Open Stack

On Mon, Jan 26, 2015 at 3:03 PM, Andrew Pashkin <apashkin at mirantis.com> wrote:
> On 23.01.2015 23:39, Ruslan Kamaldinov wrote:
>
> 1. Use ModelsMigrationsSync from [2] in tests to make sure that SQLAlchemy
> models are in sync with migrations. Usage example can be found at [3]
>
> Seems like it is a great helper, as I understand it runs all migrations and
> then compares DB state with models state and throws an error if they are out
> of sync.
> What I don't understand - why they still manually write checks for every
> migration [1]? This is redundant, because ModelsMigrationsSync already does
> the job testing that DB is in sync with models.
>
> By the way in Heat project they do the same thing [2]. What am I missing?

I think it's still important to perform migration specific checks. We
want to make sure that DB is in expected state after each specific
migration.

> [1]
> http://git.openstack.org/cgit/openstack/sahara/tree/sahara/tests/unit/db/migration/test_migrations.py#n402
> [2]
> https://github.com/openstack/heat/blob/7d4c4030c591ef5994db4327d66d353ad83c6ea8/heat/tests/db/test_migrations.py#L288

> 2. Populate DB schema from SQLAlchemy models in unit-tests which require
> access to DB
>
> You mean - using these tools [3]?
>
> [3]
> http://docs.sqlalchemy.org/en/rel_0_9/core/metadata.html#creating-and-dropping-database-tables

Yes, this one. We still have this code in source tree:
http://git.openstack.org/cgit/stackforge/murano/tree/murano/db/models.py#n289

> In what conditions 5) will fail? I see only these cases:
> - If data migrations would be introduced and Murano would require some data
> in DB to work correctly.

Actually, we already do that. We populate initial set of categories:
http://git.openstack.org/cgit/stackforge/murano/tree/murano/db/migration/alembic_migrations/versions/001_inital_version.py#n40

Would it affect anyone? I don't think so. You always can populate
these categories manually.

> - If Murano would use some database-specific features (stored procedures
> etc).

That should never happen :)

> There are good chances that these cases will never happen in reality, as I
> understand, so I tend to agree.

Agree

--
Ruslan