Linus Nordberg wrote:
> Ok. There are other reasons for renaming and separating the three
> timeouts as well. One is that these are low level timeouts needed for
> servers and proxies while clients using the struct request object of
> libradsec probably would like a single timeout spanning the whole
> connect + send + receive sequence. The 'timeout' keyword in a 'server'
> stanza should probably be reserved for such a high level timer.
Proxies are just servers glued onto clients.
The only timeout a server has is a lifetime for processing the
request. Servers don't need to do anything else, as all of the other
timers are driven by the client retransmits.
> I haven't touched on retransmissions when on UDP yet. I should've made
> that clear from the beginning. My hopes are that the library can handle
> retransmission without the need for the application to know about it.
> We'll see how this works out when I come to UDP.
If it's blocking, then the library can handle retransmissions.
Otherwise, some event callback needs to be defined.
Alan DeKok.
|