Streaming from PubSub to Firestore doesn't work on Dataflow

I came to the same error as here [https://stackoverflow.com/questions/57059944/python-package-errors-while-running-gcp-dataflow](https://stackoverflow.com/questions/57059944/python-package-errors-while-running-gcp-dataflow) but I don't see anywhere reported thus I am creating an issue just in case.

The pipeline is quite simple, reading from PubSub and writing to Firestore.

Beam version used is 2.13.0, Python 2.7

With DirectRunner works ok, but on Dataflow it throws the following message:

 
```

java.util.concurrent.ExecutionException: java.lang.RuntimeException: Error received from SDK harness
for instruction -81: Traceback (most recent call last):
File "/usr/local/lib/python2.7/dist-packages/apache_beam/runners/worker/sdk_worker.py",
line 157, in _execute
response = task()
File "/usr/local/lib/python2.7/dist-packages/apache_beam/runners/worker/sdk_worker.py",
line 190, in
self._execute(lambda: worker.do_instruction(work), work)
File "/usr/local/lib/python2.7/dist-packages/apache_beam/runners/worker/sdk_worker.py",
line 312, in do_instruction
request.instruction_id)
File "/usr/local/lib/python2.7/dist-packages/apache_beam/runners/worker/sdk_worker.py",
line 331, in process_bundle
bundle_processor.process_bundle(instruction_id))
File "/usr/local/lib/python2.7/dist-packages/apache_beam/runners/worker/bundle_processor.py",
line 538, in process_bundle
op.start()
File "apache_beam/runners/worker/operations.py", line 554,
in apache_beam.runners.worker.operations.DoOperation.start
def start(self):
File "apache_beam/runners/worker/operations.py",
line 555, in apache_beam.runners.worker.operations.DoOperation.start
with self.scoped_start_state:

File "apache_beam/runners/worker/operations.py", line 557, in apache_beam.runners.worker.operations.DoOperation.start

self.dofn_runner.start()
File "apache_beam/runners/common.py", line 778, in apache_beam.runners.common.DoFnRunner.start

self._invoke_bundle_method(self.do_fn_invoker.invoke_start_bundle)
File "apache_beam/runners/common.py",
line 775, in apache_beam.runners.common.DoFnRunner._invoke_bundle_method
self._reraise_augmented(exn)

File "apache_beam/runners/common.py", line 800, in apache_beam.runners.common.DoFnRunner._reraise_augmented

raise_with_traceback(new_exn)
File "apache_beam/runners/common.py", line 773, in apache_beam.runners.common.DoFnRunner._invoke_bundle_method

bundle_method()
File "apache_beam/runners/common.py", line 359, in apache_beam.runners.common.DoFnInvoker.invoke_start_bundle

def invoke_start_bundle(self):
File "apache_beam/runners/common.py", line 363, in apache_beam.runners.common.DoFnInvoker.invoke_start_bundle

self.signature.start_bundle_method.method_value())
File "/home/zdenulo/dev/gcp_stuff/df_firestore_stream/df_firestore_stream.py",
line 39, in start_bundle
NameError: global name 'firestore' is not defined [while running 'generatedPtransform-64']
 

```

It's interesting that using Beam version 2.12.0 solves the problem on Dataflow, it works as expected, not sure what could be the problem.

Here is a repository with the code which was used [https://github.com/zdenulo/dataflow_firestore_stream](https://github.com/zdenulo/dataflow_firestore_stream)

Imported from Jira [BEAM-7871](https://issues.apache.org/jira/browse/BEAM-7871). Original Jira may contain additional context.
Reported by: zdenulo.