Logging suggestion

[Doug Tody]

One way this could work:

     o	Each site operates a logging service which is distinct from other
     	service instances (for data access, computation, etc.).

     o	Services send lightweight messages to the local, centralized
 	logging service whenever anything happens.  This helps keeps the
 	services simple and stateless.

     o	The logging service has the ability to both push logging data to
     	subscribers, as well as respond to queries for logging data.

     o	In push or publish/subscribe mode, it would be relatively easy to
     	buffer data and send a message containing logging data only
 	periodically.  This would avoid the problems of managing very
 	large amounts of logging data, but would still be efficient due
 	to the buffering.

     o	The logging service could also expose a query interface for
     	a client to directly access logging data, e.g., by a range of
 	dates, everything since a certain date, or some other method.
 	Since it is a separate service this does not complicate other
 	service interfaces.

     o	If security is an issue it can easily be dealt with by the logging
     	service.  This is a somewhat different case than for normal
 	data access/analysis services, as only grid infrastructure needs
 	to access logging information.

It would also be possible for simple services which are not heavily used
(small sites) to send small amounts of logging information to a remote
logging service, to avoid the need for small sites to support a logging
service locally.    - Doug