Files
openteam/server/internal/channel/channel_test.go
T
Sakurasan e865f69a41 fix: 修掉模型调用链路上的四处失效逻辑
- 超时:NewGateway 里 gw_timeout*time.Second 把已是 Duration 的值二次换算,
  int64 溢出后无论 OT_PROXY_TIMEOUT 填多少都等于没有超时,渠道 timeout_ms
  也从没被读过。现在非流式用 context deadline 约束整段时长,流式只约束首包
  (Transport.ResponseHeaderTimeout),正文不限时长,长回答不会被砍。
- 断连:上游请求改用 NewRequestWithContext 绑定客户端 context,客户端断开即
  取消上游并记 canceled,不再空转生成继续付费;PATCH 已写 200 头后不再往
  响应里追加错误体。
- 白名单:APIKey.AllowedModels 以前只存不用,现在 Auth 注入 context 并参与
  判定(用户黑名单 > 密钥白名单 > 用户白名单 > 全局)。
- 选路:模型已有渠道绑定时只用绑定渠道,都不可用直接 503 no_channel,不再
  把请求发给不认识它的渠道;新增运行期健康(连续 3 次网络错误/超时/429/5xx
  进 30s 冷却,成功清零),与 60s 一次的 /models 探测互补。
- 密钥 API:PATCH 回显更新后的完整对象;限额填 0 统一视为不限制(原来
  token 配额存 0 会让密钥第一次请求就被 429 打死,而请求配额存 0 又等于
  不限制,两个字段语义相反);前端编辑框总是提交白名单,清空能真的取消。

测试:新增 proxy 链路集成测试(超时生效、流式不被截断、断连取消并记账、
密钥白名单 403、连续 429 后候选剔除)与 channel 用例(绑定模型不退全局、
冷却阈值与到期恢复、成功清零)。
2026-10-16 19:15:51 +08:00

208 lines
6.2 KiB
Go

package channel
import (
"context"
"os"
"testing"
"time"
"github.com/openteam/server/internal/pkg/crypto"
"github.com/openteam/server/internal/store"
"gorm.io/gorm"
)
func newTestDB(t *testing.T) *gorm.DB {
t.Helper()
dsn := "/tmp/channel_test_" + time.Now().Format("150405.000000000") + ".db"
db, err := store.Open("sqlite", dsn)
if err != nil {
t.Fatalf("open db: %v", err)
}
t.Cleanup(func() {
sqlDB, _ := db.DB()
sqlDB.Close()
os.Remove(dsn)
})
return db
}
func seedChannel(t *testing.T, db *gorm.DB, name string, weight, priority, maxConc int) uint64 {
t.Helper()
enc := crypto.NewEncryptor("test-master-key-1234567890")
key, _ := enc.Encrypt("upstream-key")
ch := store.Channel{
Name: name, Provider: store.ChannelProviderOpenAI, BaseURL: "http://localhost:9000",
APIKeyEnc: key, Weight: weight, Priority: priority, MaxConcurrency: maxConc,
HealthStatus: store.ChannelHealthHealthy, Enabled: true,
}
if err := db.Create(&ch).Error; err != nil {
t.Fatalf("create channel: %v", err)
}
return ch.ID
}
func TestCandidatesFiltersUnhealthy(t *testing.T) {
db := newTestDB(t)
seedChannel(t, db, "a", 1, 0, 4)
seedChannel(t, db, "b", 5, 1, 4)
// 禁用渠道不入候选
bad := store.Channel{Name: "bad", Provider: store.ChannelProviderOpenAI, BaseURL: "http://x", APIKeyEnc: "x", HealthStatus: store.ChannelHealthCooldown, Enabled: true}
db.Create(&bad)
s := NewService(db, crypto.NewEncryptor("test-master-key-1234567890"))
cands := s.Candidates("")
if len(cands) != 2 {
t.Fatalf("candidates = %d, want 2", len(cands))
}
if cands[0].Channel.Name != "a" {
t.Fatalf("first by priority should be a, got %s", cands[0].Channel.Name)
}
}
func TestPickWeighted(t *testing.T) {
db := newTestDB(t)
seedChannel(t, db, "only", 1, 0, 4)
s := NewService(db, crypto.NewEncryptor("test-master-key-1234567890"))
cands := s.Candidates("")
chosen := s.Pick(cands)
if chosen == nil || chosen.Name != "only" {
t.Fatalf("pick single: %+v", chosen)
}
// 空候选返回 nil
if p := s.Pick(nil); p != nil {
t.Fatal("expected nil for empty candidates")
}
}
func TestTryAcquire(t *testing.T) {
db := newTestDB(t)
id := seedChannel(t, db, "c", 1, 0, 1)
var ch store.Channel
db.First(&ch, id)
s := NewService(db, crypto.NewEncryptor("test-master-key-1234567890"))
release, ok := s.TryAcquire(&ch)
if !ok {
t.Fatal("first acquire should succeed")
}
if _, ok := s.TryAcquire(&ch); ok {
t.Fatal("second acquire should fail (capacity 1)")
}
release()
if _, ok := s.TryAcquire(&ch); !ok {
t.Fatal("acquire after release should succeed")
}
}
func TestHealthCooldownAndRecovery(t *testing.T) {
db := newTestDB(t)
id := seedChannel(t, db, "h", 1, 0, 4)
var ch store.Channel
db.First(&ch, id)
m := NewHealthMonitor(db, crypto.NewEncryptor("test-master-key-1234567890"), HealthConfig{
Interval: time.Minute, FailThreshold: 2, Timeout: time.Second,
})
// 两次失败 → cooldown
m.record(&ch, false)
if ch.HealthStatus != store.ChannelHealthHealthy {
t.Fatal("one failure should stay healthy")
}
m.record(&ch, false)
if ch.HealthStatus != store.ChannelHealthCooldown {
t.Fatalf("two failures should cooldown, got %s", ch.HealthStatus)
}
// 恢复
m.record(&ch, true)
if ch.HealthStatus != store.ChannelHealthHealthy {
t.Fatalf("success should recover, got %s", ch.HealthStatus)
}
}
func TestHealthStartStops(t *testing.T) {
db := newTestDB(t)
seedChannel(t, db, "probe", 1, 0, 4)
m := NewHealthMonitor(db, crypto.NewEncryptor("test-master-key-1234567890"), HealthConfig{
Interval: time.Hour, FailThreshold: 2, Timeout: time.Second,
})
ctx, cancel := context.WithCancel(context.Background())
m.Start(ctx)
cancel() // 不应 panic
}
// 模型被渠道绑定后只允许走绑定渠道;绑定渠道不可用时返回空候选(上层 503),
// 不能退化到"不认识这个模型"的全局渠道。
func TestBoundModelDoesNotFallBackToGlobal(t *testing.T) {
db := newTestDB(t)
bound := seedChannel(t, db, "bound", 1, 0, 4)
seedChannel(t, db, "other", 1, 1, 4)
m := store.Model{Name: "gpt-m", Enabled: true}
if err := db.Create(&m).Error; err != nil {
t.Fatal(err)
}
if err := db.Create(&store.ChannelModelBinding{ChannelID: bound, ModelID: m.ID, UpstreamModel: "gpt-m"}).Error; err != nil {
t.Fatal(err)
}
s := NewService(db, crypto.NewEncryptor("test-master-key-1234567890"))
cands := s.Candidates("gpt-m")
if len(cands) != 1 || cands[0].Channel.Name != "bound" {
t.Fatalf("绑定模型应只命中 bound,实际 %+v", cands)
}
if cands[0].UpstreamModel != "gpt-m" {
t.Fatalf("映射名丢失:%s", cands[0].UpstreamModel)
}
db.Model(&store.Channel{}).Where("id = ?", bound).Update("enabled", false)
if got := s.Candidates("gpt-m"); len(got) != 0 {
t.Fatalf("绑定渠道停用后应为空候选,实际 %d", len(got))
}
// 完全未绑定的模型仍然走全局
if got := s.Candidates("unmanaged"); len(got) != 1 || got[0].Channel.Name != "other" {
t.Fatalf("未纳管模型应退回全局渠道,实际 %+v", got)
}
}
// 运行期冷却:连续失败才踢,成功即清零,冷却到期自动恢复。
func TestRuntimeCooldownThresholdAndExpiry(t *testing.T) {
db := newTestDB(t)
id := seedChannel(t, db, "r", 1, 0, 4)
var ch store.Channel
db.First(&ch, id)
s := NewService(db, crypto.NewEncryptor("test-master-key-1234567890"))
s.cooldownFor = 60 * time.Millisecond
// 默认阈值 3:前两次失败不该踢,第三次才进冷却
s.ReportFailure(&ch)
s.ReportFailure(&ch)
if s.InCooldown(id) {
t.Fatal("未达阈值(3 次连续)不应进冷却")
}
s.ReportFailure(&ch)
if !s.InCooldown(id) {
t.Fatal("连续达阈值应进冷却")
}
if got := s.Candidates(""); len(got) != 0 {
t.Fatalf("冷却渠道不应进候选,实际 %d", len(got))
}
if got := s.Candidates("whatever"); len(got) != 0 {
t.Fatalf("全局候选也应过滤冷却渠道,实际 %d", len(got))
}
time.Sleep(80 * time.Millisecond)
if s.InCooldown(id) {
t.Fatal("冷却到期应恢复可用")
}
if got := s.Candidates(""); len(got) != 1 {
t.Fatalf("恢复后应重新进候选,实际 %d", len(got))
}
s.ReportFailure(&ch)
s.ReportSuccess(&ch)
s.ReportFailure(&ch)
if s.InCooldown(id) {
t.Fatal("非连续失败不应触发冷却")
}
}