Status of testing Providers that were prepared on September 09, 2026 #72902

Activity

1. vincbeck

   added

   kind:metaHigh-level information important to the communityHigh-level information important to the community

   testing statusStatus of testing releasesStatus of testing releases

   on Sep 10, 2026
2. vincbeck

   changed the title [-]Status of testing Providers that were prepared on <MONTH DD, YYYY>[/-] [+]Status of testing Providers that were prepared on September 10, 2026[/+] on Sep 10, 2026
3. vincbeck

   changed the title [-]Status of testing Providers that were prepared on September 10, 2026[/-] [+]Status of testing Providers that were prepared on September 09, 2026[/+] on Sep 10, 2026
4. 1fanwang commented on Sep 10, 2026

   

   1fanwang

   on Sep 10, 2026

   Contributor

   More actions

   Tested my changes working for on Python 3.12.14 and Airflow 3.3.1 on apache-airflow-providers-databricks==7.20.0rc1 : the behavior in #70831, related to #70340, is correct. A Dag that passes both  statement  and  statement_id  now fails during parsing with  ValueError: Provide exactly one of statement or statement_id. , including  statement_id=""  and a template that renders to  None . Both cases executed  SELECT 1  with 7.19.0. A statement-only control still ran successfully. A lone statement template that rendered to  None  failed after rendering with  ValueError: One of either statement or statement_id must be provided. 
5. 1fanwang commented on Sep 10, 2026

   

   1fanwang

   on Sep 10, 2026

   Contributor

   More actions

   Tested my changes working for on Python 3.12.14 and Airflow 3.3.1 on apache-airflow-providers-apache-iceberg==2.1.1rc1 : #72312 correctly supersedes #72173. Using a real  AssetStateStoreAccessors  and a real local Iceberg catalog and table, a restarted watcher at the stored snapshot emitted nothing. After a new snapshot, it emitted once and persisted the new watermark. The Airflow 3.3.1 accessor has no  aget , and the RC did not call it.
6. Priyaj11 commented on Sep 10, 2026

   

   Priyaj11

   on Sep 10, 2026

   Contributor

   More actions

   Tested amazon 9.36.0rc1 for #72500 (EmrContainerSensor reporting success or an unknown job state). Installed apache-airflow-providers-amazon==9.36.0rc1 into an Airflow 3.1.3
   virtualenv (Python 3.12) and exercised EmrContainerSensor.poke() against each state path with the hook mocked:
   COMPLETED -> True
   RUNNING -> False
   FAILED -> raises AirflowException
   SOME_FUTURE_STATE -> False
   None -> False
   The last two are the fix: an unrecognised state and a None response from the hook now keep the sensor poking instead of reporting success. The known states are unaffected.
   Works as expected from my side.
7. zach-overflow commented on Sep 10, 2026

   

   zach-overflow

   on Sep 10, 2026

   Contributor

   More actions

   Looks good for my change (251cd91) from main (tested using a9f6c89). Tests for pagerduty, anthropic, and google all pass; The ast-grep hook I added also passes. That said, there were 6 openlineage dataproc-injection failures which are pre-existing and unrelated to the httpx2 switch, and they reproduce on main latest as well.

   I verified the RCs from PyPI in a clean 3.10 venv: google 22.5.0rc1, pagerduty 5.2.7rc1, anthropic 1.0.0rc1. Pins are right, and I exercised the real paths, not just imports — the cloud-sql proxy download via httpcore2 succeeded with truststore / TLS. send_event had a valid, successful httpx2.Response returning the dedup key. httpx is still in that venv but only transitively via google's other deps.
8. justinpakzad commented on Sep 10, 2026

   

   justinpakzad

   on Sep 10, 2026

   Contributor

   More actions

   #72602, #70103, and #71659 are all working as expected. Tested the new changes against a live Snowflake instance with variety of success and failure paths.
9. bingqin2 commented on Sep 11, 2026

   

   bingqin2

   on Sep 11, 2026

   Contributor

   More actions

   Tested #72824 on Python 3.12.13 with apache-airflow 3.3.1 and apache-airflow-providers-amazon==9.36.0rc1: EcsTaskFailToStart, EcsOperatorError and WaiterTerminalFailure all survive a pickle.dumps / pickle.loads round-trip with their message, failures and last_response intact. Works as expected, thanks.
10. keith991001 commented on Sep 11, 2026

    

    keith991001

    on Sep 11, 2026

    Contributor

    More actions

    Hi, Tested my change working on Python 3.12.13 and Airflow 3.3.1 with apache-airflow-providers-common-compat==1.19.0rc1: #72509 behaves as expected. Using real attrs 26.1.0 entities in a clean environment, constructing two Table instances and appending a Tag to one no longer leaks into the other — tags, columns, owners, and extra are all per-instance now, and Column.tags is
    isolated the same way. Mutations also no longer bleed into the class defaults, so freshly constructed instances start empty as they should. +1 (non-binding)
11. moomindani commented on Sep 11, 2026

    

    moomindani

    on Sep 11, 2026

    Contributor

    More actions

    Tested the databricks 7.20.0 candidate wheel from dist/dev/airflow/providers/2026-09-09/ (sha512 verified) in a clean venv, so the code under test came from the packaged artifact rather than a source checkout — airflow 3.3.1, provider resolved from site-packages.

    One job covers both of my items: 101 condition_task entries push the run's task list past the API's 100-per-page limit, and a trailing always-failing notebook task with max_retries: 1 produces two attempt entries for a single task_key. Run against a live workspace, terminal result_state FAILED as intended.

    #72304 — get_run() returned 103 entries where it previously stopped at 100, get_run_tasks() agreed exactly (103 = 103), and all 102 declared task_keys were present.

    #72313 — the API returned two attempts for the failing task (attempt_number 0 and 1, distinct run ids) and extract_failed_task_errors reported it once, carrying the last attempt's run id.

    Both good from me.

    ---

    Drafted-by: Claude Code (Opus 5); reviewed by @moomindani before posting
12. nandeshkanagaraju commented on Sep 11, 2026

    

    nandeshkanagaraju

    on Sep 11, 2026

    Contributor

    More actions

    Tested amazon 9.36.0rc1 for Fix region_name being ignored by the Step Functions execution trigger (#72625) — works as expected ✅

    Installed the RC from PyPI into a clean venv (Python 3.12, apache-airflow==3.3.1, apache-airflow-task-sdk==1.3.1) and verified against the installed distribution (not a source checkout), with AWS_DEFAULT_REGION=us-east-1 so a regression would be visible:

    • StepFunctionsExecutionCompleteTrigger(execution_arn=..., region_name="eu-west-1") → trigger.region_name == "eu-west-1", and the hook the deferred waiter polls with builds a boto3 client in eu-west-1.
    • serialize() round-trip keeps region_name, and the rebuilt trigger still polls eu-west-1 (so the region survives a triggerer restart).
    • With no region_name passed, it still falls back to the connection/env default (us-east-1) — no regression for existing DAGs.
    • End-to-end via StepFunctionStartExecutionOperator(deferrable=True, region_name="eu-west-1"): the TaskDeferred trigger carries the region and polls eu-west-1.
    • The unit tests added in the PR (providers/amazon/tests/unit/amazon/aws/triggers/test_step_function.py) pass against the RC wheel: 3 passed.

    Negative control on the previous release 9.35.1 reproduces the original bug — trigger.region_name is None and the waiter client is created in us-east-1 despite region_name="eu-west-1" being passed.

    Thanks for preparing the release!
13. KidAmnesiac1 commented on Sep 11, 2026

    

    KidAmnesiac1

    on Sep 11, 2026

    Contributor

    More actions

    #72197 looks good. 🫡
14. xBis7 commented on Sep 11, 2026

    

    xBis7

    on Sep 11, 2026

    Last edited by xBis7

    Contributor

    More actions

    Tested provider apache kafka 2.0.0rc1 manually and everything looks good!

    My changes are adding an allowlist for configured callbacks in a connection. I created a connection with an unknown callback and ran a dag. The task using the connection failed as expected with the correct error.
15. 1 remaining item

    Load more

16. yuseok89 commented on Sep 11, 2026

    

    yuseok89

    on Sep 11, 2026

    Contributor

    More actions

    Tested apache-airflow-providers-amazon==9.36.0rc1 for #72497 against a real GCS bucket and a real S3 bucket. Works as expected.
17. rjgoyln commented on Sep 11, 2026

    

    rjgoyln

    on Sep 11, 2026

    Contributor

    More actions

    Tested all four items from a clean Python 3.12 venv with apache-airflow==3.3.1 and the RC wheels installed from PyPI. Each scenario was run against both the RC and previous release as a control.

    • fab 3.9.0rc1 (Fix FAB ignoring create_metadata_engine from local settings #71674): db migrate and users create/list now work with create_metadata_engine, shared engines are reused correctly, and secondary binds still get independent engines. 3.8.1 fails with sqlite3.OperationalError.
    • http 6.1.0rc1 (Keep HttpEventTrigger asset watchers polling after a failed request #72376): failed polls now continue until success or max_consecutive_failures; failure counts reset after a completed poll, serialization preserves the option, and 0 is rejected. 6.0.5 stops after the first failed request.
    • microsoft.azure 15.1.0rc1 (Add support for sending emails through Microsoft Graph #71565): verified real Graph request construction, auth, recipients, attachments, headers, dry-run/validation, repeated sends, and MSGraphNotifier. The new backend/notifier are also present in provider metadata; 15.0.1 has neither.
    • teradata 3.6.3rc1 (Fail the Teradata compute cluster trigger on an unknown operation #72696): unknown operations now immediately emit an error without polling, while all valid operations and the missing-profile case retain their previous behavior. 3.6.2 polled indefinitely instead.

    All four work as expected from my side.
18. Har1sh-k commented on Sep 11, 2026

    

    Har1sh-k

    on Sep 11, 2026

    Contributor

    More actions

    Tested apache.hive 9.6.2rc1 for #66751, installed from PyPI into a clean venv on Airflow 3.3.1 (with mysql 6.6.2 and presto 5.12.1). Works for me.

    What I exercised, driving HiveStatsCollectionOperator.execute with the metastore, Presto and MySQL hooks stubbed and capturing the SQL and parameters:

    • The Presto stats query renders the WHERE clause with the hook's own placeholder and passes the partition values through get_first(sql, parameters=...), so the values are no longer interpolated into the statement.
    • The hive_stats SELECT and the DELETE on the previous-run path both use %s with parameters=, binding table name, partition repr and dttm.
    • Identifier quoting behaves as intended: a hyphenated column and partition key come out as "weird-col" and "dt-col", db.odd-table becomes db."odd-table" with the catalog left plain, db."already.quoted" is passed through without being re-escaped, and plain word identifiers are still emitted unquoted. That is the case that used to produce an invalid statement.
19. AlejandroMorgante commented on Sep 11, 2026

    

    AlejandroMorgante

    on Sep 11, 2026

    Contributor

    More actions

    Validated apache-airflow-providers-cncf-kubernetes==10.22.0rc1 — #71244: example_kubernetes_pod_exec.py passed against a temporary Kubernetes 1.31.5 k3d cluster (1 passed). Pod creation and readiness, command execution in the existing container, stdout returned through XCom, Pod deletion, and post-test cleanup were verified.

    Environment: Breeze, Airflow 3.1.3, Python 3.10
20. Vamsi-klu commented on Sep 11, 2026

    

    Vamsi-klu

    on Sep 11, 2026 · Hidden as spam

    show commentMore actions
21. 23tae commented on Sep 12, 2026

    

    23tae

    on Sep 12, 2026

    Contributor

    More actions

    Tested apache-airflow-providers-cncf-kubernetes==10.22.0rc1 for #68890. (Environment: Breeze, Airflow 3.3.1, Python 3.12)

    Verified that KubernetesPodOperator._set_name:

    • Accepts valid pod names without errors.
    • Raises the expected validation exception (AirflowException on Airflow 3.3.1) when the name exceeds the limit, confirming backward compatibility.
22. SameerMesiah97 commented on Sep 12, 2026

    

    SameerMesiah97

    on Sep 12, 2026

    Contributor

    More actions

    apache-airflow-providers-amazon==9.36.0rc1

    PR #72455: Changes are working as expected with no observed regresssions.
23. steveahnahn commented on Sep 12, 2026

    

    steveahnahn

    on Sep 12, 2026

    Contributor

    More actions

    Tested apache-airflow-providers-snowflake==6.17.0rc1 for apache/airflow#69635 in Breeze (Python 3.10, Snowflake SQL API mocked with requests_mock). Works as expected.
24. henry3260 commented on Sep 13, 2026

    

    henry3260

    on Sep 13, 2026

    Contributor

    More actions

    apache-airflow-providers-cncf-kubernetes==10.22.0rc1

    • PR Fix XCom sidecar helper mutating the caller's pod volumes #72522: Works as expecte
    • PR Fix KubernetesPodOperator XCom loss when container_logs is a string #72502: Works as expected
      apache-airflow-providers-amazon==9.36.0rc1
apache-airflow-providers-fab==3.9.0rc1
apache-airflow-providers-keycloak==0.10.0rc1
    • PR Scope asset API responses to the assets a user may read #72682: The provider-side filter_authorized_assets / get_authorized_assets overrides work as expected.
25. amoghrajesh commented on Sep 13, 2026

    

    amoghrajesh

    on Sep 13, 2026

    Contributor

    More actions

    Installed the RC and checked my changes, none of them were functional, it was all compat related.

    #72094
    #72140
    #72446
26. aaron-y-chen commented on Sep 13, 2026

    

    aaron-y-chen

    on Sep 13, 2026

    Contributor

    More actions

    Verified #71350, it worked very well :)
27. potiuk commented on Sep 13, 2026

    

    potiuk

    on Sep 13, 2026

    Member

    More actions

    Verified the 2026-09-09 RC (38 providers) end to end — all checks pass.

    Artifacts (dist/dev/airflow/providers/2026-09-09): folder complete (231 files = 38×6 + source trio; 77 .sha512 / 77 .asc, no double-extension leftovers); SHA512 77/77; GPG 77/77 good, all on the RM's key 1E7A857875E0EE49DB99E4431B17D6C5938CED00 (in KEYS); reproducible build 77/77 byte-identical — 76 wheels+sdists rebuilt from tag providers/2026-09-09 (209e34c5022a78c7691ac68a50f60d976242e17d), plus the source tarball; Apache RAT 0.18 — 0 unapproved / 0 unknown (10146 standard files, all AL2.0).

    Consistency: all 38 rc tags point at the wave commit; 38/38 pyproject.toml versions match their changelog tops; 38/38 PyPI rc pages resolve with both wheel and sdist.

    My own changes in this wave — ticked above. Each was confirmed present in the shipped wheel, not just at the tag, and its tests run green:

    PR	Provider	Tests
    #72646	akeyless 0.3.1	37 passed
    #72645, #72657, #72198, #72199	fab 3.9.0	281 passed
    #72207, #72205	keycloak 0.10.0	202 passed, 1 skipped

    For #72646 I also exercised the namespace guard from a clean venv containing only the RC wheel: a cross-team key under multi-team + team-scoped path is refused, while a flat key, use_team_secrets_path=False, an absent team name, and multi-team disabled all still resolve.

    One non-blocking observation (pre-existing, unchanged since the 2026-08-18 wave): the source tarball ships two OFL-1.1 web fonts at registry/public/fonts/{jetbrains-mono-latin,plus-jakarta-sans-latin}.woff2 while the root LICENSE is bare AL2.0 with no third-party section. Confirmed this wave that they are source-tarball only — no provider wheel contains them. A LICENSE follow-up, not a re-cut.

    ---

    Drafted-by: Claude Opus 5; reviewed by @potiuk before posting
28. FrankYang0529 commented on Sep 14, 2026

    

    FrankYang0529

    on Sep 14, 2026

    Member

    More actions

    Tested following providers and all work as expected:

    • amazon 9.36.0rc1
      • Fix incorrect documented defaults in Amazon operators and sensors #72315: Compared the documented defaults of the five changed parameters with their inspect.signature defaults.
    • apache.beam 6.2.4rc1
      • Fix Apache Beam hook type error with mypy 2.2.0 typeshed #69946: Type-checked the installed hook with mypy 2.2.0, which reports the [type-var] error on 6.2.3 but not on the RC, and ran a small DirectRunner pipeline through BeamHook on both versions to confirm runtime behavior did not change.
    • Fix mismatched AirflowConfigException in celery and edge3 #72437: Started a Celery worker with a Redis broker and ran airflow plugins with malformed config values for [celery] ssl_active, [dag_processor] stale_bundle_cleanup_interval, and [edge] api_enabled. The previous releases crash, hang, or silently skip the edge3 plugin, while the RCs fall back as intended.
      • celery 3.24.0rc1
      • edge3 4.3.2rc1
    • databricks 7.20.0rc1
      • Add performance_target to Databricks RunNow and CreateJobs operators #72148: Ran DatabricksRunNowOperator and DatabricksCreateJobsOperator against Databricks API and checked that performance_target can work.
    • fab 3.9.0rc1
      • Remove unused FAB app-init helpers that could never run #72610: Confirmed the three removed helpers are gone from the RC. API server with the FAB auth manager on rc and previous versions can work and got the same results from the login page, token endpoint, core API, and FAB users API.
    • Make SQLAlchemy an optional dependency for JDBC and YDB providers #72373: Checked that both wheels now declare a sqlalchemy extra and that the hooks import SQLAlchemy lazily and show an install hint when the import fails. Run queries through JdbcHook against Postgres 16 and through YDBHook against a local YDB and got the same results as the previous releases.
      • jdbc 5.6.0rc1
      • ydb 2.6.0rc1
    • microsoft.azure 15.1.0rc1
      • Keep MSGraph pagination offset across paginated pages #71986: Paged through 30 messages from Microsoft Graph endpoint with $top=12, where 15.0.1 skips a page and its deferrable operator fails with KeyError: 'value' when result_processor returns the value list. The RC returns all 30 messages through both paginated_run() and the deferrable operator.
      • Restrict MSGraph deferrable pagination to the configured host #71842: Ran pagination through the deferrable operator and confirmed that the RC refuses a pagination link to a host other than the configured one before sending any request, while links to the same host and to hosts listed in allowed_hosts still work.
    • odbc 4.13.0rc1
      • Declare allow_driver_in_extra in the ODBC provider configuration metadata #72547: Checked that airflow config list and airflow config get-value now show [providers.odbc] allow_driver_in_extra. Connect through OdbcHook with the SQLite ODBC driver to confirm the handling of the driver extra did not change.
    • opensearch 1.12.1rc1
      • Do not forward an empty ca_certs to the OpenSearch log client #72552: Configured remote logging against OpenSearch 2.19.6 over HTTPS with ca_certs set to an empty value, where 1.12.0 fails to load the logging config and the RC writes and reads back a log line with certificate verification still on.
    • redis 4.6.0rc1
      • Add Redis cluster mode support to RedisHook #71067: Pointed RedisHook with cluster: true at a three-master Redis Cluster The RC reads and writes keys on every master, including through RedisKeySensor, while 4.5.0 gets MovedError for keys outside the seed node.
29. vincbeck commented on Sep 14, 2026

    

    vincbeck

    on Sep 14, 2026

    ContributorAuthor

    More actions

    Thank you everyone. Providers are released.

    I invite everyone to help improve providers for the next release, a list of open issues can be found here.
30. vincbeck

    closed this as completedon Sep 14, 2026