Status of testing of Apache Airflow 3.3.2rc1

@potiuk

We are kindly requesting that contributors to Apache Airflow RC 3.3.2rc1 help test the RC.

Please let us know by commenting if the issue is addressed in the latest RC.

Thanks to all who contributed to the release (probably not a complete list!):
@arose26 @FrankYang0529 @warreee @aaron-y-chen @eladkal @bujjibabukatta @topherinternational @raphaelauv @sortega @Aaryan123456679 @stephsi @tirkarthi @Ed3110 @kacpermuda @steveahnahn @Dev-iL @edsu @seanghaeli @ryanahamilton @SreeramaYeshwanthGowd @fat-catTW @Vamsi-klu @theJohnnyBrown @ColtenOuO @sgoel2be24-cyber @sln-jack @Lee-W @jasperjonkhans @Andrushika @SameerMesiah97 @chrispatterson @deepujain @kyungjunleeme @vatsrahul1001 @dabla @simon-hoon @JH-A-Kim @matthieuauger @potiuk @SamWheating @1fanwang @hkc-8010 @haseebmalik18 @pierrejeambrun @manish1337 @amoghrajesh @hanxdatadog @ephraimbuddy @ferruzzi @dayonet @Eason09053360 @eitan-shalev @ignacioparicio @neel-astro @Pedrinhonitz @JayalakshmiH @ehab20011 @rjgoyln @o-nikolas @kalluripradeep @kaxil @seanmuth @SHIVANSH-ux-ys @shahar1 @jedcunningham @vincbeck @jason810496 @bbovenzi @vincentpwls @dheerajturaga @6863jh @ccoliu @bramhanandlingala @henry3260 @WonYong-Jang

@potiuk

Verification of 3.3.2rc1 / Task SDK 1.3.2rc1

I ran the PMC mechanical checks from dev/README_RELEASE_AIRFLOW.md against the staged artefacts (svn r87505). Result: pass, with two non-blocking warnings that are both pre-existing in the released 3.3.1.

Check	Result
Artefact inventory	PASS — 15 Airflow files + 6 Task SDK, nothing missing or unexpected
GPG signatures	PASS — 7/7 good, signing key present in the project KEYS file
Checksums	PASS — 7/7 sha512, no deprecated md5
Licence headers (Apache RAT 0.18)	PASS — 9591 files, 0 unapproved, 0 unknown
NOTICE / LICENSE	PASS — NOTICE unchanged vs 3.3.1; LICENSE +20 lines, intentional (#73067)
Version consistency	PASS — `3.3.2` / `1.3.2` throughout
Reproducible build	PASS — all 7 artefacts bit-for-bit identical

Reproducibility was checked from a clean git clone --branch 3.3.2rc1 using the default Docker build path. All seven artefacts compare byte-identical to the staged ones.

The LICENSE addition is correct and complete: the OFL-1.1 section added by #73067 covers JetBrains Mono and Plus Jakarta Sans, both LICENSE-*.txt files ship in the source release, and both .woff2 files they cover are present. (codicon.ttf in the core wheel is a Monaco Editor asset, already covered by the existing MIT entry — not a gap.)

Two warnings — neither blocks the vote

1. Five dangling symlinks in the source release. .claude/skills/{aip-user-stories,airflow-translations,magpie-setup,prepare-providers-documentation,upgrade-fab-provider} all point at ../../.agents/skills/.... .gitattributes has .agents export-ignore but no .claude export-ignore, so the relay symlinks ship while their targets are stripped. Present since #67929 and identical in the released 3.3.1. One-line fix for a follow-up PR on main, not for this RC.

2. java-sdk/gradle/wrapper/gradle-wrapper.jar (43 KB) in the source release. RAT classifies it as an archive and does not complain. Identical at the 3.3.1 tag. Worth a policy decision eventually — ASF prefers source releases free of compiled binaries — but not grounds to -1 here.

The 23 items assigned to me

All 23 are in the 3.3.2rc1 tag, and all 23 are new in 3.3.2.

Nine of them — #70597, #71428, #71434, #71451, #71512, #71640, #71785, #71828, #71838 — do not appear under the merge-commit SHA the API reports, because v3-3-test was rebased/force-pushed after they merged. I re-confirmed each by matching the PR reference in the tag history and comparing git patch-id --stable between the original and rewritten commits: all nine carry byte-identical patches. The other fourteen (#72660, #72712, #72723, #72728, #72806, #72874, #72933, #73041, #73042, #73043, #73046, #73051, #73067, #73068) are direct ancestors of the tag.

Worth noting that #73068 (RAT excludes, move to Apache RAT 0.18) is itself in this RC, and the RAT run above used the 0.18 image it pins — so that fix is verified against the artefact it was meant to fix.

These checks are mechanical and say nothing about runtime behaviour; an install-and-exercise pass is still needed for a binding vote.

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

Tested #71952 on 3.3.2rc1, works as expected.

I tested #71585 / #71582 against the 3.3.2rc1 source tag with Go 1.25.13 on macOS arm64.

The Go SDK suite finished with 229 tests passed, 0 failed and 4 existing TODO skips. Both cross-architecture bundle-packing tests passed. I ran GOTOOLCHAIN=go1.25.13 go test -count=1 -json ./... from go-sdk/.

Both CI workflows pin Go 1.25.13. The Linux amd64 CI job also resolved 1.25.13 and passed against identical SDK and workflow sources:
https://github.com/apache/airflow/actions/runs/34749988431/job/103704727890

Tested my PRs against 3.3.2rc1, all work as expected:
#71609, #71910, #72729, #72743, #72771, #72800, #72801

Thanks for preparing the release, Rahul!

@Eason09053360

Tested all 11 of my backports on 3.3.2rc1. All verified fixed.

[v3-3-test] Show each Dag only once in airflow dags list (#70934) #71481 ✅ airflow dags list prints a Dag once. With two dag_version rows for the same dag_id, 3.3.1 printed 2 lines, rc1 prints 1.
[v3-3-test] Fix airflow config lint staying silent on conditional removal rules (#70977) #71651 ✅ airflow config lint now reports email.email_backend and elasticsearch.log_id_template. 3.3.1 said "No issues found" for both.
[v3-3-test] Stop variables export/import from silently corrupting values (#70944) #71791 ✅ variables export → import round-trip preserves {"value": "a", "description": "b"}. 3.3.1 collapsed it to a.
[v3-3-test] Export AIRFLOW_TEST_MODE from airflow tasks test without --env-vars (#72291) #72320 ✅ airflow tasks test exports AIRFLOW_TEST_MODE=True without --env-vars.
3.3.1 exported nothing — and --env-vars '{}' did not help either, since an empty dict is falsy.
[v3-3-test] Prevent Dag CLI subcommands from being silently dropped (#72340) #72365 ✅ Rebuilding the Dag parser keeps every argument (dags pause 5→5, dags test
11→11, tasks test 9→9). On 3.3.1 a rebuild dropped all of them to 1 (just -h).
[v3-3-test] Fix DAG.cli() crashing on dags pause and unpause (#72109) #72565 ✅ python my_dag.py dags pause / unpause exit 0. 3.3.1 raised
TypeError: dag_pause() takes 1 positional argument but 2 were given.
[v3-3-test] Stop airflow standalone leaking components when one fails to start (#72464) #72568 ✅ With one component failing to spawn, the shutdown loop completes and the
surviving sibling is terminated. On 3.3.1 the loop aborted with AttributeError: 'SubCommand' object has no attribute 'process' and the sibling kept running after standalone exited.
[v3-3-test] Stop airflow providers get --full mutating cached provider metadata (#72346) #72601 ✅ providers get --full leaves ProvidersManager metadata untouched when called
in-process. 3.3.1 mutated the cached description.
[v3-3-test] Trace the dag-processor job instead of its parent process with memray (#72367) #72661 ✅ With --daemon, the memray capture spans the actual run (10.3s over 8 Dags ×
20 tasks). On 3.3.1 start_time == end_time — a 0-second capture containing only
interpreter startup. Note this needs --daemon or --dev to reproduce; the plain path does not fork.
[v3-3-test] Allow airflow jobs check --allow-multiple with --limit 0 (#72677) #72744 ✅ jobs check --allow-multiple --limit 0 passes the guard; --limit 1 is still rejected, with a clearer message.
[v3-3-test] Fix airflow info --file-io uploading an empty report (#72707) #72832 ✅ AirflowInfo.render_text() returns the real report (3531 / 8403 / 3527 chars
for json / table / yaml, no ANSI). On 3.3.1 all three returned an empty string, which is
what was being uploaded. Verified offline rather than uploading to file.io.

Not covered: postgres/mysql backends, the --dev reloader path for #72661, and the ANSI
half of #72832 (the check ran in a non-tty subprocess, where color_system="auto" would
not emit escape codes anyway).

Drafted-by: Claude Opus 5 ; reviewed by @Eason09053360 before posting

Tested #71715 on 3.3.2rc1 and it works as expected

Tested #71951 using the 3.3.2rc1 airflow image, works as expected

Tested on 3.3.2rc1 with Postgres - #73152 no longer reproduces. Fixed by #71955.

Tested on 3.3.2rc1 (Docker image apache/airflow:3.3.2rc1, SQLite) with core.multi_team = True — [v3-3-test] Fix Variable.set rewriting team_name of existing variables (#71823) (#71904) works as expected.

Tested #72357 on 3.3.2rc1 (Docker image apache/airflow:3.3.2rc1, standalone). The task overview duration chart now follows the limit search param: 10 instances by default, and ?limit=25 / ?limit=5 request and render that many. Works as expected.

Installed the release bits and validated my changes.

Actual changes:

[v3-3-test] Catch general Exception when initializing a dag bundle (#71355) #71363
[v3-3-test] Authenticate only once per task process to external secrets backends (#71756) #72237 (vault already uses this too)
Improve confirmation output when connections are added via cli (#71215) #72779 (minor enhancement)

CI / non functional:

[v3-3-test] Convert check-sdk-imports to capture occurences than files (#71212) #71398
[v3-3-test] Update the docs recommending dag.test() for custom operator tests (#71542) #71587

All my change work expected

sorry for the late reply, all my change work expected too!

thanks for preparing this!

Tested #72878, #72702, #72807 with the 3.3.2rc1 image, works as expected.
Tested #71233, #71236, #71553 on the 3.3.2rc1 tag source, works as expected. These three only change breeze and prek hook scripts, which are not in the image.

3.3.2 is released. Thanks for testing efforts

Navigation Menu

Repository navigation

Status of testing of Apache Airflow 3.3.2rc1 #73089

Verification of 3.3.2rc1 / Task SDK 1.3.2rc1

Two warnings — neither blocks the vote

The 23 items assigned to me

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Status of testing of Apache Airflow 3.3.2rc1 #73089

Description

Activity

potiuk commented on Sep 13, 2026

Verification of 3.3.2rc1 / Task SDK 1.3.2rc1

Two warnings — neither blocks the vote

The 23 items assigned to me

haseebmalik18 commented on Sep 13, 2026

Vamsi-klu commented on Sep 14, 2026

1fanwang commented on Sep 14, 2026

Andrushika commented on Sep 14, 2026

Eason09053360 commented on Sep 15, 2026

ignacioparicio commented on Sep 15, 2026

stephsi commented on Sep 15, 2026

DianaHov commented on Sep 15, 2026

ccoliu commented on Sep 16, 2026

warreee commented on Sep 16, 2026

amoghrajesh commented on Sep 16, 2026

henry3260 commented on Sep 16, 2026

ColtenOuO commented on Sep 16, 2026

FrankYang0529 commented on Sep 17, 2026

vatsrahul1001 commented on Sep 17, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions