Hello,
I have a 2 client (gc1, gc2) and 2 server (gs1, gs2) cluster setup. Both the
servers have 2 x 1TB HDD in them, gs1 and gs2 are replicated.
With my configuration below, if gs2 goes offline... should I still be able to
have access to the cluster?
volume client1a
type protocol/client
option transport-type tcp/client
option remote-host gs1
option remote-port 7001
option remote-subvolume brick
end-volume
volume client2a
type protocol/client
option transport-type tcp/client
option remote-host gs2
option remote-port 7001
option remote-subvolume brick
end-volume
volume client1b
type protocol/client
option transport-type tcp/client
option remote-host gs1
option remote-port 7002
option remote-subvolume brick
end-volume
volume client2b
type protocol/client
option transport-type tcp/client
option remote-host gs2
option remote-port 7002
option remote-subvolume brick
end-volume
volume afr1
type cluster/replicate
subvolumes client1a client2a
end-volume
volume afr2
type cluster/replicate
subvolumes client1b client2b
end-volume
volume distribute
type cluster/distribute
subvolumes afr1 afr2
end-volume
volume iothreads
type performance/io-threads
option thread-count 8
subvolumes distribute
end-volume
volume readahead
type performance/read-ahead
option page-count 8
subvolumes iothreads
end-volume
On 08/12/2009 10:33 AM, Simon Liang wrote:> I have a 2 client (gc1, gc2) and 2 server (gs1, gs2) cluster setup. Both the servers have 2 x 1TB HDD in them, gs1 and gs2 are replicated. > > With my configuration below, if gs2 goes offline... should I still be able to have access to the cluster? >Yes? :-) I'd suggest try and see. I'm a proponent of the "pull the plug and see" model of testing before deploying. Too many people trust marketing material, and/or trust their own understanding and choice of configuration. It can be a big surprise for people when they are doing database backups, for instance, when the database actually does require a restore, and the restore process does not work. Oops. AFR puts the data on each of the boxes, AFR has code in it to detect and deal with volumes being unavailable, and AFR has "self-heal" capabilities to try to fix the data once the broken nodes are brought back into service. The theory is yes. Try it out and see for yourself as to whether it works for you in practice. :-) Cheers, mark -- Mark Mielke<mark at mielke.cc>