🎯 行动 1：给冷数据加 LIFECYCLE 高 ROI ⚙ 一键生成 SQL

💀 立即可关停：30 天调度 0 次成功的节点 最高 ROI · 零风险 共 {{ dwInsight.deadNodes.length }} 个

💰 行动 3：真·最贵 SQL (按扫描量计费) 高 ROI 🔄 重扫昨日 数据源 INFORMATION_SCHEMA.TASKS_HISTORY · 上次扫描 {{ dwRealSql.lastScanAt || '从未' }}

🔁 重复执行 SQL Top 10 ({{ dwRealSql.bizdate }} · 按 signature 聚合) 物化中间表可降本

💰 Owner 周费用排行 (真实 ¥, 近 7 天) 高 ROI

👥 Owner 近 7 天耗时排行 D 找出"产 SQL 最贵"的人，约谈+review

工作空间

节点列表 — {{ dataworks.selectedProject.projectIdentifier }} (projectId={{ dataworks.selectedProject.projectId }})

扫描配置

表清单 ⚙ 生成 LIFECYCLE SQL

扫描配置

汇总

业务流程列表（按僵尸评分降序）

💰 MaxCompute SQL 真实费用排行 数据源 information_schema.tasks_history · 按 input_bytes × ¥0.3/GB × complexity 估算

近 7 天费用趋势

Top {{ dwRealSql.topLimit }} 最贵 SQL · {{ dwRealSql.bizdate }}

💰 Owner 周费用排行 (近 7 天)

🔁 重复执行 SQL Top 20 · {{ dwRealSql.bizdate }} (按 signature 聚合, 同 SQL 多次跑)

🔥 输入表扫描热度 Top 20 (近 7 天，按 ¥ 排)

⚡ 实时计算 Flink 概览 刷新 workspace 从账单反查 (产品 sc) · 月费包月制 · 改规格才省钱

Workspace {{ ws.workspaceId }} ({{ ws.nick }}) ¥{{ ws.monthly_cost.toLocaleString() }} / 月 region={{ ws.region }}

SQL 计算热榜（耗时 + 错误，非真实费用）

-- ==========================================
-- 真·MaxCompute SQL 计算费用 Top 100（昨日）
-- 在 DataWorks 临时 SQL 节点执行（需 db_warehouse 项目读权限）
-- 公式: cost_cpu = cpu_cost (核·秒) × 单价
-- 公式: cost_scan = input_bytes / 1024^3 (GB) × ¥0.3 × complexity
-- ==========================================

SELECT
  task_catalog            AS project,
  task_name               AS task,
  inst_id                 AS instance_id,
  owner_name              AS owner,
  status                  AS status,
  ROUND(input_bytes/1024/1024/1024, 3) AS input_gb,
  input_records,
  ROUND(cpu_cost, 0)      AS cpu_seconds,
  complexity,
  ROUND(input_bytes/1024/1024/1024 * 0.3 * COALESCE(complexity,1), 4) AS est_cost_cny,
  start_time, end_time,
  (UNIX_TIMESTAMP(end_time) - UNIX_TIMESTAMP(start_time)) AS duration_sec
FROM INFORMATION_SCHEMA.TASKS_HISTORY
WHERE ds = '${bizdate}'                    -- 替换为昨日 yyyymmdd
  AND task_type = 'SQL'
  AND status   IN ('Terminated', 'Success')
ORDER BY est_cost_cny DESC
LIMIT 100;

-- ==========================================
-- 按 owner 聚合的周成本榜 Top 20
-- ==========================================
SELECT
  owner_name,
  COUNT(*)                                                              AS jobs,
  ROUND(SUM(input_bytes)/1024/1024/1024/1024, 2)                        AS total_input_tb,
  ROUND(SUM(input_bytes/1024/1024/1024 * 0.3 * COALESCE(complexity,1)), 2) AS est_cost_cny
FROM INFORMATION_SCHEMA.TASKS_HISTORY
WHERE ds BETWEEN '${start}' AND '${end}'   -- 替换为最近 7 天
  AND task_type = 'SQL'
  AND status   IN ('Terminated', 'Success')
GROUP BY owner_name
ORDER BY est_cost_cny DESC
LIMIT 20;

-- ==========================================
-- 按表（输入表）聚合：哪张大表被反复扫
-- ==========================================
SELECT
  inputs[0]               AS top_input_table,
  COUNT(*)                AS scan_count,
  ROUND(SUM(input_bytes)/1024/1024/1024, 2) AS total_scan_gb,
  ROUND(AVG(input_bytes)/1024/1024/1024, 2) AS avg_scan_gb
FROM INFORMATION_SCHEMA.TASKS_HISTORY
WHERE ds BETWEEN '${start}' AND '${end}'
  AND task_type = 'SQL'
  AND CARDINALITY(inputs) > 0
GROUP BY inputs[0]
ORDER BY total_scan_gb DESC
LIMIT 30;