avoid saving the azuremachine CR too often #1367
Conversation
|
Are you sure this is triggering new reconciliation loops? We were checking a case like this before, and it should not happen, as update should basically be a noop if there were no changes. If it is triggering new reconciliation loops without CR being actually changed, it might be a bug in operatorkit cc @giantswarm/sig-operator |
I might be wrong about operatorkit bug, but we did have few discussions about the topic and it came down to these updates without actual changes not having any effect. |
100% sure I can provide CRs and logs |
|
There might be some issue in the func called few lines above then: https://github.com/giantswarm/azure-operator/blob/master/pkg/helpers/azuremachine.go#L16 |
BTW I copied what you did in another place where you use the same function |
|
Touche :) Forgot now why I did that, it wasn't for avoiding reconciliation loops though. My only concern here is that there might be an issue/bug somewhere in code that is causing infinite reconciliation loops (maybe caused by me in that func 😬), and this might hide it temporarily. |
Like I possibly did with that other "changed" check. |
I'm not sure right now but I think that's the only piece of code where we actually edit the Azure machine cr. If that's true it has to be it causing the reconciliation madness |
| if changed { | ||
| err = r.ctrlClient.Update(ctx, &azureMachine) | ||
| if errors.IsConflict(err) { | ||
| r.logger.Debugf(ctx, "conflict trying to save object in k8s API concurrently", "stack", microerror.JSON(microerror.Mask(err))) |
This handler was calling
Updateon theAzureMachineCR at every reconciliation loop, regardless of any changes being made to the CR.This triggered a loop in the handler that was running in an infinite sequence.
This PR check for changes in the CR and avoid saving if there isn't any