2 Mar
2006
2 Mar
'06
10:59 a.m.
Juha Heinanen wrote:
how does passive LB know that active LB is not anymore
functional? do
you assume a third monitor process that sends SIP requests through
active LB and if not successful, make passive active or do LBs somehow
watch each other?
We currently use heartbeat for LB failover, and the passive host routes
OPTION requests (one per second) through the active one using sipsak and
a selfmade responder. If the host itself or the uplink is down,
heartbeat takes over automatically, and if three consecutive OPTION
requests are lost, hb_takeover is executed on the standby node. LBs also
have no local harddisk, but use PXE-boot and two NICs in active-backup
bonding-mode on two seperate switches to minimize sources of failures.
The setup is quite experimental right now, but especially LB failover
undergoes intensive testing right now, since it's the most critical part
in the system.
Andy