nebu 0.1.77__tar.gz → 0.1.78__tar.gz

{nebu-0.1.77/src/nebu.egg-info → nebu-0.1.78}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: nebu
-Version: 0.1.77
+Version: 0.1.78
 Summary: A globally distributed container runtime
 Requires-Python: >=3.10.14
 Description-Content-Type: text/markdown

{nebu-0.1.77 → nebu-0.1.78}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "nebu"
-version = "0.1.77"
+version = "0.1.78"
 description = "A globally distributed container runtime"
 readme = "README.md"
 requires-python = ">=3.10.14"

nebu-0.1.78/src/nebu/errors.py

@@ -0,0 +1,5 @@
+# Add this near the top with other imports/definitions
+class RetriableError(Exception):
+    """Custom exception for failures during Training resource setup."""
+
+    pass

{nebu-0.1.77 → nebu-0.1.78}/src/nebu/processors/consumer.py

@@ -16,6 +16,8 @@ import redis
 import socks
 from redis import ConnectionError, ResponseError
 
+from nebu.errors import RetriableError
+
 # Define TypeVar for generic models
 T = TypeVar("T")
 
@@ -433,6 +435,13 @@ def process_message(message_id: str, message_data: Dict[str, str]) -> None:
                     f"Subprocess for {message_id} completed successfully (return code 0)."
                 )
                 # Assume success handling (ack/response) was done by the worker
+            elif return_code == 3:
+                print(
+                    f"Subprocess for {message_id} reported a retriable error (exit code 3). Message will not be acknowledged."
+                )
+                # Optionally send an error response here, though the worker already did.
+                # _send_error_response(...)
+                # DO NOT Acknowledge the message here, let it be retried.
             else:
                 print(
                     f"Subprocess for {message_id} failed with exit code {return_code}."
@@ -550,6 +559,16 @@ def process_message(message_id: str, message_data: Dict[str, str]) -> None:
             None,
             None,
         )  # Pass None for user_id if unavailable here
+        # Acknowledge message with code load failure to prevent reprocessing loop
+        try:
+            assert isinstance(REDIS_STREAM, str)
+            assert isinstance(REDIS_CONSUMER_GROUP, str)
+            r.xack(REDIS_STREAM, REDIS_CONSUMER_GROUP, message_id)
+            print(f"Acknowledged message {message_id} due to code load failure.")
+        except Exception as e_ack:
+            print(
+                f"CRITICAL: Failed to acknowledge message {message_id} after code load failure: {e_ack}"
+            )
         return  # Skip processing
 
     return_stream = None
@@ -797,6 +816,15 @@ def process_message(message_id: str, message_data: Dict[str, str]) -> None:
         assert isinstance(REDIS_CONSUMER_GROUP, str)
         r.xack(REDIS_STREAM, REDIS_CONSUMER_GROUP, message_id)
 
+    except RetriableError as e:
+        print(f"Retriable error processing message {message_id}: {e}")
+        traceback.print_exc()
+        _send_error_response(
+            message_id, str(e), traceback.format_exc(), return_stream, user_id
+        )
+        # DO NOT Acknowledge the message for retriable errors
+        print(f"Message {message_id} will be retried later.")
+
     except Exception as e:
         print(f"Error processing message {message_id}: {e}")
         traceback.print_exc()

{nebu-0.1.77 → nebu-0.1.78}/src/nebu/processors/consumer_process_worker.py

@@ -14,6 +14,8 @@ from typing import Any, Callable, Dict, List, Optional, Tuple, TypeVar, cast
 import redis
 import socks
 
+from nebu.errors import RetriableError
+
 # from redis import ConnectionError, ResponseError # Removed unused imports
 
 # Define TypeVar for generic models
@@ -640,8 +642,24 @@ if __name__ == "__main__":
             print("[Worker] Exiting with status 0.")
             sys.exit(0)
 
+        except RetriableError as e:
+            # --- Handle Retriable Processing Error ---
+            print(f"[Worker] Retriable error processing message {message_id}: {e}")
+            tb = traceback.format_exc()
+            print(tb)
+            # Assert message_id is str before sending error
+            assert isinstance(message_id, str)
+            # Send error response (optional, consider suppressing later if too noisy)
+            _send_error_response(message_id, str(e), tb, return_stream, user_id)
+
+            # DO NOT Acknowledge the message for retriable errors
+
+            # --- 9. Exit with specific code for retriable failure ---
+            print("[Worker] Exiting with status 3 due to retriable error.")
+            sys.exit(3)
+
         except Exception as e:
-            # --- Handle Processing Error ---
+            # --- Handle Non-Retriable Processing Error ---
             print(f"[Worker] Error processing message {message_id}: {e}")
             tb = traceback.format_exc()
             print(tb)

{nebu-0.1.77 → nebu-0.1.78/src/nebu.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: nebu
-Version: 0.1.77
+Version: 0.1.78
 Summary: A globally distributed container runtime
 Requires-Python: >=3.10.14
 Description-Content-Type: text/markdown

{nebu-0.1.77 → nebu-0.1.78}/src/nebu.egg-info/SOURCES.txt

@@ -6,6 +6,7 @@ src/nebu/auth.py
 src/nebu/cache.py
 src/nebu/config.py
 src/nebu/data.py
+src/nebu/errors.py
 src/nebu/meta.py
 src/nebu/orign.py
 src/nebu.egg-info/PKG-INFO