I am posting this here in addition to a ticket on Mantis, because the ticket causes Mantis to crash ... http://bugs.unrealircd.org/view.php?id=3973.
Our network has 3 hubs, all running 126.96.36.199, with IPv6, on Linux (2 of them) and FreeBSD. The code has not been tinkered with, and we are using a few 3rd party modules ( delaylist, regexcept, quarantine, nocodes ).
There are links set to autoconnect, and this is working fine after a server start or restart. However, in time, it seems that the autoconnect feature is not kicking in after a split anymore.
This has happened with 2 different hubs about half a dozen times already: the hub that's splitted apart just remains on its own and does not even try to restore its main, autoconnect link, meaning there are no connection attempts at all (we've double checked the logs and the notices). An oper has to do it manually every time. We've seen 12-hour splits, so it's not about reconnecting too slowly or having a too low value for connfreq/pingfreq (30/90 in our case), definitely about not reconnecting at all.
The only clue I've found so far, indicating a possible synch problem, is this:
Code: Select all
[Wed Dec 1 00:19:13 2010] - WARNING: Time jumped ~23 seconds ahead! (1291159130 -> 1291159153) [Wed Dec 1 00:19:13 2010] - [TimeShift] Resetting some timers!
The other possibility is that somehow, Llinux GrSec might interfere with the way unrealircd is timing its autoconnect attempts, but I have no had evidence (I'll have to check the code).
What else can I do ?
Thanks for your help,