feat: PolarDB graph backend with decomposed mixin architecture

[anatolykoptev]

Summary

Overhaul of the graph_dbs/ module:

• PolarDB backend: decomposed from a 4200-line monolith into a mixin-based polardb/ package (11 files, ~4300 lines). Each mixin handles a logical concern (connection, nodes, edges, search, filters, etc.)
• Shared utilities (utils.py): extracted common helpers used across backends
• BaseGraphDB extended with 8 new abstract methods for full-text search, tag-based neighbors, edge queries, and user management
• Neo4j/Nebula backends updated to implement new abstract methods
• Bug fix: self.execute_query() in count_nodes() didn't exist — replaced with standard cursor pattern

PolarDB package structure

File	Lines	Responsibility
__init__.py	30	Composes all mixins into PolarDBGraphDB
connection.py	334	Pool management, health checks, retries
schema.py	172	AGE extensions, tables, indexes
nodes.py	715	Node CRUD + agtype parsing
edges.py	267	Edge CRUD
filters.py	582	Dual SQL/Cypher WHERE clause builders
search.py	361	Keyword, fulltext, embedding search
traversal.py	432	Neighbors, subgraph, paths
queries.py	658	Metadata queries, counts
maintenance.py	769	Import/export, clear, cleanup
helpers.py	14	Standalone utility functions

Import path unchanged

from memos.graph_dbs.polardb import PolarDBGraphDB  # works as before

Test plan

• All 31 BaseGraphDB abstract methods implemented in PolarDB (verified via AST)
• No cross-mixin import violations (mixins use self. for cross-calls)
• Existing test suite passes across all backends
• PolarDB search operations work end-to-end

🤖 Generated with Claude Code

[wustzdy]

Thank you for your optimization and updates to the underlying PolarDB code in the memos open-source project. Currently, PolarDB utilizes two plugins: polardb_age and vector. While vector is open-source, polardb_age is built upon Apache AGE with significant enhancements and upgrades. For instance, it supports graph_id, which serves as a bridge for connecting edges between different nodes.

As you can see from the polardb_age GitHub repository page, this plugin indeed incorporates numerous optimizations and functional extensions based on Apache AGE. It introduces new functions such as graph_i, graph_s, and graph_p, along with more efficient parallel processing and multi-graph support, among other features.

You can proceed with optimization and upgrades in postgres.py. The combination of pg_vector, Apache AGE, and multi-table joins is capable of supporting some fundamental functionalities to meet requirements.

[wustzdy]

Thank you very much for your support of the open-source database. Currently, the underlying layers of PolarDB have made many optimizations and improvements specifically for the Age and Vector plugins. You are welcome to optimize and improve PostgreSQL under this project. If you have any better ideas, let's work together to make improvements!