dashcaddy/dashcaddy-api/scripts/dashcaddy-update.sh

#!/usr/bin/env bash
# DashCaddy Host-Side Updater
# Triggered by systemd path unit when the API container writes trigger.json.
# Handles API container rebuild + restart with automatic rollback on failure.

set -euo pipefail

readonly UPDATES_DIR="/opt/dashcaddy/updates"
readonly TRIGGER_FILE="${UPDATES_DIR}/trigger.json"
readonly RESULT_FILE="${UPDATES_DIR}/result.json"
readonly BACKUP_BASE="${UPDATES_DIR}/backups"
readonly HEALTH_URL="http://localhost:3001/health"
readonly HEALTH_TIMEOUT=60
readonly MAX_BACKUPS=3

log() { echo "[dashcaddy-update] $(date '+%Y-%m-%d %H:%M:%S') $*"; }

write_result() {
  local success="$1" message="$2" version="$3" duration="$4"
  cat > "$RESULT_FILE" <<EOF
{
  "success": ${success},
  "version": "${version}",
  "message": "${message}",
  "duration": ${duration},
  "timestamp": "$(date -u +%Y-%m-%dT%H:%M:%SZ)"
}
EOF
  log "Result written: success=${success} version=${version}"
}

backup_current() {
  local api_dir="$1" version="$2"
  local backup_dir="${BACKUP_BASE}/${version}"
  mkdir -p "$backup_dir/routes"

  cp -f "${api_dir}"/*.js "$backup_dir/" 2>/dev/null || true
  cp -rf "${api_dir}/routes/"* "$backup_dir/routes/" 2>/dev/null || true
  cp -f "${api_dir}/package.json" "$backup_dir/" 2>/dev/null || true
  cp -f "${api_dir}/package-lock.json" "$backup_dir/" 2>/dev/null || true
  cp -f "${api_dir}/Dockerfile" "$backup_dir/" 2>/dev/null || true
  cp -f "${api_dir}/openapi.yaml" "$backup_dir/" 2>/dev/null || true

  log "Backed up version ${version} to ${backup_dir}"
}

restore_backup() {
  local api_dir="$1" version="$2"
  local backup_dir="${BACKUP_BASE}/${version}"

  if [ ! -d "$backup_dir" ]; then
    log "ERROR: No backup found for version ${version}"
    return 1
  fi

  cp -f "${backup_dir}"/*.js "$api_dir/" 2>/dev/null || true
  cp -rf "${backup_dir}/routes/"* "${api_dir}/routes/" 2>/dev/null || true
  cp -f "${backup_dir}/package.json" "$api_dir/" 2>/dev/null || true
  cp -f "${backup_dir}/package-lock.json" "$api_dir/" 2>/dev/null || true
  cp -f "${backup_dir}/Dockerfile" "$api_dir/" 2>/dev/null || true
  cp -f "${backup_dir}/openapi.yaml" "$api_dir/" 2>/dev/null || true

  log "Restored version ${version} from ${backup_dir}"
}

copy_new_files() {
  local staging_dir="$1" api_dir="$2"

  cp -f "${staging_dir}"/*.js "$api_dir/" 2>/dev/null || true
  [ -d "${staging_dir}/routes" ] && cp -rf "${staging_dir}/routes/"* "${api_dir}/routes/" 2>/dev/null || true
  cp -f "${staging_dir}/package.json" "$api_dir/" 2>/dev/null || true
  cp -f "${staging_dir}/package-lock.json" "$api_dir/" 2>/dev/null || true
  cp -f "${staging_dir}/Dockerfile" "$api_dir/" 2>/dev/null || true
  cp -f "${staging_dir}/openapi.yaml" "$api_dir/" 2>/dev/null || true

  log "Copied new files from ${staging_dir} to ${api_dir}"
}

wait_for_health() {
  local attempt=0
  local max_attempts=$((HEALTH_TIMEOUT / 2))

  while (( attempt < max_attempts )); do
    if curl -fsS --max-time 3 "$HEALTH_URL" >/dev/null 2>&1; then
      log "Health check passed (attempt $((attempt+1)))"
      return 0
    fi
    sleep 2
    attempt=$((attempt + 1))
  done

  log "Health check FAILED after ${HEALTH_TIMEOUT}s"
  return 1
}

find_compose_dir() {
  # Find the docker-compose.yml for dashcaddy-api
  for dir in /etc/dashcaddy/sites/dashcaddy-api /etc/dashcaddy/sites/caddy-api; do
    if [ -f "${dir}/docker-compose.yml" ] || [ -f "${dir}/docker-compose.yaml" ]; then
      echo "$dir"
      return 0
    fi
  done
  # Fallback: same as api source
  echo "$1"
}

cleanup_old_backups() {
  if [ ! -d "$BACKUP_BASE" ]; then return; fi
  local count
  count=$(ls -1d "${BACKUP_BASE}"/*/ 2>/dev/null | wc -l)
  if (( count > MAX_BACKUPS )); then
    local to_remove=$((count - MAX_BACKUPS))
    ls -1d "${BACKUP_BASE}"/*/ 2>/dev/null | head -n "$to_remove" | while read -r dir; do
      rm -rf "$dir"
      log "Cleaned old backup: $dir"
    done
  fi
}

main() {
  if [ ! -f "$TRIGGER_FILE" ]; then
    log "No trigger file found, exiting"
    exit 0
  fi

  local start_time
  start_time=$(date +%s)

  # Parse trigger file
  local action version from_version staging_dir api_dir commit
  action=$(python3 -c "import json,sys; d=json.load(open('$TRIGGER_FILE')); print(d.get('action','update'))" 2>/dev/null || echo "update")
  version=$(python3 -c "import json,sys; d=json.load(open('$TRIGGER_FILE')); print(d.get('version','unknown'))" 2>/dev/null || echo "unknown")
  from_version=$(python3 -c "import json,sys; d=json.load(open('$TRIGGER_FILE')); print(d.get('fromVersion','unknown'))" 2>/dev/null || echo "unknown")
  staging_dir=$(python3 -c "import json,sys; d=json.load(open('$TRIGGER_FILE')); print(d.get('stagingDir',''))" 2>/dev/null || echo "")
  api_dir=$(python3 -c "import json,sys; d=json.load(open('$TRIGGER_FILE')); print(d.get('apiSourceDir','/opt/dashcaddy'))" 2>/dev/null || echo "/opt/dashcaddy")
  commit=$(python3 -c "import json,sys; d=json.load(open('$TRIGGER_FILE')); print(d.get('commit','unknown'))" 2>/dev/null || echo "unknown")

  log "=== DashCaddy ${action} started: ${from_version} → ${version} (${commit}) ==="

  if [ -z "$staging_dir" ] || [ ! -d "$staging_dir" ]; then
    log "ERROR: Staging directory not found: ${staging_dir}"
    write_result "false" "Staging directory not found" "$version" "0"
    rm -f "$TRIGGER_FILE"
    exit 1
  fi

  local compose_dir
  compose_dir=$(find_compose_dir "$api_dir")

  # Step 1: Backup current version
  log "Step 1: Backing up current version (${from_version})"
  backup_current "$api_dir" "$from_version"

  # Step 2: Copy new files
  log "Step 2: Copying new files"
  copy_new_files "$staging_dir" "$api_dir"

  # Write commit hash to VERSION file
  echo "$commit" > "${api_dir}/VERSION"

  # Step 3: Rebuild container
  log "Step 3: Building new container image"
  cd "$compose_dir"
  if ! docker compose build --build-arg DASHCADDY_COMMIT="$commit" --quiet 2>&1; then
    log "ERROR: docker compose build failed, rolling back"
    restore_backup "$api_dir" "$from_version"
    local elapsed=$(( $(date +%s) - start_time ))
    write_result "false" "Build failed, rolled back to ${from_version}" "$version" "$((elapsed * 1000))"
    rm -f "$TRIGGER_FILE"
    exit 1
  fi

  # Step 4: Restart container
  log "Step 4: Restarting container"
  if ! docker compose up -d 2>&1; then
    log "ERROR: docker compose up failed, rolling back"
    restore_backup "$api_dir" "$from_version"
    docker compose build --quiet 2>&1 || true
    docker compose up -d 2>&1 || true
    local elapsed=$(( $(date +%s) - start_time ))
    write_result "false" "Container start failed, rolled back to ${from_version}" "$version" "$((elapsed * 1000))"
    rm -f "$TRIGGER_FILE"
    exit 1
  fi

  # Step 5: Health check
  log "Step 5: Waiting for health check (${HEALTH_TIMEOUT}s timeout)"
  if wait_for_health; then
    local elapsed=$(( $(date +%s) - start_time ))
    log "=== Update to ${version} SUCCESSFUL (${elapsed}s) ==="
    write_result "true" "Update successful" "$version" "$((elapsed * 1000))"
  else
    log "Health check failed — ROLLING BACK to ${from_version}"
    restore_backup "$api_dir" "$from_version"
    cd "$compose_dir"
    docker compose build --quiet 2>&1 || true
    docker compose up -d 2>&1 || true

    if wait_for_health; then
      local elapsed=$(( $(date +%s) - start_time ))
      log "Rollback to ${from_version} succeeded"
      write_result "false" "Health check failed after update. Rolled back to ${from_version}." "$version" "$((elapsed * 1000))"
    else
      local elapsed=$(( $(date +%s) - start_time ))
      log "CRITICAL: Rollback also failed. Manual intervention required."
      write_result "false" "CRITICAL: Both update and rollback failed. Manual intervention required." "$version" "$((elapsed * 1000))"
    fi
  fi

  # Cleanup
  rm -f "$TRIGGER_FILE"
  rm -rf "${UPDATES_DIR}/staging"
  cleanup_old_backups

  log "Update process complete"
}

main "$@"