sql cumulative sum

select ts,
  seats,
  sum(seats) over(order by rownum) as usage 
from (
  
select 
  row_number() over(order by '1') as rownum,
  to_timestamp(columns[0], 'yyyy-MM-dd HH:mm:ss Z z') as ts, 
  case 
    when columns[1] = 'out' then cast(columns[3] as int)
    when columns[1] = 'in' then -1 * cast(columns[3] as int)
    else 0
  end as seats
from dfs.tmp.`test1.csv`
where columns[1] = 'out' or columns[1] = 'in'

)
;