如何使用 fabric8 kubernetes Java 客户端 API 在容器上设置 GPU 资源要求

Question

我用 fabric8 kubernetes Java 客户端 API 编写了一个示例来设置容器上的 GPU 资源要求。我遇到以下运行时错误：

spec.containers[0].resources.requests[gpu]: Invalid value: "gpu": must be a standard resource type or fully qualified, 
spec.containers[0].resources.requests[gpu]: Invalid value: "gpu": must be a standard resource for containers.

fabric8 jar版本为4.3.0（最新）。貌似fabric8到现在还不支持gpu资源需求，我把"addToRequests("gpu", new Quantity("1"))"这一行去掉就可以正常工作了。

如何在 Java/Scala 应用程序中启用 GPU 资源需求？

示例的全部源码如下：

/**
 * Copyright (C) 2015 Red Hat, Inc.
 *
 * Licensed under the Apache License, Version 2.0 (the "License");
 * you may not use this file except in compliance with the License.
 * You may obtain a copy of the License at
 *
 *         http://www.apache.org/licenses/LICENSE-2.0
 *
 * Unless required by applicable law or agreed to in writing, software
 * distributed under the License is distributed on an "AS IS" BASIS,
 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 * See the License for the specific language governing permissions and
 * limitations under the License.
 */
package com.exam.docker.kubernetes.examples;


import io.fabric8.kubernetes.api.model.*;
import io.fabric8.kubernetes.client.*;
import io.fabric8.kubernetes.client.Config;
import io.fabric8.kubernetes.client.ConfigBuilder;
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;

import java.util.UUID;
import java.util.concurrent.CountDownLatch;
import java.util.concurrent.TimeUnit;


public class PodResExamples {

  private static final Logger logger = LoggerFactory.getLogger(PodResExamples.class);

  public static void main(String[] args) {
    String master = "http://127.0.0.1:8080/";
    if (args.length == 1) {
      master = args[0];
    }
    String ns = "thisisatest";
    String serviceName = "cuda-vector-add-"+ UUID.randomUUID();

    Config config = new ConfigBuilder().withMasterUrl(master).build();
    try (KubernetesClient client = new DefaultKubernetesClient(config)) {
      try {
        if(client.namespaces().withName(ns).get() == null) {
          log("Create namespace:", client.namespaces().create(new NamespaceBuilder().withNewMetadata().withName(ns).endMetadata().build()));
        }

        String imageStr = "k8s.gcr.io/cuda-vector-add:v0.1";
        String cmd = "";

        final ResourceRequirements resources = new ResourceRequirementsBuilder()
                .addToRequests("cpu", new Quantity("2"))
                .addToRequests("memory", new Quantity("10Gi"))
                .addToRequests("gpu", new Quantity("1"))
                .build();

        Container container = new ContainerBuilder().withName(serviceName)
                .withImage(imageStr).withImagePullPolicy("IfNotPresent")
                .withArgs(cmd)
                .withResources(resources)
                .build();

        Pod createdPod = client.pods().inNamespace(ns).createNew()
                .withNewMetadata()
                .withName(serviceName)
                .addToLabels("podres", "cuda-vector")
                .endMetadata()
                .withNewSpec()
                .addToContainers(container)
                .withRestartPolicy("Never")
                .endSpec().done();
        log("Created pod cuda-vector-add:", createdPod);

        final CountDownLatch watchLatch = new CountDownLatch(1);
        try (final Watch ignored = client.pods().inNamespace(ns).withLabel("podres").watch(new Watcher<Pod>() {
          @Override
          public void eventReceived(final Action action, Pod pod) {
            if (pod.getStatus().getPhase().equals("Succeeded")) {
              logger.info("Pod cuda-vector is completed!");
              logger.info(client.pods().inNamespace(ns).withName(pod.getMetadata().getName()).getLog());
              watchLatch.countDown();
            } else if (pod.getStatus().getPhase().equals("Pending")) {
              logger.info("Pod cuda-vector is Pending!");                  
            }
          }

          @Override
          public void onClose(final KubernetesClientException e) {
            logger.info("Cleaning up pod.");
          }
        })) {
          watchLatch.await(30, TimeUnit.SECONDS);
        } catch (final KubernetesClientException | InterruptedException e) {
          e.printStackTrace();
          logger.error("Could not watch pod", e);
        }

      } catch (KubernetesClientException e) {
        logger.error(e.getMessage(), e);
      } finally {
        log("Pod cuda-vector log: \n", client.pods().inNamespace(ns).withName(serviceName).getLog());
        client.namespaces().withName(ns).delete();
      }
    }
  }

  private static void log(String action, Object obj) {
    logger.info("{}: {}", action, obj);
  }

  private static void log(String action) {
    logger.info(action);
  }

}

Answer 1

参考 Kubernetes Docs 你可以尝试使用 nvidia.com/gpu 而不是 gpu:

apiVersion: v1
kind: Pod
metadata:
  name: cuda-vector-add
spec:
  containers:
    - name: cuda-vector-add
      image: "k8s.gcr.io/cuda-vector-add:v0.1"
      resources:
        limits:
          nvidia.com/gpu: 1 # requesting 1 GPU

如果您的应用程序改用 AMD GPU，请尝试 amd.com/gpu 重要说明：您不能设置 GPU-Request，除非您也将限制设置为与请求相等。

如何使用 fabric8 kubernetes Java 客户端 API 在容器上设置 GPU 资源要求

How to set GPU resource requirements on a container with the fabric8 kubernetes Java client API

kubernetes

fabric8