On Mon, Jul 30, 2001 at 12:18:12AM +0900, Junichi Uekawa wrote:
> I'd rather say, cluster.dtd needs to be modified.
> It can be modified easily.
> I think "nfsroot" and "nfsexport" entries need to be there, at least,
> <nfsroot options="ro">/tftpboot/192.168.0.1/</nfsroot>
> kind of expression comes to my mind.
I don't think we need to have an nfsroot entry for every node. We just
need a generic nfsrootbase, and then a node specific entry can
easily be created by using nfsrootbase/<ip address>
> I think that should be something different,
> that is really a specific service to spong, but maybe we could have
Similarly for spong. Just an spongbase entry.
> Hmmm... This is an interesting point.
> update-cluster-regenerate only just executes every script, and one can
> execute the script directly.
> Also, please note that running *.updatelist in postinst
> is against Debian Policy, if the configuration file is a conffile.
How would we get about doing this then ? Say, once I install mpich,
the postinst would use mpich.updatelist which would modify a conffile ?
I'm not sure of the Policy, but I guess its fine to modify conffiles
in one's own package.
> Also, another extention is to make the action non-destructive, so that
> in a configuration file we would have a region surrounded by
That is a good idea.
> > 4. update-cluster could use debconf for input. Other packages can then
> > call
> > it in their postinst to generate the new config files.
> I am feeling dubious about that...
> I don't like my configuration file automatically overwritten when
> some package gets installed.
It would be good to use debconf for getting user input about the cluster,
and storing it in the database. This can then be used to create cluster.xml,
by calling update-cluster-add on each entry. It also provides a convenient
way to modify cluster.xml.
Individual packages can then call the updatelist, or maybe the user can
call it herself.
There's someone in my head but its not me.