OpenNebula

Borja explain the approach of using the DB directly:

The current approach used in "onehost list" and in the OpenNebula/Haizea integration is to access the ONE database directly. I believe the rationale for this was that (1) the physical host information is something you typically only need to know locally (e.g., admin commands, the Haizea scheduler, etc.) and not remotely and (2) polling for a (potentially large) host list through XMLRPC could be an expensive operation, so direct database access was used instead.

I can see how rationale (1) is not really valid if you want a third-party component (like VWS, which could be in a different machine) to access the host list, so it may make sense to add a new operation to the API to retrieve the host list (even if this operation is not used by "onehost list"; at least it would be available for other components to use). Could you create a ticket in the Trac so we can track the issue there?

Finally, as a workaround, you can access the ONE database directly for now (the only limitation is that VWS and ONE will have to be running on the same node; but this assumption can be easily removed when an API call for this becomes available). There are a number of options for accessing a SQLite database from Java: http://www.sqlite.org/cvstrac/wiki?p=SqliteWrappers. If you have any trouble parsing the ONE db, don't hesitate to ask here.

I would still like to have it as an XML-RPC method. Several reasons come to mind:

a) I was under the impression that the administration tools (one{vm,host,vnet}) were using XML-RPC because that would allow them to run one a different host than oned. If that was never the intent, perhaps ONE could use a UNIX socket for access instead, which would remove the need for the session identifier in the XML-RPC interface, since access could be handled at the filesystem layer (or alternatively using SO_PEERCRED).

b) I don't really see why this operation would be much more expensive than e.g. one.hostinfo? Also, if I really do need this info, I'll query the database for it, so whether that is done using libsqlite or and xml-rpc interface shouldn't make much of a difference, should it?

c) To get access to the information, you need full read access to the database. I'm imagining that we at some point would want to let people fetch the list of hosts, but perhaps wouldn't be comfortable letting the read all the information from the database.

Re-open this, and schedule to 1.4.