Fix all deprecations for SQLAlchemy 2.0

[potiuk]

Body

Airflow is currently not compatible with SQLAlchemy 2.0 which is about to be released. We need to make a deliberate effort
to support it.

Here are some info to aid in this effort:

• Description of all removed featuers in SQLAlchemy 2.0: https://sqlalche.me/e/b8d9

• The way to see all the deprecations in latest version of SQLAlchemy 1.4 is to set environment variable SQLALCHEMY_WARN_20=1

• How to start it:

  • add SQLALCHEMY_WARN_20=1 to ci.yml at top level
  • look through warnings.txt in tests to investigate the warnings (there will also be errors in the provider imports)
  • when all of them are fixed - remove <2.0 limitation in sqlalchemy in setup.cfg
  • make sure CI is green

Local setup in PyCharm

If you use PyCharm for run tests, you might want to setup SQLALCHEMY_WARN_20=1 for all pytest runs by default

1. Go to Run -> Edit Configurations... -> click on Edit configuration templates...
   
2. In open window select Python Tests -> pytest and put SQLALCHEMY_WARN_20=1 into the environment variable section
   
3. Click Apply, Click OK
4. Thats it, every new pytest run configuration will have this option by default

Known non-compatible with SA 20

Reported in core

• airflow/models/taskinstance.py:1869
  sqlalchemy.exc.RemovedIn20Warning - Using strings to indicate column or relationship paths in loader options is deprecated and will be removed in SQLAlchemy 2.0. Please use the class-bound attribute directly. (Background on SQLAlchemy 2.0 at: https://sqlalche.me/e/b8d9)

• airflow/models/taskinstance.py:1871
  sqlalchemy.exc.RemovedIn20Warning - Using strings to indicate column or relationship paths in loader options is deprecated and will be removed in SQLAlchemy 2.0. Please use the class-bound attribute directly. (Background on SQLAlchemy 2.0 at: https://sqlalche.me/e/b8d9)

• airflow/www/views.py:845
  sqlalchemy.exc.RemovedIn20Warning - The "columns" argument to Select.with_only_columns(), when referring to a sequence of items, is now passed as a series of positional elements, rather than as a list. (Background on SQLAlchemy 2.0 at: https://sqlalche.me/e/b8d9)

• airflow/www/views.py:3455
  sqlalchemy.exc.RemovedIn20Warning - The Row.keys() method is considered legacy as of the 1.x series of SQLAlchemy and will be removed in 2.0. Use the namedtuple standard accessor Row._fields, or for full mapping behavior use row._mapping.keys() (Background on SQLAlchemy 2.0 at: https://sqlalche.me/e/b8d9)

• airflow/www/views.py:3455
  sqlalchemy.exc.RemovedIn20Warning - Retrieving row members using strings or other non-integers is deprecated; use row._mapping for a dictionary interface to the row (Background on SQLAlchemy 2.0 at: https://sqlalche.me/e/b8d9)

• airflow/www/views.py:3293
  sqlalchemy.exc.RemovedIn20Warning - The Row.keys() method is considered legacy as of the 1.x series of SQLAlchemy and will be removed in 2.0. Use the namedtuple standard accessor Row._fields, or for full mapping behavior use row._mapping.keys() (Background on SQLAlchemy 2.0 at: https://sqlalche.me/e/b8d9)

• airflow/www/views.py:3293
  sqlalchemy.exc.RemovedIn20Warning - Retrieving row members using strings or other non-integers is deprecated; use row._mapping for a dictionary interface to the row (Background on SQLAlchemy 2.0 at: https://sqlalche.me/e/b8d9)

• airflow/jobs/scheduler_job_runner.py:1642
  sqlalchemy.exc.RemovedIn20Warning - Using strings to indicate column or relationship paths in loader options is deprecated and will be removed in SQLAlchemy 2.0. Please use the class-bound attribute directly. (Background on SQLAlchemy 2.0 at: https://sqlalche.me/e/b8d9)

• airflow/models/trigger.py:141
  sqlalchemy.exc.RemovedIn20Warning - Using strings to indicate column or relationship paths in loader options is deprecated and will be removed in SQLAlchemy 2.0. Please use the class-bound attribute directly. (Background on SQLAlchemy 2.0 at: https://sqlalche.me/e/b8d9)

• airflow/utils/db_cleanup.py:200
  sqlalchemy.exc.RemovedIn20Warning - The bind argument for schema methods that invoke SQL against an engine or connection will be required in SQLAlchemy 2.0. (Background on SQLAlchemy 2.0 at: https://sqlalche.me/e/b8d9)

• airflow/cli/commands/task_command.py:202
  sqlalchemy.exc.RemovedIn20Warning - "TaskInstance" object is being merged into a Session along the backref cascade path for relationship "DagRun.task_instances"; in SQLAlchemy 2.0, this reverse cascade will not take place. Set cascade_backrefs to False in either the relationship() or backref() function for the 2.0 behavior; or to set globally for the whole Session, set the future=True flag (Background on this error at: https://sqlalche.me/e/14/s9r1) (Background on SQLAlchemy 2.0 at: https://sqlalche.me/e/b8d9)

• airflow/models/taskinstance.py:524
  sqlalchemy.exc.RemovedIn20Warning - Using strings to indicate column or relationship paths in loader options is deprecated and will be removed in SQLAlchemy 2.0. Please use the class-bound attribute directly. (Background on SQLAlchemy 2.0 at: https://sqlalche.me/e/b8d9)

Reported in providers

• airflow/providers/openlineage/utils/sql.py:152
  sqlalchemy.exc.RemovedIn20Warning - The MetaData.bind argument is deprecated and will be removed in SQLAlchemy 2.0. (Background on SQLAlchemy 2.0 at: https://sqlalche.me/e/b8d9)

• airflow/providers/fab/auth_manager/security_manager/override.py:1509
  sqlalchemy.exc.RemovedIn20Warning - "User" object is being merged into a Session along the backref cascade path for relationship "Role.user"; in SQLAlchemy 2.0, this reverse cascade will not take place. Set cascade_backrefs to False in either the relationship() or backref() function for the 2.0 behavior; or to set globally for the whole Session, set the future=True flag (Background on this error at: https://sqlalche.me/e/14/s9r1) (Background on SQLAlchemy 2.0 at: https://sqlalche.me/e/b8d9)

Reported in tests

• tests/providers/fab/auth_manager/api_endpoints/test_user_schema.py:63
  sqlalchemy.exc.RemovedIn20Warning - "User" object is being merged into a Session along the backref cascade path for relationship "Role.user"; in SQLAlchemy 2.0, this reverse cascade will not take place. Set cascade_backrefs to False in either the relationship() or backref() function for the 2.0 behavior; or to set globally for the whole Session, set the future=True flag (Background on this error at: https://sqlalche.me/e/14/s9r1) (Background on SQLAlchemy 2.0 at: https://sqlalche.me/e/b8d9)

• tests/jobs/test_backfill_job.py:1912
  sqlalchemy.exc.RemovedIn20Warning - "TaskInstance" object is being merged into a Session along the backref cascade path for relationship "DagRun.task_instances"; in SQLAlchemy 2.0, this reverse cascade will not take place. Set cascade_backrefs to False in either the relationship() or backref() function for the 2.0 behavior; or to set globally for the whole Session, set the future=True flag (Background on this error at: https://sqlalche.me/e/14/s9r1) (Background on SQLAlchemy 2.0 at: https://sqlalche.me/e/b8d9)

• tests/models/test_dagrun.py:2090
  sqlalchemy.exc.RemovedIn20Warning - "TaskInstance" object is being merged into a Session along the backref cascade path for relationship "DagRun.task_instances"; in SQLAlchemy 2.0, this reverse cascade will not take place. Set cascade_backrefs to False in either the relationship() or backref() function for the 2.0 behavior; or to set globally for the whole Session, set the future=True flag (Background on this error at: https://sqlalche.me/e/14/s9r1) (Background on SQLAlchemy 2.0 at: https://sqlalche.me/e/b8d9)

• tests/models/test_taskinstance.py:1487
  sqlalchemy.exc.RemovedIn20Warning - "TaskInstance" object is being merged into a Session along the backref cascade path for relationship "DagRun.task_instances"; in SQLAlchemy 2.0, this reverse cascade will not take place. Set cascade_backrefs to False in either the relationship() or backref() function for the 2.0 behavior; or to set globally for the whole Session, set the future=True flag (Background on this error at: https://sqlalche.me/e/14/s9r1) (Background on SQLAlchemy 2.0 at: https://sqlalche.me/e/b8d9)

• tests/ti_deps/deps/test_trigger_rule_dep.py:140
  sqlalchemy.exc.RemovedIn20Warning - "TaskInstance" object is being merged into a Session along the backref cascade path for relationship "DagRun.task_instances"; in SQLAlchemy 2.0, this reverse cascade will not take place. Set cascade_backrefs to False in either the relationship() or backref() function for the 2.0 behavior; or to set globally for the whole Session, set the future=True flag (Background on this error at: https://sqlalche.me/e/14/s9r1) (Background on SQLAlchemy 2.0 at: https://sqlalche.me/e/b8d9)

• tests/utils/test_log_handlers.py:475
  sqlalchemy.exc.RemovedIn20Warning - "Trigger" object is being merged into a Session along the backref cascade path for relationship "TaskInstance.trigger"; in SQLAlchemy 2.0, this reverse cascade will not take place. Set cascade_backrefs to False in either the relationship() or backref() function for the 2.0 behavior; or to set globally for the whole Session, set the future=True flag (Background on this error at: https://sqlalche.me/e/14/s9r1) (Background on SQLAlchemy 2.0 at: https://sqlalche.me/e/b8d9)

• tests/api_experimental/common/test_mark_tasks.py:137
  sqlalchemy.exc.RemovedIn20Warning - The eagerload construct is considered legacy as of the 1.x series of SQLAlchemy and will be removed in 2.0. Please use joinedload(). (Background on SQLAlchemy 2.0 at: https://sqlalche.me/e/b8d9)

• tests/api_connexion/endpoints/test_task_instance_endpoint.py:190
  sqlalchemy.exc.RemovedIn20Warning - "TaskInstance" object is being merged into a Session along the backref cascade path for relationship "DagRun.task_instances"; in SQLAlchemy 2.0, this reverse cascade will not take place. Set cascade_backrefs to False in either the relationship() or backref() function for the 2.0 behavior; or to set globally for the whole Session, set the future=True flag (Background on this error at: https://sqlalche.me/e/14/s9r1) (Background on SQLAlchemy 2.0 at: https://sqlalche.me/e/b8d9)

• tests/api_connexion/schemas/test_task_instance_schema.py:69
  sqlalchemy.exc.RemovedIn20Warning - "TaskInstance" object is being merged into a Session along the backref cascade path for relationship "DagRun.task_instances"; in SQLAlchemy 2.0, this reverse cascade will not take place. Set cascade_backrefs to False in either the relationship() or backref() function for the 2.0 behavior; or to set globally for the whole Session, set the future=True flag (Background on this error at: https://sqlalche.me/e/14/s9r1) (Background on SQLAlchemy 2.0 at: https://sqlalche.me/e/b8d9)

• tests/api_connexion/schemas/test_task_instance_schema.py:113
  sqlalchemy.exc.RemovedIn20Warning - "TaskInstance" object is being merged into a Session along the backref cascade path for relationship "DagRun.task_instances"; in SQLAlchemy 2.0, this reverse cascade will not take place. Set cascade_backrefs to False in either the relationship() or backref() function for the 2.0 behavior; or to set globally for the whole Session, set the future=True flag (Background on this error at: https://sqlalche.me/e/14/s9r1) (Background on SQLAlchemy 2.0 at: https://sqlalche.me/e/b8d9)

• tests/serialization/test_pydantic_models.py:162
  sqlalchemy.exc.RemovedIn20Warning - "DagScheduleDatasetReference" object is being merged into a Session along the backref cascade path for relationship "DatasetModel.consuming_dags"; in SQLAlchemy 2.0, this reverse cascade will not take place. Set cascade_backrefs to False in either the relationship() or backref() function for the 2.0 behavior; or to set globally for the whole Session, set the future=True flag (Background on this error at: https://sqlalche.me/e/14/s9r1) (Background on SQLAlchemy 2.0 at: https://sqlalche.me/e/b8d9)

• tests/serialization/test_pydantic_models.py:163
  sqlalchemy.exc.RemovedIn20Warning - "TaskOutletDatasetReference" object is being merged into a Session along the backref cascade path for relationship "DatasetModel.producing_tasks"; in SQLAlchemy 2.0, this reverse cascade will not take place. Set cascade_backrefs to False in either the relationship() or backref() function for the 2.0 behavior; or to set globally for the whole Session, set the future=True flag (Background on this error at: https://sqlalche.me/e/14/s9r1) (Background on SQLAlchemy 2.0 at: https://sqlalche.me/e/b8d9)

Committer

• I acknowledge that I am a maintainer/committer of the Apache Airflow project.

[Taragolis]

Just clarification. Our target support both 1.4.x and 2.0 or only 2.0?

[potiuk]

Just clarification. Our target support both 1.4.x and 2.0 or only 2.0?

This is a good point, and I do not know the answer. I think it will entirely depend on:

a) when we will do it
b) how difficult it will be to keep compatibility
c) some indication of the adoption of 2.0 once it is out there

I would refrain from trying to figure the answer before the final 2.0 release is out there - and possibly we should even wait for at least 2.0.1 or so, generally we should scout for stability indications. And then we should attempt to migrate and see how difficult it is to keep it compatible with both.

If we will decide to support both, due to many differences and generally backwards incompatible release of sqlalchemy I would be only heppy with supporting both if we extend our test suite in CI and add another test matrix dimention - sqlalchemy_version (on top of backend + backend version, Python version) that we have now.

But we will also have to see if that is really needed, depending on how complex the fixing will be and how different the two versions are. If the differences are small, we can likely get-by without having the extra matrix dimension.

If we see that differences are big, I think going 2.0-only is a better approach (and it basically means we will have to wait at least 6-10 months after release depending on adoption IMHO).

There is absolutely no hurry with migration yet, 1.4 will be out there for a long while and it will be supported with bugfixes - so we do not have (and will not have for a while) particular need to migrate to 2.0.

[Taragolis]

Migrate 2.0 to 1.4 potentially could have more effect rather than we have with other packages.
Some of dependency of providers also use SQLAlchemy and we need to wait until them migrate to 2.0

Current packages which explicit have SQLAlchemy as dependency from CI image.

root@6e6ad04bdb9b:/opt/airflow# pipdeptree --reverse --packages SQLAlchemy --python `which python`

sqlalchemy==1.4.46
  - alembic==1.9.1 [requires: SQLAlchemy>=1.3.0]
    - apache-airflow==2.6.0.dev0 [requires: alembic>=1.6.3,<2.0]
  - apache-airflow==2.6.0.dev0 [requires: sqlalchemy>=1.4,<2.0]
  - elasticsearch-dbapi==0.2.9 [requires: sqlalchemy]
  - eralchemy2==1.3.6 [requires: SQLAlchemy>=1.3.19]
  - Flask-AppBuilder==4.1.4 [requires: SQLAlchemy<1.5]
    - apache-airflow==2.6.0.dev0 [requires: flask-appbuilder==4.1.4]
  - Flask-SQLAlchemy==2.5.1 [requires: SQLAlchemy>=0.8.0]
    - Flask-AppBuilder==4.1.4 [requires: Flask-SQLAlchemy>=2.4,<3]
      - apache-airflow==2.6.0.dev0 [requires: flask-appbuilder==4.1.4]
  - marshmallow-sqlalchemy==0.26.1 [requires: SQLAlchemy>=1.2.0]
    - Flask-AppBuilder==4.1.4 [requires: marshmallow-sqlalchemy>=0.22.0,<0.27.0]
      - apache-airflow==2.6.0.dev0 [requires: flask-appbuilder==4.1.4]
  - snowflake-sqlalchemy==1.4.4 [requires: sqlalchemy>=1.4.0,<2.0.0]
  - sqlalchemy-bigquery==1.5.0 [requires: sqlalchemy>=1.2.0,<2.0.0dev]
  - sqlalchemy-drill==1.1.2 [requires: sqlalchemy]
  - SQLAlchemy-JSONField==1.0.1.post0 [requires: sqlalchemy]
    - apache-airflow==2.6.0.dev0 [requires: sqlalchemy-jsonfield>=1.0]
  - sqlalchemy-redshift==0.8.12 [requires: SQLAlchemy>=0.9.2,<2.0.0]
  - SQLAlchemy-Utils==0.39.0 [requires: SQLAlchemy>=1.3]
    - Flask-AppBuilder==4.1.4 [requires: sqlalchemy-utils>=0.32.21,<1]
      - apache-airflow==2.6.0.dev0 [requires: flask-appbuilder==4.1.4]

This is not complete list because some of packages could use sqlalchemy but do not have it in dependency, for example pandas use it in read_sql.

[uranusjr]

I think we should support both 1.4 and 2.0 for at least one minor release, preferrably much longer. Dependencies don’t tend to catch up very fast for this kind of migrations, and an Airflow installation generally has a lot of those.

[potiuk]

Agree.

[kaxil]

I think we should support both 1.4 and 2.0 for at least one minor release, preferrably much longer. Dependencies don’t tend to catch up very fast for this kind of migrations, and an Airflow installation generally has a lot of those.

Fully agreed, we need to support both

[infohash]

What can be the expected version that will bring SQLAlchemy 2.0 support. I am making a service that will use SQLAlchemy 2.0 APIs. I will also integrate Airflow with that service. Based on ETA, I can decide whether to use SQLAlchemy 2.0 or go ahead with SQLAlchemy 1.4 and not wait for the supported version of Airflow.

[uranusjr]

It can be the next version if you help out on it 🙂

[auvipy]

I want to contribute here

[auvipy]

SQLA 1.4 & 2.0 both can be supported at the same time.

[moiseenkov]

Hi everyone,
Do we have some kind of board or issues list that are currently in progress? I see the work is going on, so I would like to help and fix some part that nobody else is working on at the moment?
@potiuk

[VladaZakharova]

@kaxil @phanikumv @uranusjr @hussein-awala
Maybe you could help with coordination here? Do you have any plan for the process here?
Thanks!

[phanikumv]

Hi @VladaZakharova - we are currently suspending this effort till 2.7 is released. We have already completed the changes, but will keep the PR in draft mode until then. The other pending change is to fix the tests so that they use the sqlalchemy 2.0 style (but again they need to wait till 2.7 is released)

cc @jedcunningham