Re: [SSI] mkinitrd.ssi and a different /usr partition

[John B. <joh...@co...>]

Bruce Walker wrote:
> 
> Aneesh,
>   You can build a cluster without GFS and without a
> shared root in order to test other aspects of the
> SSI project code.  Later today, I hope John Byrne
> or someone else wil post the directions.  I resisted
> having them post before because the environment is
> quite limited and possibly confusing.  However, it is
> quite appropriate in getting an incremental port to
> another platform.

Thanks, Bruce.

In general, this is dead simple actually. Just make sure both systems
packages installed on both of them; modified the fstabs on both; run
mkinitrd.ssi on both. Things work fairly well. Migration, rfork have
problems with open files; we reopen the file of the same name on the
other node (except for devices opened via devfs, pipes, and sockets): so
if you are actually writing files and you migrate/rfork *bad* things
happen to your data. (We could fairly easily install remote ops for the
file structure to avoid this, but there are more issues to deal with to
make things really correct.)


> 
> The next step might be to get SSI with GFS to work
> on a single node (GFS has run on alpha so hopefully this
> is not too difficult).  Then one could try GNDB from
> the single SSI node to a server.  Then one could try
> two nodes both with GNDB to an external server (this seems
> to work better than trying to have one of the clusternodes
> actually server for the GNDB clients).


I managed to hand edit the initrd to make a non-failover cluster with
one node the GNDB server, but there were some issues trying to start the
GFS lock server there as well. So, you still need an external server for
that.


> 
> bruce
>