rake db:test:prepare

rubyonrails-core
1

I'm having fun with a legacy DB (SQL Server), and in the process found some code in rails/lib/tasks/databases.rake which feels wrong.

0) Errors which occur during this phase of a rake:test or rake:spec tend to be ignored, although they are logically fatal. In the best case, this means that the stack trace when an error finally does manage to stop the process is worthless. Best case.

1) Prepare's strategy is to dump the current schema, blow away the test db, and then rebuild the test schema from the dump. This strikes me as dangerous in the event that someone has mucked with the db outside of a migration--that is, made a change that won't get reflected in source control. Wouldn't it be safer to just blow away the test db & do the migrations to rebuild?

2) db:structure:dump for sqlserver requires the scptxfr executable, which is part of the SQLServer package--not the client. While it is likely that developers can get access to this file (and its supporting files), there is no a priori reason to believe that a developer is going to have access to this expensive bit of software. More generally, should we not restrict dependencies to client code? (The scptxfr script went away with SQLServer 2005, so this particular code is also aging.)

3) Several of the tasks in the db namespace have code of the form case abcs[blah]["adapter"]   when "mysql", "oci", "oracle" ...   when "postgresql" ...   when "sqlserver" ...

... end

This is generally considered bad practice ("hypnotic" coding). It is not extensible, and it breaks the model of separating out code for the commercial databases. Does rake support calls to computed tasks?

2

I'm having fun with a legacy DB (SQL Server), and in the process found some code in rails/lib/tasks/databases.rake which feels wrong.

0) Errors which occur during this phase of a rake:test or rake:spec tend to be ignored, although they are logically fatal. In the best case, this means that the stack trace when an error finally does manage to stop the process is worthless. Best case.

Yeah, I've actually been bitten by this in the past, please do investigate a patch to make these errors get caught and reported in a useful manner.

1) Prepare's strategy is to dump the current schema, blow away the test db, and then rebuild the test schema from the dump. This strikes me as dangerous in the event that someone has mucked with the db outside of a migration--that is, made a change that won't get reflected in source control. Wouldn't it be safer to just blow away the test db & do the migrations to rebuild?

This has been discussed in the past, cloning development (by importing schema.rb) is the preferred way of doing this. The short reasons are:

* Most migrations in a large application are for munging production data * it takes far too * and there's no guarantee that it will continue to work (referencing old models etc)

The question of when the schema gets dumped is an open one. The reason it's done in that rake task is that prior to migrations the way you made changes to your database was to issue DDL directly against the dev db, and you needed to have your tests dump the database if they were to have any chance of working.

Now we could *probably* just dump schema.rb after running a migration, and load it before test runs. This will reduce the changes of it containing 'bogus' information, without making you run all 400 migrations for every test run.

2) db:structure:dump for sqlserver requires the scptxfr executable, which is part of the SQLServer package--not the client. While it is likely that developers can get access to this file (and its supporting files), there is no a priori reason to believe that a developer is going to have access to this expensive bit of software. More generally, should we not restrict dependencies to client code? (The scptxfr script went away with SQLServer 2005, so this particular code is also aging.)

3) Several of the tasks in the db namespace have code of the form case abcs[blah]["adapter"]   when "mysql", "oci", "oracle" ...   when "postgresql" ...   when "sqlserver" ...

... end

This is generally considered bad practice ("hypnotic" coding). It is not extensible, and it breaks the model of separating out code for the commercial databases. Does rake support calls to computed tasks?

Even if rake doesn't support this we could push that behaviour into the adapters. Something along the lines of

task db:structure:dump do   current_adapter.dump!(any, arguments, needed) end

Again, if you feel like working on this, we'd be happy to help you get a patch ready to apply.

3

How timely. I've been working on just this stuff this week. I'm about to dive into patching db:test:prepare so all it does it load the schema.rb instead of dumping it from the dev db every time. That should "just work" ok, but if you want to be confident that schema.rb reflects the state of the dev db, you need to know that migrations have been run correctly. I'm working on a plugin that helps detect when you need to run migrations, even detecting when someone may have modified an old migration (by saving a snapshot of SHA1 hashes of all migrations files). The two changes should play well together.

4

For an alternative see my article at http://pivots.pivotallabs.com/users/alex/blog/articles/305-collapsing-migrations

If you make it part of your process to regularly collapse migrations (like after every deploy), you avoid both of the problems mentioned above (referencing old class definitions and taking a long time) but you also avoid the problems with relying on schema.rb, namely :

* your development and test environments will be using the same technique (migrations) for creating/altering the DB * your migrations will be exercised by developers and test builds before being deployed * you won't get burned by forgetting to export and check in schema.rb after adding a migration

Anyway, I'm not trying to change anyone's mind here, but this technique has worked really well for us on several projects.

Cheers -

- Alex