Blueking Lite AI 工作流记忆节点：让运维编排真正"记住"上下文

workflow:
  name: cpu_alert_diagnosis_and_fix
  description: CPU 告警根因分析与自动修复，通过记忆节点传递中间结论
  nodes:
    # 第一步：接收告警，提取关键信息写入记忆
    - id: alert_receiver
      type: llm
      prompt: |
        分析以下 CPU 告警信息，提取受影响主机 IP、告警阈值、当前 CPU 使用率。
        输出格式严格为 JSON：
        {"affected_ip": "...", "threshold": ..., "current_cpu": ...}
      input:
        alert_text: "{{ trigger.alert_message }}"
      next: memory_write_alert

    # 记忆节点：写入告警摘要，供后续节点读取
    - id: memory_write_alert
      type: memory_write
      entries:
        - key: affected_ip
          value: "{{ alert_receiver.output.affected_ip }}"
        - key: threshold
          value: "{{ alert_receiver.output.threshold }}"
        - key: current_cpu
          value: "{{ alert_receiver.output.current_cpu }}"
        - key: alert_time
          value: "{{ trigger.timestamp }}"
      next: root_cause_analysis

    # 第二步：根因分析，从记忆读取主机 IP，写入根因结论
    - id: root_cause_analysis
      type: llm
      prompt: |
        主机 {{ memory_read.affected_ip }} 的 CPU 使用率达到 {{ memory_read.current_cpu }}%。
        请判断根因，只返回以下之一：
        - "process_stuck"：某个进程卡死
        - "traffic_spike"：流量突增
        - "cron_storm"：定时任务集中执行
        输出格式：{"root_cause": "...", "suspect_process": "..."}
      input:
        memory_read.affected_ip: "{{ memory_write_alert.entries.affected_ip }}"
        memory_read.current_cpu: "{{ memory_write_alert.entries.current_cpu }}"
      next: memory_write_rootcause

    # 记忆节点：写入根因，决定后续分支
    - id: memory_write_rootcause
      type: memory_write
      entries:
        - key: root_cause
          value: "{{ root_cause_analysis.output.root_cause }}"
        - key: suspect_process
          value: "{{ root_cause_analysis.output.suspect_process }}"
      next: branch_by_rootcause

    # 第三步：条件分支——根据记忆中的根因走不同修复路径
    - id: branch_by_rootcause
      type: switch
      conditions:
        - when: "{{ memory_read.root_cause }} == 'process_stuck'"
          next: kill_stuck_process
        - when: "{{ memory_read.root_cause }} == 'traffic_spike'"
          next: scale_out_service
        - when: "{{ memory_read.root_cause }} == 'cron_storm'"
          next: reschedule_crons
      input:
        memory_read.root_cause: "{{ memory_write_rootcause.entries.root_cause }}"

    # 第四步（分支示例）：杀卡死进程
    - id: kill_stuck_process
      type: script
      script_type: bash
      script_content: |
        # 从记忆读取目标主机和可疑进程名
        TARGET_HOST="{{ memory_read.affected_ip }}"
        SUSPECT_PROC="{{ memory_read.suspect_process }}"

        echo "正在处理主机 $TARGET_HOST 上的卡死进程 $SUSPECT_PROC"

        # 通过 SSH 执行远程命令（实际环境中使用 Blueking 的作业平台）
        ssh ops-agent@$TARGET_HOST "ps aux | grep '$SUSPECT_PROC' | grep -v grep | awk '{print \$2}' | head -1 | xargs -r kill -9"

        # 写入执行结果到记忆，供汇报节点使用
        echo "进程 $SUSPECT_PROC 已终止"
      input:
        memory_read.affected_ip: "{{ memory_write_alert.entries.affected_ip }}"
        memory_read.suspect_process: "{{ memory_write_rootcause.entries.suspect_process }}"
      next: memory_write_fix_result

    # 记忆节点：记录修复动作，形成完整闭环
    - id: memory_write_fix_result
      type: memory_write
      entries:
        - key: fix_action
          value: "killed_stuck_process"
        - key: fix_target
          value: "{{ memory_read.affected_ip }}"
        - key: fix_status
          value: "success"
      next: final_report

    # 最终汇报：从记忆中拼装完整上下文
    - id: final_report
      type: llm
      prompt: |
        请生成运维处理报告，以下信息来自工作流记忆：
        - 告警时间：{{ memory_read.alert_time }}
        - 受影响主机：{{ memory_read.affected_ip }}
        - CPU 使用率：{{ memory_read.current_cpu }}%（阈值 {{ memory_read.threshold }}%）
        - 根因：{{ memory_read.root_cause }}
        - 修复动作：{{ memory_read.fix_action }}
        - 修复状态：{{ memory_read.fix_status }}
      input:
        memory_read.alert_time: "{{ memory_write_alert.entries.alert_time }}"
        memory_read.affected_ip: "{{ memory_write_alert.entries.affected_ip }}"
        memory_read.current_cpu: "{{ memory_write_alert.entries.current_cpu }}"
        memory_read.threshold: "{{ memory_write_alert.entries.threshold }}"
        memory_read.root_cause: "{{ memory_write_rootcause.entries.root_cause }}"
        memory_read.fix_action: "{{ memory_write_fix_result.entries.fix_action }}"
        memory_read.fix_status: "{{ memory_write_fix_result.entries.fix_status }}"