Kubernetes for People Who Have to Run It

课程概述

This is the highest level in the Library and it does not open by explaining what a container is for. It assumes you run something. Somebody else built the cluster, or you built it in a hurry two years ago, and you are now the person who gets rung when it stops. If you are still deciding whether to adopt this platform at all, read the fourth module of Securing a Cloud Estate You Didn't Build first — it works through a client who asked for a Kubernetes-scale answer to a spreadsheet-scale problem, and it gives you the words to refuse. That argument is finished and this segment does not reopen it. This one starts one step later. Everything here is taught from something that failed. You will apply an Ingress that is accepted and does nothing, schedule pods that never run, resolve a service name perfectly and connect to nothing, and fill a machine with 32 idle cores and 96 GB of untouched memory until it refuses the next twenty-six pods. That last one is the best technical moment in the source material: a node with headroom for around four hundred pods that can only be given 256 internal addresses per interface, so full-mesh communication caps density long before the processor does. The last module builds a control plane from nothing on a throwaway host. It fails four times first — a leftover file, forwarding disabled, a port already taken, and a bootstrap that hangs for four minutes and then gives up with an error that does not name the cause. The bring-up in the source notes stops at that hang, unfinished. This one finishes it. The thread running through all four modules is a single habit: in this platform a green tick means an object was accepted, and says nothing whatsoever about whether anything acted on it.

1. 01 实际运行的是什么，从外向内看

   3 节课程·1h 51m

   每个人都是从内向外学习这个平台，然后在凌晨3点的压力下从外向内调试它。这个模块按照事件到达的方向构建地图，并以任何交接文档都不会包含的那一半内容结尾。

2. 02 事物落地的位置及其原因

   3 节课程·1h 56m

   任何人都可以放置一个 pod。真正的工作是让这个安排在与从未听说过它的其他十一名工程师接触后仍然存活，这需要用到大多数人只会用到一半的机制的两个部分。

3. 03 网络与耗尽的地址空间

   3 节课程·2h 41m

   本模块中的两个实验室产生一个在监控收集的每项指标上都健康的集群，却无法启动另一个工作负载。第二个在一小时后不留任何痕迹，这才是真正被教导的部分。

4. 04 启动失败

   3 节课程·2h 52m

   大多数读到这篇文章的人永远不会构建控制平面，这是一个合理的决定而非缺陷。不过还是做一次吧：启动过程是这个平台唯一会按顺序列举自身、告诉你它由什么组成的时刻。

常见问题

我将在 Kubernetes for People Who Have to Run It 中学到什么？

For the engineer who already has this in production and did not choose it. The request path outside in, scheduling and node pools, the address space that runs out while the hardware sits idle, and a control-plane bootstrap worked through its real failure sequence to a conclusion.

我需要有先前的经验吗？

建议在开始 Kubernetes for People Who Have to Run It 前具有一定的先前知识。

Kubernetes for People Who Have to Run It 需要多长时间？

Kubernetes for People Who Have to Run It 包含 4 个模块和 12 节课程。您可以按自己的速度学习。

我如何获得访问权限？

Kubernetes for People Who Have to Run It 包含在任何付费订阅中。