[Gluster-devel] Confused on AFR, where does it happen client or server

[Kevan Benson]

matthew zeier wrote:
> 
> 
> Anand Avati wrote:
>> Here is a very nice tutorial from Paul England which gives a usage 
>> case for server side replication and high availability -
>>
>>
>> http://www.gluster.org/docs/index.php/GlusterFS_High_Availability_Storage_with_GlusterFS 
>>
>>
>> The configuration in the article can be further simplified by removing 
>> the unify layer and having only AFR and will work for most users. 
>> unify will be needed in this setup only if you want to extend this 
>> config with something more advanced.
> 
> 
> In that config the client is only mounting from one server 
> (roundrobin.gluster.local).  Seems that a failure would be user 
> impacting, especially if roundrobin.gluster.local gave me the failed 
> server's address again.
> 
> Do I just need to wait for the High Availability Translator or can the 
> client mount mailspool from two servers?

I got some very odd failover problems when using the RR DNS failover 
config (when testing failures mid-write, etc), but that was a few 
versons back.  I highly recommend using AFR on the client side as a 
failover solution.  It's very robust, and easy to deal with on the 
server end (since you have a full copy of the file system on each 
server).  If you have more than two servers it's not the most efficient 
use of your space though.

make sure to tweak the transport-timeout option for protocol/client, as 
it will define how long a request waits for a non-responding server on a 
failure.  I.e. how long a connection stalls when one server dies.

-- 

-Kevan Benson
-A-1 Networks