Fix: Add Comprehensive Retry Mechanism for Neo4j Storage Operations

[danielaskdd]

🐛 Fix: Add Comprehensive Retry Mechanism for Neo4j Storage Operations

Problem

Users were experiencing AttributeError: 'NoneType' object has no attribute 'send' errors during document processing with Neo4j storage backend. This error occurred in the merging stage when the connection pool encountered transient network issues or connection lifecycles, causing operations to fail with NoneType objects attempting to call the .send() method.

Error Stack Trace:

File "c:\develop\myenvs\lightrag-env\Lib\site-packages\neo4j\_async_compat\network\_bolt.py", line 196, in sendall
  self._write.write(data)
AttributeError: 'NoneType' object has no attribute 'send'

Solution

Implemented comprehensive retry logic using the tenacity library for all critical Neo4j database operations. The retry mechanism handles transient failures gracefully by automatically retrying failed operations with exponential backoff.

Changes Made

Modified Files

• lightrag/kg/neo4j_impl.py - Added retry decorators to 15 critical database operations

Functions Enhanced with Retry Logic

Read Operations (10 functions) - Retry on connection errors including AttributeError:

1. has_node - Check node existence
2. has_edge - Check edge existence
3. get_node - Get single node
4. get_nodes_batch - Batch get nodes
5. node_degree - Get node degree
6. node_degrees_batch - Batch get node degrees
7. get_edge - Get single edge
8. get_edges_batch - Batch get edges
9. get_node_edges - Get all edges for a node
10. get_nodes_edges_batch - Batch get node edges

Write Operations (5 functions) - Retry on write-specific errors:

11. upsert_node - Insert/update node
12. upsert_edge - Insert/update edge
13. delete_node - Delete node
14. remove_nodes - Batch delete nodes
15. remove_edges - Batch delete edges

Retry Configuration

Read Operations:

• Max attempts: 3
• Wait strategy: Exponential backoff (4-10 seconds)
• Retry on: ServiceUnavailable, TransientError, SessionExpired, ConnectionResetError, OSError, AttributeError

Write Operations:

• Max attempts: 3
• Wait strategy: Exponential backoff (4-10 seconds)
• Retry on: All read exceptions + WriteServiceUnavailable, ClientError

Benefits

✅ Improved Reliability - Automatically recovers from transient connection pool and network issues
✅ Production Ready - Handles connection lifecycle problems gracefully
✅ Zero Breaking Changes - Transparent retry logic with no API changes
✅ Better User Experience - Document processing continues successfully despite temporary failures
✅ Comprehensive Coverage - All critical database operations protected

Testing Recommendations

• Verify document processing completes successfully with Neo4j backend
• Test under network instability conditions
• Confirm retry logs appear in debug mode when transient failures occur
• Validate no performance degradation under normal operations

[danielaskdd]

@codex review