feat: 运行期渠道冷却可见并可手动恢复
运行期冷却只存在网关进程内存里,DB 的 health_status 看不见它:渠道被踢到 30s 冷却后,管理后台仍显示 healthy,模型也照常出现在 /v1/models,但请求 返回 503 no_channel,管理员只能去翻日志。 - channel.Service 增 RuntimeState(连续失败次数 + 冷却剩余)与 ClearRuntime; Gateway 包装出 ChannelRuntime / RecoverChannel 供管理端调用。 - GET /api/v1/admin/channels 每项返回 runtime_fails 与 cooldown_seconds。 - 新增 POST /api/v1/admin/channels/:id/recover:清运行期冷却与失败计数,并把 探测写的 health_status 置回 healthy;不改 enabled,上游若真挂着,下一次探测 或再连续 3 次请求失败会重新把它踢掉。 - 渠道页健康列下方显示「冷却中 Ns」/「连续失败 n/3」,仅在冷却或探测不健康时 出现「立即恢复」按钮(卡片与桌面表格两套布局都覆盖)。 仍未处理:冷却到期是裸恢复、没有 half-open 试探;计数是进程内的,多实例各自 为战,要共享得等 Redis 那一步。
This commit is contained in:
1 parent
e865f69a41
commit
d4c5263f1e
8 files changed
+166
-2
No files matched your search
@@ -30,18 +30,30 @@ func (h *Handler) AdminChannels(c *gin.Context) {
|
||||
} else if err == nil {
|
||||
masked = "****"
|
||||
}
|
||||
fails, cooldownSec := h.channelRuntime(ch.ID)
|
||||
out = append(out, gin.H{
|
||||
"id": ch.ID, "name": ch.Name, "provider": ch.Provider, "formats": ch.FormatsEffective(),
|
||||
"base_url": ch.BaseURL, "base_urls": ch.BaseURLs,
|
||||
"api_key_masked": masked, "weight": ch.Weight, "priority": ch.Priority,
|
||||
"timeout_ms": ch.TimeoutMS, "max_concurrency": ch.MaxConcurrency,
|
||||
"health_status": ch.HealthStatus, "enabled": ch.Enabled,
|
||||
"runtime_fails": fails, "cooldown_seconds": cooldownSec,
|
||||
"created_at": ch.CreatedAt,
|
||||
})
|
||||
}
|
||||
resp.OK(c, gin.H{"items": out})
|
||||
}
|
||||
|
||||
// channelRuntime 渠道运行期健康:连续失败次数与冷却剩余秒数(0 表示无冷却)。
|
||||
// 运行期冷却只存在于进程内存,DB 的 health_status 看不到它,故列表单独返回。
|
||||
func (h *Handler) channelRuntime(id uint64) (fails int, cooldownSeconds int64) {
|
||||
if h.gw == nil {
|
||||
return 0, 0
|
||||
}
|
||||
fails, remaining := h.gw.ChannelRuntime(id)
|
||||
return fails, int64((remaining + time.Second - time.Nanosecond) / time.Second) // 向上取整
|
||||
}
|
||||
|
||||
type channelBody struct {
|
||||
Name string `json:"name" binding:"required,min=1,max=64"`
|
||||
Provider string `json:"provider"` // 可选:为空时按 formats 推断(兼容旧数据)
|
||||
@@ -369,6 +381,32 @@ func (h *Handler) AdminTestChannel(c *gin.Context) {
|
||||
resp.OK(c, gin.H{"ok": true, "latency_ms": latency, "message": msg})
|
||||
}
|
||||
|
||||
// AdminRecoverChannel POST /api/v1/admin/channels/:id/recover — 立即恢复渠道:
|
||||
// 清掉运行期冷却与连续失败计数,并把探测写的 health_status 置回 healthy。
|
||||
// 不改 enabled;上游若真挂着,下一次探测或连续 3 次请求失败会重新把它踢掉。
|
||||
func (h *Handler) AdminRecoverChannel(c *gin.Context) {
|
||||
id, err := strconv.ParseUint(c.Param("id"), 10, 64)
|
||||
if err != nil {
|
||||
resp.Fail(c, http.StatusBadRequest, "invalid channel id")
|
||||
return
|
||||
}
|
||||
var ch store.Channel
|
||||
if err := h.a.DB.First(&ch, id).Error; err != nil {
|
||||
resp.Fail(c, http.StatusNotFound, "channel not found")
|
||||
return
|
||||
}
|
||||
if h.gw != nil {
|
||||
h.gw.RecoverChannel(ch.ID)
|
||||
}
|
||||
if err := h.a.DB.Model(&store.Channel{}).Where("id = ?", ch.ID).
|
||||
Update("health_status", store.ChannelHealthHealthy).Error; err != nil {
|
||||
resp.Fail(c, http.StatusInternalServerError, "failed to update channel")
|
||||
return
|
||||
}
|
||||
resp.OK(c, gin.H{"id": ch.ID, "name": ch.Name,
|
||||
"health_status": store.ChannelHealthHealthy, "runtime_fails": 0, "cooldown_seconds": 0})
|
||||
}
|
||||
|
||||
// maskAPIKey 掩码渠道密钥:保留前 7 位与后 4 位,中间固定 ****** 遮蔽。
|
||||
// 示例:xxxxxxx******Mq4Y;密钥较短时退化为仅保留后 4 位。
|
||||
func maskAPIKey(key string) string {
|
||||
|
||||
@@ -111,6 +111,7 @@ func NewRouter(a *app.App, gw *proxy.Gateway) *gin.Engine {
|
||||
admin.PUT("/channels/:id", h.AdminUpdateChannel)
|
||||
admin.DELETE("/channels/:id", h.AdminDeleteChannel)
|
||||
admin.POST("/channels/:id/test", h.AdminTestChannel)
|
||||
admin.POST("/channels/:id/recover", h.AdminRecoverChannel)
|
||||
admin.GET("/channels/:id/models/remote", h.AdminChannelRemoteModels)
|
||||
admin.GET("/channels/:id/models", h.AdminChannelModels)
|
||||
admin.POST("/channels/:id/models", h.AdminChannelAddModel)
|
||||
|
||||
@@ -215,6 +215,32 @@ func (s *Service) ReportSuccess(ch *store.Channel) {
|
||||
s.mu.Unlock()
|
||||
}
|
||||
|
||||
// RuntimeState 渠道运行期健康快照:连续失败次数与冷却剩余时间(管理后台展示用)。
|
||||
// 只读,不做过期清理——清理仍由 InCooldown 在选路时惰性完成。
|
||||
func (s *Service) RuntimeState(id uint64) (fails int, remaining time.Duration) {
|
||||
s.mu.Lock()
|
||||
defer s.mu.Unlock()
|
||||
st, ok := s.runt[id]
|
||||
if !ok {
|
||||
return 0, 0
|
||||
}
|
||||
if !st.until.IsZero() {
|
||||
if d := time.Until(st.until); d > 0 {
|
||||
return 0, d // 冷却中:计数在写入时已归零
|
||||
}
|
||||
return 0, 0 // 已到期,等下次选路清理
|
||||
}
|
||||
return st.fails, 0
|
||||
}
|
||||
|
||||
// ClearRuntime 清除运行期状态(管理后台"立即恢复")。
|
||||
// 不改 DB 的 health_status——那条由探测/管理端点负责,两者语义不同不要混在一起。
|
||||
func (s *Service) ClearRuntime(id uint64) {
|
||||
s.mu.Lock()
|
||||
delete(s.runt, id)
|
||||
s.mu.Unlock()
|
||||
}
|
||||
|
||||
// InCooldown 渠道是否处于运行期冷却;过期则清掉状态、恢复可用。
|
||||
// until 为零值表示只在累计失败、尚未触发冷却,此时不能清状态(否则计数永远攒不满阈值)。
|
||||
func (s *Service) InCooldown(id uint64) bool {
|
||||
|
||||
@@ -205,3 +205,40 @@ func TestRuntimeCooldownThresholdAndExpiry(t *testing.T) {
|
||||
t.Fatal("非连续失败不应触发冷却")
|
||||
}
|
||||
}
|
||||
|
||||
// 管理后台可见性:RuntimeState 报累计失败与冷却剩余,ClearRuntime 立即清零。
|
||||
func TestRuntimeStateAndClear(t *testing.T) {
|
||||
db := newTestDB(t)
|
||||
id := seedChannel(t, db, "s", 1, 0, 4)
|
||||
var ch store.Channel
|
||||
db.First(&ch, id)
|
||||
|
||||
s := NewService(db, crypto.NewEncryptor("test-master-key-1234567890"))
|
||||
if fails, rem := s.RuntimeState(id); fails != 0 || rem != 0 {
|
||||
t.Fatalf("初始状态 = %d/%s,期望 0/0", fails, rem)
|
||||
}
|
||||
s.ReportFailure(&ch)
|
||||
s.ReportFailure(&ch)
|
||||
if fails, rem := s.RuntimeState(id); fails != 2 || rem != 0 {
|
||||
t.Fatalf("两次失败后 = %d/%s,期望 2/0", fails, rem)
|
||||
}
|
||||
s.ReportFailure(&ch)
|
||||
fails, rem := s.RuntimeState(id)
|
||||
if fails != 0 || rem <= 0 || rem > s.cooldownFor {
|
||||
t.Fatalf("冷却中 = %d/%s,期望 0/(0,%s]", fails, rem, s.cooldownFor)
|
||||
}
|
||||
if got := len(s.Candidates("")); got != 0 {
|
||||
t.Fatalf("冷却渠道不应进候选,实际 %d", got)
|
||||
}
|
||||
|
||||
s.ClearRuntime(id)
|
||||
if fails, rem := s.RuntimeState(id); fails != 0 || rem != 0 {
|
||||
t.Fatalf("清除后 = %d/%s,期望 0/0", fails, rem)
|
||||
}
|
||||
if s.InCooldown(id) {
|
||||
t.Fatal("清除后不应仍在冷却")
|
||||
}
|
||||
if got := len(s.Candidates("")); got != 1 {
|
||||
t.Fatalf("恢复后应重新进候选,实际 %d", got)
|
||||
}
|
||||
}
|
||||
@@ -64,6 +64,16 @@ func (g *Gateway) ResetModelPolicy() {
|
||||
g.policyMu.Unlock()
|
||||
}
|
||||
|
||||
// ChannelRuntime 渠道运行期健康快照:连续失败次数 + 冷却剩余时间(管理后台展示)。
|
||||
func (g *Gateway) ChannelRuntime(id uint64) (fails int, remaining time.Duration) {
|
||||
return g.ch.RuntimeState(id)
|
||||
}
|
||||
|
||||
// RecoverChannel 立即清除渠道的运行期冷却与失败计数(管理后台手动恢复)。
|
||||
func (g *Gateway) RecoverChannel(id uint64) {
|
||||
g.ch.ClearRuntime(id)
|
||||
}
|
||||
|
||||
// globalModelRestrictions 读取全局模型允许/禁止列表(缓存 modelPolicyTTL)。
|
||||
func (g *Gateway) globalModelRestrictions() (allow, deny []string) {
|
||||
g.policyMu.Lock()
|
||||
|
||||
Reference in new issue
Block a user