Cluster mgmt of off-network (multihomed) addresses
I'm currently in the process of configuring an HA cluster to replace a single machine. The original machine had virtual interfaces plumbed for several different networks, which remote sites connect to for FTP and other data. I'd like to know how to do the same using the cluster (3.1 on Solaris 9 atm).
Part of the problem is when I attempt to define the Logical Hostname resource, I get the error "specified hostname(s) cannot be hosted by any adapter on xxx". I tried configuring a different (permanent) address on both cluster nodes that was on the same network, which allowed the initial configuration; however, when the resource was stopped, it partially unconfigured the permanent interface and brough the entire cluster to its knees.
Does anyone know how I can do this? In detail, I have interface ce3 with a network like 192.1.110, and virtual addresses that need to be failover resources in the cluster on 192.1.33, 192.1.34, etc.
[960 byte] By [
pochena] at [2007-11-27 10:44:46]
# 1
Hi,
I had run into a similar problem recently with me single node demo laptop cluster. THe solution - thanks to Thorsten - was to configure an address of the same subnet, that the logical hosts are on on the adapter. Configuring meant, add to /etc/hosts, ifconfig the address on the adapter and put the address into the ipmp group.
I have to check whether there is a regression in 3.2 as I am pretty sure I did this with earlier versions much easier. Let me try this with my "static" cluster in my lab. Will take a couple of days as I am travelling currently.
Regards
Hartmut
# 2
Thanks, I'd appreciate that. I think I may have done something similar. I tried plumbing a logical interface (ce3:6, I think it was) and configuring a fixed IP on that interface. I did not add the address to /etc/hosts and it seems I hadn't added it to DNS either. Also, I didn't put it in the IPMP group. IPMP is still a bit of a mystery to me, though I have been doing some reading.
FWIW, I'm still using 3.1, not 3.2.
# 3
I've managed to succeed in getting the multiple networks managed by the cluster, it appears. Essentially, I did an ifconfig addif to add the network, and explicitly set an appropriate netmask. I also put an entry in /etc/hosts, though I haven't done anything with IPMP. The cluster in question doesn't have multiple public interfaces (two private, one sunray, one public).
I suspect the netmask may have been the cause of my earlier troubles. Since we didn't have an entry for these networks in /etc/netmasks, it probably defaulted to 255.0.0.0, somehow making it such that when the cluster software unconfigured the managed address, it somehow partially unconfigured the permanent address and made things go pretty nuts.
I'm still interested in hearing what you've done to get it to work, however, in case I've missed something.
# 4
I finally had time to do some testing. I've found a method to make it work. I admit this is only a workaround and I'll file a bug to get this fixed.
This - long form of the scrgadm -L command does the job:
2 addresses: physical address is in the 10.x.x.x network, hatest, the new LH is in the 11.x.x.x. network:
scrgadm -a -j hatest -t SUNW.LogicalHostname -g demo-rg -x HostnameList=hatest -x NetIfList=sc_ipmp0@1,sc_ipmp0@2
This adds LH hatest to my demo-rg. Using the -L version of scrgadm did not work. The same problem exists with the new clrslh command on SC3.2
Regards
Hartmut
