On Thu, 29 Aug 2002, Yaroslav Klyukin wrote: > > Why do you say "mcast_if" does not work? > > Let me explain what i mean by "does not work" > > For example: > eth0 - local network with ganglia installed. > eth1 - external interface and default route. > > When gmond is started, it uses default route to configure multicast, > even if you specify mcast_if to be eth0. > If i am not wrong, it works perfectly fine if you swap eth0 to eth1 like: > eth1 - local network with ganglia installed. > eth0 - external interface and default route. > > So it looks like a bug. > My approach is to manually specify multicast route to ganglia, rather > then let it autodetect - this is a known issue, and it was documented on > the web site. I do not insist that my way is the best, but it works.
Another approach is to route your multicast packets "correctly". I added the following line to my /etc/sysconfig/static-routes file: eth0 net 224.0.0.0 netmask 240.0.0.0 dev where eth0 is my internal network interface. This resolved the problem. The problem I ran into with the "mcast_if" option, the external hostname was used for the multihomed machine in the ganglia database. This also caused all of the internal nodes to complain about not being able to reach the external internal address. Both approaches are valid, but routing the muilticast packets gets the naming right, and removes a lot of log messages from your client nodes. Randy Philipp
