feat: Add telemetry event for uncaught exceptions (#203)

Signed-off-by: Caden Marofke <marofke@amazon.com>

README.md

@@ -14,6 +14,18 @@ to interact with AWS Deadline Cloud and perform tasks.
 
 See [CONTRIBUTING](CONTRIBUTING.md#security-issue-notifications) for more information.
 
+## Telemetry
+
+This library collects telemetry data by default. Telemetry events contain non-personally-identifiable information that helps us understand how users interact with our software so we know what features our customers use, and/or what existing pain points are.
+
+You can opt out of telemetry data collection by either:
+
+1. Setting the environment variable: `DEADLINE_CLOUD_TELEMETRY_OPT_OUT=true`
+2. Providing the installer flag: `--telemetry-opt-out`
+3. Setting the config file: `deadline config set telemetry.opt_out true`
+
+Note that setting the environment variable supersedes the config file setting.
+
 ## License
 
 This project is licensed under the Apache-2.0 License.

pyproject.toml

@@ -8,7 +8,7 @@ dynamic = ["version"]
 dependencies = [
     "requests ~= 2.31",
     "boto3 >= 1.28.80",
-    "deadline == 0.40.*",
+    "deadline == 0.41.*",
     "openjd-sessions == 0.7.*",
     # tomli became tomllib in standard library in Python 3.11
     "tomli == 2.0.* ; python_version<'3.11'",

src/deadline_worker_agent/aws/deadline/__init__.py

@@ -12,6 +12,8 @@
 
 from deadline.client.api import get_telemetry_client, TelemetryClient
 from deadline.job_attachments.progress_tracker import SummaryStatistics
+from openjd.model import version as openjd_model_version
+from openjd.sessions import version as openjd_sessions_version
 
 from ..._version import __version__ as version  # noqa
 from ...startup.config import Configuration
@@ -35,6 +37,8 @@
     LOG_CONFIG_OPTION_STREAM_NAME_KEY,
 )
 
+__cached_telemetry_client: Optional[TelemetryClient] = None
+
 _logger = logging.getLogger(__name__)
 
 
@@ -792,7 +796,18 @@ def update_worker_schedule(
 
 def _get_deadline_telemetry_client() -> TelemetryClient:
     """Wrapper around the Deadline Client Library telemetry client, in order to set package-specific information"""
-    return get_telemetry_client("deadline-cloud-worker-agent", version)
+    global __cached_telemetry_client
+    if not __cached_telemetry_client:
+        __cached_telemetry_client = get_telemetry_client(
+            "deadline-cloud-worker-agent", ".".join(version.split(".")[:3])
+        )
+        __cached_telemetry_client.update_common_details(
+            {"openjd-sessions-version": ".".join(openjd_sessions_version.split(".")[:3])}
+        )
+        __cached_telemetry_client.update_common_details(
+            {"openjd-model-version": ".".join(openjd_model_version.split(".")[:3])}
+        )
+    return __cached_telemetry_client
 
 
 def record_worker_start_telemetry_event(capabilities: Capabilities) -> None:
@@ -802,6 +817,13 @@ def record_worker_start_telemetry_event(capabilities: Capabilities) -> None:
     )
 
 
+def record_uncaught_exception_telemetry_event(exception_type: str) -> None:
+    """Calls the telemetry client to record an event signaling an uncaught exception occurred."""
+    _get_deadline_telemetry_client().record_error(
+        event_details={"exception_scope": "uncaught"}, exception_type=exception_type
+    )
+
+
 def record_sync_inputs_telemetry_event(queue_id: str, summary: SummaryStatistics) -> None:
     """Calls the telemetry client to record an event capturing the sync-inputs summary."""
     details: Dict[str, Any] = asdict(summary)

src/deadline_worker_agent/startup/entrypoint.py

@@ -34,6 +34,7 @@
     update_worker,
     update_worker_schedule,
     record_worker_start_telemetry_event,
+    record_uncaught_exception_telemetry_event,
 )
 
 __all__ = ["entrypoint"]
@@ -161,6 +162,7 @@ def filter(self, record: logging.LogRecord) -> bool:
             raise
         else:
             _logger.critical(e, exc_info=True)
+            record_uncaught_exception_telemetry_event(exception_type=str(type(e)))
             sys.exit(1)
     finally:
         _logger.info("Worker Agent exiting")

test/unit/aws/deadline/test_client_telemetry.py

@@ -8,6 +8,7 @@
     record_worker_start_telemetry_event,
     record_sync_inputs_telemetry_event,
     record_sync_outputs_telemetry_event,
+    record_uncaught_exception_telemetry_event,
 )
 from deadline_worker_agent.startup.capabilities import Capabilities
 from deadline.job_attachments.progress_tracker import SummaryStatistics
@@ -125,3 +126,25 @@ def test_record_sync_outputs_telemetry_event():
             "queue_id": "queue-test",
         },
     )
+
+
+def test_record_uncaught_exception_telemetry_event():
+    """
+    Tests that when record_uncaught_exception_telemetry_event() is called, the correct
+    event type and details are passed to the telemetry client's record_event() method.
+    """
+    # GIVEN
+    mock_telemetry_client = MagicMock()
+
+    with patch.object(deadline_mod, "_get_deadline_telemetry_client") as mock_get_telemetry_client:
+        mock_get_telemetry_client.return_value = mock_telemetry_client
+
+        # WHEN
+        error = ValueError()
+        record_uncaught_exception_telemetry_event(str(type(error)))
+
+    # THEN
+    mock_telemetry_client.record_error.assert_called_with(
+        exception_type="<class 'ValueError'>",
+        event_details={"exception_scope": "uncaught"},
+    )

test/unit/startup/test_entrypoint.py

@@ -186,9 +186,11 @@ def test_calls_worker_run(
     mock_worker_run.assert_called_once_with()
 
 
+@patch.object(entrypoint_mod, "record_uncaught_exception_telemetry_event")
 @patch.object(entrypoint_mod.sys, "exit")
 def test_worker_run_exception(
     sys_exit_mock: MagicMock,
+    telemetry_mock: MagicMock,
     mock_worker_run: MagicMock,
 ) -> None:
     """Tests that exceptions raised by Worker.run() are logged and the program exits with a non-zero exit code"""
@@ -205,6 +207,7 @@ def test_worker_run_exception(
     logger_exception: MagicMock = logger.exception
     logger_exception.assert_called_once_with("Failed running worker: %s", exception)
     sys_exit_mock.assert_called_once_with(1)
+    telemetry_mock.assert_called_once_with(exception_type=str(type(exception)))
 
 
 def test_configuration_load(