Mark D. Roth
Mark D. Roth
The ADS call is wait_for_ready, so if the channel goes into TF before it becomes READY, the ADS call may not fail for a long time. We want a more...
1. I think we should start from index 0, and perform the usual RR algorithm from there. That usual algorithm is basically "while subchannel[i] is not READY, ++i; return subchannel[i]"....
> For (3), if we always start with 0 it seems it is baaaad for a flapping task causing the name resolver to add/remove the entry. That's a good point....
> MAX_CONNECTION_AGE and MAX_CONNECTION_IDLE don't make sense in rr I don't think that's true. In fact, you're the one who has advocated for using this approach to allow round_robin clients...
I can't comment on the question of which metadata k8s *wants* to provide here, but I do want to make a quick comment on how that metadata could be encoded....
Agreed, we still need to improve this.
I don't think we have any sanity checks for this today. I agree that we should fix that, but I don't think this is a very high priority. If all...
I've added a variant of `LoadFromJson()` that handles the case where the object is `RefCounted`, which eliminates the need for all of the LB configs to be movable. We'll see...
FWIW, note that the additions in the various `JsonLoader()` and `JsonPostLoad()` functions are actually smaller than the original code removed, which is what I would expect. Here are the additions,...
As per our discussion, there's no need for this gRFC if the C++ APIs are only experimental at this point. It's certainly a good idea for users to comment on...