feat: 为实例调度添加版本化 PCDN 策略（canary/draft/stable）与熔断回退

server/api/v1/instance/instance.go

@@ -5,6 +5,7 @@ import (
 	"github.com/flipped-aurora/gin-vue-admin/server/model/common/response"
 	instanceModel "github.com/flipped-aurora/gin-vue-admin/server/model/instance"
 	instanceReq "github.com/flipped-aurora/gin-vue-admin/server/model/instance/request"
+	instanceServicePkg "github.com/flipped-aurora/gin-vue-admin/server/service/instance"
 	"github.com/flipped-aurora/gin-vue-admin/server/utils"
 	"github.com/gin-gonic/gin"
 	"go.uber.org/zap"
@@ -264,7 +265,12 @@ func (instanceApi *InstanceApi) GetAvailableNodes(c *gin.Context) {
 		return
 	}
 
-	nodes, err := instanceService.GetAvailableNodes(ctx, specId)
+	meta := instanceServicePkg.SchedulingRequestMeta{
+		Region:   c.Query("region"),
+		ISP:      c.Query("isp"),
+		UserHash: c.Query("userHash"),
+	}
+	nodes, err := instanceService.GetAvailableNodes(ctx, specId, meta)
 	if err != nil {
 		global.GVA_LOG.Error("查询可用节点失败!", zap.Error(err))
 		response.FailWithMessage("查询可用节点失败:"+err.Error(), c)

server/config.yaml

@@ -99,6 +99,24 @@ mcp:
     url_prefix: ""
     addr: 8889
     separate: false
+pcdn:
+    enabled: true
+    version: v2-canary
+    stable-version: v1
+    release-state: canary
+    canary:
+        regions: ["华东", "华北"]
+        isps: ["电信", "联通"]
+        user-hash-percentage: 10
+    circuit-breaker:
+        failure-rate-threshold: 0.3
+        latency-threshold-ms: 800
+        min-samples: 10
+    score-weight:
+        gpu: 0.4
+        cpu: 0.25
+        memory: 0.25
+        disk: 0.1
 jumpbox:
     enabled: true
     port: 2026

server/config.yaml.bak

@@ -254,3 +254,22 @@ zap:
     show-line: true
     log-in-console: true
     retention-day: -1
+
+pcdn:
+    enabled: true
+    version: v2-canary
+    stable-version: v1
+    release-state: canary
+    canary:
+        regions: ["华东", "华北"]
+        isps: ["电信", "联通"]
+        user-hash-percentage: 10
+    circuit-breaker:
+        failure-rate-threshold: 0.3
+        latency-threshold-ms: 800
+        min-samples: 10
+    score-weight:
+        gpu: 0.4
+        cpu: 0.25
+        memory: 0.25
+        disk: 0.1

server/config/config.go

@@ -43,4 +43,7 @@ type Server struct {
 
 	// K8s管理器配置
 	K8sManager K8sManager `mapstructure:"k8smanager" json:"k8smanager" yaml:"k8smanager"`
+
+	// PCDN调度配置
+	PCDN PCDN `mapstructure:"pcdn" json:"pcdn" yaml:"pcdn"`
 }

server/config/pcdn.go

@@ -0,0 +1,42 @@
+package config
+
+// PCDN 调度配置
+// 支持策略发布状态、灰度规则以及熔断阈值动态调整
+// 可通过热更新配置文件实时生效
+// 发布状态：draft/canary/stable
+// 灰度规则：地域、ISP、用户哈希比例
+// 熔断规则：失败率、延迟阈值
+// 注意：阈值单位详见字段注释
+//
+// 该结构由配置映射驱动，字段命名与 YAML 保持一致
+// 为了避免歧义，这里保留较直观的命名。
+//
+//nolint:revive
+type PCDN struct {
+	Enabled        bool         `mapstructure:"enabled" json:"enabled" yaml:"enabled"`
+	Version        string       `mapstructure:"version" json:"version" yaml:"version"`
+	ReleaseState   string       `mapstructure:"release-state" json:"release-state" yaml:"release-state"`
+	Canary         Canary       `mapstructure:"canary" json:"canary" yaml:"canary"`
+	CircuitBreaker CircuitBreak `mapstructure:"circuit-breaker" json:"circuit-breaker" yaml:"circuit-breaker"`
+	ScoreWeight    ScoreWeight  `mapstructure:"score-weight" json:"score-weight" yaml:"score-weight"`
+	StableVersion  string       `mapstructure:"stable-version" json:"stable-version" yaml:"stable-version"`
+}
+
+type Canary struct {
+	Regions            []string `mapstructure:"regions" json:"regions" yaml:"regions"`
+	ISPs               []string `mapstructure:"isps" json:"isps" yaml:"isps"`
+	UserHashPercentage int      `mapstructure:"user-hash-percentage" json:"user-hash-percentage" yaml:"user-hash-percentage"`
+}
+
+type CircuitBreak struct {
+	FailureRateThreshold float64 `mapstructure:"failure-rate-threshold" json:"failure-rate-threshold" yaml:"failure-rate-threshold"`
+	LatencyThresholdMs   int64   `mapstructure:"latency-threshold-ms" json:"latency-threshold-ms" yaml:"latency-threshold-ms"`
+	MinSamples           int     `mapstructure:"min-samples" json:"min-samples" yaml:"min-samples"`
+}
+
+type ScoreWeight struct {
+	Gpu    float64 `mapstructure:"gpu" json:"gpu" yaml:"gpu"`
+	CPU    float64 `mapstructure:"cpu" json:"cpu" yaml:"cpu"`
+	Memory float64 `mapstructure:"memory" json:"memory" yaml:"memory"`
+	Disk   float64 `mapstructure:"disk" json:"disk" yaml:"disk"`
+}

server/service/instance/instance.go

@@ -5,6 +5,7 @@ import (
 	"errors"
 	"fmt"
 	"strconv"
+	"time"
 
 	"github.com/flipped-aurora/gin-vue-admin/server/global"
 	"github.com/flipped-aurora/gin-vue-admin/server/model/computenode"
@@ -417,7 +418,9 @@ type AvailableNode struct {
 }
 
 // GetAvailableNodes 根据产品规格获取可用的算力节点
-func (instanceService *InstanceService) GetAvailableNodes(ctx context.Context, specIdStr string) (nodes []AvailableNode, err error) {
+func (instanceService *InstanceService) GetAvailableNodes(ctx context.Context, specIdStr string, meta SchedulingRequestMeta) (nodes []AvailableNode, err error) {
+	startAt := time.Now()
+	strategyVersion, strategyState, hitRule, forceStable := chooseStrategy(meta)
 	specId, err := strconv.ParseUint(specIdStr, 10, 64)
 	if err != nil {
 		return nil, err
@@ -985,6 +988,17 @@ func (instanceService *InstanceService) GetAvailableNodes(ctx context.Context, s
 		nodes = append(nodes, availableNode)
 	}
 
+	breakerOn := pcdnSchedulerRuntime.isCircuitOpened()
+	nodes, scoreDetails := scoreAndSortNodes(nodes, strategyVersion, strategyState, hitRule, breakerOn)
+	if forceStable && len(nodes) == 0 {
+		fallback := pcdnSchedulerRuntime.stableFallbackNodes()
+		if len(fallback) > 0 {
+			nodes = fallback
+			hitRule += "_snapshot"
+		}
+	}
+	pcdnSchedulerRuntime.recordAndEvaluate(strategyVersion, strategyState, nodes, nil, time.Since(startAt))
+	writeSchedulingTraceLog(specIdStr, strategyVersion, strategyState, hitRule, meta, scoreDetails, len(nodes), nil)
 	return nodes, nil
 }