Files
aiops/backend/pkg/retention/retention.go
T
Logan.Li fd60361d93 chore: 统一本地服务器与远程仓库源码
- 合并 main 与 master 历史并以服务器结构为基线
- 保留迁移、数据保留、AWS 账单与部署安全能力
- 建立 SDD 规格和 CI 验证闭环
2026-08-05 20:03:25 +08:00

128 lines
4.2 KiB
Go

// 高频数据保留与批量清理,避免 SQLite/MySQL 大表无限增长。 // Logan.Li
package retention
import (
"fmt"
"time"
"ezops/config"
"ezops/model"
"gorm.io/gorm"
)
const defaultBatchSize = 10000
// legacyMetricsAggregate 兼容旧聚合表;新写入统一使用 monitor_metric_aggregates。 // Logan.Li
type legacyMetricsAggregate struct {
ID uint `gorm:"primaryKey"`
BucketTime time.Time `gorm:"column:bucket_time"`
}
// TableName 固定旧聚合表名称,避免历史数据在迁移完成前无限增长。 // Logan.Li
func (legacyMetricsAggregate) TableName() string {
return "monitor_metrics_aggregate"
}
// ItemResult 描述单张表的清理结果,供系统任务审计与告警使用。 // Logan.Li
type ItemResult struct {
Table string `json:"table"`
Retention int `json:"retention_days"`
RowsDeleted int64 `json:"rows_deleted"`
Skipped bool `json:"skipped,omitempty"`
}
// Result 汇总一次全量高频数据清理的可审计结果。 // Logan.Li
type Result struct {
StartedAt time.Time `json:"started_at"`
FinishedAt time.Time `json:"finished_at"`
Items []ItemResult `json:"items"`
Total int64 `json:"total_rows_deleted"`
}
// Run 仅清理已存在的高频表;每批先读取主键再删除,避免一次 DELETE 长事务。 // Logan.Li
func Run(db *gorm.DB, cfg config.RetentionConfig) (Result, error) {
result := Result{StartedAt: time.Now()}
batchSize := normalizeBatchSize(cfg.CleanupBatchSize)
items := []struct {
model interface{}
table string
column string
retentionDays int
}{
{&model.AgentReport{}, "monitor_agent_reports", "report_time", cfg.AgentReportsDays},
{&model.MetricData{}, "monitor_metric_data", "timestamp", cfg.MetricHistoryDays},
{&model.MetricHistory{}, "monitor_metric_histories", "timestamp", cfg.MetricHistoryDays},
{&model.MetricsAggregate{}, "monitor_metric_aggregates", "bucket_time", cfg.MetricAggregateDays},
{&legacyMetricsAggregate{}, "monitor_metrics_aggregate", "bucket_time", cfg.MetricAggregateDays},
{&model.SNMPMetric{}, "monitor_snmp_samples", "timestamp", cfg.SNMPSamplesDays},
{&model.SiteResult{}, "monitor_site_results", "checked_at", cfg.SiteResultsDays},
{&model.LogEntry{}, "sys_log_entries", "timestamp", cfg.LogEntriesDays},
}
for _, item := range items {
retentionDays := normalizeRetentionDays(item.retentionDays)
entry := ItemResult{Table: item.table, Retention: retentionDays}
if !db.Migrator().HasTable(item.model) {
entry.Skipped = true
result.Items = append(result.Items, entry)
continue
}
cutoff := time.Now().AddDate(0, 0, -retentionDays)
var err error
if item.table == "monitor_metric_histories" || item.table == "monitor_snmp_samples" {
err = deleteInBatches(db, item.model, item.column+" < ?", cutoff.Unix(), batchSize, &entry.RowsDeleted)
} else {
err = deleteInBatches(db, item.model, item.column+" < ?", cutoff, batchSize, &entry.RowsDeleted)
}
if err != nil {
return result, fmt.Errorf("清理 %s 失败: %w", item.table, err)
}
result.Total += entry.RowsDeleted
result.Items = append(result.Items, entry)
}
result.FinishedAt = time.Now()
return result, nil
}
// deleteInBatches 通过主键分页删除,兼容 SQLite 与 MySQL,失败时不会隐藏受影响行数。 // Logan.Li
func deleteInBatches(db *gorm.DB, target interface{}, condition string, value interface{}, batchSize int, total *int64) error {
for {
var ids []uint
if err := db.Model(target).Where(condition, value).Order("id ASC").Limit(batchSize).Pluck("id", &ids).Error; err != nil {
return err
}
if len(ids) == 0 {
return nil
}
deleted := db.Where("id IN ?", ids).Delete(target)
if deleted.Error != nil {
return deleted.Error
}
if deleted.RowsAffected != int64(len(ids)) {
return fmt.Errorf("批量删除行数不一致: 预期 %d,实际 %d", len(ids), deleted.RowsAffected)
}
*total += deleted.RowsAffected
}
}
func normalizeRetentionDays(days int) int {
if days <= 0 {
return 30
}
return days
}
func normalizeBatchSize(size int) int {
if size <= 0 {
return defaultBatchSize
}
if size > defaultBatchSize {
return defaultBatchSize
}
return size
}