The calltimeout setting sometimes does not take effect
The calltimeout setting sometimes does not take effect. For example, a 10s calltimeout is set, but sometimes it does not close after 10s, and finally closes around 1000s.
Yikes; that’s severe. I wonder if this is related to full write buffers on the outbound TCP socket. I believe that can be a problem!
@swankjesse What's the theory or observable way to reproduce or fake? A delegating Socket that blocks writes?
Yeah, a server that doesn’t read anything should do it.
(That’ll only catch a broken write timeout. A broken call timeout should still work fine.)