amoghrajesh opened a new pull request, #67715:
URL: https://github.com/apache/airflow/pull/67715
<!-- SPDX-License-Identifier: Apache-2.0
https://www.apache.org/licenses/LICENSE-2.0 -->
<!--
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: #ISSUE
-->
---
##### 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".
-->
- [x] Yes - claude sonnet 4.6
<!--
Generated-by: [Tool Name] following [the
guidelines](https://github.com/apache/airflow/blob/main/contributing-docs/05_pull_requests.rst#gen-ai-assisted-contributions)
-->
related to: https://github.com/apache/airflow/issues/24171
similar to: https://github.com/apache/airflow/pull/65991
### What problem are we solving?
When running spark applications via Airflow in Kubernetes cluster mode,
`spark-submit` blocks for the full job duration — the JVM runs and just polls
pod phase in a loop, holding `~500 MB` of heap for the entire job lifetime.
This is the same problem fixed for YARN in #65991 (yarn_track_via_rm_api). It
also prevents the operator from ever being made deferrable or crash-recoverable
via `ResumableJobMixin`.
### Current behaviour
`SparkSubmitOperator` in K8s cluster mode keeps the `spark-submit` JVM alive
until the driver pod finishes. The JVM does no actual work after pod creation —
it just watches pod status — yet holds ~500 MB of heap for the entire job
duration (minutes to hours).
### Proposed change
Introduces an opt-in flag on `SparkSubmitOperartor`, ie:
`track_driver_via_k8s_api: bool = False` flag (default False — existing K8s
users are unaffected). When set, the hook:
- Injects `spark.kubernetes.submission.waitAppCompletion=false` so
spark-submit exits ~5–10s after the driver pod is created instead of blocking
for the full job duration
- Captures the driver pod name from the submission ID `spark:<pod-name>` log
line emitted by Spark's Client class (the only output before spark-submit exits)
- Polls pod phase via the Python K8s client until terminal state, logging
`Application status for {app_id} (phase: {phase})` to mirror the existing
`LoggingPodStatusWatcherImpl` output
- Deletes the driver pod on success; leaves it alive on failure for log
inspection
Validation rejects the flag if the Spark master is not K8s, deploy_mode is
not cluster, or the user has explicitly set `waitAppCompletion=true` in their
conf (which would silently nullify the flag).
### Testing
- Spun up a kind cluster to test this
- Defined Airflow Connection:
```shell
K8S_SERVER=$(kubectl config view --minify -o
jsonpath='{.clusters[0].cluster.server}')
airflow connections add spark_default \
--conn-type spark \
--conn-host "k8s://${K8S_SERVER}" \
--conn-extra '{"deploy-mode": "cluster", "namespace": "spark"}'
```
#### Using this DAG first with the flag set to false / not set:
```python
original = SparkSubmitOperator(
task_id="submit_long_running_job",
conn_id="spark_default",
application="local:///opt/spark/examples/jars/spark-examples_2.12-3.5.3.jar",
java_class="org.apache.spark.examples.SparkPi",
application_args=["100000"],
conf={
"spark.kubernetes.container.image": "apache/spark:3.5.3",
"spark.kubernetes.authenticate.driver.serviceAccountName":
"spark",
},
retries=1,
retry_delay=datetime.timedelta(seconds=5),
)
```
Logs here:
[original.txt](https://github.com/user-attachments/files/28395810/original.txt)
#### Using this DAG next with flag set to true:
```python
new_way = SparkSubmitOperator(
task_id="submit_long_running_job",
conn_id="spark_default",
application="local:///opt/spark/examples/jars/spark-examples_2.12-3.5.3.jar",
java_class="org.apache.spark.examples.SparkPi",
application_args=["100000"],
conf={
"spark.kubernetes.container.image": "apache/spark:3.5.3",
"spark.kubernetes.authenticate.driver.serviceAccountName":
"spark",
},
retries=1,
retry_delay=datetime.timedelta(seconds=5),
track_driver_via_k8s_api=True,
)
```
<img width="2547" height="1217" alt="image"
src="https://github.com/user-attachments/assets/0b719727-1410-4edf-93af-afc2c9a2a61a";
/>
Logs:
[new_way.txt](https://github.com/user-attachments/files/28395839/new_way.txt)
---
* 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`, in
[airflow-core/newsfragments](https://github.com/apache/airflow/tree/main/airflow-core/newsfragments).
You can add this file in a follow-up commit after the PR is created so you
know the PR number.
--
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]
