[Gluster-users] AFR w/ RRDNS failover - does it work or not ? (WAS: simple AFR setup, one server crashes, entire cluster becomes unusable ?)

Daniel Maher dma+gluster at witbe.net
Tue Dec 9 09:29:42 UTC 2008

Keith Freedman wrote:

> the issue isn't reliability, it's availability.
> if a client only talks to one server and that server goes down then the 
> client has nothing to 'fail over' to.  however, if the client talks to 
> both servers then if one goes down it'll keep talking to the other one.

Either the clients will honour the RRDNS and pick another server, or 
they won't - unfortunately, we now have a case where two opposing 
possibilities are being presented.  To wit :

 From the « Gotcha » page :

Applies to server side
"The clients connect only to 1 server. You would need to implement some 
kind of load balancing or something either with round robin DNS [...]"
"If you have client1 connected to server1 and client2 connected to 
server2, and then server2 goes down, so does client2. The cluster also 
becomes unavailable."

Ok, that seems like a straightforward enough statement, however, if we 
take a look back through the mailing list archives, we find a statment 
from Mr. Anand Avati which suggests exactly the opposite :

"Or, put another way, if ClientA (by chance) resolves 
roundrobin.gluster.local to, but .1 is currently down - 
what happens ?

it will attempt on .2, and if that fails (or disconnects after a while), 
it will attempt on .3, and once all the entries are used 'once', it will 
do a fresh dns query.  it does not honor dns refresh timeouts (yet)."

The remaining basic question then is this : does AFR w/ RRDNS failover 
work or not ?  If it does, then the « Gotcha » page should be updated, 
/and/ further investigation is required to determine why it failed to 
operate as advertised in my environment.  If it does /not/, then the « 
Gotcha » page should be updated, and the wiki page i wrote (based 
largely on the suggestions of the developers) should likely be scrapped. :P

As always, thank you all for your continued discourse !

Daniel Maher <dma+gluster AT witbe DOT net>

More information about the Gluster-users mailing list