The production status machine is now a bit confusing. What is missing is:

1. VERY clear documentation.
2. A real status machine
3. The current one is valid only for the MC request: some automation would be nice for the non-MC requests.
4. The current one does some operations (i.e. removing outputs and killing MC jobs) that are far from optimal.

0. Some documentation is missing.

Clear definitions of status:

Final, irreversible states:

• Archived
• Cleaned

States reached only after manual operations:

• Stop

States

• Idle. Reached when there are only 'Done' and 'Failed' jobs, and/or when there are no 'Unused' nor 'Assigned' files.

1. A real status machine

Proposal (possibly valid for all transformation types, but replication):

For what regards implementation: there is generic state machine recently implemented for the RSS, that could probably be re-used.

The "removing output" phase can be dropped, instead we should flush the merging productions (first stopping the MC prod, and waiting for them until they are idle). Also, remove the name "Validating input" to MC which really does not makes sense.

-- FedericoStagni - 02-May-2011