ci: automate backport labeling and open PRs for failed backports - #6941
Merged
Conversation
Make release backports opt-out instead of easy-to-forget opt-in, and turn a failing backport into an actionable PR instead of a comment nobody reads. - .github/release-branches.yml: single source of truth mapping each active release branch to its release manager (release/v1.2 -> xuang7, release/v1.1 -> bobbai00), with a stdlib-only parser. - backport-auto-label: on a fix: PR into main, label it with every configured release/* target and request review from that branch's release manager. A removed label is remembered (timeline unlabeled event), so opting out sticks across later edits. - required-checks: split the pre-merge backport into a fast git-only apply-check that gates the expensive build matrix, and drop backport from the Required Checks aggregator so a red backport no longer blocks merge. - direct-backport-push: classify each target from the apply-check/build signal — green cherry-picks straight to the release branch (unchanged), red opens a draft backport PR (fix(scope, vX.Y): ...) with the conflicted tree committed, assigned to the author with the manager as reviewer. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
A fix for a feature that only exists on main should not be backported. Detect it automatically and add an explicit manual override, so neither case relies on someone remembering. - Auto-label: before labeling a target, skip it when every file the PR *modifies* (added files excluded) is absent on that release branch — the feature clearly isn't there. Any ambiguity keeps the label. - no-backport-needed label: a hard manual override that vetoes all backport work — honored by auto-label, the required-checks precheck (so apply-check and build skip), and the post-merge discover. - Red path: create-backport-branch applies the same feature-absent guard; when it fires, direct-backport-push records it on the PR instead of opening a doomed draft PR. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Contributor
Automated Reviewer SuggestionsBased on the
|
Contributor
|
| config | throughput | MB/s | latency | max Δ latest / 7d | |
|---|---|---|---|---|---|
| 🔴 | bs=10 sw=10 sl=64 | 480 | 0.293 | 21,122/24,556/24,556 us | 🔴 +40.3% / 🔴 +69.0% |
| ⚪ | bs=100 sw=10 sl=64 | 1,204 | 0.735 | 82,051/116,355/116,355 us | ⚪ within ±5% / 🟢 +23.5% |
| ⚪ | bs=1000 sw=10 sl=64 | 1,420 | 0.867 | 709,410/750,195/750,195 us | ⚪ within ±5% / 🟢 +41.5% |
Baseline details
Latest main 49f9e2c from same runner
| config | metric | PR | latest main | 7d avg | Δ latest | Δ 7d |
|---|---|---|---|---|---|---|
| bs=10 sw=10 sl=64 | throughput | 480 tuples/sec | 603 tuples/sec | 767.9 tuples/sec | -20.4% | -37.5% |
| bs=10 sw=10 sl=64 | MB/s | 0.293 MB/s | 0.368 MB/s | 0.469 MB/s | -20.4% | -37.5% |
| bs=10 sw=10 sl=64 | p50 | 21,122 us | 15,058 us | 12,502 us | +40.3% | +69.0% |
| bs=10 sw=10 sl=64 | p95 | 24,556 us | 25,148 us | 16,234 us | -2.4% | +51.3% |
| bs=10 sw=10 sl=64 | p99 | 24,556 us | 25,148 us | 18,919 us | -2.4% | +29.8% |
| bs=100 sw=10 sl=64 | throughput | 1,204 tuples/sec | 1,205 tuples/sec | 974.8 tuples/sec | -0.1% | +23.5% |
| bs=100 sw=10 sl=64 | MB/s | 0.735 MB/s | 0.736 MB/s | 0.595 MB/s | -0.1% | +23.5% |
| bs=100 sw=10 sl=64 | p50 | 82,051 us | 79,945 us | 102,449 us | +2.6% | -19.9% |
| bs=100 sw=10 sl=64 | p95 | 116,355 us | 118,043 us | 108,652 us | -1.4% | +7.1% |
| bs=100 sw=10 sl=64 | p99 | 116,355 us | 118,043 us | 116,310 us | -1.4% | +0.0% |
| bs=1000 sw=10 sl=64 | throughput | 1,420 tuples/sec | 1,452 tuples/sec | 1,004 tuples/sec | -2.2% | +41.4% |
| bs=1000 sw=10 sl=64 | MB/s | 0.867 MB/s | 0.886 MB/s | 0.613 MB/s | -2.1% | +41.5% |
| bs=1000 sw=10 sl=64 | p50 | 709,410 us | 685,868 us | 999,606 us | +3.4% | -29.0% |
| bs=1000 sw=10 sl=64 | p95 | 750,195 us | 726,383 us | 1,046,770 us | +3.3% | -28.3% |
| bs=1000 sw=10 sl=64 | p99 | 750,195 us | 726,383 us | 1,076,937 us | +3.3% | -30.3% |
Raw CSV
config_idx,batch_size,schema_width,string_len,num_batches,total_ms,total_tuples,total_bytes,tuples_per_sec,mb_per_sec,lat_p50_us,lat_p95_us,lat_p99_us
0,10,10,64,20,416.67,200,128000,480,0.293,21122.26,24555.54,24555.54
1,100,10,64,20,1661.68,2000,1280000,1204,0.735,82050.90,116354.63,116354.63
2,1000,10,64,20,14080.79,20000,12800000,1420,0.867,709409.83,750194.77,750194.77Add `actively-supporting` (boolean, defaults to true) to each release-branches entry. Only actively-supporting branches are auto-labeled on new fix: PRs; an inactive branch stays a valid manual backport target (its apply-check and post-merge backport still run when labeled by hand) but isn't offered by default. release/v1.2 is active, release/v1.1 is not. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Yicong-Huang
force-pushed
the
6940-backport-lifecycle
branch
from
July 27, 2026 21:17
76b4f99 to
7c612da
Compare
Codecov Report✅ All modified and coverable lines are covered by tests. Additional details and impacted files@@ Coverage Diff @@
## main #6941 +/- ##
============================================
- Coverage 78.96% 78.94% -0.02%
+ Complexity 3798 3794 -4
============================================
Files 1161 1161
Lines 46084 46084
Branches 5110 5110
============================================
- Hits 36388 36381 -7
- Misses 8076 8078 +2
- Partials 1620 1625 +5
☔ View full report in Codecov by Harness. 🚀 New features to boost your workflow:
|
This was referenced Jul 28, 2026
renovate-bot
pushed a commit
to renovate-bot/apache-_-texera
that referenced
this pull request
Jul 28, 2026
…ts as neutral (apache#6959) ### What changes were proposed in this PR? Follow-up to apache#6941, improving the backport CI/PRs so they stop reading as failures: - **A backport conflict no longer shows a red X.** Moved backport preflight/build out of `required-checks.yml` into a dedicated **Backport Checks** workflow, and report a conflict as a **`neutral`** (grey) check instead of a failing job. Still advisory, still never blocks merge (backport was never a required context). - **Shared precheck.** Extracted `precheck` into a reusable workflow so the label→stack mapping lives in one place. - **Nicer auto-opened draft backport PRs:** opened as `github-actions[bot]`, follow the PR template, title no longer links back to the original PR (trailing `(#N)` stripped), and the body carries the original PR's linked issue/discussion. ### Any related issues, documentation, discussions? Closes apache#6954. Follow-up to apache#6941. ### How was this PR tested? `actionlint` and `node --check` clean on all workflows; the apply-check shell/jq and the title/related-section logic were exercised locally. Full behavior runs on a real backport-labeled PR. ### Was this PR authored or co-authored using generative AI tooling? Generated-by: Claude Code (Opus 4.8)
renovate-bot
pushed a commit
to renovate-bot/apache-_-texera
that referenced
this pull request
Jul 28, 2026
### What changes were proposed in this PR? The backport auto-label workflow (`.github/workflows/backport-auto-label.yml`) decides, per actively-supported release branch, whether to add a `release/*` label to a `fix:` PR. Until now every skip was silent — visible only in the Actions log. Its cheap "is this feature on the branch?" check compares **exact file paths**, so a file moved or renamed after the branch was cut reads as *absent* and the label is dropped with no signal to the author. apache#6960 hit exactly this: its files live under `coordinator/` on `main` but `controller/` on `release/v1.2`, so a fix that genuinely applied to v1.2 was never labeled. This PR makes the workflow write its reasoning back to the PR as a single report comment: - **One row per actively-supported release branch**, with the decision and why: - ✅ **labeled** — change detected on the branch; label added (and who was requested for review); -⚠️ **skipped** — none of the modified files exist on the branch (the files are listed), with a prompt to check and add the label by hand if the fix should be backported; - 🚫 **declined** — a previously removed label (opt-out), not re-added. - The comment is **upserted in place** via a hidden marker, so `edited` re-runs update the same comment instead of stacking new ones (editing a comment sends no notification, so re-runs stay quiet). - It links the auto-label run, and is authored by `github-actions[bot]` like the existing review request (no label PAT needed). - Inactive branches and non-`fix:` PRs are unchanged (no comment); a `no-backport-needed` PR gets a one-line note instead of the table. Labeling behavior itself is unchanged — this only surfaces the decisions that were already being made. ### Any related issues, documentation, discussions? Resolves apache#6961. Follow-up to the backport lifecycle automation (apache#6941, apache#6959). Motivated by apache#6960, which was silently not labeled for `release/v1.2`. ### How was this PR tested? - YAML parses and the embedded github-script passes `node --check` (async-wrapped). - Walked the branch/label/timeline/file-status cases against apache#6960's real data to confirm the rows and skip reasons render as intended. - Full end-to-end exercise needs a live `pull_request_target` event with the org PAT/secrets, which only runs once merged. ### Was this PR authored or co-authored using generative AI tooling? Generated-by: Claude Code (Claude Opus 4.8) Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What changes were proposed in this PR?
This PR overhauls the release-backport lifecycle so it stops depending on people remembering. Three problems today, each addressed below.
Problem 1 — Backports are forgotten. The
release/*label is added by hand, so authors and committers routinely forget to request a backport, and fixes silently never reach the release branch.Solution: Backporting becomes opt-out. A new
backport-auto-labelworkflow labels everyfix:PR intomainwith the configuredrelease/*targets and requests review from each branch's release manager..github/release-branches.ymlis the single source of truth mapping each release branch to its manager and anactively-supportingflag (release/v1.2→xuang7, active;release/v1.1→bobbai00, inactive), parsed by a stdlib-only helper. Only actively-supporting branches are auto-labeled; an inactive branch stays a valid manual target but isn't offered by default. Remove a label to decline — a removal is remembered via the timeline, so a later edit never silently re-adds it.Problem 2 — Failed backports are lost. When a backport conflicts, the only trace is a PR comment that sinks to the bottom and gets forgotten; nobody is on the hook to finish it. And a conflicting cherry-pick still burns the full build matrix.
Solution: The pre-merge backport is split into a fast, git-only apply-check that gates the expensive build (a conflict is reported in seconds and no longer spins up the stacks), and
backportis removed from the Required Checks aggregator so it's advisory, not a merge gate. Post-merge,direct-backport-pushclassifies each target from that signal: green cherry-picks straight to the release branch; red (conflict, or applies-but-won't-build) auto-opens a draft backport PRfix(scope, vX.Y): …onbackport/<PR>-<slug>-<target>, with the conflicted tree committed, assigned to the author with the release manager as reviewer — an actionable task instead of a lost comment.Problem 3 — Fixes for main-only features get backported anyway. A fix for a feature that only exists on
mainshouldn't go to the release at all, but it still gets labeled and attempted.Solution: Automatic feature-absent skip — before labeling a target (and again on the red path), skip it when every file the PR modifies (added files excluded) is absent on that release branch: the feature clearly isn't there. Any ambiguity keeps the label and lets the apply-check decide. For what automation can't catch (new code inside an existing file, or pure judgment), a
no-backport-neededlabel is a hard manual veto honored end-to-end (auto-label, pre-merge precheck, and post-merge push).Note: a red apply-check now shows as a (non-blocking) red check on the PR — the intended "this needs a manual backport" signal, not a merge blocker.
Any related issues, documentation, discussions?
Resolves #6940.
How was this PR tested?
Static:
actionlinton all workflows; YAML parses; everygithub-scriptblock passesnode --check;bash -non the scripts; unit-checked the parser, branch-slug derivation,fix(scope, vX.Y)title injection, and the feature-absent git guard.End-to-end on a fork test harness (
Yicong-Huang/texera, throwawayrelease/test-clean+release/test-conflictbranches, docs-only fixes to skip the heavy build):fix:PR into main → auto-labeled with both targetsef1df63· runfix(scope, vX.Y): …, conflict markers committed, assigned to authorno-backport-needed→ all backport work vetoedNot exercised on the fork: the request-review success path (the test manager was the PR author, so it correctly took the skip branch) and a full heavy-build green path (docs-only by design).
Was this PR authored or co-authored using generative AI tooling?
Generated-by: Claude Code (Claude Opus 4.8)