feat(publisher): degrade PubSubPublisher to noop when GCP credentials are absent

CHANGELOG.md

@@ -31,6 +31,7 @@ For more information about each release including git tags and artifacts, see [R
 - Event stream shows sensible defaults so first-load isn't empty ([#297](https://github.com/roostorg/osprey/pull/297) by [@haileyok](https://github.com/haileyok))
 - Replace `react-scripts` with `rsbuild`/`rspack` for UI builds ([#235](https://github.com/roostorg/osprey/pull/235) by [@chimosky](https://github.com/chimosky))
 - Migrate from npm to pnpm via Corepack ([#252](https://github.com/roostorg/osprey/pull/252) by [@haileyok](https://github.com/haileyok))
+- `PubSubPublisher` degrades to a noop when GCP credentials are absent, so Osprey runs without GCP config instead of failing ([#388](https://github.com/roostorg/osprey/pull/388) by [@julietshen](https://github.com/julietshen))
 
 ### Fixed
 

osprey_worker/src/osprey/worker/lib/data_exporters/test/conftest.py

@@ -14,12 +14,19 @@ def monkeypatch_session(request):
 
 
 @pytest.fixture(scope='function', autouse=True)
-def pubsub_client_mock(monkeypatch_session) -> MagicMock:
+def pubsub_client_mock(monkeypatch_session, monkeypatch) -> MagicMock:
     # tagging as potential opensource item
     from google.cloud import pubsub_v1
+    from osprey.worker.lib import publisher
 
     pubsub_client_mock = MagicMock()
     publisher_class_mock = MagicMock(return_value=pubsub_client_mock)
 
     monkeypatch_session.setattr(pubsub_v1, 'PublisherClient', publisher_class_mock)
+    # PubSubPublisher degrades to noop without GCP credentials, which CI lacks;
+    # force detection so the mocked client is actually exercised. Function-scoped
+    # monkeypatch so it reverts after each test instead of leaking into later
+    # modules (e.g. test_publisher.py, whose disabled-publisher tests would then
+    # build a real client and hang on stop()).
+    monkeypatch.setattr(publisher, '_check_gcp_credentials', lambda: True)
     return pubsub_client_mock

osprey_worker/src/osprey/worker/lib/publisher.py

@@ -1,13 +1,19 @@
 import abc
+import logging
+import threading
 from typing import TypeVar
 
+import google.auth
+from google.auth.exceptions import DefaultCredentialsError
 from google.cloud import pubsub_v1
 from osprey.worker.lib.instruments import metrics
 from osprey.worker.lib.pubsub.publisher_client import BatchPubsubPublisherClient
 from pydantic import BaseModel
 
 _PydanticModelT = TypeVar('_PydanticModelT', bound=BaseModel)
 
+logger = logging.getLogger(__name__)
+
 
 class BasePublisher(abc.ABC):
     """
@@ -40,6 +46,15 @@ def stop(self) -> None:
 
 
 class PubSubPublisher(BasePublisher):
+    """Publishes Pydantic models to a Google Cloud Pub/Sub topic.
+
+    Degrades to noop mode when GCP credentials cannot be resolved at construction
+    time (e.g. local dev or adopter environments without GCP). In noop mode no
+    underlying client is built, publish() and stop() return immediately, and a
+    `PubSubPublisher.publisher.noop` metric is incremented per call so the inert
+    state is visible in dashboards. A one-time warning is logged at construction.
+    """
+
     def __init__(
         self,
         project_id: str,
@@ -54,15 +69,23 @@ def __init__(
             topic_id=topic_id,
         )
         self._project = project_id
+        self._raise_on_error = raise_on_error
+        self._tags = [f'project:{self._project}', f'topic:{self._topic_name}']
+        self._enabled = _check_gcp_credentials()
+        if not self._enabled:
+            logger.warning(
+                'GCP credentials not detected, PubSubPublisher running in noop mode (project=%s, topic=%s)',
+                project_id,
+                topic_id,
+            )
+            return
         batch_settings = pubsub_v1.types.BatchSettings(
             max_bytes=max_bytes,  # default 1MB
             max_messages=max_messages,  # default 100 messages
             max_latency=max_latency,
         )
         # self._publisher = pubsub_v1.PublisherClient(batch_settings=batch_settings)
         self._publisher = BatchPubsubPublisherClient(batch_settings=batch_settings)
-        self._raise_on_error = raise_on_error
-        self._tags = [f'project:{self._project}', f'topic:{self._topic_name}']
 
     def prepare_data(self, data: _PydanticModelT) -> bytes:
         """
@@ -72,6 +95,9 @@ def prepare_data(self, data: _PydanticModelT) -> bytes:
         return data.json(exclude_none=True).encode()
 
     def publish(self, data: _PydanticModelT, attributes: dict[str, str] | None = None) -> None:
+        if not self._enabled:
+            metrics.increment(f'{self.__class__.__name__}.publisher.noop', tags=self._tags)
+            return
         if attributes is None:
             attributes = {}
 
@@ -89,6 +115,8 @@ def publish(self, data: _PydanticModelT, attributes: dict[str, str] | None = Non
         metrics.increment(f'{self.__class__.__name__}.publisher.success', tags=self._tags)
 
     def stop(self) -> None:
+        if not self._enabled:
+            return
         self._publisher.stop()
 
 
@@ -103,3 +131,21 @@ def publish(self, data: str, attributes: dict[str, str] | None = None) -> None:
             attributes = {}
 
         super().publish(data, attributes)  # type: ignore[type-var]
+
+
+_gcp_credentials_available: bool | None = None
+_gcp_credentials_lock = threading.Lock()
+
+
+def _check_gcp_credentials() -> bool:
+    global _gcp_credentials_available
+    if _gcp_credentials_available is None:
+        with _gcp_credentials_lock:
+            # Re-check under the lock so concurrent constructors probe google.auth.default() only once.
+            if _gcp_credentials_available is None:
+                try:
+                    google.auth.default()
+                    _gcp_credentials_available = True
+                except DefaultCredentialsError:
+                    _gcp_credentials_available = False
+    return _gcp_credentials_available

osprey_worker/src/osprey/worker/lib/tests/test_publisher.py

@@ -0,0 +1,76 @@
+from typing import Iterator
+from unittest.mock import MagicMock, patch
+
+import pytest
+from google.auth.exceptions import DefaultCredentialsError
+from osprey.worker.lib import publisher
+from osprey.worker.lib.publisher import PubSubPublisher, _check_gcp_credentials
+
+
+@pytest.fixture(autouse=True)
+def reset_cred_cache() -> Iterator[None]:
+    publisher._gcp_credentials_available = None
+    yield
+    publisher._gcp_credentials_available = None
+
+
+def test_check_gcp_credentials_true_when_default_resolves() -> None:
+    with patch.object(publisher.google.auth, 'default', return_value=(MagicMock(), 'proj')):
+        assert _check_gcp_credentials() is True
+
+
+def test_check_gcp_credentials_false_when_default_raises() -> None:
+    with patch.object(publisher.google.auth, 'default', side_effect=DefaultCredentialsError()):
+        assert _check_gcp_credentials() is False
+
+
+def test_check_gcp_credentials_is_cached() -> None:
+    with patch.object(publisher.google.auth, 'default', return_value=(MagicMock(), 'proj')) as default_mock:
+        _check_gcp_credentials()
+        _check_gcp_credentials()
+        _check_gcp_credentials()
+    assert default_mock.call_count == 1
+
+
+def test_pubsub_publisher_constructs_client_when_creds_present() -> None:
+    with (
+        patch.object(publisher.google.auth, 'default', return_value=(MagicMock(), 'proj')),
+        patch.object(publisher, 'BatchPubsubPublisherClient') as client_cls,
+    ):
+        pub = PubSubPublisher('proj', 'topic')
+    assert pub._enabled is True
+    assert client_cls.called
+
+
+def test_pubsub_publisher_noops_when_creds_absent(caplog: pytest.LogCaptureFixture) -> None:
+    with (
+        patch.object(publisher.google.auth, 'default', side_effect=DefaultCredentialsError()),
+        patch.object(publisher, 'BatchPubsubPublisherClient') as client_cls,
+    ):
+        with caplog.at_level('WARNING', logger=publisher.logger.name):
+            pub = PubSubPublisher('proj', 'topic')
+    assert pub._enabled is False
+    assert not client_cls.called
+    assert 'noop mode' in caplog.text
+    assert 'project=proj' in caplog.text
+    assert 'topic=topic' in caplog.text
+
+
+def test_publish_short_circuits_and_emits_noop_metric_when_disabled() -> None:
+    with (
+        patch.object(publisher.google.auth, 'default', side_effect=DefaultCredentialsError()),
+        patch.object(publisher, 'metrics') as metrics_mock,
+    ):
+        pub = PubSubPublisher('proj', 'topic')
+        pub.publish(MagicMock())
+    metrics_mock.increment.assert_called_once_with(
+        'PubSubPublisher.publisher.noop',
+        tags=['project:proj', 'topic:projects/proj/topics/topic'],
+    )
+
+
+def test_stop_short_circuits_when_disabled() -> None:
+    with patch.object(publisher.google.auth, 'default', side_effect=DefaultCredentialsError()):
+        pub = PubSubPublisher('proj', 'topic')
+        pub.stop()
+    assert not hasattr(pub, '_publisher')