Leverage Atomic for the in-memory states in Scheduler

[yahoNanJing]

Is your feature request related to a problem or challenge? Please describe what you are trying to do.

Currently we leverage read write lock for the in-memory states in scheduler in global level, which is too coarse. When tens of thousands of states need to be updated, it will downgrade the whole system performance because of the write lock.

Describe the solution you'd like

Therefore, we propose to introduce Atomic for the states to change the write lock to the read lock.

Describe alternatives you've considered

Additional context