如何在单元测试时正确测试 pcollection 长度 Apache Beam

Question

I am wondering which is the best way to test check the lenght of the output produced by a beam pipeline.

I have some testcode like this:

test_data = [
    {'kind': 'storage#object', 'name': 'file1.doc', 'contentType': 'application/octet-stream', 'bucket': 'bucket123' },
    {'kind': 'storage#object', 'name': 'file2.pdf', 'contentType': 'application/pdf','bucket': 'bucket234'},
    {'kind': 'storage#object', 'name': 'file3.msg', 'contentType': 'message/rfc822', 'bucket': 'bucket345'}
]

with TestPipeline() as p:
   output = (p 
             | beam.Create(test_data)
             | beam.ParDo(DoFn_To_Test()).with_outputs('ok','error')
   )

I want to test sure that all elements in the test_data list go to 'output.ok'. I think way to do it is to count them like this:

with TestPipeline() as p:
   output = (p 
             | beam.Create(testdata) 
             | beam.ParDo(DoFn_To_Test()).with_outputs('ok','error')
   )

   okay_count = (output.ok | beam.Map(lambda x: ('dummy_key',x)) 
                 | beam.GroupByKey()  # This gets ('dumm_key',[element1,element2....])
                 | beam.Map(lambda x: len(x[1]) )  # Drop the key and get the lengh of the list
   )

   # And finally check^H^H^H^H^H^H assert the count is correct:
   assert_that(okay_count, equal_to([len(test_data)])

This works; but I don't feel this is the best way to do it, and I am sure there are more ways to do it.

Best option (by now)

This is the best option suggested up to date: using beam.combiners.Count.Globally()

with TestPipeline() as p:
   output = (p 
             | beam.Create(testdata) 
             | beam.ParDo(DoFn_To_Test()).with_outputs('ok','error')
   )

   okay_count = output | beam.combiners.Count.Globally()
   assert_that(okay_count, equal_to([len(test_data)])

Answer 1

You answered your own question in the the question. Writing it here as an answer:

with TestPipeline() as p:
   output = (p 
             | beam.Create(testdata) 
             | beam.ParDo(DoFn_To_Test()).with_outputs('ok','error')
   )

   okay_count = output | beam.combiners.Count.Globally()
   assert_that(okay_count, equal_to([len(test_data)])