RayCluster 커스텀 리소스를 사용하면 GKE가 Kubernetes 포드로 배포하는 Ray 클러스터를 지정할 수 있습니다. Ray 클러스터는 일반적으로 단일 헤드 포드와 여러 작업자 포드로 구성됩니다.
RayJob 커스텀 리소스
RayJob 커스텀 리소스를 사용하면 단일 Ray 작업을 실행할 수 있습니다. KubeRay는 작업의 컴퓨팅 리소스를 제공하기 위해 RayCluster를 만든 다음 RayCluster의 헤드 포드에 Ray 작업을 제출하는 Kubernetes 작업을 만듭니다.
효율적인 리소스 관리를 위해 작업이 성공적으로 완료된 후 RayCluster를 자동으로 삭제하도록 KubeRay를 구성할 수 있습니다.
RayService 커스텀 리소스
RayService 커스텀 리소스를 사용하면 모델 서빙 및 추론을 위한 애플리케이션과 같은 Ray Serve 애플리케이션을 구성할 수 있습니다. KubeRay는 컴퓨팅 리소스를 제공하기 위해 RayCluster를 만든 다음 Ray Serve 구성에 지정된 대로 Ray Serve 애플리케이션을 배포합니다.
GKE의 Ray 공유 책임
Ray 연산자를 사용하여 GKE에서 Ray 워크로드를 실행하기로 선택하는 경우 Google Cloud와 사용자, 즉 고객 간에 책임을 분담하는 방식을 이해해야 합니다.
Google의 책임
KubeRay 연산자의 안정성과 업타임 유지관리
KubeRay 연산자의 버전 업그레이드 관리
RayCluster, RayJob, RayService 커스텀 리소스를 관리하기 위한 KubeRay 관련 기능
[[["이해하기 쉬움","easyToUnderstand","thumb-up"],["문제가 해결됨","solvedMyProblem","thumb-up"],["기타","otherUp","thumb-up"]],[["이해하기 어려움","hardToUnderstand","thumb-down"],["잘못된 정보 또는 샘플 코드","incorrectInformationOrSampleCode","thumb-down"],["필요한 정보/샘플이 없음","missingTheInformationSamplesINeed","thumb-down"],["번역 문제","translationIssue","thumb-down"],["기타","otherDown","thumb-down"]],["최종 업데이트: 2025-09-01(UTC)"],[],[],null,["# About Ray on Google Kubernetes Engine (GKE)\n\n[Autopilot](/kubernetes-engine/docs/concepts/autopilot-overview) [Standard](/kubernetes-engine/docs/concepts/choose-cluster-mode)\n\n*** ** * ** ***\n\nThis page provides an overview of the Ray Operator and relevant custom\nresources to deploy and manage Ray clusters and applications on\nGoogle Kubernetes Engine (GKE).\n\n[Ray](https://www.ray.io/) is an open-source unified compute framework for\nscaling AI/ML and Python applications. Ray provides a set of libraries to\ndistribute the compute runtime for AI/ML across multiple compute nodes.\n\nTo learn how to enable the Ray operator on GKE, see\n[Enable the Ray operator on GKE](/kubernetes-engine/docs/add-on/ray-on-gke/how-to/enable-ray-on-gke).\n\nWhy use the Ray Operator on GKE\n-------------------------------\n\nThe Ray Operator is the recommended way to deploy and manage Ray clusters on\nGKE. When you run the Ray Operator on GKE, you\nbenefit from Ray's support for Python and GKE's enterprise-grade\nreliability, portability, and scalability.\n\nThe Ray Operator on GKE is based on\n[KubeRay](https://github.com/ray-project/kuberay), which provides\ndeclarative Kubernetes APIs specifically designed for managing Ray clusters.\nThis means that you can provision, scale, and manage your Ray Deployments with\nother containerized workloads on GKE.\n\nHow the Ray Operator on GKE works\n---------------------------------\n\nWhen you enable the Ray Operator in your GKE clusters,\nGKE automatically installs and hosts the KubeRay operator.\n\nKubeRay provides Kubernetes custom resources to manage Ray Deployments\non Kubernetes, including:\n\n- [RayCluster](#raycluster)\n- [RayJob](#rayjob)\n- [RayService](#rayservice)\n\n### RayCluster custom resource\n\nThe RayCluster custom resource lets you specify a Ray cluster that\nGKE deploys as Kubernetes Pods. A Ray cluster typically consists\nof a single head Pod and multiple worker Pods.\n\n### RayJob custom resource\n\nThe RayJob custom resource lets you execute a single Ray job. KubeRay\ncreates a RayCluster to provide compute resources for the job, then creates a\nKubernetes Job that submits the Ray job to the head Pod of the RayCluster.\n\nFor efficient resource management, you can configure KubeRay to\nautomatically clean up the RayCluster after your job is successfully completed.\n\n### RayService custom resource\n\nThe RayService custom resource lets you configure\n[Ray Serve applications](https://docs.ray.io/en/latest/serve/index.html),\nsuch as applications for model serving and inference. KubeRay creates a\nRayCluster to provide the compute resources and then deploys the Ray Serve\napplication as specified by the Ray Serve configuration.\n\nRay on GKE shared responsibility\n--------------------------------\n\nWhen you choose to run Ray workloads on GKE with the Ray operator, it's\nimportant to understand how responsibilities are divided between Google Cloud\nand you, the customer:\n\n### Google's responsibilities\n\n- Maintaining the reliability and uptime of the KubeRay operator.\n- Managing version upgrades for the KubeRay operator.\n- Capabilities specific to KubeRay for managing the RayCluster, RayJob, and RayService custom resources.\n\n### Customer's responsibilities\n\n- Maintaining container images used for the Ray head and Ray worker Pods.\n- Maintaining versioning and upgrades for the Ray head and Ray worker Pods.\n- Configuring resource requirements (CPU, GPU, memory, etc.) for your Ray clusters.\n- Following best practices for [securing Ray clusters](https://docs.ray.io/en/latest/ray-security/index.html).\n- Reliability and monitoring for your Ray applications.\n\nSee [GKE shared responsibility](/kubernetes-engine/docs/concepts/shared-responsibility) to learn more.\n\nWhat's next\n-----------\n\n- Learn how to [Enable the Ray operator on GKE](/kubernetes-engine/docs/add-on/ray-on-gke/how-to/enable-ray-on-gke).\n- Explore the [Ray on Kubernetes](https://docs.ray.io/en/latest/cluster/kubernetes/index.html) documentation."]]
