Open Stack

> The remaining work is to have a way of preventing database contracts
> from running until data migrations that the column interacts with are
> complete, this is not an easy problem to solve as the goal of the
> online migrations is to do pure model based schema migrations and
> there is no way of currently identifying when a data migration has
> taken place and a contract is safe to perform.

I'd still like to see what we have done so far merged as soon as
possible. That facilitates testing the main goal of it, which is
actually moving the schema along. It's experimental and completely
optional, and should have a "may eat your lunch" warning anyway.

> The first thought I had was to default to a requirement of a data
> migration removing data from a column and have the default check of
> the contract be one of verifying that the column is currently empty.
> This has the issue of not working in all cases, especially around
> Foreign Keys.

I'm not sure if SQLAlchemy will balk at this or not, but we could do
something like:


  class NameRemovedCondition(object):
      def satisfied(self):
          # Check to see if "name" has been fully migrated


  class Instance(sqla.Model):
      id = sqla.Integer()
      uuid = sqla.String()
      name = RemovedField(original=sqla.String(),
                        condition=NameCondition())


Then the schema migration bit can catch RemovedField entries, know what
they were (so that it can leave that bit alone), and only remove them
when condition is satisfied. Complex migrations that affect multiple
columns can share a condition that checks the full situation.

Like I said, I'd like to get the existing patch merged so that we can
discuss options for this remaining bit in a smaller scope.

--Dan