Hi!

So no, no native NFS support; that's a bit orthogonal. You could run NFS on 
the storage nodes; for booting purposes this is probably fine, but it's not 
high-speed storage yet. OCFS2 is a trick we haven't tried (and it wouldn't 
work, currently, block devices get locked when mounted to prevent people 
from doing the wrong thing; using something like OCFS2 is possible, but we 
need to do a relatively small amount of work to allow folks like yourself 
to relax that restriction). 

Since high availability is your goal, and it's explicitly two nodes, I 
suppose you could use Torus as effectively a replication tool, such that if 
one of the hosts goes down, the other still has the data (and can serve 
it). There's no built in failover for the mount point -- that's more in the 
realm of Kubernetes or another scheduler, same with the floating IP notion. 

I'm glad you're interested, and welcome giving it a try and finding bugs, 
but it's also a brand new prototype project, so don't expect too much from 
it. ;) PXE booting is within scope for what it might be able to help you 
with, though. 

--Barak

On Friday, June 3, 2016 at 7:34:41 AM UTC-7, Jelle Holtkamp wrote:
>
> Hi all,
>
> I am interested in your new product Torus. I have no experience with 
> CoreOS yet, so I do not know the full capabilities or restrictions it has. 
> Anyway, the blog post 
> <https://coreos.com/blog/torus-distributed-storage-by-coreos.html> states 
> that it uses 'multiple ways' to expose itself to user applications. I am 
> looking into numerous distributed storage software solutions, which have to 
> be able to:
>
>
>    - Provide constant high speed remote storage
>    - Provide highly available PXE boot capabilities for Linux clients 
>    using NFS
>    - Preferably be able provide these services in a *2 node *cluster
>    
> The PXE boot does not necessarily need to be extremely fast, as long as 
> the OS can boot in a reasonable time (at most 3 minutes). So some added 
> overhead of running NFS over a block level distributed solution would not 
> be a big problem, because once the OS is loaded it does not really do much 
> besides writing data to the high speed remote storage. Of course if the NFS 
> storage can provide the necessary I/O it would be perfect to use as both 
> PXE boot and high speed remote storage. It is also a big preference if this 
> can be done in a two node cluster, meaning that there is no external NFS 
> server that has the storage devices mounted and exports it to the clients. 
>
>
> Anyway, my question is: 
>
>
> Does Torus provide native NFS support, or
>
> Would it be possible to run an NFS server on the storage nodes and use ip 
> failover (pacemaker/corosync?) to make it highly available?
>
>
> Unless Torus has built-in failover capabilities for an active/passive 
> setup, one possible way I see is creating an active/active setup by having 
> the storage nodes themselves mounting the NBD devices  format it as OCFS2, 
> NFS export it on all nodes and use corosync+pacemaker to provide a floating 
> IP to which the clients can connect. 
>
>
> Looking forward to the response, as Torus seems very promising!
>

Reply via email to