On Tue, 22 May 2012, Dan Frincu wrote:
Best choice would have been to use the actual IP address in the config file, rather than using the network address. This would lead to the same effect (bind on the right IP) without having to modify code. This is also the recommended way of doing things when having overlapping subnets. And, yes, it will always go for the highest numerical IP it finds.
Using the actual IP address in the config doesn't help anything. Corosync still applies each interface address's network mask to see whether the address matches, and 192.168.0.1/24 and 192.168.0.0/24 end up being masked to the same thing.
Also, it's not necessarily the "highest" IP address; it will just be whatever address comes last when enumerated by getifaddrs(). As far as I know, getifaddrs() doesn't guarantee any ordering, so this is potentially non-deterministic. (On Linux, it seems that the "primary" IP address for a particular interface will always be enumerated before its "secondary" IP addresses, however.)
So this problem still stands: there's no way to tell Corosync to bind to a particular IP and ignore any other IPs it might see in the same subnet. And (as my earlier Question 1 mentioned) there seem to be certain conditions when Corosync might re-evaluate which IP to use *whilst* it is running.
totemip_iface_check() could be modified to match on a specific IP if it matches exactly the value from the config file, and only falling back to "any IP in the same subnet" if that IP isn't found. Would a patch along those lines be welcome?
- Michael _______________________________________________ discuss mailing list discuss@xxxxxxxxxxxx http://lists.corosync.org/mailman/listinfo/discuss
