您需要做的是为1和0的每个部分创建“组”。您可以通过创建一个布尔标志，然后对该列进行累加求和来获得组。从那里你可以累积你的原始数据 tag 按在子查询中创建的每个组列出。
查询：

SELECT customer
  , txn_date
  , tag
  , SUM(tag) OVER (PARTITION BY customer, flg_sum ORDER BY txn_date ROWS BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW) AS running_sum
FROM (
  SELECT *
    , SUM(tag_flg) OVER (PARTITION BY customer ORDER BY txn_date ROWS BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW) AS flg_sum
  FROM (
    SELECT *
      , CASE WHEN  tag = 1 THEN 0 ELSE 1 END AS tag_flg
    FROM database.table ) x ) y

输出：

customer        txn_date        tag     running_sum
A               2017-01-01      1       1
A               2017-01-02      1       2
A               2017-01-03      1       3
A               2017-01-04      1       4
A               2017-01-05      1       5
A               2017-01-06      1       6
A               2017-01-07      0       0
A               2017-01-08      1       1
A               2017-01-09      1       2
A               2017-01-10      1       3
A               2017-01-11      0       0
A               2017-01-12      0       0
A               2017-01-13      1       1
A               2017-01-14      1       2
A               2017-01-15      0       0