[Gluster-devel] HA failover question.

[Kevan Benson]

Chris Johnson wrote:
> On Wed, 17 Oct 2007, Kevan Benson wrote:
>
>      Hm.  Ok, different question.
>
>      If I have two file severs/nodes.  I can put the AFR stuff in the
> client end (stills seems odd as desktops get rebooted all the time).
> Can multiple clients set up for AFR across both these servers access
> them and the files at the same time?  Can the access the same file at
> the same time.  Reads obviouly, writes would have the same issues as
> EXT3 or ReiserFS or whatever with that.  GlusterFS team?  Really
> curious how self healing works with the client end doing the AFR.

Yes, multiple clients can access the files at the same time, just like 
multiple programs could access a file at the same time on a single 
system.  File locking is used to handle the cases where you need 
exclusive access, and the locking is the same as the locking provided if 
it weren't glusterfs, Fcntl and flock.  See 
http://gluster.org/docs/index.php/GlusterFS_Translators_v1.3#posix-locks.

I'm not sure exactly what you are asking, but I believe all read and 
write operations stall until the file replication/writes etc finish.  I 
don't believe you'll ever find inconsistent data to what you just wrote 
as long as the system hasn't just experienced some sort of failure 
condition.  Two separate clients might have a problem when 
accessing/writing the same data, but that's what locking's for.  You 
would have the same problem with two different processes on a single system.

>      Another question.  Could two servers each doing local unify be
> AR'ed from the client side?

I'm not sure if it matters if the unify is exported before AFR'd, all I 
have to go on is what Krishna Srinivas said in that other email, which 
is that AFR of unify volumes is not currently supported.

-- 

-Kevan Benson
-A-1 Networks