Improve cluster creation success in Korea

[jim-minter]

gcs.prod.monitoring.core.windows.net in Korea Central currently has a bad server which when our internet checker hits it causes our cluster creations to fail irrevocably due to golang/go#36026. Workaround.

[github-actions]

Please rebase pull request.

[hawkowl]

Seems good.

I'm not sure about never raising the error from the internet checker in the logs, though -- it seems like it will always requeue it, forever?

[mjudeikis]

If I read this right, it will try 6 times, and again after 1 hours return reconcile.Result{RequeueAfter: time.Hour, Requeue: true}, err So should be acceptable

[mjudeikis]

Seems good.

I'm not sure about never raising the error from the internet checker in the logs, though -- it seems like it will always requeue it, forever?

Is it not a case that requeue on failure will be logged out by the framework?

[mjudeikis]

As discussed, we might want to change the waiting logic a bit

[olga-mir]

Is this necessary? Can't we use err instead of thisErr? The only real error-generating statement this this function is c.Check(ctx) so one variable should be enough

[mjudeikis]

good questions, why this is not enough?

for _, c := range r.checkers {
		err := c.Check(ctx)
		if err != nil {
			// do all checks even if there is an error
			if err != errRequeue {
				r.log.Errorf("checker %s failed with %v", c.Name(), err)
			}
		}
	}

[jim-minter]

It's subtle. Consider the following case: checker 0 returns an error, and checker 1 returns nil. In that case by your code we would return foo, nil when we exit the function, but in fact we want to return foo, err.

[nilsanderselde]

good explanation