Kubernetes 集群性能优化实战：从资源分配到调度策略

resources:
  requests:
    memory: "8Gi"
    cpu: "2"
  limits:
    memory: "16Gi" 
    cpu: "4"

# 安装 metrics-server
kubectl apply -f https://github.com/kubernetes-sigs/metrics-server/releases/latest/download/components.yaml

# 配置 VPA
cat <<EOF | kubectl apply -f -
apiVersion: autoscaling.k8s.io/v1
kind: VerticalPodAutoscaler
metadata:
  name: risk-engine-vpa
spec:
  targetRef:
    apiVersion: "apps/v1"
    kind: Deployment
    name: risk-engine
  updatePolicy:
    updateMode: "Auto"
EOF

kubectl get vpa risk-engine-vpa -o yaml

containerRecommendations:
- containerName: risk-engine
  lowerBound:
    cpu: 500m
    memory: 2Gi
  target:
    cpu: 1200m 
    memory: 6Gi
  upperBound:
    cpu: 2
    memory: 10Gi
  uncappedTarget:
    cpu: 1200m
    memory: 9Gi

apiVersion: autoscaling.k8s.io/v1
kind: VerticalPodAutoscaler
spec:
  resourcePolicy:
    containerPolicies:
    - containerName: "*"
      minAllowed:
        cpu: "500m"
        memory: "2Gi"
      maxAllowed:
        cpu: "4"
        memory: "12Gi"
      controlledResources: ["cpu", "memory"]

spec:
  affinity:
    nodeAffinity:
      requiredDuringSchedulingIgnoredDuringExecution:
        nodeSelectorTerms:
        - matchExpressions:
          - key: topology.kubernetes.io/zone
            operator: In
            values:
            - zone-a
  topologySpreadConstraints:
  - maxSkew: 1
    topologyKey: kubernetes.io/hostname
    whenUnsatisfiable: DoNotSchedule
    labelSelector:
      matchLabels:
        app: high-freq-trading

func (ds *DynamicScheduler) prioritizeNodes(
    ctx context.Context,
    pod *v1.Pod,
    nodes []*v1.Node,
) (framework.NodeScoreList, error) {
    
    scores := make(framework.NodeScoreList, len(nodes))
    
    for i, node := range nodes {
        // 计算 GPU 碎片率
        fragScore := calculateGPUFragmentation(node)
        
        // 评估节点负载均衡
        loadScore := getNodeLoadScore(node)
        
        // 结合亲和性得分
        affinityScore := ds.affinityScore(pod, node)
        
        // 综合评分（权重可配置）
        totalScore := fragScore*0.4 + loadScore*0.3 + affinityScore*0.3
        scores[i] = framework.NodeScore{
            Name:  node.Name,
            Score: int64(totalScore * 100),
        }
    }
    
    return scores, nil
}

apiVersion: batch/v1
kind: Job
metadata:
  name: data-processing
spec:
  parallelism: 1000
  completions: 10000
  backoffLimit: 0
  podFailurePolicy:
    rules:
    - action: Terminate
      onExitCodes:
        containerName: main
        operator: In
        values: [1, 2, 137]

# 压测结果对比
kubectl run --rm -i --tty load-test --image=busybox --restart=Never -- \
    ab -c 100 -n 10000 http://apiserver:8080/api/v1/pods

# etcd 启动参数
- --auto-compaction-retention=1h
- --quota-backend-bytes=8589934592  # 8GB
- --max-request-bytes=15728640      # 15MB

# 调整 TCP 缓冲区
sysctl -w net.core.rmem_max=16777216
sysctl -w net.core.wmem_max=16777216
sysctl -w net.ipv4.tcp_rmem="4096 87380 16777216"
sysctl -w net.ipv4.tcp_wmem="4096 65536 16777216"

# 启用 BBR 拥塞控制
echo "net.core.default_qdisc=fq" >> /etc/sysctl.conf
echo "net.ipv4.tcp_congestion_control=bbr" >> /etc/sysctl.conf

fio --name=test --ioengine=libaio --rw=randread --bs=4k \
    --numjobs=16 --size=10G --runtime=60 --time_based \
    --group_reporting

osd_pool_default_size: 3
osd_pool_default_min_size: 2
osd_max_backfills: 4
osd_recovery_max_active: 6
osd_op_num_threads_per_shard: 4
bluestore_cache_autotune: true

- alert: HighAPILatency
  expr: histogram_quantile(0.99, sum(rate(apiserver_request_duration_seconds_bucket[5m])) by (le, verb)) > 1
  for: 10m
  labels:
    severity: critical
  annotations:
    summary: "API latency high ({{ $value }}s)"
    
- alert: UnbalancedNodes
  expr: stddev(node_memory_Utilization) > 0.3
  for: 30m
  labels:
    severity: warning