Which fault-tolerant distributed network file system (linux, need advice) to choose?

Question

linux

Hello comrades.
I am building a HA web cluster. the scheme is approximately as follows:

i.e. I have interchangeable nginx at the entrance, then a pool of web servers and a cluster database is connected to them through the master-backup balancer.
Everything is on fedora20 or centos 6.5
Accordingly, I only have a question with NFS. all web servers must be connected to it.
The most important component is the lack of spof.
Easy scalability is also important.
Those. if I have 10 nfs servers, I need half of them to fall and the data to be saved, the loss of speed is not important.
But if I see that there is not enough space / performance, then I just added 10 more hosts and everything got synchronized and smeared.
I have not yet had time to test anything, but from what I found out.
pNfs 4.1 - unmaintained.
wiki.linux-nfs.org/wiki/index.php/Configuring_pNFS...
here it is advised to refuse.
ceph - seems to be simple, but I don't find any mention of HA in them.
Moose fs - in doubt, looked casually.
Hadoop fs - they claim that they have a slave master (name) node, the Internet scolds. they say
luster, ocfs2, gfs2 crashes are too complex to build and hard to scale.
gluster fs - now I'm collecting for a test.
I seem to remember everything.
Comrades, tell me where to dig, what is better in your opinion and experience of use?
Sincerely, Alexander sm1ly.

Answer 1

if anyone is curious, then initially I still decided to stop at glusterfs, now I'm testing nfs from netapp with redundancy at the netapp level

Answer 2

Hello. NFS is not stable in combat. There are freezes and glitches, or I don't know how to cook...
In my case, I managed to solve the problem using rsync.
I wonder what more experienced comrades will advise.

Answer 3

Well, how are you doing there? what about the gluer? or the old fashioned way all screwed