Codis proxy resp PONG with large delay（Codis proxy返回PONG并且有巨大延迟）

Hi，all
I found a problem with Codis-proxy with redis-benchmark. I have a 3-node pika cluster and single proxy, if I run benchmark to proxy with high pressure, performance will drop dramatically, and then use redis-cli to connect the proxy, I find GET command will return PONG message, and I'm trying to show proxy processing log, I find response message of command GET is mix up to the command PING. It seems like a multi-thread problem.
My pika cluster has one master and two slaves, sharding mode with 16 slots, and consensus-level = 1, proxy config only change backend_primary_only = true and backend_primary_parallel = 32.
What's more, I try to reduce cluster slots to 1 and proxy parallel to 1, and has no PONG problem. If I increase the proxy parallel or increase the slot number, PONG issue occur. I also set pika consensus-level = 0, and proxy work fine and performance is greatly improved.
我发现Codis代理使用redis-benchmark压测时存在问题。 我有一个3节点的pika群集和单个代理，如果我对Proxy进行高强度压测，性能将急剧下降，然后使用redis-cli连接代理，我发现GET命令将返回PONG消息，并且我 在尝试显示代理处理日志时，我发现命令GET的响应消息与命令PING混淆。 这似乎是一个多线程问题。
我的pika集群有一个主设备和两个从设备，sharding模式，具有16个分片，并且consensus-level= 1，代理配置仅更改backend_primary_only = true和backend_primary_parallel = 32。
此外，我还尝试将群集插槽减少到1个，并将代理并行减少到1个，并且没有出现PONG问题。 如果我增加代理并行或增加插槽号，则会发生PONG问题。另外我还测试了consensus-level = 0的场景，发现proxy运行稳定，并且性能优异。