Config as code runs too early in the lifecycle

Your checklist for this issue

🚨 Please review the guidelines for contributing to this repository.

• Jenkins version (any / 2.176.2)

• Plugin version (any / 1.29)

• OS (any)

Description

Config as code applies the configuration after InitMilestone.EXTENSIONS_AUGMENTED. However the configuration of Jenkins is loaded after this and before JOBS_LOADED and as such as Config-As-Code will be manipulating state it can and will end up save()ing to disk whilst Jenkins is loading up. This will mean that potential changes that Config-As-Code has made may be lost and you will start up with a Jenkins that is not correctly configured.

see

• JENKINS-58993
• Jenkins reactor
• Config as Code config loader

[timja]

@jtnord there’s a pull request for it but we’re having a weird test failure on CI for java 11 which we can’t reproduce locally 😢

https://github.com/jenkinsci/configuration-as-code-plugin/pull/1262

[jtnord]

looked at the Jenkins code init groovy hook and casc are in different init stages, they should not interfere with each other?

correct.

depending on what configuration you are loosing you may just be hitting this race condition that, or there could be other plugins trying to save some jenkins config during startup. given (mostly) jcasc would be trying to set the configuration that is already on disk on restart I would be surprised you are loosing things (I would expect more a situation where new updates do not get correctly applied - but given the race condition all manner of things could go haywire).

I would suggest you raise a specific issue for your scenario with all of the CasC files and details of the plugins/versions and jenkins version you are using. It may well be that this is the root cause, but it could also be something different (and having a set of files where it does reproduce is extremely helpful. Given timing may be involved if you are running in a public cloud provider providing details of the instance type and disk etc may also help in tracking down the issue (I spent many weeks tracking down a similar issue elsewhere that one person could reliably hit but no one else could reproduce)