blinkseb opened a new pull request, #30392:
URL: https://github.com/apache/airflow/pull/30392

   <!--
   Thank you for contributing! Please make sure that your code changes
   are covered with tests. And in case of new features or big changes
   remember to adjust the documentation.
   
   Feel free to ping committers for the review!
   
   In case of an existing issue, reference it using one of the following:
   
   closes: #ISSUE
   related: #ISSUE
   
   How to write a good git commit message:
   http://chris.beams.io/posts/git-commit/
   -->
   
   Hello,
   
   While debugging an issue on why our DAGs were running without tasks (issue 
https://github.com/apache/airflow/issues/18880), I found a possible "race 
condition" in `DagBag.get_dag`.
   
   Since DAGs are serialized and written to database in a different process 
than the scheduler, it's possible to end up in a situation where the serialized 
DAG `last_updated` field is before the DagBag `last_fetched` datetime, while 
the cached version in the DagBag is not the latest one. Once this situation is 
reached, the only was to refresh the DAG is either to modify it, or to restart 
the scheduler.
   
   The diagram below should help to illustrate the problem:
   
   ```mermaid
   sequenceDiagram
       DagProcessor->>DagProcessor: Parse and collect DAGs
       activate DagProcessor
       Note left of DagProcessor: Begin Transaction
       DagProcessor->>DagProcessor: Serialize DAG A (last_updated = 10:00:00)
       DagProcessor->>DagProcessor: serialize DAG B (last_updated = 10:00:01)
       DagProcessor->>DagProcessor: serialize DAG C (last_updated = 10:00:02)
       par
           DagProcessor->>DagProcessor: serialize DAG D (last_updated = 
10:00:03)
       and
           Scheduler->>Scheduler: read DAG A (last_fetched = 10:00:03)
           Note right of Scheduler: previous version of DAG A was fetched
       end
       DagProcessor->>DagProcessor: serialize DAG E (last_updated = 10:00:04)
       Note left of DagProcessor: End Transaction
       deactivate DagProcessor
   ```
   
   If you have a lot of complex DAGs in a single file, the serialization step 
can take a long time. If, during this time, the scheduler has to run a DAG 
currently being serialized, you can have a situation where the `last_fetched` 
value of the DagBag is after the `last_updated` value of the serialized DAG 
while still having the previous version of the DAG in memory.
   
   This PR fixes the issue by also checking the serialized DAG hash in addition 
to the last updated datetime. I'm wondering if maybe we can just use the hash 
here instead of the updated datetime, but I'm not sure of the possible 
side-effects.
   
   I've added a new unit tests replicating the issue. Without the patch, the 
test fails.
   
   Thanks,
   
   ---
   **^ Add meaningful description above**
   
   Read the **[Pull Request 
Guidelines](https://github.com/apache/airflow/blob/main/CONTRIBUTING.rst#pull-request-guidelines)**
 for more information.
   In case of fundamental code changes, an Airflow Improvement Proposal 
([AIP](https://cwiki.apache.org/confluence/display/AIRFLOW/Airflow+Improvement+Proposals))
 is needed.
   In case of a new dependency, check compliance with the [ASF 3rd Party 
License Policy](https://www.apache.org/legal/resolved.html#category-x).
   In case of backwards incompatible changes please leave a note in a 
newsfragment file, named `{pr_number}.significant.rst` or 
`{issue_number}.significant.rst`, in 
[newsfragments](https://github.com/apache/airflow/tree/main/newsfragments).
   


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]

Reply via email to