digitalkin 1.0.0.dev14__tar.gz → 1.0.0.dev16__tar.gz

{digitalkin-1.0.0.dev14 → digitalkin-1.0.0.dev16}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: digitalkin
-Version: 1.0.0.dev14
+Version: 1.0.0.dev16
 Summary: SDK to build kin used in DigitalKin
 Author-email: "DigitalKin.ai" <contact@digitalkin.ai>
 License: Attribution-NonCommercial-ShareAlike 4.0 International

{digitalkin-1.0.0.dev14 → digitalkin-1.0.0.dev16}/pyproject.toml

@@ -37,7 +37,7 @@
         "pydantic-settings>=2.14.1",
         "redis[hiredis]>=7.4.0,<8",
     ]
-    version = "1.0.0.dev14"
+    version = "1.0.0.dev16"
 
     [project.optional-dependencies]
         performance = [ "uvloop>=0.21" ]
@@ -305,5 +305,6 @@
         "smoke: marks critical path tests that should always pass",
         "stability: marks long-running stability and soak tests",
         "stress: marks stress/load tests for performance under pressure",
+        "unit: marks a plain unit test with no other applicable category",
         "validation: marks tests for input validation and schema checking",
     ]

{digitalkin-1.0.0.dev14 → digitalkin-1.0.0.dev16}/src/digitalkin/__version__.py

@@ -5,4 +5,4 @@ from importlib.metadata import PackageNotFoundError, version
 try:
     __version__ = version("digitalkin")
 except PackageNotFoundError:
-    __version__ = "1.0.0.dev14"
+    __version__ = "1.0.0.dev16"

{digitalkin-1.0.0.dev14 → digitalkin-1.0.0.dev16}/src/digitalkin/grpc_servers/utils/grpc_client_wrapper.py

@@ -48,6 +48,14 @@ class GrpcClientWrapper:
         grpc.StatusCode.DEADLINE_EXCEEDED,
     }
 
+    # Codes that count toward opening the circuit (service-health failures).
+    # Application-level codes (NOT_FOUND, INVALID_ARGUMENT, …) mean the service
+    # responded, so they never trip the breaker.
+    _CIRCUIT_FAILURE_CODES: ClassVar[set[grpc.StatusCode]] = _RETRYABLE_CODES | {
+        grpc.StatusCode.UNKNOWN,
+        grpc.StatusCode.RESOURCE_EXHAUSTED,
+    }
+
     @staticmethod
     def _build_channel_credentials(config: ClientConfig) -> grpc.ChannelCredentials | None:
         """Build SSL channel credentials from config if secure mode.
@@ -242,7 +250,18 @@ class GrpcClientWrapper:
                 response = await rpc_method(request, timeout=timeout)
             except grpc.RpcError as e:
                 last_error = e
-                if e.code() not in self._RETRYABLE_CODES or attempt == max_retries:
+                if e.code() in self._RETRYABLE_CODES and attempt < max_retries:
+                    logger.warning(
+                        "gRPC transient error on %s.%s [%s] (attempt %d/%d), retrying in %.0fms",
+                        self.service_name,
+                        query_endpoint,
+                        e.code().name,
+                        attempt + 1,
+                        max_retries + 1,
+                        backoff_delays[attempt] * 1000,
+                    )
+                    continue
+                if e.code() in self._CIRCUIT_FAILURE_CODES:
                     logger.warning(
                         "circuit-breaker tick: %s.%s [%s]",
                         self.service_name,
@@ -250,16 +269,9 @@ class GrpcClientWrapper:
                         e.code().name,
                     )
                     cb.record_failure()
-                    break
-                logger.warning(
-                    "gRPC transient error on %s.%s [%s] (attempt %d/%d), retrying in %.0fms",
-                    self.service_name,
-                    query_endpoint,
-                    e.code().name,
-                    attempt + 1,
-                    max_retries + 1,
-                    backoff_delays[attempt] * 1000,
-                )
+                else:
+                    cb.record_success()
+                break
             else:
                 cb.record_success()
                 return response

{digitalkin-1.0.0.dev14 → digitalkin-1.0.0.dev16}/src/digitalkin/services/storage/grpc_storage.py

@@ -4,6 +4,7 @@ from agentic_mesh_protocol.storage.v1 import data_pb2, storage_service_pb2_grpc
 from google.protobuf.struct_pb2 import Struct
 from pydantic import BaseModel
 
+from digitalkin.grpc_servers.exceptions import CircuitOpenError
 from digitalkin.grpc_servers.utils.grpc_client_wrapper import GrpcClientWrapper
 from digitalkin.logger import logger
 from digitalkin.models.grpc_servers.models import ClientConfig
@@ -21,6 +22,22 @@ class GrpcStorage(StorageStrategy, GrpcClientWrapper):
 
     service_name: str = "StorageService"
 
+    @staticmethod
+    def _is_circuit_open(error: Exception) -> bool:
+        """Whether ``error`` is a fast-fail from an open circuit breaker.
+
+        An open breaker is an expected, already-logged condition (the
+        CLOSED -> OPEN transition is logged once), so per-call rejections are
+        logged quietly to avoid flooding logs during an outage window.
+
+        Args:
+            error: The exception raised by ``exec_grpc_query``.
+
+        Returns:
+            True if the error's cause is a ``CircuitOpenError``.
+        """
+        return isinstance(error.__cause__, CircuitOpenError)
+
     def _build_record_from_proto(self, proto: data_pb2.StorageRecord) -> StorageRecord:
         """Convert a protobuf StorageRecord message into our Pydantic model.
 
@@ -83,11 +100,10 @@ class GrpcStorage(StorageStrategy, GrpcClientWrapper):
             resp = await self.exec_grpc_query("StoreRecord", req)
             return self._build_record_from_proto(resp.stored_data)
         except Exception as e:
-            logger.exception(
-                "gRPC StoreRecord failed for %s:%s",
-                record.collection,
-                record.record_id,
-            )
+            if self._is_circuit_open(e):
+                logger.debug("gRPC StoreRecord skipped (circuit open) for %s:%s", record.collection, record.record_id)
+            else:
+                logger.exception("gRPC StoreRecord failed for %s:%s", record.collection, record.record_id)
             raise StorageServiceError(str(e)) from e
 
     async def _read(self, collection: str, record_id: str) -> StorageRecord | None:
@@ -106,7 +122,10 @@ class GrpcStorage(StorageStrategy, GrpcClientWrapper):
             resp = await self.exec_grpc_query("ReadRecord", req)
             return self._build_record_from_proto(resp.stored_data)
         except Exception as e:
-            logger.info("gRPC ReadRecord failed for %s:%s: %s", collection, record_id, e)
+            if self._is_circuit_open(e):
+                logger.debug("gRPC ReadRecord skipped (circuit open) for %s:%s", collection, record_id)
+            else:
+                logger.info("gRPC ReadRecord failed for %s:%s: %s", collection, record_id, e)
             return None
 
     async def _update(
@@ -138,7 +157,10 @@ class GrpcStorage(StorageStrategy, GrpcClientWrapper):
             resp = await self.exec_grpc_query("UpdateRecord", req)
             return self._build_record_from_proto(resp.stored_data)
         except Exception as e:
-            logger.warning("gRPC UpdateRecord failed for %s:%s: %s", collection, record_id, e)
+            if self._is_circuit_open(e):
+                logger.debug("gRPC UpdateRecord skipped (circuit open) for %s:%s", collection, record_id)
+            else:
+                logger.warning("gRPC UpdateRecord failed for %s:%s: %s", collection, record_id, e)
             return None
 
     async def _remove(self, collection: str, record_id: str) -> bool:
@@ -160,12 +182,10 @@ class GrpcStorage(StorageStrategy, GrpcClientWrapper):
             )
             await self.exec_grpc_query("RemoveRecord", req)
         except Exception as e:
-            logger.warning(
-                "gRPC RemoveRecord failed for %s:%s: %s",
-                collection,
-                record_id,
-                e,
-            )
+            if self._is_circuit_open(e):
+                logger.debug("gRPC RemoveRecord skipped (circuit open) for %s:%s", collection, record_id)
+            else:
+                logger.warning("gRPC RemoveRecord failed for %s:%s: %s", collection, record_id, e)
             return False
         return True
 
@@ -187,7 +207,10 @@ class GrpcStorage(StorageStrategy, GrpcClientWrapper):
             resp = await self.exec_grpc_query("ListRecords", req)
             return [self._build_record_from_proto(r) for r in resp.records]
         except Exception as e:
-            logger.warning("gRPC ListRecords failed for %s: %s", collection, e)
+            if self._is_circuit_open(e):
+                logger.debug("gRPC ListRecords skipped (circuit open) for %s", collection)
+            else:
+                logger.warning("gRPC ListRecords failed for %s: %s", collection, e)
             return []
 
     async def _remove_collection(self, collection: str) -> bool:
@@ -206,7 +229,10 @@ class GrpcStorage(StorageStrategy, GrpcClientWrapper):
             )
             await self.exec_grpc_query("RemoveCollection", req)
         except Exception as e:
-            logger.warning("gRPC RemoveCollection failed for %s: %s", collection, e)
+            if self._is_circuit_open(e):
+                logger.debug("gRPC RemoveCollection skipped (circuit open) for %s", collection)
+            else:
+                logger.warning("gRPC RemoveCollection failed for %s: %s", collection, e)
             return False
         return True
 

{digitalkin-1.0.0.dev14 → digitalkin-1.0.0.dev16}/src/digitalkin.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: digitalkin
-Version: 1.0.0.dev14
+Version: 1.0.0.dev16
 Summary: SDK to build kin used in DigitalKin
 Author-email: "DigitalKin.ai" <contact@digitalkin.ai>
 License: Attribution-NonCommercial-ShareAlike 4.0 International