[workspace-user] Nimbus + context broker + passing arguments

[Ismael Farfán]

2011/8/18 Tim Freeman <tfreeman at mcs.anl.gov>:
> On Thu, 18 Aug 2011 10:15:15 -0700
> Ismael Farfán <sulfurfff at gmail.com> wrote:
>
>> Hello Tim
>>
>> 2011/8/18 Tim Freeman :
>> > On Wed, 17 Aug 2011 10:50:25 -0700
>> > Ismael Farfán  wrote:
>> >
>> >> Hello list
>> >>
>> >> I have been reading the clusters and clusters2 page of nimbus,
>> >> specially about the context broker, but I don't seem to see how
>> >> to do what I need.
>> >>
>> >> Basically I want to create two custom roles and *pass them
>> >> parameters* through the context broker or other means, not
>> >> only the provide/requires ip and host name, but custom ones.
>> >>
>> >> For example one thing that I need to do:
>> >> Create a custom user on the fly with a given GID and UID (to
>> >> match that of a nfs shared system), this user will be different
>> >> every time to, so just creating a static one won't work.
>> >
>> > Is the data that needs to be sent to other nodes coming from the nodes?  I
>>
>> If I'm not mistaken, the ctx broker is my "base-cluster.xml", so I'd like to
>> put there the params before running the cluster, like this
>> <role name="CreateUser" user group shell uid gid />
>> I don't know if it's possible, all the VMs will receive the same parameters.
>
> I think this is doable.  If you add a data element to the xml input, a script
> matching the name will be invoked on *each* VM in the cluster.  Here's from
> the clusters doc page:
>
> "A script matching the data name is called. Only one argument is passed to it,
> an absolute path to a file with the data contents. You can consume that file
> in any way. This can be a powerful mechanism since you can provide data as the
> remote client. The "data" could even be a zero-length value that triggers some
> behavior merely by calling a particularly named data script."
>

Here I was wondering what that weird string after ![CDATA[ in base-cluster.xml
was about when I first saw it.

Now I just put a bash script there that my role executes and which is giving me
troubles already, so I thing I'm on the right way : )

Thanks

>
>>
>> > don't think the ctx-agent can send that kind of data back to the context
>> > broker yet.  The data elements can only be seeded by a separate client of
>> > the context broker, this could coordinate with the NFS VM somehow but that
>> > seems clunky.  There is an NFS example in the base cluster, if I recall,
>> > does that technique not work?  It must know the NFS account ahead of time.
>>
>> I checked the nfs server and client and torque master and slave under 1 and 4
>> and them all read this:
>> "called with IP, short hostname and hostname (args $1, $2, and $3
>> respectively)"
>>
>> Torque seemed to receive parameters:
>> hostname="true" pubkey="true"
>> but I haven't found where are they used in the scripts.
>
> This hostname="true" setting requires that the delivered hostname (not just
> the IP) is sent back by a ctx-agent to fulfill a "requires".
>
> The pubkey="true" setting is a signal that the client (outside of the cluster,
> typically the cloud client) is interested in getting the generated ssh hostkey
> of that node.
>
>>
>> >
>> >>
>> >> One way to do this would be to use kerberos, but I don't have
>> >> a login-server that I can use for testing purposes...
>> >>
>> >> Any ideas of how to do this?
>> >>
>> >> --
>> >> [============================]
>> >> How do you create an operating system on Linux Ubuntu 9.10?
>> >> Y luego que por qué no me gusta usar Ubuntu...
>> >> [============================]
>> >
>>
>>
>>
>> --
>> [============================]
>> How do you create an operating system on Linux Ubuntu 9.10?
>> Y luego que por qué no me gusta usar Ubuntu...
>> [============================]
>



-- 
[============================]
How do you create an operating system on Linux Ubuntu 9.10?
Y luego que por qué no me gusta usar Ubuntu...
[============================]