Implementation:LMCache LMCache Audit Connector

Knowledge Sources	LMCache
Domains	Caching, Observability, Storage Connectors
Last Updated	2026-02-09 00:00 GMT

Overview

AuditConnector is a transparent auditing wrapper for RemoteConnector that dynamically intercepts and logs all connector method calls with timing, checksums, and error tracking.

Description

The AuditConnector uses a custom metaclass (AuditConnectorMeta) to automatically generate wrapper methods for every method defined on the RemoteConnector interface. Each wrapped method logs the start, success (with elapsed time in milliseconds), or failure of the operation. Methods decorated with @NotAudit are forwarded to the underlying connector without any logging overhead. The connector also supports optional SHA-256 checksum calculation and verification for put and get operations, controlled via the audit_calc_checksum and audit_verify_checksum configuration flags. A configurable set of excluded commands can bypass auditing as well.

Usage

Use AuditConnector when you need to diagnose remote storage performance issues, verify data integrity between put and get operations, or gather detailed operational logs for remote connector interactions. It is typically instantiated by wrapping an existing RemoteConnector instance and is configured through the LMCacheEngineConfig.extra_config dictionary.

Code Reference

Source Location

• Repository: LMCache
• File: lmcache/v1/storage_backend/connector/audit_connector.py
• Lines: 1-320

Signature

class AuditConnectorMeta(abc.ABCMeta):
    def __new__(mcs, name, bases, namespace): ...

class AuditConnector(RemoteConnector, metaclass=AuditConnectorMeta):
    def __init__(
        self, real_connector: RemoteConnector, lmcache_config: LMCacheEngineConfig
    ): ...
    def _calculate_checksum(self, data: bytes) -> str: ...
    async def _audit_put(self, key: CacheEngineKey, memory_obj: MemoryObj): ...
    async def _audit_get(self, key: CacheEngineKey) -> Optional[MemoryObj]: ...

Import

from lmcache.v1.storage_backend.connector.audit_connector import AuditConnector

I/O Contract

Inputs

Name	Type	Required	Description
real_connector	RemoteConnector	Yes	The underlying remote connector whose methods will be audited
lmcache_config	LMCacheEngineConfig	Yes	Engine configuration containing extra_config for audit settings (audit_verify_checksum, audit_calc_checksum, audit_exclude_cmds)

Outputs

Name	Type	Description
AuditConnector	RemoteConnector	A wrapper that transparently delegates all RemoteConnector methods while logging performance and optionally verifying checksums

Usage Examples

from lmcache.v1.storage_backend.connector.audit_connector import AuditConnector
from lmcache.v1.storage_backend.connector.redis_connector import RedisConnector

# Wrap an existing connector with auditing
redis_connector = RedisConnector(url, loop, local_cpu_backend)
audit_connector = AuditConnector(redis_connector, lmcache_config)

# All operations are now audited with timing and optional checksums
await audit_connector.put(key, memory_obj)  # Logs PUT|START, PUT|SUCCESS|Cost:X.XXms
result = await audit_connector.get(key)      # Logs GET|START, GET|SUCCESS or GET|MISMATCH