Inconsistencies in Model Registration and Unregistration Functions in models.py

[Geogouz]

I noticed couple of inconsistencies in the models.py related to how models are registered and unregistered.

For example:

1. Duplicate Entries: The model ReplicationRule is listed twice in the register_models function (even though SQLAlchemy shouldn't have a problem handling this).

rucio/lib/rucio/db/sqla/models.py

Lines 1740 to 1743 in 3397448

	ReplicaLock,
	ReplicationRule,
	ReplicationRule,
	ReplicationRuleHistory,

2. Missing Model in Unregistration Function: The VirtualPlacements model is included in the register_models function but is absent from the unregister_models function.

3. Missing Models from both Functions: The models BadPFNs, TransferLimit, RSETransferLimit, DatasetLock, Distance, TransferStats are not considered at all.

Are (any of) these intensional?

[erlingstaff]

The current implementation is very confusing and not ideal. Even though there are models missing in register and unregister, as you noted, they all get created and deleted correctly (the commands in psql screenshot are run before/after running ./tools/destroy_database.py, which calls unregister_models).

When unregister or register_models is run now, it loops over all the listed models and calls model.metadata.drop_all(engine), where metadata is a db schema, from which all tables (models) are dropped. However, they all inherit the same metadata and schema from the BASE class.

TL;DR:
Currently register_models tries to register the same schema many times (one time for each model listed), likewise with unregister_models. The schema contains all the models. Simplified down to a single call BASE.metadata.drop_all(engine) / BASE.metadata.create_all(engine) to register or unregister the schema. #6712