|
On Tue, 2007-05-08 at 17:12 +0900, Simon Horman wrote:
> How are you starting the sync daemon? It sounds like however you
> are doing it, its only starting one. I recomend just starting
> both deamons up using an (ipvsadm?) init script.
Heartbeat is starting it up (supposedly). Let me just double check to
make sure I'm not going insane with Symptom 1. Nope. On a fresh reboot,
[ipvs_syncbackup] doesn't start up, even after a connection comes in.
ipvsadmin -lnc shows the connection on the master, but not on the
backup. This part isn't really much of an issue, because I can get it to
start up, and can kludge a fix if heartbeat won't start it up, but I
figured that, because it has to do with the connection syncing, it may
be part of the greater problem.
The main problem, really, is the failover issue. Connections to the
master node's service fail when I shutdown hearbeat, but connections to
the backup's service stay up when I fail the master. This applies in
either direction, so if I connect to what was the backup, now the master
while the normal master is down, and it directs me to it's own service,
then I start the normal master back up, the connection is dropped. But
if I was directed to the normal master's service (which is still up, as
I just failed the heartbeat) the connection stays up.
mcast traffic seems to send out the connection as it comes in. Watching
tcpdump shows a mcast packet exactly when I connect to the master. The
entry is made into the connection table, with the time discrepancies I
described earlier. Any ideas to help me track this down? Thanks!
--
Sal Tepedino <stepedino@xxxxxxxxxxxxxx>
|
