It doesn't look like the library is setting any TCP keep alive settings on the connection. I confirmed this by checking netstat -apno on my application. It appears the etcdctl command line utility sets pretty aggressive TCP keep alive settings by default.

  --keepalive-time=2s                       keepalive time for client connections
  --keepalive-timeout=6s                    keepalive timeout for client connections

I need similarly aggressive settings when using the etcd-cpp-apiv3 as I need to detect connection issues quickly so that I can failover to a different etcd node before my lease expires and continue to keep that lease alive. I'm not familiar with GRPC but it looks like it may have some TCP keep alive configuration settings. Maybe it's just configuration you need to expose and pass onto GRPC?