Description
When a node fails, the channel is marked as inactive and the endpoint marks itself for reconnection once. If said reconnection fails, the generic handling is to schedule a reconnection retry.
During this connection phase however, a few netty errors are caught which bypasses the scheduling of a retry (for the generic error case). Since 1.2.1, this includes the handling of a socket connection timeout.
Effectively, the endpoint never recovers from such a situation where the node is unreachable, because Netty's socket timeout stops it from retrying the connection.