|
If I understand you correctly, you want the RDP connections to survive a
failover situation. No matter how quick your heartbeat failover happens,
all users will still get disconnected, since the backup node doesn't
know anything about the connections that were passing through the master
node.
The only way around this is to use the sync-daemon support in ipvs so
that connections through the master are synced to the backup node, which
upon failover can continue handling the existing connections.
If I remember correctly though the fallback to the master could still
disconnect a lot (all) users, since it's connection state is at best
outdated but more likely flushed. So it is more sensible to make the
backup node master in case of failover and the master node should return
as a backup node.
Regards,
justin....
Leon Keijser wrote:
> Hi,
>
>
> Maybe a bit unusual question, but i have to ask it. Today we tested the
> LVS's in production and there was a problem with (we think) the timeout
> values for heartbeat. Upon failover, everybody connected using RDP
> disconnected. Now, these are the values i have in ha.cf :
>
> keepalive 1
> deadtime 30
> warntime 10
> initdead 120
>
> These are the default values, i think. So after the massive disconnect (140
> users) we changed it to:
>
> keepalive 10ms
> deadtime 300ms
> warntime 100ms
> initdead 120
>
> But how do i make these changes active, without another massive failure? I
> was thinking about:
>
> - shutdown heartbeat on the inactive (backup) node
> - change ha.cf
> - start heartbeat again on backup node
> - stop heartbeat on active (master) node
> * failover occurs to backup node
> - change ha.cf
> - start heartbeat again on master node
> * fallback occurs and master node takes over
>
> Unfortunately, this didn't go as smooth as i thought. First of all because
> the timeout values were too low (the LVS's did a failover every couple of
> seconds!) so new clients couldn't connect and the ones that were connected
> couldn't do much work because the connection stalled all the time. Second,
> i'm not sure if this was the right way of doing this.
>
> We've set the values back to the default ones (k:1, w:30, d:10, i:120)
> again, and by start/stopping heartbeat, killing processes and praying alot,
> things worked again. But we still have to configure the heartbeat timeout
> values though. I'm going to clone these 2 LVS's so we can experiment with
> different values until we find the right one. After that i will want to
> change the production LVS's too. How can i accomplish this?
>
> Thanks in advance,
>
>
> Léon
>
|
