see on GitHub

Cluster : Cluster Node Health Check

A health check sends an HTTP request to the /databases/[Database Name]/stats endpoint. If the request is successful, it will reset node failure counters which will cause the client to try sending operations to that specific node again.

When Does it Trigger?

Any time a low-level operation fails to connect to a node, the client spawns a health check thread for that particular node. The thread will periodically ping the not responding server until it gets a proper response. The frequency of pinging the non-responsive server will start from 100ms and will gradually increase until it reaches 5sec intervals.