[Gluster-devel] Bug Reports

[Gordan Bobic]

On Tue, 10 Feb 2009 15:23:00 +0530, Krishna Srinivas
<krishna at zresearch.com>
wrote:
>> Is there a way to specify --disable-direct-io-mode in fstab?
> 
> 
>  volume fuse
>    type mount/fuse
>    option direct-io-mode 1
>    option entry-timeout 1
>    option attr-timeout 1
>    option mount-point /mnt/storage
>    subvolumes unify
>  end-volume
> 
> you can give the options to fuse in the vol file.

So, --disable-direct-io-mode is equivalent to "option direct-io-mode 1"?
Shouldn't it be 0? I noticed there is a "direct-io-mode" parameter
understood by mount.glusterfs, but this appears to be for enabling, rather
than disabling it. Can anyone clarify?

I was also hoping that --disable-direct-io-mode might also fix the curses
issue with the kernel and the nVidia installer, but unfortunately, it
didn't. :(

>> OK, just managed to clear this weird condition and return things to some
>> appearance of normality. Stopping all the servers and explicitly
deleting
>> the offending symling/directory (it seems to have splitbrained - some
>> servers had the symling (which wasn't viewable!) and others had the
>> directory (which was also not viewable). Deleting them all from the
>> backing
>> store, and remounting seems to have fixed the problem. Really weird how
>> it split-brained while all the servers were up, though. :-/
> 
> It should not happened. If you get any clues on the reproducing steps
> it will help us.

I suspect you could reproduce the second stage by making a directory on one
server (in the backing store) and a symlink on the other. This will make
the entry non-viewable on both nodes, but you'll still be able to cd into
it. I've no idea how the initial splitbrain might be re-creatable. Maybe
creating a dot-prefixed symling pointing to a directory outside glusterfs
might do it? I'll do some more experimenting tonight and see if I can
reliably re-create the problem.

Gordan