scripts/openclaw-backup-cron.sh

#!/usr/bin/env bash
#
# OpenClaw Backup Cron Script with Health Check Integration
# 集成健康检查的备份脚本
#

set -euo pipefail

BACKUP_ENV_FILE_PATH="${OPENCLAW_BACKUP_ENV_FILE_PATH:-/root/.env.d/openclaw-backup.env}"
BACKUP_LOG_DIR="${OPENCLAW_BACKUP_LOG_DIR:-/var/log/openclaw}"
BACKUP_LOG_FILE="${BACKUP_LOG_DIR}/backup.log"
HEALTH_CHECK_ENABLED="${OPENCLAW_BACKUP_HEALTH_CHECK_ENABLED:-false}"
HEALTH_CHECK_BEFORE="${OPENCLAW_BACKUP_HEALTH_CHECK_BEFORE:-false}"
HEALTH_CHECK_AFTER="${OPENCLAW_BACKUP_HEALTH_CHECK_AFTER:-false}"

# Timestamp function
timestamp() {
    date '+%Y-%m-%dT%H:%M:%S'
}

# Logging function - 只写入文件，避免Supervisor双重记录
log() {
    echo "[$(timestamp)] $*" >> "$BACKUP_LOG_FILE"
}

is_true() {
    local value
    value="$(printf '%s' "${1:-}" | tr '[:upper:]' '[:lower:]')"
    [[ "$value" == "1" || "$value" == "true" || "$value" == "yes" || "$value" == "on" ]]
}

# Load environment
load_env() {
    if [[ -f "$BACKUP_ENV_FILE_PATH" ]]; then
        # shellcheck disable=SC1091
        if ! source "$BACKUP_ENV_FILE_PATH"; then
            log "ERROR: Failed to source $BACKUP_ENV_FILE_PATH"
            return 1
        fi
        log "INFO: Loaded environment from $BACKUP_ENV_FILE_PATH"
    else
        log "WARN: Environment file not found: $BACKUP_ENV_FILE_PATH"
    fi
}

# Run health check
run_health_check() {
    if [[ "$HEALTH_CHECK_ENABLED" != "true" ]]; then
        log "INFO: Health check is disabled"
        return 0
    fi
    
    log "INFO: Running pre-backup health check..."
    
    if /usr/local/bin/openclaw-backup-health.sh --check >> "$BACKUP_LOG_FILE" 2>&1; then
        log "INFO: Health check passed"
        return 0
    else
        log "WARN: Health check failed, attempting repair..."
        
        # Try to repair
        if /usr/local/bin/openclaw-backup-health.sh --repair >> "$BACKUP_LOG_FILE" 2>&1; then
            log "INFO: Repair successful, retrying health check..."
            
            # Retry health check
            if /usr/local/bin/openclaw-backup-health.sh --check >> "$BACKUP_LOG_FILE" 2>&1; then
                log "INFO: Health check passed after repair"
                return 0
            else
                log "ERROR: Health check still failing after repair"
                return 1
            fi
        else
            log "ERROR: Repair failed"
            return 1
        fi
    fi
}

# Run backup with retry
run_backup_with_retry() {
    local max_retries="${OPENCLAW_BACKUP_MAX_RETRIES:-3}"
    local retry_count=0
    local backup_success=false
    
    while [[ $retry_count -lt $max_retries ]]; do
        ((retry_count++))
        log "INFO: Backup attempt $retry_count/$max_retries"
        
        local exit_code=0
        python3 /opt/openclaw-hf/openclaw_hf/backup.py backup >> "$BACKUP_LOG_FILE" 2>&1 || exit_code=$?
        if [[ $exit_code -eq 0 ]]; then
            log "INFO: Backup completed successfully on attempt $retry_count"
            backup_success=true
            break
        else
            log "ERROR: Backup failed on attempt $retry_count (exit code: $exit_code)"

            if [[ $retry_count -lt $max_retries ]]; then
                local retry_delay=$((retry_count * 10))
                log "INFO: Waiting ${retry_delay}s before retry..."
                sleep $retry_delay

                # Try repair before next attempt
                log "INFO: Attempting repair before retry..."
                /usr/local/bin/openclaw-backup-health.sh --repair >> "$BACKUP_LOG_FILE" 2>&1 || true
            fi
        fi
    done
    
    if [[ "$backup_success" == "true" ]]; then
        return 0
    else
        log "ERROR: Backup failed after $max_retries attempts"
        return 1
    fi
}

# Post-backup health check
post_backup_check() {
    if [[ "$HEALTH_CHECK_AFTER" != "true" ]]; then
        return 0
    fi
    
    log "INFO: Running post-backup health check..."
    
    # Check if backup file was created
    if [[ -f "$BACKUP_LOG_FILE" ]]; then
        local last_backup
        last_backup=$(grep -i "backup complete" "$BACKUP_LOG_FILE" | tail -1)
        
        if [[ -n "$last_backup" ]]; then
            log "INFO: Post-backup check: Backup log shows completion"
            return 0
        else
            log "WARN: Post-backup check: No completion found in logs"
            return 1
        fi
    fi
    
    return 0
}

# Main backup function
main() {
    # Ensure log directory exists
    mkdir -p "$BACKUP_LOG_DIR"

    log "========================================"
    log "OpenClaw Backup Starting"
    log "========================================"

    # Load environment
    load_env

    # Check if backup is enabled
    if ! is_true "${OPENCLAW_BACKUP_ENABLED:-false}"; then
        log "INFO: Backup is disabled (OPENCLAW_BACKUP_ENABLED=false), skipping"
        exit 0
    fi

    # Pre-backup health check
    if [[ "$HEALTH_CHECK_BEFORE" == "true" ]]; then
        if ! run_health_check; then
            log "ERROR: Pre-backup health check failed, aborting backup"
            exit 1
        fi
    fi
    
    # Run backup
    if run_backup_with_retry; then
        log "INFO: Backup process completed successfully"
        
        # Post-backup check
        post_backup_check
        
        log "========================================"
        log "OpenClaw Backup Finished Successfully"
        log "========================================"
        exit 0
    else
        log "ERROR: Backup process failed"
        
        # Run emergency repair on complete failure
        log "INFO: Running emergency repair..."
        /usr/local/bin/openclaw-backup-health.sh --emergency >> "$BACKUP_LOG_FILE" 2>&1 || true
        
        log "========================================"
        log "OpenClaw Backup Failed"
        log "========================================"
        exit 1
    fi
}

# Run main
main "$@"