> The message exchange timeout is not configurable but it's also 2mn, which is > fairly conservative. At this point I'm wondering if timing out is not the > correct behavior. Maybe there's a use case that I'm not seeing but beside > pushing load testing at its max, I can't see when you'd want to wait for > more than 2mn for a non-reliable transport over a request/response exchange. > Don't you think?
In terms of semantics, I am not sure if at this point timing out is the correct behavior either in this situation. But I can see that as a possible approach for sure. I think some of it goes to where you want to put the fault tolerance for excessive delayed response times. I guess it is true that excessive load might be the only case when we need longer timeouts. I guess otherwise clients can also use async response as a way for longer response times or variable response times from individual services. My concern was largely that with slightly complex workflows I might start seeing this even with lesser concurrent clients but I can probably experiment to find out if this is true. And I can probably tweak the value in code directly for when I am doing experiments with larger number of clients for now. thanks! Lavanya
