The code and data behind xeiaso.net
kube-ps1 post
Signed-off-by: Xe Iaso <me@xeiaso.net>
+87
-9
+85
lume/src/notes/2025/kube-ps1.mdx
+85
lume/src/notes/2025/kube-ps1.mdx
···
1
+
---
2
+
title: "Life pro tip: put your active kubernetes context in your prompt"
3
+
desc: "kube_ps1 is love, kube_ps1 is life"
4
+
date: 2025-04-05
5
+
hero:
6
+
ai: "Photo by Xe Iaso, Canon EOS R6 Mark ii, 16mm wide angle lens"
7
+
file: touch-grass
8
+
prompt: "A color-graded photo of a forest in Gatineau Park, the wildlife looks green and lush"
9
+
---
10
+
11
+
Today I did an oopsie. I tried to upgrade a service in my homelab cluster (`alrest`) but accidentally upgraded it in the production cluster (`aeacus`). I was upgrading `ingress-nginx` to patch [the security vulnerabilities released a while ago](https://kubernetes.io/blog/2025/03/24/ingress-nginx-cve-2025-1974/). I should have done it sooner, but [things have been rather wild lately](https://arstechnica.com/ai/2025/03/devs-say-ai-crawlers-dominate-traffic-forcing-blocks-on-entire-countries/) and now [kernel.org runs some software I made](https://social.kernel.org/notice/Asir7LiPevX6XcEVJQ).
12
+
13
+
<Conv name="Cadey" mood="coffee">
14
+
<Picture
15
+
path="notes/2025/kube-ps1/domino-meme"
16
+
desc="A domino effect starting at 'Amazon takes out my git server' ending in 'software running on kernel.org'."
17
+
/>
18
+
</Conv>
19
+
20
+
Either way, I found out that [Oh my ZSH](https://ohmyz.sh/) (the ZSH prompt toolkit I use) has a plugin for [kube_ps1](https://github.com/ohmyzsh/ohmyzsh/blob/master/plugins/kube-ps1/README.md). This lets you put your active Kubernetes context in your prompt so that you're less likely to apply the wrong manifest to the wrong cluster.
21
+
22
+
To install it, I changed the `plugins` list in my `~/.zshrc`:
23
+
24
+
```diff
25
+
-plugins=(git)
26
+
+plugins=(git kube-ps1)
27
+
```
28
+
29
+
And then added configuration at the end for kube_ps1:
30
+
31
+
```sh
32
+
export KUBE_PS1_NS_ENABLE=false
33
+
export KUBE_PS1_SUFFIX=") "
34
+
35
+
PROMPT='$(kube_ps1)'$PROMPT
36
+
```
37
+
38
+
This makes my prompt look like this:
39
+
40
+
```text
41
+
(⎈|alrest) ➜ site git:(main) ✗
42
+
```
43
+
44
+
Showing that I'm using the Kubernetes cluster Alrest.
45
+
46
+
<ConvP>
47
+
<Conv name="Aoi" mood="wut">
48
+
Wouldn't it be better to modify your configuration such that you always have
49
+
to pass a `--context` flag or something?
50
+
</Conv>
51
+
<Conv name="Cadey" mood="coffee">
52
+
Yes, but some of the tools I use don't have that support universally. Until
53
+
I can ensure they all do, I'm willing to settle for tamper-evident instead
54
+
of tamper-resistant.
55
+
</Conv>
56
+
</ConvP>
57
+
58
+
## Why upgrading ingress-nginx broke my HTTP ingress setup
59
+
60
+
Apparently when I set up the Kubernetes cluster for my website, the [Anubis docs](https://anubis.techaro.lol) and other things like my Headscale server, I did a very creative life decision. I started out with the "baremetal" self-hosted ingress-nginx install flow and then manually edited the `Service` to be a `LoadBalancer` service instead of a `NodePort` service.
61
+
62
+
I had forgotten about this. So when the upgrade hit the wrong cluster, Kubernetes happily made that `Service` into a `NodePort` service, destroying the cloud's load balancer that had been doing all of my HTTP ingress.
63
+
64
+
Thankfully, Kubernetes dutifully recorded logs of that entire process, which I have reproduced here for your amusement.
65
+
66
+
| Event type | Reason | Age | From | Message |
67
+
| :--------- | :------------------- | :-- | :----------------- | :----------------------- |
68
+
| Normal | Type changed | 13m | service-controller | LoadBalancer -> NodePort |
69
+
| Normal | DeletingLoadBalancer | 13m | service-controller | Deleting load balancer |
70
+
| Normal | DeletedLoadBalancer | 13m | service-controller | Deleted load balancer |
71
+
72
+
<ConvP>
73
+
<Conv name="Cadey" mood="facepalm">
74
+
OOPS!
75
+
</Conv>
76
+
<Conv name="Numa" mood="smug">
77
+
Pro tip if you're ever having trouble waking up, take down production.
78
+
That'll wake you up in [a
79
+
jiffy](https://en.wikipedia.org/wiki/Jiffy_(time))!
80
+
</Conv>
81
+
</ConvP>
82
+
83
+
Thankfully, getting this all back up was easy. All I needed to do was change the `Service` type back to LoadBalancer, wait a second for the cloud to converge, and then change the default DNS target from the old IP address to the new one. [external-dns](https://kubernetes-sigs.github.io/external-dns/latest/) updated everything once I changed the IP it was told to use, and now everything should be back to normal.
84
+
85
+
Well, at least I know how to do that now!
+2
-9
lume/src/styles.css
+2
-9
lume/src/styles.css
···
2
2
@tailwind components;
3
3
@tailwind utilities;
4
4
@import url("https://cdn.xeiaso.net/file/christine-static/static/font/inter/inter.css");
5
-
@import url("https://fonts.googleapis.com/css2?family=Podkova:wght@400..800&display=swap");
5
+
@import url(https://cdn.xeiaso.net/static/css/iosevka/family.css);
6
+
@import url(https://cdn.xeiaso.net/static/css/podkova/family.css);
6
7
7
8
@layer base {
8
-
@font-face {
9
-
font-family: "Podkova";
10
-
font-style: normal;
11
-
font-weight: 400 800;
12
-
font-display: swap;
13
-
src: url("/static/font/Podkova.woff2") format("woff2");
14
-
}
15
-
16
9
a {
17
10
@apply text-link-light-normal hover:text-link-light-hover hover:bg-link-light-hoverBg visited:text-link-light-visited visited:hover:text-link-light-visitedHover visited:hover:bg-link-light-visitedHoverBg underline;
18
11
}