diff --git a/CHANGES.md b/CHANGES.md
index dac9654f2331..4296436544a8 100644
--- a/CHANGES.md
+++ b/CHANGES.md
@@ -77,7 +77,7 @@
 
 ## Bugfixes
 
-* Fixed X (Java/Python) ([#X](https://github.com/apache/beam/issues/X)).
+* Fixed DirectRunner bug in Python SDK where GroupByKey gets empty PCollection and fails when pipeline option `direct_num_workers!=1`. ([#27373](https://github.com/apache/beam/pull/27373))
 
 ## Known Issues
 
diff --git a/sdks/python/apache_beam/runners/portability/fn_api_runner/fn_runner.py b/sdks/python/apache_beam/runners/portability/fn_api_runner/fn_runner.py
index 8d957068d08b..be7f99dc61f4 100644
--- a/sdks/python/apache_beam/runners/portability/fn_api_runner/fn_runner.py
+++ b/sdks/python/apache_beam/runners/portability/fn_api_runner/fn_runner.py
@@ -825,10 +825,11 @@ def _execute_bundle(self,
 
     buffers_to_clean = set()
     known_consumers = set()
-    for _, buffer_id in bundle_context_manager.stage_data_outputs.items():
-      for (consuming_stage_name, consuming_transform) in \
-          runner_execution_context.buffer_id_to_consumer_pairs.get(buffer_id,
-                                                                   []):
+    for transform_id, buffer_id in (
+      bundle_context_manager.stage_data_outputs.items()):
+      for (consuming_stage_name, consuming_transform
+           ) in runner_execution_context.buffer_id_to_consumer_pairs.get(
+               buffer_id, []):
         buffer = runner_execution_context.pcoll_buffers.get(buffer_id, None)
 
         if (buffer_id in runner_execution_context.pcoll_buffers and
@@ -840,6 +841,11 @@ def _execute_bundle(self,
           # so we create a copy of the buffer for every new stage.
           runner_execution_context.pcoll_buffers[buffer_id] = buffer.copy()
           buffer = runner_execution_context.pcoll_buffers[buffer_id]
+        # When the buffer is not in the pcoll_buffers, it means that the
+        # it could be an empty PCollection. In this case, get the buffer using
+        # the buffer id and transform id
+        if buffer is None:
+          buffer = bundle_context_manager.get_buffer(buffer_id, transform_id)
 
         # If the buffer has already been added to be consumed by
         # (stage, transform), then we don't need to add it again. This case
@@ -854,7 +860,7 @@ def _execute_bundle(self,
         # MAX_TIMESTAMP for the downstream stage.
         runner_execution_context.queues.watermark_pending_inputs.enque(
             ((consuming_stage_name, timestamp.MAX_TIMESTAMP),
-             DataInput({consuming_transform: buffer}, {})))  # type: ignore
+             DataInput({consuming_transform: buffer}, {})))
 
     for bid in buffers_to_clean:
       if bid in runner_execution_context.pcoll_buffers:
diff --git a/sdks/python/apache_beam/runners/portability/fn_api_runner/fn_runner_test.py b/sdks/python/apache_beam/runners/portability/fn_api_runner/fn_runner_test.py
index ed09bb8f2236..b55c7162aea7 100644
--- a/sdks/python/apache_beam/runners/portability/fn_api_runner/fn_runner_test.py
+++ b/sdks/python/apache_beam/runners/portability/fn_api_runner/fn_runner_test.py
@@ -1831,6 +1831,15 @@ def create_pipeline(self, is_drain=False):
     p._options.view_as(DebugOptions).experiments.remove('beam_fn_api')
     return p
 
+  def test_group_by_key_with_empty_pcoll_elements(self):
+    with self.create_pipeline() as p:
+      res = (
+          p
+          | beam.Create([('test_key', 'test_value')])
+          | beam.Filter(lambda x: False)
+          | beam.GroupByKey())
+      assert_that(res, equal_to([]))
+
   def test_metrics(self):
     raise unittest.SkipTest("This test is for a single worker only.")