"region cache", &model.RegionCache{}},
{"compartment cache", &model.CompartmentCache{}},
}
for _, step := range steps {
if err := deleteWhere(tx, step.value, "oci_config_id = ?", id); err != nil {
return fmt.Errorf("delete tenant %s: %w", step.name, err)
}
}
if err := tx.Where("key = ?", secretKey(id)).Delete(&model.Setting{}).Error; err != nil {
return fmt.Errorf("delete tenant webhook secret: %w", err)
}
return nil
}
func deleteTenantConfig(tx *gorm.DB, id uint) error {
res := tx.Delete(&model.OciConfig{}, id)
if res.Error != nil {
return fmt.Errorf("delete tenant row: %w", res.Error)
}
if res.RowsAffected == 0 {
return gorm.ErrRecordNotFound
}
return nil
}
func deleteWhere(tx *gorm.DB, value any, query string, args ...any) error {
return tx.Where(query, args...).Delete(value).Error
}
func (s *OciConfigService) afterTenantDelete(ctx context.Context, result *tenantDeleteResult) {
// 数据已提交,客户端断开不应中断必需的运行时对齐。
ctx = context.WithoutCancel(ctx)
s.InvalidateAuditCache(result.config.ID)
if client, ok := s.client.(tenancyCacheInvalidator); ok {
client.InvalidateTenancy(result.config.TenancyOCID)
}
if s.cleanupTasks != nil {
s.cleanupTasks.ApplyTenantCleanup(ctx, result.deletedTaskIDs, result.channelsGone)
}
}