dabla opened a new pull request, #60651:
URL: https://github.com/apache/airflow/pull/60651

   <!--
   Thank you for contributing!
   
   Please provide above a brief description of the changes made in this pull 
request.
   Write a good git commit message following this guide: 
http://chris.beams.io/posts/git-commit/
   
   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 (in general) for the review if you do not see reaction for 
a few days
   (72 Hours is the minimum reaction time you can expect from volunteers) - we 
sometimes miss notifications.
   
   In case of an existing issue, reference it using one of the following:
   
   * closes: #ISSUE
   * related: [#60330](https://github.com/apache/airflow/pull/60330)
   -->
   
   ---
   
   ##### Was generative AI tooling used to co-author this PR?
   
   <!--
   If generative AI tooling has been used in the process of authoring this PR, 
please
   change below checkbox to `[X]` followed by the name of the tool, uncomment 
the "Generated-by".
   -->
   
   - [ ] Yes (please specify the tool below)
   
   <!--
   Generated-by: [Tool Name] following [the 
guidelines](https://github.com/apache/airflow/blob/main/contributing-docs/05_pull_requests.rst#gen-ai-assisted-contributions)
   -->
   
   **Motivation**
   
   Recently, we encountered an issue where TaskInstances were being prematurely 
killed by the scheduler.
   
   We initially tried the fix proposed in PR 
[#60330](https://github.com/apache/airflow/pull/60330) made by @ephraimbuddy, 
but unfortunately this did not help in our case. After further investigation, 
we discovered that this behaviour only occurred in DAGs using the WinRMOperator.
   
   **Problem**
   
   We use the WinRMOperator to launch remote processes on Windows servers. Some 
of these processes can take a significant amount of time to complete.
   
   The root cause is that the WinRMOperator currently performs polling 
synchronously inside the worker, via the run method of WinRMHook. This has 
several drawbacks:
   
   - The worker is blocked while waiting for the remote command to complete.
   - Long-running polling increases the risk of scheduler heartbeats being 
missed.
   - This can lead to TaskInstances being marked as failed or killed 
prematurely.
   
   Overall, this is not an efficient or scalable execution model in Airflow.
   
   **Solution**
   
   This PR refactors the WinRMOperator to support deferrable execution.
   
   When deferrable=True:
   
   - The worker is only responsible for launching the remote command on the 
Windows server.
   - The operator retrieves the shell_id and command_id from the WinRM session.
   - Polling for command output and the final result_code is deferred to a new 
WinRMCommandTrigger.
   - The triggerer performs this polling asynchronously, freeing the worker 
immediately.
   
   This aligns the WinRMOperator with Airflow’s recommended architecture for 
long-running or polling-based operations.
   
   **Benefits**
   
   Workers are no longer blocked by long-running WinRM commands.
   
   Polling is handled asynchronously by the triggerer.
   
   Prevents scheduler-induced task termination for long-running WinRM jobs.
   
   Improves scalability and resource utilization.
   
   Matches the deferrable operator pattern used across Airflow providers.
   
   **Example Usage**
   
   ```
   WinRMOperator(
       task_id="run_job",
       ssh_conn_id="ssh.winrm",
       command="python C:\\Temp\\python_scripts\\job_trigger.py -j TEST",
       pool="winrm",
       poll_interval=10,
       timeout=timedelta(minutes=360),  # Avoid infinite polling if something 
goes wrong
       deferrable=True,
   )
   ```
   
   **Conclussion**
   
   With this refactoring in place, we no longer experience TaskInstances being 
prematurely killed. Polling is handled asynchronously by the triggerer, which 
is the preferred and more robust approach for this type of workload in Airflow 
and we don't block workers for polling unnecessarily.
   
   ---
   
   * Read the **[Pull Request 
Guidelines](https://github.com/apache/airflow/blob/main/contributing-docs/05_pull_requests.rst#pull-request-guidelines)**
 for more information. Note: commit author/co-author name and email in commits 
become permanently public when merged.
   * For fundamental code changes, an Airflow Improvement Proposal 
([AIP](https://cwiki.apache.org/confluence/display/AIRFLOW/Airflow+Improvement+Proposals))
 is needed.
   * When adding dependency, check compliance with the [ASF 3rd Party License 
Policy](https://www.apache.org/legal/resolved.html#category-x).
   * For significant user-facing changes create newsfragment: 
`{pr_number}.significant.rst` or `{issue_number}.significant.rst`, in 
[airflow-core/newsfragments](https://github.com/apache/airflow/tree/main/airflow-core/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