[Gluster-users] fencing mechanism

[Giovanni Toraldo]

Hi,

when using GlusterFS in a virtualized infrastructure (OpenNebula in my
case), it would be nice to implement a simple fencing mechanism to
prevent data corruption when the high-availability hook trigger in,
and VM instances that were running on a failing host are restarted on
a new (working) host.

I was wondering if I can use the auth.reject volume option to inhibit
access from a host, that have an already mounted the gluster volume
early, but I fear that it's used only on volume mount time and is not
enforced for already mounted volumes.

I would be grateful if someone has some other idea to force the
disconnection of a client from a particular host (alternatives that I
can think of would be using iptables blocking IP traffic for that
particular host). Remember that this should be done from other hosts
and not remotely on the failing host itself (because it's failing and
I cannot connect to it remotely :))

Thanks for any idea.

--
Giovanni Toraldo - LiberSoft
http://www.libersoft.it