fix: Handle TransportError Exceptions thrown from gapic_publish by mukund-ananthu · Pull Request #1318 · googleapis/python-pubsub

mukund-ananthu · 2025-01-06T01:18:56Z

Current behavior:

If TransportError is thrown by google.auth library:

https://github.com/googleapis/google-auth-library-python/blob/c3ea09fd8b9ee8f094263fdc809c59d9dfaa3124/google/auth/compute_engine/_metadata.py#L187-L189

when doing a _gapic_publish:

python-pubsub/google/cloud/pubsub_v1/publisher/_batch/thread.py

Lines 322 to 327 in f79d35a

    
           response = self._client._gapic_publish( 
        
               topic=self._topic, 
        
               messages=[wrapper.message for wrapper in self._message_wrappers], 
        
               retry=self._commit_retry, 
        
               timeout=self._commit_timeout, 
        
           )

Then,

The exception is not caught:

python-pubsub/google/cloud/pubsub_v1/publisher/_batch/thread.py

Line 345 in f79d35a

except google.api_core.exceptions.GoogleAPIError as exc:

Status of batch not set to ERROR:

python-pubsub/google/cloud/pubsub_v1/publisher/_batch/thread.py

Line 348 in f79d35a

self._status = base.BatchStatus.ERROR

Exceptions not set in the publish futures

python-pubsub/google/cloud/pubsub_v1/publisher/_batch/thread.py

Lines 364 to 369 in f79d35a

    
           if self._batch_done_callback is not None: 
        
               # Failed to publish batch. 
        
               self._batch_done_callback(batch_transport_succeeded) 
        
           for future in self._futures: 
        
               future.set_exception(exc)

The thread in which the batch gapic_publish was occuring dies without setting the correct batch state / future exceptions:

python-pubsub/google/cloud/pubsub_v1/publisher/_batch/thread.py

Lines 237 to 239 in f79d35a

    
           commit_thread = threading.Thread( 
        
               name="Thread-CommitBatchPublisher", target=self._commit, daemon=True 
        
           )

State of Batch remains IN_PROGRESS indefinitely

python-pubsub/google/cloud/pubsub_v1/publisher/_batch/thread.py

Line 288 in f79d35a

self._status = base.BatchStatus.IN_PROGRESS

Subsequent publishes of messages result in creation of a new batch:

python-pubsub/google/cloud/pubsub_v1/publisher/_batch/thread.py

Lines 449 to 450 in f79d35a

    
           if self.status != base.BatchStatus.ACCEPTING_MESSAGES: 
        
               return None

python-pubsub/google/cloud/pubsub_v1/publisher/_sequencer/ordered_sequencer.py

Lines 323 to 326 in f79d35a

    
           while future is None: 
        
               batch = self._create_batch(commit_retry=retry, commit_timeout=timeout) 
        
               self._ordered_batches.append(batch) 
        
               future = batch.publish(wrapper)

Messages that were in the batch whose commit thread died:

Will not have the TransportError set as an exception in their publish futures
The publish future will either hang or timeout

Behavior expected with change

TransportError during gapic_publish will be caught in addition to the existing GoogleAPIError

python-pubsub/google/cloud/pubsub_v1/publisher/_batch/thread.py

Line 345 in f79d35a

except google.api_core.exceptions.GoogleAPIError as exc:

Existing behavior when GoogleAPIError is caught during gapic_publish would also now apply to TransportError, i.e:

Batch status will be set to ERROR, future exceptions will be set:

python-pubsub/google/cloud/pubsub_v1/publisher/_batch/thread.py

Lines 345 to 371 in f79d35a

    
           except google.api_core.exceptions.GoogleAPIError as exc: 
        
               # We failed to publish, even after retries, so set the exception on 
        
               # all futures and exit. 
        
               self._status = base.BatchStatus.ERROR 
        
               if self._client.open_telemetry_enabled: 
        
                   if self._rpc_span: 
        
                       self._rpc_span.record_exception( 
        
                           exception=exc, 
        
                       ) 
        
                       self._rpc_span.set_status( 
        
                           trace.Status(status_code=trace.StatusCode.ERROR) 
        
                       ) 
        
                       self._rpc_span.end() 
        
                   for wrapper in self._message_wrappers: 
        
                       wrapper.end_create_span(exc=exc) 
        
               batch_transport_succeeded = False 
        
               if self._batch_done_callback is not None: 
        
                   # Failed to publish batch. 
        
                   self._batch_done_callback(batch_transport_succeeded) 
        
               for future in self._futures: 
        
                   future.set_exception(exc) 
        
               return

Subsequent publish of messages on the batch object will cause an AssertionError:

python-pubsub/google/cloud/pubsub_v1/publisher/_batch/thread.py

Lines 444 to 447 in f79d35a

    
           with self._state_lock: 
        
               assert ( 
        
                   self._status != base.BatchStatus.ERROR 
        
               ), "Publish after stop() or publish error."

which will be bubbled up through the Ordered / Unordered Sequenceer's publish method:

python-pubsub/google/cloud/pubsub_v1/publisher/_sequencer/ordered_sequencer.py

Lines 321 to 328 in f79d35a

    
           batch = self._ordered_batches[-1] 
        
           future = batch.publish(wrapper) 
        
           while future is None: 
        
               batch = self._create_batch(commit_retry=retry, commit_timeout=timeout) 
        
               self._ordered_batches.append(batch) 
        
               future = batch.publish(wrapper) 
        
           return future

to the Publisher client's publish method:

python-pubsub/google/cloud/pubsub_v1/publisher/client.py

Lines 477 to 495 in f79d35a

    
               future = sequencer.publish( 
        
                   wrapper=wrapper, retry=retry, timeout=timeout 
        
               ) 
        
               future.add_done_callback(on_publish_done) 
        
           except BaseException as be: 
        
               # Exceptions can be thrown when attempting to add messages to 
        
               # the batch. If they're thrown, record them in publisher 
        
               # batching and create span, end the spans and bubble the 
        
               # exception up. 
        
               if self._open_telemetry_enabled: 
        
                   if wrapper: 
        
                       wrapper.end_publisher_batching_span(be) 
        
                       wrapper.end_create_span(be) 
        
                   else:  # pragma: NO COVER 
        
                       warnings.warn( 
        
                           message="PublishMessageWrapper is None. Hence, not recording exception and ending publisher batching span and create span", 
        
                           category=RuntimeWarning, 
        
                       ) 
        
               raise be

which would bubble the exception back to the user

Additional details can be found here: #1173 (comment)

Fixes #1173 🦕

fix: Handle TransportError Exceptions thrown from gapic_publish

f83caa9

mukund-ananthu requested review from a team as code owners January 6, 2025 01:18

product-auto-label bot added size: s Pull request size is small. api: pubsub Issues related to the googleapis/python-pubsub API. labels Jan 6, 2025

mukund-ananthu requested review from hongalex and kamalaboulhosn January 6, 2025 03:09

hongalex approved these changes Jan 6, 2025

View reviewed changes

mukund-ananthu merged commit 0e058c7 into main Jan 6, 2025
27 checks passed

mukund-ananthu deleted the gceMeta branch January 6, 2025 19:35

release-please bot mentioned this pull request Jan 6, 2025

chore(main): release 2.27.2 #1319

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

fix: Handle TransportError Exceptions thrown from gapic_publish#1318

fix: Handle TransportError Exceptions thrown from gapic_publish#1318
mukund-ananthu merged 1 commit intomainfrom
gceMeta

mukund-ananthu commented Jan 6, 2025 •

edited

Loading

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

	response = self._client._gapic_publish(
	topic=self._topic,
	messages=[wrapper.message for wrapper in self._message_wrappers],
	retry=self._commit_retry,
	timeout=self._commit_timeout,
	)

	if self._batch_done_callback is not None:
	# Failed to publish batch.
	self._batch_done_callback(batch_transport_succeeded)

	for future in self._futures:
	future.set_exception(exc)

	commit_thread = threading.Thread(
	name="Thread-CommitBatchPublisher", target=self._commit, daemon=True
	)

	if self.status != base.BatchStatus.ACCEPTING_MESSAGES:
	return None

	while future is None:
	batch = self._create_batch(commit_retry=retry, commit_timeout=timeout)
	self._ordered_batches.append(batch)
	future = batch.publish(wrapper)

	except google.api_core.exceptions.GoogleAPIError as exc:
	# We failed to publish, even after retries, so set the exception on
	# all futures and exit.
	self._status = base.BatchStatus.ERROR

	if self._client.open_telemetry_enabled:
	if self._rpc_span:
	self._rpc_span.record_exception(
	exception=exc,
	)
	self._rpc_span.set_status(
	trace.Status(status_code=trace.StatusCode.ERROR)
	)
	self._rpc_span.end()

	for wrapper in self._message_wrappers:
	wrapper.end_create_span(exc=exc)

	batch_transport_succeeded = False
	if self._batch_done_callback is not None:
	# Failed to publish batch.
	self._batch_done_callback(batch_transport_succeeded)

	for future in self._futures:
	future.set_exception(exc)

	return

	with self._state_lock:
	assert (
	self._status != base.BatchStatus.ERROR
	), "Publish after stop() or publish error."

	batch = self._ordered_batches[-1]
	future = batch.publish(wrapper)
	while future is None:
	batch = self._create_batch(commit_retry=retry, commit_timeout=timeout)
	self._ordered_batches.append(batch)
	future = batch.publish(wrapper)

	return future

	future = sequencer.publish(
	wrapper=wrapper, retry=retry, timeout=timeout
	)
	future.add_done_callback(on_publish_done)
	except BaseException as be:
	# Exceptions can be thrown when attempting to add messages to
	# the batch. If they're thrown, record them in publisher
	# batching and create span, end the spans and bubble the
	# exception up.
	if self._open_telemetry_enabled:
	if wrapper:
	wrapper.end_publisher_batching_span(be)
	wrapper.end_create_span(be)
	else: # pragma: NO COVER
	warnings.warn(
	message="PublishMessageWrapper is None. Hence, not recording exception and ending publisher batching span and create span",
	category=RuntimeWarning,
	)
	raise be

Conversation

mukund-ananthu commented Jan 6, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Current behavior:

Behavior expected with change

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

mukund-ananthu commented Jan 6, 2025 •

edited

Loading