Add more clarity to how to re-run failed AWS GPU jobs

.github/workflows/aws_gpu_benchmarks.yml

@@ -131,6 +131,44 @@ jobs:
           uvx asv compare --split ${{ inputs.base_ref }} ${{ inputs.ref }}
           exit 2
 
+      - name: Re-run instructions
+        if: failure()
+        run: |
+          # Create error annotations (appear at top of job summary)
+          echo "::error::DO NOT use 'Re-run failed jobs' - the EC2 runner no longer exists and your job will be queued forever."
+          echo "::error::USE 'Re-run all jobs' instead to start a fresh EC2 runner."
+
+          # Write to job summary (appears in Summary tab)
+          cat >> "$GITHUB_STEP_SUMMARY" << 'EOF'
+          ## ⚠️ How to Re-run This Workflow
+
+          This workflow uses **ephemeral EC2 runners** that are terminated after each run.
+
+          | | Option | Result |
+          |---|--------|--------|
+          | ❌ | **Re-run failed jobs** | Runner no longer exists → job queued forever |
+          | ✅ | **Re-run all jobs** | Starts new EC2 runner → benchmarks re-run |
+          EOF
+
+          # Also print to log for completeness
+          cat << 'EOF'
+
+          ================================================================================
+          ⚠️  IMPORTANT: HOW TO RE-RUN THIS WORKFLOW
+          ================================================================================
+
+          This workflow uses ephemeral EC2 runners that are terminated after each run.
+
+          ❌ DO NOT select "Re-run failed jobs"
+             → The runner no longer exists and your job will be queued forever.
+
+          ✅ DO select "Re-run all jobs"
+             → This will start a new EC2 runner and re-run the benchmarks.
+
+          ================================================================================
+
+          EOF
+
   stop-runner:
     name: Stop self-hosted EC2 runner
     runs-on: ubuntu-latest

.github/workflows/aws_gpu_tests.yml

@@ -118,6 +118,44 @@ jobs:
           flags: unittests
           token: ${{ secrets.CODECOV_TOKEN }}
 
+      - name: Re-run instructions
+        if: failure()
+        run: |
+          # Create error annotations (appear at top of job summary)
+          echo "::error::DO NOT use 'Re-run failed jobs' - the EC2 runner no longer exists and your job will be queued forever."
+          echo "::error::USE 'Re-run all jobs' instead to start a fresh EC2 runner."
+
+          # Write to job summary (appears in Summary tab)
+          cat >> "$GITHUB_STEP_SUMMARY" << 'EOF'
+          ## ⚠️ How to Re-run This Workflow
+
+          This workflow uses **ephemeral EC2 runners** that are terminated after each run.
+
+          | | Option | Result |
+          |---|--------|--------|
+          | ❌ | **Re-run failed jobs** | Runner no longer exists → job queued forever |
+          | ✅ | **Re-run all jobs** | Starts new EC2 runner → tests re-run |
+          EOF
+
+          # Also print to log for completeness
+          cat << 'EOF'
+
+          ================================================================================
+          ⚠️  IMPORTANT: HOW TO RE-RUN THIS WORKFLOW
+          ================================================================================
+
+          This workflow uses ephemeral EC2 runners that are terminated after each run.
+
+          ❌ DO NOT select "Re-run failed jobs"
+             → The runner no longer exists and your job will be queued forever.
+
+          ✅ DO select "Re-run all jobs"
+             → This will start a new EC2 runner and re-run the tests.
+
+          ================================================================================
+
+          EOF
+
   stop-runner:
     name: Stop self-hosted EC2 runner
     runs-on: ubuntu-latest

.github/workflows/push_aws_gpu.yml

@@ -13,5 +13,6 @@ on:
 
 jobs:
   run-tests:
+    if: github.repository == 'newton-physics/newton'
     uses: ./.github/workflows/aws_gpu_tests.yml
     secrets: inherit