From adf562ffa15d31f8dfefb7c6765abc7c6a4c2140 Mon Sep 17 00:00:00 2001 From: resolvewang <1796246076@qq.com> Date: Sun, 19 Nov 2017 15:36:06 +0800 Subject: [PATCH] fix typo --- concepts/pod.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/concepts/pod.md b/concepts/pod.md index 45a8faf27..20de7e1ed 100644 --- a/concepts/pod.md +++ b/concepts/pod.md @@ -73,7 +73,7 @@ Pod也可以用于垂直应用栈(例如LAMP),这样使用的主要动机 Pod在设计支持就不是作为持久化实体的。在调度失败、节点故障、缺少资源或者节点维护的状态下都会死掉会被驱逐。 -通常,用户不需要手动直接创建Pod,而是应该使用controller(例如[Deployments](./deployment.md)),即使时创建单个Pod的情况下。Controller可以提供集群级别的自愈功能、复制和升级管理。 +通常,用户不需要手动直接创建Pod,而是应该使用controller(例如[Deployments](./deployment.md)),即使是在创建单个Pod的情况下。Controller可以提供集群级别的自愈功能、复制和升级管理。 The use of collective APIs as the primary user-facing primitive is relatively common among cluster scheduling systems, including [Borg](https://research.google.com/pubs/pub43438.html), [Marathon](https://mesosphere.github.io/marathon/docs/rest-api.html), [Aurora](http://aurora.apache.org/documentation/latest/reference/configuration/#job-schema), and [Tupperware](http://www.slideshare.net/Docker/aravindnarayanan-facebook140613153626phpapp02-37588997).