Re: OPS and raw filesystems

> From my class notes, "Media recovery uses (archived) redo log
> files from all threads".
 

> I take this to mean that should a disk device fail,
> the recovering instance will need access to all threads'
> redo logs.
 

> Therefore, it would need shared access to redo log files.
> In the SUN PDB case, that means a raw device for all
> instances to have access.
 

> From the Oracle7 Parallel Server Concepts and Admin guide,
> page 5-7 "Log files ... Although the redo log files are
> independent per instance, each of the log files must still
> be accessible by all the instances so that recovery is provided
> for."
 

> However, in our environment we will be temporarily storing
> our archived log files in a file system on each node.
 

> Since we can't share file systems across PDB cluster nodes (yet)
> we would have to make the failed thread's archives available
> to the recovering node (FTP them, import the file system, etc.)

that's assuming that the machine with the failed instance is up for you to ftp or import from. there are a few alternatives to consider:

o make sure the filesystem for your archive threads are on the shared   SparcArray, not the individual node's local disk. If they are on   a shared volume, you can mounted the failed instance's archive FS   on to the surviving instance's node, if the failed instance's node is   down (ie, not holding an exclusive filesysystem mount on the archive FS).   If they are on the local disk of the failed machine, you are pretty much   hosed unless you can reboot the machine or reconnect some disk cables.

o set up a separate NFS server node whose job is to act as the fileserver   for the PDB nodes' archive files. Thus, each node would NFS mount both   archive thread FS. The caveat here is the network traffic you will introduce   and the bandwidth issues of dealing with a very high redo switch rate.

Johnny Chan
Independent Oracle Specialist Received on Fri Mar 21 1997 - 00:00:00 CST