Ignorable Errors

[DirectXMan12]

We have a couple classes of errors that should not trigger requeues (not-found, already-exists), as they're not errors that will be solved by requeuing (not-found will remain not-found until a cache updates, at which point you should get a requeue from the cache update, for instance, and already-exists is the same way).

The quickest, most-likely-to-not-break-users way to solve this is to have helpers for ignorable errors, along the lines of

func IgnoreNotFound(err error) error {
    if apierrors.IsNotFound(err) {
        return nil
    }
    return err
}

An easier/possibly better solution is to apply that logic on the results of reconciler by default, and allow an option to force overriding this in Result, but this could potentially break people in the case of non-watched/incorrectly-watched resources.

At any rate, the helpers are a good first step -- I find myself writing them a lot by hand.

/kind feature
/priority important-longterm.

[adracus]

Related: Comment in #357 + Fix in #428

[adohe-zz]

/close

[k8s-ci-robot]

@adohe: Closing this issue.

In response to this:

/close

Instructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the kubernetes/test-infra repository.

[DirectXMan12]

Not quite actually fixed -- the helper helps, but I'd like to figure out something automatic

[adohe-zz]

sure I will give a try to find something automatic.

[fejta-bot]

Issues go stale after 90d of inactivity.
Mark the issue as fresh with /remove-lifecycle stale.
Stale issues rot after an additional 30d of inactivity and eventually close.

If this issue is safe to close now please do so with /close.

Send feedback to sig-testing, kubernetes/test-infra and/or fejta.
/lifecycle stale

[DirectXMan12]

/lifecycle frozen

[negz]

https://godoc.org/github.com/crossplaneio/crossplane-runtime/pkg/resource#Ignore

By way of an experience report, the pattern proposed by this issue is exactly what we do in Crossplane and it has been working well for us.

the helper helps, but I'd like to figure out something automatic

While I agree that you pretty much never want to requeue a NotFound error pertaining to the resource under reconciliation I'm a little worried about the existing requeue logic becoming "cleverer" to handle such situations. We've found the existing logic to be one of the biggest foot-guns in controller-runtime, which has lead to most of our reconcilers having large comment blocks at every return so that uninitiated readers can understand our intent around the reconcile.Result, error tuple we're returning in each scenario.

My suspicion is that having reconciler authors explicitly ignore such errors by convention will lead to reconciler logic that is more readable and easier to reason about than if controller-runtime tried to solve this magically.

[DirectXMan12]

I'm a little worried about the existing requeue logic becoming "cleverer" to handle such situations

Yeah, that's my hesitance too, but I've seen a lot of people not realizing that requeuing on not found errors might cause them to see repeated backoff, and it's super-common boiler plate to write if IgnoreNotFound(err) == nil { ... }.

Still, the real experience report is super useful. Thanks 👍

[vincepri]

Closing in favor of #617

/close

[k8s-ci-robot]

@vincepri: Closing this issue.

In response to this:

Closing in favor of #617

/close

Instructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the kubernetes/test-infra repository.