1. Wire and state serialization alone can easily cost a lot of your job’s performance if not executed correctly and thus, whenever you look into the profiler output of your Flink job, you will most likely see serialization in the top places for using CPU cycles.
2. 序列化器对比：https://www.cnblogs.com/lonelywolfmoutain/p/5563985.html
3. flink序列化框架实践https://blog.csdn.net/lixinkuan328/article/details/104174266