Server selection by weight flawed?

From: Chris Elsworth <>
Date: Wed, 21 Jun 2006 14:08:09 +0100


I think I've found a logic problem in the pen code when you're using
server selection by weight.

I noticed that when using weighted servers, after a server temporarily
died and came back, it wasn't used again.

If I'm following the code properly this appears to be because
try_server() is the function that marks servers as 'up' again, once
it's successfully connected, but server_by_weight() will never return
a broken server because of the status != 0 check. So broken servers
are never retested.

Am I reading that right?

Received on Wed Jun 21 2006 - 15:32:55 CEST

This archive was generated by hypermail 2.2.0 : Wed Jun 21 2006 - 15:32:57 CEST