With round robin, the Gateway rotates through the server list on a request-by-request basis (round-robin) from the first server, to the second server, and so on. When the end of the server list is reached, the cycle continues from the top of the list. In order to determine that say server 2 is down, a request will need to be sent to it. Depending on how server 2 is actually down, the request may be sent multiple times based on the specified number of retries (3) first before attempting the next server on the list.
Additionally, the cluster property io.failoverServerRetryDelay controls the delay before the Gateway retries a failed server.
The default is to wait 5 minutes when using the "Round Robin" strategy.