数据流 - 函数未被调用 - 错误 - 名称未定义

Question

我正在 Google Dataflow 上使用 Apache Beam，并通过 lambda 函数调用函数情感，但收到错误消息：函数名称未定义。

output_tweets = (lines
                     | 'decode' >> beam.Map(lambda x: x.decode('utf-8'))
                     | 'assign window key' >> beam.WindowInto(window.FixedWindows(10))
                     | 'batch into n batches' >> BatchElements(min_batch_size=49, max_batch_size=50)
                     | 'sentiment analysis' >> beam.FlatMap(lambda x: sentiment(x))
                     )

这是我的 Apache Beam 调用，在最后一行中提到了函数情绪，这给我带来了问题。

函数代码如下（我认为这不重要）：

def sentiment(messages):
    if not isinstance(messages, list):
        messages = [messages]

    instances = list(map(lambda message: json.loads(message), messages))
    lservice = discovery.build('language', 'v1beta1', developerKey = APIKEY)
    for instance in instances['text']:
        response = lservice.documents().analyzeSentiment(
            body ={
                'document': {
                    'type': 'PLAIN_TEXT',
                    'content': instance
                }
            }
        ).execute()
        instance['polarity'] = response['documentSentiment']['polarity']
        instance['magnitude'] = response['documentSentiment']['magnitude']

    return instances

我得到以下回溯

  File "stream.py", line 97, in <lambda>
NameError: name 'sentiment' is not defined [while running 'generatedPtransform-441']

任何想法？

Answer 1

发生此问题的原因有多种

1. 函数sentiment定义是否存在于与 Beam 管道相同的 Python 文件中。
2. 函数的定义sentiment是在光束管道中调用之前吗？

我做了一个快速测试，如下所示，如果遵循上述两项，则可以按预期工作

def testing(messages):
    return messages.lower()

windowed_lower_word_counts = (windowed_words
                              | beam.Map(lambda word: testing(word))
                              | "count" >> beam.combiners.Count.PerElement())

ib.show(windowed_lower_word_counts, include_window_info=True)

0   b'have'     3   2020-04-19 06:04:39.999999+0000 2020-04-19 06:04:30.000000+0000 (10s)   Pane 0
1   b'ransom'   1   2020-04-19 06:04:39.999999+0000 2020-04-19 06:04:30.000000+0000 (10s)   Pane 0
2   b'let'      1   2020-04-19 06:04:39.999999+0000 2020-04-19 06:04:30.000000+0000 (10s)   Pane 0
3   b'me'       1   2020-04-19 06:04:39.999999+0000 2020-04-19 06:04:30.000000+0000 (10s)   Pane 0

如果函数是在调用之后定义的，那么我们会得到如下错误

windowed_lower_word_counts = (windowed_words
                              | beam.Map(lambda word: testing_after(word))
                              | "count" >> beam.combiners.Count.PerElement())

ib.show(windowed_lower_word_counts, include_window_info=True)

ERROR:apache_beam.runners.direct.executor:Exception at bundle <apache_beam.runners.direct.bundle_factory._Bundle object at 0x7f478f344820>, due to an exception.
 Traceback (most recent call last):
  File "apache_beam/runners/common.py", line 954, in apache_beam.runners.common.DoFnRunner.process
  File "apache_beam/runners/common.py", line 552, in apache_beam.runners.common.SimpleInvoker.invoke_process
  File "/root/apache-beam-custom/packages/beam/sdks/python/apache_beam/transforms/core.py", line 1482, in <lambda>
    wrapper = lambda x: [fn(x)]
  File "<ipython-input-19-f34e29a17836>", line 2, in <lambda>
    | beam.Map(lambda word: testing_after_new(word))
NameError: name 'testing_after' is not defined

def testing_after(messages):
    return messages.lower()

更新

而不是将函数作为beam.FlatMap(lambda x : fn(x))传递函数作为beam.FlatMap(x)

我相信在第一种情况下，beam 尝试在工作机器中查找 fn，但无法找到它。实现细节可以在这里找到 - https://github.com/apache/beam/blob/fa4f4183a315f061e035d38ba2c5d4b837b371e0/sdks/python/apache_beam/transforms/core.py#L780