feat: 十轮网关优化 - 安全加固/可观测性/性能/可靠性
- SSE Keepalive Ping (15s心跳防止代理断连) - Timing HTTP 头 (X-Timing-Queue/Inference/Total-Ms) - Adapter Request-ID 传播到后端 - Session 清理日志回调 - Server 安全加固 (ReadHeaderTimeout/MaxHeaderBytes 防 slowloris) - Usage Tracker 数据保留清理 (retentionDays + 定期清理) - Config Reload 后 Adapter Registry 更新 (RegisterIfAbsent + RWMutex) - Rate Limiter 空闲 Bucket 清理 (30分钟过期) - Shutdown Drain 超时可配置 (ShutdownDrainSeconds) - Config 模型字段校验增强 (provider/endpoint/actual_model) - Auth 过期 Key 自动清理 (5分钟扫描) - Admin API Rate Limiting - Adapter Health Check 独立超时 (每个 adapter 3s) - TCP 连接阶段超时 (DialContext 5s + KeepAlive 30s) - 幂等键缓存、审计日志、Gzip 中间件、CORS Expose Headers - Backpressure 响应头、熔断器 Prometheus 指标 - 连接池优化、Trace-ID 全链路传播
This commit is contained in:
selfrelease
@@ -7,27 +7,27 @@ import (
|
||||
|
||||
// CircuitBreaker implements a sliding-window circuit breaker for adapter health.
|
||||
type CircuitBreaker struct {
|
||||
mu sync.Mutex
|
||||
mu sync.Mutex
|
||||
errorRateThreshold float64
|
||||
minRequests int
|
||||
windowSeconds int
|
||||
openDuration time.Duration
|
||||
halfOpenMax int
|
||||
minRequests int
|
||||
windowSeconds int
|
||||
openDuration time.Duration
|
||||
halfOpenMax int
|
||||
|
||||
// sliding window state
|
||||
requests []time.Time
|
||||
requests []time.Time
|
||||
errors []time.Time
|
||||
|
||||
// breaker state
|
||||
state breakerState
|
||||
openedAt time.Time
|
||||
state breakerState
|
||||
openedAt time.Time
|
||||
halfOpenCount int
|
||||
}
|
||||
|
||||
type breakerState int
|
||||
|
||||
const (
|
||||
breakerClosed breakerState = iota
|
||||
breakerClosed breakerState = iota
|
||||
breakerOpen
|
||||
breakerHalfOpen
|
||||
)
|
||||
@@ -126,6 +126,53 @@ func (cb *CircuitBreaker) State() string {
|
||||
return "unknown"
|
||||
}
|
||||
|
||||
// BreakerStats 熔断器统计信息。
|
||||
type BreakerStats struct {
|
||||
State string `json:"state"`
|
||||
TotalRequests int `json:"total_requests"`
|
||||
TotalErrors int `json:"total_errors"`
|
||||
ErrorRate float64 `json:"error_rate"`
|
||||
WindowSeconds int `json:"window_seconds"`
|
||||
OpenDuration string `json:"open_duration"`
|
||||
Threshold float64 `json:"error_rate_threshold"`
|
||||
MinRequests int `json:"min_requests"`
|
||||
}
|
||||
|
||||
// Stats 返回熔断器的详细统计信息。
|
||||
func (cb *CircuitBreaker) Stats() BreakerStats {
|
||||
cb.mu.Lock()
|
||||
defer cb.mu.Unlock()
|
||||
|
||||
now := time.Now()
|
||||
cb.prune(now)
|
||||
|
||||
total := len(cb.requests)
|
||||
errors := len(cb.errors)
|
||||
var errorRate float64
|
||||
if total > 0 {
|
||||
errorRate = float64(errors) / float64(total)
|
||||
}
|
||||
|
||||
stateName := "closed"
|
||||
switch cb.state {
|
||||
case breakerOpen:
|
||||
stateName = "open"
|
||||
case breakerHalfOpen:
|
||||
stateName = "half_open"
|
||||
}
|
||||
|
||||
return BreakerStats{
|
||||
State: stateName,
|
||||
TotalRequests: total,
|
||||
TotalErrors: errors,
|
||||
ErrorRate: errorRate,
|
||||
WindowSeconds: cb.windowSeconds,
|
||||
OpenDuration: cb.openDuration.String(),
|
||||
Threshold: cb.errorRateThreshold,
|
||||
MinRequests: cb.minRequests,
|
||||
}
|
||||
}
|
||||
|
||||
// prune removes entries outside the sliding window.
|
||||
func (cb *CircuitBreaker) prune(now time.Time) {
|
||||
cutoff := now.Add(-time.Duration(cb.windowSeconds) * time.Second)
|
||||
|
||||
@@ -14,29 +14,33 @@ import (
|
||||
|
||||
// Scheduler manages task queuing and execution with priority-based scheduling.
|
||||
type Scheduler struct {
|
||||
mu sync.Mutex
|
||||
queue *priorityQueue
|
||||
running map[string]*task.Task
|
||||
maxRunning int
|
||||
maxQueued int
|
||||
notifyCh chan struct{}
|
||||
logger *observability.Logger
|
||||
ctx context.Context
|
||||
cancel context.CancelFunc
|
||||
mu sync.Mutex
|
||||
queue *priorityQueue
|
||||
running map[string]*task.Task
|
||||
maxRunning int
|
||||
maxQueued int
|
||||
notifyCh chan struct{}
|
||||
logger *observability.Logger
|
||||
ctx context.Context
|
||||
cancel context.CancelFunc
|
||||
agingSeconds int // 优先级老化阈值（秒），0 表示禁用
|
||||
agingMinPrio int // 老化生效的最低优先级（仅 P2-P4 老化，P0/P1 不老化）
|
||||
}
|
||||
|
||||
// NewScheduler creates a new scheduler.
|
||||
func NewScheduler(cfg *config.SchedulerConfig, logger *observability.Logger) *Scheduler {
|
||||
ctx, cancel := context.WithCancel(context.Background())
|
||||
s := &Scheduler{
|
||||
queue: &priorityQueue{},
|
||||
running: make(map[string]*task.Task),
|
||||
maxRunning: cfg.MaxRunningTasks,
|
||||
maxQueued: cfg.MaxQueuedTasks,
|
||||
notifyCh: make(chan struct{}, 1),
|
||||
logger: logger,
|
||||
ctx: ctx,
|
||||
cancel: cancel,
|
||||
queue: &priorityQueue{},
|
||||
running: make(map[string]*task.Task),
|
||||
maxRunning: cfg.MaxRunningTasks,
|
||||
maxQueued: cfg.MaxQueuedTasks,
|
||||
notifyCh: make(chan struct{}, 1),
|
||||
logger: logger,
|
||||
ctx: ctx,
|
||||
cancel: cancel,
|
||||
agingSeconds: cfg.PriorityAgingSeconds,
|
||||
agingMinPrio: 2, // P2 及以上优先级才会老化
|
||||
}
|
||||
heap.Init(s.queue)
|
||||
return s
|
||||
@@ -71,6 +75,10 @@ func (s *Scheduler) Submit(t *task.Task) error {
|
||||
func (s *Scheduler) GetNext(ctx context.Context) (*task.Task, error) {
|
||||
for {
|
||||
s.mu.Lock()
|
||||
// 优先级老化：提升等待过久的低优先级任务
|
||||
if s.agingSeconds > 0 {
|
||||
s.applyPriorityAging()
|
||||
}
|
||||
if s.queue.Len() > 0 && len(s.running) < s.maxRunning {
|
||||
t := heap.Pop(s.queue).(*task.Task)
|
||||
s.running[t.ID] = t
|
||||
@@ -88,6 +96,35 @@ func (s *Scheduler) GetNext(ctx context.Context) (*task.Task, error) {
|
||||
}
|
||||
}
|
||||
|
||||
// applyPriorityAging 对队列中等待超过 agingSeconds 的低优先级任务提升一级优先级。
|
||||
// 必须在持有 s.mu 锁的情况下调用。
|
||||
func (s *Scheduler) applyPriorityAging() {
|
||||
if s.agingSeconds <= 0 || s.queue.Len() == 0 {
|
||||
return
|
||||
}
|
||||
now := time.Now()
|
||||
aged := 0
|
||||
for i := 0; i < s.queue.Len(); i++ {
|
||||
t := (*s.queue)[i]
|
||||
if int(t.Priority) < s.agingMinPrio {
|
||||
continue // P0/P1 不老化
|
||||
}
|
||||
waitSec := int(now.Sub(t.CreatedAt).Seconds())
|
||||
if waitSec >= s.agingSeconds {
|
||||
t.Priority-- // 提升一级（数值越小优先级越高）
|
||||
if t.Priority < 0 {
|
||||
t.Priority = 0
|
||||
}
|
||||
aged++
|
||||
}
|
||||
}
|
||||
if aged > 0 {
|
||||
heap.Init(s.queue) // 重新堆化
|
||||
s.logger.Info("priority aging applied",
|
||||
observability.F().Event("priority_aging").Set("aged_count", aged))
|
||||
}
|
||||
}
|
||||
|
||||
// Complete marks a task as completed and removes it from running.
|
||||
func (s *Scheduler) Complete(taskID string) {
|
||||
s.mu.Lock()
|
||||
@@ -130,6 +167,35 @@ func (s *Scheduler) GetTask(taskID string) (*task.Task, bool) {
|
||||
return nil, false
|
||||
}
|
||||
|
||||
// ListTasks 返回所有运行中和排队中的任务（支持分页）。
|
||||
// status 过滤：running、queued、空字符串表示全部。
|
||||
func (s *Scheduler) ListTasks(status string, limit, offset int) ([]*task.Task, int) {
|
||||
s.mu.Lock()
|
||||
defer s.mu.Unlock()
|
||||
|
||||
var all []*task.Task
|
||||
if status == "" || status == "running" {
|
||||
for _, t := range s.running {
|
||||
all = append(all, t)
|
||||
}
|
||||
}
|
||||
if status == "" || status == "queued" {
|
||||
for i := 0; i < s.queue.Len(); i++ {
|
||||
all = append(all, (*s.queue)[i])
|
||||
}
|
||||
}
|
||||
|
||||
total := len(all)
|
||||
if offset >= total {
|
||||
return []*task.Task{}, total
|
||||
}
|
||||
end := offset + limit
|
||||
if end > total {
|
||||
end = total
|
||||
}
|
||||
return all[offset:end], total
|
||||
}
|
||||
|
||||
// Stop shuts down the scheduler.
|
||||
func (s *Scheduler) Stop() {
|
||||
s.cancel()
|
||||
|
||||
Reference in New Issue
Block a user