Status of testing Providers that were prepared on August 08, 2026

[potiuk]

I have a kind request for all the contributors to the latest provider distributions release.
Could you please help us to test the RC versions of the providers?

The guidelines on how to test providers can be found in

Verify providers by contributors

Let us know in the comments, whether the issue is addressed.

Note

Both providers here are re-cuts of candidates that were pulled from the
2026-08-06 wave, so nearly everything below was already verified against
google==22.3.0rc3 / cncf-kubernetes==10.21.0rc2 in
#71246. Please carry those results
over rather than re-testing them. Only two things are genuinely new:

• google — Workaround bug in google-claude-aiplatform evaluation extra (Python 3… #71273, which makes the provider installable on Python 3.14
  (it now requires the google-cloud-aiplatform "evaluation" extra's
  dependencies directly; upstream bug Litellm limits are too constraining for Python 3.14 googleapis/python-aiplatform#7057).
• cncf.kubernetes — Fix version_added for the concurrent pod creation options #71157, which corrects version_added to 10.21.0
  for async_pod_creation and pod_creation_max_concurrency. Metadata
  only, no behaviour change.

The 2026-08-06 wave has since been released (44 providers); the two waves
share no providers, so nothing here was affected by it.

Important

Checklist updated 2026-08-09. 25 items already verified against the 2026-08-06 wave in
#71246 are now ticked here — the code behind them is
unchanged, so please do not re-test them.

The 7 items left unticked are the ones that genuinely need attention, above all:

• Workaround bug in google-claude-aiplatform evaluation extra (Python 3… #71273 — the Python 3.14 install fix that is the whole reason for this wave
• Add opt-in concurrent pod creation to KubernetesExecutor #68480 — whose version_added metadata was corrected by Fix version_added for the concurrent pod creation options #71157

These are providers that require testing as there were some substantial changes introduced:

Provider cncf.kubernetes: 10.21.0

• Add --min-completed-minutes to cleanup-pods to prevent KPO race condition (#70595): @noamst-monday
  Linked issues:
  • Linked Issue #69269: @noah-gil
• Add opt-in concurrent pod creation to KubernetesExecutor (#68480): @1fanwang
• Prevent deferrable KubernetesPodOperator log parsing from blocking the triggerer event loop (#69661): @aeroyorch
• Fix KubernetesPodOperator dry_run requiring live Kubernetes API client (#70234): @Aaryan123456679
  Linked issues:
  • Linked Issue #45812: @baryluk
• Build KubernetesInstallKueueOperator manifest URL after rendering (#70335): @1fanwang
  Linked issues:
  • Linked Issue #70296: @shahar1
• Mark KubernetesPodOperator and AgentOperator as durable capable (#70289): @amoghrajesh
• Validate KubernetesResourceBaseOperator yaml_conf after rendering (#70337): @1fanwang
  Linked issues:
  • Linked Issue #70296: @shahar1
• Document Kubernetes RBAC permissions (#68716): @dhkim1920
  Linked issues:
  • Linked Issue #40210: @pykenny
• Documenting effect if task state store cleanup for KPO (#70722): @amoghrajesh

Provider google: 22.3.0

• Emit per-statement OpenLineage events for BigQuery script jobs (#69234): @aaron-y-chen
  Linked issues:
  • Linked Issue #47579: @luke-hoffman1
• Add Vertex AI Agent Engine operators (Create, Get, Query, Update, Delete) (#68479): @AlejandroMorgante
• Adding durable execution to BigQueryInsertJobOperator (#69542): @amoghrajesh
• Register Stackdriver remote task logging handler (#70549): @fat-catTW
  Linked issues:
  • Linked Issue #70267: @jason810496
• Add CloudSQLNoOperationInProgressSensor for parallel admin ops (#68151): @Vamsi-klu
  Linked issues:
  • Linked Issue #68040: @radhwene
  • Linked Issue #68361: @radhwene
• Enable provider-based GCS remote logging resolution (#70504): @aaron-y-chen
  Linked issues:
  • Linked Issue #67056: @jason810496
  • Linked Issue #70265: @jason810496
  • Linked Issue #70266: @jason810496
• Make schema_fields templated in GCSToBigQueryOperator (#69108): @nikolauspschuetz
  Linked issues:
  • Linked Issue #31481: @quantumlicht
• Fix Google Links for Sovereign Cloud (#69805): @molcay
• Fix Google Cloud Batch error messages to include job name (#69783): @parkhojeong
• Validate Gemini batch results_folder in the worker that writes results (#70393): @shahar1
  Linked issues:
  • Linked Issue #70362: @daesung9511
• Fail deferred Cloud Composer tasks when the GCP operation errors (#70430): @SEPURI-SAI-KRISHNA
• Apply the team scope in the Google Secret Manager backend (#70869): @potiuk
• Delay BigQueryToMsSqlOperator source table parsing (#70493): @fat-catTW
  Linked issues:
  • Linked Issue #70296: @shahar1
• Emit GCSToGCSOperator deprecation warnings after rendering (#70449): @haseebmalik18
  Linked issues:
  • Linked Issue #70296: @shahar1
• Emit GCSListObjectsOperator delimiter deprecation warning after rendering (#70533): @yuseok89
  Linked issues:
  • Linked Issue #70296: @shahar1
• Move Compute Engine validation out of constructors (#70454): @daesung9511
  Linked issues:
  • Linked Issue #70296: @shahar1
• Prevent premature validation in Gemini batch job operators (#70362): @daesung9511
  Linked issues:
  • Linked Issue #70296: @shahar1
• Only refuse team scoped like secret ids when multi_team is on (#71078): @amoghrajesh
• Strip only the literal properties/ prefix from GA property link IDs (#70919): @Eason09053360
• Update system test to the latest api version of google-ads (#69865): @olegkachur-e
• Drop redundant BigQueryInsertJobOperator job ID snapshot in init (#70494): @fat-catTW
  Linked issues:
  • Linked Issue #70296: @shahar1
• Workaround bug in google-claude-aiplatform evaluation extra (Python 3… (#71273): @potiuk
• Document effect of state-store cleanup for operators with durable execution (#70721): @amoghrajesh

All users involved in the PRs:
@SEPURI-SAI-KRISHNA @1fanwang @aaron-y-chen @AlejandroMorgante @daesung9511 @potiuk @noamst-monday @nikolauspschuetz @parkhojeong @Eason09053360 @Vamsi-klu @haseebmalik18 @amoghrajesh @aeroyorch @olegkachur-e @Aaryan123456679 @shahar1 @dhkim1920 @fat-catTW @yuseok89 @molcay

[SEPURI-SAI-KRISHNA]

Tested apache-airflow-providers-google==22.3.0rc4 for #70430 (Fail deferred Cloud Composer tasks when the GCP operation errors) — works as expected.

Re-ran everything against the re-cut rather than only carrying over my rc3 result from #71246 (comment). Installed the RC wheel from PyPI into a clean Python 3.12.3 virtualenv with Airflow 3.3.0; pip check is clean. Wheel SHA-256 is f1a1476759aa3177ec171ec3810bb8dacb8b4057b5864153148a09297fab213a, matching the PyPI artifact.

Unit tests against the installed RC — 6 passed, including the two added by the PR:

TestCloudComposerExecutionTrigger::test_run_raises_when_operation_finished_with_error PASSED
TestCloudComposerExecutionTrigger::test_run_yields_event_when_operation_finished_without_error PASSED

Behavioural check with real LRO objects. The unit tests use a stand-in, so I also drove the trigger with genuine google.longrunning.operations_pb2.Operation values:

Scenario	rc4
done=True with error.message set	raises AirflowException: Cloud Composer Environment error: quota exceeded
done=True, no error	yields {'operation_name': ..., 'operation_done': True}
polls done=False twice, then finishes with error	raises after polling

Control against the last released provider. I installed 22.2.2 (pre-fix) in a separate venv and ran the identical scenarios. The bug reproduces there — a failed operation yields a success event:

CASE1 BUG REPRODUCED: failed operation yielded SUCCESS -> {'operation_name': 'op', 'operation_done': True}
CASE3 BUG REPRODUCED: poll-then-fail yielded SUCCESS -> {'operation_name': 'op', 'operation_done': True}

and the PR's regression test fails on it:

test_run_raises_when_operation_finished_with_error FAILED - Failed: DID NOT RAISE AirflowException

So the change is confirmed to be the thing that fixes it, not an incidental pass.

For completeness, airflow/providers/google/cloud/triggers/cloud_composer.py is byte-identical between rc3 and rc4 (sha256 4990060e9847b783c1739115292990bb3e125626984ed7ea889bfbd0bb06f4b8), so the re-cut did not disturb this change.

I have not tested the Python 3.14 installability fix (#71273) — no 3.14 interpreter available here — and I don't have a live Composer environment, so this is verification against the released artifact rather than a real environment creation/update failure.

---

Drafted-by: Claude Code (Opus 5); reviewed by @SEPURI-SAI-KRISHNA before posting

[aaron-y-chen]

All of my parts are working well on my local machine :)

[nikolauspschuetz]

Tested apache-airflow-providers-google==22.3.0rc3 for #69108 (make schema_fields templated in GCSToBigQueryOperator).

Installed the RC from PyPI into a clean Python 3.13 venv against apache-airflow==3.3.0, verified against the installed artifact:

• schema_fields is now in GCSToBigQueryOperator.template_fields with a json renderer, and
• a templated value (schema_fields="{{ var.value.schema_fields }}") renders to the expected list at runtime (render_template_as_native_obj).

Works as expected. 👍

_{Drafted-by: Claude Code (Opus 4.8) — I ran the verification and reviewed the result.}

[potiuk]

The 2026-08-08 wave has been released.

Both providers are now available on PyPI:

• apache-airflow-providers-google==22.3.0
• apache-airflow-providers-cncf-kubernetes==10.21.0

The vote passed with 4 binding +1 votes and no -1. Thanks to everyone who tested these release candidates — and in particular for catching the Python 3.14 installation problem in google 22.3.0rc3 that led to this follow-up wave.

---

Drafted-by: Claude Code (Opus 5); reviewed by @potiuk before posting