Matt Cary

We also want to backoff if the detach fails due to the node not existing or something like that, right? I think that the assumption the sanity tests make that...

@judemars in that test you reference, it looks like it only fails at the end of the It()?

@pwschuurman There's only one quota for all api calls, so if we wait until the detach calls hit quota, we'll have other operations unnecessarily fail. Also given the problems we...

@judemars Making them rpc-specific might be a solution, but since quotas are counted over all RPCs I think we'd get complicated quickly trying to track stuff. We've already had churn...

@judemars Oh, interesting, I didn't realize it was the cleanup function that was failing. In many of the other tests----the GKE guitar ones in particular---we make errors in the cleanup...

Fair point. If we retried on the cleanup, would that fix it?

That could be a good idea. Since the backoff appears to break the sanity test's assumption of how the driver works, maybe it's okay to set the threshold very low....

k8s.gcr.io is going away in the 1.27-1.28 timeframe. [https://groups.google.com/a/kubernetes.io/d/msgid/dev/CAAusGiMGAC1XSr2VejrBXtB20KmvnguUR12WQ1zbKX1-uCizag%40mail.gmail.com](https://groups.google.com/a/kubernetes.io/d/msgid/dev/CAAusGiMGAC1XSr2VejrBXtB20KmvnguUR12WQ1zbKX1-uCizag%40mail.gmail.com?utm_medium=email&utm_source=footer)

/lifecycle frozen

@arkadeepsen @bharath-b-rh /cc @msau42 Do we think that adding driver command line flags is the right way to go? Those tend to be hard to configure as they're typically set...