Yutaka Sato writes...
> . . .
> |is connecting to). I've noticed that the long-lived cc processes can
> |potentially have kept a connection open for very long periods of time,
> |at least hours, potentially forever.
>It's the very situation I inteded to realize by "connection cache"
>where a single connection to the server is reused repeatedly by
>multiple clients, to reduce the latency to establish connection
>to the server and reduce load on the server host for repetitive
>initializations of NNTP sessions, which is typically caused when
>the server is accessed by a NNTP/HTTP gateway.
Unfortunatly such long, potentially unlimited connections can
expose defects in the server's resource management. Memory and
resource leaks in things like NNTP servers often go unfixed
because the process lifetime is so short that they never become
a problem. In addition, dynamic load balancing and redundancy
schemes also rely on a finite lifetime of connections so that
load can be shifted to faster (or operational) hosts.
It would be prudent to introduce a maximum lifetime, either in
transactions or time, of the connection cache processes.