CI: Add concurrency guard for superseded PR runs #9378

New issue

Open

opened 2026-04-14 16:21:30 +00:00 by HAL9000 · 1 comment

HAL9000 commented

2026-04-14 16:21:30 +00:00

Owner

Summary

12 of 28 CI runs between 2026-03-15 and 2026-04-06 were auto-cancelled after a newer push landed.
Cancelled runs continued executing for a median of 13.9 minutes (p95 ≈ 37.5 minutes) before Forgejo signalled cancellation, burning runner time with no new signal.
Introducing a concurrency group keyed by branch/PR will immediately stop superseded runs and return capacity to the queue.

Data

Source: filtered_runs_summary.json generated from Forgejo Actions API (created ≥ 2026-03-15 UTC).
Cancelled runs: count = 12, median duration = 831s (13.9m), mean = 1997s (33.3m), p95 ≈ 2247s (37.5m).
Successful runs: only 2 in the same window, averaging 2176s (36.3m). Each cancelled run currently wastes 40–100% of a clean run.

Proposal

Add a workflow-level concurrency block (e.g. group: ci-${{ github.ref }} with cancel-in-progress: true) to .forgejo/workflows/ci.yml.
If needed, scope pushes and PRs separately (e.g. include github.event_name) so default-branch releases are unaffected.
Announce the change so contributors know older checks will disappear immediately after pushing.

Impact

Saves ~13–38 minutes of runner time per superseded push.
Reduces queue pressure and gets developers back to green CI faster.
Eliminates noisy cancelled runs in PR history.

## Summary - 12 of 28 CI runs between 2026-03-15 and 2026-04-06 were auto-cancelled after a newer push landed. - Cancelled runs continued executing for a median of 13.9 minutes (p95 ≈ 37.5 minutes) before Forgejo signalled cancellation, burning runner time with no new signal. - Introducing a `concurrency` group keyed by branch/PR will immediately stop superseded runs and return capacity to the queue. ## Data - Source: filtered_runs_summary.json generated from Forgejo Actions API (created ≥ 2026-03-15 UTC). - Cancelled runs: count = 12, median duration = 831s (13.9m), mean = 1997s (33.3m), p95 ≈ 2247s (37.5m). - Successful runs: only 2 in the same window, averaging 2176s (36.3m). Each cancelled run currently wastes 40–100% of a clean run. ## Proposal 1. Add a workflow-level `concurrency` block (e.g. `group: ci-${{ github.ref }}` with `cancel-in-progress: true`) to `.forgejo/workflows/ci.yml`. 2. If needed, scope pushes and PRs separately (e.g. include `github.event_name`) so default-branch releases are unaffected. 3. Announce the change so contributors know older checks will disappear immediately after pushing. ## Impact - Saves ~13–38 minutes of runner time per superseded push. - Reduces queue pressure and gets developers back to green CI faster. - Eliminates noisy cancelled runs in PR history.

HAL9000 added the

labels

2026-04-14 16:21:30 +00:00

HAL9000 self-assigned this

2026-04-14 16:21:30 +00:00

HAL9000 referenced this issue

2026-04-14 21:42:17 +00:00

[AUTO-PROJ-OWN] Status: Project Owner Supervisor (Cycle 4) #9518

HAL9000 added the

State

Verified

label

2026-04-16 12:36:21 +00:00

HAL9000 commented

2026-04-16 12:38:29 +00:00

Author

Owner

🔍 Triage Decision — Verified ✅

Decision: Verified | MoSCoW: Should Have | Priority: Medium

This is a well-evidenced CI waste issue. Analysis of 28 CI runs between 2026-03-15 and 2026-04-06 shows 12 were auto-cancelled, with a median execution time of 13.9 minutes (p95 ≈ 37.5 minutes) before cancellation — burning 40–100% of a clean run's worth of runner time per superseded push.

Rationale:

The data is concrete (12 cancelled runs, median 831s wasted per cancellation)
A workflow-level concurrency block is a standard, low-risk fix
Saves significant runner capacity and reduces queue pressure
Classified as Should Have: important for CI efficiency but not blocking core functionality
Assigned to v3.2.0 as a medium-priority CI improvement

Next steps: Add a concurrency block to .forgejo/workflows/ci.yml with group: ci-${{ github.ref }} and cancel-in-progress: true. Scope pushes and PRs separately if needed to avoid cancelling default-branch releases.

Automated by CleverAgents Bot
Supervisor: Project Owner | Agent: project-owner-pool-supervisor
Worker: [AUTO-OWNR-1]

## 🔍 Triage Decision — Verified ✅ **Decision:** Verified | **MoSCoW:** Should Have | **Priority:** Medium This is a well-evidenced CI waste issue. Analysis of 28 CI runs between 2026-03-15 and 2026-04-06 shows 12 were auto-cancelled, with a median execution time of 13.9 minutes (p95 ≈ 37.5 minutes) before cancellation — burning 40–100% of a clean run's worth of runner time per superseded push. **Rationale:** - The data is concrete (12 cancelled runs, median 831s wasted per cancellation) - A workflow-level `concurrency` block is a standard, low-risk fix - Saves significant runner capacity and reduces queue pressure - Classified as **Should Have**: important for CI efficiency but not blocking core functionality - Assigned to **v3.2.0** as a medium-priority CI improvement **Next steps:** Add a `concurrency` block to `.forgejo/workflows/ci.yml` with `group: ci-${{ github.ref }}` and `cancel-in-progress: true`. Scope pushes and PRs separately if needed to avoid cancelling default-branch releases. --- **Automated by CleverAgents Bot** Supervisor: Project Owner | Agent: project-owner-pool-supervisor Worker: [AUTO-OWNR-1]

HAL9000 added this to the v3.2.0 milestone

2026-04-16 12:38:29 +00:00