Disable SQLAlchemy connection pooling

[RazerM]

The Database class provides no way to use it, and the current
implementation means that a connection is held by the pool even after
db.close().

See also #67.

[kennethreitz]

aren't there major advantages to using the connection pooling with the current implementation?

[RazerM]

No, Database creates an engine then immediately asks that engine for a connection:

        self._engine = create_engine(self.db_url, **kwargs)

        # Connect to the database.
        self.db = self._engine.connect()

Later, Database.close returns that connection to the pool. This means that the engine held by Database has a connection open for as long as it lives, even after calling close.

Connection pooling doesn't help since within the lifecycle of Database, one connection is checked out, then later returned.

[dmitrytokarev]

@RazerM please resolve conflicts. then we can ping Kenneth to merge this

[dmitrytokarev]

@kennethreitz could you review this PR?

[kennethreitz]

Sounds like we should fix db.close() then

[RazerM]

The way to fix db.close() is by not creating a connection pool that isn't used. It's only keeping a connection open because create_engine's default arguments include pool_size=5, which is "the number of connections to keep open inside the connection pool".

[kennethreitz]

One of the goals of this library was to provide proper connection pooling, @RazerM. It sounds like I messed that up. Can we fix that easily?