>I don't really see how this approach is covering all cases. First, a very
slow
>connect would also need to be able to get cancelled, or even if the connect
is
>done fine and then just the following TCP packages take a hundred years...
Or
>am I wrong?
[anlin] The tcp keepalive option I added on the soket will take care of any
non-responding machine. If the server machine's tcp stack doesn't response
the keepalive message, then the local socket will be shutdown. However, tcp
keepalive message is tcp layer message, it does not cover the case if server
machine's tcp stack works fine but the name server is slow. That shall
already be covered by ares' timeout feature.
BTW, how do I directly answer your posting on the web? so that it looks like
a cascade follow-up?
Received on 2007-08-09