[Gluster-devel] Unify without AFR

[Christopher Hawkins]

My goal is to design a fault tolerant server pair that can both:

A. Remove gluster single points of failure for filesystem access 
B. Double the bandwidth for serving client requests when both servers are
online 

Since the HA translator is not available yet, I am thinking of this as a
work around and I'm wondering if the devel team or anyone else sees any
problems with it or has suggestions...

What if I take two servers that export directory /example and use drbd v8
(which allows concurrent read / write) to mirror /example on server one to
/example on server 2? Any issues with using the client side to unify them,
say with RR or Random scheduler and no AFR, since the files are already
being mirrored in the back end? That way the clients would essentially
choose either server for reads and writes, and whichever one they chose
would have a complete namespace and handle the request. 

Another alternative would be just giving the clients a single server IP with
no unify or AFR, and using an external load balancer to distribute inbound
connections among the two gluster servers. I don't know gluster well enough
to know if there are any "gotchas" that I am missing. Will be happy to
report back with test results. And a big thank you to the team for a great
cluster filesystem... If it wasn't so flexible, I wouldn't even be able to
think about doing this! 

Thanks,

Chris