When I want to restart the master node, it'll always cause the cluster unavailable for query and indexing requests for 3 seconds. Even though I have three master nodes, it still need 3 seconds to elect a new master. How can I restart the cluster without any influence on the request and indexing process?
Queries are served during this time, and indexing requests only block waiting for a new master, up to a timeout.
You can not, but as I explained above, queries should still be served, and indexing requests should only block briefly.