Re: [SSI-users] failover testing of "/"
Brought to you by:
brucewalker,
rogertsang
From: David B. Z. <dav...@hp...> - 2003-04-08 21:07:26
|
See comments below. Goetzman, Dan wrote: > David, > > Just trying to understand how all this SSI stuff works... > > What if I had 3 or 4 nodes (all direct attached to storage)? > Is there any connection to the CLMS master node and the node that is the > current CFS server for a given filesystem? / will be on the CLMS master node, since it is remounted in the kernel. > Could / be CFS served by node1, /usr by node2, /var by node3 (just for > example)? Other filesystems could be anywhere. However, some kind of configuration file (not yet designed), would have to indicate the order of server nodes for non-root hard mounted filesystems. > How can I tell the current CFS server for a given filesystem? "cat /proc/mounts" will indicate a device name if it is run on the server of that filesystem. A number indicates its ssidev for a remote filesystem. # onall cat /proc/mounts (node 1) rootfs / rootfs rw 0 0 /dev/sdb1 / cfs rw 0 0 <--- server none /cluster/nodenum1/dev cfs rw 0 0 none /dev cfs rw 0 0 /proc /proc proc rw 0 0 /dev/sda1 /introot cfs rw 0 0 <---- server 00000034 /cluster/nodenum2/dev cfs rw 0 0 00000034 /dev cfs rw 0 0 (node 2) rootfs / rootfs rw 0 0 00000021 / cfs rw 0 0 <---- remote 0000002a /cluster/nodenum1/dev cfs rw 0 0 0000002a /dev cfs rw 0 0 none /cluster/nodenum2/dev cfs rw 0 0 none /dev cfs rw 0 0 0000002e /introot cfs rw 0 0 <---- remote /proc /proc proc rw 0 0 > > I guess that all this has to be figured out (automatically) for the loss of > a node to be transparent? Of course, but it is for future design and implementation to provide the "automatic" feature. For a simple two node cluster, /etc/rc.d/rc.nodedown could be made to remount filesystems by code by hand. > > Dan -- David B. Zafman | Hewlett-Packard Company Linux Kernel Developer | Open SSI Clustering Project mailto:dav...@hp... | http://www.hp.com "Thus spake the master programmer: When you have learned to snatch the error code from the trap frame, it will be time for you to leave." |