[Linux-cluster] GFS Filesystem check

[Bob Peterson]

On Thu, 2007-12-13 at 23:43 -0600, isplist at logicore.net wrote:
> Other than unmounting the filesystem from each node, is there any other way of 
> doing a quick, regular check from one of the nodes for possible problems? Say 
> a cron job or some other tool which might report potential errors.
> 
> Mike

Hi Mike,

There isn't really a good way to do what you are asking.
There really isn't a way to see if an fsck is needed without,
as you said, unmounting from all nodes and running gfs_fsck.

However, while the file system is running, gfs always does a
bunch of sanity checking as it goes.
The idea is that most kinds of corruption are detected right
away, and cause the file system to be withdrawn.

For example, when most metadata blocks are read, gfs checks
that it's the correct metadata block type it expects.  So
if, for example, an disk inode is somehow overwritten by
something else (e.g. another machine attached to the san outside
of the cluster is writing data to that area of disk without the
knowledge of the cluster), gfs will check that the disk inode
is really a disk inode whenever it is read from disk.  If it
finds something else, it will withdraw.

Someone else may have written a tool I don't know about though...

Bob Peterson