Cross-Repository Workflow Coordination Skill

Overview

GitHub Actions workflows in the project-beta ecosystem use self-hosted runners. When runner configurations change, ALL repositories using those runners need coordinated updates.

Architecture

matchpoint-github-runners-helm
├── Defines runnerScaleSetName: "arc-beta-runners"
└── ArgoCD deploys runners with this label

project-beta-frontend
project-beta-api           } Must use: runs-on: arc-beta-runners
project-beta

Critical Rule: Workflow runs-on: MUST EXACTLY match Helm runnerScaleSetName

The Coordination Problem

Issue #121 Example

Change: Update runnerScaleSetName from arc-runners to arc-beta-runners

Impact:

matchpoint-github-runners-helm
  ✅ runnerScaleSetName: "arc-beta-runners"

project-beta-frontend (15 workflows)
  ❌ runs-on: arc-runners  # OLD label - jobs stuck!

project-beta-api (13 workflows)
  ❌ runs-on: arc-runners  # OLD label - jobs stuck!

project-beta (3 workflows)
  ❌ runs-on: arc-runners  # OLD label - jobs stuck!

Result: All CI jobs stuck in "queued" state until workflows updated.

Affected Repositories

Repository	Workflows	Runner Labels	Priority
project-beta-frontend	15 files	arc-beta-runners	P0 - Blocks deploys
project-beta-api	13 files	arc-beta-runners	P0 - Blocks deploys
project-beta	3 files	arc-beta-runners	P0 - Blocks infra

Coordination Workflow

Phase 1: Planning

Before changing runnerScaleSetName, audit all repositories:

# Search for current runner label usage
for repo in project-beta-frontend project-beta-api project-beta; do
  echo "=== $repo ==="
  cd /path/to/$repo
  grep -r "runs-on:" .github/workflows/ | grep -v "ubuntu-latest" | sort -u
done

Output example:

=== project-beta-frontend ===
.github/workflows/ci.yaml:    runs-on: arc-runners
.github/workflows/deploy.yaml:    runs-on: arc-runners
...

=== project-beta-api ===
.github/workflows/test.yaml:    runs-on: arc-runners
...

Document the changes needed:

• Count of files per repository
• Specific workflow files affected
• Any workflows using different labels

Phase 2: Create Migration Plan

Option A: Dual Runner Pools (Zero Downtime)

Deploy BOTH old and new runner pools during transition:

# matchpoint-github-runners-helm/argocd/applicationset-runners.yaml
generators:
- list:
    elements:
    - name: arc-runners           # OLD - for existing workflows
      valuesFile: examples/runners-values-old.yaml
    - name: arc-beta-runners      # NEW - for updated workflows
      valuesFile: examples/runners-values-new.yaml

Timeline:

1. Deploy both runner pools
2. Update workflows in all repos (can be done gradually)
3. Remove old runner pool after all workflows migrated

Pros:

• Zero downtime
• Safe rollback (revert workflow changes)
• Can update repos independently

Cons:

• 2x runner costs during migration
• Need to track which repos migrated

Option B: Coordinated Single Cutover

Update runner AND all workflows simultaneously:

1. Prepare PRs in ALL repositories (don't merge)
2. Merge runner config change
3. Wait for ArgoCD sync (~3 min)
4. Merge ALL workflow PRs quickly
5. Monitor for stuck jobs

Pros:

• No extra runner costs
• Clean cutover

Cons:

• ~3-5 minute CI outage
• Requires coordination across repos
• Risky if issues arise

Recommended: Option A for production, Option B for dev/test

Phase 3: Update Workflows

For each repository, create a PR that updates ALL workflow files:

# Script: update-runner-labels.sh
#!/bin/bash

OLD_LABEL="arc-runners"
NEW_LABEL="arc-beta-runners"
REPO=$1

cd /path/to/$REPO

# Find all workflow files
WORKFLOWS=$(find .github/workflows -name "*.ya*ml")

# Update each file
for workflow in $WORKFLOWS; do
  if grep -q "runs-on: $OLD_LABEL" "$workflow"; then
    echo "Updating: $workflow"
    sed -i "s/runs-on: $OLD_LABEL/runs-on: $NEW_LABEL/g" "$workflow"
  fi
done

# Create PR
git checkout -b fix/update-runner-label-to-$NEW_LABEL
git add .github/workflows/
git commit -m "ci: Update runner label from $OLD_LABEL to $NEW_LABEL

Aligns with runner configuration change in matchpoint-github-runners-helm.

Refs: matchpoint-ai/matchpoint-github-runners-helm#121"

git push -u origin fix/update-runner-label-to-$NEW_LABEL

gh pr create \
  --title "ci: Update runner label from $OLD_LABEL to $NEW_LABEL" \
  --body "Updates all workflows to use the new runner label \`$NEW_LABEL\`.

## Context
matchpoint-github-runners-helm changed \`runnerScaleSetName\` to \`$NEW_LABEL\`.

## Changes
- Updates all \`.github/workflows/*.yaml\` files
- Changes \`runs-on: $OLD_LABEL\` → \`runs-on: $NEW_LABEL\`

## Testing
- [ ] Verify workflows use correct runner label
- [ ] Confirm CI jobs execute (not stuck in queue)

Related: matchpoint-ai/matchpoint-github-runners-helm#121"

Usage:

./update-runner-labels.sh project-beta-frontend
./update-runner-labels.sh project-beta-api
./update-runner-labels.sh project-beta

Phase 4: Verification

After merging workflow updates:

# Check that runners are picking up jobs
gh run list --repo Matchpoint-AI/project-beta-frontend --limit 5

# Verify no jobs stuck in queue
gh run list --repo Matchpoint-AI/project-beta-frontend --status queued

# Check runner status
gh api /orgs/Matchpoint-AI/actions/runners --jq '.runners[] | {name, status, busy, labels: [.labels[].name]}'

Success criteria:

• ✅ No jobs stuck in "queued" for > 2 minutes
• ✅ Jobs transition to "in_progress" quickly
• ✅ Runners show "busy: true" when jobs running

Common Scenarios

Scenario 1: Adding New Runner Pool

Example: Add dedicated runners for frontend with GPU support

Steps:

1. Add runner pool in matchpoint-github-runners-helm:

   # argocd/applicationset-runners.yaml
   - name: arc-frontend-gpu
     valuesFile: examples/frontend-gpu-values.yaml
   

2. Update ONLY affected workflows in project-beta-frontend:

   # .github/workflows/e2e-visual-tests.yaml
   jobs:
     visual-tests:
       runs-on: arc-frontend-gpu  # NEW pool
   

3. Keep other workflows on existing pool:

   # .github/workflows/ci.yaml
   jobs:
     test:
       runs-on: arc-beta-runners  # Existing pool
   

Impact: Only workflows explicitly updated use new pool

Scenario 2: Removing Runner Pool

Example: Deprecate arc-runners in favor of arc-beta-runners

Steps:

1. Ensure NO workflows reference old label:

   for repo in project-beta-frontend project-beta-api project-beta; do
     cd /path/to/$repo
     grep -r "runs-on: arc-runners" .github/workflows/ && echo "❌ Found old label in $repo"
   done
   

2. Remove runner pool from matchpoint-github-runners-helm:

   # argocd/applicationset-runners.yaml
   # Remove the arc-runners entry
   

3. Verify no queued jobs after removal:

   gh run list --status queued --limit 20
   

Scenario 3: Emergency Runner Failover

Example: Primary runner pool down, need to switch to backup

Steps:

1. Deploy backup runner pool (if not already deployed):

   # Quick deploy via ArgoCD
   kubectl apply -f argocd/applications/arc-backup-runners.yaml
   

2. Bulk update workflows in critical repo:

   # Emergency script
   find .github/workflows -name "*.yaml" -exec sed -i 's/runs-on: arc-beta-runners/runs-on: arc-backup-runners/g' {} \;
   git add .github/workflows/
   git commit -m "EMERGENCY: Switch to backup runners"
   git push
   

3. Monitor job execution:

   watch -n 5 'gh run list --limit 10'
   

Validation Scripts

Pre-Merge Validation

Run before merging runner configuration changes:

#!/bin/bash
# scripts/validate-runner-labels.sh

set -euo pipefail

RUNNER_LABEL=$1
REPOS=("project-beta-frontend" "project-beta-api" "project-beta")

echo "🔍 Checking if workflows use runner label: $RUNNER_LABEL"

for repo in "${REPOS[@]}"; do
  echo ""
  echo "=== $repo ==="

  if [ ! -d "../$repo" ]; then
    echo "⚠️  Repository not found: ../$repo"
    continue
  fi

  cd "../$repo"

  MATCHES=$(grep -r "runs-on: $RUNNER_LABEL" .github/workflows/ 2>/dev/null | wc -l)

  if [ "$MATCHES" -gt 0 ]; then
    echo "✅ Found $MATCHES workflow jobs using $RUNNER_LABEL"
    grep -r "runs-on: $RUNNER_LABEL" .github/workflows/ | head -5
  else
    echo "❌ No workflows use $RUNNER_LABEL"
  fi

  cd - > /dev/null
done

Usage:

cd matchpoint-github-runners-helm
./scripts/validate-runner-labels.sh arc-beta-runners

Post-Merge Validation

Run after merging workflow updates:

#!/bin/bash
# scripts/verify-ci-not-stuck.sh

set -euo pipefail

REPOS=("Matchpoint-AI/project-beta-frontend" "Matchpoint-AI/project-beta-api" "Matchpoint-AI/project-beta")

echo "🔍 Checking for stuck CI jobs..."

for repo in "${REPOS[@]}"; do
  echo ""
  echo "=== $repo ==="

  QUEUED=$(gh run list --repo "$repo" --status queued --limit 50 --json databaseId,createdAt,status | jq -r '.[] | select(.status == "queued") | "\(.databaseId) - queued since \(.createdAt)"')

  if [ -z "$QUEUED" ]; then
    echo "✅ No queued jobs"
  else
    echo "⚠️  Found queued jobs:"
    echo "$QUEUED"

    # Check if any queued > 5 minutes
    STUCK=$(echo "$QUEUED" | jq -r 'select(now - (.createdAt | fromdateiso8601) > 300)')
    if [ -n "$STUCK" ]; then
      echo "❌ Jobs stuck for > 5 minutes!"
    fi
  fi
done

Usage:

./scripts/verify-ci-not-stuck.sh

Troubleshooting

Error: Jobs Stuck After Runner Change

Symptom: CI jobs stuck in "queued" after runner label change

Diagnosis:

# Check what label runners have
kubectl get autoscalingrunnerset -A -o jsonpath='{.items[*].spec.runnerScaleSetName}'

# Check what label workflows use
for repo in project-beta-frontend project-beta-api project-beta; do
  cd ../$repo
  grep -h "runs-on:" .github/workflows/* | sort -u
done

Fix:

# If mismatch found, update workflows
cd ../project-beta-frontend
find .github/workflows -name "*.yaml" -exec sed -i 's/runs-on: OLD_LABEL/runs-on: NEW_LABEL/g' {} \;
git commit -am "fix: Update runner label to match deployed runners"
git push

Error: Some Repos Updated, Others Not

Symptom: CI works in some repos but not others

Diagnosis:

# Check each repo's workflows
for repo in project-beta-frontend project-beta-api project-beta; do
  echo "=== $repo ==="
  cd ../$repo
  grep -h "runs-on:" .github/workflows/* | sort -u
  cd -
done

Fix: Update remaining repos using update script

Error: Runners Deployed But Not Registering

Symptom: Runners deployed but GitHub doesn't show them

Diagnosis:

# Check GitHub runners
gh api /orgs/Matchpoint-AI/actions/runners --jq '.runners[] | {name, labels: [.labels[].name]}'

# Check Kubernetes runners
kubectl get pods -n arc-beta-runners -l app.kubernetes.io/component=runner

Fix: See arc-runner-troubleshooting

Best Practices

1. Plan multi-repo changes in advance - Don't surprise developers with stuck CI
2. Use dual runner pools during migration - Eliminates downtime
3. Communicate changes - Post in team chat before merging
4. Verify in dev first - Test runner changes in development repo
5. Monitor after deployment - Watch for queued jobs for 30 minutes post-change
6. Document runner labels - Keep README updated with current label names
7. Automate validation - Run validation scripts in CI for runner config changes

Coordination Checklist

Before changing runnerScaleSetName:

• Audit all repos for workflow label usage
• Document count of files per repo needing updates
• Choose migration strategy (dual pool vs cutover)
• Prepare PRs for all affected repos
• Communicate change timeline to team
• Deploy runner config change
• Wait for ArgoCD sync (verify runners online)
• Merge workflow PRs
• Verify CI jobs execute successfully
• Monitor for stuck jobs (30 minutes)
• Clean up old runner pool (if dual pool strategy)

Related Skills

• arc-runner-troubleshooting - Runner registration issues
• argocd-bootstrap - Runner deployment via ArgoCD
• infrastructure-cd - Automated deployment workflow

Related Issues

• #121 - releaseName/runnerScaleSetName mismatch causing empty labels
• #123 - Cross-repo label update coordination
• #112 - CI jobs stuck investigation
• project-beta-api#798 - Workflow label update
• project-beta-frontend#886 - CI blocked by label mismatch

References

• GitHub Actions: runs-on
• ARC: Using Runners in Workflows