Fallback when gc job fails

[zifeo]

If something bad happens within https://github.com/enix/kube-image-keeper/blob/main/helm/kube-image-keeper/templates/garbage-collection-cron-job.yaml, the registry might end up locked forever. A fallback scenario should be provided.

Additionally, the job is using bitnami/kubectl which should maybe also be cached but kuik seems to ignore its namespace.

[Nicolasgouze]

Hi @zifeo,

Thanks for your feedbacks !

Regarding your 1st comment :
We are aware of this limitation & already worked the best we could on this topic. :(
See : #209.
And additional info here : https://github.com/enix/kube-image-keeper?tab=readme-ov-file#garbage-collection-and-limitations

I'm aware that maybe it will be "enough" for you but maybe not (we know we'll not cover 100% with this PR).
A couple of months back, we "gave a round" and tried to find a way to do better, potentially with an other registry but did not coverge. We 'll try again in the future !

Regarding the 2nd comment :
Yes, kuik, by default, ignores its own namespace and this is "by design" / a choice we made.
For kuik v2, we plan to facilitate to pre-populate (and also easily let the users add the ones they like) toolbox images inside the tool. Maybe in line with your suggestion ...

[zifeo]

@Nicolasgouze thanks for getting back. I might be oversimplifying: would not a try/catch-like + timeout managed those rare cases? I would especially expect the job to fail in those circumstances (which then can be taken care manually).

For the images, wouldn't mirror kubectl using ghcr be a simple solutions to still avoid the docker limitations?