feat(routing): 引入多执行池智能调度

将 Worker 发现、路由画像、容量与执行传输抽象为平台无关接口，新增 Kubernetes 和静态容量适配器，并以 shadow 模式接入生产配置。实现网络与容量评分、路由防抖、池队列、同步 Worker 租约、一次性执行令牌，以及提交状态不明时禁止重复分配的安全语义。新增 0105 兼容迁移、管理接口、指标、OpenAPI 和回归测试。已执行全量 Go 测试、go vet、OpenAPI、迁移安全、Compose 与 Kustomize 验证。
2026-08-05 22:25:37 +08:00
parent 03c0873649
commit 7786692d32
58 changed files with 4510 additions and 348 deletions
@@ -0,0 +1,143 @@
+package httpapi
+
+import (
+	"net/http"
+	"time"
+
+	"github.com/easyai/easyai-ai-gateway/apps/api/internal/executionpool"
+)
+
+type AdminExecutionPoolWorker struct {
+	WorkerID        string         `json:"workerId"`
+	InstanceID      string         `json:"instanceId"`
+	Revision        string         `json:"revision,omitempty"`
+	ProtocolVersion string         `json:"protocolVersion"`
+	Capabilities    map[string]any `json:"capabilities"`
+	Allocated       int            `json:"allocated"`
+	SafeCapacity    int            `json:"safeCapacity"`
+	ActiveTasks     int            `json:"activeTasks"`
+	PressureState   string         `json:"pressureState"`
+	HeartbeatAt     time.Time      `json:"heartbeatAt"`
+}
+
+type AdminExecutionPool struct {
+	PoolID       string                      `json:"poolId"`
+	Labels       map[string]string           `json:"labels"`
+	Capabilities map[string]any              `json:"capabilities"`
+	State        string                      `json:"state"`
+	Capacity     *AdminExecutionPoolCapacity `json:"capacity,omitempty"`
+	Workers      []AdminExecutionPoolWorker  `json:"workers"`
+}
+
+type AdminExecutionPoolCapacity struct {
+	WorkerCount             int       `json:"workerCount"`
+	Allocated               int       `json:"allocated"`
+	SafeCapacity            int       `json:"safeCapacity"`
+	ActiveTasks             int       `json:"activeTasks"`
+	HeavyCapacity           int       `json:"heavyCapacity"`
+	HeavyTasks              int       `json:"heavyTasks"`
+	QueueDepth              int       `json:"queueDepth"`
+	EstimatedQueueWaitMilli int64     `json:"estimatedQueueWaitMs"`
+	ResourceHeadroom        float64   `json:"resourceHeadroom"`
+	Stability               float64   `json:"stability"`
+	Critical                bool      `json:"critical"`
+	SampledAt               time.Time `json:"sampledAt"`
+}
+
+type AdminRouteHealth struct {
+	PoolID               string    `json:"poolId"`
+	RouteProfileKey      string    `json:"routeProfileKey"`
+	State                string    `json:"state"`
+	SuccessRate          float64   `json:"successRate"`
+	ConnectTLSP95Milli   int64     `json:"connectTlsP95Ms"`
+	FirstByteP95Milli    int64     `json:"firstByteP95Ms"`
+	UploadBytesPerSecond float64   `json:"uploadBytesPerSecond"`
+	JitterP95Milli       int64     `json:"jitterP95Ms"`
+	ConsecutiveFailures  int       `json:"consecutiveFailures"`
+	ConsecutiveSuccesses int       `json:"consecutiveSuccesses"`
+	SampleCount          int       `json:"sampleCount"`
+	SampledAt            time.Time `json:"sampledAt"`
+	ExpiresAt            time.Time `json:"expiresAt"`
+}
+
+type AdminExecutionPoolResponse struct {
+	Pools       []AdminExecutionPool `json:"pools"`
+	RouteHealth []AdminRouteHealth   `json:"routeHealth"`
+	CapturedAt  time.Time            `json:"capturedAt"`
+}
+
+// listExecutionPools godoc
+// @Summary 查询逻辑执行池、Worker 容量和上游路由健康状态
+// @Tags Admin Runtime
+// @Produce json
+// @Success 200 {object} AdminExecutionPoolResponse
+// @Failure 500 {object} ErrorEnvelope
+// @Security BearerAuth
+// @Router /api/admin/runtime/execution-pools [get]
+func (s *Server) listExecutionPools(w http.ResponseWriter, r *http.Request) {
+	now := time.Now()
+	pools, err := s.coordinationStore.ListExecutionPools(r.Context())
+	if err != nil {
+		writeError(w, http.StatusInternalServerError, err.Error())
+		return
+	}
+	workers, err := s.coordinationStore.ListWorkers(r.Context(), now)
+	if err != nil {
+		writeError(w, http.StatusInternalServerError, err.Error())
+		return
+	}
+	capacities, err := s.coordinationStore.ListCapacity(r.Context(), now)
+	if err != nil {
+		writeError(w, http.StatusInternalServerError, err.Error())
+		return
+	}
+	health, err := s.coordinationStore.ListAllRouteHealth(r.Context())
+	if err != nil {
+		writeError(w, http.StatusInternalServerError, err.Error())
+		return
+	}
+	capacityByPool := make(map[string]executionpool.CapacitySnapshot, len(capacities))
+	for _, capacity := range capacities {
+		capacityByPool[capacity.PoolID] = capacity
+	}
+	workersByPool := make(map[string][]AdminExecutionPoolWorker)
+	for _, worker := range workers {
+		workersByPool[worker.PoolID] = append(workersByPool[worker.PoolID], AdminExecutionPoolWorker{
+			WorkerID: worker.WorkerID, InstanceID: worker.InstanceID, Revision: worker.Revision,
+			ProtocolVersion: worker.ProtocolVersion, Capabilities: worker.Capabilities,
+			Allocated: worker.Allocated, SafeCapacity: worker.SafeCapacity,
+			ActiveTasks: worker.ActiveTasks, PressureState: worker.PressureState,
+			HeartbeatAt: worker.HeartbeatAt,
+		})
+	}
+	response := AdminExecutionPoolResponse{
+		Pools: make([]AdminExecutionPool, 0, len(pools)), RouteHealth: make([]AdminRouteHealth, 0, len(health)), CapturedAt: now,
+	}
+	for _, item := range health {
+		response.RouteHealth = append(response.RouteHealth, AdminRouteHealth{
+			PoolID: item.PoolID, RouteProfileKey: item.RouteProfileKey, State: string(item.State),
+			SuccessRate: item.SuccessRate, ConnectTLSP95Milli: item.ConnectTLSP95.Milliseconds(),
+			FirstByteP95Milli: item.FirstByteP95.Milliseconds(), UploadBytesPerSecond: item.UploadBytesPerSecond,
+			JitterP95Milli: item.JitterP95.Milliseconds(), ConsecutiveFailures: item.ConsecutiveFailures,
+			ConsecutiveSuccesses: item.ConsecutiveSuccesses, SampleCount: item.SampleCount,
+			SampledAt: item.SampledAt, ExpiresAt: item.ExpiresAt,
+		})
+	}
+	for _, pool := range pools {
+		item := AdminExecutionPool{
+			PoolID: pool.ID, Labels: pool.Labels, Capabilities: pool.Capabilities,
+			State: string(pool.State), Workers: workersByPool[pool.ID],
+		}
+		if capacity, ok := capacityByPool[pool.ID]; ok {
+			item.Capacity = &AdminExecutionPoolCapacity{
+				WorkerCount: capacity.WorkerCount, Allocated: capacity.Allocated, SafeCapacity: capacity.SafeCapacity,
+				ActiveTasks: capacity.ActiveTasks, HeavyCapacity: capacity.HeavyCapacity, HeavyTasks: capacity.HeavyTasks,
+				QueueDepth: capacity.QueueDepth, EstimatedQueueWaitMilli: capacity.EstimatedQueueWait.Milliseconds(),
+				ResourceHeadroom: capacity.ResourceHeadroom, Stability: capacity.Stability,
+				Critical: capacity.Critical, SampledAt: capacity.SampledAt,
+			}
+		}
+		response.Pools = append(response.Pools, item)
+	}
+	writeJSON(w, http.StatusOK, response)
+}