[Added ha-clusters-discuss to this thread] Sumit Gupta wrote: > Michael Greenberg wrote: >> Hello again, >> Imagine this scenario: I have two servers connected to the same storage >> array. I got a zpool on the array, and one of the servers "shares" the pool >> to the FC fabric using COMSTAR. >> >> If one machine fails, I imagine the pool can fail over to the other machine >> with Sun Cluster (please bear with my Sun Cluster ignorance - We're a >> VERITAS shop). >> > > This is something which has been discussed internally but there are > a lot of details to work out before a fully automated failover between > target nodes can be supported. But I believe it is possible to do so > manually. You need to create to identical configurations of the target > side using the same zpool backend (of course at any given time only one > of the machines can have zpool imported). Keep the stmf service offline > on one of the target and keep it online on the other target (the one > with the zpool). To do the failover, offline the stmf service on the > original target machine, move your zpools to the other target mode > machine and online the stmf service on it.
To add, one way you could automate the above config using Sun Cluster would be to create one cluster resource that moves the zpools from one machine to the other and another resource that is a SUNW.Proxy_SMF_failover of the stmf service. The SUNW.Proxy_SMF_failover resource will ensure that the stmf service is running on only one node at a time. The SMF resource can have the dependency on the zpool resource so that the stmf service waits for the zpool to come online during failover. Augustus. > >> But here what I wanted to ask: >> 1. Can a configuration be created on the cluster for the stmf and sbd >> services to fail over the same way the pool does? >> 2. Let's say I got the COMSTAR service configuration syncronized between my >> target machines, can it fail over seamlessly to the nodes using the LUNs? >> > > I think the target will failover but the initiator has to retry to > get the I/O going. > >> 3. What will happen to the WWNs of the LUNs? >> > > The target PWWNs will change. In a multipath configuration this > should not be an issue. > >> 4. Or will I have to issue devfsadm to all initiator nodes in the fabric >> once the other machine takes over the pool :( ? >> > > If a FC switch is involved, online of stmf service will cause RSCNs > to go to the initiators. This should force a rediscovery. But I have > seen bugs on the initiators where they will drop RSCNs. In which case > some sort of a rediscovery action will be needed. > > Sumit > >> Any thoughts? Ideas? Technical facts? >> It's just that I don't know well enough how COMSTAR works internally. >> >> Thanks. >> -- >> This message posted from opensolaris.org >> _______________________________________________ >> storage-discuss mailing list >> [email protected] >> http://mail.opensolaris.org/mailman/listinfo/storage-discuss >> > > _______________________________________________ > storage-discuss mailing list > [email protected] > http://mail.opensolaris.org/mailman/listinfo/storage-discuss -- Augustus Franklin, Solaris Cluster http://blogs.sun.com/augustus _______________________________________________ storage-discuss mailing list [email protected] http://mail.opensolaris.org/mailman/listinfo/storage-discuss
