[AIRFLOW-2761] Parallelize enqueue in celery executor

[KevinYang21]

Jira

• My PR addresses the following Airflow Jira issues and references them in the PR title.

Description

• Here are some details about my PR, including screenshots of any UI changes:
  The change is depending on this PR: [AIRFLOW-2760] Decouple DAG parsing loop from scheduler loop #3873

Summary of major changes:

• Celery executor will create a process pool to enqueue celery task in parallel.
• Scheduler job will not pass TaskInstance to executor when trying to queue command, instead it passes SimpleTaskInstance, in order to enable parallel enqueuing.
• Scheduler no longer checks task instance state before enqueuing because worker already checks task instance state before executing. Our cluster has been requeuing everything for a long time and there should be no risk removing that redundant check.

Tests

• My PR adds the following unit tests OR does not need testing for this extremely good reason:

tests/executors/test_celery_executor.py:TestCeleryExecutor.test_error_sending_task
tests/jobs.py:SchedulerJobTest.test_change_state_for_tasks_failed_to_execute

Also updated existing failing unit test.

The change has been running in Airbnb internal cluster for 3 months+.

Before:
Before (32k tasks should run on 13:30 but didn't get >30k running tasks until 13:41):

After (32k tasks should run on 15:10 and got them all running at 15:14):

Commits

• My commits all reference Jira issues in their subject lines, and I have squashed multiple commits if they address the same issue. In addition, my commits follow the guidelines from "How to write a good git commit message":
  1. Subject is separated from body by a blank line
  2. Subject is limited to 50 characters (not including Jira issue reference)
  3. Subject does not end with a period
  4. Subject uses the imperative mood ("add", not "adding")
  5. Body wraps at 72 characters
  6. Body explains "what" and "why", not "how"

Documentation

• In case of new functionality, my PR adds documentation that describes how to use it.
  • When adding new operators/hooks/sensors, the autoclass documentation generation needs to be added.

Code Quality

• Passes git diff upstream/master -u -- "*.py" | flake8 --diff

[codecov-io]

Codecov Report

Merging #4234 into master will increase coverage by 0.01%.
The diff coverage is 82.48%.

@@            Coverage Diff             @@
##           master    #4234      +/-   ##
==========================================
+ Coverage   77.82%   77.84%   +0.01%     
==========================================
  Files         201      201              
  Lines       16367    16455      +88     
==========================================
+ Hits        12738    12809      +71     
- Misses       3629     3646      +17

Impacted Files	Coverage Δ
airflow/utils/timeout.py	78.57% <100%> (+0.79%)	⬆️
airflow/models.py	92.29% <100%> (ø)	⬆️
airflow/executors/base_executor.py	95.16% <100%> (+1.41%)	⬆️
airflow/executors/celery_executor.py	76.92% <68.51%> (-3.69%)	⬇️
airflow/utils/dag_processing.py	59.53% <86.11%> (+1.67%)	⬆️
airflow/jobs.py	77.66% <93.93%> (+0.29%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 2fd409d...554b4cf. Read the comment docs.

[KevinYang21]

Finally got this long overdue PR out :D
@Fokko @ashb @feng-tao @XD-DENG @kaxil @saguziel @aoen @YingboWang PTAL

[ashb]

Awesome, always love the detail in your PR descriptions 💖

[XD-DENG]

Hi @KevinYang21 , haven’t got time to check details now. A few very minor points above first.

[kaxil]

@Fokko Thoughts on including this to 1.10.2 ???

[KevinYang21]

I would vote +1 to include so that we have a complete story around scaling in 1.10.2

[feng-tao]

hey @KevinYang21 , I discussed with Max and heard that Airbnb run 2000+ dag in prod daily. I wonder what's the setup look like? Do you guys do multi airflow clusters?

[feng-tao]

BTW, pr nicely done!

[KevinYang21]

@feng-tao ty and ty for the reviews you have done. We have ~1600 active DAGs after some recent pruning and we don't do multi cluster. I shared some of our setup in a thread from the dev mail list. I'm happy to share more details if you'd like. In fact I think it would be a good idea for us to have a short meet up( maybe lunch on either side) to discuss more efficiently. Or if the schedule is tight we can do a webex too.

[feng-tao]

thanks @KevinYang21 . Yeah, let me go back and check with the team. It definitely helps to share and learn more from Airbnb's experience.

[will-beta]

@KevinYang21 hi, how do you guys get >30k running tasks in Airbnb internal cluster?
We can only get 20 or so running tasks in our internal cluster, because the 16-vCores PostgreSQL server our airflow is using keeps at 90% CPU consumption all the time, and we don't know how to make this down a little bit.
So curious!
Thanks in advance!