Is it possible to get posix locks in a high availability configuration
with glusterfs?
Specifically, is it sufficient to have locking enabled on the server side
below the AFR layer?
|glusterfsd....| |.......glusterfs client..
[u1]---[u1-lock]---+
:=:--afr0--+
[u3]---[u3-lock]---+ |
+----client/unify-----
[u2]---[u2-lock]---+ | |
:=:--afr0--+ |
[u4]---[u4-lock]---+ |
|
[u5-ns]--------------------------+
Here is part of the server configuration ...
volume u1
type storage/posix
option directory /glusterfs/u1/export
end-volume
volume u2
type storage/posix
option directory /glusterfs/u2/export
end-volume
volume u1-lock
type features/posix-locks
subvolumes u1
end-volume
volume u2-lock
type features/posix-locks
subvolumes u2
end-volume
I have a bad feeling that in a race condition, half of the clients will check
the locks on one brick while the other half will check the other brick ...
and that simultaneous transactions may be quite happy with getting one out of
two flocks, or only release one of the two when they are done ...
&:-)
