Skip to content

Commit

Permalink
feat: allow disabling response stream pre-fetch (#30)
Browse files Browse the repository at this point in the history
Closes #25.

This PR adds the ability to disable automatically pre-fetching the first item of a stream returned by `*-Stream` gRPC callables. This hook will be used in PubSub to fix the [stalled stream issue](googleapis/python-pubsub#93), while also not affecting Firestore, since the default behavior is preserved.

I realize the fix is far from ideal, but it's the least ugly among the approaches I tried, e.g. somehow passing the flag through `ResumableBidiRpc` (it's a messy rabbit hole). On the PubSub side monkeypatching the generated SubscriberClient will be needed, but it's a (relatively) clean one-liner:
```patch
diff --git google/cloud/pubsub_v1/gapic/subscriber_client.py google/cloud/pubsub_v1/gapic/subscriber_client.py
index e98a686..1d6c058 100644
--- google/cloud/pubsub_v1/gapic/subscriber_client.py
+++ google/cloud/pubsub_v1/gapic/subscriber_client.py
@@ -1169,6 +1169,8 @@ class SubscriberClient(object):
                 default_timeout=self._method_configs["StreamingPull"].timeout,
                 client_info=self._client_info,
             )
+            # TODO: explain this monkeypatch!
+            self.transport.streaming_pull._prefetch_first_result_ = False
 
         return self._inner_api_calls["streaming_pull"](
             requests, retry=retry, timeout=timeout, metadata=metadata
```

If/when we merge this, we should also release it, and then we can add `!= 1.17.0` to the `google-api-core` version pin in PubSub.

### PR checklist
- [x] Make sure to open an issue as a [bug/issue](https://github.com/googleapis/python-api-core/issues/new/choose) before writing your code!  That way we can discuss the change, evaluate designs, and agree on the general idea
- [x] Ensure the tests and linter pass
- [x] Code coverage does not decrease (if any source code was changed)
- [x] Appropriate docs were updated (if necessary)
  • Loading branch information
plamut committed Jun 9, 2020
1 parent 945bafc commit 74e0b0f
Show file tree
Hide file tree
Showing 2 changed files with 22 additions and 3 deletions.
12 changes: 9 additions & 3 deletions google/api_core/grpc_helpers.py
Expand Up @@ -62,14 +62,15 @@ def error_remapped_callable(*args, **kwargs):


class _StreamingResponseIterator(grpc.Call):
def __init__(self, wrapped):
def __init__(self, wrapped, prefetch_first_result=True):
self._wrapped = wrapped

# This iterator is used in a retry context, and returned outside after init.
# gRPC will not throw an exception until the stream is consumed, so we need
# to retrieve the first result, in order to fail, in order to trigger a retry.
try:
self._stored_first_result = six.next(self._wrapped)
if prefetch_first_result:
self._stored_first_result = six.next(self._wrapped)
except TypeError:
# It is possible the wrapped method isn't an iterable (a grpc.Call
# for instance). If this happens don't store the first result.
Expand Down Expand Up @@ -141,7 +142,12 @@ def _wrap_stream_errors(callable_):
def error_remapped_callable(*args, **kwargs):
try:
result = callable_(*args, **kwargs)
return _StreamingResponseIterator(result)
# Auto-fetching the first result causes PubSub client's streaming pull
# to hang when re-opening the stream, thus we need examine the hacky
# hidden flag to see if pre-fetching is disabled.
# https://github.com/googleapis/python-pubsub/issues/93#issuecomment-630762257
prefetch_first = getattr(callable_, "_prefetch_first_result_", True)
return _StreamingResponseIterator(result, prefetch_first_result=prefetch_first)
except grpc.RpcError as exc:
six.raise_from(exceptions.from_grpc_error(exc), exc)

Expand Down
13 changes: 13 additions & 0 deletions tests/unit/test_grpc_helpers.py
Expand Up @@ -80,6 +80,19 @@ def test_wrap_stream_okay():
assert responses == expected_responses


def test_wrap_stream_prefetch_disabled():
responses = [1, 2, 3]
iter_responses = iter(responses)
callable_ = mock.Mock(spec=["__call__"], return_value=iter_responses)
callable_._prefetch_first_result_ = False

wrapped_callable = grpc_helpers._wrap_stream_errors(callable_)
wrapped_callable(1, 2, three="four")

assert list(iter_responses) == responses # no items should have been pre-fetched
callable_.assert_called_once_with(1, 2, three="four")


def test_wrap_stream_iterable_iterface():
response_iter = mock.create_autospec(grpc.Call, instance=True)
callable_ = mock.Mock(spec=["__call__"], return_value=response_iter)
Expand Down

0 comments on commit 74e0b0f

Please sign in to comment.