Files
tbphp bd539713da perf: 优化测试性能与 CI/发布流水线耗时 (#465)
* test(control): 消除测试 fixture 中的重复密钥派生

- internal/control:fixture 构造中 388 次重复执行 PBKDF2(10 万轮
  SHA256)密钥派生,改为共享单例,仅派生一次
- 新增 internal/testutil/encryptiontest,提供按 key material 值
  memoize 的共享 encryption.Service,横向覆盖 gateway、subscription、
  execution/cpa 中同样存在的重复派生(PBKDF2 占比最高达 79%)
- aesService 构造后只读,跨 fixture(含并行测试)共享安全;container
  包同类调用经排查确认属于被测装配流程本身,不适用本次改造

* test(state): 迁移测试改用共享 schema 模板

state/loader 中 openMigratedDatabase 被 21 次调用,每次都完整走一遍
迁移注册表;改为复用 sqlitetest.OpenMigrated 的模板缓存,与 control、
requestlog 部分测试已采用的方式一致。app、container、requestlog 中同
类调用点经逐点排查,均属被测对象本身或依赖真实文件/账本状态,不适用。

* ci(release): 消除 CI 与发布流水线中的耗时浪费

- race-tests 独立 setup-go 缓存命名空间,避免与 test/database-contract
  共用 key 导致 race-instrumented GOCACHE 永远无法写回
- 移除 publish-images 中零命中的 GHA buildx image cache,同时消除其
  失败会拖累发布状态判定为 partial/blocked 的关联风险
- 三个只读 job(仅用 buildx imagetools 查询)移除多余的 buildx 初始化
- native-artifact-smoke 改为按需下载单个 binary + 独立 checksums,
  不再下载完整 release-assets

* test(control): 第一批测试标记 t.Parallel(35 个文件)

internal/control 651 个顶层测试里只有 5.8% 标了 t.Parallel(),其余
在包内是排队串行执行的;race-tests 总耗时现在完全由该包的最长执行时
间决定(213s),是唯一剩余的瓶颈。

本批覆盖 35 个文件、299 个函数。排除 18 个不适用的函数:
- 16 个直接操作全局 logrus 状态的测试(catalog_sync_test.go 等 8 个
  文件),继续保持串行;Go 的 t.Parallel() 调度语义保证未标记的测试
  会在所有并行测试开始前完整跑完,两者不会交叉执行
- external_database_integration_test.go 的 2 个测试依赖
  GPT_LOAD_DATABASE_TEST_DSN:race-tests 中因未设置而被 t.Skip,标记
  并行无收益;database-contract job 中会共享同一个外部数据库实例真
  实执行,标记并行有正确性风险(唯一索引冲突/断言假设独占)

已排查确认安全:t.Setenv/os.Setenv 均为 0 处,无硬编码端口占用,每个
fixture 走独立 :memory: 数据库;本地 -count=3 重复验证无 flaky。

剩余约 297 个函数(36 个文件)作为第二批,待本批 CI 实测确认收益后
再决定是否推进。

* test(control): 第二批测试标记 t.Parallel,补齐排除说明(36 个文件)

覆盖剩余 297 个函数、36 个文件,加上第一批的 299 个,internal/control
除 18 个明确排除的函数外全部标记 t.Parallel()。

为 18 个排除函数逐一补充一行注释说明原因,避免日后被误认为遗漏而被
顺手加上 t.Parallel():
- 16 个劫持全局 logrus 状态的测试
- 2 个依赖 GPT_LOAD_DATABASE_TEST_DSN 共享外部数据库的测试

本地 -count=3 重复验证、全量 go test -count=1 . ./internal/...(58
包 ok)均通过。

* fix(release): 采纳 Review 反馈,修正文档措辞与产物保留期

- encryptiontest.Service 的文档注释修正为明确说明缓存仅在单个包的
  测试二进制内有效,不跨包共享(每个包的 go test 是独立进程)
- native-artifact-smoke 依赖的 binary-*、release-checksums 保留期
  从 1 天提高到与 release-assets 一致的 14 天:失败的 release run
  若延迟超过 1 天才重试,此前较短的保留期会导致产物过期、下载失败,
  阻塞发布流程

以下反馈经核实判定为不成立,未采纳:
- validation_test.go 中两个测试被指"竞争全局并发槽位"——经查
  newValidationWorkerForTest 每次调用构造全新独立实例(worker、
  channel、mutation coordinator 等),validationConcurrency 只是
  编译期常量,两个测试之间不存在任何运行时共享状态
2026-08-24 20:07:11 +08:00

332 lines
11 KiB
Go

package control
import (
"bytes"
"context"
"encoding/hex"
"encoding/json"
"errors"
"log"
"strings"
"testing"
"time"
"github.com/sirupsen/logrus"
"gorm.io/gorm"
"gorm.io/gorm/logger"
"gpt-load/internal/channel"
"gpt-load/internal/platform/encryption"
"gpt-load/internal/state"
"gpt-load/internal/storage/models"
)
const bootstrapMarkerForTest = models.InternalSystemSettingPrefix + "bootstrap.default_access_key.v1"
var errBootstrapEncryption = errors.New("forced bootstrap encryption failure")
func TestEnsureInitialStateCreatesDefaultAccessKeyAndMarker(t *testing.T) {
t.Parallel()
fixture := newServiceFixture(t)
fixture.service.random = bytes.NewReader([]byte{
0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07,
0x08, 0x09, 0x0a, 0x0b, 0x0c, 0x0d, 0x0e, 0x0f,
})
beforeRevision := fixture.manager.Current().Revision
if err := fixture.service.EnsureInitialState(context.Background()); err != nil {
t.Fatalf("EnsureInitialState() error = %v", err)
}
var row models.AccessKey
if err := fixture.db.First(&row).Error; err != nil {
t.Fatalf("query default AccessKey: %v", err)
}
plaintext, err := fixture.encryption.Decrypt(row.KeyValue)
if err != nil {
t.Fatalf("Decrypt(default key) error = %v", err)
}
if !strings.HasPrefix(plaintext, "sk-gl-") || len(plaintext) != len("sk-gl-")+32 {
t.Fatal("default plaintext shape is invalid")
}
if _, err := hex.DecodeString(strings.TrimPrefix(plaintext, "sk-gl-")); err != nil {
t.Fatalf("default plaintext suffix is not hex: %v", err)
}
if row.KeyHash != fixture.encryption.Hash(plaintext) {
t.Fatal("default access key hash does not match plaintext")
}
if row.Name != "Default" || row.Status != string(state.AccessKeyStatusActive) ||
row.RPMLimit != 0 {
t.Fatalf("default row = %#v", row)
}
var filters AccessKeyFilters
if err := json.Unmarshal(row.Filters, &filters); err != nil {
t.Fatalf("decode default filters: %v", err)
}
if filters.Groups == nil || filters.Protocols == nil || filters.Models == nil ||
len(filters.Groups) != 0 || len(filters.Protocols) != 0 || len(filters.Models) != 0 {
t.Fatalf("default filters = %#v, want three non-nil empty slices", filters)
}
var marker models.SystemSetting
if err := fixture.db.First(&marker, "key = ?", bootstrapMarkerForTest).Error; err != nil {
t.Fatalf("query bootstrap marker: %v", err)
}
if marker.Key != bootstrapMarkerForTest || marker.Value != "true" {
t.Fatalf("bootstrap marker = %#v, want exact JSON true", marker)
}
if got := fixture.manager.Current().Revision; got != beforeRevision {
t.Fatalf("snapshot revision = %d, want unchanged %d", got, beforeRevision)
}
if len(fixture.manager.Current().AccessKeysByHash) != 0 {
t.Fatal("EnsureInitialState() published a runtime snapshot")
}
}
func TestEnsureInitialStateWithExistingAccessKeyOnlyWritesMarker(t *testing.T) {
t.Parallel()
fixture := newServiceFixture(t)
const plaintext = "gl-existing-access-key"
ciphertext, err := fixture.encryption.Encrypt(plaintext)
if err != nil {
t.Fatalf("Encrypt(existing key) error = %v", err)
}
existing := models.AccessKey{
Name: "Existing",
KeyValue: ciphertext,
KeyHash: fixture.encryption.Hash(plaintext),
KeySuffix: "0000",
Status: string(state.AccessKeyStatusActive),
Filters: models.JSON(`{"groups":[],"protocols":[],"models":[]}`),
}
if err := fixture.db.Create(&existing).Error; err != nil {
t.Fatalf("create existing AccessKey: %v", err)
}
if err := fixture.service.EnsureInitialState(context.Background()); err != nil {
t.Fatalf("EnsureInitialState() error = %v", err)
}
var rows []models.AccessKey
if err := fixture.db.Order("id ASC").Find(&rows).Error; err != nil {
t.Fatalf("query AccessKeys: %v", err)
}
if len(rows) != 1 || rows[0].ID != existing.ID {
t.Fatalf("AccessKeys = %#v, want only existing row", rows)
}
decrypted, err := fixture.encryption.Decrypt(rows[0].KeyValue)
if err != nil || decrypted != plaintext {
t.Fatalf("existing credential = %q, %v, want unchanged", decrypted, err)
}
assertBootstrapMarkerCount(t, fixture, 1)
}
func TestEnsureInitialStateDoesNotLogExpectedMarkerMissAsError(t *testing.T) {
t.Parallel()
fixture := newServiceFixture(t)
var logs bytes.Buffer
fixture.service.db = fixture.db.Session(&gorm.Session{
Logger: logger.New(log.New(&logs, "\r\n", log.LstdFlags), logger.Config{
SlowThreshold: 200 * time.Millisecond,
LogLevel: logger.Warn,
ParameterizedQueries: true,
Colorful: true,
}),
})
if err := fixture.service.EnsureInitialState(context.Background()); err != nil {
t.Fatalf("EnsureInitialState() error = %v", err)
}
if strings.Contains(logs.String(), "record not found") {
t.Fatalf("expected marker miss was logged as an error: %q", logs.String())
}
}
func TestEnsureInitialStateIsIdempotentWhenMarkerExists(t *testing.T) {
t.Parallel()
fixture := newServiceFixture(t)
if err := fixture.db.Create(&models.SystemSetting{
Key: bootstrapMarkerForTest, Value: "true",
}).Error; err != nil {
t.Fatalf("create bootstrap marker: %v", err)
}
if err := fixture.db.Exec("DROP TABLE access_keys").Error; err != nil {
t.Fatalf("drop AccessKey table: %v", err)
}
if err := fixture.service.EnsureInitialState(context.Background()); err != nil {
t.Fatalf("EnsureInitialState() with marker error = %v", err)
}
assertBootstrapMarkerCount(t, fixture, 1)
}
func TestEnsureInitialStateDoesNotRecreateDeletedFinalKey(t *testing.T) {
t.Parallel()
fixture := newServiceFixture(t)
fixture.service.random = bytes.NewReader(make([]byte, 16))
if err := fixture.service.EnsureInitialState(context.Background()); err != nil {
t.Fatalf("first EnsureInitialState() error = %v", err)
}
if err := fixture.db.Where("1 = 1").Delete(&models.AccessKey{}).Error; err != nil {
t.Fatalf("delete final AccessKey: %v", err)
}
if err := fixture.service.EnsureInitialState(context.Background()); err != nil {
t.Fatalf("second EnsureInitialState() error = %v", err)
}
assertAccessKeyCount(t, fixture, 0)
assertBootstrapMarkerCount(t, fixture, 1)
}
func TestEnsureInitialStateRollsBackWhenEncryptionFails(t *testing.T) {
t.Parallel()
fixture := newServiceFixture(t)
fixture.service.random = bytes.NewReader(make([]byte, 16))
fixture.service.encryption = bootstrapFailingEncryptService{Service: fixture.encryption}
err := fixture.service.EnsureInitialState(context.Background())
if !errors.Is(err, errBootstrapEncryption) {
t.Fatalf("EnsureInitialState() error = %v, want encryption failure", err)
}
assertAccessKeyCount(t, fixture, 0)
assertBootstrapMarkerCount(t, fixture, 0)
}
func TestEnsureInitialStateRollsBackAccessKeyWhenMarkerWriteFails(t *testing.T) {
t.Parallel()
fixture := newServiceFixture(t)
fixture.service.random = bytes.NewReader(make([]byte, 16))
if err := fixture.db.Exec(`
CREATE TRIGGER reject_bootstrap_marker
BEFORE INSERT ON system_settings
WHEN NEW.key = '_internal.bootstrap.default_access_key.v1'
BEGIN
SELECT RAISE(ABORT, 'marker rejected');
END;
`).Error; err != nil {
t.Fatalf("create marker rejection trigger: %v", err)
}
if err := fixture.service.EnsureInitialState(context.Background()); err == nil {
t.Fatal("EnsureInitialState() error = nil, want marker rejection")
}
assertAccessKeyCount(t, fixture, 0)
assertBootstrapMarkerCount(t, fixture, 0)
}
func TestEnsureInitialStateDoesNotLogPlaintext(t *testing.T) {
// 不标记 t.Parallel():本测试劫持了全局 logrus 输出/格式,与其他并行测试同时运行会互相覆盖断言。
fixture := newServiceFixture(t)
randomBytes := bytes.Repeat([]byte{0xab}, 16)
fixture.service.random = bytes.NewReader(randomBytes)
expectedPlaintext := "sk-gl-" + hex.EncodeToString(randomBytes)
var logs bytes.Buffer
logger := logrus.StandardLogger()
previousOutput, previousFormatter, previousLevel := logger.Out, logger.Formatter, logger.GetLevel()
logrus.SetOutput(&logs)
logrus.SetFormatter(&logrus.JSONFormatter{DisableTimestamp: true})
logrus.SetLevel(logrus.DebugLevel)
t.Cleanup(func() {
logrus.SetOutput(previousOutput)
logrus.SetFormatter(previousFormatter)
logrus.SetLevel(previousLevel)
})
if err := fixture.service.EnsureInitialState(context.Background()); err != nil {
t.Fatalf("EnsureInitialState() error = %v", err)
}
if strings.Contains(logs.String(), expectedPlaintext) {
t.Fatalf("bootstrap logs exposed plaintext: %q", logs.String())
}
}
func TestEnsureInitialStateRecoversInterruptedSubscriptionAuth(t *testing.T) {
t.Parallel()
fixture := newServiceFixture(t)
now := time.UnixMilli(1_800_000_000_000)
fixture.service.now = func() time.Time { return now }
group := validControlGroup("subscription-recovery")
group.ChannelID = string(channel.Codex)
group.ConnectionType = models.ConnectionTypeSubscription
if err := fixture.db.Create(group).Error; err != nil {
t.Fatal(err)
}
credential := models.Credential{
GroupID: group.ID, Data: "cipher", Fingerprint: "secret", IdentityFingerprint: "identity",
SecretVersion: 2, AuthState: models.CredentialAuthStateRefreshing,
Status: models.CredentialStatusActive, CreatedAtMS: now.Add(-time.Hour).UnixMilli(), UpdatedAtMS: now.Add(-time.Hour).UnixMilli(),
}
if err := fixture.db.Create(&credential).Error; err != nil {
t.Fatal(err)
}
stage := models.CredentialStage{
ID: "00000000-0000-4000-8000-000000000901", ChannelID: "codex",
ConnectionType: models.ConnectionTypeSubscription, AuthorizationMethod: "browser_oauth",
Status: models.CredentialStageExchanging, EncryptedPayload: "encrypted-stage", PayloadSchemaVersion: 1,
SafeSummaryJSON: models.JSON(`{}`), ExpiresAtMS: now.Add(time.Minute).UnixMilli(),
CreatedAtMS: now.Add(-time.Minute).UnixMilli(), UpdatedAtMS: now.Add(-time.Minute).UnixMilli(),
}
if err := fixture.db.Create(&stage).Error; err != nil {
t.Fatal(err)
}
if err := fixture.service.EnsureInitialState(t.Context()); err != nil {
t.Fatal(err)
}
if err := fixture.db.Take(&credential, credential.ID).Error; err != nil {
t.Fatal(err)
}
if credential.AuthState != models.CredentialAuthStateOutcomeUnknown || credential.AuthErrorCode != "refresh_interrupted" {
t.Fatalf("credential = %#v", credential)
}
if err := fixture.db.Take(&stage, "id = ?", stage.ID).Error; err != nil {
t.Fatal(err)
}
if stage.Status != models.CredentialStageOutcomeUnknown || stage.EncryptedPayload != "" || stage.ErrorCode != "authorization_exchange_interrupted" {
t.Fatalf("stage = %#v", stage)
}
}
type bootstrapFailingEncryptService struct {
encryption.Service
}
func (bootstrapFailingEncryptService) Encrypt(string) (string, error) {
return "", errBootstrapEncryption
}
func assertAccessKeyCount(t *testing.T, fixture serviceFixture, want int64) {
t.Helper()
var count int64
if err := fixture.db.Model(&models.AccessKey{}).Count(&count).Error; err != nil {
t.Fatalf("count AccessKeys: %v", err)
}
if count != want {
t.Fatalf("AccessKey count = %d, want %d", count, want)
}
}
func assertBootstrapMarkerCount(t *testing.T, fixture serviceFixture, want int64) {
t.Helper()
var count int64
if err := fixture.db.Model(&models.SystemSetting{}).
Where("key = ?", bootstrapMarkerForTest).Count(&count).Error; err != nil {
t.Fatalf("count bootstrap marker: %v", err)
}
if count != want {
t.Fatalf("bootstrap marker count = %d, want %d", count, want)
}
}
func assertModelPriceCount(t *testing.T, fixture serviceFixture, want int64) {
t.Helper()
var count int64
if err := fixture.db.Model(&models.ModelPrice{}).Count(&count).Error; err != nil {
t.Fatalf("count ModelPrice rows: %v", err)
}
if count != want {
t.Fatalf("ModelPrice count = %d, want %d", count, want)
}
}