This is an automated email from the ASF dual-hosted git repository.

potiuk pushed a commit to branch main
in repository https://gitbox.apache.org/repos/asf/airflow.git


The following commit(s) were added to refs/heads/main by this push:
     new 245304d6bf5 Skip the ready-for-review CI run when the head already 
passed (#77118)
245304d6bf5 is described below

commit 245304d6bf57b73a8467de75beb4f75f7004d666
Author: Jarek Potiuk <[email protected]>
AuthorDate: Sun Nov 15 13:59:27 2026 +0100

    Skip the ready-for-review CI run when the head already passed (#77118)
    
    Draft pull requests already get CI when they are opened and pushed to, so
    marking one ready for review repeats a full Tests (AMD) run on a head that
    was tested and passed. The `ready_for_review` trigger is still needed for
    pull requests whose runs never passed, such as the scheduled CI upgrade
    pull requests that maintainers undraft to get a fresh run.
    
    Generated-by: Claude Opus 5
    Claude-Session: https://claude.ai/code/session_01NyQq9TvfJuBFXZg3jrCSZW
---
 .github/workflows/ci-amd.yml      | 34 ++++++++++++++++++++++++++++++++--
 .github/workflows/ci-arm.yml      | 34 ++++++++++++++++++++++++++++++++--
 dev/breeze/doc/ci/05_workflows.md |  9 +++++++++
 3 files changed, 73 insertions(+), 4 deletions(-)

diff --git a/.github/workflows/ci-amd.yml b/.github/workflows/ci-amd.yml
index 33ff3aff9ab..840e1c7df03 100644
--- a/.github/workflows/ci-amd.yml
+++ b/.github/workflows/ci-amd.yml
@@ -81,7 +81,8 @@ jobs:
     name: "Check if the run is needed"
     # A scheduled canary is not needed when the last completed full run of 
this workflow on the
     # branch tested the same commit and succeeded: it would only repeat that 
run. A push run tests
-    # the changes since the last successful run of this workflow on the 
branch, found here.
+    # the changes since the last successful run of this workflow on the 
branch, found here. Marking
+    # a draft pull request ready is not needed when its head already has a 
passing `CI passed`.
     # This job runs for every event, with no steps for the others: every other 
job depends on it
     # through `build-info`, and GitHub skips the whole dependency chain below 
a skipped job unless
     # each job in it has a status function such as `always()` in its `if:`.
@@ -89,12 +90,41 @@ jobs:
     timeout-minutes: 5
     permissions:
       actions: read
+      checks: read
       contents: read
     outputs:
-      run-needed: ${{ steps.schedule.outputs.run-needed || 'true' }}
+      run-needed: ${{ steps.schedule.outputs.run-needed || 
steps.ready.outputs.run-needed || 'true' }}
       # Empty in forks, whose pushes selective checks compare against 
apache/airflow.
       push-base-commit: ${{ steps.push.outputs.push-base-commit }}
     steps:
+      - name: "Check the earlier runs of the pull request's head"
+        if: github.event_name == 'pull_request' && github.event.action == 
'ready_for_review'
+        id: ready
+        shell: bash
+        env:
+          GH_TOKEN: ${{ github.token }}
+          REPOSITORY: ${{ github.repository }}
+          HEAD_SHA: ${{ github.event.pull_request.head.sha }}
+        run: |
+          set -euo pipefail
+          # Draft pull requests get CI on `opened` and `synchronize`, so 
marking one ready usually
+          # repeats a run that already passed. Only a job that ran as `CI 
passed` counts: a gate-only
+          # run, a run skipped here and a cancelled run all succeed under 
another name of the job,
+          # and they did not test the head. A head whose runs failed, were 
cancelled or wait for
+          # approval, such as a scheduled CI upgrade pull request's, is tested 
in full.
+          if ! passed="$(gh api 
"repos/${REPOSITORY}/commits/${HEAD_SHA}/check-runs" -X GET \
+            -f check_name="CI passed" -f filter=latest -F per_page=100 \
+            --jq '[.check_runs[] | select(.conclusion == 
"success")][0].html_url // ""')"; then
+            # Fail open: a failed lookup must not skip the run.
+            echo "::warning::Could not check the earlier runs of ${HEAD_SHA}, 
running CI"
+            passed=""
+          fi
+          if [[ -n "${passed}" ]]; then
+            echo "run-needed=false" >> "${GITHUB_OUTPUT}"
+            echo "The head of the pull request already passed CI: ${passed}" 
>> "${GITHUB_STEP_SUMMARY}"
+          else
+            echo "run-needed=true" >> "${GITHUB_OUTPUT}"
+          fi
       - name: "Find the last successful run of the branch"
         if: github.event_name == 'push' && github.repository == 
'apache/airflow'
         id: push
diff --git a/.github/workflows/ci-arm.yml b/.github/workflows/ci-arm.yml
index d719834a287..14f5d4ed8e2 100644
--- a/.github/workflows/ci-arm.yml
+++ b/.github/workflows/ci-arm.yml
@@ -60,7 +60,8 @@ jobs:
     name: "Check if the run is needed"
     # A scheduled canary is not needed when the last completed full run of 
this workflow on the
     # branch tested the same commit and succeeded: it would only repeat that 
run. A push run tests
-    # the changes since the last successful run of this workflow on the 
branch, found here.
+    # the changes since the last successful run of this workflow on the 
branch, found here. Marking
+    # a draft pull request ready is not needed when its head already has a 
passing `CI passed`.
     # This job runs for every event, with no steps for the others: every other 
job depends on it
     # through `build-info`, and GitHub skips the whole dependency chain below 
a skipped job unless
     # each job in it has a status function such as `always()` in its `if:`.
@@ -68,12 +69,41 @@ jobs:
     timeout-minutes: 5
     permissions:
       actions: read
+      checks: read
       contents: read
     outputs:
-      run-needed: ${{ steps.schedule.outputs.run-needed || 'true' }}
+      run-needed: ${{ steps.schedule.outputs.run-needed || 
steps.ready.outputs.run-needed || 'true' }}
       # Empty in forks, whose pushes selective checks compare against 
apache/airflow.
       push-base-commit: ${{ steps.push.outputs.push-base-commit }}
     steps:
+      - name: "Check the earlier runs of the pull request's head"
+        if: github.event_name == 'pull_request' && github.event.action == 
'ready_for_review'
+        id: ready
+        shell: bash
+        env:
+          GH_TOKEN: ${{ github.token }}
+          REPOSITORY: ${{ github.repository }}
+          HEAD_SHA: ${{ github.event.pull_request.head.sha }}
+        run: |
+          set -euo pipefail
+          # Draft pull requests get CI on `opened` and `synchronize`, so 
marking one ready usually
+          # repeats a run that already passed. Only a job that ran as `CI 
passed` counts: a gate-only
+          # run, a run skipped here and a cancelled run all succeed under 
another name of the job,
+          # and they did not test the head. A head whose runs failed, were 
cancelled or wait for
+          # approval, such as a scheduled CI upgrade pull request's, is tested 
in full.
+          if ! passed="$(gh api 
"repos/${REPOSITORY}/commits/${HEAD_SHA}/check-runs" -X GET \
+            -f check_name="CI passed" -f filter=latest -F per_page=100 \
+            --jq '[.check_runs[] | select(.conclusion == 
"success")][0].html_url // ""')"; then
+            # Fail open: a failed lookup must not skip the run.
+            echo "::warning::Could not check the earlier runs of ${HEAD_SHA}, 
running CI"
+            passed=""
+          fi
+          if [[ -n "${passed}" ]]; then
+            echo "run-needed=false" >> "${GITHUB_OUTPUT}"
+            echo "The head of the pull request already passed CI: ${passed}" 
>> "${GITHUB_STEP_SUMMARY}"
+          else
+            echo "run-needed=true" >> "${GITHUB_OUTPUT}"
+          fi
       - name: "Find the last successful run of the branch"
         if: github.event_name == 'push' && github.repository == 
'apache/airflow'
         id: push
diff --git a/dev/breeze/doc/ci/05_workflows.md 
b/dev/breeze/doc/ci/05_workflows.md
index 7b29cc1d431..ad967d30b89 100644
--- a/dev/breeze/doc/ci/05_workflows.md
+++ b/dev/breeze/doc/ci/05_workflows.md
@@ -68,6 +68,15 @@ for these builds, we only execute Python tests if 
significant files have
 changed. For instance, if the PR involves a "no-code" change, no tests
 will be executed.
 
+Draft PRs get these runs too, on `opened` and `synchronize`. Marking a draft
+ready for review starts another run, which finishes at once when the head of
+the PR already has a passing `CI passed` check: its first job, `Check if the 
run
+is needed`, finds that check and every other job is skipped. The earlier
+`CI passed` stays on the commit, so the PR can merge on it. When there is no
+passing `CI passed` for the head, the run tests it in full. That is how a PR 
whose
+earlier runs failed, were cancelled or still wait for approval, such as the
+scheduled CI upgrade PRs, gets a fresh run when a maintainer marks it ready.
+
 Regular PR builds run in a "stable" environment:
 
 - fixed set of constraints (constraints that passed the tests) - except

Reply via email to