I'm using Microsoft SQL server 2016 as metadata db (for the sql_alchemy_conn), and pymssql as the Python driver.
When running airflow initdb, airflow creates a table named `TaskInstance` with column `pool` that is initially Nullable (when the table is created).
Later, an index named `ti_pool` on a few columns, including `pool` column in table `TaskInstance`.
Then, airflow will try to alter table `TaskInstance` and change the column `pool` to `NOT NULL`.
This does not work on Microsoft SQL Server since a column with an index defined on it cannot be changed, unless the index is deleted before.