hive超大数据量排序

当使用hive或spark对超大数据量（几十亿）数据进行排序的时候，直接使用row_number函数会导致数据严重倾斜，都在一个reduce任务上执行，导致很难跑出来，一定要排序的还可以参考如下sql

SELECT
	a.uuid,
	a.rank + NVL(b.max_num, 0) AS seq
FROM
	(
		SELECT
			uuid,
			ABS(hash(uuid)) % 10000 AS part,
			row_number() over(PARTITION BY ABS(hash(uuid)) % 10000 order by uuid) AS rank
		FROM
			{tab_seq}
		WHERE
			dt = '2022-10-06'
	)
	a
LEFT JOIN
	(
		SELECT
			part,
			SUM(max_num) over(order by part) AS max_num
		FROM
			(
				SELECT
					ABS(hash(uuid)) % 10000 AS part,
					COUNT(1) AS max_num
				FROM
					{tab_seq}
				WHERE
					dt = '2022-10-06'
				GROUP BY
					ABS(hash(uuid)) % 10000
			)
			tmp2
	)
	b
ON
	(a.part-1) = b.part