[AG-TECH] notes and thoughts on AG2.1.2b on win32

Thomas D. Uram turam at mcs.anl.gov
Sun Nov 13 14:25:05 CST 2005

Hi Lewis:

Have you been listening in on the node-related conversations we've been 
having here?  ;-)

You have successfully identified the bugs and enhancements we've been 
talking about lately to make shared nodes more "infrastructurable".  We 
really want nodes to be always available and require a minimum of user 
interaction to function.

Let me pull your points out and discuss them individually:

- Node service does not load the default config on startup
    This is a bug that showed up in the 2.1 timeframe.  It's very simple 
to fix, and will be handled in the next release.

- All users must configure services themselves
    In 2.1.2, when a user runs a node service for the first time, the 
system node configs are copied to the user space, and the system default 
node configuration will be loaded.  This allows an admin to build a 
default configuration and a set of additional configurations, and have 
new users automatically inherit these.  (Clearly this is only true when 
the node service is run within the venue client, per the point above.)

- Single video producer/consumer service for PIGs
    Bug 493 identifies--I think--the issue you are pointing out here.  A 
combined video producer/consumer service would be nearly trivial to put 
together (a one or two line change to the code).  This should show up 
very soon.

- Autorun services at boot
    The node service could run at boot right now.  The problem lies with 
the service managers:  since they start the media tools,. and the media 
tools require a display, a user must be logged in to the machine to run 
them.  We are working to solve this problem with versions of the media 
tools that can be controlled via RPC instead of a graphical interface.

-  Certificate handling for shared nodes; proxy timeouts
    When the services can run at boot, they'll need a certificate with 
which to start.  This will be a "service certificate" instead of the 
identity certificates that we are now issuing.  Service certificates 
have no passphrase, so there is no proxy to timeout.

Additionally, we've talked about node services and service managers 
advertising themselves, so a user can select from a list of those 
available.  Of course, this should be subject to authorization, so users 
can't just control them willy nilly.

Let me know if I've missed anything, or if you have questions beyond 
these.  Also, be sure to request enhancements for us to consider for the 
next-release node software.

Thanks for your comments,

Tom Uram

Lewis Grantham wrote:

>dear AG
>been working through AG2.1.2 again, and have some more notes and comments for
>(Bugzilla Bug 742 i guess): defaultWindows configurations for AGServiceManager
>only loaded when using VenueClient --personalNode. (and an ANL/AG issued cert..
>others proxy ok, but you cant manage node)
>Starting AGServiceManager and AGNodeService manually wont load any config, or
>add any services at all.
>Yes, the menu shortcut on windows includes the --personalNode argument, and self
>starts AGServiceMAnager and AGNodeservice on a single machine I presume. 
>But what of trying (mistakenly it seems) to test rigs for multimachine set ups
>(i.e. the majority of GRID nodes built to recommended AG specs)?
>how do we fully test a multimachine config and successful (default) service
>it is possible to start things manually, add services from the venue client and
>correctly connect to at least the ag2 server it seems, but not something I can
>see catching on for ease of use.
>But a personal node would look wrong too, how to link the producer and consumer
>for example?  
>Users will only truly navigate with the one display machine, so am stuck with no
>synchronisation with other servers (our encoder for example).
>and thats my main concern here. how do I establish reliable service placements
>across a multimachine node?  
>Seems I cannot autorun services at boot time - they will all eventually fail if
>a short proxy is in use, and with many aspects of node management now in the
>venueclient, these are easily changed by individuals.
>with no grid proxy init, PASSPHRASE and proxy creation is commandline based on a
>variable user ID.. ok - now whos going to bother fixing up the other machines no
>one ever runs the venue client on..  
>each facility user must by necessity start and configure all services themselves
>to make sure a valid and consistent proxy is about?
>these points are starting to place VERY high training and staffing requirements
>on any planned (multi machine) node use, and would seem more suited to self
>customisable PIGs and single machine nodes only (say floating, departmental
>laptops or similar).
>I cant visualise this on a multimachine set up at the moment without a lot of
>manual intervention - have I missed a big part or hint here? 
>many thanks for your time amnd trouble in this

More information about the ag-tech mailing list