name: Node Failover Simulation

on:
  workflow_dispatch:

concurrency:
  group: node-failover-simulation-${{ github.ref }}
  cancel-in-progress: true

jobs:
  failover-test:
    runs-on: debian
    timeout-minutes: 15

    steps:
      - name: Clone repository
        run: |
          WORKSPACE="/var/lib/aitbc-workspaces/node-failover-simulation"
          rm -rf "$WORKSPACE"
          mkdir -p "$WORKSPACE"
          cd "$WORKSPACE"
          git clone --depth 1 http://gitea.bubuit.net:3000/oib/aitbc.git repo

      - name: Initialize job logging
        run: |
          cd /var/lib/aitbc-workspaces/node-failover-simulation/repo
          bash scripts/ci/setup-job-logging.sh

      - name: Setup Python environment
        run: |
          cd /var/lib/aitbc-workspaces/node-failover-simulation/repo
          
          # Remove any existing venv to avoid cache corruption issues
          rm -rf venv
          
          bash scripts/ci/setup-python-venv.sh \
            --repo-dir "$PWD" \
            --venv-dir "$PWD/venv" \
            --skip-requirements \
            --extra-packages "requests psutil"

      - name: Run node failover simulation
        run: |
          cd /var/lib/aitbc-workspaces/node-failover-simulation/repo
          bash scripts/multi-node/failover-simulation.sh

      - name: Failover simulation report
        if: always()
        run: |
          echo "=== Node Failover Simulation Report ==="
          if [ -f /var/log/aitbc/failover-simulation.log ]; then
            tail -50 /var/log/aitbc/failover-simulation.log
          fi

      - name: Cleanup
        if: always()
        run: rm -rf /var/lib/aitbc-workspaces/node-failover-simulation