24 lines
1.3 KiB
Markdown
24 lines
1.3 KiB
Markdown
|
---
|
||
|
|
title: Fixing a redis chart
|
||
|
|
date: 2022-03-14
|
||
|
|
description: I contributed a preStop script to trigger a failover before a redis master is terminated
|
||
|
|
tags:
|
||
|
|
- helm
|
||
|
|
- kubernetes
|
||
|
|
- redis
|
||
|
|
---
|
||
|
|
|
||
|
|
## Introduction
|
||
|
|
|
||
|
|
Back in January I contributed a fix to a redis chart : https://github.com/DandyDeveloper/charts/pull/178. I should have blogged about it at the time, but life happened and I did not.
|
||
|
|
|
||
|
|
## The problem
|
||
|
|
|
||
|
|
This charts deploys a redis cluster and I noticed that when a redis master gets terminated there is a downtime before a new one gets elected.
|
||
|
|
|
||
|
|
One of the challenges in fixing this behaviour was to do it in a way that would get accepted upstream. That meant respecting the style in place to manage the chart : scripts are deployed in the images as configmap values and are written inside a helm template... yummy! This allows to use helm variables easily, and avoids maintaining custom images but made linting or shellcheck validation a pain to do.
|
||
|
|
|
||
|
|
## The solution
|
||
|
|
|
||
|
|
This restart behaviour can be fixed with a preStop script that checks if the terminating pod is a redis master. If that is the case, we trigger a failover before shutting then wait for it to complete. My fix was accepted and merged upstream, I am proud I had the opportunity to contribute back to this chart in the community spirit of open source software.
|