你试过这样的CombinePerKey吗？

import apache_beam as beam

p = beam.Pipeline()

test = (
    p
    | beam.Create([(0, "ttt"), (0, "ttt1"), (0, "ttt2"), (1, "xxx"), (1, "xxx2"), (2, "yyy")])
    | beam.CombinePerKey(lambda v: ",".join(v))
    | beam.Map(print)
)

像链接示例中那样，唯一ID是从0到n_groups的整数对您来说重要吗？
如果没有，那么我认为这里没有任何使用分组操作的必要。

import apache_beam as beam

def make_unique_id(row):
  """
  Example function for extracting a unique ID from the row.

  You could wrap the value in uuid.UUID to make a more standard format for the ID.
  """
  return ",".join([row[0], row[1]])

pipe = beam.Pipeline()

id = (pipe
            | beam.io.ReadFromText('data.csv')
            | beam.Map(lambda x: x.split(","))
            | beam.Map(lambda x: [make_unique_id(x)] + x)
            | beam.Map(print))