The process is almost the same as bootstrapping.
The leaving node state transits from NORMAL to LEAVING and finally to LEFT.
It waits for the ring delay as part of each state transition in order to
propagate the entire cluster.
Pending ranges are updated.
In the case of leaving, there will be
Thank you Yifan for the details. I have a related question: when we issue
a command to remove a node A from the ring, there could be a time that
some node B thinks node A is removed, but some node C still thinks node A
is in the ring and could reach node A. What happens if node C sends a
write
