添加with_output_types()方法以显式指定BatchElements()输出类型，如下示例所示。

import apache_beam as beam

class MyDoFn(beam.DoFn):
    def process(self, element):
        return [(i,1) for i in element]

with beam.Pipeline() as p:
    data = p | "Create data" >> beam.Create([(1,2),(3,4),(5,6),(7,8)])
    
    batched_data = (
        data 
        | "Batch data" >> beam.BatchElements(
            min_batch_size=2, 
            max_batch_size=3, 
            max_latency=1))
        
    reshuffled_data = (
        batched_data 
        | "Reshuffle data" >> beam.Reshuffle())
        
    processed_data = (
        reshuffled_data 
        | "Process data" >> beam.ParDo(MyDoFn())
        | "Grouping data" >> beam.GroupByKey())
    
    processed_data | "Print result" >> beam.Map(print)

注：beam.BatchElements()输出类型默认情况下是Tuple[Any,Union[Timestamp,int]]，这可能是导致DataflowRunner无法正常工作的原因之一。