service-recovery.yml

name: Service Recovery

on:
  workflow_dispatch:
  pull_request:
    paths:
      - "Cargo.toml"
      - "Cargo.lock"
      - "src/**"
      - "tests/**"
      - "resources/**"
      - ".github/workflows/service-integration.yml"
      - ".github/workflows/service-recovery.yml"
  push:
    branches:
      - "**"
    paths:
      - "Cargo.toml"
      - "Cargo.lock"
      - "src/**"
      - "tests/**"
      - "resources/**"
      - ".github/workflows/service-integration.yml"
      - ".github/workflows/service-recovery.yml"

permissions:
  contents: read

jobs:
  recovery:
    name: Verify service recovery (${{ matrix.os }})
    runs-on: ${{ matrix.os }}
    timeout-minutes: 20
    strategy:
      fail-fast: false
      matrix:
        os: [ubuntu-latest, macos-latest, windows-latest]
    steps:
      - name: Checkout repository
        uses: actions/checkout@v6

      - name: Install Rust toolchain
        uses: dtolnay/rust-toolchain@master
        with:
          toolchain: stable

      - name: Build service binaries
        run: cargo build --release --features "standalone client"

      - name: Install service (Unix)
        if: runner.os != 'Windows'
        run: sudo target/release/clash-verge-service-install --debug

      - name: Kill service and verify recovery (Linux)
        if: runner.os == 'Linux'
        run: |
          set -euo pipefail

          service="clash-verge-service.service"
          target/release/service-integration-driver ping

          old_pid="$(systemctl show "$service" --property=MainPID --value)"
          if [ -z "$old_pid" ] || [ "$old_pid" = "0" ]; then
            sudo systemctl status "$service" --no-pager
            exit 1
          fi

          sudo kill -9 "$old_pid"

          for _ in {1..40}; do
            sleep 1
            new_pid="$(systemctl show "$service" --property=MainPID --value)"
            active_state="$(systemctl show "$service" --property=ActiveState --value)"
            if [ "$active_state" = "active" ] && [ -n "$new_pid" ] && [ "$new_pid" != "0" ] && [ "$new_pid" != "$old_pid" ]; then
              target/release/service-integration-driver ping
              echo "Recovered service PID $old_pid -> $new_pid"
              exit 0
            fi
          done

          sudo systemctl status "$service" --no-pager
          exit 1

      - name: Kill service and verify recovery (macOS)
        if: runner.os == 'macOS'
        run: |
          set -euo pipefail

          service_id="io.github.clash-verge-rev.clash-verge-rev.service"
          target/release/service-integration-driver ping

          get_pid() {
            sudo launchctl print "system/$service_id" 2>/dev/null | awk -F'= ' '/^[[:space:]]*pid = / { print $2; exit }' || true
          }

          old_pid="$(get_pid)"
          if [ -z "$old_pid" ] || [ "$old_pid" = "-" ]; then
            sudo launchctl print "system/$service_id"
            exit 1
          fi

          sudo kill -9 "$old_pid"

          for _ in {1..40}; do
            sleep 1
            new_pid="$(get_pid)"
            if [ -n "$new_pid" ] && [ "$new_pid" != "-" ] && [ "$new_pid" != "$old_pid" ]; then
              target/release/service-integration-driver ping
              echo "Recovered service PID $old_pid -> $new_pid"
              exit 0
            fi
          done

          sudo launchctl print "system/$service_id"
          exit 1

      - name: Install service (Windows)
        if: runner.os == 'Windows'
        shell: pwsh
        run: |
          & "$env:GITHUB_WORKSPACE\target\release\clash-verge-service-install.exe"
          Start-Sleep -Seconds 2

      - name: Kill service and verify recovery (Windows)
        if: runner.os == 'Windows'
        shell: pwsh
        run: |
          $serviceName = "clash_verge_service"
          $driver = "$env:GITHUB_WORKSPACE\target\release\service-integration-driver.exe"

          & $driver ping
          if ($LASTEXITCODE -ne 0) {
            throw "IPC was not reachable before killing the service process"
          }

          $service = Get-CimInstance Win32_Service -Filter "Name='$serviceName'"
          $oldPid = [int]$service.ProcessId
          if ($oldPid -le 0) {
            throw "Service is not running"
          }

          Stop-Process -Id $oldPid -Force

          $deadline = (Get-Date).AddSeconds(60)
          while ((Get-Date) -lt $deadline) {
            Start-Sleep -Seconds 1
            $service = Get-CimInstance Win32_Service -Filter "Name='$serviceName'"
            $newPid = [int]$service.ProcessId
            if ($service.State -eq "Running" -and $newPid -gt 0 -and $newPid -ne $oldPid) {
              & $driver ping
              if ($LASTEXITCODE -ne 0) {
                throw "IPC did not respond after service restart"
              }
              Write-Host "Recovered service PID $oldPid -> $newPid"
              exit 0
            }
          }

          sc.exe queryex $serviceName
          throw "Service did not restart after forced process exit"

      - name: Uninstall service (Unix)
        if: always() && runner.os != 'Windows'
        run: |
          if [ -x target/release/clash-verge-service-uninstall ]; then
            sudo target/release/clash-verge-service-uninstall --debug
          fi

      - name: Uninstall service (Windows)
        if: always() && runner.os == 'Windows'
        shell: pwsh
        run: |
          $service = Get-Service -Name "clash_verge_service" -ErrorAction SilentlyContinue
          if ($null -ne $service) {
            & "$env:GITHUB_WORKSPACE\target\release\clash-verge-service-uninstall.exe"
          }