From 0856fb79d3abbacd6e526f0718d4c7e53177297d Mon Sep 17 00:00:00 2001
From: Ludvig Liljenberg <4257730+ludfjig@users.noreply.github.com>
Date: Wed, 11 Nov 2026 10:30:14 -0800
Subject: [PATCH] Stop restarting benchmarks on PR metadata edits

Benchmark and Benchmark Policy ran on edited and ready_for_review events. Editing a PR description or marking a draft ready restarted all 36 measurement jobs and cancelled any measurement in progress, though neither changes the head, base or skip label that eligibility reads.

Both workflows now run only when a PR opens, reopens, gets new commits or changes labels. A PR retargeted to main runs its checks on its next push.

Signed-off-by: Ludvig Liljenberg <4257730+ludfjig@users.noreply.github.com>
---
 .github/workflows/benchmark-policy.yml  | 2 +-
 .github/workflows/benchmark-trigger.yml | 2 +-
 docs/README.md                          | 4 ++--
 3 files changed, 4 insertions(+), 4 deletions(-)

diff --git a/.github/workflows/benchmark-policy.yml b/.github/workflows/benchmark-policy.yml
index 4074b7c..bc115ae 100644
--- a/.github/workflows/benchmark-policy.yml
+++ b/.github/workflows/benchmark-policy.yml
@@ -3,7 +3,7 @@ name: Benchmark Policy
 on:
   pull_request:
     branches: [main]
-    types: [opened, synchronize, reopened, labeled, unlabeled, edited]
+    types: [opened, synchronize, reopened, labeled, unlabeled]
 
 permissions:
   contents: read
diff --git a/.github/workflows/benchmark-trigger.yml b/.github/workflows/benchmark-trigger.yml
index 2cbfe4c..200d37c 100644
--- a/.github/workflows/benchmark-trigger.yml
+++ b/.github/workflows/benchmark-trigger.yml
@@ -5,7 +5,7 @@ on:
     branches: [main]
   pull_request:
     branches: [main]
-    types: [opened, synchronize, reopened, labeled, unlabeled, edited, ready_for_review]
+    types: [opened, synchronize, reopened, labeled, unlabeled]
 
 permissions:
   contents: read
diff --git a/docs/README.md b/docs/README.md
index a3c641c..22f6dfe 100644
--- a/docs/README.md
+++ b/docs/README.md
@@ -107,7 +107,8 @@ Pages skips a preview whose build lacks `index.html`, contains symbolic links, h
 ## CI Rules
 
 * `benchmarks: skip` replaces publishable measurements with one-request smoke coverage on KVM and MSHV. Smoke results are not published. Dependabot applies the label automatically.
-* Required checks evaluate the current PR policy on every subscribed event. Label and description edits restart the benchmark workflow. Measurements run unless `benchmarks: skip` is present.
+* Benchmark Policy and Benchmark run when a PR opens, reopens, gets new commits or changes labels. Any label change restarts measurement. Measurements run unless `benchmarks: skip` is present.
+* A PR retargeted to `main` needs a new push before its required checks run.
 * Each of the 36 measurement jobs runs all three strategies sequentially with a fresh server process for each. Retrying a job repeats its three strategies.
 * Preview builds run independently of benchmarks. Draft PRs have no preview.
 * Production uses published history. Previews include matching pending results when available.
@@ -119,7 +120,6 @@ after a direct push to `data`. The deploy job fails if the site still serves an
 older deployment after 5 minutes. Rerun the deploy job.
 
 Publication and Pages use serial queues with up to 100 pending jobs or runs each.
-Benchmark checks ignore unrelated labels and title or body edits.
 
 ## Security