Dubhe/dubhe-server/yaml/common-k8s.yaml

train-job:
  docker-dataset-path: "/dataset"

k8s:
  # k8s集群配置文件
  kubeconfig: kubeconfig_test
  # nfs服务暴露的IP地址 如需测试需修改为合适的地址
  nfs: 127.0.0.1
  #nfs服务端 共享目录
  nfs-root-path: /nfs/
  nfs-root-windows-path: "Z:"
   # 文件存储服务暴露的IP地址 如需测试需修改为合适的地址
  file-store: 127.0.0.1
  #文件存储服务端 共享目录
  file-store-root-path: /nfs/
  file-store-root-windows-path: "Z:"
  # 命名空间关键字
  namespace: namespace
  # k8s ingress域名 如需测试需修改为合适的域名,注意：需要为此域名配置泛域名解析到 k8s集群master节点的ip
  host: notebooktest.dubhe.club
  # k8s ingress-controller 对外port
  port: 30865
  # k8s ingress-controller 对外grpc port
  https-port: 31287
  # k8s 模型部署配置
  serving:
    # 在线服务模型部署后容器域名(k8s ingress域名),解析地址为k8s集群地址,如需测试需修改为合适的域名
    host: servingtest.dubhe.club
    # tls 证书 crt
    tls-crt: 
    # tls 证书 key
    tls-key: 
  # elasticsearch暴露的服务地址
  elasticsearch:
    hostlist: ${eshostlist:127.0.0.1:30498}
    # 日志采集配置信息
    log:
      type: _doc
      # 过滤源字段
      source_field: log,@timestamp,kubernetes.pod_name
  # 异步回调
  callback:
    # boot 单机部署即回调本机实例
    url: localhost:${server.port}
    token:
      # 秘钥
      secret-key: 1qaz2wsx
      # 过期时间（秒）
      expire-seconds: 300
  # 展示Pod的CPU使用率,Memory使用量,GPU使用率的grafana地址
  pod:
    metrics:
      grafanaUrl: http://127.0.0.1:30006/d/job/monitor?orgId=1&refresh=5s&kiosk&var-pod=
  prometheus:
    url: http://127.0.0.1:30003/
    query: api/v1/query
    query-range: api/v1/query_range
    gpu-query-param: sum(DCGM_FI_DEV_GPU_UTIL{pod="pod-name-placeholder"})by(pod,UUID)
    gpu-mem-total-query-param: sum(DCGM_FI_DEV_FB_TOTAL_MEGABYTES{pod="pod-name-placeholder"})by(pod,UUID)
    gpu-mem-use-query-param: sum(DCGM_FI_DEV_FB_USED{pod="pod-name-placeholder"})by(pod,UUID)
    cpu-range-query-param: sum(rate(container_cpu_usage_seconds_total{image!="",pod="pod-name-placeholder"}[1m])) by (pod) / (sum(container_spec_cpu_quota{image!=""}/100000) by (pod)) * 100
    mem-range-query-param: sum(container_memory_rss{image!="",pod="pod-name-placeholder"})
    gpu-usage-query-param: sum by(Hostname,gpu)(DCGM_FI_DEV_GPU_UTIL{Hostname="node-name-placeholder",pod!=""})
    gpu-range-query-param: sum(DCGM_FI_DEV_GPU_UTIL{pod="pod-name-placeholder"}) by (pod,UUID)
    gpu-mem-total-range-query-param: sum(DCGM_FI_DEV_FB_TOTAL_MEGABYTES{pod="pod-name-placeholder"}) by (pod,UUID)
    gpu-mem-use-range-query-param: sum(DCGM_FI_DEV_FB_USED{pod="pod-name-placeholder"}) by (pod,UUID)
    gpu-usage-rate-query-param: topk(10,sort_desc(max_over_time(namespace:DCGM_FI_DEV_GPU_UTIL:sumn[usage-rate-day])))
    cpu-usage-rate-query-param: topk(10,sort_desc(max_over_time(namespace:container_cpu_user_seconds_total_sumn:raten[usage-rate-day])))
    mem-usage-rate-query-param: topk(10,sort_desc(max_over_time(namespace:CONTAINER_MEMERY_USAGE_BYTES:sumn{namespace=~"namespace.*"}[usage-rate-day])))
    gpu-usage-namespace-query-param: max_over_time(namespace:DCGM_FI_DEV_GPU_UTIL:sumn{namespace=~"namespace-placeholder"}[usage-rate-day])
    cpu-usage-namespace-query-param: max_over_time(namespace:container_cpu_user_seconds_total_sumn:raten{namespace=~"namespace-placeholder"}[usage-rate-day])
    mem-usage-namespace-query-param: max_over_time(namespace:CONTAINER_MEMERY_USAGE_BYTES:sumn{namespace=~"namespace-placeholder"}[usage-rate-day])
  nfs-storage-class-name: zjlab-nfs-storage
  namespace-limits:
    cpu: 10
    memory: 32
    gpu: 2
#配置harbor
harbor:
  address: harbor.dubhe.ai
  username: admin
  password: Harbor12345
  model-name: train

# minio配置
minio:
  url: http://127.0.0.1:9000/
  accessKey: admin
  secretKey: 123@abc.com
  bucketName: dubhe-prod
  presignedUrlExpiryTime: 300
  annotation: /annotation/

docker:
  remote-api-port: 2375