[ENBD] Re: 2 sided enbd

[Peter T. Breuer]

"A month of sundays ago Chas Wareing wrote:"
> Wow, that is quite a response; If I understand correctly, ENBD cannot do
> what I want; which I will re-state below;

Proably not. I think what you are trying to do is like writing to
a scsi disk on a bus shared between two machines. It wouldn't
work because each machine would have a different image of what's
on the disk.

Let's see ...

> I have 2 servers
> I want Server A and Server B to write to the same 100+GB filesystem.
> I want a filesystem that stays up if either server goes offline;

But that's a normal failover situation, isn't it? Oh, no, it's
not. The usual situation is that two servers write to two different
file systems, and if either server dies, then the other takes
over for it and writes to the other FS as well as to its own.

Still, it's no big deal. Just entails dividing your one FS into two.

> I have many 2GB files (VMWare)
> 
> Scenario:
> 	Server A dies.  The filesystem lives, Server B can take over
> VMWare processes (via scripting);
> 
> Possible?

Let's see if the normal failover situation can be mutated to operate
with only one file system. Well, yes, but it's not elegant.

Server A talks to a raid mirror with a local 100GB device and an
enbd. The other end of the enbd is a 100GB image on server B.
But server A exports the raid mirror via NFS to server B, which
also serves from it (the NFS mount).

If Server B dies, no sweat. Server A takes its IP and continues
serving. When server B comes back up we have to resync its image,
but it's not being served anyway.

If Server A dies, then we have to start serving from the uptodate
image on B instead. When Server A comes back up we swap roles and
treat it as secondary  until the resync is complete. Server B exports
its image via NFS and server A serves from that until its image
is resynced. At that point we either swap the roles back or continue as
before.

Why not just have the normal two FS's and two mirrors? It's more
symmetric!


Peter