Long time consuming operations in RedisCodec

[1ou]

Hello, I'm using lettuce as a client to cache storage. I store in Redis deserialized to bytes protobuf format objects which are zipped as well. The average size of value is 50kb. On a daily basis everything working charming, we're using Netty epoll event loop group for lettuce.
Lettuce creates 24 lettuce-epollEventLoop threads for itself in a 4 cores container.

Sometimes after high load at some pods we see through profiling that threads of lettuce-epollEventLoop pool are busy inside RedisCodec.encode(). I also see in async profiler that io.lettuce.core.protocol.SharedLock#doExclusive is called before RedisCodec.encode(). Am I correct that if operation in RedisCodec.encode() long time-consuming, then there is no progress in this lettuce connection with another reads from socket due to exclusive lock? Maybe it is good idea to move our logic of serialization/deserialization from codec to business level of our application (will be handled on another working pool). Will it increase the lettuce throughput?

This is how I say lettuce to use netty based working pool:

        return clientResourceBuilder
            .eventExecutorGroup(transport.getWorkerEventLoopGroup())
            .eventLoopGroupProvider(new NettyEventLoopGroupProvider(transport.getWorkerEventLoopGroup()))

I can see that only a few(5-6 threads from lettuce-epollEventLoop are 100% busy in RedisCodec.encode() others are not in use)

[mp911de]

Care to provide more details, such as your stack traces or profiling data? Without further information, we cannot investigate what your arrangement is.