Set default observedGeneration to -1

[sbernheim]

Sets a default value of -1 for the observedGeneration field of the ImageUpdateAutomations type status.observedGeneration attribute. This ensures that tools like kstatus do not consider the resource to be in a Ready state prematurely because the generation and observedGeneration attributes are briefly initialized with 0 values.

• Implement defaulting of observedGeneration to -1 in all controllers flux2#2007
• Set default observedGeneration to -1 #275

Credits to @tomhuang12 for his prior work on fluxcd/image-reflector-controller#189

[sbernheim]

Attempting to correct for the failed check now.

[stefanprodan]

@sbernheim you shouldn’t edit edit the manifests manually as they are generated from code. Please run make manifests.

[sbernheim]

l had to do a bunch of things to get the tests to run.

The tests don't seem to want to work without Docker which I was trying to replace with containerd and podman. In the end I gave up on figuring out how to make that work and reinstalled Docker for Desktop on my MacBook again.

I also had to add an .envrc file with this env var export:

export PKG_CONFIG_PATH="/usr/local/Cellar/openssl@1.1/1.1.1m/lib/pkgconfig"

That directory is where brew install openssl@1.1 put the openssl.pc file that needs to be linked for the tests.

Now of course the tests are failing but with a different error message that the GitHub checks which indicates that the default observedGeneration value is supposed to be an integer and not a string. The make manifests command generates the CRD file with the default value set to a string "-1" instead of an integer -1 and I'm not sure how to fix that.

All the long string values in the generated CRD's description fields are also wrapped at 90 characters, which seems like it'd be a linter setting.

However, the same tests fail on my local and the generated CRD has the same problem when I run it against the main branch for the image-reflection-controller project. So I have some sort of version conflict between my environment and yours - probably the wrong kubebuilder version installed on my local. I installed the latest today while setting all this up, and more likely need to drop back to an older version.

I'll dig a bit more and try to resolve these issues over the next week or so.

[sbernheim]

Found an open issue for the line-wrapping problem, which has a suggested mitigation in the most recent comment kubernetes-sigs/controller-tools#514

Putting the link here so I don't forget where I saw that later when I have time to try again.

[sbernheim]

Pinning controller-gen to v0.5.0 seems to have resolved the problem of the default value being set as a string in the generated CRD.

go get sigs.k8s.io/controller-tools/cmd/controller-gen@v0.5.0

It's still wrapping all the long text lines at 90 chars, though. I pushed what I have to see if the tests pass, but obviously I'd rather not change all the description fields like this. I'll have to circle back and figure out how to fix that another day.

[makkes]

@sbernheim I created a commit on my own fork that should fix CI: makkes@28fc2cb

Feel free to cherry-pick it into this PR's branch.

[sbernheim]

@makkes - Thanks for that! I cherry-picked your PR into my branch, but the "dirty tree" check still failed!

It appears that @relu merged a commit to main while updating the Flux pkg components that wrapped text lines at 90 characters that match the ones I'm generating on my local.

Took me a while to figure out why that last round of testing was failing. At first I thought I must be reading the diff backwards and had somehow dropped your commit again after merging it to my branch. I figured out what had happened once I took a little dive into the blame history for the generated CRD file.

Hopefully the tests will succeed this time and we can finally put this one to bed. All's well that ends well AFAIC!

[relu]

LGTM, thank you @sbernheim! ✔️

[sbernheim]

🎉 !!!