Doug Fawley

> Yeah, that definitely seems like a bug in the child policy, even completely independent of xDS. I think the child policy should report TF and its Update() method should...

> The way we're handling this is: I was expecting this needs to be handled by some custom logic in clusterresolver that detects when there are no addresses and substitutes...

> I think the child policy should report TF Oops, I guess I misread this part the first time. I don't think the child policy is supposed to report TF...

> Where did you run into this? And what exact error were you seeing? It's a theoretical case that I quickly modified the existing `clusterresolver` e2e test of `TestEDS_EndpointsHealth` to...

Also, an update: we may actually want to change the expected behavior of some of these things. I'll follow up here when I know move.

It looks like my understanding of our PF implementation was wrong, and that (1) is false. We _will_ accept a zero-address update from the resolver and apply it, making future...

It seems like we should probably be doing something here to help. Reopening as we discuss internally what our options are.

@peterchenadded we already cut the release branch for 1.74, which is happening next week. So this would make it into 1.75 which would release on/around Aug 19th.

We are doing this (and the server-side RST_STREAM) but I don't believe we have tests for it -- I know we don't have tests for server-side, at least.

(Sorry, hit send too soon; updated comment above.)