perf: 流式错误判定预过滤、再减一个写索引，加可选明细保留期
- 流式放行前的上游错误判定加廉价子串预过滤（"error" / response.failed）： 长思考阶段几十上百条事件不再逐条 json.Unmarshal。 - usage_logs 去掉 idx_channel_created：没有任何查询按 channel_id 过滤该表 （渠道统计走 created_at 时间窗后在内存分组），复合索引只是每行一次多余的 B-tree 插入；schema v9 迁移里显式 DROP 旧库索引。至此该表只剩 (user_id, created_at)、created_at 两个真被用到的索引。 - 新增可选保留期（OT_USAGE_RETENTION_DAYS，默认 0=永久保留）：每天分批 （2000 行/批、批间 50ms）清理过期 usage_logs，避免长期运行后时间窗查询 与备份随表增长变慢；余额流水 balance_logs 不在清理范围。 参数复测（记账稳态吞吐，3 次取中位）： - batch 64（当前）30.2k rows/s；batch 128 19.0k、batch 256 14.3k（更差，保持 64） - wal_autocheckpoint(4000)、cache_size(-64MB) 与当前配置无差别，不引入 验证：go vet、go test ./...、-race（usage/proxy/store）全绿；新增保留期三项测试 （只删明细不动流水、未开启不动数据、跨分批删完）。
This commit is contained in:
Sakurasan committed
1 parent
8e5fe16611
commit
531a03c836
9 files changed
+224
-20
No files matched your search
@@ -0,0 +1,86 @@
|
||||
package usage
|
||||
|
||||
import (
|
||||
"context"
|
||||
"log"
|
||||
"time"
|
||||
|
||||
"gorm.io/gorm"
|
||||
)
|
||||
|
||||
// 记账明细的保留期清理。
|
||||
//
|
||||
// usage_logs 只增不减：长期运行的实例上，渠道统计这类「按时间窗扫描」的查询会
|
||||
// 随表增长线性变慢，备份也会越来越大。余额流水（balance_logs）不在清理范围内——
|
||||
// 那是财务记录，需要长期保留。
|
||||
//
|
||||
// 默认关闭（retention_days=0）：删数据是不可逆操作，由部署者显式开启。
|
||||
// 清理分批进行（每批 2000 行、批间小睡），避免长时间持锁影响在线请求。
|
||||
|
||||
const (
|
||||
retentionBatch = 2000
|
||||
retentionTick = 24 * time.Hour
|
||||
)
|
||||
|
||||
// StartRetention 周期性删除超过 days 天的用量明细；days<=0 时不启动。
|
||||
// 返回停止函数（进程退出时调用）。
|
||||
func StartRetention(ctx context.Context, db *gorm.DB, days int) (stop func()) {
|
||||
if days <= 0 || db == nil {
|
||||
return func() {}
|
||||
}
|
||||
done := make(chan struct{})
|
||||
go func() {
|
||||
purgeOlderThan(ctx, db, days)
|
||||
t := time.NewTicker(retentionTick)
|
||||
defer t.Stop()
|
||||
for {
|
||||
select {
|
||||
case <-ctx.Done():
|
||||
return
|
||||
case <-done:
|
||||
return
|
||||
case <-t.C:
|
||||
purgeOlderThan(ctx, db, days)
|
||||
}
|
||||
}
|
||||
}()
|
||||
return func() {
|
||||
select {
|
||||
case <-done:
|
||||
default:
|
||||
close(done)
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// purgeOlderThan 分批删除 cutoff 之前的明细，返回删除行数。
|
||||
func purgeOlderThan(ctx context.Context, db *gorm.DB, days int) int64 {
|
||||
cutoff := time.Now().UTC().AddDate(0, 0, -days)
|
||||
var total int64
|
||||
for {
|
||||
if ctx.Err() != nil {
|
||||
return total
|
||||
}
|
||||
res := db.Exec(
|
||||
"DELETE FROM usage_logs WHERE id IN (SELECT id FROM usage_logs WHERE created_at < ? LIMIT ?)",
|
||||
cutoff, retentionBatch)
|
||||
if res.Error != nil {
|
||||
log.Printf("usage: retention purge failed: %v", res.Error)
|
||||
return total
|
||||
}
|
||||
n := res.RowsAffected
|
||||
total += n
|
||||
if n < retentionBatch {
|
||||
break
|
||||
}
|
||||
select {
|
||||
case <-ctx.Done():
|
||||
return total
|
||||
case <-time.After(50 * time.Millisecond):
|
||||
}
|
||||
}
|
||||
if total > 0 {
|
||||
log.Printf("usage: retention removed %d rows older than %s", total, cutoff.Format("2006-01-02"))
|
||||
}
|
||||
return total
|
||||
}
|
||||
@@ -0,0 +1,84 @@
|
||||
package usage
|
||||
|
||||
import (
|
||||
"context"
|
||||
"fmt"
|
||||
"testing"
|
||||
"time"
|
||||
|
||||
"github.com/openteam/server/internal/store"
|
||||
)
|
||||
|
||||
// 保留期只删过期明细，余额流水一条不动；未开启（0）时什么都不做。
|
||||
func TestRetentionPurgesOldLogsOnly(t *testing.T) {
|
||||
db := openTestDB(t)
|
||||
u := seedUser(t, db, 100)
|
||||
|
||||
old := time.Now().UTC().AddDate(0, 0, -40)
|
||||
fresh := time.Now().UTC()
|
||||
for i := 0; i < 5; i++ {
|
||||
db.Create(&store.UsageLog{TraceID: fmt.Sprintf("old-%d", i), UserID: u, ModelName: "m1",
|
||||
Cost: 0.01, Status: store.UsageStatusSuccess, CreatedAt: old})
|
||||
}
|
||||
for i := 0; i < 3; i++ {
|
||||
db.Create(&store.UsageLog{TraceID: fmt.Sprintf("new-%d", i), UserID: u, ModelName: "m1",
|
||||
Cost: 0.01, Status: store.UsageStatusSuccess, CreatedAt: fresh})
|
||||
}
|
||||
db.Create(&store.BalanceLog{UserID: u, Change: -0.05, BalanceAfter: 99.95,
|
||||
Type: store.BalanceTypeUsage, RefID: "usage:old-0"})
|
||||
|
||||
n := purgeOlderThan(context.Background(), db, 30)
|
||||
if n != 5 {
|
||||
t.Fatalf("应删除 5 条过期明细，实际 %d", n)
|
||||
}
|
||||
var logs, ledgers int64
|
||||
db.Model(&store.UsageLog{}).Count(&logs)
|
||||
db.Model(&store.BalanceLog{}).Count(&ledgers)
|
||||
if logs != 3 {
|
||||
t.Fatalf("应保留 3 条新明细，实际 %d", logs)
|
||||
}
|
||||
if ledgers != 1 {
|
||||
t.Fatalf("余额流水不应被清理，实际 %d 条", ledgers)
|
||||
}
|
||||
}
|
||||
|
||||
// 未配置保留期时不启动、不删数据。
|
||||
func TestRetentionDisabled(t *testing.T) {
|
||||
db := openTestDB(t)
|
||||
u := seedUser(t, db, 100)
|
||||
db.Create(&store.UsageLog{TraceID: "x", UserID: u, ModelName: "m1",
|
||||
Cost: 0.01, Status: store.UsageStatusSuccess,
|
||||
CreatedAt: time.Now().UTC().AddDate(0, 0, -400)})
|
||||
|
||||
stop := StartRetention(context.Background(), db, 0)
|
||||
stop()
|
||||
var n int64
|
||||
db.Model(&store.UsageLog{}).Count(&n)
|
||||
if n != 1 {
|
||||
t.Fatalf("未开启保留期时不应删除数据，实际剩 %d", n)
|
||||
}
|
||||
}
|
||||
|
||||
// 分批删除：行数超过一批时应全部删完。
|
||||
func TestRetentionBatches(t *testing.T) {
|
||||
db := openTestDB(t)
|
||||
u := seedUser(t, db, 100)
|
||||
old := time.Now().UTC().AddDate(0, 0, -90)
|
||||
logs := make([]store.UsageLog, 0, retentionBatch+50)
|
||||
for i := 0; i < retentionBatch+50; i++ {
|
||||
logs = append(logs, store.UsageLog{TraceID: fmt.Sprintf("b-%d", i), UserID: u,
|
||||
ModelName: "m1", Cost: 0.001, Status: store.UsageStatusSuccess, CreatedAt: old})
|
||||
}
|
||||
if err := db.CreateInBatches(logs, 500).Error; err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
n := purgeOlderThan(context.Background(), db, 30)
|
||||
if n != int64(len(logs)) {
|
||||
t.Fatalf("应删除 %d 条，实际 %d", len(logs), n)
|
||||
}
|
||||
var left int64
|
||||
db.Model(&store.UsageLog{}).Count(&left)
|
||||
if left != 0 {
|
||||
t.Fatalf("应全部删完，实际剩 %d", left)
|
||||
}
|
||||
}
|
||||
Reference in new issue
Block a user