Update localization guidelines (#10485)
* Update localization guidelines for language labels
Continuing work
Continuing work
Continuing work
More work in progress
Add local OWNERS folders
Add an OWNERS file to Chinese
Remove shortcode for repos
Add Japanese
Alphabetize languages, change weights accordingly
More updates
Add Korean in Korean
Add English to languageName
Feedback from gochist
Move Chinese content from cn/ to zh/
Move OWNERS from cn/ to zh/
Resolve merge conflicts by updating from master
Add files back in to prep for resolution
After rebase on upstream/master, remove files
Review and update localization guidelines
Feedback from gochist, tnir, cstoku
Add a trailing newline to content/ja/OWNERS
Add a trailing newline to content/zh/OWNERS
Drop requirement for GH repo project
Clarify language about forks/branches
Edits and typos
Remove a shortcode specific to a multi-repo language setup
Update aliases and owners
Add explicit OWNERS for content/en
Migrate content from Chinese repo, update regex in config.toml
Remove untranslated strings
Add trailing newline to content/en/OWNERS
Add trailing newlines to OWNERS files
add Jaguar project description (#10433)
* add Jaguar project description
[Jaguar](https://gitlab.com/sdnlab/jaguar) is an open source solution for Kubernetes's network based on OpenDaylight.
Jaguar provides overlay network using vxlan and Jaguar CNIPlugin provides one IP address per pod.
* Minor newline tweak
blog post for azure vmss (#10538)
Add microk8s to pick-right-solution.md (#10542)
* Add microk8s to pick-right-solution.md
Microk8s is a single-command installation of upstream Kubernetes on any Linux and should be included in the list of local-machine solutions.
* capitalized Istio
Add microk8s to foundational.md (#10543)
* Add microk8s to foundational.md
Adding microk8s as credible and stable alternative to get started with Kubernetes on a local machine. This is especially attractive for those not wanting to incur the overhead of running a VM for a local cluster.
* Update foundational.md
Thank you for your suggestions! LMK if this works now?
* Rewrote first paragraph
And included a bullet list of features of microk8s
* Copyedit
fix typo (#10545)
Fix the kubectl subcommands links. (#10550)
Signed-off-by: William Zhang <warmchang@outlook.com>
Fix command issue (#10515)
Signed-off-by: mooncake <xcoder@tenxcloud.com>
remove imported community files per issue 10184 (#10501)
networking.md: Markdown fix (#10498)
Fix front matter, federation command-line tools (#10500)
Clean up glossary entry (#10399)
update slack link (#10536)
typo in StatefulSet docs (#10558)
fix discription about horizontal pod autoscale (#10557)
Remove redundant symbols (#10556)
Fix issue #10520 (#10554)
Signed-off-by: William Zhang <warmchang@outlook.com>
Update api-concepts.md (#10534)
Revert "Fix command issue (#10515)"
This reverts commit c02a7fb9f9.
Update memory-constraint-namespace.md (#10530)
update memory request to 100MiB corresponding the yaml content
Blog: Introducing Volume Snapshot Alpha for Kubernetes (#10562)
* blog post for azure vmss
* snapshot blog post
Resolve merge conflicts in OWNERS*
Minor typo fix (#10567)
Not sure what's supposed to be here, proposing removing it.
* Feedback from gochist
Tweaks to feedback
* Feedback from ClaudiaJKang
This commit is contained in:
Zach Corleissen
committed by
k8s-ci-robot
k8s-ci-robot
parent
753f57f0e6
commit
abcee2dccd
@@ -0,0 +1,179 @@
|
||||
---
|
||||
approvers:
|
||||
- vishh
|
||||
title: 调度 GPU
|
||||
content_template: templates/task
|
||||
---
|
||||
|
||||
{{% capture overview %}}
|
||||
|
||||
|
||||
Kubernetes 提供对分布在节点上的 NVIDIA GPU 进行管理的**实验**支持。本页描述用户如何使用 GPU 以及当前使用的一些限制
|
||||
|
||||
{{% /capture %}}
|
||||
|
||||
{{% capture prerequisites %}}
|
||||
|
||||
|
||||
1. Kubernetes 节点必须预先安装好 NVIDIA 驱动，否则，Kubelet 将检测不到可用的GPU信息；如果节点的 Capacity 属性中没有出现 NIVIDA GPU 的数量，有可能是驱动没有安装或者安装失败，请尝试重新安装
|
||||
|
||||
2. 在整个 Kubernetes 系统中，feature-gates 里面特定的 **alpha** 特性参数 `Accelerators` 必须设置为 true：`--feature-gates="Accelerators=true"`
|
||||
|
||||
3. Kuberntes 节点必须使用 `docker` 引擎作为容器的运行引擎
|
||||
|
||||
|
||||
上述预备工作完成后，节点会自动发现它上面的 NVIDIA GPU，并将其作为可调度资源暴露
|
||||
|
||||
{{% /capture %}}
|
||||
|
||||
{{% capture steps %}}
|
||||
|
||||
## API
|
||||
|
||||
|
||||
容器可以通过名称为 `alpha.kubernetes.io/nvidia-gpu` 的标识来申请需要使用的 NVIDIA GPU 的数量
|
||||
|
||||
```yaml
|
||||
apiVersion: v1
|
||||
kind: Pod
|
||||
metadata:
|
||||
name: gpu-pod
|
||||
spec:
|
||||
containers:
|
||||
-
|
||||
name: gpu-container-1
|
||||
image: k8s.gcr.io/pause:2.0
|
||||
resources:
|
||||
limits:
|
||||
alpha.kubernetes.io/nvidia-gpu: 2 # requesting 2 GPUs
|
||||
-
|
||||
name: gpu-container-2
|
||||
image: k8s.gcr.io/pause:2.0
|
||||
resources:
|
||||
limits:
|
||||
alpha.kubernetes.io/nvidia-gpu: 3 # requesting 3 GPUs
|
||||
```
|
||||
|
||||
|
||||
- GPU 只能在容器资源的 `limits` 中配置
|
||||
|
||||
- 容器和 Pod 都不支持共享 GPU
|
||||
|
||||
- 每个容器可以申请使用一个或者多个 GPU
|
||||
|
||||
- GPU 必须以整数为单位被申请使用
|
||||
|
||||
- 所有节点的 GPU 硬件要求相同
|
||||
|
||||
|
||||
如果在不同的节点上面安装了不同版本的 GPU，可以通过设置节点标签以及使用节点选择器的方式将 pod 调度到期望运行的节点上。工作流程如下：
|
||||
|
||||
|
||||
在节点上，识别出 GPU 硬件类型，然后将其作为节点标签进行暴露
|
||||
|
||||
```shell
|
||||
NVIDIA_GPU_NAME=$(nvidia-smi --query-gpu=gpu_name --format=csv,noheader --id=0)
|
||||
source /etc/default/kubelet
|
||||
KUBELET_OPTS="$KUBELET_OPTS --node-labels='alpha.kubernetes.io/nvidia-gpu-name=$NVIDIA_GPU_NAME'"
|
||||
echo "KUBELET_OPTS=$KUBELET_OPTS" > /etc/default/kubelet
|
||||
```
|
||||
|
||||
|
||||
在 pod 上，通过节点[亲和性](/docs/concepts/configuration/assign-pod-node/#affinity-and-anti-affinity)规则为它指定可以使用的 GPU 类型
|
||||
|
||||
```yaml
|
||||
kind: pod
|
||||
apiVersion: v1
|
||||
metadata:
|
||||
annotations:
|
||||
scheduler.alpha.kubernetes.io/affinity: >
|
||||
{
|
||||
"nodeAffinity": {
|
||||
"requiredDuringSchedulingIgnoredDuringExecution": {
|
||||
"nodeSelectorTerms": [
|
||||
{
|
||||
"matchExpressions": [
|
||||
{
|
||||
"key": "alpha.kubernetes.io/nvidia-gpu-name",
|
||||
"operator": "In",
|
||||
"values": ["Tesla K80", "Tesla P100"]
|
||||
}
|
||||
]
|
||||
}
|
||||
]
|
||||
}
|
||||
}
|
||||
}
|
||||
spec:
|
||||
containers:
|
||||
-
|
||||
name: gpu-container-1
|
||||
resources:
|
||||
limits:
|
||||
alpha.kubernetes.io/nvidia-gpu: 2
|
||||
```
|
||||
|
||||
|
||||
上述设定可以确保 pod 会被调度到包含名称为 `alpha.kubernetes.io/nvidia-gpu-name` 的标签并且标签的值为 `Tesla K80` 或者 `Tesla P100` 的节点上
|
||||
|
||||
|
||||
### 警告
|
||||
|
||||
|
||||
当未来的 Kubernetes 版本能够更好的支持GPU以及一般的硬件加速器时，这里的 API 描述**将会随之做出变更**
|
||||
|
||||
|
||||
## 访问 CUDA 库
|
||||
|
||||
|
||||
到目前为止，还需要预先在节点上安装 CUDA 库
|
||||
|
||||
|
||||
为了避免后面使用库出现问题，可以将库放到 ``/var/lib/`` 下的某个文件夹下，或者直接改变库目录的权限(以后的版本会自动完成这一过程)
|
||||
|
||||
|
||||
Pods能够通过 `hostPath` 卷来访问库
|
||||
|
||||
```yaml
|
||||
kind: Pod
|
||||
apiVersion: v1
|
||||
metadata:
|
||||
name: gpu-pod
|
||||
spec:
|
||||
containers:
|
||||
- name: gpu-container-1
|
||||
image: k8s.gcr.io/pause:2.0
|
||||
resources:
|
||||
limits:
|
||||
alpha.kubernetes.io/nvidia-gpu: 1
|
||||
volumeMounts:
|
||||
- mountPath: /usr/local/nvidia/bin
|
||||
name: bin
|
||||
- mountPath: /usr/lib/nvidia
|
||||
name: lib
|
||||
volumes:
|
||||
- hostPath:
|
||||
path: /usr/lib/nvidia-375/bin
|
||||
name: bin
|
||||
- hostPath:
|
||||
path: /usr/lib/nvidia-375
|
||||
name: lib
|
||||
```
|
||||
|
||||
|
||||
## 未来
|
||||
|
||||
|
||||
- Kubernetes 对硬件加速器的支持还处在早期阶段
|
||||
|
||||
- GPU 和其它的加速器很快会成为系统的本地计算资源
|
||||
|
||||
- 将引入更好的 API 以可扩展的方式提供和使用加速器
|
||||
|
||||
- Kubernets 将会自动确保应用在使用 GPU 时得到最佳性能
|
||||
|
||||
- 类似访问 CUDA 库这种关键的可用性问题将得到解决
|
||||
|
||||
{{% /capture %}}
|
||||
|
||||
|
||||
Reference in New Issue
Block a user