[Linux-HA] 3 node cluster ...help
sfarwell at neb.rr.com
Fri Jun 4 14:25:57 MDT 2004
Can anybody help ? I need to have it working by the end of the weekend.........
I know that more than 2 nodes are not officially supported, but I don't have a viable alternate solution so I'm trying to finish getting this solution working.
I'm trying to make a 3 node cluster work and I'm using the same eth0 port for ethernet and heartbeat. This configurations seems to work fine for a two node cluster, it's just the 3rd node that starts causing problems with failback.
Right now using the configuration files below, if I pull the ethernet link from one of the machines, the other two machines BOTH takeover requests for the down machine. For example: If I unplug web1, then requests from the outside world are routed to web3, requests from web3 go to web3, and requests made from web2 go to web2 (not web3). This is all livable.....
The problem is when I plug web1 back in, it comes up and backs up web 2 & 3, but web2 and web3 continue to reroute requests away from web1 the same as before. It's like web1 came up to form it's own private cluster and web2/3 are on their own separate cluster.
The only way to fix it is to reboot all 3 machines.
I'm sure I just have some configuration part oversimplified or left out. I know a 3-node cluster isn't supported "officially", but I haven't found an alternative solution and I have to make sure at least 2 of the 3 nodes are always active, otherwise a full site-to-site (city to city) failover will occur when 2 nodes go down.
FYI: The 2 node cluster version of this seems to be working OK with about a 45 second failback time.
I will be calling a shellscript (or service) on failover. I will also (in the next week or two) be trying to use the bonding driver to link eth0 & eth1 in failover mode so I can have redundant network paths in case a network switch fails. I will then have heartbeat cluster the bonding driver's virtual IP.
I have listed below the ha.cf and haresources files I am using on all 3 machines (exactly duplicated on each machine).
These are the only uncommented lines in the /etc/ha.d/ha.cf file
bcast eth0 # Linux
These are the only uncommented lines in the /etc/ha.d/haresources file.
The only other file I have modified is the authkeys file with the following lines at the bottom :
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the Linux-HA