Models should be the same as the state in DB after migration
We are using declarative_base in cinder.
For creation we are using migrations.
We don't have any tests that checks, that our models are up-to-date. Also we are testing it only on sqlite, which couldn't test such things as nullable constraints.
So We should:
1) Use always explicit nullable parameter for columns. There is a lot of mistakes in current cinder models implementation. (sometimes in schema is nullable=True and in model nullable=False and vise versa). Also at this moment when you see Column description without nullable, you couldn't be sure that is author forget to set nullable='False' or it is really nullable='True'. So the easiest way to track all this and to fix and to avoid such things is to have one rule "always use explicit nullable".
2) Add in __table_args__ indexes and unique constraints
3) Fix all mistakes in models
4) Fix all mistakes in migrations
5) Sync effects of migrations in different backends.
6) Add tests that ensures that models are up-to-date.
This will allow us to find probably some mistake, or missing indexes and make work with db more clean.
Blueprint information
- Status:
- Complete
- Approver:
- John Griffith
- Priority:
- Medium
- Drafter:
- Sergey Kraynev
- Direction:
- Approved
- Assignee:
- Ivan Kolodyazhny
- Definition:
- Obsolete
- Series goal:
- Accepted for alpha-trunk
- Implementation:
-
Unknown
- Milestone target:
-
ongoing
- Started by
- Completed by
- Sean McGinnis
Related branches
Related bugs
Sprints
Whiteboard
Gerrit topic: https:/
Addressed by: https:/
Update Lockutils from oslo
Addressed by: https:/
Add testing modules of migrations from oslo
Addressed by: https:/
Fix sqlalchemy models
Addressed by: https:/
Fix sqlalchemy migrations
Addressed by: https:/
Sync database table iscsi_targets with model IscsiTarget
Addressed by: https:/
Sync database table quotas with model state Quota
Addressed by: https:/
Sync database with state of models Encryption and QuotaUsages
<jdg>
I have no idea how to track this any longer. Overlapped projects, multiple patch sets with no explanation or updates to the whiteboard leaves things in a state where I have no idea what's completed, if you're complete if this is ongoing etc etc.
We really need to work on blueprints and updating info going forward. I'm removing this from target for now simply because I have no idea what's going on here any more and due to the confusion created by having the same BP name in multiple projects (things fall apart when you do this in terms of the auto-tracking and updating). I'd ask that in the future when creating the same bp in multiple projects we use the project name as a prefix or a suffix to denote the different contexts.
