Re: [SSI] Remount and CFS.
Brought to you by:
brucewalker,
rogertsang
From: David B. Z. <dav...@hp...> - 2002-05-20 17:35:34
|
A remount is implemented to perform base linux remounting which allows options to be changed for an existing mount. It does not let you move the CFS server from one node to another. In non-stop cluster for Unixware we had a umount -o remount which allowed you to detach a hard mounted filesystem. This feature would allow moving the CFS server by then mounting again from another node. We don't have hard mounts yet, and the remount code I just checked in was for the base Linux remount a mount option. The current state of CFS is as follows: 1. If you boot a multi-node cluster and you perform a CFS mount all nodes are notified. 2. If you unmount all nodes are notified as you would expect. EBUSY is properly detected on a cluster basis. 3. If you perform a remount all nodes are converted (I've only tried ro -> rw and rw -> ro. EBUSY is detected on a cluster basis. 4. As a test I've booted a root with CFS (single-site only). This required a handcrafted ramdisk image. What is NOT working yet: Nodes don't yet learn about CFS mounts if they join an existing cluster with CFS mounts. Since we don't really boot simultaneously, this has to be implemented to get shared root filesystem with CFS. I haven't tested nodes leaving a cluster with CFS mounts. I haven't implemented cluster-wide file locking. Aneesh Kumar K.V wrote: > Hi, > Ok I have two nodes with same distribution, That means all the system > binaries are same. Now I have a I/O intensive application running on > /home. That is it is doing I/O on the files that reside in /home > partition. I have /home on the master and the master comes up properly.. > Now i get node2 up and it will get all the partitions ( One CFS can be > used for root mount ).Now i want to free the node1 from request for > files in the /usr/ partition. So on node2 i do a remount of /usr . That > means /usr/ is now serviced by node2 and /home by node1. > > Is it possible. ? > > > -- 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." |