This is an automated email from the ASF dual-hosted git repository.
potiuk pushed a commit to branch v3-4-test
in repository https://gitbox.apache.org/repos/asf/airflow.git
The following commit(s) were added to refs/heads/v3-4-test by this push:
new c6cb20ff2ef [v3-4-test] Describe where the dependency upgrade summary
runs now (#76439) (#76454)
c6cb20ff2ef is described below
commit c6cb20ff2ef5507c1098d23760abdebb8487029a
Author: github-actions[bot]
<41898282+github-actions[bot]@users.noreply.github.com>
AuthorDate: Wed Nov 4 15:04:47 2026 +0000
[v3-4-test] Describe where the dependency upgrade summary runs now (#76439)
(#76454)
Since #76251 the CI docs sent readers to the wrong workflow and claimed
that canaries with failed
jobs save no report inputs. Those canaries now get a report, with its
explanation jobs, so anyone
estimating report runs or their cost from the docs would come out too low.
The docs also still
credited basic tests with upgrade monitoring, which left since #64836, so a
reader looking for
this check got two answers.
(cherry picked from commit 629729a9913dfd617fad5d70a5425c5b073e0673)
Generated-by: Claude Opus 5
Co-authored-by: Jarek Potiuk <[email protected]>
---
dev/breeze/doc/ci/05_workflows.md | 20 ++++++++++++++++----
1 file changed, 16 insertions(+), 4 deletions(-)
diff --git a/dev/breeze/doc/ci/05_workflows.md
b/dev/breeze/doc/ci/05_workflows.md
index a371d363004..f956ee81c4d 100644
--- a/dev/breeze/doc/ci/05_workflows.md
+++ b/dev/breeze/doc/ci/05_workflows.md
@@ -393,7 +393,6 @@ Before running expensive tests, we validate basic
functionality:
- Breeze development environment tests
- React UI component tests
- Code formatting and style checks
-- Dependency upgrade monitoring
These quick checks catch obvious problems early.
@@ -458,6 +457,7 @@ Here's what each workflow group does and when it runs:
| **Kubernetes Tests** | Tests deployment in Kubernetes
environments | Yes | Yes | Yes (1) |
| **Distribution Tests** | Tests Task SDK and CLI tools (AMD only)
| Yes | Yes | Yes |
| **Agent Framework Tests** | Tests Common AI's adapters for other
agent frameworks (4) | Yes | Yes | No |
+| **Dependency Upgrade Summary** | Checks how far constraints lag new
releases (AMD only) | Yes (5) | Yes (5) | Yes (5) |
| **Finalize Tests** | Publishes results and updates shared
resources | Yes | Yes (2) | Yes (2) |
#### AMD-Only Workflows
@@ -533,6 +533,12 @@ Both install the newest framework releases older than the
repository's uv `exclu
release that breaks an adapter fails this job about that long after it ships.
The job is not a
dependency of **Finalize Tests**, so such a release does not stop the image
cache from being pushed.
+**`(5)` Dependency Upgrade Summary**
+
+The summary runs only when the run upgrades to newer dependencies: on
scheduled and manually
+dispatched runs, when a `pyproject.toml` file changes, or on a PR with the
+`upgrade to newer dependencies` label. It needs the CI image, so it never runs
without one.
+
## Runners
Two kinds of GitHub-hosted runner are used, and which one a job gets depends
on what
@@ -566,14 +572,16 @@ When adding a job, reach for `ubuntu-slim` if it only
shuffles metadata around,
### Dependency upgrade diagnostics
Scheduled and manually dispatched AMD canaries on `main` keep the dependency
freshness
-summary in `finalize-tests.yml`, but run it with `--no-explain-why`. Tests,
constraint
+summary in `dependency-upgrade-summary.yml`, but run it with
`--no-explain-why`. It starts
+once the CI images are built, so it runs beside the tests instead of after
them. Tests, constraint
generation and validation, and image publication still run in the canary.
Release-branch
and other run types retain their existing explanation behavior. The summary
saves the
constraints it checked, and the constraints-branch commit they came from, with
`--save-constraints-to`; a failure to save them only warns and never fails the
canary.
After the canary completes, `dependency-upgrade-report.yml` runs the detailed
explanations
-independently, from the inputs the canary saved; a canary with failed jobs
saves none. It checks out
+independently, from the inputs the canary saved. The summary does not wait for
the tests, so a
+canary whose tests failed still saves its inputs and is explained too. The
report checks out
the original source SHA and downloads the original image and saved constraints
by artifact
ID from that run. It never substitutes the latest branch image or constraints:
an entry whose
saved constraints or image are missing or expired is skipped with a warning
and listed in the
@@ -663,7 +671,7 @@ Each composite workflow is stored in a separate file under
`.github/workflows/`.
#### Basic Validation
-- `basic-tests.yml` - Handles Breeze tests, UI validation, static analysis,
and upgrade monitoring
+- `basic-tests.yml` - Handles Breeze tests, UI validation, and static analysis
#### Image Management
@@ -689,6 +697,10 @@ Each composite workflow is stored in a separate file under
`.github/workflows/`.
- `k8s-tests.yml` - Tests direct Kubernetes integration
- `airflow-distributions-tests.yml` - Validates package distributions
+#### Dependency Monitoring
+
+- `dependency-upgrade-summary.yml` - Checks how far the constraints are behind
the newest releases
+
#### Finalization
- `finalize-tests.yml` - Publishes results and updates shared resources
