Files
gpt-load/internal/control/service_integration_test.go
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

148 lines
4.4 KiB
Go

package control
import (
"bytes"
"encoding/json"
"io"
"net/http"
"net/http/httptest"
"strings"
"testing"
"time"
"github.com/gin-gonic/gin"
"gpt-load/internal/channel"
"gpt-load/internal/dialect"
"gpt-load/internal/gateway"
"gpt-load/internal/health"
"gpt-load/internal/state"
)
func TestControlWriteLockDoesNotBlockDataPlane(t *testing.T) {
t.Parallel()
requestReached := make(chan bool, 2)
upstream := httptest.NewServer(http.HandlerFunc(func(writer http.ResponseWriter, request *http.Request) {
body, _ := io.ReadAll(request.Body)
streaming := bytes.Contains(body, []byte(`"stream":true`))
requestReached <- streaming
if streaming {
writer.Header().Set("Content-Type", "text/event-stream")
_, _ = writer.Write([]byte("data: {\"id\":\"chunk\"}\n\n"))
writer.(http.Flusher).Flush()
return
}
writer.Header().Set("Content-Type", "application/json")
_, _ = writer.Write([]byte(`{"id":"response"}`))
}))
t.Cleanup(upstream.Close)
fixture := newServiceFixture(t)
const (
groupID = uint(10)
upstreamKeyID = uint(20)
accessKey = "gl-data-plane-client"
upstreamKey = "sk-data-plane-upstream"
)
if _, err := fixture.manager.Publish(state.CompileInput{
ChannelRegistry: fixture.channelRegistry,
Groups: []state.GroupConfig{{ConnectionType: "api_key", ID: groupID, Name: "data-plane", ChannelID: channel.OpenAICompatible,
Params: json.RawMessage(`{"base_url":"` + upstream.URL + `/v1"}`),
Models: []state.ModelConfig{{ID: "gpt-4o"}}, Enabled: true,
}},
AccessKeys: []state.AccessKeyConfig{{
ID: 1, Name: "client", KeyHash: fixture.encryption.Hash(accessKey),
Status: state.AccessKeyStatusActive,
}},
}); err != nil {
t.Fatalf("Publish(data plane) error = %v", err)
}
credentialData := `{"api_key":"` + upstreamKey + `"}`
ciphertext, err := fixture.encryption.Encrypt(credentialData)
if err != nil {
t.Fatalf("Encrypt(upstream key) error = %v", err)
}
if err := fixture.registry.ReplaceCredentials([]state.CredentialEntry{{
ID: upstreamKeyID, GroupID: groupID, Version: 1, IdentityGeneration: 1,
Fingerprint: fixture.encryption.Hash(credentialData), Status: state.CredentialStatusActive,
EncryptedValue: ciphertext,
}}); err != nil {
t.Fatalf("ReplaceCredentials() error = %v", err)
}
openAI := dialect.NewOpenAI()
handler := gateway.NewHandler(
fixture.manager,
fixture.registry,
fixture.encryption,
gateway.NewExecutionForwarder(controlHTTPExecutor{}),
dialect.NewSet(openAI),
health.NewStatsStore(),
health.NewMutationCoordinator(),
nil,
nil,
nil,
)
gin.SetMode(gin.ReleaseMode)
engine := gin.New()
registerGatewayRoutes(t, engine, handler)
fixture.service.writeMu.Lock()
locked := true
defer func() {
if locked {
fixture.service.writeMu.Unlock()
}
}()
type response struct {
streaming bool
recorder *httptest.ResponseRecorder
}
responses := make(chan response, 2)
for _, streaming := range []bool{false, true} {
streaming := streaming
go func() {
body := `{"model":"gpt-4o"}`
if streaming {
body = `{"model":"gpt-4o","stream":true}`
}
request := httptest.NewRequest(http.MethodPost, "/v1/chat/completions", strings.NewReader(body))
request.Header.Set("Authorization", "Bearer "+accessKey)
recorder := httptest.NewRecorder()
engine.ServeHTTP(recorder, request)
responses <- response{streaming: streaming, recorder: recorder}
}()
}
for range 2 {
select {
case <-requestReached:
case <-time.After(2 * time.Second):
t.Fatal("data-plane request did not reach upstream while control writeMu was held")
}
}
for range 2 {
select {
case got := <-responses:
if got.recorder.Code != http.StatusOK {
t.Fatalf("streaming=%t response = %d %s", got.streaming, got.recorder.Code, got.recorder.Body.String())
}
if got.streaming && !bytes.Contains(got.recorder.Body.Bytes(), []byte("data:")) {
t.Fatalf("streaming body = %q, want SSE data", got.recorder.Body.Bytes())
}
case <-time.After(2 * time.Second):
t.Fatal("data-plane request did not complete while control writeMu was held")
}
}
if got, ok := fixture.registry.ActiveEncryptedCredentialDataIfMatch(state.CredentialRef{
ID: upstreamKeyID, GroupID: groupID, Version: 1, IdentityGeneration: 1,
Fingerprint: fixture.encryption.Hash(credentialData), EncryptedValue: ciphertext,
}); !ok || got != ciphertext {
t.Fatalf("Registry value = %q, %t, want active ciphertext", got, ok)
}
fixture.service.writeMu.Unlock()
locked = false
}