Thanks for Daniel and Soumya!
Putting whole cluster into grace period, will it cause performance issue?
Is it OK that we save all the session informations into backend. When ServerA crashed and
client connect ServerB, ServerB can resume session from backend including open and locking
resource.