bug(graph): One transient failure permanently disables FalkorDB for the whole process #1388

core/__init__.py:24, 51-59, 174, 238; core/database_falkordb.py:84, 165-168, 234, 237

_FALKORDB_DISABLED = False
def mark_falkordb_unavailable() -> None:
    global _FALKORDB_DISABLED
    _FALKORDB_DISABLED = True     # never reset

class FalkorDBManager:
    _startup_failed = False       # class attribute, never reset

Reproduced. A failing path followed by the known-good global path in the same process:

1st manager: KuzuDBManager
_FALKORDB_DISABLED: True
2nd manager (should be FalkorDB): KuzuDBManager  /home/claude/.codegraphcontext/global/db/kuzudb

Impact. In the long-running MCP server, switch_context_tool (server.py:517, 583) switching to a per-repo context whose FalkorDB can't start poisons the flag. Every subsequent context — including switching back to the working global one — silently returns a KuzuDBManager. The user's indexed graph becomes invisible and all further writes go to a second store.

Fix. Scope the failure flag to the specific (db_path, socket_path) pair and clear it on close_driver(teardown=True) / context switch.

Related existing issues

• fix remote FalkorDB reconnect after failed first connect #1222 — the remote-FalkorDB analogue: driver state left set after a failed first connect

Filed separately because the root cause and fix differ; close as a duplicate if maintainers judge otherwise.

_{Audit ID G-H03 · severity high · from an end-to-end audit of CGC 0.5.2 (63 findings). Reported with the reproduction as run.}