Reset sync client long-poll failure tracking

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Author: Bernd Verst

CHANGELOG.md

@@ -22,8 +22,11 @@ ADDED
 - Added optional `resiliency_options` parameters to `TaskHubGrpcClient`,
   `AsyncTaskHubGrpcClient`, and `TaskHubGrpcWorker` so applications can pass
   gRPC resiliency settings through constructor APIs.
-- Added `get_orchestration_history()` and `list_instance_ids()` to the sync and async gRPC clients.
-- Added in-memory backend support for `StreamInstanceHistory` and `ListInstanceIds` so local orchestration tests can retrieve history and page terminal instance IDs by completion window.
+- Added `get_orchestration_history()` and `list_instance_ids()` to the sync
+  and async gRPC clients.
+- Added in-memory backend support for `StreamInstanceHistory` and
+  `ListInstanceIds` so local orchestration tests can retrieve history and page
+  terminal instance IDs by completion window.
 
 FIXED
 
@@ -32,8 +35,9 @@ FIXED
   SDK-owned channels are cleaned up on shutdown and full resets, and
   caller-owned channels are never recreated or closed during worker reconnects.
 - Fixed sync `TaskHubGrpcClient` transport resiliency so SDK-owned channels are
-  recreated after repeated transport failures without counting long-poll
-  timeout deadlines against the recreation threshold.
+  recreated after repeated transport failures while long-poll timeout
+  deadlines, successful replies, and application-level RPC errors reset the
+  failure tracker.
 
 ## v1.4.0
 

durabletask/client.py

@@ -234,7 +234,7 @@ def _invoke_unary(
                 should_recreate = self._client_failure_tracker.record_failure()
                 if should_recreate:
                     self._maybe_recreate_channel()
-            elif status_code != grpc.StatusCode.DEADLINE_EXCEEDED:
+            else:
                 self._client_failure_tracker.record_success()
             raise
         else:

tests/durabletask/test_client.py

@@ -359,10 +359,20 @@ def test_sync_client_recreates_sdk_owned_channel_after_repeated_unavailable():
     timer.start.assert_called_once_with()
 
 
-def test_sync_client_does_not_count_long_poll_deadline():
+@pytest.mark.parametrize(
+    ("stub_method_name", "client_method_name"),
+    [
+        ("WaitForInstanceStart", "wait_for_orchestration_start"),
+        ("WaitForInstanceCompletion", "wait_for_orchestration_completion"),
+    ],
+)
+def test_sync_client_resets_failure_tracking_after_long_poll_deadline(
+        stub_method_name: str,
+        client_method_name: str,
+):
     stub = MagicMock()
     stub.GetInstance.side_effect = FakeRpcError(grpc.StatusCode.UNAVAILABLE)
-    stub.WaitForInstanceStart.side_effect = FakeRpcError(grpc.StatusCode.DEADLINE_EXCEEDED)
+    getattr(stub, stub_method_name).side_effect = FakeRpcError(grpc.StatusCode.DEADLINE_EXCEEDED)
 
     with patch("durabletask.client.shared.get_grpc_channel", return_value=MagicMock()), patch(
             "durabletask.client.stubs.TaskHubSidecarServiceStub", return_value=stub
@@ -373,8 +383,8 @@ def test_sync_client_does_not_count_long_poll_deadline():
         with pytest.raises(FakeRpcError):
             client.get_orchestration_state("abc")
         with pytest.raises(TimeoutError):
-            client.wait_for_orchestration_start("abc")
-        assert client._client_failure_tracker.consecutive_failures == 1
+            getattr(client, client_method_name)("abc")
+        assert client._client_failure_tracker.consecutive_failures == 0
 
 
 def test_sync_client_does_not_recreate_caller_owned_channel():