[Gluster-users] Synchronous replication, or no?

[Jeff Darcy]

> Ok, that made a lot of sense. I guess what I was expecting was that the
> writes were (close to) immediately consistent, but Gluster is rather
> designed to be eventually consistent.

All distributed file systems are, to some extent; we just try to be
clearer than most about what the guarantees are.  For example, some
buffer at the client *despite* fsync or O_SYNC.  The temptation is
obvious; POSIX "single system image" behavior is far more expensive
in a distributed file system than in a local one, and everyone has
to compete on performance.  We're actually far stricter than most
when it comes to durability, and the performance disadvantage has
been difficult to bear sometimes.  Hopefully, now that we have the
"upcall" facility (developed for NFSv4) we can improve consistency
as well without having to give up more performance.