Do not create dagruns for DAGs with import errors #19367

ephraimbuddy · 2021-11-02T13:36:57Z

An active dag can suddenly have import errors as a result of the DAG file being changed. Currently, we
do not consider this before creating dagruns.

This PR adds the has_import_errors to dag_model so dags with import errors are not set to the scheduler to
create dagruns. Also, dag runs can no longer be created on the UI/API when the dag has import errors

^ Add meaningful description above

Read the Pull Request Guidelines for more information.
In case of fundamental code change, Airflow Improvement Proposal (AIP) is needed.
In case of a new dependency, check compliance with the ASF 3rd Party License Policy.
In case of backwards incompatible changes please leave a note in UPDATING.md.

github-actions · 2021-11-02T17:13:02Z

The PR most likely needs to run full matrix of tests because it modifies parts of the core of Airflow. However, committers might decide to merge it quickly and take the risk. If they don't merge it quickly - please rebase it to the latest main at your convenience, or amend the last commit of the PR, and push it with --force-with-lease.

kaxil

Can we add a test for this case please?

Separately let's check why max_active_runs is not set via tests too

ephraimbuddy · 2021-11-03T16:50:57Z

Can we add a test for this case please?

Separately let's check why max_active_runs is not set via tests too

I can't figure a test to add.
My guess for the issue is that they removed max_active_runs_per_dag from their configuration.
To check this, I have added a fallback and also updated dag's test to ensure once DagModel is instantiated, it'll have a max_active_runs

airflow/models/dag.py

ephraimbuddy · 2021-11-03T21:14:09Z

I discovered that this happens because the DAG has import error and is active after upgrading. So scheduler was trying to create dagruns for it.
I have updated the code and commit message.
cc: @jedcunningham @kaxil

Significant changes since I approved.

airflow/models/dag.py

ashb · 2021-11-04T10:57:27Z

This feels like only a partial fix.

If a DAG is currently working (and has max_active_runs) and then the file is updated and now has an error, the DAG will still be scheduled, but it probably shouldn't be.

To do that without needing to join against ImportError perhaps we need to add a new column to DAG -- we have is_active which is close, but setting that to false hides it from the UI which isn't the behaviour we want here.

ephraimbuddy · 2021-11-04T22:39:47Z

This feels like only a partial fix.

If a DAG is currently working (and has max_active_runs) and then the file is updated and now has an error, the DAG will still be scheduled, but it probably shouldn't be.

To do that without needing to join against ImportError perhaps we need to add a new column to DAG -- we have is_active which is close, but setting that to false hides it from the UI which isn't the behaviour we want here.

I have updated it and it looks good. Thanks for the review!!

tests/models/test_dag.py

airflow/dag_processing/processor.py

ephraimbuddy · 2021-11-11T12:02:30Z

Happy to discuss about if we still need the max_active_runs condition or not too.

I agree we no longer need it.

I restricted the creation of dagruns on the UI and API too. Let me know if I should remove these restrictions

ashb · 2021-11-11T15:05:20Z

Happy to discuss about if we still need the max_active_runs condition or not too.

I agree we no longer need it.

I restricted the creation of dagruns on the UI and API too. Let me know if I should remove these restrictions

If you want to reduce duplication, but keep the rollback you can go from this:

     def test_dags_needing_dagruns_not_too_early(self):
         ...
        session = settings.Session()

to this:

     def test_dags_needing_dagruns_not_too_early(self, session):

And remove the session = settings.Session(). That session pytest fixture defined in tests/conftest.py already does the rollback.

But that should probably be a separate change to this PR.

ephraimbuddy · 2021-11-11T15:50:29Z

Happy to discuss about if we still need the max_active_runs condition or not too.

I agree we no longer need it.
I restricted the creation of dagruns on the UI and API too. Let me know if I should remove these restrictions

If you want to reduce duplication, but keep the rollback you can go from this:
     def test_dags_needing_dagruns_not_too_early(self):
         ...
        session = settings.Session()
to this:
     def test_dags_needing_dagruns_not_too_early(self, session):
And remove the session = settings.Session(). That session pytest fixture defined in tests/conftest.py already does the rollback.

But that should probably be a separate change to this PR.

That's true, will work on it

An active dag can suddenly have import errors as a result of the DAG file being changed. Currently, we do not consider this before creating dagruns. This PR adds the has_import_errors to dagmodel so dags with import errors are not set to the scheduler to create dagruns fixup! Add has_import_errors to DagModel and ensure Dags with import errors are not sent to the Scheduler fixup! fixup! Add has_import_errors to DagModel and ensure Dags with import errors are not sent to the Scheduler Update airflow/models/dag.py Co-authored-by: Ash Berlin-Taylor <[email protected]> Update docs/apache-airflow/migrations-ref.rst Co-authored-by: Ash Berlin-Taylor <[email protected]> Apply suggestions from code review Co-authored-by: Ash Berlin-Taylor <[email protected]> fixup! Apply suggestions from code review restrict manual dagrun creation if dagrun has import errors fixup! restrict manual dagrun creation if dagrun has import errors use session rollback

ashb · 2021-11-22T13:50:25Z

docs/apache-airflow/migrations-ref.rst

@@ -23,7 +23,9 @@ Here's the list of all the Database Migrations that are executed via when you ru
 +--------------------------------+------------------+-----------------+---------------------------------------------------------------------------------------+
 | Revision ID                    | Revises ID       | Airflow Version | Description                                                                           |
 +--------------------------------+------------------+-----------------+---------------------------------------------------------------------------------------+
-| ``7b2661a43ba3`` (head)        | ``142555e44c17`` | ``2.2.0``       | Change ``TaskInstance`` and ``TaskReschedule`` tables from execution_date to run_id.  |
+| ``be2bfac3da23`` (head)        | ``7b2661a43ba3`` | ``2.2.3``       | Add has_import_errors column to DagModel                                              |


I'm tempted to say this should be 2.3.0 -- it's maybe a bug fix, but this behaviour has been the same for all of 2.x so far.

What do others think here?

I am for adding it to 2.2.3. There is hardly a use of DagRuns for DAGs with import errors, especially that you could not see whether there were errors via API. Even if it requires data model change, that's fine as patch-level. This is OK for a bugfix to require new field in the DB.

Since this issue will crash the scheduler, I lean toward 2.2.3.

ashb

LGTM now, just the 2.2.3 vs 2.3 decision.

tests/dag_processing/test_processor.py

tests/models/test_dag.py

Co-authored-by: Jed Cunningham <[email protected]>

uranusjr · 2021-11-23T10:15:49Z

Do we really need a new field? I think we can instead join ImportError on import_error.filename = dag.fileloc and error if the result is not empty.

ephraimbuddy · 2021-11-23T13:16:03Z

Do we really need a new field? I think we can instead join ImportError on import_error.filename = dag.fileloc and error if the result is not empty.

It looks like it'll not be that efficient. See
#19367 (comment)

Not sure the impact the join would have.
cc: @ashb

ashb · 2021-11-23T13:26:18Z

Yeah, we could do it with a join, but this is in the hot path for the main scheduler loop so it seems better to denormalise it here.

uranusjr · 2021-11-23T20:33:16Z

Denormalise to optimise makes sense.

Fixed.

An active dag can suddenly have import errors as a result of the DAG file being changed. Currently, we do not consider this before creating dagruns. This PR adds the has_import_errors to dagmodel so dags with import errors are not sent to the scheduler to create dagruns Co-authored-by: Ash Berlin-Taylor <[email protected]>

An active dag can suddenly have import errors as a result of the DAG file being changed. Currently, we do not consider this before creating dagruns. This PR adds the has_import_errors to dagmodel so dags with import errors are not sent to the scheduler to create dagruns Co-authored-by: Ash Berlin-Taylor <[email protected]> (cherry picked from commit 3ccb794)

ephraimbuddy requested review from ashb, kaxil and XD-DENG as code owners November 2, 2021 13:36

boring-cyborg bot added the area:Scheduler including HA (high availability) scheduler label Nov 2, 2021

jedcunningham added this to the Airflow 2.2.2 milestone Nov 2, 2021

jedcunningham previously approved these changes Nov 2, 2021

View reviewed changes

github-actions bot added the full tests needed We need to run full set of tests for this PR to merge label Nov 2, 2021

kaxil reviewed Nov 2, 2021

View reviewed changes

ephraimbuddy force-pushed the use-dag-max-active-runs branch from f186abd to e5b0112 Compare November 3, 2021 16:46

jedcunningham reviewed Nov 3, 2021

View reviewed changes

airflow/models/dag.py Outdated Show resolved Hide resolved

ephraimbuddy force-pushed the use-dag-max-active-runs branch from e5b0112 to 4ffb9d2 Compare November 3, 2021 21:10

ephraimbuddy changed the title ~~Use dag.max_active_runs instead of dag_model.max_active_runs~~ Do not send DAGs with import errors to the scheduler Nov 3, 2021

ephraimbuddy requested a review from jedcunningham November 3, 2021 21:14

ephraimbuddy force-pushed the use-dag-max-active-runs branch from 4ffb9d2 to 5b49c57 Compare November 3, 2021 21:22

kaxil reviewed Nov 3, 2021

View reviewed changes

airflow/models/dag.py Outdated Show resolved Hide resolved

ephraimbuddy requested a review from kaxil November 4, 2021 00:18

ashb requested changes Nov 4, 2021

View reviewed changes

airflow/models/dag.py Outdated Show resolved Hide resolved

airflow/models/dag.py Outdated Show resolved Hide resolved

ephraimbuddy force-pushed the use-dag-max-active-runs branch from 575cc31 to bcb3822 Compare November 4, 2021 10:32

ephraimbuddy changed the title ~~Do not send DAGs with import errors to the scheduler~~ Do not send DAGs with null DagModel.max_active_runs to the scheduler Nov 4, 2021

ephraimbuddy force-pushed the use-dag-max-active-runs branch from bcb3822 to 2e3913e Compare November 4, 2021 22:25

ephraimbuddy requested a review from ashb November 4, 2021 22:39

ephraimbuddy commented Nov 4, 2021

View reviewed changes

tests/models/test_dag.py Outdated Show resolved Hide resolved

kaxil mentioned this pull request Nov 5, 2021

dag.max_active_runs is nullable but the scheduler dies if it is null #19415

Closed

2 tasks

jedcunningham reviewed Nov 5, 2021

View reviewed changes

airflow/dag_processing/processor.py Outdated Show resolved Hide resolved

ephraimbuddy force-pushed the use-dag-max-active-runs branch from 937c592 to 895b152 Compare November 11, 2021 15:49

ephraimbuddy force-pushed the use-dag-max-active-runs branch from 895b152 to 4a4e513 Compare November 22, 2021 13:22

ephraimbuddy requested review from jedcunningham and ashb November 22, 2021 13:22

ashb reviewed Nov 22, 2021

View reviewed changes

jedcunningham reviewed Nov 22, 2021

View reviewed changes

tests/dag_processing/test_processor.py Outdated Show resolved Hide resolved

tests/dag_processing/test_processor.py Outdated Show resolved Hide resolved

tests/models/test_dag.py Show resolved Hide resolved

tests/models/test_dag.py Outdated Show resolved Hide resolved

ephraimbuddy and others added 3 commits November 22, 2021 19:16

Apply suggestions from code review

925fee0

Co-authored-by: Jed Cunningham <[email protected]>

fixup! Apply suggestions from code review

11dfc7e

fixup! Apply suggestions from code review

59b56cf

jedcunningham approved these changes Nov 22, 2021

View reviewed changes

uranusjr approved these changes Nov 23, 2021

View reviewed changes

ephraimbuddy requested a review from ashb November 24, 2021 16:36

ephraimbuddy merged commit 3ccb794 into apache:main Nov 29, 2021

ephraimbuddy deleted the use-dag-max-active-runs branch November 29, 2021 10:21

jedcunningham added the type:bug-fix Changelog: Bug Fixes label Dec 8, 2021

jedcunningham mentioned this pull request Dec 10, 2021

Status of testing of Apache Airflow 2.2.3rc2 #20208

Closed

38 tasks

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Do not create dagruns for DAGs with import errors #19367

Do not create dagruns for DAGs with import errors #19367

ephraimbuddy commented Nov 2, 2021 •

edited

Loading

github-actions bot commented Nov 2, 2021

kaxil left a comment

ephraimbuddy commented Nov 3, 2021

ephraimbuddy commented Nov 3, 2021

ashb commented Nov 4, 2021

ephraimbuddy commented Nov 4, 2021

ephraimbuddy commented Nov 11, 2021

ashb commented Nov 11, 2021

ephraimbuddy commented Nov 11, 2021

ashb Nov 22, 2021

potiuk Nov 22, 2021

jedcunningham Nov 22, 2021

ashb left a comment

uranusjr commented Nov 23, 2021

ephraimbuddy commented Nov 23, 2021

ashb commented Nov 23, 2021

uranusjr commented Nov 23, 2021

Do not create dagruns for DAGs with import errors #19367

Do not create dagruns for DAGs with import errors #19367

Conversation

ephraimbuddy commented Nov 2, 2021 • edited Loading

github-actions bot commented Nov 2, 2021

kaxil left a comment

Choose a reason for hiding this comment

ephraimbuddy commented Nov 3, 2021

ephraimbuddy commented Nov 3, 2021

ashb commented Nov 4, 2021

ephraimbuddy commented Nov 4, 2021

ephraimbuddy commented Nov 11, 2021

ashb commented Nov 11, 2021

ephraimbuddy commented Nov 11, 2021

ashb Nov 22, 2021

Choose a reason for hiding this comment

potiuk Nov 22, 2021

Choose a reason for hiding this comment

jedcunningham Nov 22, 2021

Choose a reason for hiding this comment

ashb left a comment

Choose a reason for hiding this comment

uranusjr commented Nov 23, 2021

ephraimbuddy commented Nov 23, 2021

ashb commented Nov 23, 2021

uranusjr commented Nov 23, 2021

ephraimbuddy commented Nov 2, 2021 •

edited

Loading