Subject: Re: [vserver] sharing /vserver
From: Herbert Poetzl <herbert@13thfloor.at>
Date: Thu, 4 Sep 2008 18:50:49 +0200

On Thu, Sep 04, 2008 at 11:00:23PM +1000, Matt Paine wrote:
> Hi Herbert
> 
> <snip>
> >>a glusterfs one ?
> >
> >I doubt that glusterfs is Linux-VServer aware :)
> >
> >best,
> >Herbert
> 
> Can I ask what glusterfs would need to be able to Linux-VServer aware?

yep, you want to add the following things:

 - xattr barrier flag (required)
 - xattr ixunlink flag (for unification and CoW link breaking)
 - internal tagging (not really required)
 - disk space accounting (on shared partition)

> I have been away from both (linux-vserver and glusterfs) projects for a 
> while, but I like to keep an eye on things to see whats happening. 
> GlusterFS is a project that I see to be an integral part in a project 
> that I am about to experiment with at home. Linux-VServer is also a part 
> of that project so as soon as I get some spare time I am going to setup 
> a cluster to hopefully run vserver guests on top of a GlusterFS so I can 
> essentially have hot spares on any glusterfs aware system.
> 
> I know that GlusterFS uses extended attributes to keep track of 

extended attributes != xattrs

> versioning. I also know that vserver extends the linux kernel to add 
> further attributes on files. GlusterFS also intends to be fully posix 
> complient. I would like to present the developers with some 
> questions/information on what would make GlusterFS work "out of the box" 
> (quote from another email) with Linux-VServer. 

I doubt that this is a good idea, because the feature above
would break a non Linux-VServer setup

but, if they are interested, it should be easy to prepare
a patch to make it work with (and support) Linux-VServer
features ...

> With my limited understanding of file systems I may assume (i know I
> shouldn't) that perhaps if the kernel that the GlusterFS server is
> running is also patched with the relevant linux-vserver file system
> attributes, it should work correctly.

could be, if the architecture is similar to NFS, this is
definitely required, if it is more like GFS, then it is
kind of optional, but usually you want to use similar
kernels on all nodes ...

> I'm probably completely wrong though :)
> 
> I am happy to liase with the GlusterFS developers in regards to this. 
> I used to know them quite well :)

fine for me, if there is somebody doing the testing I sure
can (and will) help with the Linux-VServer side

best,
Herbert

> Matt